text
stringlengths 4
2.52M
| meta
dict |
|---|---|
1.. Introduction
================
For couples living in socially disorganized neighborhoods, alcohol outlets can act with neighborhood conditions to increase their risks for intimate partner violence (IPV). Greater numbers of alcohol outlets within a neighborhood may be a sign of loosened normative constraints against violence; promote problem alcohol use among at-risk couples; and provide environments where groups of persons at risk for IPV may form and mutually reinforce IPV-related attitudes, norms, and problem behaviors. This paper begins with an introduction to IPV research and the role of drinking, and then reviews the theoretical and empirical literatures relevant to identifying specific social mechanisms linking IPV to alcohol use in community settings. Understanding these mechanisms is of critical public health importance for developing environmental strategies aimed at IPV prevention, such as changes in zoning, community action and education, and policing.
2.. Methods
===========
The PubMed database was searched (1980 to 2009) for epidemiological studies related to intimate partner violence, neighborhoods, alcohol outlets, drinking, and social disorganization. Bibliographies of certain articles provided additional papers. Articles were screened for their relevance to the specific topic of neighborhoods, alcohol outlets, and intimate partner violence on the basis of the title and abstract.
3.. Intimate Partner Violence
=============================
3.1.. Definition
----------------
The American Psychological Association Presidential Task Force on Violence and the Family defines domestic violence as "...the physical, sexual, and emotional maltreatment of one family member by another" \[[@b1-ijerph-07-00799]\]. While the term *domestic violence* typically encompasses all types of family violence, including elder abuse, marital rape, child sexual and physical abuse, and child psychological maltreatment, the term *intimate partner violence* (IPV) refers to those acts of aggression between adult married or cohabiting intimate partners. Aggression may occur in many ways. Psychological aggression (coercive verbal and nonverbal behaviors that are not directed at the partner's body, such as slamming doors or smashing objects) has been found to predict physical aggression in longitudinal studies of married couples \[[@b2-ijerph-07-00799],[@b3-ijerph-07-00799]\]. Physical aggression, including sexual coercion, refers to coercive acts directed at the partner's body that may or may not cause injury. This paper focuses on physical aggression between intimate partners.
3.2.. Prevalence & Consequences
-------------------------------
IPV remains a significant public health problem. Based on a national probability sample of married or cohabiting couples that participated in the 1995 National Alcohol Survey, annual prevalence estimates for any partner-to-partner violence (*i.e.*, male-to-female or female-to-male) ranged from 7.8% to 21.5% \[[@b4-ijerph-07-00799]\]. A large body of research among general population samples has shown that IPV prevalence is highest among younger couples, members of racial/ethnic minorities, and couples with household indicators of lower socioeconomic status (SES), such as unemployment, lower education and income levels \[[@b5-ijerph-07-00799]--[@b8-ijerph-07-00799]\].
3.3.. Male-to-Female and Female-to-Male Partner Violence
--------------------------------------------------------
Because women are more likely than men to sustain injuries as a result of IPV \[[@b9-ijerph-07-00799]--[@b11-ijerph-07-00799]\], male-to-female partner violence (MFPV) has been regarded as the more urgent public health issue, and has received considerably more research attention than female-to-male partner violence (FMPV). Survey evidence from nationally representative samples, however, suggests that rates of FMPV equal or exceed MFPV rates among couples in the general household population \[[@b4-ijerph-07-00799],[@b12-ijerph-07-00799]--[@b14-ijerph-07-00799]\], and that approximately half of IPV events are bi-directional (*i.e.*, male-to-female and female-to-male), with the remainder divided between male-to-female only and female-to-male only partner violence \[[@b15-ijerph-07-00799],[@b16-ijerph-07-00799]\]. It is therefore important to address the contribution of individual- and environmental-level factors to both types of IPV in order to further public health prevention efforts.
3.4.. IPV Typologies
--------------------
Over the past decade, considerable progress has been made towards identifying different types or contexts of IPV \[[@b17-ijerph-07-00799]--[@b20-ijerph-07-00799]\]. For example, Johnson \[[@b18-ijerph-07-00799]\] argues that there are at least two distinct types of IPV: common couple violence and 'patriarchal terrorism'. The former is theorized to characterize the type of situational outbursts that may occur between couples, typically in the course of conflict. Common couple violence (also known as situational couple violence) can be bi-directional, and usually involves 'moderate' acts (e.g., pushing, shoving, grabbing, slapping), although escalation to severe episodes (e.g., hitting with fist, kicking) is possible. Patriarchal or intimate terrorism is characterized by a pattern of more severe violence typically associated with terms such as 'wife beating' and 'battered women'. This type of violence, theorized as being rooted in patriarchal ideology and tradition, is a form of terroristic control of women by their male partners. It involves the systematic use of violence, as well as other control tactics, such as threats, emotional abuse, isolation, and economic dependency \[[@b17-ijerph-07-00799]\].
3.5.. *Common Couple Violence* vs. *Intimate Terrorism*
-------------------------------------------------------
These distinctions are methodologically and theoretically important. Methodologically, one would expect the overwhelming majority of IPV reported by couples sampled from the general household population to consist of common couple violence \[[@b21-ijerph-07-00799]\]; cases drawn from shelter, clinical, or treatment populations are more likely to represent intimate terrorism \[[@b17-ijerph-07-00799]\]. Theoretically, the distal and proximal correlates of IPV are thought to differ based on male batterer typology \[[@b19-ijerph-07-00799],[@b20-ijerph-07-00799]\]. Although its consequences are not as severe as those for intimate terrorism, situational couple violence has deleterious health consequences. For example, Johnson and Leone \[[@b17-ijerph-07-00799]\] found that women who experienced situational couple violence experienced significantly more depressive symptoms, and were significantly more likely to use antidepressants, compared to women who did not experience any couple violence. Second, common couple violence consisting of moderate acts (e.g., pushing, shoving, grabbing) can potentially progress over time to more severe levels of IPV \[[@b22-ijerph-07-00799]\]. Given that common couple or situational violence comprise most IPV events in the general population, focusing on the individual and environmental-level factors associated with common couple or situational violence has significant public health implications for IPV prevention.
4.. Epidemiology
================
4.1.. Individual Risk Factors for IPV: Drinking
-----------------------------------------------
Although not a "necessary or sufficient cause" of IPV, problem drinking (e.g., heavy or binge drinking; intoxication) on the part of the male often precedes or accompanies acts of IPV \[[@b23-ijerph-07-00799]\]. In addition, some research suggests that problematic drinking patterns on the part of the male and female are associated with both MFPV and FMPV among couples in the general household population \[[@b24-ijerph-07-00799],[@b25-ijerph-07-00799]\]. Context of drinking and other potential moderator variables may be of critical importance for understanding why alcohol contributes to IPV for some couples under some circumstances but not others \[[@b26-ijerph-07-00799]\]. Several theoretical explanations of the alcohol-IPV relationship have been proposed. While a full discussion of these theories is beyond the scope of this article, Klostermann and Fals-Stewart \[[@b27-ijerph-07-00799]\] recently reviewed the evidence for three proposed mechanisms underlying the alcohol-IPV association: the spurious cause model in which the alcohol-IPV relationship is the result of these variables being related to other factors that influence both drinking and IPV; the indirect effects model in which alcohol use has detrimental effects on relationship quality by increasing marital discord, which in turn increases the likelihood of IPV; and the proximal effects model in which alcohol intoxication is a proximal causal agent of IPV via the psychopharmacologic effects of alcohol on cognitive processing or through alcohol-related expectancies \[[@b28-ijerph-07-00799]\].
The preponderance of evidence for the spurious cause model is weak in that the association between alcohol and IPV remains significant even when a range of psychosocial and sociodemographic variables related to both behaviors are controlled for \[[@b23-ijerph-07-00799]\]. Likewise, the indirect effects model is not well supported empirically because the alcohol-IPV association remains significant even when marital discord and similar variables are statistically accounted for \[[@b29-ijerph-07-00799]\]. Klostermann and Fals-Stewart \[[@b27-ijerph-07-00799]\] suggest that there is now considerable empirical support for the proximal effects model, including longitudinal studies that have found that the husband's alcohol use predicted subsequent marital aggression \[[@b3-ijerph-07-00799],[@b30-ijerph-07-00799],[@b31-ijerph-07-00799]\]. Research conducted among male alcoholics has shown that the occurrence of IPV was significantly reduced after the men completed treatment for alcohol dependence \[[@b32-ijerph-07-00799]\]. Testa, Quigley and Leonard found that the husband's acts of IPV that occurred when the husband was drinking involved more acts of aggression and greater severity compared to sober IPV events \[[@b33-ijerph-07-00799]\]. Despite the empirical evidence linking alcohol to IPV, it is important to note that IPV can and does occur in the absence of drinking or alcohol problems. Context of drinking and other moderator variables may be of critical importance for understanding why alcohol contributes to IPV for some couples under some circumstances but not others \[[@b26-ijerph-07-00799]\].
4.2.. Drugs and IPV
-------------------
It is important to note that illicit drug use on the part of the male and female partner is also associated with increased risk of IPV \[[@b26-ijerph-07-00799],[@b34-ijerph-07-00799],[@b35-ijerph-07-00799]\]. Particularly in treatment populations, high rates of IPV are found among women drug users; likewise, rates of drug use are elevated among women in domestic violence shelter populations \[[@b36-ijerph-07-00799]\]. Because drug use is typically low in general population samples, studying the effects of drug use in relation to IPV is more difficult than that of alcohol. Several mechanisms have been hypothesized. For example, women's drug use within abusive relationships may represent attempts at self-medication \[[@b36-ijerph-07-00799]\]. Psychopharmacologic properties of particular drugs, such as cocaine, may interact with its social correlates, such as greater propensity to use violence as a means to conflict resolution, resulting in increased likelihood of partner violence on days of use \[[@b37-ijerph-07-00799]\]. Alternatively, the association of men's and women's drug use with IPV may represent a marker for risky lifestyle choices and personality characteristics associated with risk-taking (e.g., impulsivity) that can lead to aggression, especially in the context of couple conflict \[[@b37-ijerph-07-00799]\].
4.3.. Psychosocial Correlates of IPV
------------------------------------
Numerous psychosocial correlates are significantly associated with risk for IPV perpetration, victimization, or both. These include measures of impulsivity \[[@b24-ijerph-07-00799],[@b38-ijerph-07-00799]\], anger expression \[[@b39-ijerph-07-00799],[@b40-ijerph-07-00799]\], approval of marital aggression \[[@b12-ijerph-07-00799]\], low marital satisfaction \[[@b41-ijerph-07-00799],[@b42-ijerph-07-00799]\], and family history of violence and other adverse childhood exposures \[[@b24-ijerph-07-00799],[@b43-ijerph-07-00799],[@b44-ijerph-07-00799]\]. Many of these factors (e.g., family history of violence, approval of marital aggression, low marital satisfaction) have been conceptualized as distal influences on the occurrence of IPV, with substance use (*i.e.*, alcohol and drugs) acting as proximal influences \[[@b35-ijerph-07-00799],[@b45-ijerph-07-00799]\]. Path model analysis among married or cohabiting couples sampled from the U.S. general household population suggests that childhood experiences with violence victimization are associated with impulsivity and drinking problems later in life, all of which are associated with higher levels of IPV \[[@b38-ijerph-07-00799]\].
4.4.. Neighborhood Influences on IPV
------------------------------------
Violence or aggression between intimates falls under the rubric of family violence. Like child maltreatment or elder abuse, it is typically a 'private' event that takes place behind closed doors \[[@b46-ijerph-07-00799]\]. Because of this, most IPV research over the past thirty five years has focused on the interpersonal characteristics of one or both members of the couple. With few exceptions \[[@b47-ijerph-07-00799]\], little attention was paid to how environmental factors may influence risk for IPV. Aided by theoretical and methodological advances in multilevel research on disease risk and health behaviors \[[@b48-ijerph-07-00799],[@b49-ijerph-07-00799]\], researchers have begun to examine the contribution of neighborhood factors to risk for engaging in IPV.
4.5.. Empirical Studies
-----------------------
To a large extent, these studies have consisted of empirical tests of cross-sectional data demonstrating that couples residing in disadvantaged neighborhoods are at elevated IPV risk. For example, among a national sample of white, black, and Hispanic couples, Cunradi *et al.* \[[@b50-ijerph-07-00799]\] found that black couples who lived in impoverished (≥20% of households below poverty line) neighborhoods were three times as likely to report past-year male-to-female partner violence, and twice as likely to report female-to-male partner violence, than black couples who did not live in impoverished neighborhoods. White couples who lived in impoverished neighborhoods were nearly four times likelier to report female-to-male partner violence than white couples who did not live in impoverished neighborhoods. O'Campo *et al.* \[[@b51-ijerph-07-00799]\] and Cunradi *et al.* \[[@b25-ijerph-07-00799]\] found that women who lived in neighborhoods characterized by high unemployment rates were at significant risk for male-perpetrated IPV. Van Wyk *et al.* \[[@b52-ijerph-07-00799]\], in an analysis of Wave 2 of the National Survey of Families and Households, found that rates of MFPV (*i.e.*, hitting, shoving or throwing things at the partner) were lowest in the least disadvantaged neighborhoods (3.5%) and highest in the most disadvantaged neighborhoods (7.9%).
4.6.. Perceived Neighborhood Disorder, Drinking, & Mutual IPV
-------------------------------------------------------------
In an analysis of over 18,000 married and cohabiting respondents who participated in the 2000 National Household Survey on Drug Abuse, Cunradi \[[@b13-ijerph-07-00799]\] found that the relationship between drinking level and mutual (*i.e.*, respondent report of both male-to-female and female-to-male partner violence) IPV varied by level of perceived neighborhood social disorder among women, but not men. These interactions were probed by estimating the impact of drinking level on mutual IPV conditional on neighborhood disorder being set to high and then low values, with all other variables in the model held constant \[[@b13-ijerph-07-00799]\]. The results showed that compared to women abstainers, risk for mutual IPV among women who reported recent hazardous drinking was fairly constant (Odds Ratio\~6.0) across levels of neighborhood social disorder. In contrast, the magnitude of effect between drinking level and mutual IPV significantly increased under conditions of high neighborhood social disorder, but decreased to insignificant levels under conditions of low neighborhood social disorder among women in more moderate drinking categories, compared to women abstainers. In other words, women whose drinking has reached dangerous levels are at significantly elevated risk for mutual IPV regardless of their neighborhood environment; the drinking level of women at less hazardous levels puts them at significant risk only if they reside in highly disordered neighborhoods. These findings are partially explained by the dual-hazard hypothesis proposed by Fox and Benson \[[@b53-ijerph-07-00799]\], in which the accumulation and interaction of individual- and environmental-level risk factors exacerbate risk for IPV.
Among men in the study sample, however, no evidence was found for the moderating role of neighborhood social disorder. Instead, a direct effects model indicated that neighborhood social disorder was significantly associated with likelihood of men reporting past-year mutual IPV (OR = 1.61; 95% CI 1.39, 1.87). Independent effects were also seen for patterns of alcohol consumption. For example, men who were recent heavy drinkers (drank 5 or more drinks on the same occasion on each of 5 days in the past 30 days) were more than six times as likely to report mutual IPV compared to men who did not drink in the past year. Men who were recent binge drinkers (drank 5 or more drinks on the same occasion on at least one day in the past 30 days) were approximately three times as likely to report past-year mutual IPV compared to men who were past-year abstainers.
5.. Social Disorganization Theory in Relation to IPV
====================================================
Socially disorganized neighborhoods have been characterized as having three components: low collective efficacy, weak informal local friendship networks, and low participation of residents in local organizations \[[@b54-ijerph-07-00799]\]. Aggregate neighborhood factors that inhibit community social organization include concentrated disadvantage, immigrant concentration, and residential instability. Weak or nonexistent social ties among residents of such neighborhoods helps create an environment where residents are unlikely to intervene in problem behaviors, such as public drunkenness or family violence. Under these conditions, higher rates of problem behaviors will be found in neighborhoods that lack the structure or resources to either prevent or combat these problems when they arise.
5.1.. Neighborhood Disorganization, Alcohol Outlets, & IPV
----------------------------------------------------------
Neighborhood social disorganization may independently, and in concert with high densities of alcohol outlets, lead to IPV. IPV occurs in a social and physical context. Neighborhoods that have high levels of social disorganization have greater concentrated disadvantage, residential instability, and immigrant concentrations. These neighborhoods may also have a relatively high density of alcohol outlets. Greater levels of social disorganization and a high density of alcohol outlets may promote 'cognitive landscapes' that result in more aggressive behavior among area residents, both in terms of alcohol consumption and norms \[[@b55-ijerph-07-00799]\], leading to increased IPV.
5.2.. Alcohol Outlets & IPV: Research Evidence
----------------------------------------------
To date, three ecological studies have examined the contribution of alcohol outlet density to the occurrence of police-reported IPV. All three found that alcohol outlet density was significantly correlated with IPV \[[@b56-ijerph-07-00799]--[@b58-ijerph-07-00799]\]. Moreover, one of the studies \[[@b58-ijerph-07-00799]\] had a longitudinal design, and the findings suggest that outlet density is associated with rates of IPV over time. Because of their ecological designs, however, a major limitation of these studies was their lack of individual-level data concerning drinking, respondent characteristics, and IPV.
McKinney and colleagues recently examined the relation between alcohol outlet density and IPV, and whether binge drinking or alcohol-related problems moderated the relationship between alcohol outlet density and IPV, among a sample of 1,597 couples obtained from the general household population \[[@b59-ijerph-07-00799]\]. In adjusted multilevel analyses, they found that an increase of ten alcohol outlets per 10,000 persons was associated with a 34% increased risk of MFPV; the finding for FMPV was not significant. Moreover, they found that the relationship between alcohol outlet density and MFPV was stronger among couples reporting alcohol-related problems than those reporting no alcohol-related problems. Contrary to their expectations, on-premise alcohol outlet density was positively associated with risk of MFPV; estimates concerning off-premise outlets with MFPV and FMPV were unstable, limiting their ability to interpret the findings.
6.. Potential Social Mechanisms
===============================
Potentially synergistic interactions of alcohol outlets with aspects of neighborhood disorganization may be related to the occurrence of IPV, but these environmentally modifiable relationships have not been systematically examined. Understanding the social mechanisms that underlie the association between neighborhood context, alcohol outlets, and the occurrence of IPV is needed in order to translate research findings into policy changes or other environmental interventions aimed at IPV prevention. The following section suggests likely mechanisms by which neighborhood conditions, in concert with alcohol outlet density, increase risk for IPV.
6.1.. Alcohol Outlets as a Sign of Loosened Normative Constraints against Violence
----------------------------------------------------------------------------------
Greater alcohol outlet density, especially in disorganized neighborhoods, may contribute to increased IPV risk through a number of pathways. For example, Bennett *et al.* \[[@b60-ijerph-07-00799]\] suggest that alcohol outlets, particularly off-premise packaged goods liquor stores, are often surrounded by signs of physical disorder, such as empty or broken bottles, loiterers, and publicly intoxicated patrons. Together with other deleterious neighborhood conditions, the presence of alcohol outlets signals to residents that the mechanisms of informal social control are not working \[[@b60-ijerph-07-00799]--[@b62-ijerph-07-00799]\]. Under such conditions, residents may be less likely to become involved if they witness or hear a couple involved in IPV, either through personal intervention, calling the police, or through any sort of public acknowledgement of the IPV behavior \[[@b52-ijerph-07-00799]\]. Lack of informal social control may also lead residents of disorganized neighborhoods to be less concerned about social consequences of engaging in IPV (e.g., neighbor or police intervention), and therefore less constrained in their behavior towards their spouse or partner. Furthermore, residents in these neighborhoods may be unwilling to interfere in domestic conflicts among their neighbors due to community nonintervention norms concerning "family" or "private" disputes \[[@b63-ijerph-07-00799]\].
6.2.. Alcohol Outlets Promote Problem Alcohol Use among At-Risk Couples
-----------------------------------------------------------------------
Especially for couples in socially disorganized neighborhoods, it is quite plausible that greater alcohol availability provided by bars and off-premise packaged goods stores will result in heavier drinking on the part of one or both members of the couple, and thereafter increased IPV risk. Alcohol availability theory \[[@b64-ijerph-07-00799]\] proposes that as the physical availability of alcohol increases, so too will actual alcohol use at the individual level. Thus, a relatively high concentration of bars and/or liquor stores in a particular area may increase the risk of violence such as IPV. An individual whose barriers to aggression are lowered when drinking may not have the same opportunity in a low density area compared to a high density area and thus IPV may be less. The disproportionate distribution of off-premise liquor stores in low-income African American communities may exacerbate this potential by providing a ready source of alcohol that is marketed for immediate consumption in chilled, large bottles \[[@b65-ijerph-07-00799]\]. Couples residing in neighborhoods that have greater outlet density may adopt patterns of venue use associated with heavier drinking that results in higher levels of IPV, and these patterns of venue use may be greater in neighborhoods characterized by higher levels of social disorganization.
6.3.. Alcohol Outlets Provide Environments where High-Risk Groups Form
----------------------------------------------------------------------
Greater numbers of alcohol outlets within a neighborhood may provide environments where groups of persons at risk for IPV may form and mutually reinforce IPV-related attitudes, norms, and problem behaviors. A number of scenarios are possible. For example, men who drink in bars that have physical or social characteristics that makes violence more likely \[[@b66-ijerph-07-00799]\] may return home to their spouse/partner in a disinhibited, aggressive state in which conflict can rapidly escalate to IPV. Another possibility is that the opportunity afforded to drink by the presence of off-premise outlets increases the chances that some men will purchase alcohol, consume it in the company of other intoxicated men in a public setting (e.g., street corner, park), and thereafter return home in a disinhibited, aggressive state that likewise makes IPV probable in the context of spouse/partner conflict. In addition, bars and off-premise liquor stores may help promote and/or strengthen aggressive norms. Barriers to aggression may be lowered not only by actual alcohol use but also by drinking in a setting that poorly regulates or encourages aggression. The niche theory and assortative drinking hypotheses posit that alcohol sellers 'niche market' to select social strata; drinkers return to outlets frequented by people like themselves; and consequent social stratification of drinkers across contexts will result in greater levels of problems in some outlets \[[@b67-ijerph-07-00799]\]. Social disorganization theory \[[@b54-ijerph-07-00799]\] suggests that higher rates of 'deviant' behavior, such as public intoxication and IPV, will be found in disorganized neighborhoods that lack a structure to help maintain social controls over these problem outcomes. Through these mutually reinforcing mechanisms, the presence of alcohol outlets in socially disorganized neighborhoods may compound both the effects of social disorganization and patterns of venue use and drinking. Furthermore, ambiguous or even supportive norms concerning the use of force or violence to resolve disputes may be sanctioned in socially disadvantaged neighborhoods \[[@b55-ijerph-07-00799],[@b68-ijerph-07-00799],[@b69-ijerph-07-00799]\].
7.. Conclusions
===============
7.1.. Future Directions
-----------------------
The current state of IPV research suggests that couples living in socially disorganized neighborhoods are at increased risk for IPV compared to couples that do not live in socially disorganized neighborhoods, net of other individual- and couple-level characteristics. Increased alcohol outlet density appears to be associated with risk for MFPV, and this association varies depending on the presence of alcohol-related problems among the couple. Future IPV studies need to identify the mechanisms as suggested in this paper that underlie these associations. Ideally, such studies will take into account reports about IPV and drinking from both members of the couple, and will be able to assess exposure to neighborhood characteristics and alcohol outlet density over time in order to establish temporality. Collecting dyadic data has several advantages over data obtained from one partner per couple. First, dyadic data allows for the drinking behaviors and other characteristics of both partners to be modeled. Second, IPV prevalence estimates based on dyadic reports helps reduce bias associated with estimates based on reports from one partner per couple \[[@b38-ijerph-07-00799],[@b70-ijerph-07-00799]\]. Multilevel studies of IPV need to account for spatial autocorrelations (measurement error related to the spatial proximity of sampled units one to another that can bias statistical estimates of effects) using techniques to control for potential Type 1 error (as in positive spatial autocorrelation) \[[@b71-ijerph-07-00799]\] or Type II error (as in negative spatial autocorrelation) \[[@b72-ijerph-07-00799]\]. Attention to geographic unit is also important. Key conceptual and methodological challenges include defining the geographic area (e.g., 'neighborhood') whose characteristics may be relevant to the outcome or processes under study, and operationalizing areas in a way that allows linkage of administrative data and individual-level data \[[@b73-ijerph-07-00799]\]. Some researchers have suggested that sub-divisions of cities, such as Census tracts, may be the most appropriate geographic unit to investigate the relationship between alcohol availability and violence \[[@b74-ijerph-07-00799]\]. To date, significant associations between alcohol outlet density and IPV have been identified at the postcode or zip code level \[[@b58-ijerph-07-00799],[@b59-ijerph-07-00799]\]. Additional research is needed to determine the geographic unit of analysis that is conceptually and methodologically best suited to testing the hypothesized associations between neighborhood characteristics, alcohol outlets, and IPV. Finally, although McKinney and colleagues \[[@b59-ijerph-07-00799]\] did not find a significant association between alcohol outlet density and FMPV, this issue warrants further investigation. Similarly, level of IPV severity in relation to alcohol outlets and neighborhood conditions needs to be explored. Future research will also need to test whether there are gender differences in the impact of neighborhood social disorganization and alcohol outlet density on IPV, as suggested by the findings of Cunradi \[[@b13-ijerph-07-00799]\].
7.2.. Environmental IPV Prevention Strategies are Needed
--------------------------------------------------------
Despite progress that has been made over the past decades in understanding the factors that put couples at risk for engaging in IPV, little progress has occurred in the area of prevention. Since marital aggression, by its definition, takes place between intimates apart from public surveillance, most research has focused on the interpersonal characteristics that put couples at risk for engaging in IPV. But just as environmental strategies aimed at reducing alcohol-related problems are most effective at a population level (e.g., raising the minimum drinking age to age 21 from age 18; lowering legal blood alcohol concentration (BAC) limits; enforcement of underage sales to minors laws) \[[@b64-ijerph-07-00799]\], environmental strategies may be most effective on a population level for reducing and preventing IPV. In this regard, a recent review by Popova *et al.* concluded that restricting availability of alcohol (*i.e.*, alcohol outlet density; hours and days of sale) is an effective measure to prevent alcohol-attributable harm \[[@b75-ijerph-07-00799]\]. Understanding the environmental context in which drinking and IPV occurs can lead to the design of prevention and intervention efforts that address the confluence of individual and community factors that may put couples at elevated risk for IPV. Such an approach to prevention and intervention may therefore be a promising strategy for reducing IPV occurrence.
The project described was supported by Grant Number 1R01AA017705-02 from the National Institute on Alcohol Abuse and Alcoholism; Carol Cunradi, M.P.H., Ph.D., Principal Investigator. The content is solely the responsibility of the author and does not necessarily represent the National Institute on Alcohol Abuse and Alcoholism or the National Institutes of Health.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-sensors-18-01383}
===============
Multimodal human and computer interaction (HCI) has been actively researched over the last few years. One outstanding issue is affective computing, designing devices that communicate with humans by interpreting emotions. Emotion recognition has been attracting attention as a next-generation technology in many fields, from the development of humanistic robots to consumer analysis and safe driving. Most previous research has classified emotions using only facial expressions. However, facial expressions only represent part of the overall human emotional response, and emotion discriminators can sometimes make significant mistakes. For example, classifying an athlete\'s image as displaying happy emotion, when actually the smiling athlete is nervous prior to an important game \[[@B1-sensors-18-01383]\]. On the other hand, biological signals from the central (CNS) and the peripheral (PNS) nervous systems are hard for humans to mentally control, and can accurately represent emotions. Previous studies have shown that changes in skin signals (i.e., galvanic skin response (GSR)) are closely related to changes in peripheral nerves with emotional changes \[[@B2-sensors-18-01383]\], and electroencephalogram (EEG) signals from the frontal lobe are strongly related to emotional changes \[[@B3-sensors-18-01383],[@B4-sensors-18-01383]\]. Therefore, the current study classified emotions using biological signals, including EEG and GSR.
Electroencephalogram signals are used in brain computer interface research, measuring brain electrical activity using an electrode that is attached to the scalp. However, reduced accuracy due EEG signal instability remains a major problem, and EEG signals are untrustworthy, even when they are employing expensive and reliable equipment. The solution is to use as many heterogeneous sensors as possible to provide reliable multiple data. Therefore, we designed a data adaptive CNN model to improve the emotion classification accuracy, reducing current model instabilities, using EEG and GSR data. We also implemented effective spectrogram feature extraction and designed a multimodal classifier that takes two features as input at the first layer of a fully connected network.
This paper is organized as follows. [Section 2](#sec2-sensors-18-01383){ref-type="sec"} discusses previous research methodologies and results. [Section 3](#sec3-sensors-18-01383){ref-type="sec"} discusses the current paper's main contributions, including the details of label processing, EEG signal transformation, GSR data feature extraction, and introduces the proposed CNN model architecture and training strategy. [Section 4](#sec4-sensors-18-01383){ref-type="sec"} analyzes the results and compares them with the current best practice models. Finally, [Section 5](#sec5-sensors-18-01383){ref-type="sec"} summarizes and concludes the paper.
2. Related Work {#sec2-sensors-18-01383}
===============
The related research fields of emotion classification and EEG preprocessing have achieved remarkable results. In general, preprocessing EEG data consists of selecting data while considering the frequency and the location of the brain. Fast Fourier transform (FFT) is the most common frequency analysis method for raw EEG data \[[@B5-sensors-18-01383],[@B6-sensors-18-01383],[@B7-sensors-18-01383],[@B8-sensors-18-01383],[@B9-sensors-18-01383]\], and it was adopted here to extract EEG features. However, FFT cannot reflect temporal information in the frequency data, requiring additional methods to recognize emotions over time. Therefore, short time Fourier transform (STFT), which can express frequency per hour \[[@B10-sensors-18-01383],[@B11-sensors-18-01383],[@B12-sensors-18-01383],[@B13-sensors-18-01383]\], was also used to analyze EEG signals.
Classifying EEG features by frequency is the most common method to differentiate alpha, beta, theta, and gamma waves. Liu et al. \[[@B14-sensors-18-01383]\] presented a table of emotions by frequency and electrode location within the brain region. [Figure 1](#sensors-18-01383-f001){ref-type="fig"} shows the location of the electrodes that are attached to the scalp using the 10-20 system, which is the international standard. Electrodes F3 and F4 distinguish between negative and positive emotional states, and AF3 andAF4 distinguish positive emotions from the surrounding emotions. Wavelet analysis is one of the best ways to express frequency and time, and has also been employed in EEG classification \[[@B15-sensors-18-01383],[@B16-sensors-18-01383],[@B17-sensors-18-01383],[@B18-sensors-18-01383]\].
Various previous studies considered emotional classification methods. Mollahosseini et al. \[[@B19-sensors-18-01383]\] designed a CNN based face recognition module. Gerard Pons et al. \[[@B20-sensors-18-01383]\] enhanced facial image classification performance by supervised hierarchical learning. Ding et al. \[[@B21-sensors-18-01383]\] performed deep face recognition that was based on a two steps model. Poria et al. \[[@B22-sensors-18-01383]\] implemented multimodal visual and audio data analysis beyond the focus on text-based emotional analysis. They also succeeded in feature fusion through deep learning based heterogeneous data dimension reduction.
The Database for Emotion Analysis using Physiological signals (DEAP) dataset has been widely employed for emotion classification models using biomedical signals. Koelstra et al. \[[@B23-sensors-18-01383]\] used the DEAP data set to classify PNS and CNS sensor data, and measured the emotional classification performance. Liu and Sourina \[[@B24-sensors-18-01383]\] studied EEG valence levels for real-time applications. Naser et al. \[[@B25-sensors-18-01383]\] predicted emotions extracted from music videos. Chen et al. \[[@B26-sensors-18-01383]\] applied ontology and datamining techniques for EEG based emotion analysis.
Bayesian networks, unsupervised deep running, and deep belief networks have also been applied \[[@B27-sensors-18-01383],[@B28-sensors-18-01383],[@B29-sensors-18-01383]\].
3. Methods {#sec3-sensors-18-01383}
==========
3.1. Multiple Label Classification {#sec3dot1-sensors-18-01383}
----------------------------------
A label was constructed using the self-assessment value that was provided in the DEAP dataset, including valence, arousal, dominance, liking, and familiarity. Emotional states are typically evaluated using arousal and valence, and are divided into four sections: high arousal, high valence (HAHV); high arousal, low valence (HALV); low arousal, low valence (LALV); and, low arousal, high valence (LAHV) \[[@B30-sensors-18-01383]\], as shown in [Figure 2](#sensors-18-01383-f002){ref-type="fig"}. Thus, emotional states can be classified according to arousal and valence levels.
Labeling was based on a threshold value for the two-dimensional (2D) plane. We implemented k-means clustering on self-assessed arousal and valence levels to find the most appropriate threshold. Previous studies have employed one shot encoding for labeling as a 2D vector, i.e., \[HV, LV\] and \[HA, LA\] using k-means clustering with k = 2 \[[@B31-sensors-18-01383]\]. Therefore, we performed independent valence and arousal classifications in order to compare with previous models.
However, independent classification fails to consider arousal and valence correlations, and since the data is arousal and valence levels, rather than emotion level, it cannot be implemented for end to end learning, since it must be mapped onto the two-dimensional (2D) plane ([Figure 2](#sensors-18-01383-f002){ref-type="fig"}) for emotion judgment.
Therefore, we propose k-means clustering with k = 4 to provide a four-dimensional (4D) label vector. [Figure 3](#sensors-18-01383-f003){ref-type="fig"} compares clustering for k = 2 and k = 4. Point (5, 5) is the approximate center mean for both k = 2 and k = 4, hence we use (5, 5) as the threshold.
Thus, labeling included 2D and 4D vectors through one shot encoding for learning.
3.2. EEG Signal Transformation to Time to Frequency Axes {#sec3dot2-sensors-18-01383}
--------------------------------------------------------
The data was preprocessed to reflect EEG temporal and frequency characteristics. Since the EEG data measuring human emotions are time series data, time information must be reflected in the frequency data. Although the STFT has been widely used to add time information to frequency data \[[@B10-sensors-18-01383],[@B11-sensors-18-01383],[@B12-sensors-18-01383],[@B13-sensors-18-01383]\], it has disadvantages for time-frequency analysis, in that temporal resolution decreases as the window increases; and, frequency resolution decreases as window size decreases.
Therefore, we propose using a wavelet transform to represent the frequency axis, using the open toolbox EEG lab. The extracted spectrogram was 42 × 200 pixel, width × height, where width (200 pixel) represents time, and height (42 pixel) represents EEG sensor frequency (4.0--45 Hz), as shown in [Figure 4](#sensors-18-01383-f004){ref-type="fig"}. Total transformed data include 40960 spectrograms. At this time, the number of batch data for training is 32 spectrogram data that means 32 electrodes that were derived from one stimulus. Therefore, the total amount of data set used in this study is 1280, with data labels, as shown in [Table 1](#sensors-18-01383-t001){ref-type="table"} and [Table 2](#sensors-18-01383-t002){ref-type="table"}.
Conventional EEG based emotion classification analyzes the degree of activity in a specific area of the brain (e.g. the frontal lobe), using electrodes that were attached to the head close to the frontal lobe and some other lobes (e.g., AF3, AF4, P7). Frequency bands for specific electrodes were typically subdivided into alpha, beta wave, gamma, etc. waves to allow for simple and shallow classification models, such as support vector machines (SVMs). However, sensor selection and subdivision ignores emotion related signal changes in other brain regions. Recent advanced deep learning techniques can improve emotional analysis accuracy by incorporating all sensor data for each experiment.
3.3. GSR Preprocessing Using Short Time Zero Crossing Rate {#sec3dot3-sensors-18-01383}
----------------------------------------------------------
To extract the feature, we divide the GSR waveform into defined windows and calculate the short time zero crossing rate (STZCR), i.e., the number of times the signal crosses zero within a given window. That is, we intend to use the change in amplitude of the GSR as the input feature vector for deep running. STZCR indicates the rate of signal change,$$STZCR = \frac{1}{N}\sum\limits_{n}^{m}\frac{|{sig{\{{s\left( n \right)}\}} - sig{\{{s{({n - 1})}}\}}}|}{2}w{({m - n})}$$ where $N$ is the sampled signal, and $w$ represents the window. We highlighted features using the extracted zero crossing rate vector with threshold$$T = \frac{\sum GSR_{stzcr}}{N_{stzcr}}$$ where $GSR_{stzcr}$ is a vector column and $N_{stzcr}$ is the number of vectors. If the data is greater than the threshold, it outputs max, otherwise it outputs zero. GSR amplitude is generally sensitive to arousal changes and less sensitive to valence changes, hence, it can positively affect EEG features to focus on arousal in the classifier model.
3.4. Fusion Convolution Neural Network Model for EEG Spectrograms and GSR Features {#sec3dot4-sensors-18-01383}
----------------------------------------------------------------------------------
Many neural networks have been developed for classification in recent studies. The first thing to consider when designing a CNN is data characteristics. Therefore, we designed the CNN to use the spectrogram image from the wavelet transformation of all the channels.
Tabar and Halici \[[@B32-sensors-18-01383]\] considered CNN classification problems using EEG spectrograms, and designed a single layer CNN using one-dimensional filtering to provide good classification performance based on motor imagery EEG signals. However, a single filtering through the single convolutional layer does not efficiently extract features for emotion classification, since it is not deep enough to extract emotion data.
Therefore, we propose a neural network based on the extracted data as described above, which allows for deep convolution layers, while also reflecting temporal effects, as shown in [Figure 5](#sensors-18-01383-f005){ref-type="fig"}. We first normalized the data, making the cost function a spherical contour, and helping to increase the learning rate. We then designed a deep convolution layer that reflects time, using a 3 × 2 filter rather than conventional square filters, such as 2 × 2 or 3 × 3. The spectrogram frequency per hour can be reflected by increasing the filter height. Since the filter is a feature identifier that extracts the information from the manifold state, the shape of the filter is related to the content of the feature to be extracted from the receptive field. Our proposed filter can identify data in a region that is relatively longer than a square filter. Thus, the data containing the vertical meaning is repeatedly transmitted to the input of the next layer. As a result, the frequency per hour of the spectrogram image can be learned in CNN. Setting stride = \[2, 1\] with no padding, the filter can be extracted based only on the image time base. We used a fully connected layer for the final classification. The classifier is trained on the spectrogram features of 32 electrodes extracted through CNN. In continuous training, the classifier learns similar patterns extracted from 32 individual electrodes, and can be classified as a label through the last softmax layer. The entire model consists of four convolutional layers and seven fully connected layers.
Batch normalization \[[@B33-sensors-18-01383]\] was performed before each value was passed to the activation function, except for the last convolutional layer, in order to prevent the model gradient vanishing during training. It has the effect of preventing internal covariance shift by reducing activation function variation that is caused by the previous layer's variation. Batch normalization was implemented, as follows.(1)Normalize the batch data using the batch mean, $\mu_{\beta}$, and variance, $\sigma_{\beta}^{2}$,$${\hat{x}}_{i} = \frac{x_{i} - \mu_{\beta}}{\sqrt{\sigma_{\beta}^{2} + \epsilon}}$$(2)Use the r and d values for scale and shift operations,$$y_{i} = \gamma{\hat{x}}_{i} + \beta$$
Updating $\gamma$ and $\beta$ by training allows for the CNN to better reflect the model characteristics model in normalized variables, rather than simple normalization, such as whitening. Testing uses average $\gamma$ and $\beta$ obtained.
Feature maps are generated as the image passes through each convolution layer. The layer activation function is a rectified linear unit (ReLU), which is a function that makes the value of the part where *x* \< 0 in the linear function *y* = *x* is 0,$${ReLU\left( u \right) = max{({u,0})},}{max{({u,0})} = \begin{cases}
{u,} & {if~u > 0} \\
& \\
{0,} & {otherwise} \\
\end{cases}}$$
The ReLU function is computationally efficient because its activation is not restricted to \[−1, 1\], as for the hyperbolic tangent function, but is used as it is. Therefore, training speed for large spectrogram images is increased, and outputting 0 prevents overfitting due to training many weights, hence training regularization can be expected.
After passing through the final 2 × 2 pooling layer, the image is flattened and combined with GSR. To positively influence EEG data performance classification, GSR data uses the data average as the thresh hold to remove noise. It also reduces the computation burden for training a fully connected network by transmitting a zero value to each neuron's perceptron. The final layer neuron returns the model's probability distribution using softmax, and performs classification by changing the number of neurons according to the experimental environment, such as \[HV, LV\], \[HA, LA\], or \[HVHA, LVHA, LVLA, HVAL\].
3.5. Training Strategy {#sec3dot5-sensors-18-01383}
----------------------
We use maximum likelihood estimation (MLE) in order to train the proposed CNN model. MLE maximizes $P{({Y|X;\theta})}$ by optimizing *θ* in the probability model for a given data point $X$ and label $Y$. Cross entropy is the most commonly used MLE loss function, and it calculates the difference between two probability distributions. Let $p\left( x \right)$ be the actual and $q\left( x \right)$ be the predicted probability distribution for the label. Then, cross entropy, $L{({p,q})}$, is$$L{({p,q})} = \int p{(x)} \cdot \ln q\left( x \right)dx$$
CNN training proceeds by back propagation using the gradient decent. We update the weights using the partial derivative of cross entropy loss $L$ for weight matrix $W$,$$W = W - \gamma \cdot \frac{\partial L}{\partial W}$$ where $z_{j} = \sum w_{ij}o_{i} + b$ is the sum of inner products, and we calculate the gradient as$$\frac{\partial L}{\partial W} = \frac{\partial L}{\partial p{(z_{j})}} \cdot \frac{\partial p{(z_{j})}}{\partial z_{j}} \cdot \frac{\partial z_{j}}{\partial w_{ij}}$$ where $\frac{\partial L}{\partial p{(z_{j})}}$ is the magnitude of the influence of function p on L and $p{(z_{j})}$ is the softmax result.
Generally, to find the optimal training point, we find the bias variance trade off point using validation loss, as shown in [Figure 6](#sensors-18-01383-f006){ref-type="fig"} for the 4 class case.
After 400 iterations, validation loss increases, whereas training continues to decrease. Thus, we can conclude the model becomes over-fitted beyond 400 iterations, providing the optimal training point. Test data should be applied with this level of iteration to measure model accuracy.
4. Results and Discussions {#sec4-sensors-18-01383}
==========================
4.1. Experiment Environment {#sec4dot1-sensors-18-01383}
---------------------------
[Table 3](#sensors-18-01383-t003){ref-type="table"} shows the hardware and framework specifications for the experiment.
4.2. Dataset {#sec4dot2-sensors-18-01383}
------------
The DEAP dataset \[[@B23-sensors-18-01383]\] was used to provide bio-signal data, containing CNS and PNS data. PNS data comprised GSR, skin temperature, respiration, blood volume (by plethysmograph), and electrooculogram (EOG). GSR data was the skin resistance of the middle and forefinger, skin temperature, and breath change by emotion, including body tension and irritating fear. Plethysmograph measured blood flow changes in the finger. EOG signal was measured by eye blinking, which is related to anxiety. CNS data was the EEG signal.
Data were collected from 32 subjects for 1 m for each of 40 selected music videos. Data was recorded on 48 channels with 512 Hz sampling frequency. We used preprocessed data version of MATLAB and numpy formats that were provided by the DEAP dataset, down-sampled to 128 Hz with a 4.0--45 Hz band pass filter applied.
4.3. Performance Analysis {#sec4dot3-sensors-18-01383}
-------------------------
In this section, we analyzed the performance of the model in two ways. In first evaluation, we analyzed the classification performance for each label using hold-out validation. To construct a hold-out validation set, test, verification, and learning datasets were created 1:1:9 ratio for each label, with batch size = 32 to reflect data from one stimulus. Second, for the LOOCV, we constructed the dataset that was measured by one-video as test set and the other video data as training set. The DEAP dataset consists of a data set for 40 videos per participants. In other words, the second evaluation was performed with 39 video stimuli as training dataset, and the data that was extracted by the other one stimulus was used as a test dataset.
The desired ideal model would accurately distinguish data patterns and generalize them even when testing data are considered, i.e., we want to find a model between over and under fitting. The proposed model does not apply L2 regularization to prevent overfitting, because there is a batch normalization layer. In addition, cross entropy loss was measured for each iteration to find the optimal training point, as shown in [Figure 6](#sensors-18-01383-f006){ref-type="fig"}. [Table 4](#sensors-18-01383-t004){ref-type="table"} shows the predicted accuracy for methods of label based and video based classification using each validation method.
4.4. Comparison with Existing Models {#sec4dot4-sensors-18-01383}
------------------------------------
We used two class labels that were commonly adopted in previously studies to compare performance, as measured by arousal and valence classification accuracy for the DEAP dataset. [Table 5](#sensors-18-01383-t005){ref-type="table"} shows the performance compared with the existing models measured using the same dataset. The performance of our model is shown by the result of LOOCV in [Section 4.3](#sec4dot3-sensors-18-01383){ref-type="sec"}, to validate the generalized performance of the model.
The considered methods used a variety of approaches: Koelstra et al. \[[@B23-sensors-18-01383]\] used CNS and PNS sensors; Liu and Sourina \[[@B24-sensors-18-01383]\] used a fractal algorithm to reflect signal complexity that was based on a threshold value; Naser and Saha \[[@B25-sensors-18-01383]\] extracted features using a dual-tree complex wavelet transform and used SVM for classification; Chen et al. \[[@B26-sensors-18-01383]\] used decision trees; Yoon and Chung \[[@B27-sensors-18-01383]\] used Bayesian and perceptron convergence; and, Wang et al. \[[@B28-sensors-18-01383]\] and Li et al. \[[@B29-sensors-18-01383]\] used deep belief networks to automatically extract features and to classify them. [Figure 7](#sensors-18-01383-f007){ref-type="fig"} show the proposed model has better performance than all compared models
Although EEG data is easier to classify into two classes \[[@B34-sensors-18-01383]\], increasing the number of classes not only enables end-to-end learning, but it also includes correlations between arousal and valence. Therefore, we compared the proposed model performance against previous four class models. Generally, when data quantity is limited, the model accuracy decreases as the number of labels to be classified increases. The performance of our model is shown by the result of LOOCV in [Section 4.3](#sec4dot3-sensors-18-01383){ref-type="sec"}, in order to validate the generalized performance of the model. [Table 6](#sensors-18-01383-t006){ref-type="table"} shows that the proposed model has high performance when compared to current models.
A variety of approaches were employed in the comparison models: Zubair and Yoon \[[@B35-sensors-18-01383]\] used a discrete wavelet transform, and also applied the mRMR algorithm to enhance the feature correlations; Jadhav et al. \[[@B36-sensors-18-01383]\] extracted EEG features using the gray level co-occurrence matrix, and classified emotion using K-nearest neighbor; Hatamikia et al. \[[@B37-sensors-18-01383]\] used using nonlinear extraction and self-organized classification; Martínez-Rodrigo et al. \[[@B38-sensors-18-01383]\] extracted biological signal features using quadratic sample entropy, performed feature selection, and classified the extracted features by SVM; Zhang et al. \[[@B39-sensors-18-01383]\] used wavelet feature extraction that was based on a smoothed pseudo Winger-Ville distribution and classification using SVM; Mei et al. \[[@B40-sensors-18-01383]\] extracted features by constructing a connection matrix of the brain structure, with subsequent classification using CNN. [Figure 8](#sensors-18-01383-f008){ref-type="fig"} show a bar graph that the proposed model has high performance when compared to the current models.
5. Conclusions {#sec5-sensors-18-01383}
==============
This study devised data labeling according to emotion criteria, and proposed a data preprocessing methodology to increase the emotional classification performance. Emotion classification was performed using single and multiple sensor based models. Particular focus was overall analysis and CNN filter design according to input data characteristics and noise removal for data processing.
Feature extraction performance was remarkably improved through the proposed filter design, providing significantly improved classification performance when compared with previous models.
This study paves the way for combining data and designing corresponding deep running models. Future research directions will investigate further changes to the emotion analysis framework, such combining multiple neural networks. One approach would be to improve concatenation of simple convolution layers. It may be possible to construct convolution layers for each data characteristic and improve the classification performance using multiple convolution layers.
Yea-Hoon Kwon and Shin-Dug Kim conceived and designed the experiments; Yea-Hoon Kwon performed the experiments; Yea-Hoon Kwon and Sae-Byuk Shin analyzed the data; Yea-Hoon Kwon and Sae-Byuk Shin wrote the paper.
This work was supported by The Institute for Information & Communications Technology Promotion funded by the Korean Government (MSIP) (R0124-16-0002, Emotional Intelligence Technology to Infer Human Emotion and Carry on Dialogue Accordingly).
The authors declare no conflict of interest.
{#sensors-18-01383-f001}
{#sensors-18-01383-f002}
{#sensors-18-01383-f003}
{#sensors-18-01383-f004}
{#sensors-18-01383-f005}
{#sensors-18-01383-f006}
{#sensors-18-01383-f007}
{#sensors-18-01383-f008}
sensors-18-01383-t001_Table 1
######
The number of extracted wavelet transformed data for two types of labels.
Label ^1^ Data quantity
----------- ---------------
HAHV 458
HALV 294
LAHV 255
LALV 273
Total 1280
^1^ H: high, L: low, V: valence, A: arousal, e.g., HAHV: high arousal, high valence.
sensors-18-01383-t002_Table 2
######
The number of extracted wavelet transformed data for four types of labels.
Label ^1^ Data quantity Label ^2^ Data quantity
----------- --------------- ----------- ---------------
HA 752 HV 713
LA 528 LV 567
Total 1280 Total 1280
^1,\ 2^ H: high, L: low, V: valence, A: arousal, e.g., HA: high arousal.
sensors-18-01383-t003_Table 3
######
Hardware specifications.
---------------- ---------------------------------
**CPU** Intel Core i5-6600
**GPU** NVIDIA GeForce GTX 1070 8GBytes
**RAM** DDR4 16GBytes
**OS** Ubuntu 16.04.
**Frameworks** Tensorflow1.3\
MATLAB/ EEG toolbox
---------------- ---------------------------------
sensors-18-01383-t004_Table 4
######
Emotion classification accuracy.
Results A ^1^ Results B ^2^
--------------------------- --------------------------------------- ---------------------------------------- --------------------------------------- ---------------------------------------- -------- --------
**Clssification Methods** Two Class Classification Accuracy ^3^ Four Class Classification Accuracy ^4^ Two Class Classification Accuracy ^3^ Four Class Classification Accuracy ^4^
Arousal Valence Arousal Valence
**Proposed Fusion Model** 0.7812 0.8125 0.7500 0.7656 0.8046 0.7343
^1^ Results A: label based classification using hold-out validation.
^2^ Results B: video based classification using leave-out one cross validation.
^3^ Arousal: HA, LA; Valence: HV, LV.
^4^ HAHV, HALV, LALV, LAHV.
sensors-18-01383-t005_Table 5
######
Two class classification performance.
Model Accuracy
----------------------------------- ----------------------------- ------------ ------------
CNS feature based single modality \[[@B23-sensors-18-01383]\] 0.6200 0.5760
PNS feature based single modality \[[@B23-sensors-18-01383]\] 0.5700 0.6270
Liu and Sourina \[[@B24-sensors-18-01383]\] 0.7651 0.5080
Naser and Saha \[[@B25-sensors-18-01383]\] 0.6620 0.6430
Chen et al. \[[@B26-sensors-18-01383]\] 0.6909 0.6789
Yoon and Chung \[[@B27-sensors-18-01383]\] 0.7010 0.7090
Li et al. \[[@B29-sensors-18-01383]\] 0.6420 0.5840
Wang and Shang \[[@B28-sensors-18-01383]\] 0.5120 0.6090
Proposed fusion CNN model **0.7656** **0.8046**
sensors-18-01383-t006_Table 6
######
Four class classification performance.
Model Accuracy
--------------------------- ----------------------------- ------------
M Zubair and C Yoon \[[@B35-sensors-18-01383]\] 0.4540
N Jadhav et al. \[[@B36-sensors-18-01383]\] 0.4625
Hatamikia et al. \[[@B37-sensors-18-01383]\] 0.5515
Martínez-Rodrigo et al. \[[@B38-sensors-18-01383]\] 0.7250
Zhang et al. \[[@B39-sensors-18-01383]\] 0.7162
Mei et al. \[[@B40-sensors-18-01383]\] 0.7310
Proposed fusion CNN model **0.7343**
| {
"pile_set_name": "PubMed Central"
} |
Scientific Reports 6: Article number: 30410; 10.1038/srep30410 published online: 07 26 2016; updated: 12 22 2014.
In the Supplementary Information file originally published with this Article, Figures S8 and S12 were omitted.
These errors have been corrected in the Supplementary Information file that now accompanies the Article.
| {
"pile_set_name": "PubMed Central"
} |
Sir,
The knowledge of the ancient concepts regarding the nature of the mind are increasingly being taken note of in contemporary psychiatry.\[[@CIT1][@CIT2]\] In light of this, the article by C. Shamasundar\[[@CIT3]\] on ancient Indian wisdom and its relevance to modern mental health is timely. I believe this article also has implications for the philosophical foundations of present-day psychiatry.
Each school of philosophy is based on certain principles that differentiate it from other schools. Though cross-pollination between different schools is often desirable and productive, it is sometimes not desirable or possible to extrapolate directly from one to the other simply because they are not the same. In this regard, I would like to comment on some of the examples and conclusions that have been drawn by the author of the article.
The mind, as contemporary psychiatrists and other bio-medical professionals understand it, is a collective term for aspects of the mental state that are functions of the brain\[[@CIT4]\] and are manifested as combinations of thought, perception, memory, emotion, motivation, imagination and consciousness. These concepts are important in the study of the mind because they are easily definable, and more importantly are available for objective scrutiny and can be measured with relative ease. The author asserts that the mind is a subject of 'academic apartheid'. It must be understood that the mind is a complex manifestation of different interlocking processes as has been pointed out above and it is not possible to talk about the mind as a unity without difficulty. Owing to this complexity, the study of the mind necessarily has to be broken down into components to make it simpler. It is expected that enquiry into the nature of these components can lead us to 'bits' of truth that can then be put together to get the 'whole' truth. As has been pointed out by the proponents of the 'Decade of the Mind' project,\[[@CIT5][@CIT6]\] any such study would have to focus on four broad intertwined areas of mental health, higher cognitive functions, education and computational applications. Today the study of the mind is the study of the components of the mind and numerous examples can easily be found in any neuroscience journal. The notion of 'academic apartheid' is difficult to maintain in this light. What is becoming increasingly uncommon is research into the nature of the mind from the point of view of thinkers such as Freud. Concepts such as the structural theory of the mind have been highly influential in psychiatry for many decades and have led to the development of many insights, but zeal for these should be tempered by the knowledge of fundamentals of science. Science is the effort to discover and increase human knowledge of the basis of the physical world by means of observation, experimentation and interpretation of the results of the former. These models of the mind have lost some of their influence because of the fact that they simply could not stand up to scientific scrutiny in any replicable manner.\[[@CIT4]\] One could argue that psychiatric nosological systems today are not valid\[[@CIT7]\] but they have undeniable usefulness as regards etiological assumptions, course and outcome and treatment implications of the nosological entities that has led us to practice psychiatry the way we do today.
The author has also described the attributes of the mind as per the different literary sources. While it cannot be denied that these concepts are valid from the points of view of the systems from which they are drawn, drawing an analogy with the aspects of the biomedical construct of the mind is difficult. I would particularly mention the examples of subliminal stimulation, mind-mind interactions, mind-body interactions and reincarnation that have been cited. These phenomena are not accepted as within the fringe of science and as per contemporary knowledge cannot be taken as scientific fact. No one can have any objections to unfettered and unconventional thinking as a way to advancement of science. Indeed many important discoveries have been made as a result of serendipity and out-of-the-box thinking, but these have entered mainstream science only after validation by a scientific method. Indeed, the quest for proof is as old as civilization itself\[[@CIT8]\] and claims such as those cited should come with a caveat about their current status. This also leads us to the question of what should be included in psychiatry and what should not. While it is difficult to exclude subjectivity from what a psychiatrist regards as the truth, it is important to be able to make a reasonable distinction between knowledge that can guide our actions and knowledge that can enable us to be aware of a contrary point of view.
The author makes a distinction between the 'material' and the 'immaterial mind' and claims that the latter cannot be recorded. Before recording the material or the immaterial mind, it is important to ask whether such a distinction exists or is indeed even desirable. The most striking example of such a distinction was propounded by Descartes who held that the mind is a non-physical substance endowed with self-awareness and consciousness that is distinct from the brain which is the seat of all intellect. This concept of dualism and its corollary that the brain is not the seat of the mind has had deleterious effects on psychiatry and the management of the mentally ill.\[[@CIT9][@CIT10]\] Contemporary psychiatry would be best served by the model proposed by Kandel\[[@CIT4]\] because it does away with all etiological assumptions except that the mind is a function of the brain and that epiphenomenonalism is false.
I would congratulate the writer for his deep knowledge lucid exposition of body-mind relationships and definition of mental health as per Indian systems of thought regarding the nature of the mind and the body. Indeed, such concepts are likely to be useful in the psychotherapeutic management of selected patients.
| {
"pile_set_name": "PubMed Central"
} |
Sir,
Leaks in the low-pressure segment of an anaesthesia machine can affect patients by causing hypoxia, awareness and hypercarbia.\[[@ref1][@ref2]\] The low-pressure segment includes the flow tubes, vaporizer manifold, vaporizers and one-way check valve on most modern anaesthesia machines. Leaks in these components are difficult to identify.
Different methods are used to check the low-pressure segment for leaks. The 1993 Food and drug administration universal negative-pressure leak test is one of them. It was named "universal" because at that time it could be used to check all anaesthesia machines, regardless of the presence or absence of check valves in the low-pressure segment. At present, most anaesthesia machines are compatible with this test. It is the most sensitive of all leak tests because it is independent of flow. It can detect leaks as small as 30 mL/min. Several mishaps have resulted from application of a wrong leak test to the anaesthesia machine.\[[@ref3]\]
Negative-pressure leak test is performed by creating negative pressure in the low-pressure segment with a negative-pressure leak-testing device. This device is made of a suction bulb and tubing connecting suction bulb to a 15 mm adaptor, which fits the common gas outlet.\[[@ref4]\] Suction bulb has unidirectional air outlet valve (as shown in \[[Figure 1](#F1){ref-type="fig"}\]) to evacuate air from the machine side of the bulb and to create a negative pressure of 65 cm of water.
{#F1}
ANAESTHESIA APPARATUS CHECKOUT RECOMMENDATIONS, 1993 {#sec1-1}
====================================================
Step 5. Leak Check of Machine Low-Pressure System\[[@ref5]\] -
Verify that the machine master switch and flow control valves are OFFAttach a "suction bulb" to the common (fresh) gas outSqueeze the bulb repeatedly until it is fully collapsedVerify that the bulb stays fully collapsed for at least 10 sOpen one vaporizer at a time and repeat steps c and d as aboveRemove the suction bulb and reconnect the fresh gas hose.
Only few anaesthesia machines are supplied with a negative-pressure leak-testing device. With prolonged use, the supplied device gets damaged or lost. One improvised device is shown in [Figure 1](#F1){ref-type="fig"} to perform the leak test in such situations. A simple suction bulb, used for neonatal resuscitation, is connected to the "a" end of a three-way stopcock connector and the "c" end is connected to a 15 mm adaptor by a piece of rubber tubing used in a mercury sphygmomanometer cuff. All junctions are made leak proof with adhesive.
The negative-pressure leak test with this device is performed by manipulating the three-way connector (as shown in \[[Figure 1](#F1){ref-type="fig"}\]) and squeezing the suction bulb. This device generates negative pressure of more than 250mmHg, which was tested by connecting it to a pressure transducer.
STEPS {#sec1-2}
=====
Verify that the machine master switch and flow control valves are OFFConnector is adjusted to close "b" end and open "a--c" ends. Suction bulb is squeezed completely and device is connected to common gas outletIf bulb fills with air, connector is adjusted to close "c" end and open "a--b" ends to remove air by squeezing itWith squeezed suction bulb, connector is turned to close "b" end and open "a--c" endsSteps 3 and 4 are repeated until the bulb is fully collapsedVerify that the bulb stays fully collapsed for at least 10 sOpen one vaporizer at a time and repeat steps 3--6Remove the device and reconnect the fresh gas hose This improvised device is effective. It is easy to make and cheap because the required materials are of low cost and easily available in the operation theatre.
| {
"pile_set_name": "PubMed Central"
} |
John Bowlby opened his 1948 paper *The Study and Reduction of Group Tensions in the Family* by writing: "Child guidance workers all over the world have come to recognise more and more clearly that the overt problem which is brought to the clinic in the person of the child is not the real problem; the problem as a rule we need to solve is the tension between all the different members of the family" (Bowlby, 1948, p.123). The clinical approach we describe in this paper is something of a return to the systemic emphasis we find in this comment. Such a return to the systems level of the family can be distinguished from the internalized cognitive model of attachment based on a representational and therefore individuated model of attachment theory. When attachment theory is thought of as a discursive-relational model, it fits neatly with both interpersonal and systemic clinical approaches. As we can hear in the comment above, from the outset, Bowlby clearly emphasized the *child and family* in his clinical thinking ([@B1]).
The current paper is focused on the elaboration of the key principles of such a discursive-relational approach, and a description of the treatment techniques of *Behaviour Exchange and Systems Therapy* (BEST). Our research team based in Melbourne and Perth have been developing the theory and practice of BEST in different forms for two decades. These interventions are family-based interventions and can be delivered either to individual families or with small groups of families. Initially, these interventions focused on the parents of adolescents presenting with substance abuse, but the model evolved over time into a whole of family approach and was adapted to also serve as a treatment for adolescent depression and anxiety ([@B2], [@B3]). Our work now is extending the approach to interventions for children under 12 years of age.
Clinical trials our team have being running in Australia have accumulated good evidence to show that the approach can effectively treat a range of adolescent mental disorders. The studies show that improvement in adolescent mental health is typically accompanied by improvements in family functioning and notably improvements in the parent's mental health ([@B2]--[@B8]). Previously our research group has also published qualitative studies of participant experiences as well as a description of the main features of the program for the treatment of adolescent depression called BEST-Mood ([@B5], [@B9], [@B10]). These make up a rich quantitative and qualitative dataset which forms the background to this theoretical paper.
Discourse, Narrative, and Dialogue {#s2}
==================================
There is an increasing use of the term "attachment-based therapy" referring to relational approaches and these are generally considered to have the broad goal of promoting attachment security between parents and children ([@B11]). An attachment-based approach minimally adopts a dyadic view of inter-subjective communication ([@B12]) rather than treating an individual. Such models of therapy have their origins in the clinical approach originally described by Bowlby ([@B1]), but the clinical application of attachment theory has been elaborated by many others, usually, but not exclusively, in relation to a broadly psychodynamic framework ([@B13]--[@B18]). Alongside the work of other groups focused on adolescent mental health, we are interested in how attachment patterns are perpetuated within a family system and how such an understanding can inform interventions ([@B19], [@B20]).
Our clinical interests in an integration of attachment theory and family systems has led us to propose a conceptual shift from a representational model to a discourse model. The most common way of clinically interpreting attachment is derived from the "working model" concept. This is thought to be an individuated representational-cognitive model ([@B21]) and it extends the cybernetic notions of signalling in Bowlby's evolutionary-development framework. The concept of an "internal working model" originates in the attempt to explain how early relationship experiences are carried forward as enduring styles of interpersonal relations and modes of regulating affects. Attachment theory's next major development occurred with Mary Main's work on the manifestation of attachment patterns within adult narratives and in developing this theory it is of importance to recall that she was drawing very directly on H.P. Grice's categories of conversational coherence ([@B22], [@B23]). Attachment classifications based on the coding of the Adult Attachment Interview (AAI) were able to reliably identify very specific discursive features of language use. For example, this includes the mode of recall of early attachment memories, narrative accounts of separation, loss or challenging interpersonal experiences, the subject's capacity to mentalise about aspects of their parent's relationship. Overall patterns of autonomous, dismissive and preoccupied conversational styles emerge across the full interview. These components are rated in terms of an overall *coherence* of discourse, reflecting the integration, and consistency of the narrative.
This shift to the level of representation has given rise to a range of discourse-based measures of attachment and generated a substantial body of evidence to validate the concept of attachment discourse. Discourse based assessments of adult attachment have been more recently developed to analyse responses to images (Adult Attachment Projective- AAP) ([@B24]), a secure-base script method ([@B25]), and similar ideas have been applied to the analysis of child play narratives in response to structured attachment stimuli ([@B26]). This discourse model in particular is fundamental to many clinical applications of attachment theory, and certainly attracted a renewed exchange with psychoanalytic theory in the 1990s ([@B27]--[@B29]). Empirically, a number of important studies have now shown relationships between attachment discourse measures and broader aspects of family discourse. For example, studies found that mother's scripts of secure narratives were related to both the child's degree of attachment security, and the mother's narrative style and emotional language when reminiscing about shared experiences ([@B30]). The researchers suggested that their findings should be understood in terms of the way mother-child dyads discuss emotion-laden content. Similar findings have been reported in high risk samples with histories of child maltreatment ([@B31])
Now, in some clinical models, attachment theory has been applied to a family by supposing that each family member interacts with the other members on the basis of their internalised model of prior relationships. In effect, this view sees family interactions as reflecting individual attachment histories preserved as a generalised "Attachment State of Mind". However, by shifting this framework to the level of a systemic approach, a family therapy can more effectively focus on the family as a single discursive system. To elaborate this idea, we can say a single-family discourse, viewed synchronically, consists of the set of statements in a given family. However, the term "discourse" does not simply refer to an individual's speech acts, but its reception within a given social context. In this sense, discourse requires dialogue. In our therapeutic application, the social context is considered to be the family and the dialogue includes not only speech, but also any actions which have a communicative effect. Such styles of interacting constitute the family discourse which we suggest has consequences for the formation and perpetuation of attachment relationships.
From a diachronic perspective, the family discourse has an historical legacy in the discourses of the parent's own family of origin. Such histories are subjected to a continual process of integration over time into the current family discourse. For example, once a parental couple is formed there is a major integration of two family histories. Similarly, when children are born, there is a further elaboration of the discourse in terms of the experiences of parenting each of their offspring. At any given moment, the family discourse *constructs* a position and role for each family member. Each family discourse consists of an implicit set of rules for what can and cannot be said, and what can and cannot be done ([@B32]). This is quite a different perspective to seeing a family as a conglomerate of "internal working models".
The difference between family discourse and internal working models has a number of consequences. First of all, a discourse is not an internalizes representation of a relationship, it is an external articulation or set of communicative actions. The family's discourse is derived from historical experiences and material which is intergenerational, but as a synchronic function, it is always updating itself and seeking to retrospectively make sense of the past. The discourse is also able to adjust to new circumstances in the present. The family narrative is the process whereby a family draws upon the resources available within its current discourse to construct a temporal account of its history. So, discourse and narrative are closely related, but distinct concepts. The family discourse at any given time is a major work of integration and an attempt to reach a degree of coherence through a process of dialogue, but coherence is only an ideal or a goal. The family discourse is analogous to a myth and could be described in terms of Levi-Strauss's celebrated concept of *bricolage*, since it is pieced together from various threads of narrative, a reconstructive and a retrospective process in which there are always revisions and contested attempts to renegotiate the meaning and significance of the past ([@B33]). There is no possibility of testing the correspondence of the narrative account with the actual historical events in the therapeutic setting. There is only the degree of coherence and consistency of statements within the discourse.
On this basis, we conceptualise our treatment goal as firstly to improve the degree of organization and discursive coherence of the attachment-family system. Any family discourse is on a continuum of being more or less coherent at any given time. The clinical goal is a pragmatic one: for the family's discourse to be coherent enough to provide a platform for family life. Second, the approach is based on the assumption that targeted and strategic interventions designed to promote changes in the relationship between parents and children can modulate both communicative actions and affective states for both parents and children ([@B34]). Changes in ways of speaking, modes of interacting, and different ways of experiencing affects lead to overall shifts in the functional operation of the family. This entails identifying impasses where the dialogical process has broken down or "frozen". We conceptualize therapeutic changes as shifts in the family discourse. There are two major ways in which the dialogue breaks down--- both of which fall outside discourse as such. These are the experience of unresolved trauma or loss, and second the enactment of uncontained affect.
The Limits of Discourse: Trauma, Loss, and Enactment {#s3}
====================================================
A major theme in our clinical work is the predominance of experiences of loss and trauma when undertaking our clinical work with families. There are painful memories, attempts to represent raw events, traumas, loss, bereavement, illnesses--- and these may constitute gaps and elisions, discursive 'black holes' in the realm of what is unspeakable. We find there is particularly rich material in attachment theory to draw on here, especially research on disorganized/unresolved attachment in both the behavior of infants, but particularly the attachment discourse of unresolved adults. Bowlby's work makes a major contribution to the psychology of loss and trauma by showing how permanent losses, prolonged separation from the primary attachment figure, experiences of abuse and neglect are experienced as major assaults on the coherence and function of the attachment system ([@B35]). Later research on adult attachment revealed that the transmission to infants of unresolved experiences of loss and trauma can be predicted even from the attachment discourse of pregnant women ([@B36]). This implies that the origins of an offspring's disorganized attachment are somehow present in the mother's attachment related discourse, prior to even interacting with their infant ([@B37], [@B38]). Therapeutically, the fundamental question here is how to intervene to prevent or reverse such transmission. This is one of the core questions of any attachment-based therapy.
Explanations of this transmission of experiences of trauma and loss across generations generally refer to Main and Solomon's characterization of disorganized infants. These authors employed the ethological concept of "conflict behavior" to explain the paradoxical dyadic interactions of disorganized mother-infant dyads ([@B39]). Others have pointed out the similarity between this concept and the systems theory concept of the double-bind ([@B40]). Bateson referred to the double bind as "some sort of tangle in the rules" or a confusion between the object language and the metalanguage such that several contradictory statements simultaneously direct a behaviour ([@B41], [@B42]). The disorganized-disoriented infant provides a good example of a double bind: the infant is motivated to respond to a threat by seeking the protection and proximity of their primary attachment figure, but in doing so, they encounter not comfort and assuagement, but threat, fear, helplessness, alarm, panic, aggression, and so on--- their attachment system is frozen by an unresolvable paradox due to self-contradictory statements. The point made by attachment theories is that the impasse in the infant's behavior is both precipitated and maintained by the contradictory interactions and communications of the attachment figure.
Mary Main's 1991 paper provides a cognitive explanation by introducing the idea that disorganized discourse results from lapses in the metacognitive monitoring of conversational rationality. She distinguished between single versus multiple models of attachment ([@B43]) referring to the cognitive underpinnings which allow multiple and contradictory models of the same aspect of reality. In effect, Main is using the same kind of explanation as Bateson: a confusion of object language and meta-language. The metacognitive monitoring of the coherence of discourse fails at the point where it needs to provide a consistent and coherent account of trauma or losses.
We generalize this idea to the family discourse and note that contradictory or segregated accounts of a given traumatic experience are often encountered in the clinical setting. Mary Main notes the vivid examples of segregated models of attachment given in Bowlby's discussion of parent's denial and distortion of traumatic events which a child has directly observed: a child may have witnessed a parent's suicide, only to be told that he had died of an illness or accident ([@B43]). Bowlby also referred to examples of a child who found her father's body hanging in a closet only to be told he had died in a car accident ([@B35]). Much of this has been articulated in similar terms within psychoanalytic theory, but our application to work with families is to add the suggestion that the split is not simply internal to the ego, and we do not conceptualise it as an "intrapsychic defense" but think of these contradictions as frozen elements in the family discourse.
The failure to integrate such experiences into a family discourse impacts the family's mode of communication and interaction. Instead of being integrated into the narrative process, sometimes these experiences repeat as triggered enactments and incongruous displays of affect. Enactment can be thought of as a pre-representational means of processing affect through a non-communicative action. Our view on the relationship between discourse, which is by definition social, and affect, which is individually embodied, is related to our concept of enactment. Enactment as a concept has its origins in the psychoanalytic tradition where it is related to repetition compulsion ([@B44]). A great deal more would need to be said about the relationships between attachment models of affect and the psychoanalytic drive theory, but that is well beyond the scope of this paper. The key point clinically is that the management of contradictory family discourses is closely related to conflicted and threat activated emotional systems. The escalation in parent-child conflict is well known in the literature as a very strong predictor of adolescent mental disorder ([@B45]). Families often present with narratives of contests for domination, patterns of threat and counter-threat, adolescents testing their power in response to threat, or using withdraw. Adolescence brings new modes of enactment such as threats to leave home, self-harm, suicide attempts, taking drugs, and so on. Such acts typically occur in the absence of family dialogue and proximity seeking. Addressing enactment, promotion of dialogue and resolving contradiction, defusing patterns of threat and counter-treat, are therefore crucial concepts in the clinical model.
Review of Attachment Related Predictors of Addiction {#s4}
====================================================
Before elaborating these ideas, it is valuable to very briefly review the evidence that can be used to justify a focus on the whole family in relation to adolescent substance abuse. This requires looking broadly across several areas of research in order to understand the kind of experiences and histories which should be the focus of family interventions where adolescent substance abuse is a salient feature. It is important that any psychological theory be posed in terms that are consistent with the most current neurobiological findings of the corresponding phenomena. A number of researchers have pointed out the parallel between psychological processes related to attachment figures, both parental and romantic, and similar mental dispositions in states of addiction ([@B46], [@B47]). Papers are now emerging integrating neurobiological and psychodynamic perspectives into a developmental model on the basis of the findings linking attachment and addiction ([@B48]). One neurobiological model of addiction suggests that deficits in a person's ability to derive rewards from sustained interpersonal or intimate relationships impels reward seeking through the repeated use of psychoactive substances which stimulate these same dopaminergic brain regions ([@B49]). There are animal studies in which exposure to early life stressors predispose to vulnerability to later substance use which point to neural mechanisms involving alteration of neural reward pathways and separation distress regulation ([@B50]). Another line of animal research has proposed gender specific pathways beginning in adolescence. Females predisposed to a heightened stress response are more liable to seek substances as a means of ameliorating high stress reactivity. Males are more likely to respond to chronic stressors with a blunted stress reactivity and their attraction is to substances which increase arousal, increase social capacity, or provide novel sensation such as cocaine and methamphetamine which block dopamine reuptake, and increase dopaminergic activity ([@B51], [@B52]).
The psychological and developmental literature already contains several excellent reviews that have examined the empirical findings showing the relationship between a variety of measures of attachment and different kinds of addiction ([@B53], [@B54]). While it is well accepted that addiction results in the deterioration of the quality of close relationships, Fairbairn's review showed that longitudinal studies have established that attachment insecurity prospectively predicts the development of later substance problems irrespective of the type of measure used. Another interesting finding to come from this review was that the relationship between insecure attachments and substance use was less pronounced in older age groups. The same pattern has been observed in other reviews on the wider relationship between attachment and psychopathology ([@B55]) pointing to the particular importance of the interaction of attachment and developmental processes in adolescence. There have also been interesting findings suggesting that different types of insecure attachment may influence preferences for different substances of abuse ([@B56]).
Unfortunately, at his point in time, the current evidence includes only a handful of studies examining the attachment related discourse of substance using adolescents *via* their performance in the AAI or AAP. These include findings of a strong association between preoccupied-enmeshed and substance use in a sample of orphans ([@B57]). The other adolescent studies of this type have found associations between avoidant-dismissing and unresolved-disorganized representations in a variety of different substance using groups ([@B54]). Adult studies of substance abuse have found associations with Lyons Ruth's hostile-helpless pattern and also with the Main coding of unresolved/disorganized ([@B58]). The main findings of discourse-based measures in adolescence suggest associations between substance use and dismissing forms of insecurity and reasonably consistent findings of high rates of unresolved/disorganized attachments.
The place of trauma and loss in the clinical treatment of patients with substance abuse is also well documented in other studies. It is well established that both Posttraumatic Stress Disorder and bereavement predict increases in substance use and the development of substance use disorders ([@B59], [@B60]). Such findings are consistent with studies on relationship qualities within families showing that adolescent substance abuse is predicted by factors such as low family cohesion, family member enmeshment, and a parenting style known as affectionless control ([@B61], [@B62]). Such findings provide evidence to support the relevance of treatment and prevention goals designed to improve a person's capacity to form and preserve close relationships, be those within a family context or in other close relationships, as a means of either prevention or treatment of substance abuse ([@B63]). With these factors in mind we can now elaborate five therapeutic strategies that have been developed in our clinical work.
The Adolescent as Proxy: The Referral and Presenting Problem {#s4_1}
------------------------------------------------------------
A first area to comment on is the referral process where adopting a systems approach has substantial advantages over the individual model typically used in adolescent mental health services. There often are major challenges in engaging adolescents in any form of psychological treatment and, at the time of initial referral by parents or professionals, the adolescents themselves are sometimes not willing to present for treatment. Within our model the sessions can commence with whichever members of the family are willing to attend. An adolescent's refusal to attend sessions can become a powerful position in the system and can be thought about clinically as a form of communicative action. Refusal may be a signal of a wider refusal to be part of the family's everyday life. This is because underlying the referral of the adolescent and the presenting problems of "substance abuse" is a clinical encounter with a family who often are at a point of fragmentation. At the point of referral, the typical situation is one of breakdown in the major attachment relationships across the family. This is consistent with the empirical findings of a bidirectional relationship between attachment insecurity as both an antecedent predictor of substance use disorder, but also that substance use induces further deterioration in the quality and functioning of close relationships ([@B54]). In some cases, there is strong intergenerational transmission and one is dealing with the adolescent offspring of parents with a history of substance abuse ([@B64]). In the context of the treatment of adolescents still residing in their family of origin, clinical referral often comes at the end of this vicious cycle of deteriorating relationships generating a point of crisis in the family-attachment system. There are important conceptual and clinical questions to be considered even at the point of referral. Who is actually making the referral for treatment? Who in the family system is most willing to consider change? What impasse within the family does the adolescent represent? Referral is therefore not considered to be the referral of an individual with a "mental disorder" requiring that individual to attend and receive treatment. Instead we consider referral to be the referral of a family, as a system, at a point of crisis in that family's history.
A Letter of Invitation: From Helplessness to Action {#s4_2}
---------------------------------------------------
One of the most common comments from a parent at the commencement of the treatment is "It feels like there is *nothing* I can do." ([@B5]). Our clinical work suggests that at the commencement of sessions parents have often adopted a helpless position and probably for quite a long time prior. The concept of helplessness (*Hilflosigkeit*) has deep roots in the psychoanalytic tradition and was revived as an attachment concept by Lyons-Ruth ([@B65]). In the parent's helplessness, one can also recognize a specific dynamic, common in child and family therapy, in which the more helpless the parent, the more domineering the child. It is a family situation of great isolation and disconnection. From a relational perspective, we can see that substance abuse acts as a freezing point in the family discourse and its dialogical movement. On the one hand the adolescent is focused on addictions and these are a one-sided affair, that is substances, while generally reliable, do not "relate back" or make relational demands ([@B66]). Addiction for an adolescent belies a breakdown in the trust that another is capable or willing to respond to their interpersonal and relational needs. On the other hand, the parental helplessness and withdrawal is the parental counterpart and complicit with this freezing in the family system's dialogue.
The first response to this sense of helplessness and isolation is to discuss with parents, either alone or in a small group, the many small ways that they can be effective in relation to their adolescent's problem, how change is incremental and requires persistence, and how they can take action to contribute to improvement in family life. It is critical to do this in a positive way which is very distinct from implying that parents are somehow responsible for their adolescent's disorder. One approach that has been used with some success is to ask parents to write a letter of invitation to their adolescent, telling them that they are attending a group, the concerns they have about current family life, and expressing a desire for change and inviting the adolescent to join with them in attending sessions. The parents work on this letter over the initial sessions of treatment, consulting with the therapists and sharing drafts for comment. Often the parents will be lacking in confidence to produce the letter, feel it will be a useless gesture, or use the writing as a vehicle to vent their own anger and frustration. All this is worked through. The adolescent often receives the letter with surprise and it generates some curiosity. It is both a challenge to the state of helplessness and serves as a gesture of sending a message indicating that the parents are taking the initiative, stepping up as open to dialogue and agents of change.
The Oxygen Mask: Rebuilding the Family Secure Base {#s5}
==================================================
In many cases, what Bowlby referred to as the "emotional atmosphere" of the family is characterised by a vicious cycle of uncontained affect and its behavioral enactment ([@B67]). In other literature, a similar idea might be presented under the concept of expressed emotion. There are common parent-child dynamics in which adolescent withdrawal or aggression triggers parental distress and helplessness. It is clear that the situation with the adolescent is activating basic affective systems in the parent including panic, catastrophic or escalating fear, despair, and anger/aggression ([@B68]). As mentioned above, the attachment perspective understands these affective systems as threat activated affects which trigger basic survival systems. They do so by shutting down affiliative and care-giving motivational systems. We also find that systemically these vicious cycles of affect and enactment can escalate to such a degree that they precipitate a premature rupture in the family-attachment system. The adolescent seeks to achieve a kind of pseudo-independence in which they sometimes leave or sometimes remain physically within the family, but are psychologically cut off within the family, unable to access any sense of security *via* intersubjective relations within the family. This may take the form of an externalising presentation in the context of substance use which often consists of various conduct and "anti-social" problems, taking up with their peer group, in some cases spending little or no time in the family unit. Another permutation is the withdrawal of a depressed adolescent within the family--- the parents describe them as moody, difficult to reach or living in a virtual world of social media ([@B10]).
An important concept derived from attachment thinking which we use to both understand and respond to such situations is John Byng-Hall's concept of the "secure family base". He uses this term to describe the family foundation from which an adolescent can safely explore their social world ([@B11], [@B69]). The notion of a secure family base refers to the parental function and it assumes to some degree a unified parental position. We have encountered several obstacles to the parents facilitating the family operating as a secure base.
First, we often encounter a challenge within the parental couple itself who, under enormous stress, find it difficult to present a unified front. Instead it is common that they turn on each other and split off into polarised reactions to a challenging situation. This is understandable within a context where each parent brings their own attachment history, styles of defense, and their own ways of having traversed adolescence and the position of parenthood. Therapeutic discussion of these three moments: the parent's own childhood attachment histories, their traversal of adolescence, and their assumption of parenthood--- can be a source of significant therapeutic gain. The therapist needs to be looking out for when this parent has been able to make use of a reparative attachment experience with a reflective other. For example, it is not unusual that a parent may have worked-through an adolescent period of rupture with their own parents, but later made reparation when they formed a new couple relationship by making use of new capacities derived from their romantic relationship.
Second, we frequently encounter within disorganized family dynamics histories of role reversal emerging over early and middle childhood. The same dynamic has been described using various different terms in psychoanalytic, systems, and attachment theory ([@B70], [@B71]). Role confusion and reversal begins with the primary attachment figure not providing care to the infant, but in numerous different ways and circumstances seeking or requiring that care themselves. As noted by Lyons-Ruth, it is not unusual to also uncover in such parents histories of victim/aggressor relational patterns, patterns of withdrawal in the face of the child's attachment demands, and a critical failure to regulate the child's attachment need in those moments where assuagement of distress is most needed ([@B72]). The child's defensive positioning within this dynamic as "parentified" takes up a subjective position of control in their relation with others, an objectification of others as objects to be controlled, and perceives that there is an absence of any anyone else "taking control". There are elements of both grandiosity and narcissism at play in the child's position and in adulthood this can develop into a personality style which seems to exude a high degree of "competence". However, from a clinical point of view, the predominance of role reversals between caregiver and care-receiver bellies a major alteration in family structures by placing the child in the dominant and controlling position. The most obvious form this takes in adolescence is a control that takes an aggressive and commanding form, but equally the adolescent's withdraw in the more internalizing presentations can be seen as a mode of control.
The therapeutic response is twofold. First, to rebuild the family as a secure base and this entails numerous different techniques designed to allow parents to contain their distress, redefine their roles as supportive, see themselves as setting examples of coping with stressors and taking responsibility for problems. There are also a range of techniques to rejuvenate the dialogue by parents showing they are willing to change and adapt, communicate, compromise, and negotiate and to expect the same of their adolescents. Here amongst other approaches, we make use of a metaphor based on the use of the Oxygen mask--- "In the aeroplane, safety instructions suggest that the parent secure their oxygen mask *before* assisting their children..."
The idea of this and various other components of the treatment is to promote the parent's adoption of the position of the family secure base. This foundational position could also be likened to a sounding board, which facilitates the reconstruction of family discourse. The second aspect presupposes the first has been achieved to some degree and is based on an encouragement that in the context where the family has achieved a secure base, an adolescent will seek to explore. We reframe such "explorations" in terms of their importance in the adolescent resuming their developmental pathway towards autonomy, to re-negotiating their relationship with their parents as they enter adulthood, and are able to make use of the availability and receptivity of their parents in new ways.
Red Buttons: Intersubjective Regulation of Affect {#s6}
=================================================
There are important systemic factors which perpetuate family conflict. It is also apparent that conflict based on mutual aggression can become a vicious cycle of affective dysregulation, where aggression triggers the escalation of threat which in turn triggers further aggression, described by Bateson using the concept of schismogenesis ([@B73]). This cycle can become particularly vicious when, as occurs for many parents in our intervention, the idea of the adolescent's separateness generates panic. Separateness is not greeted as a developmental achievement, but as a threat to their child and the integrity of the family. Parents then see it as their role to intervene to "prevent damage" occurring, and this can be very acute in relation to drug use, but this implicitly sends a message to their adolescent that they are considered "incompetent" or cannot "cope on their own". It also sends a message that their adolescent is in grave danger, but the adolescent is considered to lack the skills to keep themselves safe. In our experience, this can generate an emotional atmosphere which does have features of alexithymia, but with a heightened sense of panic and an aggressive battle for control. This drives the adolescent further away and undermines the adolescent's developmental process of autonomy seeking, building social confidence, and sometimes taking risks.
These are themes discussed in our intervention around a series of metaphors. These take the form of stories of separation, autonomy, risk, and adventure designed to evoke discussions of separation as a key developmental process in adolescence. The generation of the family as a secure base requires that the parents stabilise their affective responses to these sometimes-threatening themes. Such stabilization occurs through discursive coherence and the capacity to speak about and think about, rather than enact these powerful affective experiences.
Clinically addressing family conflict and aggression is a core part of our approach. Above all we emphasize that effective communications cannot occur in the context of conflict and hostility. The first step is often to help families recognise the degree of aggression and conflict inherent in many of their interactions. It is critical to have therapeutic discussions naming the kinds of emotional experiences the parents are having. We also discuss common "hot spots" which are points in family life which tend to generate conflict- getting out of bed, going to bed, getting to school on time, etc. We often appeal to the idea that parents need to model taking control not of their child, but of their own emotions, to recognise when they are feeling "out of control" and to curtail interactions based on that recognition. Parents are encouraged to regather their self-control and then return to seek dialogue. Often this is discussed, modelled and even roll played with the therapists.
Sometimes, a "circuit breaker" is needed by way of intervention. One idea arose from recounting the experience of one of the participants. One of our dads used to talk a lot in the sessions about his "Red buttons"--- these were the ones his daughter knew well how to push! And the two of them were often triggering aggression in each other to the point where they were often unable to inhabit the same space. One day, anticipating an argument, the dad came into his daughter's room with an actual red button stuck on his sleeve and said to her "Do you just want to push it and get that bit over with and then we can discuss the issue" She laughed and there was a shift ... This has become a story we tell within sessions since it very nicely illustrates how a parent can redirect what was typically an aggressive enactment onto a discursive level through the use of humour.
Bumps in the Road: Integrating the Narrative of Loss and Trauma {#s7}
===============================================================
Therapeutically, there is great benefit in addressing segregated systems at a family level. There are a variety of techniques that encourage a family unit to collectively work through their narrative of traumatic experiences, or family losses or other major setbacks. The approach is to ensure that this is done in a manner in which all family members can contribute and where therapists are proactive in seeking clarity, in asking for other's versions of events, and to encourage a goal of "setting the record straight". In our model, we used a simple drawing technique called "bumps in the road" in which family units are asked to draw their family road trip along a "rocky road" with the bumps and pitfall labeled along the way. Who is driving their car, who are the passengers? When has it needed repairs? It generally takes some time before the family is ready for this task following earlier work to build the family secure base and a sense of trust in the therapeutic process. Often there are highly impactful sessions where a sense of both clarity and the theme of "how did we survive it all" emerges.
Conclusions {#s8}
===========
To conclude, this paper has tried to give a sense of how attachment research and theory can be used to inform and develop a family-based treatment approach for adolescents with mental health issues--- including substance use. There is compelling evidence that there are higher rates of attachment insecurity in substance using adolescents and also strong evidence for histories of trauma, loss, and family conflict. Alongside what we now know of the neurological processes involved in addiction and their links to social affiliative systems, this justifies the need for an attachment approach to such family-based treatments.
The basic theoretical commitment of the BEST approach is based on the claim that the "move to the level of representation" in attachment theory, can be reconsidered as a properly inter-subjective and linguistic model, compatible with family systems theory. This is broadly consistent with those approaches which could be called the "linguistic turn" in psychotherapy. These approaches all emphasize language as a means to generate meaning in shared patterns of communication, and that meaning can take the forms of action and interaction. There are limits to meaning generation in terms of enactment and overwhelming experiences of affect. The concern with language, meaning, dialogue, and narrative are widely shared by systemic approaches such as narrative therapy ([@B74]) postmodern therapy ([@B75]) and dialogic family therapy ([@B76]), and by contemporary psychodynamic approaches such as Lacanian and Neo-Kleinian psychoanalysis. This paper has attempted to bring these elements of systems and psychoanalytic thinking together with discourse-oriented research within attachment theory ([@B40], [@B77], [@B78]).
Certainly, BEST is not the only family systems model to draw on attachment theory and comparison can be made to Attachment Based Family Therapy (ABFT), which is a similarly manualized and evidenced based approach, which has shown impressive results with depressed and suicidal adolescents ([@B20]). Very briefly the main theoretical differences between BEST and ABFT would appear to be the former's emphasis on discourse and narrative as the aspects it draws from attachment theory. However, both approaches have similar overall goals and what appear to be some similar techniques to reduce family conflict, promote affect regulation, build attachment relationships, and encourage adolescent autonomy on the basis of strengthened family relationships. A detailed comparison of the two approaches would be a promising avenue for future research.
Treatments for adolescent substance use will clearly benefit from strategies designed to enhance not only attachment security, but the organisation of attachment related discourse. Such changes provide the secure family-base which enables an adolescent's continuation of the developmental process into adulthood. Underlying attachment vulnerabilities are maintained not only in the representational models of the individual members, but also as interactional patterns and modes of communication within families. We propose that this discursive level of the family system can be a target of a number of specific techniques. Families as a whole can be engaged in these techniques and new approaches to patterns of communication and connectedness used as a means of engaging substance using, depressed, or suicidal adolescents.
Author Contributions {#s9}
====================
The author confirms being the sole contributor of this work and has approved it for publication.
Funding {#s10}
=======
Evaluation studies of BEST have been funded by the Australian Research Council (grant number LP110200167) and also Beyondblue, Drummond St Services, Australian Drug Foundation, Turning Point and WA Department of Justice and supported by Deakin and Murdoch Universities.
Conflict of Interest {#s11}
====================
The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The current paper is derived from a presentation to the conference *Sucht und Bindung* held in Vienna, 16^th^--18^th^ May 2018 and hosted by the Grüner Kreis Society. I would like to thank the conference organizing committee and in particular Dr Human-Fredrich Unterrainer for their kind invitation to present this work. Many thanks also to conference participants who offered comments and suggestions for this paper. Evaluation studies of BEST have been funded by the Australian Research Council (grant number LP110200167) and also Beyondblue, Drummond St Services, Australian Drug Foundation, Turning Point and WA Department of Justice and supported by Deakin and Murdoch Universities. Many colleagues over the years have contributed to the development of these clinical techniques and the evaluation of BEST. These include Irene Serfaty, Kim Kho, Michelle Benstead, Renita Almeida, Lucy Poole, Tess Knight, John Toumbourou, John Bamberg, Nic Cecic, Campbell Paul, Reima Prior, Dan Lubman, Melanie Bertino, Joanna Skewes, and Karen Field.
[^1]: Edited by: Marc N. Potenza, Yale University, United States
[^2]: Reviewed by: Mauricio Alvarez-Monjaras, University College London, United Kingdom; Domenico De Berardis, Azienda Usl Teramo, Italy
[^3]: This article was submitted to Addictive Disorders, a section of the journal Frontiers in Psychiatry
| {
"pile_set_name": "PubMed Central"
} |
1.. Introduction {#sec1}
==================
The macromolecular crystallography (MX) experiment lends itself perfectly to high-throughput technologies, automation and remote experimentation. The experiment comprises a series of distinct steps, beginning in the wet laboratory with protein expression, purification, crystallization and crystal mounting using flash-cooling in liquid nitrogen, and progressing through to the screening of crystals for diffraction quality, the collection of diffraction data, data processing and structure determination. Most of these steps have been fully automated, and in many cases it is now possible to go from expressed protein to fully determined three-dimensional structure with only minimal intervention. However, several steps still require expert human intervention, including the choice of crystal for data collection. Since the ultimate goal of the experiment is to produce a high-quality high-resolution structure of the protein in question, this relies heavily upon the choice of the best possible crystal for data collection and the most appropriate data-collection strategy. In this regard, the careful training and education of students and novices is of fundamental importance to these aspects of the process and cannot be overlooked, however much automation and remote access are involved in the experiment.
Some of the most important developments in the automation of protein expression, purification and crystallization have taken place under the auspices of the NIH-funded Protein Structure Initiative (Burley *et al.*, 2008[@bb1]). With regard to high-throughput crystal screening and data collection, many facilities and groups worldwide have developed automated sample changers, including Abbot Laboratories in Illinois, USA (Muchmore *et al.*, 2000[@bb12]), DORIS in Hamburg, Germany (Karain *et al.*, 2002[@bb7]; Pohl *et al.*, 2004[@bb21]), the Spring8 synchrotron in Japan (Ueno *et al.*, 2004[@bb29]), the European Synchrotron Radiation Facility in Grenoble, France (Ohana *et al.*, 2004[@bb14]; Cipriani *et al.*, 2006[@bb2]) and the Advanced Light Source (ALS) in Berkeley, California, USA (Snell *et al.*, 2004[@bb26]). In an effort to produce a true high-throughput crystal-screening and data-collection facility, and to improve the efficiency of the synchrotron radiation resource, the Stanford Synchrotron Radiation Lightsource (SSRL) Structural Molecular Biology (SMB) Group and the Structure Determination Core of the Joint Center for Structural Genomics (JCSG) (Lesley *et al.*, 2002[@bb9]) worked together to develop the Stanford auto-mounting (SAM) system (Cohen *et al.*, 2002[@bb3]). In addition to complete automation of the experiment, SSRL has also implemented fully remote access to the MX beamlines (Soltis *et al.*, 2008[@bb27]).
Notwithstanding the obvious increase in throughput and efficiency, the advent of automation and remote access at the SSRL MX beamlines has generated substantial spinoffs for the scientific user community by providing increased opportunities for collaboration between research groups and allowing scientists who might not typically have had access to a national user facility to obtain valuable beam time. It has also introduced many young scientists to synchrotron radiation science by providing educational and training opportunities for graduate students and postdoctoral researchers in user laboratories. The scientific staff at SSRL offer in-house training workshops and have run remote-access workshops around the US and at international sites. Attending one of these workshops is strongly encouraged before taking part in remote-access beamtime. Furthermore, often the most effective training is from the experiences gained during remote-access beamtime, when new researchers conduct their own experiments under the advice and encouragement of other members of the home laboratory and of SSRL User Support scientists, who are readily available *via* cellular telephone, email and a 'chat' feature (instant messaging) in the *BLU-ICE*/*DCS* beamline control system.
2.. Synchrotron radiation research at SSRL {#sec2}
============================================
SSRL has a long history of excellence in structural biology research, including some of the first reports of X-ray absorption spectra from a biological sample (Kincaid *et al.*, 1975[@bb8]), the first published report of single-crystal diffraction from protein crystals using synchrotron radiation (Phillips *et al.*, 1976[@bb20]), fundamental studies of what would become the multiple-wavelength anomalous diffraction phasing experiment (Phillips *et al.*, 1977[@bb19], 1978[@bb18]; Phillips & Hodgson, 1980[@bb17]; Templeton *et al.*, 1980[@bb28]) and the development of insertion devices as sources of high-intensity radiation (Doniach *et al.*, 1997[@bb4]).
SSRL is a national user facility funded by the US Department of Energy Office of Basic Energy Science, the National Institutes of General Medical Sciences (NIGMS) and the National Center for Research Resources, the latter two being components of the US National Institutes of Health (NIH). SSRL provides extremely bright X-ray and UV photon beams produced by the third-generation 3 GeV SPEAR3 storage ring, for applications in materials science, environmental science, chemistry and structural biology research, utilizing scientific techniques including photoelectron spectroscopy, small-angle X-ray scattering (SAXS), X-ray absorption spectroscopy (XAS), total X-ray reflection fluorescence and MX.
The SMB group at SSRL (<http://smb.slac.stanford.edu>) operates and maintains ten beamlines, seven for MX (BL1-5, BL7-1, BL9-1, BL9-2, BL11-1, BL12-2 and BL14-1), two for biological XAS (BL7-3 and BL9-3) and one for biological SAXS (BL4-2). All seven MX beamlines at SSRL are fully automated, employing the SAM system which has been integrated into the *BLU-ICE*/*DCS* beamline control system and graphical user interface developed earlier at SSRL (McPhillips *et al.*, 2002[@bb11]). Up to 288 crystals can be screened in a matter of hours without manual intervention using this reliable and robust robotic system. The use of the SAM system has not only seen an increase in throughput by research groups but also an improvement in the overall quality of the diffraction data being collected. Researchers are now able to screen all their crystals reliably and take advantage of the automated image-analysis tools developed at SSRL, prior to choosing the best quality crystals for subsequent diffraction data collection. These tools include the Crystal Analysis server, which will automatically analyze test images and feed relevant parameters and statistics back to the researcher *via BLU-ICE*, and the browser-based *WEB-ICE* interface (González *et al.*, 2008[@bb6]), where diffraction and video images of the samples can be viewed, crystals ranked and data-collection strategies calculated.
2.1.. Automation {#sec2.1}
------------------
The seven SSRL MX beamlines are all very similar, in that the experimental table, front-end beam-conditioning system, kappa goniometer, cryosystem and detector positioner are nearly all identical. The undulator micro-focus beamline (BL12-2) differs somewhat in design to meet the more demanding hardware requirements for microbeam and microcrystal experiments, but is still compatible with the SAM system and standard beamline control software. Every aspect of beamline control inside the experimental hutch, and also on the upstream optics elements (mirrors, monochromators and slits), is motorized to the extent that it is unnecessary to enter the hutch to change any of the experimental parameters (X-ray energy, beam size, X-ray detector position, fluorescence detector position, beamstop position, attenuation and lighting), to mount or dismount samples, or to anneal or wash ice from samples. This degree of automation of the beamlines is absolutely critical to the implementation of fully remote access; if there remains a single task that requires human intervention inside the hutch during the normal course of crystal screening and data collection then remote access is not practical.
Automated sample mounting was made available to general experimenters during the first SPEAR3 run of 2004 on three beamlines. Since its inception, use of the SAM system has accelerated such that, during the last scheduling period (2009), 110 out of 121 research groups (91%) were using SAM during their experiments. The SAM system has been described in detail previously (Cohen *et al.*, 2002[@bb3]; Smith & Cohen, 2008[@bb25]; Soltis *et al.*, 2008[@bb27]). During the first year of operation (2004), 30 research groups used the automated mounter and over the course of 60 experimental starts mounted over 3500 crystals. The JCSG, one of the original SAM test user groups, mounted an additional 2000 or more crystals from 125 target proteins that year, and were successful in solving 30 new structures from 36 unique proteins (Smith & Cohen, 2008[@bb25]). The number of crystals mounted using the SAM system has also increased dramatically since it was first introduced, such that currently well over 300 000 crystals have been screened by researchers (Fig. 1[▶](#fig1){ref-type="fig"}).
2.2.. The remote-access experiment {#sec2.2}
------------------------------------
Fully remote access was made available to research groups during the 2005 scheduling period. During the first two years the number of research groups choosing to conduct their experiments remotely rose from 24 to 44%, and has continued rising each year (Fig. 2[▶](#fig2){ref-type="fig"} *a*) until the last scheduling period, which saw 105 of the 121 groups (87%) screening their crystals and collecting their data using remote-access tools. Most noticeably, the total number of remote starts saw an almost exponential growth in 2007 (Fig. 2[▶](#fig2){ref-type="fig"} *b*), which can be primarily attributed to an increase in beamline efficiency (fewer beam-hours per start) as the coupled use of the SAM system and remote access became more popular. This increase in beamline efficiency can also be seen in the total number of crystals mounted *via* the SAM system since its inception, which also experienced a dramatic rise in 2007 (Fig. 1[▶](#fig1){ref-type="fig"}).
The remote-access experiment at SSRL has been described previously (Smith & Cohen, 2008[@bb25]; Soltis *et al.*, 2008[@bb27]). Scientists ship their cryo-cooled samples to SSRL in 96-port cassettes custom-designed at SSRL for use with the SAM system, or in 16-port Uni-pucks (<http://smb.slac.stanford.edu/robosync/Universal_Puck>). The cassettes have been designed such that two can be shipped in a standard dry shipper (192 crystals in total). Up to seven Uni-pucks (112 crystals in total) may be shipped in a standard dry shipper. The Uni-pucks have been designed as part of a collaboration between developers at synchrotrons throughout the United States, allowing research groups to take advantage of automated sample-mounting systems at different synchrotron facilities (<http://smb.slac.stanford.edu/robosync/>). The Uni-pucks are based upon the ALS-style puck, and are currently used with the SAM robot at SSRL, with many ALS-style robots at the three other large DOE-funded synchrotrons in the US (ALS, the Advanced Photon Source and the National Synchrotron Light Source), with the ACTOR robot (Rigaku, USA), and with various other sample-mounting robots in Europe, Australia and Asia. At SSRL, four Uni-pucks are mounted in an adaptor cassette such that the sample pins can be accessed by the SAM system in the same way as it accesses sample pins in an SSRL cassette.
During their allotted beam time, the remote researchers connect to the beamline computers *via* an NX server/client application (<http://www.nomachine.com>). The NX client is downloaded for free onto the researchers' home computers, and they can then connect to an NX server running on an SSRL computer. The client uses minimal CPU and memory resources on the host computer, with the entire computational load on the SSRL server. Once connected, the researchers see a remote desktop (Fig. 3[▶](#fig3){ref-type="fig"} *a*), identical in all aspects to the environment they would see on a computer at the beamline. They can then use the *BLU-ICE* control interface (McPhillips *et al.*, 2002[@bb11]) and/or the *WEB-ICE* interface (González *et al.*, 2008[@bb6]) to screen their crystals and obtain results directly back into the *BLU-ICE* screening interface (Fig. 3[▶](#fig3){ref-type="fig"} *b*), collect monochromatic diffraction data, measure absorption edges prior to multiple- (MAD) or single-wavelength anomalous diffraction (SAD) data collection, monitor all aspects of the experiment, and connect to User Support staff and collaborators *via* a real-time chat feature. In fact, everything that a crystallographer would typically do during a synchrotron data-collection visit can be achieved in the remote-access experiment.
The remote desktop also gives researchers access to all the crystallographic software installed on the SSRL computers, for data processing, structure solution and analysis. Although experimental control, decision making and strategy calculation are carried out in the home laboratory by the researchers and their students, research associates, postdoctoral fellows and/or collaborators (Soltis *et al.*, 2008[@bb27]), SSRL User Support staff are available to troubleshoot experiments, help analyze the screened crystals or advise on data-collection strategy if required. This contrasts with the options that other synchrotrons offer, known as 'service', 'mail-in' or 'FedEx' crystallography, whereby researchers send their cryo-cooled samples to the synchrotron but the decision making and data collection are carried out solely by beamline staff (Robinson *et al.*, 2006[@bb22]), or the more limited telepresence described for a small-molecule crystallography beamline at Daresbury (Warren *et al.*, 2008[@bb30]).
3.. Training and collaboration {#sec3}
================================
Based upon feedback from recent SSRL remote-access workshops, remote-access demonstrations at national and international meetings and conferences, anecdotal evidence from informal discussions with research groups, and a recent remote-access survey sent to research groups who regularly use SSRL, the remote-access capabilities have not only revolutionized the way in which diffraction data at synchrotrons are collected but also changed the way in which graduate students and postdoctoral researchers, new to crystallography or synchrotron data collection, are introduced to the area and trained. The general consensus is that the remote-access capabilities at SSRL are a useful tool in training graduate students and postdoctoral fellows in the collection of good quality diffraction data.
Prior to automation and remote access, a research group comprising, on average, three laboratory members (perhaps one or two experienced people and some graduate students) would undertake a synchrotron data-collection trip and spend 48--72 h continuously screening crystals and collecting diffraction data. Since the first beamlines were developed and made available to the general scientific community, a synchrotron data-collection trip has almost been viewed as a rite of passage for scientists, young postdoctoral fellows and graduate students. It is quite likely that most, if not all, synchrotron beamline users can remember the first time they set foot in one of these laboratories. In recent years, with the increased pressure on funding, the use of research grants to take a large group of scientists to a synchrotron beamline has become uneconomical, particularly given the trend towards increased numbers of crystals being produced in some laboratories, which necessitates more and more access to beamlines. Although the use of a national user facility such as SSRL has no direct cost associated with it (it is mandated that such facilities give free access to US and international scientists at academic institutions), there are still significant costs involved with travel and accommodation (Table 1[▶](#table1){ref-type="table"}). With the advent of remote-access data collection, new students or other laboratory members who would not normally be sent on a data-collection trip are now exposed to the synchrotron resource, and this access provides valuable experience for their future careers in science.
Fatigue from travel and prolonged presence at the beamline form a hurdle which has, on occasion, given rise to errors and mistakes during mounting of the crystals, analysis of the diffraction or determination of the optimum collection strategy. Prior to the incorporation of the robotic sample mounter, the screening of flash-cooled crystals typically involved manual mounting using cryo-tongs pre-cooled in liquid nitrogen, which enclose the crystal (mounted in a fiber loop at the end of a sample pin) inside a hollow cavity (Parkin & Hope, 1998[@bb15]; Rodgers, 2001[@bb23]; Pflugrath, 2004[@bb16]; Smith & Cohen, 2008[@bb25]) to maintain the crystal at cryogenic temperatures during transfer into the experimental hutch and onto the goniometer. Although this method has proved to be very reliable since its inception in the 1990s (Pflugrath, 2004[@bb16]), it becomes laborious and tedious when repeated many times. The skill and patience of the experimenter, rather than the number of samples available, have often dictated the quality of the crystal selected for data collection; crystals were screened manually until a crystal deemed 'good enough' to collect a complete diffraction data set was found. In cases like this, other crystals from the same project would go unscreened; if a better quality crystal were among those which were unscreened, it would go undetected and uncollected.
The process of crystal screening, crystal selection and data-collection strategy determination has become significantly easier with the implementation of the SAM system, the Crystal Analysis server and *WEB-ICE*. As noted above, useful crystal parameters and statistics \[including the Bravais lattice, the unit-cell parameters, the estimated mosaicity, the predicted resolution, the r.m.s. fit from *MOSFLM* (Leslie, 1992[@bb10]) and an overall score\] are continually fed back into the *BLU-ICE* spreadsheet (Fig. 3[▶](#fig3){ref-type="fig"} *b*), and these are also accessible through *WEB-ICE*, where researchers can also inspect the diffraction images and crystal video images. The availability of screening results and the crystal analysis have provided a new resource for training novice crystallographers during the experiment. Researchers can easily access and compare diffraction images, video images of each crystal and the results of the Crystal Analysis server to decide how best to proceed. For example, a crystal may need to be rescreened because the best part of the crystal was not in the beam, or perhaps the crystal may need washing as it was covered with surface ice (visible on the crystal images and as strong ice rings on the diffraction images), or the automated strategy may be confirmed as a good approach for subsequent data collection. Access to all this information through *WEB-ICE* makes it easier to teach novice crystallographers when to use automated results and when to question them.
It is undeniable that hands-on experience with the control systems of a synchrotron beamline, and the ability to analyze and monitor the data as they come off the detector in real time, are vital not only to the collection of the best possible diffraction data (which will ultimately lead to the best possible structures) but also in the training of the next generation of synchrotron beamline users. Our contention, which is thoroughly backed up by the feedback we have received over the past five years, is that the training being received by students and novices *via* SSRL User Support staff and the SSRL remote-access tools is fully comparable with the on-site training they would have received had they made an actual trip to SSRL or other synchrotron facilities. In most cases this is a guided participation approach, whereby an experienced researcher, principal investigator (PI) or SSRL User Support person will demonstrate the fundamental aspects of the system to perhaps a small group of students or novice group members, and then guide them through the experiment as they take control of the *BLU-ICE* or *WEB-ICE* interface. It is well understood that people learn by different methods, whether it be through observation, analysis, discussion or activity, or a combination of these. The remote-access tools available to the SSRL user groups offer something to all types of learner and therefore provide a very effective method of teaching the new user the best possible ways in which to collect the highest quality diffraction data, this being the ultimate goal of any X-ray diffraction experiment.
Direct contact with SSRL User Support staff is strongly emphasized as being the important first step in remote training for any research group. The User Support staff have a vast amount of knowledge and expertise with the SSRL beamline systems, the SAM robot and the remote-access capabilities, and can direct researchers to the appropriate information and resources to make their group training, and ultimately their valuable beam time, a most effective and efficient process. Moreover, SSRL User Support staff can effectively facilitate remote training with a research group over the telephone, employing all the remote-access tools available to the research group. These tools include (i) access to the *SSRL User Guide*, (ii) access to a number of video tutorials which illustrate various steps in a remote-access data-collection experiment, (iii) connection to a 'simulated' beamline, facilitated through SSRL User Support staff, (iv) information on software packages installed and supported on SSRL computers (<http://smb.slac.stanford.edu/public/facilities/software/>), (v) access to test images and data sets so that the processing software and structure-solution software and scripts can be tested by or demonstrated to students and novices, (vi) use of the chat feature in *BLU-ICE*, and (vii) use of the shared desktop capabilities of the NX server/client interface, whereby SSRL support staff can demonstrate the *BLU-ICE* or *WEB-ICE* interfaces while a remote research group follows on their local computers. The full capabilities of the NX desktop-sharing tools are described on the developer's website (<http://www.nomachine.com>).
3.1.. SSRL User Support {#sec3.1}
-------------------------
The SSRL User Support staff are a group of expert crystallographers and engineers who are available before, during and after beam time for consultation and practical help. Typically, one staff member is responsible for a given beamline for a specified period, and research groups can determine who their particular support person will be from the online User Support schedule (<http://smb.slac.stanford.edu/schedule/sch_staff.cgi>). As noted above, research groups are strongly encouraged to contact the responsible staff member by either telephone or email prior to upcoming remote-access beam time to discuss beamline characteristics, sample preparation, and experimental design and strategy, to gain access to the simulated beamlines, to test connectivity through the NX server/client system, and to organize either pre-beam remote training or training once their beam time starts. The use of remote training as a teaching tool in research laboratories assumes the presence in the research group of an experienced user of the SSRL beamlines and the *BLU-ICE* or *WEB-ICE* interfaces who can facilitate this training. If the research group is new to SSRL then this may not be the case, and under these circumstances we strongly recommend that the group send at least one representative to either an on-site or a remote SSRL workshop to gain hands-on experience with *BLU-ICE* and *WEB-ICE*, the SSRL computing systems, and in the use of the cryo-tools associated with the SAM system, the storage and transport options available, and the proper sample preparation techniques. Sample preparation is absolutely critical to the success of the experiment, irrespective of whether it is on-site or remote. These trained scientists can then return to their laboratories and facilitate the training of group members in the use of these systems, with the assistance of SSRL User Support staff. A comprehensive description of the tools and their use, along with correct sample-pin selection and preparation, is also available through the SMB website (<http://smb.slac.stanford.edu/public/users_guide/manual/Using_SSRL_Automated_Mounti.html>).
Once screening and data collection are underway, staff are also on hand to help with connectivity problems or beamline troubleshooting, to give *BLU-ICE* or *WEB-ICE* help, and to give direct experiment-related advice regarding crystal selection, data-collection strategy determination, processing software help and data backup. Staff can contact remote scientists by telephone, by email or using the chat feature in *BLU-ICE*, and researchers can contact staff using the same methods. SSRL User Support staff contact details are available on the SMB website (<http://smb.slac.stanford.edu/public/staff/index.shtml>).
3.2.. *SSRL User Guide* {#sec3.2}
-------------------------
The SMB group website (Fig. 4[▶](#fig4){ref-type="fig"}; <http://smb.slac.stanford.edu>) contains up-to-date information for research groups on the state of the MX beamlines, the beamline schedule and the SPEAR accelerator status, with links to the computing and software resources available (through the *Facilities* tab), and to the *User Guide* (<http://smb.slac.stanford.edu/public/users_guide/index.shtml>). The *User Guide* is available online to all users at any time, irrespective of whether they have beam time, and can be downloaded as a PDF file. The guide gives a detailed description of all aspects of MX experiments at SSRL, from becoming an SSRL user, to detailed instructions on the use of the SAM system and the preparation of samples, and how to use the *BLU-ICE* and *WEB-ICE* interfaces effectively to set up and carry out a crystal-screening and data-collection experiment. The differences between an on-site and a remote experiment are clearly defined, such that novices and first-time remote-access users have all the information at hand prior to the start of their beam time. Information specific to the collection of MAD data and high-resolution monochromatic data are presented, and the data-processing software packages available to researchers are described, along with short tutorials on the most effective use of these programs. A set of detailed answers to frequently asked questions (FAQs) is also included at the end of the *User Guide* to aid users in their experiments, and to help with programs and with questions should they arise.
3.3.. Video tutorials {#sec3.3}
-----------------------
The video tutorials can be accessed from the *User Guide* page of the SMB website as given above, or *via* the link <http://smb.slac.stanford.edu/public/users_guide/tutorials/>). This project is constantly being developed and updated as new beamline capabilities and tools become available. Current tutorials include those that give information on tasks that can be carried out prior to beam time, such as (i) downloading and installing the NX client software, (ii) the best ways to fill in the *Excel* spreadsheet with crystal information for a remote-access or on-site SAM-assisted experiment, and (iii) instructions on how to upload the completed spreadsheet to the crystal database prior to or at the beginning of the user beam time. Three additional videos describe (iv) the SAM-assisted remote-access experiment in detail, demonstrating how to use the SAM system to screen crystals in a cassette, (v) how to interpret the screening results subsequently to select crystals for data collection and (vi) a simulated *WEB-ICE* strategy calculation. A strategy calculation for a MAD or SAD data collection is also demonstrated.
3.4.. Simulated beamlines {#sec3.4}
---------------------------
Prior to the start of beam time, the members of a research group can connect to the SSRL computers and gain access to a 'simulated' beamline. The seven SSRL beamlines each have a simulated counterpart which can be accessed in exactly the same way as the 'real' beamlines. Access is only possible by contacting one of the SSRL User Support staff beforehand and asking for authorization on one of the simulated beamlines. Following authorization, the remote user connects to the simulated beamline through a *BLU-ICE* interface indistinguishable from the one that will be used later to screen crystals and collect data. All the motors that control experiment variables, such as beam size, detector distance, X-ray energy and the beamstop position, can be moved. Since the remote user is not actually connected to a real beamline, this does not affect experiments currently being carried on the real counterpart of the simulated beamline. The cassette spreadsheet can be uploaded and new users can then be taken through the steps involved in crystal screening by the experienced users in the group.
The simulated beamlines are an extremely valuable resource for a research group that may be new to remote-access data collection, the SSRL beamlines or synchrotron data collection in general. The best use of these simulated beamlines involves the inclusion of a member of the SSRL User Support staff in the remote training exercise, whereby the use of the *BLU-ICE* interface on the simulated beamline can be fully described and discussed in detail with all members of the group. This can be facilitated by a telephone call or by use of the desktop-sharing tools available with the NX server/client software (<http://www.nomachine.com>).
3.5.. Multiple NX connections {#sec3.5}
-------------------------------
In most remote-access experiments, there are generally several experienced people in the home laboratory responsible for the data collection. Because the NX client system allows multiple connections with the same user account, experienced users can passively monitor the screening and data collection being carried out by students or postdoctoral researchers, which still allows the students their independence and involvement in the decision-making process, yet allows for the correction of mistakes or the suggestion of alternative strategies. This capability also makes it easy for SSRL User Support staff to monitor the screening and data collection, and to step in if they see a potential problem. Multiple connections under the same user account can have the name and telephone number of the scientist associated with each one in the *Users* tab of *BLU-ICE*, making it easy to identify who is currently active should User Support staff wish to contact the researcher.
This can be extended beyond the home laboratory to the laboratories of collaborators, who can also connect during active beam time, again with a name and telephone number associated with the connection on the *Users* tab, either passively to monitor the data collection, or actively to play a role in the screening, analysis and choice of crystals, or the data collection. The general consensus amongst SSRL research groups is that providing beamline access to collaborators under the auspices of their proposals has given these collaborating scientists and their group members exposure to synchrotron beamlines that they would never have been able to obtain without remote access. In some cases, this exposure has led to these collaborating scientists writing their own successful proposals for synchrotron beam time. A prime example of this is the beam-time proposal submitted by the Center for Molecular Structure (CMolS) at the California State Polytechnic University Pomona campus, which is part of the California State University (CSU). This was not a single-user proposal, as are the majority of proposals, but a wide-ranging one encompassing at least five CSU campuses and several different co-PIs. The CSU campuses are traditionally undergraduate institutions which have not typically had access to synchrotron resources in the past, either because of a lack of funding or because it was not something that was ever thought of as being a possibility. Remote connection to the MX beamlines at SSRL is now giving these researchers and their undergraduate students continued access to state-of-the-art facilities, and is having a positive impact on their approach to science and research.
4.. Education and outreach {#sec4}
============================
4.1.. Remote-access workshops {#sec4.1}
-------------------------------
Scientific staff from the SMB group not only are regularly involved in one-to-one user support *via* email and telephone (before, during and after the experiment), but also facilitate remote-access workshops to train new researchers in the use of *BLU-ICE* and *WEB-ICE*, and in the practical aspects of sample mounting and cryo-cooling, synchrotron data collection, and data processing. Several of these remote-access workshops have been held locally at SSRL, and scientists from the group have also traveled both nationally and internationally to hold remote-host workshops (Table 2[▶](#table2){ref-type="table"}).
The SSRL local workshops started in June 2006. They are usually scheduled at the start of the user run, or more often, depending on demand. Occasionally, these workshops also take place in conjunction with the Annual SSRL Users' Meeting (see Table 2[▶](#table2){ref-type="table"}). A typical workshop lasts half a day and includes a thorough introduction to the experimental facilities for MX users, including hands-on tutorials on the optimal use of the SAM robot tools, data collection with *BLU-ICE*, analysis and strategy calculations with *WEB-ICE*, and data processing with the available locally installed software packages.
The remote-host locations have included the Hauptmann--Woodward Medical Institute (HWI) in Buffalo, New York, USA (August 2006), the University of Melbourne, Australia (February 2007), the University of Pittsburgh, Pennsylvania, USA (October 2008), and the California Institute of Technology (CalTech), Pasadena, USA (June 2009). During the University of Melbourne workshop, one of the participants screened crystals that had previously been shipped to SSRL, identified the best quality crystal, collected a MAD data set and solved a novel protein structure (Schmidberger *et al.*, 2008[@bb24]), completely remotely, fully utilizing the computational resources made available to researchers at SSRL. SSRL remote access has also been incorporated into two workshops sponsored by the Center for Workshops in Chemical Sciences (<http://chemistry.gsu.edu/CWCS>) at CMolS, which were aimed at faculty from predominantly undergraduate institutions. Additional workshops at which SSRL staff have presented the remote-access tools and capabilities are listed in Table 2[▶](#table2){ref-type="table"}.
4.2.. Remote-access demonstrations, seminars and lectures {#sec4.2}
-----------------------------------------------------------
Another important method of disseminating information regarding the SSRL remote-access tools to the user community is through seminars and live remote-access demonstrations at conferences and meetings (Table 2[▶](#table2){ref-type="table"}). This turns out to be a perfect test of the capabilities of the NX client system, because generally at conference locations the wireless internet access can be somewhat intermittent and with variable speed or bandwidth, particularly as conference participants continually connect and disconnect to the system. Since the NX client system is designed to run on only 20 kbps of network bandwidth, good performance is generally maintained in the seminar locations, even on a busy wireless network. The use of a remote-access connection to either an SSRL MX beamline or a simulated beamline, when combined with conference lectures or seminars, workshop presentations, or in a formal university teaching environment, is a powerful pedagogical tool. We strongly encourage and support such use of the SSRL systems by the scientific community.
5.. Conclusions {#sec5}
=================
The SAM system has been used to screen a total of over 300 000 crystals for diffraction quality in the past seven years, and has most certainly proved its worth. When coupled with the remote-access capabilities that have been available to scientific user groups (general users) for the past five years, this system has led to the MX beamlines at SSRL becoming a true high-throughput facility. The efficiency of the research groups who use remote access has increased remarkably, which has in turn given synchrotron access to more user groups than ever before and resulted in a surge in the number of user starts at SSRL. Researchers are now easily able to screen all crystals being grown in the laboratory, in order to choose the best possible crystals for data collection, whereas before they may have limited themselves to the crystals that simply appeared to be the best, or else spent innumerable hours on a home source screening crystals. It has become increasing clear that many user groups are forgoing in-house screening, and simply cryo-cooling as many crystals as they can fit into a cassette or Uni-pucks and letting the robust efficient SAM system do the work for them. This is exactly the vision the developers of the Stanford auto-mounter had in mind for the system: to provide a true high-throughput platform for the screening of large numbers of protein crystals.
The ways in which remote access to the SSRL beamlines can facilitate training and collaboration have most certainly not gone unnoticed by the scientific community. All research groups who collect their data remotely use the available tools provided by SSRL to train and educate their laboratory members in the most effective ways to collect the best possible diffraction data. Approximately 60% of researchers with active proposals and current beam time have at some point had collaborators participate in remote-access data collection, where they take either a passive or an active role, and in some cases have even used the time to train or educate members of their own laboratory. The way in which remote access to SSRL beamlines serves to bring collaborators together is one of the most fundamental examples of what has been described as a 'cultural community', as noted by the Director of the NSF report *Cyberinfrastructure Vision for 21st Century Discovery* (NSF Cyberinfrastructure Council, 2007[@bb13]). This idea is something that we at SSRL will continue to foster and promote. At SSRL we are dedicated to making the remote-access experience as easy, efficient and instructive as possible, and making a synchrotron beamline accessible to anyone in the scientific community who has a need for a high-intensity X-ray beam and expects high-quality diffraction data.
The authors acknowledge the entire SAM and remote-access development teams, which include members of the Joint Center for Structural Genomics and the SSRL Structural Molecular Biology group. Special thanks are extended to Lisa Dunn for help with analysis of the user statistics. Operations funding for the Stanford Synchrotron Radiation Lightsource is provided by the US Department of Energy Office of Basic Energy Sciences. The SSRL Structural Molecular Biology Program is supported by the Biomedical Technology Program of the National Center for Research Resources of the US National Institutes of Health, by the US Department of Energy Office of Biological and Environmental Research, and by the National Institute of General Medical Sciences of the US National Institutes of Health. We also thank Katherine Kantardjieff at CMolS, Eddie Snell at HWI, Peter Turner at the University of Sydney, Guillermo Calero and JoAnne Yeh at the University of Pittsburgh, and Doug Rees at CalTech for organizing and facilitating remote-access workshops.
{#fig1}
{#fig2}
![(*a*) Screen capture of a typical remote-access NX session showing multiple windows open, including *BLU-ICE* in the top left background, the *MOSFLM* graphical user interface on the bottom right, *COOT* (Emsley & Cowtan, 2004[@bb5]) at the top right and a *WEB-ICE* session in the left foreground. (*b*) Screen capture of the *Screening* tab from the *BLU-ICE* software. The spreadsheet at the top left has been loaded by the experimenter, and during initial screening the Crystal Analysis server updates the table with results, as shown.](j-43-01261-fig3){#fig3}
{#fig4}
###### Cost comparison between a visit to SSRL and remote-access data collection
Costs are in US dollars.
US domestic[†](#tfn1){ref-type="table-fn"} International[‡](#tfn2){ref-type="table-fn"} Remote access
------------------------------------------------ -------------------------------------------- ---------------------------------------------- ------------------------------------------------------------------------
Airfares 432.90 1210.00 0
Sample shipping 0 0 200[§](#tfn3){ref-type="table-fn"}/1000[¶](#tfn4){ref-type="table-fn"}
Meals 191.25 191.25 0
Accommodation 195.00 195.00 0
Taxes 19.50 19.50 0
Rental car 148.00 148.00 0
Parking 24.00 0 0
Communications[††](#tfn5){ref-type="table-fn"} 0 200.00 20/200
Total per person 1010.65 1763.75 0
Total (3 people) 2735.95[‡‡](#tfn6){ref-type="table-fn"} 5195.25[‡‡](#tfn6){ref-type="table-fn"} 220/1200
Three-day data-collection trip from Huntsville, Alabama, USA.
Three-day data-collection trip from Auckland, New Zealand.
US domestic Dewar shipping by FedEx from Huntsville.
International Dewar shipping by FedEx from New Zealand.
Includes telephone calls, internet and ftp data backup.
Total includes three times the airfare, meals, accommodation and taxes only.
###### A selection of the many remote-access workshops, seminars, lectures and demonstrations facilitated or presented by SSRL scientific staff
Type Meeting/workshop Location Date Notes
----------------------- ---------------------------------------- ------------------------------------------------------- ---------------- --------------------------------------------------------------------------------------
Workshop SSRL Menlo Park, California, USA October 2004 In conjunction with the Annual SSRL Users' Meeting
Workshop SSRL Menlo Park, California, USA October 2005 In conjunction with the Annual SSRL Users' Meeting
Workshop Canadian eScience Workshop Saskatoon, Saskatchewan, Canada November 2005
Lecture/demonstration MBC 1 Fullerton, California, USA June 2005 Sponsored by the Center for Workshops in Chemical Sciences
Seminar ACA Annual Meeting Honolulu, Hawaii, USA July 2006
Workshop HWI Buffalo, New York, USA August 2006
Seminar NoBUGS 2006 Berkeley, California, USA October 2006
Workshop Joint SSRL/ALS Workshop Menlo Park, California, and Berkeley, California, USA October 2006 Uni-Puck and *WEB-ICE*
Workshop MacCHESS, Cornell Ithaca, New York, USA December 2006 Led from SSRL with participants at CHESS in a conference room
Workshop University of Melbourne Melbourne, Australia February 2007
Demonstration Rotorua Proteins Meeting Rotorua, New Zealand February 2007
Demonstration BSR9 Manchester, UK August 2007 Biology and Synchrotron Radiation Meeting
Seminar RAMC San Diego, California, USA September 2007 Recent Advances in Macromolecular Crystallization
Seminar Laboratory Automation Palm Springs, California, USA January 2008
Seminar CLS Saskatoon, Saskatchewan, Canada June 2008 In conjunction with the Canadian Light Source Annual Users' Meeting
Seminar Protein Crystallography Europe Amsterdam, The Netherlands June 2008
Lecture/demonstration MBC 2 Fullerton, California, USA June 2008 Sponsored by the Center for Workshops in Chemical Sciences
Lecture ACA Summer Course Indiana, Pennsylvania, USA July 2008
Seminar GRC, Bates College Lewiston, Maine, USA July 2008 Diffraction Methods in Structural Biology, Gordon Research Conference
Workshop CEI2008 Arlington, Virginia, USA July 2008 Cyber-Enabled Instruments 2008 Strategic Planning Workshop
Workshop SSRL Menlo Park, California, USA October 2008 In conjunction with the Annual SSRL Users' Meeting
Workshop University of Pittsburgh Pittsburgh, Pennsylvania, USA October 2008 In conjunction with the Pittsburgh Diffraction Society Annual Meeting
Lecture/demonstration AstraZeneca/MedImmune Research Meeting Gaithersburg, Maryland, USA February 2009
Workshop NIGMS Workshop Bethesda, Maryland, USA March 2009 Enabling Technologies for Structural Biology
Lecture/demonstration ACA Summer Course Indiana, Pennsylvania, USA June 2009
Workshop CalTech Pasadena, California, USA June 2009
Seminar SRI Melbourne, Australia September 2009 10th International Conference on Synchrotron Radiation Instrumentation
Lecture CSHL Course Cold Spring Harbor, New York, USA October 2009 Cold Spring Harbor Laboratory, X-ray Methods in Structural Biology Course
Workshop SSRL Menlo Park, California, USA October 2009 In conjunction with the Annual SSRL Users' Meeting
Seminar BSR10 Melbourne, Australia February 2010 Biology and Synchrotron Radiation Meeting
Workshop NSLS Brookhaven, New York, USA May 2010 Frontiers in Automated Crystal Handling, in conjunction with the NSLS Users' Meeting
Lecture/demonstration ACA Summer Course Indiana, Pennsylvania, USA June 2010
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-cells-09-00148}
===============
Prune dwarf virus (PDV) is a viral pathogen distributed worldwide \[[@B1-cells-09-00148],[@B2-cells-09-00148]\]. PDV, a member of the *Bromoviridae* family and the genus *Ilarvirus*, infects fruit trees, generates enormous decrease in fruit yield, and reduces the effectiveness of vegetative reproduction of various species of orchard trees \[[@B3-cells-09-00148]\]. For example, a decrease in fruit yield has been estimated at the level of 80--90% for sweet cherry \[[@B3-cells-09-00148],[@B4-cells-09-00148],[@B5-cells-09-00148]\]. Vegetative reproduction is crucial for orchard trees. A decrease during vegetative reproduction can reach up to 90% in nurseries \[[@B3-cells-09-00148],[@B6-cells-09-00148]\]. PDV can also be transmitted via seeds and pollen of infected plants \[[@B3-cells-09-00148]\].
The PDV genome has a complex structure ([Scheme S1](#app1-cells-09-00148){ref-type="app"}) with three positive ssRNA segments, named RNA1, RNA2, and RNA3. RNA1 encodes P1 protein (replicase), which is the first element of the viral replication complex \[[@B1-cells-09-00148]\]. RNA2 encodes P2 protein, which serves as the RNA-dependent RNA polymerase (RdRp), a component of the replication complex \[[@B1-cells-09-00148],[@B2-cells-09-00148]\]. The third segment, RNA3, encodes two proteins, coat protein (CP) and movement protein (MP). CP supports the "genome activation" process, which is crucial for replication of the viral RNA genome and also creates capsid for viral particles \[[@B7-cells-09-00148]\]. MP enables generation of tubular structures during cell-to-cell transport and supports translocation of viral particles via plasmodesmata \[[@B8-cells-09-00148]\].
In an infected host plant, PDV is transported both cell-to-cell and systemically throughout the whole plant. During local spreading (cell-to-cell movement), the virus is transported as viral particles via MP-generated tubular structures through plasmodesmata, which modify their size exclusion limit (SEL) \[[@B8-cells-09-00148]\]. In the case of systemic movement, our new data from infected tobacco and cucumber reveal that the transport is mainly associated with phloem and xylem cells \[[@B8-cells-09-00148],[@B9-cells-09-00148],[@B10-cells-09-00148]\].
There is very little knowledge about resistance genes or resistant reactions of plant hosts during PDV infection. Based upon Fulton's research \[[@B11-cells-09-00148],[@B12-cells-09-00148]\] and the Plant Viruses Online Database \[[@B13-cells-09-00148]\], we know that many species are incompatible with PDV infection and replication. These test plants are then potential resources for resistant reaction/genes for plant immunity. One of the most promising test plants is quinoa (*Chenopodium quinoa*). This plant is compatible with many members of the *Bromoviridae* family, for example, *Prunus necrotic ringspot virus* (PNRSV) of genus *Ilarvirus* \[[@B13-cells-09-00148]\]. Since PNRSV and PDV are similar viruses \[[@B2-cells-09-00148]\], the potential incompatibility of PDV may be due to a resistance reaction. Therefore, the aim of this study was to determine the reactions of quinoa to PDV inoculation in the context of ultrastructural changes that are associated with resistance. By using: PDV CP-targeted double antibody sandwich enzyme-linked immunosorbent assay (DAS-ELISA), immunofluorescence localization of viral P1 protein (the main component of the viral replication complex) and real-time quantitative polymerase chain reaction (qPCR) of PDV movement protein (MP) gene and localization of MP, the presence of PDV in inoculated leaves was confirmed. Moreover, by using microscopic methods, we demonstrated that changes in both types of vascular tissues are crucial for abolition of PDV systemic transport/spreading. Alterations in cell organelles were observed only in inoculated leaves, whereas DAS-ELISA, qPCR, and immunofluorescence excluded the presence of PDV in the stem and upper leaves.
2. Materials and Methods {#sec2-cells-09-00148}
========================
2.1. Virus Inoculation and DAS-ELISA {#sec2dot1-cells-09-00148}
------------------------------------
Before electron transmission and light microcopy, DAS-ELISA was used on quinoa plants (*C. quinoa*). According to the Plant Virus Online database \[[@B13-cells-09-00148]\], this species is not susceptible to a broad spectrum of different isolates/strains of PDV. Fifty quinoa plants were mechanically inoculated as was presented in \[[@B9-cells-09-00148]\]. Mock- and PDV-inoculated leaves along with leaves and stems above were checked for the presence of PDV by using DAS-ELISA in 3 repeats for each time point, with primary antibodies against the PDV CP (Bioreba, Reinach, Switzerland), followed by purified anti-rabbit antibodies conjugated with alkaline phosphatase (Bioreba, Reinach, Switzerland) \[[@B14-cells-09-00148]\]. Each repeat was a new ELISA plate with samples. For each test, we took samples from 15 mock-inoculated and 15 PDV-inoculated plants. Readings of OD~405nm~ values were performed after 60 min. All DAS-ELISA tests were performed using the same reagents. The measurements were performed at 3 time intervals, 7, 14, and 20 dpi, for mock- and PDV-inoculated quinoa leaves and for systemic leaves. After 20 dpi, only systemic leaves were tested. The mean OD~405nm~ values from 3 DAS-ELISA tests were statistically assessed with one-factor analysis of variance (ANOVA) and evaluated at the *p* \< 0.05 level of significance using Tukey's post hoc honestly significant difference (HSD) test in Statistica software (version 13.0; StataSoft and TIBCO Software Inc., Palo Alto, CA, USA). For more precise assessment of DAS-ELISA results, we computed corrected mean OD~405nm~. To do this, we subtracted from the mean OD~405nm~ of sample PDV-inoculated plants a sum of mean OD~405nm~ of buffer and appropriate mock-inoculated plants. These data were also statistically assessed as above. As suggested by Paduch-Cichal et al. \[[@B15-cells-09-00148]\] and Paduch-Cichal and Sala-Rejczak \[[@B16-cells-09-00148]\], absorbance above 0.2 confirmed the presence of the virus. Significance threshold values of DAS-ELISA were determined according to \[[@B6-cells-09-00148],[@B15-cells-09-00148],[@B16-cells-09-00148]\]. Absorbance from mock-inoculated plants was much lower than this threshold.
2.2. Isolation of RNA and Genomic DNA (gDNA) and qPCR Analysis of Expression of MP Gene of PDV in Quinoa Plants {#sec2dot2-cells-09-00148}
---------------------------------------------------------------------------------------------------------------
Parallel DAS-ELISA molecular analysis of *MP* gene expression based on qPCR was performed. This analysis was conducted with the same time intervals on a group of mock- and virus-inoculated plants as the DAS-ELISA test. Stem and leaf samples (the weight of each sample was 0.05 g) from mock- and virus-inoculated plants at the inoculation point and above (at 7, 14, and 20 dpi) were collected. From each plant we collected 6 leaf and 6 stem samples at each time point after inoculation. We repeated the whole experiment 3 times. During the experiment, we gathered samples from 90 mock- and 90 virus-inoculated plants. RNA from these samples was isolated by use of GeneMATRIX Universal RNA Purification Kit (EURx Sp. z o.o., Gdansk, Poland) according to the manufacturer's protocol. From 6 selected samples, RNA and gDNA were isolated by use of GeneMATRIX Universal DNA/RNA/Protein Purification Kit (EURx Sp. z o.o., Gdansk, Poland) according to the manufacturer's protocol. gDNA obtained in this way was used for preparation of calibration curves. In the next step, calibration curves were used to determine the efficiency of qPCR reaction for low-expression transcripts. Isolated RNA was purified from gDNA by on-column DNase I digestion (EURx Sp. z o.o., Gdansk, Poland). After, reaction samples were purified by use of GeneMATRIX Universal RNA Purification Kit (EURx Sp. z o.o., Gdansk, Poland) according to the manufacturer's protocol. RNA concentration after purification was measured by NanoDrop 2000 (Thermo Fisher Scientific, Waltham, MA, USA). Quality of extracted RNA was checked with the use of electrophoresis in 1% agarose gel in denaturation conditions. Additional lack of RNA contamination was checked by RT-PCR reaction with *actin* (reference gene) primers on matrix of obtained RNA. This reaction showed that RNA did not have gDNA contamination. After contamination analysis, cDNA was obtained by use of NG dART RT kit (EURx Sp. z o.o., Gdansk, Poland) according to the manufacturer's protocol. Reverse transcription reaction was performed in a volume of 10 µL (for reaction we used 700 ng of RNA formulation).
Reaction of qPCR was performed in a Bio-Rad CFX96 Touch^TM^ (Bio-Rad Poland Sp. z.o.o., Warsaw, Poland), using SsoAdvanced^TM^ Universal SYBR^®^ Green Supermix (Bio-Rad Polska Sp. z.o.o., Warsaw, Poland) with 2 already prepared 5-point calibration curves (based on cDNA and gDNA). Expression of PDV *MP* gene (gene ID: HM015770.1) in quinoa was investigated at different times after inoculation in different parts of the plant (stem and leaves). As reference gene, we used quinoa *actin* (gene ID: LOC110724665) in quinoa, as suggested in \[[@B17-cells-09-00148]\]. Primers for reference and investigated genes were designed in Primer3 software v. 0.4.0 (Primer3Plus, Free Software Foundation, Inc., Boston, MA, USA). Concentrations of primer sequences (for both genes) used for reaction and connection temperature are presented in [Table S1](#app1-cells-09-00148){ref-type="app"}. The starting solution of cDNA (used in preparation of calibration curves) was a 4× diluted mix of 12 randomly selected cDNA formulations. For calibration curves based on gDNA, it was a 10× diluted solution of gDNA. In all cases, other subsequent points in calibration curves were prepared by a series of 4× dilutions of mix. Reaction was performed in a volume of 15 µL, and 5 µL 10× diluted cDNA formulation of each analyzed gene was added. Conditions of qPCR reaction are presented in [Table S2](#app1-cells-09-00148){ref-type="app"}.
The most important parameters of RT-PCR reaction for all sequences are presented in [Supplementary Table S3](#app1-cells-09-00148){ref-type="app"}. Reaction efficiency and R^2^ factor, which define the quality of calibration curves for transcripts, remained within normal range (90--110%, \>0.98, respectively). Analysis of melting curves indicated the presence of only one PCR product in each reaction. Level of expression of PDV *MP* in context (comparison) of expression of reference gene (quinoa actin) was calculated by use of the Gene Study tool in Bio-Rad CFX Connect software v. 1.1 (Bio-Rad Polska Sp. z.o.o., Warsaw, Poland). Statistical analysis of the results, which included calculation of relative gene expression levels and the significance of differences between tested samples (ANOVA method), was performed using the Gene Study tool. Results of this analysis were normalized using one reference gene encoding quinoa actin.
2.3. Immunofluorescence Localization of PDV-P1 in Quinoa Tissues and Assessment of Quantitative Fluorescence Signal {#sec2dot3-cells-09-00148}
-------------------------------------------------------------------------------------------------------------------
Two weeks after PDV inoculation, fragments of quinoa leaf blades that were analyzed with DAS-ELISA were treated, as previously described by Kozieł et al. \[[@B8-cells-09-00148]\]. To check the distribution of PDV, P1 protein was localized by immunofluorescence according to a procedure described by Kozieł et al. \[[@B9-cells-09-00148]\] with modification regarding secondary antibody. Purified rabbit polyclonal antibody anti-P1-PDV (GeneCust, Boynes, Luxemburg) served as primary antibody \[[@B9-cells-09-00148]\], while the secondary one was anti-rabbit antibody IgG with the attached AlexaFluor**^®^**488 (Abcam, Cambridge, UK). For better contrast of plant tissues, we added 4′,6-diamidine-2′-phenylindole dihydrochloride (DAPI). Slides were imaged on a PROVIS AX70 fluorescent microscope with Olympus UP90 high-definition camera (Olympus, Warsaw, Poland) using Olympus Cell Sense Standard Software (version 1.18; Olympus, Center Valley, PA, USA). The intensity of the green fluorescent signal from regions of localization of P1 was further analyzed by use of a quantitative measuring method, corrected total cell fluorescence (CTCF) \[[@B18-cells-09-00148]\]. For CTCF, 25 selected areas of every sample were analyzed. To measure the fluorescent signal levels, we used ImageJ (version 1.51k; National Institutes of Health., Bethesda, MD, USA). Measurements of the green immunofluorescence signal gained from ImageJ were calculated with CTCF at 20× magnification with 1.00 zoom factor by using the formula previously presented by Otulak-Kozieł et al. \[[@B18-cells-09-00148]\]:
Estimated CTCF values were then analyzed statistically at selected time intervals for all plants by using one-factor analysis of variance (ANOVA) as presented by \[[@B18-cells-09-00148]\].
2.4. Preparation of Leaf Material for Light and Transmission Microscopy {#sec2dot4-cells-09-00148}
-----------------------------------------------------------------------
Fragments of quinoa leaf tissue (at 7, 14, and 20 dpi) were cut out, fixed, and embedded with EPOXY resin exactly as described by Kozieł et al. \[[@B8-cells-09-00148]\], followed by slicing into thin sections for light microscope or ultrathin sections (100 nm) for TEM, and mounted on slides or copper grids, respectively \[[@B19-cells-09-00148]\]. Slides were stained with crystal violet, while ultrathin sections were contrasted with uranyl acetate/lead citrate (Sigma-Aldrich, St. Louis, MO, USA) for TEM.
2.5. Preparation of Leaf Material and Immunogold Localization of PDV Movement Protein (PDV MP) {#sec2dot5-cells-09-00148}
----------------------------------------------------------------------------------------------
Fragments of quinoa leaf tissue was prepared according to [Section 2.4](#sec2dot4-cells-09-00148){ref-type="sec"}. Immunogold localization was performed exactly according to the procedure presented by \[[@B8-cells-09-00148]\] with use of primary antibodies targeted PDV MP used previously in \[[@B8-cells-09-00148]\] and secondary anti mouse antibodies associated with 18 nm gold particles (JaksonImmunoResearch Europe, Cambridgeshire, UK).
2.6. Localization and Quantification of H~2~O~2~ by Corrected Total Electron Density (CTED) Method {#sec2dot6-cells-09-00148}
--------------------------------------------------------------------------------------------------
Hydrogen peroxide was detected at selected time points by the method of Bestwick et al. \[[@B20-cells-09-00148]\] as modified by Otulak and Garbaczewska \[[@B21-cells-09-00148]\]. Briefly, quinoa leaf tissue was preincubated in 50 mM (*w*/*v*) 3-morpholinopropane-1-sulfonic acid (MOPS) buffer (pH 7.2) containing 5 mM CeCl~3~, washed for an hour with the same buffer, and fixed in 2% (*w*/*v*) paraformaldehyde/2% (*v*/*v*) glutaraldehyde in 0.05 M cacodylate buffer (pH 7.2--7.4) \[[@B21-cells-09-00148]\] for 2 h at room temperature. Samples were contrasted and fixed in 2% (*w*/*v*) OsO~4~ in cacodylate buffer and also dehydrated in a series of increasingly strong ethanol--water solutions as was presented by \[[@B21-cells-09-00148]\]. The material was gradually saturated with Epon 812 (Fluka) resin and polymerized for 24 h at 60 °C. Observations were made as previously described in \[[@B8-cells-09-00148]\]. To quantify the level of H~2~O~2~, negative photos were analyzed for distribution of electron-dense cerium (IV) perhydroxide precipitates by using the CTED method \[[@B22-cells-09-00148]\], with a general formula as follows:
CTED = Integrated Density − (Area of Selected Cell Region × Mean Density of Background Readings).
In the next step, CTED values were analyzed for statistical significance by using ANOVA). The mean CTED values were evaluated at the *p* \< 0.05 level of significance using Tukey's post hoc HSD test in Statistica software (version 13.0; StataSoft and TIBCO Software Inc., Palo Alto, CA, USA).
2.7. Measurment of H~2~O~2~ Concentration {#sec2dot7-cells-09-00148}
-----------------------------------------
The hydrogen peroxide (H~2~O~2~) levels in mock- and PDV-inoculated quinoa leaves and leaves above the PDV inoculation point were determined according to Velikova et al. \[[@B23-cells-09-00148]\] at 7 and 14 dpi. Frozen leaves (0.5 g) were extracted in an ice bath with 5 mL 0.1% (*w*/*v*) tri-chloro-acetic acid (TCA). The homogenate was centrifuged at 15,000× *g* for 15 min. Then, an aliquot (0.5 mL) of the supernatant was added to a mixture of 0.5 mL potassium phosphate buffer (10 mM, pH 7.0) and 1 mL 1 M KI. After 20 min of incubation in darkness at room temperature, the absorbance of the samples was spectrophotometrically determined at 390 nm \[[@B23-cells-09-00148]\] by use of a SmartSpec™ 3000 spectrophotometer (Bio-Rad, Philadelphia, PA, USA). The concentration of H~2~O~2~ was calculated from a standard curve.
3. Results {#sec3-cells-09-00148}
==========
3.1. Symptoms of PDV Infection and Immunofluorescent Localization of P1 Protein in a Context of PDV CP Distribution and Relative Expression of MP {#sec3dot1-cells-09-00148}
-------------------------------------------------------------------------------------------------------------------------------------------------
PDV first caused symptoms approximately 7 days post-inoculation (dpi) on inoculated leaves, and chlorotic lesions developed near the edges of leaf blades ([Figure 1](#cells-09-00148-f001){ref-type="fig"}A), which did not occur on mock-inoculated plants ([Figure 1](#cells-09-00148-f001){ref-type="fig"}B). After 14 dpi, the inoculated leaves became necrotized ([Figure 1](#cells-09-00148-f001){ref-type="fig"}C,D), but mock-inoculated leaves ([Figure 1](#cells-09-00148-f001){ref-type="fig"}E) and upper leaves of PDV-inoculated plants were unchanged and had no symptoms of infection ([Figure 1](#cells-09-00148-f001){ref-type="fig"}F). The inoculated necrotized leaves got detached from the plant about 20 dpi ([Figure 1](#cells-09-00148-f001){ref-type="fig"}G). To check whether the observed symptoms were caused by ongoing PDV infection, P1 protein was identified/localized by immunofluorescence; the presence of P1 was not detected in mock-inoculated leaves after 7 dpi ([Figure 1](#cells-09-00148-f001){ref-type="fig"}H). Green fluorescent signal of PDV P1 protein was spotted starting at 7 dpi inside both palisade and spongy mesophyll, but not in phloem and xylem tissue ([Figure 1](#cells-09-00148-f001){ref-type="fig"}I). At 14 dpi, the P1 fluorescent signal was less intense than at 7 dpi. P1 protein deposition was detected not only in both types of mesophyll but also in the phloem ([Figure 1](#cells-09-00148-f001){ref-type="fig"}J). Moreover, numerous palisade and spongy mesophyll cells were deformed ([Figure 1](#cells-09-00148-f001){ref-type="fig"}J). Such alterations were clearly visible as cell wall invaginations ([Figure 1](#cells-09-00148-f001){ref-type="fig"}J). In contrast to PDV-inoculated leaves at 14 dpi, P1 was not detectable in upper leaves even at 20 dpi ([Figure 1](#cells-09-00148-f001){ref-type="fig"}K). Such deposition patterns suggest that PDV replication inside quinoa plants is associated with and limited to inoculated leaves.
Moreover, quantitative measurement of the fluorescence signal based on corrected total cell fluorescence (CTCF) confirmed a statistically significant decrease of P1 protein signal (approximately 62%) between 7 and 14 dpi and the absence of P1 in upper leaves of PDV-inoculated plants ([Figure 2](#cells-09-00148-f002){ref-type="fig"}).
To quantify relative virus concentration, DAS-ELISA was performed by using anti-PDV CP antibodies. This test confirmed the presence of PDV in the inoculated leaves at 7 and 14 dpi ([Table S4](#app1-cells-09-00148){ref-type="app"}, [Figure S1A,B](#app1-cells-09-00148){ref-type="app"}), but PDV was not detected in upper leaves or stems at 14 and 20 dpi ([Figure S1](#app1-cells-09-00148){ref-type="app"}). Moreover, statistical analysis of the optical density (OD~405~) and corrected optical density (corrected OD~405~) data from DAS-ELISA confirmed a significantly decreased relative concentration of CP/virus by 55.36% (mean OD~405~) or 56.74% (corrected OD~405~) between 7 and 14 dpi time points ([Figure S1A,B](#app1-cells-09-00148){ref-type="app"}).
Furthermore, real-time quantitative polymerase chain reaction (qPCR) expression analysis of PDV movement protein was performed. Statistical analysis of qPCR results revealed that PDV *MP* gene was expressed only in virus-inoculated quinoa leaves at 7 and 14 dpi ([Figure 3](#cells-09-00148-f003){ref-type="fig"}). Normalized relative expression level of *MP* gene statistically significantly decreased (approximately 86%) in inoculated leaves between 7 and 14 dpi. Moreover, in stem and leaves above the inoculation point, no expression was observed at 7, 14, and 20 dpi, as well as in all parts of mock-inoculated plants ([Figure 3](#cells-09-00148-f003){ref-type="fig"}).
3.2. Anatomical and Ultrastructural Changes in PDV-Infected Quinoa Plants and Immunogold Labbleing of PDV MP {#sec3dot2-cells-09-00148}
------------------------------------------------------------------------------------------------------------
Mock-inoculated quinoa plants (at 7 dpi) displayed unaffected cells in all tissues of the leaf ([Figure 4](#cells-09-00148-f004){ref-type="fig"}A). Anatomical changes were observed only in the inoculated leaves ([Figure 4](#cells-09-00148-f004){ref-type="fig"}B). In this case, high levels of deformation were observed in spongy mesophyll and in parenchyma cells (14 dpi) ([Figure 4](#cells-09-00148-f004){ref-type="fig"}B). At 14 dpi, we also spotted invagination of cell wall into this tissue ([Figure 4](#cells-09-00148-f004){ref-type="fig"}B,C) as well as necrosis in both phloem and xylem parenchyma inside major vascular bundles ([Figure 4](#cells-09-00148-f004){ref-type="fig"}C). More severe changes were observed in PDV-inoculated leaves at some distance from major vascular bundles at 14 dpi in the form of numerous deformations and cell-wall invaginations, not only in spongy mesophyll but also in epidermis and palisade mesophyll ([Figure 4](#cells-09-00148-f004){ref-type="fig"}D). In PDV-inoculated leaves (20 dpi), the level of deformations and alterations related to local necrotization was increased in both xylem and spongy mesophyll ([Figure 4](#cells-09-00148-f004){ref-type="fig"}E). Such anatomical alterations and cell deformations were not observed at 14 dpi in upper non-inoculated leaves ([Figure 4](#cells-09-00148-f004){ref-type="fig"}F).
For deeper analysis of the course of PDV infection at the ultrastructural level, the tissues were examined by transmission electron microscopy (TEM). Changes were first observed in chloroplasts in PDV-inoculated leaves at 7 dpi ([Figure 5](#cells-09-00148-f005){ref-type="fig"}A). Many chloroplasts in palisade mesophyll cells accumulated electron-dense substances (lipids). Later, alterations in chloroplasts became more intense, with spotted invaginations of chloroplast envelope at 14 dpi ([Figure 5](#cells-09-00148-f005){ref-type="fig"}B); likewise, invaginations of cell wall in spongy mesophyll cells were spotted. At 20 dpi, many disintegrated chloroplasts and deep invaginations of cell wall could be found ([Figure 5](#cells-09-00148-f005){ref-type="fig"}C). In comparison, in mock-inoculated plants, changes in chloroplasts did not occur ([Figure 5](#cells-09-00148-f005){ref-type="fig"}D,E).
During infection, especially at 7 dpi, we observed the presence of PDV viral particles (VPs) in parenchyma cells ([Figure 6](#cells-09-00148-f006){ref-type="fig"}A,B) alongside vesicle pockets ([Figure 6](#cells-09-00148-f006){ref-type="fig"}A) and enlarged endoplasmic reticulum (ER) cisterns ([Figure 6](#cells-09-00148-f006){ref-type="fig"}B). PDV VPs were also often observed near plasmodesmata, which had an extended size exclusion limit (SEL) at 14 dpi, which thus enabled PDV cell-to-cell transport ([Figure 6](#cells-09-00148-f006){ref-type="fig"}C,D). More importantly, PDV VPs were found near potential movement protein-induced tubular structures (MTs) that passed through to the cell wall ([Figure 6](#cells-09-00148-f006){ref-type="fig"}E), which was not observed in mock-inoculated parenchyma cells ([Figure 6](#cells-09-00148-f006){ref-type="fig"}F). To check potential association of plasmodesmata and MT changes, immunogold labeling of PDV MP was performed. During infection we observed presence of PDV MP epitopes inside plasmodesmata ([Figure 6](#cells-09-00148-f006){ref-type="fig"}G,H) and also inside potential MTs in parenchyma cells ([Figure 6](#cells-09-00148-f006){ref-type="fig"}I).
In PDV-inoculated leaves at 14 dpi, viral particles were present in phloem parenchyma and companion cells ([Figure 7](#cells-09-00148-f007){ref-type="fig"}A,B). In phloem parenchyma, the nearby sieve tube cell wall was often dented and had tubular structures accompanied by PDV VPs ([Figure 7](#cells-09-00148-f007){ref-type="fig"}A). In companion cells, the observed changes at 14 dpi concerned high accumulation of viral particles and formation of spherules in vacuoles ([Figure 7](#cells-09-00148-f007){ref-type="fig"}B). Moreover, VPs were mostly present near the nucleus ([Figure 7](#cells-09-00148-f007){ref-type="fig"}C) in companion cells, followed by enlarged endoplasmic reticulum (ER) cisterns and Golgi apparatus ([Figure 7](#cells-09-00148-f007){ref-type="fig"}D). Spherules and VPs were not noticed in cells from mock-inoculated tissues ([Figure 7](#cells-09-00148-f007){ref-type="fig"}E).
At 14 dpi, necrotic changes occurred in phloem companion cells ([Figure 8](#cells-09-00148-f008){ref-type="fig"}A) concomitant with the presence of viral particles ([Figure 8](#cells-09-00148-f008){ref-type="fig"}B). In companion cells of mock-inoculated plants, the presence of PDV was not noticed ([Figure 8](#cells-09-00148-f008){ref-type="fig"}C). Moreover, changes were noticed in xylem tissue, whereas viral particles were detected only in xylem parenchyma ([Figure 9](#cells-09-00148-f009){ref-type="fig"}A,B) with vesicle pockets or near tubular structures ([Figure 9](#cells-09-00148-f009){ref-type="fig"}B).
Furthermore, reactions were also observed in xylem tracheary elements, accumulating electron-dense material (probably phenols), starting from 7 dpi ([Figure 9](#cells-09-00148-f009){ref-type="fig"}C). Such electron-dense material could be observed first along internal parts of cell walls ([Figure 9](#cells-09-00148-f009){ref-type="fig"}D). Finally, at 20 dpi this electron-dense substance filled the xylem tracheary elements completely ([Figure 9](#cells-09-00148-f009){ref-type="fig"}E,F). Electron-dense material did not accumulate in mock-inoculated xylem ([Figure 9](#cells-09-00148-f009){ref-type="fig"}G).
To define the source of ultrastructural changes, the tissue was tested for hydrogen peroxide (H~2~O~2~) because reactive oxygen species (ROS) may induce the above alterations during transduction of the infection signal. Detection of H~2~O~2~ in quinoa cells was accomplished by using CeCl~3~, which reacts with H~2~O~2~, generating an electron-dense cerium (IV) perhydroxide precipitate. For mock-inoculated plants, deposition of such precipitate was observed rarely and in small amounts in vacuoles ([Figure 10](#cells-09-00148-f010){ref-type="fig"}A). However, precipitates with H~2~O~2~ were spotted as layers on the surface of altered chloroplasts in the mesophyll at 7 dpi ([Figure 10](#cells-09-00148-f010){ref-type="fig"}B). At 14 dpi, the presence of H~2~O~2~ was confirmed near disintegrated chloroplasts ([Figure 10](#cells-09-00148-f010){ref-type="fig"}C) and at high levels in mesophyll vacuoles ([Figure 10](#cells-09-00148-f010){ref-type="fig"}D). Moreover, these depositions were also related to changes in cells of both types in vascular tissues. High accumulation of H~2~O~2~ was first observed in companion cells at 14 dpi as an electron-dense layer along the tonoplast near viral particles in cytoplasm ([Figure 10](#cells-09-00148-f010){ref-type="fig"}E,F). Later on, H~2~O~2~ was not only in tonoplast but also in necrotically changed cytoplasm ([Figure 10](#cells-09-00148-f010){ref-type="fig"}G) or in disintegrated cytoplasm of companion cells at 20 dpi ([Figure 10](#cells-09-00148-f010){ref-type="fig"}H). Furthermore, in xylem we observed H~2~O~2~ in tracheary element at 14 dpi ([Figure 10](#cells-09-00148-f010){ref-type="fig"}I), and the level of H~2~O~2~ increased with time at 20 dpi ([Figure 10](#cells-09-00148-f010){ref-type="fig"}J).
The distribution of H~2~O~2~ was quantified statistically based on the corrected total electron density (CTED) tool ([Figure S2](#app1-cells-09-00148){ref-type="app"}) by measuring the density of cerium (IV) perhydroxide precipitate. In mock-inoculated plants from 7 to 14 dpi, CTED slightly increased but not to a statistically significant level ([Figure S2](#app1-cells-09-00148){ref-type="app"}), whereas the level of H~2~O~2~ was higher in PDV-inoculated leaves than in mock-inoculated plants, increasing with time between 7 and 14 dpi ([Figure S2](#app1-cells-09-00148){ref-type="app"}) to over 92%.
CTED results were also confirmed by spectrophotometric analysis of H~2~O~2~ concentration in mock- and PDV-inoculated leaves. These data clearly indicated that PDV inoculation significantly increased H~2~O~2~ levels at 7 and 14 dpi (from 12 to 19 µmol H~2~O~2~) compared to mock-inoculated quinoa plants ([Figure 11](#cells-09-00148-f011){ref-type="fig"}. Moreover, H~2~O~2~ concentration was significantly higher in PDV-inoculated leaves at 7 and 14 dpi ([Figure 11](#cells-09-00148-f011){ref-type="fig"}).
4. Discussion {#sec4-cells-09-00148}
=============
4.1. Symptoms of PDV Infection and Immunofluorescent Localization of P1 Protein Within a Context of PDV CP Distribution {#sec4dot1-cells-09-00148}
-----------------------------------------------------------------------------------------------------------------------
There are two major ways to detect/identify PDV infection. First, DAS-ELISA is generally useful for natural hosts \[[@B24-cells-09-00148],[@B25-cells-09-00148],[@B26-cells-09-00148],[@B27-cells-09-00148],[@B28-cells-09-00148],[@B29-cells-09-00148]\]. Second, biological inoculation of susceptible test plants can be done, including *Cucumis sativus* (especially cv. Wisconsin and cv. Polan), *Cucurbita maxima* (for example, cv. Buttercup), and *Nicotiana tabacum* cv. Samsun \[[@B3-cells-09-00148],[@B11-cells-09-00148],[@B12-cells-09-00148],[@B24-cells-09-00148]\]. Here, PDV multiplies to higher levels, easily detectable by DAS-ELISA \[[@B6-cells-09-00148]\]. In our research we have started analyzing the PDV test plant quinoa (*C. quinoa*). This species, according to the Plant Virus Online Database \[[@B13-cells-09-00148]\], is generally considered to be unsusceptible to all isolates/strains of PDV, while other *Ilarviruses* are able to infect quinoa \[[@B3-cells-09-00148]\]. Quinoa is a commonly distributed weed, but is also cultivated as an alternative to cereals \[[@B30-cells-09-00148]\]. In this work we investigate the immunity-like reaction of quinoa to PDV by using CTCF, DAS-ELISA, and immunofluorescent localization of PDV as well as anatomical and ultrastructural observations. We demonstrate that PDV generates chlorotic lesions on inoculated leaves at 7 dpi, also previously reported by Fulton \[[@B11-cells-09-00148]\] and Kozieł et al. \[[@B8-cells-09-00148]\] for cucumber cotyledons or inoculated leaves of tobacco \[[@B10-cells-09-00148]\]. However, for the latter two hosts, infection symptoms were also observed on upper leaves at 14 dpi by Fulton \[[@B11-cells-09-00148]\] and Kozieł et al. \[[@B9-cells-09-00148]\]. Moreover, Kozieł et al. \[[@B9-cells-09-00148]\] also observed deformation of leaf blades resulting from PDV infection. The fact that PDV is absent from systemic leaves suggests that it cannot move systemically to other parts of the plant. Fulton \[[@B11-cells-09-00148]\] and Waterworth and Fulton \[[@B31-cells-09-00148]\] suggested that in the case of the *Bromoviridae* family (of which PDV is a member), the absence of symptoms above the inoculation point strongly implies a lack of systemic viral transport. Nevertheless, we were also aware of asymptomatic chronic PDV infection in some natural hosts. Indeed, immunofluorescent localization of P1 protein, CTCF analysis, and detection of PDV CP by using DAS-ELISA, as well as qPCR for *MP* gene, strongly indicate the ability of PDV to replicate in quinoa tissues.
PDV CP accumulates to higher levels than other PDV proteins \[[@B1-cells-09-00148],[@B2-cells-09-00148],[@B7-cells-09-00148],[@B32-cells-09-00148]\]. CP is engaged in "genome activation" needed at stages before viral RNA replication, but also is crucial for creation of PDV viral particles \[[@B33-cells-09-00148]\]. PDV is transported in the form of viral particles \[[@B8-cells-09-00148]\], so CP is also needed for this process. Immunofluorescent localization of PDV P1 epitope showed deposition inside palisade and spongy mesophyll (7 dpi) but not in phloem and xylem tissues in virus-inoculated quinoa leaves. P1 reached phloem after 14 dpi, but the fluorescent signal was less intense than at 7 dpi. Similar patterns of P1 distribution were observed in tobacco leaves \[[@B9-cells-09-00148]\]. Moreover, in mock-inoculated and upper quinoa leaves, no P1 was localized. However, our observations in PDV-infected tobacco \[[@B9-cells-09-00148]\] revealed stronger fluorescence from P1 than in inoculated quinoa leaves, although for quinoa, changes (deformations) in palisade and spongy mesophyll cells were more severe than in tobacco \[[@B9-cells-09-00148]\]. The presence of P1 protein is an efficient marker of ongoing PDV replication \[[@B8-cells-09-00148],[@B9-cells-09-00148]\]. Similarly, the absence of P1 protein in upper leaves potentially supports our hypothesis about a blockage of systemic PDV transport. Moreover, CTCF levels of PDV also decreased in already virus-inoculated leaves during the time of infection.
The DAS-ELISA results clearly show that the relative level of PDV CP in quinoa plants significantly decreased over time. More interestingly, DAS-ELISA did not detect CP not only in leaves above inoculation but also in the stem. These results show that CP was unable to translocate through stem to/from inoculated and upper leaves. Pallas et al. \[[@B1-cells-09-00148],[@B2-cells-09-00148],[@B32-cells-09-00148]\] stated that levels of CP in *Ilarviruses* may serve as a parameter of virus quantity, building the viral capsids. If so, we believe we have demonstrated that the relative level of PDV decreases in quinoa plants.
Moreover, decreased normalized relative expression of *MP* gene in PDV-inoculated leaves and the lack of expression in organs above the inoculation point indicate that quinoa has an immunity-like reaction. Based on observations of Kozieł et al. \[[@B8-cells-09-00148]\] in PDV-infected cucumber, both CP and MP proteins are equally important in cell-to-cell transport. Therefore, changes in CP and MP expression levels indicate that PDV infection has less mobility in quinoa plants, which represents the first evidence of an immunity-like reaction against PDV in *Chenopodium quinoa* plants.
4.2. Anatomical and Ultrastructural Changes in PDV-Infected Quinoa Plants {#sec4dot2-cells-09-00148}
-------------------------------------------------------------------------
We observed high levels of deformations in spongy mesophyll and parenchyma cells at 14 dpi and 20 dpi but only in inoculated leaves. This is a unique feature of quinoa plants that was not observed in any natural or test hosts \[[@B1-cells-09-00148],[@B3-cells-09-00148],[@B7-cells-09-00148],[@B11-cells-09-00148],[@B29-cells-09-00148]\]. At 14 dpi we also spotted invaginations of cell walls and necrosis in phloem and xylem parenchyma tissue inside major vascular bundles. Such necrosis in different vascular tissues due to viral infection has been frequently noted not only for PDV \[[@B9-cells-09-00148]\] but also for other plant viruses, e.g., *Potato virus Y* (PVY, *Potyviridae*) \[[@B34-cells-09-00148]\]. Interestingly, changes in PDV-inoculated leaves were more severe at some distance from major vascular bundles, including deformations in almost all structural tissues and numerous cell wall invaginations. Our results clearly indicate that the anatomical response to PDV did engage almost all leaf tissues in quinoa. The following TEM analysis revealed changes in chloroplast ultrastructure at 7 dpi also in PDV-inoculated leaves. The electron-dense substance, probably lipids, accumulated first on the surface of many palisade mesophyll chloroplasts, and at 14 dpi evaginations of chloroplast envelope, and invaginations of cell wall were spotted in spongy mesophyll cells; at 20 dpi the chloroplasts became completely disintegrated. The localized alterations in chloroplasts in the early stages of infection were similar between susceptible (for example, cucumber) \[[@B8-cells-09-00148]\] and "resistant" (*C. quinoa*) hosts. However, the appearance of these changes varied over time. In susceptible cucumber hosts, the first stage of chloroplast alteration was translucent regions in the stroma between thylakoids, which was not observed in "resistant" *C. quinoa*. It was quickly followed by disintegration of whole chloroplasts in cucumber leaves \[[@B8-cells-09-00148]\]. Moreover, the disintegration process of chloroplasts was much more intense in cucumber than in quinoa. This had a significant influence on the chlorosis level, which was greater in susceptible cucumber than in quinoa. Moreover, in cucumber, severe alteration of mesophyll chloroplasts was accompanied by changes of mitochondria, showing reduced cristae and large electron-translucent regions \[[@B8-cells-09-00148]\]. In quinoa plants, such changes have not been observed. The endoplasmic reticulum (ER) cisterns were also enlarged, especially in companion cells in the phloem. Alterations in chloroplasts and ER by PDV were observed in cucumber \[[@B8-cells-09-00148]\], with numerous electron-translucent regions \[[@B8-cells-09-00148]\] not observed in quinoa. Favali and Conti \[[@B35-cells-09-00148]\] found electron-translucent regions in bean chloroplasts infected with alfalfa mosaic virus (AMV). Such severe alterations in quinoa chloroplasts could be an effect of ROS generation \[[@B36-cells-09-00148]\]. One ROS involved in signal transduction during viral infection \[[@B29-cells-09-00148]\] is H~2~O~2~, and we observed the presence of H~2~O~2~ in changed quinoa chloroplasts. In resistant plants, ROS (also H~2~O~2~) is frequently associated with a hypersensitive response (HR) \[[@B36-cells-09-00148]\], which may induce high levels of cell deformation. As we mentioned above, cell deformation and high levels of H~2~O~2~ are also linked to PDV infection, so deformations in quinoa cells could be a possible result of H~2~O~2~ accumulation. This is supported by statistical analysis showing the level of H~2~O~2~ to increase over 92% (based on CTED) and 63% in spectrophotometric analysis.
As in tobacco and cucumber \[[@B8-cells-09-00148],[@B9-cells-09-00148]\], we observed PDV VPs and MP epitope inside quinoa cytoplasm near plasmodesmata with extended SEL. Moreover, PDV VPs and MP epitope were found near potentially movement protein-induced tubular structures that pass through to the cell wall. Changes in plasmodesmata and generation of microtubules likely reflect the intercellular PDV transport \[[@B8-cells-09-00148]\] in this host, whereas the presence of vesicle pockets and spherules where P1 protein was localized via immunofluorescence serves as evidence of active replication of PDV, similar to other ilarviruses \[[@B9-cells-09-00148]\].
Ultrastructural analysis also reveals changes in vascular tissues such as phloem, where PDV was observed in both parenchyma and companion cells, likely involved in the production of PDV VPs. The resulting viral particles presence could cause necrosis of companion cells. This necrotization could be potentially responsible for blocking virus transport through sieve tubes. The presence of PDV in xylem parenchyma generated clearly visible sequential modification of xylem vessels, filled initially with electron-dense substance (possible/putative phenolic compounds), finally clogging the whole cells. As has been shown by Kozieł et al. \[[@B8-cells-09-00148],[@B9-cells-09-00148]\], systemic movement of PDV in tobacco and cucumber is associated with phloem and xylem cells. If so, then absence of PDV and MP in quinoa leaves above the inoculation site (based on DAS-ELISA, immunofluorescence, and expression of *MP*) could be an effect of simultaneously combining three types of characteristic ultrastructural change that influence quinoa reaction to PDV. First, large-scale cell deformation potentially stalls intercellular transport to the vascular tissues. The second is necrotization of companion cells, possibly indicating that PDV is no longer able to undergo systemic transportation via sieve tubes. The last one is deposition of electron-dense substance in xylem vessels, which is likely to block systemic PDV transport. These elements taken together may contribute to the immunity-like reaction against PDV in quinoa. Observed ultrastructural changes in quinoa cells are also supported by high levels of H~2~O~2~, which may potentially increase the effectiveness of antiviral reactions. Moreover, not only the character of changes in xylem and phloem but also the rate of change could have an influence on quinoa resistance.
5. Conclusions {#sec5-cells-09-00148}
==============
Our study combines immunolocalization of P1, analysis of immunofluorescence signal, statistical assessment of PDV-CP detection, and corrected total electron density assessment of H~2~O~2~ to better understand the reaction of *C. quinoa* to PDV. TEM ultrastructural analysis demonstrated several specific types of ultrastructural changes in quinoa associated with different types of vascular tissues, which could induce/retain PDV in the inoculation zone of the leaf. It was noticed that quinoa plants had symptoms of PDV inoculation. The localization of the main component of replication complex-P1 protein confirmed the potential and possible PDV replication process in inoculated leaves. Interestingly, the deposition of PDV P1 protein as well as virus amount was significantly decreased from 7 and 14 dpi in inoculated leaves. Moreover, the completely lack of the P1 protein deposition was noticed in leaves above inoculation point. Similar observations to P1 protein were detected in the case of PDV MP. The relative expression level of *MP* was detectable only in inoculated leaves but with 80% decrease tendency between 7 and 14 dpi. Moreover, immunolabeling of PDV MP confirmed deposition inside plasmodesmata and potential tubular structures. Therefore, between 7 and 14 dpi after inoculation, the virus could move through plasmodesmata. Despite these observations, PDV MP was not present in tissues above inoculation site, which was confirmed by relative expression of *MP*. Intensive necrotizations of companion cells, electron-dense substance inside xylem tracheary elements, and deformation of plant cells accompanied by decreased amount of virus with increased deposition of H~2~O~2~ influence the restriction of virus systemic translocation through vascular tissues. Because of such specific and quick reactions, the inoculum is eliminated from the leaf, and PDV is not transported systemically to other plant organs. DAS-ELISA and qPCR tests confirmed the absence of the virus in the stem and other leaves and the decreased level of virus in inoculated leaves. *C. quinoa* reactions strongly suggest the presence of some immunity-like reaction to PDV. Further studies based on molecular techniques are required to investigate more deeply the virus fitness and thus the resistance potential in *C. quinoa* plants. Likewise, quinoa--PDV interactions should be investigated in near the future from the point of view of genetic and/or molecular factors possibly engaged in the resistance response. These results could then serve as models for searching the types of antiviral resistance in other plants/crops.
The authors would like to thank Professor Benham E. L. Lockhart from the University of Minnesota (USA) for his help with critical comments on the manuscript and his expert skills with performance and interpretation of DAS-ELISA.
The following are available online at <https://www.mdpi.com/2073-4409/9/1/148/s1>, Scheme S1, Tables S1--4, Figures S1A,B and S2. Scheme S1: Genome structure of PDV, Tables S1 and S2: Sequence of primers and qPCR conditions. Table S3: Characteristic of qPCR reaction. Table S4 and Figure S1A,B: PDV detection and relative virus concentration assessment using DAS-ELISA in quinoa leaves. Figure S2: Corrected total electron density (CTED) of cerium (IV) perhydroxide precipitate in mock- and PDV-inoculated quinoa leaves (7 and 14 dpi) combined with ANOVA.
######
Click here for additional data file.
E.K. and K.O.-K. wrote the manuscript, conceived the idea of research, and performed all experiments. J.J.B. helped in analyzing data and writing the manuscript and performed qPCR analysis. All authors have read and agreed to the published version of the manuscript.
This research was partially funded by National Science Center, Poland, NCN project number: 2019/03/X/NZ9/00499 given to E.K.
The authors declare no conflict of interest.
{#cells-09-00148-f001}
{#cells-09-00148-f002}
{#cells-09-00148-f003}
{#cells-09-00148-f004}
{#cells-09-00148-f005}
{#cells-09-00148-f006}
{#cells-09-00148-f007}
{#cells-09-00148-f008}
{#cells-09-00148-f009}
{#cells-09-00148-f010}
{#cells-09-00148-f011}
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION {#s1}
============
Nasopharyngeal carcinoma (NPC) is a unique head and neck cancer which is highly prevalent in Southern China and throughout Southeast Asia \[[@R1]\], with an incidence of 25--30 cases per 100,000 persons annually \[[@R2]\]. NPC pathogenesis is a multistep process driven by an accumulation of genetic alterations, including the loss of tumor suppressor genes and activation of oncogenes \[[@R3]\]. Malignant cell proliferation and apoptosis inhibition are the main factors influencing the development and progression of NPC and lead to the poor overall survival of NPC patients \[[@R4], [@R5]\]. It is essential to elucidate the molecular mechanisms underlying tumorigenesis and invasiveness in NPC in order to identify novel therapeutic targets and develop new modalities of treatment.
It has been reported that *ARHGEF3*, a Rho-guanine nucleotide exchange factor (GEF) upregulated in acute myeloid leukemia (AML), modulates AML differentiation through activation of RhoA and pathways directly controlled by small GTPase family proteins \[[@R6]\]. The human gene *ARHGEF3* is located at chromosome 3p13-21 and encodes a polypeptide of 526 amino acids with homology to neuroepithelial transforming gene 1 (NET1) \[[@R7]--[@R9]\]. *ARHGEF3* belongs to the family of Rho-GEFs which specifically activates two members of the Rho-GTPase family, RHOA and RHOB, and accelerates Rho-GTPase activity by conversion of GTP to GDP \[[@R10], [@R11]\]. Mutations in some members of the GEF family, such as DOCK2, DOCK8 and *ARHGEF6*, are associated with invasiveness and metastasis of human malignancies \[[@R12]--[@R15]\]. However, the potential roles and biological mechanisms of the GEF family gene *ARHGEF3* in human cancers have not been studied. To investigate if abnormalities in *ARHGEF3* are involved in NPC pathogenesis, we examined *ARHGEF3* protein levels in a series of carcinomatous and non-neoplastic human nasopharyngeal cells and tissues, assessed the clinicopathologic/prognostic significance of *ARHGEF3* expression in our NPC cohort, and investigated the mechanisms underlying the oncogenic and tumorigenic role of *ARHGEF3* in NPC.
We found that high expression of *ARHGEF3* in NPCs is important in the acquisition of an aggressive phenotype. Silencing *ARHGEF3* in NPC cells was sufficient to inhibit cell growth, migration, and invasion *in vitro*, while overexpression of *ARHGEF3* supported the tumorigenic and metastatic capacities of NPC cells *in vivo*. Further, we demonstrated that depletion of *ARHGEF3* in NPC cells promoted caspase3-induced apoptosis. We also identified the anti-apoptosis factor *BIRC8* as a critical downstream target of *ARHGEF3*. Collectively, our results provide an explanation for the malignant nature of NPC involving *ARHGEF3* overexpression and the underlying mechanism that links *ARHGEF3* to *BIRC8* in NPC cell apoptosis.
RESULTS {#s2}
=======
Analysis of *ARHGEF3* protein levels in NPC cells and nasopharyngeal tissues {#s2_1}
----------------------------------------------------------------------------
We analyzed endogenous *ARHGEF3* protein levels in 8 human nasopharyngeal cell lines by Western blotting and found that *ARHGEF3* was overexpressed in 5 NPC cell lines (CNE2, SUNE1, 5-8F, 6-10B and C666), while the other 2 NPC lines (CNE1 and HONE1) and the immortalized normal nasopharyngeal cell line NP69 exhibited low *ARHGEF3* protein levels (Figure [1A](#F1){ref-type="fig"}, left). At the same time, we found that *ARHGEF3* protein expression was higher in 8 primary NPC tissues, compared with adjacent non-neoplastic nasopharyngeal tissues. But there were no difference between the tumor and adjacent tissues in 1 case. (Figure [1A](#F1){ref-type="fig"}, right).
{#F1}
IHC staining of *ARHGEF3* expression in NPC tissues and its correlation with NPC patients' pathological features and survival {#s2_2}
-----------------------------------------------------------------------------------------------------------------------------
Using IHC staining, we observed high expression of *ARHGEF3* (Figure [1B](#F1){ref-type="fig"}, left) in 111 of 192 (57.8%) primary NPC tissues (Table [1](#T1){ref-type="table"}). 17 cases of NPC were not informative due to unrepresentative samples or lost samples. We used the whole NPC tissue slides of these cases to improve the limitation of TMA technology in our study. Correlation analysis demonstrated that high expression of *ARHGEF3* was positively associated with an increased T status, distant metastasis, and/or a more advanced clinical stage of NPCs (*P*\< 0.05, Table [1](#T1){ref-type="table"}). Kaplan-Meier survival curves showed that the mean disease-free survival time in NPC patients with high expression of *ARHGEF3* was significantly shorter than in patients with low expression of *ARHGEF3* (*P*=0.001, long-rank test, Figure [1C](#F1){ref-type="fig"}, Table [2](#T2){ref-type="table"}). Multivariate Cox proportional hazards regression analysis demonstrated that high expression of *ARHGEF3* was a significant and independent prognostic factor for poor survival of NPC patients (relative risk: 1.709, confidence interval: 1.002-2.913, *P*=0.049, Table [2](#T2){ref-type="table"}).
###### Correlation between the clinicopathological features and expression of ARHGEF3 in NPCs
cases ARHGEF3 protein
------------------------------------------- ------- ----------------- ------------ -------
Sex 0.532
Female 57 26 (45.6%) 31 (54.4%)
Male 135 55 (40.7%) 80 (59.3%)
Age at diagnosis (years) 0.299
≤ 47[^†^](#tfn_002){ref-type="table-fn"} 103 47 (45.6%) 56 (54.4%)
\> 47 89 34 (38.2%) 55 (61.8%)
Histological classification (WHO) 0.303
Type II 50 18 (36.0%) 32 (64.0%)
Type III 142 63 (44.4%) 79 (55.6%)
T classification 0.044
1 23 10 (43.5%) 13 (56.5%)
2 65 31 (47.7%) 34 (52.3%)
3 67 32 (47.8%) 35 (52.2%)
4 37 8 (21.6%) 29 (78.4%)
N classification 0.203
0 38 15 (39.5%) 23 (60.5%)
1 89 43 (48.3%) 46 (51.7%)
2 50 20 (40.0%) 30 (60.0%)
3 15 3 (20.0%) 12 (80.0%)
Distant metastasis 0.005
0 152 72 (47.4%) 80 (52.6%)
1 40 9 (22.5%) 31 (77.5%)
Clinical stage 0.002
I 9 3 (33.3%) 6 (66.7%)
II 50 26 (52.0%) 24 (48.0%)
III 83 42 (50.6%) 41 (49.4%)
IV 50 10 (20.0%) 40 (80.0%)
Chi-square test;
median age.
Abbreviation: ARHGEF3, Rho guanine nucleotide exchange factor3; NPC, nasopharyngeal carcinoma; T, tumor; N, node.
###### Univariate and multivariate analysis of different prognostic parameters in 192 patients with NPC
Variable Univariate analysis Multivariate analysis
-------------------------------------------- --------------------- ----------------------- ---------- --------------------- -----------
Sex 0.868
Female 57 1.0
Male 135 0.959 (0.581-1.581)
Age at surgery (years) 0.511
≤ 47[^\*^](#tfn_003){ref-type="table-fn"} 103 1.0
\> 47 89 0.856 (0.539-1.360)
Histological classification (WHO) 0.126
Type II 50 1.0
Type III 142 1.577 (0.881-2.824)
T classification \<0.0001 0.744
T1-T2 88 1.0 1.0
T3-T4 104 2.465 (1.484-4.094) 1.106 (0.605-2.020)
N classification \<0.0001 0.050
N0-N1 127 1.0 1.0
N2-N3 65 2.829 (1.781-4.492) 1.668 (1.001-2.779)
Distant metastasis \<0.0001 \< 0.0001
0 152 1.0 1.0
1 40 4.462 (2.791-7.135) 2.602 (1.575-4.298)
Clinical stage \<0.0001 0.034
I-II 59 1.0 1.0
III-IV 133 6.291 (2.727-14.515) 3.103 (1.087-8.859)
ARHGEF3 expression 0.001 0.049
Low 81 1.0 1.0
high 111 2.334 (1.395-3.905) 1.709 (1.002-2.913)
median age; HR indicates hazards ratio; CI indicates confidence interval.
Abbreviation: NPC, nasopharyngeal carcinoma; ARHGEF3, Rho guanine nucleotide exchange factor3; T, tumor; N, node.
Knockdown of *ARHGEF3* suppresses NPC cell growth, migration, and invasion *in vitro* {#s2_3}
-------------------------------------------------------------------------------------
The above observations prompted us to explore the biological function of *ARHGEF3* in NPC tumorigenesis and progression. The capacity for colony formation was evaluated in two NPC cell lines (CNE2 and SUNE1) that were transfected with si*ARHGEF3* or control siNC. The efficiency of *ARHGEF3* knockdown by si*ARHGEF3* was examined by Western blotting (Figure [2A](#F2){ref-type="fig"}). Both *ARHGEF3*-silenced CNE2 and SUNE1 cells had fewer and smaller colonies than that siNC-transfected cells (Figure [2B](#F2){ref-type="fig"}), indicating that depletion of *ARHGEF3* inhibits growth in NPC cells. Next, the effect of *ARHGEF3* levels on NPC cell migration and invasion capacities were characterized by the wound-healing and Matrigel invasion assays, respectively. Knockdown of *ARHGEF3* in both CNE2 and SUNE1 cells caused a dramatic suppression of cell migration and invasion abilities as compared to control cells (Figure [2C](#F2){ref-type="fig"} and [2D](#F2){ref-type="fig"}).
{#F2}
Upregulated expression of *ARHGEF3* supports the tumorigenic and metastatic capacities of NPC cells *in vivo* {#s2_4}
-------------------------------------------------------------------------------------------------------------
To investigate whether levels of *ARHGEF3* influence the tumorigenic function of NPC cells *in vivo,* we first constructed a HONE1-*ARHGEF3* cell line which stably overexpressed *ARHGEF3* (Figure [2E](#F2){ref-type="fig"}, right, upper) . Next, HONE1-*ARHGEF3* cells were transplanted into the backs of BALB/C-nu athymic nude mice, while HONE1-vector cells were used as a negative control (n = 5 mice per group). Thirty days after cell injection mice were sacrificed and the size and the weight of the subcutaneous tumors were examined. Tumors developed from HONE1-*ARHGEF3* cells were significantly larger and heavier (*P*=0.02) than those arising from control cells (Figure [2E](#F2){ref-type="fig"}). To investigate if increased expression of *ARHGEF3* in NPC cells is causative in an *in vivo* experimental metastasis model, we injected HONE1-*ARHGEF3* or control HONE1-vector cells into the tail vein of BALB/C-nu athymic nude mice (n = 5 mice per group). Eight weeks after injection, mice were killed and metastatic tumor nodules formed in lung and liver were examined. We did not detect tumor nodule formation in the livers of all mice examined, but overexpression of *ARHGEF3* significantly increased metastasis in lung (*P*\<0.01, Figure [2F](#F2){ref-type="fig"}).
Expression level of *ARHGEF3* influences the apoptosis of NPC cells *in vitro* {#s2_5}
------------------------------------------------------------------------------
To further study the effect of *ARHGEF3* on NPC cell apoptosis, we transfected CNE2 and SUNE1 cells with either si*ARHGEF3* or si*BIRC8* for 48h and found that knockdown of *ARHGEF3* and *BIRC8* increased apoptosis in both CNE2 and SUNE1 cells compared with control cells. Then in rescue experiment, three days after the transfection of si*ARHGEF3*, cells were transfected with pcDNA3.1(+)-*BIRC8*. Finally, the apoptotic assay was performed with flow cytometry. Ectopic expression of *BIRC8* in CNE2 and SUNE1 cells with the knockdown of *ARHGEF3* reversed the pro-apoptotic function of si*ARHGEF3* (Figure [3A](#F3){ref-type="fig"}). We also determined that the levels of active, cleaved caspase-3 were substantially increased in si*ARHGEF3*-CNE2 and si*ARHGEF3*-SUNE1 cells when compared to matched control cells (Figure [3B](#F3){ref-type="fig"}). These data suggest that the attenuation of *ARHGEF3* expression promotes NPC cell apoptosis.
{#F3}
*ARHGEF3* regulates apoptosis-related gene expressions in NPC cells {#s2_6}
-------------------------------------------------------------------
In an effort to determine the potential downstream targets of *ARHGEF3* that are involved in the promotion of NPC cell apoptosis, we compared mRNA expression profiles of si*ARHGEF3*-CNE2 cells with those of control siNC-CNE2 cells using a Human Tumor Apoptosis RT^2^ Profiler™ PCR Array containing 84 apoptosis-related genes. We identified a total of 6 downregulated genes (*NAI*P, *BIRC2*, *BIRC3, XIAP*, *BIRC6*, and *BIRC8*) and 1 upregulated gene (*FASLG*) in si*ARHGEF3*-transfected CNE2 cells, which showed more than a twofold change in mRNA levels compared to control siNC-CNE2 cells (Figure [3C](#F3){ref-type="fig"} and Table [3](#T3){ref-type="table"}). Downregulation of *XIAP* and *BIRC8 (ILP-2)* was further validated by Western blotting assay in CNE2 and SUNE1 cells after *ARHGEF3* knockdown (Figure [3D](#F3){ref-type="fig"}). Further, we found a significant positive correlation between the expression of *ARHGEF3* and *BIRC8* in our large cohort of NPC tissues (*P*=0.015, Table [4](#T4){ref-type="table"}). There was no significant difference in *XIAP* expression between the *ARHGEF3* high-expressing and low-expressing groups (*P* = 0.321, Table [4](#T4){ref-type="table"}).
###### List of genes differentially expressed in NPC CNE2 cells after ARHGEF3 knockdown using a human tumor apoptosis real-time PCR array
Gene Fold Change Location Function
------------------------- ------------- ------------- ---------------------------------
**Downregulated genes**
ABL1 −1.47 9q34.12 Induces cell division, adhesion
AKT1 −1.37 14q32.33 Inhibits cell apoptosis
BAD −1.40 11q13.1 Induces cell apoptosis
BAG1 −1.68 9p13.3 Inhibits cell apoptosis
BAG3 −1.83 10q26.11 Inhibits cell apoptosis
BAG4 −1.45 8p11.23 Inhibits cell apoptosis
BAK1 −1.73 6p21.31 Induces cell apoptosis
BAX −1.65 19q13.33 Induces cell apoptosis
BCL2A1 −1.93 15q25.1 Inhibits cell apoptosis
BCL2L1 −1.63 20q11.21 Inhibits cell apoptosis
BCL2L10 −1.03 15q21.2 Inhibits cell apoptosis
BCL2L11 −1.76 2q13 Induces cell apoptosis
BCL2L2 −1.91 14q11.2 Inhibits cell apoptosis
BFAR −1.47 16p13.11 Inhibits cell apoptosis
BID −1.01 22q11.21 Induces cell apoptosis
BIK −1.95 22q13.2 Induces cell apoptosis
**BIRC2** **-2.22** **11q22.2** **Inhibits cell apoptosis**
**BIRC3** **-3.55** **11q22.2** **Inhibits cell apoptosis**
**BIRC6** **-2.27** **2p22.3** **Inhibits cell apoptosis**
**BIRC8** **-5.04** **6q21** **Inhibits cell apoptosis**
BNIP1 −1.03 5q35.1 Inhibits cell apoptosis
BNIP2 −1.69 15q22.2 Inhibits cell apoptosis
BNIP3 −1.65 10q26.3 Inhibits cell apoptosis
BNIP3L −1.60 8p21.2 Inhibits cell apoptosis
BRAF −1.97 7q34 Inhibits cell apoptosis
CARD6 −1.07 5p13.1 Induces cell apoptosis
CARD8 −1.67 19q13.33 Induces cell apoptosis
CASP3 −1.11 4q35.1 Induces cell apoptosis
CASP4 −1.13 11q22.3 Induces cell apoptosis
CASP5 −1.58 11q22.3 Induces cell apoptosis
CASP6 −1.52 4q25 Induces cell apoptosis
CASP7 −1.56 10q25.3 Induces cell apoptosis
CASP8 −1.29 2q33.1 Induces cell apoptosis
CD40 −1.00 20q13.12 Induces cell apoptosis
CFLAR −1.54 2q33.1 Inhibits cell apoptosis
CIDEA −1.13 18p11.21 Induces cell programmed death
CIDEB −1.47 14q12 Induces cell programmed death
CRADD −1.99 12q22 Induces cell apoptosis
DAPK1 −1.02 9q21.33 Induces cell programmed death
FADD −1.66 11q13.3 Induces cell programmed death
FAS −1.50 10q23.31 Induces cell programmed death
GADD45A −1.17 1p31.3 Induces cell apoptosis
HRK −1.01 12q24.22 Induces cell apoptosis
IGF1R −1.74 15q26.3 Inhibits cell apoptosis
LTA −1.47 6p21.33 Induces cell apoptosis
MCL1 −1.86 1q21.3 Inhibits cell apoptosis
**NAIP** **-6.12** **5q13.2** **Inhibits cell apoptosis**
NOL3 −1.45 16q22.1 Inhibits cell apoptosis
RIPK2 −1.58 8q21.3 Induces cell apoptosis
CD70 −1.88 19p13.3 T cell activator
TNFSF8 −1.26 9q32 Induces cell apoptosis
TP53 −1.89 17p13.1 Induces cell apoptosis
TP53BP2 −1.13 1q41 Induces cell apoptosis
TP73 −1.27 1p36.32 Induces cell apoptosis
TRADD −1.38 16q22.1 Induces cell apoptosis
TRAF2 −1.27 9q34 Induces cell apoptosis
TRAF3 −1.85 14q32.32 Induces cell apoptosis
**XIAP** **-4.8** **Xq25** **Inhibits cell apoptosis**
B2M −1.23 15q21.1 Immune response
HPRT1 −1.17 Xq26.1 Nucleotide metabolism
ACTB −1.09 7p22.1 Cytoskeleton actin
**Upregulated genes**
APAF1 1.16 12q23.1 Induces cell apoptosis
BCL10 1.79 1p22.3 Induces cell apoptosis
BCL2 1.17 18q21.33 Induces cell apoptosis
BCLAF1 1.37 6q23.3 Induces cell apoptosis
NOD1 1.75 7p14.3 Induces cell apoptosis
CASP1 1.18 11q22.3 Induces cell apoptosis
CASP10 1.96 2q33.1 Induces cell apoptosis
CASP14 1.08 19p13.12 Induces cell apoptosis
CASP2 1.08 7q34 Induces cell apoptosis
CASP9 1.67 1p36.21 Induces cell apoptosis
CD40LG 1.44 Xq26.3 Inhibits cell apoptosis
DFFA 1.55 1p36.22 Induces DNA damage
**FASLG** **2.99** **1q24.3** **Induces cell apoptosis**
LTBR 1.77 12p13.31 Inhibits cell apoptosis
PYCARD 1.72 16p11.2 Induces cell apoptosis
TNF 1.74 6p21.33 Induces cell apoptosis
TNFRSF10A 1.11 8p21.3 Induces cell apoptosis
TNFRSF10B 1.08 8p21.3 Induces cell apoptosis
TNFRSF11B 1.66 8q24.12 Induces cell apoptosis
TNFRSF1A 1.06 12p13.31 Induces cell apoptosis
TNFRSF21 1.40 6p21.1 Induces cell apoptosis
TNFRSF25 1.09 1p36.31 Induces cell apoptosis
CD27 1.39 12p13.31 Induces cell apoptosis
TNFRSF9 1.51 1p36.23 Induces cell apoptosis
TNFSF10 1.59 3q26.31 Induces cell apoptosis
RPL13A 1.14 19q13.3 Protein metabolism
GAPDH 1.02 12p13.31 Glycometabolism
Abbreviation: NPC, nasopharyngeal carcinoma; ARHGEF3, Rho guanine nucleotide exchange factor3.
###### Correlation between expression of ARHGEF3 and that of BIRC8 and XIAP in 192 patients with NPC
All cases BIRC8 expression XIAP expression
-------------------- ----------- ------------------ ----------------- ------- ------------ ------------ -------
ARHGEF3 expression 0.015 0.321
Low 81 48 (59.3%) 33 (40.7%) 46 (56.8%) 35 (43.2%)
High 111 46 (41.4%) 65 (58.6%) 55 (49.5%) 56 (50.5%)
Chi-square test.
Abbreviation: NPC, nasopharyngeal carcinoma; ARHGEF3, Rho guanine nucleotide exchange factor3.
DISCUSSION {#s3}
==========
*ARHGEF3* is a key activator of Rho-GTPases including, in particular RhoA, Rac1, and CDC42, which function as molecular switches in a variety of cellular signaling pathways \[[@R16], [@R17]\]. Increasing evidence suggests that Rho-GTPases are frequently deregulated during tumor progression, which promotes malignant phenotypes in cancer cells and is correlated with poor patient prognosis \[[@R18]--[@R20]\]. However, the molecular status of *ARHGEF3* and its potential function in the underlying mechanisms of NPC are unclear.
In the present study, we demonstrated that the majority of human NPC cell lines and tissues expressed high levels of endogenous *ARHGEF3* protein compared to control non-neoplastic nasopharyngeal cells and tissues. These findings suggest that upregulated expression of *ARHGEF3* provides a selective advantage in NPC pathogenic processes.
Our analyses also found that high *ARHGEF3* expression in our NPC cohorts was positively correlated with tumor T status, distant metastasis, and advanced clinical stage, suggesting that high expression of *ARHGEF3* facilitates a malignant phenotype in NPC. Further, we observed that expression of *ARHGEF3* was a strong and independent prognostic predictor for NPC patients. These findings underscore a potentially important role of *ARHGEF3* in the development and progression of NPC, and suggest that examination of *ARHGEF3* expression by IHC could be used as an additional tool in identifying those NPC patients at increased risk of tumor growth and/or metastasis.
Our investigation of the mechanisms through which *ARHGEF3* regulates NPC cell malignancy demonstrated that knockdown of *ARHGEF3* in CNE2 and SUNE1 NPC cells dramatically repressed cell growth, migration, and invasion *in vitro*. In contrast, *ARHGEF3* overexpression in HONE1 NPC cells induced cell tumorigenicity *in vivo*. Moreover, in a tail vein injection mouse model of cancer metastasis ectopic overexpression of *ARHGEF3* in HONE1 cells led to a significant increase in the number of metastatic lung lesions. These data support our emerging view that increased *ARHGEF3* expression is a critical molecular event in the process of NPC pathogenesis.
It has been reported that cancer cells exhibit deficiencies in the induction of apoptosis, resulting in accelerated tumor development and reduced responsiveness to anti-cancer therapies \[[@R21]--[@R23]\]. Conversely, increased apoptosis in cancer cells may provide therapeutic benefits, especially in apoptosis-defective cancers \[[@R24], [@R25]\]. In the current study, we found that silencing *ARHGEF3* in NPC cells could induce apoptosis, as measured by an increased percentage of annexin V positive cells and increased cleavage of caspases3. Gene expression profiling also revealed that silencing of *ARHGEF3* resulted in downregulation of a number of genes, most notably *BIRC8*. Further, we did observe a significant positive correlation between expression of *ARHGEF3* and *BIRC8* in our large cohort of NPC clinical tissues.*BIRC8 (ILP-2)* belongs to the inhibitors of apoptosis protein (IAP) family, which are apoptosis inhibitors that may protect against apoptotic stimuli and suppress apoptosis \[[@R26]\]. It has been reported that *BIRC8* exerts its effects by association with an inhibition of specific caspases \[[@R27]\]. In addition, studies have documented that *BIRC8 (ILP-2)* is a tumor biomarker and promotes cancer progression \[[@R28], [@R29]\]. These results suggest that *ARHGEF3* regulates cell apoptosis via control of BIRC8 expression, which is in turn involved in attenuation of caspases3-induced apoptosis in the pathogenesis of NPC.
Our study demonstrates, for the first time, the protein expression dynamics of *ARHGEF3* in a large cohort of clinical NPC tissues. High expression of *ARHGEF3* may be important in tumorigenesis and acquisition of a poor prognostic phenotype of human NPC. Importantly, our functional and mechanistic studies suggest an important oncogenic role for *ARHGEF3* in the suppression of NPC cell apoptosis by regulating *BIRC8* expression and caspases3-induced apoptosis, activities that might be responsible for the development and progression of human NPCs.
MATERIALS AND METHODS {#s4}
=====================
Nasopharyngeal cell lines and tissue specimens {#s4_1}
----------------------------------------------
Seven human NPC cell lines (CNE1, CNE2, HONE1, SUNE1, 5-8F, 6-10B and C666) and one immortalized normal nasopharyngeal cell line (NP69) were cultured in RPMI-1640 supplemented with 10% fetal bovine serum. 192 specimens of NPC and 50 specimens of non-neoplastic nasopharyngeal mucosa were collected at Sun Yat-Sen University Cancer Center, Guangzhou, China, between January 1991 and August 2000. A nasopharyngeal tissue microarray (TMA) was then constructed. In addition, 9 pairs of fresh NPC tissues and adjacent non-neoplastic nasopharyngeal mucosa specimens were collected at Guangdong Provincial People\'s Hospital (Guangzhou, China) in 2012. None of the NPC patients had received preoperative radiation or chemotherapy before diagnosis. A pathological diagnosis for all specimens was confirmed according to the 2005 WHO histological classification of NPC. Tumor stage was defined according to the criteria of the sixth edition of the TNM classification of the Union for International Cancer Control (UICC, 2002). The study was approved by the Institute Research Medical Ethics Committee of Sun Yan-Sun University Cancer Center (Guangzhou, China).
Western blotting assay {#s4_2}
----------------------
Equal amounts of whole cell and tissue lysates were resolved by SDS-polyacrylamide gel electrophoresis and electrotransferred on a polyvinylidenedifluoride membrane (Pall Corp., Port Washington, NY). The samples were then incubated with primary antibodies against *ARHGEF3* (Abgent Laboratories, San Diego, CA), caspase-3, cleaved caspase-3, caspase-8 (BD Biosciences, San Jose, CA), *PARP, NAIP*(Abcam, Cambridge, UK), *BIRC2, BIRC3, XIAP, BIRC6, BIRC8* (Proteintech, Chicago, IL), *FASLG* (Abnova, Taibei City, Taiwan), α-tubulin (Santa Cruz Biotech, Dallas, TX), or *GAPDH* (BD Biosciences, San Jose, CA). The immunoreactive signals were detected with an enhanced chemiluminescence reagent kit (Amersham Biosciences, Uppsala, Sweden). The procedures were conducted in accordance with the manufacturer\'s instructions.
Immunohistochemistry (IHC) staining {#s4_3}
-----------------------------------
IHC studies were performed using a standard streptavidin biotin-peroxidase complex method. For antigen retrieval, TMA slides were microwave treated in 10 mM citrate buffer (pH 6.0) for 10 min. The TMA slides were incubated with anti-*ARHGEF3* (1:100 dilution; Abgent Laboratories, San Diego, CA), in a moist chamber overnight at 4°C. A negative control was obtained by replacing the primary antibody with a normal rabbit IgG.
A semi-quantitative scoring criterion for IHC of *ARHGEF3* was used, in which both staining intensity and positive areas were recorded. A staining index (values 0 to 12) obtained as the intensity of positive staining (negative = 0, weak = 1, moderate = 2, or strong = 3 scores) and the proportion of immunopositive cells of interest (\< 25% = 1, 25 to 50% = 2, 51% to 75% = 3, \> 75% = 4 scores) were calculated. Because the staining index of expression of *ARHGEF3* in all 50 non-neoplastic nasopharyngeal mucosa tissues was no more than 4, we designated staining index scores of 0-4 as "low" expression of *ARHGEF3* (Figure [1B](#F1){ref-type="fig"} right) and staining index scores of 6-12 was as "high" expression of *ARHGEF3* (Figure [1B](#F1){ref-type="fig"} left). The cutoff scores for determining "high" level expression of *XIAP* and *BIRC8 (ILP-2)* were determined at staining index scores of \>4 and \>6, respectively.
Small interfering RNA (siRNA) {#s4_4}
-----------------------------
CNE2 and SUNE1 cells were cultured in six-well plates. The cells were transfected with anti-*ARHGEF3* siRNA or anti-control siRNA (Ambion, Austin, Texas) using Lipofectamine 2000 reagent (Invitrogen, Carlsbad, CA) according to the manufacturer\'s instructions. The gene silencing effect was measured by Western blotting 48 h post-transfection.
Colony formation assay {#s4_5}
----------------------
Five hundred infected cells were placed in a fresh six-well plate and maintained in RPMI1640 containing 10%FBS for 10 days. Colonies were fixed with methanol and stained with 0.1% Giemsa in 20% methanol for 15 minutes.
Wound-healing and matrigel invasion assays {#s4_6}
------------------------------------------
Cell migration was assessed by measuring the movement of cells into a scraped cellular area created by 200μL pipette tip. The spread of wound closure was observed after 48h and photographed under a microscope. We measured the fraction of cell coverage across the line to calculate migration rate.
For invasion assays, 1×10^4^ cells were added to a Matrigel invasion chamber (BD Biosciences, Becton Dickson Labware, Franklin Lakes, NJ) present in the insert of a 24-well culture plate. FBS was added to the lower chamber as a chemoattractant. After 48h, the non-invading cells were gently removed with a moist cotton swab. Invasive cells located on the lower side of the chamber were fixed with paraformaldehyde and then stained with crystal violet, air dried, and photographed. For colorimetric assays, the samples were treated with 150μL 10% acetic acid and absorbance was measured with a spectrophotometer at 560nm.
Plasmid constructs and transfection {#s4_7}
-----------------------------------
*ARHGEF3* and *BIRC8* complementary DNA (Fulengen, Guangzhou, China) was cloned into a pcDNA3.1 plasmid. HONE1 Cells were transfected with pcDNA*-ARHGEF3* or the control plasmid pcDNA3.1(+) using Lipofectamine 2000 (Invitrogen, Carlsbad, CA) according to the manufacturer\'s instructions. For the establishment of the *ARHGEF3*-HONE1 cell line stably expressing *ARHGEF3*, the cells were split at a ratio of 1:10 48 h after transfection. Next, cells were maintained in Leibovitz\'s L-15 medium containing 200 μg/mL of G418 (Calbiochem, San Diego, CA). After 6 weeks of selection, resistant colonies stably transfected with pcDNA-*ARHGEF3* (HONE1 pcDNA-*ARHGEF3*) or pcDNA3.1(+) \[HONE1pcDNA3.1(+)\] were pooled.
*In vivo* tumorigenesis and metastasis assays {#s4_8}
---------------------------------------------
For the *in vivo* assays of subcutaneous tumorigenesis of NPC cells, 1×10^6^ mixed populations of HONE1-*ARHGEF3* cells stably overexpressing *ARHGEF3* or the control HONE1--vector were injected subcutaneously into the backs of 4-week-old male BALB/C-nu athymic nude mice. At day 30, the mice were sacrificed, the tumors were removed, and tumor weight was calculated.
To evaluate the metastasis potential of NPC cells *in vivo*, five 4-week-old male BALB/C-nu athymic nude mice in each experimental group were injected with HONE1-*ARHGEF3* and HONE1-vector cells, separately. Briefly, each mouse received 2×10^5^ cells via tail vein injection. Eight weeks after cell injection mice were killed and the lungs and the liver were removed from each mouse and fixed with phosphate-buffered neutral formalin. Consecutive tissue sections were made for each block of the tissues, which were then stained with haematoxylin-eosin. Finally, the slides of the lungs and the liver were carefully examined under a microscope. All experimental procedures involving animals were are accordant with the Guidelines for the Care and Use of Laboratory Animals (NIH publications Nos. 80-23, revised 1996).
Apoptosis assay {#s4_9}
---------------
To assess the rate of apoptosis, transfected cells were harvested and washed twice with cold PBS, and the Annexin V-PI Kit (Nanjing Keygen, Nanjing, China) was used according to the manufacturer\'s guidelines. The detection was performed with a FACS Calibur using Cell Quest software (BDIS, San Jose, CA).
Real-time PCR gene array {#s4_10}
------------------------
RNA was extracted from si*ARHGEF3*-CNE2 and siNC-CNE2 cells using Trizol (Invitrogen, Carlsbad, CA) and was cleaned them using the RNeasy MinElute cleanup kit (Qiagen, Valencia, CA). Messenger RNA expression levels were quantified with a Human Tumor Apoptosis RT^2^ Profiler PCR array (Super Array Bioscience, Frederick, MD). Si*ARHGEF3*-CNE2 cells were compared with control siNC-CNE2 cells using a Human Apoptosis RT^2^ Profiler PCR array containing 84 key genes involved in programmed cell death.
Statistical analysis {#s4_11}
--------------------
Statistical analysis was performed using the SPSS software package (SPSS Standard version18.0, SPSS Inc). Differences between variables were assessed by the Chi-square test or Fisher\'s exact test. For survival analysis, we analyzed all patients with NPC by Kaplan-Meier analysis. A log rank test was used to compare different survival curves. Multivariate survival analysis was performed on all parameters that were found to be significant in univariate analysis using the Cox regression model. Data derived from cell line experiments are presented as mean ±SD and assessed by the two-tailed Student\'s *t* test. *P* values of \< 0.05 were considered statistically significant.
**CONFLICTS OF INTEREST**
The authors declare that there is no conflict of interest.
**GRANT SUPPORT**
This work was supported by grants from the Nature Science Foundation of China (No. 81225018 and 81572848), the Science and Technology Planning Project of Guangdong Province, China (No. 2013B021800103)and the Foundation of Guangzhou Science and Technology Innovation Commission, China (No.2014J4100181).
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .507}
{#sp2 .508}
{#sp3 .509}
{#sp4 .510}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#sec1}
============
Preventing the transmission of COVID-19 within prisons and other prescribed places of detention (PPDs) is an integral part of the public health response to the current pandemic.[@ref1] PPDs concentrate individuals who may be vulnerable to severe infection due to poor health profiles, and the confined conditions in which detainees live can encourage person-to-person spread, increasing the basic reproduction number (R~0~) above that seen in the general population.[@ref1]^,^[@ref2]^,^[@ref3] Furthermore, outbreaks of COVID-19 within these settings have the potential to overwhelm prison healthcare services and place additional burden on hospital critical care units in the community.
Guidance for how to prevent and control COVID-19 outbreaks within PPDs is commonly adapted from the guidance for the general public, with additional measures included to reflect the increased susceptibility of the population and likelihood of explosive outbreaks. Therefore, the approach to COVID-19 in PPDs often reflects the approach for the country as a whole. Social distancing measures adopted by many countries across the world are challenging to implement in prisons, where individuals are already deprived of their liberty. There have been various approaches to facilitate this such as implementing 'restricted regimens' within prisons and by enabling single cell accommodation through movement of people in prison to alternative accommodation, temporary or early release of detainees, and through increasing community disposals.[@ref4]
As countries look to move forward and consider easing social distancing measures, alternative approaches will need to be explored in order to prevent a resurgence of cases. Contact tracing is an important contributor to the longer-term management strategy for COVID-19 both in the community and prisons. Contact tracing aims to rapidly identify secondary cases that may arise through transmission from known cases, allowing for intervening action to take place to interrupt further onward transmission. In a prison setting, this outbreak management tool has the potential to rapidly deploy sufficient resources and bring outbreaks to a close quickly. However, the resource implications of these activities in terms of personnel are significant.[@ref5]
Successes in managing COVID-19 to date have been informed by international experiences, learning from countries where strategies and actions have already been established. In Ireland, contact tracing is central to the response strategy and is currently being implemented alongside social distancing measures.[@ref6] This approach has been extended to all prisons within Ireland, with the adoption of a broader case definition to provide further protection against prison-level outbreaks.
This paper describes the approach to contact tracing in the Irish Prison Service (IPS).
Methods {#sec2}
=======
Aim of the programme {#sec3}
--------------------
Within Ireland, contact tracing for cases of COVID-19 is primarily managed by the national public health agency, Public Health (PH). However, similar to many countries, the rapid increase in the number of cases has impacted on the capacity of PH to manage the volume of contact tracing required in all areas of the wider community. A Contact Management Programme (CMP) was established in prisons to enable contact tracing to operate at scale and to support the work being done by PH, with the National Quality Improvement (QI) team providing an education and training enabling function within the CMP. In response to the contact tracing requirements in prisons, the IPS National Infection Control Team (NICT), with the support of PH and the National QI team, commenced a programme to develop and train contact tracing teams (CTTs) within prisons. CTTs were intended to be prison-based and run by members of staff from within each prison. The aim of this programme was to enable the IPS to assist PH in the early identification of people that may have been exposed to COVID-19 and take action to prevent onward transmission.
Contact tracing teams {#sec4}
---------------------
The NICT manager initially contacted all Governors within the IPS to outline the importance of contact tracing in the management of COVID-19 within prisons and the proposed programme for achieving this. Governors were then asked to suggest names of personnel within their prison staff who they considered would be interested in being invited to be part of a CTT.
Selection criteria for CTT members included:
- Experience in the use of CCTV within the prison.
- Ability to use the prison IT system, including use of Microsoft® Excel®.
- Ability to use the IPS rostering/clocking system.
These prerequisites were essential to ensure that the CTTs had the expertise required to establish the movements of detainees and staff and therefore identify any potential contacts of cases.
Each CTT was made up of a minimum of four members of staff either side of the roster, which consisted of Security Chiefs, Assistant Chief Officers, Prison Officers, an Assistant Psychologist and clerical staff. The final membership of the CTT within each prison was determined locally, taking into account the skill sets of members.
Development and delivery of training {#sec5}
------------------------------------
The NICT, PH and the National QI team jointly developed a contact tracing training package. This tripartite collaboration was crucial to ensuring that the required standard of contact tracing would be accomplished.
Training then occurred in two phases. The first phase involved delivery of a 'Train-the-Trainer' package for all members of the NICT, under the guidance of PH and the National QI team. Once trained, NICT members then delivered face-to-face training to the 158 staff across the IPS who had been selected to be members of a CTT. CTTs training consisted of the following:
1. \(i\) An overview of COVID-19 and modes of transmission.
2. \(ii\) A description of the contact tracing process.
3. \(iii\) An introduction to the interview scripts.
On completion of the training, all members of the CTTs were provided with contact tracing protocols, interview scripts for undertaking contact tracing of both staff and detainee cases and an Excel® template for collection of information regarding close and casual contacts of cases.
{#f1}
Process of contact tracing within prisons {#sec6}
-----------------------------------------
Within the IPS, contact tracing is commenced for all confirmed and highly probable cases of COVID-19 amongst both detainees and prison staff (see [Box 1](#box01){ref-type="boxed-text"} for case definitions).
######
Case definitions.
**Confirmed case:** An individual with laboratory confirmed COVID-19.
**Highly probable case:** An individual with either a cough, a fever of 38.0 °C or above or shortness of breath.
The process of contact tracing is summarized in [Fig. 1](#f1){ref-type="fig"}. Once a case is identified by, or reported to, a member of prison staff, they immediately notify the CTT. The case is then interviewed by a member of the CTT to establish all close and casual contacts during the 48 hours prior to, and during the period since, symptom onset in the case (see [Box 2](#box02){ref-type="boxed-text"} for definitions of contacts).
A member of the CTT also reviews any CCTV footage from within the prison, which contains footage of the case during the 48 before, and period since, symptom onset. This process is used to identify any additional contacts not reported during the interview with the case and to provide further details on the nature and proximity of contact events.
Information relating to the case and all close and casual contacts is collected and recorded in an Excel® spreadsheet and saved securely on the prison IT system. This data comprises name of the case, date of symptom onset, the result of their COVID-19 test and symptom status of any close and casual contacts. This information is securely emailed to the NICT and PH. This data is collated centrally by PH in the Health Service Executive (HSE).
Management of contacts {#sec7}
----------------------
All close contacts are considered 'at risk'. Detainee close contact are placed in isolation within the prison, and staff close contacts are advised that they need to self-quarantine at home. The period of isolation/quarantine is dependent on the COVID-19 test result of the case. If the case returns a negative result, the close contacts are informed that they no longer need to be in isolation/quarantine, provided they are asymptomatic. If the case returns a positive result, the close contacts are informed that they need to remain in isolation/quarantine for 14 days from the date of contact with the case. During this period detainee close contacts receive daily clinical monitoring. Any staff close contacts are notified to PH and followed up by the community CMP. If close contacts develop symptoms consistent with COVID-19, they are then required to remain in isolation/quarantine, testing is arranged, and tracing of their contacts is also undertaken.
Casual contacts identified are provided with an information leaflet, which advises that they do not need to isolate/self-quarantine but that they should self-monitor for signs and symptoms of COVID-19.
######
Definitions of close and casual contacts.
Contact tracing includes contact from 48 hours before symptom onset.
**Close contact:**
- Any person who has shared a space with for longer than 2 hours with a case.
```{=html}
<!-- -->
```
- Any person who has had face-to face contact with a case for a total of 15 minutes over the period of a day.
```{=html}
<!-- -->
```
- Any person who has not worn appropriate PPE or had a breach of PPE when dealing with a case.
**Casual contact:**
- Any person who has shared a closed space with a case for less than 2 hours.
```{=html}
<!-- -->
```
- Any person who has worn appropriate PPE and taken recommended infection control precautions and that has direct contact with a case or their body fluids.
```{=html}
<!-- -->
```
- Any person who has shared a closed space with a case for longer than 2 hours but, following a risk assessment, does not meet the definition of a close contact.
Supporting resources {#sec8}
--------------------
To support contact tracing, information leaflets about COVID-19 and contact tracing were developed for both detainees and prison staff, as well as posters for display around the prisons.
Flow charts for use within the IPS detailing the procedures for reporting cases, contact tracing and follow-up were jointly developed and agreed by the IPS, the National QI team and HSE. Scripts were created for use by the CTTs for each step of the contact tracing process: the initial interview with the case to establish contacts, the discussion with contacts regarding isolation/quarantine and symptoms and the follow-up discussion with contacts following the test result of the case. Leaflets for close contacts were developed to provide information about what activities are and are not permitted during isolation/quarantine; how to safeguard themselves and others, including those who may be at higher risk of complications from COVID-19; and what to do if they become symptomatic themselves.
Results {#sec9}
=======
Contact tracing teams {#sec10}
---------------------
All 12 prisons and the two support agencies, Operational Support Group (OSG) and Prison Service Escort Corp (PSEC), within the IPS now have fully functional in-prison CTTs. Every CTT has responded to at least one case of highly probable or confirmed COVID-19, undertaken contact tracing and instigated quarantine of contacts.
Detainee cases {#sec11}
--------------
Between 6 April (date of establishment of the first CTT) and 22 May 2020 (time of writing), there have been 66 highly probable cases of COVID-19 identified amongst prison detainees, all of which have subsequently tested negative for COVID-19. There have been no confirmed cases of COVID-19 amongst the detainee population in the IPS during this period.
A total of 84 close contacts (50 detainees, 9 staff and 25 externals) of detainee cases were identified and followed up by prison CTTs, resulting in a mean of 1.3 close contacts per detainee case.
Staff cases {#sec12}
-----------
Between 6 April and 22 May 2020, there have been 119 highly probable and 45 confirmed cases of COVID-19 identified amongst prison staff. This includes 13 historic cases, which predated the establishment of the CTT in every prison.
A total of 448 close contacts of staff cases were identified and followed up by prison CTTs, all of whom were prison staff members, resulting in a mean of 2.8 close contacts per staff case.
Discussion {#sec13}
==========
Main finding of this study {#sec14}
--------------------------
We have described an approach to implementing contact tracing for cases of COVID-19 within prisons, with the aim of preventing and controlling outbreaks. By taking a collaborative approach, the prison service and the national public health agency in Ireland were able to achieve the rapid creation and deployment of in-prison CTTs in every estate within the IPS. Working to agreed contact tracing protocols, CTTs have undertaken contact tracing for 230 cases within the IPS to date.
What is already known on this topic {#sec15}
-----------------------------------
Contact tracing can be a highly effective approach to controlling the spread of COVID-19 and preventing outbreaks.[@ref3] However, this activity is highly labour intensive and can quickly place a strain on the resources of public health organizations.
What this study adds {#sec16}
--------------------
To our knowledge, there has not been any scientific literature published on the subject of contact tracing for cases COVID-19 in PPDs. This case study provides an example of a partnership approach to contact tracing that could be adopted by other countries over the next few months as they look for alternatives to the highly restrictive social distancing measures currently in place across Europe and elsewhere.
The approach taken in Ireland demonstrates that prison-based CTTs, run by prison staff, are ideally placed for undertaking contact tracing within prisons. Prison staff are experienced in working with detainees and fellow staff members. They are physically located within prisons and therefore have the ability to rapidly respond to notifications, complete interviews with cases and implement isolation/quarantine of the case and contacts. The ability to conduct contact tracing without delay, starting on the day that a case is identified, avoids missing opportunities to prevent further onward transmission from potentially infected contacts.
Furthermore, prison staff selected for CTTs in the IPS were experienced at accessing and interpreting prison CCTV. This means they are able to use this information to enhance intelligence gathered through interview and establish whether there are any additional contacts not previously discovered. An additional benefit of prison staff specifically undertaking this activity is that their familiarity with detainees and staff enables them to identify individuals by sight on the CCTV footage.
The combination of contact tracing interviews with CCTV footage has also provided an opportunity to identify 'hotspots' within prisons where advised social distancing is not being observed. CTTs have been able to use this information to help inform activities aimed at improving adherence to social distancing advice.
Limitations of this study {#sec17}
-------------------------
The epidemiology of COVID-19 in both the detainee and staff populations within the IPS is reflective of the number of cases of the disease within Ireland, and this may limit the applicability of this case study to other countries where the incidence may be different.[@ref7] Furthermore, the absence of any confirmed cases of COVID-19 amongst detainees in the IPS means that it is not possible to draw any conclusions about the effectiveness of this contact tracing approach in preventing transmission or outbreaks within the prison estate in Ireland.
The number of prisons and size of the prison population within the IPS is smaller than many countries. It is possible that larger prison services may find it challenging to adopt the approach taken in Ireland. However, this approach was designed with the aim of enabling prisons to take a key role in the contact tracing process and therefore has the ability to be scaled up.
Contact tracing of cases in prisons is only part of the IPS's response to the COVID-19 pandemic. Every country is likely to need to implement a range of measures to prevent and control outbreaks of COVID-19 in PPDs.
Conclusions {#sec18}
===========
A partnership approach involving community public health expertise, QI input and support and prison resources can provide an effective mechanism for contact tracing of COVID-19 cases within the vulnerable prison setting. As countries look to explore alternatives to stringent social distancing measures, in-prison CTTs offer a potential solution to the significant resource burden of implementing contact tracing in prisons and other PPDs.
Funding {#sec19}
=======
No funding was received for this study.
Conflicts of interest {#sec20}
=====================
No conflicts of interest to declare.
The authors would like to thank Elaine Dunne (IPS), Liam Philips (IPS), Philip Kennedy (IPS), Padraic Carty (IPS), Tom Malone (IPS), Mark Farrelly (IPS), Darren McDonnell (IPS), Dr Mary Browne (HSE), Lisa Toland (HSE), Lorraine Murphy (HSE) and Dr Philip Crowley (HSE).
**Mattea Clarke,** Specialist Registrar in Public Health
**John Devlin,** Executive Clinical Lead
**Emmett Conroy,** Infection Prevention & Control Manager
**Enda Kelly,** National Operational Nurse Manager
**Sunita Sturup-Toft,** Public Health Specialist
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction
===============
Glucose that comes from the mother provides fuel for fetal growth \[[@B1-nutrients-05-00001]\]. High maternal blood glucose levels (BGL), even within the current recommended range, have been associated with higher infant body fat \[[@B2-nutrients-05-00001]\]. The peak postprandial BGL (PBGL) occurs later in pregnant women than in the non-pregnant state \[[@B3-nutrients-05-00001]\], *i.e.*, at 60 *vs.* 30 min, and is an important contributor to the risk of fetal overgrowth \[[@B4-nutrients-05-00001],[@B5-nutrients-05-00001]\]. Treatment of gestational diabetes mellitus (GDM), where maternal glucose homeostasis is impaired, is therefore generally based on a combination of pre-meal BGL and PBGL 1 or 2 h after meals as monitored by self-blood glucose monitoring \[[@B6-nutrients-05-00001],[@B7-nutrients-05-00001]\].
The glycemic index (GI) is a measurement of the glycemic quality of the carbohydrates in foods, where a low GI indicates that the carbohydrates are digested and absorbed slowly \[[@B8-nutrients-05-00001]\]. The limited evidence available suggests that GI of meals is of relevance in the setting of GDM \[[@B9-nutrients-05-00001]\]. A low GI diet was shown to reduce the requirement for insulin in the glycemic management of women with GDM \[[@B10-nutrients-05-00001]\], and was equally effective in improving pregnancy outcomes in GDM when compared to a conventional high fibre diet \[[@B11-nutrients-05-00001]\].
While the GI concept may be valid in diabetic pregnancy, scepticism of the efficacy of a low GI diet in GDM remains. This could partly be attributed to the lack of scientific evidence to demonstrate that low GI meals actually reduce, but not delay, the peak PBGL in GDM. A recent analysis of glycemic responses to over 1000 foods indicated that the timing of the peak was the same for high *vs.* low GI foods in non-diabetic individuals \[[@B12-nutrients-05-00001]\]. Whether this is true in pregnancy, especially those complicated by GDM, remains unclear because changes in gastric motility \[[@B13-nutrients-05-00001]\] and insulin sensitivity \[[@B14-nutrients-05-00001]\] during pregnancy may alter digestion and absorption.
In this study we investigated whether two bread-based breakfasts with different GI produced different postprandial peaks and peak time points in a group of women with GDM. Our hypothesis was that a low GI bread-based breakfast would produce a lower but not delayed PBGL peak in GDM when compared to an energy and macronutrient matched high GI bread-based breakfast.
2. Experimental Section
=======================
2.1. Subjects
-------------
All women who attended the Royal Prince Alfred Hospital GDM antenatal clinic during June 2010 to November 2011 were approached for recruitment, and ten women aged 18--45 years, who had been diagnosed with GDM by a 75 g oral glucose tolerance test (OGTT) using the following criteria: fasting glucose 5.5 mmol/L or more and/or 1 h post-load glucose of 10.0 mmol/L or more and/or 2 h post-load glucose of 8.0 mmol/L or more; were between 30 and 32 weeks of gestation, with no known food allergy and/or special dietary requirement and not currently using insulin were enrolled in the study. Information about demographics and ethnicity was gathered. All women in this study received a standard GDM group education session with an experienced diabetes dietitian, which covered carbohydrate counting, importance of even distribution of carbohydrates throughout the day, and food sources of carbohydrate, with no specific emphasis of GI.
2.2. Anthropometry and Self Blood Glucose Monitoring (SBGM)
-----------------------------------------------------------
Subjects were weighed at the first study session in light clothing and with shoes off, and their height was obtained from their electronic medical record. Subjects were instructed to self-monitor their BGL using a glucometer (AccuChek Performa, Roche Diagnostic, Castle Hill, NSW, Australia). Fasting BGL on the 7 previous days was obtained from the electronic record of the glucometer.
2.3. Test Meals
---------------
Subjects were required to fast for at least 8 h prior to the start of the test sessions. They consumed a carbohydrate controlled, low GI bread-based breakfast on one occasion, and an energy and macronutrient matched high GI bread-based breakfast on the other occasion one to two weeks apart. The order of test meals was randomized, and the allocation sequence was unpredictable and concealed from the recruiter. The subjects were asked to complete the meals within 15 min. The composition and nutritional content of the two test meals ([Table 1](#nutrients-05-00001-t001){ref-type="table"}) were analysed with FoodWorks Professional (version 2009, Xyris Software, Brisbane, Australia), using AUSNUT2007 as the source of nutrition composition \[[@B15-nutrients-05-00001]\].
nutrients-05-00001-t001_Table 1
######
Composition and nutritional analysis of the test meals. ^a^ Burgen fruit and muesli bread; ^b^ Flora mono-sun margarine; ^c^ Devondale light dairy blend; ^d^ Benefibre fibre supplement; ^e^ Tip Top Sunblest wholemeal; ^f^ Lucozade orange flavour.
Low GI High GI
---------------------------- ------------------------------- --------------------------
Foods included 70 g fruit bread ^a^ 60 g wholemeal bread ^e^
3 g margarine ^b^ 134 g Fizzy glucose drink ^f^
3 g light dairy blend ^c^ 1 hardboiled egg
200 mL skim milk
1.7 g fibre supplement ^d^
*Nutritional Analyses*
Energy (kcal) 328 328
Protein (% kcal) 18.9 18.3
Fat (% kcal) 22.1 24.7
Carbohydrates (% kcal) 54.5 52.1
Dietary Fibre (g) 4.2 3.9
Glycemic index 45 82
Glycemic load 21 36
2.4. Quantification of Fasting and Postprandial Blood Glucose Levels
--------------------------------------------------------------------
Finger prick blood samples were collected before the start of the meal, and at 15, 30, 45, 60, 75, 90 and 120 min after the start of the meal. The blood samples were analysed immediately after collection using a portable blood analyser (HemoCue Glucose Analyzer 201, HemoCue Australia Pty Ltd., Wamberal, Australia).
2.5. Subjective Satiety
-----------------------
Participants were asked to indicate their subjective satiety rating on a 15 cm visual analogue scale at each blood sample collection, with 0% representing "extremely hungry" and 100% representing "extremely full".
2.6. Statistical Analyses
-------------------------
All statistical analyses were performed in IBM SPSS version 19 (IBM Australia, St Leonards, Australia). Two women withdrew from the study after the first session, and their results were used only in the overall analysis ([Figure 1](#nutrients-05-00001-f001){ref-type="fig"}). Mean ± SEM BGL for all subjects ([Figure 1](#nutrients-05-00001-f001){ref-type="fig"}), as well as that of the individual subjects ([Figure 2](#nutrients-05-00001-f002){ref-type="fig"}) were plotted against time to obtain postprandial blood glucose curves, and the incremental area under the glucose curve (iAUC~glucose~) was calculated by the trapezoidal rule. Paired sample *t*-test was used to compare differences in PBGL and subjective satiety between the breakfasts, and independent sample *t*-test was used to compare differences in postprandial iAUC~glucose~ between the breakfasts as two subjects only provided data for one breakfast. Their data were included in the analysis because the results did not differ significantly when they were excluded. The time point with the highest mean blood glucose level was deemed as the time of peak PBGL for the overall analysis ([Figure 1](#nutrients-05-00001-f001){ref-type="fig"}), and the peak PBGL time for individual subjects were also identified ([Figure 2](#nutrients-05-00001-f002){ref-type="fig"}).
{#nutrients-05-00001-f001}
{#nutrients-05-00001-f002}
2.7. Ethics Approval
--------------------
This study was conducted according to the guidelines laid down in the Declaration of Helsinki and all procedures involving human subjects/patients were approved by the Human Research Ethics Committee of the Sydney South West Area Health Service (RPAH Zone). Written, informed consent was obtained from all subjects in this study.
3. Results
==========
The study sample included five South Asians (Indian, Nepalese and Bangladeshi), four Chinese, and one Caucasian. The subjects attended the first and second session at a mean ± SEM of 33.5 ± 0.5 and 35.1 ± 0.6 weeks of gestation respectively. There was no difference in BMI (first visit: 24.9 ± 0.4 kg/m^2^*vs.* second visit: 26.1 ± 0.9 kg/m^2^; *p* = 0.212) and mean 7-day fasting BGL (first visit: 4.8 ± 0.1 mmol/L *vs.* second visit: 4.8 ± 0.1 mmol/L; *p* = 0.461) on the test days. No subject started insulin therapy during the study period.
Overall the consumption of a high GI bread-based breakfast resulted in higher postprandial glycemia ([Figure 1](#nutrients-05-00001-f001){ref-type="fig"}). The mean ± SEM iAUC~glucose~ after a low GI breakfast was significantly lower than that after the consumption of a high GI breakfast (212.7 ± 22.9 *vs.* 340.8 ± 23.4; *p* = 0.001). The mean ± SEM peak BGL was 6.7 ± 0.3 mmol/L for the low GI breakfast and 8.6 ± 0.3 mmol/L for the high GI breakfast (*p* \< 0.001). However, there was large inter-subject variability in the timing of the peak between the two test meals ([Figure 2](#nutrients-05-00001-f002){ref-type="fig"}): the peak occurred between 45 and 75 min for the low GI breakfast (mean ± SEM: 60.0 ± 4.0 min), and between 30 and 60 min for the high GI breakfast (mean ± SEM: 43.1 ± 3.4 min; *p* = 0.015). In the eight subjects who provided data for both breakfasts, six had a delayed peak PBGL time after consuming the low GI breakfast when compared to that of the high GI breakfast. There was no significant difference in subjective satiety throughout the 2-h test period ([Figure 3](#nutrients-05-00001-f003){ref-type="fig"}).
{#nutrients-05-00001-f003}
4. Discussion
=============
This study is the first attempt to examine the timing of the peak PBGL in women with GDM after consumption of breakfasts of different GI values. We expected to see differences in the peak concentration but similar timing of the peak. However we found that the low GI breakfast produced a peak at closer to 60 min after the start of the meal compared with \~45 min after the high GI breakfast.
It has been previously shown that a low GI diet produces comparable outcomes as a conventional high fibre diet in pregnancy complicated with GDM \[[@B11-nutrients-05-00001]\]. The limited evidence on the efficacy of a low GI diet for the management of GDM has suggested that a low GI diet may improve postprandial glycemia \[[@B10-nutrients-05-00001],[@B16-nutrients-05-00001]\], therefore reducing excessive transfer of maternal blood glucose to the fetus. In GDM, breakfast was shown to have the greatest variability in postprandial glycemic response \[[@B17-nutrients-05-00001]\]. The results from our study suggested that a low GI breakfast may be of benefit in the management of post-breakfast glycemia.
Since peak PBGL has been shown to be more strongly associated with outcomes of GDM pregnancy \[[@B18-nutrients-05-00001]\] than 2-h PBGL, it is important to accurately time the postprandial SBGM testing to capture the peak BGL. In healthy, non-pregnant subjects low and high GI meals reach peak concentrations at the same time, *i.e.*, at 30 min \[[@B12-nutrients-05-00001]\]. However during pregnancy, especially those complicated with GDM, changes in gastric motility \[[@B13-nutrients-05-00001]\] and insulin sensitivity \[[@B14-nutrients-05-00001]\] can be expected to alter the rate of carbohydrate digestion and absorption, and hence shape of the postprandial glucose curve. Indeed, Wolever *et al.* \[[@B19-nutrients-05-00001]\] showed that the peak PBGL after a standard test meal occurred later and later as glucose tolerance worsened in subjects with diabetes.
A previous study had found that the peak PBGL in GDM occurs at about 60 min post meal \[[@B7-nutrients-05-00001]\]. Therefore SBGM that tests the 2 h PBGL may miss the glucose peak, and the clinical decision to commence insulin therapy was usually based on a cut-off of 1-h PBGL \[[@B6-nutrients-05-00001]\]. We found that while a low GI breakfast delayed the peak PBGL in GDM, the peak occurs closer to 60 min after the start of the meal than a high GI breakfast, which produced a peak at \~45 min post meal. Therefore our results suggested the findings of the study by Moses *et al.*\[[@B10-nutrients-05-00001]\], which demonstrated that a low GI diet reduces the need for insulin in GDM for BGL management, was indeed due to the fact that low GI meals produce a lower peak PBGL at 60 min post meal. Our results also suggest that the 1-h postprandial SBGM reading is likely to underestimate the actual postprandial glycemic response to a high GI meal. Therefore our finding raises the question whether the GI of the patient's diet should be considered in the interpretation of SBGM results.
Our study has limitations, including a small sample size and a high proportion of Asian subjects, which compromises the generalizability of the findings. In addition, the non-continuous nature of blood sample collection did not allow accurate determination of the actual peak time. However, more frequent fingerprick blood sampling would be impractical. Although continuous glucose monitoring might be helpful in this context, it measures changes in the interstitial fluid rather than capillary blood, and by nature is subject to a delay \[[@B20-nutrients-05-00001],[@B21-nutrients-05-00001],[@B22-nutrients-05-00001]\].
5. Conclusions
==============
The low GI breakfast produces lower postprandial glycemia than a macronutrient matched high GI breakfast. The timing of the peak BGL varies within and between breakfasts of different GI. The peak PBGL after a high GI breakfast occurs at \~45 min, 15 min earlier than that of a low GI breakfast. The peak PBGL of a low GI breakfast occurs closer to the time recommended for PBGL monitoring in GDM. Since many women consume high GI meals throughout pregnancy, there are implications for clinical practice.
This study was funded by internal revenue. The authors would like to thank the subjects for their participation.
JCBM is a co-author of The New Glucose Revolution book series (Hodder and Stoughton, London, UK; Marlowe and Co., New York, NY, USA; Hodder Headline, Sydney; and elsewhere), is the director of a not-for-profit GI-based food endorsement program in Australia, and manages the University of Sydney GI testing service. All other authors declare they have no conflict of interest.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Central venous catheterization plays an important role in modern medical practice.
It is estimated that approximately 8% of hospitalized patients require central venous access during the course of their hospital stay, and it has been estimated that more than five million CVCs are inserted in patients in the United States each year \[[@B1], [@B2]\]. Indications for CVC placement are diverse. Some of the more common indications include invasive hemodynamic monitoring, parenteral nutrition support, dialysis, chemotherapy, fluid resuscitation, drug administration, and renal replacement therapy.
Although CV catheterization is a simple and relatively safe procedure, many complications have been reported during or after the procedure. Malposition is one of the complications observed.
In this report, we describe a rare case of malposition in a 10-year-old girl.
2. Case Presentation {#sec2}
====================
The patient was a 10-year-old girl, recently diagnosed with leukemia and hospitalized for treatment. She required central venous (CV) line placement for chemotherapy. Other than her diagnosed leukemia, she had no other significant medical history.
After the procedure was explained thoroughly to the patient and her parents, a consent form was completed by her parents, and she was transferred to the CV line room. The patient was very alert and cooperative. Based on the local protocols, all patients requiring CV line placement are transferred to a room dedicated for CV line placement (the intravenous access room), which is located in the operating room. Standard monitoring including an electrocardiogram (ECG), noninvasive blood pressure, and pulse-oximetry were initiated. A 20-gauge cannula was inserted into the vein on the dorsum of the patient\'s left hand.
The right internal jugular vein was selected for CV cannulation.
Propofol was used for sedation, and after adequate sedation, a single-lumen 14-gauge catheter was inserted in the right internal jugular vein using ultrasound sonography under sterile conditions by an experienced anesthesiologist. No problem was encountered during the procedure, and after blood was aspirated, the catheter was fixed at 13 cm. Normal saline infusion was initiated through the CV. Central venous waveforms were not used for catheter position confirmation.
A chest radiograph was immediately arranged to confirm the catheter position.
On chest radiograph, the catheter could be seen looping back and going upward at the junction of the right internal jugular vein and the right subclavian vein ([Figure 1](#fig1){ref-type="fig"}). The team decided to use ultrasonography to verify that the catheter was in the jugular vein and had not punctured the dorsal wall of the vein. During scanning of the right internal jugular vein, only a single lumen of the catheter could be seen until we scanned the bottom third of the jugular vein ([Figure 2](#fig2){ref-type="fig"}).
We decided to pull back the catheter under the guidance of ultrasonography until only one lumen could be visualized and then pass a guidewire over the catheter to reposition the catheter. While doing so and after only one lumen was visualized following extraction of the central venous line under the guidance of ultrasonography, passage of the guidewire was attempted, but resistance was encountered. Therefore, we decided to completely remove the catheter and reinsert another catheter.
When the catheter was removed, it was observed that the catheter had bent at the distal end ([Figure 3](#fig3){ref-type="fig"}).
Subsequently, we placed another CVC through the internal jugular vein without any complications. The position of the CVC was confirmed by radiograph.
3. Discussion {#sec3}
=============
CV catheterization is being performed every day in medical centers around the world. The internal jugular vein is one of the most common sites that anesthesiologists use, and this site is chosen because CVC can be securely inserted in this location.
It is difficult to estimate the rate of early and late complications that occur during insertion of CV lines. Many of the complications may go unseen, and many are unreported. Some complications may be life threatening or may cause morbidity, and some may not be recognized as a complication at all \[[@B3]\]. The incidence and occurrence of complications depend on various factors, such as the experience of the operator, the site of insertion, and the placement technique \[[@B4]\].
At our center, Shariati Hospital, we insert over 2000 CV lines every year. The Hematology-Oncology Research Center and Stem Cell Transplantation (HORCSCT) Center, which is affiliated to Tehran University of Medical Sciences (TUMS), is based in Shariati Hospital, and over two-thirds of our patients are individuals who have hematological cancer and require CV line placement for chemotherapy.
The low price of CV lines compared to other options for CV access has favored their use at our center. However, using CV lines for chemotherapy has its own hazards, especially in the pediatric population.
We use single-lumen 14-gauge CV lines for all our patients (aged 2 months and above), and the main route of insertion and technique of insertion for adults at our center is the subclavian vein via the landmark technique. Regarding the route and technique of insertion, we have found that patients are much more satisfied and can handle their daily tasks much easier when the catheter is fixed at a level below the clavicle.
In the pediatric patients, we usually use an ultrasound-guided approach to the right internal jugular vein, as we are inserting a very large catheter and the risk of complications is much higher via the subclavian vein approach.
In the case described here, the procedure went smoothly with no resistance, and the blood was successfully aspirated at the end of the first attempt at placement, which led us to refrain from scanning the catheter placement with ultrasound. This example shows that although insufficient blood flow from the catheter during aspiration is a possible warning sign of misplacement of the CV line, adequate blood aspiration cannot be totally relied on as a sign of successful placement.
Ultrasonography has aided in the placement of CVCs in many ways, especially in pediatric patients. However, as can be seen in this case, relying on ultrasound only for finding the vein and guiding the needle is not enough and did not prevent the malposition of the catheter. Based on this report, it is advised to scan along the vein to localize the catheter, even after a seemingly uneventful catheter placement. This is especially important when placing catheters via the subclavian vein, which is a site where a catheter risks going upward into the internal jugular vein.
Chest radiographs have always helped us determine early and late complications. The immediate chest radiograph that was taken in this patient proved to be vital in showing the complication.
Although we have had different types of malposition of catheters, this is the first time that we encounter a catheter looping back on itself. A probable mechanism could have been the angle of the internal jugular vein and subclavian vein, facilitating the malposition and pushing the j-shaped tip of the guidewire upward and back on itself. The right-sided bevel of the needle at the time of internal jugular vein puncture and guidewire insertion may also have contributed to the looping back of the guidewire and, consequently, the catheter.
Another issue that needs to be highlighted is the need for a specific facility for intravenous access procedures. Due to the large number of CVCs placed at our center, a few years ago we decided to dedicate a room with trained personnel for CV placement. Having a dedicated room for this purpose has not only decreased the time spent during each procedure but also decreased the rate of complications and increased patient satisfaction.
A variety of rare complications such as perforation of the left brachiocephalic vein and massive hemothorax, chylothorax, internal mammary artery malposition of catheter, and inadvertent placement of a CVC in the left pericardiophrenic vein have been reported previously \[[@B5]--[@B8]\].
It should be noted that a bending catheter has the risk of occlusion or perforation of the vein which fortunately did not occur in the described report above.
Many practical techniques such as using surface landmarks for estimating the length of catheter insertion, ultrasound-guided localization of the vein and guidance of the needle, echocardiography, electrocardiographic guided catheter tip placement using NaHCO3-filled catheters, and immediate postprocedure X-rays have been proposed for aiding a safe placement of CVC \[[@B9], [@B10]\].
We believe that to decrease the rate of complications associated with CVC placement, a multimodal approach is required. An appropriate setting with trained personnel, in combination with ultrasound guidance during and after the procedure, is helpful but not enough. Making sure the operator is focused throughout the procedure with attention to any atypical events such as resistance during any of the stages of catheter placement may help to decrease complications.
CV line placement should always be looked on as a procedure that could become complicated, even in the hands of the most experienced operators.
Therefore, it should be remembered that follow-up and checking of the correct function and placement of the CVC are as important as the procedure itself.
Conflicts of Interest
=====================
The authors declare no conflicts of interest.
{#fig1}
{#fig2}
{#fig3}
[^1]: Academic Editor: Kuang-I Cheng
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Cyanobacteria are photosynthetic prokaryotic organisms comprising the major biomass of living organisms in earth oceans, and are believed to be related to the ancestor of higher plant chloroplasts. They share characteristics from bacteria and chloroplasts regarding mRNA degradation. *Synechocystis* PCC6803 is a model organism for cyanobacteria, but not much is known about the mechanism of RNA degradation in this class of organisms.
The *Synechocystis* genome contains genes that have a high homology to RNase E, RNase J, PNPase, RNase II/R and nucleotidyltransferase/poly(A)-polymerase (PAP), the proteins involved in mRNA degradation [@pone.0032690-Rott1]. The polyadenylation pathway was already studied and it was shown that the product of the putative PAP gene has nucleotidyltransferase and not polyadenylation activity. Instead, the reaction of polyadenylation in *Synechocystis* is performed by PNPase, which generates heterogeneous poly(A)-rich tails [@pone.0032690-Rott1]. In the chloroplast of higher plants, PAP together with PNPase contributes to the polyadenylation activity [@pone.0032690-YehudaiResheff1], [@pone.0032690-Lisitsky1]. In *Synechocystis* the absence of PNPase is lethal for the cell [@pone.0032690-Rott1]. The same behaviour was observed by disrupting the genes for RNase II/R or RNase E [@pone.0032690-Rott1]. In *Synechocystis*, the RNase E homologue is more related to RNase G and it is not part of a multicomponent complex, unlike RNase E of *Escherichia coli* [@pone.0032690-Rott1]. However, *in vivo* assays showed that RNase E homologue from *Synechocystis* is able to complement the functions of both RNase E and RNase G in *E. coli*. Moreover, its endonucleolytic activity was confirmed *in vitro*, showing that cleavage is dependent on the primary target sequence and on the secondary structure of the mRNA [@pone.0032690-Horie1]. In addition to RNase E, the *Synechocystis* genome contains an RNase J homologue, a ribonuclease that has both endonucleolytic and 5′ to 3′ exonucleolytic activity and was originally described in *Bacillus subtilis* [@pone.0032690-Even1], [@pone.0032690-Mathy1].
Analysis of the *Synechocystis sp*. genome has revealed a single member of the RNase II-family of proteins, from now on named synRNB. Proteins from this family are present in all domains of life, are involved in several processes, and play a central role in the mechanism of gene expression [@pone.0032690-Arraiano1]. In eukaryotes, RNase II homologues are part of a multiprotein complex, called the exosome, where they provide the catalytic subunit [@pone.0032690-Dziembowski1]. Mutants in these homologues have shown defects in development, mitotic control and chloroplast biogenesis [@pone.0032690-Arraiano2]. In prokaryotes, they are important for growth and stress responses and they are also involved in virulence [@pone.0032690-Arraiano1], [@pone.0032690-Arraiano2]. RNase II is the prototype of this family of enzymes, which also comprises RNase R in *E. coli*. They both act hydrolytically, degrading RNA molecules in the 3′ to 5′ direction, releasing 5′-nucleoside monophosphates [@pone.0032690-Cannistraro1]. The resolution of RNase II crystal structure showed that it is composed of two N-terminal Cold Shock Domains (CSD) domains and a C-terminal S1 domain involved in RNA binding [@pone.0032690-Frazo1], [@pone.0032690-Amblar1]. The central RNB domain is responsible for the catalytic activity of the protein [@pone.0032690-Frazo1]--[@pone.0032690-Matos1]. This domain organization is present in all family members [@pone.0032690-Frazo1], [@pone.0032690-Mian1]. In the RNase II catalytic cavity, the first five nucleotides of the RNA molecule (counting from the 3′ end) are stacked between the two aromatic residues, Tyr253 and Phe358. After the last cleavage event, the 4 nt fragment is no longer "clamped" by those residues, and then the 4 nt is the end-product released by RNase II [@pone.0032690-Frazo1]. Mutational analysis showed that the Tyr253 is crucial for setting the final end product, not only in RNase II [@pone.0032690-Barbas1] but also in RNase R [@pone.0032690-Matos1], indicating that its role may be conserved in all members of this family of enzymes. The active site is formed by four highly conserved aspartates (Asp201, Asp207, Asp209 and Asp210) and an arginine (Arg500), which were shown to be important for the activity of the protein and proper RNA binding/orientation [@pone.0032690-Barbas1]--[@pone.0032690-Barbas2]. From these residues, Asp209 was the most critical for the activity of RNase II, since its substitution for an asparagine totally abolished its activity without affecting the RNA binding ability [@pone.0032690-Amblar2]. The same results were obtained for other members of this family [@pone.0032690-Dziembowski1], [@pone.0032690-Matos1], [@pone.0032690-Awano1], [@pone.0032690-Charpentier1]. It was postulated that Glu542 was involved in nucleotide elimination. However, when Glu542 was changed into an alanine, the resultant protein had more affinity for RNA and it was more than 100 times more active than the wild type protein. Therefore, the E542A mutant was called a "super-enzyme" [@pone.0032690-Barbas2].
*E. coli* RNase II and RNase R belong to the same family, but act differently on RNA substrates and have different specificities. RNase R was shown to have an intrinsic ability to unwind double-stranded RNAs [@pone.0032690-Awano1], [@pone.0032690-Matos2], while RNase II is sensitive to these structures and stalls a few nucleotides before reaching the double-stranded region [@pone.0032690-Arraiano1]. RNase II is responsible for 90% of the hydrolytic activity in *E. coli* [@pone.0032690-Deutscher1]; however, in stationary phase and stress conditions (like cold-shock induction), RNase R levels increase in the cell [@pone.0032690-Cairro1]. This can be related to the crucial role that RNase R has in RNA quality control, namely in the degradation of defective tRNA and rRNA [@pone.0032690-Cheng1], and also in protein quality control [@pone.0032690-Cairro1]. In contrast, RNase II is mainly involved in the terminal stages of mRNA degradation and its activity can be replaced by PNPase. It was also suggested that the main function of RNase II is the protection of some mRNA transcripts by rapidly removing the short poly(A) tail [@pone.0032690-Mohanty1].
In cyanobacteria nothing is known about this family of enzymes, except that SynRNB protein is essential for the viability of these organisms [@pone.0032690-Rott1]. This indicates that RNase II/R homologue may have a crucial role in *Synechocystis* metabolism. The aim of this work was to characterize the activity of synRNB as a first approach for the study of the role of this enzyme in RNA metabolism in cyanobacteria. The results showed that the *Synechocystis* protein behaves like an RNase II-like protein: the final product released is a 4 nt fragment and the protein is not able to degrade structured substrates. This is the first case reported where the only member of the RNase II family of enzymes present in an organism behaves like an RNase II and not like RNase R. However, while RNase II prefers polyadenylated substrates [@pone.0032690-Barbas1], [@pone.0032690-Barbas2], RNase II from *Synechocystis* does not demonstrate such preference. This may happen because in this genus there is no PAP and the tails produced by PNPase are heterogeneous [@pone.0032690-Rott1].
Results and Discussion {#s2}
======================
RNB protein from *Synechocystis sp.* PCC6803 is a hydrolytic exoribonuclease {#s2a}
----------------------------------------------------------------------------
SynRNB, purified by immobilised metal affinity chromatography, was incubated with petD3 RNA to test the activity of the protein after purification. After only 5 minutes of incubation we were already able to see the formation of degradation products, which increased in abundance with time ([Figure 1A](#pone-0032690-g001){ref-type="fig"}). To confirm that this activity was due to RNB from *Synechocystis* and did not result from a contamination with the *E. coli* PNPase, we analysed the reaction products by TLC (Thin Layer Chromatography) ([Figure 1B](#pone-0032690-g001){ref-type="fig"}) [@pone.0032690-Portnoy1]. As shown in [figure 1B](#pone-0032690-g001){ref-type="fig"}, the products detected were exclusively nucleotide monophosphates. In this case, since the RNA was \[^32^P\] UTP labelled, the obtained radioactive product co-migrated with the UMP marker. With the results obtained we were able to confirm that the recombinant RNB from *Synechocystis* has hydrolytic activity.
{#pone-0032690-g001}
Salt and pH preference of the synRNB protein {#s2b}
--------------------------------------------
In order to characterize the activity of synRNB protein we first determined the optimal conditions for the catalysis. We empirically determined the effects of changing the salt, its concentration and the pH of the reaction buffer.
To assess the effect of salt type and concentration, we performed activity assays with 10, 50, 100, 150 and 200 mM of NaCl and KCl as described in the [Materials and Methods](#s3){ref-type="sec"}. The activity of synRNB was measured by quantifying the disappearance of the substrate with time. The conditions used for this and the following determinations were adjusted such that less than 25% of the substrate was degraded. The results obtained are presented in [figure 2](#pone-0032690-g002){ref-type="fig"}. It should be noted that the gels showed in this figure and the ones that follow were not used to quantify the degradation of the substrate. In the ones that are shown, the reactions have been allowed to proceed further to enable detection of the final products. The assays quantified were performed in different conditions (lower protein concentration and reaction time) to ensure that less than 25% of the substrate was degraded. When the salt used was NaCl, the protein seemed inactive for all the concentrations tested, while in the presence of KCl we were able to detect activity with all the concentrations used, although the proteins preferred 50 mM KCl ([Figure 2A](#pone-0032690-g002){ref-type="fig"}). However, following the reactions ate different time points, we were able to see that, in fact, synRNB was active in NaCl, but the cleavage efficiency was 10 times lower when compared to its activity in KCl ([Figure 2B](#pone-0032690-g002){ref-type="fig"}). We were also able to confirm that the preferred KCl concentration was 50 mM. For higher salt concentrations, the activity of synRNB starts to decrease ([Figure 2B](#pone-0032690-g002){ref-type="fig"}). Taking these results into consideration, we used an activity buffer with a salt concentration of 50 mM of KCl for the remaining experiments.
{#pone-0032690-g002}
After establishing the optimal salt concentration for the activity of the protein, we analysed the effect of pH on catalysis. For this purpose, we tested the activity of synRNB using pH ranging from 6.5 to 9 in the presence of 50 mM of KCl. The results obtained showed that the activity of synRNB peaked at a pH of 8.0 ([Figure 3](#pone-0032690-g003){ref-type="fig"}). However, the peak is relatively broad and the activity at pH 7.5 and 8.5 was not substantially lower ([Figure 3B](#pone-0032690-g003){ref-type="fig"}).
{#pone-0032690-g003}
RNB protein from *Synechocystis sp.* PCC6803 prefers Mg^2+^ for its activity {#s2c}
----------------------------------------------------------------------------
Exoribonucleases from the RNase II-family of enzymes need a divalent ion in order to proceed with catalysis. For *E. coli* RNase II and RNase R, the presence of Mg^2+^ is crucial for the activity of the proteins [@pone.0032690-Frazo1], however, catalysis can also occur in the presence of other divalent ions (Matos et al, manuscript submitted). We tested the activity of the *Synechocystis* homologue with different divalent metal ions: Mg^2+^, Mn^2+^, Ca^2+^, Zn^2+^, Cu^2+^, Co^2+^, and Ni^2+^. As shown in [figure 4A](#pone-0032690-g004){ref-type="fig"}, incubating the RNA with 10 nM of protein during 5 minutes we can only detect strong activity in the presence of Mg^2+^. In the presence of Ca^2+^, it seems that the protein has some residual activity. To confirm this, we increased the time of the reaction, and saw that, after one hour of incubation in a buffer with Ca^2+^, the protein was also able to cleave the substrate ([Figure 4A](#pone-0032690-g004){ref-type="fig"}). Under the same conditions no activity was detected for any of the other divalent ions that were tested ([Figure 4A](#pone-0032690-g004){ref-type="fig"}). From these experiments we conclude that the synRNB needs an Mg^2+^ ion for the catalysis, but is also able to cleave RNA in the presence of Ca^2+^, although with less efficiency. We then tested the effect of changing the Mg^2+^ concentrations (1, 2.5, 5 and 10 mM). As shown in [figure 4B and C](#pone-0032690-g004){ref-type="fig"}, we found that synRNB is most active in the presence of 1 mM of Mg^2+^. At higher Mg^2+^ concentrations, the activity of synRNB decreases ([Figure 4B and C](#pone-0032690-g004){ref-type="fig"}).
{#pone-0032690-g004}
For the following experiments, we used reaction conditions that were optimised as described above; 50 mM KCl and 1 mM MgCl~2~ in 20 mM Tris-HCl buffer with a pH of 8.0. To prevent degradation of substrates during SPR experiments, Mg^2+^ was omitted and EDTA added to the buffer.
RNB protein from *Synechocystis sp.* PCC6803 displays RNase II-like properties {#s2d}
------------------------------------------------------------------------------
RNase II and RNase R of *E. coli* differ with regard to the final product released and their ability to degrade double-stranded RNAs [@pone.0032690-Matos1], [@pone.0032690-Barbas2]. While the *E. coli* RNase II releases an end-product of 4 nt and is sensitive to secondary structures, stalling 5 to 7 nt before it reaches the double-stranded region, RNase R degrades RNA releasing fragments of 2 nt of length and is able to overcome structured RNAs [@pone.0032690-Matos1], [@pone.0032690-Barbas2]. We tested how the *Synechocystis* protein behaved regarding the degradation of two single-stranded substrates, poly(A) and 16 ss, and also using the double-stranded substrate 16--30 ds ([Figure 5](#pone-0032690-g005){ref-type="fig"}). In order to compare synRNB with *E. coli* RNase II and RNase R, the three proteins were assayed at the same time. The activity of RNase II and RNase R was assayed using the conditions described previously [@pone.0032690-Amblar2], [@pone.0032690-Amblar3], [@pone.0032690-Arraiano3]. For the *Synechocystis* protein, the conditions used were those described above. The results obtained are represented in [Figure 5](#pone-0032690-g005){ref-type="fig"}. Different conditions were used for the three proteins (salt and/or Mg^2+^ concentrations) to ensure that all were acting in their optimal conditions.
{#pone-0032690-g005}
It was already shown that *E. coli* RNase II and RNase R prefer poly(A) substrates [@pone.0032690-Matos1], [@pone.0032690-Barbas2]. For that reason, this was one of the substrates that we used to test the activity of the *Synechocystis* protein from this family. We also decided to use the 16-mer RNA, which had a mixed composition of all ribonucleotides, in order to analyse if, like its *E. coli* homologues, synRNB had some preference for poly(A) substrates. As already described, and as observed for both single-stranded substrates tested in this work, RNase II is able to degrade ssRNA substrates releasing a 4 nt fragment, while RNase R is able to proceed with catalysis until it reaches the 2 nt of length ([Figure 5](#pone-0032690-g005){ref-type="fig"}) [@pone.0032690-Arraiano1]. When we tested the recombinant protein from *Synechocystis* with the poly(A) substrate, we were able to see that, with 50 nM of protein, the final product released was a 5 nt fragment ([Figure 5](#pone-0032690-g005){ref-type="fig"}). However, when we used a higher protein concentration, 250 nM, the protein was now able to release a 4 nt end-product ([Figure 5](#pone-0032690-g005){ref-type="fig"}). We also assayed this substrate with higher protein concentrations but no more cleavage events were observed (data not shown), confirming that the final product released is 4 nt, as produced by *E. coli* RNase II. However, 50 nM of the *Synechocystis* protein were sufficient to degrade the 16 ss substrate until the 4 nt of length. Together, these results show that by leaving a 4 nt degradation fragment, the *Synechocystis* enzymes reacts like the RNase II of *E. coli*
We then asked if the synRNB degrades poly(A) better than randomized sequenced RNA. To this end, we compared the activity of the *Synechocystis* protein with poly(A) and 16 ss substrates. The results showed that the protein had similar activities for both substrates ([Figure 6](#pone-0032690-g006){ref-type="fig"}). This indicates that, in contrast to the *E. coli* RNase II, which prefers poly(A) substrates [@pone.0032690-Barbas1], [@pone.0032690-Barbas2], the synRNB does not have such preference. Interestingly, while in *E. coli* the polyadenylation is performed mainly by PAP (which generates homopolymeric tails), and to some extent also by PNPase, in cyanobacteria those tails are exclusively synthesized by PNPase and are heteropolymeric [@pone.0032690-Rott1]. Therefore, the substrate preference of members of the RNase II family may reflect the composition of 3′ tails in their organism of origin. To further examine the preference for 3′ poly(A) tails, we performed a competitive degradation assay. In this experiment, *petD3* RNA was incubated with synRNB in the presence of poly(N) oligomers. The aim was to verify if the addition of unlabelled poly(N) oligomers would compete for *petD3* degradation. The results confirmed that, in fact, poly(A) is not a preferred substrate for synRNB, since the inhibition of the degradation by both poly(A) and poly(U) was similar. However, poly(G) was able to strongly inhibit the degradation activity ([Figure 7](#pone-0032690-g007){ref-type="fig"}). These results confirmed that synRNB does not prefer poly(A) substrates and had higher preference for poly(G), in contrast to what was shown for the *E. coli* counterparts [@pone.0032690-Barbas1], [@pone.0032690-Barbas2]. In fact, for *E. coli* RNase R, poly(G) was practically inactive as substrate [@pone.0032690-Cheng2]. Also, the activity of both *E. coli* RNase II and RNase R is higher for poly(A), then poly(U) and finally poly(C) [@pone.0032690-Cheng2], [@pone.0032690-Cannistraro2].
{#pone-0032690-g006}
{#pone-0032690-g007}
To verify that the binding affinity of the enzyme to poly(A) and other RNA is similar, we determined the dissociation constants by SPR using different single-stranded substrates. One of the substrates was constituted only by adenosines and the second with a random sequence. The results presented in [Table 1](#pone-0032690-t001){ref-type="table"} showed that the *K* ~D~ values obtained for synRNase II with both single-stranded substrates are similar (3.9±0.3 nM for the poly(A) and 3.3±0.5 nM for the 25 ss), and the same was observed regarding the association (*k* ~a~) and dissociation (*k* ~d~) rates. In contrast, the *K* ~D~ values previously obtained for *E. coli* RNase II and RNase R showed that these proteins have an increased affinity for poly(A) substrates when compared to a random sequence [@pone.0032690-Matos1], [@pone.0032690-Barbas1], [@pone.0032690-Barbas2].
10.1371/journal.pone.0032690.t001
###### Kinetic parameters of RNase II-like protein from *Synechocystis sp*.
{#pone-0032690-t001-1}
k~a~ (1/Ms) k~d~ (1/s) K~D~ (nM)
--------------- ------------- -------------- -----------
**PolyA** 3,0±0,4 E03 1,9±0,1 E-05 3,9±0,3
**25 ss** 3,7±0,4 E03 1,1±0,1 E-05 3,3±0,5
**16--25 ds** 2,4±0,2 E05 1,4±0,2 E-04 2,1±0,1
The kinetic constants were determined by Surface Plasmon Resonance using Biacore 2000 with a 25-nt RNA oligomer (5′-Biotin-CCCGACACCAACCACUAAAAAAAAA-3′) and 30 nts PolyA as substrates.
The activity results presented above were obtained at 37°C, which is the optimal activity for the *E. coli* enzymes. However, the *Synechocystis* PCC6803 cyanobacteria lives in freshwater and its optimal temperature for growth is around the 30°C. For this reason, we compared the activity of synRNB with the synthetic oligomers at 30°C and 37°C. As shown in [Figure 6](#pone-0032690-g006){ref-type="fig"}, the specific activity of the protein is very similar at both temperatures with all substrates tested. However, the degradation of the poly(A) substrate is different at 30°C ([Figure S1](#pone.0032690.s001){ref-type="supplementary-material"}). While at 37°C we needed 250 nM of protein to reach the end-product of 4 nt, at 30°C we were able to detect the 4 nt fragment with 50 nM of protein ([Figure 5](#pone-0032690-g005){ref-type="fig"} and [S1](#pone.0032690.s001){ref-type="supplementary-material"}). Moreover, while at 37°C we can observe the presence of high amounts of the intermediary fragment of 6 nt in the degradation of the 16-mer substrate, similarly to what is observed for RNase II ([Figure 5](#pone-0032690-g005){ref-type="fig"}), in the same conditions but at 30°C the 6 nt fragment is no longer visible after 20 min of reaction ([Figure S1](#pone.0032690.s001){ref-type="supplementary-material"}). These results indicate that the synRNB may have a higher affinity for small fragments at 30°C, which can explain why it is able to reach the final product more rapidly when compared to the activity at 37°C.
We also tested the activity of the protein against the synthetic double-stranded substrate, 16--30 ds. *E. coli* RNase II is sensitive to secondary structures, and is not able to degrade this RNA, stalling 7 nt before it reaches the double-stranded region, releasing a 23 nt fragment ([Figure 5](#pone-0032690-g005){ref-type="fig"}) [@pone.0032690-Cannistraro3], [@pone.0032690-Spickler1]. In contrast, *E. coli* RNase R is able to overcome the secondary structures, releasing the typical 2 nt fragments ([Figure 5](#pone-0032690-g005){ref-type="fig"}) [@pone.0032690-Cheng2]. When we tested the synRNB we could observe that it was not able to overcome secondary structures. It stalled 4 nt before it reached the double-stranded region, releasing a fragment of 20 nt, which is shorter when compared to the 23 nt fragment released by *E. coli* RNase II ([Figure 5](#pone-0032690-g005){ref-type="fig"}). The resolution of the crystal structure from *E. coli* RNase II showed us that the catalytic cavity of the protein is only accessible to single stranded RNA due to the steric hindrance at its entrance caused by the RNA-binding domains [@pone.0032690-Frazo1]. The synRNB showed to be able to move closer to the double-stranded junction when compared to the *E. coli* protein, since that the final product released is shorter (20 nt vs. 23 nt, respectively). This may indicate that the RNA-binding domains may have a different rearrangement in this protein. In order to address this question, we modelled synRNB and compared it with *E. coli* RNase II 3D structure ([Figure 8](#pone-0032690-g008){ref-type="fig"}). Both proteins seemed to have a similar overall structure arrangement, with the important residues for catalysis located in the same spatial position ([Figure 8D](#pone-0032690-g008){ref-type="fig"}). The active site of RNase II is composed by four highly conserved aspartates and an arginine, which are important for catalysis [@pone.0032690-Matos1], [@pone.0032690-Barbas1], [@pone.0032690-Barbas2]. Tyrosine is a residue responsible for setting the final product in the RNases from this family [@pone.0032690-Matos1], [@pone.0032690-Barbas1], [@pone.0032690-Barbas2] ([Figure 8D](#pone-0032690-g008){ref-type="fig"}). In synRNB, these residues can also be found and are located in an equivalent position ([Figure 8D](#pone-0032690-g008){ref-type="fig"}). If we look closer to the RNA binding domains, it is possible to see that the CSD1 of synRNB ([Figure 8A](#pone-0032690-g008){ref-type="fig"}) is quite different from the one from *E. coli* RNase II ([Figure 8B](#pone-0032690-g008){ref-type="fig"}). When we superposed both structures, that difference is more noticeable ([Figure 8C](#pone-0032690-g008){ref-type="fig"}). Moreover, the superposition of both structures also showed that the S1 domain from SynRNB (red) lacks at least two β-sheets when compared to the one from RNase II (green) ([Figure 8C](#pone-0032690-g008){ref-type="fig"}). Therefore, the CSD1 from *Synechocystis* protein is more distant from the S1 domain, which could result in a wider anchoring region which in turn might allow the substrate to move nearer to the catalytic cavity, explaining why this protein is able to get closer to the double-stranded junction ([Figure 8C](#pone-0032690-g008){ref-type="fig"}). The activity of this protein was also determined with the 16--30 ds at both temperatures, 30°C and 37°C. Similarly to what was observed to the single-stranded substrates, the activity of the protein for the 16--30 ds is the same as observed for the other substrates and does not change with temperature ([Figure 6](#pone-0032690-g006){ref-type="fig"}). At 30°C, the degradation pattern of the protein remained unaltered, and the protein was still not able to overcome secondary structures, releasing a final product of 20 nt similarly to what was observed at 37°C ([Figures 5](#pone-0032690-g005){ref-type="fig"} and [S1](#pone.0032690.s001){ref-type="supplementary-material"}). It is known that, at lower temperatures, the RNA molecules form more stable secondary structures. In *Synechocystis* PCC6803, low temperatures highly induce the expression of an RNA helicase, CrhR [@pone.0032690-Suzuki1], which may be involved in the degradation of the transcripts at these temperatures by helping to unwind the secondary structures. In a strain defective in this helicase, the PNPase levels are increased up to ∼2-fold. This would help to eliminate the transcripts with cold-induced excessive secondary structures [@pone.0032690-Rowland1]. No changes were observed for synRNB protein. Together with the results described here, these findings indicate that this protein may not be involved in the degradation of double-stranded substrates at environmental temperatures. Moreover, when we determined the dissociation constants for this substrate, the value is very similar to the ones obtained for the single-stranded substrates ([Table 1](#pone-0032690-t001){ref-type="table"}). However, the protein associates and dissociates more rapidly to the double-stranded substrate when compared to the other two ([Table 1](#pone-0032690-t001){ref-type="table"}, *k* ~a~ and *k* ~d~ values).
{#pone-0032690-g008}
Considering that we were using a synthetic substrate, we also decided to test the activity of the recombinant protein using an mRNA and a tRNA as substrates ([Figure 9](#pone-0032690-g009){ref-type="fig"}). The results obtained with these substrates confirm the ones that we obtained with the synthetic 16--30 ds substrate ([Figure 5](#pone-0032690-g005){ref-type="fig"}). For the *petD3* RNA, which has a stem loop structure near the 3′ end, synRNB protein was only able to cleave a few nucleotides, stalling when approached the stem loop region ([Figure 9](#pone-0032690-g009){ref-type="fig"}). When the substrate used was the tRNA-Glu, which is a highly structured RNA molecule, the protein was not able to cleave it ([Figure 9](#pone-0032690-g009){ref-type="fig"}). Together, these results showed us that the *Synechocystis* homologue behaved like an RNase II protein, since that the final product released was a 4 nt fragment for the single-stranded substrates and that the protein was shown to be sensitive to double-stranded substrates ([Figures 5](#pone-0032690-g005){ref-type="fig"} and [9](#pone-0032690-g009){ref-type="fig"}). However, in contrast to what happens with *E. coli* RNB family members, in *Synechocystis* RNase II does not prefer poly(A) substrates. This may be related to the fact that there is no polyadenylation by a PAP enzyme and the tails are synthesized by PNPase and are heteropolymeric [@pone.0032690-Rott1]. This protein is the only member of the RNB-family present in *Synechocystis*. To date, when only a member of this family is described in an organism, it was shown to behave like RNase R. This was the case of *Mycoplasma genitalium* [@pone.0032690-Lalonde1], *Legionella pneumophila* [@pone.0032690-Charpentier1], and *Streptococcus pneumoniae* [@pone.0032690-Domingues1]. *Synechocystis sp* is, the first organism where such observation was now shown not to be the case. In order to clarify why *Synechocystis* evolved to have an RNase II and not an RNase R, we decided to perform a phylogenetic analysis.
{ref-type="sec"} using petD3 and tRNA-Glu as substrates. The concentration of protein used is referred. Samples were taken during the reaction at the time points indicated.](pone.0032690.g009){#pone-0032690-g009}
To analyze the phylogenetic relationship between cyanobacteria RNase II/R homologues and defined proteobacteria RNase II and RNase R enzymes, we first created a multiple sequence alignment with MUSCLE [@pone.0032690-Edgar1]. Amino acid identities between cyanobacteria RNase II/R homologues and RNase II or RNase R enzymes are restricted to the central RNB catalytic domain of these proteins. The unrooted phylogenetic tree prepared with the 57 RNase II/R homologues reveals the existence of three clusters with different evolutionary lineages, consisting of the cyanobacteria RNase II/R homologous and the proteobacteria RNase II and RNase R groups ([Figure 10](#pone-0032690-g010){ref-type="fig"}). Interestingly, the cyanobacterial RNase II/R group is subdivided into three subclusters indicating considerable diversity between the species. It may represent an ancestral condition for the phylum with a subsequent convergence into two specialized family of exoribonucleases, namely RNase II and RNase R ([Figure 10](#pone-0032690-g010){ref-type="fig"}). Moreover, the phylogenetic analysis revealed that the RNase II and RNase R members are approximately equally distant from the RNase II/R homologue present in *Synechocystis* sp. PCC6803 ([Figure 10](#pone-0032690-g010){ref-type="fig"}). In this work, we proved that biochemically the RNase from *Synechocystis* sp. PCC6803 behaves as an RNase II and not like RNase R, although it is equally distant from both. It is possible that the ancestor would have both homologues and one of the enzymes was lost during evolution (some organisms maintained only an RNase R-like member, while for others, like *Synechocystis*, it was more propitious and favourable to maintain the RNase II-like protein). Other hypothesis was the presence of a unique enzyme, which evolved according to the environmental conditions.
{#pone-0032690-g010}
Materials and Methods {#s3}
=====================
Overexpression and purification of recombinant RNase II from *Synechocystis sp.* {#s3a}
--------------------------------------------------------------------------------
The plasmid used for expression of *Synechocystis sp.* PCC6803 histidine-tagged RNase II protein was pQE31synRNB. The *rnb* gene of *Synechocystis sp.* was amplified by PCR from genomic DNA using the primers synrnb1 (5′-GGCGAATTCATGGAAAAAGGACAACTAAT-3′) and synrnb2 (5′-GGCAGATCTAGGCYYCATTGGCCAACA-3′). This fragment was then digested and ligated into a pQE31 expression vector using *Pst*I and *Sph*I restriction sites. This allows the expression of the (His)~6~-tagged Syn RNase II fusion protein.
The plasmid was transformed into *E. coli* M15 (REP4) strain to allow the expression of the recombinant protein. Cells were grown at 37°C in 100 ml LB medium supplemented with 150 µg/ml ampicillin to an OD~600~ of 0.5 and induced by addition of 0.5 mM IPTG for 2 h. Cells were pelleted by centrifugation and stored at −80°C. *E. coli* RNase II and RNase R overexpression were performed as described previously [@pone.0032690-Amblar2], [@pone.0032690-Amblar3], [@pone.0032690-Arraiano3].
Purification was performed by histidine affinity chromatography using HiTrap Chelating HP columns (GE Healthcare) and AKTA FPLC system (GE Healthcare) following the protocol described previously [@pone.0032690-Amblar2], [@pone.0032690-Arraiano3]. The fractions containing the purified *Synechocystis* protein were pooled and loaded to an anion exchange monoQ column (GE Healthcare) equilibrated in buffer B composed by 20 mM Tris pH 8, 60 mM KCl, 2 mM MgCl~2~ and 0.2 mM EDTA. Protein elution was achieved by a continuous KCl gradient (from 60 mM to 1 M) in buffer B. Protein concentration was determined by spectrophotometry by using the ND100 Spectrophotometer from Nanodrop and 50% (v/v) glycerol was added to the final fractions prior storage at −20°C. 0.5 µg of the purified protein was applied to 8% SDS-PAGE and visualized by Coomassie blue staining (data not shown).
Exoribonucleolytic activity assays {#s3b}
----------------------------------
The exoribonucleolytic activity was determined using different synthetic substrates: a poly(A) oligomer of 30 nts, a 16-mer oligoribonucleotide (5′-CCCGACACCAACCACU-3′), and a 30-mer oligoribonucleotide (5′-CCCGACACCAACCACUAAAAAAAAAAAAAA-3′). The 30-mer was hybridized to the complementary unlabelled 16-mer oligodeoxiribonucleotide (5′-AGTGGTTGGTGTCGGG-3′), in order to obtain the double-stranded substrate 16--30 ds. The hybridization was performed in a 1∶1 (mol∶mol) ratio by 5 min incubation at 100°C followed by 45 min at 37°C. These RNA molecules were labelled at 5′-end with \[γ-^32^ATP\] and T4 polynucleotide kinase. The RNA oligomers were then purified using Microcon YM-3 Centrifugal Filter Devices (Millipore) to remove the unincorporated nucleotides. *In vitro* transcribed substrates were also used for the degradation assays. *petD3* and tRNA-Glu were transcribed from pBlueScript KS-psbA3′ [@pone.0032690-Lisitsky2] or from a PCR product of the coding sequence of tRNA-glu [@pone.0032690-Portnoy1] using T7 RNA polymerase RIBOMAX kit from Promega (following the instructions given by manufacturers) in a 20 µl volume, containing 20 µCi of \[α-^32^P\] UTP. Radioactively labelled RNA transcripts were purified on a 6% PAA/7M urea gel as previously described [@pone.0032690-Conrad1].
The exoribonucleolytic reactions were carried out in a final volume of 12.5 µl containing 5 nM of substrate, 20 mM Tris-HCl (pH tested from 6.5 to 9), KCl or NaCl (from 10 to 200 mM), MgCl~2~ (from 1 to 10 mM), and 1 mM DTT. The amount of each enzyme added to the reaction was adjusted to obtain linear conditions and is indicated in the figures and respective legends. Reactions were started by the addition of the enzyme and incubated at 30°C or 37°C. Samples were withdrawn at the time points indicated and the reaction was stopped by adding formamide-containing dye supplemented with 10 mM EDTA. Reaction products were resolved in a 20% polyacrylamide/7 M urea, or TLC chromatography [@pone.0032690-Portnoy1] and detected by using the Fuji Phosphorimager Analyzer TLA-5100 from GE Healthcare. The exoribonucleolytic activity of the enzyme was determined by measuring and quantifying the disappearance of the substrate in several distinct experiments. In the quantifications, the protein concentration was adjusted and less than 25% of substrate was degraded. Each value obtained represents the mean for these independent assays.
Surface plasmon resonance analysis - BIACORE {#s3c}
--------------------------------------------
Biacore SA chips were obtained from Biacore Inc. (GE Healthcare). The Flow cells of the SA streptavidin sensor chip were coated with a low concentration of the following substrates. On flow cell 1, no substrate was added so this cell could be used as the control blank cell. On flow cell 2, a 5′ biotinylated 25-nucleotide RNA oligomer (5′-CCCGACACCAACCACUAAAAAAAAA-3′) was added to allow the study of the protein interaction with a single-stranded RNA molecule. On flow cell 3, a 5′ biotinylated 30-mer PolyA substrate. On flow cell 4, the biotinylated 25-mer hybridized with the complementary 16-mer oligodeoxiribonucleotide (5′-AGTGGTTGGTGTCGGG-3′) was immobilized, originating the double-stranded substrate 16--25 ds. The target substrates were captured on flow cells 2 to 4 by manually injecting 20 µl of a 500 nM solution of the substrates in the reaction buffer at a 20 µl/min flow rate. The biosensor assay was run at 4°C in the buffer with 20 mM Tris-HCl pH7.5, 50 mM KCl, 1 mM DTT and 25 mM EDTA. The proteins were injected over flow cells 1, 2, 3 and 4 for 2.5 min at concentrations of 10, 20, 30, 40 and 50 nM using a flow rate of 20 µl/min. All experiments included triple injections of each protein concentration to determine the reproducibility of the signal and control injections to assess the stability of the RNA surface during the experiment. Bound protein was removed with a 30 s wash with 2 M NaCl. Data from flow cell 1 were used to correct for refractive index changes and non-specific binding. Rate and equilibrium constants were calculated using the BIA EVALUATION 3.0 software package, according to the fitting model 1∶1 Languimir Binding.
Phylogenetic analysis {#s3d}
---------------------
Based on sequence similarity to the *Escherichia coli* RNase II (Accession number NP_415802.1), we have identified 47 non-redundant representative homologues among cyanobacteria. Sequence similarity searches were performed using BLASTP 2.0 against cyanobacterial genomes database of the National Center for Biotechnology Information (NCBI). Next, we selected from the NCBI database five representative members from RNase II and RNase R family of enzymes, respectively. A multiple sequence alignment of the 57 RNase II/R homologues was generated by MUSCLE [@pone.0032690-Edgar1] and curated with default parameters of GBlocks [@pone.0032690-Castresana1]. Subsequently, a phylogenetic tree was constructed using the default option of the "advanced mode" implemented in the Phylogeny.fr platform (<http://www.phylogeny.fr/>) [@pone.0032690-Dereeper1]. A bootstrap analysis of 500 replicates was used to provide confidence of the constructed tree.
Protein modelling {#s3e}
-----------------
The model structures were built using the Swiss Model web server (<http://swissmodel.expasy.org/> [@pone.0032690-Arnold1]--[@pone.0032690-Schwede1]). 3D model for the synRNB protein was based on the crystal structures of wild-type RNase II and the RNase II D209N mutant complexed with a 13-nucleotide poly(A) RNA (Protein Data Bank entries 2IX1 and 2IX0 [@pone.0032690-Frazo1]). Figures of the structure and models were generated with Pymol [@pone.0032690-DeLano1].
Supporting Information {#s4}
======================
######
**Exoribonucleolytic activity at 30°C of** ***Synechocystis*** **protein: comparison with** ***E. coli*** **RNase II and RNase R.** Activity assays were performed using the three synthetic substrates: 30-mer poly(A), the 16-mer and the double-stranded substrate 16--30 ds. The concentration of proteins used is indicated in the figure. Samples were taken during the reaction at the time points indicated. Control reactions with no enzyme added (*Ctrl*) were incubated at the maximum reaction time for each protein. Length of substrates and degradation products are labelled.
(JPG)
######
Click here for additional data file.
We thank Andreia Aires for technical support in the lab.
**Competing Interests:**The authors have declared that no competing interests exist.
**Funding:**Work at ITQB was supported by grants from Fundação para a Ciência e a Tecnologia (FCT) and grant PEst-OE/EQB/LA0004/2011, also from FCT. Work at IBB/IST had financial support from FCT. Work in GS\'s lab was supported by grants from the BARD (Binational Agricultural Research and Development Fund) and BSF(Binational Science Foundation) foundations. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
[^1]: Conceived and designed the experiments: RGM AMF MG GS CMA. Performed the experiments: RGM AMF MG. Analyzed the data: RGM AMF MG GS CMA. Contributed reagents/materials/analysis tools: RGM AMF MG GS CMA. Wrote the paper: RGM AMF MG GS CMA.
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .168}
{#sp2 .169}
{#sp3 .170}
{#sp4 .171}
{#sp5 .172}
{#sp6 .173}
{#sp7 .174}
{#sp8 .175}
{#sp9 .176}
{#sp10 .177}
{#sp11 .178}
{#sp12 .179}
{#sp13 .180}
{#sp14 .181}
{#sp15 .182}
{#sp16 .183}
{#sp17 .184}
{#sp18 .185}
{#sp19 .186}
{#sp20 .187}
{#sp21 .188}
{#sp22 .189}
{#sp23 .190}
{#sp24 .191}
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-ijerph-14-00689}
===============
Antimony (Sb) and its compounds are widely used in some industries, including alloy, plastic, glass, and textile industries \[[@B1-ijerph-14-00689],[@B2-ijerph-14-00689],[@B3-ijerph-14-00689],[@B4-ijerph-14-00689],[@B5-ijerph-14-00689],[@B6-ijerph-14-00689]\]. The permissible exposure limit (PEL) established by the US Occupational Safety and Health Administration for exposure to antimony and compounds is 0.5 mg/m^3^ of antimony in the workplace based on a working 8 h shift and a 40 h week \[[@B7-ijerph-14-00689]\]. Taiwan has also adopted the 8 h time weighted average concentration of antimony and its compounds in the air of workplace at a level of 0.5 mg/m^3^ \[[@B8-ijerph-14-00689]\]. Antimony and its compounds may irritate eyes, skin, and respiratory systems, and are suspected to be carcinogenic and teratogenic substances \[[@B1-ijerph-14-00689],[@B9-ijerph-14-00689],[@B10-ijerph-14-00689],[@B11-ijerph-14-00689],[@B12-ijerph-14-00689],[@B13-ijerph-14-00689]\]. In an animal study, among rats exposed to substances containing Sb, only female rats developed lung cancer; 27% in those rats exposed to antimony trioxide and 25% in those rats exposed to antimony ore had the cancer \[[@B10-ijerph-14-00689]\]. Grosskopf et al. reported that trivalent antimony was responsible for genotoxicity in the cellular system because antimony could partly impair the pathway of nucleotide repair \[[@B11-ijerph-14-00689]\]. DNA damage has been detected for workers with occupational exposure to antimony trioxide \[[@B12-ijerph-14-00689],[@B13-ijerph-14-00689]\]. International Agency for Research on Cancer (IARC) has classified trivalent antimony as a possible human carcinogen \[[@B14-ijerph-14-00689]\].
Studies investigating the impact of antimony exposure on the human immune system are limited. A survey on occupational contact dermatitis and sensitization among 126 workers employed in the ceramics industry showed that 48 (25.3%) workers were found to be sensitized to various exposures. Among them, two persons were sensitized to antimony trioxide \[[@B1-ijerph-14-00689]\]. Huang et al. found that urinary levels of metals, including antimony, were higher in patients with asthma than subjects without asthma \[[@B15-ijerph-14-00689]\]. In an earlier study in workers at an antimony trioxide manufacturing plant, Kim et al. found the serum immunoglobulin levels, such as IgG1 and IgE, were lower in workers exposed to antimony than in controls \[[@B2-ijerph-14-00689]\].
In 2010, Taiwan imported 5501 tons of antimony trioxide and produced 14,100 tons of the chemical. Workers in the antimony industry are likely exposed to a large amount of antimony than general population, particularly for workers exposed to antimony trioxide at the nanoparticle size of 0.6 µm emitted with nano-technology. This study investigated antimony exposure levels among workers and administrative staff employed in three types of industry associated with manufacturing antimony compounds or using them, including industries producing antimony trioxide, and industries using antimony for manufacturing glass and engineering plastic products. In the glass manufacturing industry, sodium antimonite is used to decolorize and refine glass \[[@B5-ijerph-14-00689]\]. Antimony trioxide is also used as a polycondensation catalyst in the plastic manufacturing industry to synthesize polyethylene terephthalate \[[@B5-ijerph-14-00689]\]. Workers at these three types of industry may be exposed to various levels of antimony. We attempted to investigated whether the immune factors were associated with the antimony exposure levels among these workers. Administrative staff served as general population controls.
2. Materials and Methods {#sec2-ijerph-14-00689}
========================
2.1. Study Groups {#sec2dot1-ijerph-14-00689}
-----------------
This study was conducted at one antimony trioxide manufacturing plant, two glass manufacturing plants, and two engineering plastic manufacturing plants, after obtaining the approval from the Research Ethics Committee at China Medical University & Hospital (DMR99-IRB-142(FR)). Ninety-one male workers were recruited at worksites as the metal exposure group, and 42 male administrators from these five plants as controls. With consent, we collected air samples at worksites and administrative offices at the 5 plants. Each participant provided three tubes of blood samples in the mornings: one 3 mL blood sample in a purple head tube for metal analysis, one 3 mL blood sample in a purple head tube for white blood cell (WBC) count, and a 10 mL blood sample in a red head tube for an immunoglobulin assay. We also collected urine and hair samples from participants to assess metal concentrations. All samples were shipped to laboratories at 4 °C. Samples for white blood cell counts and the immunoglobulin assay were sent to the hospital for analyses in 24 h. Blank samples were prepared on sites for all types of sample. Samples for metal tests were stored at −70 °C until analysis. We used Inductively Coupled Plasma Mass Spectrometer (ICP-MS) (Perkin-Elmer SCIEX ELAN DRC II, Concord, ON, Canada) to determine the antimony concentrations in the blanks, samples, and standards (ICP multi element Standard Solution, Merck, Darmstadt, Germany). The limit of detection (LOD) and the limit of quantification (LOQ) were performed. The spike recovery test or reference materials recovery test was performed as well. Analyzers were blinded to sample identifications. A questionnaire was used to collect information on personal characteristics, including birth date, work histories, lifestyle (i.e., smoking, drinking, betel nut chewing, etc.), and physician diagnosed allergy status.
2.2. Sample Collection and Analysis for Antimony {#sec2dot2-ijerph-14-00689}
------------------------------------------------
### 2.2.1. Air Sample {#sec2dot2dot1-ijerph-14-00689}
The environmental air sampling devices were set up at a 120 cm height at worksites and administrative offices to collect the particulates in the air using a 37 mm filter cartridge containing a 0.5 μm PolyVinyl Chloride (PVC) filter. Using a flow rate of 2 L/min, personal samplers (Gillian Instrument Corp., West Caldwell, NJ, USA) were set for the sampling time from 5 to 7 h during work. Filter blanks and field blanks were prepared for quality control and were analyzed together with the samples. The filter in the sampling tube was treated with a mixture of 5 mL of 37% hydrochloric acid and 1 mL of 70% nitric acid, followed by ultrasonic shock for 30 min, and filtered using a Milipore filter membrane with a 0.22 μm pore size. The filtrate was diluted with 1% (*v*/*v*) of hydrochloric acid and nitric acid \[[@B16-ijerph-14-00689]\]. We used ICP-MS (Perkin-Elmer SCIEX ELAN DRC II, Waltham, MA, USA) to determine the antimony concentration in the blanks, samples, and standards (ICP multi element Standard Solution, Merck, Darmstadt, Germany). LOD and LOQ were 4.4 µg and 15 µg, respectively. The spike recovery test of antimony was 84.5%.
### 2.2.2. Blood Sample {#sec2dot2dot2-ijerph-14-00689}
For analyzing antimony in the blood, a 0.5 mL blood sample was placed into a vial with 3 mL of 70% nitric acid and digested using microwave. We took 1 mL of digested solution, added 1 mL of Indium standard solution (as an internal standard), mixed it with 8 mL of 1% (*v*/*v*) hydrochloric acid to make a 10 mL solution, and quantified blood antimony using ICP-MS (Perkin-Elmer SCIEX ELAN DRC II, Waltham, MA, USA) \[[@B17-ijerph-14-00689]\]. LOD and LOQ for the blood antimony determination were 0.06 µg/L and 0.12 µg/L, respectively. Seronorm Trace Elements Blood L-3 (ref. 102405) (Seronorm Pharmaca, Billingstad, Norway) was used as a reference material, and the recovery rate was 90.0%.
### 2.2.3. Urine Sample {#sec2dot2dot3-ijerph-14-00689}
A clean glass vial was sent to each participant in advance. Each participant provided a vial of the first void urine specimen in the morning. We took a 1 mL urinary sample to measure creatinine. The remaining urinary samples were stored at −70 °C until ready for analysis. At room temperature, 10 mL of urinary samples were centrifuged at 3000 rpm for 5 min. We took 1 mL of upper solution mixed with 1 mL of Indium standard solution (as an internal standard) and 8 mL of 1% (*v*/*v*) of hydrochloric acid and nitric acid and then quantified urinary antimony using ICP-MS (Perkin-Elmer SCIEX ELAN DRC II, Waltham, MA, USA) \[[@B18-ijerph-14-00689]\]. The levels of urine antimony were adjusted for urinary creatinine (cre.) and expressed as ug/g cre. The LOD and LOQ for the urine antimony determination were 0.03 µg/L and 0.12 µg/L, respectively. Seronorm Trace Elements Urine L-2 (ref. 201205) (Seronorm Pharmaca, Billingstad, Norway) was used as a reference material, and the recovery rate was 90.2%.
### 2.2.4. Hair Sample {#sec2dot2dot4-ijerph-14-00689}
A pinch of hairs near neck cut from each participant was collected in a sealed bag for analyzing the antimony concentration. We took 0.4 g of hairs and washed it twice with 1:200 (*v*/*v*) Triton X-100 solutions, followed by acetone, and finally washed twice with deionized water. The washed hairs were dried in an oven at 75 °C for 24 h, and then stored in an electronic dry cabinet (AD-51, EDRY Enterprise Co, Taipei, Taiwan) at room temperature for 12 h or longer until digestion. We measured 0.2 g of dried hairs in a vial with 3 mL of 70% nitric acid and digested using microwave. The digested hair solution was diluted to 10 mL with 1% (*v*/*v*) hydrochloric acid. We took 1 mL solution, mixed with 1 mL of Indium standard solution (as an internal standard) and 8 mL of 1% (*v*/*v*) hydrochloric acid and quantified hair antimony using ICP-MS (Perkin-Elmer SCIEX ELAN DRC II, Waltham, MA, USA) \[[@B19-ijerph-14-00689]\]. The LOD and LOQ for the hair antimony determination were 0.0004 µg/g and 0.012 µg/g, respectively. Certified reference hair (CRM GBW-09101-Human Hair, Shanghai Institute of Nuclear Research Academia Sinica, Shanghai, China) was used as a reference material, and the recovery rate was 81.0%.
2.3. White Blood Cell and Immunoglobulins Determination {#sec2dot3-ijerph-14-00689}
-------------------------------------------------------
The white blood cell counts were performed using flow cytometry (Automated Hematology Analyzer of Beckman Coulter LH series). The 10 mL blood sample was centrifuged to obtain serum for the measurement of IgA, IgG, and IgE. Serum IgA and IgG were measured using turbidimetry (Nephlometer, Hitachi 747, Tokyo, Japan), and serum IgE was quantified by Enzyme-linked immunoassay \[[@B20-ijerph-14-00689]\].
2.4. Statistical Analysis {#sec2dot4-ijerph-14-00689}
-------------------------
Data analysis first compared the personal characteristics between all operation workers and all administrative staff recruited at the 5 plants. Distributions of age, employment history, lifestyles, and allergy history were examined using chi-square. Average antimony concentrations in air, blood, urine, and hair samples were compared between workers and administrative staff by industry type. Differences were examined using the Kruskal--Wallis test because antimony concentrations in the air samples among the 5 plants, and in the blood, urine, and hair samples of the participants, were not normally distributed. Counts of WBC, lymphocyte, monocyte, IgA, IgG, and IgE were stratified into 2 or 3 levels based on the range of reference values, and compared between all workers and all administrative staff, examined using chi-square. We also calculated and compared means of serum WBC, lymphocyte, monocyte, IgA, IgG, and IgE between workers and administrative staff using the Kruskal--Wallis test. The Spearman's correlation coefficients, ρ (rho), were calculated between levels of antimony and of immunological indicators for all participants. IBM SPSS Statistics version 18 software (IBM Corp., Armonk, NY, USA) was used for data analyses, and the *p*-value was set at 0.05 as significant.
3. Results {#sec3-ijerph-14-00689}
==========
3.1. The Attributes of Subjects {#sec3dot1-ijerph-14-00689}
-------------------------------
[Table 1](#ijerph-14-00689-t001){ref-type="table"} shows that workers were younger and had shorter employment history than were administrators. However, the prevalence rates of smoking, drinking, betel nut chewing, and allergic history of metal-exposed workers and administrative staff were alike. Near 30% of study subjects smoked, used alcohol, and chewed betel nuts, and 21.8% of them had been diagnosed with an allergic disorder.
3.2. The Antimony Levels in the Air of Worksite and in Blood, Urine and Hairs Samples {#sec3dot2-ijerph-14-00689}
-------------------------------------------------------------------------------------
[Table 2](#ijerph-14-00689-t002){ref-type="table"} shows antimony levels in samples of air, blood, urine, and hair by industry type for workers and administrative staff. The mean antimony concentration in air samples measured for the antimony trioxide manufacturing plant was the highest (2.51 ± 0.57 mg/m^3^), near 18-fold higher than that for glass plants or 12-fold higher than that for engineering plastic manufacturing plants. The antimony concentrations in blood, urine, and hair measured for workers of antimony trioxide manufacturing plant were also the highest, at levels of 3.88 ± 1.10 μg/L, 27.15 ± 6.00 μg/g cre., and 0.10 ± 0.01 μg/g, respectively. The Spearman's correlation analysis showed that antimony concentrations in blood, urine, and hair of participants were significantly associated with the concentrations in air samples with coefficients of 0.713, 0.870, and 0.865 (*p* \< 0.01), respectively (data not shown). The measured antimony levels in air and in blood, urine, and hair samples were much lower for all administrative staff than for all workers (all *p* \< 0.01).
3.3. White Blood Cell Count and Immunoglobulin Indicators {#sec3dot3-ijerph-14-00689}
---------------------------------------------------------
Immunoglobulin levels of most participants in this study were in normal physiological reference ranges ([Table 3](#ijerph-14-00689-t003){ref-type="table"}). However, 9.0% of participants had the WBC levels higher than the reference values, and 24.1% of participants had lymphocyte levels below the reference values. The monocyte levels, IgA and IgE of workers and staff were in normal reference value ranges. However, the mean serum IgG, IgA, and IgE levels among workers were lower than that among administrative staff (*p* ≤ 0.001).
3.4. The Correlation between Immunological Levels and Antimony Levels {#sec3dot4-ijerph-14-00689}
---------------------------------------------------------------------
[Table 4](#ijerph-14-00689-t004){ref-type="table"} shows correlations between immunological indicators of all participants and antimony levels in air, blood, urine, and hair samples. WBC levels had a positive relationship with antimony exposures, but not significant. The monocyte levels were negatively correlated with antimony levels in blood and urine, with the corresponding coefficients of −0.300 and −0.175 (*p* \< 0.05), respectively. The serum IgG levels were negatively correlated with antimony levels in air samples at worksites and in hairs of participants (*p* \< 0.05). The serum IgA and IgE levels also had significant negative correlations with antimony levels in air and in blood, urine, and hair. The Spearman's ρ (rho) values were stronger for IgA, with coefficients of −0.366, −0.291, −0.355 and −0.370 (*p* \< 0.001), associated with antimony levels in air, and in blood, urine, and hair, respectively.
4. Discussion {#sec4-ijerph-14-00689}
=============
This study surveyed the antimony exposure levels for workers and administrative staff at manufacturing plants with antimony exposures and evaluated relationships between levels of antimony exposure and immunologic characteristics of participants. We surveyed glass, antimony trioxide, and engineering plastics manufacturing plants and found that the environmental antimony concentration in the air samples collected at these five worksites was the highest at the antimony trioxide manufacturing plant, more than five times the legal limit (PEL) of 0.5 mg/m^3^ of Taiwan. The antimony levels in blood, urine, and hair were also the highest in samples from workers of the antimony trioxide manufacturing plant, in response to the exposure from the air.
In this study, the antimony measured in blood, urine, and hair for participants were strongly associated with the antimony concentrations in the air to which they were exposed to. Our further data analysis showed that the relationship was stronger for levels in urine and in hair (coefficients of 0.870 and 0.865, respectively) than for levels in blood (a coefficient of 0.713) (data not shown). Antimony in urine and in hairs could be appropriate biomarkers for evaluating the exposure of antimony at worksites. However, in an occupational survey for antimony exposure in textile factory, Iavicoli et al. found that the air antimony levels of personal exposure ranged from 0.01 to 0.55 µg Sb/m^3^ and that the mean urinary antimony level of workers was 0.35 ± 0.29 µg Sb/L \[[@B3-ijerph-14-00689]\]. They considered the correlation between low environmental exposure and human burden is negligible. In an earlier survey at a lead battery factory, Kentner et al. found that the mean antimony levels in the air were 4.5 µg Sb/m^3^ in the grid casting area and 12.4 µg Sb/m^3^ in the lead plate stibine formation area \[[@B21-ijerph-14-00689]\]. The corresponding mean urinary levels in workers at the end of a week of exposure were 3.9 and 15.2 µg Sb/g creatinine, respectively. Their air and urinary antimony levels were greater than those we found in our study at the antimony plants. Lüdersdorf et al. assessed trivalent antimony exposure among glass refining workers and found that the urinary antimony levels were associated with the concentrations in the air samples of the worksites. This suggested that urinary antimony levels were useful in monitoring the exposure of antimony in work places \[[@B22-ijerph-14-00689]\].
Metals and organic chemicals have been associated with immunity \[[@B23-ijerph-14-00689],[@B24-ijerph-14-00689],[@B25-ijerph-14-00689],[@B26-ijerph-14-00689]\]. Fewer studies have investigated the immunomodulatory associated with antimony exposure. We found that the serum IgG, IgA, and IgE levels were significantly lower among workers than among administrative staff and were negatively correlated with the antimony levels in the worksite air, and in the blood, urine, and hair of study participants. Our results are consistent with findings of an earlier study \[[@B2-ijerph-14-00689]\]: Kim et al. also found that the antimony exposure had an association with lower serum IgG1 and IgE levels \[[@B2-ijerph-14-00689]\]. However, we did not evaluate the relationship between the subclasses of IgG1 and antimony exposure. Immunoglobulins play an important role in anti-infection and in lowering the chance of cancer \[[@B23-ijerph-14-00689],[@B24-ijerph-14-00689]\]. Whether the serum levels of IgG, IgA, and IgE suppressed in workers exposed to antimony increase the risk of infections or chronic disorders deserves further study.
This is one of the few studies exploring the correlation between antimony exposure and immunoglobulin levels, but has some limitations. The causal relationship between antimony exposure and immunological indicators cannot be established in this study because of its cross-sectional design. However, the antimony levels in hairs represent a historical exposure; there could be a negative relationship between antimony levels in the hair and serum levels of IgG, IgA, and IgE. Levels of neutrophils and eosinophils in white blood cells were not measured in this study, we were unable to measure whether levels of neutrophils and eosinophils are associated with the antimony exposure. Our sample size was not large enough to analyze these associations by age stratum or by work history.
5. Conclusions {#sec5-ijerph-14-00689}
==============
The antimony levels in blood, urine, and hair were useful in evaluating the antimony exposure from worksites. Our study demonstrated that the high heterogeneity in antimony exposures from the air of five plants provided clear Spearman's correlations with human immunity markers. The serum levels of IgG, IgA, and IgE were lower among workers exposed to antimony than among administrative staff and were negatively associated with antimony levels in hair. Whether the suppression of serum levels of IgG, IgA, and IgE associated with antimony exposure is detrimental to health deserves study.
We are grateful for the cooperation from glass, antimony trioxide, and engineering plastic manufacturing factories and all participants who had donated samples of blood, urine, and hair. This work was supported by the National Science Council in Taiwan (grant No. NSC 99-2314-B-039-032-MY2).
Chin-Ching Wu designed and performed the experiments. Yi-Chun Chen analyzed the data and revised and finished the paper.
The authors declare no conflict of interests.
ijerph-14-00689-t001_Table 1
######
Demographic and lifestyle characteristics of workers exposed to antimony and administrative staff.
Variables Workers Administrators Total *p*-Value \*
------------------- ----------- ---------------- ------------ --------------
Age, years *n* (%) *n* (%) *n* (%)
\<30 16 (17.6) 0 16 (12.0) \<0.001
30--39 30 (33.0) 9 (21.4) 39 (29.3)
40--49 31 (34.1) 19 (45.3) 50 (37.6)
50--59 14 (15.3) 8 (19.1) 22 (16.5)
≥60 0 6 (14.2) 6 (4.5)
Years at work
\<10 19 (20.9) 4 (9.5) 23 (17.3) 0.001
10\~19 50 (55.0) 18 (42.9) 68 (51.1)
20\~29 22 (24.1) 15 (35.7) 37 (27.8)
≥30 0 5 (11.9) 5 (3.8)
Smoking
Yes 26 (28.6) 13 (31.9) 39 (29.3) 0.78
No 65 (71.4) 29 (69.1) 94 (70.7)
Drinking
Yes 30 (33.0) 14 (33.3) 44 (33.1) 0.97
No 61 (67.0) 28 (66.7) 89 (66.9)
Betel nut use
Yes 28 (30.8) 12 (28.6) 40 (30.1) 0.80
No 63 (69.2) 30 (71.4) 93 (69.9)
Diagnosed allergy
Yes 17 (18.7) 12 (28.6) 29 (21.8) 0.20
No 74 (81.3) 30 (71.4) 104 (78.2)
\* Chi-square test.
ijerph-14-00689-t002_Table 2
######
Average antimony concentrations in samples of air, blood, urine, and hair of metal-exposed workers and administrative staff by type of industry.
Factory Antimony Concentration
-------------------------------- ------------------------ ----------------- --------------- ---------------
**Glass**
Workers (*n* = 55) 0.14 ± 0.01 0.78 ± 0.21 5.60 ± 1.24 0.10 ± 0.01
Administrativestaff (*n* = 20) 0.007 ± 0.001 0.60 ± 0.11 2.55 ± 0.71 0.06 ± 0.01
*p*-value \* \<0.001 \<0.001 \<0.001 \<0.001
**Antimony trioxide**
Workers (*n* = 14) 2.51 ± 0.57 3.88 ± 1.10 27.15 ± 6.00 5.66 ± 3.66
Administrativestaff (*n* = 9) 0.04 ± 0.01 1.07 ± 0.87 2.09 ± 0.55 0.04 ± 0.004
*p*-value \* \<0.001 \<0.001 0.001 \<0.001
**Engineering plastic**
Workers (*n* = 22) 0.21 ± 0.06 2.17 ± 0.48 7.48 ± 1.30 0.32 ± 0.05
Administrativestaff (*n* = 13) 0.004 ± 0.001 0.49 ± 0.05 1.86 ± 0.55 0.04 ± 0.004
*p*-value \* \<0.001 \<0.001 \<0.001 \<0.001
**Total**
Workers (*n* = 91) 0.52 ± 0.88 1.61 ± 1.25 9.28 ± 6.31 1.00 ± 2.35
Administrativestaff (*n* = 42) 0.012 ± 0.015 0.602 ± 0.0.140 2.26 ± 0.0.68 0.048 ± 0.041
*p*-value \* \<0.001 \<0.001 \<0.001 \<0.001
\* Kruskal--Wallis test. ^a^ cre.: creatinine.
ijerph-14-00689-t003_Table 3
######
Distributions of levels of white blood cell and immunological indicators compared between workers and administrative staff.
Immunological Indicators Workers Administrators Total *p*-Value
-------------------------- --------------- ---------------- --------------- -----------
WBC, 10^3^/μL *n* (%) *n* (%) *n* (%)
\<4 1 (1.1) 0 1 (0.8) 0.69
4--10 ^a^ 81 (89.0) 39 (92.9) 120 (90.2)
\>10 9 (9.9) 3 (7.1) 12 (9.0)
Mean (SD) 6.59 (1.99) 6.00 (1.71) 6.41 (1.92) 0.051
Lymphocyte, %
\<30 26 (28.6) 6 (14.3) 32 (24.1) 0.20
30--40 ^a^ 61 (67.0) 34 (81.0) 95 (71.4)
\>40 4 (4.4) 2 (4.7) 6 (4.5)
Mean (SD) 32.3 (4.78) 33.2 (3.99) 32.6 (4.56) 0.160
Monocyte, %
\<4 0 0 0 \-
4--10 ^a^ 91 (100.0) 42 (100.0) 133 (100.0)
\>10 0 0 0
Mean (SD) 6.71 (0.76) 6.80 (0.82) 6.74 (0.78) 0.899
IgG, mg/dL
\<700 1 (1.1) 0 1 (0.8) 0.50
700--1600 ^a^ 90 (98.9) 42 (100.0) 132 (99.2)
\>1600 0 0 0
Mean (SD) 925.7 (131.5) 989.6 (94.7) 945.4 (124.5) 0.001
IgA, mg/dL
\<70 0 0 0 \-
70--400 ^a^ 91 (100.0) 42 (100.0) 133 (100.0)
\>400 0 0 0
Mean (SD) 225.7 (32.9) 248.3 (26.3) 232.6 (32.6) \<0.001
IgE, mg/dL
0\~200 ^a^ 91 (100.0) 42 (100.0) 133 (100.0) \-
\>200 0 0 0
Mean (SD) 123.6 (18.7) 135.7 (16.6) 127.3 (18.9) \<0.001
WBC, white blood cell. \* chi-square and Kruskal--Wallis tests. ^a^ the normal reference value range.
ijerph-14-00689-t004_Table 4
######
Spearman's correlation coefficients of antimony exposure levels and immunological indicators of all study participants. (*N* = 133).
Immunological Indicators Antimony in
-------------------------- ------------- ------------- ------------- -------------
WBC 0.135 0.010 0.126 0.143
Lymphocyte −0.104 −0.106 −0.121 −0.137
Monocyte −0.117 −0.300 \*\* −0.175 \* −0.164
IgG −0.260 \* −0.026 −0.157 −0.187 \*
IgA −0.366 \*\* −0.291 \*\* −0.355 \*\* −0.370 \*\*
IgE −0.236 \* −0.171 \* −0.175 \* −0.217 \*
^a^ Spearman's correlation coefficients. \* *p* \< 0.05. \*\* *p* \< 0.001.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-polymers-11-01443}
===============
Our daily life is filled with diverse forms of ambient energies that are continuously generated and observable, but mostly wasted without being exploited. Various types of energy harvesters have been developed in order to harness such energies. For example, many promising technologies have been developed to convert solar and wind energies to electrical energy \[[@B1-polymers-11-01443]\]. Following the demonstration of triboelectric nanogenerators (TENGs) consisting of polyester and Kapton polymer thin films by Wang et al. in 2012, TENGs have attracted significant interest as a promising energy harvesting technology \[[@B2-polymers-11-01443],[@B3-polymers-11-01443],[@B4-polymers-11-01443],[@B5-polymers-11-01443],[@B6-polymers-11-01443]\]. TENGs can convert mechanical energy into electrical energy through the phenomenon termed the triboelectric effect \[[@B7-polymers-11-01443],[@B8-polymers-11-01443],[@B9-polymers-11-01443]\]. For example, electrostatic charges are generated in our shirts, coats, and dresses by the triboelectric effect when their textiles are rubbed against each other. Interestingly, the electrostatic charges generated can be utilized as a useful energy source with the aid of TENGs. Since cotton fibers are one of the most common materials used for manufacturing clothing, it would be useful to investigate the electrical performance of cotton-textile-based TENGs by employing various approaches. For example, investigations of TENGs with differently-woven textiles can provide useful information about the suitability of different textile morphologies for energy harvesting. In the field of textile engineering, previous studies have reported that mechanical properties such as stiffness, compressibility, and tensile strength, which are associated with yarn interlacement, differ among differently-woven textiles \[[@B10-polymers-11-01443],[@B11-polymers-11-01443],[@B12-polymers-11-01443]\]. Many different patterns, such as plain, matt, twill, warp rib, herringbone, and satin are used for weaving yarns into textiles \[[@B11-polymers-11-01443]\]. The mechanical differences among differently-woven cotton textiles may be reflected in electrical differences among cotton-textile-based TENGs employing such textiles, since triboelectric charges can be generated by mechanical interactions such as the friction among cotton fibers. In a previous study, silver- and polytetrafluoroethylene-textile-based TENGs with different weave patterns, showing the output voltage of 23.50 V, were investigated in terms of the textiles′ mechanical properties \[[@B13-polymers-11-01443]\]. In particular, the relationship between the mechanical properties of the cotton textiles employed and the device performance in cotton-textile-based TENGs has not been fully explored in the context of electronic engineering, despite the physical processes involved with triboelectric charge generation on organic surfaces and molecules being moderately well understood. In addition, it is noteworthy that owing to their mechanical flexibility, fiber-type TENGs are more suitable candidates for wearable electronics compared to film-type TENGs \[[@B14-polymers-11-01443],[@B15-polymers-11-01443]\].
In this work, a comparative study of the output voltages of TENGs fabricated with plain- and 2/1 twill-woven cotton textiles was conducted. The microstructures of the cotton fibers were examined using atomic force microscopy (AFM) and scanning electron microscopy (SEM) analyses. The difference in the output voltages of the TENGs was elucidated on the basis of triboelectric charge generation in the cotton textiles.
2. Materials and Methods {#sec2-polymers-11-01443}
========================
TENGs were fabricated by using cotton textiles as active triboelectric layers. [Figure 1](#polymers-11-01443-f001){ref-type="fig"}a shows a schematic of the TENGs. For use as substrates and spacers, polydimethylsiloxane (PDMS) films were prepared using Sylgard-184, an elastomeric PDMS kit manufactured by Dow Corning (Midland, MI, USA). A 10:1 PDMS base/curing agent mixture, which was stored in a vacuum desiccator to remove air bubbles, was poured onto a flat plate and subjected to thermal treatment at 100 °C for 1 h on a hot plate. As shown in [Figure 1](#polymers-11-01443-f001){ref-type="fig"}a, cotton textiles, a PDMS spacer film, and copper-tape electrodes were stacked and fixed on a PDMS substrate using an adhesive for fabricating the TENG. A photo of the fabricated TENG is shown in [Figure 1](#polymers-11-01443-f001){ref-type="fig"}b. Plain- and 2/1 twill-woven cotton textiles were used as the active triboelectric layers. [Figure 1](#polymers-11-01443-f001){ref-type="fig"}c shows schematics of plain- and 2/1 twill-woven cotton textiles \[[@B16-polymers-11-01443]\]. The cotton textiles consisted of warp and weft yarns, each of which was a strand of cotton fibers. The thicknesses of the PDMS spacer film, cotton textile, and copper-tape electrode were 1 mm, 280 μm and 50 μm, respectively. The PDMS spacer film had a 2.5 cm × 2.5 cm square gap. The cotton textiles were purchased from Sombe (Congo). The output voltages of the TENGs were measured using a low-noise current preamplifier (SR570; Stanford Research Systems, Sunnyvale, CA, USA).
3. Results and Discussion {#sec3-polymers-11-01443}
=========================
The chemical composition of cotton fibers should be considered in order to understand triboelectric charge generation in cotton textiles. Cotton fibers contain various organic compounds, including cellulose, waxes, pectins, organic acids, and some inorganic substances; cellulose accounts for approximately 90% of the dry weight of cotton fibers \[[@B17-polymers-11-01443]\]. [Figure 1](#polymers-11-01443-f001){ref-type="fig"}d shows the molecular structures of cellulose, wax, and pectin \[[@B18-polymers-11-01443]\]. When mechanical stresses are applied to cotton textiles, energetic interactions may occur among the various molecules on their surface. Collisions among the different molecules are likely to induce electron exchange due to the difference in electron affinity among them \[[@B19-polymers-11-01443]\].
The surface microstructures of cotton fibers were examined to understand the basic mechanical interaction among the cotton fiber yarns in the TENGs. Unlike film-type TENGs based on film-to-film interactions, examining the morphological properties of cotton fibers would be important and helpful in understanding the working mechanisms of fiber-type TENGs based on fiber-to-fiber interactions. [Figure 2](#polymers-11-01443-f002){ref-type="fig"}a shows a SEM image of cotton fibers in a single yarn. The observed cotton fibers have elongated and curved shapes, which is in conformance with the general morphologies of cotton fibers \[[@B20-polymers-11-01443]\]. [Figure 2](#polymers-11-01443-f002){ref-type="fig"}b shows an AFM image of a cotton fiber. The cotton fiber exhibited an uneven and bumpy surface; the root-mean-square surface roughness of the cotton fiber was 3.247 nm. [Figure 2](#polymers-11-01443-f002){ref-type="fig"}c shows the triboelectric interaction between the surfaces of adjacent cotton fibers. When friction occurs between adjacent cotton fibers, the molecules in the surface microstructures of the cotton fibers collide with each other, thereby generating triboelectric charges. It should be noted that air gaps are present between adjacent cotton fibers in a single yarn and between warp and weft yarns in a single textile since the fibers as well as the warp and weft yarns are not completely interfaced with each other. Presumably, triboelectric charges are generated not only at the colliding surfaces between the upper and lower textiles, but also at the colliding surfaces of cotton fibers in each yarn.
The output voltages of the TENGs with plain- and 2/1 twill-woven cotton textiles were measured by applying a pressure of 0.47 N/cm^2^ on the textiles at a frequency of 5.8 Hz for 5 s. [Figure 3](#polymers-11-01443-f003){ref-type="fig"}a,b show the output voltages of the TENGs, respectively. The positive peak output voltages of the TENGs with plain- and 2/1 twill-woven cotton textiles were 1.59 ± 0.08 and 12.47 ± 0.62 V, respectively, and the negative peak output voltages were −0.86 ± 0.05 and −4.13 ± 0.21 V, respectively. These values were obtained from tens of devices for each case. Clearly, the TENG with 2/1 twill-woven cotton textiles exhibited higher peak output voltages. [Figure 3](#polymers-11-01443-f003){ref-type="fig"}c and d show the successive two output voltage waveforms of the TENGs with plain- and 2/1 twill-woven cotton textiles, respectively. Considering the stabilizing time of the measured voltage from the negative peak to 0 V, the time interval, corresponding to the 5.8 Hz measurement frequency, was properly chosen to carry out the repetitive measurement with the prevention of the electrical interference between successive measurements. A single output voltage waveform shows a positive peak resulting from pressure application and a negative peak generated by pressure release. When pressure is applied on the textiles, triboelectric charges are generated in them, as shown in [Figure 3](#polymers-11-01443-f003){ref-type="fig"}e. The generated mobile triboelectric charges, which are electrons, move along the surface of cotton fibers and are subsequently gathered near the top electrode by diffusion, resulting in a positive peak in the output voltage. Cellulose molecules would form the triboelectric charge pathway, considering that cellulose is the major component of cotton fibers and that cellulose-to-cellulose charge transfer is the electron movement between the identical energy levels of cellulose molecules. Previously, it was found that triboelectric charges can diffuse laterally on the surface of silicon dioxide which is one of the most well-known insulators with high resistivity \[[@B21-polymers-11-01443]\]. At the tip of the bottom electrode, electric repulsion among triboelectric charges possibly hindered their accumulation near the electrode, as shown in [Figure 3](#polymers-11-01443-f003){ref-type="fig"}e. The recombination of triboelectric charges and the continuous charge flow caused the electrodes to have opposite polarities, which lead to the negative peak in the output voltage \[[@B2-polymers-11-01443]\]. In addition, the output voltages of the TENGs were measured repeatedly with repeated press and release operations to determine the stability of voltage generation. [Figure 3](#polymers-11-01443-f003){ref-type="fig"}f shows the normalized positive peak output voltages of Test \#1, \#2, \#3, and \#4. Each test, consisting of twenty waveforms, was carried out after one thousand press and release operations. During the whole measurement, the TENGs exhibited no significant variation in the voltage generation. In other words, the TENGs generated voltage at a stable level with thousands of pressing and releasing applications. The stable voltage generation can be attributed to the chemical and morphological stabilities of cotton fibers under mechanical stresses.
To explain the difference in output voltage between the TENGs, an understanding of the relationship between the mechanical properties of the cotton textiles and the triboelectric charge density in the cotton textiles is necessary. [Figure 4](#polymers-11-01443-f004){ref-type="fig"} shows schematics of plain- and 2/1 twill-woven cotton textiles to which compressive forces were applied. The textiles contain the critical contact points of warp and weft yarns, which are resistant to compressive deformation due to the vertical components of the contact forces. The difference in the density of the critical contact points would create the difference in compressibility between the two types of textiles. Note that in previous studies, the difference in compressibility between plain and twill weaves was proven experimentally \[[@B11-polymers-11-01443],[@B12-polymers-11-01443]\]. The degree of compressive deformation in the 2/1 twill-woven cotton textiles is likely to be higher than that in plain-woven cotton textiles, leading to greater mechanical interaction among cotton fibers. Consequently, under a compressive force, the density of the generated triboelectric charge in the 2/1 twill-woven cotton textiles would be higher than that in the plain-woven cotton textiles. Thus, the different output voltages of the TENGs with plain- and 2/1 twill-woven cotton textiles appear to result from the difference in the mechanical property of the textiles. Meanwhile, immobile positive triboelectric charges presumably interfere with the flow of triboelectric electrons, restricting the performance of the TENGs. Our future work will be specific studies on the structural modifications of the TENGs such as use of mesh-type electrodes for improving electron collection efficiency.
4. Conclusions {#sec4-polymers-11-01443}
==============
TENGs were fabricated using plain- and 2/1 twill-woven cotton textiles, and their output voltages were compared. The TENGs with plain- and 2/1 twill-woven cotton textiles exhibited average positive peak output voltages of 1.59 and 12.47 V, respectively. Their average negative peak output voltages were −0.86 and −4.13 V, respectively. The difference in the output voltages is explained on the basis of the difference in the mechanical properties of plain- and 2/1 twill-woven cotton textiles. The results of this study are expected to improve our understanding of triboelectric charge generation on cotton fiber surfaces and help in designing the structure of cotton textiles from the viewpoints of triboelectric engineering and wearable electronics.
J.J. (Jaebum Jeong) and J.-H.K. (Jin-Hyuk Kwon) performed the experiments, analyzed the experimental results, and wrote the paper; K.L., S.B., A.T., S.L. (Suwoong Lee), H.J.O., J.-H.K. (Jong-Hyoung Kim), J.K., D.-W.L., H.C., P.L., J.J. (Jaewon Jang) and S.L. (Sohee Lee) also contributed to the analysis of the experimental results; J.-H.B. and H.K. supervised the whole procedure, analyzed the data, and wrote the paper.
This work was supported by the Korea Institute of Industrial Technology as Research Source Technique Project \[KITECH EO180026\]; Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT \[2018R1A2B6008815\]; Korea Electric Power Corporation \[R19XO01-05\]; and the BK21 Plus project funded by the Ministry of Education, Korea \[21A20131600011\].
The authors declare no conflict of interest.
{#polymers-11-01443-f001}
{#polymers-11-01443-f002}
{#polymers-11-01443-f003}
{#polymers-11-01443-f004}
[^1]: These authors contributed equally to this work.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Matrix-assisted laser desorption/ionization (MALDI) is commonly used as a soft ionization technique to study a wide range of biomolecules. A powerful application of MALDI-mass spectrometry (MS) is the ability to determine the spatial distribution of molecules in a tissue slice by mass spectrometry imaging (MSI) ([@B5]). MALDI-MSI has been used to study a wide range of biomolecules, from small molecule metabolites ([@B33]; [@B17]), to neuropeptides ([@B9]), and intact proteins ([@B7]).
More recently, atmospheric pressure-MALDI (AP-MALDI) was introduced, increasing the ease of sample preparation and allowing for analysis of volatile molecules, as the sample no longer needs to be placed under vacuum prior to analysis ([@B31],[@B32]). Since then, AP-MALDI has been used to detect tryptic peptides ([@B47]; [@B20]; [@B42]), pesticides ([@B36]), oligosaccharides ([@B10]), and proteolytic fragments ([@B19]). Tandem MS has also been coupled with AP-MALDI ionization, which provides the ability to fragment molecules and use the fragmentation patterns to identify biomolecules ([@B38]; [@B23]). AP-MALDI is also capable of performing imaging experiments. The handling of samples at atmospheric pressure (AP) is an advantage of the technique, as shown by the imaging of lipids with a matrix that sublimes under higher vacuum ([@B24]). As lipids ionize readily, multiple studies have reported imaging lipids with AP-MALDI ([@B48]; [@B8]; [@B12]). Other applications of AP-MALDI-MSI include imaging of secondary metabolites in licorice rhizome ([@B34]) and neuropeptides in crustaceans ([@B8]).
Recent developments in MALDI-MSI have been directed at lowering the minimum spatial resolution. Lowering spatial resolution allows for increased resolution of fine molecular features and for single-cell MALDI-MSI ([@B2]). Although historically MALDI-MSI imaging experiments have been carried out above the low μm spatial resolution requirement for single cell imaging, recently instrument advances have lowered the minimum raster step size to allow imaging at higher spatial resolution ([@B57], [@B58]; [@B30]). AP-MALDI sources have also been introduced with optimized geometry to allow for high spatial resolution, including a scanning microprobe AP-MALDI (TransMIT GmbH, Griessen, Germany). The AP-SMALDI source has allowed for imaging of metabolites and lipids in a variety of samples at 5--20 μm spatial resolution ([@B48]; [@B3], [@B4]; [@B34]; [@B29]).
The AP-MALDI (ng) UHR system (MassTech Inc., Columbia, MD, United States) is another AP-MALDI source capable of imaging at high spatial resolutions. The source is compact, allowing for easy and fast switching between ESI and AP-MALDI. High spatial resolution is achieved through an Nd:YAG 355 nm laser with a laser spot size of 10 μm and a maximum output frequency of 10 kHz. In the source, the sample plate is approximately 2 mm away from the heated MS inlet capillary ([@B28]; [@B51]; [@B40]). The laser operates as a continuous raster along the rows of the sample. TARGET software (MassTech Inc., Columbia, MD, United States) controls the source settings. ImageQuest software (Thermo Fisher Scientific, Waltham, MA, United States) is used to correlate the XY coordinates to the MS spectra in the raw file to create a molecular map of target analytes distributed on the tissue section.
Matrix-assisted laser desorption/ionization-mass spectrometry imaging is now commonly used to investigate metabolite distribution in various plant tissues ([@B27]; [@B33]). In *Medicago truncatula* (Medicago), which forms a symbiotic relationship with rhizobia for biological nitrogen fixation, MALDI-MSI has been applied to root nodules to provide insight into the metabolites involved in biological nitrogen fixation ([@B55]; [@B18]). In addition to studying metabolites involved in biological nitrogen fixation, metabolite distribution changes in the root nodules due to stress can also be investigated with MALDI-MSI. Salt stress results in an energy cost to plants, as they reallocate more of their energy to physiological changes that allow continued function under stress ([@B41]). This cost of energy for plants under salt stress translates into economic costs to farmers due to reduced yields ([@B41]). The ability of legumes to form symbiotic root nodules is highly sensitive even to mild concentrations of salt that do not affect other aspects of plant growth ([@B50]; [@B56]). Consequently, identifying metabolite changes within symbiotic nodules under salt stress or non-stress conditions may help us to understand why this symbiosis is so strikingly affected by moderate levels of salt stress. Salinity tolerance in Medicago has been studied by adding NaCl and monitoring metabolite changes with activity assays ([@B35]). Gas Chromatography-MS has also been used to investigate the metabolic profile of severe salt stress ([@B11]).
Here, the ability of an AP-MALDI (ng) UHR source coupled to a high resolution accurate mass platform to study metabolites in Medicago root nodules will be investigated. As a stand-alone source that can attach to multiple instruments, the AP-MALDI (ng) UHR source is a promising alternative to a traditional dedicated MALDI source for labs that might not have the ability to obtain a dedicated MALDI platform. Thus, a study to compare the performance of the AP-MALDI (ng) UHR source to a traditional MALDI platform and to demonstrate the application of the source to investigate metabolite changes due to stress is a valuable evaluation of the performance of the source. Initially, optimized AP-MALDI MSI of root nodules was compared to MALDI-MSI of root nodules on a commercial MALDI LTQ Orbitrap XL system. The AP-MALDI system was then used to study the metabolic response to salt stress through imaging at high spatial resolution. This study analyzed the localization changes of metabolites in Medicago root nodules during salt stress.
Materials and Methods {#s1}
=====================
Materials
---------
2,5-Dihydroxybenzoic acid (DHB) was purchased through Acros Organics (Thermo Fisher Scientific), and α-Cyano-4-hydroxycinnamic acid (CHCA) through Sigma-Aldrich. Methanol, acetonitrile, chloroform, and formic acid were purchased through Fisher Chemical (Fisher Scientific). A Millipore system was used for double distilled water. Plain microscope slides were obtained from Fisher Scientific, and indium tin oxide coated glass slides from Delta Technologies.
Plant Growth
------------
Seeds of *M. truncatula* cv. Jemalong A17 were acid scarified, surface sterilized, and vernalized for two overnights at 4°C. Seedlings were germinated at room temperature and transferred to sterilized growth pouches which contained 10 ml of Modified Nodulation Medium (MNM) which was modified from Buffered Nodulation Medium (BNM) ([@B14]) with addition of 1 mM KCl containing 100 mM of sodium chloride. Plants grown only in MNM medium were used as controls. The pouches were placed in a transparent box in a growth chamber with 16 h light for 4 days. The roots were inoculated with 1 ml of *Sinorhizobium meliloti* (Rm1021) (OD~600~ of 0.1), grown for another 3 weeks and nodules were harvested for subsequent analysis.
MSI Sample Preparation
----------------------
Root nodules from control and high salt plants were trimmed from the plants with 2--4 mm of the surrounding root. Nodules were embedded in 100 mg/mL gelatin and frozen on dry ice. Nodules were sectioned at 16 um thickness on a Microm HM 525 cryostat (Thermo Fisher Scientific) at -20°C. Sections were thaw-mounted onto plain glass microscope slides for analysis on the MALDI LTQ Orbitrap XL or indium tin oxide coated glass slides for analysis on the AP-MALDI QE-HF system. A TM Sprayer (HTX Technologies, LLC, Carrboro, NC, United States) was used to apply DHB and CHCA matrix. DHB matrix (40 mg/mL in 50% methanol, 0.1% formic acid) was applied with a 24 pass TM Sprayer method (30 s dry time in between passes, 90° rotation between passes and the spacing offset in between every two passes, 3 mm spacing, 1250 velocity, 80°C temperature, and 0.05 mL/min flow rate). CHCA matrix (10 mg/mL in 70% acetonitrile, 0.1% formic acid) was applied with a 4 pass TM Sprayer method (30 s dry time in between passes, 90° rotation between passes and the spacing offset in between every two passes, 1.5 mm spacing, 1200 velocity, 75°C temperature, and 0.24 mL/min flow rate). Matrix covered samples were stored in a dry box at -20°C until analysis.
Vacuum MALDI MSI
----------------
Matrix-assisted laser desorption/ionization-mass spectrometry imaging was performed on a MALDI LTQ Orbitrap XL (referred to as MALDI) mass spectrometer (Thermo Fisher Scientific, Waltham, MA, United States) equipped with a nitrogen laser in positive ion mode. LTQ Tune software (Thermo Fisher Scientific, Waltham, MA, United States), and Xcalibur (Thermo Fisher Scientific, Waltham, MA, United States) were used to select the imaging region and step size and the instrument parameters, respectively. The laser energy for DHB was set at either 15 or 20 μJ (later replicates needed higher laser energy to get the same signal level as earlier replicates) and the laser energy for CHCA was set at 10 μJ. Imaging was performed on three biological replicates with technical replicates at 75 μm raster step size. The mass range was set to 100--1000 *m/z* and the resolution to 60,000. Two microscans were averaged at each pixel.
AP-MALDI MSI
------------
Atmospheric pressure-MALDI experiments were performed on an AP-MALDI (ng) UHR ion source (MassTech Inc., Columbia, MD, United States) coupled to a Q Exactive-HF (Thermo Fisher Scientific, Waltham, MA, United States). Initially, the S-lens RF value, capillary temperature, and spray voltage parameters were optimized on-tissue. Imaging experiments were conducted in positive ion mode for 100--1000 *m/z* with 60,000 resolution, two microscans, 1E6 AGC target, 100 ms maximum injection time, 3.25 kV spray voltage, 350°C capillary temperature, and 70% for the S-lens RF value. For DHB covered sections 40% laser energy was used, and for CHCA sections 25% laser energy was used on the AP-MALDI control software. Experiments were conducted at 30 μm raster size. TARGET ng software (MassTech Inc., Columbia, MD, United States) was used to set the imaging area, raster size, and laser energy. Tune software (Thermo Fisher Scientific, Waltham, MA, United States) was used to acquire data.
MSI Data Analysis
-----------------
MSiReader software ([@B45]) was used to create peak lists and generate images from the data. Briefly, the interrogated zone was drawn around the nodule and root and compared to the reference zone of a matrix only area. Each technical replicate was analyzed individually in MSiReader, and all data was normalized to the total ion current. For each nodule, *m/z* in more than 15% of the total area of the interrogated zone (the root and root nodule) and less than 5% of the total area of the reference zone (matrix area) were pulled out for the MALDI data sets. *M/z* in 10% of the interrogated zone and less than 5% of the reference zone were pulled out for AP-MALDI data. The analysis was performed using a ± 5 ppm window. Low numbers for the interrogated zone percentages were selected to ensure that peaks localized to a small region of the sample (and not just peaks localized to the entire sample) were pulled out. A low percentage for the reference zone was used to have a strict cut-off for removing matrix peaks. Different interrogated region percentages were used for the two platforms due to the difference in signal intensity and number of peaks pulled out between the two. For the MALDI system, 15% was used over 10% as using 10% pulled out many more noise peaks compared to 15%. Also, as the AP-MALDI system detected hundreds of peaks (compared to the over 2,000 mass spectral peaks detected by the MALDI system), the extra *m/z* were easier to manually verify for the AP-MALDI platform. As the AP-MALDI system produced fewer images, a lower cutoff threshold was used to generate as many good images as possible. Peak lists for biological replicates were generated by combining the technical replicate peak lists and combining duplicates (*m/z* within 5 ppm). Peak lists from the three biological replicates were combined and duplicates combined (5 ppm error) to create peak lists for the two platforms with each matrix. All peak lists were manually validated by visual inspection of the resulting ion images. Peak lists were imported into SCiLS software (Bruker, Bremen, Germany) along with the data for statistical analysis. Centroid data was imported with linear interpolation at a mass accuracy of 0.0005 Da for the mass axis settings. Data was normalized to the total ion current after importing, and all analysis were performed with normalization to the total ion current. The discriminative analysis was performed using receiver operating characteristic (ROC) on both the nodules and root together and the nodule and root separately. For the analysis, individual spectra from three biological replicate nodules were used for each class. The two classes were control and salt. As the control and salt nodules are not necessarily the same size, a random subset of 500 spectra for each class was used for the analysis. A 5 ppm interval width and the validated peak lists were used for the analysis. Hypothesis tests were performed on all individual spectra of the same three biological replicates as the ROC test. The entire root nodule and root area for the control and salt samples were used for the test. Specifically, the *t*-test was used with a 5 ppm window. The peak lists generated in MSiReader was again used for the test. Principal component analysis (PCA) was performed using the mean spectra of each region (each region was drawn around the root and root nodule from a technical replicate) with a 5.0 ppm interval width, five components, and unit variance scaling.
Sample Extraction
-----------------
Approximately 100 control nodules and 100 salt treated nodules (with 2--4 mm of surrounding root) were trimmed from the plants and flash frozen. Nodules were ground with a mortar and pestle under liquid nitrogen. A methanol/chloroform/water (Milli-Q) extraction was performed by adding in order three parts methanol (600 μL), one part chloroform (200 μL), and four parts water (800 μL). Samples were vortexed briefly and centrifuged for 10 min at 5000 × *g* and 4°C. The upper aqueous layer was removed, four parts methanol were added, and the extraction was vortexed briefly. The extraction was centrifuged again at 1500 × *g* for 5 min and 4°C. The supernatant (organic layer) was removed. The aqueous and organic fractions were dried down in a speedvac and saved in a -80°C freezer prior to analysis.
LC-MS/MS for Identifications
----------------------------
Aqueous samples were resuspended in optima grade water with 0.1% FA at 10 mg/mL. LC-MS/MS was performed with a Dionex Ultimate 3000 UHPLC system (Thermo Fisher Scientific, Waltham, MA, United States) equipped with a Kintex C18 column (2.1 mm internal diameter × 150 mm length, 1.7 μm particle size; Phenomenex, Torrance, CA, United States) with a corresponding guard column, and a Q Exactive MS (Thermo Fisher Scientific, Waltham, MA, United States). For separation, the column temperature was 35°C, and the mobile phases were optima grade water with 0.1% formic acid (A) and acetonitrile with 0.1% formic acid (B). A 35-min gradient at a flow rate of 0.3 mL/min with the following conditions was used: 0--5 min, held at 1% B; 5--10 min, linear gradient from 1--3% B; 10--18 min, linear gradient from 3--40% B; 18--22 min, linear gradient from 40--80% B; 22--27 min, column cleaning at 95% B; and 27--35 min, re-equilibration at 1% B. The injection volume was 4 μL, and the samples were kept at 10°C during analysis. The MS was operated in the positive ion mode with a scan range of *m/z* 100--1500 using a top five method for MS/MS. A target list, which included *m/z* more prevalent in either the control nodules or salt nodules, was used to acquire MS/MS on target *m/z*. If less than 5 *m/z* on the target list were found, then the most abundant *m/z* were chosen. The MS parameters were as follows: 70,000 resolution, 1E6 AGC, and 100 ms maximum injection time. The settings for HCD MS/MS were as follows: 35,000 resolution, 1E5 AGC, 100 ms max inject time, 15 s dynamic exclusion, and collision energies of 30, 35, and 40 for injections 1, 2, and 3, respectively. MetFrag ([@B46]) was used to analyze the MS/MS results by searching the \[M+H\]^+^, \[M+Na\]^+^, and \[M+K\]^+^ adducts against the KEGG database with 5 ppm error tolerance. The *in silico* fragmentation was matched up to the top 20--40 experimental fragments of the MS/MS spectra at a 5 ppm and 0.01 Da tolerances. MS/MS spectra from the mzCloud high-resolution MS/MS database was used where possible to validate the MetFrag identification. For mzCloud analysis, LC-MS/MS results were loaded into Compound Discoverer software (Thermo Fisher Scientific). Briefly, raw files were aligned with adaptive curve setting with 5 ppm mass and 1.0 min retention time tolerances. Unknown compounds were detected with a 5 ppm mass tolerance, three signal to noise ratio, and 1,000,000 minimum peak intensity, and then grouped with 5 ppm mass and 0.1 min retention time tolerances. A search against the mzCloud database was then performed against all activation types with a 25 activation energy tolerance, and the intensity threshold set to true. Identifications were made if the top result in MetFrag explained almost all the major fragments and there were no other strong results in the lower scoring MetFrag results (score less than 0.8 for all other hits) or if the top result in MetFrag explained almost all the major fragments, and the compound discoverer MS/MS for this compound matched almost exactly. Arginine, soyasaponin I, asparagine, and adenosine standards were obtained to verify identifications. MS/MS parameters were the same as described for the extractions.
Results
=======
AP-MALDI Parameter Optimization
-------------------------------
Initially, the AP-MALDI source and QE-HF MS parameters were optimized by metabolite profiling on tissue sections. The laser energy was optimized on matrix areas by increasing the laser energy until increasing the energy no longer increased yield of matrix ions. A wide range of S-lens, capillary temperature, and spray voltage values were tested by profiling on sections of control nodules. Ions initially enter the instrument through a heated capillary. The spray voltage is applied to the source, in this case to the plate, to assist ions into the MS. The S-lens is an ion guide behind the heated capillary consisting of a series of stacked rings that operates as a radio frequency (RF) device to capture and focus ions into a beam. Typically, larger molecules need a higher S-lens value to be efficiently transferred into the mass analyzer ([@B15]). The temperature on the heated capillary, the voltage on the plate, and the RF value on the S-lens, all play a role in optimal detection of ions. **Figures [1A--C](#F1){ref-type="fig"}** shows the profiling results from adjusting the S-lens, spray voltage, and capillary temperature with CHCA as the matrix. Adjusting the S-lens and spray voltage values did not reveal any clear trend during the profiling experiments, although most *m/z* had an increase in signal at 80% for the S-lens RF value. The capillary temperature showed a stronger trend, as increasing the temperature resulted in an increased signal. **Supplementary Figures [1A--C](#SM1){ref-type="supplementary-material"}** shows similar optimization graphs for optimizing instrument parameters with DHB as the matrix.
{#F1}
A smaller subset of instrument parameters was further tested by performing imaging experiments on control nodules. The base parameters were 70% S-lens, 3.0 kV spray voltage, and 300°C capillary temperature. Parameters were then individually adjusted above and below these values. **Figures [1D--F](#F1){ref-type="fig"}** shows the imaging results from adjusting the S-lens, spray voltage, and capillary temperature with CHCA as the matrix. **Supplementary Figures [1D--F](#SM1){ref-type="supplementary-material"}** shows the imaging optimization results with DHB. The S-lens value still showed inconsistent results as higher *m/z* tended to increase slightly with higher S-lens while lower *m/z* decreased slightly with higher S-lens. Thus, a middle value of 70% was chosen for future experiments. In the imaging experiments, increasing the spray voltage did tend to increase the signal, especially for higher *m/z* (above 600), which were not very noticeable with the lower spray voltages. Furthermore, increasing the capillary temperature increased signal, which was consistent with the profiling results. Parameters of 3.25 kV spray voltage and 350°C capillary temperature were chosen to give the best signal, especially for higher *m/z*. The same instrument parameters were chosen for both DHB and CHCA matrices as there was not a noticeable difference between the two matrices. Spray voltages above 3.25 kV were not attempted as higher voltages could cause discharge on the AP-MALDI electronics.
α-Cyano-4-hydroxycinnamic acid matrix showed higher overall signal and better coverage of the lipid region (*m/z* above 600) compared to imaging experiments with DHB as the matrix. In **Supplementary Figure [1](#SM1){ref-type="supplementary-material"}**, only *m/z* 104.1073 and *m/z* 133.0607 had a signal above 10,000, and this occurred at higher spray voltages and capillary temperatures. In the CHCA imaging experiments, all but the two highest *m/z* (682.0023 and 719.9575) had a signal above 10,000 at high capillary temperatures and spray voltages. In the DHB graphs in **Supplementary Figure [1](#SM1){ref-type="supplementary-material"}**, *m/z* above 600 were not included as there was minimal signal in this region with DHB matrix.
Comparison of AP-MALDI and Vacuum MALDI Sources
-----------------------------------------------
Imaging experiments using control nodules on both the AP-MALDI and MALDI platforms were conducted to compare the performance of the two instruments for MSI. **Figures [2A--H](#F2){ref-type="fig"}** compares the MALDI and AP-MALDI platforms. **Figure [2A](#F2){ref-type="fig"}** shows the number of *m/z* detected that resulted in good images (i.e., not matrix peaks and signal predominantly in plant sample) for control nodules in both the MALDI and AP-MALDI systems with both DHB and CHCA matrices. It should be noted that if a *m/z* is not detected in one platform but is detected in the other, it is likely the case that the *m/z* is present in both samples, but the signal intensity in one platform was below the threshold needed to pull out the *m/z* with MSiReader. Interestingly, DHB provided better results on the MALDI, and CHCA gave better results on the AP-MALDI system. While instrumentation differences could play a role in why different matrices were best for the two platforms, the fact that the CHCA matrix application method was a wetter method than the DHB method could also play a role. The wetter CHCA method could potentially extract metabolites form the tissue at a higher concentration. As the AP-MALDI had a lower signal than the MALDI platform, improved extraction could benefit the AP-MALDI platform more than the MALDI platform. The MALDI system detected significantly more *m/z* than the AP-MALDI system as it detected almost twice the number of *m/z* using CHCA and over six times the number of *m/z* using DHB as matrix, respectively. **Figures [2B,C](#F2){ref-type="fig"}** shows the overlap between the *m/z* detected in MALDI and AP-MALDI systems for DHB matrix and CHCA matrix. Both matrices had similar numbers of *m/z* shared between the two instruments. However, there were more unique *m/z* than shared *m/z*, especially for CHCA. **Figures [2D,E](#F2){ref-type="fig"}** compares the DHB and CHCA matrices for both the MALDI and AP-MALDI platforms. While both platforms have many *m/z* that are shared in both matrices, there is a high number of *m/z* only detected in one of the matrices, especially for DHB on the MALDI and CHCA on the AP-MALDI due to their higher number of detected *m/z* compared to the other matrix. Thus, the DHB and CHCA matrices were complementary to each other. The PCA plot in **Figure [2F](#F2){ref-type="fig"}** shows that the different matrix and platform conditions (MALDI DHB, MALDI CHCA, AP-MALDI DHB, AP-MALDI CHCA) all separate out into groups. The technical replicates group close together in most cases. While there is more variation in the biological replicates, the biological replicates from each matrix/platform experimental group are close enough together to separate them from the other matrix/platform experimental groups. **Figures [2G,H](#F2){ref-type="fig"}** shows example spectra for control nodules with DHB matrix for the MALDI (G) and AP-MALDI (H) platforms. The spectra show clear differences in *m/z* and intensity, which supports the separation of the different experimental conditions in the PCA plot. Example spectra averaged over control nodules with CHCA as the matrix for each platform are shown in **Supplementary Figure [2](#SM1){ref-type="supplementary-material"}**. **Supplementary Tables [1](#SM1){ref-type="supplementary-material"}, [2](#SM1){ref-type="supplementary-material"}** list the *m/z* unique to control nodules imaged with the AP-MALDI system for DHB and CHCA, respectively. These *m/z* were compared to the matches to the mzCloud database from the LC-MS/MS data. The putative identifications from this analysis are shown in **Supplementary Table [3](#SM1){ref-type="supplementary-material"}**. From these putative identifications, the AP-MALDI data is potentially detecting more acids as nicotinic acid/picolinic acid, pyroglutamic acid, aspartic acid, DL-α-aminosuberic acid, and pantothenic acid were all putatively identified from the *m/z* unique to AP-MALDI control nodules. Further MS/MS data collection and analysis would be necessary to verify the identity of these acids or identify additional compounds unique the AP-MALDI control nodules. The high number of unique *m/z* between the two sources is potentially due to the instrumental differences. One hypothesis for the differences is that the AP-MALDI had fewer in-source fragmentation products as other studies observed that AP-MALDI is a soft ionization method with decreased and more consistent fragmentation ([@B31]; [@B49]). As the MALDI detected many more *m/z* and had a higher overall signal compared to the AP-MALDI, the *m/z* solely detected in the MALDI experiments could be due to an increased sensitivity. Also, the MALDI has a nitrogen laser, which operates at 337 nm, whereas the AP-MALDI uses an Nd/YAG laser (355 nm). The differing beam profiles of these lasers ([@B22]) could be causing some of the differences in *m/z* detected between the sources. The different efficiencies of the two instruments could also affect the detected ions. As the two platforms have two different Orbitrap instruments, the differences in ion transfer, detection, and fragmentation efficiencies can potentially result in some of the observed differences.
{#F2}
Despite the lower number of shared *m/z* between the MALDI and AP-MALDI, the distributions of the shared *m/z* were similar. **Figure [3](#F3){ref-type="fig"}** compares the spatial distribution of *m/z* detected in both sources. Two representative images were selected from the shared *m/z* (see the Venn diagrams in **Figures [2B,C](#F2){ref-type="fig"}**) for each matrix. The *m/z* were chosen for their good normalized signal intensity in both platforms to make the images easy to compare, and an attempt was made to get an *m/z* spread evenly across the 100--1,000 range. For each *m/z*, the images acquired with the AP-MALDI have similar distributions as the images obtained on the MALDI system. However, the box and whisker plots showing the unnormalized intensity reveal a wide gap in the intensity of the signals between the AP-MALDI and MALDI. The optical images for the samples shown in **Figure [3](#F3){ref-type="fig"}** are shown in **Supplementary Figure [3](#SM1){ref-type="supplementary-material"}**, and the box and whisker plots for three biological replicates of the AP-MALDI control nodules, and three biological replicates of the MALDI control nodules are shown in **Supplementary Figure [4](#SM1){ref-type="supplementary-material"}**. After normalization to the total ion current, signals between the AP-MALDI and MALDI are much more comparable in the box and whisker plots. Although the overall signal on the AP-MALDI was lower than the MALDI, the AP-MALDI QE-HF instrument was still capable of determining the spatial distribution of small molecules in Medicago root nodules.
{#F3}
Metabolites Changing Due to Salt Stress
---------------------------------------
Overall, the quality of MS spectra obtained from salt nodules was consistent with the quality of MS spectra obtained from control nodules despite the abundance of sodium in the salt nodules. The total ion current was very similar between the control and salt nodules. For example, in one biological replicate the total ion current was 2.6E5 for control nodules versus 2.3E5 for salt nodules with CHCA and 9.6E4 for control nodules versus 4.0E4 for salt nodules with DHB. Example spectra for salt nodules with both matrices are shown in **Supplementary Figure [5](#SM1){ref-type="supplementary-material"}** (control nodule spectra are located in **Figure [2](#F2){ref-type="fig"}** for DHB and **Supplementary Figure [2](#SM1){ref-type="supplementary-material"}** for CHCA). The largest difference between the control and salt samples was the abundance of sodium adducts in the salt samples. The higher tolerance of MALDI systems to salt could potentially account for the fact that ion suppression due to the high salt concentrations in this study did not severely decrease the signal in the high salt samples. SCiLS software was used for statistical analysis of MSI data obtained from control and salt root nodules. ROC analysis was performed to generate area under the curve (AUC) values for specific *m/z*. ROC curves are generated by plotting the sensitivity (true positive rate) versus 100-specificity (false positive rate) for the ability of a single *m/z* value to discriminate between two conditions. AUC values, which range from 0 to 1, are calculated from the ROC curve for a specific *m/z*. By importing a peak list, an AUC value was generated from its respective ROC curve for each *m/z* in the list. An AUC cut-off of 0.75 was utilized as this resulted in a list of *m/z* that showed distinct differences between the control and salt nodules. As AUC values closest to 0.5 are less discriminative, an AUC value halfway in between 0.5 and 1 was chosen to give numerous *m/z* that were different between the control and salt root nodules. The ROC test was run on the entire nodule and root sample, just the nodule, and only the root to find *m/z* values that are discriminative to either the salt or control condition in specific regions of the sample (compared to *m/z* values present in the entire root and nodule sample). The discriminative analysis was compared to the *t*-test results, which were only performed on the entire root and nodule region as the *t*-test was less sensitive to the area selected. **Figure [4](#F4){ref-type="fig"}** compares the number of *m/z* selected from either control or salt samples using the three analysis methods: manual analysis, ROC analysis, or the *t*-test. The number of *m/z* by manual analysis was determined by looking through images for each biological replicate and selecting *m/z* that only showed signal in either the control or salt condition. The final number for the manual analysis only shows *m/z* selected in all three biological replicates.
{#F4}
In **Figure [4](#F4){ref-type="fig"}**, the results of the SCiLS analysis to find *m/z* solely in the control nodules (**Figures [4A--C](#F4){ref-type="fig"}**) and *m/z* strictly in the salt nodules (**Figures [4D--F](#F4){ref-type="fig"}**) are shown. The *t*-test found the highest number *m/z* specific to either the control or salt root nodules, with well over half of the input *m/z* having *p*-values less than 0.001. A number of these significant *m/z* did not appear to be changed in the images by naked eye, making it very hard to sort some *m/z* into either the control or salt group. The discriminative analysis test and manual analysis provided a more practical number of *m/z* to focus on. In most cases, the *m/z* selected in the manual and ROC analysis were found to be significant by the *t*-test. Differences between the manual and ROC analysis can likely be attributed to low signal in one or more biological replicates and inconsistencies in the manual sorting. Consequently, the ROC test was selected to look at the differences between the control and salt roots and root nodules. **Supplementary Table [4](#SM1){ref-type="supplementary-material"}** lists the *m/z* and AUC values for *m/z* with AUC \> 0.75 for the ROC analysis on the control roots and root nodules. **Supplementary Table [5](#SM1){ref-type="supplementary-material"}** provides the *m/z* and AUC values for *m/z* with AUC \> 0.75 in the ROC analysis on the salt roots and root nodules.
After combining the DHB and CHCA results, removing isotope peaks, and removing images with high background signal, 44 targets from control samples, and 77 targets from salt samples were selected. Overall, a minority of images with an AUC above 0.75 were removed due to high signal in the background. **Figure [5](#F5){ref-type="fig"}** shows representative images from control targets, and **Figure [6](#F6){ref-type="fig"}** shows images for selected targets from the salt treated root nodules. Most *m/z* with an AUC higher than 0.75 show signal uniformly distributed throughout the nodule or throughout the nodule and root. Only a couple of *m/z* values, which had an AUC higher than 0.75 just in the roots, did not show any distribution in the nodule. The images show distinct differences between the control and salt nodules with AUC's above 0.75, demonstrating the power of the ROC analysis.
{#F5}
{#F6}
Identification of metabolites from the LC-MS/MS data was performed by searching *m/z* against the KEGG database and using a combination of *in silico* fragmentation (MetFrag) and matching to the mzCloud high-resolution MS/MS database. For the MetFrag analysis, compounds that yielded theoretical fragments matching the highest number of fragments in the experimental MS/MS spectra were considered putative identifications. If more than one compound matched the major fragments, then an attempt was made to narrow down to one candidate with MS/MS spectra in the mzCloud database. **Table [1](#T1){ref-type="table"}** shows the identifications from the control list of *m/z* with AUC \> 0.75. Adenosine was the best option in MetFrag results, and nicotianamine was the only KEGG hit within 5 ppm for its *m/z* (the *in silico* fragmentation results did match well), but asparagine was the highest scored MetFrag result with two good options behind it. Glycylglycine was second but was ruled out with its MS/MS spectra in mzCloud. The third MetFrag result, *N*-carbamoylsarcosine, was not in mzCloud. The MS/MS spectra for asparagine in mzCloud was nearly identical to the experimental MS/MS, so it was putatively identified. Both asparagine and nicotianamine had AUC values higher than 0.75 in the nodules, while adenosine only had an AUC value higher than 0.75 in the roots, although it was also detected in the nodules. **Table [2](#T2){ref-type="table"}** shows the identifications in salt roots and root nodules. Arginine was detected with an AUC higher than 0.75 in the salt nodules with MS/MS that closely matched the database spectra in mzCloud. For *m/z* 365.1045, the AUC was very high in the root nodules, roots, and roots and nodules combined, but the MS/MS was only able to distinguish the *m/z* as the sodium adduct of a disaccharide as multiple sugars ranked very high in the MetFrag analysis. Soyasaponin I was also detected as a sodium adduct and interestingly was only located to the outer portion of the root nodules and in the roots in salt nodules. **Figure [7](#F7){ref-type="fig"}** shows the AP-MALDI images for the *m/z* identified in **Tables [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}**. The experimental MS/MS spectra for the identifications are shown in **Supplementary Figure [6](#SM1){ref-type="supplementary-material"}**. Arginine, soyasaponin I, asparagine, and adenosine experimental MS/MS spectra were compared to that of obtained standards for verification of the identification. The MS/MS spectra for the standards are in **Supplementary Figure [7](#SM1){ref-type="supplementary-material"}**. Retention times matched closely between the experimental data and the obtained standards (values are provided in **Tables [1](#T1){ref-type="table"}, [2](#T2){ref-type="table"}** and **Supplementary Figure [7](#SM1){ref-type="supplementary-material"}**).
######
Identifications from control roots and root nodules with AUC \> 0.75.
*m/z*; retention time (min) Distribution AUC \> 0.75 location Identification; adduct identified Literature molecular weight Delta ppm
----------------------------- ----------------- ---------------------- ----------------------------------- ----------------------------- -----------
133.0606; 1.05 Nodule Nodule and root Asparagine \[M+H\]^+^ 132.0535 -1.48
268.1034; 3.55 Nodule and root Root Adenosine \[M+H\]^+^ 267.0968 -2.53
304.1493; 1.05 Nodule and root Nodule Nicotianamine \[M+H\]^+^ 303.1430 -3.31
######
Identifications from salt treated roots and root nodules with AUC \> 0.75.
*m/z*; retention time (min) Distribution AUC \> 0.75 location Identification; adduct identified Literature molecular weight Delta ppm
----------------------------- ----------------------- ---------------------- ----------------------------------- ----------------------------- -----------
175.1186; 1.04 Nodule Nodule Arginine \[M+H\]^+^ 174.1117 -1.81
365.1045; 1.18 Root and nodule Root and nodule Disaccharide \[M+Na\]^+^ 342.1162 2.74
965.5076; 20.46 Root and outer nodule Root Soyasaponin I \[M+Na\]^+^ 942.5188 -0.44
{ref-type="table"}** **(B--D)** and **Table [2](#T2){ref-type="table"}** **(F--H)**. **(A)** Shows the optical image for the control identification **(B--D)** and **(E)** shows the optical images for salt identifications **(F--H)**. Images in **(B--D,F--G)** were with CHCA as the matrix and **(H)** was with DHB as the matrix. The white scale bar corresponds to 1 mm.](fpls-09-01238-g007){#F7}
Discussion
==========
Here, an AP-MALDI (ng) UHR source was utilized for imaging of Medicago root nodules at 30 μm spatial resolution. The spatial resolution provided by the AP-MALDI source is much higher than the conventional MALDI, which is 75 μm spatial resolution without oversampling. The AP-MALDI source is also compatible with multiple mass spectrometers. Here, a high-resolution accurate mass QE-HF Orbitrap instrument is utilized, offering even higher mass accuracy and resolution compared to the commercial MALDI system. Furthermore, the coupling of the AP-MALDI system to a high-resolution accurate mass Orbitrap system offers distinct advantage over commercial MALDI-TOF instruments, in terms of its high mass accuracy and resolution for confident identification of small molecule metabolites.
To maximize the *m/z* detected with the AP-MALDI source, parameters were carefully optimized. The parameters selected for imaging (high capillary temperature and spray voltage) maximized the detection of most *m/z* ions. However, even with the optimized parameters, the signal in the current AP-MALDI setup was at least one order of magnitude lower than the signal with the MALDI. A previous study comparing AP and vacuum MALDI on peptides and protein digests spots revealed that although signal increased twofold in the vacuum system, the noise level increased at a similar rate, resulting in a similar signal to noise ratios between the two ([@B47]). While a full limit of detection and signal to noise analysis was not conducted here, the MALDI detected significantly more *m/z* than the AP-MALDI, indicating a higher sensitivity for the MALDI system. The MALDI's superior performance regarding signal intensity and detection of *m/z* provides a powerful instrument for comprehensive analysis of tissue sections. However, the lower signal did not prevent imaging of many ions with the AP-MALDI system and its higher spatial resolution provides the ability to analyze samples with fine molecular features that may be difficult to resolve with the lower spatial resolution of the MALDI instrument. In addition, the 10 kHz laser on the AP-MALDI significantly increases the speed of image acquisition compared to the 60 Hz laser on the MALDI. A 50 × 70 pixel grid on the AP-MALDI took 26.60 min to image, resulting in 2.192 pixels/s acquisition speed. However, on the MALDI, a 29 × 34 pixel grid took 65.02 min, giving an acquisition speed of 0.2527 pixels/s. Thus, the AP-MALDI is more than eight times faster than the MALDI. To acquire the 50 × 70 grid of the AP-MALDI, the MALDI would take 230.6 min compared to the 26.60 min of the AP-MALDI. Therefore, the AP-MALDI has an advantage over the MALDI system regarding the speed of acquisition. Furthermore, as at best half of the *m/z* detected with the AP-MALDI were also detected with the MALDI, the AP-MALDI-MSI results are complementary to the MALDI imaging results. The AP-MALDI source allows for the detection of additional small molecules and potentially labile small molecules that are not compatible with vacuum MALDI sources. By performing MALDI-MSI studies with both sources, one could increase the coverage of the metabolome in MALDI-MSI studies.
The AP-MALDI QE-HF system was used to study the metabolite changes due to salt stress with high spatial resolution and high mass accuracy. SCiLS software was used to perform statistical analysis on the MSI data to confidently assign *m/z* discriminative to the control and salt conditions. Although the *t*-test (*p*-value \<0.001) gave the largest number of *m/z* as its output, the percentage of input *m/z* that were selected as significant was very high, and for some *m/z*, it was not apparent to the naked eye which group (either control or salt) was higher. Here, discriminative analysis using an ROC test was chosen as this test gave *m/z* with a signal that was consistently distinctive to either the salt or control group and mostly avoided *m/z* with only slight changes or changes in only one biological replicate. The discriminative analysis is also beneficial over manual analysis as it avoids potential inconsistencies in sorting. A random subset of spectra was used for the analysis as the salt nodules were typically smaller than the control nodules, meaning that using all spectra would result in a different number of spectra in each class. Although using multiple spectra per sample creates a large subset to generate ROC curves, it should be noted that individual spectra from the same sample are not independent. Furthermore, the ROC curve analysis and the *t*-test have two different meanings. The ROC curve is looking for *m/z* that discriminate between conditions (often healthy versus diseased tissue) whereas the *t*-test looking for *m/z* that have significant changes between the two conditions. While a *p*-value \<0.001 and an AUC \> 0.75 are not the same and provide different explanations about the data, the objective here was to compare their ability to select whichever *m/z* are changing between the conditions. As manual analysis is laborious, a statistical test to select changing *m/z* to focus identification efforts on is beneficial.
Previous studies have found changes in amino acids, organic acids, and sugars due to salt stress ([@B35]; [@B11]). Although sugars and amino acids were identified here as differing in the salt and control nodules, a potential pitfall of this study is that some of the metabolic differences observed could be due to the increased sodium levels in the salt samples. This creates a paradox observation where the same compound is higher in the control nodules for the \[M+H\]^+^ and \[M+K\]^+^ adducts, but higher in the salt treated samples for the \[M+Na\]^+^ adduct. The identifications of asparagine and nicotianamine in control nodule samples show this fluctuation as they had an AUC \> 0.75 in the control nodules for the \[M+H\]^+^ adduct but the *m/z* that accurate mass matched to the sodium adduct was higher in the salt target list (MS/MS data was not able to confirm presence in salt nodules). Similarly, the disaccharide sodium adduct was identified in the salt treated samples, but based on accurate mass matching, the potassium adduct was shown upregulated in the control target list. In addition, the inability of traditional MALDI-MSI to separate isobaric compounds prevented identification of different sugars. As a result, the changes in sugar content was difficult to determine as the changes in the availability of sodium for adduction and the isobaric nature of the sugars complicated assignment significantly. However, in most cases, one can still identify metabolites changing due to salt stress (and not due to the differences in sodium adduct formation). For example, both the \[M+H\]^+^ and \[M+Na\]^+^ adducts of arginine were on the salt target list, indicating that this change is due to the effects of the stress and not due to changes in sodium availability. Although some of the compounds with AUC \> 0.75 are likely due to changes in sodium levels and not due to the salt stress, there are still many targets discovered that do not show the relative intensity change between the control and salt nodules with different adducts (i.e., \[M+H\]^+^ adduct higher in control and \[M+Na\]^+^ higher in salt). These changes in the relative intensity between different adduct species can be determined by looking at the images for the control and salt nodules on the same intensity scale for each adduct species. Thus, AP-MALDI-MSI provides a viable technique to study metabolite changes in salt stress in Medicago nodules.
We observed increased accumulation of arginine in the salt-stressed nodules. Accumulation of arginine is often seen in plants subjected to various environmental stresses, and exogenous arginine helps to tolerate the harmful effects of salt stress ([@B44]; [@B25]). Arginine metabolism plays a crucial role in salt tolerance in plants as discussed below. Arginine is synthesized from the non-proteinogenic amino acid ornithine. *N*-acetylglutamate synthase (NAGS) is an enzyme that catalyzes the first reaction during ornithine biosynthesis, and overexpression of the gene encoding NAGS improves salt tolerance in tomato plants ([@B52]). Arginase catalyzes the initial reaction of arginine degradation, and a loss of activity of this enzyme is associated with increased salt tolerance, presumably via accumulation of beneficial molecules, such as, nitric oxide (NO) and polyamines ([@B39]; [@B52]; [@B37]). Ornithine δ-aminotransferase, another enzyme involved in arginine catabolism shows increased activity under salt stress ([@B52]). The arginine decarboxylase (ADC) enzyme converts arginine to agmatine, which is a precursor of polyamines. Spermine is a polyamine often involved in salt tolerance, and its deficiency leads to salt hypersensitivity ([@B54]). Spermine accumulation is low in salt-treated roots in a genetic background where arginine decarboxylase activity is reduced compared to the wild-type, implicating this enzyme in salt-acclimation ([@B26]). In salt-tolerant rice, expression of the *ADC* gene is induced in the presence of salinity ([@B6]). Single nucleotide polymorphisms (SNPs) associated with *ADC* showed a strong correlation with multiple environmental factors, such as, salinity, drought, and soil nitrogen, placing this enzyme as an essential regulator of plant-environment interactions ([@B21]). Arginine is also involved in the production of NO with the latter implicated in salt tolerance ([@B16]; [@B13]). Exogenous NO, in the form of its donor S-nitroso-*N*-acetylpenicillamine (SNAP), alleviates the adverse effects of salt stress, presumably by upregulating Reactive Oxygen Species (ROS)-scavenging enzymes and enhancing the accumulation of osmolytes ([@B1]). It is suggested that the accumulation of NO and other Reactive Nitrogen Species (RNS) cause nitrosative stress, which is essential for salt "priming" ([@B39]). Altogether, these results suggest an essential position of arginine metabolism in salt stress responses.
We also found an enhanced accumulation of soyasaponin I. Saponins are amphipathic glycosides found in many plant species ([@B43]). A salt-tolerant genotype of soybean accumulates high amounts of group B saponin, alluding to its possible role in salt tolerance ([@B53]). These findings validate our technique and demonstrate that it can be used to address significant biological questions.
Here, the AP-MALDI-MSI analysis of metabolites in salt stress demonstrated the ability of the AP-MALDI (ng) UHR source to image metabolites with high resolution in both mass and space. Despite the lower number of detected compounds due to a reduced sensitivity compared to the vacuum MALDI MS platform, a respectable number of *m/z* values were found to change in root nodules between the control and salt conditions. The spatial resolution used here was not quite at the level of single-cell imaging, but with further optimization higher spatial resolutions could be achieved as the source has the potential for 5--10 μm imaging. Overall, the AP-MALDI QE-HF platform is a robust system for analyzing small molecules, and when combined with the ease of changing between AP-MALDI and ESI on a single mass spectrometer, the source makes for a useful alternative to a traditional dedicated MALDI instrument. The custom-designed source is a cost-effective substitute for a traditional MALDI platform, allowing labs to perform imaging experiments on mass spectrometers currently used with ESI. Furthermore, the complementary detection of *m/z* between the AP-MALDI and MALDI platforms allows for wider coverage of metabolites. On-going development for a new generation of a sub-AP-MALDI source from MassTech will offer improved sensitivity, and with continued ease of switching between ESI and MALDI operation, would allow for more comprehensive metabolome characterization of these important model systems.
Author Contributions
====================
CK performed all the experiments, sample preparation and analysis, and wrote the manuscript. SC optimized the salt stress treatment. JM grew the plants in all conditions. DJ, MS, JH, J-MA, and LL developed the research project and wrote the manuscript.
Conflict of Interest Statement
==============================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Funding.** This work was supported in part by funding from the National Science Foundation (NSF) Division of Integrative Organismal Systems (IOS) RESEARCH PGR award \#1546742, University of Wisconsin--Madison Graduate School and the Wisconsin Alumni Research Foundation (WARF), a Vilas Distinguished Achievement Professorship to LL and a NSF grant to J-MA (NSF\#0701846). LL acknowledges funding support from NIH through grants R56MH110215 and R01 DK071801. The MALDI LTQ Orbitrap XL and Q Exactive instruments were purchased through an NIH shared instrument grant (NCRR S10RR029531 to LL).
The authors would like to acknowledge MassTech Inc., for providing the AP-MALDI (ng) source.
Supplementary Material
======================
The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpls.2018.01238/full#supplementary-material>
######
Click here for additional data file.
[^1]: Edited by: Zhibo Yang, The University of Oklahoma, United States
[^2]: Reviewed by: Thanh Duc Do, University of Illinois at Urbana-Champaign, United States; Pietro Franceschi, Fondazione Edmund Mach, Italy; Cheng-Chih Hsu, National Taiwan University, Taiwan
[^3]: This article was submitted to Plant Metabolism and Chemodiversity, a section of the journal Frontiers in Plant Science
| {
"pile_set_name": "PubMed Central"
} |
Public health concerns over the potential for a devastating influenza pandemic in the near future are well known. Surveillance efforts have increased throughout the world, and much time and money have been directed toward preparedness for such a pandemic. Given that vaccination rates vary greatly among the nonmilitary population and that influenza diagnostics are sporadically available, annual influenza vaccine effectiveness studies based on laboratory-confirmed diagnoses are rare. However, evidence of locally circulating strains evading the vaccine-induced protection could be critical for early recognition and intervention. In addition, the emergence of pandemic strains within military populations has been noted. The first documented influenza outbreak in the spring of 1918, before the great influenza pandemic of 1918--19, was among recruits at Fort Riley, Kansas ([@R1]). In 1976, a unique strain of influenza (H1N1) caused an outbreak at Fort Dix, New Jersey, causing 1 death, and creating concern over spread of this nonvaccine strain ([@R2]). Highly vaccinated military populations, under close surveillance, provide the opportunity for annual calculation of influenza vaccine effectiveness, thereby benefiting global pandemic preparedness.
The Study
=========
The Naval Health Research Center (NHRC) began conducting tri-service surveillance for febrile respiratory illness at military training centers in 1996; by 1999, this surveillance network had expanded to include 8 of the largest military basic training centers in the United States ([@R3]). This surveillance includes the systematic collection of throat swab specimens and clinical data (including but not limited to gender, date of birth, symptoms, influenza vaccination status, type of vaccine received, and date of vaccination) from consenting US military trainees meeting the case definition for febrile respiratory illness (oral temperature ≥100.5°F \[38.0°C\] and a cough or sore throat). Samples are stored locally at each site at −70°C until they are forwarded to the Naval Respiratory Disease Laboratory at NHRC for viral culture and molecular diagnostic processing. Research personnel at participating surveillance sites report the weekly number of trainees who sought care for febrile respiratory illness and total trainee populations for their respective sites, and rates for such illnesses are calculated.
During the 2003--04 influenza season, we recognized the opportunity of using data from this ongoing active surveillance to estimate influenza vaccine effectiveness in protecting against both laboratory-confirmed influenza and febrile respiratory illness of any cause among US military basic trainees. Despite concerns that vaccine effectiveness during the 2003--04 season would be low because of the poor match between the components of the vaccine and the circulating strain ([@R4]), the vaccine provided good protection (94.4%) against laboratory-confirmed influenza that season ([@R5]). Annual vaccine effectiveness calculations are important as we heighten our preparedness for pandemic influenza strains; therefore, we performed similar calculations for the 2004--05 and 2005--06 seasons.
During the late fall and winter seasons, all active-duty military forces are required to receive the influenza vaccine, and this policy is strictly enforced in training camps. Upon arrival, all incoming trainees receive mandatory influenza vaccination, either the trivalent inactivated influenza vaccine by injection (FluZone, Sanofi Pasteur, Lyon, France) or intranasal cold-adapted, live, attenuated influenza vaccine (CA-LAIV) spray (FluMist, MedImmune, Gaithersburg, MD, USA).
For this analysis, vaccine protection was assumed to begin 14 days postvaccination. Therefore, in an 8-week training program, 25% of trainees were considered "unvaccinated" at any given time, assuming immunity takes 14 days to develop. Likewise, 33% of trainees in a 6-week training program were considered unprotected by the vaccine at any time. These assumptions allow estimates of denominator data for "vaccinated" and "unvaccinated" person-weeks in calculations of vaccine effectiveness..
From January through March 2006 all new trainees arriving for basic training received the influenza vaccine; all recruits already present had been vaccinated. The observation period for this analysis included January 1---March 31, 2006. However, 2 sites, Naval Service Training Command, Great Lakes, and Marine Corps Recruit Depot, San Diego, had completed vaccination by December 2005. Therefore, December was included in the observation period for those sites as well. Total person-weeks in recruit training during the observation period were obtained directly from the participating training centers. Vaccine effectiveness was calculated for both laboratory-confirmed influenza and any cause of febrile respiratory illness as follows: 100 × (1 -- relative risk = 1 -- \[rate in vaccinated group\]/\[rate in unvaccinated group\]).
During the observation period, 6 of 8 surveillance sites had influenza activity and were included in this analysis. In 479,181 person-weeks of observation, 4,052 cases of febrile respiratory illness were reported from these 6 sites, and 722 patients were enrolled into the surveillance study (includes throat swab specimen, case data, and consent). Seventy (9.7%) specimens tested positive for influenza, by either culture or molecular techniques.
Rates of laboratory-confirmed influenza were higher among unvaccinated trainees at all sites except Fort Benning, Georgia, which had only 3 cases ([Figure](#F1){ref-type="fig"}). Overall, influenza vaccine effectiveness among US military trainees was 92% (confidence interval \[CI\] 85.4--95.6%) during the 2005--06 season ([Table](#T1){ref-type="table"}). Vaccine effectiveness against laboratory-confirmed influenza was high (range 86%--94%) in each of the past 3 seasons. Vaccine effectiveness against non--laboratory-confirmed febrile respiratory illness was lower, ranging from −10% in 2005--06 to 52% in 2004--05.
{#F1}
###### Vaccine effectiveness against laboratory-confirmed influenza among US military basic trainees, 2005--06\*†
Site Vaccinated person-weeks Unvaccinated person-weeks Cases in vaccinated trainees Cases in unvaccinated trainees Vaccine effectiveness (%) 95% CI
---------------------- ------------------------- --------------------------- ------------------------------ -------------------------------- --------------------------- ----------------
Fort Jackson, SC 77,874 25,958 7 13 82.1
Fort Wood, MO 67,513 22,504 2 11 93.9
Fort Benning, GA 68,652 22,884 3 0 --
Lackland AFB, TX 37,435 18,690 1 10 95.0
NSTC Great Lakes, IL 67,763 22,588 0 13 100.0
MCRD San Diego, CA 35,490 11,830 0 10 100.0
Total 354,727 124,454 13 57 92.0 (85.4%, 95.6%)
\*CI, confidence interval; SC, South Carolina; MO, Missouri; GA, Georgia; AFB, Air Force base; TX, Texas; NSTC, Naval Service Training Command; IL, Illinois; MCRD, Marine Corps Recruit Depot; CA, California.
†Assuming 14 d before vaccine is protective.
Conclusions
===========
This analysis suggests that the 2005--06 influenza vaccine was highly effective in protecting US military basic trainees against laboratory-confirmed influenza. Furthermore, these data suggest that both the trivalent inactivated vaccine injection and the CA-LAIV intranasal spray were equally effective, because the Marine Corps Recruit Depot in San Diego vaccinated its trainees with CA-LAIV almost exclusively, and vaccine effectiveness at that site was 95% (vaccine effectiveness at all other sites combined = 90%).
These estimates of effectiveness were supported by results of additional analyses that would be expected to bias the outcome toward the null hypothesis. For example, a 7-day lag period before immune response was considered in an alternative analysis, and it yielded similar results: the calculated vaccine effectiveness changed only slightly, from 92% to 90%. We also analyzed vaccine effectiveness, assuming that 10% fewer trainees were vaccinated at any given point, yet the calculated vaccine effectiveness was only reduced to 87%.
In contrast to the consistently high effectiveness of the vaccines against laboratory-confirmed influenza, the effectiveness against febrile respiratory illness of any cause was much lower and varied with each season (13.9% in 2003--04, 52.1% in 2004--05, and −10% in 2005--06). This lower effectiveness in 2005--06 is most likely due to the generally high proportion of adenovirus infection seen in this population ([@R6]), and the lesser effectiveness is further exacerbated by the tendency for adenoviral infections to occur beyond the second week of training. The lower vaccine effectiveness seen against febrile respiratory illness of any cause gives credence to the estimates of high vaccine effectiveness against laboratory-confirmed influenza. If a measurement bias existed, both estimates would be affected.
As a highly vaccinated population, military personnel, and basic trainees in particular, can provide critical information regarding the effectiveness of each year's influenza vaccine formulations. Because of the annual variations of both the vaccine formulations and the circulating strains, influenza vaccine effectiveness should be evaluated annually. With the ever-rising concerns of an imminent influenza pandemic, reliable and rigorous influenza surveillance is paramount. Our existing surveillance network will allow us to repeat the methods used in this analysis each year, thus providing valuable estimates of influenza vaccine effectiveness to the public health community.
*Suggested citation for this article*: Strickler JK, Hawksworth AW, Myers C, Irvine M, Ryan MAK, Russell KL. Influenza vaccine effectiveness among US military basic trainees, 2005--06. Emerg Infect Dis \[serial on the Internet\]. 2007 Apr \[*date cited*\]. Available from <http://www.cdc.gov/eid/content/13/4/617.htm>
Contributions from the following persons are gratefully acknowledged: Viola Paulk, Laura Pacha, Sharon Cole-Wainwright, Johnnie Conolly, R.J. Newsom, Robert Greenup, Susan Wolf, Shelly Oates, Mimms Mabee, John Gomez, Patricia Rohrbeck, Lorie Brosch, Edgar Tuliao, Josephine Genese, Annie Wang, Richard Skinner, staff from the Naval Respiratory Disease Laboratory, Naval Health Research Center; the Department of Defense Global Emerging Infection Surveillance and Response System; and the Henry M. Jackson Foundation for the Advancement of Military Medicine. Finally, we thank Gregory Gray for his original leadership of recruit respiratory infection surveillance.
Ms Strickler has coordinated epidemiologic studies for the Department of Defense Center for Deployment Health Research at Naval Health Research Center since 2000. Her research interests focus on respiratory illness among military populations.
| {
"pile_set_name": "PubMed Central"
} |
The author contributions are incorrect. The correct author contributions should read as follows: Conceived and designed the experiments: XY RGS QY. Performed the experiments: XY MA, FS. Analyzed the data: XY RKMK PT. Contributed reagents/materials/analysis tools: RKMK FS KY LDM PT. Wrote the paper: XY QY.
| {
"pile_set_name": "PubMed Central"
} |
Neurons in the hippocampal formation exhibit a variety of spatially tuned firing patterns. The mechanisms by which these different patterns emerge are not fully resolved, although competing computational models exist for several of them. Here we present a new model that can generate all observed spatial firing patterns by a single mechanism. The model consists of a feedforward network with a single output neuron. Its essential ingredients are i) spatially tuned excitatory and inhibitory inputs \[e.g., 1\] and ii) interacting excitatory and inhibitory Hebbian plasticity. The inhibitory plasticity homeostatically controls the output firing rate by balancing excitation and inhibition \[[@B2]\]. We show in simulations and by a mathematical analysis that the output neuron develops periodic firing patterns along a stimulus dimension if inhibitory inputs are more broadly tuned than excitatory inputs along this dimension. More generally, depending on the relative spatial auto-correlation length of the excitatory and inhibitory inputs, the model exhibits firing patterns that are similar to those of place cells, grid cells (see Figure 1) or band cells (neurons that fire on spatially periodic bands \[[@B3]\]). For inputs with combined spatial and head direction tuning, the same mechanism leads to output firing patterns reminiscent of head direction cells and conjunctive cells (neurons that fire like grid cells in space but only at a particular head direction). A linear stability analysis of the homogeneous steady state accurately predicts the spatial periodicity obtained from simulations. The model combines the robust pattern formation of attractor models \[e.g., 4\], with the spatial (rather than neural) structure formation of models based on synaptic plasticity \[[@B5]\]. In contrast to attractor models \[[@B6]\], our model predicts that the grid spacing should be robust to global modifications in inhibitory synaptic strength, a distinction which could be experimentally verified.
In conclusion, we propose a feedforward network model that generates all known spatial firing patterns in the hippocampal formation through a single self-organizing mechanism.
{#F1}
Acknowledgements
================
Funded by the German Federal Ministry for Education and Research, FKZ 01GQ1201.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Facilitating recovery post stroke is an important goal of rehabilitation \[[@B1],[@B2]\]. The recovery patterns over time and outcomes at specific time points are variables which are often investigated in research into stroke rehabilitation \[[@B3]-[@B5]\]. In addition research in this field also investigates variables such as process of rehabilitation \[[@B4]\] and the content and intensity of rehabilitation \[[@B5],[@B6]\]. These variables have been investigated in a number of different settings and contexts where the rehabilitation approaches could differ.
The body of literature on stroke rehabilitation clearly establishes that there are different approaches to stroke rehabilitation in developed \[[@B4],[@B5]\] and developing countries \[[@B7]-[@B10]\]. The differential approaches do not necessarily represent preferred models or best practice. Although rehabilitation in developed countries tends to follow recommended stroke rehabilitation guidelines \[[@B9],[@B11]\], in developing countries rehabilitation provided is often dependent on the availability of resources, which are often limited \[[@B10]\].
The majority of studies investigating recovery and outcome after stroke have been conducted in developed high-income countries \[[@B4],[@B5],[@B12]\]. It emerges that admission to in-patient rehabilitation facilities is the norm in developed countries \[[@B9],[@B13]\] where factors that enhance the outcome of rehabilitation are more readily available. In contrast, only limited literature is available that reports empirically on outcomes and recovery patterns of stroke patients living in developing or under-resourced countries such as Africa, including South Africa. The lack of resources often results in the lack of adherence to global best practice guidelines, which could influence the outcomes and recovery post stroke. Rehabilitation at out-patient facilities is more common in developing countries. Also a multidisciplinary approach is seldom applied because some disciplines are not being employed at the centres \[[@B14]\].
The differences in stroke rehabilitation between developed and developing countries are readily acknowledged in common perception, but are not empirically supported. Research into rehabilitation has maintained the binary construct of stroke rehabilitation, by focusing separately on treatment aspects in developed and developing countries. The typical outcomes and recovery patterns across both types of country are seldom directly compared. According to our knowledge, there is no documented information comparing the outcomes of stroke rehabilitation in developed (well-resourced) and developing (low-resourced) countries.
The aim of the present study is to compare motor and functional recovery patterns, as well as functional outcomes, in stroke patients receiving rehabilitation in different treatment regimes relating to the process of rehabilitation in centres from developed and developing countries. This comparison could provide empirical support for the perceived differences in the recovery and outcomes of stroke patients living in developed and developing countries.
Methods
=======
Patients and settings
---------------------
Patients were included from two previous studies, which represent stroke rehabilitation in developing and developed countries. Out-patient rehabilitation offered at Community Health Centres in South Africa is an example of rehabilitation in developing countries, and an in-patient rehabilitation centre in Germany is an example of rehabilitation in developed countries. The primary motivation for selecting these settings is that they represent the typical practice model in developing and developed countries respectively.
The German cohort was selected from the CERISE study (Collaborative Evaluation of Rehabilitation in Stroke across Europe) that compared stroke care and recovery patterns in four European rehabilitation centres \[[@B4]\]. In-patient multidisciplinary care was provided in a stroke rehabilitation unit (SRU). Patients were recruited consecutively, using the following inclusion criteria: first-ever stroke as defined by WHO \[[@B15]\], age 40 to 85 years, and Rivermead Motor Assessment scores; \[[@B16]\] gross function (RMA-GF) ≤ 11, and/or leg and trunk function (RMA-LT) ≤ 8; and/or arm function (RMA-A) ≤12 on admission to the centre. Exclusion criteria were: other neurological impairments with permanent damage; stroke-like symptoms attributable to subdural hematoma, tumour, encephalitis or trauma; admission to the centre \> 6 weeks after stroke; no informed consent; and pre-stroke Barthel Index \[[@B17]\] \<50. The sample in this study comprised 135 patients. Figure [1](#F1){ref-type="fig"} illustrates that in the German (GE) sample, two patients were lost-to-follow-up at two months (one died and one refused) and another five at six months after stroke (two died and three refused).
{#F1}
The South African study (SA) \[[@B18]\] aimed at documenting the recovery patterns of 100 patients with stroke who were admitted to 21 Community Health Centres (CHCs) in the Cape Town Metropolitan district of the Western Cape Province. Physiotherapy services were provided at all the CHCs (n = 21), while occupational therapy and speech therapy services were provided in 16 and two CHCs respectively. Patients were recruited consecutively from the 21 CHSs using identical criteria for inclusion and exclusion to those in the CERISE study, except for age. In South Africa, stroke occurs in a younger population \[[@B14]\], therefore the age range was widened to 35--85 years in order to include a representative patient sample in the South African CHC study. In this sample, 10 patients were lost-to-follow-up at two months (four died, three refused and three could not be traced), and another seven patients were lost at six months after stroke (two died, one refused and four could not be traced) see Figure [1](#F1){ref-type="fig"}.
For inclusion in the present study, matched patient pairs were identified from the South African and German groups. Patients were matched on age at stroke onset (plus/minus 5 years), gender, and RMA-GF score (plus/minus 1 point) on admission. The final sample comprised 73 matched pairs.
Participant assessment
----------------------
On admission to the SRU or CHC, patients' age, gender, urinary incontinence (defined as a score \<10 on item 'bladder' of the BI), aphasia (defined as a score \>0 on item 9 of the National Institute of Health Stroke Scale, (NIHSS) \[[@B19]\] (and dysarthria (defined as a score \>0 on item 10 of the NIHSS) were assessed. In addition, motor and functional recovery was assessed at admission to the SRU or CHC, and at two and six months after stroke onset, using the Rivermead Motor Assessment \[[@B16]\] (sections; gross motor function = RMA-GF, leg and trunk = RMA-LT and arm = RMA-A) and the Barthel Index \[[@B19]\] respectively. A six-month follow-up period was used because the majority of motor recovery occurs before that time point \[[@B20]\]. A researcher in the European centre collected the data. At the start of the study, the researcher was trained in the assessments during a workshop. A manual was provided to ensure standardization. The project manager (LDW) visited the centre four times to recalibrate the researchers' work. The same training and recalibration method was provided to the South African researcher (AR).
Statistical analysis
--------------------
Demographic and clinical data of both matched patient samples was presented in terms of means with standard deviations, medians with interquartile ranges, or frequencies with percentages, as appropriate. The Mc-Nemar test was used to compare the prevalence of urinary incontinence, aphasia and dysarthria with both matched samples, while the T-test for dependent samples was used to test for significant differences based on age.
The Wilcoxon rank sum test was used to assess significant differences in motor and functional ability (scores) between the South African and the German patient sample at admission to the centre, two and six months after stroke. In addition, motor and functional recovery patterns were compared between both patient samples, using a generalized linear mixed methods model (GLIMMIX). GLIMMIXs are mixed models that can be used with discrete outcomes. Such models correct for the correlation between repeated observations with subjects. They also provide valid inferences for missing observations, provided that their absence does not depend on unobserved outcomes (i.e. assuming missingness at random) \[[@B21]\]. It should be further noted that GLIMMIX models compare the steepness of the recovery slope between the two groups taking into account the longitudinal study design, whereas the Wilcoxon rank sum test only compares outcomes at a certain point in time, not taking into account the patients' initial scores.
The models were fitted with the GLIMMIX procedure. All statistical analyses were performed using SAS, version 9.2, and tested for significance at a 0.05 alpha level (p \< 0.05).
Ethics
------
Ethical clearance for the South African study was obtained from the University of the Western Cape's Senate Ethics Committee and for the German study from the ethics committee of the German Rehabilitation Centre where the study was conducted.
Results
=======
Participants
------------
Demographic and clinical characteristics were presented and compared between both samples (Table [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). No significant differences existed for age, gender and aphasia on admission. Urinary incontinence (p = 0.03) and dysarthria (p = 0.02) occurred significantly more in the German than in the South African sample.
######
Comparisons of clinical data between matched South-African (SA) (n = 73) and German (GE) patient samples (n = 73)
**Parameters** **SA n = 73** **GE n = 73** **p-value**
----------------------------- --------------- --------------- -----------------
Age in years: mean (SD) 63.4 (10.0) 63.9 (9.2) 0.15^a^
Gender: male: n (%) 28 (38.4) 28 (38.4) 1.00^b^
Female: n (%) 45 (61.6) 45 (61.6)
Urinary incontinence: n (%) 11 (15.1) 22 (30.1) **0.03**^**b**^
Dysarthria: n (%) 25 (34.2) 39 (53.4) **0.02**^**b**^
Aphasia: n (%) 15 (20.5) 20 (27.4) 0.35^b^
TSO median days (q1 -- q3) 21 (15--31) 20 (16--27) 0.32^c^
^a^paired T-test, ^b^Mc-Nemar test, ^c^Wilcoxon signed rank test, TSO: time since stroke onset, IQR: Interquartile range, p-values \< 0.05 are in bold.
######
Comparisons of admission, two and six months post stroke data between matched South-African (SA) (n = 73) and German (GE) patient samples (n = 73)
**Parameters** **SA** **GE** **p-value**
------------------------------------------ ------------- -------------- ---------------------
RMA-GF
On admission to CHC/SRU: median (IQR) 8 (4--11) 8 (4--10) 0.28^c^
At two months after stroke: median (IQR) 11 (6--11) 10 (6--11) **0.03**^**c**^
At six months after stroke: median (IQR) 11 (8--11) 11 (9--12) **0.03**^**c**^
RMA-LT
On admission to CHC/SRU: median (IQR) 5 (3--7) 7 (5--8) **0.0001**^**c**^
At two months after stroke: median (IQR) 7 (3--8) 9 (7--9) **\<0.0001**^**c**^
At six months after stroke: median (IQR) 7 (5--9) 9 (7--10) **\<0.0001**^**c**^
RMA-A
On admission to CHC/SRU: median (IQR) 4 (1--9) 6 (1--10) 0.09^c^
At two months after stroke: median (IQR) 8 (2--11) 9 (3--14) **0.03**^**c**^
At six months after stroke: median (IQR) 8 (1.5-11) 12 (5--14) **0.0003**^**c**^
BI
On admission to CHC/SRU: median (IQR) 65 (50--80) 80 (45--90) 0.05^c^
At six months after stroke: median (IQR) 85 (65--95) 95 (80--100) **0.003**^**c**^
^c^Wilcoxon signed rank test.
SA indicates South-African patient sample, GE: German patient sample; CHC: Community Health Centre; SRU: Stroke Rehabilitation Unit; RMA-GF: Rivermead Motor Assessment- Gross Function, RMA-LT: Rivermead Motor Assessment- Leg and Trunk; RMA-A: Rivermead Motor Assessment Arm Function, BI: Barthel Index, IQR: Interquartile range, p-values \< 0.05 are in bold.
Motor and functional outcome
----------------------------
On admission to the CHC/SRU, RMA-GF scores did not differ significantly between the two samples. At two and six months after stroke, RMA-GF scores differed significantly between patient samples, with higher scores for the South African patients at two months after stroke, and higher scores for the German patients at six months after stroke. On admission to the CHC/SRU, at two and six months after stroke, the RMA-LT scores were significantly lower in the South African than in the German sample. No significant differences existed in RMA-A and BI-scores on admission to the CHC/SRU. At two and six months after stroke, both the RMA-A and BI-scores were significantly lower in the South African than the German sample.
Motor and functional recovery patterns
--------------------------------------
The results of the GLIMMIX modelling are shown in Table [3](#T3){ref-type="table"}, and the least square means are visually presented in Figures [2](#F2){ref-type="fig"}a,b,c and d. For the RMA-GF, the interaction term *'time\*center'* was found to be significant (p = 0.006), indicating that the RMA-GF recovery slope was significantly steeper in the German than the South African sample. For the RMA-A, the significant interaction term *'time\*center'* was found to be significant (p = 0.01), indicating that the RMA-A recovery slope was significantly steeper in the German sample than the South African one. The interaction term *'time\*center'* proved not be significant in the RMA-LT (p = 0.07) and the BI-model (p = 0.35) indicating that the recovery slopes for both RMA-LT and BI did not differ significantly between patient samples.
######
Results of the GLIMMIX modelling
**RMA-GF** **RMA-LT** **RMA-A** **BI**
------------------ -------------- ------------ -------------- ---------- -------------- ---------- -------------- ----------
**Center** 0.54 (0.67) 0.42 −1.05 (0.50) 0.04 −0.40 (0.83) 0.63 −3.93 (4.50) 0.38
**Time** 1.35 (0.13) \<0.0001 1.02 (0.11) \<0.0001 1.67 (0.19) \<0.0001 7.55 (1.09) \<0.0001
**Center\*time** −0.55 (0.20) **0.006** −0.30 (0.17) 0.07 −0.73 (0.28) **0.01** −1.51 (1.60) 0.35
RMA-GF: Rivermead Motor Assessment- Gross Function, RMA-LT: Rivermead Motor Assessment- Leg and Trunk; RMA-A: Rivermead Motor Assessment Arm Function, BI: Barthel Index, \*: with.
{#F2}
Discussion
==========
In the present study, motor and functional recovery patterns were compared between stroke patients admitted to an in-patient treatment centre in Germany (developed/well-resourced) and their matched counterparts admitted to an out-patient treatment facility in South Africa (developing/under-resourced) for a six-month post stroke recovery period.
The results of the GLIMMIX showed that the recovery patterns for gross motor functioning (RMA-GF) were significantly steeper in the German patients than the South African patients. The RMA-GF has a total of 13 items that include the assessment of the ability to sit, perform transfers (lying to sitting, sitting to standing and wheelchair to chair) and walk independently. The significant differences in RMA-GF scores therefore imply that the German participants were better at performing these activities than the South African participants. Similarly, German patients demonstrated significantly faster recovery of arm function (RMA-A). These findings could be linked to differences in the process of rehabilitation associated with the typical treatment regimes in developed versus developing countries, but they do not imply a causal link.
The German centre provided in-patient treatment of high intensity, meaning that the patients received daily therapy on an average of 2 hours and 20 minutes \[[@B22]\]. In contrast, the South African patients received out-patient therapy at an average of once a week \[[@B18]\]. Thus the intensity of treatment differed significantly between the developed and developing countries, potentially impacting outcome and recovery patterns \[[@B6]\].
The content of physiotherapy received by the South African and German patients was similar. In both the South African \[[@B23]\] and the German centres \[[@B24]\] the most frequently practised activities were selective movements, exercises, and balance in sitting and standing. Ambulatory exercises were, however, practised less by the South African sample, which could have contributed to the improved ability of the German participants to perform these activities. Task-specific exercise is known to be effective in the rehabilitation of stroke patients \[[@B25]\]. The German sample received more occupational therapy (OT) than the South African participants \[[@B22]\]. In the study conducted at the CHCs, 99 % of the stroke patients received physiotherapy, while only 21 % received occupational therapy. The patients in the German centre engaged in domestic and activities of daily living in the OT sessions \[[@B24]\], activities in which the upper limb is more involved \[[@B26],[@B27]\] than when performing ambulatory activities, for example \[[@B24]\]. The activities practised in the occupational therapy sessions are intended to contribute to upper limb recovery, which may explain the steeper RMA-A recovery curves in the German sample.
No significant differences were found in the recovery patterns for functional recovery as measured by the Barthel Index, though the German sample produced higher median scores at two and six months after stroke. This non-significant finding could be attributed to the suboptimal recovery of the South African sample with regards to basic activities of daily living as measured by the Barthel Index. With regards to the German sample the ceiling effect of the Barthel Index could have affected the recovery, the median scores of the BI of the German sample was 90 and 95 at two and six months respectively \[[@B28]\]. The findings relating to the recovery of the RMA-LT could not be compared, as the two groups were significantly different with regard to this outcome at baseline, which would have affected the comparison of the recovery patterns.
Limitations of the study
------------------------
As the study compared the outcomes of stroke patients from two different countries, cultural differences which are intrinsic to the patients could have affected the findings. The matching process used in the study also decreased the size of sample that could be compared. The process of matching meant that a number of participants from both settings were excluded from the study, which could have affected the findings. A major limitation is that the study used secondary data, which limited the researcher's ability to determine what the factors were that could have influenced the recovery patterns and outcomes.
Conclusion
==========
The findings indicated that the German stroke population reported statistically significantly better recovery patterns for RMA-GF and RMA-A. Well-resourced rehabilitation in a German rehabilitation centre generated moregross motor and upper limb recovery when compared to less resourced outpatient services in South Africa. The findings therefore provide empirical support for perceptions held by rehabilitation professionals. The findings of this study, using secondary data, should be further investigated using prospective designs.
Abbreviations
=============
BI: Barthel Index; CERISE: Collaborative evaluation of rehabilitation in stroke across Europe; CHCs: Community health centres; GLIMMIX: Generalized linear mixed methods model; NIHSS: National institute of health stroke scale; RMA: Rivermead motor assessment scale; RMA-GF: Rivermead motor assessment scale gross motor function; RMA-LT: Rivermead motor assessment scale leg and trunk function; RMA-A: Rivermead motor assessment scale arm function; SA: South Africa; SRU: Stroke rehabilitation unit.
Competing interests
===================
The authors declare that they have no competing interests.
Authors' contributions
======================
AR and LDW were the main researchers in conceptualization of the study and writing the article. MS and RM contributed to the conceptualization and finalization of the South African study, and commented on the article. WD contributed to conceptualization and finalization of both South African and European studies, and commented on the article. KP was involved in the European study and commented on the articles; he also developed the matching process used in the comparison. All authors read and approved the final manuscript.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<http://www.biomedcentral.com/1472-6963/14/82/prepub>
Acknowledgments
===============
Ms H. Ellen technical and language editor for reviewing this article and providing editing support. European Commission and Sekretariat f¨r Bildung und Forschung SBF (C.H.) for financial assistance for co-authors' PhD Thesis as part of the CERISE study. University of the Western Cape Research Fund and VLIR for financial assistance for corresponding author's PhD thesis.
| {
"pile_set_name": "PubMed Central"
} |
All relevant data are within the paper and its Supporting Information files.
Introduction {#sec005}
============
Renal interstitial fibrosis is the common pathway in progressive kidney disease, which leads to deterioration and eventual failure of renal function, irrespective of the diverse causes of disease \[[@pone.0149242.ref001],[@pone.0149242.ref002]\]. The interstitial fibrosis process is often accompanied by epithelial-mesenchymal transition (EMT) of tubular epithelial cells, by which tubular epithelial cells are converted into the phenotype of myofibroblasts and produce interstitial matrix components \[[@pone.0149242.ref001]--[@pone.0149242.ref003]\]. Many studies have shown that transforming growth factor-β1 (TGF-β1) and its downstream transcription factor Snail are the key molecules that trigger the process of tubular EMT \[[@pone.0149242.ref001]--[@pone.0149242.ref005]\].
Our previous *in vitro* studies using cell culture techniques, including cell co-culture, have revealed that tubular epithelial cells can be activated by aristolochic acid (AA) and then secrete fibrogenic factors including TGF-β1, which can in turn act on renal interstitial fibroblasts through cell-cell cross talking to enhance collagen type I synthesis \[[@pone.0149242.ref006],[@pone.0149242.ref007]\]. In addition, we have successfully created a rat model of chronic aristolochic acid nephropathy (CAAN) with typical interstitial fibrosis \[[@pone.0149242.ref008], [@pone.0149242.ref009]\] and demonstrated that EMT of tubular epithelial cells is involved in the development of interstitial fibrosis in the CAAN rat model *in vivo* \[[@pone.0149242.ref010]\]. Therefore, in the present study, we choose the cultured tubular epithelial cells stimulated by AA and the rat model of CAAN to investigate the antagonistic effects of *Hirsutella sinensis* (HS) on tubular EMT.
HS is the anamorph of *Cordyceps sinensis* \[[@pone.0149242.ref011]\]. According to the record in the "Pharmacopoeia of People's Republic of China", *Cordyceps sinensis* is a kind of fungus that belongs to Clavicipitaceae \[[@pone.0149242.ref011]\]. *Cordyceps sinesis*, as a traditional Chinese medicine herb, and its anamorph HS are widely used for kidney disease treatment in China and have been identified to have therapeutic effects on delaying the progression of renal function damage in patients with chronic kidney diseases \[[@pone.0149242.ref012],[@pone.0149242.ref013]\]. Our previous *in vitro* and *in vivo* studies also have shown that HS is able to antagonize the fibrogenic actions of AA \[[@pone.0149242.ref014], [@pone.0149242.ref015]\]. However, so far, the underlying mechanism of the protective effects of HS has not been fully elucidated. Does HS have an inhibitory effect on EMT in its anti-fibrotic mechanism? If so, is there a change of Snail expression involved in this process? Therefore, the aim of this research project is to answer the above questions in order to improve the understanding of the HS anti-fibrotic mechanism.
Materials and Methods {#sec006}
=====================
Animals and Ethics Statement {#sec007}
----------------------------
Sprague-Dawley rats were purchased from Vital River Laboratory Animal Technology Co. (Beijing, China). Rats were maintained under specific-pathogen-free conditions in the animal facility at the Beijing Heart Lung and Blood Vessel Diseases Institute. The rats were given a standard rodent chow and water ad libitum. All animal care and experimental protocols complied with the US National Institutes of Health Guide for the Care and Use of Laboratory Animals (publication no. 85--23, 1996) and were approved by the Institutional Animal Care and Use Committee of Capital Medical University.
Animal experiment {#sec008}
-----------------
Eighteen male Sprague-Dawley rats with weight of 200±10g at the age of 8 weeks were randomly and equally divided into the following 3 groups: (1) CAAN model group: the rats intermittently received extract of *Aristolochia manshuriensis Kom* (AmK) by gavages in the morning. In the beginning 5 days, the gavage dosage of AmK equaled to AA 20 mg/kg•d, and then gavage was stopped for 9 days; afterward, AmK dosage was reduced to equivalent dose of AA 15 mg/kg•d and daily gavage of every other week was performed until the end of 12th week. The preparation of AmK extract was carried out according to our previously described method \[[@pone.0149242.ref008]\]. (2) Intervention group: besides receiving AmK as mentioned above, the rats were also administered HS (Zhongmei Huadong Pharmaceutical Co.) in a dose of 1.5 g/kg•d by gavage every afternoon. (3) Control group: the rats were given tap water by gavage every morning.
24h urine samples were collected for the measurement of urinary protein excretion every 4 weeks. The urinary protein was quantified using Bradford Protein Assay Kit (Beyotime Biotechnology). Serum creatinine (SCr) and urinary creatinine (UCr) were detected by the picric acid method with automatic biochemistry analyzer (Hitachi 7170) at 0 and 12th week, and creatinine clearance (CCr) was calculated according to the following formula: CCr (ml/min) = UCr (μmol/L) × urine volume (ml/min) / SCr (μmol/L). All the rats were sacrificed after anesthetized with pentobarbital at the end of 12th week. A part of kidney tissue was fixed in 4% neutral formaldehyde solution for pathological and immunohistochemical examinations, and another part of kidney tissue was rapidly reserved in liquid nitrogen for real time quantitative polymerase chain reaction (PCR) analysis.
Experiment of HKC cells {#sec009}
-----------------------
Human kidney proximal tubular epithelial cell line, HKC, was purchased from the American Type Culture Collection (ATCC). The cells were cultured in DMEM/F12 media (Life technologies, USA) supplemented with 10% inactivated fetal bovine serum (Life technologies, USA) at 37°C in humidified air with 5% CO2. Cultured HKC cells were divided into the following 4 groups: (1) AA stimulation group: the cells were incubated with 10μmol/L AA (Sigma); (2) HS intervention group: incubated with 10 μmol/L AA and 10 mg/L HS (Zhongmei Huadong Pharmaceutical Co.); (3) HS control group: incubated with 10mg/L HS; (4) Control group: only incubated with media. After 12 h and 36 h of incubation, the cells were harvested for real time quantitative PCR analysis and Western blot assay, respectively.
Experiment of siRNA interference of HKC cells {#sec010}
---------------------------------------------
The HKC cells were transiently transfected with 4 μl human Snail siRNA (Santa Cruz, sc-38398) or 4 μl control siRNA-A (Santa Cruz, sc-37007) using Lipofectamine 2000 (Life technologies, USA) according to the manufacturer's instruction. After that, the transfected cells were cultured in DMEM/F12 media (Life technologies, USA) with 10% inactivated fetal bovine serum (Life technologies, USA) for 24 h, and then incubated with or without AA (10 μmol/L, Sigma). After 12 h and 36 h of incubation, the transfected cells were harvested for real time quantitative PCR analysis and Western blot assay, respectively.
The experiment was grouped as follows: (1) Control group: non-transfected HKC cells were incubated with media; (2) AA stimulation group: non-transfected HKC cells were incubated with 10 μmol/L AA; (3) AA-stimulated Snail siRNA transfection group: HKC cells transfected with Snail siRNA were incubated with 10μmol/L AA; (4) AA-stimulated control siRNA-A transfection group: HKC cells transfected by control siRNA-A were incubated with 10μmol/L AA.
Experiment of Snail overexpression of HKC cells {#sec011}
-----------------------------------------------
Cultured HKC cells were transiently transfected with 3μg pGV167-Snail or 3μg vector pGV167 (Shanghai Genechem Co.) using Lipofectamine 2000 (Life technologies, USA) according to the manufacturer's instruction. After that, the transfected cells were cultured in DMEM/F12 media (Life technologies, USA) with 10% inactivated fetal bovine serum (Life technologies, USA) for 24 h, and then the cells were continuously incubated with media only, 10 μmol/L AA or 10 μmol/L AA and 10 mg/L HS, respectively. After 12 h or 36 h incubation the cells were harvested for real time quantitative PCR analysis and Western blot assay, respectively.
Pathological examination {#sec012}
------------------------
The kidney tissue was conventionally dehydrated, embedded, cut into 3 μm-thick sections and stained with Masson trichrome reagent for light microscopy. The tubulointerstitial images of renal cortex, which did not contain glomeruli and arterioles, in 15 random visual fields (×100 times) were separately taken, and then were analyzed by Motic Med 6.0 digital medical image analysis system. The relative renal interstitial fibrosis area (%) = renal interstitial green dye area/visual field area ×100%.
Immunohistochemical staining {#sec013}
----------------------------
3 μm-thick tissue sections of renal cortex were microwave heating-induced epitope retrieval (95°C for 10 min). Rabbit anti-TGF-β1 polyclonal antibody (Santa Cruz, 1:150), rabbit anti-Snail polyclonal antibody (Abcam, 1:125), mouse anti-α-smooth muscle actin (α-SMA) monoclonal antibody (1:30) and mouse anti-cytokeratin-18 monoclonal antibody (Zhong Shan Golden Bridge Biotech, 1:50) were used as primary antibodies and incubated with renal tissue sections respectively at 4°C over night. Then, the immunostaining of TGF-β1, Snail and cytokeratin-18 was performed using the EnVision detection kit (Dako), and the immunostaining of α-SMA was carried out using labeled streptavidin- biotin (LSAB) method \[[@pone.0149242.ref010]\].
The immunohistochemical staining images of TGF-β1, α-SMA and cytokeratin-18 were analyzed using the method similar to the image analysis of aforementioned Masson staining. The relative tubulointerstitial positive area = tubulointerstitial brown dye area/visual field area × 100%. The images of Snail were analyzed by the following procedures: the tubulointerstitial images of renal cortex, which did not contain glomeruli and arterioles, in 20 random visual fields (×400 times) were separately taken under light microscopy, and then the number of tubular epithelial cells with Snail positive staining in each visual field was counted. The Snail positive cell number per square millimeter area (number/mm^2^) = positive cell number per visual field/area of per visual field.
Reverse transcription and real time quantitative PCR {#sec014}
----------------------------------------------------
Total RNA was isolated from rat renal cortex tissue or HKC cells using Trizol reagent (BBI) following the manufacturer's instructions. 2 μg total RNA from each sample was reverse-transcribed to cDNA with Moloney murine leukemia virus reverse transcriptase (Huamei Biotech). Gene-specific primers [see [S1](#pone.0149242.s002){ref-type="supplementary-material"} and [S2](#pone.0149242.s003){ref-type="supplementary-material"} Tables]{.ul}, respectively. Real time quantitative PCR was performed using SYBR Green Realtime PCR Master Mix (TOYOBO) following the manufacturer's instructions. The GAPDH and β-actin fragments were also amplified as the internal control genes for the animal and cellular experiments, respectively. The relative quantity of mRNA expression was calculated according to the formula 2^-(target\ gene\ Ct--control\ gene\ Ct)^×10^3^, in which Ct is threshold cycle number.
Western blot assay {#sec015}
------------------
The HKC cells were lysed using RIPA buffer (ComWin Biotec Co) and then the cell lysate was boiled for 5 min. Equivalent amounts of boiled cell proteins were separated by electrophoresis on sodium dodecyl sulfate-polyacrylamide gel (SDS-PAGE) and transferred to nitrocellulose membranes (General Electric Co). After blocking with 5% skim milk in phosphate-buffered saline with 0.1% Tween 20 for 1 h, the membranes were incubated with primary antibody at 4°C over night, and then incubated with secondary antibody in room temperature for 1 h. Details regarding primary and secondary antibodies are listed in [[S3 Table](#pone.0149242.s004){ref-type="supplementary-material"}]{.ul}. The blotted proteins were quantified using Odyssey Infrared Imaging System (LI-COR Biosciences). β-actin was used as an internal loading control and the relative expression level of each target protein was displayed as a ratio of target protein/β-actin protein. All the assays were performed at least in triplicate.
Statistical Analysis {#sec016}
--------------------
All the data of continuous variables were expressed as the mean ± SD and analyzed by using SPSS 15.0 statistical software. One-way ANOVA was used for the comparison of multiple continuous variables. Correlation between two variables was inspected by using the Pearson correlation analysis. P\<0.05 was considered statistically significant.
Results {#sec017}
=======
The effects of HS on rat kidney injury caused by AA {#sec018}
---------------------------------------------------
Urinary protein excretion at baseline between the three animal groups had no statistical difference (*P*\>0.05). Compared with control group, the urinary protein excretion in CAAN model group and intervention group was significantly increased at 4th, 8th and 12th week (*P*\<0.01). Compared with CAAN model group, the urinary protein excretion in intervention group was decreased at 4th, 8th and 12th week, and the difference reached statistical significance at 12th week (*P*\<0.05) ([Fig 1A](#pone.0149242.g001){ref-type="fig"}).
{#pone.0149242.g001}
CCr levels at baseline among the three animal groups were not statistically different (*P*\>0.05). At the 12th week, CCr levels in CAAN model and intervention groups were significantly lower than that in control group (*P*\<0.05 or *P*\<0.01), and in intervention group was significantly higher than that in CAAN model group (*P*\<0.05) ([Fig 1B](#pone.0149242.g001){ref-type="fig"}).
Pathological examination of kidney tissue at the 12th week showed that the relative renal interstitial fibrosis areas in CAAN model and intervention groups were significantly larger than that in control group (*P*\<0.05), and in intervention group was significantly smaller than that in CAAN model group (*P*\<0.05) ([Fig 1C](#pone.0149242.g001){ref-type="fig"}).
The above results suggest that HS has antagonism effects on AA-induced kidney injury *in vivo*.
Effects of HS on the expression of α-SMA and cytokeratin-18 in rat renal cortex tissues {#sec019}
---------------------------------------------------------------------------------------
Immunohistochemical staining of rat renal cortex tissue displayed that α-SMA mainly expressed on arteriolar smooth muscle cells in control group, but also on tubular epithelial cells and in interstitium in CAAN model and intervention groups; cytokeratin-18 only expressed on tubular epithelial cells in the three groups ([Fig 2](#pone.0149242.g002){ref-type="fig"}).
{#pone.0149242.g002}
The animal experiment showed that α-SMA mRNA and protein expression in renal cortex tissue was significantly up-regulated in CAAN model and intervention groups compared with control group (*P*\<0.05 or *P*\<0.01), and was significantly down-regulated in intervention group compared with CAAN model group (*P*\<0.05) (Figs [2](#pone.0149242.g002){ref-type="fig"} and [3](#pone.0149242.g003){ref-type="fig"}).
{#pone.0149242.g003}
In contrast, cytokeratin-18 mRNA and protein expression in renal cortex tissue was significantly down-regulated in CAAN model and intervention groups compared with control group (*P*\<0.05 or *P*\<0.01), and was significantly up-regulated in intervention group compared with CAAN model group (*P*\<0.05) (Figs [2](#pone.0149242.g002){ref-type="fig"} and [3](#pone.0149242.g003){ref-type="fig"}).
The above results suggest that the EMT of tubular epithelial cells occurs in the disease process of CAAN, and HS can antagonize the tubular EMT.
Effects of HS on the expression of TGF-β1 and Snail in rat renal cortex tissues {#sec020}
-------------------------------------------------------------------------------
Immunohistochemical staining of rat renal cortex tissue in the three experimental groups displayed that TGF-β1 mainly expressed on tubular epithelial cells and weakly in glomeruli and arteriolar wall; Snail expressed in tubular epithelial cells ([Fig 4](#pone.0149242.g004){ref-type="fig"}).
{#pone.0149242.g004}
The animal experiment showed that mRNA and protein expression of the TGF-β1 and Snail in renal cortex tissue was significantly up-regulated in CAAN model and intervention groups compared with control group (*P*\<0.05 or *P*\<0.01), and was significantly down-regulated in intervention group compared with CAAN model group (*P*\<0.05) (Figs [4](#pone.0149242.g004){ref-type="fig"} and [5](#pone.0149242.g005){ref-type="fig"}).
{#pone.0149242.g005}
The above results suggest that there is the up-regulation of TGF-β1 and Snail expression in tubular epithelial cells in the CAAN rat model during the same time of tubular EMT, and HS can antagonize the effects.
The correlation analyses between various parameters in CAAN rat model {#sec021}
---------------------------------------------------------------------
The results of immunohistochemical staining on rat renal cortex tissues showed that the expression of TGF-β1 was positively correlated with expression of Snail and α-SMA, and negatively correlated with expression of cytokeratin-18 (correlation coefficients 0.715, 0.739 and -0.696, respectively. *P*\<0.01); the expression of Snail was positively correlated with expression of α-SMA and negatively correlated with expression of cytokeratin-18 (correlation coefficients 0.843 and -0.740, respectively. *P*\<0.01); the expression of α-SMA was negatively correlated with expression of cytokeratin-18 (correlation coefficient -0.735. *P*\<0.01).
The results of Masson trichrome staining and immunohistochemical staining on rat renal cortex tissues showed that the relative renal interstitial fibrosis area was positively correlated with the expression of TGF-β1, Snail and α-SMA, and negatively correlated with the expression of cytokeratin-18 (correlation coefficients 0.787, 0.805, 0.915 and -0.814, respectively. *P*\<0.01).
The above results suggest that the up-regulation of TGF-β1 and Snail expression in tubular epithelial cell is correlated with tubular EMT and renal interstitial fibrosis in the CAAN rat model.
Effects of HS on the expression of α-SMA and cytokeratin-18 in cultured HKC {#sec022}
---------------------------------------------------------------------------
The HKC cell experiment showed that α-SMA mRNA and protein expression was significantly up-regulated in AA stimulation group compared with control group (*P*\<0.05), and was significantly down-regulated in HS intervention group compared with AA stimulation group (*P*\<0.05 or *P*\<0.01) ([Fig 6A and 6B](#pone.0149242.g006){ref-type="fig"}).
{#pone.0149242.g006}
On the contrary, cytokeratin-18 mRNA and protein expression in HKC cells was significantly down-regulated in AA stimulation group compared with control group (P\<0.01), and was significantly up-regulated in HS intervention group compared with AA stimulation group (P\<0.05 or P\<0.01) ([Fig 6A and 6B](#pone.0149242.g006){ref-type="fig"}).
The results of cell experiment are quite similar to those in animal experiment, and both suggest that HS can antagonize the tubular EMT caused by AA.
Effects of HS on the expression of TGF-β1 and Snail in cultured HKC {#sec023}
-------------------------------------------------------------------
The HKC cell experiments showed that mRNA and protein expression of TGF-β1 and Snail was significantly up-regulated in AA stimulation group compared with control group (*P*\<0.05 or *P*\<0.01), and was significantly down-regulated in HS intervention group compared with AA stimulation group (*P*\<0.05 or *P*\<0.01) ([Fig 7A and 7B](#pone.0149242.g007){ref-type="fig"}).
{#pone.0149242.g007}
The results of cell experiment, which are quite similar to the results of animal experiment, suggest that HS can antagonize the up-regulation of TGF-β1 and Snail expression of HKC cells caused by AA.
The expression changes of Snail, α-SMA, cytokeratin-18 and fibronectin in HKC cells of Snail gene knockdown {#sec024}
-----------------------------------------------------------------------------------------------------------
Real time quantitative PCR analysis revealed that, compared with control group, the Snail mRNA expression was significantly down-regulated in Snail siRNA transfection group (*P*\<0.05), but not changed in control siRNA-A transfection group ([Fig 8A](#pone.0149242.g008){ref-type="fig"}). The results suggest the siRNA transfection is successful.
{#pone.0149242.g008}
The results of gene knockdown experiment showed that the mRNA and /or protein expression of Snail, α-SMA and fibronectin was significantly down-regulated (*P*\<0.05 or *P*\<0.01), and the mRNA and protein expression of cytokeratin-18 was significantly up-regulated (*P*\<0.05 or *P*\<0.01) in AA-stimulated Snail siRNA transfection group, compared with AA stimulation group. However, there were no similar changes of mRNA and protein expression in AA-stimulated control siRNA-A transfection group compared with AA stimulation group ([Fig 8B, 8C and 8D](#pone.0149242.g008){ref-type="fig"}).
These results, that down-regulation of Snail expression could significantly attenuate AA-induced EMT and matrix production of HKC cells, suggest that AA-induced fibrogenic actions might be mediated by transcription factor Snail.
The expression changes of Snail, α-SMA, cytokeratin-18 and fibronectin in HKC cells of Snail overexpression {#sec025}
-----------------------------------------------------------------------------------------------------------
Western blot assay revealed that, compared with control group, the Snail protein expression was significantly up-regulated in pGV167-Snail transfection group (*P*\<0.01), but not changed in vector pGV167 transfection group ([Fig 9A](#pone.0149242.g009){ref-type="fig"}). The results suggest the pGV167-Snail transfection is successful.
{#pone.0149242.g009}
The experimental results of Snail overexpression showed that the up-regulated mRNA and/or protein expression of Snail, α-SMA and fibronectin, and the down-regulated mRNA and protein expression of cytokeratin-18 were not able to inhabited by HS in AA-stimulated pGV167-Snail transfection group (*P*\>0.05). However, the above changes of mRNA and/or protein expression were significantly reversed by HS in AA-stimulated vector pGV167 transfection group (*P*\<0.05 or *P*\<0.01) ([Fig 9B to 9E](#pone.0149242.g009){ref-type="fig"}).
These results, that HS could not weaken AA-induced EMT and matrix production in Snail overexpression HKC cells, suggest that the antagonistic effects of HS on AA-induced fibrogenic actions might be implemented by inhibiting Snail.
All original experimental data see [S1 Data](#pone.0149242.s001){ref-type="supplementary-material"}.
Discussion {#sec026}
==========
Myofibroblasts in renal interstitium, as predominant effector cells, play a vital role in the development of interstitial fibrosis by synthesizing and secreting extracellular matrix (ECM) \[[@pone.0149242.ref002]\]. During the past decades, the origins of renal interstitial myofibroblasts have been investigated extensively. Currently, they have been proposed to be derived from the following one or more sources: (1) activation of local resident fibroblasts; (2) differentiation of bone marrow-derived mesenchymal stem cells and fibrocytes; (3) transdifferentiation of pericytes; (4) tubular EMT; and (5) endothelial-mesenchymal transition \[[@pone.0149242.ref016]--[@pone.0149242.ref020]\]. Although the understanding has improved, the relative contribution of different cell sources to renal interstitial myofibroblasts remains controversial, especially for tubular EMT \[[@pone.0149242.ref019], [@pone.0149242.ref020]\]. Lebleu and colleagues reported that the cells of tubular epithelial origin undergoing EMT only contributed 5% to the renal interstitial myofibroblast pool \[[@pone.0149242.ref017],[@pone.0149242.ref018]\], but this result could not be reproduced by subsequent studies \[[@pone.0149242.ref019]\], and the contribution of EMT even reached 36% in a previous report \[[@pone.0149242.ref021]\]. However, tubular EMT has been confirmed by many independent studies and its role in renal interstitial fibrosis is undisputed \[[@pone.0149242.ref019]--[@pone.0149242.ref021]\]. Therefore, we choose tubular EMT as the focus of this investigation to study the antagonistic effects of HS on renal interstitial fibrosis.
*Cordyceps sinensis*, also known by the Chinese name of *Dong Chong Xia Cao*, is a precious Chinese herbal medicine \[[@pone.0149242.ref012]\]. *Cordyceps sinensis* as a medicinal herb was first recorded in the Chinese classical medicine book "Ben Cao Bei Yao" ("Essentials of Materia Medica") in 1694 \[[@pone.0149242.ref022]\], and its full scientific name is "*Cordyceps sinensis* (Berkeley) Saccardo," named by Italian scholar Saccardo in 1878 \[[@pone.0149242.ref022]\]. HS is the anamorph of *Cordycep sinensis* \[[@pone.0149242.ref023], [@pone.0149242.ref024]\], which was verified by the studies of individual developmental biology \[[@pone.0149242.ref025]\] and molecular systems biology \[[@pone.0149242.ref026]--[@pone.0149242.ref028]\]. *Cordycep sinesis* and its anamorph HS are widely used to treat kidney diseases in China \[[@pone.0149242.ref012], [@pone.0149242.ref013]\]. Our previous *in vitro* and *in vivo* studies have shown that HS is able to antagonize the fibrogenic actions of AA by down-regulating the expression of TGF-β1, connective tissue growth factor (CTGF), plasminogen activator inhibitor-1 (PAI-1) and tissue inhibitor of metalloproteinases-1 (TIMP-1), which can promote ECM synthesis and/or inhibit ECM degradation \[[@pone.0149242.ref014],[@pone.0149242.ref015]\]. However, the effect of HS on tubular EMT in renal interstitial fibrosis has not been investigated. In the process of EMT, the phenotypic conversion of tubular epithelial cells takes place. Cells lose epithelial cell markers such as E-cadherin and cytokeratin, acquire myofibroblastic markers such as α-SMA and vimentin, and produce interstitial matrix components including collagens type I and type III. In addition, the cell morphology and function are altered. The cell shape transforms from a cobblestone-like to a spindle-like appearance, and the cells lose their polarity and adhesion properties and increase their migration and invasion abilities \[[@pone.0149242.ref002], [@pone.0149242.ref003], [@pone.0149242.ref029]--[@pone.0149242.ref032]\]. In this study, we observed changes in the mRNA and protein expression levels of α-SMA and cytokeratin-18 both *in vitro* and *in vivo*. Our results showed that AA could give rise to the down-regulation of cytokeratin-18 expression and *de novo* α-SMA expression of tubular epithelial cells during the fibrogenic process, while HS significantly antagonized the effects of EMT induced by AA.
TGF-β1 and its downstream zinc finger transcription factor Snail are the key molecules that trigger the process of tubular EMT \[[@pone.0149242.ref001]--[@pone.0149242.ref005], [@pone.0149242.ref033]--[@pone.0149242.ref038]\]. TGF-β1 binds with its type I and type II transmembrane receptors, and then the cytoplasmic latent signal transduction proteins Smad 2/3 are phosphorylated by the type I receptor serine kinase. Phosphorylated Smad 2/3 partner with Smad 4 and subsequently translocate into the nuclei and up-regulate Snail expression \[[@pone.0149242.ref003], [@pone.0149242.ref033], [@pone.0149242.ref036], [@pone.0149242.ref039]\]. Snail binds to specific DNA sequences (CANNTG, where N is any nucleotide), called E-boxes in the promoter of the E-cadherin gene, and then represses E-cadherin transcription \[[@pone.0149242.ref004], [@pone.0149242.ref035], [@pone.0149242.ref038], [@pone.0149242.ref039]\]. E-cadherin, as a major constituent of adherens-type junctions, plays an essential role in the assembly of the junctional complex, maintaining the structural integrity and polarity of epithelial cells \[[@pone.0149242.ref003], [@pone.0149242.ref038]\]. Down-regulation or loss of E-cadherin expression would immediately induce *de novo* α-SMA expression and cause an early event of EMT \[[@pone.0149242.ref003], [@pone.0149242.ref034], [@pone.0149242.ref035], [@pone.0149242.ref039]\].
In this study, we observed changes in the mRNA and protein expression levels of TGF-β1 and Snail in tubular epithelial cells, and analyzed the relationship between their expression and tubular EMT as well as renal interstitial fibrosis in an animal experiment, and ECM production in a cell experiment. Both the in vitro and in vivo research results showed that AA significantly up-regulated the expression of TGF-β1 and Snail, while HS significantly antagonized the above AA-induced fibrogenic effects. Statistical correlation analysis in the animal experiment revealed that the expression of TGF-β1 was positively correlated with the expression of Snail, and that the expression of TGF-β1 and Snail was negatively correlated with the expression of cytokeratin-18 and positively correlated with the expression of α-SMA as well as the relative renal interstitial fibrosis area.
In order to confirm the effects of Snail on EMT and ECM production of tubular epithelial cells, a siRNA transfection experiment of HKC cells was performed. Real time quantitative PCR analysis and Western blot assay showed that Snail gene knockdown significantly up-regulated the AA-induced low expression of cytokeratin-18 and down-regulated the AA-induced high expression of Snail, α-SMA and fibronectin. In addition, a pGV167-Snail transfection experiment of HKC cells was also carried out. Real time quantitative PCR analysis and Western blot assay showed that AA-induced high expression of Snail, α-SMA and fibronectin, and low expression of cytokeratin-18 were not able to reverse by HS in the Snail overexpression HKC cell. These experimental results suggest that transcription factor Snail might play an important role in AA-induced fibronenic actions and the antagonistic effects of HS might be realized by suppressing Snail expression.
In conclusion, HS, the anamorph of *Cordyceps sinensis*, is able to antagonize effectively AA-induced tubular EMT and renal interstitial fibrosis. Transcription factor Snail might be one of potential targets of HS effect.
Supporting Information {#sec027}
======================
###### Original experimental data.
(DOC)
######
Click here for additional data file.
###### Primer sequences for real time quantitative RT-PCR analysis in animal experiment.
(DOC)
######
Click here for additional data file.
###### Primer sequences for real time quantitative RT-PCR analysis in cell experiment.
(DOC)
######
Click here for additional data file.
###### Primary and secondary antibodies for Western blot assay.
(DOC)
######
Click here for additional data file.
[^1]: **Competing Interests:**The authors have declared that no competing interests exist.
[^2]: Conceived and designed the experiments: Y-PC HC X-YX J-JC H-LR. Performed the experiments: X-YX J-JC H-LR Y-YW H-RD Y-LM. Analyzed the data: X-YX J-JC H-LR Y-PC. Contributed reagents/materials/analysis tools: X-YX J-JC H-LR. Wrote the paper: Y-PC X-YX J-JC H-LR HC.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Organismal form and function emerge during ontogeny through complex interactions between gene products, environmental conditions, and ontogenetic processes \[[@B1], [@B2]\]. The causes, nature, and consequences of these interactions are the central foci of epigenetics \[[@B3]\]. Broadly, epigenetics seeks to understand how phenotypes emerge through developmental processes, and how that emergence is altered to enable evolutionary modification, radiation, and innovation. Epigenetic mechanisms can operate at any level of biological organization above the sequence level, from the differential methylation of genes to the somatic selection of synaptic connections and the integration of tissue types during organogenesis. Here, we take this inclusive definition of epigenetics and apply it to the phenomenon of developmental plasticity, defined as a genotype\'s or individual\'s ability to respond to changes in environmental conditions through changes in its phenotypes \[[@B4]\]. All developmental plasticity is, by definition, epigenetic in origin, as the genotype of the responding individual remains unaltered in the process. It is the nature, origins, and consequences of the underlying epigenetic mechanisms that we focus on in this review. We do so with specific reference to horned beetles, an emerging model system in evo-devo in general and the evolutionary developmental genetics of plasticity in particular.
We begin our review with a general introduction to the concept of developmental plasticity. We then introduce our focal organisms, horned beetles, summarize the most relevant forms of plasticity that have evolved in these remarkable organisms, review what is known about the underlying epigenetic mechanisms, and highlight future research directions. Lastly, we discuss how studies in *Onthophagus* species could provide meaningful insight into three major foci in evo-devo research: the development and evolution of shape, the process of evolution via genetic accommodation, and the origin of novel traits. We begin, however, with a brief introduction of the significance of plasticity in development and evolution.
2. The Biology of Developmental Plasticity {#sec2}
==========================================
Developmental plasticity refers to an individual\'s ability to respond to environmental changes by adjusting aspects of its phenotype, often in an adaptive manner. In each case a single genotype is able, through the agency of environment-sensitive development, to give rise to vastly different phenotypes. Developmental plasticity is perhaps most obvious in the expression of alternative morphs or polyphenisms, as in the seasonal morphs of butterflies, winged or wingless adult aphids, aquatic or terrestrial salamanders, or the different castes of social insects (reviewed in \[[@B2]\]). However, developmental plasticity is also inherent in more modest, often continuous changes in response to environmental conditions, such as tanning (in response to sun exposure), muscle buildup (in response to workouts) or immunity (following an infection resulting in an immune response). Lastly, developmental plasticity is a necessary prerequisite for developmental canalization, or the production of an *in*variant phenotype in the face of environmental fluctuation. Here, plastic compensatory adjustments on some level of biological organization enable the homeostatic maintenance of developmental outputs at others, such as the maintenance of blood sugar levels in the face of fluctuating nutrition and activity, or the maintenance of scaling relationships despite nutrition-dependent variation of overall body size in most organisms. Developmental plasticity is thus a ubiquitous feature of organismal development, applicable to all levels of biological organization, and rich in underlying mechanisms.
Developmental plasticity not only enables coordinated and integrated responses in development but also has great potential to affect evolutionary processes and outcomes (reviewed in \[[@B4], [@B5]\]). Developmental plasticity enables organisms to adaptively adjust their phenotype to changing environmental conditions. On one side, developmental plasticity may thus impede genetic divergences that might otherwise evolve between populations subject to disparate environmental conditions. On the other, plasticity buffers populations against local extinctions, thus increasing the opportunity for the evolution of local adaptations and diversification.
Similarly, developmental plasticity may both impede and facilitate evolutionary diversification by providing additional targets for selection to operate on, by offering modules for the regulation of development that can be reused across developmental contexts, and by creating novel trait interactions. In each case, developmental plasticity may result in pleiotropic constraints on adaptive evolution, but also has the potential to shift the evolutionary trajectories available to lineages into phenotypic space that otherwise would remain unexplored \[[@B5]\].
The role of developmental plasticity in evolution is perhaps most important when we consider the consequences of organisms encountering novel environments, for instance during the natural colonization of a new habitat or the anthropogenic alteration of ecosystems due to global climate change, habitat degradation, and the invasion of alien species \[[@B5]\]. Here, developmental plasticity enables the production of functional, integrated phenotypes, despite development occurring in previously unencountered, or greatly altered, conditions. Moreover, such novel conditions may result in the formation of novel traits or trait variants previously unexpressed, alongside the release of previously cryptic, conditionally neutral genetic variation. Developmental plasticity thus has the potential to determine which phenotypic and genetic variants become visible to selection in a novel environment, thus delineating the nature and magnitude of possible evolutionary responses. Consistent with a long-assumed role of developmental plasticity in evolution (reviewed in \[[@B2]\]), a growing number of artificial selection experiments on a broad range of organisms (*Drosophila*: \[[@B6]\]; but see \[[@B7]\], Arabidopsis \[[@B8]\], fungi \[[@B9]\], and Lepidoptera \[[@B10]\]) have now demonstrated unequivocally that developmental systems confronted with challenging or novel environments can indeed expose novel phenotypic and genetic variants that, in turn, provide ample substrate for rapid, selective evolution of novel phenotypes. Similarly, studies on natural populations are providing growing evidence that ancestral patterns of plasticity have enabled and guided more refined evolutionary responses in derived populations (e.g., \[[@B11]\]).
Developmental plasticity thus plays a central role in the production and evolution of phenotypic variation. Further understanding of the nature of this role likely requires a thorough understanding of the epigenetic mechanisms that enable plastic responses to environmental variation. As outlined in the following sections horned beetles have begun to provide diverse opportunities to investigate the mechanisms underlying the epigenetic regulation of developmental plasticity and to probe their significance in the developmental origin and evolutionary diversification of form and behavior. We begin with a brief introduction of the biology of these organisms.
3. The Biology of Horned Beetles {#sec3}
================================
Beetles are holometabolous insects and constitute the most diverse insect order on the planet. Horned beetles comprise a polyphyletic group of diverse beetle families marked by the development of horns or horn-like structures in at least some species (reviewed in \[[@B12], [@B13]\]). Horn evolution has reached its extremes, both in terms of exaggeration and diversity, in two subfamilies within the Scarabaeidae, the Dynastinae (i.e., rhinoceros beetles), and the Scarabaeinae, or true dung beetles ([Figure 1](#fig1){ref-type="fig"}). In both subfamilies, thousands of species express horns and have diversified with respect to location, shape, and number of horns expressed. In extreme cases, horn expression more than doubles body length and may account for approximately 30% of body mass.
Despite the remarkable morphological diversity that exists among horned beetle species, horns are used invariably for very similar purposes: as weapons in aggressive encounters with conspecifics (reviewed in \[[@B12]\]). In the vast majority of species, horn expression is restricted to, or greatly exaggerated in, males, and absent or rudimentary in females. In these cases horns are used by males as weapons in male combat over access to females (e.g., \[[@B14]\]). In all species studied to date, body size has emerged as the most significant determinant of fighting success. In a subset of species, horns are expressed by both sexes. Here, males and females use horns as weapons in defense of mates and nesting opportunities, respectively (e.g., \[[@B15]\]). Lastly, in a very small number of species, horn expression is exaggerated in females and greatly reduced in males. Such reversed sexual dimorphisms are rare and the ecological conditions that have facilitated their evolution are largely unknown \[[@B16], [@B17]\].
We know most about the biology of horned beetles through studies on one particular genus in the Scarabaeinae: *Onthophagus.*Adults of the *Onthophagus* genus colonize dung pads of a variety of dung types, consume the liquid portions and bury the more fibrous fraction in subterranean tunnels as food provisions for offspring in the form of brood balls. Brood balls typically contain a single egg and constitute the sole amount of food available to a developing larva. Variation in the quantity or quality of parental provisions or abiotic factors such as soil moisture can greatly affect the amount of food that is effectively available to sustain larval development, which in turn results in substantial variation in larval mass at pupation and final adult body size, as detailed below.
Also similar to many other horned beetles species, *Onthophagus* frequently have to contend with high levels of male-male competition for females and female-female competition over breeding resources such as dung and tunneling space \[[@B18]\]. This unique combination of developmental conditions (marked by partly unpredictable larval resources) and ecological conditions (marked by intense intraspecific competition) has facilitated the evolution of a remarkable degree of plasticity in development, physiology, and behavior in *Onthophagus* beetles, as overviewed in the next section.
4. Developmental Plasticity in Onthophagus {#sec4}
==========================================
4.1. Plasticity in Timing of Life History Transitions {#sec4.1}
-----------------------------------------------------
Larval *Onthophagus* develop in a partly unpredictable resource environment, as their feeding conditions depend on the quantity and quality of dung provisioned for them by their parents and the physical properties of the nesting site. Unlike the highly mobile larval stages of many other holometabolous insects, larval *Onthophagus* cannot change their location or add to the resources made available to them. *Onthophagus* larvae meet these unpredictable conditions with a striking degree of plasticity in the timing of life history transitions, specifically by molting to the pupal stage at a range of larval body sizes far greater than what has been observed for other insects (reviewed in \[[@B19]\]). For instance, *Onthophagus taurus* larvae will routinely feed for 15 days during the third and final larval instar under *ad libitum*conditions, but are capable of completing metamorphosis if food deprived after just 5 days of feeding. The resulting larvae pupate at a fraction of the body mass of larvae fed *ad libitum* and eclose as tiny adults. Such striking flexibility in the dynamics of larval development and the body mass at pupation allows *Onthophagus* larvae to respond to unpredictable variation in larval feeding conditions while ensuring eclosion to a viable adult capable of reproducing. As a consequence of this phenomenon, natural populations of adult *Onthophagus* commonly display a remarkable amount of intraspecific variation in male and female body sizes.
4.2. Morphological Plasticity {#sec4.2}
-----------------------------
Recall that in the vast majority of species horn expression is restricted to males, which use horns in male combat over access to females or nesting sites. Recall also that body size is the most important determinant of fighting success, yet ecological conditions generate males of a wide range of body sizes, many of which are too small to succeed in aggressive encounters. In many horned beetle species, these conditions have led to the evolution of alternative male phenotypes, with large males relying on the use of horns and aggressive fights to secure mating opportunities, while smaller males rely on nonaggressive sneaking behaviors (discussed in detail below). *Morphologically*, male polyphenism has a range of manifestations.
First, in numerous species horn expression is restricted to, or greatly exaggerated in, large males only, whereas smaller males express greatly reduced or rudimentary horns. On the population level, this results in a bimodal distribution of horn lengths and thus two more or less discrete morphs ([Figure 2](#fig2){ref-type="fig"}). Intermediate morphologies do exist, but are rare in most species. As a consequence, populations of conspecific males express a characteristic scaling relationship, or allometry, between body size and horn length ([Figure 3](#fig3){ref-type="fig"}). Different species have diversified greatly in the degree of male horn polyphenism and the exact shape of the associated allometry \[[@B20]\], in extreme cases causing alternative conspecific morphs to be classified as different species \[[@B21]\].
Second, smaller males (often referred to as "hornless males" "minor males," or "sneaker males") do not invest in horns and fights as means of securing matings, but instead invest in non-aggressive tactics, including the use of enlarged testes and ejaculate volumes to aid in sperm competition \[[@B22]\]. As with horns, morph-specific differences in testes development differ greatly from one species to the other, but comparative studies have not been able to identify any general relationship between the relative sizes of horns and testes \[[@B23], [@B24]\].
Third, the facultative enlargement of horns in large males appears to tradeoff with a variety of other structures. The precursors of adult horns develop, just like the precursors of wings, legs, and mouthparts, right before the larval-pupal transition, but *after*all larval feeding has ceased. As such, the development of horns is, like that of all other adult traits, largely enabled by a finite amount of resources accumulated during the larval stage \[[@B25]\]. Structures that develop in the same body location or at the same developmental time may therefore find themselves competing for a limited pool of resources to sustain their growth \[[@B26]\]. When faced with resource allocation tradeoffs, developmental enlargement of one structure may only be possible through the compensatory reduction of another. As such, resource allocation tradeoffs have the potential to not only alter developmental outcomes, but to also bias evolutionary trajectories. In horned beetles, resource allocation tradeoffs have been implicated in antagonistic coevolution of horn length and the relative sizes of eyes, wings \[[@B27]\], and copulatory organs \[[@B28]\], although the exact nature of these tradeoffs remains to be investigated.
4.3. Behavioral Plasticity {#sec4.3}
--------------------------
Alternative horned and hornless male morphs employ different behavioral repertoires to maximize breeding opportunities \[[@B12]\]. In many species, horned males rely exclusively on fighting behaviors including the use of horns as weapons. Body size is the most important determinant of fight outcome, and among similar-sized males, relative horn length predicts fight outcome in most contests (e.g., \[[@B14]\]). Fights can be long, appear energetically expensive, but are rarely injurious (but see \[[@B29]\]). Horned losers typically withdraw from fights.
Hornless males also engage in prolonged fights when confronted with other hornless males, but quickly withdraw from fights against large, horned conspecifics and switch to a set of non-aggressive sneaking behaviors. For instance, in *Onthophagus taurus*, perhaps the best studied horned beetle species, sneaking behaviors include the use of naturally occurring tunnel interceptions to locate and mate with females without being detected by a guarding male \[[@B14]\]. Small males may also dig their own shallow intercept tunnel to access females underneath guarding males, or wait for females above ground as they emerge periodically to collect dung provisions. Lastly, small males may simply wait next to tunnel entrances for opportunities to temporarily gain access to females while the guarding male is distracted, for instance by fighting off a second intruder. Studies have provided evidence consistent with the hypothesis that hornlessness increases maneuverability inside tunnels, suggesting that the absence of horns may be adaptive in the particular behavioral niche inhabited by small, sneaking males \[[@B30]\].
Male morphs also differ distinctly in nature and extent of paternal investment. Horned males generally assist females in tunneling and brood ball production, whereas small, hornless males invest most to all of their time into tunnel defense and the securing of additional mating opportunities \[[@B31]\].
Lastly, behavioral plasticity is not limited to males but also exists in females. Two contexts are especially relevant. First, females typically reproduce by provisioning food for their offspring in the form of brood balls buried underground. In the process, females of at least some species utilize a wide range of dung types and qualities. For instance, *O. taurus* females routinely utilize horse and cow dung in the field. Both dung types differ substantially in quality, and nearly twice as much cow dung than horse dung is needed to rear an adult of similar body size in the laboratory \[[@B32]\]. Individual mothers respond to this variation in dung quality by roughly doubling brood ball masses when offered cow instead of horse dung. Second, females facultatively switch from brood-provisioning behavior to brood-parasitic behavior and the utilization of brood balls constructed by other females \[[@B33]\]. In most cases, a brood-parasitic female will consume the egg inside and either replace it with one of her own while leaving the remainder of the brood ball intact, or incorporate the brood ball into a new, larger brood ball she is constructing herself. Under benign, *ad lib* laboratory breeding conditions up to 13% of brood balls may be affected by such facultative brood-parasitic behavior. This incidence rate roughly doubles when breeding conditions are made adverse by increasing dung desiccation rates \[[@B33]\].
4.4. Physiological Plasticity {#sec4.4}
-----------------------------
Recent studies have discovered an unexpected amount of plasticity in thermoregulatory properties and preferences among morphs, sexes, and species of horned beetles. Specifically, Shepherd et al. \[[@B34]\] observed that the ability to be active at high temperatures increased substantially with male and female body size in a species with a modest sexual and male dimorphism. This was also observed in a second species except for large males, which express extremely large thoracic horns, yet exhibited the thermoregulatory behavior of small, hornless males and females. Using these and additional observations, Shepherd et al. \[[@B34]\] suggested that horn development and possession adversely affect the thermoregulatory abilities of male beetles, and that the magnitude of this effect depended on the degree of horn exaggeration. Specifically, they proposed that large, heavily horned males lack the thermoregulatory ability of their large female counterparts, possibly due to a tradeoff between horn production and investment into thoracic musculature, which plays an important role in the shedding of excess heat in scarab thermoregulation \[[@B35]\]. If so, large horned males may be forced to be active at lower temperatures to avoid risking overheating. Preliminary biochemical analyses of thorax protein content are at least partly consistent with such a scenario (Snell-Rood, Innes, and Moczek, unpublished).
In summary, developmental plasticity pervades the biology of horned beetles, providing rich opportunities to investigate the epigenetic mechanisms underlying plastic responses alongside the ecological and behavioral contexts within which they function and diversify. One genus in particular, *Onthophagus*, has emerged as an especially accessible study system, in large part due to a growing toolbox of developmental genetic and genomic resources. In the next section, we review what we have learned from the application of these tools in the study of the epigenetic regulation of developmental plasticity in these charismatic organisms.
5. Epigenetic Mechanisms Underlying Developmental Plasticity in *Onthophagus* {#sec5}
=============================================================================
5.1. Gene Expression {#sec5.1}
--------------------
Microarray applications to *Onthophagus*horned beetle development have been used to quantify and characterize the degree to which the plastic expression of alternative male phenotypes is associated with changes in gene expression \[[@B36], [@B37]\]. For instance, Snell-Rood et al. \[[@B36], [@B37]\] used microarrays to examine single-tissue transcriptomes of first-day pupae to contrast male morph-specific gene expression with sex- and tissue-specific gene expression. Several important findings emerged from this work. First, if the same tissue type was examined across alternative morphs (and sexes), transcriptional similarities overall far outweighed differences. Second, for those genes that were significantly differentially expressed across morphs, the frequency and magnitude of differential expression paralleled or exceeded that observed between sexes. In other words, if differential expression is used as a metric of developmental decoupling, the development of alternative morphs appeared just as decoupled as did the development of males and females. Lastly, degree and nature of differential expression varied in interesting ways by tissue type. For instance, the transcriptomes of developing head horns in *O. taurus* were more similar between hornless males and females than to the corresponding tissue region in presumptive horned males. In other words, the head horn transcriptome of small, hornless males appeared feminized, which may not be surprising as both females and small males inhibit horn expression. In contrast to head horns, thoracic horns are enlarged in all*O. taurus* males compared to females but develop transiently, such that they are only visible in pupae yet become resorbed prior to the pupal-adult molt. Transcriptomes of thoracic horns for both male morphs were more similar to each other compared to that of females, and a similar pattern was observed in developing legs. Lastly, brain gene expression patterns of large horned males were more similar to females than to small hornless males. In other words, opposite to the situation for head horns, brain transcriptomes of *horned* males appeared more feminized. Combined, these data demonstrate that the development of alternative male morphs is associated with an appreciable amount of differential gene expression, the nature and magnitude of which differs significantly by tissue type.
Additional array experiments (\[[@B38]\]; Moczek et al. in preparation) and a growing number of candidate gene studies (e.g., \[[@B12], [@B39]--[@B44]\]) have now begun to investigate the possible functional significance of genes that are expressed in a morph-specific (on/off) or morph-biased (up/down) manner. Several important findings have emerged from these studies. First, the development of horns appears to rely, at least in part, on the function of conserved developmental pathways such as the establishment of proximodistal axis through leg gap genes \[[@B39]\], growth regulation through TGF*β*- and insulin-signaling \[[@B41], [@B43], [@B45]\], cell-death mediated remodeling during the pupal stage \[[@B40]\], or positioning through Hox- and head gap-genes (\[[@B42]\]; Simonnet and Moczek, unpublished). Second, not all genes expressed during the development of large horns are functionally significant. For instance, the transcription factor *dachshund (dac)*is expressed prominently during the development of both head and thoracic horns, yet RNAi mediated *dac*transcript depletion does not result in any detectable horn phenotypes, despite pronounced phenotypic effects in nonhorn traits \[[@B39]\]. Third, different horn types, whether expressed by different species, sexes, or in different body regions of the same individuals, rely at least partly on different developmental mechanisms and thus may have had different and independent evolutionary histories \[[@B46]\]. Combined, these findings illustrate that the evolution and diversification of horn development have been enabled by the differential recruitment of preexisting developmental mechanisms into new contexts, resulting in a surprising functional diversity within and between species.
5.2. Gene Expression---Future Directions {#sec5.2}
----------------------------------------
Except for a few well-studied models such as the honey bee \[[@B47]\] or *Daphnia* water fleas \[[@B48]\], little is known about the overall genome-wide magnitude and nature of conditional gene expression. Similarly, we know little about how conditional gene expression compares to other forms of context-dependent gene expression, such as tissue-, stage-, or sex-specific expression. Such comparative data are critical to evaluate whether (a) differential expression of largely similar or different gene-sets underlie different types of context-dependent changes in gene expression; (b) the extent of pleiotropic constraints that might delineate evolution of context-dependent gene expression; (c) the degree to which environment-specific gene expression may result in relaxed selection and mutation accumulation.
Studies on *Onthophagus*beetles have made a first attempt to address a subset of these questions. As detailed above, preliminary array studies identified that the development of alternative, nutritionally cued male morphs is associated with a considerable amount of morph-biased gene expression, the nature and magnitude of which exceeded that of sex-biased gene expression for some tissue but not others, a level of complexity likely to be overlooked by whole-body array comparisons \[[@B36]\]. Furthermore, genes with morph-biased expression were more evolutionarily divergent than those with morph-shared expression, consistent with predictions from population-genetic models of relaxed selection \[[@B36], [@B49], [@B50]\] as well as results from other studies (*Drosophila*: \[[@B51]\]; aphids: \[[@B52]\]; bacteria: \[[@B53]\]).
Additionally, recent work has raised the possibility that conditional gene expression, rather than resulting in relaxed selection, is instead enabled by it. Studies on both Hymenoptera \[[@B54]\] and amphibians \[[@B55]\] show that genes expressed in a morph-biased manner exhibit patterns of sequence evolution consistent with relaxed selection not only in polyphenic taxa, but also related taxa lacking alternative morphs. This suggests that genes exhibiting relaxed selection (for whatever reason) may preferentially be recruited into the expression of alternative phenotypes. If correct this would suggest the possibility for positive feedback, as conditional expression would further relax selection, hence further increasing the probability of recruitment into a plasticity context. Lastly, it is conceivable that the initial relaxation of selection that might enable recruitment of genes for the expression of alternative morphs was facilitated by more subtle forms of plasticity and conditional-gene expression in ancestral, monomorphic taxa, such as season- or sex-biased expression. Ultimately, evaluating the relative significance of the *plasticity-first* versus the *relaxed selection-first* hypotheses (and the potential interplay between them) will require a more thorough sampling of transcriptomes across clades, and most importantly, a more thorough understanding of the developmental functions and fitness consequences of conditional gene expression. Research on *Onthophagus* beetles has the potential to contribute to these efforts through the use of recently developed next-generation transcriptomes and corresponding microarrays \[[@B56]\] as well as studies currently under way to analyze patterns of SNP diversity and sequence evolution within and between species.
5.3. Endocrine Regulation {#sec5.3}
-------------------------
Endocrine mechanisms play a critical and well-established role in the epigenetic regulation of insect plasticity (reviewed in \[[@B57], [@B58]\]). Findings supporting a role of endocrine factors in the regulation of polyphenism in *Onthophagus* are derived primarily from hormone manipulation experiments, hormone titer profiling, and more recently, gene expression and gene function manipulation studies, as summarized below.
Juvenile hormone (JH) is a sequiterpenoid hormone secreted by the insect *corpora allata* that maintains the current developmental stage across molts. Applications of a JH analog, methoprene, during *Onthophagus* development provided some of the first evidence that endocrine factors may regulate the expression of alternative nutritionally cued male morphs. Specifically, applications of JH analogs induced ectopic horn expression in *Onthophagus taurus*larvae fated to develop into small, hornless males \[[@B59]\]. In addition, *O. taurus* populations that have diverged in the body size threshold for horn induction showed corresponding changes in the degree and timing of JH sensitivity \[[@B60]\]. Subsequent work on other species has provided additional evidence that JH applications can alter aspects of horn expression, and do so differently for different species, sexes, and horn types \[[@B61]\].
Ecdysteroids play a critical role in initiating the onset of the molting cycle, and for this class of hormones direct titer measurements do exist for a single *Onthophagus* species, *O. taurus*\[[@B59]\]. Expectedly, ecdysteroid titers were observed to increase in male and female *O. taurus* approaching the larval-pupal molt. However, Emlen and Nijhout \[[@B59]\] also observed a small ecdysteroid peak several days earlier during the feeding phase of the last larval instar. This particular peak in ecdysteroid titers was found in female larvae and male larvae fated to develop into the small, hornless morph, but not in males fated to develop into the large, horned morph. Ecdysteroids have been shown to play a major role in inducing changes in gene expression in developing tissues \[[@B62]\] and Emlen and Nijhout \[[@B59]\] therefore suggested that the low ecdysteroid titers observed in female and small male larvae may facilitate development of a hornless morphology in both groups of individuals via a shared endocrine regulatory process. However, ecdysteroid titers have never been replicated in this or any other *Onthophagus* species, and functional tests using ectopic ecdysteroid applications failed to confirm a function of the early ecdysteroid peak in both females and small males (D.J. Emlen, personal communication).
Most recently, transcriptional profiling combined with candidate gene studies have provided additional, albeit somewhat indirect support for a role of endocrine regulators during horned beetle development. For instance, Kijimoto et al. \[[@B40]\] investigated the dynamics of programmed cell death during horn remodeling using cell death-specific bioassays. Integrating findings from a companion microarray study, the authors also showed that several genes known to be associated with ecdysteroid signaling in *Drosophila* were expressed in a manner consistent with a role of ecdysteroid signaling in the regulation of horn-specific programmed cell death. Similarly, a combination of candidate gene expression data \[[@B45]\] and array-based transcriptional profiling \[[@B37], [@B40]\] has begun to implicate signaling via insulin-like growth factors in the regulation of male horn polyphenism. A subsequent functional analysis of *FoxO*\[[@B43]\], a key growth inhibitor in the insulin pathway, has now provided the first functional data in support of such a role (and see below).
5.4. Endocrine Regulation---Future Directions {#sec5.4}
---------------------------------------------
Despite the progress summarized above, our understanding of how endocrine mechanisms influence *Onthophagus* development and behavior lag far behind what is known in other insect model systems, such as photoperiodically cued wing dimorphism in crickets (reviewed in \[[@B63]--[@B65]\]) and nutritionally cued caste-development in honey bees (reviewed in \[[@B66], [@B67]\]). Furthermore, most insights, in particular pertaining to juvenile hormone, have been derived solely from hormone manipulation experiments, whose lack of precision and possible pharmacological side effects limit confidence in the results \[[@B63]\]. While these data are consistent with a functional role of JH in the regulation of developmental plasticity in horned beetles, it is worth noting that direct JH titer profiles have yet to be empirically determined across morphs and sexes for any *Onthophagus* species. Furthermore, direct functional interactions between JH and potential targets relevant for the development of alternative male morphs have yet to demonstrated. Consequently, existing models of JH\'s role in the development and evolution of horn polyphenism remain largely hypothetical and await critical experimental validation. A recent study by Gotoh et al. \[[@B68]\] is now the first to combine observations of hormone titers with manipulation experiments to demonstrate the role of juvenile hormone in promoting mandible length in a stag beetle, a group of beetles closely related to the Scarabaeidae. These findings motivate complementary studies in horned beetles, which now appear particularly feasible given the recent development of many critical resources.
Research advances in determining gene function and comparative gene expression have raised the possibility that work in the near future will be able to ascertain more clearly the role of hormones in *Onthophagus* ontogeny, characterize the interplay between genetic and endocrine regulators of development, and examine their respective evolution across species that have diverged in nature and magnitude of developmental plasticity. For example, RNA interference protocols now work routinely and reliably in *Onthophagus* beetles and have already permitted comparative gene function analyses of a variety of key developmental regulators \[[@B39], [@B41], [@B42], [@B69]\], including components of endocrine pathways \[[@B43]\], providing numerous avenues for future research. Furthermore, next-generation transcriptomes \[[@B56]\] of at least two species have massively increased access to relevant sequence information, with additional transcriptomes of other *Onthophagus* species forthcoming.
5.5. DNA Methylation {#sec5.5}
--------------------
The role of DNA methylation in development and developmental plasticity of *Onthophagus* beetles is still poorly understood, but preliminary evidence suggests that these organisms could be an important system in which to better understand the genetic underpinnings and evolutionary consequences of methylation. First, *O. taurus* has joined the ranks of other emerging insect models, including honeybees, aphids, and parasitic wasps, in containing a complete set of methylation machinery, such as the *de novo* methyltransferase (dnmt3) and the maintenance methyltransferase (dnmt1) \[[@B56], [@B70]--[@B72]\]. Second, a pilot study now suggests that differential methylation is associated with nutritional environment in at least one species, *O. gazella*, and correlated with performance across nutritional environments \[[@B73]\]. This study used a methylation-specific AFLP analysis to survey methylation patterns in family lines derived from a wild population and reared in two different dung types across successive generations. Two major findings emerged. First, methylation state was most heavily influenced by genotype (family line), then rearing environment (dung type), as well as genotype-by-environment interactions (different lines tended to be methylated at different sites when reared on different dung types). Second, methylation state had a significant effect on performance, measured as body size, but in a surprisingly sex- and environment-specific manner: methylation state affected the performance of males (but not females) on cow dung, with the reversed pattern observed on horse dung. Intriguingly, the family line with the greatest flexibility in methylation across environments also showed the highest consistent performance across those environments. Combined, these data are consistent with the hypothesis that facultative methylation underlies adaptive, plastic responses to variation in nutritional environment.
5.6. DNA Methylation---Future Directions {#sec5.6}
----------------------------------------
The patterns, function, and phenotypic consequences of DNA methylation in insects have received increased attention in recent years, in part for two major reasons. First, insects were once thought to be devoid of methyltransferase enzymes as found in mammals due to the lack of such machinery in the model insect *D. melanogaster*. Subsequent studies have shown that DNA methylation is also absent in two other major invertebrate models, the beetle *T. castaneum* and the nematode *C. elegans* \[[@B74]\]. Phylogenetic reconstructions now suggests rather than reflecting ancestral states, all three lineages have lost aspects of DNA methylation independently \[[@B75]\]. This now provides a unique opportunity to determine the relevance of DNA methylation in development and evolution of phenotypic diversity, plasticity, and integration. Second, genomic methylation patterns and their impact upon transcription in insects are very different from patterns in other taxa. In mammals, genomes are heavily methylated, both in intergenic and intragenic regions, and are generally associated with gene silencing (reviewed in \[[@B76], [@B77]\]). In many invertebrates, however, genomes appear to be mosaically methylated, with methylation occurring disproportionately in intragenic regions of constitutively expressed housekeeping genes (reviewed in \[[@B77]\]). Thus, studies in emerging and nonmodel insects could allow further understanding of the function of DNA methylation in transcriptional and posttranscriptional regulation \[[@B78]\].
In establishing a correlation between methylation patterns, diet, and performance (body size), the study by Snell-Rood et al. \[[@B73]\] summarized above raised the intriguing possibility that methylation patterns influenced by diet could mediate plastic responses during development in *O. gazella*. If correct, the incredible diversity in nutritional responses that exist within and among *Onthophagus* species would provide a remarkable opportunity to explore the evolutionary diversification of methylation-mediated nutritional plasticity. Such studies would be especially powerful if methylation patterns could be linked to gene regions (e.g., through the use of bisulfite sequencing approaches) and replicated separately for different tissue types, such as gut, epithelium, and brain tissue.
5.7. Conditional Crosstalk between Developmental Pathways {#sec5.7}
---------------------------------------------------------
The growing number of studies investigating the genetic regulation of horned beetle development has begun to provide the first insights into how different developmental pathways and processes might interact, including facultative interactions depending on nutritional conditions. For instance, Kijimoto et al. \[[@B44]\] investigated the role of *Onthophagus doublesex (dsx)*, a transcription factor known to regulate the sex-specific expression of primary and secondary sexual traits in diverse insects (reviewed in \[[@B79]\]). As in other taxa, *Onthophagus dsx* is alternatively spliced into male- and female-specific isoforms, and consistent with findings from other studies, *male-dsx (mdsx)* and *female-dsx (fdsx)* isoforms promote horn development in male and inhibit it in female *O. taurus*, respectively. Remarkably, *O. taurus mdsx* appears to have evolved the additional function to regulate the development of male horn polyphenism, as evidenced by the following observations. First, *mdsx* is expressed at much higher levels in the head and thoracic horn primordia of large males compared to their legs or abdomen, or when compared to any tissue examined in smaller males. Second, *mdsx*RNA*i* dramatically reduced horn expression in large males only, but left smaller males unaffected. Intriguingly, downregulation of *fdsx* in female *O. taurus* resulted in the nutrition-dependent *induction* of ectopic head horns. Combined, these data suggest that sex- and tissue-specific *dsx* expression and function underlie not only sexual dimorphism, but also male polyphenism in horn expression \[[@B44]\]. The utilization of *dsx* as a regulator of both sexual and male dimorphism may also explain the tight coevolution of both patterns of phenotype expression as reported by earlier phylogenetic studies \[[@B20]\], which found that 19/20 instances of gain or loss in sexual dimorphism were paralleled by a corresponding gain or loss of male dimorphism. Exactly how *dsx* expression and function may be coupled to nutritional input, however, is presently unclear, though several promising candidate mechanisms exist.
One such candidate is signaling via insulin-like peptides, a pathway well-known for its role in coupling nutritional variation to a wide range of developmental responses, including growth \[[@B80]\]. Differential expression of members of the insulin signaling pathway during facultative horn development have been documented by both a candidate gene study on the insulin receptor \[[@B45]\] as well as array-based transcriptional profiling \[[@B37], [@B40]\]. The latter studies identified a particularly intriguing member of this pathway, the *forkhead box subgroup O* gene, also known as *FoxO*, as being differentially expressed across several tissue types and nutritional responses. *FoxO* is a growth inhibitor which is typically activated during poor nutritional conditions. Array-based expression evaluations suggested that, relative to abdominal tissue of the same individual, the horn primordia of insipient large males showed much lower *FoxO* expression than the horn primordia of small males, consistent with a role of *FoxO* inhibiting horn growth in small, but not large, males. More detailed qRT-PCR-based expression analyses revealed that contrary to these initial inferences, *FoxO* was not differentially expressed in the horn primordia of large and small male *O. taurus*, but was instead overexpressed in the abdomen of large males, in particular in regions associated with the development of genitalia, including testes. In comparison, the abdomen of small males showed reduced *FoxO* expression. Thus, *FoxO* expression differences in the abdomen of large (high) and small (low) males, rather than expression differences in their horn primordia, accounted for the initial array-based expression data.
Recall that small males, while reducing investment into horns, invest heavily into genital development, in particular testes mass and ejaculate volumes \[[@B23], [@B81], [@B82]\]. Low *FoxO* expression in presumptive testes tissue is consistent with a role of *FoxO* in the upregulation of testicular growth in small males relative to more inhibited growth, marked by elevated *FoxO* expression, in large males. Subsequent RNAi-mediated depletion of *FoxO* transcripts resulted in extended development time and larger body size at eclosing, consistent with a general disinhibition of growth. Moreover, *FoxO-*RNA*i*disrupted the proper scaling of male body size with copulatory organ size, further supporting that *FoxO* may regulate morph-specific genitalia development in horned beetles \[[@B43]\]. In particular, small male genitalia lost their body size dependence whereas large male genitalia exhibited reduced development. Lastly, *FoxO-*RNA*i*modestly but significantly increased the length of horns in large males. Since *FoxO* is *not* differentially expressed in different horn primordia, this finding suggests that elevated horn development observed in large RNAi males might be a secondary consequence of *FoxO*-RNAi-mediated reduction in genitalia development in those same males. More generally, these results raise the possibility that *FoxO* regulates relative growth and integration of nutrition-dependent development of body size, horn length, and genitalia size.
5.8. Conditional Crosstalk---Future Directions {#sec5.8}
----------------------------------------------
How different body parts and tissue types communicate with each other during development, and how their varied scaling relationships are enabled along a continuum of body sizes and in the face of nutritional variation, represent long-standing questions at the interface of developmental and evolutionary biology. Answering these questions is critical to our understanding of the nature of phenotypic integration. Horned beetles are now uniquely positioned as a model taxon in which to identify, on one side, nutrition-responsive developmental pathways and the nature of their interactions with other pathways during development of different body parts and tissues. On the other, the diversity of nutritional responses that exist within and among sexes, populations, and species all provide fantastic substrate for future research efforts into the developmental causes and evolutionary consequences of phenotypic integration.
6. Opportunities and Challenges in *Onthophagus* Epigenetics {#sec6}
============================================================
6.1. Stepping Back {#sec6.1}
------------------
Adaptive developmental plasticity allows organisms to modulate their phenotype in response to external environmental cues, permitting developing organisms to better cope with variation in resource availability, physical environment, and social contexts \[[@B2]\]. Plasticity has been of interest to biologists for over a century, and the increased accessibility of molecular data and technology is now enabling an exploration of the molecular underpinnings of this developmentally, ecologically, and evolutionarily central phenomenon \[[@B83]\]. Epigenetic processes have emerged as a diverse and important collection of mechanisms that mediate the interaction between environment and the genome at multiple scales, enabling the expression of developmentally plastic phenotypes (reviewed in \[[@B83], [@B84]\]). Studies of traditional model organisms have provided powerful insights into the nature and consequences of epigenetic mechanisms. For example, through murine models we have learned that endocrine disruptors, such as the pesticide vinclozolin, can impact not only an exposed individual, but can lead to physiological and behavioral changes in unexposed offspring and grand-offspring. Furthermore, gene knockout lines have subsequently allowed researchers to elucidate some of the molecular underpinnings of this particular phenomenon, mainly epimutations in the germline (reviewed in \[[@B85], [@B86]\]). Although model organisms are clearly useful for investigating mechanisms underlying epigenetic processes, studies in these organisms have limited power to investigate the relative significance of epigenetics in naturally occurring populations. For instance, many laboratory strains of model organisms are highly inbred, and likely fail to capture the richness of genetic and epigenetic variation found in natural populations \[[@B87]\]. Similarly, one reason that many model organisms were initially selected is that they are phenotypically resilient to variation in the environment, making the study of plasticity in these organisms difficult \[[@B87]\]. New models will thus be important in addressing questions regarding the role of various epigenetic processes in regulating developmental plasticity.
Here, diet-induced plasticity stands out as a particularly important and widespread form of plastic development. Variation in diet quality represents a challenge faced by most, if not all, heterotrophic organisms, and numerous diverse developmental strategies have evolved to cope with diet variation. Moreover, understanding how diet and genes interact during development to form adult phenotypes is essential to understanding how experiences in early life can promote trajectories toward disease later on. Here, we contrast these findings to what is known about the epigenetic control of plasticity in other emerging and established insect models, and close by highlighting several research areas in which future research on *Onthophagus*beetles could potentially contribute to the growing knowledge of the role of epigenetics in regulating developmental plasticity in general and diet-induced plasticity in particular.
6.2. The Development and Evolution of Shape {#sec6.2}
-------------------------------------------
Much variation in organismal shape is the product of evolutionary tinkering in the location, allometry, or function of preexisting structures. Thus, the ultimate factors that promote diversification of shape, as well as the proximate underpinnings that coordinate adaptively proportioned traits, are both of fundamental interest in evolutionary-developmental biology. Adaptive radiations, textbook examples of extensive phenotypic variation stemming from a single ancestral phenotype, have long been used as models to address questions of both ultimate and proximate causes of shape evolution (reviewed in \[[@B88]\]). For instance, the flexible stem hypothesis, pioneered by West-Eberhard \[[@B2]\], suggests that phenotypic diversification observed in adaptive radiations results from selection upon ancestral phenotypes made possible by developmental plasticity. Specifically, ancestral plasticity links the expression of conditional phenotypic variants to particular inducing conditions, thus delineating the nature of phenotypic variation that selection can later act upon in different environments. The flexible stem hypothesis therefore has the potential to explain the common observation of very similar phenotypes arising repeatedly yet independently during adaptive radiations (e.g., \[[@B11], [@B89]\]). More generally, this hypothesis highlights the potential importance of preexisting plasticity in enabling any kind of evolutionary change, including changes in shape and scaling, by creating the potential for facultatively expressed trait variants to become genetically stabilized and accommodated in descendent generations (see also next section).
*Onthophagus* beetles provide several interesting opportunities to explore the role of plasticity in the diversification of shape and scaling relationships. For instance, adult thoracic horns emerge during development from pupal precursors that originally carried out a very different function \[[@B90]\]. Ancestrally, pupal thoracic horns were resorbed prior to the adult molt, yet descendent species have evolved various ways of partially or fully retaining thoracic horns into adulthood and shaping them into sex- and species-specific weapons. In a subset of species, degree of resorption itself is nutrition dependent \[[@B91]\]. Furthermore, spontaneous retention of thoracic horns also can be observed on occasion in laboratory colonies of species that normally constitutively resorb horns, possibly in response to stressful environmental conditions \[[@B40]\]. This raises the possibility that the diversification of thoracic horn shape and size may have been made possible by harnessing some of the condition dependency of horn retention that existed in ancestral taxa.
A second example involves the well-defined body size thresholds separating alternative horned and hornless male morphs in many species. The exact location of this threshold has diversified greatly among species ([Figure 3](#fig3){ref-type="fig"}) as well as some populations. In *O. taurus*, for instance, exotic populations in the Eastern United States, Eastern Australia, and Western Australia have diverged remarkably from their Mediterranean ancestor since introduction approximately 50 years ago \[[@B92]\]. Some of these divergences are similar in magnitude to those observed between well-established species. Intriguingly, body size thresholds are also subject to seasonal or geographic fluctuations in larval nutrition \[[@B60], [@B93]\] brought about by changes in dung quality and/or changes in the intensity of competition over breeding resources. Again, this raises the possibility that some of the threshold divergences observed between populations and species may have been facilitated initially by conditional responses to altered growth or social conditions.
6.3. Evolution via Genetic Accommodation {#sec6.3}
----------------------------------------
Genetic accommodation posits that environmental conditions interacting with developmental processes generate phenotypic transformations that can subsequently be stabilized genetically through selection operating on genetic variation in a population. Genetic accommodation does not require new mutations to occur, but will take advantage of them alongside standing genetic variation. Evolution of novel traits and norms of reaction by genetic accommodation have been demonstrated repeatedly and convincingly in artificial selection experiments (reviewed in \[[@B5]\]). Similarly, studies on ancestral plasticity and cases of contemporary evolution provide growing evidence consistent with a role of genetic accommodation in diversification of natural populations (e.g., \[[@B11], [@B94]\]). However, exactly how important environmental induction really is in the origin and diversification of novel phenotypes remains largely to be determined, in particular in natural populations. Similarly, the proximate mechanisms underlying plasticity-mediated diversification are largely unknown.
The preceding section highlighted two examples, the diversification of thoracic horn size and shape and the diversification of size thresholds, where research on horned beetles has the potential to generate valuable case studies on the mechanisms and consequences of genetic accommodation of initially environment-induced phenotypic variation. Many additional opportunities exist. For instance, female *Onthophagus* facultatively engage in intra- and possibly interspecific brood parasitism \[[@B33]\]. Interspecific brood parasitism is the dominant reproductive strategy in other dung beetle genera, raising the possibility that it may have evolved initially as a conditional alternative that became subsequently stabilized in a subset of descendent lineages \[[@B13]\]. Similarly, extent of maternal care (brood ball size and depth of burial) vary greatly among females, in part as a function of female body size and thus the nutritional conditions a mother herself experienced when she was a larva. Importantly, *O. taurus* populations obtained from different latitudes within the Eastern US have diverged significantly in the extent of investment mothers provide, again raising the possibility that some of these divergences were enabled initially by plastic responses to environmental conditions (Snell-Rood and Moczek, unpublished data). As highlighted in the last section, *Onthophagus*beetles also provide great opportunities to begin exploring some of the proximate genetic, developmental, and physiological mechanisms that may facilitate accommodation of conditionally expressed phenotypes.
6.4. The Origin of Novel Traits {#sec6.4}
-------------------------------
How complex novel traits, such as the eye, the firefly lantern, or the turtle shell, originate is among the most fundamental yet unresolved questions in evolutionary biology \[[@B46]\]. Evolution operates within a framework of descent with modification---anything new and novel must have descended from something old and ancestral. Yet novelties are generally defined as lacking obvious correspondence, or homology, to preexisting traits. How then, do novel traits originate from within the confines of ancestral variation? Studies of epigenetic mechanisms in general, and those focusing on non-model organisms in particular, have likely much to offer to address this question.
Traditional developmental biology and evo-devo are focused on the identification of genes and gene networks that regulate development and developmental outcomes. At times, this view is expanded to make room for environmental influences by viewing gene function as environment dependent, and viewing genotypes as possessing a reaction norm---that is, the range of phenotypes produced across a range of environmental conditions. The study of epigenetics takes a radically broader and far less gene-centric view. Here, phenotypes (from nucleotide sequences to cells, tissues, organisms, and social groups) emerge as the products of developmental processes to which genes contribute important interactants. In this view, genes are critical and genetic changes can make important differences, but they do not make traits or organisms. Instead, those emerge through the actions of development. This more integrative perspective has many important consequences, three of which are especially critical here. First, epigenetic processes facilitate the production of integrated and functional phenotypes through a wide variety of mechanisms operating well above the sequence level \[[@B95], [@B96]\]. Second, the integration put in place by epigenetic mechanisms allows development---when confronted with environmental perturbations---to give rise to possibly novel but nevertheless integrated, functional, and on occasion adaptive phenotypes. Third, the same integration enabled by epigenetic mechanisms allows random and modest genetic change to give rise to nonrandom, functional phenotypic changes. In short, the integrity and functionality of phenotypes in development and evolution are facilitated through the chaperoning action of epigenetic mechanisms. As such epigenetics likely plays a central role in facilitating innovation and diversification in nature.
*Onthophagus* beetles have begun to contribute to our understanding of innovation through epigenetic mechanisms through a series of studies focused on the origin and diversification of horns, themselves novel structures lacking any obvious homology to other insect traits (reviewed in \[[@B97]\]). Through a combination of observational, comparative, and manipulation studies it has now become clear that at least some horns originated from pupal-specific structures that originally functioned in completely unrelated contexts (reviewed in \[[@B13]\]). Innovation was enabled initially through the potentially accidental maintenance of normally pupal-specific projections into the adult stage. Similar events can be observed at low frequency in laboratory cultures of species lacking adult horns \[[@B40]\]. Diversification between species, sexes, and morphs was then made possible through the recruitment of preexisting developmental pathways and their targets into a novel context, for instance enabling morph-specific elaboration of horns via preexisting endocrine mechanisms \[[@B59], [@B61], [@B92], [@B98]\] or sex-specific horn expression via sex-specific activation of programmed cell death \[[@B40]\]. Exactly how such recruitment was made possible and by what kind of genetic and environmental variation (and what interactions between them) remain unclear, however, posing some of the many intriguing question for future research in these organisms and the field in general.
7. Conclusions {#sec7}
==============
The study of epigenetic mechanisms in development and evolution promises to fill an otherwise abstract genotype-phenotype map with biological reality. Epigenetic mechanisms feature especially prominently in developmental plasticity and its evolutionary consequences. We hope to have shown in this review that the study of horned beetles provides rich and promising opportunities to investigate the role of epigenetics in the evolution of adaptations, phenotypic diversification, and the origin of novel traits. The remarkable degree of plasticity inherent in the biology of horned beetles, combined with the stunning phenotypic diversity that exists both within and among species, and the growing experimental toolbox available for a subset of these organisms makes horned beetles a promising emerging model system in the study of epigenetic mechanisms, their nature, causes, and consequences.
The authors thank the Editors of this special issue for the opportunity to contribute this paper, and Amy Cash and two anonymous reviewers for constructive comments on earlier drafts. Research presented here was funded in part by NSF Grants IOS 0445661, IOS 0718522, and IOS 0820411 to A. P. Moczek. The content of this paper does not necessarily represent the official views of the National Science Foundation.
{#fig1}
{#fig2}
{#fig3}
[^1]: Academic Editor: Eveline Verhulst
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Retinitis pigmentosa (RP; MIM\# 268,000) is the most frequent subtype of inherited retinal disease and is clinically and genetically a highly heterogeneous disorder \[den Hollander at al., 2010\]. Fifty-two genes are known to be associated with nonsyndromic RP, involving all modes of inheritance \[Berger et al., [@b6]\]. A further 59 genes are known to underlie other subtypes of syndromic and nonsyndromic retinal diseases (RetNet; http://www.sph.uth.tmc.edu/Retnet/). With a few exceptions, there are no ophthalmologic characteristics specifically associated with the genetic subtypes of RP, impeding the prioritization of genes for analysis by Sanger sequencing.
Molecular diagnostics is particularly challenging for isolated RP patients, who constitute majority of RP cases \[van den Born et al., [@b52]; Najera et al., [@b40]; Hayakawa et al., [@b22]\]. Because of the unknown mode of inheritance, mutations in any of the 52 known RP genes may be causative. The most widely applied diagnostic test for allelic and genetic heterogeneous diseases, arrayed primer extension (APEX) chip technology, is only able to detect known mutations \[Ávila-Fernández et al., [@b2]\]. In addition, these chips are designed to separately test for the presence of mutations in autosomal dominant or recessive RP genes, resulting in a diagnostic yield for autosomal recessive RP of only ∼10% \[Ávila-Fernández et al., [@b2]\]. Altogether, the yield of diagnostic testing has remained disappointingly low for RP patients, despite many important disease gene discoveries in the last two decades \[Berger et al., [@b6]\].
In this study, we analyzed all known inherited retinal dystrophy (RD) genes in parallel by targeted next-generation sequencing (NGS) in a cohort of 100 RP patients. All identified genetic variants entered a systematic data analysis pipeline that included Sanger sequencing validation, segregation analysis, and a bioinformatic prediction of pathogenicity.
Materials and Methods
=====================
Clinical Diagnosis of RP
------------------------
The diagnosis of RP was made in all individuals on the basis of ophthalmologic examination, including best-corrected visual acuity, slit-lamp biomicroscopy, ophthalmoscopy, and fundus photography. Electroretinograms, recorded according to the protocol of the International Society for Clinical Electrophysiology of Vision \[Marmor et al., [@b34]\], and Goldmann visual field measurements were available for majority of patients.
Previous Genotyping and Patient Ascertainment
---------------------------------------------
A cohort of 234 RP patients had been collected over a period of 15 years (Supp. Fig. S1). Mutation screening by APEX analyses and Sanger sequencing had identified a molecular diagnosis in 20 patients \[Cremers et al., [@b14]; den Hollander et al., [@b18]; Maugeri et al., [@b35]; den Hollander et al., [@b17]; Yzer et al., [@b56]; Klevering et al., [@b28]\]. In 186 of the remaining probands, genome-wide homozygosity mapping had been performed, which had resulted in the identification of mutations in three novel autosomal recessive RP genes, *EYS* (MIM\# 612,424), *C2ORF71* (MIM\# 613,425), and *IMPG2* (MIM\# 607,056) \[Collin et al., [@b11]; Bandah-Rozenfeld et al., [@b4]; Collin et al., [@b12]\] in 13 patients, together with 24 homozygous mutations in previously known RD genes \[Collin et al., [@b13]\]. From the remaining 177 probands, 100 were selected for the targeted NGS analysis. This selection was based on the availability of DNA samples from both patients and their family members. It included 78 cases with nonsyndromic isolated RP and 22 cases with autosomal recessive RP (e.g., unaffected parents and two or more affected siblings), all without a molecular diagnosis. No apparent dominant RP cases were present in the cohort. Affected and nonaffected relatives were either included prior to this study or requested to participate after the identification of potentially causative variants in the proband. Some patients were clinically re-evaluated upon identification of potentially causative genetic variants.
After explaining the nature of this study, informed consents adhering to the tenets of the Declaration of Helsinki were obtained from all patients and their relatives. DNA was extracted from peripheral blood using standard procedures \[Miller et al., [@b37]\].
In addition to the 100 selected patients, DNAs from 12 cases with various types of autosomal recessive retinal diseases, carrying known compound heterozygous mutations in genes represented on the NGS array (Supp. Table S1) were investigated to design and optimize the current approach.
Array-Based Sequence Capture and Targeted Resequencing
------------------------------------------------------
To enrich multiple DNAs in a single procedure, a 12-plex NimbleGen sequence-capture array (Roche NimbleGen, Madison, WI) consisting of 12 subarrays of 135K oligonucleotides was used ([Fig. 1A](#fig01){ref-type="fig"}). Probes targeting all coding exons, noncoding exons, and untranslated regions of 111 known blindness genes were included on these arrays, as well as probes for a fragment of intron 26 of *CEP290* (MIM\# 610,142) that harbors a frequent mutation causing Leber congenital amaurosis (MIM\# 204,000; Supp. Table S2). The *RPGR* (MIM\# 312,610) exon ORF15 was not included because of highly repetitive regions hampering enrichment and sequencing. The array design comprised a total of 723,662 bases targeting 111 genes consisting of 2,011 individual regions (each individual target region mostly targets the sequence of a known exon and is at least 250 bp long). Sequence capture was performed following the "Titanium Optimized Sequence Capture Array Delivery" protocol (version 1.0), as supplied by Roche and optimized for sequence capture by NimbleGen arrays. Minor changes were made to adapt this protocol (for 385K arrays) to the 12-plex format. In brief, 5 μg of genomic DNA per sample were used in the preparation of DNA for sequence-capture hybridization. DNA was sheared using the Covaris S2 system (Covaris Inc., Woburn, MA) according to the instructions by the manufacturer for 500 bp fragments. Molecular-identifier-adapter ligation was performed as described in Technical Bulletins TCB \#004-2009 and TCB \#005-2009 (Roche NimbleGen). After pre-ligation-mediated polymerase chain reaction (LM-PCR), a final mass of 1.125 μg prepared sample and 35.5 μg Cot1 DNA was co-hybridized to each subarray. The reagent volume used for hybridization was reduced in proportion to the smaller loading volume of the 12-plex format (6 μl). Samples were eluted 72 hr after hybridization and subsequently amplified by post LM-PCR, resulting in a 12-plex sequencing library. Small-volume emulsion PCRs (emPCRs) were performed for each library using four different DNA concentrations as input to generate a titration curve, to determine the optimal input of DNA for each run. Based on that titration curve, a large-volume emPCR was performed and sequencing of each library was carried out on Roche GS FLX (454 Life Sciences, Branford, CT) sequencer with Titanium series reagents.
{#fig01}
Assembly and Variant Calling
----------------------------
Sequence reads were mapped against the human reference genome (hg18) using the Roche Newbler software (version 2.3; 454 Life Sciences). Signal processing parameters were changed to use less stringent quality filtering. Additional mapping and coverage statistics were extracted from the mapping output files using custom software. Variations were considered as high-quality differences when either (1) the variation was sequenced in at least three nonduplicate reads including at least one forward and one reverse read, or (2) the variation was seen in at least five reads with quality scores greater than 20.
Variants were annotated using a custom annotation pipeline adding the following annotation for each detected variant: annotation with known polymorphism data from dbSNP130, genomic feature annotation (e.g., exon and intron), amino acid translations, and PhyloP conservation score (based on conservation of 44 vertebrate species) \[Hoischen et al., [@b25]; Vissers et al., [@b53]\]. Additionally, variants were annotated with frequencies of the variants in (1) all 100 blindness samples, (2) an internal database of 86 exomes, (3) an internal database with known pathogenic blindness mutations, and (4) an internal database with mutations from the professional version of the Human Gene Mutation Database (HGMD).
Nucleotide numbering reflects cDNA numbering with +1 corresponding to the A of the ATG translation initiation codon in the reference sequence, according to the journal guidelines (www.hgvs.org/mutnomen). The initiation codon is codon 1.
Quality Control and Coverage
----------------------------
Coverage was calculated by counting the number of sequenced bases mapping to the target regions. Bases mapping to regions within a 500 bp range of a target were considered "near target." From this, the average coverage across all the regions was calculated for each sample (Supp. Tables S3A and S3B). To evaluate the performance of targets, we calculated the average coverage of each target across all 100 samples (Supp. Table S2). Evenness of coverage was calculated for all samples according to Mokry et al. \[[@b38]\] ([Fig. 1D](#fig01){ref-type="fig"} and Supp. Tables S3A and S3B).
Prioritization of Variants
--------------------------
Variants were selected for follow-up when all of the following criteria applied: Variants were either nonsynonymous variants or splice site variants (8 bp splice acceptor, 20 bp splice donor).Variants were not included in dbSNP130 unless they were reported in HGMD or were previously defined as a pathogenic blindness mutation.Variants did not occur with the 86 in-house exomes with a frequency greater than 5%.Variants did not occur with a frequency of more than 15% within all 100 blindness samples.Variants were covered by at least 10 reads with at least 20% variation reads or at least 5 reads with 80--100% variation reads.Variants were consistent with the known pattern of inheritance of the respective gene (i.e., homozygous/compound heterozygous or heterozygous).
If prioritization resulted in a single strong candidate allele (a nonsense or canonical splice site variant) in a recessively acting gene, a manual search for a second variant allele was performed in the unfiltered variant set. Variants that were selected for further analysis were validated using Sanger sequencing and analyzed in all available relatives.
Data Validation and Segregation Analysis
----------------------------------------
Variant validation was performed using conventional Sanger sequencing. Dependent on the selection criteria described above, between one and 17 variants per sample were selected for validation (on average, 3.6 variants per sample; Supp. Tables S3A and S3B), leading to a total of 359 validated variants in the cohort of 100 samples. Where available, DNAs from additional family members were sequenced to enable segregation analysis.
Determination of Pathogenicity of Variants
------------------------------------------
To systematically determine the pathogenicity of genetic variants, segregating variants were sorted based of the type of variation. Nonsense, frameshift, and canonical splice site variants were considered to be pathogenic. Based on existing guidelines \[Bell et al., [@b5]\], we developed a classification system for missense changes as well as for noncanonical splice-site variants. Evidence was based on (1) in silico evidence and (2) co-occurrence of two mutations in a recessive gene. With regard to in silico evidence, three different features were considered important: (a) missense or splice-site prediction software (acting on amino acid level), (b) evolutionary conservation (acting on nucleotide level), and (c) population frequencies (Supp. Fig. S2).
### Missense/splice-site prediction tools
In the case of missense prediction programs (SIFT \[Ng & Henikoff, [@b41]\], PolyPhen \[Adzhubei et al., [@b1]\], MutPred \[Li et al., [@b29]\]), the different scores for the three different tools were combined by a majority vote resulting in a single classification as "pathogenic," "unknown," or "benign." For the splice-site prediction programs (Splice Site Finder, MaxEntScan, NNSplice), wild-type score (wt) and mutation score (mut) were compared using the following cutoffs. Splice-site finder: if wt \> 50 and mut \< 50, then the prediction is "pathogenic." If the difference between wt and mut is \>5, then the prediction is "unknown," otherwise the prediction is "benign." MaxEntScan: if the difference between wt and mut is \>0.8, then the prediction is "pathogenic." If there is no difference between wt and mut, then the prediction is "benign," else the prediction is "unknown." NNSplice: if wt \> 0.5 and mut \< 0.5, then the prediction is "pathogenic." If the difference between wt and mut is \>0.05, it is "unknown,"; otherwise it is "benign" (see Supp. Fig. S2). The different scores for the three different tools were combined by a majority vote, resulting in a single classification as "pathogenic," "unknown," or "benign."
### Evolutionary conservation
For a classification based on evolutionary conservation, all variants with a PhyloP (44 vertebrate species) score of less than 1 were considered "benign," all variants with a PhyloP value above 2.5 were considered to be "pathogenic," and variants with intermediary values were classified as "unknown" (cutoffs were based upon a comparison of evolutionary conservation scores of dbSNP \[build 130\] and HGMD, as described in Vissers et al. \[[@b53]\]).
### Population frequencies
We established the frequencies of variants within 86 whole-exome sequencing samples of patients with unrelated disorders, as well as in the 100 RP samples. Variants found at a frequency greater than 3% in the exome samples were classified as "benign," and variants with an exome frequency between 1% and 3% or variants with a frequency of 3% or more in the 100 blindness samples were classified "unknown." The remaining variants were classified as "pathogenic." The three classifications (i.e., prediction tools, conservation, and frequency) were combined according to the following rules: If the prediction based on frequencies was "benign," then the final classification of the variant was "probably benign" regardless of the other predictions. In all other cases, the three predictions were combined into a single prediction using a majority vote. In case of three different votes, the variant was predicted as "unknown."
As a result, all segregating (pairs of) variants were classified as either probably pathogenic, unknown, or benign.
Biostatistical Analyses
-----------------------
The diagnostic performance in this study was evaluated by calculating the mutation detection rate as a percentage of mutations detected for the cohort of 12 patients with 24 known mutations, and by calculating the diagnostic yield as the percentage of patients for which a molecular diagnosis was obtained (for details, see *Results*).
Results
=======
Targeted Resequencing of 111 RD Genes
-------------------------------------
This study has rigorously tested the power of large-scale targeted resequencing to perform molecular genetic analysis for RP, one of the most genetically heterogeneous human diseases. The coding regions of 111 RD genes were enriched by target capture and screened for mutations by NGS in 112 subjects---12 RD patients carrying two known heterozygous mutations in one of the retinal disease genes and 100 RP patients without a molecular diagnosis. The average amount of mappable sequence data per sample was 31 Mb, resulting in an evenly distributed average coverage of 26× per exon per sample (evenness score = 97.3%; [Figs. 1B--1D](#fig01){ref-type="fig"}). On average, 89% of all bases were covered at least 10×. Among the 2,011 targeted exons, only 15 exons (0.7%) were covered poorly (less than five times on average), either due to high GC contents or the abundant presence of repetitive sequences ([Figs. 1E and 1F](#fig01){ref-type="fig"}).
Development and Application of a Variant Prioritization Pipeline
----------------------------------------------------------------
Automated variant detection for all 112 samples resulted in an average of 1,274 variants per sample (Supp. Tables S3A and S3B). A systematic variant prioritization tool was developed to identify the pathogenic mutation(s) amongst this substantial number of variants. To assess and optimize the performance of this methodology, 12 retinal disease samples, each carrying two known mutations, were analyzed. In these 12 patients, a total of 14,144 genetic variants were automatically detected, including 21 of the 24 known mutations ([Fig. 2A](#fig02){ref-type="fig"}), resulting in a technical mutation detection rate of 87.5%. Two of the three mutations that were not detected were located in an exon with an extreme high GC content (exon 1 of *GRM6*; MIM\# 604,096), included in this study with the explicit purpose of determining whether the approach can deal with extreme GC content. Systematic filtering of variants was accomplished as described in *Materials and Methods* and is summarized in [Fig. 2A](#fig02){ref-type="fig"}. Filtering reduced the total number of putative pathogenic variants from 14,144 to 48 (99.7% reduction), whereas only one of the known causative variants was eliminated during the course of the filtering. Hence, filtering enriched the percentage of disease-causing mutations from 0.15% (21/14,144) to 41.6% (20/48).
{#fig02}
The same systematic prioritization process was subsequently applied to all variants in the molecularly undiagnosed RP cohort of 100 patients. In total, 128,557 variants (including six large deletions) were identified ([Fig. 2B](#fig02){ref-type="fig"}). After applying the same filtering steps as for the 12 known RD samples, 359 variants remained ([Fig. 2B](#fig02){ref-type="fig"}), achieving the same reduction of putative causative variants of 99.7% (from 128,557 to 359).
Confirmation of Variants and Segregation Analysis
-------------------------------------------------
Systematic Sanger sequencing of all 359 variants confirmed 283 variants (79%) in the total RP cohort. The majority of nonconfirmed variants (76%) were falsely called indels, many in the vicinity of homopolymer stretches, a known pitfall of the Roche 454 pyrosequencing technology. On the other hand, a number of larger homozygous and heterozygous deletions (≥43 bp) were detected and validated by Sanger sequencing, underlining an advantageous feature of the Roche long-read technology. Subsequently, all confirmed variants in a given sample were sequenced in available samples from relatives (Supp. Fig. S3).
Interpretation of Genetic Variants
----------------------------------
To further determine whether a variant might be pathogenic, a systematic classification scheme was developed (*Materials and Methods*; Supp. Fig. S2). In brief, classification of pathogenicity was performed using a combination of prediction programs, evolutionary conservation, and frequency data. This method was validated with 100% sensitivity on an independent set of 20 functionally proven pathogenic missense mutations (Supp. Table S4A), whereas it reached a specificity of 94% for a set of 36 putatively benign missense mutations (selected based on frequency data, Supp. Table S4B). This classification method suggested disease-causing variants in 39 of the 100 investigated families ([Table 1](#tbl1){ref-type="table"}, Supp. Table S5). After checking all putative pathogenic variants in the Exome Variant Server of the NHLBI Exome Sequencing Project (ESP) (<http://evs.gs.washington.edu/EVS/>, release version: v.0.0.10), two presumed mutations in dominant genes (c.1730C\>A in *TOPORS* (MIM\# 609507) and c.1724C\>T in *GUCY2D* (MIM\# 600179)), present in three of our families, were re-classified to 'benign' due to frequency data of these variants. Of the remaining 36 families, the pathogenicity was supported in 21 by a full segregation in available relatives. Of all 36 families, inheritance was recessive in 27 cases, X-linked in three cases, and dominant in six cases. In two samples from isolated cases (samples 22,315 and 27,790), disease-causing mutations in the dominant genes *RHO* (MIM\# 180,380) and *PRPF31* (MIM\# 606,419), respectively, occurred as de novo events ([Figs. 3A and 3B](#fig03){ref-type="fig"}). In addition, in one case (family 28,557), a de novo mutation in the recessive gene *USH2A* (MIM\# 608,400) was identified, with the second mutation being inherited from the mother ([Fig. 3C](#fig03){ref-type="fig"}). For all three cases, parental testing proved paternity (Supp. Fig. S4). Additional de novo mutations in dominant genes are likely in two other cases (samples 17,792 and 31,035), although the absence of data from at least one parent hampers a definite conclusion.
{#fig03}
######
Diagnostic Results of 36 RP Patients with Validated Pathogenic Variants
Patient Gene Inheritance in gene Inheritance in family Validated mutation M1 Validated mutation M2
--------- ----------- --------------------- ------------------------- ---------------------------------------------------------------------- -------------------------------------------------------------------------------------
7,554 *ABCA4* ar Isolated female c.1554G\>A (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.4254-2A\>G (p.(?))[a](#tf1-1){ref-type="table-fn"}
8,625 *IMPG2* ar Familial recessive c.379G\>A (p.\[R127^\*^\])[a](#tf1-1){ref-type="table-fn"} c.3423-8_c.3423-5del (p.(?))[a](#tf1-1){ref-type="table-fn"}
9,437 *USH2A* ar Familial recessive c.486-14G\>A (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.12729G\>A (p.(W4243^\*^))[a](#tf1-1){ref-type="table-fn"}
9,470 *RP2* xl Isolated male c.323G\>A (p.\[C108Y\])[b](#tf1-2){ref-type="table-fn"} --
9,472 *PDE6B* ar, ad Familial recessive c.1920+2T\>C (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.1920+2T\>C (p.(?))[a](#tf1-1){ref-type="table-fn"}
*PRPH2* ad Familial recessive c.424C\>T (p.\[R142W\])[c](#tf1-3){ref-type="table-fn"} --
9,493 *PDE6B* ar, ad Isolated male c.1043_1044insCG (p.\[A349fs\])[a](#tf1-1){ref-type="table-fn"} c.1927_1969delinsGG (p.(N643fs))[a](#tf1-1){ref-type="table-fn"}
9,518 *RP1* ar, ad Familial recessive c.515T\>G (p.\[L172R\])[b](#tf1-2){ref-type="table-fn"} c.515T\>G (p.(L172R))[b](#tf1-2){ref-type="table-fn"}
13,480 *RPE65* ar Isolated male c.208T\>G (p.\[F70V\])[d](#tf1-4){ref-type="table-fn"} c.1102T\>C (p.(Y368H))[d](#tf1-4){ref-type="table-fn"}
15,569 *PRPH2* ad Familial female sibship c.441del (p.\[P147fs\])[a](#tf1-1){ref-type="table-fn"} --
17,597 *USH2A* ar Isolated male c.10525A\>T (p.\[K3509^\*^\])[a](#tf1-1){ref-type="table-fn"} c.\[12343C\>T;13274C\>T\] (p.\[(R4115C);(T4425M)\])[d](#tf1-4){ref-type="table-fn"}
17,792 *RHO* ar, ad Isolated female c.403C\>T (p.\[R135W\])[d](#tf1-4){ref-type="table-fn"} --
18,060 *SPATA7* ar Isolated female c.3G\>A (p.\[M1?\])[a](#tf1-1){ref-type="table-fn"} c.322C\>T (p.(R108^\*^))[a](#tf1-1){ref-type="table-fn"}
18,336 *PDE6B* ar, ad Isolated male c.2326G\>A (p.\[D776N\])[b](#tf1-2){ref-type="table-fn"} c.1927_1969delinsGG (p.(N643fs))[a](#tf1-1){ref-type="table-fn"}
19,693 *PDE6B* ar, ad Isolated male c.299G\>A (p.\[R100H\])[b](#tf1-2){ref-type="table-fn"} c.1927_1969delinsGG (p.(N643fs))[a](#tf1-1){ref-type="table-fn"}
19,733 *USH2A* ar Isolated female c.(4957C\>T; 7379G\>A) (p.\[(R1653^\*^);(R2460H)\])^a,b^ c.10073G\>A (p.(C3358Y))[b](#tf1-2){ref-type="table-fn"}
19,735 *RPE65* ar Isolated female c.271C\>T (p.\[R91W\])[d](#tf1-4){ref-type="table-fn"} c.715T\>G (p.(Y239D))[d](#tf1-4){ref-type="table-fn"}
20,984 *PDE6B* ar, ad Isolated female c.1107+3A\>G (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.1107+3A\>G (p.(?))[a](#tf1-1){ref-type="table-fn"}
21,141 *CEP290* ar Isolated male c.3559del (p.\[L1187fs\])[a](#tf1-1){ref-type="table-fn"} c.4705-1G\>T (p.(?))[a](#tf1-1){ref-type="table-fn"}
21,334 *CACNA1F* xl Isolated male c.220T\>C (p.\[C74R\])[d](#tf1-4){ref-type="table-fn"} --
21,933 *RP1* ar, ad Isolated male c.368_369dup (p.\[P214fs\])[a](#tf1-1){ref-type="table-fn"} c.4241_4242del (p.(H1414fs))[a](#tf1-1){ref-type="table-fn"}
22,315 *RHO* ar, ad Isolated male c.538C\>T (p.\[P180S\])[b](#tf1-2){ref-type="table-fn"} --
22,393 *PDE6B* ar, ad Isolated male c.892C\>T (p.\[Q298^\*^\])[a](#tf1-1){ref-type="table-fn"} c.892C\>T (p.(Q298^\*^))[a](#tf1-1){ref-type="table-fn"}
22,777 *USH2A* ar Isolated female c.1256G\>T (p.\[C419F\])[d](#tf1-4){ref-type="table-fn"} c.\[12343C\>T;13274C\>T\] (p.\[(R4115C);(T4425M)\])[d](#tf1-4){ref-type="table-fn"}
27,790 *PRPF31* ad Isolated male c.553G\>T (p.\[E185^\*^\])[a](#tf1-1){ref-type="table-fn"} --
28,557 *USH2A* ar Isolated female c.2276G\>T (p.\[C759F\])[d](#tf1-4){ref-type="table-fn"} c.5576T\>G (p.(F1859C))[b](#tf1-2){ref-type="table-fn"}
31,035 *NR2E3* ar, ad Isolated male c.95G\>A (p.\[W32^\*^\])[a](#tf1-1){ref-type="table-fn"} --
31,124 *USH2A* ar Male sibship c.486-14G\>A (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.2276G\>T (p.(C759F))[d](#tf1-4){ref-type="table-fn"}
31,343 *USH2A* ar Familial recessive c.917_918insGCTG (p.\[S307fs\])[a](#tf1-1){ref-type="table-fn"} c.11007C\>A (p.(Ser3669Arg))[b](#tf1-2){ref-type="table-fn"}
31,723 *ARL6* ar Isolated female c.185+1G\>A (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.185+1G\>A (p.(?))[a](#tf1-1){ref-type="table-fn"}
31,933 *CRB1* ar, ad Isolated male c.1602G\>T (p.\[K534N\])[b](#tf1-2){ref-type="table-fn"} c.2234C\>T (p.(T745M))[d](#tf1-4){ref-type="table-fn"}
31,994 *RHO* ar, ad Familial dominant c.641T\>A (p.\[I214N\])[b](#tf1-2){ref-type="table-fn"} --
32,594 *NRL* ar Male sibship c.508C\>A (p.\[R170S\])[d](#tf1-4){ref-type="table-fn"} c.654del (p.(C219fs))[a](#tf1-1){ref-type="table-fn"}
32,655 *RP2* xl Male sibship c.318_319delAG (p.\[D107fs\])[a](#tf1-1){ref-type="table-fn"} --
33,626 *USH2A* ar Isolated female c.1227G\>A (p.\[W409^\*^\])[a](#tf1-1){ref-type="table-fn"} c.12575G\>A (p.(R4192H))[b](#tf1-2){ref-type="table-fn"}
33,672 *RDH12* ar Isolated male c.658+591\_^\*^603delinsCT (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.658+591\_^\*^603delinsCT (p.(?))[a](#tf1-1){ref-type="table-fn"}
37,370 *RLBP1* ar Isolated female c.525_945del (p.\[?\])[a](#tf1-1){ref-type="table-fn"} c.525_945del (p.(?))[a](#tf1-1){ref-type="table-fn"}
Patient, patient identifier; Gene, RefSeq gene name of the gene in which mutations were identified; Inheritance, possible inheritance of phenotype for known mutations in the corresponding gene; Validated mutation M1 and Validated mutation M2, cDNA and protein annotation of the identified mutations.
Nonsense/splicing/frameshift mutation.
Variant predicted to be pathogenic.
Known mutation identified in proband, not present in other affected siblings, but contributing to a more severe phenotype in the proband.
Known mutation.
ar, autosomal recessive; ad, autosomal dominant; xl, X-linked.
Cumulative Effect of Multiple Pathogenic Alleles on the Phenotype
-----------------------------------------------------------------
In sample 9,472, a homozygous canonical splice-site mutation in *PDE6B* (MIM\# 180,072, c.1920+2T\>C \[p.(?)\]) and a known (dominant) missense mutation in *PRPH2* (MIM\# 179,605, c.424C\>T \[p.(R142W)\]) \[Boon et al., [@b7]\] were identified. Although the homozygous splice-site mutation in *PDE6B* was segregating with the disease in four affected siblings, the dominant missense mutation was not present in any of the affected siblings. This may help to explain the earlier onset of macular abnormalities, which are known to be associated with *PRPH2* mutations, in patient 9,472; at age 20, a bull\'s eye maculopathy with significant abnormalities of the retinal pigment epithelium was observed, whereas his sibling 12,273 at age 22 did not show any significant macular abnormality.
Diagnostic Yield for an Unscreened RP Cohort
--------------------------------------------
The mutation detection rate for the current NGS approach was 87.5% for the 12 RD patients with 24 known mutations (21/24 = 87.5%). In 10 out of these 12 patients, this would have resulted in a molecular diagnosis, resulting in a solved rate (diagnostic yield) of 83% (10/12 = 83%).
We estimate that for an unscreened RP population, the diagnostic yield would be ∼50%, based on the following calculation. The original cohort of 234 patients had undergone previous selected genotyping, resulting in the identification of genetic defects in 57 patients (Supp. Fig. S1). As a result of a solved rate of 83% for the NGS-based approach in 12 control patients, 47 out of these 57 patients (83%) would have been solved when analyzed by NGS, assuming autosomal recessive inheritance. Of the remaining 177 patients, 100 were investigated by our approach, of which 36 were diagnosed (36%). The residual 77 samples (177--100) have so far not been analyzed by NGS. Assuming the same diagnostic yield of 36% with NGS in those, we would diagnose additional 28 patients out of these 77 (36%) with NGS. Together, our data indicate that this approach could be used to establish a molecular diagnosis in nearly 50% of a random RP cohort (\[47 + 36 + 28\]/234 = 47%).
Overall, the results of this study underscore the extreme genetic heterogeneity in RP by identifying putatively causative mutations in no fewer than 20 different genes (Supp. Fig. S5). Mutations in *USH2A* and *PDE6B* appear to be slightly overrepresented in the mutational spectrum, as has been described previously \[Hartong et al., [@b21]\]. In five patients that were initially diagnosed with RP (individuals 7,554, 9,437, 21,141, 21,334, and 31,723), identification of the genetic defect led to a reappraisal of the phenotype to either a different subtype of nonsyndromic RD or to a (mild) syndromic form of RP (Supp. Table S6). For the 64 cases without a clear molecular diagnosis, either no interesting variants were detected among the 111 targeted genes (eight cases), the identified variants were predicted to be benign or did not segregate completely (48 cases), or segregating variants were identified, but their pathogenicity remained uncertain (eight samples).
Discussion
==========
In this study, we developed a comprehensive diagnostic tool for RP, consisting of a massive parallel sequencing approach for all known retinal disease genes, with systematic analysis and interpretation of all detected genetic variants. Our interpretation workflow provides a general approach for the interpretation of the vast amount of data generated by NGS-based molecular diagnostics. The enormous potential and clinical utility of NGS is highlighted by a high diagnostic yield achieved by the identification of mutations of all modes of inheritance, including de novo mutations. The high degree of (inherited and de novo) autosomal dominant and X-linked mutations confers a significant risk for transmitting the disease to the patients\' offspring, thereby illustrating not only the molecular diagnostic power of NGS, but also the huge impact of this method on the affected families.
The strength of this study lies in the parallel analysis of all known retinal disease genes in 100 RP patients with an unknown molecular cause of disease by NGS. Systematic variant prioritization reduced the initially high number of identified variants (∼1,200 per patient) by 99.7% to a number that is manageable to validate by Sanger sequencing (approximately four variants per patient). Combined with a systematic assessment of pathogenicity of the validated variants, this NGS approach resulted in a molecular diagnosis in no fewer than 36 patients. The usage of a completely unscreened, prospective cohort would probably have led to a more accurate estimation of the success rate of this approach. Although such prospective studies can be performed in a research setting, it is, however, impossible within DNA diagnostics, where every effort needs to be made to solve a sample in a timely fashion.
Importantly, the 100 patients investigated here originally belonged to a cohort of 234 patients that had previously undergone selected genotyping \[Collin et al., [@b13]\], resulting in the identification of genetic defects in 57 of the patients before the start of this study (Supp. Fig. S1). Taking into consideration a solved rate of 83%, we calculated a potential diagnostic yield of nearly 50% in RP for this NGS-based diagnostic approach (see *Results* for details), thereby outperforming previously used approaches such as traditional Sanger sequencing and APEX analysis \[Hartong et al., [@b21]; Ávila-Fernández et al., [@b2]\]. As the number of 12 samples used for finding known mutations is quite low and technology is still improving, we cannot rule out a certain variability of this given percentage.
Apart from the ability to analyze all known RP genes simultaneously, thus increasing effectiveness of the genetic analysis, this approach has added value and superior clinical relevance because it allowed the identification of a substantial number of cases with apparent de novo mutations, and it has the potential to identify additional mutations potentially contributing to the phenotype---that is, to determine the cumulative mutational load. The 100 investigated samples comprised 78 isolated cases and 22 cases from recessive multiplex families. Among the 22 multiplex families, a molecular diagnosis was achieved in eight families: seven families carried mutations in autosomal recessive RP genes and one carried a mutation in the X-linked gene *RP2* (MIM\# 300,757). Of the 78 isolated RP cases, a molecular diagnosis was established in 28 cases, including 20 families that harbored mutations in autosomal recessive RP genes and two male cases with mutations in X-linked genes. In two presumed isolated cases, mutations were found in autosomal dominant RP genes, but re-evaluation of both families revealed that one of the parents also had RP. Intriguingly, de novo mutations were proven to be present in three further isolated cases. Two of these were located in dominant genes, whereas in one patient, one of the two autosomal recessive mutations in *USH2A* occurred as a de novo event. Additional de novo mutations are likely in two other cases because these detrimental variants occur in dominant genes, whereas no phenotypes are reported for the parents. However, the absence of DNA of at least one parent hampers a definite conclusion. Although de novo mutations have recently been shown to play a major role in human diseases with reduced reproductive fitness \[Hamdan et al., [@b20]; Hoischen et al., [@b25]; Vadlamudi et al., [@b51]; Vissers et al., [@b53]\], the identification of a significant fraction of de novo dominant mutations in RP is a surprising finding. We speculate that de novo mutations may have been underappreciated as a cause of autosomal dominant RP as the result of a bias in ascertainment towards sizeable families with a clear autosomal dominant inheritance pattern \[Daiger et al., [@b15]\]. Given a transmission risk of 50% and a recurrence risk of \<1% for autosomal dominant de novo mutations, compared with a transmission risk of \<1% and a recurrence risk of 25% for autosomal recessive inheritance, the relatively high number of de novo mutations in our cohort will have significant implications for genetic counseling of our patients and their relatives. An additional advantage of an NGS approach that targets all known RD genes is the unprecedented possibility to study the cumulative effect of multiple pathogenic alleles in different genes on the phenotype \[Ng et al., [@b42]\]. This could be observed in family 9,472, where the proband carried three disease-causing mutations in two different genes (one recessive and one dominant gene), whereas his four affected siblings carried only the mutations in the recessive gene. This led us to the conclusion that the additional mutation could explain the earlier onset of macular abnormalities in the proband when compared with his siblings. The characterization of the genetic defect, however, is not only important for understanding the patients\' phenotypes, but also confers benefits concerning eligibility for gene therapy. Gene replacement therapy has been proven to be beneficial in clinical trials for patients with *RPE65* (MIM\# 180,069) mutations \[Bainbridge et al., [@b3]; Hauswirth et al., [@b23]; Maguire et al., [@b33]\]. Furthermore, preclinical gene therapy studies in animal models are ongoing for a substantial set of RD genes \[den Hollander et al., [@b16]\] and are likely to enter clinical trials in near future.
Further improvements in sequencing technology together with the identification of novel RP genes will undoubtedly boost the success rate of NGS-based diagnostic approaches in RP in near future. Recently, others have also reported on the use of NGS for the identification of disease-causing mutations in genetically heterogeneous disorders, including that of dominant \[Bowne et al., [@b8]\] and recessive \[Simpson et al., [@b48]\] RP, although the number of genes that were analyzed and/or the cohort sizes were considerably smaller as compared with our study \[Shearer et al., [@b47]; Jones et al., [@b27]; Otto et al., [@b44]\]. An alternative for parallel, targeted resequencing of known disease genes is whole-exome or whole-genome sequencing \[Choi et al., [@b10]; Lupski et al., [@b32]; Worthey et al., [@b55]\], permitting a standardized laboratory workflow that can be used for any type of genetic disorder. We think that the diagnostic interpretation workflow developed in this study can also be applied to these approaches, especially given the fact that most diagnostic laboratories will focus on the interpretation and reporting of variants in the coding part of our genome. Importantly, three patients in this study were found to carry pathogenic deletions that were easily detected by the long reads of the Roche 454 sequencing technology used in this study. This stresses the fact that these indels represent a diagnostically important group of genome variants that is not robustly called by the most commonly used whole-exome or whole-genome sequencing approaches.
Together, these studies support our conclusion that NGS will become increasingly important for the identification of disease-causing genes, and underscore the impact of this technology for the future of DNA diagnostics. For this, improvements in sequencing performance and interpretation should be combined with further cost reductions and improvements in turnaround times.
We are very grateful to all participating patients and their relatives. We also thank Peer Arts, Tom Hofste, and Saskia D. van der Velde-Visser for excellent technical assistance. We thank Esther Kok for large bioinformatical efforts; Ton Feuth for biostatistical analyses; Marcel Nelen for management of the diagnostic facilities; Suzanne Ijzer, Mary van Schooneveld, Bert de Vries, and John R. Heckenlively for clinical support; and Han G. Brunner for strategic guidance and proofreading of the manuscript.
Additional Supporting Information may be found in the online version of this article.
[^1]: Communicated by Paolo M. Fortina
[^2]: These authors contributed equally to this work
[^3]: Contract grant sponsors: European Community\'s Seventh Framework Program FP7/2007-2013 (\#223143; TECHGENE); the Netherlands Organization for Health Research and Development (ZonMW grants 917-66-363 and 911-08-025); TOP-grant (40-00812-98-09047); Algemene Nederlandse Vereniging ter Voorkoming van Blindheid; Gelderse Blinden Stichting; Landelijke Stichting voor Blinden en Slechtzienden; Retina Nederland; Stichting Oogfonds Nederland; Rotterdamse Stichting Blindenbelangen; Foundation Fighting Blindness.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-geriatrics-01-00010}
===============
Delirium, an acute neuropsychiatric syndrome, is a common, severe complication in the elderly and is associated with poor clinical outcomes including increased morbidity and mortality, prolonged hospital stay, loss of independence, and increased rates of cognitive decline \[[@B1-geriatrics-01-00010],[@B2-geriatrics-01-00010]\]. The pathophysiological mechanisms underlying delirium are still poorly understood, but it is widely accepted that delirium occurs due to a complex interplay among several biochemical pathways. Therefore, it might be required to interrupt in multiple biochemical pathways at the same time to prevent, treat, or to lower the severity of a delirium.
Activation of the immune system, oxidative stress, and disturbances in the serotonergic neurotransmission may all contribute to the development of a delirium. Recently, we found that acutely ill hospitalized elderly patients with a delirium have increased levels of neopterin \[[@B3-geriatrics-01-00010]\]. Neopterin is produced primarily by activated monocytes and macrophages in response to the pro-inflammatory cytokine interferon-gamma (IFN-γ) and its levels reflect the amount of cell-mediated immune activation and oxidative stress \[[@B4-geriatrics-01-00010],[@B5-geriatrics-01-00010]\]. Furthermore, we have found that patients with a delirium have a decreased availability of tryptophan to the central nervous system. This decreased availability might result in a decreased serotonin production in the brain, since tryptophan is the precursor of serotonin \[[@B6-geriatrics-01-00010]\].
A study of Schroecksnadel *et al*. showed that treatment of stimulated peripheral blood mononuclear cells with aspirin significantly decreased neopterin production and tryptophan degradation *in vitro* \[[@B7-geriatrics-01-00010]\]. Therefore, the aim of the present study was to evaluate the possible association between aspirin use and mean levels of neopterin and tryptophan in patients with and without a delirium and additionally, whether the use of aspirin is associated with a decreased prevalence of delirium.
2. Methods {#sec2-geriatrics-01-00010}
==========
2.1. Participants {#sec2dot1-geriatrics-01-00010}
-----------------
The present study was performed within the Delirium In The Old (DITO) study in which mean plasma/serum levels of several biochemical parameters, including neopterin and tryptophan, were compared between patients with and without a delirium \[[@B3-geriatrics-01-00010],[@B6-geriatrics-01-00010]\]. In the DITO study, a cross-sectional study, we included patients who were admitted to the wards of Internal Medicine and Geriatrics of the Erasmus University Medical Center and the ward of Geriatrics of the Havenziekenhuis, Rotterdam, The Netherlands. All acutely admitted patients aged ≥65 years were eligible to participate. Exclusion criteria were a diagnosis of Lewy Body dementia, Parkinson's disease, neuroleptic malignant syndrome, tardive dyskinesia, ongoing treatment with antipsychotics or other psychiatric medications, except haloperidol and benzodiazepines, aphasia, insufficient understanding of the Dutch language, and a Mini Mental State Examination (MMSE) score \< 10 points out of 30. Patients with a MMSE \< 10 were not included because it can be quite difficult to distinguish between features of severe dementia and delirium at admission, as well as to measure improvement of cognitive function in this group. Additional exclusion criteria for the present study were unclear data regarding the use of non-steroidal anti-inflammatory drugs (NSAIDs) in the days preceding hospital admission as well as the use of aspirin concomitantly with other NSAIDs (as it might be possible that other NSAIDs interfere with aspirin's potential effect on neopterin and tryptophan levels).
Written informed consent was obtained from all participants. In case of a delirium or cognitive impairment at the time of admission, informed consent was obtained from a representative of the patient. The Medical Ethics Committee of the Erasmus University Medical Center approved the study protocol.
2.2. Procedures {#sec2dot2-geriatrics-01-00010}
---------------
All participants were observed daily by the nursing and medical staff and by members of the research team until discharge. To screen for a change in behavior, the 13-item Delirium Observation Screening scale was used during the first five days of admission \[[@B8-geriatrics-01-00010]\]. The diagnosis of delirium was made by a geriatrician, according to the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV) \[[@B9-geriatrics-01-00010]\], and was based on the psychiatric examination of the patient, the medical and nursing records, including the Delirium Observation Screening scale scores, and information given by the patient's closest relative. When the diagnosis of delirium was doubtful, the case was discussed with the geriatric consultation team to gain consensus.
Demographic and clinical data were collected at admission. Age and gender were documented. Cognitive functioning was assessed in absence of a delirium using the MMSE \[[@B10-geriatrics-01-00010]\]. When it was impossible to score the MMSE during admission because the patient was too ill, the cognitive functioning was discussed with a clinician or assessed with information from the available medical records. When the clinical opinion was that the patient would have a MMSE score ≥ 10, the patient was not excluded from the study. Severity of comorbidities was scored using the Charlson Comorbidity Index. This index encompasses 19 medical conditions and each condition is weighted with a score of 1 to 6 by severity \[[@B11-geriatrics-01-00010]\]. The physical functionality was assessed using the six-item Katz Activities of Daily Living (ADL) scale and the Barthel Index \[[@B12-geriatrics-01-00010],[@B13-geriatrics-01-00010]\]. The instrumental functionality was assessed using the 7-items Older Americans Resource Scale for Instrumental ADL (OARS-IADL) \[[@B12-geriatrics-01-00010]\]. Frailty was measured with the Identification of Seniors at Risk (ISAR) questionnaire \[[@B14-geriatrics-01-00010]\]. For all participants the medication at hospital admission was reviewed for the use of NSAIDs (including low dose acetylsalicylic acid and the equivalent drug carbasalate calcium), beta-blockers, diuretics, angiotensin converting enzyme (ACE) inhibitors, angiotensin II receptor antagonists, calcium channel blockers, nitrates, statins, and dipyridamole.
Blood samples of all patients were collected within 48 h after admission. When a patient developed a delirium during the hospital stay, new blood samples were collected within 24 h after the onset of the delirium and were used, instead of the first blood samples, for the statistical analyses.
2.3. Biochemical Measurements {#sec2dot3-geriatrics-01-00010}
-----------------------------
Non-fasting blood was collected preferably between 8 and 10 a.m. in an 8-mL tube containing ethylene diamine tetra-acetic acid. After blood sampling, the tubes were protected from light to prevent oxidative loss of neopterin \[[@B15-geriatrics-01-00010]\], and stored at room temperature to prevent changes in the transfer of amino acids between plasma and blood cells \[[@B16-geriatrics-01-00010]\]. Within 3 h, the blood was centrifuged for 20 min at 2650 g and 20 °C. The obtained plasma was stored at −80 °C until analysis.
Plasma neopterin levels were determined by high-performance liquid chromatography after acid oxidation, as previously described \[[@B17-geriatrics-01-00010]\]. Tryptophan levels were determined by high-performance liquid chromatography with automated pre-column derivatization with ortho-phthalaldehyde \[[@B16-geriatrics-01-00010]\].
2.4. Statistical Analyses {#sec2dot4-geriatrics-01-00010}
-------------------------
Depending on the distribution of the data, differences in demographic and clinical baseline characteristics between patients with and without a delirium were evaluated using the chi-square test or the Fisher's exact test for categorical variables and the Mann--Whitney *U*-test or the Student's *t*-test for continuous variables.
Levels of neopterin and tryptophan were not normally distributed and were, therefore, logarithmically transformed. Univariate one-way analysis of variance was used to investigate the association between mean levels of neopterin and tryptophan (dependent variable) and the use of aspirin in both patients with and without a delirium. For this purpose, analyses were stratified for aspirin use. Age, gender, Charlson Comorbidity Index and statin use were used as covariates. The Charlson Comorbidity Index was added since neopterin levels are found to be increased and tryptophan levels decreased in several medical conditions. Statin use was added since statins might also inhibit neopterin production and tryptophan degradation \[[@B18-geriatrics-01-00010]\]. The model including neopterin was additionally adjusted for eGFR, since neopterin is excreted mainly by the kidneys \[[@B5-geriatrics-01-00010]\]. All mean levels and 95% confidence intervals (CI) of neopterin and tryptophan presented in this manuscript are the back-transformed log-values. A two-tailed *p* \< 0.05 was defined as statistically significant. All statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS), version 21.0 (IBM Corp., Armonk, NY, USA). GraphPad Prism 5.01 for Windows (GraphPad Software, San Diego, CA, USA) was used to draw the graphs.
3. Results {#sec3-geriatrics-01-00010}
==========
3.1. Participant Characteristics {#sec3dot1-geriatrics-01-00010}
--------------------------------
Of the 86 patients enrolled in the DITO study, 80 were included in the stratified analyses to examine the effect of aspirin use on neopterin and tryptophan levels. Three patients were excluded due to unclear data regarding the use of NSAIDs in the days preceding hospital admission, 1 patient used diclofenac and 2 patients used carbasalate calcium concomitantly with another NSAID (diclofenac and etoricoxib respectively). [Table 1](#geriatrics-01-00010-t001){ref-type="table"} represents the baseline characteristics of the included patients. Twenty-two patients were diagnosed with a delirium, of which 21 were admitted to the hospital with a delirium and 1 developed a delirium during admission. No difference was found in the number of aspirin users between patients with and without a delirium (46.6% *versus* 40.9%, *p* = 0.651, respectively).
3.2. Analysis of Biochemical Parameters {#sec3dot2-geriatrics-01-00010}
---------------------------------------
Mean levels and corresponding 95% CI of neopterin and tryptophan in patients with and without a delirium, stratified for the use of aspirin, are presented in [Table 2](#geriatrics-01-00010-t002){ref-type="table"} and [Table 3](#geriatrics-01-00010-t003){ref-type="table"}.
In the group without a delirium, no significant difference was found in the adjusted mean levels of neopterin between patients who used aspirin (43.6 nmol/L, 95% CI: 34.5--55.0) and patients who did not use aspirin (47.0 nmol/L, 95% CI: 37.8--58.3) (*p* = 0.645). Also no difference was found in the adjusted mean levels of tryptophan between patients who used aspirin (33.9 µmol/L, 95% CI: 29.6--38.8) and patients who did not use aspirin (33.1 µmol/L, 95% CI: 29.2--37.6) (*p* = 0.816).
In the group with a delirium, unadjusted levels of neopterin seemed to be lower in patients who used aspirin than in those who did not, as shown in [Figure 1](#geriatrics-01-00010-f001){ref-type="fig"}. However, in this small group the adjusted mean levels of neopterin were not statistically significantly lower in patients who used aspirin (71.1 nmol/L, 95% CI: 43.8--115.6) than in patients who did not use aspirin (77.8 nmol/L, 95% CI: 52.4--115.6) (*p* = 0.779). In addition, unadjusted levels of tryptophan seemed to be higher in patients who used aspirin than in those who did not ([Figure 1](#geriatrics-01-00010-f001){ref-type="fig"}). However, the adjusted mean levels of tryptophan were not statistically significantly higher in patients who used aspirin (27.3 µmol/L, 95% CI: 18.4--40.5) than in those who did not (22.4 µmol/L, 95% CI: 16.2--30.9) (*p* = 0.439).
4. Discussion {#sec4-geriatrics-01-00010}
=============
In the present study we found that the use of aspirin (exclusively low dose aspirin) was not associated with a decreased prevalence of delirium. Furthermore, we found that mean neopterin and tryptophan levels were not statistically significant affected by the use of aspirin in patients with and without a delirium.
As far as we are aware, this is the first study investigating the possible association between aspirin use and mean neopterin and tryptophan levels in patients with and without a delirium. We found that neopterin and tryptophan levels were not statistically significant affected by the use of aspirin. Therefore, we could not confirm the relationship found *in vitro* by Schroecksnadel *et al.* that aspirin decreases neopterin production as well as tryptophan degradation. These controversial findings may be caused by several factors. First, Schroecksnadel *et al.* found that the inhibition of neopterin production and tryptophan degradation by aspirin was dose-dependent \[[@B7-geriatrics-01-00010]\]. In our study, all aspirin users used acetylsalicylic acid or carbasalate calcium in a low dose within cardiovascular risk management (80 and 100 mg per day, respectively). It might be possible that these dosages are too low for having a significant effect on neopterin and tryptophan levels. Another possibility is that we did not find an association due to the small sample size. However, in the group with a delirium, neopterin levels seemed to be lower and tryptophan levels seemed to be higher in patients who used aspirin than in those who did not. Therefore, it might be possible that in a larger group these differences would become statistically significant. On the other hand, this trend was not seen in the group without a delirium. Schroecksnadel *et al.* found that aspirin did not influence tryptophan degradation and only minimally affected neopterin production in resting cells \[[@B7-geriatrics-01-00010]\]. It might be possible that in patients without a delirium neopterin production and tryptophan degradation was not stimulated enough and that we, therefore, did not see a trend in this group.
The potential influence of aspirin on neopterin production and tryptophan degradation in patients with a delirium might be the result of a modulating effect of aspirin on the cytokine IFN-γ. Both the production of neopterin as well as the degradation of tryptophan is IFN-γ dependent. During immune activation, IFN-γ induces in macrophages the enzyme guanosine triphosphate cyclohydrolase-I, which is among others responsible for the production of neopterin \[[@B4-geriatrics-01-00010],[@B5-geriatrics-01-00010]\]. IFN-γ also induces the enzyme indoleamine-2,3-dioxygenase which converts tryptophan to kynurenine \[[@B19-geriatrics-01-00010]\]. In a previous study performed in a chimeric mouse model of giant cell arteritis, aspirin has been demonstrated to be highly effective in suppressing IFN-γ production at doses of 20--100 mg/kg \[[@B20-geriatrics-01-00010]\]. However, the doses used in that study were much higher than the dose used by our participants and therefore it might be expected that the findings are only generalizable to a lesser extent to the dose used in our study. Interestingly, they also found that another NSAID, indomethacin, was not able to reduce IFN-γ transcription \[[@B20-geriatrics-01-00010]\]. This might suggest that NSAIDs which are structurally unrelated to aspirin are not able to affect neopterin and tryptophan levels. In line with this hypothesis, Forrest *et al.* found in patients with osteoporosis after two years of drug treatment that additional pain treatment with a NSAID did not decrease neopterin levels and did not increase tryptophan levels in comparison with patients who did not use NSAIDs \[[@B21-geriatrics-01-00010]\]. The authors note that patients taking NSAIDs might be among the more severely affected patients in whom disease control could be difficult and this could have influenced their results \[[@B21-geriatrics-01-00010]\].
Furthermore, we found that the use of aspirin was neither associated with a decreased nor with an increased prevalence of delirium, despite it has been speculated that NSAIDs increase the risk of a delirium. In a systematic review it was found that research on the association of NSAIDs with delirium is limited and that the association remains uncertain \[[@B22-geriatrics-01-00010]\]. It is important to note that in the present study the association between the use of low dose aspirin and delirium was evaluated. Therefore, it might be still possible that other NSAIDs are associated with an increased risk of a delirium.
Limitations and Strengths
-------------------------
This study has some limitations. First, the cross-sectional design limits the ability to identify a causal relationship between aspirin use, neopterin and tryptophan levels, and the prevalence of delirium. Therefore, the results of this study should be considered as hypothesis generating. Second, the relatively small sample size decreased the ability to detect a possible association between aspirin use and mean neopterin and tryptophan levels in patients with a delirium. Third, delirium severity was not scored in our study. It might be possible that the use of aspirin does not prevent delirium, but that it will decrease delirium severity (as aspirin really inhibits neopterin production and tryptophan degradation as it seems). Fourth, tryptophan levels could be influenced by dietary intake. In this study, we were not able to adjust for dietary status and this might have influenced our results. However, since possible food intake was random and blood was collected between 8 and 10 a.m., we think that our results are only minimally influenced by this. Finally, we were not able to evaluate whether the use of NSAIDs other than aspirin will have a potential influence on neopterin and tryptophan levels. Since other NSAIDs, used to treat inflammation and pain, are only limited prescribed to elderly patients due to their negative effects on renal function, it would probably only be interesting to investigate this for diseases in a younger population in which neopterin and tryptophan are also involved and not for delirium in elderly patients.
The present study has several strengths. First, the intensive monitoring of clinical symptoms of patients with a delirium until discharge and the DSM-IV diagnosis by a geriatrician makes it less likely that we missed a delirium or misdiagnosed symptoms. Second, we have performed statistical analyses in a relatively homogeneous group of patients, since all of them used low dose aspirin.
5. Conclusions {#sec5-geriatrics-01-00010}
==============
In this study in older, acutely ill hospitalized patients, we did not find a statistically significant effect of aspirin use on neopterin and tryptophan levels in patients with and without a delirium. However, in patients with a delirium, neopterin levels seemed to be lower and tryptophan levels seemed to be higher in patients who used aspirin compared with those who did not. Larger studies might be needed to investigate this potential influence of aspirin use on neopterin production and tryptophan degradation in patients with a delirium.
We thank all patients who participated in the study as well as Eline Wijnbeld and Milly van der Ploeg for their contribution to the data collection. We are also very grateful to Ans Voskuilen-Kooijman for her assistance in processing the blood samples and high-performance liquid chromatography analyses. This study was supported by a research grant of Fund NutsOhra (project number 0902-047). We did not receive funds for covering the costs to publish in open access.
A.E., D.F., G.Z., T.J.M.C. and F.U.S.M.R. were responsible for the study concept and design. A.E., G.Z. and T.J.M.C. recruited patients. A.E., D.F., G.Z. and T.J.M.C. collected data. A.E., D.F., G.Z., T.J.M.C. and F.U.S.M.R. contributed to the data analysis and interpretation of the results. A.E. wrote the manuscript. D.F., G.Z., T.J.M.C. and F.U.S.M.R. critically revised the manuscript.
The authors declare no conflict of interest.
{#geriatrics-01-00010-f001}
geriatrics-01-00010-t001_Table 1
######
Demographic and clinical baseline characteristics of the study participants.
Variable No Delirium (*n* = 58) Delirium (*n* = 22) *p*-Value
------------------------------------- ------------------------ --------------------- -----------
Gender male 28 (48.3) 9 (40.9) 0.555 \*
Age in years 80.4 ± 7.5 85.8 ± 4.1 0.002 ^‡^
MMSE ^\|\|^ 25.0 (22.0--28.0) 20.0 (17.3--24.3) 0.000 ^†^
Katz ADL score ^¶^ 0.0 (0.0--3.0) 3.5 (1.0--11.3) 0.013 ^†^
OARS-IADL score ^\#^ 5.0 (0.0--10.0) 10.0 (3.0--14.0) 0.037 ^†^
Barthel Index \*\* 18.0 (13.0--20.0) 16.0 (9.0--19.0) 0.050 ^†^
ISAR score ^††^ 4.0 (2.5--6.0) 6.0 (5.0--7.0) 0.000 ^†^
Charlson Comorbidity Index ^‡‡^ 2.00 (1.00--3.00) 2.00 (1.00--3.25) 0.202 ^†^
eGFR (mL/min) 64.3 ± 25.3 48.0 ± 24.7 0.011 ^‡^
Aspirin at admission 27 (46.6) 9 (40.9) 0.651 \*
Type of aspirin:
Acetylsalicylic acid 12 (44.4) 5 (55.6)
Carbasalate calcium 15 (55.6) 4 (44.4)
Beta-blockers 17 (29.3) 6 (27.3) 0.857 \*
Diuretics 22 (37.9) 7 (31.8) 0.612 \*
ACE inhibitors 14 (24.1) 6 (27.3) 0.772 \*
Angiotensin II receptor antagonists 8 (13.8) 2 (9.1) 0.719 ^§^
Calcium channel blockers 13 (22.4) 3 (13.6) 0.536 ^§^
Nitrates 5 (8.6) 1 (4.5) 1.000 ^§^
Statins 27 (46.6) 3 (13.6) 0.007 \*
Dipyridamole 5 (8.6) 0 (0.0) 0.315 ^§^
Values are expressed as mean ± SD for normally distributed continuous variables, median (interquartile range) for not normally distributed continuous variables and *n* (percentages) for categorical variables. \* Chi-square test. ^†^ Mann--Whitney *U*-test; ^‡^ Student's *t*-test; ^§^ Fisher's exact test; ^\|\|^ range 0 (severe cognitive impairment) to 30 (no cognitive impairment); ^¶^ range 0 (no disability) to 12 (severe disability); ^\#^ range 0 (no disability) to 14 (severe disability); \*\* range 0 (severe disability) to 20 (no disability); ^††^ scores ≥ 2 indicate a high risk for functional decline; ^‡‡^ range 0 to 37 (severe burden of comorbidities).
geriatrics-01-00010-t002_Table 2
######
Neopterin levels (nmol/L) in patients with and without a delirium, stratified for the use of aspirin.
No Delirium No Aspirin (*n* = 31) Aspirin (*n* = 27) *p*-Value
-------------- --------------------------- ----------------------- ---------------
Model 1 45.8 (36.2--57.9) 44.9 (34.9--57.8) 0.908
Model 2 47.0 (37.8--58.3) 43.6 (34.5--55.0) 0.645
**Delirium** **No Aspirin (*n* = 13)** **Aspirin (*n* = 9)** ***p*-Value**
Model 1 88.3 (57.7--135.2) 59.3 (35.6--98.9) 0.228
Model 2 77.8 (52.4--115.6) 71.1 (43.8--115.6) 0.779
Values are expressed as mean (95% CI) and are the back-transformed log~10~ values. Model 1: not adjusted. Model 2: adjusted for age, gender, Charlson Comorbidity Index, eGFR, statin use.
geriatrics-01-00010-t003_Table 3
######
Tryptophan levels (µmol/L) in patients with and without a delirium, stratified for the use of aspirin.
No Delirium No Aspirin (*n* = 31) Aspirin (*n* = 27) *p*-Value
-------------- --------------------------- ----------------------- ---------------
Model 1 33.8 (29.6--38.6) 33.1 (28.7--38.2) 0.835
Model 2 33.1 (29.2--37.6) 33.9 (29.6--38.8) 0.816
**Delirium** **No Aspirin (*n* = 13)** **Aspirin (*n* = 9)** ***p*-Value**
Model 1 21.6 (15.4--30.3) 28.8 (19.2--43.2) 0.269
Model 2 22.4 (16.2--30.9) 27.3 (18.4--40.5) 0.439
Values are expressed as mean (95% CI) and are the back-transformed log~10~ values. Model 1: not adjusted. Model 2: adjusted for age, gender, Charlson Comorbidity Index and statin use.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Human studies concerning the effects of peripheral retinal loss on adult visual cortical structure and function are scarce. Previous studies have mainly addressed central retinal disorders such as macular degeneration or other hereditary retinal dystrophies and diseases such as glaucoma \[[@B1]--[@B3]\]. Peripheral and central visual information is differentially routed in the brain \[[@B4], [@B5]\]. Thus, neural adaptation mechanisms might differ when central or peripheral visual degeneration occurs. Previously, we found evidence for visual retinotopic reorganization in RP (peripheral regions responding to more central representations) \[[@B6]\]. It is nevertheless also important to assess the impact of peripheral retinal loss on visual attentional mechanisms. This may provide useful information in the context of low and high level strategies for treating different retinal diseases. Additionally, there is still an ongoing debate on the nature of adult brain functional reorganization induced by retinal diseases \[[@B7]--[@B10]\].
Our study is aimed at determining the interaction between attentional mechanisms and peripheral retinal dystrophy caused by Retinitis Pigmentosa (RP) on brain function using magnetic resonance imaging (MRI). RP is an inherited degeneration of photoreceptors that initially affects the peripheral retina and later advancing towards the central retina. The onset age varies from infancy to adulthood. The disease manifestations comprise night blindness, tunnel vision, and possibly blindness in severe stages \[[@B11], [@B12]\]. In a functional magnetic resonance imaging (fMRI) case series study of three RP patients \[[@B13]\], authors reported task-dependent changes in cortical responses in the lesion projection zone (LPZ---the cortical region that no longer receives input due to a bilateral retinal lesion or scotomata \[[@B1]\]). They suggested the unmasking of feedback signals from higher-order visual areas under attentional demands when retinal input signals are lost. Another fMRI report with one RP patient found no evidence of functional alterations in the LPZ \[[@B14]\]. Our recent study in a relatively large cohort showed clear topological evidence for reorganization, which was dependent on the long-term extent of visual loss \[[@B6]\].
Contrasting with RP, central vision is primarily affected in macular degeneration. Some authors claimed that the deafferented cortical neurons in the primary visual cortex become responsive to inputs from the peripheral retina in this pathology \[[@B15]--[@B20]\]. However, other studies have questioned such visual cortical alterations by reporting the existence of a silent LPZ \[[@B14], [@B21]--[@B25]\]. Some of the previous studies showed that visual cortical alterations in macular degeneration are associated with the severity of retinal function loss, arguing that large-scale reorganization only occurs when there is a complete foveal visual loss \[[@B15], [@B18], [@B20]\]. However, other researchers did not find any signs for cortical reorganization in a large cohort of patients without foveal sparing \[[@B22]\]. Moreover, the influence of age of disease onset on the degree of cortical alterations is not clear \[[@B15], [@B22]\], although there is some evidence for larger reorganization in macular degeneration patients with earlier forms of the disease \[[@B19]\]. The reduced numbers of participants \[[@B21], [@B25]\] and the difference in stimuli and tasks used \[[@B26]\] may have also contributed to the controversy in the reported macular degeneration studies (\[[@B1]\], \[[@B7]\].
Our study was aimed at investigating the effect of peripheral retinal dystrophy caused by RP on brain attentional mechanisms using fMRI taking into account the effect of visual field extent and age of disease onset. Our hypothesis stated that, in addition to the previously demonstrated reorganization, visual cortical responses were also altered as a function of attentional demands in RP patients due to the lack of peripheral retinal bottom-up input. Moreover, we hypothesized that these alterations were more prominent for RP patients with more constricted visual fields and earlier disease onset.
2. Methods {#sec2}
==========
2.1. Participants {#sec2.1}
-----------------
The participants selected for this study were also included in a previous work from our group on visual retinotopy \[[@B6]\]. We included 13 RP individuals (8 males and 5 females; mean age 38.31 ± 12.65 years; age range 20 - 66 years; 12 right-handed and 1 left-handed; self-reported symptomatic age of onset range 2 - 39 years, resulting in symptomatic duration range 6 - 42 years) and 22 control subjects (11 males and 11 females; mean age 38.45 ± 12.29 years; age range 23 - 66 years; 21 right-handed and 1 left-handed). Both groups were matched for age (*t*~(33)~ = ‐0.03, nonsignificant *p* \> 0.050 (NS)), gender (*χ*^2^~(1)~ = 0.44, NS), and handedness (*χ*^2^~(1)~ = 0.15, NS) ratio. Patients were recruited at the Centro Hospitalar e Universitário de Coimbra. The control group participants were local volunteers. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Faculty of Medicine of the University of Coimbra. Written informed consent was obtained from all participants. Exclusion criteria were intracranial abnormalities, movement during MRI acquisitions, fixation instability, visual alterations in control subjects, or visual alterations other than RP for patients (e.g., diabetic retinopathy or glaucoma).
2.2. Ophthalmological Assessment {#sec2.2}
--------------------------------
For each participant, we measured visual acuity with a decimal chart (converted to logarithm of Minimum Angle of Resolution (logMAR) scale), average cortical thickness, and retinal nerve fiber layer (RNFL) thickness with Frequency Domain Cirrus Ocular Coherence Tomography (OCT, software version 5.1.1.6, Carl Zeiss Meditec AG, USA), and static visual fields with a MonCv3 multifunction perimeter (Metrovision, France) ([Figure 1](#fig1){ref-type="fig"}). A detailed description of the methodology used was described in our previous study \[[@B6]\] ([Table 1](#tab1){ref-type="table"}).
2.3. Brain Imaging Procedures {#sec2.3}
-----------------------------
Scanning was performed on a 3 T scanner (Magneton TrioTim, Siemens AG, Germany) at the Portuguese Brain Imaging Network, using a 12-channel birdcage head coil. Two anatomical T1-weighted Magnetization-Prepared Rapid Acquisition with Gradient Echo (MPRAGE) sequences with 1 × 1 × 1 mm^3^ voxel size, Repetition Time (TR) 2.53 s, Echo Time (TE) 3.42 ms, Flip Angle (FA) 7°, Field Of View (FOV) 256 × 256 mm^2^, and 176 slices were acquired from each participant. Functional sequences consisted of a single shot Echo-Planar Imaging (EPI) acquired in the axial plane parallel to the Anterior Commissure (AC)-Posterior Commissure (PC) plane with 2 × 2 × 2 mm^3^ voxel size, TR 2 s, TE 39 ms, interslice time (TI) 76 ms, FA 90°, FOV 256 × 256 mm^2^, 26 slices, and 128 × 128 imaging matrix.
Stimuli were presented using MRI compatible goggles with refractive correction (VisualSystem, NordicNeurolab, Norway). One eye was covered with a cotton patch while the other received the visual input (the dominant eye, except if it was the eye with the worst visual acuity). The RP and control group were matched for the selected eye ratio (*χ*^2^~(1)~ = 0.85, NS) and for the dominance of the selected eye ratio (*χ*^2^~(1)~ = 0.01, NS; 1 patient with missing data). The maximum field of view was 23 × 30 deg (resolution of 600 × 800).
2.4. MRI Stimuli {#sec2.4}
----------------
Stimuli were designed using Matlab 2011b (the MathWorks Inc., USA) with Psychophysics Toolbox 3 extensions (<http://psychtoolbox.org/>). A central red-colored cross with 0.78 deg of diameter was used for fixation \[[@B6]\].
### 2.4.1. Retinotopic Mapping Stimuli {#sec2.4.1}
Polar angle and eccentricity stimuli were employed to delineate the cortical visual areas (V1, V2, and V3) using the traveling-wave approach from the standard phase-encoded retinotopic mapping \[[@B27]\]. Polar angle maps were obtained using a black and white flickering checkerboard wedge with 45 deg rotating in an anticlockwise direction (initial angle of 22.50 deg with horizontal axis). Eccentricity maps were obtained using a black and white flickering checkerboard expanding ring (for additional details see \[[@B6]\]). Checkerboard size varied with cortical magnification factor from the center to the periphery \[[@B28], [@B29]\]. Stimuli flickering frequency was 8 Hz and contrast was \~100%. Each run comprised 2 baseline blocks (\~0% contrast; 12 s) at the beginning and end of the run with 4 cycles of polar angle or eccentricity stimuli (48 s each; total duration of a run 216 s). Two runs of polar angle and 2 runs of eccentricity were acquired for each subject.
### 2.4.2. Attentional Task Stimuli {#sec2.4.2}
The set of stimuli of the main experimental task of this study consisted of a random sequence of 2 different sized checkerboard rings pseudorandomly presented during either passive viewing (fixation only) or a one-back visual memory task condition. During the task condition, participants were instructed to press a button every time a ring was the same size as the immediately preceding one. Ring~1~ was presented at a foveal location (diameter between 0.78 and 1.90 deg), and Ring~2~ at a parafoveal location of the visual field (diameter between 6.74 and 9.52 deg). Ring thickness varied with the cortical magnification factor from the center to the periphery \[[@B28], [@B29]\]. Rings appeared during 0.50 s in random intervals of 1.50, 3.50, or 5.50 s within each block. Four passive viewing blocks and 4 task blocks (\~100% contrast; flickering frequency of 8 Hz; 10 rings; 36 s) were alternately presented, intercalated with 9 baseline blocks (\~0% contrast; 12 s; each run began and ended with one baseline block). Auditory instructions were provided to the subject before each block, depending on the condition: "Rest" for passive viewing or "Answer" for the one-back task. The average response time and percentage of errors during the task were recorded. Participants who answered during passive viewing blocks were excluded. [Figure 2](#fig2){ref-type="fig"} shows a representation of the one-back task stimulus. We aimed at performance matching between patients and controls. We preferred this choice as compared to a possible 2-back task, because we believe that this would lead to quite large executive load.
2.5. MRI Data Processing {#sec2.5}
------------------------
Brain imaging analysis was performed with BrainVoyager QX 2.6.1 (Brain Innovation B. V., Netherlands). The two anatomical images of each participant were averaged, reoriented to AC--PC plane, and transformed to Talairach (TAL) space. The image was segmented into cerebral spinal fluid, gray matter, and white matter to create inflated mesh representations of each hemisphere. A cut was manually drawn along the calcarine sulcus, and meshes were flattened for retinotopic maps projection \[[@B6], [@B30]\].
The preprocessing of functional sequences consisted of scan time correction, temporal high-pass filtering (2 cycles per run), spatial smoothing (FWHM 2 mm), and a correction for small interscan head movements. Participants were excluded if within-run movements exceeded 4 mm (-2 to 2 mm). Polar angle and eccentricity maps were obtained from the average of the two runs, created based on linear regression analysis, and projected onto the flattened surfaces of each subject (statistical maps with *r* \> 0.25; [Figure 1](#fig1){ref-type="fig"}) \[[@B6]\]. Field Sign Maps were automatically created using polar angle and eccentricity Look-Up Table maps. Retinotopic areas V1 dorsal (V1~d~), V1 ventral (V1~v~), V2 dorsal (V2~d~), V2 ventral (V2~v~), V3 dorsal (V3~d~), and V3 ventral (V3~v~) were manually defined for each subject in each hemisphere on flattened meshes.
Statistical analyses were performed on individual data in TAL space using a general linear model (GLM) (*z*-transformation, False Discovery Rate (FDR) *q* \< 0.05, correction for temporal serial correlations AR(2)) within the retinotopically defined visual areas. Response predictors for the visual memory task were obtained, and beta values evoked by each stimulus conditions were retrieved: *Ring~1~ Passive Viewing*, *Ring~2~ Passive Viewing*, *Ring~1~ Task*, and *Ring~2~ Task*. [Figure 3](#fig3){ref-type="fig"} represents the visual cortical responses for all predictors during the visual memory task for RP patients and control participants.
A multistudy GLM (random fixed effects, *z*-transformation, FDR *q* \< 0.05, AR(2)) was also run in two different regions of interest along the calcarine sulcus (V1). The functional projection zone (FPZ) represented the preserved visual field region, and the LPZ represented either the visual field scotomata in patients or the unstimulated visual field in controls. These cortical regions were manually defined considering the retinotopic eccentricity maps of each participant. [Figure 4](#fig4){ref-type="fig"} illustrates the location of the two regions of interest, the FPZ and the LPZ, in one control participant. Response predictors for the visual working memory task were obtained, and beta values evoked by each stimulus conditions were retrieved inside these regions of interest: *Ring~1~ Passive Viewing*, *Ring~2~ Passive Viewing*, *Ring~1~ Task*, and *Ring~2~ Task*.
Cortical thickness was calculated on the retinotopic areas using the standard procedure of BrainVoyager (see \[[@B30]\] for a complete description). To allow an accurate segmentation of white matter--gray matter and gray matter--cerebral spinal fluid boundary, TAL anatomical data were converted to high-resolution 0.5 × 0.5 × 0.5 mm^3^. The subcortical structures and the ventricles were filled as white matter. After computation, cortical thickness maps were superimposed on cortical meshes, and mean cortical thickness values of all visual areas were extracted using Matlab BVQXtools toolbox extensions (<http://support.brainvoyager.com/available-tools/52-matlab-tools-bvxqtools.html>).
2.6. Subgroup Analysis {#sec2.6}
----------------------
To understand the influence of the level of peripheral degeneration and disease onset age on visual cortex response, the RP group was divided accordingly to these factors. The RP group was divided accordingly to the extent of visual field measured by the static perimetry test in two subgroups: RPsvf (small visual field) (*n* = 6, patients RP 1 to RP 6) with bilateral visual field diameter under 9.52 deg and RPlvf (large visual field) (*n* = 7, patients RP 7 to RP 13) with bilateral visual field diameter over 9.52 deg (see [Table 1](#tab1){ref-type="table"}). In this way, the RPlvf patients were expected to see the complete visual working memory task stimuli (Ring~1~ and Ring~2~, because the maximum diameter of Ring~2~ was 9.52 deg), whereas most RPsvf patients would only partially see Ring~2~. Both subgroups and the control group were matched for age (*F*~(2,\ 32)~ = 1.00, NS), gender (*χ*^2^~(2)~ = 0.56, NS), handedness (*χ*^2^~(2)~ = 1.37, NS), selected eye (*χ*^2^~(2)~ = 1.44, NS), and selected eye dominance ratio (*χ*^2^~(2)~ = 1.65, NS; 1 patient with missing data). Additionally, the disease onset age (*F*~(1,\ 11)~ = 2.80, NS) and the disease symptomatic duration (*F*~(1,\ 11)~ = 0.00, NS) were not different between the subgroups RPsvf and RPlvf.
The RP group was also divided accordingly to the disease onset age into two subgroups: RPeo (early onset) (*n* = 6, patients RP 5, 7, 8, 9, 11, and 13) with an onset age lower than 14 years and RPlo (late onset) (*n* = 7, patients RP 1, 2, 3, 4, 6, 10, and 12) with an onset age greater than or equal to 14 years (see [Table 1](#tab1){ref-type="table"}). Patients from subgroup RPeo with earlier disease onset age were expected to have more prominent alterations in visual cortex responses than RPlo patients since some visual plasticity is thought to remain until 14 to 16 years of age \[[@B8], [@B31]\]. Both subgroups and the control group were matched for age (*F*~(2,\ 32)~ = 0.43, NS), gender (*χ*^2^~(2)~ = 0.57, NS), handedness (*χ*^2^~(2)~ = 1.82, NS), selected eye (*χ*^2^~(2)~ = 2.96, NS), and selected eye dominance ratio (*χ*^2^~(2)~ = 0.01, NS; 1 patient with missing data). As expected, the disease onset age (*F*~(1,\ 11)~ = 19.29, *p* = 0.001) was different between the two subgroups of patients RPeo and RPlo, but the disease symptom duration (*F*~(1,\ 11)~ = 2.81, NS) was not different. [Table 2](#tab2){ref-type="table"} summarizes the information of visual field extent and onset age for all subgroups.
2.7. Statistical Analysis {#sec2.7}
-------------------------
Statistical analysis was performed with IBM SPSS Statistics, Version 22 (IBM Corporation, USA). Normality and homogeneity of variance were tested using Shapiro-Wilk\'s test and Levene\'s test, respectively. For data in accordance with these assumptions, statistical parametric tests were performed. Otherwise, nonparametric methods were applied. Bonferroni correction was applied for multiple comparisons (*p* values presented as *correctedp*). The epsilon value was used for correction of nonspherical data (Huynh-Feldt for epsilon higher than 0.75 and Greenhouse-Geisser for epsilon lower than 0.75). The significance level was 0.05, and the statistical power was higher than 0.80 for all presented results.
3. Results {#sec3}
==========
3.1. Visual Assessment {#sec3.1}
----------------------
The first two parts of the results section (Visual Assessment and Behavioral Data) are focused on ophthalmological features---visual acuity, visual field extent, average retinal thickness, RNFL thickness-, and performance in visual memory task in patient and clinical groups and subgroups. The subsequent parts are focused on the main hypotheses of the article.
### 3.1.1. All Groups (No Stratification according to Visual Field Extent or Age of Onset) {#sec3.1.1}
Visual acuity (left eye (LE) *U* = 275.00, *p* = 2.642 × 10^−7^ and right eye (RE) *U* = 286.00, *p* = 1.355 × 10^−9^) and average retinal thickness (LE *U* = 275.00, *p* = 1.883 × 10^−7^ and RE *U* = 283.00, *p* = 9.483 × 10^−9^) were reduced in both eyes for patients as compared to control participants. Visual field deficit volume (LE *U* = 3.00, *p* = 9.483 × 10^−9^ and RE *U* = 2.00, *p* = 5.419 × 10^−9^) was higher in both eyes for patients as compared to the control group. No differences were found for RNFL thickness in both eyes between the two groups (LE *U* = 125.00, NS and RE *U* = 120.00, NS). Moreover, no statistically significant differences were found between the left and right eyes within groups for visual acuity (RP *Z* = 1.33, NS and control *Z* = 1.41, NS), average retinal thickness (RP *Z* = −0.84, NS and control *Z* = −0.88, NS), RNFL thickness (RP *Z* = 0.00, NS and control *Z* = −0.07, NS), visual field deficit volume (RP *Z* = −1.24, NS and control *Z* = −0.28, NS), and visual field extent (RP *Z* = −1.18, NS; controls have a constant visual field extent equal to 48 deg corresponding to the maximum diameter covered by our static perimetry) \[[@B6]\].
### 3.1.2. Visual Field (Large vs. Small Visual Field) Subgroup Analysis {#sec3.1.2}
Visual acuity (LE *χ*^2^~(2)~ = 22.62, *p* = 1.200 × 10^−5^ and RE *χ*^2^~(2)~ = 25.64, *p* = 3.000 × 10^−6^) and average retinal thickness (LE *χ*^2^~(2)~ = 20.91, *p* = 2.900 × 10^−5^ and RE *χ*^2^~(2)~ = 23.01, *p* = 1.000 × 10^−5^) were reduced in both eyes for the two RP subgroups as compared to control participants. No differences were found for RNFL thickness in both eyes between the three groups (LE *χ*^2^~(2)~ = 3.00, NS and RE *χ*^2^~(2)~ = 0.65, NS). Visual field deficit volume (LE *χ*^2^~(2)~ = 23.00, *p* = 1.000 × 10^−5^ and RE *χ*^2^~(2)~ = 23.29, *p* = 9.000 × 10^−6^) was higher in both eyes for both patients\' subgroups as compared to the control group. According to the subgroup division, visual field extent was different between the two subgroups (LE *χ*^2^~(1)~ = 9.10, *p* = 0.003 and RE *χ*^2^~(1)~ = 9.02, *p* = 0.003), with RPsvf (small visual field) patients presenting smaller visual fields than RPlvf (large visual field/more preserved). Moreover, no statistically significant differences were found between the left and right eyes within groups for visual acuity (RPsvf *Z* = 0.73, NS, RPlvf *Z* = 1.26, NS, and control *Z* = 1.41, NS), average retinal thickness (RPsvf *Z* = 1.15, NS, RPlvf *Z* = 0.42, NS, and control *Z* = 0.88, NS), RNFL thickness (RPsvf *Z* = 1.58, NS, RPlvf *Z* = −1.76, NS, and control *Z* = −0.07, NS), visual field deficit volume (RPsvf *Z* = −0.36, NS, RPlvf *Z* = −1.35, NS, and control *Z* = −0.28, NS), and visual field extent (RPsvf *Z* = −0.73, NS and RPlvf *Z* = −0.95, NS; controls have a constant visual field extent equal to 48 deg corresponding to the maximum diameter covered by static perimetry).
### 3.1.3. Age of Onset Subgroup Analysis {#sec3.1.3}
Visual acuity (LE *χ*^2^~(2)~ = 22.55, *p* = 1.300 × 10^−5^ and RE *χ*^2^~(2)~ = 25.42, *p* = 3.000 × 10^−6^) and average retinal thickness (LE *χ*^2^~(2)~ = 20.77, *p* = 3.100 × 10^−5^ and RE *χ*^2^~(2)~ = 22.89, *p* = 1.100 × 10^−5^) were reduced in both eyes for the two RP subgroups (RPeo, early onset; RPlo, late onset) as compared to control participants. No differences were found for RNFL thickness in both eyes between the three groups (LE *χ*^2^~(2)~ = 0.38, NS and RE *χ*^2^~(2)~ = 1.48, NS). Visual field deficit volume (LE *χ*^2^~(2)~ = 22.88, *p* = 1.100 × 10^−5^ and RE *χ*^2^~(2)~ = 23.19, *p* = 9.000 × 10^−6^) was higher in both eyes for both patients\' subgroups as compared to the control group. The visual field extent was not different between the two subgroups of patients (LE *χ*^2^~(1)~ = 2.50, NS and RE *χ*^2^~(1)~ = 2.95, NS). Moreover, no statistically significant differences were found between the left and right eyes within groups for visual acuity (RPeo *Z* = 1.08, NS, RPlo *Z* = 0.13, NS, and control *Z* = 1.41, NS), average retinal thickness (RPeo *Z* = −0.21, NS, RPlo *Z* = −1.18, NS, and control *Z* = −0.88, NS), RNFL thickness (RPeo *Z* = −1.21, NS, RPlo *Z* = −1.16, NS, and control *Z* = −0.07, NS), visual field deficit volume (RPeo *Z* = −0.40, NS, RPlo *Z* = −1.36, NS, and control *Z* = −0.28, NS), and visual field extent (RPeo *Z* = −0.52, NS and RPlo *Z* = −1.29, NS; controls have a constant visual field extent equal to 48 deg corresponding to the maximum diameter covered by static perimetry).
In sum, ophthalmological tests showed decreased patients\' visual acuity, visual field extent, and average retinal thickness, while RNFL thickness was preserved. Importantly, subgroups defined by visual field extent and age of onset only differ in the visual field extent or in the age of onset of the disease, respectively. [Table 2](#tab2){ref-type="table"} shows a summary of participants\' ophthalmologic characterization, presenting the visual parameter values for each group and subgroups.
3.2. Visual Memory Task: Behavioral Data {#sec3.2}
----------------------------------------
The mean response time (*U* = 110.00, NS) and response error (*U* = 99.00, NS) were not different between the RP and control groups during the performance of the one-back visual memory task, and both groups were actually near ceiling levels. Concerning the visual field subgroups, we did not find differences among the RPsvf and RPlvf subgroups and the control group in response time (*F*~(2,\ 32)~ = 0.03, NS) and error (*F*~(2,\ 32)~ = 1.00, NS). For the onset age subgroups, the analysis did not show differences among the RPeo and RPlo subgroups and the control group in response time (*F*~(2,\ 32)~ = 0.02, NS) and error (*F*~(2,\ 32)~ = 0.23, NS). Groups were therefore behaviorally matched.
3.3. Visual Memory Task: Responses in Functional (FPZ) and Lesion Projection Zone (LPZ) {#sec3.3}
---------------------------------------------------------------------------------------
Here, we tested the hypothesis that compensatory allocation of visual attention mechanisms do occur in patients.
### 3.3.1. All Group {#sec3.3.1}
The beta values of the task predictors (Ring~1~ Passive Viewing, Ring~2~ Passive Viewing, Ring~1~ Task, and Ring~2~ Task) were analyzed between groups for each condition (*Condition* Passive Viewing and Task), each ring (*Ring* Ring~1~ and Ring~2~), each cortical zone (*Zone* FPZ and LPZ in V1, see [Figure 4](#fig4){ref-type="fig"}), and each hemisphere (*Hemisphere* Left and Right) with repeated measures ANOVA. We found a significant effect of the predictor beta values between groups (*F~(1,33)~* = 4.47, *p* = 0.042). Importantly, within-subject effects for *Condition* (*F~(1,33)~* = 97.18, *p* = 2.313 × 10^−11^) and *Condition×Group* (*F~(1,33)~* = 9.66, *p* = 0.004) were present. To analyze the effect of the interaction *Condition×Group*, we used ANOVA between groups for each condition (Passive Viewing and Task). Results showed that the cortical responses during the task were higher for RP patients when compared to the control group (*p* = 0.002), whereas no differences were found for passive viewing condition between groups, thereby corroborating the main hypothesis.
### 3.3.2. Analyses by Subgroups Defined by Extent of Visual Field Loss {#sec3.3.2}
The statistical analysis described above was conducted to study differences among the two subgroups of patients according to the extent of visual field loss and the control group. Importantly, we found the effects for *Condition* (*F~(1,32)~* = 135.11, *p* = 5.054 × 10^−13^), *Condition×Group* (*F~(2,32)~* = 13.86, *p* = 4.600 × 10^−5^), *Zone×Ring* (*F~(1,32)~* = 11.52, *p* = 0.002), and *Zone×Ring×Group* (*F~(2,32)~* = 6.86, *p* = 0.003). ANOVA for each condition (Passive Viewing and Task) was applied to analyze the interaction effect of *Condition×Group*. Results showed that the cortical responses for task were higher for RPsvf (small visual field) patients when compared to the RPlvf (large visual field) patients (*correctedp* = 0.013) and the control group (*correctedp* = 8.500 × 10^−5^), whereas no differences were found for passive viewing condition among the three groups. In this way, patients with smaller intact visual fields presented higher cortical response during the task than patients with larger visual field extent and controls. To study the effect of the interaction *Zone×Ring×Group*, we used ANOVA among groups for values of the ring (Ring~1~ and Ring~2~) in each cortical zone (FPZ and LPZ). We found increased cortical responses for RPsvf patients for Ring~2~ in the FPZ when compared to controls (*correctedp* = 0.014). In this way, patients with smaller visual fields presented higher cortical response (2.37 ± 2.21) than controls (−0.72 ± 2.21) in the FPZ during the stimulation of paracentral visual field, while patients with larger visual field extent (0.06 ± 2.21) had similar activation to controls. These observations suggest that larger damage (present in small visual field patients) lead to larger allocation of visual attention mechanisms.
### 3.3.3. Analyses Based on Age of Onset Defined Subgroups {#sec3.3.3}
The statistical analysis described above was conducted to study differences among the two subgroups of patients according to the age of disease onset and the control group. Effects for *Condition* (*F~(1,32)~* = 90.76, *p* = 7.325 × 10^−11^) and *Condition×Group* (*F~(2,32)~* = 5.35, *p* = 0.010) were present. ANOVA for each condition (Passive Viewing and Task) was applied to analyze the interaction effect of *Condition×Group*. Results showed that the cortical responses for the task were surprisingly higher for RPlo (later onset) patients when compared to the control group (*correctedp* = 0.006), whereas no differences were found for passive viewing condition among the three groups. In sum, patients with later disease onset presented surprisingly higher cortical response during task than controls, suggesting that lower activation may represent more "efficient" brain activity patterns in early onset patients (with more prolonged disease evolution and more time to recruit mechanisms where "efficiency" dominates).
3.4. Visual Memory Task: Responses in Retinotopic Regions (V1, V2, and V3) {#sec3.4}
--------------------------------------------------------------------------
In this section, we investigate how different retinotopic areas contribute to the observed patterns of activity (overall and according to the subgroups defined above).
### 3.4.1. All Group {#sec3.4.1}
The beta values of the task predictors (*Ring~1~ Passive Viewing*, *Ring~2~ Passive Viewing*, *Ring~1~ Task*, and *Ring~2~ Task*) were analyzed between groups for each condition (*Condition* Passive Viewing and Task), each ring (*Ring* Ring~1~ and Ring~2~), each visual area (*Area* V1, V2, and V3), each cortical visual region (*Region* Ventral and Dorsal), and each hemisphere (*Hemisphere* Left and Right). We found the main effects of *Area* (*F~(1.74,57.42)~* = 50.45, *p* = 1.700 × 10^−12^), *Condition* (*F~(1,33)~* = 44.27, *p* = 1.427 × 10^−7^), and *Ring* (*F~(1,33)~* = 53.81, *p* = 2.019 × 10^−8^). We also found the effects of *Area×Condition* (*F~(1.76,58.20)~* = 25.22, *p* = 4.532 × 10^−8^), *Area×Region×Ring* (*F~(1.72,56.92)~* = 9.12, *p* = 0.001), and *Area×Region×Ring×Group* (*F~(1.72,56.92)~* = 5.98, *p* = 0.006). We can summarize these interactions by the change in visual cortical responses from V1 to V3 (*Area*) between passive viewing and the one-back task condition (*Condition*) and between Ring~1~ (central visual field) and Ring~2~ (paracentral visual field) (*Ring*) for both groups.
We then hypothesized that age of onset and extent of visual field loss could influence the effects in patients.
### 3.4.2. Visual Field Subgroup {#sec3.4.2}
The statistical analysis described above was conducted to study differences between the two subgroups of patients accordingly to the extent of visual field loss and the control group. We found the effects for *Area* (*F~(1.82,58.10)~* = 33.96, *p* = 5.461 × 10^−10^), *Condition* (*F~(1,32)~* = 58.91, *p* = 9.513 × 10^−9^), *Ring* (*F~(1,32)~* = 48.37, *p* = 7.057 × 10^−8^), *Condition×Group (F~(2,32)~* = 8.53, *p* = 0.001), *Area×Condition* (*F~(1.80,57.65)~* = 20.06, *p* = 5.902 × 10^−7^), *Area×Hemisphere×Group* (*F~(3.95,63.20)~* = 3.71, *p* = 0.009), and *Area×Region×Ring* (*F~(1.78,57.00)~* = 11.04, *p* = 1.590 × 10^−3^). An ANOVA for each condition (Passive Viewing and Task) was applied to analyze the critical interaction effect of *Condition×Group*. Results showed that the cortical responses for task were higher for RPsvf (small visual field) patients when compared to the control group (*correctedp* = 0.034), whereas no differences were found for the passive viewing condition among groups. In this way, patients with smaller visual fields presented higher cortical response during task than controls, while patients with larger visual field extent showed responses that were similar to the control participants. To study the effect of the interaction *Area×Hemisphere×Group*, we used ANOVA between groups for each visual area (V1, V2, and V3) in each hemisphere (Left and Right). We found increased cortical responses for RPsvf patients in the right V1 when compared to controls (*correctedp* = 0.044), but the remaining visual areas presented similar cortical activation. In this way, patients with smaller visual fields presented higher cortical response in the right primary visual cortex (3.17 ± 1.93) when compared to controls (0.87 ± 1.93), while patients with larger visual field extent (1.07 ± 1.93) showed similar activation to controls.
### 3.4.3. Subgroup Defined by Distinct Age of Onset {#sec3.4.3}
The statistical analysis described above was conducted to study differences among the two subgroups of patients according to the age of onset and the control group. We found the effects for *Area* (*F~(1.89,60.64)~* = 42.83, *p* = 5.413 × 10^−12^), *Condition* (*F~(1,32)~* = 40.44, *p* = 3.853 × 10^−7^), *Ring* (*F~(1,32)~* = 51.74, *p* = 3.615 × 10^−8^), *Area×Group* (*F~(3.79,60.64)~* = 4.46, *p* = 0.004), *Area×Condition* (*F~(1.82,58.365)~* = 20.28, *p* = 4.572 × 10^−7^), *Area×Hemisphere×Group* (*F~(3.96,63.40)~* = 3.99, *p* = 0.006), *Region×Ring×Group* (*F~(2,32)~* = 5.66, *p* = 0.008), and *Area×Region×Ring* (*F~(1.79,57.38)~* = 11.51, *p* = 1.100 × 10^−4^).
In sum, this analysis focused on retinotopic areas that essentially mimics the findings observed for the LPZ and FPZ zones, corroborating the main hypothesis of preferential attentional allocation in patients, and distinct effects of the visual field lesion extent and age of onset.
3.5. Cortical Thickness of Visual Areas (V1, V2, and V3) {#sec3.5}
--------------------------------------------------------
Finally, we investigated whether functional changes were associated with structural alterations.
Visual cortical thickness differences were evaluated using repeated measures ANOVA with three within-subject factors (*Area* V1-V3, *Region* Ventral vs Dorsal, and *Hemisphere* Left Vs Right), one between-subject factor *Group* (RP vs. Control), with the average brain cortical thickness as a covariate to account for variability across participants. Cortical thickness of the individually defined visual areas was not different between the two groups (*F~(1,32)~* = 0.047, NS), and no within-subject effects or interactions were found. This was also true for subgroup analyses.
4. Discussion {#sec4}
=============
We investigated whether visual cortical responses in a disorder of peripheral vision are related to recruitment of attentional mechanisms. To test this hypothesis, we used a visual one-back task and passive viewing conditions with a visual stimulus covering the central (Ring~1~) and paracentral (Ring~2~) visual field in a group of RP patients (*n* = 13) with peripheral retinal loss and matched healthy controls (*n* = 22). Cortical responses were studied in visual retinotopic areas (V1, V2, and V3) and in two different regions of interest in V1: the FPZ representing the preserved visual field and the LPZ representing the visual field scotomata. To understand the influence of the level of peripheral degeneration and the disease age of onset, the analysis was further conducted for two distinct RP subgroups: subcategories defined by the extent of visual field loss (RPsvf, remaining small field, and RPlvf, remaining large field patients with bilateral visual field diameters under or over 9.50 deg, respectively), and subgroups defined by distinct ages of disease onset (RPeo, early onset, and RPlo, late onset---patients with age of onset of the disease lower or equal/greater than 14 years, respectively).
Our results demonstrated that RP patients have overall preserved visual cortical responses under central and paracentral visual field stimulation. Visual cortical responses (V1, V2, and V3) to the visual memory task stimuli were also overall preserved. A critical interaction with task condition was however found: RP patients presented higher overall cortical responses during the task condition than control participants in FPZ and LPZ regions. This was further highlighted when extent of visual loss was taken into account.
Concerning the role of the extent of visual loss, RPsvf patients, with smaller visual fields, presented higher overall visual cortical responses during the task condition than control participants, while responses for RPlvf patients, with larger visual field extent, were similar to control participants. Additionally, RPsvf patients had significantly higher cortical activation in the right V1 when compared to controls, while responses for RPlvf patients were similar to healthy participants. In line with these results, RPsvf patients presented higher overall cortical responses during the task condition than control participants and RPlvf patients in LPZ and FPZ regions, while responses for RPlvf patients were similar to control participants. Additionally, the cortical activation in the FPZ was higher for RPsvf patients during the stimulation of the paracentral visual field (Ring~2~) when compared to the healthy subjects. Because visual stimuli were the same in both task and passive viewing conditions, these responses for RPsvf patients seem to be related to attentional demands during the one-back task. Masuda et al. \[[@B13]\] found a similar increase in striate cortical responses of three RP participants related to changes in task demands and suggested that unmasking of feedback signals from the extrastriate cortex occurs when retinal signals are absent. Such unmasking might come from activation of previously silent synapses \[[@B32]\]. These feedback signals might be associated with attention, visual imagery, and task-related visual processing \[[@B26]\]. In our work, enhanced attentional top-down modulation may compensate for the lack of retinal input from the peripheral visual field in RPsvf patients with greater visual field loss \[[@B9], [@B24], [@B33]\]. A recent study presented evidence for increased functional connectivity between afferent early visual areas and cortical regions involved in visual processing (middle occipital gyrus and superior temporal gyrus/sulcus) in RP patients, suggesting a possible compensatory mechanism for peripheral visual loss. These authors also found enhanced functional connectivity between the deafferented visual cortex and higher-order regions (inferior parietal lobe/sulcus and middle frontal gyrus) involved in top-down control, attentional processes, and multisensory integration \[[@B34]\].
However, in the three patients reported in Masuda et al.\'s work, the increased responses during task were found in the V1 LPZ \[[@B13]\]. Here, we report an overall increase in cortical responses under task demands while analyzing V1, V2, V3, and V1 FPZ and LPZ for patients with more severe visual field degeneration. In this way, feedback signals might influence both striate and extrastriate visual cortex when there is a severe lack of peripheral retinal input, not being restricted to the LPZ. A recent work from our group provided evidence of functional remapping in V1 in the same group of patients studied here. This functional reorganization was also more prominent in RP patient with larger visual field damage \[[@B6]\].
Previous works with macular degeneration also showed increased cortical responses for V1 LPZ while patients performed a one-back task with peripheral stimulation \[[@B15]--[@B19], [@B24]\], contrary to passive viewing stimulation \[[@B19], [@B21]--[@B25]\]. Some of the authors showed that these V1 responses were higher for more severe central retinal loss without foveal sparing \[[@B15], [@B18]\], in accordance with our work. Recently, Plank et al. \[[@B26]\] reported that patients with central scotomata presented enhanced cortical activation in areas beyond the retinotopic cortex for complex images with naturalist scenes, supporting an increased top-down modulation of the deprived visual cortex. This result was further supported by the work of Sabbah et al. \[[@B34]\] showing increased functional connectivity between the LPZ and high-level regions in central retinal disease patients.
Several studies with glaucoma patients found reduced amplitude of cortical responses in V1 during passive viewing stimulation associated with structural damage of the optic disk, the RNFL thickness, and/or the visual field scotomata \[[@B35]--[@B39]\]. A recent study demonstrated reduced cortical activity within the LPZ in V1 and V2 in glaucoma patients under passive viewing \[[@B40]\]. Here, we did not find decreased activity during passive viewing in the cortical regions studied, which might indicate that RP patients have preserved visual cortical responses even for severe visual field damage, possibly due to a similar compensation mechanism by increased attentional modulation.
The cortical responses in visual areas were not globally significantly different among the RPeo and RPlo patients\' subgroups with different onset ages and the control participants. Thus, overall visual cortical responses do not seem to be influenced by disease onset age, in contrast to the extent of visual loss. However, RPlo patients with later disease onset presented higher overall cortical responses during the task condition than control participants in FPZ and LPZ regions, while responses for RPeo patients with earlier forms of the disease were similar to control participants. This result was unexpected considering our initial hypothesis that earlier onset ages would lead to larger brain alterations. Two sorts of mechanisms might be operating: the first requiring long-term circuit modifications and, more present in early-onset patients, the second entailing stronger frontoparietal recruitment which tends to manifest more in patients with more recent changes in visual experience. There is indeed evidence showing that higher top-down modulation may indeed be stronger in participants with more recent changes in visual experience \[[@B41], [@B42]\]. Moreover, it may reflect the fact that longer disease durations may lead to efficient compensatory mechanisms and decrease of frontoparietal activation. A second study from Rosa et al. \[[@B41], [@B42]\] shows that frontoparietal activation decreases over time as patients\' vision becomes more adapted. Less fMRI activation might actually indicate "more efficient" compensation \[[@B41], [@B42]\]. Studies in macular degeneration patients did not report the effects of the age of onset on visual cortical responses \[[@B15], [@B22]\], while others showed that juvenile-macular degeneration patients with earlier disease forms have stronger cortical activation than age-related macular degeneration patients with later onset age \[[@B19]\]. Future studies should address the discrimination between age-dependency of neuroplasticity and disease-duration effects, which are separable. Disease onset age is often difficult to determine, which can make this an imprecise measure of RP severity \[[@B11], [@B43]\].
Given the evidence for cortical reorganization in our prior study \[[@B6]\], remodeling at the retinal level is unlikely. This issue can be further clarified in the future by explicitly computing population receptive fields or alternatively running experiments with artificial scotomata.
Our results did not find evidence for visual cortical structural alterations in this cohort of RP patients, showing that the visual loss level was not sufficient to produce significant cortical atrophy in the visual areas studied (V1 to V3). To our knowledge, few structural MRI studies have been conducted with low vision RP patients \[[@B3], [@B44]\]. In our study, the RP patients did not present RNFL thickness atrophy which is in line with the preservation of visual cortical thickness. Nonetheless, in more advanced stages of RP disease with larger photoreceptor loss and retinal ganglion cell degeneration, disuse-driven mechanisms may lead to the visual cortical atrophy pattern that is often seen in macular degeneration, glaucoma, and also late-blindness.
5. Conclusion {#sec5}
=============
We found that cortical visual areas (V1, V2, and V3) responses under attentional demands were increased in patients with larger degeneration of visual field. Moreover, activation during the task condition was increased for patients in both cortical regions corresponding to the preserved (FPZ) and the damaged visual field (LPZ), specifically for patients with severe visual field loss. These findings were identified in the presence of preserved visual cortical structure. The age of onset of the disease did not seem to be associated with visual cortical alterations. We conclude that RP patients may have relatively preserved visual cortical responses due to feedback attentional modulation in the absence of cortical atrophy, despite their retinal degeneration. The unmasking of corticocortical feedback signals from higher level visual regions involved in attentional processes might explain the increased cortical responses \[[@B1]\]. Such unmasking might lead to activation of previously silent synapses.
These results might be considered in the context of strategies for treating retinal diseases \[[@B21], [@B45], [@B46]\]. This is quite relevant given previous evidence that attentional cueing improves vision restoration therapy in patients with visual field loss \[[@B47], [@B48]\]. The role of higher-level neuronal networks \[[@B49]\] and their functional connectivity \[[@B50]\] cannot be underestimated in this context. This suggests that visual responses can be dynamically adapted as a function of flexible mechanisms requiring the interaction between high-level regions that implement attentional control.
This work was supported by the Portuguese *Funding Agency for Science and Technology* (FCT) grants E-Rare2-SAU/0001/2008, E-Rare4/0001/2012, COMPETE, POCI-01-0145-FEDER-007440, FCT. UID/NEU/04539/2013--2020, MEDPERSYST, POCI-01-0145-FEDER-016428, and BIGDATIMAGE, CENTRO-01-0145-FEDER-000016 financed by Centro 2020 FEDER, COMPETE. We would like to thank Carlos Ferreira, João Marques, and Sónia Afonso for technical assistance in the magnetic resonance acquisitions. We express our thanks to the patients and their families and the other participants for their collaboration.
RP:
: Retinitis Pigmentosa
MRI:
: Magnetic resonance imaging
fMRI:
: Functional magnetic resonance imaging
RNFL:
: Retinal nerve Fiber layer
LPZ:
: Lesion projection zone
FPZ:
: Function projection zone
deg:
: Degrees.
Data Availability
=================
The data used to support the findings of this study are available from the corresponding author upon request.
Disclosure
==========
This work was presented in an abstract form in the 2017 European Association for Vision and Eye Research Conference.
Conflicts of Interest
=====================
The authors declare that no conflicts of interest exist.
{#fig1}
{#fig2}
{#fig3}
{#fig4}
######
Summary of the participants\' characterization and ophthalmological test results for the Retinitis Pigmentosa group (adapted from \[[@B6]\]).
Patient Age (years) Gender Eye dominance Onset age (years) Disease duration (years) Visual acuity (logMAR) Retinal thickness (*μ*m) RNFL thickness (*μ*m) Visual field deficit volume (dB·deg^2^) Visual field extent (\~diameter in deg)
--------- ------------- -------- --------------- ------------------- -------------------------- ------------------------ -------------------------- ----------------------- ----------------------------------------- ----------------------------------------- -------- --------- --------- ------- -------
RP 1 66 F a 27 39 0.30 0.30 254.00 248.00 95.00 108.00 1349.00 1349.00 6.50 6.50
RP 2 42 M LE 18 24 0.22 0.30 194.00 208.00 100.00 116.00 1718.00 1757.00 8.00 4.50
RP 3 50 M RE 16 34 0.30 0.30 201.00 203.00 70.00 62.00 1168.00 1156.00 8.00 8.00
RP 4 23 M RE 16 7 0.30 0.18 254.00 266.00 130.00 143.00 1240.00 1232.00 8.00 8.50
RP 5 35 M LE 6 29 0.05 0.30 228.00 243.00 89.00 91.00 1700.00 1700.00 9.50 10.50
RP 6 45 F LE 39 6 0.10 0.18 192.00 184.00 76.00 81.00 1204.00 1181.00 10.50 8.50
RP 7 20 M LE 7 13 0.22 0.22 225.00 228.00 106.00 99.00 1639.00 1666.00 14.50 13.00
RP 8 35 F RE 3 32 0.52 0.22 281.00 265.00 101.00 101.00 1650.00 1609.00 19.00 15.50
RP 9 50 M LE 8 42 0.18 0.40 205.00 207.00 79.00 73.00 1414.00 1402.00 20.50 18.50
RP 10 38 F RE 32 6 0.00 0.40 249.00 258.00 128.00 128.00 1520.00 1423.00 21.50 21.00
RP 11 38 F RE 6 32 0.10 .55 216.00 221.00 133.00 109.00 1118.00 1030.00 23.00 29.00
RP 12 25 M RE 14 11 0.40 0.30 242.00 245.00 101.00 95.00 802.00 539.00 43.00 43.00
RP 13 31 M LE 2 29 0.10 0.40 273.00 268.00 91.00 92.00 72.00 126.00 47.50 47.00
RP = Retinitis Pigmentosa; F = female; M = male; LE = left eye; RE = right eye; RNFL = retinal nerve fiber layer; logMAR = logarithm of Minimum Angle of Resolution; ^a^missing information.
######
Ophthalmological characterization of the participants from the patients\' and controls\' groups and for the patients from visual field extent subgroups (RPsvf and RPlvf) and the disease onset age subgroups (RPeo and RPlo). The visual field and onset age subgroups only differ in the visual field extent and in the age of onset of the disease, respectively. Results showed a severe decrease of patients\' visual acuity, visual field extent, and average retinal thickness when compared to the control group (RNFL thickness was unchanged).
Visual parameters Eye RP group (*n* = 13) Control group (*n* = 22) Visual field extent Disease age of onset
----------------------------------------- ------------------ --------------------- -------------------------- --------------------- ---------------------- ------------------ ------------------
Visual acuity (logMAR) LE 0.22 (0.20) 0.00 (0.11) 0.26 (0.22) 0.18 (0.30) 0.14 (0.21) 0.30 (0.20)
RE 0.30 (0.18) 0.00 (0.11) 0.30 (0.12) 0.40 (0.18) 0.35 (0.21) 0.30 (0.12)
Retinal thickness (*μ*m) LE 228.00 (51.00) 288.50 (21.50) 214.50 (60.50) 242.60 (57.00) 226.50 (61.75) 242.00 (60.00)
RE 243.00 (54.00) 285.00 (20.75) 225.50 (54.25) 245.00 (44.00) 235.50 (48.25) 245.00 (55.00)
RNFL thickness (*μ*m) LE 100.00 (33.00) 94.50 (10.50) 92.00 (33.00) 101.00 (37.00) 96.00 (26.25) 100.00 (52.00)
RE 99.00 (26.50) 95.00 (17.25) 99.50 (46.50) 99.00 (17.00) 95.50 (16.50) 108.00 (47.00)
Visual field deficit volume (dB·deg^2^) LE 1349.00 (501.50) 30.00 (24.25) 1294.50 (509.50) 1414.00 (837.00) 1526.50 (806.00) 1240.00 (353.00)
RE 1349.00 (544.50) 27.50 (44.00) 1290.50 (539.50) 1402.00 (1070.00) 1505.50 (870.50) 1233.00 (267.00)
Visual field extent (\~diameter; deg) LE 14.50 (14.25) 48^a^ 8.00 (2.13) 21.50 (24.00) 19.75 (15.88) 8.00 (13.50)
RE 13.00 (16.75) 48^a^ 8.25 (3.00) 21.00 (27.50) 17.00 (21.13) 8.50 (14.50)
Onset age onset (years) --- 14.92 ± 11.58 --- 20.33 ± 11.32 10.29 ± 10.34 5.33 ± 2.34 23.14 ± 9.63
Disease duration (years) --- 23.38 ± 13.07 --- 23.17 ± 13.85 23.57 ± 13.48 29.50 ± 9.40 18.14 ± 14.09
Data are median (interquartile range) and mean ± standard deviation; RP = Retinitis Pigmentosa; RPsvf = small visual field; RPlvf = large visual field; RPeo = early onset; RPlo = late onset; LE = left eye; RE = right eye; RNFL = retinal nerve fiber layer; logMAR = logarithm of Minimum Angle of Resolution. ^a^Visual field extent for the control group is the maximum diameter tested during the static perimetry (48 deg).
[^1]: Academic Editor: Stuart C. Mangel
| {
"pile_set_name": "PubMed Central"
} |
Pancreatic cancer (PC) has a dismal prognosis and is currently the fourth leading cause of cancer-related mortality, and it is expected to become the second within the next 20 years ([@bib23]; [@bib4]; [@bib29]). Owing to its asymptomatic nature and high metastatic potential, the diagnosis of PC is only possible for those in an advanced state, and the prognosis of PC remains the worst of the major malignancies ([@bib13]; [@bib36]). The medium survival rate of PC after diagnosis is \<6% and the 5-year survival rate has remained at ∼5--7% for decades ([@bib37]; [@bib35]; [@bib30]). The ability to diagnose PC in asymptomatic patients would allow many patients to be actively treated, thereby greatly improving their prognosis ([@bib19]).
Many researchers have aimed to identify effective biomarkers for the early detection of PC ([@bib6]). In our previous work, novel prognostic predictors of PC and PC-associated diabetes mellitus were investigated based on the analysis of surgically resected fresh PC tissues and adjacent non-tumour tissues ([@bib32], [@bib33], and so on). Ideally, a blood-based biomarker or biomarker panel would be more optimal as it would be more feasible and minimally invasive. The Food and Drug Administration-approved blood-based biomarker CA 19--9 has demonstrated only modest effectiveness for the diagnosis of PC, with variable sensitivity (SN, 60--90%) and specificity (SP, 68--91%) ([@bib24]; [@bib14]; [@bib15]; [@bib18]). It also showed false negative results in the Lewis negative phenotype (5--10%) ([@bib24]) and false positive results in the presence of obstructive jaundice (10--60%) ([@bib11]). These limitations of CA 19-9 have led to the urgent search for alternative biomarkers. The development of new methodologies for the discovery of biomarkers is an ongoing endeavor ([@bib9]).
A typical proteomics-based biomarker pipeline starts with a discovery stage, followed by verification and validation of the candidate biomarker for its intended clinical use ([@bib27]). Discovery proteomics analyses have rapidly developed to detect and comprehensively quantitate proteins expressed in complex biological systems, generating hundreds of candidate biomarkers of differential abundance ([@bib28]; [@bib40]; [@bib39]; [@bib7]). Immunoassays and stable isotope dilution-multiple reaction monitoring mass spectrometry (SID-MRM) offers robust, high-throughput, and absolute quantification of targeted peptide(s) across different samples ([@bib1]; [@bib21]; [@bib25]). However, lacking of commercial available antibodies for specific proteins and posttranslational modifications hinders the development of newly discovered biomarkers ([@bib22]). Besides, it is both time consuming and expensive to verify dozens or hundreds of candidate biomarkers for both antibody-based and SID-MRM assays. Additionally, because of the wide dynamic range of protein content in serum samples, high-abundant protein depletion strategies and extensive separation of enzymatically digested peptides are utilised for better coverage of protein identification in the discovery stage. Therefore, a general approach is needed to verify and prioritise the subset of candidate biomarkers that are detectable in the whole serum sample using one-dimensional liquid chromatography MRM-MS (1D LC-MRM-MS) analyses, which can validate biomarkers with high throughput and high efficiency.
Herein, we employed isobaric tags for relative and absolute quantitation (iTRAQ)-based comparative proteomics analysis, 1D targeted LC-MS/MS, a prime MRM without SIS peptides and SID-MRM in an integrated workflow for biomarker candidate discovery, verification and validation, respectively. The data from each stage can systematically inform the next stage without discrimination. To the best of our knowledge, this is the first construction of a coherent and MS-intensive pipeline for biomarker development in PC. A total of 150 serum samples from healthy people (normal control, NC), patients with benign diseases (BD) and PC patients were analysed, and a new panel of candidate biomarkers consisting of apolipoprotein E (APOE), inter-alpha-trypsin inhibitor heavy chain H3 (ITIH3), apolipoprotein A-I (APOA1) and apolipoprotein L1 (APOL1) showed significant differences between PC *vs* NC and BD groups. The combined diagnosis of the four proteins and CA19-9 outperformed CA19-9 alone in the diagnosis of PC and thus could serve as a potential predictive biomarker panel.
Materials and methods
=====================
Patients and specimens
----------------------
A total of 150 blood samples were recruited (using informed consent) at the Zhongshan Hospital between June 2010 and January 2012, and categorised as follows: NC (*n*=40), BD (*n*=30, pancreatitis (4), pancreatic cysts (13), benign tumours (13)), and PC (*n*=80). The research followed the tenets of the Declaration of Helsinki and was approved by the Ethics Committee of the Fudan University Shanghai Zhongshan hospital. Blood samples were collected in the morning after an overnight fast using Vacutainer tubes (Becton Dickinson, Franklin Lakes, NJ, USA) without anticoagulant and allowed to clot at room temperature for 1 h before centrifugation at 1500 g for 10 min. The serum was removed, immediately aliquoted in sterile centrifuge tubes and stored at −80 °C for future analysis.
2D LC-MS/MS analysis of iTRAQ-labelled peptides
-----------------------------------------------
In the discovery stage, every 10 serum samples were pooled together in each group for subsequent analysis. The high-abundant proteins from the pooled serum sample were depleted using a Human 14 Multiple Affinity Removal System Column (Agilent Technologies, Santa Clara, CA, USA). Proteins were then digested ([@bib34]), followed by iTRAQ labelling, according to the manufacture's instructors. As a consequence, two sets of iTRAQ 8-Plex (NC with 113 and 114 tags, BD with 115 and 116 tags, and PC with 117, 118, 119 and 121 tags) and 1 set of iTRAQ 5-Plex (NC with 113 tags, BD with 115 and 116 tags, and PC with 117 tags) were constructed to provide multiple biological replicates. The labelled peptides were then fractionated with high pH reversed-phase liquid chromatography on a UPLC system (Waters, Milford, MA, USA).
Nano-LC-MS/MS analyses were performed on a Nanoeasy system with a 50- cm-long column (75 um id × 50- cm-long,C18, Thermo Fisher Scientific, San Jose, CA, USA) connected to a Q Exactive Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific), and a 1D Plus nano LC system (Eksigent of Sciex, Framingham, MA, USA) coupled with the Triple TOF 5600 system (Sciex, Framingham, MA, USA). The detailed methods may be found in the [Supplementary section](#sup1){ref-type="supplementary-material"}.
Protein identification and quantitation analysis were performed with Proteome Discovery (v.1.3, Thermo Fisher Scientific) and ProteinPilot (version 4.5, Sciex). All the data were searched against the Swiss-Prot human database (20,238 entries) with MS tolerance set at 20 ppm, and MS/MS tolerance set at 0.1 Da.
In this study, a false discovery rate (FDR) lower than 1% was used to control protein level identification based on the target-decoy strategy. Proteins with at least one unique peptide with confidence higher than 95% were used for quantitation. Student\'s *t*-test (PC *vs* control (NC and BD)) was applied to compare the protein expression levels between the PC group and the control group. The mean value of the ratio of each group was used to calculate the fold change. Proteins with a fold change larger than 1.2 or less than 0.8 with a Student\'s *t*-test *P*-value \<0.05 were selected as differently expressed proteins. A total of 142 proteins met these criteria.
1D-targeted LC-MS/MS
--------------------
Referring to the spectrums generated above, the precursor-ion intensities of all the unique peptides of the 142 proteins were analysed. The six most intense unique peptides of each protein were selected and their m/z values were set as the inclusion list for 1D-targeted LC-MS/MS detection on a Triple TOF 5600 system. A total of 2 μg of enzymatically digested crude serum sample without high abundant protein depleted was used for the analysis. A short list of 49 proteins was identified with high confidence (FDR\<1%, peptide confidence \>95%).
P-MRM analysis
--------------
For the P-MRM analysis, a total of 96 crude unique peptides corresponding to the 49 proteins were synthesised and used to optimise the transition selection, method building, retention time scheduling for the MRM assay development. An exogenous peptide was added to each digested crude serum sample working as an internal standard. In total, 1 pmol of each crude peptide and 2 μg of digested peptides from 52 crude serum samples were analysed separately.
The MRM analyses were performed on a 6500 QTRAP hybrid triple quadrupole/liner ion trap mass spectrometer (Sciex) interfaced with a UPLC system (Eksigent of Sciex) using a 15-cm-long column (75 μm id × 150, C18). The MRM data were processed using Skyline software (v 3.1) resulting in 4 significantly changed candidate biomarkers (*P*\<0.05%, comparing PC with NC and BD groups; fold change \>1.2 or fold change \<0.8).
SID-MRM analysis
----------------
For the SID-MRM analysis, four stable isotope-labelled peptides corresponding to the best performing peptides of the targeted proteins were synthesised (Bankpeptide, Ltd., China). SIS peptide of 4--5 orders of magnitude were added to the digested serum proteins and tested in triplicate to construct a standard curve ([Supplementary Information S8](#sup1){ref-type="supplementary-material"}). A certain concentration of SIS peptide were spiked in each sample and the absolute quantitation of interested peptides were carried out in 100 serum samples (34 NCs, 26 BDs and 40 PCs). The concentrations of the endogenous peptides were calculated as follows: *C*~endogenous~=*C*~SIS~ × peak area~endogenous~/peak area~SIS.~ Detailed information can be found in the [Supplementary Methods Section](#sup1){ref-type="supplementary-material"}.
Immunoassay measurement of CA19-9 level
---------------------------------------
CA19-9 levels of the 100 serum samples tested in SID-MRM analysis were determined by electrochemiluminescence immunoassay (double-antibody sandwich ELISA) on a Roche cobas e 602 module according to the manufacturer's instructions (Roche Diagnostics, Mannheim, Germany).
Statistical construction of a diagnostic model
----------------------------------------------
The quantitative results from the P-MRM and SID-MRM analyses were compared and visualised using Prism 5.0 (GraphPad Software Inc., La Jolla, CA, USA). The peptide concentrations of APOE, ITIH3, APOA1, APOL1 and the expression level of CA19-9 in the serum of the NC group (34 cases), BD (26 cases) and the PC group (40 cases) were used to construct the diagnostic model. The statistical analyses were performed using SPSS (v24.0, IBM, Armonk, NY, USA), and *P*\<0.05 was considered statistically significant. Receiver operation characteristic curves (ROCs) were calculated to determine the specificity and sensitivity, as well as to compare the area under the curve (AUC) of single candidate biomarkers and their combinations using a binary logistic regression analysis ([@bib10]).
Immunohistochemistry
--------------------
A total of 4 μm of whole formalin-fixed and paraffin-embedded tissue section samples (cancer and para-cancer tissues of pancreas) were prepared. The samples were deparaffinised with xylene, followed by rehydration in a series of four graded alcohols (70, 80, 90 and 100%). Rabbit polyclonal antibodies, anti-APOL1 (1 : 300 dilution, Proteintech, Wuhan, China), anti-APOA1 (1 : 50 dilution, Proteintech), anti-ITIH3 (1 : 50 dilution, Proteintech), and anti-APOE (1 : 1000 dilution, Proteintech) were incubated for 1 h at room temperature, detected with ImmPRESS-HRP anti-rabbit IgG reagent (Beyotime, China) and visualised using DAB+substrate (Dako). An Aperio Scanscope XT (Leica Biosystems, Vista, CA, USA) was used to digitally scan the slides.
Results
=======
Discovery MS
------------
[Figure 1](#fig1){ref-type="fig"} shows the overall workflow for the discovery, verification and validation of the candidate biomarkers for PC. In the whole pipeline, a series of MS-based methods were applied as follows: in stage I, iTRAQ-2DLC-MS/MS was applied to analyse the expression level for up to a thousand proteins in the serum samples (with highly abundant proteins depleted) from the NC, BD and PC groups. In stage II, 1D-targeted LC-MS/MS was utilised to ascertain the detection of the 142 altered protein in 1D LC-MS/MS, resulting in a shorten list of 49 interesting proteins. For cost savings and non-discriminant selection of candidate biomarkers, in stage III, the 49 proteins were further verified using P-MRM in 52 crude serum samples, resulting in a prioritised panel of proteins that were further absolutely quantitated and validated by SID-MRM in 100 serum samples in stage IV. The performance of the biomarker candidates was evaluated using the ROC curves, which was based on the quantitated concentration of these proteins.
iTRAQ 2D-LC-MS/MS enables an unbiased quantitative comparison of the expression levels of proteins in different samples and is widely used in biomarker discovery. However, protein content of serum samples has a wide dynamic range which spans \>12 orders of magnitude ([@bib31]). High abundant proteins such as albumin masks or sequesters the detection of lower abundant proteins ([@bib22]). To extensively identify serum proteome, we depleted high abundant proteins prior to comparative proteomics analysis. Besides, we used extremely high pressures in LC and a long column packed with small particles to improve the separation efficiency ([@bib38]). In this study, a total of 1,217 proteins were identified with a FDR of \<1%, of which 142 proteins were differentially expressed according to the criteria mentioned above ([Supplementary Information S1--S3](#sup1){ref-type="supplementary-material"}). Among these, 78 proteins were increased \>1.20-fold in serum samples from PC group compare with NC and BD groups with *P*-value \<0.05, and 64 proteins were decrease \<0.8-fold in the PC group ([Supplementary Information S4](#sup1){ref-type="supplementary-material"}). All the up and down regulated proteins were applied for further analysis.
1D-targeted LC-MS/MS
--------------------
To ascertain which of the proteins discovered in a iTRAQ- 2D-LC-MS/MS method could also be detected in 1D-LC-MS/MS analysis of crude serum sample without high abundant protein depletion, a 1D-targeted LC-MS/MS analysis was incorporated in the pipeline. As a result, a total of 49 proteins were identified; other proteins were not detected due to the weak signal and ion suppression that resulted from the high dynamic range of the crude serum proteins ([Supplementary Information S4](#sup1){ref-type="supplementary-material"}).
Relative quantification of candidate proteins using the P-MRM assay
-------------------------------------------------------------------
To ensure the quality of P-MRM method analysis, crude peptides corresponding to the 49 proteins were used for the assay development and another exogenous peptide was monitored as an internal standard. The relative quantitation and comparison of each peptide was based on the integration of the areas of the chromatography peaks of the transitions for each peptide ([Supplementary Information S5](#sup1){ref-type="supplementary-material"}). The coefficient of variation (CV) of the summed area of transitions of the internal standard peptide was 14% in all 52 samples, which indicated that it is reliable to approximately evaluate the relative amount of peptides in the different samples according to the P-MRM results.
According to the results, 47 peptides corresponding to 27 proteins showed significant changes between the PC group and NC group ([Supplementary Information S6](#sup1){ref-type="supplementary-material"}). Some of the proteins, such as fibronectin (FINC, *P*=0.001), thrombospondin-1 (TSP1, *P*\<0.001), lumican (LUM, *P*\<0.001), retinol-binding protein 4 (RET4, *P*\<0.001), and gelsolin (GELS, *P*\<0.001) ([Figure 2](#fig2){ref-type="fig"}), showed marked differences between the PC and NC groups (p-values as above). However, the concentrations of these proteins were not significantly different comparing PC with BD groups. These proteins can help us to distinguish patients with pancreatic disorders from healthy people, but they are not appropriate biomarkers for PC.
Of all the results, APOE and ITIH3 expression was significantly increased in PC ([Figure 2C and D](#fig2){ref-type="fig"}), whereas APOA1 and APOL1 expression was apparently decreased in PC compared with that in the controls (BD and NC groups) ([Figure 2A and B](#fig2){ref-type="fig"}), which was consistent with the iTRAQ findings. These four proteins were selected as candidates for further confirmation and absolute quantification using the SID-MRM assay.
Absolute quantification of 4 candidate proteins using SID-MRM
-------------------------------------------------------------
On the basis of the relative quantification results, four SIS peptides corresponding to the 4 selected proteins (APOE, ITIH3, APOA1, and APOL1) were synthesised for absolute quantification. The details of the confirmed peptides are shown in [Supplementary Information S7](#sup1){ref-type="supplementary-material"}. Standard curves were tested based on the SIS peptides. The correlation coefficients of the weighted calibration curves of the four SIS peptides ranged from 0.9 to 1 ([Figure 3A, D, G and J](#fig3){ref-type="fig"}). The lower limit of quantitation (LLOQ, S/N\>10) of the 4 SIS peptides was also determined ([Supplementary Information S8](#sup1){ref-type="supplementary-material"}). The good linearity and reproducibility of the 4 SIS peptides ([Figure 3B, E, H and K](#fig3){ref-type="fig"}) proved the reliability of the SID-MRM method we developed. Group comparisons were performed according to the concentration of each endogenous peptide ([Figure 3C, F, R, L](#fig3){ref-type="fig"}, [Supplementary Information S10](#sup1){ref-type="supplementary-material"}). The results showed that the concentrations of APOA1, APOL1, APOE, and ITIH3 were significantly different (*P*\<0.004) between the PC *vs* NC and BD groups ([Supplementary Information S9](#sup1){ref-type="supplementary-material"}).
Immunoassay measurement of CA19-9 level
---------------------------------------
The result of CA19-9 expression level of the 100 serum samples are shown in [Supplementary Information S9](#sup1){ref-type="supplementary-material"}.
Combination biomarker models outperform CA19-9 alone
----------------------------------------------------
To gain a further insight to the utility of these markers, binary logic regression was performed to produce predictive models that were then analysed by ROC curves. [Figure 4](#fig4){ref-type="fig"} shows the performance, in terms of the area under the curve (AUC), sensitivity, specificity values, of the serum factors such as APOE, ITIH3, APOA1, APOL1, CA19-9. The assessment of the combination of the 4 newly discovered proteins (Com-4 proteins) and the panel of the combination of all the five elements (Com-all) were carried out as well.
For analysis that uses all the samples, including 34 NCs, 26 BDs and 40 PCs, we set NC and BD group together as the control. To differentiate PC from the control group, the plot demonstrates a significant improvement of AUC and Youden Index for the Com-4 proteins and Com-all compare with CA19-9 alone ([Figure 4B](#fig4){ref-type="fig"}). The Com-4 proteins and Com-all panels outperformed CA19-9 alone for the differentiation of PC *vs* NC &BD group.
Comparing PC with NC group, the AUC values for APOE, ITIH3, APOA1, APOL1, CA19-9 were 0.669 (*P*=0,013), 0.784 (*P*\<0.001), 0.896 (*P*\<0.001), 0.803 (*P*\<0.001), and 0.78(*P*\<0.001), respectively. The Com-4 proteins robustly increased the AUC to 0.937 (*P*\<0.001), and the sensitivity and specificity were 85.0 and 94.1% ([Figure 4A](#fig4){ref-type="fig"}). Incorporating with CA19-9, the multi-marker panels named Com-all, remarkably elevated the AUC to 0.99 with a sensitivity of 95% and specificity of 94.1%. The combination of proteins discovered in our analysis and CA19-9 proved to be highly discriminatory between the PC and NC groups.
Biomarker validation by immunohistochemistry in the tissue samples
------------------------------------------------------------------
Biomarker profiles of a specific cancer are factors generated by the tumour itself or by the systemic response to the growing and progressing tumour. The new panel of biomarkers identified in the serum samples was further validated by assessing the expression level using immunohistochemistry. The antibody staining demonstrated that APOA1 and APOL1 expression was strong in para-carcinoma tissues. In contrast, APOE and ITIH3 expression was higher in PC tissues ([Figure 5](#fig5){ref-type="fig"}). These results were consistent with the differential expression levels of the four proteins in the serum samples.
Discussion
==========
A pipeline consisting of an extensive discovery stage followed by a timely verification and validation of altered proteins is becoming increasingly essential for the putative discovery of candidate biomarkers. However, the lack of a highly efficient verification method for the evaluation of multiple altered proteins has hindered the clinical application of candidate biomarkers identified through research ([@bib8]). Due to its multiplexing capability and antibody independence, robust and high-throughput MRM assays can be developed to verify and quantify hundreds of targeted proteins across large sample sets. The combination of large-scale proteome screens and the high-throughput MRM evaluation of interesting proteins show the potential to increase the efficiency of biomarker development.
On the basis of the MS methods mentioned above, in the current work, we developed a high-throughput and non-discriminatory pipeline for biomarker discovery, verification and validation, where each step systematically informed the next stage. The following aspects were monitored carefully to ensure the functionality of the pipeline. First, several approaches were applied for high resolution biomarker discovery. For example, most of the high-abundant proteins in the serum samples were selectively depleted using affinity columns; a long column (50 cm) was utilised to improve of the separation efficiency of the peptides; and three subsets of iTRAQ--2DLC-MS/MS experiments were analysed separately. The combination of these three approaches highly increased the number of proteins identified in the serum samples. In the discovery stage, a total of 1217 serum proteins were identified, among which 142 circulating proteins were revealed to be differentially expressed in PC compared with the controls. For cost savings and non-discriminant selection of biomarker candidates, 1D-targeted LC-MS/MS was used to confirm the peptide detection; P-MRM was conducted to relative quantification of the targeted proteins, thus bridging the gap between the high-throughput discovery stage and the large-scale targeted validation of samples. Strict quantity control was employed in the whole MRM analysis. Fragmentation properties (such as the retention time and chromatography traces of fragment ions) of the synthesised peptides were used as a constraint for the correct detection of the targeted peptides in the complex background. An exogenous peptide was added to each sample to monitor the reproducibility of the MRM runs and to normalise the results of each sample. A standard curve was constructed to ensure the high performance of the final SID-MRM assays. Thus, reliable relative quantification of the targeted proteins was achieved, which helped to prioritise the candidates for further validation. Finally, the significantly altered proteins were absolutely quantitated and evaluated in large-scale serum samples using SID-MRM. Based on the newly developed pipeline and excellent management of each step, a panel of proteins was observed and the results of the ROC analysis highlight the superiority of the newly developed 4 proteins and the combination of the new panel with CA19-9 for the diagnosis of PC.
Our results also indicate that PC is associated with circulating alterations in a number of proteins that represent a diverse set of biological families, particularly proteins with functions related to retinoid regulation ([@bib5]), inflammation and multi-molecular metabolism. For example, APOE (2.27 times higher in PC *vs* NC) is a highly abundant protein in serum and is essential for the normal catabolism of triglyceride-rich lipoprotein constituents ([@bib26]). Previous studies revealed that during tumour progression, APOE is overexpressed in ovarian carcinomas to maintain cell growth and prevent apoptosis ([@bib26]). ITIH3 (1.68 times higher in PC *vs* NC) can be found in the extracellular matrix of various organs as well as in the blood circulation. One study proposed that the ITIH family acts an important factor to stabilise hyaluronic acid on the extracellular matrix. When tumours grow, the epithelial hyaluronic complex increases in size ([@bib17]); thus, ITIH3 may play an important role in extracellular matrix remodelling during tumour progression. APOA1 (1.59 times lower in PC *vs* NC) is a key component of the reverse cholesterol transport pathway, binding to prion inflammatory phospholipids, thereby giving it anti-inflammatory properties ([@bib12]). Furthermore, APOL1 (1.34 times lower in PC *vs* NC) possesses both extra- and intra-cellular functions that are crucial in host defense and cellular homeostatic mechanisms ([@bib16]).
Although we developed a powerful and high-throughput pipeline for biomarker development, there are still some limitations. As mentioned above, not all of the candidate biomarkers discovered using 2D LC-MS/MS can be detected with 1D LC-MRM. Some efforts could be made to improve the assays developed, such as special enrichment of interested proteins with very low abundance prior to MRM analysis ([@bib3], [@bib2]; [@bib20]). Nonetheless, in terms of time and cost, this MS-intensive pipeline may still be one of the most powerful analytical approaches for biomarker discovery, with high throughput and high efficiency. The panel of these 4 proteins discovered based on this pipeline was identified to have a high predictive value with good sensitivity (85%) and specificity (94.1%),when combining with CA19-9, the sensitivity significantly increased to 95%, which outperformed CA19-9 alone for highly discriminate the PC group from the NC group. These proteins have potential value as novel predictive circulating biomarkers for PC.
We thank the funding supported by the Special Project on Precision Medicine under the National Key R&D Program (SQ2017YFSF090210), the National Key Research and Development Program of China (2017YFA0505100), National Basic Research Program of China (2013CB910802), the National High Technology Research and Development Program of China (2014AA020902), the National Natural Science Foundation of China (21675033) and the China Postdoctoral Science Foundation (2015M570324).
[Supplementary Information](#sup1){ref-type="supplementary-material"} accompanies this paper on British Journal of Cancer website (http://www.nature.com/bjc)
This work is published under the standard license to publish agreement. After 12 months the work will become freely available and the license terms will switch to a Creative Commons Attribution-NonCommercial-Share Alike 4.0 Unported License.
The authors declare no conflict of interest.
Supplementary Material {#sup1}
======================
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
{#fig1}
{#fig2}
{#fig3}
{#fig4}
{#fig5}
[^1]: These authors contributed equally to this work.
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
Laparoscopic nephrectomy (LN) is currently accepted as a standard minimally invasive procedure at many institutes. At centers with a high turnover, a considerable number of candidates for nephrectomy have a previous history of surgery. Traditionally, previous abdominal surgery has been considered a relative contraindication for laparoscopy, because of the high likelihood of access-related complications, vital organ injury, and difficulties with tissue dissection.^[@B1],[@B2]^ The procedure may be even more challenging when LN is needed at the site of previous ipsilateral kidney surgery, because dense postoperative scarring may make hilar dissection and kidney mobilization more cumbersome. In a recent prospective trial, we showed that transperitoneal LN was feasible for benign renal pathologies in the setting of previous ipsilateral open or percutaneous renal surgery.^[@B3]^ Although difficulties with tissue dissection and hilar control may result in a longer operative time, no increased risk was observed for intraoperative or postoperative complications.
In the present study, we prospectively compared the outcome of LN in patients with a previous history of open renal surgery (ORS) or percutaneous nephrolithotomy (PCNL).
PATIENTS AND METHODS
====================
Between March 2008 and December 2011, all patients with previous open or percutaneous flank surgery who required LN were enrolled in the study. Written informed consent was obtained from all patients. The benefits and risks of laparoscopic surgery and the possible need for open conversion, especially in the setting of previous surgery, were explained to each patient before surgery.
All patients had a symptomatic small or hydronephrotic poorly functioning kidney caused by chronic pyelonephritis or chronic obstructive uropathy resulting from missed ureteropelvic junction obstruction or obstructive stone disease. Functioning of the target kidney was evaluated by preoperative intravenous (IV) urogram and technetium-99m dimercaptosuccinic acid scintigraphy.
After the preoperative administration of a single IV dose of ceftriaxone and light bowel preparation, all patients underwent LN by the same surgeon (ARA).
Surgical Technique
------------------
The patient was placed in a flank position and supported by adequate padding. A 4-port transperitoneal laparoscopy was made with a 10-mm camera port at the umbilicus, two 5-mm working ports at the subcostal area and midway between the umbilicus and anterior superior iliac spine, and another 10-mm trocar lateral to the rectus muscle at the level of the umbilicus. When needed, another 5-mm port was placed on the midline at the level of the xiphoid process for liver retraction. The camera port was made with an open-access technique. After medial mobilization of the colon (and duodenum on the right side), the ureter, gonadal vein, and renal pedicle were exposed. After complete dissection of the renal vein and artery, they were double-clipped separately with 10-mm Hem-o-Lok clips (Weck Closure Systems, Research Triangle Park, NC). If the renal artery was encased in dense scars, it was double-clipped en block with surrounding scars. If severe hydronephrosis was present, exposure of the renal pedicle was facilitated by draining the collecting system percutaneously with a Chiba needle. After the renal pedicle was divided, the kidney was mobilized outside Gerota\'s fascia; if dense surrounding fibrosis was observed, Gerota\'s fascia was incised and the kidney was mobilized in this fascia. Then, the specimen was extracted through a tiny incision over the site of the previous flank incision or by extending the site of the 10-mm trocar. The site of specimen retrieval as well as the trocar sites were then closed.
Study Outcome
-------------
Patients were classified into 2 groups. Patients in group 1 had previous ORS (open retroperitoneal flank approach) and those in group 2 had previous PCNL. Demographic data as well as perioperative variables and major postoperative complications (higher than grade 1 according to Clavien\'s classification^[@B4]^) were recorded.
The outcome of LN was compared between the 2 groups by analyzing perioperative variables with Mann-Whitney\'s and Student *t* tests. SPSS version 15 was used for data analysis (SPSS, Inc., Chicago, IL).
RESULTS
=======
During the study period, 38 patients (18 \[47.4%\] men) with a previous history of ipsilateral renal surgery underwent LN. Mean age of the patients was 57.6 y (range, 15 to 77). Of these, 22 (57.9%) patients had a previous history of ORS and the other 16 (42.1%) had a previous history of ipsilateral PCNL. Both groups were age and sex matched.
In all patients, dense and loose fibrous tissue was found around the hilum and the kidney. In all but 2 patients, the hilar scars were released and the renal pedicle could be controlled safely. The other 2 patients had a history of open nephrolithotomy, and the procedure was converted to open nephrectomy to facilitate dissection of the renal pedicle. Mean operative time was longer in group 1, but the difference was not statistically significant (111 versus 97 min; *P* = .22). Major intraoperative complications occurred in 2 (5.3%) patients (1 in each group). One patient with previous ORS developed capnothorax that compromised her hemodynamics. This patient was managed by insertion of a chest tube for 48 h. In another patient in group 2, the diaphragm ruptured during dissection of the left kidney. The diaphragm was repaired laparoscopically, and no chest tube was required postoperatively.
Intraoperative blood loss and mean postoperative hematocrit drop were statistically similar in the two groups. However, the need for blood transfusion was higher in group 1 (40.9% versus 12.4%). No statistically significant differences were observed between groups with respect to postoperative variables, including time to oral intake, hospital stay, and time to ambulation. Demographic data and perioperative parameters are shown in **[Table 1](#T1){ref-type="table"}**. No major postoperative complications occurred in any patient.
######
Patient Characteristics, Intraoperative and Postoperative Data
Group 2 (LN^[a](#TF1-1){ref-type="table-fn"}^ with prior PCNL^[c](#TF1-3){ref-type="table-fn"}^) (n=16) Group 1 (LN^[a](#TF1-1){ref-type="table-fn"}^ with prior ORS^[b](#TF1-2){ref-type="table-fn"}^) (n=22) Total (n=38) *P* Value
--------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------- -------------------------------------------- -----------
Blood Transfusion (%) 2 (12.4%) 9 (40.9%) 11 (28.9%) --
Intraoperative complication (%) 1 (6.2%) 1 (4.5%) 2 (5.3%) --
Open conversion^[d](#TF1-4){ref-type="table-fn"}^ (%) 0 (0%) 2^[d](#TF1-4){ref-type="table-fn"}^ (9.1%) 2^[d](#TF1-4){ref-type="table-fn"}^ (5.3%) --
Mean Hospital Stay (days) (range) 3.1 (2--5) 3.27 (2--7) 3.1 (2--7) 0.94
Mean Time to Oral Intake (days) (range) 1 (1--1) 1.2 (1--5) 1.1 (1--5) 0.83
Mean Operative Time^[e](#TF1-5){ref-type="table-fn"}^ (range) 97 (60--150) 111 (60--180) 108 (60--180) 0.22
Mean Preop/Postop Cr^[f](#TF1-6){ref-type="table-fn"}^ 0.84/1.0 1.15/1.3 1.03/1.17 0.25/0.33
Mean ΔPostop-Preop Hct^[g](#TF1-7){ref-type="table-fn"}^ (±SD) −5.81 ±7.21 −8.35 ±6.19 −7.28 ±6.67 0.25
Indications for LN Nf small kidney^[h](#TF1-8){ref-type="table-fn"}^/Nf HN kidney^[i](#TF1-9){ref-type="table-fn"}^ 9 (36%)/7 (53.8%) 16 (64.0%)/6 (46.2%) 25 (65.8%)/13 (34.2%) 0.40
Right/Left 7/9 15/7 22/16 0.6
Male/Female 8/8 10/12 18/20 0.82
Mean age (range) (years) 44.8 (20--77) 43.8 (15--70) 57.6 (15--77) 0.87
LN=Laparoscopic nephrectomy.
ORS=Open renal surgery.
PCNL=Percutaneous nephrolithotomy.
Due to difficulties in pedicle dissection.
Calculated in minutes from insertion of the first trocar to closure of the skin.
Cr: Creatinine (md/dL).
Hct: Hematocrit.
Nonfunctioning (Nf) small size kidneys presented with intractable flank pain/recurrent urinary tract infection (UTI) or uncontrolled renovascular hypertension.
Nonfunctioning (Nf) hydronephrotic (HN) kidneys due to chronic obstructive stone disease, missed ureteropelvic junction obstruction (UPJO), secondary UPJO following previous operation or failed previous pyeloplasty and presenting with flank pain/flank mass or recurrent UTI.
DISCUSSION
==========
Laparoscopy in the setting of previous abdominal surgery needs special precautions; however, as a result of extensive research in the field of general surgery, previous abdominal surgery is no longer considered a contraindication for laparoscopic cholecystectomy.^[@B5]^ Previous abdominal surgery may be associated with intraabdominal adhesions in up to 90% of patients. The urology literature contains little evidence regarding the effect of previous surgery on the outcome of LN. Likewise, little is known regarding the role of percutaneous kidney surgery on adhesion formation and the influence of previous PCNL on the performance and outcomes of LN.
Because the use of LN has become increasingly popular at academic centers, the number of candidates for laparoscopic urologic procedures who have a previous history of surgery has increased. Parsons et al.^[@B6]^ noted that during a 6-y period, 48% of their 700 patients who underwent a laparoscopic procedure had a history of abdominal surgery, and 15% of them had had surgery for the same target kidney. These investigators showed that this history was associated with a longer operative time and hospital stay after LN, but not with a higher rate of open conversion or complications.^[@B6]^ Similar outcomes were later reported in a prospective cohort trial.^[@B3]^
The retroperitoneoscopic approach for nephrectomy or adrenalectomy was suggested by Viterbo et al.^[@B7]^ for patients with previous open abdominal surgery or irradiation to avoid intraperitoneal adhesions and limit the risk of visceral injury. Seifman et al.^[@B8]^ also showed that a retroperitoneoscopic approach for renal surgery was safer in the setting of previous open abdominal surgery. Nonetheless, this approach might be difficult in patients with previous ipsilateral ORS or PCNL. In these situations, a transperitoneal approach may be more appropriate, because the scar tissue usually lies behind the renal pedicle. Pautler et al.^[@B9]^ retrospectively compared the outcome of laparoscopic renal and adrenal procedures in patients with and without previous abdominal surgery. Despite the presence of adhesions caused by previous open surgery, LN could be accomplished without an increased risk of complications. These investigators recommended measures to prevent access-related complications, such as using an open technique to insert the first trocar and selecting the site of the first trocar as far as possible from the site of the previous incision.
PCNL, as a minimally invasive approach, might be associated with less adhesion formation than ORS is, which usually requires complete mobilization of the kidney and its hilum. However, during and after PCNL, leakage of the irrigation fluid and urine into the retroperitoneum as well as perinephric hematoma can result in future scarring. To our knowledge, few studies have compared the outcome of LN in patients with previous PCNL versus previous ORS. Recently, Turna et al.^[@B10]^ documented the feasibility of laparoscopic partial nephrectomy in patients with previous ipsilateral renal surgery. They stratified their sample (n = 25) into those with previous percutaneous surgery (n = 13) and ORS (n = 12) and found that in both groups, the operation can be challenging and should be limited to centers with appropriate experience with laparoscopic procedures. They observed no significant differences between the 2 groups regarding operative time, surgical complications, and postoperative morbidity. In our experience, and somewhat unexpectedly, we found that in patients with previous PCNL, the difficulties associated with LN were similar to those in patients with previous ORS. However, with due care, LN was feasible in a timely manner in both groups.
The present study highlights that during LN in patients with previous ipsilateral renal surgery (whether open or percutaneous), perinephric and/or perihilar adhesions can be expected. Transperitoneal laparoscopy in these patients provides excellent exposure and visualization of the renal pedicle, because the fibrotic bands are usually located posterior to the renal pedicle. Because of the obliteration of tissue planes in these settings, meticulous dissection is crucial to avoid penetration into the renal parenchyma. Meticulous technique is also important during hilar dissection. In our experience, we were usually able to dissect the renal vein circumferentially, but in some cases, in both groups, skeletonizing the renal artery was difficult and risky, so the renal artery was double-clipped with its surrounding tissue. We believe that if the hilum is encased with fibrous tissue (as in 2 patients in this series), the surgeon should not hesitate to convert the procedure to open nephrectomy to ensure safe pedicle control.
Some limitations of this study deserve mention. The relatively small sample size may mask potential differences between our 2 subgroups. Moreover, we included only patients with benign renal pathologies in this cohort. Obviously, laparoscopic surgery for cancer in the setting of previous ipsilateral open or percutaneous renal surgery would be more challenging, because precise dissection of the scar tissue would be needed to avoid penetrating Gerota\'s fascia or the renal capsule.^[@B10]^ However, a comparative study in a group of patients with laparoscopic radical nephrectomy would be worthwhile to define the extent of obliteration of the tissue planes caused by previous ORS or PCNL more precisely.
CONCLUSIONS
===========
Despite its technical challenges, transperitoneal LN in patients with a previous history of ipsilateral PCNL or ORS is a feasible, rewarding minimally invasive procedure. Given adequate laparoscopic experience, the perioperative outcome is similar for patients who previously underwent ORS or PCNL. When LN is used, the precautions that need to be considered are similar for patients with previous PCNL and those with previous open flank surgery.
[^1]: The authors thank Massoumeh Khosravi for data analysis and Karen Shashok (AuthorAID in the Eastern Mediterranean) for improving the use of English in the manuscript for this article.
[^2]: This study was supported by Shiraz University of Medical Sciences.
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .3}
{#sp2 .4}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#sec1-0271678X17732508}
============
Mild cognitive impairment (MCI) marks the transition between normal cognitive function and Alzheimer's disease (AD) dementia. However, some MCI patients remain stable throughout their entire observed clinical course, and may even revert to normal cognition.^[@bibr1-0271678X17732508]^ Discriminating between non-converters and patients with pre-dementia AD is important for patient management and for future clinical trials. However, this is not easily achieved by clinical evaluation alone.
^18^F-Fluoro-deoxyglucose positron emission tomography (^18^F-FDG-PET) may aid in this differentiation.^[@bibr2-0271678X17732508]^ ^18^F-FDG-PET provides an index of brain glucose metabolism, which reflects synaptic activity and integrity.^[@bibr3-0271678X17732508]^ AD pathology induces synaptic dysfunction in specific, connected brain regions. These downstream brain changes result in decreased ^18^F-FDG uptake in posterior temporo-parietal regions.
Patterns of altered ^18^F-FDG uptake can be extracted with the Scaled Subprofile Model and Principal Component Analysis (SSM/PCA), a well-studied multivariate method.^[@bibr4-0271678X17732508]^ Disease-related patterns identified by SSM/PCA not only delineate the disease topography, but can also be used to quantify new ^18^F-FDG-PET scans. For quantification, normalized scans are projected onto a previously identified disease pattern to calculate a subject score. The subject score reflects the degree to which the pattern is present in a subject's scan. This method was previously shown to be successful in the differential diagnosis of Parkinsonian disorders,^[@bibr5-0271678X17732508],[@bibr6-0271678X17732508]^ in predicting disease onset in prodromal individuals with Parkinson's disease,^[@bibr7-0271678X17732508]^ and in evaluating disease progression and treatment effects.^[@bibr8-0271678X17732508][@bibr9-0271678X17732508]--[@bibr10-0271678X17732508]^
In a previous study, we identified an AD-related pattern (ADRP) in 15 AD patients and 18 controls studied at the University Medical Centre Groningen using SSM/PCA. The ADRP was characterized by relative hypometabolism of the posterior temporo-parietal cortical areas ([Figure 1](#fig1-0271678X17732508){ref-type="fig"}), and was expressed in new patients with AD, but not in healthy elderly. In addition, ADRP subject scores correlated significantly with neuropsychological test performance.^[@bibr11-0271678X17732508]^ The ADRP has also been identified by two other groups.^[@bibr12-0271678X17732508],[@bibr13-0271678X17732508]^ It has been shown that the ADRP is superior at identifying AD patients compared to univariate, region-of-interest approaches.^[@bibr12-0271678X17732508]^ Further validation of our ADRP^[@bibr11-0271678X17732508]^ in a larger cohort is necessary. Moreover, it is unknown whether ADRP subject scores can discriminate between non-AD and AD at the MCI stage. Figure 1.Topography of the ADRP. The ADRP was identified in ^18^F-FDG-PET data (anatomically registered to an ^18^F-FDG-PET template) as described previously.^[@bibr11-0271678X17732508]^ Stable voxels in the ADRP were overlaid onto a T1 MRI template to show the most salient regions in the pattern. Stable voxels in the ADRP were determined with bootstrap resampling.^[@bibr12-0271678X17732508]^ In this procedure, the pattern identification process (SSM/PCA) is repeated multiple times on randomly sampled data with replacement. This yields multiple slightly different patterns and thus a distribution of weights per voxel. Using this distribution, confidence intervals (CIs) per voxel can be determined. Voxels with CIs straddling zero can be interpreted as non-informative and are therefore excluded from the visualization. Here, we performed 1000 repetitions and applied a one-sided CI threshold of 90% (percentile method). For a discussion of pattern topography, we refer to a previous publication.^[@bibr11-0271678X17732508]^ L = Left. Relatively hypermetabolic areas are color-coded red, and relatively hypometabolic areas are color-coded blue. ADRP: Alzheimer's disease-related metabolic brain pattern.
In this study, we further validate the ADRP^[@bibr11-0271678X17732508]^ by computing its expression in a large cohort of healthy controls, MCI patients with long-term clinical follow-up, and AD patients. Our main objective was to determine whether ADRP subject scores could discriminate between MCI-converters and MCI non-converters at baseline.
Material and methods {#sec2-0271678X17732508}
====================
Participants {#sec3-0271678X17732508}
------------
^18^F-FDG-PET data from healthy aged subjects (NA; *n* = 42), patients with MCI (*n* = 122), and patients with AD dementia at the time of the PET scan (*n* = 55) were analysed from a previous study (Supplementary Table 1).^[@bibr14-0271678X17732508],[@bibr15-0271678X17732508]^ Patients with MCI (*n* = 122) were separated into three groups: patients who did not progress during follow-up (non-converter MCI; ncMCI, *n* = 27), patients who progressed to AD after ≥ 2 years of follow-up (early MCI; eMCI; *n* = 34), and MCI patients who progressed to AD within two years of follow-up (late MCI, lMCI; *n* = 61). The study was approved by the institutional review board of the University of Genoa, and all subjects gave written informed consent to undergo ^18^F-FDG-PET in the framework of a long-term observational study, in accordance with the Declaration of Helsinki.
^18^F-FDG-PET data analysis {#sec4-0271678X17732508}
---------------------------
^18^F-FDG-PET data were acquired and pre-processed as described previously.^[@bibr14-0271678X17732508]^ We calculated ADRP subject scores in all ^18^F-FDG-PET scans as follows: First, ^18^F-FDG-PET images were masked to remove out-of-brain voxels. Next, each image was log-transformed and both the subject mean and reference group mean were removed. The reference group mean was previously determined in the ADRP identification cohort.^[@bibr11-0271678X17732508]^ These operations resulted in a subject residual profile (SRP) for each scan. Finally, a subject score (SS) for each subject was computed based on the ADRP by taking the inner product of the two vectors, the ADRP and SRP (SS = SRP × ADRP). For further details, we refer to an excellent overview of the method by Spetsieris and Eidelberg.^[@bibr4-0271678X17732508]^
The ^18^F-FDG-PET data investigated in this study were acquired on a different PET system than the cohort which was originally used for ADRP identification.^[@bibr11-0271678X17732508]^ To account for the effect of camera differences on ADRP expression, ADRP subject scores were z-transformed to the NA group, such that the NA mean was 0 with a standard deviation of 1.
In the ADRP identification cohort,^[@bibr11-0271678X17732508]^ we determined the threshold ADRP z-score with optimum sensitivity and specificity using a receiver-operating curve. This threshold was determined to be z = 0.8 and subsequently applied to the new ADRP z-scores in this study. ADRP z-scores ≥ 0.8 were considered to be indicative of AD. A visual representation of the ADRP^[@bibr11-0271678X17732508]^ that was used in this study is provided in [Figure 1](#fig1-0271678X17732508){ref-type="fig"}.
Statistical analysis {#sec5-0271678X17732508}
--------------------
ADRP subject z-scores were compared across NA, ncMCI, eMCI, lMCI, and AD subjects with a one-way analysis of variance (ANOVA) with post hoc Bonferroni corrections. With the threshold of z = 0.8, we identified the number and percentage of subjects correctly classified in each group. Sensitivity, specificity, accuracy, and area under the receiver-operating characteristic curve (AUC-ROC) were determined for the comparisons: NA versus AD patients, NA versus MCI-converters + AD patients, NA versus MCI-converters alone, ncMCI versus MCI-converters + AD patients, and finally, ncMCI versus MCI-converters alone.
Pearson's R correlation coefficient was used to test the correlation between the parametric variables of ADRP z-scores and age. Correlations between ADRP z-scores and non-parametric variables (MMSE corrected for educational level and age, time to conversion, and education) were tested for significance with a Spearman rank correlation coefficient. All analyses were performed using SPSS software version 23 (SPSS Inc., Chicago, IL), and results were considered significant when *P* \< 0.05 (two-tailed).
Results {#sec6-0271678X17732508}
=======
ADRP z-scores were significantly different between groups (F = 36.33, *P* \< 0.0001). ADRP z-scores were significantly higher in MCI-converters and AD patients compared with both NA and ncMCI patients ([Figure 2](#fig2-0271678X17732508){ref-type="fig"}). [Table 1](#table1-0271678X17732508){ref-type="table"} shows the sensitivity, specificity, accuracy, and AUC-ROC for the different group comparisons. Specificity in the ncMCI category was limited (66.66%). Figure 2.ADRP z-scores across groups. All ADRP subject scores were z-transformed to NA. Group differences were tested for significance with a one-way ANOVA; post hoc comparisons were Bonferroni-corrected. AD: Alzheimer\'s disease; ADRP: Alzheimer's disease-related metabolic brain pattern; MCI: mild cognitive impairment; NA: normal ageing; ncMCI: non-converting MCI; eMCI: early MCI; lMCI: late MCI. Table 1.Diagnostic performance of the ADRP.NA versus AD dementiaNA versus MCI-converters + AD dementiaNA versus MCI-convertersncMCI versus MCI-converters + AD dementiancMCI versus MCI-convertersSensitivity90.9086.6684.2184.6184.21Specificity85.5785.5785.7166.6666.66Accuracy0.8786.460.8481.9780.33AUC-ROC curve0.950.910.890.840.80[^2]
Supplementary Table 1 shows the percentage of correctly classified subjects per category based on ADRP z-scores. Nine out of 27 ncMCI had a supra-threshold ADRP z-score. Follow-up time in these nine individuals ranged from a minimum of 6.8 to a maximum of 9.8 years.
ADRP z-scores were not significantly correlated to time-to-conversion in MCI-converters (ρ = −0.05, *P* = 0.66). A significant relationship was observed between ADRP z-scores and MMSE (MCI-converters + AD patients; ρ = −0.341; *P* \< 0.0001; Supplementary Figure 1).
In the ncMCI group, ADRP z-scores appeared to be higher in patients with a higher education (Supplementary Figure 2(a)), with borderline statistical significance (ρ = 0.375, *P* = 0.054). Furthermore, a significant relationship between age and ADRP z-scores was only present in the ncMCI group (*r* = 0.502, *P* = 0.008; Supplementary Figure 2(b)). ADRP z-scores did not correlate significantly to age or education in the other groups.
Discussion {#sec7-0271678X17732508}
==========
We studied expression of the ADRP in baseline ^18^F-FDG-PET scans of a large cohort of MCI patients with clinical follow-up. ADRP subject z-scores were significantly higher in MCI patients who progressed to AD dementia compared with both healthy elderly and non-converting MCI patients. In line with a previous study, ADRP z-scores were significantly correlated to disease severity in AD (measured by the MMSE score).^[@bibr11-0271678X17732508]^ Compared to healthy elderly, we found good sensitivity (84.2%) and specificity (85.7%) of the ADRP for the detection of early brain dysfunction in AD (i.e. NA versus MCI-converters).
Specificity of the ADRP was limited in the non-converting MCI group (66.7%), as one-third of ncMCI patients (9/27) had a supra-threshold ADRP z-score. Even though clinical follow-up in these patients was long (6.8--9.8 years), such cases cannot simply be interpreted as false-positives. Some may still develop AD dementia on further follow-up. To illustrate, one late-converting MCI patient had a baseline ADRP z-score of 3.11 and only developed clinical AD eight years later.
It could also be hypothesized that non-converting MCI patients with a supra-threshold ADRP score have a larger cognitive reserve. Such individuals are able to maintain a certain level of cognitive functioning, despite having temporo-parietal hypo-metabolism.^[@bibr16-0271678X17732508]^ Individuals with a higher education have a larger cognitive reserve, and are thought to have the ability to recruit compensatory networks involving the dorso-lateral prefrontal cortex.^[@bibr17-0271678X17732508]^ In line with this, we found higher ADRP z-scores in MCI non-converters with a higher education. The non-linear relationship between hypo-metabolism and the clinical manifestation of the disease may also explain why ADRP z-scores did not correlate significantly to time-to-conversion in the pre-dementia AD group.
An alternative explanation for the limited specificity in the ncMCI group may be that the ADRP reflects the underlying abnormalities in neuronal networks in AD, but is not pathology-specific.^[@bibr2-0271678X17732508]^ MCI is a common manifestation in many conditions.^[@bibr18-0271678X17732508]^ The finding that ADRP z-scores were only correlated to age in the ncMCI group may indicate that ncMCI patients with a supra-threshold ADRP score have a non-AD pathology which progresses with age, and affects cortical areas which partially overlap with the ADRP topography. Although patients who met the criteria for vascular cognitive impairment were excluded,^[@bibr14-0271678X17732508]^ mild cerebrovascular disease in combination with other factors (drug therapy, chronic disease, and depression) may have resulted in ADRP-like metabolic changes in the ncMCI group.
While our main objective was to examine whether ADRP scores could adequately differentiate between MCI-converters and non-converters, a second objective of this study was to validate the ADRP which was previously identified in a cohort of AD patients and controls in Groningen.^[@bibr11-0271678X17732508]^ We successfully applied the ADRP to a completely independent dataset. We note that other multivariate approaches were also applied to the data presented in this study,^[@bibr14-0271678X17732508],[@bibr15-0271678X17732508]^ and gave similar sensitivity and specificity for AD. An important advantage of the current approach is that the ADRP could be applied to new subjects on a single case-by-case basis, despite these subjects having been scanned on a different PET system.
Along with anatomical brain imaging, amyloid PET, and cerebrospinal fluid analysis, expert visual reading of ^18^F-FDG-PET scans is an accepted ancillary investigation in the diagnostic work-up of cognitive decline.^[@bibr2-0271678X17732508],[@bibr19-0271678X17732508]^ A single-case analysis in which separate brain regions are identified where ^18^F-FDG uptake levels deviate from normal can also be achieved with univariate, semi-quantitative SPM-based methods.^[@bibr20-0271678X17732508]^ A limitation of semi-quantitative methods is that it is difficult to objectively quantify progression of metabolic changes. The ADRP may be especially useful to measure disease progression, and may thus provide important complementary information to a semi-quantitative visual reading of ^18^F-FDG-PET in clinical practice.
Supplementary Material
======================
###### Supplementary material
We thank Rosalie V Kogan for proofreading the manuscript. We thank the Dutch 'Stichting ParkinsonFonds' for financial support.
Funding {#sec8-0271678X17732508}
=======
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests {#sec9-0271678X17732508}
====================================
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Authors' contributions {#sec10-0271678X17732508}
======================
SKM designed the study, analysed the data, and drafted the manuscript; MP designed the study, acquired the data, and revised the manuscript critically for important intellectual content; DA designed the study, acquired the data, and revised the manuscript critically for important intellectual content; FDC designed the study and acquired the data; BD designed the study and acquired the data; SM designed the study and acquired the data; GS designed the study and acquired the data; CJ designed the study and acquired the data; KLL supervised interpretation of data and revised the manuscript critically for important intellectual content; FN designed the study, supervised interpretation of data, and revised the manuscript critically for important intellectual content.
Supplementary material {#sec11-0271678X17732508}
======================
Supplementary material for this paper can be found at the journal website: <http://journals.sagepub.com/home/jcb>
[^1]: Shared last authors.
[^2]: MCI: mild cognitive impairment; AD: Alzheimer's disease; ADRP: Alzheimer's disease-related metabolic brain pattern; ncMCI: non-converting MCI.
| {
"pile_set_name": "PubMed Central"
} |
{.55}
{.56}
{.57}
{.58}
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Cardiac fibrosis, a hallmark of most cardiomyopathies, is characterized by excessive extracellular matrix accumulation contributing to the destruction of normal tissue architecture and progressive organ dysfunction [@B1], [@B2]. Cardiac fibrosis is a strong driver of adverse ventricular remodeling and heart failure that occurs after a variety of different cardiac injuries, such as myocardial infarction (MI) and hemodynamic stress as seen in hypertrophic and dilated cardiomyopathies [@B3], [@B4]. Although acetyl choline esterase (ACE) inhibition, angiotensin II receptor antagonists, and recently LCZ696 (an angiotensin II type 1 receptor-neprilysin inhibitor) can partially reverse remodeling, no effective anti-fibrotic therapeutic strategies are currently available [@B1], [@B5], [@B6]. The lack of an effective therapy for cardiac fibrosis and cardiac remodeling is in part responsible for the morbidity, mortality, and healthcare expenditure attributable to heart failure [@B2], [@B5]. Therefore, novel anti-fibrotic strategies represent a critical unmet clinical need [@B2], [@B5].
MicroRNAs (miRNAs, miRs) are small noncoding RNAs, which repress gene expression by degradation or translational inhibition of target mRNAs [@B7]. A single mRNA can be regulated by multiple miRNAs, while individual miRNAs are capable of regulating tens to hundreds of distinct target genes [@B7], [@B8]. As approximately 60% of protein-coding genes are regulated by miRNAs, they have emerged as powerful regulators for almost all essential biological processes including cellular proliferation, differentiation, apoptosis, development, and metabolism [@B9], [@B10]. Emerging data have suggested that aberrant expression of miRNAs could lead to a profound disturbance of target gene network and signaling cascades that participate in many pathological phenotypes. One such example is of adverse cardiac remodeling and fibrosis [@B1], [@B11], [@B12]. Increased pro-fibrotic miRNAs such as miR-21, 22, and 34a and decreased anti-fibrotic miRNAs such as miR-24, 15 family, 26a, and 29b have been reported to contribute to cardiac fibrosis [@B13]-[@B20]. These observations indicate that manipulation of miRNAs may serve as a novel potential therapeutic approach to combat cardiac fibrosis. An unexplored candidate located on chromosome 12, miR-433, has been reported to be up-regulated in renal fibrosis and liver fibrosis [@B21], [@B22]. However, the role of miR-433 in the heart and especially in cardiac fibrosis is unclear.
In the present study, based on miRNA arrays, we noted that miR-433 was significantly increased in ventricle samples at 21-days following MI in mice. We further validated up-regulation of miR-433 in a rodent model of doxorubicin-induced cardiomyopathy and human dilated cardiomyopathy (DCM). Also, over-expression of miR-433 increased the proliferation of cardiac fibroblasts and promoted their differentiation into myofibroblasts, whereas knockdown of miR-433 suppressed these responses upon transforming growth factor-β (TGF-β) or Angiotensin II (Ang II) stimulation. Our work further identified AZIN1 and JNK1 as two target genes of miR-433. Importantly, treatment with miR-433 antagomir or adeno-associated virus 9 (AAV9)-mediated cardiac transfer of a miR-433 sponge improved post-MI cardiac function and attenuated cardiac fibrosis in adult mice. Collectively, our findings indicate that miR-433 promotes cardiac fibrosis and therefore inhibition of miR-433 might be useful for the treatment of cardiac fibrosis.
Materials and Methods
=====================
Ethics Statement
----------------
All animals were raised at the Experimental Animal Center of Nanjing Medical University (Nanjing, China) or Shanghai University (Shanghai, China). All procedures with animals were in accordance with the guidelines on the use and care of laboratory animals for biomedical research published by National Institutes of Health (No. 85-23, revised 1996). The experimental protocol was reviewed and approved by the ethical committees of Nanjing Medical University and Shanghai University. All human investigations conformed to the principles outlined in the Declaration of Helsinki and was approved by the institutional review committees of Nanjing Medical University. All participants gave written informed consent before enrollment in the study. Human left ventricular tissue samples were obtained from 4 patients with dilated cardiomyopathy (DCM) undergoing cardiac transplantation and 4 healthy donors (The First Affiliated Hospital of Nanjing Medical University).
Isolation of Cardiac Fibroblasts, Culture, and Transfection
-----------------------------------------------------------
Cardiac fibroblasts were isolated from 1 to 3-day-old SD rats. Ventricles were finely minced and digested in trypsin buffer (60% trypsin and 40% collagenase). Cell suspensions were centrifuged, resuspended in DMEM (Gibco, Grand Island, CA, USA) with 10% fetal bovine serum (FBS), 100 U/ml penicillin and 100 μg/ml streptomycin, and plated for 2 h under standard culture conditions (37°C in 5% CO~2~ and 95% O~2~) which allowed fibroblast attachment to the culture plates.
All transfections and assays on cardiac fibroblasts were conducted in low serum medium (1% FBS). Cardiac fibroblasts at passage 2 were exposed to either miRNA agomir *versus* negative control (100 nM), or antagomir *versus* negative control (200 nM) (RiboBio, Guangzhou, China) for 48 h, and treated with 10 ng/ml recombinant human TGF-β1 for 24 h (Peprotech, Rocky Hill, NJ, USA) or 100 nM Ang II for 48 h (Sigma, St. Louis, MO, USA), respectively. siRNAs for AZIN1, JNK1, and negative controls were purchased from Invitrogen Carlsbad, CA. Plasmids over-expressing AZIN1 or JNK1 were purchased from Sangon Biotech, Shanghai, China. Transfections with siRNAs (50 nM) or plasmids (50 nM) for 48 h were carried out using Lipofectamine RNAiMAX Transfection Reagent (Invitrogen). p38 MAP kinase inhibitor SB202190 (Sigma, 10 μM, 1 h), ERK inhibitor U0126 (Sigma, 10 μM, 1 h), and Smad3 inhibitor SIS3 (Millipore, 1 μM, 48 h) were used to treat cells in the presence or absence of miR-433 agomir.
Animal Models
-------------
Eight-week-old male C57BL/6 mice were used in this study. MI was generated by ligating the left anterior descending coronary artery (LAD) using a 7/0 silk thread while sham was created by the same process but without LAD ligation. Doxorubicin-induced cardiomyopathy mouse model was induced by chronically treating mice with either doxorubicin or phosphate-buffered saline (PBS) by four intraperitoneal (i.p) injections (day 0, 2, 4 and 6) at a dose of 4 mg/kg. All mice were sacrificed after 4 weeks.
To determine if inhibition of miR-433 is sufficient to prevent cardiac fibrosis *in vivo*, mice were injected via tail vein with 80 mg/kg antagomir (a 2\'OME+ 5\'chol modified miR-433 inhibitor) or the scramble control (Ribobio, Guangzhou, China) for 3 consecutive days and subjected to LAD ligation. AAV represents an efficient and safe vector for *in vivo* gene transfer and serotype 9 is significantly cardiotropic [@B23]-[@B26]. Thus, besides miR-433 antagomir, the cardiotropic miR-433 sponge AAV9 was used to determine further if cardiac inhibition of miR-433 is sufficient to prevent fibrosis *in vivo.* In brief, mice were randomly chosen to receive a single-bolus tail vein injection of either miR-433 sponge AAV9 or miR-scramble (Hanheng Biotechnology, Shanghai, China) at 1\*10^11^ vg (viral genomes) per animal. After 1 week, mice were subjected to LAD ligation and finally sacrificed at 3 weeks post-MI.
miRNA Array and Gene-Chip Analysis
----------------------------------
Total RNA extracted from ventricular tissues 21 days post-MI or sham control was used for miRNA arrays based on Affymetrix 4.0 (OE Biotech\'s, Shanghai, China). Additionally, total RNA extracted from ventricular tissues 21 days post-MI injected with miR-433 antagomir or scramble control was used for gene-chip analysis based on Agilent SurePrint G3 Mouse GE (8\*60K, Design ID: 028005) Microarray (OE Biotech\'s, Shanghai, China). The MIAME compliant data have been submitted to Gene Expression Omnibus (GEO, platform ID: GSE74135 for miRNA array and GSE74206 for gene-chip analysis, respectively).
Quantitative Real-time Polymerase Chain Reactions (qRT-PCRs)
------------------------------------------------------------
Total RNAs were extracted from cardiac fibroblasts and heart samples by using miRNeasy Mini Kit (Qiagen, Hilden, Germany) according to manufacturer\'s instructions. Total RNAs (400 ng) were reverse transcribed using Bio-Rad iScript^TM^ cDNA Synthesis Kit (Bio-Rad, Hercules, CA, USA) to obtain cDNAs. The expression levels of TGF-β, α-SMA, Col1a1, and Col3a1 were analyzed by using Bio-Rad SYBR qPCR (Bio-Rad, Hercules, CA, USA) on ABI-7900 Real-Time PCR Detection System (7900HT, Applied Biosystems, CA, USA). 18S RNA was used as an internal control for gene expressions. Primer sequences used in the study are listed in Supplemental Table [1](#SM1){ref-type="supplementary-material"}. For quantitative miRNA analysis, the Bulge-Loop^TM^ miRNA qPCR Primer Set (RiboBio) was used to determine the expression levels of miRNAs with Takara SYBR Premix Ex Taq^TM^ (Tli RNaseH Plus) on ABI-7900 Real-Time PCR Detection System (Applied Biosystems). U6 was used as an internal control for miRNA template normalization.
Pharmacokinetics of miRNA
-------------------------
miR-433 antagomir or the scramble control (Ribobio, Guangzhou, China) was prepared in PBS and administered via tail vein at a dose of 7.5 mg/kg for each mice. Subsequently, mice were sacrificed and plasma and heart tissues were collected immediately at different time points of 5, 10, 15, 30, 60, 120, 240, 480, 1320 and 1440 minutes after injection (n=5 per group for each time point) [@B27]. miR-433 expression levels in plasma and heart samples were determined using qRT-PCRs as described above.
Immunofluorescence and EdU Staining
-----------------------------------
Cardiac fibroblasts were fixed in 4% paraformaldehyde (PFA) for 20 min at room temperature. Cells were then permeabilized with 0.2% Triton X-100 for 20 min and blocked with 10% goat serum in PBS-Tween for 1 h at room temperature. Subsequently, cardiac fibroblasts were incubated with α-SMA-Cy3 antibody (1:500, Sigma, St. Louis, MO, USA) diluted in 10% goat serum overnight at 4°C. To detect proliferation, EdU assays were performed using Click-iT Plus EdU Alexa Fluor 488 Imaging Kit (Invitrogen) according to manufacturer\'s instructions. Cell nuclei were counterstained with DAPI and the number of EdU-positive nuclei was calculated. Fifteen fields/sample (200 x magnification) were viewed under a confocal microscope (Carl Zeiss, Thuringia, Germany).
Sections of heart samples were cut at a thickness of 5-6 μm. Subsequently, the sections were fixed in 4% PFA for 20 min at room temperature, permeabilized with 0.2% Triton X-100 for 20 min, and then blocked with 10% goat serum in PBS-Tween for 1 h at room temperature. Next, the sections were incubated with diluted primary antibodies at 4°C overnight; the following antibodies were used: α-SMA-Cy3 antibody (1:500, Sigma), Vimentin antibody (1:100, Abcam), Ki67 antibody (1:100, Abcam), and pHH3 antibody (1:100, Abcam). After three washes with PBS for 5 min each, the sections were incubated with secondary antibodies or other dyes at room temperature for 2 h. Fifteen fields/sample (400x magnification) were viewed under a confocal microscope (Carl Zeiss).
Western Blotting Analysis
-------------------------
Cardiac fibroblasts and heart samples were lysed using RIPA buffer (Beyotime Institute of Biotechnology, Nantong, China), which contained a protease inhibitor cocktail (Sigma). The concentration of protein samples was evaluated by Bicinchoninic Acid Protein Assay Kit (Thermo Fisher, Waltham, MA, USA). Equal amounts of protein were separated in SDS-PAGE and blotted onto PVDF membranes. The primary antibodies used were from the following sources: α-SMA (1:1000, Sigma), TGF-β (1:1000, Cell Signaling Technology, Boston, MA, USA), p38 (1:1000, CST), p-p38 (1:1000, CST), ERK (1:1000, CST), p-ERK (1:1000, CST), p-Smad3 (1:1000, CST), Smad3 (1:1000, CST), JNK1 (1:1000, CST), AZIN1 (1:500, Proteintech, Wuhan, China), CTGF (1:500, Proteintech, Wuhan, China), Col1a1 (1:500, Proteintech), Col3a1 (1:500, Proteintech), MMP2 (1:500, Proteintech), MMP9 (1:500, Proteintech) and GAPDH (1:10000, Kangchen, Shanghai, China). All proteins were visualized by ECL Chemiluminescence Kit (Thermo Fisher) and the quantification of each band was performed using Imagelab Software (Bio-Rad) with GAPDH as a loading control.
Luciferase Reporter Assay
-------------------------
A fragment of the 3\'UTRs of AZIN1 or JNK1 containing the target site of miR-433 was obtained by PCR amplification and then cloned into the pGL3-Basic Vector (Promega, Madison, WI, USA) to generate the AZIN1 or JNK1 wt-luc vector. The AZIN1 or JNK1 mutant-luc vector was generated by using the MutaBest kit (Takara, Tokyo, Japan). Forty-eight hours after transfection, luciferase activities were measured using a dual luciferase reporter assay system (Promega) following a standard procedure.
Echocardiography
----------------
Three weeks after the injection of miR-433 antagomir, mice were anesthetized with 1.5-2% isoflurane and then evaluated by Vevo 2100 echocardiography (VisualSonics Inc, Toronto, Ontario, Canada) with a 30 MHz central frequency scan head to detect cardiac function. The following parameters were measured from M-mode images taken from the parasternal short-axis view at papillary muscle level: left ventricular fractional shortening (FS) and left ventricular ejection fraction (EF). The left ventricle internal diameter (LVID), interventricular septum (IVS), and left ventricle posterior wall (LVPW) in diastole or systole were also measured. At least three measurements were obtained and averaged for each mouse.
TTC staining
------------
At 3 days\' post LAD ligation, mice were anesthetized with intraperitoneal injection of 0.5 mg/g tribromoethanol. Subsequently, 1 ml Evans blue (BioSharp, Anhui, China) was slowly injected into inferior vena and the heart was removed immediately. After storage for 15 minutes at -20°C, the heart was cut into 5 transverse slices at 1 mm thickness across the long axis. The slices were then stained with 1% triphenyltetrazolium chloride (TTC, Amresco, OHIO, USA) in PBS for 10 min at 37°C following which the slices were fixed with 4% PFA and analyzed. The final infarct size was calculated by Image J Software (National Institutes of Health).
Masson\'s Trichrome Staining
----------------------------
Heart samples were fixed in 4% PFA and then embedded in paraffin. Five μm-thick sections were subjected to Masson\'s trichrome staining following a standard procedure. Images of the left ventricular area of each section were taken by Nikon model (200x magnification) with Spot Insight camera. Image J Software (National Institutes of Health) was used to quantify fibrotic region in each section. The percentage of fibrosis was measured as fibrosis areas/total left ventricular areas x 100%.
Collagen content assay
----------------------
A quantitative dye-binding method was used to determine the collagen content. Analysis of heart tissues was performed using Sircol assay (Biocolor, Carrickfergus, UK) according to manufacturer\'s instructions. In this assay, each heart sample was weighed and homogenized with pepsin. The BioTek Software (Hercules, CA, USA) was used to quantify collagen content in each sample.
Statistical Analysis
--------------------
Data were presented as mean ± SE. A Student\'s t-test, Chi-squares test or one-way ANOVA followed by Bonferroni\'s post-hoc test was used to compare the one-way layout data when appropriate. *P* values less than 0.05 were considered to be statistically different. All analyses were performed using GraphPad Prism 5.
Results
=======
miR-433 is Increased in Cardiac Fibrosis
----------------------------------------
miRNA arrays were used to determine aberrant expressions of miRNAs, which might contribute to cardiac fibrosis in the post-MI ventricle at a time point notable for prominent fibrosis. A total of 26 miRNAs were found to be dysregulated (Fold change \>2.0; *P*\<0.05; Figure [1](#F1){ref-type="fig"}A and Supplemental Table [2](#SM1){ref-type="supplementary-material"}). Interestingly, the top 3 dysregulated miRNAs including miR-34b-3p, 34c-5p, and 34c-3p belong to the miR-34 family, whose inhibition has been shown to attenuate pathological cardiac remodeling [@B13]. Since miR-433 (number fourth) has previously been reported to participate in kidney and liver fibrosis [@B21], [@B22] but has not so far been explored in the myocardium and during cardiac fibrosis, we explored its function further.
Based on the qRT-PCR analysis, we confirmed that miR-433 was upregulated in heart samples with fibrosis from mice 3 weeks post-MI (Figure [1](#F1){ref-type="fig"}B). To exclude the possibility that increased miR-433 is specific to cardiac fibrosis post-MI, we also determined its expression in doxorubicin-induced cardiomyopathy rodent model and in human dilated cardiomyopathy (DCM) (Figure [1](#F1){ref-type="fig"}B). The clinical information and echocardiography parameters for DCM patients are presented in Supplemental Table [3](#SM1){ref-type="supplementary-material"}. The DCM sample size is small due to the difficulty of acquiring human heart tissues. Interestingly, miR-433 was consistently upregulated in all three models, i.e., in heart tissues with fibrosis, in doxorubicin-induced cardiomyopathy, and in patients with DCM (Figure [1](#F1){ref-type="fig"}B). Thus, there appeared to be a strong correlation between the presence of cardiac fibrosis and an increase in miR-433 expression in several different cardiac diseases. Furthermore, miR-433 was also increased in cultured neonatal rat cardiac fibrosis models stimulated by TGF-β or Ang II (Figure [1](#F1){ref-type="fig"}C-D). Taken together, these data supported a potential role for miR-433 in cardiac fibrosis.
*In vivo* Inhibition of miR-433 Preserves Cardiac Function and Prevents Fibrosis
--------------------------------------------------------------------------------
Next, we determined the relative expression level of miR-433 in isolated neonatal rat cardiac fibroblasts *versus* cardiomyocytes, and demonstrated higher expression level in fibroblasts compared to cardiomyocytes (Figure [1](#F1){ref-type="fig"}E). Forced expression of miR-433 in cardiomyocytes did not lead to an elevation of markers for pathological hypertrophy (ANP, BNP, and Myh7) or extracellular matrix proteins (CTGF, TSP-1, Col1a1 and Col3a1) (Figure [1](#F1){ref-type="fig"}F) supporting a more prominent role for miR-433 in fibroblasts rather than cardiomyocytes.
To evaluate the effect of miR-433 inhibition on cardiac fibrosis, we administrated miR-433 antagomir in mice via tail vein to downregulate miR-433 *in vivo*. First, the pharmacokinetic analysis for miR-433 antagomir was performed by measuring miR-433 expression level in both plasma and heart samples at different time points after mice were administrated with a single bolus of miR-433 antagomir at the dose of 7.5 mg/kg as previously reported [@B27]. The pharmacokinetic analysis showed that miR-433 was significantly downregulated in plasma and heart samples at 10 min post injection maintaining the low expression level thereafter (Supplemental Figure [1](#SM1){ref-type="supplementary-material"}). Next, to explore whether antagonizing miR-433 attenuates cardiac fibrosis and preserves ventricular function post-MI, we treated mice with miR-433 antagomir or scrambled negative control via tail vein injection for 3 consecutive days and subjected them to MI or sham surgery. Then mice were sacrificed 3 weeks after MI and the loss of miR-433 in the heart was confirmed by qRT-PCRs (Figure [2](#F2){ref-type="fig"}A). Echocardiography showed that miR-433 antagomir preserved cardiac function including FS and EF (Figure [2](#F2){ref-type="fig"}B), and also reversed MI-induced increase in systolic left ventricle internal diameter (LVID;s) and diastolic left ventricle internal diameter (LVID;d) as shown in Supplemental Table [4](#SM1){ref-type="supplementary-material"}. Importantly, inhibition of miR-433 also attenuated cardiac fibrosis as evidenced by reduced collagen deposition and content in MI heart tissues (Figure [2](#F2){ref-type="fig"}C-D). In particular, we evaluated the effect of miR-433 inhibition on cardiac infarction 3 days after MI; the purpose was to determine whether miR-433 inhibition predominantly protects against cardiac fibrosis in the remodeling phase after MI or prevents cardiac infarction in the acute phase after MI. Based on TTC staining, there was no difference in the infarct size between mice treated with miR-433 antagomir or negative control, strongly suggesting that miR-433 inhibition predominantly protects against cardiac fibrosis in the remodeling phase after MI (Figure [2](#F2){ref-type="fig"}E).
To further confirm the effect of miR-433 inhibition in preventing cardiac fibrosis, we used a cardiotropic AAV9 delivery system to achieve cardiac inhibition of miR-433 *in vivo*. Mice received a single-bolus tail vein injection of either miR-433 sponge AAV9 or miR-scramble. After 1 week, mice were subjected to LAD ligation and sacrificed at 3 weeks post-MI. Using qRT-PCR, we confirmed that miR-433 sponge AAV9 efficiently reduced miR-433 expression level in heart tissues (Figure [3](#F3){ref-type="fig"}A). Furthermore, our data showed that AAV9-mediated inhibition of miR-433 could significantly preserve left ventricular EF and FS (Figure [3](#F3){ref-type="fig"}B), and reduce increased systolic LVID and diastolic LVID in mice 3 weeks post-MI (Supplemental Table [5](#SM1){ref-type="supplementary-material"}). Cardiac inhibition of miR-433 also reduced collagen deposition and collagen content in hearts post-MI (Figure [3](#F3){ref-type="fig"}C-D). These data provide strong evidence that inhibition of miR-433 has cardioprotective effect against fibrosis.
Inhibition of miR-433 Attenuates Cardiac Fibroblast Proliferation and Myofibroblast Differentiation *In Vivo* and *In Vitro*
----------------------------------------------------------------------------------------------------------------------------
The transformation of fibroblasts into myofibroblasts is a critical event in the genesis of cardiac fibrosis [@B28], [@B29]. We determined the effects of miR-433 inhibition on cardiac fibroblasts proliferation and their differentiation into myofibroblasts in both post-MI mice and cultured cardiac fibroblasts. Based on the heart samples from *in vivo* experiments, immunofluorescence analysis revealed that antagonizing miR-433 decreased cardiac fibroblast proliferation as evidenced by reduced Ki-67/Vimentin or phospho-HistoneH3 (pHH3)/Vimentin double positive cells (Figure [4](#F4){ref-type="fig"}A-B). Furthermore, miR-433 inhibition also attenuated the differentiation of cardiac fibroblasts into myofibroblasts as shown by decreased number of cells double-positive for α-SMA and Vimentin (Figure [4](#F4){ref-type="fig"}C). Consistent with this, the expression levels of α-SMA, Col1a1, and Col3a1 in the ventricle following MI were also attenuated by miR-433 inhibition (Figure [4](#F4){ref-type="fig"}D). Agilent gene arrays were used to compare the difference of gene expressions between ventricle samples from miR-433 antagomir or scrambled negative control post-MI (Supplemental Tables [6](#SM1){ref-type="supplementary-material"}-7). The KEGG pathway analysis based on dysregulated genes showed that extracellular matrix (ECM) receptor interaction was the most affected pathway (Figure [4](#F4){ref-type="fig"}E). Also, the protein levels of pro-fibrotic genes (TGF-β, α-SMA, CTGF, Col1a1, and Col3a1) were decreased, while genes responsible for collagen degradation (MMP2 and MMP9) were further increased by miR-433 inhibition in post-MI hearts (Figure [4](#F4){ref-type="fig"}F). Similar results were obtained for fibrosis-associated genes in miR-433 sponge AAV9-treated MI mice (Supplemental Figure [2](#SM1){ref-type="supplementary-material"}).
To gain mechanistic insight into the role of miR-433 in regulating fibrosis, we investigated the effect of miR-433 overexpression in cardiac fibroblasts *in vitro*. miR-433 overexpression promoted proliferation and differentiation of cardiac fibroblasts, as evidenced by an increase in EdU and α-SMA staining and increased expression levels of α-SMA, Col1a1, Col3a1, CTGF, and TSP-1 (Figure [5](#F5){ref-type="fig"}). However, up-regulation of miR-433 failed to further enhance cardiac fibroblasts proliferation and differentiation in the presence of either TGF-β or Ang II stimulation (Figure [5](#F5){ref-type="fig"}). Contrary to the effects of miR-433 overexpression, inhibition of miR-433 decreased cardiac fibroblasts proliferation and differentiation (Figure [6](#F6){ref-type="fig"}). Collectively, these data indicate that inhibition of miR-433 attenuates proliferation of cardiac fibroblasts and their differentiation into myofibroblasts both *in vitro* and *in vivo*.
AZIN1 and JNK1 are Identified as Two Target Genes of miR-433
------------------------------------------------------------
AZIN1 is reported to be a target gene of miR-433 in renal fibrosis [@B21]. However, its role in cardiac fibroblasts is not known. We first performed luciferase reporter assays to confirm that miR-433 could directly target the 3\'UTR of AZIN1 in both 293T cells and cardiac fibroblasts (Figure [7](#F7){ref-type="fig"}A). Next, we investigated whether AZIN1 could potentially mediate the effects of miR-433 in cardiac fibrosis. In cardiac fibroblasts, the expression level of AZIN1 was decreased by miR-433 agomir but increased by miR-433 antagomir as determined by Western blotting (Figure [7](#F7){ref-type="fig"}B-C), indicating that miR-433 could regulate endogenous AZIN1 expression levels. We next used AZIN1 overexpression plasmid to determine AZIN1\'s role in mediating the effect of miR-433 on cardiac fibroblasts proliferation and differentiation into myofibroblasts. Our data illustrated that overexpression of AZIN1 could attenuate the pro-fibrotic effect of miR-433 agomir on cardiac fibroblasts (Figure. 7D-F). Also, AZIN1 knockdown via siRNA failed to have an additive effect on fibroblast proliferation and myofibroblast differentiation in cells co-treated with miR-433 agomir (Supplemental Figure [3](#SM1){ref-type="supplementary-material"}). These data strongly suggest that AZIN1 is a target gene of miR-433 mediating its effect in cardiac fibrosis.
As AZIN1 was previously reported to be linked to TGF-β signaling in both kidney and liver fibrosis [@B21], [@B22], we further examined the modulatory effect of AZIN1 on TGF-β and its downstream effector Smad3. Our data revealed that knockdown of AZIN1 could upregulate TGF-β expression and activate Smad3 phosphorylation, while overexpressing AZIN1 had an opposite effect (Figure [8](#F8){ref-type="fig"}), indicating a potential relationship between AZIN1 and TGF-β/Smad3 signaling in the regulation of cardiac fibrosis.
Besides AZIN1, bioinformatic analysis using Targetscan indicated that JNK1 might be an additional potential target gene of miR-433 (Figure [9](#F9){ref-type="fig"}A). Luciferase reporter assays further confirmed that miR-433 led to a reduction in luciferase activity for the wild-type 3\'UTR construct for JNK1, but had no effect when the miR-433 binding site in the JNK1 3\'UTR was mutated, implying that JNK1 is a direct target of miR-433 (Figure [9](#F9){ref-type="fig"}A). To check if miR-433 could regulate endogenous JNK1 expression in cardiac fibroblasts, miR-433 agomir, antagomir, or their negative controls were transfected into cardiac fibroblasts. As determined by Western blotting, miR-433 agomir decreased, while miR-433 antagomir increased JNK1 expression (Figure [9](#F9){ref-type="fig"}B-C), confirming that miR-433 could regulate endogenous JNK1 expression levels in cardiac fibroblasts. In addition, we used JNK1 overexpression plasmid to determine its role in the miR-433-mediated cardiac fibroblasts proliferation and differentiation into myofibroblasts. Our results clearly indicated that overexpression of JNK1 could attenuate the pro-fibrotic effect of miR-433 agomir on cardiac fibroblasts (Figure [9](#F9){ref-type="fig"}D-F). JNK1 knockdown via siRNA, on the other hand, did not further increase fibroblast proliferation, though myofibroblast differentiation was slightly enhanced in cells co-treated with miR-433 agomir (Supplemental Figure [4](#SM1){ref-type="supplementary-material"}). These data identify JNK1 as a novel target gene of miR-433 contributing to cardiac fibroblast proliferation and myofibroblast differentiation.
As a member of mitogen-activated protein kinase (MAPK) family, JNK1 may have functional cross-talk with two other members of MAPK family, namely ERK and p38 kinase [@B30], [@B31]. To confirm this, JNK1 was knocked down by siRNA and the expression levels of ERK and p38 kinase were determined by Western blotting. We observed that JNK1 knockdown significantly activated ERK and p38 kinase as evidenced by increased ratios of p-ERK/ERK and p-p38/p38, paralleling with the activation of Smad3 (Figure [10](#F10){ref-type="fig"}A). However, the introduction of the JNK1 overexpression plasmid resulted in reduced phosphorylation levels of ERK1/2, p38, and Smad3 (Figure [10](#F10){ref-type="fig"}B). Interestingly, inhibition of p38, ERK or Smad3 could block the positive effects of miR-433 agomir on cardiac fibroblasts proliferation and differentiation, as determined by α-SMA and EdU staining, and reduce the expression levels of α-SMA, Col1a1, and Col3a1 (Figure [10](#F10){ref-type="fig"} C-D).
To examine whether AZIN1 and JNK1 could be regulated during cardiac fibrosis and/or miR-433 inhibition, we first examined their expression levels *in vivo* in the heart samples from post-MI rodent model, doxorubicin-induced cardiomyopathy rodent model, and human dilated cardiomyopathy. Consistently, both AZIN1 and JNK1 were down-regulated in these three fibrotic conditions at both protein and mRNA levels (Figure [11](#F11){ref-type="fig"}A-B). We next tested whether AZIN1 and JNK1 were increased in ventricle samples of miR-433 antagomir-injected mice. As determined by gene arrays, we did not detect changes of AZIN1 and JNK1 at the mRNA level. However, considering the fact that miRNAs regulate their target genes mostly at posttranscriptional levels, we also determined AZIN1 and JNK1 protein levels by Western blotting. Our results clearly showed that treatment with miR-433 antagomir increased the expression of AZIN1 and JNK1 in the presence or absence of MI (Figure [11](#F11){ref-type="fig"}C). These data are consistent with the hypothesis that one or both genes are target genes of miR-433*in vivo*. Interestingly, during MI, miR-433 antagomir inactivated ERK and p38 kinase as evidenced by the decreased ratio of p-ERK/ERK and p-p38/p38, together with Smad3 (Figure [11](#F11){ref-type="fig"}C).
Discussion
==========
Myocardial fibrosis is a common hallmark in a variety of cardiomyopathies [@B4]. Consequently, anti-fibrotic therapies are increasingly considered as an extremely promising approach for the treatment of heart failure [@B3]. Unfortunately, effective strategies to attenuate cardiac fibrosis are not available [@B2]. Aberrant expression of various miRNAs has been shown to play a crucial role in cardiac fibrosis and heart failure [@B1], [@B32]. These small non-coding miRNAs with conserved sequences have become promising therapeutic candidates from a drug development standpoint [@B7]. Recently, manipulating miRNAs for developing anti-fibrotic therapies has emerged as a novel treatment strategy for fibrotic changes [@B1], [@B33].
According to the miRBase 21 release, 1881 miRNAs have been identified in humans. Numerous studies have demonstrated the involvement of many of these miRNAs in vital cellular processes. However, the role of miRNAs in the heart and especially for cardiac fibrosis is unclear. It has been suggested that dysregulated miRNAs such as miR-21 and miR-29b, contribute to cardiac fibrosis [@B16], [@B18]. Using miRNA arrays, we identified elevated levels of miR-433 in post-MI cardiac fibrosis. The same modulation was also observed in other cardiac pathologies, including doxorubicin-induced cardiomyopathy in a rodent model, and in a limited number of human DCM samples, indicating that upregulation of miR-433 might be a common feature of adverse cardiac remodeling. Besides cardiac pathologies, miR-433 has been reported to be downregulated in human gastric carcinoma. Ectopic expression of miR-433 in the gastric cancer cell line HGC-27 could inhibit cellular proliferation, migration, invasion, and cell cycle progression [@B34]. miR-433 also inhibits liver cancer cell migration and oral squamous cell carcinoma (OSCC) cell growth and metastasis [@B35], [@B36], indicating that miR-433 acts as a tumor suppressor. In other studies, miR-433 has been shown to promote renal fibrosis and also TGF-β-dependent fibrogenesis in liver and kidney [@B21], [@B22]. In another report, miR-433 has been described to promote resistance to paclitaxel through the induction of cellular senescence in ovarian cancer cells [@B37]. These data point to the complex tissue- and cell-based specific roles of miR-433 in various cancers.
The role of miR-433 in the heart and during cardiac fibrosis had not been investigated previously. The proliferation and transformation of cardiac fibroblasts into myofibroblasts are key events for cardiac fibrosis [@B29]. Fibroblast proliferation and myofibroblast differentiation can be differentially regulated by growth factors such as TGF-β, EGF, PDGF, CTGF, and IGF [@B38]. Herein, we demonstrated that miR-433 over-expression enhanced both cardiac fibroblast proliferation and their differentiation into myofibroblasts, whereas inhibition of miR-433 attenuated these processes, indicating the critical stimulatory effect of miR-433 on cardiac fibroblast activation. We also observed that miR-433 was enriched in cardiac fibroblasts compared to cardiomyocytes. Furthermore, overexpression of miR-433 in cardiomyocytes does not appear to play a role in cardiomyocyte biology, as seen by the lack of effect on markers for pathological hypertrophy and extracellular matrix proteins.
A previously reported target gene of miR-433 in renal fibrosis, AZIN1, has been linked to TGF-β signaling in both kidney and liver fibrosis [@B21], [@B22]. It is an ornithine decarboxylase (ODC) homolog that binds to antizyme with a higher affinity [@B21], [@B22]. Suppression of AZIN1 expression results in antizyme repression followed by a decline of polyamine levels and consequent activation of the TGF-β signaling pathway to promote fibrosis [@B21], [@B22]. To date, very little information is available on the role of AZIN1 in cardiac pathologies. In this study, AZIN1 appeared to be responsible for the effects of miR-433 in cardiac fibroblasts. It was downregulated in the heart tissues from post-MI mice, doxorubicin-induced cardiomyopathy rodent model, and human dilated cardiomyopathy, indicating its potential role in the diseased myocardium with fibrosis. Furthermore, knockdown of AZIN1 could promote proliferation and differentiation of cardiac fibroblasts into myofibroblasts accompanied with an activation of TGF-β/Smad3 signaling pathway. However, the direct relationship between AZIN1 and TGF-β1 and their functional roles in the regulation of cardiac fibrosis needs to be further clarified through the function-rescue assay. Taken together, these results suggest that AZIN1 is a target gene of miR-433 in cardiac fibrosis and also provide evidence for the functional role of AZIN1 in the heart that needs to be explored in the future.
Besides AZIN1, based on bioinformatic analysis and experimental validation, JNK1 was identified as a novel target gene of miR-433 in cardiac fibroblasts. Jun NH2-terminal kinases, including three isoforms (JNK1, JNK2, and JNK3), belong to the MAPK family and play major roles in development, cell proliferation, differentiation, and apoptosis [@B31], [@B39]. JNK1 and JNK2 are abundant in myocardium while JNK3 is most abundant in the brain [@B31], [@B39]. In this study, luciferase assays demonstrated that JNK1 was a direct target of miR-433 and Western blot analysis confirmed that miR-433 could endogenously regulate JNK1 expression in cardiac fibroblasts. Functional studies in cardiac fibroblasts further indicated that reduction of JNK1 was responsible for the pro-fibrotic effects of miR-433 in cardiac fibroblasts. Furthermore, JNK1 may also have a functional cross-talk with ERK and p38 kinase, two other members of MAPK family [@B30]. ERK and p38 kinase pathways were activated while JNK1 was inhibited in the heart samples from post-MI mice, doxorubicin-induced cardiomyopathy rodent model, and human dilated cardiomyopathy. Also, reduction of JNK1 in cardiac fibroblasts activated ERK and p38 kinase and inhibition of ERK and p38 kinase attenuated the biological effects of miR-433 agomir on the proliferation and differentiation of cardiac fibroblasts. Collectively, these results suggest that miR-433 downregulates JNK1 and subsequently activates ERK and p38 kinase promoting cardiac fibrosis.
The protective effects of miR-433 inhibition against cardiac fibrosis were confirmed by antagonizing miR-433 or inhibiting miR-433 via cardiotropic AAV9, which attenuated cardiac fibrosis and preserved ventricular function post-MI. Although several lines of evidence presented here strongly supports the functional role of miR-433 in regulating cardiac fibrosis, more rigorous approaches are required to support this contention. These may include intra-myocardial rather than systemic delivery with a cardiac fibroblast-specific promoter and/or using a miR-433 transgenic mouse model created by using the cardiac fibroblast-specific promoter.
It is of note that cardiac fibrosis was decreased with the miR-433 antagomir but not abolished indicating the involvement of other pathways. For example, some clustered miRNAs of miR-433 including miR-431, miR-434 and miR-127 were also elevated in our initial miRNA array based on fibrotic heart samples post-MI. These miRNAs might work coordinately to promote cardiac fibrosis. It would also be interesting to further determine *in vivo* therapeutic roles for each of miR-433 targets, alone or in combination, by gain-of-function and loss-of-function studies. Furthermore, the therapeutic effects of miR-433 reduction on cardiac fibrosis in an established model need to be determined in the future. Last but not least, as cardiac fibrosis in the acute phase post-MI may protect the ischemic heart from structural rupture [@B40], the effect as well as the safety of miR-433 inhibition in the treatment of cardiac fibrosis must be carefully evaluated during the early phase post-MI. Notably, the data from the present study demonstrated that antagonizing miR-433 *in vivo* did not impact the infarct size 3 days after MI surgery suggesting that inhibition of miR-433 does not affect infarct size during the early phase post-MI.
In summary, our study has shown that miR-433 is induced by cardiac fibrosis, subsequently reducing the expression of AZIN1 and JNK1. Decreased AZIN1 activates TGF-β1 pathway while down-regulated JNK1 leads to activation of ERK and p38 kinase stimulating Smad3 and ultimately leading to cardiac fibrosis
This work was supported by the grants from National Natural Science Foundation of China (81570362 and 81200169 to JJ Xiao, 81370332 and 81170201 to XL Li, 81270314 and 81470515 to JH Xu, 81472158 to L Che, 81400647 to YH Bei, 81370362 to JC Zhong), the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD20102013 to XL Li), the National Basic Research Program of China (2014CB542300), the National Major Research Plan Training Program (91339108), Shanghai Municipal Education Commission-Gaofeng Clinical Medicine Grant (20152509), Shanghai Medical Guide Project from Shanghai Science and Technology Committee (134119a3000 to JH Xu), Natural Science Foundation of Shanghai (14ZR1437900 to L. Che), the Netherlands Cardiovascular Research Initiative (CVON): the Dutch Heart Foundation, Dutch Federation of University Medical Centers, the Netherlands Organization for Health Research and Development, and the Royal Netherlands Academy of Science (to JPG Sluijter) and the National Institutes of Health (NCATS Grant UH3 TR000901 to S Das). Dr XL Li is an Associate Fellow at the Collaborative Innovation Center for Cardiovascular Disease Translational Medicine.
Author Contributions
====================
J.X. designed the study, instructed all experiments and drafted the manuscript. X.L. participated in the design of the study and coordination of the whole work. L.T., Y.B., P.C., Z.L., S.F., H.Z., J.X. and L.C. performed the experiments and analyzed the data. X.C., X.B., J.Z., J.PG.S., S.D. helped to perform the experiments, provided technical assistance and revised the manuscript.
Supplementary Material {#SM0}
======================
######
Supplementary tables and figures.
######
Click here for additional data file.
{#F1}
{#F2}
{#F3}
{#F4}
{#F5}
{#F6}
{#F7}
{#F8}
{#F9}
{#F10}
{#F11}
[^1]: ^\*^These two authors contributed equally to this work.
[^2]: Competing Interests: The authors declare no competing financial interests.
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
With the developments of affordable and reliable high-throughput genotyping and next-generation sequencing platforms, many genome-wide association studies (GWAS) have been successfully conducted to identify DNA variants associated with many complex human diseases and traits, such as cancer, autoimmune diseases, height, blood pressure, body mass index, among others. As of 11/16/13, there were 11,907 single nucleotide polymorphisms (SNPs), 940 traits with 15,052 associations documented in the GWAS catalog, maintained by the National Human Genome Research Institute ([@B10]). These studies have uncovered many novel genes and implicated unexpected pathways associated with disease mechanisms, leading to great insights on disease etiology.
In spite of these accomplishments, many challenges remain in GWAS design and analysis. The first limitation is the limited statistical power to identify all disease-associated loci. Although many susceptible loci have been identified, they only explain a small fraction of the overall heritability, with the majority of heritability remaining unexplained. One possible reason is that the missing heritability is due to the lack of coverage of genetic variations on the genotyping platforms, such as those rare or even private variations. Another explanation is that most disease-associated variants have small effect sizes that are not likely detected due to low statistical power, even with thousands of subjects. To better identify these variants, more powerful and cost-effective designs and statistical methods are desired. Several approaches have proved cost-effective to enrich signals and increase statistical power. For example, a number of customized genotyping platforms have been used to target certain genomic regions with high density to fine map disease-associated variants. The successful examples include the use of the ImmunoChip ([@B35]) to fine map 186 distinct loci associated with 12 autoimmune diseases, and the use of the MetaboChip ([@B36]) to fine map established trait-associated loci. As for rare variant analysis, a number of whole exome sequencing studies have enjoyed success for diseases like autism and schizophrenia ([@B40]; [@B29]). It has also been found that studies focusing individuals with extreme phenotypes can increase statistical power because of the enriched genetic signals in the study subjects ([@B13]). In addition to improved platforms and study designs, many statistical methods have also been developed to increase statistical power. Meta-analysis is commonly applied to leverage all the information from separate studies to increase statistical power to identify disease-associated loci. A number of methods aim to investigate the combinatorial effects of a group of SNPs, including both marginal and interaction effects. These are accomplished by explicitly modeling the interactions of two or more SNPs; joint analysis of all the SNPs in a gene or defined region; and joint consideration of the SNPs in proximity of all the genes annotated in a specific pathway/network. The advances in study designs and statistical methods have led to many novel discoveries. For example, a recent study of the inflammatory bowel disease (IBD; [@B11]), which is a meta-analysis of both ImmunoChip and GWAS datasets, increased the number of IBD loci to 163, where these loci account for 13.6% of the genetic risk of Crohn's disease (CD) and 7.5% of ulcerative colitis. See [@B4] for a comprehensive review of some of the topics.
The second limitation is the difficulty to interpret the biological relevance of susceptible loci and link them with the disease etiology. Because the ultimate goal of association studies is to understand disease etiology and develop effective strategies to prevent and treat diseases, it would be desirable that GWAS results can implicate functional variants and disease pathways that can be experimentally studied in follow-up functional studies. However, a large proportion of disease-associated variants fall into non-coding regions of the genome, with 88% of the associated SNPs in GWAS catalog non-coding ([@B10]), making it difficult to form testable hypothesis. Even when the variants in the coding region, it is often not clear whether they are functional due to the presence of several closely linked variants. To address this problem, many statistical methods have been proposed to prioritize GWAS signals by incorporating diverse functional evidence, so that variants with small effect sizes but possessing functional features may be prioritized over variants with similar effect sizes but less likely to be functional. GWAS signals can be prioritized at both the SNP level and gene level, depending on the biological features considered and the input signals. Statistical approaches that prioritize at the SNP level are especially helpful in pinpointing the causal variants with sequence data, where essentially all the variations in the genome can be identified. This is contrast to earlier GWAS that only interrogated a subset of SNPs, such as tag SNPs, in genotyping platforms. One benefit of such approaches is that the functional evidence provides paths to derive plausible and testable hypotheses for the prioritized genes or loci. Moreover, with the incorporation of other data informative about disease association, the prioritized genes/loci are more likely to be truly associated with disease. For example, it has been observed that trait-associated loci are more concentrated in regions with certain genomic features, such as protein coding regions and expression quantitative trait loci (eQTL). In this review, we will review (1) biological/genomic features that are informative for prioritizing GWAS results; and (2) computational methods and tools that prioritize disease-associated SNPs by integrating these biological/genomic features.
BIOLOGICAL FEATURES USED IN PRIORITIZATION AND THEIR JUSTIFICATIONS
===================================================================
The first step, and sometimes the only step of many SNP prioritization approaches is to annotate the candidate SNPs by intersecting GWAS signals with desired genomic features, such as eQTLs, transcription factor binding sites, DNase hypersensitive sites, histone modifications, and others. For CD, [@B9] showed that *cis*-eQTL SNPs were enriched in known CD-associated SNPs. Based on this observation, the authors proposed to select a subset of SNPs to follow-up a public CD GWAS dataset, to intersect the top 500 GWAS hits with *cis*-eQTLs in an eQTL database. The SNPs thus selected, *cis*-eQTLs of the genes *UBE2L3* and *BCL3*, were replicable in two independent replication cohorts. This represents a successful application of the annotation-based prioritization methods.
The genomic features may implicate functional roles of the prioritized SNPs to disease etiology, and these hypotheses can be formally tested through molecular studies. These include variants both in coding and non-coding regions. Through these filtering/intersecting methods, researchers can focus on a much smaller number of SNPs in follow-up studies. Although the proximity of a SNP with a documented genomic feature may suggest a functional role of the SNP, it may not necessarily increase the probability that it will affect the phenotype of interest, nor the probability that the locus is truly susceptible. In general, the genomic features discussed above are considered biologically plausible and extensively used to prioritize SNPs, but whether each feature is informative on a SNP's functional relevance is disease and context dependent. In a recent study, [@B20] tried to identify features that are important in selecting SNPs for follow-up studies by surveys in experts. They sent questionnaires to ten experts who conducted GWAS studies, and asked their opinion on the importance of a set of selected features. The features included relative position of the SNP to a nearby transcript, whether the SNP causes an amino acid change, etc. (see Table 2 in their paper). The result was not surprising, as experts considered gene level evidence more important, such as the SNP in a gene that is previously associated with the phenotype, or that encodes a protein in a phenotype related pathway, or that has gene/protein interaction relevant to the phenotype. Experts opinions are valuable, however, they might be biased toward existing knowledge and also expertise in specific diseases. Nevertheless, this paper highlighted the need for understanding what features should be considered in prioritization. In the following, we review these features and statistical methods to use these features, to inform human geneticists in their applications of the annotation-based approaches.
EXPRESSION QUANTITATIVE TRAIT LOCI
----------------------------------
By contrasting the SNPs documented in the GWAS catalog ([@B10]) with those randomly sampled SNPs with matching minor allele frequency distribution, [@B24] showed that complex trait-associated SNPs are more likely to be eQTLs. The conclusion remained valid for a linkage disequilibrium (LD)-pruned subset of SNPs in the GWAS catalog. Since the eQTL annotation considered by [@B24] was derived from an expression dataset of lymphoblastoid cell lines, it was of interest to investigate whether cell line-specific eQTLs showed different levels of enrichments across diseases with different focal tissues, including cancer, neurological/psychiatric disorders, and autoimmune disorders. By tissue of relevance, the lymphoblastoid cell lines should be a good proxy for autoimmune disorders, and relatively poor for cancer and neurological/psychiatric disorders. As expected, there was greater enrichment of eQTLs in the group of autoimmune disorders, while only moderate enrichment for the other two groups of diseases. Furthermore, in the examination of the results in the Wellcome Trust Case Control Consortium (WTCCC) GWAS dataset of CD, eQTLs were enriched in SNPs with association *p*-value less than 0.01, but not for the missense SNPs, indicating potential loss of information if non-coding SNPs are ignored.
PROTEIN DELETERIOUSNESS PREDICTIONS
-----------------------------------
Polymorphisms in the coding region may have different effects on protein function. Synonymous SNPs do not change the corresponding protein sequence; non-synonymous SNPs change amino acid composition, or truncate the protein sequence by causing an early codon; indels can change protein sequence with varying consequence depending on whether the indel is in-frame or frame-shifting; SNPs and indels can also interrupt splicing sites, thus change the mRNA isoform. In other words, mutations in the coding region may be benign or deleterious to protein function, with deleterious mutations more likely to have a phenotypic effect. Many computational tools have been developed to predict "deleteriousness" of SNPs and indels ([@B21]; [@B1]). These methods generally take features like biochemical property of the altered amino acid, conservation and sequence homology as input, and use machine learning technique to train a classifier. These methods have been comprehensively reviewed by [@B8] and [@B22].
The most extreme case of protein function interruption is the loss of function mutations. However, genome sequencing studies found that all human carry loss of function mutations without obvious phenotypic effect, and common loss of function variants were depleted in polymorphisms associated with complex disease like CD and rheumatoid arthritis ([@B15]). The results indicate that the "deleteriousness" feature should be interpreted with caution, since disruption of protein function does not necessarily have a phenotypic effect. In this regard, the "residual variance intolerance score" has been defined quantitatively measure the tolerance of a protein to mutations ([@B25]). The number of missense and non-sense variants found in each gene in the cohort of the National Heart, Lung, and Blood Institute (NHLBI) exome sequencing project was compared to the number of functionally neutral variants, and genes with variants fewer than expected are assigned a negative score, indicating they are less tolerant to variations.
DIFFERENTIAL GENE EXPRESSIONS
-----------------------------
Gene expression microarrays and RNA-seq are commonly used to study gene expression profiles in disease cases and matched controls, and differentially expressed genes thus identified may suggest disease mechanisms and potential biomarkers that can be further explored in follow-up studies. [@B5] analyzed 476 expression datasets in the Gene Expression Omnibus (GEO), and calculated the frequency that a gene was differentially expressed in these datasets, which they called "differential expression ratio." They found that differential expression ratio is positively correlated with the likelihood that a gene harbors disease-associated variants, where the list of disease-associated genes was created by combining information from Genetic Association Database (GAD; [@B2]) and Human Gene Mutation Database (HGMD; [@B30]). In addition, they found that among the genes discovered in the initial scan of the WTCCC type 1 diabetes mellitus GWAS dataset, the differential expression ratio was higher in genes that were replicable than those not replicable in follow-up studies. These authors have developed an online server, FitSNPs, to incorporate this feature (see **Table [1](#T1){ref-type="table"}**).
######
A list of online SNP prioritization tools.
Name Website Reference
--------------- ------------------------------------------- -----------
FASTSNP <http://fastsnp.ibms.sinica.edu.tw> [@B41]
FitSNPs <http://fitsnps.stanford.edu/fitSNPs.php> [@B5]
SNPranker 2.0 <http://www.itb.cnr.it/snpranker> [@B19]
SPOT <http://spot.cgsmd.isi.edu> [@B27]
DNase I HYPERSENSITIVE SITES
----------------------------
DNase I hypersensitive sites (DHSs) are markers of accessible chromatin, which indicate regulatory roles in the transcription process. DHS have been mapped in 349 cell and tissue samples genome-wide by next-generation sequencing ([@B33]). Enrichment analysis showed that trait-associated SNPs in the GWAS catalog ([@B10]) are more concentrated within DHS regions, excluding confounding factors such as allele frequency and distance from the nearest transcriptional start site ([@B17]).
OTHERS
------
There are many more genomic features collected and annotated in large community projects, such as the Encyclopedia of DNA Elements (ENCODE; [@B7]), which are potentially valuable for SNP prioritization. [@B12] examined enrichment or depletion of trait-associated SNPs in 58 genomic features. The features investigated covered genic and regulatory features, conservation features, and chromatin state features (see Table 1 in [@B12]). Among those features, genomic regions annotated as "heterochromatin" and "low expression signals" are depleted of trait-associated SNPs, while eQTLs and "strong enhancer" showed the highest level of enrichment.
The biological features discussed so far are measured/inferred from laboratory cell lines and the sequence and annotation of the human genome, which do not provide trait-specific information. However, trait-relevant features are intrinsically helpful for prioritization. For example, a DNase-seq experiment in intestine tissues and immune cells of CD patients would be more informative for prioritizing variants associated with CD than those measured in brain tissues. [@B16] and [@B26] reviewed recent progress in mapping the epigenome (including DNA methylation and histone modification), showing that epigenetic modifications play important roles in human diseases, including cancer, neurodevelopmental disorder, neurodegenerative disease, neurological disease, and autoimmune diseases. Thus, epigenome data in disease states is valuable for understanding disease and prioritize disease susceptible loci. However, efforts in disease-specific epigenome mapping and systematic database to document such data are still lacking and greatly needed. For DNA methylation alone, a database, DiseaseMeth, has incorporated methylation data for 72 human diseases ([@B14]).
SNP PRIORITIZATION APPROACHES AND AVAILABLE WEB SERVERS
=======================================================
Here we review methods and tools that prioritize GWAS signals at single SNP level. There are mainly two steps in these methods. The first step applies annotations or filters based on whether or not the candidate SNP has the desirable features and the second step scores the candidate SNPs by integrating evidence from multiple sources. [@B28], [@B27] developed an online prioritization tool, SPOT, which systematically combines multiple biological databases to prioritize SNPs by the genomic information network (GIN) model (see **Table [1](#T1){ref-type="table"}**). In their model, each SNP is assigned a prioritization score, which is a linear sum of scores derived from pathway information, comparative genomics, linkage scan, and results of other independent GWAS studies. The weights are decided by the strength of the link between the SNP and the annotations. For example, for a SNP that is in LD with a non-synonymous SNP of a susceptible gene, the assigned weight will be less than that of SNPs physically in the gene. The methodology prioritized SNPs with increased biological relevance in a GWAS study of nicotine dependence. [@B32] incorporated biological features in a Bayesian framework, where the prior probability that a SNP is associated with the phenotype is determined by its annotations. They first curated a training set, including SNPs that were confirmed replicable as the positive set, and 1,000 randomly selected SNPs as control set. For a selected set of features, a logistic regression model was fit for each disease. Thus, the log odds ratio that a test SNP is associated with the disease can be estimated through the model.
There are also web servers that perform SNP prioritization in an annotation fashion. They annotate the candidate SNPs by single or multiple features, but do not combine the results. They differ by the features and strategies they use in prioritization. A list of SNP prioritization resources are provided in **Table [1](#T1){ref-type="table"}**. Most of these tools are only applicable to SNPs, and tools that can prioritize indels are still lacking.
*FASTSNP* uses a decision tree framework to assign different risk level to SNPs by considering the genomic location and functional effect of the SNPs ([@B41]).
*FitSNPs* calculated a differential expression ratio for all genes in the genome, and prioritize SNPs by the differential expression ratio of their associated genes ([@B5]).
*SNPranker 2.0* first annotates the SNPs with different features, and then user a user interactive way to integrate features ([@B5]). Users can specify the features they want to include and the weight of each feature, which would give the users an opportunity to enforce their biological priors. But they also provide an optimal set of weights by default. The optimal weights were determined by a cross-validation approach.
TOOLS FOR VARIANT ANNOTATION
============================
Besides the SNP prioritization tools, there are also many web servers and software for variant annotation (**Table [2](#T2){ref-type="table"}**), which could provide useful information for prioritization. Basically, these tools take a list of query variants as input, and annotate them with their in-house databases. Among these, HaploReg ([@B38]) and RegulomeDB ([@B3]) provide annotation for variations in non-coding regions. HaploReg annotates variations by their chromatin state, conservation across mammals, and computationally predicted transcription factor binding sites. Besides, by utilizing LD information from the 1000 Genomes Project, HaploReg automatically reports, and annotates all variations within a user-specified LD threshold of the query variant. RegulomeDB has incorporated many data sources, including the ENCODE project, available transcription factor ChIP-seq data, and eQTL datasets. The other tools are designed for variations in the whole genome. They annotate the query variations by dbSNP ID, allele frequency in different ethnic groups, position in a transcript (intron, exon, 5′ UTR, etc.,), and the resultant amino acid change if any. SeattleSeq ([@B23]) and Variant Effect Predictor (VEP; [@B18]) has convenient web interface, suitable for users who are not familiar with scripts and programming languages, while ANNOVAR ([@B37]) and Snpeff ([@B6]) are stand-alone software packages, so that they can be easily incorporated into variant analysis pipelines. Discussions on variant annotation tools can also be found in [@B39].
######
list of tools for variant annotation.
Name Website Reference
------------ ---------------------------------------------------------------- -----------
ANNOVAR <http://www.openbioinformatics.org/annovar/> [@B37]
HaploReg <http://www.broadinstitute.org/mammals/haploreg/haploreg.php> [@B38]
RegulomeDB <http://regulome.stanford.edu/> [@B3]
SeattleSeq <http://snp.gs.washington.edu/SeattleSeqAnnotation137/> [@B23]
Snpeff <http://snpeff.sourceforge.net> [@B6]
VEP <http://useast.ensembl.org/info/docs/variation/vep/index.html> [@B18]
DISCUSSION
==========
In this review, we have focused on biological and genomic features that are informative for SNP prioritization. In the second phase of association studies, researchers can use these databases or tools to choose SNPs in follow-up studies. The observation that eQTLs and open chromatin regions are enriched of trait-associated SNPs highlights the potential rich information in the non-coding regions of the genome ([@B24]). Nonetheless, the gene-centric approaches may be helpful in disease gene discovery, and there are many approaches that perform prioritization at the gene level. A review of the methods and tools for gene-based prioritization can be found in [@B34].
The prioritization methods discussed here take as input a list of candidate SNPs, which is usually derived by taking all the SNPs achieving a specified significance level in GWAS, e.g., all SNP with *p*-values less than 0.01. The SNPs are treated equally regardless of the association *p*-values. However, the *p*-value, the effect size, and other statistics that summarize the association level of individual SNP could be informative for SNP selection. A computational framework that incorporates the significance level with the biological/genomic features discussed above might improve the performance of the prioritization scheme. A discussion of different signal measures of association was given by [@B31].
Although there are many web servers and databases for SNP prioritization, most of them provide annotations of different types of features, but do not rank these SNPs through integrating GWAS and annotation information. Also, these methods do not employ disease-specific information. There is still a great need for statistical methods that select and integrate multiple annotations in a disease-specific manner, and re-rank SNPs under a coherent statistical framework.
Conflict of Interest Statement
==============================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Author Contributions
====================
Lin Hou and Hongyu Zhao conceived and wrote the paper.
Supported in part by the NIH grants R01 GM59507, the VA Cooperative Studies Program of the Department of Veterans Affairs, Office of Research and Development, and the Clinical and Translational Science Award UL1 RR024139 from the National Center for Research Resources, National Institutes of Health.
[^1]: Edited by: *Shuang Wang, Columbia University, USA*
[^2]: Reviewed by: *Yun Li, University of North Carolina, USA; Jinming Li, Southern Medical University, China*
[^3]: This article was submitted to Statistical Genetics and Methodology, a section of the journal Frontiers in Genetics.
| {
"pile_set_name": "PubMed Central"
} |
{.48}
{.49}
{.50}
{.51}
{.52}
{.53}
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction
===============
Yellow perch (*Perca flavescens*, Mitchell 1814), widely distributed in fresh waters of the USA and Canada, is an important ecological and aquacultural species, especially in the Great Lakes Region and the Midwestern states of the U.S. This species has been considered one of the most flavorful species among all panfish and carries special advantages in exhibiting a mild taste and firm flesh with low fat and phospholipid content \[[@b1-ijms-10-00018]\]. Due to commercial and recreational overexploitation, harvests have declined since the 1990s. To supply continued high market demand, breeding programs such as O'GIFT (Ohio Genetic Improvement of Farmed-fish Traits) have been launched to improve growth rate and disease resistance of yellow perch. Further investigation of these and other important multigenic traits depends on the availability of molecular genetic markers and use of such markers in efficient breeding programs (e.g. marker assisted selection).
Microsatellite analysis based on polymerase chain reaction (PCR) offers the finest resolution to date for studying molecular variation in perch. Most of the microsatellite DNA markers are Type II markers, which are developed from anonymous genomic sequences. Previously, Type II markers were isolated and characterized to perform the landscape genetic analysis \[[@b2-ijms-10-00018]\] and to evaluate broodstock populations of yellow perch \[[@b3-ijms-10-00018]\]. Comparatively, Type I markers, which are associated with genes of known functions, are more useful for comparative genome mapping \[[@b4-ijms-10-00018]\]. Type I markers often serve as anchorage points for genomic segments. Lack of Type I markers in yellow perch and many other aquaculture species has hindered major progress in genomics and genetic studies in aquatic animals \[[@b4-ijms-10-00018]\]. Although an amount of Type II microsatellite DNA markers were developed for yellow perch, the number of markers is still insufficient for planned QTL analysis of traits such as growth or disease resistance. To increase the numbers of independent simple sequence repeat (SSR) loci available for genomic studies in *Perca flavescens*, we evaluated loci from several microsatellite-enriched libraries and mined online cDNA databases as suggested by a number of researchers \[[@b5-ijms-10-00018], [@b6-ijms-10-00018], [@b7-ijms-10-00018]\]. Of particular interest was whether microsatellite markers developed from expressed sequence tag (EST) sequences (i.e. Type I markers) of species related to *P. flavescens* would be sufficiently polymorphic \[[@b8-ijms-10-00018]\].
In the present study, we report bioinformatic mining of the EST database of a related species, European perch (*Perca fluviatilis*), from which we developed polymorphic EST microsatellites for yellow perch. The rates of polymorphism recorded for these markers, both genomic and EST microsatellites, were evaluated by genotyping 30 individuals sampled from a wild population. Additionally, the cross utility of these markers was tested in a related species, the walleye (*Sander vitreus*).
2. Experimental Section
=======================
2.1. EST database mining
------------------------
To develop EST-SSRs for yellow perch, European perch EST sequences were obtained from GenBank dbEST (<http://www.ncbi.nlm.nih.gov/dbEST/index.html>). All data were scanned using the software SSR Hunter version 1.3 (<http://www.biosoft.net/dna/SSRHunter.htm>) using search parameters set to more than seven repetitions for di-nucleotide repeats, five for tri-, four for tetra-, and three for penta-and hexanucleotide repeats.
2.2. Microsatellite-enriched library construction
-------------------------------------------------
Microsatellite-enriched libraries were conducted using the method described by Li *et al*. \[[@b3-ijms-10-00018]\]. Briefly, the genomic DNA isolated from fin tissue was digested with a restriction enzyme *Sau*3A at 37°C for 3 hours. The fragments with the size range of 0.5 -- 2 kb were recovered from an agarose gel. A synthesized adaptor SAUL (A: 5′-GCGGTACCCGGGAAGCTTGG-3′ and B: 5′-GATCCCAAGCTTCCCGGGTACCGC-3′) was ligated to the fragments using T4 DNA ligase. Microsatellite-containing fragments were selectively coupled to biotinylated repeat motifs \[(CA)~n~, (GT)~n~, (AAC)~11~, (GAAT)~10~, (ACAT)~11~, (AAAG)~11~, (GTA)~15~, and (AAT)~15~\], captured, and washed. Fragments containing microsatellites were ligated to a TOPO vector (Invitrogen) and transformed into competent *Escherichia coli* cells. Positive clones were selected for PCR amplification using M13 universal primers and the PCR products were sequenced.
2.3. Sequence analysis and primer design
----------------------------------------
To exclude duplicates, all sequences, including genomic and EST sequences, were subjected to BioEdit Sequence Alignment Editor Software for grouping clusters using multiple sequence alignment. Microsatellites with the same flanking regions were considered as the same loci. The independent sequences were submitted to the DNA Data Bank of Japan (DDBJ) for homology searches using BLASTN (<http://blast.ddbj.nig.ac.jp/top-e.html>) against the vertebrate DNA databases to exclude loci previously reported. Sequences with the longest perfect repeats and flanking regions were selected for PCR primer design (Primer Premier version 5.0 software; <http://www.PremierBiosoft.com/faq.html>). One primer of each primer pair was modified at the 5′-end with an M13 universal tail (5′-CAGTCGGGCGTCATCA-3′) as described by Boutin *et al*. \[[@b9-ijms-10-00018]\].
2.4. DNA extraction, PCR amplification and genotyping
-----------------------------------------------------
A total of 30 adult yellow perch were collected live from a wild population in Lake Wallenpaupack in Pennsylvania, U.S. Individual fin-clips were stored immediately into 95% ethanol. For each specimen, DNA was extracted from 50 mg of tissue according to the methods described by Waters *et al*. \[[@b10-ijms-10-00018]\]. Amplification of microsatellite loci was performed with three primers, the tailed primer, the nontailed primer, and the M13 universal 5′-labelled (FAM, TET, or NED) primer that contained the same sequence as the M13 universal tail. The PCR reaction mix contained approximately 50 ng of genomic DNA, 3 μL of JumpStart RedMix (Sigma), 1.5 pmol of both nontailed and labelled primers, 0.1 pmol of the tailed primer, and 100 μM of spermidine in a total volume of 6 μL. The PCR conditions were programmed as one cycle of denaturation at 95°C for 3 min, followed by 35 cycles of 30s at 95°C, 30s at locus-specific annealing temperature ([Table 1](#t1-ijms-10-00018){ref-type="table"}), and 45s at 72°C, ending with a final step at 72°C for 5 min. Amplification products were separated using an ABI 3130 Prism DNA genetic analyzer and the genotyping results were analyzed using Genemap® 4.0 software.
2.5. Genetic data analysis
--------------------------
For a certain locus, the allele size range (*S*) was directly obtained from the Genemap® 4.0 software. The number of alleles (*A*) and their frequency (*F*), the observed heterozygosity (*H~o~*) and the expected heterozygosity (*H~e~*) were calculated using the computer program POPGENE 32. The Markov chain method \[[@b11-ijms-10-00018]\] was used to estimate the probability of significant deviation from Hardy--Weinberg equilibrium (HWE) and pairwise tests for linkage disequilibrium (LD) were performed using the program GENEPOP online version (<http://genepop.curtin.edu.au/>) using the default parameters. Significance criteria were adjusted for the number of simultaneous tests using Bonferroni correction \[[@b12-ijms-10-00018]\].
2.6. Cross utility
------------------
To determine the potential for cross utility, amplification of the identified markers was assessed in one related species, the walleye (*Sander vitreus*). The same PCR conditions and genotyping methods were used as described above except that annealing temperature was re-optimized at each locus.
3. Results and Discussion
=========================
3.1. Genomic-SSRs
-----------------
A total of 16 sequences derived from the microsatellite-enriched libraries were selected for primer design. The optimization results showed that eight primer pairs could successfully amplify target fragments of the expected sizes. All eight loci exhibited polymorphism in the individuals tested. The numbers of alleles varied from 3 -- 14 with an average of 8.5 alleles per locus. The observed and expected heterozygosities ranged from 0.07 to 0.81 and from 0.20 to 0.95, respectively ([Table 1](#t1-ijms-10-00018){ref-type="table"}). None of the loci showed significant linkage disequilibrium. After sequential Bonferroni correction for multiple tests, five loci were found to depart significantly from Hardy--Weinberg equilibrium (HWE). To exclude the impact of short allele dominance (large allele dropout), data were subject to analysis with Micro-Checker \[[@b13-ijms-10-00018]\]. No evidence for large allele drop-out was found for any of the loci. Further tests indicated that heterozygote deficiency at these loci was responsible for the departure ([Table 1](#t1-ijms-10-00018){ref-type="table"}). Another possible explanation for the departure from HWE is the dramatic contemporary decline in spawning populations, and consequent non-random mating and genetic bottlenecks \[[@b14-ijms-10-00018], [@b15-ijms-10-00018]\]. A final possibility is subpopulation structure which cannot be ruled out without further analysis.
3.2. EST-SSRs
-------------
In the process of EST database mining, a total of 2,226 EST sequences were deposited in GenBank. The mining results showed that 110 (4.93%) sequences contained microsatellites that conformed to our mining criteria ([Table 2](#t2-ijms-10-00018){ref-type="table"}). As found in other species, di-nucleotide repeats were the most abundant, accounting for 73.64% of all repeats located. This ratio is much higher than has been reported for some other aquatic species such as shrimp, bivalves \[[@b5-ijms-10-00018], [@b6-ijms-10-00018], [@b16-ijms-10-00018]\], and other freshwater fish \[[@b7-ijms-10-00018]\]. Surprisingly, the most abundant di-nucleotide repeat type was AG/CT, which is not consistent with reported findings for other fish such as common carp *Cyprinus carpio* \[[@b7-ijms-10-00018]\], pufferfish *Fugu rubripes* \[[@b17-ijms-10-00018]\] and catfish *Ictalurus punctatus* \[[@b18-ijms-10-00018]\] where, in general, the AC/GT repeat type is the most abundant di-nucleotide microsatellite. Similar to our findings, whole genome scanning has indicated that AG/CT is the most abundant type in some aquatic animals such as scallop \[[@b19-ijms-10-00018], [@b20-ijms-10-00018]\]. Biased sampling, due primarily to the small number of EST sequences examined in this study, may explain the current findings. To further confirm or refine the observation for yellow perch, more sequences or whole genome scanning are needed.
Twenty-three EST-derived sequences were chosen for PCR primer design. Among them, 13 primer pairs (56.5%) amplified products of the expected size. The presence of long introns between primers in genomic DNA, primer sequences spanning across introns and/or mutations, and indels (insertions or deletions) in the primer annealing sites between the two perch species may explain the non-amplification \[[@b5-ijms-10-00018]\]. However, the success ratio we observed for EST-derived microsatellites is slightly higher than those in other studies using the same strategy, such as development of Japanese sea urchin (*Strongylocentrotus intermedius*) using the EST sequences of a related species of purple sea urchin (*S. purpuratus*) \[[@b8-ijms-10-00018]\].
Although we selected relatively long microsatellite regions, the polymorphism assessment results revealed low levels of genetic diversity at these loci. Two or three alleles were detected at most loci and only one locus displayed 5 alleles in the individuals tested ([Table 1](#t1-ijms-10-00018){ref-type="table"}). Similarly low genetic diversity was also observed in terms of heterozygosity ([Table 1](#t1-ijms-10-00018){ref-type="table"}). In previous studies where levels of polymorphism have been compared between Type I and Type II microsatellite DNA markers in the same species, the level of polymorphism of Type I markers has usually been observed to be slightly lower \[[@b20-ijms-10-00018]\] but not nearly so dramatic as the differences we observed in polymorphism between EST-SSRs and genomic-SSRs ([Table 1](#t1-ijms-10-00018){ref-type="table"}). Evolutionary conservation and lower mutation rates within gene-coding sequences is a possible explanation for our observation but does not account for the prior published results. The practical implications are as described by Eujayl *et al*. \[[@b21-ijms-10-00018]\], that a suite of more mutationally-stable EST-SSRs could complement highly variable genomic-SSRs to reconstruct past evolutionary events and to identify regions of genomes that are identical by descent. This would be of particular utility in genomic, gene mapping, and QTL studies across species within *Perca*.
3.3. Cross utility
------------------
Of eight genomic-SSRs and 12 EST-SSRs, three (37.5%) and eight (66.7%) loci were successfully cross-amplified in the walleye ([Table 1](#t1-ijms-10-00018){ref-type="table"}), respectively. The cross utility results confirmed that Type I microsatellite markers have higher success ratio than that of Type II microsatellites in the cross-species amplifications among closely-related species. Although the walleye belongs to a different genus, the high cross-amplification ratio was also observed at both genomic-SSR and EST-SSR loci.
4. Conclusions
==============
In the present study, a total of 21 novel genomic-SSRs and EST-SSRs for yellow perch (*Perca flavescens*) were developed using the methods of construction of SSR-enrichment libraries and EST database mining of a related species. Compared with the genomic-SSRs, the EST-SSRs for yellow perch displayed a relatively lower level of genetic variability not only in number of alleles but also in heterozygosity. As described in other publications, mining EST databases provides an efficient and low-cost approach to obtaining new microsatellite markers for species of interest. Furthermore, the results also demonstrated the feasibility of microsatellite marker development by EST database mining of a genetically related species in fish.
This study was supported by the Cooperative State Research, Education, and Extension Service, U.S. Department of Agriculture, under Agreement No. 2005-38879-02357 and 2006-38879-03684. Salaries and research support were provided by state and federal funds appropriated to The Ohio State University, Ohio Agricultural Research and Development Center. We thank Hong Yao for genotyping.
######
Characterization of genomic-SSRs and EST-SSRs for yellow perch (*Perca flavescens*).
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Locus name Accession Primer sequence (5′-3′) *Ta* Repeats *S* (bp) *A* *Ho* *He* *P*-value Cross utility (*Ta*; *A*)
------------------------------------------------------ ---------------------------------- ------ ---------------------------------- ---------- ----- -------- -------- ------------------------------------------------------ ---------------------------
YP23[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: M13-TTGGACAAAAATAACTCACT\ 55 (TTC)~16~ 180--210 10 0.8077 0.8620 0.8312 52; 3
FJ547096 R: AGAGTAGAAATGCGGTTGCT
YP72[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: AAAGAGAGCAAAGGGGAAGA\ 55 (GGT)~5~GAA (GGT)~5~GAA(GGT)~16~ 255--264 3 0.3846 0.4970 0.4615 54; 2
FJ547097 R: M13-TGTGTAAGAAACAGGCAGGT
YP86[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: M13-CCGGCTACTTCATGTTAAAA\ 55 (AGAT)~14~ 331--387 12 0.5185 0.9371 0.0093[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} ---
FJ547098 R: GTGGGAATAAGGGTTAGGCT
YP89[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: ATGGAGATTTACAGCCCCTA\ 55 (CA)~5~GA(CA)~18~ 191--227 6 0.1238 0.6260 0.0000[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} ---
FJ547099 R: M13-ACTAATAACCACCATCCTGC
YP90[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: M13-AGAAAAGAGGGAAAGAAGG\ 52 (GAAA)~16~ 123--171 11 0.5556 0.7596 0.8084 ---
FJ547100 R: CCGCTATTTCACTCTGTTTT
YP94[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: M13-TTCACATTCAATAGGAGTAGAGT\ 50 (ACAT)~15~ 331--407 9 0.0714 0.8331 0.0003[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} ---
FJ547101 R: CTGTAAAACCATTGCCGATAAA
YP95[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: GTGCCCTTTGTCACCCAT\ 55 (CA)~14~ 127--133 3 0.0870 0.3710 0.0001[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} 52; 1
FJ547102 R: M13-GCCCTCATTTATGTCTCTCC
YP105[†](#tfn8-ijms-10-00018){ref-type="table-fn"}\ F: M13-TAGAAGCAAAACCCGTGA\ 55 (CTA)~14~ 169--214 14 0.4815 0.9511 0.0028[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} ---
FJ547103 R: TGTCCCTCACCAGCCAGT
PFE01[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-CTCCCAAAATAAAGCCAATGTC\ 54 (TC)~10~ 250--268 2 0.0714 0.0701 0.8907 54; 2
DR730576 R: ACAGAGTTTCAGGCACTTGTGG
PFE03[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-GCAGAAATGCTACATAGATCCT\ 52 (GT)~16~ 124--136 5 0.5714 0.5396 0.8719 50; 3
DR730639 R: AGTCAATATCCTCCAAATGTGC
PFE06\#\ F: M13-TTGCCTGAGGTTGTATTGAGAA\ 52 (AG)7 164--176 2 0.0357 0.0357 1.0000 52; 2
DV671343 R: ACAGTCGTAGCAGAGGGTCAC
PFE07[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-CGGCACGAGGGGACTGTAATC\ 50 (AAC)~6~ 109--121 3 0.0357 0.1045 0.0018[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} 54; 1
DV671312 R: TGTGCTCTTTCCCTTGTGACCG
PFE08[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-GTCTTAAACAAGTCTTCATAGCAC\ 56 (TAA)~11~ 160--168 2 0.0357 0.0357 1.0000 50; 1
DV671070 R: GGACAGAGAACACATAGAGAATC
PFE11[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-CTTAGACAGACCGACCTACAG\ 50 (TGA)~12~ 220--223 2 0.0357 0.0357 1.0000 ---
DW985750 R: ATGTCAGCCAAGATGTAATG
PFE12[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-TGCGTGCCAAGGGCGGTGTT\ 54 (CCT)~5~ 131--149 3 0.0357 0.0708 0.0018[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} 54; 1
DV752650 R: CCGTCCCCTCAACAAATACC
PFE14[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-AGCCACAAAGCTGAACATAG\ 52 (AT)~10~ 258--264 3 0.1429 0.1351 0.7270 50; 1
DV671188 R: TGCCATGTTGTATCTCCCAC
PFE15[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-GTATTAGTCTATGTATATTGCC\ 55 (TATC)~17~ 292--296 2 0.0357 0.0357 1.0000 50; 1
DR731110 R: CGGGATGTCACTTACTTCTC
PFE19[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-TGTCTAACGATTGCTTTTCCT\ 56 (AT)~10~ 80--82 2 0.0000 0.0701 0.0016[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} ---
DV671307 R: CAATGAAAAATAAACATGCGTGACC
PFE20[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-GATCCATCCTGCTCAGACTC\ 56 (TC)~23~ 281--283 2 0.0000 0.0701 0.0016[\*](#tfn9-ijms-10-00018){ref-type="table-fn"} ---
DR731052 R: AAGAGATTGAGTTTGGTAGC
PFE22[\#](#tfn7-ijms-10-00018){ref-type="table-fn"}\ F: M13-ATACAGAGGCCTTCATTTGT\ 56 (TA)~9~ 280--282 3 0.0714 0.0701 0.8907 ---
DR730585 R: CAGCTACAGTTCATTCTACCT
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
*T~a~*: annealing temperature (°C); *S*: allele size range (M13 universal tail included);
*A*: number of alleles;
*H~o~*: observed heterozygosity;
*H~e~*: expected heterozygosity;
*P*-value: *P*-values for exact test for Hardy--Weinberg equilibrium (HWE); M13: universal M13 tail (5′-CAGTCGGGCGTCATCA-3′);
Cross utility: primers cross amplified for the walleye (*Sander vitreus*) (*N* = 4);
EST-SSRs developed for yellow perch;
genomic-SSRs derived from microsatellite-enriched library;
departure from HWE after Bonferroni correction.
######
The fates of EST sequences of European perch (*Perca fluviatilis*) used for Type I marker development for yellow perch (*P. flavescens*).
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
The problem-solving competence is defined as the capacity to engage in cognitive processing to understand and resolve problem situations where a solution is not immediately obvious. It includes the willingness to engage in these situations in order to achieve one\'s potential as a constructive and reflective citizen (OECD, [@B25]; Kurniati and Annizar, [@B18]). Problem solving can be conceptualized as a sequential process where the problem solver must understand the problem, devise a plan, carry out the plan, and monitor the progress in relation to the goal (Garofalo and Lester, [@B11]; OECD, [@B24]). These problem-solving skills are key to success in all pursuits, and they can be developed in school through curricular subjects. Therefore, it is no surprise that the problem-solving competency is increasingly becoming the focus of many testing programs worldwide.
Advances in technology have expanded opportunities for educational measurement. Computer-based assessments, such as simulation-, scenario-, and game-based assessments, constantly change item design, item delivery, and data collection (DiCerbo and Behrens, [@B7]; Mislevy et al., [@B21]). These assessments usually provide an interactive environment in which students can solve a problem through choosing among a set of available actions and taking one or more steps to complete a task. All student actions are automatically recorded in system logs as coded and time-stamped strings (Kerr et al., [@B17]). These strings can be used for instant feedback to students, or for diagnostic and scoring purposes at a later time (DiCerbo and Behrens, [@B7]). And they are called process data. For example, the problem solving assessment of PISA 2012, which is computer-based, used simulated real-life problem situations, such as a malfunctioning electronic device, to analyze students\' reasoning skills, problem-solving ability, and problem-solving strategies. The computer-based assessment of problem solving not only ascertains whether students produce correct responses for their items, but also records a large amount of process data on answering these items. These data make it possible to understand students\' strategies to the solution. So far, to evaluate students\' higher order thinking, more and more large-scale assessments of problem solving become computer-based.
Recent research has focused on characterizing and scoring process data and using them to measure individual student\'s abilities. Characterizing process data can be conducted via a variety of approaches, including visualization, clustering, and classification (Romero and Ventura, [@B29]). DiCerbo et al. ([@B8]) used diagraphs to visualize and analyze sequential process data from assessments. Bergner et al. ([@B3]) used cluster analysis to classify similar behaving groups. Some other researchers used decision trees, neural networks, and Bayesian belief networks (BBNs) (Romero et al., [@B30]; Desmarais and Baker, [@B6]; Zhu et al., [@B38]), to classify the performance of problem solvers (Zoanetti, [@B39]) and to predict their success (Romero et al., [@B28]). Compared to characterizing process data, the research of scoring process data is very limited. Hao et al. ([@B14]) introduced "the editing distance" to score students\' behavior sequences based on the process data in a scenario-based task of the National Assessment of Educational Progress (NAEP). Meanwhile, these process data have been used in psychometric studies. Researchers analyzed students\' sequential response process data to estimate their ability by combining Markov model and item response theory (IRT) (Shu et al., [@B34]). It is noteworthy that all these practices have examined process data that describe students\' sequential actions to solve a problem.
All the actions, recorded as process level data, which are nested in individual level, are logically interconnected. This interdependency allows a straightforward modeling in a multi-level framework (Goldstein, [@B12]; Raudenbush and Bryk, [@B27]; Hox, [@B15]). This framework is similar to those used in longitudinal studies, yet with some differences. In longitudinal studies, measurements are typically consistent to show the development pattern of certain traits. For process data, however, actions are typically different within each individual. These successive actions are used to characterizing individuals\' problem solving strategies.
It is common in computer-based assessments that a nested data structure exists. To appropriately analyze process data (e.g., time series actions) within a nested structure (e.g., process within individuals), the multi-level IRT model can be modified by allowing process data to be a function of the latent traits at both process and individual levels. It is noteworthy that in the modified model, the concept of "item" in IRT changed to each action in individuals\' responses, which was scored based on certain rules.
With respect to the assessment of problem solving competency, the focus of this study is the ability estimate at the student level. We were not concerned with individual\'s ability reflected from each action at the process level, since the task needs to be completed by taking series actions. Even for individuals with high problem solving ability, the first few actions may not accurately reflect test takers\' ability. As a result, more attention was put on the development of ability at the process level because it can reveal students\' problem solving strategies. Mixture item response theory (MixIRT) models have been used in describing important effects in assessment, including the differential use of response strategies (Mislevy and Verhelst, [@B22]; Rost, [@B32]; Bolt et al., [@B4]). The value of MixIRT models lies in that they provide a way of detecting different latent groups which are formed by the dimensionality arising directly from the process data. These groups are substantively useful because they reflect how and why students responded the way they did.
In this study, we incorporated the multilevel structure into a mixture IRT model and used the modified multilevel mixture IRT (MMixIRT) model to detect and compare the latent groups in the data that have differential problem solving strategies. The advantage of this approach is the usage of latent groups. Although they are not immediately observable, these latent groups, which are defined by certain shared response patterns, can help explain process-level performance about how members of one latent group differ from another. The approach proposed in this study was used to estimate abilities both at process and student levels, and classify students into different latent groups according to their response strategies.
The goal of this study is to illustrate steps involved in applying the modified MMixIRT model in a computer-based problem solving assessment then to further present and interpret the results. Specifically, this article focuses on (a) describing and demonstrating the modified MMixIRT model using a task of PISA 2012 problem-solving process data; (b) interpreting the different action patterns; (c) analyzing the correlation between characteristics of different strategies and task performance, as well as some other operational variables such as the number of resets or clicks. All the following analysis was based on one sample data set.
Measurement material and dataset {#s2}
================================
Problem solving item and log data file
--------------------------------------
This study illustrates the use of the modified MMixIRT model in analyzing process data through one of the problem-solving tasks in PISA 2012 (Traffic CP007Q02). The task is shown in Figure [1](#F1){ref-type="fig"}. In this task, students were given a map and the travel time on each route, and then they were asked to find the quickest route from Diamond to Einsten, which takes 31 min.
{#F1}
The data are from the task\'s log file (CBA_cp007q02_logs12_SPSS.SAV, data source: <http://www.oecd.org/pisa/data/>) (an example of log data file is shown in Appendix [1](#SM1){ref-type="supplementary-material"}). The data file contains four variables associated with the process. The "event" variable refers to the type of event, which may be either system generated (start item, end item) or student generated (e.g., ACER_EVENT, Click, Dblclick). The "time" variable is the event time for this item, given in seconds since the beginning of the assessment, with all click and double-click events included. The "event_value" variable is recorded in two rows, as a click event involves selecting or de-selecting a route of the map. For example, in the eleventh row where the state of the entire map is given, 1 in the sequence means that the route was selected, and 0 means that it was not; the twelfth row records an event involving highlighting, or un-highlighting. A route of the map represents the same click event, and it is in the form "hit_segment name" (The notes on log file data can be downloaded from <http://www.oecd.org/pisa/data/>). All the "click" and "double-click" events represent that a student performs a click action that is not related to select a route. Table [1](#T1){ref-type="table"} shows the label, the route and the correct state of the entire selected routes.
######
The routes of the map.
**Label** **Route** **Included or not in the correct routes**
----------- ------------------ -------------------------------------------
P1 Diamond-Nowhere 1
P2 Diamond-Silver 0
P3 Emerald-Lincoln 0
P4 Emerald-Unity 0
P5 Lee-Mandela 1
P6 Lincoln-Sato 0
P7 Mandela-Einstein 1
P8 Market-Lee 1
P9 Market-Park 0
P10 Nobel-Lee 0
P11 Nowhere-Einstein 0
P12 Nowhere-Emerald 0
P13 Nowhere-Sakharov 1
P14 Nowhere-Unity 0
P15 Park-Mandela 0
P16 Park-nowhere 0
P17 Sakharov-Market 1
P18 Sakharov-Nobel 0
P19 Sato-nowhere 0
P20 Silver-Market 0
P21 Silver-nowhere 0
P22 Unity-Park 0
P23 Unity-Sato 0
*1, Yes; 0, No*.
Sample
------
The study sample was drawn from PISA 2012 released dataset, consisting of a total of 413 students from 157 American schools who participated in the traffic problem-solving assessment (47.2% as females). The average age of students was 15.80 years (*SD* = 0.29 years), ranging from 15.33 to 16.33 years.
For the traffic item response, the total effective sample size under analysis was 406, after excluding seven incomplete responses. For the log file of the process record, there were 15,897 records in the final data file, and the average record number for each student was 39 (*SD* = 33), ranging from 1 to 183. The average response time was 672.64 s (*SD* = 518.85 s), ranging from 58.30 to 1995.20 s.
The modified mmixirt model for process data {#s3}
===========================================
Process-level data coding
-------------------------
In this task log file, "ACER_EVENT" is associated with "click." However, in this study we only collected the information of ACER_EVENT and deleted the redundant click data. Then, we split and rearranged the data by routes, making each row represent a step in the process of individual students, and each column represent a route (0 for de-selecting, and 1 for selecting). Table [2](#T2){ref-type="table"} shows part of the reorganized data file, indicating how individual student selected each route in each step. For example, the first line represents that student 00017 selected P2 in his/her first step.
######
Example of the reorganized data file.
**StIDStd** **Time** **Event_number** **Event_value** **P1** **P2** **P3** **P4** **P5** **P6** **P7** **P8** **...** **P21** **P22** **P23**
------------- ---------- ------------------ --------------------------- -------- -------- -------- -------- -------- -------- -------- -------- --------- --------- --------- ---------
00017 837.6000 2.00 \'01000000000000000000000 0 1 0 0 0 0 0 0 **...** 0 0 0
00017 839.8000 4.00 \'11000000000000000000000 1 1 0 0 0 0 0 0 **...** 0 0 0
00017 841.1000 7.00 \'11000000000010000000000 1 1 0 0 0 0 0 0 **...** 0 0 0
00017 841.7000 9.00 \'11000000000010000100000 1 1 0 0 0 0 0 0 **...** 0 0 0
00017 842.7000 11.00 \'11000000010010000100000 1 1 0 0 0 0 0 0 **...** 0 0 0
00017 844.8000 13.00 \'11000000010010000101000 1 1 0 0 0 0 0 0 **...** 0 0 0
00017 846.4000 15.00 \'11000000010000000101000 1 1 0 0 0 0 0 0 **...** 0 0 0
00017 847.4000 17.00 \'01000000010000000101000 0 1 0 0 0 0 0 0 **...** 0 0 0
00017 848.4000 19.00 \'01000000010000000001000 0 1 0 0 0 0 0 0 **...** 0 0 0
00017 850.6000 21.00 \'01000000000000000001000 0 1 0 0 0 0 0 0 **...** 0 0 0
00017 851.6000 23.00 \'01000000010000000001000 0 1 0 0 0 0 0 0 **...** 0 0 0
00017 852.5000 25.00 \'01000000000000000001000 0 1 0 0 0 0 0 0 **...** 0 0 0
00017 853.4000 27.00 \'01000000100000000001000 0 1 0 0 0 0 0 0 **...** 0 0 0
00017 853.7000 29.00 \'01000000100000010001000 0 1 0 0 0 0 0 0 **...** 0 0 0
Process data were first recoded for the analysis purpose. Twenty-three variables were created to represent a total number of available routes that can possibly be selected (similar to 23 items). The right way for solving this problem is to select the following six routes: Diamond--Nowhere--Sakharov--Market--Lee--Mandela--Einstein (i.e., P1, P5, P7, P8, P13, and P17). For the correct routes, the scored response was 1 if one was selected, and 0 otherwise; for the incorrect routes, the scored response was 0 if one was selected, and 1 otherwise. Each row in the data file represents an effective step (or action) a student took during the process. In each step, when a route was selected or not, the response for this route was recoded accordingly. When a student finished an item, all the steps during the process were recorded. Therefore, for the completed data set, the responses of the 23 variables were obtained and the steps were nested within students.
The modified MMixIRT model specification
----------------------------------------
The MMixIRT model has mixtures of latent classes at the process level or at both process and student levels. It assumes that possible heterogeneity exists in response patterns at the process level and therefore are not to be ignored (Mislevy and Verhelst, [@B22]; Rost, [@B32]). Latent classes can capture the interactions among the responses at the process level (Vermunt, [@B36]). It is interesting to note that if no process-level latent classes exist, there are no student-level latent classes, either. The reason lies in that student-level units are clustered based on the likelihood of the processes belonging to one of the latent classes. For this particular consideration, the main focus in this study is to explore how to classify the process-level data, and the modified MMixIRT model only focus on latent classes at the process level.
The MMixIRT model accounts for the heterogeneity by incorporating categorical or continuous latent variables at different levels. Because mixture models have categorical latent variables and item response models have continuous latent variables, latent variables at each level may be categorical or continuous. In this study, the modified MMixIRT includes both categorical (latent class estimates) and continuous latent variables at the process level and only continuous (ability estimates) latent variables at the student level.
The modified MMixIRT model for process-level data is specified as follows:
Process-Level
P
(
y
j
k
i
=
1
\|
θ
j
k
g
,
C
j
k
=
g
)
=
exp
(
α
i
g
.
W
θ
j
k
g
\-
β
i
g
)
1
\+
exp
(
α
i
g
.
W
θ
j
k
g
\-
β
i
g
)
P
(
y
j
k
1
=
ω
1
,
y
j
k
2
=
ω
2
,
⋯
,
y
j
k
I
=
ω
I
)
=
∑
g
=
1
G
γ
j
k
g
∏
i
=
1
I
P
(
y
j
k
i
=
1
\|
θ
j
k
g
,
C
j
k
=
g
)
ω
i
(
1
\-
P
(
y
j
k
i
=
1
\|
θ
j
k
g
,
C
j
k
=
g
)
)
(
1
\-
ω
i
)
Student-Level
P
(
y
k
i
=
1
\|
θ
k
)
=
exp
(
α
i
.
B
θ
k
\-
β
i
)
1
\+
exp
(
α
i
.
B
θ
k
\-
β
i
)
For the process level, in Equation (1), ***i*** is an index for *i*th route (*i* = 1, ..., *I*), ***k*** is an index for a student (*k* = 1,..., *K*), ***j*** is an index for the *j*th valid step of a student during the response process (*j* = 1, ..., *J*~*k*~),(*J* is the total steps of the *k*th student) and *g* indexes the latent classes (*C*~*jk*~ = 1, ..., *g*...*G*, where *G* is the number of latent classes), ***C***~*jk*~ is a categorical latent variable at the process level for the *j*th valid step of student *k*, which captures the heterogeneity of the selections of routes in each step. *P*(*y*~*jki*~ = 1\|θ~*jkg*~, *C*~*jk*~ = *g*) is the probability of selecting an route *i* in the *j*th step of student *k*, which is predicted by the two-parameter logistic (2PL) model, and α~*ig*.*W*~ is the discrimination parameter of process-level in class *g, W* means within-level, β~*ig*~ is the location parameter in class *g*, and θ~*jkg*~ is the latent ability of examinee ***k*** for a specific step ***j*** during the process of selecting the route, which is called the process ability in this study (θ~*jkg*~ \~***N(***μ~*jkg*~**,** $\sigma_{jkg}^{2}$***)***). The process abilities across different latent classes are constrained to follow a normal distribution ***(***θ~*jk*~ \~***N(0, 1))*.** In Equation (2), *P*(*y*~*jk*1~ = ω~1~, *y*~*jk*2~ = ω~2~, ⋯ , *y*~*jkI*~ = ω~*I*~) is the joint probability of the actions in the *j*^th^ step of student *k*. ω~*i*~ denotes either selected or not selected for *i*th route. For the correct routes, 1 represents that the route was selected, and 0 otherwise; for the incorrect routes, 0 represents that the route was selected, and 1 otherwise. γ~*jkg*~ is the proportion of the *j*th step in each latent class and $\sum\limits_{g = 1}^{G}\gamma_{jkg} = 1$. As can be seen from the Equation (2), the probability of the actions (*y*~*jki*~) are assumed to be independent from each other given class membership, which is known as the local independence assumption for mixture models.
For the student level, in Equation (3), α~*i*.*B*~ is the item discrimination parameter where ***B*** represents between-level. β~*i*~ is the item location parameter which is correlated with the responses of the final step of the item. θ~*k*~ is the ability estimate at the student level based on the final step of the process, which also represents the problem-solving ability of student ***k*** in this study ***(***θ~*k*~ \~***N(0*,** ***1))***.
Figure [2](#F2){ref-type="fig"} demonstrates a modified two-level mixture item response model with within-level latent classes. The squares in the figure represent item responses, the ellipses represent latent variables, and 1 inside the triangle represents a vector of 1 s. As is shown in the figure, the response for each route of the jth step \[***y***~*jk*1~**,...,** ***y***~*jki*~**,...,** ***y***~*jkI*~\] is explained by both categorical and continuous latent variables (***C***~*jk*~ and θ~*jkg*~, respectively) at the process level; and the final response of students for each route \[***y***~*k*1~**,...,** ***y***~*ki*~**,...,** ***y***~*kI*~\] is explained by a continuous latent variable (θ~*k*~) at the student level. The arrows from the continuous latent variables to the item (route) represent item (route) discrimination parameters (α~*ig,\ W*~ at the process level and α~*i,\ B*~ at the student level), and the arrows from the triangle to the item responses represent item location parameters at both levels. The dotted arrows from the categorical latent variable to the other arrows indicate that all item parameters are class-specific.
{#F2}
It should be noted that the MMixIRT model is different from the traditional two-level mixture item response model in the definition of the latent variables at the between-level. In the standard MMixIRT model, the between-level latent variables are generally obtained from the measurement results made by within-level response variables \[***y***~*jk*1~**,...,** ***y***~*jki*~**,...,** ***y***~*jkI*~\] on between-level latent variables (Lee et al., [@B19]). In this study, the process-level data mainly reflect the strategies for problem solving, while the responses at the last step represent students\' final answers on this task. Therefore, students\' final responses are used to estimate their problem-solving abilities (latent variable at the between-level, i.e., ability of the student level) in the modified MMixIRT model.
Mplus Software (Muthén and Muthén, [@B23]) was used to estimate the parameters of the modified MMixIRT model, as specified above. In addition, the detailed syntax are presented in Appendix [5](#SM1){ref-type="supplementary-material"}.
Results {#s4}
=======
Results of descriptive statistics
---------------------------------
Table [3](#T3){ref-type="table"} shows the proportion of each route selected by the students in the correct group and in the wrong group, respectively. The correct group consists of students who selected the right routes, and the wrong group refers to students who failed to do so. There are a total of 476 students, with 377 in the correct group and 99 in the wrong group. The results show that most of the students in the correct group selected the right routes, while a large number of students in the wrong group selected the wrong routes. To further explore the differences of the proportion of students selecting the wrong routes in the two groups, χ^2^-tests were conducted. No significant differences were found between the correct group and the wrong group in terms of the proportion of students who clicked four wrong routes, including P4 \[χ^2^~(1)~ = 0.370, *P* \> 0.05\], P9 \[χ^2^~(1)~ = 3.199, *P* \> 0.05\], P10 \[χ^2^~(1)~ = 3.636, *P* \> 0.05\], and P15 \[χ^2^~(1)~ = 2.282, *P* \> 0.05\]. This further suggests that it was difficult for the correct group to avoid these routes during their response process, and even quite a number of students in the correct group experienced trial and error before eventually solving the problem.
######
The proportion of route selection.
**Route** **Selected proportion**
----------- ------------------------- ------------
**P1** **40.023** **69.504**
P2 38.158 19.872
P3 3.290 1.688
P4 0.635 0.815
**P5** **16.055** **25.148**
P6 2.481 1.287
**P7** **15.699** **22.260**
**P8** **4.340** **21.953**
P9 25.379 23.435
P10 12.586 12.007
P11 16.559 10.819
P12 4.304 2.601
**P13** **36.846** **64.109**
P14 8.404 3.622
P15 5.182 6.886
P16 19.122 12.771
**P17** **16.653** **43.530**
P18 17.629 13.157
P19 4.884 1.923
P20 17.579 10.732
P21 15.369 7.211
P22 5.531 1.759
P23 4.296 1.377
*The right routes are printed in bold*.
Results of the modified MMixIRT model
-------------------------------------
### Model selection
The determination of the number of latent classes has been discussed in many studies (Tofighi and Enders, [@B35]; Li et al., [@B20]; Peugh and Fan, [@B26]). Several statistics of the mixture IRT models are often computed to compare relative fits of these models. Akaike\'s ([@B1]) information criterion (AIC) incorporates a kind of penalty function for over-parameterization on model complexity. A criticism of AIC has been that it is not asymptotically consistent because the sample size is not directly involved in its calculation (Janssen and De Boeck, [@B16]; Forster, [@B9]). Schwarz ([@B33]) proposed BIC as another information-based index, which attains asymptotic consistency by penalizing over-parameterization by using a logarithmic function of the sample size. For the sample size in BIC, the number of persons is used in multilevel model (Hamaker Ellen et al., [@B13]) and in multilevel item response model (Cohen and Cho, [@B5]). Most studies suggested the BIC value as the best choice because it was a sample-based index that also penalized the sophisticated model. However, Tofighi and Enders ([@B35]) indicated in their simulation study that a sample size-adjusted BIC (aBIC) was an even better index. Smaller AIC, BIC, and aBIC values indicate a better model fit for mixture IRT models. Besides, entropy value has been used to measure how well a mixture model separates the classes; an entropy value close to 1 indicates good classification certainty (Asparouhov and Muthén, [@B2]).
The model selection results for the modified MMixIRT models are given in Table [4](#T4){ref-type="table"}. The model fit indicates that *LL, AIC, BIC*, and *aBIC* decreased consistently as the class number increased to eight classes, and the nine-class model did not converge. As noted above, the best fit for *AIC, BIC*, and *aBIC* was determined or dictated by the smallest value in the ordered set of models from the least to the most complex. As suggested by Rosato and Baer ([@B31]), selecting a robust latent class model is a balance between the statistical result of the model fit and the substantive meaning of the model. The model that fits best and yields meaningful classes should be retained. In this study the proportions of latent classes were examined to ensure the empirical significance, and the interpretability of each class was considered accordingly. For the 6-class model, the proportion of each class was 18.1, 30.7, 18.1, 20.1, 7.2, and 5.9%. And for the 7-class model, the proportion was 19.9, 13.4, 6.0, 12.3, 13.5, 27.4, and 7.5%. Compared to the 6-class model, in the 7-class model, the extra class of the steps was similar to class 2 of the 6-class model, while mixing class 4 at the same time. This makes the 7-class model hard to interpret. For the 8-class model, the proportion of one of the classes was too small (only 2.7%). Taking into account both the model fit index and the interpretability of each class, the 6-class model was retained in this study.
######
Model comparison and selection.
**No of class** **No of Free parameters** ***LL Value*** ***Akaike (AIC)*** ***Bayesian (BIC)*** ***Sample-Size Adjusted BIC*** ***Entropy***
----------------- --------------------------- ---------------- -------------------- ---------------------- -------------------------------- ---------------
1 46 −112745.581 225583.161 225936.108 225789.923
2 95 −99334.232 198858.463 199587.375 199285.472 0.957
3 144 −92723.338 185734.676 186839.552 186381.931 0.860
4 193 −89375.035 179134.070 180607.239 179997.077 0.920
5 242 −87186.912 174857.823 176714.629 175945.571 0.936
6 291 −85974.117 172530.234 174763.005 173838.228 0.908
7 340 −84864.882 170409.764 173018.500 171938.004 0.904
8 389 −83821.533 168421.066 171405.766 170169.552 0.893
### Description of class characteristics
The most likely latent class membership are displayed in Table [5](#T5){ref-type="table"}. In this matrix, steps from each class have an average probability of being in each class. Large probabilities are expected on the diagonal. The numbers on diagonal are greater than 0.9. It can be concluded from the results that the modified MMixIRT model can classify students properly based on process data.
######
Most likely latent class membership of each class.
**Most likely latent class membership**
--------- ----------------------------------------- ------- ------- ------- ------- -------
Class 1 0.945 0.000 0.006 0.033 0.004 0.012
Class 2 0.001 0.936 0.002 0.033 0.013 0.015
Class 3 0.002 0.020 0.949 0.011 0.017 0.001
Class 4 0.029 0.004 0.007 0.949 0.002 0.010
Class 5 0.002 0.007 0.018 0.002 0.969 0.002
Class 6 0.016 0.014 0.001 0.025 0.002 0.942
Figure [3](#F3){ref-type="fig"} presents the characteristics of route selection for each class based on the 6-class mixture IRT model, with ➀, ➁, ➂....indicating the order of the routes. Based on the results of the modified MMixIRT model, the number of clicks of the 23 routes (P1--P23) in each class is listed in Appendix [2](#SM1){ref-type="supplementary-material"}. The characteristics of route selection can be obtained pursuant to routes that get more clicks than others in each class, as well as the relations among routes shown in Figure [1](#F1){ref-type="fig"}. For example, P17, P13, P1, P8, P5, P16, and P7 in Class 1 were clicked more than other routes; however, Figure [1](#F1){ref-type="fig"} shows that there is no obvious relationship between P16 and other routes. Therefore, the characteristic of Class 1 was defined as P1-P13-P17-P8-P5-P7 and P16 was removed. These routes were sequenced by the number of clicks they got, with the most clicked routes taking the lead. As indicated in Figure [3](#F3){ref-type="fig"}, different latent classes have typical characteristics depending on the similarity of the correct answers. For example, the route selection strategy of Class 1 best approximated the ideal route required by the item. Based on their last click, almost all the students in Class 1 gave the correct answer. Therefore, Class 1 could be regarded as the correct answer class, while the rest classes took different wrong routes.
{#F3}
The numbers in circles (➀, ➁, ➂....) indicate the order of the routes.
As is illustrated in Table [6](#T6){ref-type="table"}, different classes demonstrated different means of process-level ability. It is obvious that the mean process ability in Class 1 is the highest (0.493), followed by Class 6, Class 2, Class 4, yet Class 5 and Class 3 with the lowest process-level ability. A closer check of these classes in Figure [3](#F3){ref-type="fig"} indicates that the selected routes of Class 5 and Class 3 were incredibly far away from the correct one, and they took far more than 31 min. Therefore, it is no surprise that the mean process-level ability estimates of these two classes were the lowest and were both negative (−1.438 and −0.935, respectively). In addition, as can be seen in the number of students, almost all the students in Class 1 provided the right answer, demonstrating that different latent classes had different probabilities of the correct answer. In summary, the process-level ability is different across latent classes, which is related to different strategies of students\' route selection or cognitive process.
######
Means and standard deviations of process level abilities.
**Latent class size for process-level** **No of Students** **Process-level ability**
--------- ----------------------------------------- -------------------- --------------------------- ---- -------- -------
Class 1 2875 18.1 307 3 0.493 0.678
Class 2 4867 30.7 0 41 0.323 0.903
Class 3 2867 18.1 0 14 −0.935 0.386
Class 4 3192 20.1 0 26 0.292 0.556
Class 5 1138 7.2 0 12 −1.438 0.404
Class 6 940 5.9 0 3 0.424 0.698
Total 15879 100 307 99 0.000 0.934
*In the column of no of Students, the last step of the process within each student is classified into one of the six latent classes. Then, the numbers of students who gave the correct or wrong answer are summarized based on the latent classes*.
### The sequence of latent classes at the process level
Based on the results of the modified MMixIRT model, the characteristics of the strategy shifts between step-specific classes were explored and summarized. To capture the characteristics of students\' strategy shifts during the response, it is necessary to identify the typical route selection strategy of each class in the first place. In this study, if a student applied the strategy of a certain class three or more times consecutively, it was considered that the student had employed the strategy of this class at the process level. Three times was chosen as the rule of thumb because it demonstrated enough stability to classify a solution behavior. Then the strategy shifts of each student during their clicking procedure could be obtained in orders. The typical route selection strategy of different classes and the class shifts of students in the correct group are presented in Appendixes [3](#SM1){ref-type="supplementary-material"}, [4](#SM1){ref-type="supplementary-material"}, respectively. The results in Appendix [4](#SM1){ref-type="supplementary-material"} provide useful and specific information about the strategy shifts used by students over time. For example, in the correct group, 58 students shifted from one class to another, including 22 from Class 2 to Class 1, 3 from Class 3 to Class 1, 30 from Class 4 to Class 0, and 3 from Class 6 to Class 1. It is noteworthy that when students did not apply any strategies for more than three times consecutively, it was regarded as class 0 in this study.
The relationship of the two level ability estimates and operational variables
-----------------------------------------------------------------------------
To validate whether students with different patterns of actions will have different process-level ability, the descriptive statistics were conducted of operational variables such as the number of route clicks and resets and their correlation with the mean ability estimate of process-level ability (See Table [7](#T7){ref-type="table"} for details). To further explore the differences of click actions between the correct group and the wrong group, several *T*-tests were conducted. The results indicate that students in the correct group did significantly fewer resets than their counterparts in the wrong group \[*t*~(404)~ = 2.310, *P* \< 0.05\]. No significant differences were detected of the number of routes clicked or the response time between the correct group and the wrong group \[*t*~(404)~ = 1.656, *P* = 0.099; *t*~(404)~ = −0.199, *P* = 0.843\]. The results in Table [7](#T7){ref-type="table"} suggest two things. Firstly, positive correlation existed between the estimate of student-level ability and that of process-level ability. This means that the process-level ability estimate provides consistency and auxiliary diagnostic information about the process. The students with higher process-level ability had higher ability estimates of student level. Secondly, for the process-level ability, a significant negative correlation existed between the mean process-level ability estimate and variables such as the valid number of route clicks and the number of resets for students in the correct group. It is concluded that in the correct group, the less frequently a student clicks the routes and resets the whole process, the higher process-level ability he or she is likely to obtain. For students in the wrong group, however, no significant correlations were observed between the mean ability estimate and the variables discussed above. Instead, a significant negative correlation was found between the mean process-level ability estimate and the absolute time of difference from 31 min. For these students, their process-level ability decreased as the time cost by the wrong routes increased. Third, the mean process-level ability estimate for the correct group was 0.310, in contrast to −0.175 for the wrong group, which reveals a significant difference between the two groups \[*t*~(404)~ = 8.959, *P* \< 0.001\]. In terms of student-level ability, the estimate for the correct group was significantly higher than for the wrong group \[*t*~(404)~ = 112.83, *P* \< 0.001\].
######
Correlation between ability estimates and operational variables in process.
**Item response result** **Click action variable** **Mean ability of process level** **Ability of student level** **Mean** ***SD***
-------------------------- -------------------------------------- ------------------------------------------- --------------------------------------------- ---------- ----------
Correct (*N* = 307) No of Route Clicks −0.657[^\*\*^](#TN2){ref-type="table-fn"} / 79.760 63.874
No of Resets −0.467[^\*\*^](#TN2){ref-type="table-fn"} / 0.919 1.737
Absolute Time of Difference from 31 **/** / 0.000 0.000
Response Time 0.048 / 675.540 525.710
Mean Ability of Process Level / / 0.310 0.447
Ability of Student Level **/** / 1.371 0.000
Wrong (*N* = 99) No of Route Clicks −0.050 0.142 93.030 84.138
No of Resets −0.124 0.098 1.394 1.910
Absolute Time of Difference from 31 −0.248[^\*^](#TN1){ref-type="table-fn"} −0.179 5.210 10.869
Response Time −0.087 0.022 663.620 499.466
Mean Ability of Process Level / 0.597[^\*\*^](#TN2){ref-type="table-fn"} −0.175 0.530
Ability of Student Level 0.597[^\*\*^](#TN2){ref-type="table-fn"} / −0.432 0.281
Total (*N* = 406) No of Route Clicks −0.439[^\*\*^](#TN2){ref-type="table-fn"} −0.066 83.000 69.484
No of Resets −0.378[^\*\*^](#TN2){ref-type="table-fn"} −0.103[^\*^](#TN1){ref-type="table-fn"} 1.035 1.790
Absolute Value of Difference from 31 −0.269[^\*\*^](#TN2){ref-type="table-fn"} −0.407[^\*\*\*^](#TN3){ref-type="table-fn"} 1.300 5.802
Response Time 0.015 0.012 672.640 518.849
Mean Ability of Process Level / 0.454[^\*\*\*^](#TN3){ref-type="table-fn"} 0.192 0.512
Ability of Student Level 0.454[^\*\*^](#TN2){ref-type="table-fn"} / 0.931 0.787
Correct Responses 0.407[^\*\*^](#TN2){ref-type="table-fn"} 0.985[^\*\*\*^](#TN3){ref-type="table-fn"} 0.756 0.430
"*31" indicated in "absolute value of difference from 31" in Column 8 refers to the time taken in walking the right route for the item CP007Q02*.
*p \< 0.05*,
*p \< 0.01*,
*p \< 0.001*.
The result in Table [8](#T8){ref-type="table"} indicates that the sequence of latent classes are consistent with the ability estimates at both process and student levels. For students in the correct group, the mean process-level ability estimate decreased as the number of class shifts, clicks and resets increased. Students with higher process-level ability tended to select the correct route immediately or after a few attempts. Consequently, these students clicked and reset for fewer times because they had a clearer answer in mind and therefore were more certain about it. In contrast, for students in the wrong group, the mean ability estimates at both process and student levels were rather small when the number of class shifts were 0 and 1. When the number of class shifts was 0, students failed to stick with a specific strategy to solve the problem during the process. It took them a longer response time with about two resets on average; as a result, the time cost for their route selection was nearly twice the target time. When the number of class shifts was 1, these students simply stuck to a totally wrong route for the entire time, with shorter response time and fewer numbers of clicks. However, unlike the correct group, the number of class shifts in the wrong group showed a non-linear relationship with the mean ability at both process and student levels. At first, when the number of class shifts increased from 0 to 4, the ability estimates at both levels increased as well. The explanation was that because these students figured out the right routes, they should have higher abilities than the 0 shift group that sticks to the wrong route all the time. For example, students with four shifts all ended up using strategy of Class 1, which was the right strategy class (Appendix [4](#SM1){ref-type="supplementary-material"}). Therefore, they were supposed to have the highest process ability in the wrong group. However, when the number of class shifts increased from 5 to 6, the process-level ability estimate dropped. This has much to do with the fact that too many shifts reflected little consideration and a lack of deep cognitive processing.
######
Ability estimates and the operational variables in the different numbers of class shifts in the correct group and wrong group.
**Correct or wrong answer group** **No of class shifts** **No of students** **Process-level ability (Mean)** **Student-level ability (Mean)** **Response time (Mean)** **Valid No of click (Mean)** **Absolute value of difference from 31 (Mean)** **No of Reset (Mean)**
----------------------------------- ------------------------ -------------------- ---------------------------------- ---------------------------------- -------------------------- ------------------------------ ------------------------------------------------- ------------------------
Correct group (*N* = 307) 1 32 0.650 1.371 714.941 19.375 0 0.156
2 58 0.692 1.371 609.116 31.655 0 0.121
3 69 0.468 1.371 814.619 60.667 0 0.275
4 73 0.196 1.371 601.215 93.795 0 1.192
5 63 −0.141 1.371 649.711 134.143 0 1.937
6 12 −−0.279 1.371 679.617 212.25 0 3.5
Wrong group (*N* = 99) 0 11 −0.453 −0.548 991.7 36.909 29.182 2.091
1 15 −0.439 −0.552 377.713 22.867 1.067 1.067
2 12 0.139 −0.312 470.392 37.75 1.417 0.5
3 12 0.466 −0.275 552.042 71.917 0.917 0.667
4 20 −0.151 −0.438 784.455 94.4 1.250 0.85
5 24 −0.343 −0.492 690.038 170.292 5.042 2.292
6 5 −0.348 −0.162 921.02 234 1.000 2.6
Discussion {#s5}
==========
A modified MMixIRT model was described for modeling response data at process and student levels. The model developed in this study combined the features of an IRT model, a latent class model, and a multilevel model. The process-level data provide an opportunity to determine whether latent classes or class shifts differ in their response strategies to solve the problem. The student-level data can be used to account for the differences of students\' problem solving abilities. The ability estimate at both process and student levels are different across latent classes. The modified MMixIRT model makes it possible to describe differential strategies based on process-level and student-level characteristics. If a student\'s specific strategies and their strengths and weaknesses can be described in the process of solving a problem, then the assessment of a student\'s proficiency in problem solving can guide instructional interventions in target areas.
As process data from various computer-based assessment or educational learning system have become common, there is an urgent call for analyzing such data in an accurate way. The psychometrical model-based approach has a great potential in this aspect. Latent classes and the characteristics of latent class shifts obtained from process data can reveal students\' reasoning skills in problem-solving. The findings of characteristics of process-level latent classes make it easy to uncover meaningful and interesting action patterns from the process data, and to compare patterns from different students. These findings provide valuable information to psychometricians and test developers, help them better understand what distinguishes successful students from unsuccessful ones, and eventually lead to better test design. In addition, as shown in this study, some operational variables such as the number of resets and the number of clicks or double clicks are related to the ability estimates at both process and student levels and therefore can predict student scores on problem solving assessment. Since students\' different abilities capture individual patterns in process data, it can be used to score or validate the rubrics. Williamson et al. ([@B37]) explain that a "key to leveraging the expanded capability to collect and record data from complex assessment tasks is implementing automated scoring algorithms to interpret data of the quantity and complexity that can now be collected" (p. 2).
The extension of the modified MMixIRT approach proposed in this study can be implemented in several ways. Firstly, it can be simplified in removing the process-level ability parameters, and also be extended to include student-level latent classes instead of abilities. Secondly, one of the advantages of this proposed model is that item parameters can be constrained to be equal across the process-level and student-level. So the abilities of both levels are on the same scale and can be compared and evaluated. Lastly, the main benefits of multilevel IRT modeling lie in the possibility of estimating the latent traits (e.g., problem solving) at each level. More measurement errors can be accounted for by considering other relevant predictors such as motivations (Fox and Glas, [@B10]).
The psychometrical model-based approach also has its limitations. First, even though latent class shifts preserve the sequential information in action series, they do not capture all the related information. For instance, for the purpose of convenient analysis in this study, some unstable characteristics of a latent class such as random shifts were not used in our definition of class characteristics and class shifts. Fortunately, in many cases, as in this study, this missing information does not affect the results. If it becomes an issue in some cases, it can be addressed by considering more details about the latent class shifts to minimize the ambiguity. Second, this study only takes a single route as an analysis unit, yet failing to consider possible route combinations. For example, in some cases two routes are available, it makes full sense to combine these two routes into one to conduct analysis, because the link between these routes is exclusive. In the future, we may consider the transition model for different route combinations, such as Bi-Road. In terms of the generalizability of the modified MMixIRT model for solving complicated problems, if the process data for another single task can be recoded or restructured as the data file in this study, similar models can be applied to explore the latent classes and characteristics of the problem solving process. However, the difficulty during the analysis lies in how to recode the responses into dichotomous data. For multiple tasks, a three-level model can be applied, with the first level as the process level, the second as the task level and the third as the student level. If there are plenty of tasks, the ability estimates of the student will stay stable. Therefore, while the generalizability of the model may be conditional, the main logic of the MMixIRT approach can be generalized.
Author contributions {#s6}
====================
HL research design, data analysis, and paper writing. YL paper writing. ML data analysis, and paper writing.
Conflict of interest statement
------------------------------
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Funding.** Supported by National Natural Science Foundation of China (31571152); Special Found for Beijing Common Construction Project (019-105812).
Supplementary material {#s7}
======================
The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpsyg.2018.01372/full#supplementary-material>
######
Click here for additional data file.
[^1]: Edited by: Qiwei He, Educational Testing Service, United States
[^2]: Reviewed by: Yunxiao Chen, Emory University, United States; Matthias Stadler, University of Luxembourg, Luxembourg
[^3]: This article was submitted to Quantitative Psychology and Measurement, a section of the journal Frontiers in Psychology
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
The production of haploid gametes is required for sexual reproduction. These gametes are produced by meiosis, a specialized cell division program during which a single round of DNA replication is followed by two rounds of chromosome segregation (the meiotic divisions), Meiosis I (MI) and Meiosis II (MII). In contrast, mitotically--dividing cells maintain their ploidy by strictly alternating rounds of DNA replication and chromosome segregation. The lack of DNA replication between MI and MII is essential for the reduction in ploidy inherent to meiosis, but it is unclear how the meiotic program differs from mitosis to allow for two sequential chromosome segregation events without an intervening S phase.
In mitotic cells, both DNA replication and chromosome segregation require cyclin-dependent kinase (CDK) activity to oscillate during the cell cycle. A low--CDK state during G1 phase allows both events to initiate, and a high--CDK state is required for their completion. During meiosis, the CDK--oscillation dependence of both events presents a unique problem between MI and MII, a period known as the MI--MII transition ([Figure 1A](#fig1){ref-type="fig"}). After MI has been completed, CDK activity decreases, and then increases again upon entry into MII ([@bib13]). This oscillation is required for multiple essential chromosome--segregation events, including duplication of the spindle pole body (SPB, the yeast centrosome) ([@bib11]; [@bib27]; [@bib45]). However, the DNA replication program must remain inhibited between MI and MII to achieve the hallmark of meiosis, reductive cell division. Given that an oscillation of CDK activity is sufficient for re--replication of the entire genome in mitotic cells ([@bib20]), it is not fully understood how meiotic cells reset the chromosome segregation program while retaining inhibition of the DNA replication program.
![Mcm2--7 loading onto replication origins is inhibited during the MI--MII transition.\
(**A**) The DNA replication program and chromosome segregation program are uncoupled during the MI--MII transition. Relative CDK activity at various stages of the meiotic cell cycle are shown ([@bib13]). The dashed boxes highlight the oscillations of low-to-high CDK activity during meiosis, and show the discrepancy between CDK-regulation of SPB duplication and DNA replication. See text for details. (**B**) ORC is bound to origins of replication throughout the meiotic divisions. The strain yDP71 was put through meiosis. ChIP--qPCR was used to detect ORC binding at the early--firing origin *ARS1* (top graph, dark blue), the late--firing origin *ARS1413* (top graph, light blue), and at the re--replication prone origins ARS305 (bottom graph, dark blue) and ARS418 (bottom graph, light blue). The time after transfer into sporulation medium and the associated meiotic stages are indicated below each lane. For cell--cycle stage quantification for this experiment, see [Figure 1---figure supplement 1A](#fig1s1){ref-type="fig"}. The peak % of input DNA immunoprecipitated (set to arbitrary unit (A.U.) =1.0) was 9.1% for ARS1, 2.2% for ARS1413, 46.7% for ARS305, and 10.9% for ARS418. (**C**) Mcm2--7 is bound to origins of replication in G1 phase but does not reassociate with origins during or between the meiotic divisions. The strain yDP71 was put through meiosis. ChIP--qPCR was used to detect Mcm2--7 binding at the early--firing origin *ARS1* (top graph, red), the late--firing origin *ARS1413* (top graph, orange), and at the re--replication prone origins ARS305 (bottom graph, red) and ARS418 (bottom graph, orange). The time after transfer into sporulation medium and the associated meiotic stages are indicated below each lane. For cell--cycle stage quantification for this experiment, see [Figure 1---figure supplement 1B](#fig1s1){ref-type="fig"}. The peak % of input DNA immunoprecipitated (set to A.U. = 1.0) was 7.4% for ARS1, 5.0% for ARS1413, 35% for ARS305, and 19.5% for ARS418.\
10.7554/eLife.33309.006Figure 1---source data 1.Raw values used for the quantification of [Figure 1B and C](#fig1){ref-type="fig"}.](elife-33309-fig1){#fig1}
Mitotic cells use oscillations of CDK activity to ensure that the genome is replicated exactly once per cell division. During G1 phase, low CDK activity allows for the Mcm2--7 complex, the core enzyme of the replicative helicase, to be loaded onto origins of replication in an inactive state. This event, known as origin licensing or helicase loading, cannot occur in the presence of high CDK activity and requires the cooperative action of three proteins: Cdc6, Cdt1, and the Origin Recognition Complex (ORC) ([@bib25]; [@bib56]). Upon S--phase entry, S--CDK (CDK bound to S-phase cyclins Clb5/6) is activated and impacts DNA replication in two ways. First, S--CDK phosphorylates two essential proteins, Sld2 and Sld3, that subsequently promote helicase activation, replisome assembly, and chromosome duplication ([@bib46]; [@bib63]; [@bib72]). Second, both S--CDK and M-CDK (CDK bound to mitotic cyclins Clb1-4) inhibit new helicase loading during S, G2, and M phases. These kinases directly phosphorylate Cdc6, Mcm3, and ORC to trigger the proteolytic degradation of Cdc6, the nuclear export of Mcm2--7--Cdt1, and inhibition of ORC helicase--loading activity, respectively ([@bib12]; [@bib14]; [@bib23]; [@bib40]; [@bib50]).
CDK oscillations also ensure that chromosome segregation occurs once per mitotic cell cycle ([@bib68]). At the end of G1 phase, G1--CDK (CDK bound to G1 cyclins Cln1-3) is required for duplication of the SPB ([@bib36]). Later in the cell cycle, S--CDK and M--CDK prevent re--duplication of SPBs, and M-CDK is essential for the assembly of metaphase spindles ([@bib3]; [@bib24]; [@bib30]). Finally, downregulation of CDK activity is required for anaphase spindle disassembly upon completion of chromosome segregation ([@bib58]; [@bib65]; [@bib69]). From this point forward, we will refer only to total CDK activity without specifying G1--, S--, or M--CDK, as the events we will be discussing are similarly regulated by all three kinases.
Two models have been proposed to explain how meiotic cells uncouple DNA replication and chromosome segregation during the MI--MII transition. The CDK--balance model suggests that partially inactivating CDK is sufficient to reset the chromosome segregation program while still inhibiting Mcm2--7 loading and replication initiation ([@bib35]). In contrast, the alternative--kinase model suggests that a second kinase inhibits Mcm2--7 loading during the MI--MII transition, allowing the oscillation of CDK activity to reset the chromosome segregation program without resetting the DNA replication program. Ime2, a yeast meiosis--specific kinase that is evolutionarily related to CDK ([@bib39]), has been proposed to have this role ([@bib33]).
We set out to systematically address how DNA replication is inhibited between the meiotic divisions using a combination of in vivo and in vitro approaches. We found that Mcm2--7 loading is the earliest inhibited step of replisome assembly during the MI--MII transition, and that this inhibition can be bypassed by simultaneous inhibition of CDK and Ime2. Furthermore, we identified two previously uncharacterized mechanisms used by Ime2 to inhibit origin licensing. First, Ime2 phosphorylation of Mcm2--7 directly inhibited its ability to be loaded onto replication origins. Second, we found that Ime2 and CDK cooperate to repress expression of Cdc6. In addition to the inhibition of helicase loading, meiotic cells promote degradation of Sld2, an essential helicase--activation protein, using phosphorylation sites for CDK and the polo--like kinase Cdc5. Together, these data show that meiotic cells use multiple kinases to inhibit both Mcm2--7 loading and activation, ensuring that MI and MII occur sequentially without an intervening S--phase.
Results {#s2}
=======
Mcm2--7 loading is the first inhibited step of DNA replication initiation {#s2-1}
-------------------------------------------------------------------------
To address how DNA replication is inhibited between MI and MII, we sought to identify the earliest inhibited step of replisome assembly. During replication initiation, the first proteins to stably associate with replication origins are ORC followed by the Mcm2--7 complex ([@bib5]). To analyze ORC and Mcm2--7 binding to origins in populations of cells, it was necessary for us to obtain synchronized cultures of cells undergoing meiosis. To this end, we used a previously--described block--release procedure ([@bib6]; [@bib13]). This method allows meiotic cells to proceed through G1 and S phase but arrest in Prophase I (hereafter referred to as G2 phase). Subsequent release from this cell cycle arrest results in cells progressing synchronously through Metaphase I, Anaphase I, Metaphase II, and Anaphase II with \>90% of cells completing meiosis ([Figure 1---figure supplement 1](#fig1s1){ref-type="fig"}).
Using ChIP--qPCR, we found that ORC was bound to a representative early--firing (*ARS1*) and late--firing (*ARS1413*) replication origin throughout both meiotic divisions ([Figure 1B](#fig1){ref-type="fig"}, top). ORC binding to DNA is therefore not limiting for replication initiation during meiosis. In contrast, although Mcm2--7 was present at origins during pre--meiotic G1 phase, this complex did not associate with either *ARS1* or *ARS1413* throughout MI and MII ([Figure 1C](#fig1){ref-type="fig"}, top). We also tested two additional origins, *ARS305* and *ARS418*, that are prone to Mcm2--7 reloading and DNA re--replication in mitotic cells ([@bib29]; [@bib51]; [@bib64]). As with the other origins tested, ORC bound to these origins throughout the meiotic divisions but Mcm2--7 loading was not observed after pre--meiotic G1 phase ([Figure 1B and C](#fig1){ref-type="fig"}, bottom). Both ORC and Mcm2-7 associated specifically with origin DNA sequences compared to non-origin DNA ([Figure 1---figure supplement 2](#fig1s2){ref-type="fig"}). The lack of Mcm2--7 binding to origins was not due to the absence of Mcm2--7 proteins or the essential helicase--loading protein Cdt1 ([Figure 1---figure supplement 3](#fig1s3){ref-type="fig"}). Thus, Mcm2--7 loading onto origins of replication is inhibited during the MI--MII transition.
CDK-dependent mechanisms inhibiting Mcm2--7 loading are weakened during the MI--MII transition {#s2-2}
----------------------------------------------------------------------------------------------
We considered two reasons that Mcm2--7 complexes were not reloaded upon decreased CDK activity during the MI--MII transition: (1) the known CDK-dependent mechanisms remain active enough to completely inhibit helicase loading (the CDK-balance model); or (2) meiosis--specific mechanisms inhibit helicase loading while CDK activity is reduced (the alternative-kinase model). To test the first possibility, we asked whether any of the CDK--dependent mechanisms preventing Mcm2--7 loading in mitotic cells were weakened during the MI--MII transition. In mitotically dividing cells, CDK inhibits helicase loading by three mechanisms: inhibition of ORC function, Cdc6 protein degradation, and Mcm2--7 nuclear export ([@bib1]). We found that at least two of these mechanisms were transiently weakened during the MI--MII transition.
Phosphorylation of Orc2 and Orc6 by CDK prevents ORC from facilitating helicase loading ([@bib14]; [@bib51]). Consistent with robust helicase loading during pre--meiotic G1 phase ([Figure 1C](#fig1){ref-type="fig"}), Orc2 and Orc6 were not phosphorylated during G1 when CDK is inactive ([Figure 2A and B](#fig2){ref-type="fig"}) ([@bib13]). In contrast, both subunits were phosphorylated throughout MI when CDK is highly active. Interestingly, we observed partial de--phosphorylation of both Orc2 ([Figure 2A](#fig2){ref-type="fig"} lane 9) and Orc6 ([Figure 2B](#fig2){ref-type="fig"} lane 6) at the MI--MII transition, before returning to a fully phosphorylated state that persisted until the end of MII.
{ref-type="fig"} (Orc2) and [Figure 2---figure supplement 1A](#fig2s1){ref-type="fig"} (Orc6). (**C**) *CDC6* protein and mRNA transiently reaccumulate during the MI--MII transition (strain yDP71). Top: Cdc6 immunoblots (short and long exposures) during meiosis. Bottom: *CDC6* mRNA levels were detected by northern blots during meiosis. The time after transfer into sporulation medium and the associated meiotic stages are indicated above each lane. For cell--cycle stage quantification, see [Figure 2---figure supplement 1B](#fig2s1){ref-type="fig"}.](elife-33309-fig2){#fig2}
Unlike ORC inhibition, CDK--phosphorylation of Cdc6 targets it for proteolytic degradation ([@bib12]; [@bib23]) and CDK is also partially responsible for repression of *CDC6* transcription ([@bib48]; [@bib54]). We found that Cdc6 protein was present during pre--meiotic G1 phase but became undetectable in MI when CDK is highly active ([Figure 2C](#fig2){ref-type="fig"}). During the MI--MII transition, however, Cdc6 protein partially reaccumulated ([Figure 2C](#fig2){ref-type="fig"} lane 8) before decreasing again in MII. As with Cdc6 protein levels, although *CDC6* mRNA was low during most of MI and MII, expression increased briefly during the MI--MII transition ([Figure 2C](#fig2){ref-type="fig"} lane 7). Taken together, these findings suggested that reduced CDK activity during the MI--MII transition results in a partial but detectable decrease of ORC phosphorylation and a slight reaccumulation of Cdc6.
Ime2 inhibits the Mcm2--7 complex by an intrinsic mechanism to prevent helicase loading {#s2-3}
---------------------------------------------------------------------------------------
The transient weakening of the CDK-dependent inhibitory mechanisms suggested the existence of meiosis--specific mechanisms to inhibit helicase loading. A strong candidate to mediate these potential mechanisms was Ime2, a CDK--related meiosis--specific kinase ([@bib39]). Previous studies found that Ime2 is active during the meiotic divisions ([@bib7]) and that, like CDK, this kinase can promote Mcm2--7 nuclear export upon completion of meiotic S phase ([@bib33]). However, if Ime2 were replacing CDK--dependent inhibition of helicase loading during the MI--MII transition, we hypothesized that it would inhibit Mcm2--7 loading by more than one mechanism (as CDK is known to do).
To identify additional mechanisms by which Ime2 inhibits Mcm2--7 loading, we asked if Ime2 could inhibit helicase loading in vitro. To this end, we used an assay that reconstitutes helicase loading on origin--containing DNA with four purified proteins (ORC, Cdc6, Cdt1, and Mcm2--7; see reaction scheme in [Figure 3A](#fig3){ref-type="fig"}) ([@bib25]; [@bib56]). We found that pre-treating the helicase-loading proteins with purified Ime2 fully inhibited Mcm2--7 loading ([Figure 3B and C](#fig3){ref-type="fig"}). To demonstrate that this inhibition depended on Ime2 kinase activity, we purified an analog--sensitive Ime2 protein (Ime2--AS). Analog--sensitive kinases are active in the presence of ATP but are inhibited by specific bulky ATP analogs ([@bib8]). In the case of Ime2--AS, addition of the ATP analog 1--NA--PP1 strongly inhibits its kinase activity ([@bib6]). In the presence of ATP, Ime2--AS inhibited Mcm2--7 loading to the same extent as wild--type (WT) Ime2 (compare [Figure 3---figure supplement 1](#fig3s1){ref-type="fig"} to [Figure 3C](#fig3){ref-type="fig"}). However, the addition of 1--NA--PP1 to assays treated with Ime2--AS (but not WT Ime2) fully restored helicase loading ([Figure 3D](#fig3){ref-type="fig"} lanes 3--10). Consistent with Ime2 phosphorylation being responsible for the inhibition, the extent of helicase-loading inhibition correlated with the extent of phosphorylation of helicase--loading proteins for both WT Ime2 and Ime2--AS, with and without 1--NA--PP1 treatment ([Figure 3---figure supplement 2](#fig3s2){ref-type="fig"}). Taken together, these data demonstrate that Ime2-phosphorylation of one or more helicase-loading proteins directly inhibits origin licensing. Furthermore, because these experiments use only purified proteins, the mechanism preventing helicase loading must be due to the intrinsic inhibition of a specific protein's function, as opposed to indirect inhibitory mechanisms such as nuclear export or protein degradation.
{#fig3}
To more precisely elucidate the mechanism of Ime2-inhibition of helicase loading, we sought to determine the step of this reaction that Ime2 inhibits. During helicase loading, origin--bound ORC--Cdc6 complexes recruit Cdt1--Mcm2--7 heptamers to form a short-lived complex called the OCCM (for ORC-Cdc6-Cdt1-Mcm2-7) ([@bib55]; [@bib61]; [@bib67]; [@bib71]). After OCCM formation, multiple conformational changes and ATP hydrolysis events are required for the first Mcm2--7 complex to be stably loaded around the DNA and a second Mcm2--7 to be recruited and loaded ([@bib17]; [@bib18]; [@bib38]; [@bib66]; [@bib71]; [@bib73]). To determine whether Ime2 inhibits initial Mcm2--7 recruitment, we conducted in vitro association assays with the slowly hydrolyzable analog ATPγS instead of ATP, stalling helicase loading after OCCM formation (reaction scheme in [Figure 3A](#fig3){ref-type="fig"}) ([@bib55]). Prior phosphorylation of the helicase--loading proteins with Ime2 (in the presence of ATP) did not prevent subsequent OCCM formation in the presence of excess ATPγS ([Figure 3E](#fig3){ref-type="fig"}). Thus, Ime2 phosphorylation does not block the protein--protein interactions necessary for initial Mcm2--7 recruitment, but instead must inhibit a downstream step during Mcm2--7 loading.
Next, we sought to identify the Ime2-target(s) that result in the inhibition of helicase loading. In vitro kinase assays showed that Ime2 phosphorylates Cdc6, Cdt1, and subunits of both ORC and the Mcm2--7 complex ([Figure 4A](#fig4){ref-type="fig"}). To identify which of these phosphorylation events inhibits helicase loading, we used Ime2--AS to phosphorylate each of these proteins separately. We then inhibited Ime2--AS by the addition of 1--NA--PP1 before adding the three remaining, non--phosphorylated proteins and origin DNA to initiate helicase loading. Strikingly, we found that phosphorylation of the Mcm2--7 complex alone resulted in a \>90% decrease in helicase loading ([Figure 4B and C](#fig4){ref-type="fig"}). In contrast, Ime2 phosphorylation of ORC resulted in a \~50% reduction in loading, whereas reactions with phosphorylated Cdc6 and Cdt1 showed only minor defects ([Figure 4B and C](#fig4){ref-type="fig"}). Although CDK also inhibits helicase loading in vitro, the equivalent experiment using CDK confirmed previous results showing that CDK only strongly inhibits ORC activity ([Figure 4---figure supplement 1](#fig4s1){ref-type="fig"}) ([@bib14]). Together, these data demonstrate that Ime2 phosphorylation of the Mcm2--7 complex is sufficient to inhibit helicase loading. Furthermore, although Ime2 and CDK both directly inhibit helicase loading, the critical target required for their direct inhibition is distinct.
![Ime2--phosphorylation of the Mcm2--7 complex intrinsically inhibits its loading onto replication origins.\
(**A**) Ime2 can phosphorylate Cdc6, ORC, Cdt1 and Mcm2--7 in vitro. Buffer control (lanes 1, 3, and 5) or 50 nM Ime2 (lanes 2, 4, and 6) were incubated with the indicated substrate proteins. The substrates were purified Cdc6 (lanes 1 and 2), ORC (lanes 3 and 4) or Mcm2--7--Cdt1 (lanes 5 and 6). Asterisk (\*) marks Ime2 autophosphorylation. Top: total protein (Krypton stain). Bottom: phosphorylated protein (modified with \[γ-^32^P\] ATP). (**B**) Ime2--phosphorylation of each protein separately shows that the primary target of Ime2--mediated inhibition is the Mcm2--7 complex (compare lanes 7 and 8). Top: Flowchart of experiment. Bottom: Helicase--loading assay after prior Ime2--phosphorylation of indicated protein. (**C**) Quantification of (**B**) from three independent experiments. Inhibition ratio was calculated as total Mcm2--7 loading from the +Ime2 reactions divided by amount of loading in the corresponding reaction lacking Ime2. The mean is represented by the height of the bar. Error bars represent the standard deviation of three independent experiments.\
10.7554/eLife.33309.014Figure 4---source data 1.Raw values used for the quantification of [Figure 4C](#fig4){ref-type="fig"}.](elife-33309-fig4){#fig4}
CDK and Ime2 cooperate to inhibit Mcm2--7 loading and Cdc6 expression during the MI--MII transition {#s2-4}
---------------------------------------------------------------------------------------------------
A critical question we sought to answer was which kinase inhibits helicase loading during the MI--MII transition in vivo. To address this question, we employed yeast strains with analog--sensitive alleles of CDK (*cdk1--as*) and Ime2 (*ime2--as*) in place of their respective wild-type alleles. As cells were entering the MI--MII transition (≥45% of cells in anaphase I; [Figure 5---figure supplement 1](#fig5s1){ref-type="fig"}), we inhibited these kinases and examined Mcm2--7 loading, *CDC6* mRNA and protein expression, and ORC phosphorylation. As a control, we confirmed that cells with wild--type *CDK1* and *IME2* were unaffected by addition of inhibitors (compare [Figure 5A](#fig5){ref-type="fig"} to [Figures 1](#fig1){ref-type="fig"} and [2](#fig2){ref-type="fig"}).
{ref-type="fig"}, see [Figure 5---figure supplement 1A--1D](#fig5s1){ref-type="fig"}, respectively. Mcm2--7 loading was analyzed at ARS305 (red) and ARS418 (orange). The peak % of input DNA immunoprecipitated (set to A.U. = 1.0) was 15.5% for ARS305 and 5.3% for ARS418. (**E**) Ime2 directly phosphorylates Cdc6 phospho--degron domains. Purified Cdc6 was treated with purified Ime2 or buffer--control in the presence of ATP. Quantitative mass spectroscopy was used to identify Ime2-dependent phosphorylation sites on Cdc6. Phosphorylation sites detected (with \>4--fold enrichment upon Ime2 treatment) as well as the location of the Cdc6 phospho--degron domains are illustrated. Yellow markers indicate unique Ime2 sites. Orange markers indicate Ime2 sites that are also CDK sites based on previous work ([@bib12]; [@bib23]). Phospho-degron domains are based on previous work ([@bib53]). For phosphorylation--site enrichment values, see [Supplementary file 1](#supp1){ref-type="supplementary-material"}.\
10.7554/eLife.33309.019Figure 5---source data 1.Raw values used for the quantification of [Figure 5A--5D](#fig5){ref-type="fig"}.](elife-33309-fig5){#fig5}
We inhibited CDK and Ime2 separately to determine the impact of each kinase on helicase loading. Inhibition of CDK at the end of MI promoted only limited Mcm2--7 reloading and reaccumulation of *CDC6* mRNA and protein, although ORC was substantially dephosphorylated in this condition ([Figure 5B](#fig5){ref-type="fig"}). In mitotic cells, CDK-inhibition bypasses all of these inhibitory mechanisms and promotes robust Mcm2--7 reloading ([@bib20]; [@bib23]; [@bib51]), and we have recapitulated this result with the *cdk1--as* allele ([Figure 5---figure supplement 2](#fig5s2){ref-type="fig"}). Consequently, there must be CDK--independent mechanisms to inhibit origin licensing and *CDC6* expression that are specific to meiosis. To test whether Ime2 fulfills these functions, we inhibited Ime2 as cells were entering the MI--MII transition. Similar to CDK, however, Ime2-inhibition only caused limited Mcm2--7 reloading, although it did promote more significant reaccumulation of *CDC6* mRNA and protein than CDK-inhibition ([Figure 5C](#fig5){ref-type="fig"}). Therefore, neither CDK nor Ime2 are solely responsible for inhibiting helicase loading during the MI--MII transition.
Do CDK and Ime2 cooperate to repress helicase reloading? Strikingly, simultaneous inhibition of both CDK and Ime2 at the end of MI resulted in much higher levels of Mcm2--7 reloading than we observe with either kinase alone ([Figure 5D](#fig5){ref-type="fig"}). Furthermore, co--inhibition of these kinases restored expression of *CDC6* mRNA and protein to levels observed in pre--meiotic G1 cells ([Figure 5D](#fig5){ref-type="fig"}). Consistent with repression of Cdc6 by both kinases, mass spectrometry analysis showed that Ime2 phosphorylates multiple sites on Cdc6 within its phosphorylation--responsive degron domains in vitro, and two of these sites directly overlap with CDK--sites known to contribute to Cdc6 degradation ([Figure 5E](#fig5){ref-type="fig"}, [Supplementary file 1](#supp1){ref-type="supplementary-material"}) ([@bib12]; [@bib23]). The transcriptional and proteolytic inhibition of Cdc6 by both CDK and Ime2 illustrates that it is a critical target of inhibition to prevent helicase loading. Thus, neither CDK nor Ime2 is capable of full inhibition of helicase loading but together they are a potent inhibitor of origin licensing and *CDC6* expression during the MI--MII transition.
Cdc5 and CDK promote the degradation of Sld2, an essential helicase--activation protein {#s2-5}
---------------------------------------------------------------------------------------
We considered the possibility that downstream steps of DNA replication could also be inhibited during the meiotic divisions. Despite the numerous mechanisms inhibiting Mcm2--7 loading, the weakening of ORC-- and Cdc6--dependent controls ([Figure 2](#fig2){ref-type="fig"}) revealed a degree of leakiness in at least a subset of these mechanisms without any kinase perturbation. Therefore, we analyzed the abundance of proteins required for Mcm2--7 activation, which is the step after Mcm2--7 loading during replication initiation. Most helicase--activation proteins were present throughout the meiotic divisions, including Cdc45, Psf2 (a member of the GINS complex), Sld3, and Dpb11 ([Figure 6---figure supplement 1](#fig6s1){ref-type="fig"}). In contrast, Sld2 was robustly degraded upon entry into MI and did not reaccumulate until the completion of MII ([Figure 6A](#fig6){ref-type="fig"}, [Figure 6---figure supplement 2](#fig6s2){ref-type="fig"}). Sld2 is essential for replication initiation ([@bib37]; [@bib70]) and thus, its degradation represents a robust mechanism to inhibit helicase activation.
{ref-type="fig"}. (**B**--**D**) Mutation of either Cdc5-- or CDK--phosphorylation sites on Sld2 results in stabilization of Sld2 throughout the meiotic divisions: Immunoblots of Sld2--13myc during meiosis with the following mutations: (**B**) Cdc5--phosphorylation sites (strain yDP473: 2TA -- T122A/T143A), (**C**) CDK--phosphorylation sites (strain yDP642: 2SA -- S128A/S138A), or (**D**) Cdc5-- and CDK--phosphorylation sites (strain yDP644: 4A -- T123A/S128A/S138A/T143A). The time after transfer into sporulation medium and the associated meiotic stages are indicated above each lane. For cell--cycle synchrony, refer to [Figure 6---figure supplement 2B--2D](#fig6s2){ref-type="fig"}. (**E**) Top: Samples from (**A**--**D**) with the peak number of cells in G2, Metaphase I, Anaphase I, Metaphase II, and Anaphase II were run side--by--side. Middle: Mean of Sld2 levels normalized to PGK1 levels from three independent experiments. Bottom: Graph of Sld2/PGK1 quantification from three independent experiments. The mean is represented by the height of the bar. Error bars represent the standard deviation.\
10.7554/eLife.33309.025Figure 6---source data 1.Raw values used for the quantification of [Figure 6E](#fig6){ref-type="fig"}.](elife-33309-fig6){#fig6}
Both CDK and the polo--like kinase Cdc5 were candidates to regulate meiotic Sld2 protein levels ([@bib57]). During mitotic divisions, total Sld2 protein levels do not change dramatically until the end of mitosis, at which point CDK-- and Cdc5-- phosphorylation sites within a phospho--degron domain on Sld2 become important for its degradation. In addition, Cdc5 and the M--phase cyclins Clb1, Clb3, and Clb4 are transcriptionally induced by the meiosis-specific transcription factor Ndt80 ([@bib15]), and both Cdc5 and CDK are active during the meiotic divisions ([@bib2]; [@bib6]; [@bib13]; [@bib16]; [@bib21]; [@bib41]; [@bib60]).
To test whether Cdc5 and CDK contribute to Sld2 degradation during meiosis, we mutated their previously identified phosphorylation sites on Sld2 ([@bib57]). Mutation of either Cdc5--phosphorylation sites (*sld2-2TA* = T122A and T143A), CDK--phosphorylation sites (*sld2-2SA* = S128A and S138A), or both together (*sld2-4A*) resulted in stabilization of Sld2 during the meiotic divisions ([Figure 6B--D](#fig6){ref-type="fig"}, [Figure 6---figure supplement 2](#fig6s2){ref-type="fig"}). To directly compare the effects of these mutations, we ran samples from all four strains side--by--side at each stage of meiosis and quantified Sld2 levels relative to a PGK1 loading control ([Figure 6E](#fig6){ref-type="fig"}, [Figure 6---figure supplement 3](#fig6s3){ref-type="fig"}). No more than a 2--fold difference in Sld2 levels was detected during meiotic G2 between all four strains. From metaphase I until metaphase II, wild type (WT) Sld2 abundance decreased to \~10% of the levels observed in G2 phase. In contrast, the abundance of the Sld2-4A mutant remained almost unchanged during this time. The levels of the Sld2-2TA and Sld2-2SA protein were marginally less than the Sld2-4A protein at these same times. Because both Sld2-2SA or Sld2-2TA were expressed at much higher levels than the WT protein during the meiotic divisions, both CDK-- and Cdc5--phosphorylation are required to drive substantial Sld2 degradation. Upon entry into anaphase II, WT Sld2 protein levels began to recover while the mutant protein levels decreased, suggesting that other mechanisms were impacting Sld2 expression. Together, these data indicate that CDK and Cdc5 inhibit Mcm2--7 activation during the meiotic divisions by promoting Sld2 degradation. Thus, if an origin escapes the inhibition of Mcm2-7 loading during the meiotic divisions, Sld2 degradation would prevent activation of the associated helicases.
Discussion {#s3}
==========
In this study, we address a fundamental question concerning the regulation of meiosis; how do meiotic cells undergo two sequential rounds of chromosome segregation without an intervening S--phase? We found that meiotic cells prevent DNA replication between the meiotic divisions using CDK, Ime2, and Cdc5 to inhibit both helicase loading and activation. Ime2 and CDK cooperate to inhibit helicase loading, and their co--inhibition was sufficient for aberrant origin licensing during the MI--MII transition. Compared with CDK, Ime2 inhibits origin licensing using both overlapping and distinct mechanisms. In particular, we found that unlike CDK, Ime2 phosphorylation of Mcm2--7 directly inhibits its participation in helicase loading. In addition to the inhibition of origin licensing, meiotic cells use CDK and the polo--like kinase Cdc5 to promote degradation of Sld2, a key helicase-activation protein. Together, these data reveal that multiple kinases inhibit DNA replication between the two meiotic divisions by targeting both many components of the helicase-loading machinery and at least one helicase-activation protein. These mechanisms combine to ensure the hallmark reduction in ploidy associated with meiotic cell division.
Elements of both the CDK--balance model ([@bib35]) and the alternative--kinase model ([@bib33]) are evident in our results ([Figure 7A](#fig7){ref-type="fig"}). In support of the CDK--balance model, the amount of CDK activity present during the MI--MII transition is sufficient to inhibit most origin licensing, because Ime2 inhibition does not result in strong Mcm2--7 reloading. Thus, the decrease in CDK activity that resets the chromosome segregation program ([@bib9]; [@bib11]; [@bib13]; [@bib27]; [@bib45]) is not sufficient to fully reset the DNA replication program. In support of the alternative--kinase model, we found that CDK inhibition similarly did not result in extensive Mcm2--7 reloading during the MI-MII transition. Instead, we show that two additional kinases contribute to the inhibition of DNA replication between the meiotic divisions. Ime2 inhibits helicase loading, and Cdc5 helps stimulate the degradation of Sld2.
![Model of how meiotic cells inhibit DNA replication during the MI--MII transition.\
(**A**) Graphical representation of CDK (blue) ([@bib13]) and Ime2 (red) ([@bib7]) kinase activities during meiosis, and how they regulate the chromosome segregation and DNA replication programs. Chromosome segregation is regulated by CDK, whereas DNA replication is regulated by CDK, Ime2, and Cdc5. During the MI-MII transition, CDK activity decreases enough to reset the chromosome segregation program for MII. Although CDK remains active enough to mostly inhibit the DNA replication program, the decreased activity is a significant threat to the inhibition of origin licensing. Ime2 is also mostly sufficient to inhibit Mcm2-7 loading during the MI-MII transition. Cdc5 activity has not been precisely determined and is thus not shown, but it contributes to the inhibition of DNA replication by limiting Mcm2--7 activation. (**B**) The mechanisms and effector proteins used to inhibit DNA replication during the meiotic divisions. CDK and Ime2 cooperate to inhibit helicase loading by promoting Mcm2--7 nuclear export and the repression of *CDC6* by proteolytic degradation and transcriptional inhibition. Additionally, Ime2 phosphorylates and directly inhibits the Mcm2--7 complex, whereas CDK directly inhibits ORC. To inhibit helicase activation, CDK and Cdc5 promote the proteolytic degradation of Sld2.](elife-33309-fig7){#fig7}
The inhibition of DNA replication by Ime2 and Cdc5 allows for CDK to oscillate and reset the chromosome segregation program without allowing a new round of DNA replication. That meiotic cells use Ime2 and Cdc5 to inhibit helicase loading and activation, respectively, suggests that CDK--dependent inhibition of origin licensing is not adequate to ensure genome stability during meiosis. Furthermore, a subset of the mechanisms by which CDK and Ime2 inhibit helicase loading are non--overlapping, providing additional avenues to prevent inappropriate replication ([Figure 4](#fig4){ref-type="fig"}). It is important to note that our helicase reloading results are population--based experiments. Accordingly, the slight helicase reloading and Cdc6 reaccumulation that occurred upon inhibition of each kinase separately suggests that, at some point during the MI-MII transition or at a subset of replication origins, each kinase is required to prevent origin licensing.
Why do meiotic cells use so many mechanisms to inhibit DNA replication between the meiotic divisions ([Figure 7B](#fig7){ref-type="fig"})? We speculate that this is due to the difficulty of preventing \>300 replication origins from initiating while CDK activity oscillates. Emphasizing the importance of completely inhibiting DNA replication at unwanted periods of the cell cycle, previous studies have shown that reinitiating replication from even a single origin can cause gene amplification and chromosome missegregation ([@bib28]; [@bib31]). The transient weakening of both Cdc6 repression and ORC phosphorylation during the MI--MII transition ([Figure 2](#fig2){ref-type="fig"}) indicates that the associated decrease in CDK activity ([@bib13]) is a significant threat to origin licensing. Even the additional repression of Cdc6 by Ime2 is not sufficient to fully eliminate the protein during the MI-MII transition, although Ime2 has a larger role than CDK for Cdc6 repression at this time ([Figure 5](#fig5){ref-type="fig"}). It is worth noting that in multicellular eukaryotes, preventing DNA re-replication in meiotic cells is more important than in mitotic cells to ensure that the inherited genome remains intact.
In addition to helicase loading, our studies strongly suggest that meiotic cells inhibit downstream steps of DNA replication. In mitotic cells, prevention of DNA re--replication relies on inhibiting Mcm2--7 loading ([@bib1]). Inhibiting early steps of replication makes sense, as there is less danger of aberrant DNA unwinding and polymerase recruitment. Although helicase loading is an earlier event in replication initiation, we note that Sld2 is required to form the active eukaryotic replicative helicase, the Cdc45--Mcm2--7--GINS (CMG) complex ([@bib32]; [@bib37]; [@bib49]; [@bib70]). Activation of the replicative helicase is the committed step of replication initiation ([@bib5]) and preventing this step would still stop replication before initial DNA unwinding and synthesis. Previous studies found that Sld2 degradation in mitotic cells is important for ensuring genome stability at the M--G1 transition, but not during other parts of the mitotic cell cycle ([@bib57]). Our finding that Sld2 is absent during both meiotic divisions suggests that mechanisms preventing DNA re--replication at the mitotic M--G1 transition also function during meiosis at the MI--MII transition, a partially G1-like state. Consistent with this idea, Dbf4, which binds to Cdc7 to form the kinase DDK and is also required for helicase activation, is degraded during both mitotic anaphase ([@bib26]) as well as meiotic anaphase I ([@bib47]). The degradation of these two proteins strongly suggests that meiotic cells further protect themselves from replication initiation during the meiotic divisions by inhibiting multiple steps during Mcm2--7 activation.
Ime2 is not just a backup for the known inhibitory mechanisms used by CDK. We found that in addition to using at least two mechanisms similar to CDK (Cdc6 degradation ([Figure 5](#fig5){ref-type="fig"}) ([@bib23]) and Mcm2-7 nuclear export ([@bib33])), Ime2-phosphorylation of Mcm2-7 results in a distinct mechanism of inhibition not seen after CDK-phosphorylation ([Figure 4](#fig4){ref-type="fig"}). Origin licensing is inhibited by a number of different mechanisms across eukaryotic evolution, but previous studies have not identified a mechanism that directly prevents Mcm2-7 from completing helicase loading ([@bib1]; [@bib59]). Understanding how helicase loading is inhibited by CDK-phosphorylation of ORC and Ime2-phosphoryation of Mcm2-7 will reveal whether these two kinases target the same or different steps of this process.
The ability of Ime2 to inhibit helicase loading also suggests an important role for this kinase during the meiotic G1--S transition. During the mitotic G1--S transition, helicase loading is inhibited by G1--CDK before helicase activation is stimulated by S--phase CDK ([@bib23]; [@bib40]). This 'insulation' prevents cells from being in a state that permits both helicase loading and activation (e.g. at an intermediate level of CDK activity). During meiosis, G1 cyclins are not expressed and it is Ime2 that triggers activation of S--CDK at the G1--S transition ([@bib6]; [@bib22]). Our data and previous studies ([@bib33]) show that Ime2 can robustly inhibit helicase loading by multiple mechanisms, similar to G1-CDK. These observations strongly suggest that Ime2 insulates helicase loading from helicase activation during the meiotic G1--S transition in an analogous manner to G1--CDK during the mitotic G1--S transition.
The problem of uncoupling DNA replication and chromosome segregation during meiosis is conserved in other eukaryotes. Metazoans also use CDK to regulate both chromosome segregation and DNA replication ([@bib1]; [@bib44]; [@bib59]). Although metazoans mostly rely on CDK--independent mechanisms to inhibit Mcm2--7 loading during S and G2 phase, these mechanisms are not as potent during M--phase at which point Cdk1 becomes a critical inhibitor of origin licensing ([@bib1]; [@bib59]). Transient inactivation of Cdk1 in human cells is sufficient for DNA re--replication, just as it is in yeast ([@bib4]; [@bib20]; [@bib34]). An oscillation of Cdk1 activity is also required for correct chromosome segregation in mammalian cells ([@bib44]). With regard to how DNA replication and chromosome segregation are uncoupled during mammalian meiosis, previous studies raise the intriguing possibility that mammalian Cdk2 may have a similar role during meiosis as Ime2 does in yeast. Human Cdk2 can rescue some meiotic defects associated with ime2∆ in yeast ([@bib62]). Additionally, Cdk2 is not required for mitosis in mice, but it is required for both male and female meiosis ([@bib52]). Together, our results provide new insight into how meiotic cells use multiple cell--cycle regulators and mechanisms to uncouple DNA replication and chromosome segregation.
Materials and methods {#s4}
=====================
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
Reagent type (species)\ Designation Source or reference Identifiers Additional information
or resource
---------------------------------- ----------------- ---------------------- ------------------------ --------------------------------------------------------------
strain, strain background\ yDP71 This paper Cdc6-3V5 *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*Saccharomyces cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*CDC6-3V5::KANMX6/CDC6-3V5::KANMX6*
strain, strain background\ yDP120 This paper Orc6-3V5 *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*ORC6-3V5::KANMX6/ORC6-3V5::KANMX6*
strain, strain background\ yDP152 This paper cdk1-as, Cdc6-3V5 *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*cdc28-as1(F88G)/cdc28-as1(F88G)*\
*CDC6-3V5::KANMX6/CDC6-3V5::KANMX6*
strain, strain background\ yDP159 This paper Ime2 Purification *W303 MATa bar1::hisG pep4::unmarked*\
(*S. cerevisiae* W303) *LEU2::pGAL1,10-IME2(1--404)−3xFLAG*
strain, strain background\ yDP176 This paper ime2-as, Cdc6-3V5 *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*ime2-as1(M146G)/ime2-as1(M146G)*\
*CDC6-3V5::KANMX6/CDC6-3V5::KANMX6*
strain, strain background\ yDP177 This paper cdk1-as, ime2-as,\ *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) Cdc6-3V5 *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*cdc28-as1(F88G)/cdc28-as1(F88G)*\
*ime2-as1(M146G)/ime2-as1(M146G)*\
*CDC6-3V5::KANMX6/CDC6-3V5::KANMX6*
strain, strain background\ yDP329 This paper Dpb11-3V5 *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*DPB11-3V5::KANMX6/DPB11-3V5::KANMX6*
strain, strain background\ yDP330 This paper Psf2-3V5 *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*PSF2-3V5::KANMX6/PSF2-3V5::KANMX6*
strain, strain background\ yDP335 This paper Cdc45-13myc *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*CDC45-13myc::KANMX6/CDC45-13myc::KANMX6*
strain, strain background\ yDP336 This paper Sld2-13myc *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*SLD2-13myc::KANMX6/SLD2-13myc::KANMX6*
strain, strain background\ yDP337 This paper Sld3-13myc *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*SLD3-13myc::KANMX6/SLD3-13myc::KANMX6*
strain, strain background\ yDP473 This paper Sld2-2TA-13myc *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*SLD2(T122A T143A)−13myc::KANMX6/*\
*SLD2(T122A T143A)−13myc::KANMX6*
strain, strain background\ yDP554 This paper Ime2-AS Purification *W303 MATa bar1::hisG pep4::unmarked*\
(*S. cerevisiae* W303) *LEU2::pGAL1,10-IME2(1--404, M146G)−3xFLAG*
strain, strain background\ yDP642 This paper Sld2-2SA-13myc *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*SLD2(S128A S138A)−13myc::KANMX6/*\
*SLD2(S128A S138A)−13myc::KANMX6*
strain, strain background\ yDP644 This paper Sld2-4A-13myc *SK1 MATa/alpha ura3::pGPD1-GAL4(848).*\
(*S. cerevisiae* SK1) *ER::URA3/ura3::pGPD1- GAL4(848).ER::URA3*\
*GAL-NDT80::TRP1/GAL-NDT80::TRP1*\
*SLD2(T122A S128A S138A T143A)−13myc::*\
*KANMX6/SLD2(T122A S128A S138A T143A)*\
*−13myc::KANMX6*
strain, strain background\ ySK119 Lõoke *et al*, 2017\ Cdk1-Clb5 Purification W303 MATa bar1::hisG pep4::unmarked\
(*S. cerevisiae* W303) (PMID: 28270517) URA3::pGAL1,10-Cdc28-His,Δ1--95-Clb5-Flag
strain, strain background\ yST135 This paper Mcm2-7 Purification *W303 MATa bar1::hisG pep4::unmarked*\
(*S. cerevisiae* SK1) *TRP1::pSKM003(pGAL1,10-MCM6,MCM7)*\
*HIS3::pSKM004-(pGAL1,10-MCM2,Flag-MCM3)*\
*LYS2::pSKM002-(pGAL1,10-MCM4,MCM5)*
strain, strain background\ yST144 [@bib67]\ Mcm2-7-Cdt1\ *W303 MATa bar1::hisG pep4::unmarked*\
(*S. cerevisiae* W303) (PMID: 25892223) Purification *TRP1::pSKM003(pGAL1,10-MCM6,MCM7)*\
*HIS3::pSKM004-(pGAL1,10-MCM2,Flag-MCM3)*\
*LYS2::pSKM002-(pGAL1,10-MCM4,MCM5)*\
*URA3::pALS1(pGAL1,10-Cdt1,GAL4)*
strain, strain background\ A4370 Angelika Amon cdk1-as *W303 MATa bar1::hisG cdc28-as1(F88G)*
(*S. cerevisiae* W303)
strain, strain background\ ySDORC John Diffley ORC purification *W303 MATa bar1::hyg pep4::kanMX*\
(*S. cerevisiae* W303) *TRP1::pGAL1,10-ORC5,ORC6*\
*HIS3::pGAL1,10-ORC3,ORC4 URA3::pGAL1,*\
*10-CBP-TEV-ORC1,ORC2*
antibody poly ORC\ HM1108 (Bell Lab)
(Orc1 and Orc2\
western blots\
and ORC ChIP)
antibody Cdt1 HM5353 (Bell Lab)
antibody poly MCM\ UM174 (Bell Lab)
(Mcm3 and\
Mcm6 western\
blots)
antibody poly MCM\ UM185 (Bell Lab)
(Mcm2-7 ChIP)
antibody Mcm2 Santa Cruz, yN-19\ RRID:[AB_648843](https://scicrunch.org/resolver/AB_648843)
(code sc-6680)
antibody Mcm7 Santa Cruz, yN-19\ RRID:[AB_647936](https://scicrunch.org/resolver/AB_647936)
(code SC-6688)
antibody PGK1 Invitrogen\ RRID:[AB_2532235](https://scicrunch.org/resolver/AB_2532235)
(catalog \#459250)
recombinant DNA\ pSKM033 [@bib38]\ Cdc6 purification pGEX-GST-3C-FLAG-*CDC6*
reagent (plasmid) (PMID: 25087876)
recombinant DNA\ pALS16 This study Cdt1 purification pGEX-GST-3C-*CDT1*
reagent (plasmid)
chemical compound, drug 1-NM-PP1 Toronto Research\
Chemicals\
(catalog \#A603003)
chemical compound, drug 1-NA-PP1 Cayman\
Chemical Co.,\
(catalog \#NC1049860)
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
Yeast strains and plasmids {#s4-1}
--------------------------
All *S. cerevisiae* strains are summarized in the Key Resources Table. Strains used for meiosis were diploids isogenic with SK1 *ho::LYS2/ho::LYS2, lys2/lys2, ura3/ura3, leu2::hisG/leu2::hisG, his3:hisG/his3:hisG, trp1::hisG/trp1::hisG*. Other strains were isogenic with W303 *ade2--1 trp1--1 leu2--3112 his3--11,15 ura3--1 can1--100*. Epitope tagging was done by homologous recombination as previously described ([@bib42]). Protein expression plasmids are described in the Key Resources Table.
Meiotic time--courses {#s4-2}
---------------------
Meiotic time courses were done as described ([@bib7]). Briefly, saturated YPD cultures were diluted to an OD~600~ = 0.25 in BYTA medium and grown for 18--22 hr. Cells were then washed once with water and resuspended to OD~600~ = 1.9 in Sporulation medium before taking the 0--hour sample. All strains used for meiotic time courses included *P~GPD~--GAL4--ER, P~GAL~--NDT80.* Cultures were shaken at 30°C for 6 hr to allow cells to accumulate at the *NDT80* block. Addition of 1 µM β--estradiol (5 mM stock in ethanol \[Sigma, E2758\]) released cells from the *NDT80* block. Protein, RNA, ChIP, and immunofluorescence samples were harvested in parallel at the indicated time points.
Immunofluorescence {#s4-3}
------------------
Tubulin immunofluorescence was performed as described ([@bib7]). For spindle and nuclei scoring, 100 cells were counted per time point. Metaphase--I cells were defined as having a short, thick, bipolar spindle; Anaphase--I cells as having a long, bipolar spindle; Metaphase--II cells as having two short, thick, bipolar spindles; and Anaphase--II cells as having two long, bipolar spindles. Nuclei were counted as the number of separated DNA masses.
Immunoblots {#s4-4}
-----------
Cells were pelleted at 136,000 x g, resuspended in 5% TCA, and left at 4°C overnight. Pellets were washed with acetone, dried, and resuspended in 100 µL of 50 mM Tris \[pH = 7.6\],1 mM EDTA, 2.75 mM DTT, 1 mM PMSF, and 1x cOmplete Protease Inhibitors (Roche). Cells were lysed three times with glass beads using a FastPrep (MP Biomedicals), and boiled for 5 min after addition of 75 uL 5x sample buffer. Ponceau staining was used as a loading control. Proteins were detected with antibodies recognizing the epitope indicated or with the following antibodies: Orc1 and Orc2 (HM1108), Cdt1 (HM5353), Mcm3 and Mcm6 (UM174), Mcm2 (Santa Cruz, yN--19), Mcm7 (Santa Cruz, yN--19), and PGK1 (Invitrogen).
Northern blots {#s4-5}
--------------
Total RNA was isolated using a (400 µL:400 µL) mixture of TES buffer (10 mM Tris \[pH7.6\], 10 mM EDTA, 0.5% SDS) and acid phenol while shaking (Thermomixer, Eppendorf) with glass beads at 65°C for 30 min. After ethanol precipitation and resuspension in DEPC--treated water, equal amounts of RNA (between 10--14 µg) were loaded in each lane. rRNA was used as a loading control and detected with methylene blue. *CDC6*--specific ^32^P--labeled probes were made by Klenow extension (GE Healthcare, RPN1605) Template: 3'--end specific *CDC6* PCR--product. Primers: random hexamers.
Helicase loading and OCCM formation assays {#s4-6}
------------------------------------------
Helicase loading was done as described in [@bib38], except using 200 mM potassium glutamate (KGlut) instead of 300 mM. Briefly, 50 nM ORC, 100 nM Cdc6, 150 nM Mcm2--7/Cdt1 were combined in a 40 µL reaction containing 25 nM bead--bound 1.3 kB DNA including the *ARS1* origin. After mixing the proteins and DNA, reactions were shaken at 1,250 rpm at 25°C for 30 min (Thermomixer, Eppendorf), and then washed three times. For helicase--loading experiments, the three washes had buffer containing 300 mM K--Glut, 500 mM NaCl, and 300 mM KGlut respectively. DNA--bound proteins were eluted using DNase, run on SDS--PAGE gels, and detected using Krypton fluorescent stain (Fisher, PI--46629). To monitor OCCM formation, 5 mM ATPγS was used in place of ATP, and only washed with buffers containing 300 mM KGlut. Ime2 or CDK (150 nM or buffer control) was pre--incubated with the four purified proteins for 45 min before adding the bead--bound *ARS1* DNA. For experiments examining the effects of phosphorylating individual proteins, 150 nM kinase was pre--incubated with the indicated protein in the '+kinase' reactions, and the other three proteins were mock phosphorylated. After 1 hr, the kinase inhibitor (1--NA--PP1 for Ime2, Sic1 for CDK) was added to both the '+kinase' and '--kinase' reactions, and the kinase itself was then added to the '--kinase' reactions to control for any effects that didn't depend on kinase activity. Phosphorylated and mock phosphorylated proteins were then combined and added to *ARS1* DNA to start the helicase--loading assay.
Protein purifications {#s4-7}
---------------------
Ime2 (strain yDP159) and Ime2--AS (strain yDP554) were purified from yeast strains containing the *P~GAL~--IME2^stable^--3xFLAG* construct, which expressed amino acids 1--404 of *IME2* fused to a 3xFLAG epitope at the C--terminus. The Ime2--AS protein contains a M146G mutation. Eight liters of yeast were grown in YEP--Glycerol to OD~600~ = 1.0 and induced with 2% galactose for 5 hr. Cells were lysed using a freezer mill in Buffer H (25 mM Hepes \[pH = 7.6\], 5 mM magnesium acetate \[MgAc\], 1 mM EDTA, 1 mM EGTA, 10% glycerol) containing 1 M Sorbitol, 0.02% NP--40, 2 mM ATP, 0.5 M KCl, 1x cOmplete Protease Inhibitors (Roche), and PhosSTOP phosphatase inhibitors (Roche). The lysate was clarified by centrifugation at 150,000 x g. KCl concentration was adjusted to 300 mM and the lysate was clarified again by centrifugation at 25,000 x g. Lysate was incubated with 1 mL M2--resin (Sigma, A2220) and washed with Buffer H with 300 mM KGlut and 0.01% NP--40 before elution with 3xFLAG peptide. Eluted protein was concentrated using a spin column (10 kDa cutoff, Vivaspin) and injected onto a Superdex 75 column (GE Healthcare) equilibrated with the same buffer. Peak Ime2--containing fractions were pooled and aliquoted.
Cdt1 (plasmid pALS16) was purified from *E. coli* Rosetta 2 cells induced overnight at 18°C. Cells were resuspended in 1x PBS, 10% glycerol, 1 mM DTT, and 300 mM NaCl, and lysed with lysozyme and sonication. The lysate was clarified by centrifugation at 150,000 x g. GST--Cdt1 was incubated with glutathione resin (Fisher Scientific, 17--5132--01) and washed with 50 mM Tris \[pH7.6\], 300 mM NaCl, 0.05% NP--40, 1 mM DTT, 1 mM EDTA, and 10% glycerol. Cdt1 was cleaved off the resin with Prescission Protease.
Clb5--CDK (ySK119) was purified as previously described ([@bib43]). Cdc6 (pSKM033), ORC (ySDORC), Mcm2--7 (yST135), and Mcm2--7--Cdt1 (yST144) were purified as previously described ([@bib38]).
In vivo kinase inhibition {#s4-8}
-------------------------
For strains containing *cdk1--as*, the kinase was inhibited with 10 µM 1--NM--PP1 (Toronto Research Chemicals, A603003). For strains containing *ime2--as*, the kinase was inhibited with 20 µM 1--NA--PP1 (Cayman Chemical Co., NC1049860). For experiments comparing the effects of inhibiting one or both kinases, both inhibitors were added to each strain regardless of genotype to normalize for the effect of the inhibitor without the corresponding analog--sensitive allele. Both inhibitors were prepared from 20 mM stocks in DMSO.
ChIP--qPCR {#s4-9}
----------
Chromatin immunoprecipitations were performed as described with minor modifications ([@bib10]). 10 mL of cells were harvested for each sample, and 4% of the lysate was removed as an input control after sonication. Mcm2--7 was immunoprecipitated with 1.5 µL of UM185 (rabbit polyclonal antibody), whereas ORC was immunoprecipitated with 1 µL HM1108 (rabbit polyclonal antibody). Input and immunoprecipitated DNA from each sample were run in triplicate on a Light Cycler 480 II Real--Time PCR system (Roche). The relative amount of immunoprecipitated DNA vs. input DNA was calculated, and the highest sample, or the average of all G1 samples in [Figure 5](#fig5){ref-type="fig"}, was normalized to 1.0 within each experiment. Error bars represent the standard deviation from three PCR replicates.
In vitro kinase assay {#s4-10}
---------------------
Ime2 was incubated with 100--200 nM of purified substrate protein (Cdc6, ORC, or Mcm2--7/Cdt1) in Buffer H + 200 mM K--Glut, 1 mM ATP, and 5 µCi \[γ--^32^P\] ATP. Reactions were terminated after 45 min by boiling in sample buffer. Samples were loaded onto an SDS--PAGE gel, and total protein was visualized by Krypton fluorescent stain. ^32^P--modified proteins were detected by autoradiography.
iTRAQ LC--MS/MS {#s4-11}
---------------
All samples were analyzed with two biological replicates. 150 nM Ime2 (or buffer control) was incubated with 400 nM Cdc6 under the same buffer conditions and for the same time as in the helicase--loading assay. Proteins were reduced, alkylated, and digested with trypsin. Peptides were labeled using 4 of the 6 channels from the TMT 6plex kit (Thermo) performed per manufacturer's instructions. Samples labeled with the four different isotopic TMT reagents were combined and concentrated to completion in a vacuum centrifuge. house, 6 cm of 10 µm C18) and a self--pack 5 µm tip analytical column (12 cm of 5 µm C18, New Objective) over a 140 min gradient before nanoelectrospray using a QExactive Plus mass spectrometer (Thermo). The parameters for the full scan MS were: resolution of 70,000 across 350--2000 *m/z*, AGC 3e^6^, and maximum IT 50 ms. The full MS scan was followed by MS/MS for the top 10 precursor ions in each cycle with a NCE of 28 and dynamic exclusion of 30 s. Raw mass spectral data files (.raw) were searched using Proteome Discoverer (Thermo) and Mascot version 2.4.1 (Matrix Science). Mascot search parameters were: 10 ppm mass tolerance for precursor ions; 15 mmu for fragment ion mass tolerance; 2 missed cleavages of trypsin; fixed modification were carbamidomethylation of cysteine and TMT 6plex modification of lysines and peptide N--termini; variable modifications were methionine oxidation, tyrosine phosphorylation, and serine/threonine phosphorylation. TMT quantification was obtained using Proteome Discoverer and isotopically corrected per manufacturer's instructions, and were normalized to the mean of each TMT channel. Only peptides with a Mascot score greater than or equal to 25 and an isolation interference less than or equal to 30 were included in the data analysis. Mascot peptide identifications, phosphorylation site assignments, and quantification were verified manually with the assistance of CAMV ([@bib19]). Phosphorylation sites were assigned based on having an average enrichment of \>4--fold in the Ime2--treated samples compared to control samples.
Funding Information
===================
This paper was supported by the following grants:
- http://dx.doi.org/10.13039/100000011Howard Hughes Medical Institute Investigator Award to Stephen P Bell.
- http://dx.doi.org/10.13039/100000054National Cancer Institute Biopolymer Facility Support to Stephen P Bell.
- http://dx.doi.org/10.13039/100000057National Institute of General Medical Sciences Gradaute Student Fellowship to David V Phizicky.
We thank Angelika Amon, Iain Cheeseman, Audra Amasino, Ishara Azmi, Caitlin Blank, and Annie Zhang for comments on the manuscript. We thank all members of the Bell laboratory for helpful discussions. We thank Angelika Amon and John Diffley for yeast strains. D.V.P. was supported in part by a NIH Pre-Doctoral Training Grant (GM007287). SPB is an investigator with the Howard Hughes Medical Institute. This work was supported in part by the Koch Institute Support Grant P30-CA14051 from the NCI. We thank the Koch Institute Swanson Biotechnology Center for technical support, specifically the Biopolymers core.
Additional information {#s5}
======================
No competing interests declared.
Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing---original draft.
Supervision, Methodology, Writing---review and editing.
Conceptualization, Supervision, Funding acquisition, Project administration, Writing---review and editing.
Additional files {#s6}
================
10.7554/eLife.33309.027
###### In vitro Ime2--phosphorylation sites on Cdc6 from iTRAQ LC--MS/MS.
Related to [Figure 5E](#fig5){ref-type="fig"}. Shown are the phosphorylated Cdc6 peptides, the specific phosphorylated residue(s), the relative amount of those phosphopeptides detected in both biological replicates of buffer--treated and Ime2--treated Cdc6, and the average enrichment upon Ime2--treatment. See Materials and Methods for iTRAQ LC--MS/MS details.
10.7554/eLife.33309.028
10.7554/eLife.33309.031
Decision letter
Stillman
Bruce
Reviewing Editor
Cold Spring Harbor Laboratory
United States
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting your article \"Multiple kinases inhibit origin licensing and helicase activation to ensure reductive cell division during meiosis\" for consideration by *eLife.* Your article has been favorably evaluated by Andrea Musacchio (Senior Editor) and three reviewers, one of whom, Bruce Stillman (Reviewer \#1), is a member of our Board of Reviewing Editors.
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
Summary:
This paper addresses a fundamental, unresolved question concerning the regulation of DNA replication during progression through meiosis: How do cells carry out two rounds of chromosome segregation without an intervening round of DNA replication? The authors elucidate several distinct mechanisms which inhibit replication initiation between two consecutive meiotic nuclear divisions (MI and MII). In mitotically dividing cells, oscillation of CDK ensures that replication origins fire once per cell cycle: MCM helicase loading occurs when the cellular CDK level is low, and MCM is activated while loading is prevented when CDK level is high. In meiosis, the CDK level drops once at the MI to MII transition, but replication initiation does not occur. In frog oocyte extracts, addition of a CDK inhibitor promotes S phase between MI and MII, and in yeast simultaneous inhibition of CDK and Ime2 (a meiosis specific kinase related to CDK) leads to nuclear accumulation of an MCM protein during nuclear division stages. Thus, it is suggested that the two kinases function to prevent replication initiation when MCM associates to replication origins. But it was not known precisely how these two kinases do so.
The authors employed a previously developed system to synchronize meiosis progression in which they use a yeast strain with inducible Ndt80 (a transcription factor that triggers exit from meiotic prophase). This strain arrests in meiotic prophase and resumes progression upon Ndt80 induction. Using this system, authors found that MCM ChIP signals at replication origins are decreased after the release from prophase arrest and that either CDK or Ime2 activity is required to keep MCM ChIP signal at a low level. Results of an in vitro MCM loading assay demonstrate that Ime2 and CDK prevent MCM loading by phosphorylating both overlapping and distinct targets. Furthermore, during nuclear meiotic division stages, CDK and Ime2 repress Cdc6 (a protein that recruits MCM to origins) in part by phosphorylating Cdc6 within phospho-degron domains, and Sld2 (an MCM activator) degradation is dependent on phosphorylation sites for CDK and Cdc5 (the polo-like kinase).
This elegant paper provides detailed mechanistic insight into how replication is inhibited at the MI-MII transition stage in meiosis. The data are excellent and the paper is very well written. The paper is highly appropriate for *eLife*. However, there are a few areas that should be addressed to strengthen the paper.
Minor points:
1\) The dephosphorylation of Orc2 and Orc6 are very subtle and it is not clear if there is \"a partial but detectable reversal of ORC inhibition\" since the authors do not know if partial dephosphorylation is sufficient for reversing inhibition of ORC function. This statement should be qualified.
2\) [Figure 3C](#fig3){ref-type="fig"}. The amount of Ime1 kinase that inhibits Mcm207 loading is near stoichiometric with the pre-RC assembly proteins. Similarly, the kinase assay in [Figure 3---figure supplement 2B](#fig3s2){ref-type="fig"} suggests stoichiometric phosphorylation. These data might suggest that like CDK, Ime2 binds one or more of the pre-RC proteins. Has this been tested?
3\) A rather obvious question is whether DNA re-replication does in fact occur when the mechanisms delineated here are compromised. Have the authors tried to see if replication occurs when CDK and Ime2 are artificially oscillated by inhibition followed by washing out of the inhibitors in the non-phosphorylatable sld2 mutant background? The discussion point about Cdc7-Dbf4 regulation raises the possibility of yet other mechanisms to inhibit re-replication, but it would strengthen the paper to test just how redundant these mechanisms are.
4\) Please show representative micrographs to illustrate the cell cycle staging. This could be provided in [Figure 1---figure supplement 1](#fig1s1){ref-type="fig"} (would not be needed for all).
5\) [Figure 1B](#fig1){ref-type="fig"} and C. No negative control is presented to evaluate ChIP signal specificity. This is more important for the ORC ChIP, but would have been good to show for MCM as well. Also, it would be better if the y-axis in this and other ChIP plots showed efficiency (% of input), rather than the max-normalized value (arbitrary units).
6\) [Figure 5---figure supplement 2B and D](#fig5s2){ref-type="fig"}. Experiments here and elsewhere were conducted in an artificially synchronized meiosis using inducible Ndt80. Normally, the pachytene checkpoint monitors failures in homologous recombination and chromosome synapsis, and inhibits NDT80 expression until the checkpoint is satisfied. But even in wild-type meiosis, normal recombination processes delay Ndt80 activation to prevent premature meiotic entry. Thus, it is important to consider that the artificial synchrony system allows for cells to be driven out of prophase before they have completed recombination and other events. Normally, the majority of cells have completed prophase by 6 hr after the induction of sporulation. However, the sporulation is intrinsically variable. In the case of cells that are progressing somewhat slowly, \"premature\" induction of Ndt80 could cause a substantial number of achiasmate chromosomes, which would lead to activation of the spindle assembly checkpoint. This could be problematic especially when the next manipulations take place within a fixed time interval. Such a scenario perhaps explains why a lower fraction of cells completed MI in [Figure 5---figure supplement 2B and D](#fig5s2){ref-type="fig"}. If possible, it might be useful to supplement or replace these data with better cultures or with cultures synchronized with an alternative method (e.g., pCUP1-IME4/IME4 strain or cdc7-as inhibition followed by inhibitor washout). At the least, it is not appropriate to claim the 7h45\' time point as MI-MII, since nearly half of the cells are still in metaphase I, so the text should be modified accordingly.
7\) [Figure 6E](#fig6){ref-type="fig"} needs to be bolstered with additional data. The western signals need to be normalized to input control, for which Ponceau staining is probably sub-optimal. A dilution series to demonstrate linearity of the western blot quantification is needed (can be supplemental). There are no error bars on this experiment; how many times was this repeated?
10.7554/eLife.33309.032
Author response
> \[...\] This elegant paper provides detailed mechanistic insight into how replication is inhibited at the MI-MII transition stage in meiosis. The data are excellent and the paper is very well written. The paper is highly appropriate for eLife. However, there are a few areas that should be addressed to strengthen the paper.
>
> Minor points:
>
> 1\) The dephosphorylation of Orc2 and Orc6 are very subtle and it is not clear if there is \"a partial but detectable reversal of ORC inhibition\" since the authors do not know if partial dephosphorylation is sufficient for reversing inhibition of ORC function. This statement should be qualified.
We understand and agree with the reviewers' concern with this statement. We have now qualified our statement to read "a partial but detectable decrease of ORC phosphorylation".
> 2\) [Figure 3C](#fig3){ref-type="fig"}. The amount of Ime1 kinase that inhibits Mcm207 loading is near stoichiometric with the pre-RC assembly proteins. Similarly, the kinase assay in [Figure 3---figure supplement 2B](#fig3s2){ref-type="fig"} suggests stoichiometric phosphorylation. These data might suggest that like CDK, Ime2 binds one or more of the pre-RC proteins. Has this been tested?
The most direct test for Ime2 binding to pre-RC proteins we have done is to look for the association of Ime2 with the ORC-Cdc6-Cdt1-MCM complex formed in ATPyS. Because this is a stable complex containing all the components required for helicase loading we felt this was the best test for Ime2 binding. In contrast to a model in which Ime2 inhibits helicase loading through binding to helicase-loading proteins, we found that Ime2 did not stoichiometrically bind to these proteins in this experimental condition (see [Author response image 1](#respfig1){ref-type="fig"}). We realize that this does not exclude binding of Ime2 to these proteins at other times in the reaction. On the other hand, we show that Ime2-mediated inhibition of helicase loading is completely dependent on its kinase activity ([Figure 3D](#fig3){ref-type="fig"}), indicating that any potential role for Ime2-binding to its substrates is not sufficient for inhibition.
{#respfig1}
> 3\) A rather obvious question is whether DNA re-replication does in fact occur when the mechanisms delineated here are compromised. Have the authors tried to see if replication occurs when CDK and Ime2 are artificially oscillated by inhibition followed by washing out of the inhibitors in the non-phosphorylatable sld2 mutant background? The discussion point about Cdc7-Dbf4 regulation raises the possibility of yet other mechanisms to inhibit re-replication, but it would strengthen the paper to test just how redundant these mechanisms are.
We agree with the reviewers that this would be a very exciting experiment. Unfortunately, we found that inhibition of Cdk1-as during meiosis followed by washing away the inhibitor does not allow Cdk1-as to be reactivated. These results are consistent with previous experiments showing that transient inhibition of Cdk1-as at any point during meiosis caused cells to be unable to progress any further in meiosis (Holt et al., 2007 -- see Discussion). Like Holt et al., we believe that these data suggest a requirement for basal CDK activity during meiosis. It is possible that complete CDK inhibition allows for Sic1 to accumulate preventing CDK reactivation upon inhibitor washout.
> 4\) Please show representative micrographs to illustrate the cell cycle staging. This could be provided in [Figure 1---figure supplement 1](#fig1s1){ref-type="fig"} (would not be needed for all).
We have now included representative images of cells in Metaphase I, Anaphase I, Metaphase II, and Anaphase II in [Figure 1---figure supplement 1](#fig1s1){ref-type="fig"}.
> 5\) [Figure 1B](#fig1){ref-type="fig"} and C. No negative control is presented to evaluate ChIP signal specificity. This is more important for the ORC ChIP, but would have been good to show for MCM as well. Also, it would be better if the y-axis in this and other ChIP plots showed efficiency (% of input), rather than the max-normalized value (arbitrary units).
We agree with the reviewers that reporting the negative control and% of input is important. We have now included an additional figure ([Figure 1---figure supplement 2](#fig1s2){ref-type="fig"}) showing that our ChIP for both ORC and MCM is specific to origin DNA compared to non-origin DNA. The previous "[Figure 1---figure supplement 2](#fig1s2){ref-type="fig"}" is now "[Figure 1---figure supplement 3](#fig1s3){ref-type="fig"}".
We have also included the % of input detected for each protein and origin combination in the corresponding figure legends. We have not changed the y-axes of our graphs, however, to allow protein association at different origins to be compared on the same graph (the peak% input value varies significantly between origins).
> 6\) [Figure 5---figure supplement 2B and D](#fig5s2){ref-type="fig"}. Experiments here and elsewhere were conducted in an artificially synchronized meiosis using inducible Ndt80. Normally, the pachytene checkpoint monitors failures in homologous recombination and chromosome synapsis, and inhibits NDT80 expression until the checkpoint is satisfied. But even in wild-type meiosis, normal recombination processes delay Ndt80 activation to prevent premature meiotic entry. Thus, it is important to consider that the artificial synchrony system allows for cells to be driven out of prophase before they have completed recombination and other events. Normally, the majority of cells have completed prophase by 6 hr after the induction of sporulation. However, the sporulation is intrinsically variable. In the case of cells that are progressing somewhat slowly, \"premature\" induction of Ndt80 could cause a substantial number of achiasmate chromosomes, which would lead to activation of the spindle assembly checkpoint. This could be problematic especially when the next manipulations take place within a fixed time interval. Such a scenario perhaps explains why a lower fraction of cells completed MI in [Figure 5---figure supplement 2B and D](#fig5s2){ref-type="fig"}. If possible, it might be useful to supplement or replace these data with better cultures or with cultures synchronized with an alternative method (e.g., pCUP1-IME4/IME4 strain or cdc7-as inhibition followed by inhibitor washout). At the least, it is not appropriate to claim the 7h45\' time point as MI-MII, since nearly half of the cells are still in metaphase I, so the text should be modified accordingly.
We have altered the text (subsection "CDK and Ime2 cooperate to inhibit Mcm2-7 loading and Cdc6 expression during the MI-MII transition") and [Figure 5](#fig5){ref-type="fig"} to more accurately reflect the stage of meiosis upon CDK/Ime2 inhibition. We specifically highlight the percentage of cells in anaphase I and acknowledge in the text that not all cells are at this stage of meiosis (first paragraph of the aforementioned subsection). We focus on these cells because anaphase I is good marker for cells as CDK activity decreases upon entry into the MI-MII transition. Two pieces of evidence support this claim. First, previous work demonstrated that Cdk1-cyclin B complexes have decreased activity in anaphase I relative to metaphase I (Carlile and Amon, 2008). Second, the peak of anaphase I corresponds with the decreased ORC phosphorylation and the beginning of Cdc6 reaccumulation we observe in [Figure 2](#fig2){ref-type="fig"}, markers of decreased CDK activity. We note that it is likely that CDK, and potentially Ime2, inhibits helicase loading throughout the meiotic divisions and not just at the MIMII transition. Thus, the fact that we see helicase reloading at any time during the meiotic divisions (there is little doubt that the majority of the cells have at least entered MI) strongly supports our primary conclusion: that the two kinases both contribute to inhibiting helicase loading.
The order of [Figure 5---figure supplements 1](#fig5s1){ref-type="fig"} and 2 have been switched from the previously submitted version to account for changes in the text.
With regard to why fewer cells have completed MI in the current [Figure 5---figure supplement 1B](#fig5s1){ref-type="fig"} and 1D relative to 1A and 1C of the same figure, this difference is because CDK has been inhibited in the 1B and 1D cultures but not in the other two cultures. CDK is required for progression through MI, and thus its inhibition prevents MI completion in a large proportion of cells. We have now included immunofluorescence quantification from an 8h30min time point from these same cultures that was not treated with the kinase inhibitors ([Figure 5---figure supplement 1](#fig5s1){ref-type="fig"}). This data shows that \>85% of cells have completed MI at this time point in all four cultures.
> 7\) [Figure 6E](#fig6){ref-type="fig"} needs to be bolstered with additional data. The western signals need to be normalized to input control, for which Ponceau staining is probably sub-optimal. A dilution series to demonstrate linearity of the western blot quantification is needed (can be supplemental). There are no error bars on this experiment; how many times was this repeated?
We understand the reviewers' concern with the data in this form. We generally prefer Ponceau staining over an individual protein since monitoring many proteins as opposed to just PGK1 make the normalization less dependent on the stable expression of one protein. However, we do understand that it is more difficult to quantify the Ponceau staining. Thus, we have now normalized our western signals to a PGK1 loading control and found the same results. This experiment has now been repeated three times, and we report the mean and standard deviation from these three experiments. The text has been modified to reflect the quantification shown in [Figure 6E](#fig6){ref-type="fig"} (subsection "Cdc5 and CDK promote the degradation of Sld2, an essential helicase- activation protein", last paragraph). We have also included a dilution series to show that the western blot quantification is accurate within a factor of two across a 32-fold dilution range ([Figure 6---figure supplement 3](#fig6s3){ref-type="fig"}).
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Psychiatric comorbidities are relatively common in epilepsy patients. Among these comorbidities, depression and cognition impairment appear to be the major comorbidities associated with chronic epilepsy (Gaitatzis et al., [@B17]; Kanner, [@B27], [@B28]; LaFrance et al., [@B30]; Loughman et al., [@B33]; Tai et al., [@B53]). Despite accumulating epidemiological and animal model evidence suggesting a correlation between epilepsy and the psychiatric comorbidities of depression and cognitive deficit, the biological mechanisms underlying this correlation remains poorly understood. Neuronal nitric oxide synthase (nNOS) has been widely distributed in the neurons, where it produces nitric oxide (NO) in the process of converting L-arginine into citrulline, with the presence of NADPH. NO in the brain has been involved in synaptogenesis, neural transmission, learning and memory and synaptic plasticity. nNOS has been largely distributed in the brain regions of striatum, hippocampus, hypothalamus and amygdala, where it is involved in the regulation of cognition and affective behaviors (Downen et al., [@B14]; Kittner et al., [@B29]; Saavedra et al., [@B45]). Indeed, increasing evidence has shown that nNOS plays a pivotal role in multiple psychiatry disorders, including schizophrenia, depression and anxiety (Kittner et al., [@B29]; Reif et al., [@B42]; Brzustowicz, [@B6]; Delorme et al., [@B12]; Lawford et al., [@B31]; Saavedra et al., [@B45]). MAPK and PI3K/AKT are two major intracellular signaling pathways involved in the brain activities. Recent studies demonstrate that MAPK (Hutton et al., [@B24]; Ullrich et al., [@B55]) and PI3K/AKT (Papaleo et al., [@B37]) signaling pathways are associated with a number of neuropsychiatric disorders. However, whether these signaling pathways are dependent on nNOS activities remains unclear.
Pentylenetetrazole (PTZ) kindling is a chronic epilepsy model, in which a progressive seizure development is observed. PTZ kindling causes alterations in the molecular and cellular levels, which are responsible for neuronal plasticity. It has been demonstrated that PTZ kindling-induced morphological changes are usually accompanied with long-lasting changes in emotional behavior (Franke and Kittner, [@B15]; Mortazavi et al., [@B35]). A recent study reports that PTZ kindling induces depression-like behavior and cognition deficits (Choudhary et al., [@B8]), suggesting chronic epilepsy is associated with psychiatric comorbidities.
In our previous study, we demonstrated that hippocampal nNOS expression and enzymatic activity have been dramatically enhanced after the mice were kindled (Zhu et al., [@B63]). Here, in this study, we hypothesize that this increased nNOS activity acts through MAPK and PI3K/AKT signaling pathways to trigger cognition deficit and depressive-like behavior in PTZ-kindled mice.
Materials and methods {#s2}
=====================
Animals
-------
Male C57BL/6J mice (4--6 weeks old; weighing 19 ± 2 g at the beginning of the experiments) were obtained from Nanjing Biomedical Research Institute of Nanjing University (NBRI) (Nanjing, China). Mice lacking nNOS (B6;129S4-Nos1^tm1Plh^) obtained from NBRI were backcrossed to C57BL/6J strain and the heterozygotes were intercrossed to obtain mutation homozygotes. Male homozygous nNOS-null (nNOS^−/−^) and their wild-type (nNOS^+/+^) littermates (4--6 weeks old) were used in the experiment. The animals were housed in plastic cages and kept in a regulated environment (22 ± 1°C) with an artificial 12 h light/dark cycle (lighted from 7:00 a.m. to 7:00 p.m.). Food and tap water were available *ad libitum*. Procedures for PTZ induced-kindling and all subsequent experiments were approved by the Animal Care and Use Committee of Medical School of Southeast University. All efforts were made to minimize animal suffering and discomfort and to reduce the number of animals used.
PTZ kindling procedure
----------------------
PTZ kindling model was produced as we previously described (Zhu et al., [@B64]). Briefly, mice were treated with a subconvulsive dose of PTZ (Sigma Aldrich, St. Louis, MO, USA) at 35 mg/kg intraperitoneally on every second day for eleven total injections. (Figure [1A](#F1){ref-type="fig"}). Vehicle control mice received the same amount of saline. After each PTZ injection, convulsive behaviors were observed for 30 min by a video monitoring system (HK vision, Hanzhou, China). The seizure intensity was evaluated by using the following scale. Stage 0, no response; Stage 1, ear and facial twitching; Stage 2, convulsive twitching axially through the body; Stage 3, myoclonic jerks and rearing; Stage 4, turning over onto the side, wild running, and wild jumping; Stage 5, generalized tonic-clonic seizures; and Stage 6, death (Schroder et al., [@B46]; Becker et al., [@B3]; Mizoguchi et al., [@B34]).
{#F1}
Cognitive function evaluation
-----------------------------
Cognitive function evaluation is conducted by novel object recognition (NOR) test, which is based on a rodent\'s nature to preferentially investigate unfamiliar objects rather than familiar objects. NOR test was performed as previously described (Bevins and Besheer, [@B4]; Nomoto et al., [@B36]) with some modifications. Briefly, NOR testing is consisting of three phases: habituation, training and test (Figure [2A](#F2){ref-type="fig"}). During the habituation session, mice were placed in the empty arena and allowed to freely explore without objects 15 min per day for two consecutive days. During training session, mice were placed in the arena with two identical objects A and A\', and were allowed to explore the objects for 10 min. During the test session, mice were placed in the arena, where one familiar object A\' was replaced with a novel object B, and mice were allowed to explore the object for 5 min. The arena and the objects were cleaned thoroughly with 75% ethanol between trials to remove any olfactory cues. A video monitoring system (Hikvision video monitoring system, Hangzhou, China) was used to capture the animal behavior for later analysis. Time spent exploring each object was calculated by an observer blind to the experimental conditions and was expressed as a percentage of the total exploration time.
{#F2}
Sucrose preference test
-----------------------
After 4 days of NOR test, the mice were recovered for 24 h and then subjected to 2 days of sucrose preference test. The sucrose preference test was used to assess the taste preference of sweetened water. A diminished preference for the sweetened water is a sign of anhedonia, indicating depressive-like behavior. The sucrose preference test was conducted as previously described (Strekalova et al., [@B52]; Snyder et al., [@B50]; Wu et al., [@B60]) with some modifications. Briefly, mice were given a free choice between two bottles, one with 1% sucrose solution and another with tap water for 48 h. To avoid place preference in drinking behavior, the position of the bottles was switched after 24 h. The consumption of tap water and sucrose solution was estimated simultaneously in control and experimental groups by weighing the bottles. The preference for sucrose was calculated as a percentage of consumed sucrose solution of the total amount of liquid drunk. The food preference is also calculated as a control to demonstrate that mice do not show a place preference.
Tail-suspension test and forced swim test
-----------------------------------------
After the sucrose preference test, the mice were recovered for 24 h and then subjected to tail-suspension test followed by forced swim test. The tail-suspension test was used to evaluate depressive-like behaviors in animals. The method is based on the observations that mice suspended by the tails show immobility, which reflects despair of hope. The tail-suspension test was performed as previously described (Steru et al., [@B51]) with small modifications. Each mouse was suspended at a height of 50 cm using a thread tied with the tip of tail. The mice were considered to be immobile when they did not show any movement of body and hanged passively. The duration of immobility was recorded throughout the 5 min test period. The forced swim test was created as a situation of despair and allows to assessing the depressive-like behavior. Although some studies indicate that FST reflects a passive stress coping and adaption mechanism (de Kloet and Molendijk, [@B11]; Commons et al., [@B9]), it is still the most widely used tool to test depressive-like behavior. The forced swim test was performed as previously described (Porsolt et al., [@B39]). Briefly, the mice were recovered for 24 h after the tail-suspension test and were placed in a transparent glass cylinder (height, 30 cm; diameter, 10 cm) filled with 20 cm height of water at 23 ± 1°C. Mice were judged to be immobile when if floated in an upright position and made only small movements to keep its head above water for more than 2 s. The duration of immobility during the 6 min of trial was recorded using the video monitoring system (Hikvision video monitoring system, Hangzhou, China). After each test, the container was washed and refilled with fresh water.
Measurement of ROS production
-----------------------------
ROS production was measured by a cell membrane-permeable superoxide-sensitive fluorescent dye dihydroethidium (DHE) (Sigma-Aldrich, St. Louis, MO, USA) as we previously described (Zhu et al., [@B65]). Briefly, hippocampal sections were incubated in 1 μM DHE (in 0.1 M PBS, PH = 7.4) for 15 min in the dark room. Hippocampal sections were then rinsed with PBS three times and mounted on gelatin-coated slides. DHE fluorescence was detected by a confocal laser scanning microscope (Olympus LSM-GB200, Japan) using an excitation wavelength of 520--540 nm. Fluorescence was quantified with the Image J software program (NIH, Bethesda, MD, USA).
Western blotting
----------------
Hippocampal tissues were lysed for 15 min in tissue lysis buffer (Beyotime Biotechnology, China). The protein concentration was measured using a BCA protein assay kit (Pierce, Rockford, IL, USA). Hippocampal proteins were then separated by 12% acrylamide denaturing gels (SDS-PAGE) and were transferred to nitrocellulose membranes (Amersham, Little Chalfont, UK) by a Bio-Rad mini-protein-III wet transfer unit (Hercules, CA, USA) overnight at 4°C.
The membranes were incubated with 5% non-fat milk in TBST (10 mmol/l Tris pH = 7.6, 150 mmol/L NaCl, 0.01%Tween-20) for 1 h at room temperature followed by several washes, then were incubated with rabbit anti-ERK and phospho-ERK (1:2,500; Abcam, Temecula, CA, USA), rabbit anti-p38 and phospho-p38 (1:2,000; Cell signaling, Danvers, MA, USA), mouse anti-PI3K (1:2,000; Cell signaling, Danvers, MA, USA), rabbit anti-AKT and phosphor-AKT (1:2,000; Cell signaling, Danvers, MA, USA) and rabbit anti-β-actin (1:5,000; Sigma-Aldrich, St, Louis, USA) in TBST overnight at 4°C. After several washes, the membranes were incubated with HRP-linked secondary antibody (Boster Bioengineering, Wuhan, China) diluted 1:5,000 for 1 h. After several washes, the antibody was detected by an enhanced chemiluminescence (ECL) (Millipore, Billerica, MA, USA) by using a MicroChemi chemiluminescent image analysis system (DNR Bio-imaging Systems, Jerusalem, Israel). Blots were quantified using the Image J software (NIH, Bethesda, MD, USA).
Statistical analysis
--------------------
All data are presented as the means ± S.E.M. Statistical significance was determined by using unpaired two-tailed Student\'s *t*-test for two group\'s comparison and by using two-way ANOVA for multi-group comparisons, and repeated-measures ANOVA. Tukey\'s test was used for *post-hoc* comparisons. A Spearman rank correlation coefficient was used to determine any correlation between the immobility duration under the condition of forced swim and tail suspension and the seizure score during kindling development. Differences were considered to be significant for values of *p* \< 0.05.
Results {#s3}
=======
PTZ kindling induced cognitive impairment
-----------------------------------------
The PTZ kindling model was successfully established by giving the mice with PTZ at a dose of 35 mg/kg every other day for 11 doses (Figure [1A](#F1){ref-type="fig"}). PTZ-kindled mice showed a gradual increase of seizure intensity, compared with the mice in control group (Figure [1B](#F1){ref-type="fig"}). PTZ kindling as a chronic epilepsy experimental model is usually associated with neuronal plasticity and causing psychiatric comorbidities. To determine whether PTZ kindling affects cognition and depressive-like behavior in mice, we performed a variety of neurobehavior tests 24 h after the mice were fully kindled. Firstly, we examined the cognitive function of PTZ-kindled mice by using a novel object recognition (NOR) test (Figure [2A](#F2){ref-type="fig"}). During the training phase, both PTZ-kindled and vehicle control mice spent almost the same percent of time exploring the two identical objects A and A\' (Figure [2B](#F2){ref-type="fig"}). However, in the testing phase, the vehicle control mice spent more time exploring the novel object B compared to the familiar object A (Figure [2C](#F2){ref-type="fig"}), indicating that they remembered the familiar object A from the training phase and thus had a preference for the novel object B in the testing phase. In contrast, during the testing phase, PTZ-kindled mice spent almost the same amount of time exploring both the novel object B and the familiar object A (Figure [2C](#F2){ref-type="fig"}), suggesting that these mice did not remember the familiar object A during the training phase and had a cognitive deficit.
PTZ kindling induced depressive-like behavior
---------------------------------------------
We examined the depressive-like behaviors by using sucrose preference, forced swim and tail suspension test. Our data show that PTZ-kindled mice showed significantly reduced sucrose water consumption (Figure [3A](#F3){ref-type="fig"}) and the percentage of sucrose water consumption (Figure [3C](#F3){ref-type="fig"}) compared to vehicle control mice. However, the tap water (Figure [3B](#F3){ref-type="fig"}) and food consumption (Figure [3D,E](#F3){ref-type="fig"}) between these two group of mice remains similar. Immobility in forced swim and tail-suspension test represents a symptom of depression. Our data show that PTZ-kindled mice in both forced swim (Figure [3F](#F3){ref-type="fig"}) and tail-suspension test (Figure [3G](#F3){ref-type="fig"}) displayed significant increase of immobility duration, suggesting these mice had depressive-like behavior. Furthermore, we examined the immobility duration under the conditions of the forced swim and tail suspension correlated with the severity of behavioral seizures during kindling process in individual mice. Our data show a strong positive correlation between the duration of immobility and the seizure score during PTZ kindling development in forced swim as well as tail-suspension test (Figure [3H](#F3){ref-type="fig"}). Taken together, these results suggest that PTZ kindling induced cognitive impairment and depressive-like behaviors.
{#F3}
PTZ kindling-induced cognitive impairment and depressive-like behavior is dependent on nNOS activity
----------------------------------------------------------------------------------------------------
Growing body of evidence demonstrated that nNOS plays a pivotal role in psychiatry disorders, to confirm the involvement of nNOS in PTZ kindling-induced cognitive impairment and depressive-like behavior, we tested the cognition and depressive-like behavior in nNOS^−/−^ mice and their wildtype littermates under normal and PTZ kindling conditions. Firstly, we examined the cognitive function by using NOR test as described above (Figure [2A](#F2){ref-type="fig"}). During the training phase, the wildtype control, wildtype kindled, nNOS^−/−^ control and nNOS^−/−^ kindled mice spent almost the same percent of time exploring the two identical objects A and A\' (Figure [4A](#F4){ref-type="fig"}). However, in the testing phase, the vehicle control mice, nNOS^−/−^ control and nNOS^−/−^ kindled mice spent more time exploring the novel object B compared to the familiar object A (Figure [4B](#F4){ref-type="fig"}), indicating that these mice remembered the familiar object A from the training phase and thus had a preference for the novel object B in the testing phase. In contrast, during the testing phase, the wildtype kindled mice spent almost the same amount of time exploring both the novel object B and the familiar object A (Figure [4B](#F4){ref-type="fig"}), indicating that these mice did not remember the familiar object A during the training phase and had a cognitive deficit. These data suggests that PTZ kindling induced cognitive impairment. Depletion of nNOS rescued PTZ kindling-induced cognitive deficit.
{#F4}
We then examined the depressive-like behaviors by using sucrose preference, forced swim and tail suspension test. Our data show that wildtype kindled mice showed significantly reduced percentage of sucrose water consumption compared to wildtype control mice (Figure [4D](#F4){ref-type="fig"}), moreover, nNOS^−/−^ kindled mice showed increased percentage of sucrose water consumption compared to wildtype kindled mice (Figure [4D](#F4){ref-type="fig"}). nNOS^−/−^ control and wildtype control mice exhibited similar percentage of sucrose water consumption (Figure [4D](#F4){ref-type="fig"}). For sucrose water consumption, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 28)~ = 14.69, *p* \< 0.001\], but not a significant effect of genotype \[*F*~(1,\ 28)~ = 1.79, *p* = 0.191\], nor a significant effect of PTZ treatment × genotype interaction \[*F*~(1,\ 28)~ = 2.18, *p* = 0.15\]. For the percentage of sucrose water consumption, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 28)~ = 17.63, *p* \< 0.001\], but not a significant effect of genotype \[*F*~(1,\ 28)~ = 2.30, *p* = 0.14\], however, there is a significant effect of PTZ treatment × genotype interaction \[*F*~(1,\ 28)~ = 5.11, *p* = 0.032\]. A Tukey *post-hoc* test revealed that WT kindled mice showed a significant lower percentage of sucrose water consumption than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice showed a significant higher percentage of sucrose water consumption than WT kindled mice (*p* = 0.015). In forced swim and tail-suspension test, we found that wildtype kindled mice displayed significant increase of immobility duration compared to wildtype control mice in both forced swim (Figure [4E](#F4){ref-type="fig"}) and tail-suspension test (Figure [4F](#F4){ref-type="fig"}). Furthermore, nNOS^−/−^ kindled mice displayed significant decrease of immobility duration compared to wildtype kindled mice in both of these two tests (Figures [4E,F](#F4){ref-type="fig"}), while nNOS^−/−^ control and wildtype control mice have similar immobility duration in both of these two test (Figures [4E,F](#F4){ref-type="fig"}). For the immobility time in force swim test, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 28)~ = 25.06, *p* \< 0.001\], genotype \[*F*~(1,\ 28)~ = 4.94, *p* = 0.034\], as well as PTZ treatment × genotype interaction \[*F*~(1,\ 28)~ = 4.74, *p* = 0.011\]. A Tukey *post-hoc* test revealed that WT kindled mice had longer duration of immobility than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice had shorter duration of immobility than WT kindled mice (*p* = 0.002). For the immobility time in tail suspension test, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 28)~ = 24.36 *p* \< 0.001\], genotype \[*F*~(1,\ 28)~ = 5.02, *p* = 0.033\], as well as PTZ treatment × genotype interaction \[*F*~(1,\ 28)~ = 4.40, *p* = 0.045\]. A Tukey *post-hoc* test revealed that WT kindled mice had longer duration of immobility than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice had shorter duration of immobility than WT kindled mice (*p* = 0.006). These data suggests that PTZ kindling induced depressive-like behavior. Depletion of nNOS suppressed PTZ kindling-induced depressive-like behavior. Taken together, these results indicated that PTZ kindling-induced cognitive impairment and depressive-like behavior is dependent on nNOS activity.
PTZ kindling-induced hippocampal ROS production is dependent on nNOS activity
-----------------------------------------------------------------------------
To explore whether increased oxidative stress in PTZ-kindled mice is relevant to nNOS signaling, we detected hippocampal reactive oxygen species (ROS) level in nNOS^−/−^ mice as well as their wildtype littermates under normal or PTZ kindling conditions. Our results show that hippocampal ROS production, which was measured by the DHE fluorescence intensity, was remarkably enhanced in the wildtype kindled mice in comparison to wildtype control mice, while the hippocampal DHE fluorescence intensity in nNOS^−/−^ kindled mice was dramatically decreased compared to wildtype kindled mice (Figures [5A,B](#F5){ref-type="fig"}), suggesting PTZ kindling-induced hippocampal ROS production is dependent upon nNOS activity. For DHE fluorescence intensity, two-way ANOVA revealed a significant main effect of drug treatment \[*F*~(1,\ 16)~ = 43.70, *p* \< 0.001\] and genotype \[*F*~(1,\ 16)~ = 14.22, *p* = 0.002\], as well as drug treatment × genotype interaction \[*F*~(1,\ 16)~ = 12.79, *p* = 0.003\]. A Tukey *post-hoc* test revealed that WT kindled mice had significant higher level of DHE intensity than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice had significant lower level of DHE intensity than WT kindled mice (*p* \< 0.001).
{#F5}
PTZ kindling activates nNOS-dependent MAPK and PI3K/AKT signaling pathways
--------------------------------------------------------------------------
The MAPK signaling pathway is involved in modulation of various physiological and pathological events. The correlation of the MAPK signaling pathway with ROS has been investigated in many studies (Ramos-Nino et al., [@B41]; Cakir and Ballinger, [@B7]; Batra et al., [@B2]; Lee et al., [@B32]). The MAPK pathway comprises extra-cellular signal-regulated kinases (ERK 1/2), the p38 kinase, and the stress-activated protein kinase or c-Jun N-terminal kinase (SAPK/JNK) (Seger and Krebs, [@B47]). To determine whether MAPK signaling pathway is activated in the hippocampus of PTZ-kindled mice and whether this signaling pathway is dependent on nNOS activation, we detected ERK and p38 and their phosphorylation in nNOS^−/−^ mice and their wildtype littermates under normal and PTZ kindling conditions by western blot. Our results revealed that phosphorylation of p38 and ERK in the hippocampus of wildtype kindled mice was significantly increased compared to wildtype control mice. Moreover, the phosphorylation of p38 and ERK in the hippocampus of nNOS^−/−^ kindled mice was significantly decreased compared to that of wild type kindled mice (Figures [5C--E](#F5){ref-type="fig"}). For ERK phosphorylation, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 16)~ = 15.46, *p* = 0.001\], genotype \[*F*~(1,\ 16)~ = 9.6, *p* \< 0.007\], as well as PTZ treatment × genotype interaction \[*F*~(1,\ 16)~ = 9.89, *p* = 0.006\]. A Tukey *post-hoc* test revealed that WT kindled mice had higher ERK phosphorylation level than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice had significant lower ERK phosphorylation level than WT kindled mice (*p* \< 0.001). For p-38 phosphorylation, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 16)~ = 50.59, *p* \< 0.001\], genotype \[*F*~(1,\ 16)~ = 31.06, *p* \< 0.001\], as well as PTZ treatment × genotype interaction \[*F*~(1,\ 16)~ = 34.93, *p* \< 0.001\]. A Tukey *post-hoc* test revealed that WT kindled mice had higher p38 phosphorylation level than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice had significant lower p38 phosphorylation level than WT kindled mice (*p* = 0.002). These data suggests that PTZ kindling activates nNOS dependent MAPK signaling pathway.
PI3K/AKT signaling pathway is another important signaling pathway which is involved in regulating redox status (Uranga et al., [@B56]; Hambright et al., [@B21]). To determine whether PI3K/AKT signaling pathway is activated in the hippocampus of PTZ-kindled mice and whether this signaling pathway is dependent on nNOS activation, we detected PI3K, AKT and phosphorylation of AKT in nNOS^−/−^ mice and their wildtype littermates under normal and PTZ kindling conditions by western blot. Our results showed that PI3K level and phosphorylation of AKT in the hippocampus of wildtype kindled mice was significantly increased compared to wildtype control mice. Moreover, the PI3K level and phosphorylation of AKT in the hippocampus of nNOS^−/−^ kindled mice was significantly decreased compared to that of wild type kindled mice (Figures [5F--H](#F5){ref-type="fig"}). For PI3K level, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 16)~ = 14.63, *p* = 0.001\], genotype \[*F*~(1,\ 16)~ = 10.55, *p* \< 0.005\], as well as PTZ treatment × genotype interaction \[*F*~(1,\ 16)~ = 5.49, *p* = 0.032\]. A Tukey *post-hoc* test revealed that WT kindled mice had higher PI3K level than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice had significant lower PI3K level than WT kindled mice (*p* \< 0.001). For AKT phosphorylation level, two-way ANOVA revealed a significant main effect of PTZ treatment \[*F*~(1,\ 16)~ = 16.29, *p* \< 0.001\], genotype \[*F*~(1,\ 16)~ = 10.84, *p* = 0.005\], as well as PTZ treatment × genotype interaction \[*F*~(1,\ 16)~ = 15.40, *p* = 0.001\]. A Tukey *post-hoc* test revealed that WT kindled mice had higher AKT phosphorylation level than WT ctrl mice (*p* \< 0.001), and nNOS^−/−^ kindled mice had significant lower AKT phosphorylation level than WT kindled mice (*p* \< 0.001). These data suggests that PTZ kindling activates nNOS dependent PI3K/AKT signaling pathway. Taken together, these results suggest that PTZ kindling activates both MAPK and PI3K/AKT signaling pathways and the activation of these signaling pathways are dependent on nNOS activation.
Discussion {#s4}
==========
Cognitive dysfunction and depressive like behavior have been reported as main neurobehavioral comorbidities of chronic epilepsy, which significantly impact the outcomes and affects the life quality of epilepsy patients. Cognitive impairment in is evident in children with epilepsy. It is reported that children with generalized nonabsence seizures were at increased risk for learning abilities (Zalachoras et al., [@B61]). Furthermore, Children who have pharmacoresistant seizures appear to have lower IQ scores than children with well controlled seizures (Guo and Commons, [@B20]). Adults with chronic epilepsy are also reported to have cognitive impairment (Hutton et al., [@B24]; Shrestha et al., [@B49]). Depression is regarded as the most common comorbid condition of epilepsy, with prevalence in the range of 25--55% in epilepsy patients. Moreover, the incidence of depression is remarkably higher in epilepsy patients than that in the normal people (Cramer et al., [@B10]). PTZ Kindling is a well-established chronic epilepsy model that has been extensively studied to understand the pathological mechanisms of epilepsy. Here in this study, we found that PTZ kindling triggered cognition impairment and depressive like behavior, which is in agreement with previous reports (Russo et al., [@B44]; Loughman et al., [@B33]; Tai et al., [@B53]). Although epilepsy is known to be relevant to a high incidence of cognition deficits and depressive-like behavior, the responsible underlying mechanisms remains elusive. NO has been recognized as a neuronal messenger, which is involved in regulation the balance of neurotransmission (West et al., [@B58]; Garthwaite, [@B18]; Raju et al., [@B40]). It is suggested that alterations of NO signaling may contribute to the pathophysiology of cognition deficits (Walton et al., [@B57]; Funk and Kwan, [@B16]) and depression (Zhou et al., [@B62]; Gigliucci et al., [@B19]). A previous study demonstrated that nNOS accounts for approximately 90% of the overall NO production in the brain (Hara et al., [@B22]), suggesting that nNOS is mainly responsible for the NO signaling-mediated pathophysiological process in the brain. Our previous study and others\' reported that PTZ-kindling enhanced hippocampal nNOS expression and enzymatic activity (Itoh et al., [@B26]; Zhu et al., [@B63]). However, this PTZ kindling-induced increase of nNOS signaling is abolished in nNOS knockout mice. nNOS knockout mice are viable and show normal behavior, although they exhibit enlarged stomachs and dysfunction in gastrointestinal motility (Huang, [@B23]). When nNOS knockout mice are subjected to focal ischemia, they have smaller infarcts, suggesting a protective role of nNOS against neurotoxicity (Huang, [@B23]). Interestingly, genetic deletion of nNOS did not affect the PTZ kindling progress, which is consistent with a previous report (Itoh and Watanabe, [@B25]). To define a primary role of nNOS on PTZ kindling-induced psychiatric comorbidities, we measured the cognition function and depressive-like behavior in nNOS deficient mice and their wildtype littermates under normal and PTZ kindling conditions. We demonstrated that PTZ kindling-induced cognition deficit and depressive-like behavior is dependent on nNOS activity. These results suggest that nNOS plays a crucial role in PTZ kindling-induced psychiatric comorbidities.
Redox homeostasis is essential for maintain normal function of brain. Excessive production of ROS, a hallmark of redox homeostasis impairment in the brain, appears to be involved in the pathogenesis of epilepsy (Rowley et al., [@B43]; Williams et al., [@B59]). In agreement with previous studies, here we show that hippocampal ROS level are significantly increased in PTZ-kindled mice, however, depletion of nNOS suppressed PTZ kindling-induced ROS production, indicating PTZ kindling-induced ROS production is dependent on nNOS activity. Mounting evidence suggests that increased oxidative stress in the brain was usually accompanied with cognition deficit and depressive-like behavior. (de Morais et al., [@B13]; Pearson et al., [@B38]; Taiwe et al., [@B54]). Both MAPK and PI3K/AKT signaling pathways have been reported to respond to ROS stimulation in the central nervous system, thereby activating certain cellular events which contribute to pathological processes (Shah et al., [@B48]; Brobey et al., [@B5]). A recent study reported that ROS-mediated MAPK signaling pathway activation plays a pivotal role in cognition deficits in Alzheimer\'s disease (Arora et al., [@B1]). Our data show that both MAPK and PI3K/AKT signaling pathways have been activated in PTZ-kindled mice, and both signaling pathways activation are dependent on nNOS activity, suggesting nNOS may activate MAPK and PI3K/AKT signaling pathways through ROS production to trigger PTZ kindling-induced cognition deficit and depressive-like behavior.
In summary, here we have used a PTZ kindling epilepsy model, supported by a genetic nNOS deficient mice, to demonstrate nNOS as a critical signaling in PTZ kindling -induced comorbidities including cognitive impairment and depressive-like behavior. Our understanding of the role of nNOS signaling in PTZ kindling-induced cognition deficit and depressive-like behavior may provide insight into the molecular mechanism for psychiatric comorbidities in chronic epilepsy patients.
Author contributions {#s5}
====================
XZ, JC, and HY designed research; XZ, JD, BH, RH, AZ, ZX, and HC performed research; XZ analyzed data and wrote the paper.
Conflict of interest statement
------------------------------
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This work was supported by grants from the National Natural Science Foundation of China (81673413 to XZ), Natural Science Foundation of Jiangsu Province (BK20141335 to XZ), the Specialized Research Fund for the Doctoral Program of Higher Education (20130092120043 to XZ), the Fundamental Research Funds for the Central Universities and the Scientific Research Foundation of State Education Ministry for the Returned Overseas Chinese Scholars (No. 311, 2015 to XZ).
[^1]: Edited by: Nuno Sousa, Instituto de Pesquisa em Ciências da Vida e da Saúde (ICVS), Portugal
[^2]: Reviewed by: Salim Yalcin Inan, University of Konya-NE, Turkey; Millie Rincón Cortés, University of Pittsburgh, United States
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-materials-09-00149}
===============
In recent years, controlled drug delivery systems for modern drug therapy have been attracting increasing attention because they exhibit low toxicity, a wide therapeutic window, and ideal drug efficacy as compared to conventional drug delivery systems \[[@B1-materials-09-00149],[@B2-materials-09-00149]\]. The multifunctional nanocomposites combine with magnetic and luminescent properties in one entity, and they have attracted great attention in recent years owing to their potential application in the biotechnology and nanomedicine fields including magnetic resonance imaging (MRI), cell separation, drug delivery agents, cell separation, labeling, and optical probes \[[@B3-materials-09-00149],[@B4-materials-09-00149],[@B5-materials-09-00149]\]. In the choice of luminescent nanomaterials for labeling, targeting and imaging, lanthanide-doped nanomaterials possess many of advantages such as high fluorescence quantum yields, low toxicity, long lifetimes, and high stability in comparison to quantum dots and organic dyes \[[@B5-materials-09-00149],[@B6-materials-09-00149],[@B7-materials-09-00149],[@B8-materials-09-00149]\].
So far, there have been some reports of constructing multifunctional nanomaterials that were made up of Fe~3~O~4~ and lanthanide-doped nanomaterials. In these reports \[[@B8-materials-09-00149],[@B9-materials-09-00149],[@B10-materials-09-00149],[@B11-materials-09-00149],[@B12-materials-09-00149]\], if the lanthanide-doped nanomaterials are chosen as cores, their luminescent intensity may be suppressed to some extent due to the coating of the outer layers. Meanwhile, if the lanthanide-doped nanomaterials are in direct contact with Fe~3~O~4~, their luminescence may be decreased as the direct contact can cause fluorescence-quenching \[[@B13-materials-09-00149],[@B14-materials-09-00149],[@B15-materials-09-00149]\]. Therefore, a SiO~2~ mid-layer between Fe~3~O~4~ and lanthanide-doped nanomaterials is needed.
To the best of our knowledge, there are no previous reports on the combination of magnetic properties with gadolinium vanadate nanophosphors. The previous investigation results indicated that nanosized GdVO~4~:Ln^3+^ phosphors have a significant application in a high definition flat display panels and potential applications in biology \[[@B16-materials-09-00149],[@B17-materials-09-00149],[@B18-materials-09-00149],[@B19-materials-09-00149]\]. Compared with Ln^3+^-activated YVO~4~, GdVO~4~:Ln^3+^ exhibits highly efficient emitting phosphors, in which the energy transfers from the GdVO~4~ host to the incorporated Ln^3+^ ions through V^5+^--O^2−^ charge transfer (CT), yielding an efficient luminescence of Ln^3+^ activators.
Herein, we develop, for the first time, a novel and simple route to prepare Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ core-shell microspheres with excellent magnetic and luminescence properties. The good aqueous colloidal stability, low toxicity and excellent self-heating efficacy make these novel magnetic, luminescent nanomaterials suitable for the hyperthermia treatment of cancer, and the luminescent entity helps us to identify the location of magnetic nanoparticles during *in vitro* cellular imaging \[[@B20-materials-09-00149],[@B21-materials-09-00149],[@B22-materials-09-00149],[@B23-materials-09-00149],[@B24-materials-09-00149]\].
2. Results and Discussion {#sec2-materials-09-00149}
=========================
[Figure 1](#materials-09-00149-f001){ref-type="fig"} depicts the X-ray diffraction (XRD) patterns of as-synthesized Fe~3~O~4~, Fe~3~O~4~\@SiO~2~ and Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ nanoparticles. From [Figure 1](#materials-09-00149-f001){ref-type="fig"}, we can find that there are characteristic diffraction peaks of Fe~3~O~4~, with a face-centered-cubic structure in all curves according to JCPDS card No. 65-3107. Besides the corresponding peaks of Fe~3~O~4~, SiO~2~ (JCPDS card No. 29-0085) and GdVO~4~ (JCPDS card No.86-0996) can be detected in [Figure 1](#materials-09-00149-f001){ref-type="fig"}a--c, respectively. No peaks corresponding to impurities are detected, showing the adequate purity of the Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ composites.
The morphology and size details of the composites were characterized by SEM (scanning electronic microscope) and TEM (transmission electron microscopy) images. SEM investigations, as displayed in [Figure 2](#materials-09-00149-f002){ref-type="fig"}a, reveal that the magnetic cores of Fe~3~O~4~ particles are of a rough appearance and have an average size of 290 (±20) nm. Once coated with one layer of silica, the composite microspheres are slightly larger in diameter and have a relatively smooth surface, with their size increased up to 320 (±30) nm, as shown in [Figure 2](#materials-09-00149-f002){ref-type="fig"}b. The average size of the core-shell nanocomposites finally increased up to 360 (±25) nm, as illustrated in [Figure 2](#materials-09-00149-f002){ref-type="fig"}c. The representative TEM images in [Figure 2](#materials-09-00149-f002){ref-type="fig"}e,f indicate that the nanocomposites exhibit a core-shell structure.
To estimate the magnetic sensitivity, the room temperature magnetization hysteresis loops of the as-prepared cores and core-shell nanocomposites were collected and displayed in [Figure 3](#materials-09-00149-f003){ref-type="fig"}. The magnetic hysteresis loops in [Figure 3](#materials-09-00149-f003){ref-type="fig"} indicate that they have saturation magnetizations of 83.9 emu/g (Fe~3~O~4~), 27.8 emu/g (Fe~3~O~4~\@SiO~2~) and 20.4 emu/g (Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^) as well as negligible coercivity at room temperature, implying characteristics of their strong magnetism. The reduction of saturation magnetization could be attributed to the nonmagnetic shells (SiO~2~ and GdVO~4~:Dy^3+^). Our study revealed that, though the magnetism of the core-shell nanocomposites is less than that of the bare magnetic cores, it still possesses enough magnetic response for biomedical applications such as MRI, which is effectively magnetic separation.
The photoluminescence spectra of Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ are shown in [Figure 4](#materials-09-00149-f004){ref-type="fig"}. In the excitation spectra ([Figure 4](#materials-09-00149-f004){ref-type="fig"}A), the excitation band at 300--350 nm monitored with a 571 nm emission of ^4^F~9/2~--^6^H1~3/2~ electronic transition of Dy^3+^ can be attributed to a charge transfer through the V--O bond overlay of the Dy--O charge transfer band. The emission spectra of GdVO~4~:Dy^3+^ are shown in [Figure 4](#materials-09-00149-f004){ref-type="fig"}B. The main emission peaks at 481 nm and 571 nm are results of the ^4^F~9/2~--^6^H~15/2~ transition and ^4^F~9/2~--^6^H~13/2~ transition of Dy^3+^ ions. Moreover, [Figure 4](#materials-09-00149-f004){ref-type="fig"} shows the excitation spectra and emission spectra of Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ composites with different doped concentrations of Dy^3+^ ions. It is shown that the optimum doped concentration of Dy^3+^ ions in the Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ composites is 1 mol %.
To investigate the porous structure of the Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ nanocomposites, the N~2~ adsorption-desorption isotherms were investigated and are shown in [Figure 5](#materials-09-00149-f005){ref-type="fig"}. This isotherm profile can be categorized as type IV, with a small hysteresis loop observed at a relative pressure of 0.05--1.0, indicating the mesoporous features. The inset in [Figure 5](#materials-09-00149-f005){ref-type="fig"} is the pore size distribution. As calculated by the Brunauer-Emmett-Teller (BET) method, Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ nanocomposites' core-shell structure gives rise to a BET area of 30.21 m^2^·g^−1^, with a relatively high pore volume of 0.212 cm^3^·g^−1^, and the average pore diameter is 17.46 nm. The BET indicated the potential of such nanostructures for drug delivery applications.
To evaluate the cytotoxicity of the Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ nanoparticles, *in vitro* cytotoxicity tests against HeLa cells were carried out. From the 3-\[4,5-dimethylthiazol-2-y1\]-2,5-diphenyltetrazolium bromide (MTT) viability histogram, shown in [Figure 6](#materials-09-00149-f006){ref-type="fig"}, we can find that the Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ causes insignificant damage to the HeLa cells when the sample concentration increases to 200 μg·mL^−1^ for 24 h, and the cell viability remains at 90.38% even at the highest concentration, which indicates that Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ nanoparticles are biocompatible.
3. Materials and Methods {#sec3-materials-09-00149}
========================
3.1. Materials {#sec3dot1-materials-09-00149}
--------------
All reagents are of analytical reagent grade and used without further purification. Gd~2~O~3~ (99.9%) and Eu~2~O~3~ (99.9%) were purchased from Jinan Camolai Trading Company, Ferrous chloride hexahydrate (FeCl~3~·6H~2~O) (99%), tetraethyl orthosilicate (TEOS, 99.0%), sodium acetate (NaAc), Citrate acid monohydrate were purchased from Beijing Chemicals Corporation. Nitric acid, ethanol, ethylene glycol (EG), and ammonia aqueous (25%) were purchased from Tianjin Chemicals Corporation. Deionized water obtained from the Milli-Q system (Millipore, Bedford, MA, USA) was used in all experiments. The magnetic Fe~3~O~4~ nanoparticles were prepared using a modified solvothermal reaction.
3.2. Synthesis of Fe~3~O~4~ {#sec3dot2-materials-09-00149}
---------------------------
The magnetic Fe~3~O~4~ nanoparticles were prepared according to a previously reported synthetic process \[[@B19-materials-09-00149]\]. Typically, FeCl~3~·6H~2~O (1.3495 g) and NaAc (7.1926 g) were dissolved in EG solution (40 mL). Then PEG-10000 (1.0015 g) was added with vigorous stirring and the mixture was stirred for 30 min to form a homogeneous russet solution. The obtained solution was transferred to a Teflon-lined stainless-steel autoclave (50 mL capacity) and heated at 200 °C for 10 h. Subsequent cooling to room temperature yielded black magnetite particles, which were washed with ethanol and deionized water three times, respectively, and dried at 60 °C for 12 h.
3.3. Synthesis of Fe~3~O~4~\@SiO~2~ {#sec3dot3-materials-09-00149}
-----------------------------------
Fe~3~O~4~\@SiO~2~ nanoparticles were prepared according to the modified by the Stöber method. In brief, 1.0 g of Fe~3~O~4~ nanoparticles were homogeneously dispersed in a mixture of 160 mL of ethanol, 40 mL of deionized water, and 3.0 mL of 28 wt % concentrated ammonia aqueous solution, followed by the addition of 3.0 mL of tetraethyl orthosilicate (TEOS). After vigorous stirring at 40 °C for 6 h, the obtained Fe~3~O~4~\@SiO~2~ microspheres were separated with a magnet and washed repeatedly with ethanol and deionized water to remove nonmagnetic by products.
3.4. Synthesis of Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ Nanoparticles {#sec3dot4-materials-09-00149}
-----------------------------------------------------------------
Functionalization of GdVO~4~:Dy^3+^ on the template Fe~3~O~4~\@SiO~2~ was achieved according to the reported process with a doping concentration of Dy^3+^ of 0.5--4 mol % to Dy^3+^ in GdVO~4~:Dy^3+^. The typical procedure for synthesis is described as follows: stoichiometric amounts of Gd~2~O~3~, Dy~2~O~3~ and citric acid were dissolved in dilute nitric acid with heating followed by the addition of NH~4~VO~3~ in distilled water. Then PEG-10000 was added with a concentration of 0.05 g·mL^−1^. After stirring for 0.5 h, a homogenous sol was formed. Then the desired amount of Fe~3~O~4~\@SiO~2~ nanoparticles was added into the gel, after further stirring for another 3 h, the resulting material was dried at 120 °C for 12 h to obtain the precursors. Then the precursors were calcined at 700 °C for another 4 h. The obtained nanoparticles were denoted as Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ ([Scheme 1](#materials-09-00149-f007){ref-type="scheme"}).
3.5. Cytotoxicity Study of Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ Nanoparticle {#sec3dot5-materials-09-00149}
-------------------------------------------------------------------------
Cell viabilities of Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ nanoparticles at different concentrations were tested by MTT assay on HeLa (human cervical cancer cells). In the experiment, the corresponding untreated cells were used as control. First, the HeLa cells were pre-incubated in a 96-well plate (about 3000 cells per well) for 24 h. Second, 2 mg of the Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^ nanoparticles were added into 10 mL of 0.01 M phosphate buffered saline (PBS, pH = 7.4) to form a stable orange solution. Third, the above solution at concentrations of 6.25, 12.5, 25, 50, 100 or 200 μg·mL^−1^ was added to the cells. Six parallel-group experiments were simultaneously conducted for each concentration. After 24 h, the viability of HeLa cells was examined by a MTT assay.
3.6. Characterization {#sec3dot6-materials-09-00149}
---------------------
The purities of all the nanoparticles were checked by X-ray diffraction (XRD) measurements at room temperature using Cu Kα radiation (Kα = 1.54059 Å). The morphology and microscope structure of all the nanocomposites were characterized by a scanning electronic microscope (SEM, NoVa™ Nano SEM 430, FEI Co., Ltd., Hillsboro, OR, USA) and transmission electron microscopy (TEM, JEOL JEM-2010F, JEOL Co., Ltd., Tokyo, Japan). The room temperature magnetic hysteresis (M-H) loops were measured using a superconducting quantum interference device vibrating sample magnetometry (SQUID-VSM, Quantum Design Co., Ltd., San Diego, CA, USA). Luminescence spectra were recorded on a FluoroMax-4 spectrophotometer (HORIBA Jobin Yvon Co., Ltd., Paris, France). The specific surface area was determined by the Brunauer-Emmett-Teller (BET) method. The HeLa cells were assayed for viability by using a microplate reader (Bio-Rad 680, Bio-Rad Co., Ltd., Hercules, CA, USA).
4. Conclusions {#sec4-materials-09-00149}
==============
In summary, we report a novel magnetic/luminescence multifunctional nanocomposite, Fe~3~O~4~\@SiO~2~\@GdVO~4~:Dy^3+^, with a core-shell structure from a combination of hydrothermal reaction and the sol-gel process. The as-prepared nanocomposites, combining the merits of the good magnetic response of the assembled Fe~3~O~4~\@SiO~2~ microspheres and the fluorescence property of GdVO~4~:Dy^3+^, displayed high surface area and biocompatibility. Therefore, our study may provide new insight and useful information for the design of diverse, functional nanocomposites as drug carriers.
This work is supported by the National Natural Science Foundation of China (Nos. 21401112, 21301100).
Huitao Fan, and Bo Li were involved in designing the aim of this manuscript, performed the experiments and prepared the manuscript. Qiang Zhao performed the room temperature magnetic hysteresis (M-H) loops measurements and analyzed the data. Congcong Wang helped with several analyses. Each contributor was essential to the production of this work.
The authors declare no conflict of interest.
Figures and Scheme
==================
{#materials-09-00149-f001}
{#materials-09-00149-f002}
{#materials-09-00149-f003}
{#materials-09-00149-f004}
{#materials-09-00149-f005}
{#materials-09-00149-f006}
{#materials-09-00149-f007}
| {
"pile_set_name": "PubMed Central"
} |
As an example of dietary fibres, short-chain fructo-oligosaccharides (scFOS) produced from sucrose by a controlled reaction with the enzyme fructo-furanosiadase are not extensively digested nor absorbed in the small intestine, but as prebiotic fibres they will be selectively fermented in the large intestine, providing potential health benefits for the host^(^[@ref1]^)^. Because they are not digested and also have a sweet taste (30 % *v.* sucrose), scFOS can be used to reduce food sugar content or energy while maintaining the same rheological and sensory attributes^(^[@ref2]^)^. In Europe and elsewhere (e.g. Canada), all fibres are considered to provide 2 kcal (8·4 kJ) of energy per g instead of 4 kcal (17·9 kJ) for digestible carbohydrates^(^[@ref3]^,^[@ref4]^)^. Dietary fibres can also be used to lower the postprandial blood glucose response^(^[@ref5]^)^, which is a beneficial physiological effect for subjects who are intolerant to glucose^(^[@ref6]^,^[@ref7]^)^. However, this effect depends on the type of fibres used (soluble, insoluble, viscous, non-viscous, etc.) and should be verified for each of them^(^[@ref5]^)^.
The effects of scFOS on the blood glucose response in humans were studied in a pilot study involving two subjects who either consumed a single dose of pure dextrose or of pure scFOS. The consumption of scFOS did not induce an increase in blood glucose contrary to the consumption of dextrose^(^[@ref8]^)^. More recently the effects of different mixtures of maltitol and scFOS in dairy desserts on the postprandial blood glucose response were studied as a secondary objective in a randomised, double-blind reference-controlled clinical study with eighteen healthy subjects^(^[@ref9]^)^. Maltitol and scFOS were used in combination to totally replace sugars and scFOS was also tested for partial sugar replacement (−30 %). The control food contained 35 g of dextrose. As the study was not powered to study this outcome measure, there was only a trend towards reduced glucose AUC~0--120 min~ when a dairy dessert was reduced in sugars through the addition of scFOS (11 g).
The primary objective of the two present studies was to evaluate the postprandial glycaemic response to two different types of foods, in healthy adults, standard with reduced (−30 % w/w) sugars (e.g. dextrose) content through replacement with scFOS in comparison with a control food containing a standard amount of sugars. The secondary objective was to assess the effect on the insulinaemic response. Dairy dessert and pound cake (traditional sponge cake made using 1 lb (0·4536 kg) each of the four ingredients: flour, butter, eggs, sugars) were chosen as examples of popular foods eaten by both children and adults and typically providing significant amounts of sugar that is difficult to replace without compromising on taste^(^[@ref10]^)^.
Methods {#sec1}
=======
Subjects {#sec1-1}
--------
Healthy subjects were recruited for these two studies according to the same inclusion criteria: age (18--50 years), BMI (18·0 and 25·0 kg/m^2^), used to eating breakfast, non-smoker, not using medication which could affect nutrient absorption, lipid or carbohydrate metabolism. The subjects were screened according to fasting blood glucose level ≤ 1·1 g/l, fasting blood cholesterol ≤ 6·35 mmol/l, TAG ≤ 1·70 mmol/l, insulin ≤ 20 mU/l (139 pmol/l), HbA1c ≤ 7 %, and no clinically significant abnormality concerning complete blood count, and liver enzymes.
All the subjects provided written informed consent to participate after study procedures had been explained to them. The studies were approved by the ethics committee (CPP Nord-Ouest I, Rouen, France) and were performed in accordance with the guidelines of the International Conference on Harmonisation of Good Clinical Practice and the principles laid down in the current version of the Helsinki Declaration.
The two studies were powered (α 0·05; 0·80 power) on the basis of previously reported postprandial glycaemic response induced by the consumption of FOS and sugar products^(^[@ref9]^)^.
After adjustment for drop-out, twenty-five and thirty-five subjects, respectively, were required in the dairy dessert and pound cake studies.
Experimental design {#sec1-2}
-------------------
These two acute crossover, double-blind placebo-controlled, randomised studies were both performed in a single clinical centre (Institut Pasteur Lille, Lille, France).
For each study, the two different foods (standard = control or sugar-reduced) were administered orally, according to the randomisation list, to the subjects during two successive experimental sessions with at least one washout week between them. During each experimental visit, the subjects arrived at the clinical centre in the morning after a 10-h fast and were subjected to a clinical examination and a medical interview. The last dinner before fasting was a calibrated meal given to all the subjects, providing 677 kcal (2833 kJ) including 12 % of energy from protein, 54 % from carbohydrate and 34 % from fat.
Products being studied {#sec1-3}
----------------------
The studied scFOS were FOS from sucrose (Actilight^®^ 950P; Beghin Meiji), comprising about 37 % 1-kestose (GF2), 53 % nystose (GF3) and 10 % 1F-β-fructofuranosyl nystose (GF4). For the first study, the products were administered orally in a dairy dessert containing, in decreasing order of weight: dextrose, cocoa powder, maltodextrins, modified starch, milk protein, chocolate flavour, carrageenan, and sucralose for sweetness adjustment ([Table 1](#tab01){ref-type="table"}). The ingredients were mixed together and heated in two phases: 65°C for 10 min and 85°C for 15 min. After being packaged in cans, they were sterilised at 121°C for 16 min. For the second study, the products were administered in a pound cake containing, in decreasing order of weight: wheat flour, eggs, sucrose, margarine, water, glucose syrup, wheat starch, baking powder (disodium diphosphate, sodium carbonate) and emulsifying agent ([Table 1](#tab01){ref-type="table"}). After mixing all the ingredients at 30°C, the dough was cooked at 170°C for 30 min. Table 1.Nutritional composition of the dairy dessert and pound cake, by portionDairy dessert (210 g)Pound cake (100 g)ControlReduced in sugarsControlReduced in sugarsEnergy content kcal218·9193·0461443 kJ915·9807·519291854Carbohydrates (g)48·132·158·349·2Sugars[\*](#tfn1_2){ref-type="table-fn"}36·025·3[†](#tfn1_3){ref-type="table-fn"}28·119·4[‡](#tfn1_4){ref-type="table-fn"}Lipids (g)2·23·022·122·1Proteins (g)6·18·36·46·4Fibres (g)2·714·81·010·1scFOS (g)0·011·20·09·1scFOS recovery in final food (%)--100--100[^1][^2][^3][^4]
Both products were manufactured externally and labelled in exactly the same way by the person in charge of production. The investigators and subjects could not see any difference in term of food presentation and taste, as sweetness, for example, was adjusted to be equivalent in both products.
Analysis of short-chain fructo-oligosaccharide content in foods {#sec1-4}
---------------------------------------------------------------
The food samples were dissolved in pure water at 40°C, homogenised and centrifuged. The supernatant fraction was filtered (0·2 µm) and diluted before injection into the chromatograph. They were then analysed by anion-exchange chromatography (Dionex). The samples were hydrolysed by α-glucosidase and invertase and analysed again. The method has been described by Ouarne *et al*.^(^[@ref11]^)^.
Evaluation of glycaemic and insulinaemic responses {#sec1-5}
--------------------------------------------------
On the day of the test, a nurse placed a catheter on the subject\'s arm, started the kinetic test for 120 min, and then took the catheter off. The subjects were instructed to eat the dairy dessert or pound cake within 5 to 10 min, under fasting conditions, with 150 ml of water at the clinical site. The kinetic test consisted of sampling venous blood at T-5 and T-1 min before the subject ate the meal, and then at T15, T30, T45, T60, T90 and T120 min after meal intake.
For the kinetic test, the blood samples were collected in sodium fluoride and potassium oxalate for glucose determination, and serum-separating tubes for insulin. The level of blood glucose was assessed by an enzymic UV test (hexokinase method) (AU480; Beckman Coulter) and commercially available glucose reagents (OSR6121; Beckman Coulter). The blood insulin level was assessed by an immunoradiometric assay (Cisbio Bioassays).
Each subject\'s compliance was checked by the study coordinator during the sessions (meal intake according to protocol and respect of the sampling time).
Statistical analyses {#sec1-6}
--------------------
The results are presented as means and standard deviations. Variables were assessed for normality of distribution using the Shapiro--Wilk test. If the normality assumption was rejected, a log transformation of the data was performed. The incremental AUC between 0 and 120 min (AUC~0--120 min~) for blood glucose and insulin concentrations was computed following the FAO recommendation^(^[@ref12]^)^.
AUC~0--120 min~ and maximum concentration (C~max~) for glucose and insulin were analysed using a mixed-model ANCOVA with 'product' as fixed effect, 'subject' as random effect and glucose or insulin baseline value as covariate.
A statistical analysis was conducted on the modified intention-to-treat (mITT) population and on the per protocol (PP) population using SAS^®^ software version 9.1.3 (SAS Institute Inc.). For all the statistical tests, a 0·05 significance level was used to claim a statistically significant effect.
Results {#sec2}
=======
Short-chain fructo-oligosaccharide content in foods {#sec2-1}
---------------------------------------------------
Analysis of the foods after cooking and heat treatment showed that 100 % of scFOS was recovered ([Table 1](#tab01){ref-type="table"}) and the ratio between GF2, GF3 and GF4 was identical to that of the ingredients (data not shown).
Study populations {#sec2-2}
-----------------
Two different study populations were recruited to participate in the studies: one to test the dairy dessert and one to test the pound cake.
Study 1: dairy dessert {#sec2-3}
----------------------
The mITT and PP populations were equivalent in this study ([Fig. 1](#fig01){ref-type="fig"}). This population was 32·3 ([sd]{.smallcaps} 8·7) years old on average, with an average BMI of 22·3 ([sd]{.smallcaps} 1·9) kg/m^2^. Of the subjects, 28 % were male and 72 % female. The average fasting glycaemia of the ITT population was 4·88 ([sd]{.smallcaps} 0·50) mmol/l and insulinaemia was 3·68 ([sd]{.smallcaps} 1·52) mU/l (25·56 ([sd]{.smallcaps} 10·56) pmol/l), with no significant difference between the two visits. Fig. 1.Distribution of subjects among intention-to-treat and per protocol populations.
Study 2: pound cake {#sec2-4}
-------------------
The mITT population represented thirty-three out of the thirty-five randomised subjects and the PP population (results not shown) was composed of thirty-one randomised subjects who completed the study without any major deviation. Two subjects were excluded from the mITT because they left the study prematurely for no specific reason before product consumption, and two subjects were excluded from the PP population because several blood samples used to evaluate the blood glucose response could not be taken during the 2nd visit. The randomised population was 31·9 ([sd]{.smallcaps} 8·1) years old on average with an average BMI of 21·9 ([sd]{.smallcaps} 1·9) kg/m^2^. Of the subjects, 29 % were male and 71 % female. The average fasting glycaemia of the ITT population was 4·86 ([sd]{.smallcaps} 0·37) mmol/l and insulinaemia was 4·27 ([sd]{.smallcaps} 1·71) mU/l (29·66 ([sd]{.smallcaps} 11·88) pmol/l), with no significant difference between the two visits.
Glycaemic and insulinaemic responses {#sec2-5}
------------------------------------
Consumption of the dairy dessert, formulated with scFOS replacing part of dextrose (−30 %, w/w), induced a lower postprandial blood glucose response compared with dextrose in the mITT population whereas the C~max~ was not altered ([Table 2](#tab02){ref-type="table"}, [Fig. 2](#fig02){ref-type="fig"}). In parallel, the insulin response as illustrated by the AUC~0--120 min~ was also lower after consumption of the sugar-reduced dairy dessert than with the standard one without impact on the C~max~ ([Table 2](#tab02){ref-type="table"}, [Fig. 2](#fig02){ref-type="fig"}). Fig. 2.Postprandial (a) plasma glycaemic and (b) plasma insulinaemic responses over 120 min after taking the dairy dessert containing 35 g dextrose (control; --●--) or 24 g dextrose and 11 g short-chain fructo-oligosaccharides (scFOS; --□--) in the modified intention-to-treat population (*n* 24). Data are means, with standard errors represented by vertical bars. Glucose and insulin AUC were significantly lower following scFOS-containing products than following control (*P* = 0·020 and *P* = 0·003, respectively) (mixed-model ANCOVA). To convert insulin in mU/l to pmol/l, multiply by 6·945. Table 2.AUC~0--120 min~ and maximum concentration (C~max~) of plasma glucose and insulin for 2 h after consumption of dairy dessert or pound cake with standard or reduced sugar content in the modified intention-to-treat population(Mean values and standard deviations)Dairy dessert (*n* 24)Pound cake (*n* 33)ControlReduced in sugarsControlReduced in sugarsMean[sd]{.smallcaps}Mean[sd]{.smallcaps}Dessert effect: *P*Mean[sd]{.smallcaps}Mean[sd]{.smallcaps}Cake effect: *P*Glucose AUC (mmol × min/l)92·890·669·356·00·02069·069·862·055·80·340 C~max~ (mmol/l)7·31·46·91·20·1846·30·16·10·10·272Insulin AUC (mU × min/l)[\*](#tfn2_1){ref-type="table-fn"}286711902232·411210·0031885·510001695·5847·00·082 C~max~ (mU/l)[\*](#tfn2_1){ref-type="table-fn"}64·825·457·525·00·12235·016·931·615·00·245[^5]
Consumption of the pound cake formulated with scFOS replacing part of the sugars (sucrose and glucose syrup; −31 %, w/w) did not modify the postprandial blood glucose response or the insulin response compared with the standard recipe in the mITT population to any significant extent ([Table 2](#tab02){ref-type="table"}, [Fig. 3](#fig03){ref-type="fig"}). Similar results were obtained for the PP population which differed only by two subjects. Fig. 3.Postprandial (a) plasma glycaemic and (b) plasma insulinaemic responses over 120 min after eating the pound cake containing 28 g sugars (control; --●--) or 19 g sugars and 9 g short-chain fructo-oligosaccharides (--□--) in the modified intention-to-treat population (*n* 33). Data are means, with standard errors represented by vertical bars. To convert insulin in mU/l to pmol/l, multiply by 6·945.
Adverse events {#sec2-6}
--------------
No serious adverse events occurred during the two studies. Six minor adverse events were reported, not linked to the study products but possibly linked with the research procedures (bruising at the blood sampling area). No statistically significant association was made between the ingredients and the imputation of adverse events.
Discussion {#sec3}
==========
While a reduced postprandial blood glucose response is acknowledged as a beneficial effect, especially for subjects intolerant to glucose, acceptance of sugar-reduced foods is generally less than it is for standard foods^(^[@ref13]^)^ because sugars provide sweetness but also other rheological functions (structure, volume, flavour and aroma, etc.)^(^[@ref10]^,^[@ref14]^)^.
Dietary fibres such as scFOS are sometimes used to partially replace sugars in foods and beverages^(^[@ref2]^)^ because they have a sweet taste and also have a lower energy value than digestible carbohydrates^(^[@ref15]^)^. They have already been successfully used in different types of food reduced in sugars (up to 30 %) in comparison with a control product, without compromise on taste as evaluated by a sensory panel^(^[@ref2]^,^[@ref16]^)^. Furthermore, using scFOS to partially replace sugars also helps enrich foods with fibres, the consumption of which, in Europe, is generally lower than what is recommended^(^[@ref17]^)^.
The present study aimed to evaluate, in healthy adults, the postprandial glycaemic and insulinaemic responses of two types of food matrices with or without scFOS in replacement of sugars. A dairy dessert containing 35 g dextrose was chosen as a first example; the dose of dextrose was partially replaced by 11 g of scFOS in the sugar-reduced recipe. This level of reduction was chosen according to European Union regulation for the claim 'reduced in sugars' which requires at least a 30 % (w/w) reduction in comparison with a reference product. The same principle was applied to formulate the pound cake that initially contained 28 g of sugars (sucrose and glucose syrup) for the consumed portion. The test pound cake provided 9 g of scFOS.
All scFOS added to the recipe were recovered after cooking and heat treatment, confirming the stability of scFOS in food matrices. Indeed, it is generally acknowledged that scFOS are stable during heat treatment but may be sensitive to interaction with an acidic pH and high temperature^(^[@ref18]^,^[@ref19]^)^.
The postprandial glycaemic response of the dairy dessert containing the scFOS was reduced in comparison with the glycaemic response of the control dairy dessert, as illustrated by a lower AUC~0--120 min~. This lower glycaemic response was not due to hyperinsulinaemia induced by scFOS because the postprandial insulin AUC~0--120 min~ was also reduced and the insulin spike was not altered compared with the control dessert. This might be explained by the fact that in humans, scFOS are mostly neither digested nor absorbed in the small intestine but completely fermented in the large intestine^(^[@ref15]^,^[@ref20]^)^. Indeed, while pancreatic enzymes cannot hydrolyse scFOS^(^[@ref21]^,^[@ref22]^)^, bifidobacteria and some other bacterial groups possess the β-fructosidase enzyme necessary to hydrolyse the β-(2,1) glycosidic linkages in scFOS as demonstrated *in vitro*^(^[@ref23]^,^[@ref24]^)^. Their fermentation leads to the production of SCFA such as acetate, propionate, butyrate and also CO~2~. Consequently, it was previously shown in two subjects that contrary to dextrose, scFOS consumed as such do not increase postprandial blood glucose or insulin^(^[@ref8]^)^. The present study also confirms a more recent observation that when replacing dextrose by up to30 % (w/w) with scFOS could tend to reduce the postprandial glucose AUC~0--120 min~ of a dairy dessert while not increasing the insulinaemic response^(^[@ref9]^)^.
While being numerically lower with the scFOS, the postprandial glycaemic and insulinaemic responses following consumption of the pound cake reduced in sugars with scFOS did not differ from the responses induced by the control cakes. These different results highlight the fact that factors other than the relative quantities of sugars interact with the postprandial glucose response to foods. This cannot be linked to hydrolysis of scFOS during heating, because they are stable and were completely recovered in the final form of both foods. One hypothesis may be that the quantity of glycaemic carbohydrates compared might be too low. Traditionally for evaluating the glycaemic index, the test load is made with 50 g of available carbohydrates and a recommendation is made to use a test dose not less than 25 g of available carbohydrates^(^[@ref25]^)^. In the present study the quantities of food to be eaten were defined to be as closely as possible representative of reasonable consumption of the considered food. The portion of pound cake contained 28 and 19 g of sugars, respectively, for the control and the sugar-reduced versions. These quantities may not be large enough to observe a significant difference in glycaemic response when these sugars are consumed within a complex food matrix and not as single ingredients.
The two food matrices that were tested also differed in consistency and fat content and these parameters could influence the glycaemic response of food. Studies indicate than a more compact and viscous meal (like the pound cakes in our study) tends to delay gastric emptying^(^[@ref26]^,^[@ref27]^)^. Since the pound cake can be considered as a solid meal while the dairy dessert is more a semi-solid meal, this could explain why blood glucose concentration did not behave the same way with the two matrices. While the dairy dessert was also very low in fat (less than 1·5 %), the pound cake contained around 22 % of fat. Fat is well known to slow down gastric emptying and thus could indirectly make an impact on the arrival of glucose in the bloodstream^(^[@ref28]^--^[@ref31]^)^. Interestingly, the time to the glucose spike in our study was delayed by about 10 min for the pound cake compared with the dairy dessert.
This study highlights the fact that, when used in place of sugars (w/w) for partial replacement, scFOS may help reduce the postprandial glycaemic and insulinaemic responses to foods as illustrated by the dairy dessert here. This is certainly explained by the fact that scFOS are non-digestible carbohydrates and that they are used to replace sugars which are fully available, without making an impact on the palatability of the tested foods^(^[@ref2]^,^[@ref32]^)^. Longer-term studies or second meal effect studies would be of interest in order to see whether scFOS could have other effects on the regulation of blood glycaemia and insulinaemia via their modulation of the activities of gut microbiota as has already been observed in an animal model harbouring human microbiota^(^[@ref33]^)^.
The authors would like to thank Marjorie Lejeune, Laurence Baron, Anne Ehret, Xavier Coustenoble, Coralie Berthier and Nathalie Frisicale for their support to this study. We also thank Tereos Syral and Beghin Meiji for the financial support of this research study and for kind contribution of the foods for the study.
A. W. and F. R. are employees of Tereos. The remaining authors declare no conflict of interest.
[^1]: scFOS, short-chain fructo-oligosaccharides.
[^2]: 'Sugars' means all monosaccharides and disaccharides present in food but excludes polyols.
[^3]: Sugars reduced by 30 %.
[^4]: Sugars reduced by 31 %.
[^5]: To convert insulin in mU/l to pmol/l, multiply by 6·945.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Chloroplast DNA (cpDNA) has been used extensively in plant phylogenetic studies as it is maternally inherited in most angiosperms, has a low mutation rate and provides variable and informative regions over broad timescales ([@ref-20]; [@ref-24]). Moreover, cpDNA provides abundant DNA polymorphisms at inter- and intraspecific levels, providing molecular phylogenies with high resolution at different taxonomic scales ([@ref-23]).
The emergence of next generation sequencing (NGS) technologies, allowing for complete organellar DNA sequencing, has promoted the use of plastome (i.e., complete cpDNA sequences) data as a major tool in phylogenomic and evolutionary analyses ([@ref-28]; [@ref-32]), as well as an extended DNA-barcode ([@ref-16]). As a result, massive amounts of genome-scale data have been generated at a relatively low cost, with ever faster turnaround times ([@ref-32]). The availability of this expanding data volume has allowed for the exploration of chloroplast organization and chloroplast features throughout the angiosperm tree at the molecular level ([@ref-27]) and the development of novel approaches for phylogenetic studies ([@ref-6]). To date, for angiosperms alone, there are over 4,500 sequenced plastomes available in NCBI GenBank through the INSDC database (as of June 2019) and this number has seen an exponential increase in recent years.
A consequence of the rapid development of NGS techniques is the emerging need to handle and curate increasingly large organellar genome datasets. For example, a recent study ([@ref-17]), used nearly 3,000 plastomes to reconstruct angiosperm phylogeny and evolution. Because mitochondrial genomes and plastomes are usually circular, but are represented as linear text sequences in bioinformatic pipelines and online databases, any base position can represent the start of the string sequence (see [Fig. 1](#fig-1){ref-type="fig"}). Plastome sequences are typically characterized by two inverted-repeated regions (hereafter called IRs) separated by a long (LSC) and a short (SSC) single copy region ([@ref-3]; [@ref-21]; [@ref-26]). Due to the replication mode of organellar DNA, each part can be found in both directions (i.e., in 5′--\>3′ orientation, or in the reverse complement orientation) in the cell ([@ref-21]). This organization implies that orthologous regions need to be assessed for each of these parts ([Fig. 1](#fig-1){ref-type="fig"}). By general consensus and some computational constraints, most of the assembly methods start the reconstructed sequence at the first base of the large single-copy region (LSC). However, depending on both the quality and quantity of the NGS sequences (i.e., the "raw reads"), as well as the complexity of the plastome to reconstruct, a sequence may start randomly along the circular sequence ([Fig. 1B](#fig-1){ref-type="fig"}), occasionally making the identification of homologous regions among sequences laborious and time-consuming, especially for large datasets (100--1,000 s of sequences).
{#fig-1}
Despite there being some consensus in the community on the starting position of plastome sequences, a quick assessment of sequences in GenBank shows extensive disparity in the organization and order of the different regions. To circumvent this, most of the plastomic studies ([@ref-5]) (including the above-mentioned 3,000 plastomes study ([@ref-17])) only consider the coding regions of the reconstructed organelles, to remove small inversions and changes in gene order. However, such rearrangements are less likely to happen at small taxonomic scales (e.g., at intra-generic level, or in species complexes). Moreover, such an approach results in the deletion of the non-coding regions, that often contain more phylogenetic signal ([@ref-25]), especially at lower taxonomic level or in recalcitrant taxa, in which genomic approaches are usually required to fully resolve relationships ([@ref-8]; [@ref-2]). In addition, using the entire plastome sequence allows for detection of evolutionary events, such as insertion--deletions or inversions and facilitate the identification of sequences from non-organelles origins (nuclear plastid DNA-NUPTs; nuclear mitochondrial DNA-NUMTs).
Accurate and complete molecular sequence data are essential in phylogenomic reconstruction. Therefore, accurate homology detection among sequences is a vital step. Consequently, improvement in the functional annotation of organisms will become an extremely important step to delineate evolutionary processes.
Usually, three distinct steps can be identified in phylogenomic studies: (1) The reconstruction of organelles in the different species; (2) the assessment of orthology and alignment of the plastome sequences (i.e., the dataset assembly); and (3) the actual phylogenetic reconstruction ([@ref-10]). Current software focuses on organelle reconstruction ([@ref-4]; [@ref-14]), multiple sequence alignment (MSA) algorithms ([@ref-1]) and phylogenetic approaches ([@ref-9]). Software currently available can aid in assembly and analyses of plastid and mitochondrial genomes. However, orthology assessment and curation of the assembled plastomes usually remains a manual task (([@ref-19]) for a combined assembly and orientation script). Indeed, there is currently no open-source software that reconstructs and assesses plastome orthology for large datasets (numbering thousands of plastomes) in a fast and accurate way from assembled draft sequences. While orthology assessment can be easily done for datasets based on individually extracted coding regions, the specific structure of the plastome (LSC, IRb, SSC, IRa) complicates the proper alignment of orthologous regions. Indeed, any differences in organization/orientation of these parts in the reconstructed draft plastomes, can result in improper assessment of orthologous positions and thus increase the risk of discarding potentially useful data at the end of alignments. This process is either done manually (representing 3--5 min per chloroplast for a trained bioinformatician, using commercially available software, for example, Geneious; [@ref-22]), or skipped by extracting the coding regions (thus discarding the most variable and useful regions of the plastome: see above). This represents an emergent bottleneck in dataset assembly and downstream analyses. Therefore, the development of flexible and user-friendly tools that remove these labor-intensive and time-consuming components from genomic workflows has become a latent priority for the plant genomics community.
We designed *ECuADOR* to facilitate both the rapid processing of organelle genomic data as well as providing output requirements for downstream analyses. *ECuADOR* is a rapid, platform-independent and user-friendly algorithm built in Perl, that automates detection and reorganization of sequence features in newly assembled plastomes. As *ECuADOR* uses only the reconstructed plastomes, it is also independent of the sequencing technology used to generate the data and can thus be used with assemblies derived from short reads (e.g., Illumina, San Diego, CA, USA), long reads (e.g., Nanopore; PacBio, Menlo Park, CA, USA) or those acquired using other sequencing technologies. Data are generated as moving singular reciprocally-compared fragments, tracking the number of nucleotide changes for a window of a user-defined length, slide along the sequence.
The algorithm is executed with default settings for the sliding window option (but with the option to adjust these manually) and adjustments can be made to input parameters according to the desired output format file (see *ECuADOR* command line below). These options can be customized via a command-line, making it user-friendly and easily accessible.
Materials and Methods
=====================
Analysis pipeline
-----------------
*ECuADOR* is written in Perl (tested with Perl 5.18) and uses the following modules: Bio::SeqIO, IO::String, Set::IntSpan, IO::File, Bio::AlignIO, Bio::Factory::EMBOSS, File::Temp qw/tmpnam/ and Cwd. Input for *ECuADOR* is a draft plastome sequence (GenBank or fasta), the length of the sliding window, format of input file (GenBank or fasta), and output format file (fasta or GFF3). *ECuADOR* reads and analyzes single or multiple plastomes stored in a designated folder, containing one sequence per file.
*ECuADOR* is based on a user-defined sliding window and dynamic suffix array approach. The algorithm partitions the sequence into fragment intervals, of sizes defined by the "*window size*" option. This window slides along the plastome in both 5′--\>3′ and 3′--\>5′ direction (as a reverse-complement sequence). Then, a positioning array index is generated from the similarity between the generated fragments. This new array index stores a sequence of the four main regions according to the exact location where each repeated extreme was found. In later stages, the extremes are used to recover the remaining IR length between both of them.
However, when the quality of the used sequencing reads is poor, mis-assemblies are likely to be introduced in the sequence during organellar reconstruction (e.g., gapped palindromes), for example, in one of the repeats. In such cases, the sliding-window approach could retrieve different sequence lengths for each repeat region. Because the main goal of the sliding-window approach is to capture the two identical-sized repeat regions (IRa and IRb), a non-equality in any repeat sequence length could prevent the recovery of its total length. This would result in an incorrect positioning for all the main regions (LSC--IRb--SSC--IRa) and prohibit the rapid assembly of correct alignments downstream. *ECuADOR* addresses this drawback by only using the extremes of each inverted repeat to estimate the repeat size, that is, once the position of each extreme fragment is known in both repeats, *ECuADOR* will take all the remaining positions between both previously positioned extremes to recover the entire repeated sequence. This offers a flexible (allowing small discrepancy between IRs-value defined as a user-customizable setting) yet conservative approach (the borders of IRs have to perfectly match each other).
*ECuADOR* allows the user to define the size of the window (in bps) used for the sliding-window step. The optimal length of the sliding window (−w parameter) depends on the quality of the reconstructed sequence. If the sequence quality is low, it is advised to use a smaller sliding window size to adjust for the higher likelihood of finding gapped palindromes throughout the inverted repeat. These would not normally be detected using a larger sliding window size, leading to a loss of information in the length of the repeats. By adjusting the sliding window size, the user can balance the sensitivity of the IR detection. A larger window size allows for reducing the influence of misassembled reads and thus false positives---at the cost of a lower resolution (− sensitivity, + specificity). A smaller window size provides increased resolution but may also increase the number of false positives if the data is noisy (+ sensitivity, − specificity). The quality of the input plastome sequence (e.g., as a consequence of a low quality base call during sequencing or mis-assemblies) is an important factor to take into account before setting a value for −w (i.e., the window size parameter).
Finally, *ECuADOR* is executed using the following command-line perl command: *ECuADOR*.pl (−h) −i, \<folder containing the plastomes\>, −w, \<sliding window length\> (1,000 bp option by default), −f, \<input file format (option not by default, GenBank or fasta)\>, −out \<output prefix\>, --ext, \<output format\> (either fasta or GFF3), --save_regions, \<save chloroplast regions (LSC, SSC, IRs separately, or in combination)\>, --orient \<TRUE\> (reorientation of each plastome regions, providing the user with the cpDNA regions ready to use for MSA), finally the option --noIRs \<3 or 4\> (to generate files including either only the concatenated regions LSC--IRb--SSC or the entire one LSC--IRb--SSC--IRa)
The program is divided into seven main steps ([Fig. 2](#fig-2){ref-type="fig"}): Argument check and analysis of the input file, setting format files and sliding window size (if no slide window size is provided, the default option will be set to 1,000 bp.*ECuADOR* generates a reverse complement sequence of the input fasta sequence (i.e., reconstructed draft plastome).A suffix array index (permutation of index numbers giving the starting positions of suffixes of a string in alphabetical order) is built, comparing the suffix of fragment sequences with each and every suffix of the reverse complement fragments (both generated through slide windows). This new array index will be used to detect the exact location of the corresponding extremes to each IR. In the usual case (gap-free or identical palindromes), a total of two arrangements of only one sequential element will be located in the suffix arrays, thus covering the exact position of each repeat. In unusual cases (gapped palindromes) two arrangements of multi non-sequential elements will be located in the suffix arrays. In such cases these extra fragments will prevent the recovery of the repeated regions.If three or more un-sequential elements are detected in any of the two main arrangements as a result of gapped palindromes, *ECuADOR* takes the extra smallest dissimilar sequential elements and it will create a new fragment from them. This new fragment will contain the start position of the multi non-sequential elements located in the first inverted repeat with the end position of the multi non-sequential elements located in the second inverted repeat. Then, this new fragment will be concatenated with the ending homologous fragment.Once the inverted repeat regions are located, the algorithm maps the remaining positions (LSC and SSC respectively) through a sequence sweep from start to end.When all the four main regions are found through mapping-by similarity-analysis (LSC, IRa, SSC and IRb respectively), *ECuADOR* reorganizes the structure of the entire sequence in the same order for all the analyzed plastomes.*ECuADOR* generates five output files. The first one is a summary of the main annotation features of the regions, including the length of the IRs. The second one is the reordered sequences (i.e., LSC--IRb--SSC--IRa) in either fasta or GFF3 format, depending on user options (for GenBank files, it will provide a complete reorientation and extraction for all its established annotations in a new GFF3 output file). If a sequence cannot be reliably curated, *ECuADOR* generates a third file containing a problem description for that particular sequence. If the option "all-orient TRUE" was selected, an additional step is performed, that will homogeneize the direction of the sequences for each plastome fragment before generating the final results files (for this purpose, we integrated in the *ECuADOR* code a modified version of the script *seqOrient.pl*, available at). Finally, we included an additional option number of inverted repeats ("-noIRs") which allows the user to get either one or two IRs. The default value is 1 (thus outputting the LSC--IRb--SSC), while "-noIRs 2" will output the complete plastome sequence (LSC--IRb--SSC--IRa, useful e.g., before annotation for GenBank submission). The output (fasta file) can be immediately used in software for downstream phylogenetic and genomic analysis (e.g., Geneious, CLC workbench or MAFFT, PHYML and BEAST).
{#fig-2}
Taxon selection and dataset construction
----------------------------------------
To test and assess performance of *ECuADOR*, two main datasets and a reference case were generated. The first (dataset1: *data control*) was used to reevaluate sequences corresponding to 161 published plastomes from a selection of 51 major angiosperm groups ([@ref-23]). Fasta sequences of the plastomes were downloaded from GenBank and analyzed using *ECuADOR* with default parameters and a sliding window adjusted to 1,500 bp. We reviewed the accuracy of IR identification by evaluating the similarity and position of each retrieved region compared to their original annotation, using Geneious R9 v.9.0.5 ([@ref-22]), with the "Find repeats" function (1,500 bp as minimum repeat length; no allowed mismatch between repeats). To further evaluate the quality of the obtained locations (LSC, SSC, IRs), we built a phylogenetic tree based on the 161 reanalyzed plastomes and compared this with the previously published topology ([@ref-23]). The sequences were aligned using MAFFT v.7 ([@ref-15]) under the FFT-NS-1 option and PhyML v3.3.20 ([@ref-11]) was used to build a maximum-likelihood (ML) tree with the GTR DNA substitution model and the fast likelihood based method (aLRT SH-like).
Secondly, to assess the robustness of our algorithm, we simulated low quality/noisy plastomes by introducing random substitutions with different percentages of variation in the plastome of *Arabidopsis thaliana* (dataset2: *data testing*). Eleven levels of variation were chosen, ranging from 0.01% to 5.31% and 1,000 simulations were generated for each variation level, respectively, using an in-house script (available at <https://github.com/BiodivGenomic/ECuADOR/>). These "low quality" plastomes were then evaluated through *ECuADOR* with default parameters.
Finally, in order to evaluate the applicability of our algorithm to any kind of chloroplast data (not only for families previously analyzed) and taking advantage of its fast detection and extraction speed, a dataset corresponding to 4,541 angiosperm chloroplasts was downloaded from GenBank (database INSDC accessed on 2019/06/20-dataset3: *mass data evaluation*) and analyzed using default options in the same format. This further served as an additional survey to detect cases with missing or poorly uploaded regions currently available in GenBank. The goal of this analysis was to evaluate the percentage of negative cases (i.e., without any IR identified), using the largest number of chloroplasts available for angiosperms to date and to determine the underlying causes of these events.
Results
=======
Prediction of performance analysis (Dataset1: data control)
-----------------------------------------------------------
To evaluate *ECuADOR*, we compiled 161 plastome sequences from a total of 51 major angiosperm groups, as previously used in [@ref-23]. *ECuADOR* ran fluently for each plastome, with regions identified in almost all sequences ([Table S1](#supp-1){ref-type="supplementary-material"}). It provided basic information regarding the location and inverted repeat lengths, reordering of the main plastome regions (LSC--IRb--SSC--IRa) as well as the repositioning for all the protein coding genes (CDS) in gff3 format file for all plastomes analyzed. Furthermore, it substantially eases the post-processing analyses of plastomes reconstructed from NGS data. Results obtained with *ECuADOR* and Geneious were very similar, validating the performance and accuracy of our approach ([Table S1](#supp-1){ref-type="supplementary-material"}). Retrieved annotations were identical for 150 sequences out of 161. This number increased to 160 after manually setting the first position of the LSC as the start of the plastome (contrary to *ECuADOR*, in which this is done automatically). IR annotations were not retrieved accurately for one draft plastome sequence (GU592211), due to very poor sequence quality.
*ECuADOR* took 15 min to analyze this dataset using a MacBook Pro, 2.2 GHz Core 2 Duo, 16 Gb RAM. When compared to manual individual treatment (e.g., average of 3--5 min handling per plastome), this would have taken between 8 and 13 h.
The topology of the obtained phylogenetic tree ([Fig. 3](#fig-3){ref-type="fig"}) was similar for 49 families compared to previously identified relationships ([@ref-23]). An inconsistency was found in the placement of Ranunculaceae (*Ranunculus macranthus*), which grouped together with Piperaceae, Dioscoreaceae and Chloranthaceae.
![Phylogenetic tree constructed with 161 cpDNAs, using fast likelihood-based method (aLRT SH-like) as implemented in PhyML ([@ref-11]).\
Numbers on nodes indicate probability values. Families highlighted in red show an inconsistency found in the placement of Ranunculaceae (*Ranunculus macranthus*), which groups together with Piperaceae, Dioscoreaceae and Chloranthaceae.](peerj-08-8699-g003){#fig-3}
Performance of the introduced variation simulation (Dataset2: data testing)
---------------------------------------------------------------------------
The introduction of mismatches between both repeats is based on the loss of information, thus reducing the identity of the IRs and altering the final reorganization of the plastome. This analysis allowed us to understand how the introduced error for the different simulation sets affects the recovery of the original positions of the inverted repeats and therefore the ability of the algorithm to retrieve the reordinated sequence completely. Thus, for each mismatch level, we scored and investigated cases where *ECuADOR* failed to retrieve the original IRs locations (due to excessive variation in the base pair numbers within the inverted repeats), using as a model the *A. thaliana* chloroplast genome ([Fig. 4](#fig-4){ref-type="fig"}).
{#fig-4}
*ECuADOR* was able to recover and reorder the main regions of the plastomes (LSC--IRb--SSC--IRa) for each altered dataset. As expected, the mismatching percentage affecting the true annotation increased as the alteration for each data set increased. *ECuADOR* showed an accuracy above 90% with 22 or fewer alterations ([Table S2](#supp-2){ref-type="supplementary-material"}). Such a high level of mismatch between the two IRs, likely represents major misassembled positions and we recommend such low-quality draft plastomes should first be carefully checked to assess the origin of such mismatches. Moreover, the user can easily modify the stringency of the detection process by specifying the sliding window fragment size to improve search precision. This should however only be done for noisy datasets which are known to contain high levels of mismatching or highly similar fragments throughout the plastome.
Reference data set (Dataset3: mass data evaluation)
---------------------------------------------------
To evaluate the potential and flexibility of *ECuADOR*, we analyzed a total of 4,541 angiosperm plastomes using default parameters with fasta format as input. The main regions were successfully detected in 4,446 sequences (97.90%, [File S1](#supp-3){ref-type="supplementary-material"}) whereas identification and organization of the plastome structure failed in 95 sequences (2.09%, [File S2](#supp-4){ref-type="supplementary-material"}).
To further assess the causes of these failed instances, the 95 sequences were evaluated in Geneious ([@ref-22]) and we found manual curation was needed to identify the main regions in these plastomes. Indeed, these plastomes had missing regions or poorly formatted annotations in GenBank, or were of extremely poor quality. Finally, several plastomes were characterized by an absence of inverted repeats (e.g., gymnosperm plastomes and several species in parasitic plants ([@ref-30], [@ref-31])).
*ECuADOR* requires a correct match between both repeats to be able to recover the newly generated regions (LSC--IRb--SSC--IRa) and score the analysis as successful. The script will not identify IRs in sequences of extremely poor quality. Nevertheless, it has several detection error and correction mechanisms---such as a wide detection of more than one inverted repeat region and recognition of ambiguous characters in the plastome.
Discussion
==========
*ECuADOR* is a novel algorithm designed for the identification, reorganization and reordering of homologous regions (LSC, SSC and IRs) on large-scale plant datasets using the specific location of the inverted repeats in any point of the circular genome as a starting point. Other software aimed at helping with circular genomes exist, but none handles the post-assembly curation in a standardized manner. For example, *Circlator* ([@ref-13]) was designed to use with long sequencing reads (e.g., PacBio data, Menlo Park, CA, USA) in bacterial chromosomes and plasmids and the plastid and mitochondrial genomes of eukaryotes to reconstruct circular genomes. However, *Circlator* is an assembler and thus works with the filtered, long sequencing reads (FASTQ files) to generate a circular genome. In that way, it is not different from the many assemblers or scripts specifically designed to generate organelles genomes (e.g., ORG.asm, NovoPlasty, GetOrganelle) that can be used before *ECuADOR*. As most phylogenomic studies combine both new sequences and plastomes mined from GenBank, original raw data can be missing for a significant part of the sampling. Although long sequencing reads are quickly emerging as a powerful tool in genomics, the vast majority of generated data available are short reads (\<300 bp) from Illumina platforms, prohibiting the use of a long-read-specific solution. As *Circlator* could be involved in the assembly process, *GenomeRing* ([@ref-12]) is a visualization tool that allows an overview of several plastomes in the same coordinate system. In that sense, it could be considered as a good complement to *ECuADOR*, to visualize the results after alignment. But given that an alignment is a requirement for *GenomeRing* and the program does not generate any files for immediate downstream analyses, we cannot consider *GenomeRing* as an alternative for *ECuADOR*. The closest algorithm to *ECuADOR* would be *MARS* ([@ref-18]), as it can homogeneize the starting point of a set of sequences from a circular genome, a function *ECuADOR* also performs as a side-effect of the reorganization of the plastome. However, *ECuADOR* is able to not only move the starting point, but also to keep order and orientation of the main regions of the plastome. In addition, *MARS* outputs a FASTA file, while *ECuADOR* can keep track of all annotations previously included in the input files. Fast-Plast ([@ref-19]), despite being a "all-in-one" pipeline including assembly and ordering of the plastome, is not able to handle GenBank sequences, limiting its use to newly generated draft plastomes. In addition, its structure implies to work with each plastome separately, while *ECuADOR* is able to homogeneize the output order and direction for a complete set of sequences, making downstream analyses easier.
A possible explanation for the incongruence observed in dataset1 could be that [@ref-23] used only protein coding data of 78 genes from 360 taxa to build their phylogenetic tree. Gene conflicts in plastome-based phylogenies have recently been highlighted as a major cause of incongruence among studies ([@ref-7]; [@ref-29]), and differences in methodology between our study and the study by [@ref-23] could explain the observed incongruence. The analyses performed here were based on complete plastome data and include non-coding, fast evolving regions. Despite that these regions have been proven useful at smaller taxonomic scales, they are expected to saturate at very large scales, resulting in nonspecific phylogenetic signals in deeper parts of the tree and causing incongruence with signals inherent in coding-regions. It is beyond the scope of our study to provide a detailed analysis of the problems inherent in the reconstruction of the phylogeny of angiosperms but worthy to note that the level of saturation effects in the plastome dataset seems low, with only one incongruence compared to the analysis using coding-regions only.
The introduction of *ECuADOR* has provided a major step forward in our ability to quickly identify and extract the main plastome regions in a coordinated, standardized arrangement. This new reference system not only will allow to define a global reorganization for the main plastome regions but can also be employed to generate a chain of re-repositioning for all the remaining GenBank annotations in the sequence. This last condition applies to all plastomes in GenBank in which the string starts randomly along the circular sequence. This allows to recast the previous reference annotation into a new coordinate system for all the available annotations and print this out in a GFF3 output file. In addition, the generation of reorganized and reoriented datasets, thus already formatted for downstream analyses (e.g., alignment) will greatly improve the utility of plastomes, either newly sequenced or mined from GenBank. Finally, the new coordinate system can be used to implement exploratory methods for a more accurate and faster analysis of phylogenetic comparative data, either using complete regions or concrete molecular markers in case of using GenBank files. This, in turn, will allow us to advance the development of more accurate hypotheses in the reconstruction of the evolutionary history of extant plant groups.
Conclusions
===========
Curating draft plastomes and formatting them for downstream phylogenetic analyses is laborious, time-consuming and error-prone. We developed *ECuADOR* for the robust extraction and mass reorganization of plastome regions. The proposed algorithm is based on sliding windows and dynamic suffix array approaches to track inverted repeat locations, followed by extraction and repositioning of the main chloroplast regions. In addition, the user can generate datasets in which all sequences are similarly oriented, allowing a direct inference of the homology through sequence alignment. This facilitates post-processing analyses of extra-nuclear genomes from NGS data, optimizing handling times and reducing error. We demonstrated its accuracy, especially in handling poorly reconstructed plastomes, when repeats are interrupted by misassembled positions (resulting in fragments poorly positioned throughout the sequence), preventing recovery of IRs. This method significantly reduces handling time and complexity in the analysis of large plastome datasets and allows for error free processing of high quantities of data. Our study not only underscores the importance of developing new tools for detecting and characterizing inverted repeated sequences, but also provides a new approach to systematically identify complete regions within plastomes. *ECuADOR* will be maintained and regularly improved to add new features, according to the emergence of new needs with the development of innovative approaches in NGS. For example, future scheduled improvements include the extraction of the homologous genes and non-coding regions (i.e., intergenic spacers, introns and ribosomal RNA) to generate locus-specific alignments, as currently widely used to avoid the laborious steps involved in manual curation. We believe *ECuADOR* has the potential to be useful and widely applicable to the plant science community that handles large genomic datasets on a daily basis, whether this is for genomics/phylogenomics, evolution, ecology or bioinformatics.
Supplemental Information
========================
10.7717/peerj.8699/supp-1
###### ECuADOR results for Dataset 1.
Dataset1 containing 53 major angiosperm groups used in our study obtained from [@ref-23] as well as the analysis of their 161 evaluated species. Positions and length of the large single copy (LSC), inverted repeat-B (IRb), short single copy (SSC) and inverted repeat-A (IRa) are shown. In addition, these results were compared with the "Find repeats" function in Geneious R9 v.9.0.5 to confirm successful matches, validating the performance and accuracy.
######
Click here for additional data file.
10.7717/peerj.8699/supp-2
###### Analysis results of 1,000 simulations with dataset 2 and the *Arabidopsis thaliana* reference plastome sequence (NC_000932).
######
Click here for additional data file.
10.7717/peerj.8699/supp-3
###### Dataset 3, corresponding to 4446 angiosperm chloroplasts, was used in our evaluation.
These chloroplast genomes were downloaded from GenBank, using the database INSDC (accessed on 2019/06/20-dataset 3: mass data evaluation).
######
Click here for additional data file.
10.7717/peerj.8699/supp-4
###### Dataset 3, corresponding to 95 angiosperm chloroplasts where the main regions were not detected.
######
Click here for additional data file.
We kindly thank F. Areces-Berazain for providing useful comments to an earlier version of this manuscript.
Additional Information and Declarations
=======================================
The authors declare that they have no competing interests.
[Angelo D. Armijos Carrion](#author-1){ref-type="contrib"} performed the experiments, analyzed the data, prepared figures and/or tables, authored or reviewed drafts of the paper, designed the algorithm, wrote the program, and approved the final draft.
[Damien D. Hinsinger](#author-2){ref-type="contrib"} conceived and designed the experiments, authored or reviewed drafts of the paper, designed the algorithm, and approved the final draft.
[Joeri S. Strijk](#author-3){ref-type="contrib"} conceived and designed the experiments, authored or reviewed drafts of the paper, and approved the final draft.
The following information was supplied regarding data availability:
The program is available at GitHub: <https://github.com/BiodivGenomic/ECuADOR/>
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Cell hyper-proliferation has long been considered as an important etiological factor of many cancers, and self-sufficiency proliferation signals are sustaining, such as Ras signal pathway \[[@CR1], [@CR2]\]. Mitofusin 2 (Mfn2) is involved in regulation of cell survival and has been of interest in cancer field \[[@CR3]--[@CR6]\]. It may be involved in cervical cancer pathogenesis and might serve as a biomarker of cervical SCC in the future \[[@CR7]\]. In our and others studies, Mfn2 could induce Hela cells into mitochondrial apoptosis \[[@CR8]\], pancreatic cancer into cell autophagy \[[@CR9]\], and breast cancer into DNA methylation \[[@CR10]\], as a tumour suppressor in many cancers. Mfn-2 is a mitochondrial outer-membrane protein with GTPase activity involved in mitochondrial fusion and fission and apoptosis regulation \[[@CR11]--[@CR13]\]. *Mfn2* gene encodes for a 757-amino-acid protein containing a p21Ras domains near the N-ter \[[@CR14]\]. The p21Ras domain makes Mfn2 as an anti-Ras protein in VSMCs, and regulating the VSMCs cell-cycle \[[@CR15]\]. In human breast cancer, overexpression of Mfn2 inhibited the Ras-ERK1/2 signaling pathway, but with deletion of the p21Ras motif partially reduced the anti-tumor function of Mfn2 \[[@CR10]\].
In our early studies, we found that PTD4-apoptin fusion protein could upgrade Mfn2 expression in cervical carcinoma cells \[[@CR16]\]. Then, we found that when the expression of Mfn2 increased, the Hela cells were induced into apoptosis via mitochondrial pathway \[[@CR8]\]. In this study, we aimed to investigate whether Mfn2 was involved in proliferation in Hela cells, and activated Ras signaling pathways to inhibit Hela cells proliferation. Our findings provide a new target of cervical carcinoma and suggest candidates for potential use in cervical carcinoma therapy in the future.
Materials and methods {#Sec2}
=====================
Antibodies {#Sec3}
----------
Antibodies were as follows: Anti-Mfn2 (D2D10) (Cell Signaling, 9482) directed against Mfn2 protein, Ras (Cell Signaling, 3965), Cyclin D1 (Cell Signaling, 2922), p44/42 MAPK (Erk1/2, Cell Signaling, 9102), PCNA (PC10) (Cell Signaling, 2586), Myc (Cell Signaling, 5605), mTOR (Cell Signaling, 2972), STAT3 (Cell Signaling, 9139) and NF-κB p65 GAPDH (G-9) (Santa Cruz, sc-365062, monoclonal, mouse) and β-actin (Tianjin Sungene Biotech, China) antibodies were used as the loading controls.
Construction of mfn2 expression adenoviral vector {#Sec4}
-------------------------------------------------
Rno-mfn2 precursor DNA (*Homo sa*piens (human), Gene ID: 9927) was synthesized by Genechem (Shanghai, China). The adenovirus expressing mfn2 (Adv-mfn2), or control adenovirus expressing control (Adv-control) was generated using the AdMax system (Microbix Biosystems, Canada) according to Wang's in 2018 \[[@CR8]\].
Cell culture {#Sec5}
------------
The human cervical carcinoma cell line HeLa was purchased from the Chinese Culture Tissue Collection Center (CCTCC, China). The cells were cultured in DMEM (Hyclone, USA) supplemented with 10% FBS (fetal bovine serum; Hyclone, USA) at 37 °C and 5% CO~2~.
qRT-PCR analysis {#Sec6}
----------------
The total RNA from Hela cells was extracted using TRIzol@ Reagent (Invitrogen, USA). Reverse transcription and qRT-PCR were performed as described previously \[[@CR8], [@CR17], [@CR18]\]. Amplification and detection of specific products were performed with the ABI stepone plus (PE Applied Biosystems). The mfn2 mRNA expression was measured by RevertAid Reverse Transcriptase (Thermo scientific, EP0442) and qPCR Master Mix (Fermentas,K0221), and the GAPDH was used as an internal control. The 2−ΔΔCt method was used to measure the realtime PCR Data. The following sequence-specific primers of Mfn2 were as follow: F: 5′-ATCTGTGCCAGCAAGTTGACA-3′ and R: 5′-AAGTGAATCCAGAGCCTCGAC-3′.
CCK-8 test {#Sec7}
----------
The cells, seeded into the 96-well plate at 3000 cells per well, were incubated with 50, 100 and 150 pfu/cell of Mfn2 or as negative control with PBS for 48 and 60 h in 5% CO~2~ at 37 °C according to Wang's paper in 2018 \[[@CR8]\].
Cell cycle {#Sec8}
----------
The cells, seeded into the 6-well plate at 1 × 10^5^ cells per well, were incubated with 100 pfu/cell Adv-mfn2 or Adv-control of Adv-mfn2 or Adv-control for 60 h in 5% CO~2~ at 37 °C. Cells were then washed twice in ice-cold PBS, stained with Cell Cycle and Apoptosis Analysis Kit (Beyotime, C1052, China), for 30 min at room temperature and analyzed with a BD FACSort flow cytometer (BD Biosciences, USA). Cell cycle data was analyzed by ModFit LT 3.2 software (Verity Software House, Topsham, USA).
Immunofluorescence {#Sec9}
------------------
HeLa cells were seeded in 6-well plates at a ratio of 10,000 cells per well. After 12 h, Adv-mfn2 or Adv-control was added into the medium of 50 pfu/cell and incubated at 37 °C. After 60 h, the cells were washed 3 times with PBS and fixed with 4% paraformaldehyde for 10 min at room temperature. Subsequently, the cells were washed 3 times with PBS and permeabilized for 5 min with PBS containing 0.2% Triton X-100. Anti-mfn2 antibodies were used to detect the presence and cellular localization of mfn2 protein in HeLa cells, as recently reported \[[@CR16]\]. The appropriate Rhodamine-conjugated goat anti-mouse IgG antibodies (Pierce, 31569) were used as secondary antibodies. The cellular nuclei were stained with 4, 6-diamidino-2-phenylindole (DAPI, 1 μg/L in PBS, Roche, 10236276001). The cells were analyzed by means of confocal fluorescence microscopy Fluoview FV50 (Olympus, Japan).
Western blotting {#Sec10}
----------------
The Whole cell proteins and tissue proteins from model mouse were extracted by ice-cold SDS lysis buffer. The BCA protein assay kit (Pierce, 23227) was used to determine the protein concentrations according to the manufacturer's instructions. The proteins were fractionated on 12% and 15% SDS-polyacrylamide gel and electroblotted onto Immobilon-P PVDF transfer membranes (Millipore, IPVH08130), as recently reported \[[@CR8]\]. The blots were incubated with Mfn-2, anti-Ras, Cyclin D1, ERK1/2, PCNA, c-Myc, mTOR, STAT3 and NF-κB p65. GAPDH and β-actin were used as the loading controls. The positive signals were visualized by Odyssey^®^ Two-Color Infrared Imaging System (Li-Cor, USA).
Xenografted cervix carcinoma mouse model {#Sec11}
----------------------------------------
BALB/c nude mice (4--5 weeks old, female) were obtained from the Hubei Provincial Center for Disease Control and Prevention (HBCDC, China). 5 × 10^5^ Human cervix carcinoma HeLa cells were collected and injected subcutaneously under alar of nude mice as recently reported \[[@CR8]\]. When the tumors were visible, the mice were divided randomly into 2 groups, consisting of 5 tumor-bearing mice per group for a 2-week-treatment with Adv-mfn2 or Adv-control, respectively. The solution samples were infected into the tumor tissue. Every 3 day 5×10^8^ pfu/ml Adv-mfn2 or Adv-control were applied per mouse as our last paper \[[@CR8]\]. All animal studies were carried out in accordance with the "Guide for the Care and Use of Laboratory Animals" and approved by the Hubei Provincial Center for Disease Control and Prevention (HBCDC, China).
Detection of xenografted tumor growth {#Sec12}
-------------------------------------
The tumor volume was measured before and after the 2-week-treatment by the following formula: volume = 0.52 × length × width^2^ \[[@CR8], [@CR16]\]. The tumor volume difference was calculated by the following formula: difference = after volume -- before volume. After 2-week-treatment, the tumor tissues were obtained, and prepared for HE stains.
HE assay {#Sec13}
--------
After a 2-week-treatment, respectively, the mice were killed and the tumors were fixed with 4% paraformaldehyde, and paraffin sections were prepared for carrying out HE assay, as recently reported \[[@CR17]\].
Statistics {#Sec14}
----------
The Student's t test was used to determine the statistical significance of data. A p value of less than 0.05 (\*) or less than 0.01 (\*\*) was considered to be significant. Data presented in the figures represent the mean ± standard error.
Ethics statement {#Sec15}
----------------
Our animal studies were carried out with the Guide for the Care and Use of Laboratory Animals of the People's Republic of China in strict compliance. All efforts were made to minimize suffering and all procedures were performed under ethylether anesthesia. The study protocol was approved by the Committee on the Ethics of Animal Experiments of the Chinese Centre for Diseases Control and Prevention.
Results {#Sec16}
=======
Over-expression of Mfn2 in Hela cells {#Sec17}
-------------------------------------
To detect whether the adenovirus took the mfn2 gene into the cervix carcinoma cells, we uesd the qRT-PCR to obverse the expression of mfn2 gene in Hela cells. Adv-mfn2 or Adv-control was given into the wells with 1 × 10^6^ Hela cells. After incubated for 60 h, the cells were collected, and lysed in ice-cold TRIzol@ Reagent to obtain the total RNA. Figure [1](#Fig1){ref-type="fig"}a shows that the adenovirus take the mfn2 gene into the cells, and increases the expression of mfn2 gene. Then, we detected the expression of Mfn2 protein in Hela cells. We dissolved the Hela cells which were incubated with Adv-mfn2 or Adv-control for 60 h, and used the western blot to obverse the expression of Mfn2 proteins. We could see that Mfn2 proteins were over-expression in Adv-mfn2 Hela cells (Fig. [1](#Fig1){ref-type="fig"}b, c).Fig. 1The overexpression of Mfn2 and its effect on cell proliferation in Hela cells. **a** HeLa cells were incubated with Ad-Mfn2 (100 or 150pfu/cell) or Adv-control. The expression of Mfn2 mRNA in the Hela cells was detected by qRT-PCR. Data are represented as mean ± SD (\**p* \< 0.05, \*\**p* \< 0.01). **b**, **c** Cell lysate was extracted from HeLa cells treated with Adv-control, Adv-Mfn2 (100 or 150pfu/cell) for Mfn2 expression by western blot analysis (**b**) and quantitation (**c**). GAPDH was used as reference. **d** Adv-mfn2 were added to the medium of the cells, and incubated for 60 h. Cellular staining of nuclear DNA (DAPI) and localization of Mfn2 protein in HeLa cells. Magnification: 200×. **e**, **f** HeLa cells were exposed to different concentrations of Adv-Mfn2 (0, 50, 100, 150 pfu/cell) for 48 h and 60 h, and then the cell viability was measured by CCK-8 assay. The Hela cells decreased after being treated with Adv-Mfn2 in a dose- and time-dependent manner. (mean ± SD from three independent experiments, \*\**p *\< 0.01). **g** After incubated with Adv-mfn2 (0, 100 and 150 pfu/cell), the expression of PCNA protein in HeLa cells was detected by western blot. β-actin was used as reference
Cellular location of Mfn2 in Hela cells {#Sec18}
---------------------------------------
Mfn2 is considered as the outer membrane protein of the mitochondria \[[@CR11]\], so we analyzed the cellular location of Mfn2 in Hela cells. The Hela cells were incubated with 50 pfu/cell Adv-mfn2 or Adv-control for 60 h in 6-well plate at 37 °C. Then, the Mfn2 antibody was incubated in the wells, as DAPI for cellular nuclear. Across the fluorescence microscope, we examined the cellular location of Mfn2 in Hela cells (Fig. [1](#Fig1){ref-type="fig"}d). The Mfn2 proteins were increased in Mfn2 group, and located in the cytoplasmic, not in the nuclear.
Mfn2 inhibits the growth of cervix carcinoma cells {#Sec19}
--------------------------------------------------
The CCK-8 was performed for detecting the relative inhibition rate of Hela cells. To obverse the inhibition of Mfn2 in Hela cells, we set the CCK-8 to detect the relative inhibition rate of Hela cells. The cells were seeded into the 96-well plates as 3000 cells per well. After 12 h, the mediums of 50, 100 and 150 pfu/cell of Adv-mfn2 or Adv-control was incubated into the wells. After 0, 48 and 60 h, the cells were analyzed by CCK-8 test. We could found that the relative inhibition of Mfn2 in Hela increased, depend on the time and the dose (Fig. [1](#Fig1){ref-type="fig"}e, f). 60 h incubation group had the highest inhibition, as well as the 150 pfu/cell Adv-mfn2. It was suggested that Mfn2 could inhibit Hela cells growth with time and dose dependence.
Proliferating Cell Nuclear Antigen (PCNA) is the major coordinator of faithful and processive replication and DNA repair at replication forks \[[@CR19]\]. It is bound up with the cells proliferation. Here, PCNA protein was examined to proof the inhibition of Mfn2 in Hela cells. The expression of PCNA in Mfn2 groups was decreased obviously in western blot result in Fig. [1](#Fig1){ref-type="fig"}g dependence on the dose of Mfn2.
Over-expression of Mfn2 altered cell-cycle in Hela cells {#Sec20}
--------------------------------------------------------
To confirm the possible mechanism of inhibition of Mfn2 in Hela cells, the cell-cycle test was performed in Hela cells incubated with Adv-mfn2. The 100 pfu/cell of Adv-mfn2 or Adv-control was added into Hela cells in 6-well plate for 60 h. Fluorescence activated cell sorting (FACS) analysis was used to examine cell-cycle distribution. Adv-control added into well of Hela cells, approximately 35.53% of Hela cells infected by Adv-control progressed into S phase. On the contrary, Hela cells infected with Adv-mfn2 remained mostly in the G0/G1 phases with only 26.58% of cells entering S phase. We speculated that Mfn2 could further block the Hela cells in G0/G1 phase due to the effects of Adv-mfn2 on Hela cells proliferation as shown in Fig. [2](#Fig2){ref-type="fig"}a, b.Fig. 2The effect and the signal pathway of Mfn2 on cell-cycle in Hela cells. **a** HeLa cells were incubated with Ad-Mfn2 for 60 h. Cells stained with Cell-Cycle assay after incubated with two concentrations (100 pfu/cell) of Adv-Mfn2 were measured by flow cytometry. **b** The percentage of every phase of cell-cycle after HeLa cells being treated with indicated concentrations of Adv-Mfn2 or Adv-control was calculated. **c** Adv-Mfn2 increased the expression of Cyclin D1 by western blot. Beta-actin was used as reference. **d** Expressions of Ras and Myc after HeLa cells being treated with Adv-Mfn2. GAPDH was used as reference. **e** Expressions of STAT3 and NF-κB after HeLa cells being treated with Adv-Mfn2. GAPDH was used as reference
Cyclin D1 is an oncogene frequently overexpressed in human cancers that has a dual function as cell cycle and transcriptional regulator, although the latter is widely unexplored \[[@CR18]\]. Western Blot of Cyclin D1 was used to further confirm that the cell-cycle of Hela cells was inhibited by Cyclin D1 down-expression in Hela cells, depend on the dose of Mfn2 in Fig. [2](#Fig2){ref-type="fig"}c.
Mfn2 activity is affected via Ras signal pathway {#Sec21}
------------------------------------------------
The function of the RAS signaling pathway is to integrate extracellular signals and coordinate a suitable response by a subsequent control of cellular growth, survival, and differentiation \[[@CR20], [@CR21]\]. Mfn-2 is known to block cell proliferation via inhibition of the Ras pathway in VSMCs \[[@CR15], [@CR22]\]. We examined whether Mfn2-mediated inhibition of cell proliferation and cell cycle in Hela cells was affected by the cellular level of Ras. HeLa cells were incubated with Adv-mfn2 or Adv-control. The results represented in Fig. [2](#Fig2){ref-type="fig"}d demonstrate that the expression of Ras are affected by Mfn2. Over-expression of Mfn2 could increase the Ras and Myc protein expression in HeLa cells. Meanwhile, the expressions of STAT3 and NF-κB p65 protein were also decreased by incubation with Adv-Mfn2 (Fig. [2](#Fig2){ref-type="fig"}d, e).
Therefore, we conclude that Mfn2 inhibits Hele cells proliferation and cell-cycle by activating Ras protein expression. Furthermore, our finding that Myc, STAT3 and NF-κB p65 protein are decreased by incubated with Adv-mfn2 in HeLa cells. The present research showed that Ras was sensitive in Mfn2-induced proliferation depressing in HeLa cells, and inactivated its downstream signal pathway as Myc, STAT3 and NF-κB p65 protein expression.
In-vivo, Mfn2 inhibits cervix carcinoma growth {#Sec22}
----------------------------------------------
In order to verify the effect of Mfn2 protein against the cervix carcinoma in vivo, we examined the therapeutic effect of Adv-mfn2 on xenografted cervix carcinoma in a mouse tumor model for a 2-week treatment. HeLa cells were injected subcutaneously and the animals were randomly divided into two groups of each 5 tumor-bearing mice. These tumor-bearing mice were treated with Adv-mfn2 or Adv-control as described in the Materials and Methods section. Before and after the treatment, the volume of the tumors was determined. During the treatment, the volume of xenografted cervix carcinomas treated with Adv-mfn2 increased slowly, whereas the volume of the tumors treated with Adv-control increased faster than the Adv-mfn2 group (Fig. [3](#Fig3){ref-type="fig"}a, b). At day 14, the mice were sacrificed and the tumors were macroscopically or histologically analyzed. The tumors treated with Adv-mfn2 were smaller than the ones treated with Adv-control (Fig. [3](#Fig3){ref-type="fig"}b). After the 2-week treatment all mice were checked, and there were not any metastases in the mice of two groups.Fig. 3Mfn2 inhibited the tumor growth in vivo of cervical carcinoma mouse model and its signal pathway. **a** Determination of the tumor volume in mice treated with Adv-mfn2 or Adv-control for the 0, 3, 6 and 9 days. Data are represented as mean ± SD (\*\**p* \< 0.01). **b** The sizes of xenografted cervix carcinoma after 14-days treatment in the Mfn2 group and the con group. **c** HE assay of tumor sections from animals treated with either Adv-control or Adv-Mfn2. Magnification: 200×. **d**--**f** The expression of Mfn2 and Cyclin D1 proteins in the tumors of mouse model was detected by western blot (**a**, **b**) and quantitation (**c**). Beta-actin was used as reference. Data are represented as mean ± SD (\*\*p \< 0.01). **g** Western blot showed Ras, Erk1/2 and Myc proteins expressions in cervix tumors treated with Adv-mfn2 after 2-weeks treatment. Total tumor tissue lysates were prepared and analyzed by Western blot for these proteins. GAPDH was used as reference. **h** Western blot showed mTOR and NF-κB proteins expressions in cervix tumors treated with Adv-mfn2 after 2-weeks treatment. Total tumor tissue lysates were prepared and analyzed by Western blot for these proteins. GAPDH was used as reference
To detect the underlying mechanism of the Mfn2-triggered tumor reduction, tumor sections of both Adv-mfn2 and Adv-control treated tumors were analyzed by means of a HE assay and Western blot. The nucleus of cervix carcinoma cells treated with Adv-mfn2 was pycnosis and fragmented into several parts in the analyzed histological section. In contrast, the tumors treated with Adv-control had no signs of nucleus pycnosis (Fig. [3](#Fig3){ref-type="fig"}c). Then, Western blot was performed to confirm that Mfn2 was over expression in the tumor tissue of Adv-mfn2 group, and the expression of Cyclin D1 was decreased in Adv-control group (Fig. [3](#Fig3){ref-type="fig"}d--f). The expressions of Ras, Erk1/2, Myc, mTOR and NF-κB p65 protein in tumor tissues of 2 groups were detected by Western blot, and these were decreased by treated with Adv-mfn2 comparing with the Adv-control group (Fig. [3](#Fig3){ref-type="fig"}g, h). We found that Mfn2 could decrease the proteins of Ras-NF-κB signaling pathway.
Discussion {#Sec23}
==========
In the current study, we confirmed that Mfn2 could inhibit Hela cells proliferation in vitro and in vivo, and arrest Hela cells cell-cycle. The cellular location of Mfn2 in Hela cells was performed by immunofluorescence. The results of CCK8 and the western blot of PCNA showed that Mfn2 inhibited the Hela cells proliferation in the dose- and time-dependence. The flow cytometry and the western blot of Cyclin D1 meant that the cell-cycle of Hela cells were arrested by Mfn2 in G0/G1 phase. It was been confirmed that Mfn2 was the inhibitor of Ras in VSMCs \[[@CR15]\]. To find the signal pathway of cell-proliferation inhibition and cell-cycle arrest in Mfn2 manner in Hela cells, we detected the Ras, Myc, NF-κB p65, STAT3 proteins by western blot. The results of these western blots suggested that Mfn2 could inhibit the Hela cells by decreasing the expression of Ras and relative proteins in Ras-NF-κB signal pathway. In xenografted mouse model, we measured the tumor size before and after the Adv-mfn2 or Adv-control treatment, and analyzed the expression of Cyclin D1, Ras, Myc, ERK1/2, NF-κB p65, mTOR proteins by western blot. These results implicated that Mfn2 could inhibit the Hela cells growth by declining the expression of Ras protein and arresting the Ras-NF-κB signal pathway.
The proliferation of cells was remarkably inhibited, thus inducing the cells into apoptosis or senescence \[[@CR23], [@CR24]\]. In our previous study, the inhibition of Mfn2 in Hela cells has been proved to be achieved through the apoptosis induction \[[@CR8]\] and cell cycle arrest. Mfn2 could inhibit Ras-Erk1/2 and PI3 k-Akt signal pathway in VSMCs \[[@CR12], [@CR15]\]. Ras-Erk1/2 and PI3k-Akt-mTOR are the classic signal pathways of cell proliferation \[[@CR25], [@CR26]\]. NF-κB has recently generated considerable interest as it has been implicated in human cancer initiation, progression and resistance to treatment \[[@CR27]\]. Mutations of upstream signaling molecules, such as Ras, often lead to constitutive activation of NF-κB in solid malignancies \[[@CR27], [@CR28]\]. NF-κB could stimulate the transcription of proliferation regulating genes like Cyclin D1 and Myc \[[@CR27], [@CR29]--[@CR31]\]. NF-kB does not function alone but is part of a network, which determines the pattern of its effects on the expression of STAT3 \[[@CR32]\]. Meanwhile, the activation of STAT3 is responsible for genes that promote cell proliferation such as Cylinc D1, Myc and so on \[[@CR33]\]. The activation of STAT3 could cause a positive feedback mechanism to NF-κB \[[@CR34]\]. STAT3 and NF-κB work together in a network for a result \[[@CR27]\]. Therefore, Mfn2 inhibits Ras- NF-κB signal pathway to arrest the cell proliferation and cell-cycle in Hela cells (Fig. [4](#Fig4){ref-type="fig"}). Based on our research, we propose that Mfn2 may be a novel target to the therapy of cervical carcinoma in the future.Fig. 4The network of Mfn2-Ras-NF-κB pathway in Hela cells
Conclusions {#Sec24}
===========
Mfn2 could inhibit the proliferation and cell cycle in Hela cells. The Ras-NF-κB signaling pathway was inactive by the expression of Mfn2 increasing. The xenografted cervical carcinoma mouse model was examined to confirm the effect of Mfn2 in Hela cells in vivo.
Mfn2
: mitofusin 2
FBS
: fetal bovine serum
CCK-8
: Cell Counting Kit-8
**Publisher\'s Note**
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Xiaowen Liu and Jun Sun contributed equally and share first authorship
Xiaowen Liu and Kangquan Shou contributed equally and share correspondence authorship
We thank Jing Zhang, Jiawang Ding, Hui Wu, Chao He, Chun Liu and Chao Luo for helpful discussion.
Conceived and designed the experiments: XWL, JSun, PY. Performed the experiments: XWL, PY, YHZ, WQG, JShe. Analyzed the data: XWL, PY, JY, JY. Contributed reagents/materials/analysis tools: JShe, JH. Contributed to the writing of the manuscript: XWL, PY, JSun, KQS. Advice of the experiments: JY, JY. Revised the manuscript: KQS. All authors read and approved the final manuscript.
National Natural Science Foundation of China for Youth to X. W. Liu. Grant number: 81402568; National Natural Science Foundation of China to J. Sun. Grant number: 81472833; Open fund of Key Laboratory of ischemic cardiovascular and cerebrovascular disease translational medicine (Three Gorges University) to J. She. Grant number: 2017KXN04.
Not applicable.
All applicable international, national, and/or institutional guidelines for the care and use of animals were followed.
Not applicable.
The authors declare that they have no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .569}
{#sp2 .570}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
Kidney cancer is one of the ten most common cancer types in both men and women, with an estimated number of 338,000 new cases per year \[[@CR1]\]. The most common tumor arising in the kidney is renal cell carcinoma (RCC) and a total of more than ten different subtypes can be identified \[[@CR2]\]. The therapeutic field of metastatic RCC (mRCC) has drastically changed in the past decade with the introduction of the VEGF signaling pathway inhibitors and inhibitors of mammalian target of rapamycin (mTOR) \[[@CR3]\]. For years, everolimus has been the standard second-line treatment after a VEGF-based treatment regimen until the arrival of axitinib as an alternative and recently nivolumab and cabozantinib, drugs that inhibit the PD-1 immune checkpoint and the MET, AXL and VEGF tyro-sine kinases, were shown to be more effective compared to everolimus \[[@CR4]--[@CR6]\]. Furthermore, the progression-free survival (PFS) of patients with mRCC was improved by addition of the multi-target tyrosine kinase inhibitor lenvatinib to everolimus \[[@CR7]\].
Everolimus has been shown to be an effective inhibitor of mTOR, resulting in inhibition of cell growth, proliferation, angiogenesis and survival of tumor cells \[[@CR8]\]. However, mTOR also plays an important role in the regulation of the immune response, by promoting the expansion of regulatory T cells (Tregs) \[[@CR9], [@CR10]\]. Since Tregs have immune suppressive capacities, this Treg-promoting effect of everolimus can be considered a detrimental effect in the treatment of cancer. In support of this notion, increased Treg numbers have been associated with poor survival in patients with cancer, including mRCC \[[@CR11]--[@CR13]\].
Several strategies have been investigated to selectively deplete Tregs, among them the use of low-dose cyclophosphamide (CTX). Administration of metronomic low-dose CTX was reported to selectively deplete Tregs, with additional beneficial effects on T and NK cell functionality \[[@CR14], [@CR15]\]. Therefore, a phase 1 clinical trial was initiated to prevent everolimus-induced detrimental Treg expansion, by adding metronomic CTX to the standard dosage of everolimus \[[@CR16]\], to achieve improved survival by modulating the immune system. Patients were treated in cohorts of five patients, with six different doses and schedules of CTX. Clinical results and results of changes in Treg frequencies in the various cohorts of this phase 1 trial were separately described \[[@CR17]\]. Here, we report on the results of the extensive and comprehensive immune monitoring that was additionally performed in this phase 1 study, where patients were treated with either everolimus alone or the combination of everolimus and different CTX administration dosages and schedules.
Materials and methods {#Sec2}
=====================
Study population {#Sec3}
----------------
Forty patients with mRCC and previously treated with a VEGF targeting regimen were treated with everolimus in combination with different doses and schedules of metronomic oral CTX. Thirty-nine patients were evaluable, since one patient was not able to complete 2 weeks of the treatment due to early toxicity. The trial was initiated by the department of medical oncology of the Amsterdam UMC, location VUmc and conducted within the context of the Netherlands Working Group on Immunotherapy of Oncology (WIN-O) with participation of 13 hospitals and enrollment of patients from January 2012 until August 2015. Clinical findings were reported separately \[[@CR17]\].
Treatment {#Sec4}
---------
Patients were treated with a fixed dose of 10 mg everolimus once daily and enrolled in one of the seven cohorts, five patients per cohort, with different doses and schedules of low-dose oral CTX. One patient in dose level 6 stopped treatment because of several toxicities (highest grade 3 nausea) within 2 weeks of enrollment and was not evaluable. CTX was scheduled week on/week off or continuously, once or twice daily, based on the previously used dose regimens reported by Ghiringhelli et al. \[[@CR15]\]. In cohort 0, patients were treated with 10 mg everolimus without CTX. In cohort 1, patients were treated with everolimus and 50 mg CTX once daily, week on/week off. In cohort 2, patients were treated with everolimus and 50 mg CTX once daily in a continuous scheme. In cohort 3, patients received 50 mg CTX twice daily, week on/week off, and in cohort 4 patients received 50 mg CTX twice daily, continuously. In the last two cohorts, cohort 5 and 6, respectively, patients received 100 mg CTX twice daily, in cohort 5 in a week on/week off regimen and in cohort 6 continuously.
Immune monitoring {#Sec5}
-----------------
At baseline and after 2, 4, and 8 weeks after the start of study treatment, 60 mL of heparinized peripheral blood was collected for immune monitoring. All materials were processed on the same day the blood was drawn. PBMC were isolated by density-gradient centrifugation with Lymphoprep (Axis-Shield, Oslo, Norway). After isolation, PBMC were stored overnight at 4 °C in RPMI 1640 (Lonza, Basel, Switzerland) supplemented with 100 IU/ml sodium penicillin (Astellas Pharma, Leiden, the Netherlands), 100 mg/ml streptomycin sulfate (Radiumfarma-Fisiofarma, Naples, Italy), 2.0 nM L-glutamine (Life Technologies, Bleiswijk, the Netherlands), 10% FBS (HyClone, Amsterdam, the Netherlands), and 0.05 mM 2-ME (Merck, Darmstadt, Germany). The next day, cells were stained for flow cytometric analysis.
Flow cytometry {#Sec6}
--------------
FITC, PE, PerCP or allophycocyanin (APC)-labeled antibodies directed against human CD3, CD4, CD8, CD11c, CD14, CD16, CD19, CD25, CD56, CD86, CD123, CTLA-4, HLA-DR, Ki-67, PD-1, (all BD Biosciences, New Jersey, USA), CD33, (Beckman Coulter Inc., California, USA), CD56 (IQ Products, Groningen, the Netherlands), and blood DC antigens BDCA1, BDCA2, BDCA3 (all from Miltenyi Biotec, Bergisch-Gladbach, Germany) and matching isotype control antibodies were used. Stainings were performed in PBS supplemented with 0.1% BSA and 0.02% sodium azide for 30 min. Intracellular staining was performed after fixation and permeabilization using a fixation/permeabilization kit according to the manufacturer's protocol (eBioscience). For staining of FoxP3, a PE-labeled Ab against FoxP3 (clone PCH101, eBioscience) or AlexaFLuor488 FoxP3 (clone 259D) (Biolegend) was used. Live cells were gated based on forward and side scatter and analyzed on a BD FACSCalibur (BD Biosciences) and analyzed using Kaluza Analysis Software (Beckman Coulter).
Statistical analysis {#Sec7}
--------------------
One-way repeated measures ANOVA was used to determine the statistical significance of differences within cohorts with Dunnett's multiple comparison test as post-test. Two-way ANOVA was used to compare the mean values between cohorts. Differences were considered statistically significant when *p* values were ≤ 0.05, as indicated with asterisks (\**p* ≤ 0.05, \*\**p* \< 0.01, \*\*\**p* \< 0.001). Statistical analyses were performed using GraphPad Prism software (version 7, 2016).
Results {#Sec8}
=======
The addition of a once daily oral dose of 50 mg CTX to treatment with everolimus results in Treg depletion and an increase in the CD8^+^ T cell: Treg ratio without changes in T-cell activation {#Sec15}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
As previously reported \[[@CR16]\], the main objective of this trial was to determine the optimal dose and schedule of orally administered CTX, when combined with 10 mg everolimus, to obtain selective Treg depletion. As shown in Fig. [1](#Fig1){ref-type="fig"}a (left graphs), cohort 2, the cohort where 10 mg everolimus was combined with 50 mg CTX continuously, showed a significant decrease in Treg percentages (within CD4^+^ T cells), both within the cohort, comparing the percentages at time point 0 to time point 4, and compared to the corresponding time point 4 in cohort 0, the everolimus only cohort, whereas CD4^+^ T-cell percentages remained stable (Fig. [1](#Fig1){ref-type="fig"}a and Supplementary Table 1). Cohort 2 was the only cohort in which this effect was observed. Except for cohort 4, in which a significant decrease in CD8^+^ T cells was observed in comparison to cohort 0 at time point 4, no major differences were observed between cohorts in CD8^+^ T-cell frequencies. On the other hand, the ratio of CD8^+^ T cells to Tregs was significantly increased in cohort 2 compared to cohort 0 at week 4 (Fig. [1](#Fig1){ref-type="fig"}a). This increase in CD8^+^ T cell:Treg ratio was only statistically significant in cohort 2. Based on the Treg-depleting data in cohort 2 and the observation that the Treg-depleting effect of CTX was less pronounced in subsequent cohorts, with even an increase in Treg percentages in cohort 5 and 6 (see Fig. [1](#Fig1){ref-type="fig"}a), the decision was made to proceed to the expansion cohort wherein an additional 5 patients were treated with the combination of 10 mg everolimus and 50 mg CTX continuously (as in cohort 2). The expansion cohort again showed a significant decrease in Treg percentages at time point 4 in comparison to Treg percentages from cohort 0 and a significant increase in the CD8^+^ T cell:Treg ratio, thereby confirming the previously observed results of cohort 2 (Fig. [1](#Fig1){ref-type="fig"}b).
Fig. 1Effect of different dosages and administration schedules of CTX when combined with a fixed dose of 10 mg everolimus on the frequency of Tregs, CD8^+^ T cells, the effector to suppressor (CD8:Treg) ratio and CD4^+^ T cells. **a** Relative percentages (to start) of Tregs, CD8^+^ T cells, the effector to suppressor ratio and CD4^+^ T cells were determined in freshly isolated PBMC from patients treated with different dosages and schedules of CTX, combined with a fixed dose of everolimus at baseline and subsequently 2, 4, and 8 weeks after start of treatment. Cohorts 1--6 correspond to the different CTX dosages and schedules investigated (black bullets, black line) and are compared to cohort 0, the everolimus only cohort (open bullet, dotted line). Tregs were determined within CD4^+^ T cells, CD8^+^ T cells and CD4^+^ T cells within CD3^+^ T cells. **b** Relative percentages of Tregs, CD8^+^ T cells, the effector to suppressor ratio and CD4^+^ T cells are shown for the expansion cohort. Patients were again treated with 50 mg CTX once daily, combined with 10 mg everolimus once daily as previously in cohort 2. Means ± SEM are shown
For T-cell activation, PD-1 and CTLA-4 expression was determined on CD4^+^ and CD8^+^ T cells. Overall, no consistent or persistent changes in either PD-1 or CTLA-4 expression on either subset of T cells could be observed (Supplementary Fig. 1a). This was also the case for cohort 2 and the expansion cohort 2E (Supplementary Fig. 1b). As Supplementary Fig. 1 shows relative values, absolute percentages of PD-1 and CTLA-4 expression on CD4^+^ and CD8^+^ T cells are shown in supplementary table 2.
As a measure of the proliferative activity of Tregs and CD4^+^ T cells, Ki-67 expression was determined in both cell types. As shown in Supplementary Fig. 2, a significant decrease in Treg Ki-67 expression was observed within cohort 3 and 5 comparing the expression at baseline to time point 2. In addition, the percentage of Ki-67^+^ Tregs in cohort 5 was significantly increased compared to cohort 0 at week 4 (Supplementary Fig. 2a, left panels) and a similar trend (not significant) was observed in cohort 6. For the CD4^+^ T cells, a significantly lower percentage of cells expressed Ki-67 at week 2 in cohort 2 as compared to week 0 (*p* \< 0.001). Although not significant, an increase in CD4^+^ T cells expressing Ki-67 was seen at time point 4 for cohort 2 and subsequent cohorts showed a similar increase at time point 4, with a significant effect in cohort 6.
The results of the expansion cohort 2E (Supplementary Fig. 2b) were similar to cohort 2, however, the Ki-67^+^ Tregs in the expansion cohort first showed a significant decrease at week 2, followed by a significant increase at week 4. Furthermore, the increase in Ki-67^+^ Treg cells at time point 4 was more abundant and significantly different compared to cohort 0. Although an increase in Ki-67 expression in CD4^+^ T cells was also observed, it failed to reach statistical significance.
CTX results in a decrease in the frequency of monocytic MDSC {#Sec9}
------------------------------------------------------------
As MDSC are key players in immune suppression in the tumor microenvironment as well as systemically, being able to contribute to tumor progression and metastasis \[[@CR18]\], the percentages of monocytic MDSC (mMDSC, defined as Lin^−^CD14^+^HLA-DR^−^) in peripheral blood were determined in all cohorts. As shown in Fig. [2](#Fig2){ref-type="fig"}a, treatment with everolimus alone resulted in a non-significant increase in mMDSC. With the exception of cohort 3, addition of CTX resulted in a decrease in the frequency of mMDSC. In cohort 2, this decrease relative to levels in cohort 0 reached statistical significance (*p* \< 0.01) at time point 4 weeks. In addition, within cohort 2, a significant difference between mMDSC percentages at baseline versus time point 4 was observed (*p* \< 0.05). Although less pronounced, the results of the expansion cohort confirmed the earlier observed decrease in the frequency of mMDSC, with a significant difference between the expansion cohort and cohort 0 at time point 4 (*p* \< 0.05, Fig. [2](#Fig2){ref-type="fig"}b).
Fig. 2Effect of different dosages and administration schedules of CTX when combined with a fixed dose of 10 mg everolimus on the frequency of mMDSC. **a** Relative percentages of mMDSC (to start) defined as Lin^−^CD14^+^HLA-DR^−^ are shown for the six investigated CTX cohorts (black bullets, black line), compared to cohort 0 (open bullet, dotted line). **b** Relative percentages of mMDSC are shown for the expansion cohort. Means ± SEM are shown
Addition of CTX reverses the effects of everolimus on blood DC subsets {#Sec10}
----------------------------------------------------------------------
To assess the effects of the combination of everolimus and CTX on blood DC subsets, the percentages and activation status of three blood DC subsets were determined, i.e., conventional DC1 (cDC1, defined as BDCA3^+^CD14^−^CD11c^+^), cDC2 (defined as BDCA1^+^CD19^−^CD14^−^CD11c^+^), and plasmacytoid DC (pDC, defined as BDCA2^+^CD123^+^) \[[@CR19]\]. The activation status of these subsets was determined by MFI measurement of CD86 (and in addition CD40, data not shown). In cohort 0, a significant decrease in the frequency of cDC1 at time point 2 and 4 weeks, and of the cDC2 subset at time point 2 weeks was noted. Addition of CTX diminished these effects (Fig. [3](#Fig3){ref-type="fig"}a). Interestingly, an actual increase in both cDC1 and cDC2 percentages was most pronounced with increasing doses of CTX. Addition of CTX to everolimus also resulted in an increase in the frequency of pDC, which reached statistical significance at week 4 in both cohort 2 and cohort 5. The expansion cohort (Fig. [3](#Fig3){ref-type="fig"}b) confirmed the changes previously noted in patients treated in cohort 2.
Fig. 3Effect of different dosages and administration schedules of CTX when combined with a fixed dose of 10 mg everolimus on the frequency of three blood DC subsets. **a** Relative percentages of cDC1 (BDCA3^+^CD14^−^CD11c^+^), cDC2 (BDCA1^+^CD19^−^CD14^−^CD11c^+^), and pDC (BDCA2^+^CD123^+^) are shown for the six investigated CTX cohorts (black bullets, black line), compared to cohort 0 (open bullet, dotted line), relative to start. **b** Relative percentages of the three subsets are shown for the expansion cohort. Means ± SEM are shown
Treatment with everolimus alone resulted in a significant decrease in the expression of CD86 on the cDC1 and pDC subsets, at week 4 and 2, respectively, both with a *p* ≤ 0.05. As shown in Fig. [4](#Fig4){ref-type="fig"}a, CTX was capable of reversing this downregulation of CD86 expression on cDC1 reaching statistical significance in all CTX cohorts except cohort 3. While CTX did not result in a significant alteration of CD86 expression on cDC2, an increase in the expression of CD86 on pDC was observed in cohorts 3, 5 and 6 when compared to the patient group treated with everolimus alone (i.e., cohort 0).
Fig. 4Effect of different dosages and administration schedules of CTX when combined with a fixed dose of 10 mg everolimus on the activation status of three blood DC subsets. **a** Relative MFI of CD86 was determined for the cDC1, cDC2, and pDC subset and shown for the six investigated CTX cohorts (black bullets, black line), compared to cohort 0 (open bullet, dotted line). **b** Relative percentages of the MFI of CD86 on cDC1, cDC2, and pDC shown for the expansion cohort. Means ± SEM are shown
Figure [4](#Fig4){ref-type="fig"}b shows the changes in MFI of CD86 on the three blood DC subsets in the expansion cohort, again confirming the results previously seen in patients treated in cohort 2.
Addition of CTX does not result in enhanced NK cell frequencies {#Sec11}
---------------------------------------------------------------
A previously published article reported beneficial effects of low-dose metronomic CTX on NK cell function \[[@CR15]\]. Two distinct NK cell subsets were monitored, the immunoregulatory CD56^bright^CD16^dim/−^ and the cytotoxic CD56^dim^CD16^+^ subset. As shown in Fig. [5](#Fig5){ref-type="fig"}a, 4 weeks of treatment with everolimus alone resulted in a significant reduction in CD56^bright^CD16^dim/−^ NK cells (Fig. [5](#Fig5){ref-type="fig"}a, left panels). Addition of CTX resulted in a dose-dependent reversal of this effect with higher doses of CTX actually inducing an increase in the frequency of immunoregulatory NK cells. While treatment with everolimus resulted in a temporary (non-significant) increase in the frequency of CD56^dim^CD16^+^ cytotoxic NK cells, this effect was attenuated by the addition of CTX and actually resulted in a significant decrease in CD56^dim^CD16^+^ NK cells in cohorts 4 and 6 with effects being most striking in cohort 6. Again, the expansion cohort confirmed the earlier observed data in patients treated in cohort 2 (Fig. [5](#Fig5){ref-type="fig"}b).
Fig. 5Effect of different dosages and administration schedules of CTX when combined with a fixed dose of 10 mg everolimus on the frequency of NK cells. **a** Relative percentages of immunoregulatory (CD56^bright^CD16^dim/−^) and cytotoxic (CD56^dim^CD16^+^) NK cells are shown for the six investigated CTX cohorts (black bullets, black line), compared to cohort 0 (open bullet, dotted line). **b** Relative percentages of the NK subsets are shown for the expansion cohort. Means ± SEM are shown
Overall effect of the addition of continuous once daily oral administration of 50 mg of cyclophosphamide on immune cell populations in patients with mRCC treated with everolimus {#Sec12}
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Figure [6](#Fig6){ref-type="fig"} shows the data from patients treated with 10 mg once daily everolimus (i.e., cohort 0) versus all patients treated with the combination of 10 mg once daily everolimus and 50 mg cyclophosphamide in a continuous scheme (i.e., cohort 2 and the expansion cohort, together designated "combined cohort 2"). In this combined cohort, a significant decrease in Tregs was observed and accompanied by an increase in the frequency of CD8^+^ T cells. Together, this resulted in a significant increase in the CD8^+^ T cell:Treg ratio (Fig. [6](#Fig6){ref-type="fig"}a). Interestingly, the combined cohort also showed a significant decrease in Ki-67^+^ Tregs at week 2, followed by a significant increase at week 4 compared to week 0. Additionally, a significant difference between the combined cohort and cohort 0 was found at week 4. The Ki-67 expression in CD4^+^ T cells decreased significantly at week 2 and although the Ki-67^+^CD4^+^ T-cell percentages were higher in the combined cohort 2 compared to cohort 0, this difference was not significant (Fig. [6](#Fig6){ref-type="fig"}b). Notable was also the observation that whereas mMDSC frequencies increased in patients treated with everolimus alone, the frequency significantly decreased in the combined cohort 2 both at time point 2 as well as 4 weeks (Fig. [6](#Fig6){ref-type="fig"}c) with a significant difference between the percentages at time point 4 when comparing the combined cohort to cohort 0. The decrease in cDC1 and cDC2 percentages that was observed with everolimus monotherapy was reversed by adding CTX to the treatment, with a significant difference when comparing to cohort 0 at time point 4 (Fig. [6](#Fig6){ref-type="fig"}d). In addition, 50 mg of CTX once daily continuously in combination with everolimus resulted in a reversal of the effect on the immunoregulatory NK cells, however, a decrease in cytotoxic NK cells was observed at time point 2 (*p* \< 0.05, Fig. [6](#Fig6){ref-type="fig"}e). No significant differences between the combined cohort 2 and cohort 0 were observed for the T-cell activation markers, although there was a notable increase in frequency of CD4^+^CTLA-4^+^ T cells by week 4 (Fig. [6](#Fig6){ref-type="fig"}f).
Fig. 6Overview of changes in immune cell subsets in cohort 0 compared to the combined cohort 2 (i.e., cohort 2 and cohort 2E). Patients were treated with 50 mg CTX once daily, combined with 10 mg everolimus once daily. **a** Tregs within CD4^+^ T cells, CD8^+^ T cells within CD3^+^ T cells and the ratio of CD8^+^ T cells versus Tregs are shown. **b** Ki-67 expression in Tregs and CD4^+^ T cells. **c** mMDSC. **d** Blood DC subsets. **e** NK cell subsets. **f** PD-1 and CTLA-4 expression on CD4^+^ and CD8^+^ T cells. Panels show the combined cohort 2 (black bullets, black line) versus cohort 0 (open bullet, dotted line). Means ± SEM are shown
Discussion {#Sec13}
==========
This is the first clinical trial in which several dosages and schedules of metronomic cyclophosphamide in combination with the standard dosage of everolimus were investigated, and where extensive and comprehensive immune monitoring was performed. Our data indicate that while the frequency of Tregs slowly increased during treatment with everolimus alone, the combination of 10 mg everolimus once daily and 50 mg CTX once daily continuously, resulted in a significant decrease of Tregs within 2 weeks of treatment. This decrease persisted up to week 4 returning to baseline levels after 8 weeks of combination treatment. The slight increase in Treg percentages that was observed at 4 and 8 weeks of treatment suggest the reduction in Tregs to be a temporary effect, as was previously also reported in advanced-stage breast cancer patients treated with single-agent 50 mg CTX p.o. daily \[[@CR20]\] and which is in line with the observation of an increase in the expression of the proliferation marker Ki-67 in Tregs at time points 4 and 8 weeks. Though this may seem to limit the rationale for combination treatment of CTX and everolimus, Ge et al. \[[@CR20]\] reported that transient depletion of Tregs can increase tumor-reactive T-cell numbers implying that even a temporary Treg depletion may sufficiently boost the anti-tumor immune response, by creating a window for T-cell priming against the tumor. In addition, in the combined cohort 2 CD8^+^ T-cell percentages significantly increased and together with the decrease in Treg percentages resulted in an increase in the effector to suppressor ratio (CD8^+^ T cell:Tregs). Since an increased effector to suppressor ratio is associated with improved survival \[[@CR21]--[@CR23]\], this may imply positive effects on the survival of mRCC patients when treated with a combination of everolimus and CTX. Of note, the mechanism behind these reduced cell amounts, e.g., by necrosis or apoptosis, has not been examined.
The combined cohort 2 analysis also revealed a significant decrease in the frequency of mMDSC after 2 and 4 weeks of treatment. As the role of MDSC in the tumor environment is diverse \[[@CR18]\] leading to promotion of tumor growth, this mMDSC-depleting effect, though temporary in nature as the decrease in mMDSC did not persist after 4 weeks of combination treatment, could further contribute to improved survival. Of interest, sunitinib has also been reported to decrease the frequency of myeloid suppressor cells \[[@CR24]\]. As lenvatinib is not only a TKI directed against the VEGF receptor \[[@CR25]\] but additionally inhibits fibroblast growth factor receptor (FGFR) which may also dampen MDSC activity \[[@CR26]\], the combination of lenvatinib and everolimus might exert similar or even more pronounced effects on the immune system as the combination of everolimus and CTX.
Previously, we reported that treatment with everolimus alone significantly reduced the frequency of the cDC1 and cDC2 blood DC subsets, while it did not affect the frequency of pDC \[[@CR27]\]. Interestingly, we here demonstrate that the addition of any dosage or scheme of CTX could reverse these everolimus-induced alterations in the frequency of cDC1 and cDC2. Furthermore, CTX was also able to increase the activation of at least two of the three blood DC subsets, the cDC1 and pDC subset. While cohort 2 already showed beneficial effects on the blood DC subsets, effects were even more pronounced when higher doses of CTX were used.
As Ghiringhelli et al. \[[@CR15]\] reported beneficial effects of CTX treatment on NK and T cell effector functions, we were interested in the effects of the combination of everolimus and CTX on both cell subsets. We found that adding CTX to everolimus could reverse the effects of everolimus monotherapy on the immunoregulatory CD56^bright^CD16^dim/−^ NK cell subset as well as on the cytotoxic CD56^dim^CD16^+^ NK cell population, overall resulting in an increase in the frequency of immunoregulatory NK cells and a decrease in the frequency of cytotoxic NK cells with combination therapy. For both NK cell subsets, functional analyses were not performed, and therefore it remains impossible to determine whether these subsets are also functionally compromised as previously reported \[[@CR28]\]. For both the CD4^+^ and CD8^+^ T cell subset the expression of PD-1 and CTLA-4 was determined. Though not significant, a minor increase in expression of PD-1 and CTLA-4 could be noted, perhaps suggesting an increase in tumor-specific effector T cells and a window for combination therapy with immune checkpoint inhibitors targeting CTLA-4 and/or PD-1 \[[@CR29]\].
In conclusion, we performed a phase 1 study in patients with mRCC treated with everolimus alone and the combination of everolimus and different doses and administration schedules of CTX and here report on the comprehensive immunomonitoring that was performed in these patients. The predefined goal of the study, i.e., to identify the dose and schedule of CTX that when combined with everolimus would result in optimal and selective depletion of Tregs, was achieved with a once daily continuous oral dose of 50 mg CTX. Addition of this dose of CTX to everolimus, resulted in depletion of Tregs, a sustained increase in CD8^+^ T cells with an increase in the effector to suppressor ratio. Furthermore, this combination therapy resulted in a depletion of mMDSC, while negative effects of monotherapy with everolimus on blood DC subsets were counteracted. All together, these observed changes in various immune cell populations may result in increased antitumor immunity and improved survival of patients with mRCC, which is currently further investigated in a phase 2 clinical trial \[[@CR16]\].
Electronic supplementary material
=================================
{#Sec14}
Below is the link to the electronic supplementary material.
Supplementary material 1 (PDF 2141 KB)
APC
: Allophycocyanin
CCMO
: Committee on Research Involving Human Subjects
cDC
: Conventional DC
CTX
: Cyclophosphamide
FGFR
: Fibroblast growth factor receptor
ICH
: International Conference on Harmonization
mTOR
: Mammalian target of rapamycin
mRCC
: Metastatic RCC
mMDSC
: Monocytic MDSC
pDC
: Plasmacytoid DC
PFS
: Progression-free survival
Tregs
: Regulatory T cells
RCC
: Renal cell carcinoma
WIN-O
: The Netherlands Working Group on Immunotherapy of Oncology
CMH, TDG and HJV analyzed all the data and prepared the manuscript. SML, ZB contributed to data acquisition. CMH, PH, MT, JBH, and HV contributed to patient data acquisition. All authors read and approved the final manuscript.
The trial was supported by the Dutch Cancer Society (Grant number: VU 2011--5144) and partly funded by a grant from Novartis Oncology Netherlands. Novartis has had no part in study design, data collection, analysis, interpretation, the writing of the manuscript, or the decision to submit for publication.
Conflict of interest {#FPar1}
====================
The authors declare that they have no conflict of interest.
Ethical approval and ethical standards {#FPar2}
======================================
ClinicalTrials.gov Identifier NCT01462214, Netherlands Trial Register number NTR3085. The study was conducted in accordance with the Declaration of Helsinki and consistent with International Conference on Harmonization (ICH) Guidelines for Good Clinical Practice. The Medical Ethical Committee of the VU University Medical Center, Amsterdam, the Netherlands and the Central Committee on Research Involving Human Subjects (CCMO) approved the study protocol.
Informed consent {#FPar3}
================
All patients gave written informed consent.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-jcm-09-00077}
===============
Adult sepsis, defined as the dysregulated host response to infection resulting in organ dysfunction, is a major global health problem and represents a challenge for physicians and health care systems all over the world due to its high incidence, mortality and economic impact \[[@B1-jcm-09-00077],[@B2-jcm-09-00077],[@B3-jcm-09-00077],[@B4-jcm-09-00077]\]. Additionally, survivors show a substantially worse quality of life, physical and cognitive decline and an increased risk of death several years after an episode \[[@B3-jcm-09-00077],[@B5-jcm-09-00077]\].
The negative impact of sepsis, including a shortened lifespan and increased burden of morbidity and disability, can be especially significant in the case of young adults under 44 years of age as it is the population-group in which the majority of individuals are at their maximum physiological, biological and cognitive-intellectual capacity and supposes the active demographic, labor and economic force of the society.
There is limited population-based information on the epidemiology and trends of sepsis in these adults, however, especially in Europe where data from 2015 reveal that adults aged 20--44 years account for 34% of the population \[[@B6-jcm-09-00077]\]. The size of this population and the results of several studies indicating that the incidence of sepsis has increased in the general population \[[@B4-jcm-09-00077],[@B7-jcm-09-00077],[@B8-jcm-09-00077]\], make that information essential to assess, considering the burden of the disease and to estimate healthcare resource requirements. Thus, this population-based study was designed to address the epidemiological characteristics and trends of sepsis in young adults in Spain.
2. Materials and Methods {#sec2-jcm-09-00077}
========================
The data for this study were compiled from the Spanish Health Ministry's National Minimum Basic Data Set (MBDS). Regarding the Spanish National Health System, when a patient is discharged from hospital, the responsible physician is required by law to record all diagnoses and clinical procedures performed according to the ICD-9-CM system. This information is then subjected to a validation process and consolidated in the MBDS. This database is considered to be representative of the national population as it includes data on over 90% of all hospitalizations produced annually in the country \[[@B9-jcm-09-00077]\].
Viewing this database, each hospitalization is treated as a single registry and includes information on patient demographics, date of admission and whether elective or non-elective, date and destiny of discharge along with diagnostic codes including the main diagnosis, 13 additional diagnoses, and up to 20 procedures conducted during the patient's hospital stay.
2.1. Case Identification {#sec2dot1-jcm-09-00077}
------------------------
All hospitalizations of adults aged 20--44 years produced over a 10-year period from January 1, 2006 to December 31, 2015 were identified. To capture sepsis, we used an ICD-9-CM widely used strategy to define infection and organ dysfunction. To define infection, the following codes were used \[[@B7-jcm-09-00077],[@B8-jcm-09-00077],[@B10-jcm-09-00077]\]: 038.0 (streptococcal septicaemia), 038.1 (staphylococcal septicaemia), 038.2 (pneumococcal septicaemia), 038.3 (septicaemia due to anaerobes), 038.4 (septicaemia due to other Gram-negative organisms), 038.8 (other specified septicaemias), 038.9 (unspecified septicaemia), 003.1 (salmonella septicaemia); 020.2 (septicaemia plague); 036.2 (meningococcal septicaemia); 036.3 (Waterhouse--Friderichsen syndrome); 054.5 (herpetic septicaemia); 098.89 (gonococcemia); 112.5 (systemic candidiasis); 112.81 (candidal endocarditis); 117.9 (other and unspecified mycoses) and 790.7 (bacteraemia). We also included the ICD-9-CM codes 995.91 (sepsis, systemic inflammatory response syndrome due to infectious process without organ dysfunction) and 995.92 (severe sepsis) \[[@B10-jcm-09-00077]\].
To identify acute organ dysfunction, the ICD-9-CM codes detailed below were captured: Respiratory: 518.81 (acute respiratory failure), 518.82 (other pulmonary insufficiency), 518.84 (acute-on-chronic respiratory failure), 518.85 (acute respiratory distress syndrome after shock or trauma), 786.09 (respiratory distress, insufficiency), 799.1 (respiratory arrest), and 96.7 (invasive mechanical ventilation); Cardiovascular: 785.5 with all subcodes (shock without mention of trauma, includes 785.51, 785.52, 785.59); renal: 584 with all subcodes (acute renal failure), 580 (acute glomerulonephritis), and 39.95 (hemodialysis); Hepatic: 570 (acute and subacute necrosis of liver), 572.2 (hepatic coma), and 573.3 (hepatitis, unspecified); Haematologic: 286.6 (defibrination syndrome), 286.9 (other and unspecified coagulation defects), and 287.3--5 (secondary thrombocytopenia, unspecified); Neurologic: 293 (acute delirium), 348.1 (anoxic brain damage), 348.3 (encephalopathy, unspecified), 357.82 (critical illness polyneuropathy), 780.01 (coma), 780.09 (drowsiness, unconsciousness, stupor), and 89.14 (electroencephalogram); and Metabolic: 276.2 (acidosis metabolic or lactic).
Comorbidity was defined using the Charlson's index adapted and validated by Deyo for the ICD-9-CM \[[@B11-jcm-09-00077]\]. This index includes specific comorbid conditions of known prognostic value and has been shown effective to assess the mortality risk in patients with sepsis \[[@B12-jcm-09-00077]\]. Further, due to the impact of comorbidity on the outcome of sepsis \[[@B13-jcm-09-00077]\], cases were stratified according to the presence or not of comorbidities as defined by the Charlson's index.
Concerning every case, the main diagnostic group at hospital admission was assessed according to the ICD-9-CM chapters: infectious disease (001--139); neoplasms (140--239); endocrine diseases (240--279); hematological diseases (280--289); neurological diseases (320--389); diseases of the circulatory system (390--459); respiratory diseases (460--519); diseases of the digestive system (520--579); diseases of the genitourinary tract (580--629); diseases of the skin and subcutaneous tissue (680--709); diseases of the musculoskeletal system and connective tissue (713--739) and injury-poisoning (800--999).
Episodes were described as surgical when this was indicated by their GRD (diagnosis-related group code). To identify specific microorganisms, code 041 was included as indicated by the ICD-9-CM coding manual for the purpose of identifying specific microorganisms in the case of diseases classified under the heading 'other' \[[@B9-jcm-09-00077]\]. Hospitalization costs were estimated from their respective GRDs.
As the data examined were sourced from anonymized records, there was no need for informed consent \[[@B14-jcm-09-00077]\].
2.2. Data Analysis and Presentation {#sec2dot2-jcm-09-00077}
-----------------------------------
During this descriptive study, the main variables examined were the incidence and mortality of sepsis and their trends over the study period. Other variables examined were demographics, comorbidities, microbiological data, number and type of organ dysfunction, and the use of hospital resources (length of hospital stay and costs).
Categorical variables are provided as their absolute frequencies and percentages, and continuous variables as means and standard deviations. Age groups were defined as 5-year intervals. In-hospital case-fatality rate (CFR) was calculated as the number of deaths divided by the number of cases of sepsis and expressed as a percentage. To identify factors associated with in-hospital mortality, an exploratory logistic regression was performed including variables of clinical relevance such as sex, age group and comorbidity burden, and others of known impact on the outcome of sepsis such as number of organ dysfunctions. Results are presented as odds ratios (OR) and 95% confidence intervals (CI).
Crude incidence rates were calculated using national population data, of those aged 20--44 years, from the Spanish institute of statistics (INE) and results expressed per 100,000 population \[[@B15-jcm-09-00077]\]. Age-adjusted rates were calculated by direct standardization referred to the European population \[[@B16-jcm-09-00077]\].
To assess possible temporal changes in incidence and in-hospital CFR, we used Joinpoint regression models \[[@B17-jcm-09-00077]\]. These generalized linear models follow a Poisson distribution and served to estimate annual average percentage change (AAPC) and its 95% CI for each trend \[[@B18-jcm-09-00077]\].
All statistical tests were performed using the package Stata^®^ version 15 (StataCorp LP, College Station, Texas, USA) and Joinpoint regression programme version 4.7.0.0 (National Cancer Institute, Bethesda, MD, USA). Significance was set at *p* \< 0.05.
2.3. Data Availability {#sec2dot3-jcm-09-00077}
----------------------
The data came from anonymized registries. According to the confidentiality agreement signed with the Ministry of Health, Consumer Affairs and Social Welfare, the data from this study cannot be shared with third parties. The authors did not have special access privileges. Should any researcher wish to gain access to these data, they can do so by applying directly to the Ministry through the following link: <https://www.mscbs.gob.es/estadEstudios/estadisticas/estadisticas/estMinisterio/SolicitudCMBDdocs/Formulario_Peticion_Datos_CMBD.pdf>.
3. Results {#sec3-jcm-09-00077}
==========
Regarding 9,271,272 hospital discharge registries corresponding to adults aged between 20 and 44 years for the period 2006--2015 in Spain, 28,351 episodes of sepsis were identified, representing 3.06‰ of all-cause hospitalizations in this age group.
The crude incidence rate for the cohort was 16.36 cases per 100,000 persons; being higher in men (18.62 per 100,000 persons) than women (13.9 per 100,000 persons). Shown in [Figure 1](#jcm-09-00077-f001){ref-type="fig"}, incidence increased with age in cases with and without comorbidities, but the increase was greater among those with comorbidities. The increase with age was observed both in men and women, although in almost every age group the incidence was higher among men.
The overall mean age was 36 years and 58% of cases were men. Roughly 41% of cases (38.2% in men, 46% in women) had a Charlson score of 0. The most frequent comorbidities in the remaining cases were liver disease, cancer and AIDS ([Table 1](#jcm-09-00077-t001){ref-type="table"}). This table also shows the demographic and clinical characteristics of the episodes stratified according to the presence or absence of comorbidities. Regarding most cases, hospital admission was non-elective via the emergency services and, in one third of cases, the cause of hospitalization was infection.
The potential source of infection in most cases was respiratory, followed by genitourinary and procedure-related infections. Found in close to 60% of cases, at least one microorganism was identified. Gram-negative bacteria were slightly more frequent than other microorganisms. The presence of bacteraemia was recorded in 21% (*n* = 5908) of the episodes, this rate being higher in those with comorbidities (24% versus 16.5%).
Seen in around 45% of cases, single organ dysfunction was present, while 26% and 25% showed the dysfunction of two or more organs, respectively. Shown in [Table 1](#jcm-09-00077-t001){ref-type="table"}, the percentage of cases with more than two organ dysfunctions was greater among those with comorbidities. The most frequently affected organs were the lungs, recorded in more than half of the cases, followed by the cardiovascular system and kidneys. Renal dysfunction was much more frequent in the subset with comorbidities. Essentially, of the 3191 cases in which patients underwent dialysis, 2444 (77%) belonged to this subset. The use of invasive mechanical ventilation was, nevertheless, more frequent in the group without comorbidities and, overall, this measure was employed in 38% of cases.
The mean hospital stay was 28.4 days and the mean cost per case was 17,878 Euros. There were no significant differences between the cases with or without comorbidities.
Six thousand, eight hundred and eight hospital deaths were recorded, which corresponds to an overall case-fatality rate CFR of 24%, but this rate varied according to various demographic and clinical characteristics. Thus, as shown in [Table 2](#jcm-09-00077-t002){ref-type="table"}, it was higher in men than in women and clearly increased with age.
The multivariate logistic regression analysis revealed that a greater age, the failure to detect the source of infection, the non-identification of the responsible microorganism, the extent of organ dysfunction and the presence of comorbidities were significantly associated with an increased risk of mortality ([Table 2](#jcm-09-00077-t002){ref-type="table"}). When we perform the same analysis stratified by the presence or absence of comorbidities, as indicated by Charlson's index, it is observed that, except for sex, mortality risk factors are the same with small differences in the odds ratios ORs ([Table 3](#jcm-09-00077-t003){ref-type="table"}).
3.1. Trends {#sec3dot1-jcm-09-00077}
-----------
### 3.1.1. Incidence {#sec3dot1dot1-jcm-09-00077}
Regarding the whole population analyzed, and as shown in [Table 4](#jcm-09-00077-t004){ref-type="table"}, the incidence of sepsis has discretely increased over the six years of the study, though not significantly. When adjusted by sex, there was no significant change in men, however, a significant increase was detected in the incidence rate in women.
Rates also increased significantly in cases with a Charlson's index of zero but not in those with comorbidities. Additionally, our data show a decreasing rate over time of cases with one or two organ dysfunctions.
### 3.1.2. In-Hospital Mortality {#sec3dot1dot2-jcm-09-00077}
The temporal analysis indicated a significant drop in case-fatality rate CFR both overall and in each of the subsets examined ([Table 4](#jcm-09-00077-t004){ref-type="table"}), however, the decrease produced was variable and lower in men than women. Similarly, [Table 4](#jcm-09-00077-t004){ref-type="table"} reveals that the fall was significant both in cases with a Charlson's index of zero, as in those with comorbidities, although it was greater in the former. When we assessed the changes produced in CFR by number of organ dysfunctions, this variable was observed to fall in all cases, although the decrease was lower in the case of the dysfunction of more than two organs.
4. Discussion {#sec4-jcm-09-00077}
=============
This is the first study to provide representative national estimates of epidemiological characteristics and incidence and mortality trends of sepsis in young Spanish adults, to our knowledge. Our findings reveal that sepsis is a common and frequently fatal condition among adults aged 20--44 years. Trends indicate that incidence rates are increasing in women but not in men, whereas in-hospital mortality is decreasing. Additionally, sepsis associates with a substantial use of hospital resources.
The incidence of sepsis observed was 16.4 cases per 100,000 persons aged 20--44 years. Our data confirmed that, even in this young age group, its frequency was defined by patient age \[[@B19-jcm-09-00077]\] and was very much higher in individuals aged 40--44 years than in those aged between 20 and 24 years. Sepsis also more frequently affects men than women \[[@B4-jcm-09-00077],[@B7-jcm-09-00077]\], something that literature has linked to a distinct immune response between men and women, suggesting an advantageous response from women to an infection \[[@B19-jcm-09-00077]\]. However, in contrast with the findings of others who report an overall increase in incidence rates of sepsis in the general population \[[@B4-jcm-09-00077],[@B7-jcm-09-00077],[@B8-jcm-09-00077]\], in our study only women showed a significant increase over the 10-year study period. Regrettably, the characteristics of our dataset do not permit us to identify the causes of this particular increase in women, nor if it may be related to differences in individual risk patterns or care-level determinants, but we feel that this finding merits further inquiry.
Among the most relevant findings of our study was the high in-hospital mortality associated with sepsis in the young adult, which amounted to 24% of all episodes. Mortality was higher in men than women and also shows a clear association with age and, especially, with the presence of comorbidities and the number of organ dysfunctions. Accordingly, age acts as an independent risk factor of mortality \[[@B20-jcm-09-00077]\] even in the young adult and, as our data show, its impact is aggravated by the presence of comorbidities.
Though the capacity of comorbidities to affect the risk of sepsis remains unclear \[[@B21-jcm-09-00077]\], our observation that around 60% of the cases in our cohort showed comorbidities with the predominance of liver disease, cancer and AIDS, confirms the results of others who suggest that chronic diseases, specifically those mentioned here, increase this risk \[[@B13-jcm-09-00077],[@B20-jcm-09-00077],[@B21-jcm-09-00077]\]. Besides a high proportion of men and a difference in age, which was greater among cases with comorbidities, the most appreciable differences between those with and without comorbidities was a greater extent of organ dysfunction along with a greater frequency of renal dysfunction and a higher mortality in the former. This corroborates the fact that the presence of comorbidities promotes the development of multiple organ dysfunction \[[@B22-jcm-09-00077]\] and has a great influence on the outcome of patients with sepsis \[[@B13-jcm-09-00077],[@B23-jcm-09-00077]\]. Found in our study, this factor increased the risk of death by 2.82-fold.
We should highlight, however, the elevated in-hospital mortality observed for cases without comorbidities. This was 13.5% and much higher than that associated with other diseases in the 20--44 years age group in our country \[[@B9-jcm-09-00077]\]. Further, we must also highlight that the proportion of cases without comorbidities has increased over time. Although we cannot establish a formal cause for this rise, it may be related to the increasing incidence in women who constitute a group with less comorbidity burden as measured by the Charlson Index.
Additionally, it was especially noticeable that both in cases with or without comorbidities, the presence of two or more organ dysfunctions was a critical event with a cumulative impact on the risk of death in the septic young adult \[[@B24-jcm-09-00077]\].
The mortality observed here doubles the 12% described by Kaukonen \[[@B25-jcm-09-00077]\], who analyzed the data of 15,471 cases of severe sepsis in patients aged ≤44 years in a retrospective observational study performed in the intensive care units (ICUs) of Australia and New Zealand between 2000 and 2012. This discrepancy could be explained in part by differences in information systems, methodology and setting. Although Kaukonen analyzed cases from ICUs and we looked at a national-level population, notable clinico-demographic differences were observed with a profile that suggests greater severity and mortality risk in our study, as our cases were older and showed greater pulmonary, cardiovascular and renal organ dysfunction and a greater comorbidity burden. Regarding this last issue, in the absence of comorbidities, the mortality observed in the present study is greater than the 8% described by Kaukonen, although the systems used to assess comorbidities differed between the two studies.
Both studies coincided, however, in observing a significant decline in hospital mortality rates of sepsis in patients aged ≤44 years. Considering our study, we observed from 2006 to 2015 an annual reduction in hospital mortality of 5.9%, a similar figure to that noted by Kaukonen. This decrease in mortality was observed both in men and women but was less marked in cases with comorbidities and in those showing two or more organ dysfunctions.
Concerning resource utilization, both mean hospital stay and costs were elevated and much above the mean figures provided by the National Health Service for all-cause hospitalizations in this age group. Over the period analysed, these figures were 4.9 days of mean stay and 3219.26 Euros per hospitalization \[[@B9-jcm-09-00077]\]. We could easily speculate that this can be related to the use of costly invasive therapeutic interventions in sepsis \[[@B26-jcm-09-00077],[@B27-jcm-09-00077]\].
While sepsis is often associated with elderly subjects, this study indicates that it is frequent and related to a high in-hospital mortality in young adults. According to data from the UN, adults aged 20--44 years in 2015 accounted for 33.7% of the Spanish population \[[@B8-jcm-09-00077]\]. This means sepsis in this age group is a real challenge for clinicians and healthcare systems. Our findings identify a need to implement measures designed to optimize its prevention and management, including early diagnostic and therapeutic measures, especially in high-risk patients with comorbidities, targeted at reducing the development and progression of organ dysfunction and, consequently, mortality. Essentially, as comorbidities are usually easily recognizable chronic conditions, if these are managed early on and adequately, it could be possible to improve outcomes in these patients \[[@B28-jcm-09-00077]\].
Clinical-administrative databases are an essential tool for health research \[[@B16-jcm-09-00077],[@B29-jcm-09-00077]\]. Spain has a large population database which is required by law and is widely representative as it covers practically all hospitalizations produced annually, allowing for accurate epidemiological estimates. During this study, we followed RECORD guidelines for observational studies' routinely-collected health data \[[@B30-jcm-09-00077]\]. Notwithstanding, due to its inherent characteristics, the database employed has some limitations. Although national guidelines exist for the use of the coding system, this may not be uniform across the national healthcare system and we cannot rule out coding errors, despite regular audits; however, we consider systematic coding errors highly unlikely. Moreover, due to its confidential nature, the database used lacks complete clinical information, precluding any causal inferences. The use of national databases is, nevertheless, well established for the epidemiological monitoring of the incidence and mortality of sepsis \[[@B31-jcm-09-00077],[@B32-jcm-09-00077]\]. Additionally, the results of a recent meta-analysis confirm the essential role of administrative data for surveillance of mortality trends in sepsis \[[@B33-jcm-09-00077]\]. Another limitation is that studies based on hospital discharge data do not include non-hospitalized cases of sepsis such that our estimates of incidence and mortality are conservative. Similarly, deaths produced after hospital discharge also will be missed \[[@B5-jcm-09-00077],[@B34-jcm-09-00077]\].
5. Conclusions {#sec5-jcm-09-00077}
==============
This population-based nationwide study shows that sepsis is a common, and frequently fatal, condition among adults aged 20--44 years. Data over time indicate that incidence rates are increasing in women but not in men and in cases without comorbidity, findings that need further research. Conversely, in-hospital mortality shows a decreasing trend. This trend was lower in cases with comorbidity and dysfunction of more than two organs. Additionally, sepsis associates with a substantial use of hospital resources.
We would like to thank the Sub-Directorate General for Healthcare Information and Innovation of the Spanish Ministry of Health, Consumer Affairs and Social Welfare for providing the data used in this study.
Conceptualization, C.B.; methodology, C.B. and T.L.-C.; software, T.L.-C.; validation, C.B. and T.L.-C.; formal analysis, T.L.-C.; investigation, C.B.; resources, C.B.; data curation, T.L.-C.; writing---original draft preparation, C.B.; writing---review and editing, C.B.; visualization, T.L.-C.; funding acquisition, C.B. All authors have read and agreed to the published version of the manuscript.
This research was supported by the Carlos III Health Institute (grant number PI14/00081).
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
{#jcm-09-00077-f001}
jcm-09-00077-t001_Table 1
######
Characteristics and outcomes of sepsis.
-----------------------------------------------------------------------------------------------------------------------------------------
All Cases\ Without Comorbidities \*\ With Comorbidities \*\ *p*-Value
(*n* = 28,351) (*n* = 11,798) (*n* = 16,553)
------------------------------------------------------ ----------------- --------------------------- ------------------------ -----------
Age, mean ± SD 35.73 ± 6.63 34.59 ± 6.83 36.55 ± 6.37 \<0.001
Age group, years \<0.001
20--24 2381 (8.4) 1307 (11.1) 1074 (6.5)
25--29 3217 (11.4) 1619 (13.7) 1598 (9.7)
30--34 4971 (17.5) 2313 (19.6) 2658 (16.1)
35--39 7287 (25.7) 3054 (25.9) 4233 (25.6)
40--44 10,495 (37.0) 3505 (29.7) 6990 (42.4)
Sex \<0.001
Men 16,501 (58.2) 6299 (53.4) 10,202 (61.6)
Women 11,848 (41.8) 5498 (46.6) 6350 (38.4)
Non-Elective Hospital Admission 24,988 (88.1) 10,916 (92.5) 14,072 (85.0) \<0.001
Medical Admission 18,543 (65.5) 11,312 (68.4) 7231 (61.4) \<0.001
Hospital Admission Main Diagnosis (ICD-9-CM chapter)
Infectious disease 9164 (32.3) 3844 (32.6) 5320 (32.1) 0.432
Respiratory disease 3774 (13.3) 1874 (15.9) 1900 (11.5) \<0.001
Trauma & poisoning 3640 (12.8) 2066 (17.5) 1574 (9.5) \<0.001
Digestive disease 2938 (10.4) 1241 (10.5) 1697 (10.2) 0.467
Cancer 2137 (7.5) 94 (0.80) 2043 (12.3) \<0.001
Genitourinary disease 1560 (5.5) 874 (7.4) 686 (4.1) \<0.001
Circulatory disease 1372 (4.8) 186 (1.6) 1186 (7.2) \<0.001
Charlson Score, mean ± SD 1.77 ± 2.29 NA 3.03 ± 2.27
Specific Comorbidities ^‡^
Liver disease 5670 (20.0) 5670 (34.2)
Cancer 4320 (15.2) 4320 (26.1)
AIDs 2298 (8.1) 2298 (13.9)
Chronic kidney disease 2327 (8.2) 2327 (14.1)
COPD 1714 (6.1) 1714 (10.4)
Cardiac insufficiency 1588 (5.6) 1588 (9.6)
Diabetes 1646 (5.8) 1646 (9.9)
Hemiplegia or Paraplegia 1213 (4.3) 1213 (7.3)
Stroke 1052 (3.7) 1052 (6.4)
Rheumatologic disease 475 (1.7) 475 (2.9)
Peripheral vascular event 373 (1.3) 373 (2.2)
Acute myocardial infarction 234 (0.8) 234 (1.4)
Infection Site ^‡^
Respiratory 8414 (29.7) 3808 (32.3) 4606 (27.8) \<0.001
Genitourinary 4953 (17.5) 2439 (20.7) 2514 (15.2) \<0.001
Procedure-related 4352 (15.4) 1830 (15.5) 2522 (15.2) 0.526
Abdominal 3271 (11.5) 1512 (12.8) 1759 (10.6) \<0.001
Soft tissue 1265 (4.5) 549 (4.7) 716 (4.3) 0.188
Central nervous system 799 (2.8) 369 (3.1) 430 (2.6) 0.008
Cardiac 375 (1.3) 124 (1.1) 251 (1.5) 0.001
Other/Not specified 7112 (25.1) 2625 (22.3) 4487 (27.1) \<0.001
Microbiological Data ^‡^ 16,225 (57.2) 6568 (55.7) 9657 (58.3) \<0.001
Gram-positive bacteria 8340 (51.4) 3336 (50.8) 5004 (51.8)
Gram-negative bacteria 9457 (58.3) 3959 (60.3) 5498 (56.9)
Fungus 815 (5.0) 266 (4.1) 549 (5.7)
No. Organ Dysfunctions ^†^ \<0.001
1 12,764 (45.0) 5367 (45.5) 7397 (44.7)
2 7365 (26.0) 3274 (27.8) 4091 (24.7)
\>2 7095 (25.0) 2565 (21.7) 4530 (27.4)
Type of Dysfunction ^‡^
Respiratory 15,159 (53.4) 6560 (55.6) 8599 (51.9) \<0.001
Cardiovascular 13,178 (46.5) 5671 (48.1) 7507 (45.4) \<0.001
Renal 10,800 (38.1) 3713 (31.5) 7087 (42.8) \<0.001
Haematological 5652 (20.0) 2383 (20.2) 3269 (19.8) 0.350
Neurological 3145 (11.1) 1358 (11.5) 1787 (10.8) 0.059
Hepatic 2491 (8.8) none 2491 (15.1) \<0.001
Metabolic 2471 (8.7) 1025 (8.7) 1446 (8.7) 0.888
Invasive Therapeutic Measures
Mechanical ventilation 10,056 (35.5) 4481 (38.0) 5575 (33.7) \<0.001
Haemodyalisis 3191 (11.3) 747 (6.3) 2444 (14.8) \<0.001
In-Hospital Death 6808 (24.0) 1588 (13.5) 5220 (31.5) \<0.001
Hospital LOS, d, mean ± SD 28.44 ± 37.70 28.06 ± 37.90 28.72 ± 37.55 0.149
Hospitalization Cost, Euros mean ± SD 17,878 ± 22,348 17,888 ± 23,486 17,871 ± 21,501 0.949
-----------------------------------------------------------------------------------------------------------------------------------------
\* Comorbidity is defined by the Charlson Index. Cases with a Charlson Index of 0 are considered "without comorbidities". Cases with a Charlson index \>0 are considered "with comorbidities". Comparisons are made between cases without and with comorbidities (*p*-Value). Values in parentheses are percentages. SD: standard deviation. COPD: chronic obstructive pulmonary disease. LOS: length of hospital stay. ^‡^ Subgroups not mutually exclusive. d: days. ^†^ Not specified No. organ dysfunctions: 1127 cases (3.98%).
jcm-09-00077-t002_Table 2
######
In-hospital deaths. General characteristics, case-fatality and risk (*n* = 6808).
------------------------------------------------------------------------------------------------------------
Characteristic Cases Case-Fatality Rate\ Bivariate\ Multivariate\
(% Severe Sepsis) OR (95% CI) OR (95% CI), *p*-Value
----------------------------- ------- --------------------- ------------------- ----------------------------
Sex
Women 2455 20.72 Reference group Reference group
Men 4353 26.38 1.37 (1.30, 1.45) 1.12 (1.05, 1.19)
Age-Group (years)
20--24 405 17.01 Reference group Reference group
25--29 611 18.99 1.14 (0.99, 1.31) 1.10 (0.94, 1.28), 0.224
30--34 1050 21.12 1.31 (1.15, 1.48) 1.19 (1.03, 1.36), 0.016
35--39 1738 23.85 1.53 (1.36, 1.72) 1.28 (1.13, 1.47), \<0.001
40--44 3004 28.62 1.96 (1.74, 2.19) 1.48 (1.30, 1.68), \<0.001
Charlson Index Comorbidity
0 1588 13.46 Reference group Reference group
\>0 5220 31.54 2.96 (2.78, 3.15) 2.82 (2.63, 3.01), \<0.001
Diagnostic Categories
Medical 4457 24.04 Reference group Reference group
Surgical 2338 23.91 0.99 (0.94, 1.05) Not applicable
Pathogens Identified
No 3663 30.21 Reference group Reference group
Yes 3145 19.38 0.56 (0.53, 0.59) 0.68 (0.64, 0.72), \<0.001
Principal Site of Infection
Respiratory 2179 25.90 1.16 (1.09, 1.23) 1.15 (1.06, 1.25), 0.001
Abdominal 804 24.58 1.04 (0.95, 1.13) Not applicable
Genitourinary 555 11.21 0.35 (0.32, 0.38) 0.53 (0.48, 0.60), \<0.001
Soft tissue 223 17.63 0.67 (0.58, 0.77) 0.80 (0.67, 0.94), 0.007
Procedure-related 755 17.35 0.62 (0.57, 0.68) 0.80 (0.73, 0.89), \<0.001
Not specified 2483 34.91 2.10 (1.98, 2.23) 1.69 (1.55, 1.84), \<0.001
No. Organ Dysfunctions
1 1634 12.80 Reference group Reference group
2 1850 25.12 2.28 (2.12, 2.46) 2.27 (2.10, 2.45), \<0.001
\>2 3171 44.69 5.50 (5.13, 5.90) 5.06 (4.70, 5.45), \<0.001
------------------------------------------------------------------------------------------------------------
OR, odds ratio; 95% CI, 95% confidence interval.
jcm-09-00077-t003_Table 3
######
In-hospital mortality according to the presence or absence of comorbidities.
Characteristic Multivariate OR (95%CI), *p*-Value
----------------------------- ------------------------------------ ----------------------------
Sex
Women Reference group Reference group
Men 1.27 (1.13, 1.43), \<0.001 1.06 (0.98, 1.14), 0.146
Age-group (yrs)
20--24 Reference group Reference group
25--29 1.09 (0.85, 1.39), 0.508 1.11 (0.92, 1.35), 0.274
30--34 1.20 (0.95, 1.51), 0.118 1.19 (0.99, 1.41), 0.059
35--39 1.14 (0.92, 1.42), 0.228 1.35 (1.14, 1.60), \<0.001
40--44 1.44 (1.16, 1.77), 0.001 1.51 (1.28, 1.77), \<0.001
Pathogens identified
No Reference group Reference group
Yes 0.79 (0.71, 0.89), \<0.001 0.63 (0.59, 0.68), \<0.001
Principal site of infection
Respiratory 1.17 (1.01, 1.36), 0.044 1.14 (1.04, 1.25), 0.007
Abdominal Not applicable Not applicable
Genitourinary 0.55 (0.45, 0.68), \<0.001 0.53 (0.47, 0.60), \<0.001
Soft tissue 0.70 (0.51, 0.97), 0.030 0.82 (0.68, 0.99), 0.047
Procedure-related 1.01 (0.85, 1.21), 0.895 0.72 (0.64, 0.81), \<0.001
Others/Not specified 1.75 (1.47, 2.06), \<0.001 1.67 (1.51, 1.84), \<0.001
No. organ dysfunctions
1 Reference group Reference group
2 2.54 (2.19, 2.95), \<0.001 2.17 (1.98, 2.37), \<0.001
\>2 5.51 (4.78, 6.36), \<0.001 4.88 (4.48, 5.32), \<0.001
OR, odds ratio; 95% CI, 95% confidence interval.
jcm-09-00077-t004_Table 4
######
Incidence and mortality trends of sepsis in adults aged 20--44 years.
Variable 2006 2015 Change Whole Period (%) Annual Average Percent Change (%, 95% CI)
---------------------------- ------ ------ ------------------------- -------------------------------------------
Rates
Adjusted incidence ^‡^
Overall 13.5 17.1 26.7 1.5 (0, 3.0)
Women 10.3 15.9 54.4 3.8 (2.1, 5.5) ^†^
Men 16.5 18.2 10.3 −0.2 (−1.6, 1.4)
Charlson Index (%)
0 38.6 44.6 15.5 1.3 (0.6, 1.9) ^†^
\>0 61.4 55.4 −9.8 −0.9 (−1.3, −0.4) ^†^
No. organ dysfunctions (%)
1 48.0 45.0 −6.2 −0.6 (−1.4, 0.2)
2 28.9 23.9 −17.3 −1.6 (−2.4, −0.9) ^†^
\>2 21.7 24.9 14.7 0.6 (−0.8, 2.1)
CRF (%)
Overall 31.1 18.6 −40.2 −5.9 (−6.6, −5.2) ^†^
Sex
Women 27.9 14.6 −47.7 −6.6 (−7.8, −5.4) ^†^
Men 33.0 21.9 −33.6 −5.1 (−6.1, −4.2) ^†^
Charlson's index
0 19.0 9.9 −47.9 −8.1 (−10.1, −6.1) ^†^
\>0 38.7 25.6 −33.8 −4.7 (−5.2, −4.2) ^†^
No. organ dysfunctions
1 18.6 11.0 −40.9 −6.6 (−8.7, −4.4) ^†^
2 33.9 17.7 −47.8 −7.0 (−8.2, −5.8) ^†^
\>2 55.1 35.7 −35.2 −4.6 (−5.4, −3.9) ^†^
^†^ The Annual percent change is significantly different from zero (Poisson regression, *p* \< 0.05). ^‡^ per 100,000 persons.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Neurodegenerative disease is a type of chronic degenerative disease in the central nervous system with the degenerative changes of the neuronal cells in brain and spinal cord. The symptoms of the neurodegenerative disease deteriorate slowly and eventually lead to death \[[@CR1], [@CR2]\]. Thereinto, the Huntington's disease is due to a triplet (CAG) repeat elongation in the Huntington gene (IT15), which further affects numerous interactions between molecules. With the accumulation of the variant Htt protein in brain, a number of molecular pathways are affected in turn, resulting in neuronal malfunction and degeneration. Changes in Htt protein and the interactions between molecules are closely associated with the abnormalities of gene expression. It has been shown that there exist abnormalities of gene expression among the genes related to nerve conduction in the striatum tissue of Huntington's disease individuals \[[@CR3], [@CR4]\]. Since the complexity of chronic disease, the molecular pathogenesis of Huntington's disease is not entirely clear. Nevertheless, identifying the key genes associated with the disease deterioration can reveal useful insights into the disease pathogenesis.
The rapid development of high-throughput sequencing technologies, especially next-generation sequencing methods, provides possibility for us to explore the molecular mechanisms of complex disease on a genome-wide scale. However, because of the complex etiology of chronic diseases \[[@CR5]\], the traditional disease-associated gene prediction methods cannot effectively identify the genes affected during the disease development. Generally, the disease-associated prediction methods roughly fall into three categories: network-based methods \[[@CR6], [@CR7]\], statistic-based methods \[[@CR8]--[@CR10]\], and machine learning methods \[[@CR11], [@CR12]\]. At present, as a branch of machine learning methods, the deep learning methods have become the most advanced technology in the field of computer vision, speech recognition and natural language processing. Deep learning methods use the hierarchical structure of deep neural network to conduct the nonlinear transfer of the input data, which could learn automatically the internal features that represent the original data \[[@CR13], [@CR14]\]. Compared with methods that are of manual designed features, the deep learning methods could improve the prediction accuracy. Recently, the deep learning methods have been introduced into the field of bioinformatics. Liang et al. \[[@CR15]\] designed a multimodal deep belief network to conduct the integrative data analysis on multi-platform genomic data including gene expression data, miRNA expression data, and DNA methylation data. They used the model to detect a unified representation of latent features, capture both intra- and cross- modality correlations, and to identify key genes that may play distinct roles in the pathogenesis between different cancer subtypes. Cheng et al. \[[@CR16]\] designed a miRNA prediction algorithm based on convolutional neural network (CNN). The CNN automatically extracts essential information from the input data while the exact miRNA target mechanisms are not well known. Experimental results demonstrated that the algorithm significantly improved the prediction accuracy.
During neurodegenerative disease development, gene expression level is affected by many factors, e.g. the environment, impaired metabolic pathways, protein mis-folding, etc \[[@CR17]--[@CR19]\]. Intuitively, identifying the key genes associated with the disease development is to screen the genes that are most seriously affected by these factors over with time. Consequently, the features that distinguish disease-related genes from non-disease-related genes could be represented by these factors. Extracting the deep hierarchical structure of the gene expression data and learning the important information represented by the decreased neurones in hidden layers are helpful to further understand the changes of gene expression during the disease development. In this paper, we designed a deep learning approach based on restricted Boltzmann machine to analyze the gene expression data \[[@CR20]\], namely stacked restricted Boltzmann machine (SRBM). We used the unsupervised contrastive divergence algorithm (CD) to learn the parameters in each restricted Boltzmann machine \[[@CR21], [@CR22]\]. By maximizing the likelihood function, the probability distribution of the hidden layer variables fitted the probability distribution of the original data well. We trained the stacked restricted Boltzmann machine in a greedy layer-wise fashion \[[@CR23]\]. Because the number of neurons in hidden layers is far smaller than that in the visible layer, we could reduce dimensions of the input data and capture useful high-level features of the input data at the same time. The gene expression level is manipulated by regulatory factors. In this work, we assume that the effects of regulatory factors can be captured by the hierarchical structure and narrow hidden layers of the SRBM. We used the model to rank the genes, aiming to make key genes that may play important roles in the pathogenesis of Huntington's disease with high rankings. First, according to the differentially activated hidden neurons obtained by gene expression datasets at different time periods, we selected disease-associated factors. Then, we selected disease-associated genes according to the changes of the gene energy in SRBM at different time periods. Experimental results demonstrated that SRBM can detect the important information for differential analysis of time series gene expression datasets. The identification accuracy of the disease-associated genes is improved to some extent. Moreover, the prediction precision of disease-associated genes for top ranking genes using SRBM is effectively improved compared with that of the state of the art methods.
The presented study is organized as follows: The deep learning approach proposed in this paper is presented in "[Methods](#Sec2){ref-type="sec"}" section. Experiments that analyze the performance of the stacked restricted Boltzmann machine and the overall discussion of the experimental results are reported in "[Results and discussion](#Sec8){ref-type="sec"}" section. Conclusions are presented in "[Conclusions](#Sec13){ref-type="sec"}" section.
Methods {#Sec2}
=======
In this section, first, the stacked restricted Boltzmann machine model and the learning method are described. Next, we detailedly describe how the SRBM is used to extract the disease-associated genes with gene expression data at different disease stages. Finally, we present the parameter setting of the SRBM.
Stacked restricted Boltzmann machine {#Sec3}
------------------------------------
### Model {#Sec4}
RBM is a kind of undirected probabilistic graphical model containing a layer of observable variables and a single layer of hidden variables \[[@CR24]\]. In the RBM model (Fig. [1](#Fig1){ref-type="fig"}), each visible variable connects to every hidden variable, but no connections are allowed between any two variables within the same layer. Fig. 1Schematic illustration of RBM
In this study, we designed a stacked restricted Boltzmann machine to extract the hierarchical structures of gene expression dataset. The schematic illustration of SRBM is shown in Fig. [2](#Fig2){ref-type="fig"}. We add another RBM (denoted as RBM2 in Fig. [2](#Fig2){ref-type="fig"}) to the original RBM (denoted as RBM1 in Fig. [2](#Fig2){ref-type="fig"}). The input of visible layer in RBM2 is the output of hidden layer in RBM1. The dimension of gene expression data can be further reduced through the SRBM. As the gene expression data is real-valued data, we assume that the expression of genes obeys Gaussian distribution \[[@CR15]\]. We use a Gaussian-Bernoulli RBM model for RBM1. However, variables in RBM2 are all binary numbers. Fig. 2Schematic illustration of SRBM
In the analysis of the gene expression dataset, the gene expression profile of a sample is *V*=(*v* ~1~,*v* ~2~,⋯,*v* ~*n*~ ~*V*~), where *v* ~*i*~ represents the expression level of gene *i* and *n* ~*V*~ is the number of genes. Here, *v* ~*i*~ represents visible variable and *V* represents a layer of visible variables. *H*=(*h* ~1~,*h* ~2~,⋯,*h* ~*n*~ ~*H*~) denotes the layer of hidden variables, where *h* ~*j*~ represents hidden variable and *n* ~*H*~ is the number of hidden variables. The weight of the corresponding connection between hidden variable *h* ~*j*~ and visible variable *v* ~*i*~ is *w* ~*ji*~. The weight matrix *W*=\[*w* ~*ji*~\]~*n*~ ~*H*~×*n* ~*V*~ represents the parameter setting of weights between the hidden layer and the visible layer. Let *B*=(*b* ~1~,*b* ~2~,⋯,*b* ~*n*~ ~*V*~) be the bias vector of visible layer, where *b* ~*i*~ stands for the bias of visible variable *v* ~*i*~. Let *C*=(*c* ~1~,*c* ~2~,⋯,*c* ~*n*~ ~*H*~) be the bias vector of hidden layer, where *c* ~*j*~ stands for the bias of hidden variable *h* ~*j*~.
In RBM1 (Gaussian-Bernoulli RBM), the conditional distribution over the visible variables is usually supposed to be a Gaussian distribution whose mean is a function of the hidden variables \[[@CR25], [@CR26]\]. The conditional probability of a visible variable is $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ p_{\theta}\left(v_{i}|H\right)= \mathcal{N}\left(\sum_{j=1}^{n_{H}}h_{j}w_{ji}+b_{i},\sigma_{i}^{2}\right), $$ \end{document}$$
where *θ*=(*W*,*B*,*C*) represents the parameter setting of the model. Symbol *σ* ~*i*~ is the standard deviation of Gaussian noise in visible variable *v* ~*i*~.
The energy function of the RBM1 with binary hidden variables and real-valued visible variables can be defined as $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ E_{\theta}(V,H)=\sum_{i=1}^{n_{V}}\frac{\left(v_{i}-b_{i}\right)^{2}}{2\sigma_{i}^{2}}-\sum_{j=1}^{n_{H}}c_{j} h_{j} -\sum_{i=1}^{n_{V}}\sum_{j=1}^{n_{H}}\frac{v_{i}}{\sigma^{2}_{i}}h_{j}w_{ji}. $$ \end{document}$$
To simplify the parameter learning of the model, we standardized the input gene expression dataset, i.e., the average value of the visible variables *v* ~*i*~ is equal to 0 and the variance of that is equal to 1 (*σ* ~*i*~=1). In this way, the energy function in Eq. [2](#Equ2){ref-type=""} can be rewritten as $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ E_{\theta}(V,H)=\sum_{i=1}^{n_{V}}\frac{\left(v_{i}-b_{i}\right)^{2}}{2}-\sum_{j=1}^{n_{H}}c_{j} h_{j} -\sum_{i=1}^{n_{V}}\sum_{j=1}^{n_{H}}v_{i}h_{j}w_{ji}. $$ \end{document}$$
The joint probability density function of (*V*,*H*) is given by $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ p_{\theta}(V,H)= \frac{1}{Z(\theta)} e^{-E_{\theta}(V,H)}, $$ \end{document}$$
where *Z*(*θ*) is a normalizing constant known as the partition function, $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$Z(\theta)=\sum _{V,H}e^{-E_{\theta }(V,H)}$\end{document}$. It is important to state that the variables are under independent identical distribution. We need to get the conditional probability distribution of the visible variables due to the unobservability of the hidden layer, thus to solve the model. The edge probability distribution of the visible variables is given by $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ p_{\theta}(V)=\sum_{H}p_{\theta}(V,H) = \frac{1}{Z(\theta)}\sum_{H}e^{-E_{\theta}(V,H)}. $$ \end{document}$$
Since the gene expression data are very noisy, we discretized the gene expression values into binary values during the Gibbs sampling process. And we used binary activations instead of the real-valued visible units sampled from a Gaussian distribution which are usually seen as their activations. Because a binary activation contains less information than a real-valued gene expression, using the binary activation to represent a gene expression is helpful to distinguish the genes. This is a straightforward way to reduce noise in the gene expression data. The conditional probability density distributions can be easily obtained according to Eqs. [4](#Equ4){ref-type=""} and [5](#Equ5){ref-type=""}. (The detail derivation process is given in Additional file [1](#MOESM1){ref-type="media"}). $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ p\left(h_{k}=1|V\right) = \frac{1}{1 + e^{-\left(c_{k} + \sum_{i=1}^{n_{V}}w_{ki}v_{i}\right)}}, $$ \end{document}$$
$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ p(v_{k}=1|H) = \frac{1}{1 + e^{-\left(-0.5+b_{k} + \sum_{j=1}^{n_{H}}h_{j}w_{jk}\right)}}. $$ \end{document}$$
In RBM2, *v*=(*v* ~1~,*v* ~2~,⋯,*v* ~*n*~ ~*v*~) represents the input layer (hidden layer 1 in Fig. [2](#Fig2){ref-type="fig"}) and *h*=(*h* ~1~,*h* ~2~,⋯,*h* ~*n*~ ~*h*~) denotes the output layer (hidden layer 2 in Fig. [2](#Fig2){ref-type="fig"}). The weight of the corresponding connection between output variable *h* ~*j*~ and input variable *v* ~*i*~ is *w* ~*ji*~. The weight matrix *w*=\[*w* ~*ji*~\]~*n*~ ~*h*×*n*~ ~*v*~ represents the parameter setting of weights between the output layer and the input layer. Let *b*=(*b* ~1~,*b* ~2~,⋯,*b* ~*n*~ ~*v*~) be the bias vector of input layer, where *b* ~*i*~ stands for the bias of variable *v* ~*i*~. Let *c*=(*c* ~1~,*c* ~2~,⋯,*c* ~*n*~ ~*h*~) be the bias vector of output layer, where *c* ~*j*~ stands for the bias output variable *h* ~*j*~.
As the variables in RBM2 are all binary, the energy function of the RBM2 model is defined as $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ E_{\theta}(v,h)=-\sum_{i=1}^{n_{v}}b_{i} v_{i}-\sum_{j=1}^{n_{h}}c_{j} h_{j} -\sum_{i=1}^{n_{v}}\sum_{j=1}^{n_{h}}h_{j}w_{ji} v_{i}. $$ \end{document}$$
In the same way, we get the following conditional probability density distributions $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ p\left(h_{k}=1|v\right) = \frac{1}{1 + e^{-\left(c_{k} + \sum_{i=1}^{n_{v}}w_{ki}v_{i}\right)}}, $$ \end{document}$$
$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ p\left(v_{k}=1|h\right) = \frac{1}{1 + e^{-\left(b_{k} + \sum_{j=1}^{n_{h}}h_{j}w_{jk}\right)}}. $$ \end{document}$$
### Learning {#Sec5}
Training the RBM model means to learn the parameters of the model, making sure that the probability density distribution of the hidden variables fit that of the variables in the visible layer well. Physically, the energy function of the system is minimized when the system reaches a steady state. Mathematically, the goal of RBM training is to maximize the logarithmic likelihood function. For such a type of optimization problem, we use gradient up method to learn the parameters of the model. $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ \theta := \theta +\eta \frac{\partial log p_{\theta}(V)}{\partial \theta}, $$ \end{document}$$
$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ \frac{\partial log p_{\theta}(V)}{\partial \theta} \,=\, -\!\left\langle \frac{\partial E_{\theta}(V,H)}{\partial \theta} \right\rangle_{p_{\theta}(H|V)} \!+ \left\langle \frac{\partial E_{\theta}(V,H)}{\partial \theta} \right\rangle_{p_{\theta}(V,H)}, $$ \end{document}$$
where *η* is learning rate, $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$\left \langle \frac {\partial E_{\theta }(V,H)}{\partial \theta } \right \rangle _{p_{\theta }(H|V)}$\end{document}$ is the expectation of energy gradient function $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$\frac {\partial E_{\theta }(V,H)}{\partial \theta }$\end{document}$ under the condition distribution *p* ~*θ*~(*H*\|*V*), and $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$\left \langle \frac {\partial E_{\theta }(V,H)}{\partial \theta } \right \rangle _{p_{\theta }(V,H)}$\end{document}$ is the expectation of energy gradient function under the joint distribution *p* ~*θ*~(*V*,*H*). Since the hidden variables cannot be directly observed, we use CD-*k* algorithm to approximately estimate the probability *p* ~*θ*~(*V*) though Gibbs sampling in *k* steps \[[@CR21], [@CR22]\], thus to obtain the solution of $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$\left \langle \frac {\partial E_{\theta }(V,H)}{\partial \theta } \right \rangle _{p_{\theta }(V,H)}$\end{document}$. For sample *V*, the initial values of visible layer is *V* ^(0)^=*V*. We use *V* ^(*k*)^ to denote the sample obtained by CD-*k*.
The gradients for sample *V* in one iterative process are given by (The detail derivation process is given in Additional file [1](#MOESM1){ref-type="media"}). $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ {} \frac{\partial log p_{\theta}(V)}{\partial w_{ij}} = p\left(h_{i}=1|V^{(0)}\right)v_{j}^{(0)}-p\left(h_{i}=1|V^{(k)}\right)v_{j}^{(k)}, $$ \end{document}$$
$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ \frac{\partial log p_{\theta}(V)}{\partial b_{i}} = v_{i}^{(0)}-v_{i}^{(k)}, $$ \end{document}$$
$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ \frac{\partial log p_{\theta}(V)}{\partial c_{i}} = p\left(h_{i}=1|V^{(0)}\right)-p\left(h_{i}=1|V^{(k)}\right). $$ \end{document}$$
In this study, we use mini-batch strategy to learn parameters in the RBM. We use sample set *S*={*V* ^1^,*V* ^2^,⋯,*V* ^*n*^} to train the model one batch. Here *n* ~*block*~=*n* represents the size of mini-batch. The gradient calculation formula for one iteration is shown below $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ \frac{\partial log L_{s}}{\partial \theta} = \sum_{t=1}^{n} \frac{\partial\left(log p(V^{t})\right)}{\partial \theta}, $$ \end{document}$$
where *L* ~*s*~=*p* ~*θ*~(*S*) is the likelihood function of product edge probability density distributions, *V* ^*t*^ represents the *t*-th sample. The gradients for *S* in one iteration are given by $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ {\begin{aligned} \frac{\partial log L_{s}}{\partial w_{ij}} = \sum_{t=1}^{n}\left[p\left(h_{i}=1|V^{t(0)}v_{j}^{t(0)}-p\left(h_{i}=1|V^{t(k)}\right)v_{j}^{t(k)}\right.\right], \end{aligned}} $$ \end{document}$$
$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ \frac{\partial log L_{s}}{\partial b_{i}} = \sum_{t=1}^{n}\left[ v_{i}^{t(0)}-v_{i}^{t(k)}\right], $$ \end{document}$$
$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ \frac{\partial log L_{s}}{\partial c_{i}} = p\left(h_{i}=1|V^{t(0)}\right)-p\left(h_{i}=1|V^{t(k)}\right). $$ \end{document}$$
In summary, the detail training process of the RBM is shown below.
We trained the stacked restricted Boltzmann machine in a greedy layer-wise fashion \[[@CR23]\]. We first trained the RBM1 according to the above training process (see Algorithm 1), then trained RBM2 in the same way.
Identification of key genes {#Sec6}
---------------------------
In our study, the regulatory factors are seen as high-level features which could be captured by the hierarchical structure and narrow hidden layers of the SRBM. On the one hand, the differentially activated hidden neurons are important for distinguishing different disease stage samples. On the other hand, the neurons differential activation indicates that the regulatory factors change greatly during the disease development. So, we select disease-related regulatory factors according to the differentially activated neurons in the hidden layers.
Biologically, the connections among neurons in one functional neural circuit are more strong. In fact, it has also been shown that the high-level hidden units in RBM tend to have strong positive weights to similar features in the visible layer \[[@CR27]\]. In an SRBM model, the connections from a visible unit in the input layer to the high-level features (disease-related regulator factors) are seen as the connections in a functional neural circuit. And we use the energy of the neural circuit in the SRBM to measure the property of the input unit (represent a gene). Since the hidden units were activated very differently along with the disease progression, the energy of the neural circuit changed greatly. It suggests that the gene expression has been greatly affected during the disease development. Based on the above analysis, we rank the genes according to the energy changes at different time periods. The higher the ranking of gene it is, the more likely the disease-related gene it is.
Let $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$x_{i}^{s}$\end{document}$ denote the activated frequency of neuron *i* in the first hidden layer, using the gene expression data of *s* time period samples. Symbol $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$y_{j}^{s}$\end{document}$ denotes the activated frequency of neuron *j* in the second hidden layer, i.e., the output layer. Let $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$E_{g}^{s}$\end{document}$ denote the energy of gene *g* at *s* time period. According to Eqs. [3](#Equ3){ref-type=""} and [8](#Equ8){ref-type=""}, the energy of gene *g* is given by $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ {{} \begin{aligned} E_{g} = \frac{\left(v_{g}-b_{1,g}\right)^{2}}{2} - \sum_{j=1}^{n_{H}}h_{1,j}w_{1,jg}v_{g}&-\sum_{i=1}^{n_{v}}b_{2,i} v_{2,i} \\&-\sum_{i=1}^{n_{v}}\sum_{j=1}^{n_{h}}h_{2,j} w_{2,ji} v_{2,i}, \end{aligned}} $$ \end{document}$$
where *b* ~1,*i*~, *h* ~1,*i*~, *w* ~1,*ji*~ represent the parameters in RBM1 and *b* ~2,*i*~, *v* ~2,*i*~, *h* ~2,*i*~, *w* ~2,*ji*~ represent the parameters in RBM2. Since the energy caused by the bias of the hidden layer in RBM1 is same for all genes, we omit the term in the calculation formula of gene energy.
The energy change of gene *g* at different time periods is computed by $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document} $$ C_{g} = \left|\frac{1}{|s_{1}|}\sum_{i=1}^{s_{1}} E^{s_{1}}_{g} - \frac{1}{|s_{2}|}\sum_{i=1}^{s_{2}} E^{s_{2}}_{g}\right|, $$ \end{document}$$
where *s* ~*i*~ denotes the samples at *i* time period. The details for identifying key genes are shown below: **Step 1.** Rank the two hidden layer neurons in descending order according to the difference of the activated frequency between different time periods, respectively. We select the top ranked neurons in the ranked lists as the differentially activated neurons, respectively. The neurons that are not differentially activated in the two hidden layers are all set to 0 in any case. **Step 2.** Compute the energy changes of gene *g* at different time periods according to Eq. [21](#Equ21){ref-type=""}. Rank genes in descending order according to the energy changes of genes.
Parameter setting {#Sec7}
-----------------
Here, we initialize parameters in SRBM according to empirical studies in deep learning literature. The initialization weights obey Gaussian distribution *N*(0,0.01). The initialization bias variables are set to 0. The learning rate *η*=0.5. The number of hidden neurons is usually about one tenth of visible neurons. In this study, the number of variables in the first hidden layer is 400 and that of the second hidden layer is 20. Moreover, the number of sampling steps in CD-*k* is set to be *k*=1.
Results and discussion {#Sec8}
======================
We used the SRBM to analyze the gene expression data of Huntington's disease mice at different time periods. In this section, first, we briefly introduce the dataset used in this study. Second, we demonstrate the experimental results using SRBM. Then, we compare the performance of SRBM with other computational methods. Finally, we analyze and discuss the results of SRBM in detail.
Gene expression data {#Sec9}
--------------------
The gene expression dataset used in this study were downloaded from <http://www.hdinhd.org>, which were obtained from the striatum tissue of Huntington's disease mice by using RNA-seq technology. The genotype of Huntington's disease mice is ployQ 111. There are 8 samples of 2-month-old mice and 8 samples of 6-month-old Huntington's disease mice. We conducted a preprocessing step to filter out noisy and redundant genes by selecting the genes with large mean value and variance of the 16 samples. Finally 4433 genes from the total 23,351 genes were left for further analysis. The data of modifier genes were from \[[@CR28]\], which contained 520 genes, including 89 disease-related genes and 431 non-disease-related genes.
The results of SRBM {#Sec10}
-------------------
Figures [3](#Fig3){ref-type="fig"} and [4](#Fig4){ref-type="fig"} show the energy changes of RBM1 and RBM2 along with every iteration during the parameter training process. From Figs. [3](#Fig3){ref-type="fig"} and [4](#Fig4){ref-type="fig"}, we can see that the changes become small with the increasing of iterations. In this study, since there are large amounts of parameters in RBM1, the iteration times of RBM1 are preset to be 50 to reduce computational time and avoid over-fit. The iteration times of RBM2 are preset to be 400 to avoid over-fit. Fig. 3The energy change of RBM1. **a** The energy change of RBM1 with gene expression data of 2-month-old Huntington's disease mice. **b** The energy change of RBM1 with gene expression data of 6-month-old Huntington's disease Fig. 4The energy change of RBM2. **a** The energy change of RBM2 with gene expression data of 2-month-old Huntington's disease mice. **b** The energy change of RBM2 with gene expression data of 6-month-old Huntington's disease
We statisticed the differentially activated frequency of neurons in the hidden layers using SRBM with gene expression datasets at different time periods. The results are shown in Table [1](#Tab1){ref-type="table"}. Compared with the differentially activated frequency of neurons in the hidden layer 1, that in the hidden layer 2 is much larger. The number of neurons, whose differentially activated frequency in hidden layer 1 is 3, is too small to be used to distinguish samples at different time periods. It is better to use the neurons with largest differentially activated frequency in the hidden layer 2 to distinguish samples at different time periods, thus to identify the key genes that may be seriously affected during the disease progression. Table 1The number of neurons that are of the same differentially activated frequency using SRBM with different time period samplesDifferentially activated frequencyHidden layer 1Hidden layer 250540234325731199401403
Furthermore, we draw heatmaps of the weight matrices of RBM2 to investigate the deep structure difference between the gene expression data of Huntington's disease mice at different time periods. The weight matrices are obtained by using SRBM with gene expression datasets of Huntington's disease mice at different time periods (Figs. [5](#Fig5){ref-type="fig"} and [6](#Fig6){ref-type="fig"}). The numbers in the left of the heatmap represent the corresponding neuron in the output layer. From Figs. [5](#Fig5){ref-type="fig"} and [6](#Fig6){ref-type="fig"}, we can clearly see that there are significant difference between the two heatmaps. It suggests that the gene expression changes complicatedly during the disease progression. Fig. 5Heatmap of weight matrix of RBM2 with 2-month-old gene expression data. The weight matrix is obtained using SRBM with gene expression data of 2-month-old Huntington's disease mice Fig. 6Heatmap of weight matrix of RBM2 with 6-month-old gene expression data. The weight matrix is obtained using SRBM with gene expression data of 6-month-old Huntington's disease mice
Performance comparison between SRBM with other methods {#Sec11}
------------------------------------------------------
To verify the performance of SRBM, we conducted other experiments using the original RBM method, t-test method \[[@CR10]\], fold change rank-product method (FC-RP) \[[@CR10]\], and joint non-negative matrix factorization meta-analysis method (jNMFMA) \[[@CR11]\] with the gene expression data. We use true positive rate (TPR), false positive rate (FPR), precision, and recall to evaluate the prediction accuracy of disease-associated genes. TPR is defined as the ratio of correctly predicted disease genes to all disease genes. FPR is defined as the ratio of incorrectly predicted disease genes to all non-disease genes. Precision is defined as the ratio of correctly predicted disease genes to all the predicted disease genes. Recall is defined as the ratio of correctly predicted disease genes to all disease genes. The receiver operating characteristic (ROC) curves were created by plotting TPR versus FPR. The precision-recall (PR) curves were created by plotting precision versus recall. The area under the ROC curve (AUC) and the area under the precision-recall curve (AUPR) were used as measures of the prediction accuracy \[[@CR29]\].
To test the reasonability of the assumption in this study, we used all neurons in hidden layers to compute the gene energy while overlooking one third weak connections that from one neuron to all the neurons of the next layer. The corresponding experiments are denoted as SRBM-I. On the other hand, we selected differentially activated neurons at different time periods as factors that manipulate the expression of all genes during the disease progression, 61 neurons were selected in the first hidden layer with differentially activated frequency larger than 1, and 5 neurons were selected in the second hidden layer with differentially activated frequency larger than 5. Then, we computed the energy for each gene. The corresponding experiments are denoted as SRBM-II. Note that we use RBM-I and RBM-II to denote the experiments using the original RBM model.
From Fig. [7](#Fig7){ref-type="fig"}, we can see that the ROC cures of the seven methods are similar. The AUCs of these methods are around 0.5. It illustrates that these methods cannot separate the disease genes from non-disease genes in the modifier gene set. It also indicates that the expression of genes change complicatedly during the disease development. Nevertheless, the AUC of SRBM-II is mildly improved compared with that of the other six methods. Fig. 7ROC curves. The ROC curves of the prediction results using t-test, FC-RP, jNMFMA, RBM-I, RBM-II, SRBM-I and SRBM-II
From Fig. [8](#Fig8){ref-type="fig"}, the PR curves of the seven methods are similar to some extent. However, the prediction precision for top ranked genes of the seven methods are clearly distinct. The prediction precision of SRBM-II is significantly higher for top ranked genes compared with that of the other six methods. Fig. 8Rank-product curves. The RP curves of the prediction results using t-test, FC-RP, jNMFMA, RBM-I, RBM-II, SRBM-I and SRBM-II
We further investigate the distributions of the rankings of top ranked 10 disease genes in the ranked lists obtained by using the seven methods, respectively (Fig. [9](#Fig9){ref-type="fig"}). From Fig. [9](#Fig9){ref-type="fig"}, we can roughly know the rankings of the top ranked disease genes. Although the distributions obtained by these methods are similar, SRBM-II makes the disease genes get mild higher rankings compared with the other six methods. Fig. 9Boxplots of the rankings of top ranked 10 disease genes. The rankings are obtained using different methods, including t-test, FC-RP, jNMFMA, RBM-I, RBM-II, SRBM-I and SRBM-II
In total, the performance of SRBM-II is moderately better than other methods. From Figs. [7](#Fig7){ref-type="fig"}, [8](#Fig8){ref-type="fig"} and [9](#Fig9){ref-type="fig"}, we can know that the performance of SRBM-II is better than SRBM-I. It suggests that we improved the prediction accuracy by selecting the differentially activated neurons, which are assumed to be disease-associated factors in our study. We can also know that the performance of SRBM methods are better than RBM methods. It verifies that we effectively separated some noisy factors from the gene expression dataset, using the deep structure of SRBM.
We also statisticed the overlapped degree of top ranked 500 genes between any two ranked lists, the results are shown in Table [2](#Tab2){ref-type="table"}. It can be clearly seen that the overlapped degrees between any two ranked lists (except for that between SRBM-I and SRBM-II) are all small. However, the overlap degrees between jNMFMA and SRBM methods are smaller than that between others. The jNMFMA assumes that the gene expression is a weighted linear combination of metagenes. The jNMFMA selects disease-associated genes through differentially regulated metagenes. SRBM selects disease-associated genes according to the energy changes at different disease states. Since the basic assumptions of the two models are greatly different, the overlapped degrees of top ranked genes between the two ranked lists are smaller. Table 2The number of overlapped genes (the degree of overlap) of top ranked 500 genes between any two ranked lists obtained using t-test, FC-RP, jNMFMA, RBM-I, RBM-II, SRBM-I, and SRBM-IIFC-RPjNMFMARBM-IRBM-IISRBM-ISRBM-IIt-test81 (16.2%)36 (7.2%)73 (14.6%)74 (14.8%)75 (15%)73 (14.6%)FC-RP114 (22.4%)28 (5.6%)22 (4.4%)38 (7.6%)40 (8.0%)jNMFMA6 (1.2%)8 (1.6%)5 (1.0%)9 (1.8%)RBM-I344 (68.8%)252 (50.4%)214 (42.8%)RBM-II245 (49.0%)248 (49.6%)SRBM-I351 (70.2%)
The top ranked 500 genes in different ranked lists share 4 common genes: Chmp1b, Poldip3, Lrrtm1 and Slc44a1. According to the annotation of Gene Ontology, the molecule function of Chmp1b is protein domain specific binding, that of Lrrtm1 is protein kinase inhibitor activity, that of Poldip3 is nudeotide binding, and that of Slc44a1 is choline transmembrane transporter activity. The functions of the four genes are all related to protein transportation. Those genes may be related to the disturbance of intracellular protein trafficking in Huntington's disease individuals \[[@CR30]\].
Enrichment analysis {#Sec12}
-------------------
According to Fig. [10](#Fig10){ref-type="fig"}, it is obvious that the changes of gene energy for the top ranked 100 genes are significantly larger. Combined with Fig. [8](#Fig8){ref-type="fig"}, we known that the higher the ranking of gene it is, the more precise the prediction accuracy of disease-related gene it is. To avoid introducing too many false positives, we chose the top ranked 100 genes in the ranked list obtained by using SRBM-II to conduct enrichment analysis. We used the functional annotation clustering tool through DAVID \[[@CR31]\] to annotate the functions of those genes, the result can be seen in Table [3](#Tab3){ref-type="table"}. The annotations listed in the table are cellular component from GOTERM. From Table [3](#Tab3){ref-type="table"}, we can see that those genes are related to membrane, synapse and cell junction. It suggests that the cellular form changes greatly during the Huntington's disease progression and deterioration. In fact, the connections between neurons get sparse, and the neurons finally died during the Huntington's disease deterioration \[[@CR32], [@CR33]\]. Fig. 10The changes of gene energy. The gene ranking is obtained by using SRBM-II based on the changes of gene energy at different time periods Table 3The functional annotation clustering of the top ranked 100 genes in the ranked list obtained using SRBM-IIAnnotationAnnotationGenes-*P*-valueBenjaminiclusterincludedAnnotationMembrane607.2E-87.3E-6cluster 1Plasma membrance425.0E-51.1E-3AnnotationSynapse148.2E-73.3E-5cluster 2Postsynaptic density102.2E-67.3E-5Dendritic spine77.8E-51.6E-3Cell junction112.3E-32.5E-2Synaptic vesicle42.3E-21.8E-1Postsynaptic membrane49.0E-24.0E-1AnnotationCell-cell adherens junction88.0E-41.3E-2cluster 3
Conclusions {#Sec13}
===========
In this paper, we designed a stacked restricted Boltzmann machine to detect the hierarchical structures and to capture the important information for differential analyzing gene expression datasets of Huntington's disease mice at different time periods. We also proposed a new framework to identify the key genes that may be affected by the disease progression. Experimental results verify the feasibility of the assumption in this study. It also demonstrates that the performance of SRBM-II is mildly better than other traditional methods. Besides the exploratory analysis of the disease molecular mechanisms through enrichment analysis, we also conducted a integrated analysis on the ranked lists obtained by the seven methods. We found that four genes (Chmp1b, Poldip3, Lrrtm1 and Slc44a1) related to protein transportation are seriously affected during the disease progression.
Additional file
===============
{#Sec14}
Additional file 1Supplementary Material. The detail derivation process for solving the gradients of RBMs learning is given in the Supplementary Material. (PDF 321 kb)
AUC
: The area under the ROC curve
AUPR
: The area under the PR curve
CD
: Contrastive divergence
CNN
: Convolutional neural network
FC-RP
: Fold change rank-product
FPR
: False positive rate
jNMFMA
: Joint non-negative matrix factorization neta-analysis method
PR
: Precision-recall
RBM
: Restricted Boltzmann machine
ROC
: Receiver operating characteristic
SRBM
: Stacked restricted Boltzmann machine
TPR
: Stacked restricted Boltzmann machine
**Electronic supplementary material**
The online version of this article (doi:10.1186/s12859-017-1859-6) contains supplementary material, which is available to authorized users.
The authors would like to thank the editor and the reviewers for their comments and suggestions, which helped improve the manuscript greatly.
Funding {#d29e4701}
=======
This work has been supported by the Natural Science Foundation of Tianjin (15JCYBJC18900), the National Natural Science Foundation of China (31728013, 61673224, 61403213), and the Key Program of Science Foundation of Tianjin (14JCZDJC31800).
Availability of data and materials {#d29e4706}
==================================
The gene expression data used in this study were downloaded from <http://www.hdinhd.org>. To make the dataset available to public, we deposit it in publicly available repository, please download at <https://figshare.com/s/071072960aa5132a6d2c>. The modifier genes were from "Langfelder P, Cantle J P, Chatzopoulou D, et al. Integrated genomics and proteomics define huntingtin CAG length-dependent networks in mice\[J\]. Nature Neuroscience, 2016. PMID: 26900923 DOI: 10.1038/nn.4256". We also deposit it in publicly available repository, please download at <https://figshare.com/s/13fdc5c17d736142dcd0>. Moreover, we also deposit the ranked lists obtained by using SRBM-II in the publicly available repository for reference, please download at <https://figshare.com/s/bf1ea56b278ef7b3f1ac>.
HZ, FD and XQ conceived the research. XJ and FD designed the research. XJ performed the experiments and analyzed the data. XJ, HZ and FD wrote the manuscript. All authors reviewed the manuscript. All authors read and approved the final manuscript.
Ethics approval and consent to participate {#d29e4737}
==========================================
Not applicable.
Consent for publication {#d29e4742}
=======================
Not applicable.
Competing interests {#d29e4747}
===================
The authors declare that they have no competing interests.
Publisher's Note {#d29e4752}
================
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION {#s1}
============
During the past 36 years, unrelated donor HCT has already become one of the most effective but complex therapy for selected patients with hematologic malignancies or certain life-threatening nonmalignant disorders \[[@R1], [@R2]\]. However, the clinical application of the HCT is limited by leukemia relapse \[[@R3]\] and life-threatening complications, such as graft-versus-host disease (GVHD) \[[@R4], [@R5]\], infection \[[@R6]--[@R8]\], conditioning regimen-related toxicities \[[@R9]--[@R11]\], and transplant-associated thrombotic microangiopathy (TMA) \[[@R12]--[@R14]\] as well, which are more common in patients receiving HLA locus mismatched grafts. In clinical practice, it is increasingly difficult to identify a HLA locus completely matched donor in the presence of highly polymorphic HLA alleles. The most loci are the HLA class I (A, B and C) and the class II (DRB1, DQB1 and DPB1) molecules \[[@R15]\].
A large number of studies assessed the impact of individual HLA mismatches on multiple clinical outcomes \[[@R16]--[@R51]\]. For a given end point, the risk of a specific HLA locus mismatches was generally inconsistent or even contradictory across studies. These discrepancies make it difficult to figure out which mismatched HLA loci contribute mainly to the incidence and severity of GVHD, TRM and mortality, and which HLA locus mismatches has minimal impact on outcomes. Despite there are several guidelines published, evidence-based recommends have been absent so far \[[@R52]--[@R55]\]. Most recently, a published meta-analysis assess the impact of HLA-DPB1 allele mismatches on overall survival of patients receiving unrelated-donor HCT \[[@R56]\]. Other important end points were not mentioned, and many studies with large populations were not included in the review. Additionally, the analysis of individual HLA locus mismatches at the antigen level was not performed in the absence of sufficient data. As such, we undertake the meta-analysis in an effort to identify potential permissive HLA locus mismatches and candidate markers for protecting against primary disease relapse by means of systematically and comprehensively assessing the impacts of both individual HLA locus mismatches and number of HLA locus mismatches on multiple outcomes, which is of great benefit for ascertaining acceptable HLA minimal mismatched grafts and for predicting prognosis of patients after unrelated-donor HCT.
RESULTS {#s2}
=======
Study and patient characteristics {#s2_1}
---------------------------------
The flow diagram of study search and selection was illustrated in Figure [1](#F1){ref-type="fig"}. The search strategy was showed in [Supplementary Table 1](#SD1){ref-type="supplementary-material"}. A total of 36 studies were included \[[@R16]--[@R51]\], of which, 15 studies analyzed 6 HLA loci \[[@R17], [@R18], [@R23], [@R24], [@R26], [@R28], [@R30], [@R31], [@R35], [@R42], [@R44]--[@R47], [@R50]\], 8 researches mentioned 5 HLA loci \[[@R16], [@R19]--[@R22], [@R27], [@R29], [@R38]\], and 14 studies investigated 4 HLA loci \[[@R25], [@R32]--[@R34], [@R37], [@R39]--[@R41], [@R43], [@R44], [@R48], [@R49], [@R51]\] (Table [1](#T1){ref-type="table"}). Study characteristics were showed in [Supplementary Table 2](#SD1){ref-type="supplementary-material"}, all of the included studies were at low risk of bias. Polymorphism and matching likelihood at individual HLA loci were showed in Figure [2](#F2){ref-type="fig"}. Class I HLA alleles are more polymorphic than Class II HLA alleles, which was also seen in terms of protein diversity. The highest mismatch likelihood was seen in patients with HLA-DPB1 and -C locus mismatches.
{#F1}
###### Patient, donor and transplantation characteristics according to number of HLA locus
Characteristic and stratum Total HLA 4 loci HLA 5 loci HLA 6 loci
---------------------------------------------- --------------- --------------- --------------- ---------------
**Number of studies** 36 14 8 15
**Patients, no. (%)** 100,072 (100) 47,837 (40.9) 10,932 (12.1) 41,303 (47.0)
**Patient age, median (range), y** 40.5 (0-81) 40.5 (0-81) 39.5 (0-79) 38.5 (0-77)
**Donor age, median (range), y** 41 (3-79) 39 (3-75) 48.5 (18-79) 35 (3-67)
**Disease at HCT, no. (%)**
Acute lymphoblastic leukemia 22,210 (22.2) 10,543 (22.0) 1,907 (17.4) 9,760 (23.6)
Acute myeloblastic leukemia 37,115 (37.1) 21,996 (46.0) 2,681 (24.5) 12,438 (30.1)
Chronic myeloid leukemia 21,027 (21.0) 8,514 (17.8) 3,704 (33.9) 8,809 (21.3)
Myelodysplastic syndrome 10,654 (10.6) 6,010 (12.6) 994 (9.1) 3,650 (8.8)
Lymphoid malignancy 2,289 (2.3) 43 (0.1) 652 (6.0) 1,594 (3.9)
Aplastic anemia 881 (0.9) 275 (0.6) 112 (1.0) 494 (1.2)
Multiple myeloma 519 (0.5) 2 (\<0.1) 280 (2.6) 237 (0.6)
Others 3,491 (3.5) 454 (0.9) 602 (5.5) 2,435 (5.9)
Missing 1,886 (1.9) 0 0 1,886 (4.6)
**Graft source, no. (%)**
Bone marrow 69,941 (69.9) 26,909 (56.3) 7,305 (66.8) 35,727 (86.5)
Peripheral blood 29,797 (29.8) 20,928 (43.7) 3,483 (31.9) 5,386 (13.0)
Missing 334 (0.3) 0 144 (1.3) 190 (0.5)
**Disease status at HCT, no. (%)**
Standard 41,857 (41.8) 24,523 (51.3) 3,572 (32.7) 13,762 (33.3)
High (intermediate and advanced) 47,092 (47.1) 21,906 (45.8) 6,647 (60.8) 18,539 (44.9)
Missing 11,123 (11.1) 1,408 (2.9) 713 (6.5) 9,002 (21.8)
**Performance status prior to HCT, no. (%)**
\<90 14,081 (14.0) 12,540 (26.2) 0 1,541 (3.7)
90-100 31,001 (31.0) 26,755 (55.9) 0 4,246 (10.3)
Missing 54,990 (55.0) 8,542 (17.9) 10,932 (100) 35,516 (86.0)
**Donor/recipient gender match, no. (%)**
Male to male 30,547 (30.5) 12,462 (26.0) 3,990 (36.5) 14,095 (34.1)
Male to female 17,655 (17.6) 7,948 (16.6) 2,330 (21.3) 7,377 (17.9)
Female to male 14,715 (14.7) 5,839 (12.2) 1,809 (16.5) 7,067 (17.1)
Female to female 14,399 (14.4) 5,868 (12.3) 1,800 (16.5) 6,731 (16.3)
Missing 22,756 (22.7) 15,720 (32.9) 1,003 (9.2) 6,033 (14.6)
**Donor/recipient CMV serostatus, no. (%)**
Negative/negative; −/− 15,474 (15.5) 11,192 (23.4) 105 (1.0) 4,177 (10.1)
Negative/positive; -/+ 14,736 (14.7) 11,622 (24.3) 74 (0.7) 3,040 (7.4)
Positive/positive; +/+ 9,265 (9.3) 6,726 (14.0) 68 (0.6) 2,471 (6.0)
Positive/negative; +/− 7,840 (7.8) 5,343 (11.2) 59 (0.5) 2,438 (5.9)
Missing 52,757 (52.7) 12,954 (27.1) 10,626 (97.2) 29,177 (70.6)
**Conditioning regimen, no. (%)**
Myeloablative 76,240 (76.2) 38,240 (79.9) 8,248 (75.4) 29,752 (72.0)
Reduced intensity 13,423 (13.4) 8,977 (18.8) 1,321 (12.1) 3,125 (7.6)
Missing 10,409 (10.4) 620 (1.3) 1,363 (12.5) 8,426 (20.4)
**Total body irradiation, no. (%)**
Yes 36,000 (36.0) 16,793 (35.1) 1,780 (16.3) 17,427 (42.2)
No 19,532 (19.5) 11,654 (24.4) 629 (5.7) 7,249 (17.6)
Missing 44,540 (44.5) 19,390 (40.5) 8,523 (78.0) 16,627 (40.2)
**GVHD prophylaxis, no. (%)**
Cyclosporine based 29,584 (29.6) 15,675 (32.8) 2,986 (27.3) 10,923 (26.4)
Tacrolimus based 34,734 (34.7) 23,205 (48.5) 210 (1.9) 11,319 (27.4)
Cyclosporine or tacrolimus based 9,391 (9.4) 3,009 (6.3) 0 6,382 (15.5)
Others 5,314 (5.3) 2,505 (5.2) 293 (2.7) 2,516 (6.1)
Missing 21,049 (21.0) 3,443 (7.2) 7,443 (68.1) 10,163 (24.6)
**T-cell depletion, no. (%)**
Yes 18,001 (18.0) 9,781 (20.4) 1,669 (15.3) 6,551 (15.9)
No 67,387 (67.3) 26,258 (54.9) 8,864 (81.1) 32,265 (78.1)
Missing 14,684 (14.7) 11,798 (24.7) 399 (3.6) 2,487 (6.0)
**Year of transplantation**
1988-2012 1988-2011 1988-2010 1988-2012
Abbreviations: HLA 4 Loci, HLA-A, -B, -C, and -DRB1 loci; HLA 5 Loci, HLA-A, -B, -C, DRB1 and -DQB1 loci; HLA 6 Loci, HLA-A, -B, -C, DRB1,-DQB1 and -DPB1 loci; HCT, hematopoietic cell transplantation; CMV, cytomegalovirus.
{#F2}
Acute GVHD {#s2_2}
----------
With respect to recipients with HLA-A, -B, -C, -DRB1, or -DPB1 locus mismatches, the risk of acute GVHD (III-IV) was significantly higher, with hazard ratios of 1.40 (95% CI, 1.28 to 1.54; *P* \< .001), 1.42 (95% CI, 1.24 to 1.62; *P* \< .001), 1.50 (95% CI, 1.33 to 1.69; *P* \< .001; I^2^ = 58.5%), 1.26 (95% CI, 1.14 to 1.40; *P* \< .001) and 1.24 (95% CI, 1.16 to 1.33), respectively, as compared to controls (Figure [3](#F3){ref-type="fig"}). However, HLA-DQB1 mismatches did not have a significant impact on acute GVHD (III-IV) (HR, 1.07; 95% CI, 0.95 to 1.20; *P* = .271) (Figure [3](#F3){ref-type="fig"}). The effect of individual HLA mismatches was replicated for acute GVHD (II-IV), with substantial heterogeneity in the analysis of HLA-DPB1 locus (I^2^ = 63.9%) (Figure [3](#F3){ref-type="fig"}). Secondly, we investigated the impact of nonpermissive HLA-DPB1 mismatches on aGVHD (Figure [4](#F4){ref-type="fig"}). Among patients with 10/10 HLA matching, nonpermissive HLA-DPB1 mismatches were associated with a significantly increased risk of acute GVHD (III-IV) (*P* \< .001), and had a trend of slight increasing risk of acute GVHD (II-IV) (*P* = .101). Conversely, matched HLA-DPB1 was significantly associated with decreased incidence of acute GVHD (II-IV) (*P* \< .001) and GVHD (III-IV) (*P* = .023). In the 9/10 HLA matching population, both nonpermissive mismatched and matched HLA-DPB1 did not have statistically significant impacts on grade II-IV or III-IV acute GVHD (all *P* \> .05). Thirdly, we assessed the risks of aGVHD for number of HLA locus mismatches (Figure [5](#F5){ref-type="fig"}). Compared with recipients with 8/8 HLA matching, those with 7/8 HLA matching had a higher risk of acute GVHD (II-IV) (*P* \< .001) and GVHD (III-IV) (*P* \< .001); those with 6/8 matches had a higher risk of acute GVHD (II-IV) (*P* \< .001), and had a trend of increased incidence of acute GVHD (III-IV) (*P* = .087; I^2^ = 78.0%). Only one study assessed the risk of the acute GVHD (II-IV) for 9/10 HLA matches, compared with 10/10 matches (*P* = .080).
{#F3}
{#F4}
{#F5}
Chronic GVHD {#s2_3}
------------
Patients with HLA-A locus mismatches had a higher risk of chronic GVHD (HR, 1.20; 95% CI, 1.04 to 1.39; *P* = .014; I^2^ = 50.4%), compared with the control (Figure [3](#F3){ref-type="fig"}). Similarly, HLA-C locus mismatches slightly increased hazard of chronic GVHD with a borderline significance (HR, 1.13; 95% CI, 1.01 to 1.27; *P* = .047; I^2^ = 67.2%) (Figure [3](#F3){ref-type="fig"}). However, mismatches at other HLA loci had no significant impact on chronic GVHD (all *P* \> .05) (Figure [3](#F3){ref-type="fig"}). Secondly, both nonpermissive mismatched and matched HLA-DPB1 had no impact on the incidence of chronic GVHD (all *P* \> .05) (Figure [3](#F3){ref-type="fig"}). Thirdly, compared with 8/8 HLA matches, neither 7/8 nor 6/8 matches showed a higher risk of chronic GVHD (*P* = .213 and .522, respectively).
Neutrophil engraftment {#s2_4}
----------------------
As shown in Figure [3](#F3){ref-type="fig"}, there was a trend of promoting neutrophil engraftment for recipients with individual HLA locus mismatches. Recipients with 7/8 HLA matching did not have an impact on neutrophil engfratment, compared with 8/8 matching (Figure [5](#F5){ref-type="fig"}).
Relapse {#s2_5}
-------
Mismatches at HLA-DPB1 locus was significantly associated with a decreased risk of primary disease relapse, compared with the control (HR, 0.74; 95% CI, 0.68 to 0.80; *P* \< .001) (Figure [3](#F3){ref-type="fig"}). HLA-C locus mismatches has a trend of decreased relapse (HR, 0.84; 95% CI, 0.69 to 1.03; *P* = .102; I^2^ = 70.9%) (Figure [3](#F3){ref-type="fig"}). Mismatches at HLA-A, -B, -DRB1 or DQB1 locus had no significant impact on disease relapse (all *P* \> .05) (Figure [3](#F3){ref-type="fig"}). Secondly, in the population with 10/10 HLA matching, nonpermissive HLA-DPB1 mismatches had a trend of reduced disease relapse, (*P* = .080) (Figure [4](#F4){ref-type="fig"}), whereas matched HLA-DPB1 had a higher risk of relapse (*P* \< .001) (Figure [4](#F4){ref-type="fig"}). That the impact of nonpermissive mismatched and matched HLA-DPB1 on disease relapse was not observed among the patients with 9/10 HLA matching (*P* = .900 and *P* = 0.736) (Figure [4](#F4){ref-type="fig"}). Thirdly, the impact on relapse was not observed in patients with 9/10, 7/8 and 6/8 HLA matching (*P* = .516, .960 and .360, respectively) (Figure [5](#F5){ref-type="fig"}).
TRM, mortality and DFS {#s2_6}
----------------------
For recipients with mismatches at HLA-A, -B, -C -DRB1 or -DQB1 locus, the risk of TRM was significantly higher, as compared to controls, with hazard ratios of 1.47 (95% CI, 1.26 to 1.71; *P* \< .001), 1.54 (95% CI, 1.29 to 1.83; *P* \< .001), 1.35 (95% CI, 1.20 to 1.51; *P* \< .001), 1.29 (95% CI, 1.02 to 1.63; *P* = .033) and 1.30 (95% CI, 1.01 to 1.67; *P* = .041), respectively (Figure [3](#F3){ref-type="fig"}). Whereas, HLA-DPB1 mismatches had no significant impact on TRM (*P* = .591) (Figure [3](#F3){ref-type="fig"}). With respect to mortality, intriguingly, similar results were observed for HLA-A, -B, -C, -DRB1 or DPB1 mismatches, with hazard ratios of 1.33 (95% CI, 1.27 to 1.40; *P* \< .001), 1.35 (95% CI, 1.21 to 1.50; *P* \< .001), 1.23 (95% CI, 1.17 to 1.29; *P* \< .001), 1.19 (95% CI, 1.07 to 1.32; *P* = .033) and 1.03 (95% CI, 0.97 to 1.09; *P* = .460), respectively (Figure [3](#F3){ref-type="fig"}). Whereas HLA-DQB1 locus mismatches had no significant impact on mortality (*P* = .460), which is inconsistent with it for TRM (Figure [3](#F3){ref-type="fig"}). These data demonstrated that the pooled point estimates of class I HLA loci were prone to be greater than those of class II HLA loci, with respect to TRM and mortality. Additionally, the effect of mismatches at HLA-A, -B, -C, -DRB1 or -DQB1 locus was replicated for DFS, with few studies investigating this end point (Figure [3](#F3){ref-type="fig"}).
Secondly, we investigated the impact of nonpermissive HLA-DPB1 mismatches on TRM and mortality (Figure [4](#F4){ref-type="fig"}). In the 10/10 HLA matching population, the risk of TRM and mortality was significantly greater for nonpermissive HLA-DPB1 mismatches (*P* \< .001 and *P* \< .001, respectively). In contrast, matched HLA-DPB1 had a marginally significant effect of protecting against TRM (*P* = .029). But for mortality, the impact of matched HLA-DPB1 was not identified (*P* = .986). Among the 9/10 HLA matching patients, matched HLA-DPB1 did not result in a decreased risk of TRM and mortality (*P* = .313 and *P* = .259, respectively); similarly, the impact on TRM was replicated in patients with nonpermissive HLA-DPB1 mismatches (*P* = .144), but increasing risk of mortality was observed for the mismatches, with a borderline significance (HR, 1.10; 95% CI, 1.01 to 1.20; *P* = .033).
Thirdly, as shown in Figure [5](#F5){ref-type="fig"}, compared with 10/10 matching, there was a significantly increased risk of TRM and mortality for both 9/10 and 8/10 HLA mismatches. Furthermore, the pooled point estimate of 8/10 HLA mismatches was greater than it of 9/10 mismatches. The findings were replicated in both 7/8 and 6/8 HLA mismatches, compared with 8/8 HLA matching. And similar results were observed in terms of DFS.
Stratified analyses {#s2_7}
-------------------
Stratified analysis was showed in [STables S3-5](#SD1){ref-type="supplementary-material"} according to combinations of HLA allele or antigen mismatches. Only one study analyzed 1 or 2 antigen mismatches so that the pooled analysis could not be performed \[[@R38]\].
DISCUSSION {#s3}
==========
Summary main results {#s3_1}
--------------------
We found HLA-DQB1 locus mismatches had no significant impact on multiple outcomes except for TRM, it is a potential candidate of permissive HLA locus mismatches. Secondly, we attempted to identify several candidates serving as remarkable graft-*versus*-tumor effects (GVT). HLA-DPB1 locus mismatches had a significantly protective effect against leukemia relapse, which was attributed to nonpermissive HLA-DPB1 mismatches. Meanwhile, mismatched HLA-DPB1 had no significant impact on chronic GVHD, TRM or mortality. But relative to permissive mismatches, nonpermissive HLA-DPB1 mismatches had a significantly increased risk of TRM and mortality in 10/10 HLA matching. Similarly, HLA-C locus mismatches had a trend of reduced risk of relapse, but had a significant increased risk of TRM and mortality and a slightly increased risk of chronic GVHD. Thirdly, mismatches at HLA-A, -B, -DRB1 loci significantly increased the risks of acute GVHD, TRM and mortality, but had no significant protection against primary disease relapse.
Agreements and disagreements with other studies {#s3_2}
-----------------------------------------------
Most recently, one meta-analysis demonstrated that 9/10 HLA matching had a higher risk of mortality, compared with 10/10 matching, with hazard ratio of 1.27 (95% CI, 1.12 to 1.45; *P* \< .001) \[[@R56]\], which was similar to ours (HR, 1.31; 95% CI, 1.14 to 1.50; *P* = .001). In addition, the risk for individual HLA allele mismatches was similar with it in our stratified analyses. However, multiple comparisons for other important outcomes were not performed in the pool-analysis. Many low-quality studies with small sample size were included, which did not meet our eligible criteria. To our knowledge, our meta-analysis is the first to systematically and comprehensively assess the impact of HLA locus mismatches on clinical outcomes in unrelated donor HCT.
Strengths and limitations of this study {#s3_3}
---------------------------------------
Our meta-analysis had several strengths. Firstly, a large number of patients were included to obtain a bigger statistical power for a given comparison. Secondly, a series of end points were assessed in an effort to obtain a comprehensive recognition about the effect of individual HLA locus mismatches. Thirdly, the risk of individual HLA locus mismatches was similar among analogous end points, which contributed to the robustness of pooled estimates. For instance, mismatches at HLA-DQB1 locus had no impact on both acute GVHD (II-IV) and acute GVHD (III-IV). Furthermore, HLA-DQB1 locus mismatches were better tolerated than other HLA loci for acute GVHD. In terms of TRM, the pooled point estimates of class I HLA molecules were greater than those of class II HLA molecules, as observed for mortality. Fourthly, given the highest mismatched likelihood of HLA-DPB1 alleles, we explored the impact of nonpermissive HLA-DPB1 locus mismatches on multiple end points. Fifthly, we assessed the impact of number mismatches of HLA loci on outcomes. Sixthly, Donor-recipient baseline characteristics were summarized together in order to demonstrate the practical application field of pooled results.
There were several limitations in our meta-analysis. First of all, all of our main pooled estimates belonged to average effects for individual HLA loci, with combining results from different studies presented separately as 1 allele mismatching, 1 or 2 allele mismatching or a single mismatching. Few studies investigated the effect of 2 antigen mismatches at individual HLA loci \[[@R38]\]. Second, clinical heterogeneity from individual studies such as donor age, patients' performance status, primary disease, disease status at HCT, intensity of conditioning regimen and GVHD prophylaxis, grafts with T cell depletion, were difficult to be completely balanced between cases and controls, especially in studies with relatively small sample size \[[@R17], [@R20], [@R22], [@R32], [@R45]\]. Third, bone marrow transplantation reduced the risk of chronic GVHD but increase the risk of graft failure, compared with peripheral blood transplantation \[[@R57], [@R58]\]. In our study, 69.9% patients received bone marrow derived hematopoietic cells, which might attenuate the risk of individual HLA locus mismatches for chronic GVHD and neutrophil engraftment. Similarly, the inclusion of anti-thymocyte globulin (ATG) into conditioning regimen for patients with leukemia resulted in significantly decreased risk of chronic GVHD after allogeneic transplantation \[[@R59], [@R60]\], but the effect of HLA locus mismatching was not analyzed according to the application of ATG in included studies. Fourth, with respect to the same primary disease, the therapy strategy has been evolving over time, which might decrease the incidence of complications \[[@R17], [@R61]\]. Fifth, HLA genes match likelihood at high resolution varied across different race and ethnicity groups \[[@R62]\], detailed information was not presented in many studies. Sixth, HLA locus mismatching had a higher risk of mortality in recipients with standard-risk disease compared those with high-risk disease \[[@R19], [@R27]\]. In our meta-analysis, less than half of the population was standard-risk disease status at HCT. Seventh, significant heterogeneity mainly existed in the mismatches at the HLA-A, -C and -DPB1 loci. We conducted a sensitivity analysis to figure out the robustness of pooled results. As shown in [Supplementary Figure S1](#SD1){ref-type="supplementary-material"}, using the trim and fill method, the robustness of all of the pooled estimates were presented in both fixed and random-effects models \[[@R55]\]. It is notable that HLA-C locus mismatches had a significantly reduced risk of relapse compared with matched control (*P* \< .001 and *P* = .017 respectively). We stratified the HLA-C locus mismatches into three groups: 1 allele, 1 antigen and 1 or 2 allele mismatches, the hazard ratios of relapse were 0.88 (95% CI, 0.70 to 1.10; *P* = .259; I^2^ = 33.7%), 1.04 (95% CI, 0.91 to 1.20; *P* = .531; I^2^ = 0.0%) and 0.70 (95% CI, 0.62 to 0.79; *P* \< .001; I^2^ = 0.0%), respectively. The heterogeneity appeared to be found, but results of the subgroups were less robust for fewer studies used in each pooled analysis. To explore the heterogeneity of HLA-DPB1 mismatches for acute GVHD (II-IV), we excluded 2 studies with small sample size and 1 study with GVH direction mismatches, and then pooled the remaining results, with hazard ratio of 1.34 (95% CI, 1.27 to 1.42; *P* \< .001; I^2^ = 0.0%), which was consistent with it from the primary analysis. We failed to reveal possible sources of heterogeneity for TRM at HLA-DPB1 locus, and for chronic GVHD and acute GVHD (III-IV) at HLA-C locus. Eighth, we were unable to assess publication bias because of relatively few studies for most of the end points. Ninth, specific HLA genotype mismatching combination among donor-recipient pairs was investigated in few studies, the pool-analysis could not be performed \[[@R31], [@R43]\]. Last but not least, in the absence of more detailed information, predefined subgroup analyses could not be conducted.
Assessment of HLA locus mismatches in terms of expression levels or amino acid substitutions {#s3_4}
--------------------------------------------------------------------------------------------
Most recent studies attempted to identify permissive mismatches in terms of expression levels of HLA-C and -DPB1 \[[@R63], [@R64]\]. With respect to HLA-C molecule, patients with higher expression of mismatched HLA-C tended to suffer from higher risk of actue GVHD (III-IV) and nonrelapse mortality compared with those with lower expression mismatches. Furthermore, the definitive correlation of most HLA-C allotypes with their corresponding expression levels will be beneficial to the selection of permissive mismatched donors in terms of HLA genotype. A similar finding was identified in HLA-DPB1 mismatches. When donors with low-expression HLA-DPB1 molecules, recipients with mismatched high-expression HLA-DPB1 had a higher risk of acute GVHD (II-IV), compared with patients with low-expression HLA-DPB1 mismatches. It is found that high-expression HLA-DPB1 correlated with its single- nucleotide variant (rs9277534G) of the sixth exon in the 3' untranslated region. In contrast, HLA-DPB1 expression was lower when with the rs9277534A variant. A possible explanation is that non-coding RNA might mediate the HLA-DPB1 RNA silencing through binding rs9277534A. In addition, some studies attempted to identify specific nonpermissive HLA locus mismatches according to amino acid substitutions (AAS) at key peptide-binding residues of HLA molecules. For example, among the population with a single HLA-C mismatches, ASS at position 116 had a significantly increased risk of acute GVHD (III-IV), compared with those without the AAS \[[@R65]\].
MATERIALS AND METHODS {#s4}
=====================
This meta-analysis is reported according to the PRISMA statement \[[@R66]\]
Eligibility criteria {#s4_1}
--------------------
Studies should be included when meeting the following criteria: (1) patients receiving unrelated-donor HCT; (2) patients with hematological disorders; (3) high-resolution typing was performed as described for HLA-A, -B, -C, -DRB1, -DQB1, and/or -DPB1 loci; (4) investigating the impact of HLA locus mismatching on clinical outcomes; (5) cohort studies. If a study meet any of the following criteria, it should be excluded: (1) HLA locus mismatch combinations; (2) analysis of HLA protein expression; (3) mismatched HLA alleles as controls except for HLA-DPB1; (4) data presented as percentage; (5) unrelated *versus* related donor HCT; (6) meeting abstract or case report.
Study searching and selection {#s4_2}
-----------------------------
We searched four databases (PubMed, Embase, Web of Science and the Cochrane Library) from inception to February 2016, with these keyword combinations involving "hematopoietic", "hematologic" or "transplantation"; "unrelated"; "human leukocyte antigen" or "major histocompatibility complex"; and "mismatch" or "mismatched". The complete search strategy is available in the appendix ([Supplementary Table 1](#SD1){ref-type="supplementary-material"}). Two reviewers (R.T. and T.Z.) independently selected studies based on the eligibility criteria. Disagreements were resolved through discussing with a third reviewer (B.Y.).
Definition of end points {#s4_3}
------------------------
Primary end points included grade II to IV aGVHD, grade III to IV aGVHD, chronic GVHD (cGVHD), neutrophil engraftment and disease relapse. The incidence of grades II-IV or III-IV acute GVHD was defined according to the Glucksberg scale \[[@R67]\]. Chronic GVHD included limited and extensive conditions and was defined according to the Seattle criteria \[[@R68]\]. Neutrophil engraftment was defined achieving an absolute neutrophil count \> 0.5×10^9^/L for 3 consecutive days after transplantation. Relapse was regarded as recurrence of primary leukemia or myelodysplastic syndrome (MDS). Secondary end points were as follows: transplant-related mortality (TRM), Mortality and disease-free survival (DFS). Overall mortality was defined as time from HCT to death from any cause. TRM was death without evidence of primary disease recurrence after HCT. DFS was defined as time to relapse of primary disease or death from any cause.
Risk of bias within and across studies {#s4_4}
--------------------------------------
Risk of bias within studies was assessed independently by the two authors (R.T. and T.Z.) using the Newcastle-Ottawa Scale (NOS) components for cohort studies \[[@R69]\]. Studies were scored to be low risk of bias (≥ 3 points) or higher risk of bias ( \< 3 points). Disagreement was resolved through discussing with a third author (B.Y.). Publication bias was not investigated, because of less than 10 studies included for most of the end points.
HLA typing {#s4_5}
----------
In our meta-analysis, high-resolution HLA typing refers to obtaining diversity of the allele sequence at individual HLA loci for donor-recipient pairs, using various methods such as sequencing based typing \[[@R70]\], sequence specific priming \[[@R71]\], reference strand conformation analysis \[[@R72]\], sequence specific oligonucleotide probing \[[@R73]\] and so on. Low-resolution HLA (antigen or serologic level) disparities are derived from converting high-resolution typing to its corresponding serologic equivalents, except for a few HLA-B alleles mapping to their specific equivalents \[[@R74]\].
HLA matching {#s4_6}
------------
Whenever assessing the effect of individual HLA locus mismatches, we predefined controls as patients with corresponding HLA allele matching adjusted with other HLA allele matching, or those with complete HLA allele matching. Individual HLA locus mismatches involved allele-level (1 or 2 alleles), antigen-level (1 or 2 antigens) and/or a single (1 allele or 1 antigen) mismatches, as presented in studies. Secondly, there were at most 6 HLA loci (HLA-A, HLA-B, HLA-C, HLA-DRB1, HLA-DQB1 and HLA-DPB1) investigated in some studies. High-resolution matching at 5 loci (except for HLA-DPB1) is designated as 10/10 HLA matching, HLA 9/10 matching refers to donor-recipient pairs with a single mismatches at any one of the 5 loci, HLA 8/10 matching includes two allele or antigen mismatches at one or two of the 5 HLA loci; and high-resolution matching at 4 loci (except for HLA-DPB1 and HLA-DQB1) is designated as 8/8 HLA matching, HLA 7/8 matching was defined as donor-recipient pairs with a single mismatches at any one of the 4 loci, HLA 6/8 matching includes two mismatch (allele or antigen) at one or two of the 4 HLA loci. Additionally, in a population with HLA-DPB1 allele mismatches, permissive HLA-DPB1 mismatches are T-cell-epitope group matches, whereas nonpermissive HLA-DPB1 mismatches belong to T-cell-epitope group mismatches in either graft-*versus*-host or host-*versus*-graft direction \[[@R30], [@R35], [@R42]\].
Data extraction {#s4_7}
---------------
Data were extracted as follows: (1) the baseline characteristics of donor-recipient pairs and individual studies; (2) data presented as multivariate-adjusted point estimates and corresponding 95% CIs for each comparison. Among the included studies, some data were showed as HR, others were presented as relative risk (RR) or odds ratio (OR). We uniformed effect measure as HR in our meta-analysis. Two reviewers (R.T. and T.Z.) independently extracted these data using a spreadsheet developed specifically for the meta-analysis. Discrepancies were resolved through discussing with a third reviewer (B.Y.).
Data synthesis and analysis {#s4_8}
---------------------------
Primary analyses compared mismatches at individual HLA loci with corresponding controls. Separately, we assessed the impact of nonpermissive mismatches and matches at HLA-DPB1 locus on multiple end points, as compared to permissive HLA-DPB1 mismatches. Secondary analyses evaluated the impact of number of HLA locus mismatches on multiple end points. In addition, we summarized the baseline characteristics of included studies and patients separately, and calculated the matching likelihood of individual HLA loci based on a larger population. We pooled HRs and 95% CIs using the Mantel-Haneszel random-effects model for each comparison \[[@R75], [@R76]\]. The magnitude of between-studies heterogeneity was assessed using the I^2^ statistic, with value ≥ 50% indicating substantial heterogeneity, I^2^ value \< 50% was not showed in the text \[[@R77]--[@R79]\].
Sensitivity and subgroup analyses {#s4_9}
---------------------------------
Firstly, with respect to pooled estimates with a substantial heterogeneity, we performed a sensitivity analysis using the trim and fill adjustment method (random and fixed effects linear estimator) in an effort to investigate the robustness of primary synthesized results \[[@R80]\]. Secondly, we did stratified analysis according to mismatched level (allele and/or antigen) at individual HLA loci. Additional subgroups analyses should be performed according to the quality of studies, direction of HLA locus mismatching, patient or donor age, disease status before HCT, intensity of conditioning regimen, grafts with T-cell depletion, GVHD treatment, cytomegalovirus serostatus of donor-recipient pairs, duration of follow up.
All statistical tests were 2-sided, *P* value \< .05 was considered statistically significant. The meta-analysis was performed using STATA/SE version 12.0.
CONCLUSIONS {#s5}
===========
We identify HLA-DQB1 locus mismatches as a permissive mismatching, which offers HCT choices for patients without all 5 HLA-locus matched grafts. HLA-DPB1 locus nonpermissive mismatches have a significantly protective effect against leukemia relapse, simultaneously have no significantly increased risk of TRM, mortality or DFS. HLA-C locus mismatches have a trend of protecting against leukemia relapse. Further researches should be conducted to confirm our findings using individual patient data meta-analysis, and should assess the impacts of individual HLA locus mismatches on multiple outcomes in terms of the allele and antigen levels respectively. Subgroup analysis should be performed according to disease category, disease status at HCT, intensity of conditioning regimen, T-cell depletion or not, and so on. More studies are needed to identify and verify permissive mismatches at HLA-C and DPB1 loci in unrelated donor HCT.
SUPPLEMENTARY MATERIALS TABLES {#s6}
==============================
**GRANT SUPPORT**
This work was supported by the National Natural Science Foundation of China (Grant No. 81471582, No. 81370647 and No. 81520108002).
**CONFLICTS OF INTEREST**
The authors declare no competing financial interests.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1}
===============
Guillain-Barré syndrome (GBS) is a heterogenous neuropathy characterized by rapid bilateral limb paresis after a triggering immunological event. GBS patients always achieve a maximum deficit within 4 weeks of onset, and most achieve nadir weakness within 2 weeks \[[@B1], [@B2]\]. The evaluation of GBS is largely based on its clinical presentation, cerebrospinal fluid analysis, and electrophysiology \[[@B3]\]. However, the optimal timing for lumbar puncture and electrophysiological examination is at least 1 and 2 weeks after the onset of weakness, respectively \[[@B4], [@B5]\]. Thus, biomarkers for the early prediction of clinical course and outcome are urgently needed.
To monitor disease progression and improve the prognosis of GBS, several markers have been identified as predictors for disease severity and outcomes, including serum albumin levels for monitoring treatment response \[[@B6]\], blood glucose levels for short-term prognosis \[[@B7]\], serum sodium levels for outcomes 1 year after onset \[[@B8]\], and plasma cortisol levels for the prediction of respiratory failure \[[@B9]\]. Compared with these markers, folate plays a more fundamental role in polyneuropathy \[[@B10]\]. Therefore, serum folate level could also be a biomarker for disease evaluation and prediction of GBS.
The primary objectives of this study were to determine whether serum folate levels can be a prognostic marker in GBS patients. We investigated the prevalence of folate deficiency in GBS patients and analyzed the association between serum folate levels on admission and GBS disease severity at nadir.
2. Methods {#sec2}
==========
2.1. Ethical Statements {#sec2.1}
-----------------------
This retrospective study was approved by the Ethics Committee of the First Hospital of Jilin University and all patients provided signed informed consent. All data were obtained from the electronic medical records at our hospital, and the patients were anonymized.
2.2. Subjects {#sec2.2}
-------------
This study included two groups of patients with recorded serum folate levels: (1) GBS patients admitted to the Department of Neurology of the First Hospital of Jilin University between January 2014 and June 2017 and (2) healthy controls who attended the Physical Examination Section of the First Hospital of Jilin University in 2017. Subjects in the control group were matched for sex and age (within a 5-year margin) with GBS patients in a 1:1 ratio. The following GBS patients were excluded: (1) those aged \<18 years; (2) those with Miller Fisher syndrome or chronic inflammatory demyelinating polyradiculoneuropathy; and (3) those with a hospital stay of \<5 days.
Motor function deficits of the patients were monitored during the study using the Medical Research Council (MRC) sum score \[[@B11]\] ranging from 0 (quadriplegia) to 60 (normal strength) and the GBS disability scale \[[@B12]\] ranging from 0 (healthy) to 6 (dead). The nadir of GBS was defined as the lowest MRC sum score \[[@B11]\]. The progression duration was defined as the number of days between the onset of symptoms and weakness at nadir \[[@B13]\]. A faster disease progression was defined as a peak illness less than or equal to seven days from onset.
2.3. Folate Measurement {#sec2.3}
-----------------------
Serum folate levels were determined by radioimmunoassay (UniCel DxI 800, Beckman Coulter Inc., USA). Folate deficiency was defined as serum folate of \<3.5 ng/mL \[[@B14]\]. In normal folate group, high-normal and low-normal levels were defined as \>6.0 g/mL and 3.5-6.0 ng/mL, respectively \[[@B15]\].
2.4. Statistics {#sec2.4}
---------------
IBM SPSS Statistics for Windows, version 21.0 software (IBM Corp., USA), and GraphPad Prism 7.0 (GraphPad Software Inc., USA) were used for data and used analysis. Case-control matching in SPSS was used to match GBS patients and healthy controls based on gender and age. Shapiro-Wilk tests were used to test the normality of the distributions. Univariate analysis was conducted using Mann--Whitney U tests (for disease duration, MRC score, GBS disability scale, laboratory results, and other variables that were not normally distributed), Kruskal-Wallis tests (for comparison of disease duration among the three groups), Student\'s t-tests (for age; normally distributed), Pearson Chi-square or Fisher\'s exact tests (for proportions), and univariable regression analysis (for identification of the predictors of disease progression). Correlations were expressed by Spearman rank correlation coefficients (r). Binary logistic regression was performed with backward selection to identify the independent predictors of disease progression duration. All tests were two-tailed, with a level of significance set to a*p* value of \<0.05.
3. Results {#sec3}
==========
3.1. Serum Folate Levels in GBS Patients and Healthy Controls {#sec3.1}
-------------------------------------------------------------
A total of 112 pairs of sex- and age-matched GBS patients and healthy subjects were included. Folate deficiency was present in 21 (18.9%) of GBS patients and two (1.8%) of the control group. Serum folate levels in the GBS patients were significantly lower than those in healthy controls ([Table 1](#tab1){ref-type="table"}).
3.2. Characteristics of GBS Patients with Folate Deficiency {#sec3.2}
-----------------------------------------------------------
The 112 GBS patients were divided into two groups depending on whether the serum folic acid level was \<3.5 ng/mL. The clinical features of the patients with or without folate deficiency are presented in [Table 2](#tab2){ref-type="table"} and [Table S1](#supplementary-material-1){ref-type="supplementary-material"}. Only two patients (9.5%) in the folate deficiency group were female, while the proportion of female patients without folate deficiency was 50.5% (*p* \< 0.001). Patients with normal folate levels had a shorter GBS progression duration than those with folate deficiency (median progression duration: 6 days \[IQR 4-11\] versus 13 days \[7-18\], *p* \< 0.001).
3.3. Association between Serum Folate Level and the Clinical Severity of GBS {#sec3.3}
----------------------------------------------------------------------------
As [Table 3](#tab3){ref-type="table"} shows, there was a significant correlation between serum folic acid level and disease progression duration (r = -0.261, *p* = 0.005). When looking at the correlation by extremities (upper versus lower limbs), the analysis showed a significant correlation between the serum folate level and the MRC score of upper limbs at nadir (r = -0.208, *p* = 0.03) but no correlation between the MRC score or GBS disability score at nadir (r = -0.172, *p* = 0.07; r = 0.114, *p* = 0.23).
3.4. Baseline Folate Level Predicts the Length of GBS Progression and Disability at Nadir {#sec3.4}
-----------------------------------------------------------------------------------------
Univariate regression analysis was performed to validate the relationship between the serum folate level and GBS disease progression ([Table 4](#tab4){ref-type="table"}). Besides folate deficiency, four other predictors of GBS progression length were also identified: superficial sense deficit, deep sensation deficit, dyspnea on admission, and MRC sum score on admission (all *p* \< 0.05). Fasting plasma glucose level, antecedent events, electrophysiology type, history of hypertension or diabetes, and GBS disability score on admission were not significant. The laboratory results of cerebrospinal fluid examinations were not included in the analysis because lumbar punctures were not performed immediately upon admission. To further determine the independent predictors of disease progression, age and gender were also included in the logistic regression analysis model ([Table 4](#tab4){ref-type="table"}). Multivariable analysis showed that GBS patients without folate deficiency were six times more likely to have a faster progression compared to those with folate deficiency (odds ratio \[OR\], 6.04; 95% confidence interval \[CI\], 1.69-21.61; *p* = 0.006).
To further categorize patients with different folate levels on admission, we divided those with normal levels of folate into two groups. As shown in [Figure 1(a)](#fig1){ref-type="fig"}, the differences in the medians of progression days between high-normal versus low (*p* = 0.001) or high-normal versus low-normal tertiles (*p* = 0.026) were statistically significant, indicating the value of serum folate level on admission in predicting the length of progression. [Figure 1(b)](#fig1){ref-type="fig"} shows that patients with baseline folate levels of \<4.0 ng/mL had a higher MRC score of the upper limbs at nadir compared to the score in those with folate levels of \>4.0 ng/mL (*p* = 0.045).
4. Discussion {#sec4}
=============
Serum folate levels have been widely measured but rarely studied in GBS patients. We confirmed the occurrence of folate deficiency in GBS patients and identified it as an independent predictor of disease progression duration, together with age, deep sensation, and baseline MRC score. Specifically, higher folate level has been shown to predict a faster disease progression and a worse strength of the upper limbs at nadir.
There are several possible explanations for our finding of insufficient folate levels in GBS patients. Folate deficiency has been shown to diminish human immune functions by affecting T and B cell differentials as well as the proliferation response of lymphocytes \[[@B16], [@B17]\]. Consequently, it is possible that patients with low folate levels are vulnerable to infections, which may provoke GBS. However, studies have shown that folate supplementation has no effect on infection reduction \[[@B18]--[@B21]\]. This suggests a poor association between deficient folate levels and infection susceptibility. On the other hand, GBS as an immune response to infection and infection itself may also cause folate deficiency because these events with rapid cell proliferation could lead to an increased need for folate \[[@B22]\]. We also noticed the impact of gender on folate deficiency in GBS. In general, serum folate level is higher in women than in men \[[@B23]\], which could explain why only two women in our study had folate deficiency. Further, the safety and efficacy of folic acid supplementation in GBS should be carefully evaluated in future studies, as randomized trials have suggested the folic acid may increase the possibility of neoplasms \[[@B24]--[@B27]\].
The results of the present study revealed a significant association between folate level on admission and the duration of GBS progression. However, the role of folate deficiency in GBS is unclear. Folate is essential to the peripheral nerves and, and in rare cases, its shortage may cause axonal sensory polyneuropathy \[[@B10]\]. Folate-deficient GBS and folate deficiency neuropathy are both slowly progressive compared to normal folate GBS and thiamine-deficiency neuropathy, respectively. However, this is probably a coincidence because their mechanisms are entirely different. One possible explanation for this association is that deficient folate levels may depress the immune response in GBS and retard the disease progression due to its crucial role in DNA synthesis \[[@B28], [@B29]\]. Similarly, the potentially concurrent deficiency of other vitamins may also explain the findings \[[@B16], [@B17]\]; however, this requires further study. Also, the effect of GBS variants cannot be ignored, but our analysis did not prove the roles of antecedent infection and electrophysiology and made GBS forms appear less relevant. Folate level also showed a less strong association with the strength of the upper limbs at nadir; thus, its underlying mechanisms require consideration.
Our study had several limitations. Because this study was retrospective in nature and folate levels were measured only on admission but not at the onset or at nadir, the causation between baseline folate levels and the progression duration require validation. In addition, the plateau phase and recovery phase durations were not available in this study, limiting the investigation of folate\'s role in the complete clinical course.
5. Conclusions {#sec5}
==============
In summary, we demonstrated that serum folate level is an independent marker associated with the duration of GBS progression. The roles of folate in GBS pathogenesis and prognosis need to be explored in prospective studies to monitor dynamic serum folate levels.
This study was supported by funding from Jilin University.
Data Availability
=================
The data generated or analyzed during this study are included in this article and the supplementary files.
Conflicts of Interest
=====================
The authors declared that they have no conflicts of interest.
Supplementary Materials {#supplementary-material-1}
=======================
######
Table S1 shows the clinical characteristics of GBS patients with and without folate deficiency.
######
Click here for additional data file.
{#fig1}
######
Demographics characteristics and serum folate levels.
Group GBS (n = 112) HCs (n = 112) *p*
---------------- ------------------ ------------------- ----------
Age (year) 52.23 (13.61) 51.16 (12.85) 0.55
Female (N, %) 48 (42.9%) 48 (42.9%) 1.00
Folate (ng/mL) 5.34 (3.98-7.99) 9.46 (5.93-12.78) \< 0.001
HCs, healthy controls.
######
Characteristics of GBS patients with and without folate deficiency^a^.
Group GBS with folate deficiency (n = 21) GBS without folate deficiency (n = 91) *p*
------------------------------------- ------------------------------------- ---------------------------------------- -------
**Demographics**
Age, mean (SD) 53.48 (17.20) 51.95 (12.73) 0.70
Female 9.5% (2/21) 50.5% (46/91) 0.001
**Weakness at nadir, median (IQR)**
GBS disability score 3 (2-4) 4 (3-4) 0.61
MRC score 48 (28.5-50) 40 (24-48) 0.30
**Disease course, median (IQR)**
Hospital days 15 (12-19) 13 (11-18) 0.55
Ventilator days (n = 10) 13 (8-13) 27 (14-101) 0.27
Progression duration, d 13 (7-18) 6 (4-11) 0.006
MRC, Medical Research Council.
^a^Data are presented as percentage of patients unless otherwise indicated. All items are shown for 112 patients unless otherwise specified.
######
Correlation of serum folate level with the severity of GBS.
Correlation Spearman r *p*
-------------------------- ------------ -------
Progression duration -0.261 0.005
Length of stay 0.135 0.18
GBS disability score 0.114 0.23
MRC score at nadir -0.172 0.07
MRC score in upper limbs -0.208 0.03
MRC score in lower limbs -0.118 0.22
MRC, Medical Research Council
######
Predictors of a faster GBS progression, defined as reaching nadir weakness in 1 week from onset.
Univariate analysis Multivariable analysis
---------------------------------------- --------------------- ------------------------ -------------------- -------
Age 0.97 (0.95, 1.00) 0.07 0.96 (0.93, 0.99) 0.02
Gender (Female vs. male) 1.16 (0.55, 2.47) 0.70 \- \-
Superficial sense deficit (no vs. yes) 2.68 (1.24, 5.80) 0.01 \- \-
Deep sensation deficit (no vs. yes) 5.13 (1.33, 19.82) 0.02 5.17 (1.16, 23.09) 0.03
Dyspnea on admission (no vs. yes) 0.31 (0.10, 0.91) 0.03 \- \-
MRC score on admission 0.96 (0.93, 0.99) 0.005 0.94 (0.90, 0.98) 0.001
Folate deficiency (no vs. yes) 4.19 (1.49, 11.83) 0.007 6.04 (1.69, 21.61) 0.006
OR, odds ratio; CI, confidence interval; MRC, Medical Research Council
[^1]: Academic Editor: Himanshu Garg
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .43}
{#sp2 .44}
{#sp3 .45}
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
Recent droughts across the United States have resulted in loss of life and billions of dollars in damage ([@R1]--[@R3]), making drought forecasting and planning a high societal priority. Yet, the paleoclimate record shows that these droughts pale in comparison to the megadroughts of the Common Era (C.E.), many of which appear to be longer lasting than the famed "Dust Bowl" of the 1930s ([@R4], [@R5]) and had similarly marked societal impacts. For instance, the drought of 1276 to 1299 C.E. likely contributed to the migration of the Anasazi people out of the American southwest near the beginning of the 14th century ([@R5], [@R6]). Because continued greenhouse gas emissions are projected to increase drought risk over much of the United States and other regions ([@R7]--[@R9]), understanding the dynamical causes of past droughts is of paramount importance.
U.S. droughts have been associated with changes in the El Niño/Southern Oscillation (ENSO) ([@R4], [@R10]--[@R12]), sea surface temperatures (SSTs) in the Atlantic Ocean ([@R10], [@R11], [@R13], [@R14]), internal atmospheric dynamics ([@R15]--[@R17]), and greenhouse forcing ([@R2]). Variations in the Atlantic Multidecadal Oscillation (AMO) have also been suggested as an influence on past "megadroughts" ([@R15]), especially in conjunction with La Niña--like SSTs ([@R18]--[@R20]). Much of this past work relies either on the instrumental record, which is relatively short ([@R21]), or on paleoclimate data processed in different ways ([@R10], [@R15], [@R22]). Here, we use a data assimilation approach called the Last Millennium Reanalysis (LMR) ([@R23], [@R24]) to combine paleoclimate proxy data with climate model output within a unified framework to explore the climate drivers of U.S. drought.
The LMR methodology (Materials and Methods) combines a network of annually resolved proxy records---including tree rings, corals, and ice cores ([Fig. 1](#F1){ref-type="fig"})---with output from a general circulation model (GCM), to estimate a gridded multivariate record of climate variability over the past two millennia ([@R23], [@R24]), similar to the Paleo Hydrodynamics Data Assimilation product (PHYDA) reconstruction ([@R25]). All temporal variations in the reanalysis are informed by the proxy records weighted against a time-independent climate estimate from the GCM prior. Fundamentally, the network of proxy records sample the climate system at different locations and seasons and for different spans of time, and climate variable relationships from a climate model are used to synthesize these diverse records to reconstruct a dynamically consistent view of past climate. The reanalysis represents a mix of information from the proxies and the GCM, relying more on GCM covariances and remote proxies where local proxies are scarce.
![Proxies used in the LMR reanalysis.\
(**A**) Proxy locations and (**B**) the number of proxies through time. Proxies come from PAGES2k v2.0.0, Breitenmoser *et al.* ([@R62]), and the National Centers for Environmental Information paleoclimate archives. In total, 2787 proxies are used: 2556 trees, 116 corals and sclerosponges, 105 ice records, 9 lake sediments, and 1 bivalve.](aay7268-F1){#F1}
RESULTS
=======
Climate reconstructions via data assimilation
---------------------------------------------
The LMR has been extensively validated against instrumental data and independent proxy data not used in assimilation ([@R24]). Compared to instrumental datasets and modern reanalysis products, global-mean temperature correlations range from 0.88 to 0.94 and coefficients of efficiency \[([@R23], [@R26]), here calculated using identical calibration and verification periods\] range from 0.77 to 0.87 (fig. S1). Spatial skill is highest over tropical and mid-latitude oceans, similar to previous findings ([@R23]). Additional verification in a similar set of experiments, including verification against withheld proxies and tests using separate calibration and verification periods, can be found in recent work ([@R24]).
Annual-mean climate indices in the LMR also compare well with observationally based datasets. Three large-scale modes are inspected: the Nino 3.4 index (SST variations in the 5°S to 5°N, 170°W to 120°W region of the Pacific Ocean), the Pacific Decadal Oscillation (PDO; calculated as the first principal component of Pacific SST anomalies north of 20°N in the LMR), and the AMO (calculated as the mean of North Atlantic SSTs detrended over the period 1856 to 2000 in the LMR). A comparison between the Niño 3.4 index derived from the LMR ensemble and from observational data ([@R27]) gives a correlation of 0.79 and a coefficient of efficiency of 0.40 over 1873 to 2000 C.E. For PDO, the LMR matches an observationally based time series ([@R28]) with a correlation of 0.65. For the AMO, the LMR matches the annualized AMO index from the Earth System Research Laboratory ([@R29]) with a correlation of 0.65 (fig. S2). This general agreement provides evidence that the LMR, using data from a network of proxy records, skillfully reconstructs large-scale climate indices given sufficient proxies.
Drought is quantified using the Palmer Drought Severity Index (PDSI), a common hydroclimate metric whose negative values are indicative of moisture deficit ([@R30]). First, we examine the full range of PDSI variations to analyze the effects of large-scale climate patterns on U.S. hydroclimate in general. Then, we define a drought threshold (here, all years with PDSI below 1 standard deviation compared to the nearest 51-year period) to evaluate the conditions that are relevant for dry periods alone, for different regions of the United States. For the LMR, PDSI is calculated in the model prior using the Penman-Monteith method for estimating the potential evapotranspiration component ([@R31]) and then reconstructed in the past through data assimilation alongside other climate variables (Materials and Methods).
For the purpose of cross-validation, the LMR-derived PDSI is compared to two PDSI datasets based on different information sources. The first dataset, the North American Drought Atlas (NADA), is a summer \[June-July-August (JJA)\] PDSI reconstruction based on 1845 tree records ([@R32], [@R33]). A comparison between LMR and NADA shows high values of correlation and coefficient of efficiency over most of the United States, especially in densely sampled regions ([Fig. 2](#F2){ref-type="fig"}). This agreement is remarkable despite the differences in seasonality between the two datasets (annual-mean values for LMR versus June to August for NADA) and the very different methodologies. When averaged regionally, the correlation between LMR and NADA from 1001 to 2000 C.E. is 0.84 for the U.S. region as a whole (30°N to 49°N, 130°W to 65°W, land only) and 0.90 for the southwest United States \[32°N to 40°N, 125°W to 105°W, land only, as in ([@R10])\]. The timing of PDSI variations compares well between the LMR and NADA datasets, especially over the southwest United States ([Fig. 2](#F2){ref-type="fig"}), although the LMR displays reduced variability at basically all time scales. This reduced variability compared to NADA could result from several differences in methodology, including NADA's use of variance restoration, different approach to tree ring detrending, or proxy prewhitening. Differences between LMR and NADA over Canada and Mexico may be due to the existence of fewer records there; NADA generally shows reduced skill in those regions \[see figure 6 in ([@R32])\]. A shortcoming of this comparison, however, is an overlap in proxy data sources informing the reconstructions (fig. S3).
![PDSI comparison between LMR and NADA.\
Coincident (**A**) correlation and (**B**) coefficient of efficiency (CE) calculated between annual-mean LMR and JJA NADA PDSI at every point for the years of common overlap during 1001 to 2000 C.E. Regional-mean time series of LMR (blue) and NADA (black) for (**C**) an approximate U.S. region (land within 30°N to 49°N, 130°W to 65°W) and (**D**) the southwest United States (land within 32°N to 40°N, 125°W to 105°W). The box in (B) represents the southwest U.S. region used in (D) and analyses throughout the paper. Blue shading in (C) and (D) represents the 95% uncertainty bands for the LMR PDSI. Uncertainty bands widen back in time, consistent with proxy attrition. A 10-year running mean has been applied to the time series for plotting clarity \[(C) and (D)\], but all correlation and coefficient of efficiency values are calculated using annual values. Local proxies used in the LMR data assimilation are shown in (A), with count totals representing proxies over the entire globe. Hatched areas in (A) are not significant at the 95% level according to an isospectral test ([@R69]).](aay7268-F2){#F2}
To compare LMR to a dataset that does not use proxy data, we also validate the LMR-derived PDSI against a PDSI product derived from instrumental observations (Dai PDSI) (fig. S4) ([@R30]). While the LMR uses observational temperature and precipitation data \[GISS Surface Temperature Analysis (GISTEMP) and Global Precipitation Climatology Centre (GPCC), respectively\] to derive relationships between proxy quantities and modern climate quantities, the LMR methodology has no additional knowledge of modern PDSI values, making this an insightful comparison despite only covering 1850 to 2000 C.E. The LMR-derived PDSI shows good agreement with Dai PDSI, with a correlation of 0.65 over the U.S. region and 0.70 for the southwest United States.
The overall agreement between the LMR and these two PDSI datasets, which are formulated using two very different approaches, lends support to the LMR methodology and increases confidence in the multivariate analysis of North American drought. Furthermore, while correlations between LMR and Dai PDSI are slightly worse than between NADA and Dai PDSI, coefficient of efficiency values are better for LMR in the southwest United States, Mexico, and Canada. This comparison shows that, despite methodological differences, the LMR produces PDSI values in line with other datasets. The advantage of LMR is that it provides dynamically consistent reconstructions of other climate fields, promoting a more in-depth multivariable analysis, which is the focus of the following sections.
Drought versus SST
------------------
Understanding the causes of drought variability is critical in water-poor regions like nearly all of the United States westward of the 100th meridian ([@R34]). Among forcing mechanisms, drought in the western United States is most often related to ENSO, with a La Niña pattern of cooler eastern equatorial Pacific SSTs correlated with drier conditions in the southwest ([@R4], [@R10]--[@R12]). Heating anomalies associated with equatorial SSTs modify Rossby wave propagation from the tropics into the extratropics; in particular, cool La Niña SSTs reduce tropical convection and upper level divergence, which affects the location of quasi-stationary waves ([@R35], [@R36]). Positive pressure anomalies over the northern Pacific (i.e., a "blocking" high) divert storm tracks northward, reducing precipitation in the southwest United States. Drought development may also depend on transient eddy activity related to the Pacific storm track, as well as land-atmosphere feedbacks such as the soil moisture feedback, with models suggesting that preexisting dry soils help exacerbate subsequent drought ([@R35], [@R37]). A cold phase of the PDO is also associated with drought in the southwest United States and wet conditions in the northwest ([@R10]), though the PDO may affect drought primarily in conjunction with ENSO rather than on its own ([@R18]).
To examine these links between drought in the contiguous United States and climate features of the surrounding basins, multivariate climate patterns are analyzed over the past millennium. It is important to note that the causes of drought can vary seasonally, with different factors affecting summer versus winter precipitation patterns, but here, we take a broader view and focus on annual-mean anomalies. Because U.S. drought may be influenced by SSTs in the preceding winter, our analysis focuses on connections between annual-mean PDSI and climate patterns in the coincident year as well as in the previous year. In observational datasets, correlation patterns are similar when comparing annual-mean PDSI to Nino 3.4 averaged over the previous December-January-February (DJF) (which is often used as a target for analysis) or over the previous full year, lending support for this approach (note S1 and fig. S5).
To isolate the patterns that constitute the largest amount of covariance between U.S. PDSI and the surrounding climate system, we conduct a maximum covariance analysis (MCA; see Materials and Methods) ([@R38]) between annual-mean PDSI over the United States and a joint field consisting of annual-mean SST and 500-hPa heights in the surrounding regions ([Fig. 3](#F3){ref-type="fig"}). MCA isolates orthogonal patterns that explain the maximum amount of covariance between the fields over the analysis period (here, the last millennium), offering a largely impartial assessment of the relationships between chosen fields in the multivariate LMR reconstruction.
{#F3}
The first mode of the MCA outlines a clear connection between PDSI and tropical Pacific SSTs, with southwest and western-central U.S. dry conditions over the last millennium corresponding with La Niña and cold PDO SST patterns ([Fig. 3](#F3){ref-type="fig"}). Geopotential heights at 500-hPa increase primarily over the North Pacific, with a band extending across the United States and part of the North Atlantic. Over the equatorial Pacific, 500-hPa heights are slightly reduced. This Pacific response---lower pressure in the tropics and higher pressure over the northern Pacific---fits the canonical view of La Niña producing a blocking high that diverts Pacific storm tracks to the north. In addition, the pattern of height anomalies, with largest increases in the North Pacific and stretching across the contiguous United States, is consistent with (and opposite to) the decreased heights associated with the type of El Niño events that produced the greatest positive precipitation anomalies in California between 1948 and 2016 ([@R39]). The first mode of the MCA is fairly robust when analyzing the individual iterations of the LMR (see Materials and Methods).
In the MCA above, the squared covariance fraction quantifies the fraction of squared covariance between two fields represented by a given mode of variability ([@R40]), and the values for fraction of variance separately quantify the fraction of total variance represented in each of these fields. These metrics indicate that the first mode of the MCA, discussed above, accounts for 39% of the variance in the U.S. PDSI field, 34% of variance in the joint SST/500-hPa height field, and 83% of the squared covariance between these two fields.
If the MCA is conducted between PDSI and the previous year's SST and 500-hPa geopotential height anomalies (rather than comparing coincident years), then the patterns are similar to those discussed above (fig. S6). Examining a time-lagged relationship is a good target for analysis, as years are reconstructed individually in the LMR data assimilation and any relationships between different years stem from the proxy records rather than covariances in the model prior. The MCA suggests that the primary link between these fields is that a La Niña/cool PDO pattern is associated with drought in the southwest United States although the correlation of the expansion coefficients is reduced in the lagged case.
As a complement to the previous analysis, which analyzes the full range of hydroclimate variability over the United States, we now explore climate conditions specific to drought states by implementing a drought threshold. "Droughts" are here defined as all years where regional PDSI is more than 1 SD below a 51-year moving window of PDSI, which accounts for any mean state shifts in PDSI values and reductions in variance with the loss of proxies further back in time. To examine the climate patterns associated with drought in different parts of the United States, we calculate drought years for four regions: the northwest, southwest, central, and southeast United States. These regions were chosen in other work to represent regions of greatest statistical drought independence ([@R10]), and the mean SST and 500-hPa height anomalies during drought years in each of these regions are shown in [Fig. 4](#F4){ref-type="fig"} along with the conditional distributions of Nino 3.4, PDO, and AMO values for drought years and nondrought years.
{#F4}
Mean climate states for drought years for each of the four U.S. regions are characterized by La Niña and cool PDO SST patterns, although these patterns are weak when considering droughts in the northwest U.S. region. Droughts in each region also generally correspond with warmer temperatures in much of the North Atlantic and increased 500-hPa heights over the North Pacific, continental United States, and North Atlantic. Analysis of climate indices shows that the Nino 3.4 and PDO indices are significantly lower during drought years compared to nondrought years for at least three of the four regions using a resampling test (Materials and Methods), and AMO is significantly higher in drought years for the central and southeast regions, but mean AMO is not significantly different for the two west coast regions. The analysis of climate indices also reveals a considerable range of values in both drought years and nondrought years, with a large degree of overlap between patterns that correspond to drought years and those that do not. This indicates that while certain climate states (i.e., La Niña and cold PDO) are associated with drought states on average, these relationships only emerge when examining mean state differences among considerable amounts of climate variability.
Similar results emerge when computing linear regressions between the full range of PDSI variations and the surrounding climate fields (not shown). Correlations between PDSI and equatorial Pacific SSTs are strongest for the southwest United States (exceeding −0.6 for SSTs just off the equator, larger than for the Nino 3.4 region itself) and weakest for the northwest United States, again indicating the differing effects of these teleconnections on different regions of the United States. To ensure that these results are not overly determined by covariances in the model prior---which may affect results for coincident years but not the lagged analysis, as mentioned above---an alternate experimental design is explored in note S2.
We use self-organizing maps (SOMs) to explore connections between SSTs, 500-hPa heights, and U.S. drought conditions (fig. S7). SOMs isolate characteristic patterns in a given climate field and identify which years are most represented by each pattern ([@R20], [@R41]). Here, eight SOM patterns are computed, as in ([@R20]) (see Methods in that paper for details), from the global SST field over years 1001 to 1925, with the post-1925 years removed to eliminate trends in the SOM patterns because of anthropogenic warming. In addition, detrended 500-hPa geopotential height and PDSI anomalies are composited over the years corresponding to each SOM pattern, revealing the geopotential height and PDSI anomalies that correspond to each SST pattern.
The primary SST pattern that emerges through this SOM analysis corresponds to ENSO. In general, drought years in each region have a higher occurrence of La Niña--like patterns and a lower occurrence of El Niño--like patterns, in agreement with the relationships described above (fig. S7I). This connection appears to be strongest for the southwest U.S. region and weakest for the northwest U.S. region. The primary non-ENSO patterns consist of warmer or cooler SST anomalies overall, with warmer SSTs connected with drought in northern North America and a lack of drought in the southern United States, although this connection is relatively weak (fig. S7, D and G).
Together, these results indicate that La Niña is a noisy predictor of reduced precipitation, but much drought variability appears unrelated to simple ENSO metrics such as Nino 3.4 (fig. S8). Considerable variability exists in the analyzed teleconnection patterns, as seen in the large overlap in climate index values for drought versus nondrought years ([Fig. 4](#F4){ref-type="fig"}). Years with drought in the southwest United States have below average Nino 3.4 75% of the time, and, when considering all years, Nino 3.4 only accounts for 13% of the variance in southwest U.S. PDSI \[i.e., coefficient of determination (*R*^2^) = 0.13; fig. S8\]. Some research has suggested that SST anomalies over longer time periods influence longer term drought ([@R10]), but only a weak correlation emerges for decadal means of the present reanalysis (*R*^2^ = 0.07 for 1001 to 2000 C.E.).
Even observational datasets \[Nino 3.4 ([@R27]) and Dai PDSI ([@R30])\] reveal considerable variability in the relationship between these two quantities; for the years 1874 to 2000 C.E., *R*^2^ between southwest U.S. PDSI and Nino 3.4 is 0.10 when Nino 3.4 is calculated during the previous DJF, 0.10 when Nino 3.4 is calculated over the previous year, and only 0.03 for coincident annual means (fig. S5). Despite this, we primarily analyze coincident years in the reanalysis results because they produce a higher correlation on these time scales, potentially a result of the reanalysis methodology. Slightly higher values can be found for PDSI near the coast of Texas, but the general weakness of these relationships suggest that ENSO, measured by standard metrics such as Nino 3.4, is a rather minor influence on U.S. drought. Stronger connections are revealed when considering the northern Pacific and Atlantic basins as a whole, as seen in the MCA analysis, but a considerable portion of drought variability still appears unrelated to the Nino 3.4 metric alone, possibly suggesting the need for a more comprehensive approach when evaluating the ocean's influence on U.S. drought.
Drought response to external forcings
-------------------------------------
According to a recent modeling study using prescribed SSTs ([@R12]), SST variations in the global oceans explain 40% of annual-mean precipitation variance in northern Mexico and the southeastern United States but much less in other regions, including the southwest United States. The remaining drought variability is a topic of much interest ([@R12]) and may help explain events such as the reduced precipitation in southern California during 2015/2016, which occurred despite a strong El Niño. Along these lines, we now consider the extent to which some drought variability may be driven by external climate forcings, such as greenhouse gases, explosive volcanism, or variations in solar irradiance, which is a question well suited for GCMs. Volcanic eruptions, for instance, reduce global-mean precipitation in the Hadley Centre Coupled Model version 3 (HadCM3) ([@R42]) and Coupled Model Intercomparison Project 5 models ([@R43]), with the largest modeled precipitation changes taking place in the tropics. Volcanism can also generate abrupt cooling, followed by a recovery of several years ([@R44]).
Because climate models can be run with specified forcings, they provide a valuable counterpart to data-driven studies like the LMR. Here, we explore the effects of external forcings on past climate in the Community Earth System Model (CESM)--Last Millennium Ensemble (LME) simulations ([@R45]). The CESM-LME consists of a set of transient GCM simulations starting at 850 C.E. and run with changes in one or all of the following: greenhouse gases, volcanic aerosols, solar forcing, orbital forcing, and land use change. In addition to analyzing an ensemble of nine simulations run with all forcings (hereafter called "All" or "fully forced"), we investigate a variety of single- forcing simulations where one forcing varies while all other forcings are set to their 850 C.E. values. These simulations focus on the effects of changes in greenhouse gases (three ensemble members), land use/land cover (three members), Earth's orbit (three members), solar irradiance (four members), and volcanic forcing (five members) \[see table 1 in ([@R45])\]. While orbital forcing, greenhouse gases, and land use change have much slower rates of change than the drought variability of interest, they are included here for completeness and to examine whether slow changes in these parameters can affect general drought statistics. These simulations present a useful complement to the LMR reconstruction because they explore climate variations in the presence or absence of certain forcings, which is impossible to fully disentangle in observations alone. In addition, multiple ensemble members allow us to sample many different expressions of internal atmospheric variability, which is important for determining which variations are endogenous to the atmosphere-ocean system and which are exogenous. We examine how different external forcings in these simulations affect PDSI in the southwest United States. Because this analysis spans the years 850 to 1849 C.E., recent anthropogenic changes will not be considered.
The nine fully forced simulations are subjected to identical forcings and differ only in their initial conditions. Because imposed forcings always occur with the same timing and magnitude across these simulations, externally forced responses should exhibit consistent timing across simulations, emerging with averaging, provided that the ensemble size is large enough. Variations that are a function of unforced atmosphere-ocean variability, on the other hand, including variability associated with ENSO or other large-scale teleconnections, need not have consistent timing between ensemble members and tend to cancel out across ensemble members. Put another way, forced responses should emerge as common signals from the otherwise distinct climate variations in each simulation.
We compute PDSI from modeled climate values in the CESM-LME (Materials and Methods). When southwest U.S. PDSI is compared across the nine fully forced simulations, considerable differences are evident, with the mean signal exhibiting relatively small variations ([Fig. 5](#F5){ref-type="fig"}). To quantify similarities between any two simulations, we compute correlations for southwest U.S. PDSI between every pair of fully forced simulations. These correlations have a mean value of 0.02, and no two time series agree with a correlation above 0.09. This lack of consistency indicates that little southwest U.S. drought variability may be explained by external forcing. The largest forced signal relates to explosive volcanism, which tends to produce wetter conditions in the southwest United States after eruptions. This wettening is particularly apparent after the Samalas eruption of 1257 C.E., which is the largest volcanic forcing in these simulations, when all nine ensemble members show positive PDSI regardless of prior conditions ([Fig. 5](#F5){ref-type="fig"}). A more detailed analysis of volcanic responses in these simulations has been presented in past work ([@R46]). This volcanically forced signal, however, only accounts for a small part of the overall variability. In other regions of the world, such as northwestern South America and northwestern Africa, volcanic eruptions have a larger relative impact in these simulations (not shown).
{#F5}
Single-forcing CESM-LME simulations are used to further investigate whether (and to what extent) particular forcings influence southwest U.S. drought. To determine whether much temporal agreement exists between single-forcing experiments and fully forced experiments, we calculate correlations between different sets of simulations (fig. S9). Of the single-forcing experiments, volcanic forcing has the largest correlation with the fully forced simulations, although the median correlation in that case is still only 0.04, suggesting that very little of the total variability is explained by these external forcings.
While external forcing appears to have little influence on the timing of droughts, it is worth investigating whether imposed forcings affect PDSI characteristics in other ways, such as the length or severity of a drought. To prevent long-term trends from overly affecting the estimated variability, we remove a linear trend from each southwest U.S. PDSI time series, and droughts are then calculated as periods beginning with two consecutive years with PDSI below 1 SD relative to the nearest 51-year periods and ending with two consecutive years with PDSI above that threshold, similar to the definition used by Coats and coauthors ([@R47], [@R48]). Using this definition, the average frequency, length, and magnitude of droughts are calculated over the 1000-year interval for each simulation. This analysis is similar to the one performed in other work ([@R16]), which showed that a considerable portion of drought variability may be unrelated to SSTs. Comparison of these drought statistics reveals a high degree of similarity across CESM-LME experiments, indicating that the applied external forcings do not have a large impact on the longevity or magnitude of droughts in this region ([Fig. 6](#F6){ref-type="fig"}). This is consistent with past work, which has shown that external forcings are not required to explain the magnitude, spatial, and temporal extent of severe droughts such as those seen in the proxy record (i.e., megadroughts), though these forcings may be necessary to explain the clustering of these droughts during the medieval era ([@R49]).
{#F6}
For comparison, statistics of the LMR show southwest U.S. droughts that are generally longer and weaker, a characteristic that can be seen in the PDSI time series (fig. S10). The statistics of southwest U.S. PDSI in NADA, on the other hand, show more intense droughts. These differences are difficult to explain, but some of the disparity may stem from the aforementioned methodological differences between LMR and NADA regarding variance restoration, tree ring detrending, and proxy prewhitening.
In general, external forcings appear to have only minor effects on southwest U.S. drought in the CESM-LME simulations. Volcanic eruptions encourage wetter conditions in the short term, but these forced variations make up only a small portion of the total PDSI variability. Our focus on the years 850 to 1849 C.E. makes it difficult to comment on the effect of anthropogenic forcing during the industrial period, but this analysis suggests that natural forcings have only exerted a minor influence on southwest U.S. drought in the centuries before 1850, when greenhouse gases were more constant.
DISCUSSION
==========
The LMR constitutes a powerful methodology for creating a physically consistent multivariate climate reconstruction from a diverse array of proxy records. The proxies provide data about specific regions, climate fields, spans of time, and seasons, and we use proxy system models (PSMs) and the covariance structure from a GCM to synthesize these diverse perspectives into a cohesive view of past climate. While more work is needed, analyses show that the LMR climate reconstruction compares well with established datasets for temperature, PDSI, and large-scale climate indices, providing evidence of reconstruction skill. In accord with a wide body of published work, the LMR reconstruction also finds a clear connection between southwest drought and a La Niña SST pattern over the past millennium. This connection emerges as the primary mode of covariability between PDSI, SST, and 500-hPa height fields, even though the analysis methodology (MCA) is not directed to focus specifically on the equatorial Pacific. While this pattern is robust, teleconnections to SST variations appear to explain only a part of U.S. drought variability, leaving the larger portion of drought variability unexplained. Fundamentally, this data assimilation approach presents data-based evidence for the importance of internal atmospheric variability in determining past hydroclimate variability, in agreement with other work ([@R15]--[@R17], [@R39], [@R50]).
The present work contains important caveats, however. In particular, the data assimilation methodology relies on GCM output to partly quantify relationships between different variables and locations, as well as provide a first estimate of past climate. Using a model prior like this is necessary, as it provides the framework for synthesizing information from diverse proxies---which differ in their climate sensitivities, locations, seasonal biases, and temporal coverage---into a physically consistent multivariate climate reconstruction. However, model bias in spatial climate covariance patterns does affect the reconstructions. While this work has used alternate analyses (e.g., lagged correlations and alternate experimental designs) to minimize the impact of these potential biases in results, future work should explore this topic in more depth. The use of model priors from several different GCMs, for instance, may help mitigate the effect of biases in any one particular model.
Another area for future improvement is the incorporation of additional proxies into the data assimilation product, particularly from poorly sampled regions and using additional archive types. One data assimilation advance, which is being explored in current and past work ([@R51]), is the incorporation of lower-resolution proxies into the data assimilation methodology. While proxies that lack annual temporal resolution will require additional considerations within the data assimilation framework, these proxies can provide information about sparsely sampled regions (such as continental margins, in the case of marine cores) and should more accurately capture low-frequency variations compared to tree rings ([@R52], [@R53]), which are heavily represented in the current data assimilation approach. This could provide additional information about slower climate variations and trends, refining our understanding of climate variations such as drought and potentially making this approach more relevant for studies of past megadroughts \[e.g., ([@R54])\]. Considering the potential for anthropogenic changes to worsen future droughts in many regions ([@R7]--[@R9]), better understanding of the climate dynamics behind drought variations is critical for future planning.
If, as we argue, internal atmospheric variability has been a leading cause of multiyear drought over the Common Era, then this bears unfavorably on the prospect for forecasting these droughts. This may have been at play in the 2011--2017 California drought, which was, by some measures, the most severe Californian drought of the past 1200 years ([@R2]). Much drought relief was expected from the 2015/2016 El Niño, which rivaled in magnitude the extreme El Niño events of 1982/83 and 1997/98 ([@R55]). However, while those previous events brought abundant rainfall to California ([@R39]), the 2015/2016 event produced average rainfall throughout most of California ([@R39], [@R55]), defying the expected teleconnection pattern in this region ([@R39]). Consequently, this event failed to end the prolonged drought, which had persisted since 2011/12. Southern California had to wait until the following year, characterized by mild La Niña conditions, to receive enough rainfall to end the drought ([@R56]). Our results suggest that this situation may have been a common occurrence throughout the past millennium, making current limitations in interannual drought predictability especially important in a warming climate with greater evaporative demand for moisture ([@R2], [@R57]).
MATERIALS AND METHODS
=====================
Experimental design: Paleoclimate data assimilation
---------------------------------------------------
Paleoclimate data assimilation offers a powerful approach for synthesizing a vast array of proxy observations (here, thousands of records) with the aid of a model's climate covariance structure. In particular, the LMR is a data assimilation approach that uses information from proxy records and output from a GCM to estimate climate variability over the past two millennia. An earlier version of this method is described in past work ([@R23]), and updates to the methodology are described in a recent paper ([@R24]). The data assimilation methodology is composed of four primary components: (i) GCM output, which serves as a "first guess" at the range of possible climate states and quantifies covariances within the climate system; (ii) PSMs, which relate the model quantities to proxy quantities; (iii) proxy records, which provide the temporal information for the reconstruction; and (iv) a Kalman filter, which is used to perform the data assimilation.
The methodology works by first randomly selecting a collection of annual-mean climate states from the output of an existing GCM simulation. Here, we use 100 years from the Community Climate System Model 4 (CCSM4) last millennium simulation ([@R58]). These randomly selected model states (i.e., the prior) are initially identical for every year of the assimilation, serving as the first guess of the real climate state for any given year. In other words, before any assimilation takes place, the real climate is suspected to be somewhere in this range of modeled climate states. The LMR is then run for 20 iterations; within a given iteration, the same prior is used for every year of the reconstruction, so the model provides no temporal information to the reconstruction. In addition to providing an initial range of plausible climate states, the 100-member prior is also used to quantify the covariances within the climate system, which forms the mathematical scaffolding that relates climate variations at different locations and between different fields.
To perform the data assimilation, proxy and model quantities must be compared in the same units, so PSMs are needed. Here, relationships between proxy quantities and climate variables are computed by regressing proxy records onto instrumental fields over the period 1880 to 2015 C.E. For all proxies except tree rings, a linear regression is computed against temperature \[GISTEMP version 4 ([@R59])\]. For tree rings, a bivariate regression is computed against both temperature (GISTEMP) and precipitation \[GPCC ([@R60])\]. Regarding seasonality, proxy records are not assumed to record annual-mean quantities. Instead, proxy records are regressed onto climate quantities averaged over the entire year as well as multiple subsets of the year: summer, winter, and four different two-season half-year periods (as well as the season specified in the proxy metadata if different from the previously mentioned seasons). The averaging windows that produce the best regression between instrumental data and proxy records are used in the data assimilation, and this is determined separately for each proxy ([@R24]). Tree ring width proxies, which are regressed onto both temperature and moisture, are allowed to have different seasonal sensitivities for temperature and moisture ([@R24]). Linear regressions are simpler than a process-based model, but they provide good results that can provide a baseline for more complex PSMs in the future ([@R24]).
The proxy network used in the present work includes records from three sources: the PAGES2k v2.0.0 database ([@R61]), several thousand tree chronologies compiled by Breitenmoser *et al.* ([@R62]), and a selection of other proxy records from the National Centers for Environmental Information (formerly National Climatic Data Center) paleoclimate archives. The proxies are available at Zenodo ([@R63]), and the non-PAGES2k records are described in a recent data report ([@R64]). To be assimilated, records must be at least annual in resolution and must have at least 25 years of overlap with the instrumental records. A total of 2787 records are assimilated in the present reanalysis, the spatial coverage of which is shown in [Fig. 1](#F1){ref-type="fig"}.
For each year of the reanalysis, the prior is used as a starting point, and the climate state is updated through assimilation of annually resolved proxy records one by one via a Kalman filter. The Kalman filter compares each proxy value against an estimate of the proxy value computed from the model prior and then adjusts the climate state to produce a better fit for the given year. Because the model prior quantifies the climate covariance structure (between locations as well as between climate variables), it provides the mathematical framework for updating more distant locations as well as a variety of climate variables in a uniform framework. In general, climate is reconstructed through comparison with both local and remote proxies. Locations closest to proxies, as well as variables that are most closely related to proxy measurements ([@R23]), are expected to be better informed by the proxy network, while other locations and climate fields rely more on model covariances. For example, previous research has shown that the LMR has higher skill in reconstructing surface temperature than 500-hPa height ([@R23]), although the skill of both has been improved with recent methodological innovations ([@R24]). These qualifications should be kept in mind when interpreting results.
The ability to reconstruct multiple variables has clear benefits and facilitates the analysis of climate teleconnections over an extended period, with some qualification (note S2). In the LMR, the proxy records provide temporal information and some spatial information (by making use of multiple records in space), while the covariance structure of the GCM prior is used to propagate information between locations and between climate fields. A localization radius is used to ensure that proxies cannot influence the climate farther than 25,000 km from their location, a value that was chosen to produce the expected variance characteristics in the reconstructed temperature \[see table 1 in ([@R24])\]. Further details of this methodology are explained in other work ([@R23], [@R24]).
In the present analysis, 20 iterations of the LMR were run. Each iteration uses a different random selection of 100 model years for the prior and a different random selection of 75% of the proxies for assimilation. Variety in the priors and assimilated proxies helps sample uncertainty in the results. All climate fields are output as annual quantities averaged from January to December. Exact methodological choices are explored in past work ([@R24]), and the data have been made available as the v2.0 release of the LMR dataset (see "Data and materials availability").
The number of annually resolved proxy records used in the reanalysis decreases back in time ([Fig. 1B](#F1){ref-type="fig"}), and larger differences emerge between the LMR results and the NADA for the first millennium compared to the second, so the analysis in this paper focuses on years 1001 to 2000 C.E. The tree proxies used in the LMR and the NADA have considerable overlap (fig. S3), although the two approaches have numerous differences: The methodologies are distinct, the LMR uses additional proxy types, and the methods used to remove tree ring growth curves are likely different as well, among other differences. Still, because trees are the most numerous proxy used in this study, both over the United States and globally, this overlap in data sources should be considered when comparing drought in the LMR and the NADA.
Using this multivariate reconstruction of past climate, relationships in the climate system can be explored over an extended period of time, with the qualifications mentioned in note S2. Pure modeling studies, which are used for exploring possible future drought changes ([@R7], [@R8]), are deficient in modeling some aspects of drought variability; models may have inadequate low-frequency hydroclimate variability ([@R65], [@R66]), although at least one study finds a similar number of long droughts in the southwest United States in models as compared to NADA ([@R48]) and the present analysis finds similar values between NADA and the CESM-LME simulations. In addition, because reanalysis uses real proxy data, this method can provide insight on actual past droughts.
PDSI calculations
-----------------
To reconstruct the PDSI over the past two millennia, PDSI values are first calculated from quantities in the CCSM4 last millennium simulation, which is used as the model prior. This was done using the Penman-Monteith equation for potential evapotranspiration and monthly climate model output of 2-m air temperature, precipitation, vapor pressure, surface pressure, net surface radiation, and surface wind (estimated from 10 m down to 2 m using the wind profile power law). The computations of PDSI were carried out using the MATLAB code from Jacobi *et al.* ([@R67]), which produces the standard formulation of PDSI as opposed to self-calibrating versions \[e.g., ([@R68])\]. Once PDSI is calculated in the model prior, it can be included in the LMR data assimilation to calculate proxy-informed PDSI values over the past two millennia. The same method was used to calculate PDSI in the CESM-LME simulations.
Maximum covariance analysis
---------------------------
To examine how U.S. drought covaries with large-scale patterns of the surrounding climate system, an MCA \[also called singular value decomposition ([@R40])\] is used to isolate the mode that explains the largest amount of covariance between two fields. Here, one field is PDSI over the United States and the second field is a concatenation of SSTs and 500-hPa heights over a larger region (see the regions displayed in [Fig. 3](#F3){ref-type="fig"}). To ensure that neither the temperature nor the 500-hPa height anomalies dominate the second term of the MCA, all climate anomalies have been standardized by the mean and SD over their entire regions, an alternate approach mentioned in past work ([@R40]). This MCA analysis is conducted on variables on their reconstructed 2° latitude-longitude grid, with spatial weighting applied. To illustrate common patterns in SST/500-hPa height and their impact on drought, figures show the homogeneous map of SST/500-hPa height and the heterogeneous map of PDSI. Maps are scaled to have standardized expansion coefficients before plotting.
Resampling test for significance
--------------------------------
The split violin plots in [Fig. 4](#F4){ref-type="fig"} show the range of annual-mean climate indices (Nino 3.4, PDO, and AMO) for years when different regions of the United States were experiencing drought or nondrought. To see whether the mean of each climate index was significantly different during drought and nondrought years for each of these cases, a resampling test was done. In this test, the continuous spans of years spent in drought or nondrought were identified for each case, and then analogous sets of values were randomly selected from the whole time series of the climate index, using sampling with replacement. This was repeated 1000 times, and the original drought versus nondrought climate index anomaly was compared against the anomalies in these 1000 randomly sampled cases. Cases in which the original difference was outside of the 2.5th or 97.5th percentile were deemed significant.
Supplementary Material
======================
###### aay7268_SM.pdf
We would like to extend thanks to E. Cook for invaluable information about tree ring proxies and R. Tardif for help with the LMR code. **Funding:** This work was supported by the National Oceanic and Atmospheric Administration (grants NA14OAR4310175 and NA14OAR4310176) and the National Science Foundation (grants NSF AGS-1702423 and NSF AGS-1805490). M.P.E. was supported by the University of Southern California and Northern Arizona University, and computational resources were also provided by both universities. LDEO contribution number 8427. **Author contributions:** M.P.E. provided most of the analysis and writing of the paper. J.E.-G. provided additional analysis and experimental design. N.S. provided PDSI calculations and discussion of drought considerations. G.J.H. and E.J.S. provided much of the methodological design of the LMR, as well as feedback on the paper. All authors contributed to the writing. **Competing interests:** The authors declare that they have no competing interests. **Data and materials availability:** The LMR v2.0 dataset, as well as documentation, proxies, and other sample data to run additional reanalyses, has been made publicly available at [www.ncdc.noaa.gov/paleo-search/study/27850](https://www.ncdc.noaa.gov/paleo-search/study/27850) and <https://atmos.washington.edu/~hakim/lmr/LMRv2/index.html>. The reanalysis data consist of spatially reconstructed climate fields for a variety of variables over the years 1 BCE to 2000 C.E. The source code for the LMR is available at <https://github.com/modons/LMR>.
Supplementary material for this article is available at <http://advances.sciencemag.org/cgi/content/full/6/32/eaay7268/DC1>
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Hepatitis delta virus (HDV) is considered to be a satellite virus of the hepatitis B virus (HBV). HDV co-infects or super-infects liver cells already infected with HBV resulting in an higher risk of cirrhosis and fulminant hepatitis, as well as increased liver tissue damage [@pone.0012512-Govindarajan1], [@pone.0012512-Jacobson1]. Hepatitis delta virus contains a ribonucleprotein core which includes a 1.7 Kb circular single-stranded RNA genome and several copies of the only virus encoded protein, the so called delta antigen (reviewed in Taylor, [@pone.0012512-Taylor1]). The clinical association between HDV and HBV is due to the fact that the outer envelope of HDV consists of the surface antigens coded by the HBV genome (HBsAgs) which are necessary for virion maturation and release from the cells (reviewed by Taylor in [@pone.0012512-Taylor2]). Therefore, productive HDV infection occurs only in the presence of HBV.
It is widely accepted that the clinical course of super-infection and co-infection displays distinct features. In most cases, super-infection of chronic HBV patients results in the development of chronic HDV infection. In general, the clinical course of HDV super-infection starts with an acute phase which is followed by the development of chronicity, and finally the elimination of HDV and HBV. During the acute phase of infection, an active replication of HDV is observed whilst HBV replication is partially suppressed. The following chronic phase is characterized by a decrease in HDV replication which is accompanied by a subsequent increase in HBV replication [@pone.0012512-Hsieh1]. It is estimated that about 70% of super-infected patients will progress from acute to chronic disease. Additionally, 60--79% of chronic HDV patients will further develop cirrhosis. This rate is 3 times higher than that found in HBV or HCV infected patients alone [@pone.0012512-Rizzetto1]. According to Fattovich et al. [@pone.0012512-Fattovich1], HDV super-infection leads to a 3 times greater increase in risk of hepatocellular carcinoma and twice greater rates of mortality in patients with compensated cirrhosis. In HDV and HBV co-infections, the clinical course is similar to that observed during acute HBV infection [@pone.0012512-Chisari1], [@pone.0012512-Guidotti1].
There is no specific treatment for HDV infection. The most common therapeutic approach is based on the administration of interferon-. However, the clinical response is variable, and in most cases reversible upon interruption of treatment [@pone.0012512-Hoofnagle1]--[@pone.0012512-Lau1]. The concomitant use of antiviral drugs like ribavirin or lamivudine, showed no significant benefits in the treatment of hepatitis delta patients [@pone.0012512-Lau2]--[@pone.0012512-Yurdaydin1]. Although these drugs may have some inhibitory effect on HBV replication, they do not suppress HDV replication probably due to the fact that HBsAgs expression, at least in part, seems not to be affected.
Vaccination against HBV protects individuals against HDV co-infection. Although vaccination programs led to a considerable reduction in both HBV and HDV prevalence, the two viruses are still endemic in a number of regions, namely the Amazon basin, and some African and Asian countries [@pone.0012512-Ponzetto1]. It is estimated that worldwide about 350 million people are infected with HBV of which 5--10% are also co-infected with HDV. HDV and HBV share the same routes of infection affecting individuals of all age groups. The most frequent routes of transmission are the sexual contact and the direct contact with blood or blood products from infected carriers [@pone.0012512-Hansson1], [@pone.0012512-Oliveira1].
The use of mathematical models to study dynamics of virus infections may represent a powerful approach to simulate the course of infection and predict the potential response to different therapies. They have been previously developed for a number of pathologies including HIV, HBV, and HCV [@pone.0012512-Brunetto1]--[@pone.0012512-DUgo1]. More recently, the mathematical simulation of the spread of HDV and HBV in a population was reported [@pone.0012512-Xiridou1]. However, a mathematical model to study HDV and HBV dynamics in infected individuals is still lacking. In this initial work we report the development of mathematical models to simulate the dynamics of HDV and HBV, both during co-infection and super-infection, in the presence and absence of any therapy. Data are presented concerning the number of predicted infected cells and viral load along the period of infection.
Methods {#s2}
=======
To model for the viral dynamics of HDV we considered the following six variables:
1. the number of uninfected cells at time ,
2. the number of HBV infected cells at time ,
3. the number of HDV infected cells at time ,
4. the number of infected cells with both HBV and HDV at time ,
5. the HBV viral load at time ,
6. the HDV viral load at time .
The model representing the hepatitis delta virus (HDV) viral dynamics can be represented in the diagram in [Figure 1](#pone-0012512-g001){ref-type="fig"}, followed by a description of the variables and parameters represented in this diagram.
{#pone-0012512-g001}
he change in the number of uninfected cells at a certain moment in time will depend on the constant rate at which these cells are generated, ; the number of deaths at that time, which is proportional to the constant death rate of uninfected cells, ; and the number of infected cells lost by infection with HBV and HDV which are proportional to the constant infection rates of HBV and HDV, and , respectively, and to the HBV and HDV viral loads at that time and , respectively.
With respect to the change in the number of , HBV infected cells, at a certain moment in time, its dependency involves the constant infection rate of HBV, , the number of uninfected cells and the HBV viral load at that time. Also, the constant death rate of HBV infected cells, ; and the infected cells with HBV that were co-infected with HDV. The later, proportional to the constant infection rate of HDV, , the number of HBV infected cells, , and the HDV viral load, , at that time.
Similarly, the change in the number of HDV infected cells at a certain time, , will depend on the constant infection rate of HDV, , the number of uninfected cells and the HDV viral load at that time; the constant death rate of HDV infected cells, ; and the infected cells with HDV that were co-infected with HBV at that time which is proportional to the constant infection rate of HBV, , the number of HDV infected cells, , and the HBV viral load, , at that time.
The change in the number of infected cells with both HBV and HDV at a certain time will depend on the number HBV infected cells that are infected with HDV at that time; the number HDV infected cells that are infected with HBV at that time; and the number of uninfected cells that are infected simultaneously with HBV and HDV, , proportional to the constant infection rates of HBV and HDV and the viral loads of HBV and HDV.
The change of HBV viral load at a certain time will depend on the constant HBV virion clearance rate , and from both sources of viral production, i.e. cells that are infected only with HBV and with both HBV and HDV. From cells only infected with HBV, the viral load is proportional to the constant HBV viral production rate and the number of HBV infected cells at that time . From cells infected with both viruses, the viral load is proportional to the constant HBV viral production rate and the number of HBV and HDV infected cells at that time .
As for the change in HDV viral load, the dependency of this viral load will also come from the constant HDV virion clearance rate , but now the viral production will only come from cells that are infected with both HBV and HDV. Therefore, the change of HDV viral load at a certain moment in time is proportional to the constant HDV viral production rate and the number of HBV and HDV infected cells at that time .
As a result of the above, the model for the viral dynamics in an individual infected with HBV and HDV can be expressed as the following system of six differential equations modeling the changes in , () and ():
Solving the Mathematical Model {#s2a}
------------------------------
The system of differential equations in (1) needs to be solved numerically. We used the software *Matlab 7.5* [@pone.0012512-MathWorks1] where we simulated the behavior of our proposed model considering two different kinds of infection: co-infection and super-infection. Co-infection occurs when an individual is simultaneously infected by HBV and HDV, while super-infection occurs in persons with an existing chronic HBV infection. Therefore, for a co-infection scenario the viral dynamics can be modeled simply by the equations in (1), while for the super-infection scenario, we assumed that the individual is first infected with HBV and that the infection of HDV occurs after 200 days (d) of being infected with HBV. Thus, before 200 d, the model (1) is simplified by the following model (2) since only the infection with HBV has occurred.
In both scenarios, and to obtain a numerical solution for the viral dynamics expressed by the models (1) and (2), we will need to provide some initial values for the functions , () and () and the 12 constant rates involved in these models. Unless mentioned otherwise, we will now present the assumptions considered in all simulations throughout this study. The values considered are summarized at the end of this section in [Table 1](#pone-0012512-t001){ref-type="table"}.
10.1371/journal.pone.0012512.t001
###### Parameters considered in this study and corresponding values.
{#pone-0012512-t001-1}
Parameter Values Reference
-------------------------------- ------------------- ----------------------------------------------------------------
Liver cell number [@pone.0012512-Tsiang1]
Blood volume 6000 mL [@pone.0012512-Tsiang1]
Cell division rate day [@pone.0012512-Tsiang1], [@pone.0012512-BlikkendaalLieftinck1]
Liver cells half-life 231 days [@pone.0012512-Tsiang1], [@pone.0012512-BlikkendaalLieftinck1]
Virion life-span in plasma 15--92 hours [@pone.0012512-Nowak1], [@pone.0012512-Lewin1]
Infected cells half-life 10--100 days [@pone.0012512-Nowak1], [@pone.0012512-Lewin1]
Infection rate mL/copies per day [@pone.0012512-Tsiang1]
Viral production rate 6.24/day [@pone.0012512-Tsiang1]
Virion clearance rate 6.24/day [@pone.0012512-Tsiang1]
Mean viral load in equilibrium copies/mL [@pone.0012512-Tsiang1]
Infected cells number cells/mL [@pone.0012512-Tsiang1], [@pone.0012512-Bianchi1]
Clearance constant rate 0.65/day [@pone.0012512-Tsiang1]
All the values for , () and () are expressed in number of copies per milliliter of blood. We will consider that at time 0 there are no cells infected with HBV or HDV, i.e. for . Let us assume three different levels of viral load at the time of infection, : a low level of infection with copies/mL; a medium level of infection with copies/mL; and a high level of infection with copies/mL for . For co-infection, we have for , and for the super-infection we have , for , and . Finally, the number of uninfected cells in an adult has been estimated to be equal to , and considering that the average blood volume of an adult is equal to 6000 mL [@pone.0012512-Tsiang1], the number of uninfected cells per mL can be estimated by cells/mL [@pone.0012512-Tsiang1].
The constants in our study were based on the work by Tsiang and Gibbs in [@pone.0012512-Tsiang1] and summarized in what follows. From [@pone.0012512-Grisham1] and [@pone.0012512-BlikkendaalLieftinck1] approximately 0.3% of the liver cells in rats go through mitosis every day. Assuming the same for humans, the constant rate at which uninfected cells are generated, , can be estimated by cells/mL per day. Since before infection the liver is in equilibrium, i.e. the number of uninfected cells is assumed to be constant. Therefore, we have that , meaning that, at , we have d. In terms of half-lives is equal to days.
Regarding the virion clearance rates and the cell death rates , it is easier to interpret them as their inverse and , representing the mean life-spans of HBV and HDV virion in plasma () and the mean life-spans of productively infected cells with HBV, HDV and both HBV and HDV (), respectively. Since there is very little known about the dynamics of the hepatitis delta virus, we considered not only what it is known in multiple studies with HBV and HDV patients [@pone.0012512-Hsieh1], [@pone.0012512-Tsiang1], [@pone.0012512-Romeo1], but also the opinions of clinicians and biologists that suggested possible values of in the range of 15 to 92 hours, and for values between 10 days and 100 days [@pone.0012512-Lewin1]. In what follows, we consider the following set of values and days for and , respectively.
In terms of the , the viral production rates of HBV from HBV infected cells (), of HDV () and HBV () from cells infected with both HBV and HDV, it is believed that, at least during some periods of infection, HDV replication will have an inhibitor effect on the production of mature HBV particles; therefore, for our study we considered that d, and when the infection of HDV occurs, the viral production rate of HBV from cells infected with HBV and HDV, , is equal to d with ( varying from 0.1 to 1 with increments of 0.1), where represents no inhibition effect. The value of 6.24 d for was suggested by Tsiang and Gibbs in [@pone.0012512-Tsiang1] considering the fact that, when infection reaches an equilibrium, we have that , i.e. the viral production rate from an infected cell is approximately equal to the virion clearance. The mean viral load in the equilibrium was considered to be copies/mL, the clearance constant rate and the number of infected cells to be approximately cells/mL. This value was obtained considering the work of Bianchi, *et al.* [@pone.0012512-Bianchi1] where it was reported that approximately 5--40% of hepatocytes are infected in chronic HBV patients. Tsiang and Gibbs in [@pone.0012512-Tsiang1] considered then the mean value of 22.5% and calculated as .
The infection rates, mL/copies per day, for , were also suggested by Tsiang and Gibbs in [@pone.0012512-Tsiang1]. They were determined such that the peak of the primary viremia occurred 56 days after the infection of an individual.
We also assumed that when the super-infection occurs there is an extra HBV inoculum at that moment. In our model we considered values for extra HBV inoculum of the order 400, and copies/mL, equals to the initial viral load of HBV.
In the next section we will discuss in detail the results obtained from the simulations in *Matlab 7.5*. Although we present only the graphs considering initial viral loads of 400 copies/mL, the viral dynamics for all other scenarios is described in this study as well.
Results {#s3}
=======
The super-infection results {#s3a}
---------------------------
The values of HDV life-span in plasma of patients and of life-span of HDV infected liver cells are, to our knowledge, unknown. Although assuming that these values should not be substantially different from those reported for HBV [@pone.0012512-Hsieh1], [@pone.0012512-Tsiang1] we first decided to perform a sensitivity analysis of the influence of these parameters on the dynamics of infection.
For days, none of the values considered for \'s produced scenarios for the super-infection that are clinically observed since they would predict a spontaneous HDV clearance soon after infection. Consider, for example, [Figure 2](#pone-0012512-g002){ref-type="fig"} where days and hours. Unfortunately, the viral load of HDV and the number of infected cells do not disappear naturally, as suggested in this figure.
{#pone-0012512-g002}
By increasing the mean life-spans of infected cells by 5 days, i.e days, and maintaining hours, we can observe the dynamics of the super-infection in [Figure 3](#pone-0012512-g003){ref-type="fig"}. Nevertheless, a biologically irrelevant scenario occurs again for all other values of . See the result of the simulations for hours in [Figure 4](#pone-0012512-g004){ref-type="fig"}.
{#pone-0012512-g003}
{#pone-0012512-g004}
For values of \'s of 50 and 100 days, all the results were compatible with reported clinical pictures. What the simulations suggest is that the smaller the values of mean life-spans of productively infected cells, , the higher the values of mean life-spans of HBV and HDV virion in plasma, , needed in order to obtain biologically relevant scenarios. Similar results were obtained when the effect of inhibition of HBV replication by HDV was tested with different values of up to 1. We decided to create a finer grid for the values of \'s to estimate the approximate minimum value for the mean life-spans of HBV and HDV virion in plasma in order to obtain possible biological observed scenarios. The results can be seen in [Table 2](#pone-0012512-t002){ref-type="table"}.
10.1371/journal.pone.0012512.t002
###### Estimated minimum values for .
{#pone-0012512-t002-2}
(hours)
---- ---------
10
15
20
The results obtained for estimating the minimum values for \'s in order to obtain observable biological scenarios were quite similar, independent of the initial viral loads for HBV and HDV and the different values of the inhibition factor for the viral production rate of HBV from a cell infected with HBV and HDV. In [Figure 5](#pone-0012512-g005){ref-type="fig"} we see an example of the grid values considered for \'s, with days, initial viral loads of 400 copies/mL, and for . It is very clear that values of \'s smaller than 60 hours generate a scenario that is not observed in patients.
{#pone-0012512-g005}
The particular effect of the inhibition factor of the HBV viral production rate from an infected cell with HBV and HDV, , can only be observed for large values of the mean life-spans of productively infected cells. For example, for days and hours, the higher the value of ( representing no inhibition), the closer the viral loads of HBV and HDV are. A more unstable behavior of the viral loads is observed for greater values of the inhibition (small values of ). The least number of infected cells is always observed for those only infected with HBV, followed by the ones only infected with HDV, and lastly the HBV and HDV infected cells. For higher values of inhibition, say, greater than 80% (), the number of cells infected only by HDV surpasses the number of cells infected with both HDV and HBV. A similar picture is observed for the corresponding values of free viral load. In [Figure 6](#pone-0012512-g006){ref-type="fig"} and [Figure 7](#pone-0012512-g007){ref-type="fig"} we can observe this behavior where the inhibition varies from 90% () to no inhibition, respectively.
{#pone-0012512-g006}
{#pone-0012512-g007}
Notice that for \'s equal to 15 days and \'s equal to 92 hours, there is no effect of the parameter on the results of the model. Notice how close [Figure 8](#pone-0012512-g008){ref-type="fig"} is to [Figure 3](#pone-0012512-g003){ref-type="fig"}, where we had 90% and 10% of inhibition of the HBV viral production in cells infected with HBV and HDV, respectively. In this case, we see from [Figure 8](#pone-0012512-g008){ref-type="fig"} that the HDV viral load is always smaller than the HBV viral load, and that the number of infected cells is highest for HBV only infected cells, followed by the ones infected with both HBV and HDV, and lastly by the number of cells only infected with HDV.
{#pone-0012512-g008}
Since all HDV infected patients are also infected with HBV, we also considered the scenario of extra HBV inoculum occurring concomitantly with the HDV infection. In our simulations we considered this extra HBV inoculum to be equal to the viral load of HDV. The results obtained were the same whether this supplement of HBV occurs or not.
For all biological relevant scenarios, the simulations here reported predict the existence of a peak of viral load and number of infected cells in the beginning of infection. This is particularly noticeable in the case of HBV. A similar observation was reported by Tsiang and Gibbs [@pone.0012512-Tsiang1] when modeling HBV infection alone. This feature may represent an artifact of the model although in woodchuck hepatitis virus (WHV), peaks of viremia in infected experimental animals were observed during the first weeks of infection (see for instance [@pone.0012512-Casey1]).
The co-infection results {#s3b}
------------------------
Co-infection occurs when the individual is infected with HBV and HDV at the same time. The results obtained for the co-infection using model (1) are, in general, very similar to the ones obtained for super-infection. The differences are essentially based on the speed at which the individual reaches the peak of infection and certain behaviors when the infection is in equilibrium. Next, we will show some of the results obtained for co-infection where we try to illustrate the similarities and the differences between these two scenarios.
Consider the example in [Figure 3](#pone-0012512-g003){ref-type="fig"} where days, hours, initial viral loads of 400 copies/mL and a 10% inhibition of the HBV viral production in cells infected with HBV and HDV. In [Figure 9](#pone-0012512-g009){ref-type="fig"} we simulate the co-infection behavior with copies/mL, i.e. the infection of HBV and HDV occurs at the same time, .
{#pone-0012512-g009}
By comparison with [Figure 3](#pone-0012512-g003){ref-type="fig"}, we see that while in super-infection it takes approximately 550 days from the day of infection with HDV to reach an equilibrium, in the co-infection case, this equilibrium is reached after 200 days. In all the examples analyzed, the slope of the line representing the viral load growth and the number of infected cells with HDV and with both HBV and HDV is higher for the co-infection.
It is also believed that for HBV and HDV co-infection the clinical course does not differ from that observed in patients infected HBV alone [@pone.0012512-Hsieh1]. In situations of no inhibition the major difference for co-infection is the stability of the viral load when the equilibrium is reached. Compare both the right graphs in [Figure 7](#pone-0012512-g007){ref-type="fig"} and [Figure 10](#pone-0012512-g010){ref-type="fig"} for days, hours. In co-infection we see that the viral loads for HBV and HDV have a very similar and rapid behavior, and that the higher number of infected cells are represented by the cells infected with both viruses, followed by the ones infected with HBV, and lastly the ones infected only with HDV, as can be observed in [Figure 10](#pone-0012512-g010){ref-type="fig"}. This contrasts with the very slow growth during the evolution of the disease when super-infection occurs, resulting in a different behavior of the number of infected cells. Recall that for the super-infection case, the smaller number of infected cells occurred for cells infected only with HBV.
{#pone-0012512-g010}
From the results obtained in this study, we can say that the development of the disease in super-infection is slower than in co-infection, suggesting that for super-infection a larger window of time is available before the beginning of therapy. Might that imply different reactions when therapies are applied? This will be explored in the following section.
The Model considering Antiviral Therapy {#s3c}
---------------------------------------
Antiviral therapy will have an impact on HDV viral dynamics depending on whether it is aimed at eliminating the virus itself including through modulation of the immune system (i.e. interferon-), or inhibiting HBV free virion production (i.e. lamivudine or ribavirine). Although there is no specific treatment for HDV infection, there has been some recent successful stories in treating patients [@pone.0012512-Gozlan1]. As mentioned before, the most common therapeutic approach is based on the administration of peggylated interferon- which helps promoting virus clearance. However, the clinical response is variable, and in most cases reversible upon interruption of treatment [@pone.0012512-Hoofnagle1]--[@pone.0012512-Lau1]. The concomitant use of antiviral drugs like ribavirin or lamivudine, which is believed to reduce the production rate of HBV free virions that are released from infected cells in the blood, showed no significant benefits in the treatment of hepatitis delta patients [@pone.0012512-Lau2]--[@pone.0012512-Yurdaydin1]. Although these drugs may have some inhibitory effect on HBV replication, they do not suppress HDV replication, probably due to the fact that HBsAgs expression is not significantly affected.
Let be the efficacy of inhibiting new virus infections as a consequence of virus clearance, and the efficacy of inhibiting viral production from infected cells, with both and in the interval . The antiviral impact from the different types of therapies can then be introduced in the initial model ((1)) as follows (changes in **bold**):
Although the mechanism of the action of interferon-, IFN, in HDV patients is not clearly understood [@pone.0012512-Niro1], some studies report improvements in patients, with IFN efficacy as high as 90% () [@pone.0012512-Chien1]. Although lamivudine (LMV) does not have a direct effect on HDV viral production (notice the absence of in the equations above) its effect on the viral production of HBV will also have an effect on the HDV viral dynamics. Studies such as [@pone.0012512-Lewin1] show efficacy levels of therapies based on LMV varying between 90% to 99% () for the HBV infection.
With this in mind, we considered 5 different scenarios of antiviral therapy responses for super-infected and co-infected individuals in our simulations: two with monotherapies with LMV and IFN alone and three others with LMV and three different efficacy levels of IFN, i.e. and (LMV antiviral therapy), and (IFN antiviral therapy) and with , and (LMV and IFN antiviral therapy). These last three scenarios represent ones where LMV is very efficient, but with different patient\'s response levels to IFN. Although the antiviral therapy responses for co-infected individuals were less pronounced, the results were analogous in nature for both types of infections; therefore, in the next five figures we only present the results for the super-infection scenario where therapy was applied for 168 days after the equilibrium of infection was obtained. The results here presented are also based on hours, , days, , , with all other constants specified in the different graphs below. We decided to omit a full discussion based on the effects of the different values considered for the parameters in the antiviral therapy model, due to the fact that the effects of these variations were the same as in the previous section, and therefore we concentrate our discussion on the new parameters of the model, and .
In the graphs that follow, the vertical dashed lines in the middle graph represent the beginning and the end of the antiviral therapy.
From the graphs, we observe the biphasic linear behavior of HBV viral load as previously reported in [@pone.0012512-Colombatto1], [@pone.0012512-Tsiang2] For the HBV viral load a marked decrease in the first days of therapy is observed, followed by a slower decrease. For HDV viral load, there is very slow decrease in the beginning of the therapy, which is not surprising due to what is believed about the no direct effect of IFN and LMV in HDV patients. The slow decrease is followed by a marked decrease, which becomes more and more parallel to the viral load of HBV as the efficacy of IFN increases.
[Figures 11](#pone-0012512-g011){ref-type="fig"} and [12](#pone-0012512-g012){ref-type="fig"} represent the dynamic behavior under monotherapy with LMV and IFN, respectively. We observe that even before the end of the therapy period, with only IFN, the HBV viral load starts to increase (right graph of [Figure 12](#pone-0012512-g012){ref-type="fig"}). The use of IFN alone seems to have only a momentary effect on the decrease of the infection.
{#pone-0012512-g011}
{#pone-0012512-g012}
Notice that when an effective response to IFN therapy is observed (values for ), as in [Figure 13](#pone-0012512-g013){ref-type="fig"}, the joint antiviral therapy is able to decrease the viral load on an order of 10 at the end of the therapy (from for HDV and for HBV only with LMV, to for HDV and for HBV for the joint therapy).
{#pone-0012512-g013}
From the simulation study we realize that as gets closer to 0.5 () as in [Figure 14](#pone-0012512-g014){ref-type="fig"}, the viral load behavior of HBV and HDV infections gets closer to the case when we only apply the lamivudine antiviral therapy (compare [Figure 11](#pone-0012512-g011){ref-type="fig"} to [Figures 13](#pone-0012512-g013){ref-type="fig"} and [14](#pone-0012512-g014){ref-type="fig"}).
{#pone-0012512-g014}
For the cases where IFN shows little efficacy (), the viral load dynamics of HBV and HDV is similar to the antiviral monotherapy where only LMV is applied, such as in [Figure 15](#pone-0012512-g015){ref-type="fig"}.
{#pone-0012512-g015}
However, in all cases there is a rebound of the infection when the therapy is terminated. Again, this feature may represent an artifact of this and other similar compartmental ODE models since a complete clearance of the virus is not predicted. However, a rebound of infection after ceasing treatment is a common observation in IFN treated patients (for a review see [@pone.0012512-Rizzetto2]).
Discussion {#s4}
==========
Mathematical models represent useful tools to predict the clinical course of virus diseases and the response to different antivirus therapies. They have been previously used for a number of human viruses, including HBV, HCV, and HIV [@pone.0012512-Brunetto1]--[@pone.0012512-DUgo1]. In this study, we describe, for the first time, a mathematical model for HDV infection. Since production of HDV infective particles is dependent on the simultaneous presence of HBV mathematical modeling of HDV infection poses an additional degree of complexity. Both HBV and HDV infect and replicate exclusively in liver cells. Since HDV co-infects or super-infects exclusively HBV infected individuals, the simultaneous behavior of both HDV and HBV and the interaction between the two viruses is considered according to the current knowledge of the biology of the viruses and clinical course of the disease. Little is still known about the mechanisms of HDV replication and its interaction of HBV *in vivo*. The woodchuck model for HDV/HBV infection was able to show some important tendencies on the clinical course of HDV infection, both during co-infection and super-infection. However, in humans the clinical course and response to antiviral therapy may largely differ between individuals and is believed to be also dependent on a number of variables, including the response of the immune system, which is still poorly understood. Since in this work we did not consider modeling the immune response of the host, the obtained results and conclusions shall be interpreted with regard to this limitation.
Six variables were taken into account in the herein proposed model: the number of uninfected cells, the number of HBV infected cells, the number of HDV infected cells, the number of cells simultaneously infected with HBV and HDV, the HBV viral load, and the HDV viral load. Six differential equations were obtained which were subsequently solved using the Matlab software [@pone.0012512-MathWorks1], and biological parameters previously described and used by others [@pone.0012512-Tsiang1]--[@pone.0012512-BlikkendaalLieftinck1], [@pone.0012512-Lewin1], [@pone.0012512-Bianchi1], [@pone.0012512-Nowak1] when modeling HBV infection. The obtained numerical solutions were consistent with those previously reported by others considering the HBV infection alone [@pone.0012512-Brunetto1], [@pone.0012512-Ribeiro1], [@pone.0012512-Tsiang1], [@pone.0012512-Lewin1]. In general, the predicted course of HDV infection is similar to that observed for HBV. Given the same initial viral loads of both viruses we observe a faster increase in the number of HBV infected cells and viral load. After reaching a peak, a small decrease in the HBV viral load and the number of infected cells is observed followed by a stabilization of these parameters of infection with small oscillations around what can be considered as a plateau. Concerning HDV, the increase in the number of infected cells and viral load is slower than the predicted for HBV. Usually, the plateau is reached between 200 and 500 days after infection depending on the initial viral load. In most tested scenarios, the number of HDV infected cells and viral load values remain below corresponding predicted values for HBV. The only exception is observed when an inhibitory factor *c* of HBV replication is introduced during super-infection. This issue is further discussed below. Previous studies aimed to evaluate HBV and HDV activity in infected patients were mainly performed using cross-sectional approaches together with qualitative analysis or low sensitivity quantitative analysis [@pone.0012512-Sakugawa1]--[@pone.0012512-Mathurin1]. This led often to contradictory conclusions with some authors showing that HBV replication may modulate HDV pathogenesis [@pone.0012512-Wu1], [@pone.0012512-Smedile1], [@pone.0012512-Su1] and others claiming that liver disease is mainly due to HDV infection [@pone.0012512-Sakugawa1], [@pone.0012512-Yamashiro1]--[@pone.0012512-Gudima1]. To our knowledge, a single quantitative longitudinal study of HBV DNA and HDV RNA dynamics, in 25 chronic patients, has been until now reported [@pone.0012512-Schaper1]. The authors show different replication profiles of HBV and HDV including fluctuating activities of both viruses with alternate predominance across 4--8 year periods of monitorization. Although these oscillations were also predicted by the present model further clinical studies are mandatory to confirm this observation.
Concerning the HDV infection, whether the co-infection or super-infection of HBV infected liver cells, we decided to consider different values for the biological parameters tested. A sensitive analysis was performed taking into account the data reported by other groups concerning both experimental animal infections and monitorization of the course of infection in human patients in the presence or absence of therapy. The parameters tested included the initial viral loads for HBV and HDV, the virion half-life, and the half-life of infected cells. The initial viral loads of an infected individual did not significantly alter the overall course of infection with the number of infected liver cells reaching a plateau which seems to remain stable in the absence of any therapy. The only noticed difference concerns the time needed to reach this plateau which is shorter when the initial viral load is larger. In contrast, the virion half-life and the time infected cells remain alive and thus secreting new infectious virus particles showed to critically influence the course of infection. We found that the combination of these two parameters influences the speed at which the number of HDV infected cells and HDV viral loads reach a plateau. In the case of HBV, it was suggested that the mean life-span of free virions in plasma could vary between 15 and 92 hours and the half-life of infected cells could reach 100 days [@pone.0012512-Lewin1], [@pone.0012512-Nowak1]. Since HBV and HDV share the same envelope proteins we decided to test the same values for the life-span of free virions. The half-life of HDV and HBV infected liver cells was also tested in the range of 10--100 days since we considered that co-infection with both viruses would not increase the half-life of cells when compared with the HBV infection alone. Surprisingly, in our model variations in the values of these two parameters showed to radically influence the possible course of infection. In general, the longer the life-span of free virions the faster the number of infected cells and free virions reaches a plateau. A similar picture was found for the different values of half-life of infected cells tested. Our simulation data suggest that a biological relevant scenario is established for values of half-lives of infected cells above 50 days or 20 days if the mean life-span of free virions will be over 32 hours. Moreover, if the mean half-life of infected cells is below 10 days, then the mean life-span of free-virions should be over 118 hours in order to be possible to observe an increase in the overall number of infected cells and HDV viral load. Since this value is higher than the mean half-life of virions in plasma calculated by others (36.9 hrs; Tsiang and Gibbs [@pone.0012512-Tsiang1]) it is possible that a therapy directed to reducing the half-life of infected cells below 10 days would significantly increase virion clearance. Identical pictures were observed for both the co-infection and super-infection scenarios. For values below those indicated by our model it seems probable that a spontaneous clearance of HDV infection would occur. However, to our knowledge, this scenario was not until now observed. It has been previously reported that during the acute phase of super-infection, HDV actively replicates while, at the same time, HBV replication is partially suppressed [@pone.0012512-Govindarajan2], [@pone.0012512-Lianjie1]. The degree of suppression of HBV replication, however, remains largely speculative. We decided to test the influence of HDV suppression of HBV replication by introducing a new variable *c* in our model. The values of *c* tested ranged from 0 (100% inhibition of HBV replication) to 1 (no inhibition of HBV replication). In general, we observed that variations in the values of *c* influenced the predicted relative number of HBV and HDV viral loads and infected cells. For high inhibition values of HBV replication () the number of HDV infected cells and HDV viral load surpasses the number of HBV infected cells and HBV viral load, respectively. This picture is not observed when *c* is set to 1 (no inhibition) or when low inhibition of HBV replication is considered ().
The potential to predict the behavior of virus infections under the presence of different antiviral therapies is one of the most important issues in mathematical modeling. Accordingly, we decided to test this model giving the most commonly used and generally accepted therapy approaches for HDV infection. These approaches are based on the use of nucleotide analogues, like lamivudine or ribavirine, peggylated -interferon or a combination of both. Nucleotide analogues are known to inhibit HBV replication with efficacies ranging between 90% and 99%, but seem not to have any effect on HDV replication *per se*. The detailed mechanisms of action of interferon- are still controversial but, in any case, it is generally accepted that it induces the expression of a large number of cellular proteins some of which have a direct antiviral effect. Moreover, interferon- has been implicated in the regulation of adaptative immune responses (reviewed by [@pone.0012512-Sadler1]).
When tested alone, lamivudine showed to be able to reduce significantly the HBV viral load (about 4 in 150 days) although a complete virus clearance could not be achieved. This may be due to limitations of the model as noticed also by Tsiang and Gibbs [@pone.0012512-Tsiang1] when modeling HBV infection alone. The observed reduction was biphasic with a first fast decrease in the beginning of treatment and a second slower in the following weeks. This biphasic behavior was previously reported for LMV treated patients [@pone.0012512-Lewin1]. Although the present model does not predict such a marked biphasic behavior, the decrease in the HBV viral load is clearly faster during the first 15 days of treatment. The biphasic decline of HBV viral load in patients under therapy aimed to inhibit virus production has been previously reported by others (Tsiang *et al.*, [@pone.0012512-Tsiang2]). This biphasic behavior was not accommodated by the initial model of HBV viral dynamics developed by Nowak *et al.* [@pone.0012512-Nowak1]. To overcome this problem Tsiang et al (1999, [@pone.0012512-Tsiang2]) introduced modifications in the assumptions of the efficacy of inhibition of viral infection. This enabled to show that increasing values of drug efficacy result in an initial faster decline of viral load followed by a slower second phase. The decline during the second phase was found to be similar independent of the values of inhibition tested.
In contrast, concerning HDV, LMV seems to reduce the HDV viral load with less efficiency. In this case, HDV virus clearance is slow in the beginning of treatment and then seems to decay exponentially.
In the case of LMV and interferon- (IFN) combination therapies we tested several scenarios that differed for the efficacy of IFN. For low (30%) and medium (70%) interferon efficacies, the combination therapy did not show a significant improvement in reducing the viral load when compared to the LMV monotherapy. However, if the efficacy of IFN is high (90%) the model predicts a 10 times reduction of HBV and HDV viral loads when compared with LMV monotherapy and low or medium efficacy interferon- combination therapies. The same pattern was observed for both co-infection and super-infection scenarios. In any case, we could not observe a complete clearance of virus infection after 6 months of therapy. Moreover, after ceasing therapy, a rebound of infection was in all cases observed. Tsiang and Gibbs [@pone.0012512-Tsiang1] reported a similar behavior when modeling HBV dynamics alone. This may be due to limitations of the present model since small amounts of free virus particles are predicted to survive in plasma even after prolonged treatment. In any case, we believe that further development of stochastic individual-based models for HDV infection is crucial to clarify this question.
Modeling the scenario for IFN monotherapy showed a significant decrease in the HDV viral load (3 in 150 days) and the number of infected cells. However, in this scenario the HBV viral load did not decrease significantly (1 in 150 days) displaying what seems to be a tendency to stabilize at high levels ( copies/mL).
Finally, both LMV monotherapy and combination therapy of LMV and IFN were predicted to more effectively reduce the HBV and HDV viral loads in the case of super-infection scenarios when compared with the co-infection. In contrast, IFN monotherapy was found to reduce the HDV viral load more efficiently in the case of super-infection while the effect on the HBV viral load was more pronounced during co-infection.
In conclusion, the combination LMV/IFN therapy seems to be more effective in reducing the number of infected cells and viral load of both viruses. LMV alone reduces the HBV viral load faster when compared with HDV, and IFN monotherapy has a significant effect in reducing solely the HDV viral load. In all tested scenarios a rebound of infection could be observed after the end of therapy.
Taken together, this model suggests that there is a need for development of high efficacy therapeutic approaches towards the specific inhibition of HDV replication. These approaches may additionally be directed to the reduction of the half-life of infected cells and life-span of newly produced circulating virions.
Further research is needed to overcome some of the limitations of the mathematical model here proposed; namely, to date, most of the constant parameters in the model are unknown for HDV infections. The present work is just a first step in trying to understand the HDV viral dynamics; however, a more in-depth look is necessary to understand the different behavior regarding co-infection and super-infection. The authors will continue their work of modeling the dynamics of HDV through a hierarchical Bayesian modeling approach. The main difference between this approach and the one here presented, hierarchical Bayesian versus mathematical, is that in a hierarchical Bayesian approach not only are the parameters no longer fixed quantities, but instead random quantities, their distribution depends on additional parameters, called the hyperparameters. The posterior distribution represents the uncertainty of the parameters after taking the data into consideration. MCMC (Markov Chain Monte Carlo) methods allow us to evaluate any characteristic of the posterior by simulating many sample values from it and then approximating any desirable characteristic from its corresponding sample value. The latest developments of free software such as R [@pone.0012512-R1] and WinBUGS [@pone.0012512-WinBUGS1] overcame some of the difficulties in the implementations of MCMC methods when fitting highly complex models.
Bearing this in mind, and working together with clinicians, we hope in the future to discover additional information regarding viral dynamics of HDV and thus contribute to a better understanding of this pathology during the different treatment therapies.
The authors would like to thank the Associate Editor and the two reviewers for their useful suggestions that improved the presentation of the paper.
**Competing Interests:**The authors have declared that no competing interests exist.
**Funding:**This work was supported in part by MCI grant MTM2008-01603. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
[^1]: Analyzed the data: BCdS CC. Wrote the paper: BCdS CC. Designed and programmed the simulation study: BCdS.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
A complex range of interactions exist between a pathogen with its host, which may include manipulation of the host for the pathogen\'s own advantage. There are several examples of viruses, such as rabies virus [@pone.0023866-Lagrue1], and parasites, including *Acanthocephala* spp. [@pone.0023866-Lefevre1] and *Toxoplasma gondii* [@pone.0023866-Berdoy1], that influence host behavior to increase their transmission efficiency. For years, scientists have been intrigued by the association between *T. gondii* infection and altered aversive behavior. The underlying mechanism(s) responsible for this behavior change are presently unknown. The aim of our study was to identify a possible explanation for this phenomenon.
*T. gondii* is a common, global protozoan parasite, which requires both a definitive host and an intermediate host to complete its life cycle. Although felines are the only definitive host of *T. gondii*, any warm-blooded animal, including humans, can be infected [@pone.0023866-Dubey1]. It is estimated that one quarter of the population (over 12 years of age) in the United States is positive for *T. gondii* infection (Center for Disease Control, USA, 2008). Prevalence in some areas can be as high as 95% in older populations. Latent, chronic infection, which is characterized by parasite encystment in the host muscle and brain cells (particularly neurons and glial cells), persists following the resolution of acute infection and continues with seropositivity throughout the host\'s lifetime [@pone.0023866-Dubey1]. Due to its high prevalence in the human population, it is critical to better understand the effects of *T. gondii* infection in the brain.
During the chronic stage of infection, infected rodents, which are a key intermediate host for *T. gondii*, exhibit a distinct repertoire of specific behavioral changes, including a loss of aversion to cat odors [@pone.0023866-Berdoy1], [@pone.0023866-Vyas1]. Infected rodents are, conversely, attracted to these odors, and this may be responsible for increased predation and for an increase in successful transmission of the parasite to the feline host; as cats are the only animal that can shed the environmentally-resistant stage of the parasite known as oocysts. This behavior change in infected rodents during the chronic stage of infection appears highly specific to feline odor, as a similar change is not evoked by other predators and has no effect on conditioned fear and anxiety [@pone.0023866-Lamberton1], [@pone.0023866-Webster1]. The underlying mechanism(s) responsible for this behavior change still remain unclear, however, it has been revealed that anti-psychotic (haloperidol) and mood-stabilizing medication (valproic acid) can prevent the development of these behavior changes [@pone.0023866-Webster1] in addition the dopamine uptake inhibitor GBR12909 modifies behavioral responses associated with latent toxoplasmosis in infected rodents [@pone.0023866-Skallova1]. Furthermore, we have recently identified an enzyme with tyrosine hydroxylase activity encoded in the *Toxoplasma* genome whose expression is induced during differentiation to tissue cyst stages [@pone.0023866-Gaskell1]. Several studies have suggested that *T. gondii* infection in humans can have serious neurological effects [@pone.0023866-Brown1]. Associations have been identified between *T. gondii* seroprevalence and schizophrenia [@pone.0023866-Brown2]--[@pone.0023866-Torrey1]. The schizophrenia-associated risk factors of *T. gondii* infection have been found to be greater than the risk factors associated with an individual\'s genes and with other environmental factors [@pone.0023866-Torrey1], [@pone.0023866-Purcell1]. Schizophrenia affects approximately 1% of the adult population and in most cases is a lifelong disease with exacerbations. Although schizophrenia is a multifactorial disease, pharmacological and genetic evidence suggest that dysregulation of dopamine metabolism is involved in schizophrenia [@pone.0023866-Howes1], [@pone.0023866-Seeman1].
Thus, it is crucial to examine whether dopamine metabolism is affected by *T. gondii* infection, particularly based on evidence of a tyrosine hydroxylase encoded by *T. gondii*. To address these questions, dopamine metabolism was monitored *in vivo* in the brains of chronically infected mammals and monitored *in vitro* during infection of neural cells.
Methods {#s2}
=======
Ethics statement {#s2a}
----------------
All animal work was performed according to national and international guidelines following approved animal procedures by the Beltsville Area Animal Care Committee, United States Department of Agriculture (Protocol no. 09-010--Toxoplasmosis in mice; approved June 4, 2009). This protocol is reviewed annually, and any amendments are approved separately.
Growth of parasites and host cells {#s2b}
----------------------------------
*T. gondii* strains were maintained in human foreskin fibroblasts (HFFs) as previously described [@pone.0023866-Gaskell1]. PC-12 cells obtained from ECACC (Salisbury) were maintained as described by the supplier.
Mouse strains {#s2c}
-------------
Female Swiss Webster mice infected with *T. gondii* VEG strain were used for histology.
Immunofluorescence assay of brain sections {#s2d}
------------------------------------------
Immunofluorescence against multiple targets was performed on paraformaldehyde-fixed, paraffin-embedded mouse brain sections. Female Swiss Webster mice were infected with *T. gondii* VEG strain oocysts 6--8 weeks prior to processing. Tissues were collected, formalin-fixed and paraffin-embedded using standard protocols and following approved guidelines. Slides were deparaffinized and rehydrated with an alcohol descending row, which was then followed by epitope retrieval in 10 mM sodium citrate buffer (pH 6.0) overnight at 60°C following sectioning. Slides were blocked with 2% normal goat sera for 1 h at room temperature. TRITC-conjugated lectin from *Dolichos biflorus* (Cat \# L9658, Sigma, St. Louis) was introduced to the slides for 4 h at room temperature, diluted 1∶200 in primary staining solution (1% BSA, 0.1% cold fish skin gelatine, 0.5% Triton X-100 in 0.1 M PBS pH 7.2). Next, samples were washed (3×10 min) in wash buffer (TBS pH 8.4 with 0.1% Triton X-100 and 1% fish skin gelatin) and blocked using a biotin-streptavidin blocking kit (Cat \# SP-2002, Vector Labs, Peterborough) according to the manufacturer\'s protocol. Samples were incubated with primary antibody (raised in rabbit) against dopamine (Cat \# ab8888, Abcam, Cambridge, MA) (diluted 1∶200) or tyrosine hydroxylase (Cat \# ab112, Abcam) (diluted 1∶500) overnight at 4°C. Samples were rinsed with wash buffer and incubated for 1 h with biotinylated anti-rabbit IgG secondary antibody (Cat \# B-1000, Vector Labs) diluted 1∶500 in secondary antibody solution (0.05% Tween in 0.1 M PBS pH 7.2). Sections were rinsed and incubated with FITC-conjugated streptavidin (Cat \# SA-5001, Vector Labs) diluted 1∶100 according to the manufacturer\'s guidelines in secondary antibody buffer at room temperature. Finally, slides were rinsed in wash buffer containing DAPI and double-distilled water prior to mounting in Fluoromount G (Southern Biotech, Birmingham). All incubation and blocking steps were carried out in a wet chamber. All slides were kept at 4°C in the dark before imaging using a Zeiss LSM510 META laser scanning inverted AxioVert 200M confocal microscope with DIC optics. 3D reconstructions of serial sections were generated with the same equipment using the LSM imaging software for the 3D deconvolution. To assess the specificity of dopamine staining, sections were incubated either without primary antibody or with primary anti-dopamine antibody in the presence of freshly prepared dopamine or serotonin for 30 min prior to and for overnight following addition to the sections.
A *T. gondii* tyrosine hydroxylase antibody custom antibody (Genscript, Piscataway) was developed to assess the parasite enzyme in animals. The affinity purified antibody is directed against a unique sequence (CIRSSPDPLDLRKMT) in the amino terminal domain that is not found in mammalian tyrosine hydroxylase and has no significant similarity to any protein in the predicted mammalian proteome or other proteins of the *T. gondii* proteome. The specificity of the antibody for *T. gondii* tyrosine hydroxylase was confirmed by Western analysis. Total protein from half mouse brains was isolated in 20 volumes (wt/vol)lysis buffer (20 mM Tris-HCl pH 8; 137 mM NaCl; 10% glycerol; 1% Triton X; 2 mM EDTA and protease inhibitors (cOmplete Mini EDTA-free cocktail, Roche)) and quantified using Bradford reagent (Sigma) as per manufacturer\'s instructions. Expression and purification of *T. gondii* tyrosine hydroxylase was as previously described [@pone.0023866-Gaskell1]. SDS-PAGE was following standard protocols with 2--20 µg protein separated on a 12% sodium dodecylsulphate- polyacrylamide gel. The proteins were transferred to nitrocellulose membrane, blocked with 5% non-fat dried milk in PBS containing 0.05% Tween-20 (vol/vol) for 1 hour. Incubation with the custom antibody (1∶2500) 4°C overnight was followed by washing in PBS-Tween (0.05%) and incubation with an anti-rabbit (1∶5000) conjugated horseradish peroxidase antibody (Sigma) at room temperature for 1 hour. Blots were then washed as above and developed using Supersignal West Pico Chemiluminescent kit (Pierce, Rockford, IL). Bands were visualised with an X-Omat film system. The membrane was stripped and re-probed with mouse anti-β-actin (1∶25,000; Sigma) overnight at 4°C followed by anti-mouse (1∶10,000) conjugated horseradish peroxidase antibody (AutogenBioclear, Wiltshire, UK) at room temperature for 1 hour and subsequent visualisation. The anti-*T. gondii* tyrosine hydroxylase antibody was used for immunofluorescence (diluted 1∶500) following a similar protocol as described above.
Immunofluorescence of tyrosine hydroxylase in cultured parasites was performed with paraformaldehyde-fixed cell cultures. Cultures of *T. gondii* stably expressing RFP-conjugated GRASP protein (kindly donated by Manami Nishi from David Roos laboratory, University of Philadelphia, USA) in human foreskin fibroblasts grown on polylysine-coated coverslips were alkaline induced for differentiation as published and differentiation monitored by counting the number of parasites in the vacuoles in the normal and pH shifted cultures. These conditions yielded bradyzoite forms as shown by RT-qPCR with the bradyzoite markers BAG1 and SAG4 [@pone.0023866-Gaskell1]. After five days, coverslips were paraformaldehyde-fixed and probed with tyrosine hydroxylase antibody (Cat \# ab112, Abcam) (diluted 1∶500) with visualisation as described above.
Immunohistochemical assay of brain sections {#s2e}
-------------------------------------------
For immunohistochemical assays, sections were treated as described above, except for the following steps: for washing and dilution buffers, 0.1 M PBS supplemented with 0.1% Tween was used. After antigen retrieval, slides were incubated in 0.3% H~2~O~2~ (in 0.1 M PBS) to quench endogen peroxidases. Following secondary antibody treatment, 5 µg/ml HRP-conjugated streptavidin (Cat \# SA-5004, Vector Labs) was applied, and next, the peroxidase substrate kit (Vector Labs, ImmPACT™DAB, Cat \# SK-4105) was used according to the manufacture\'s protocol. Prior to mounting, sections were stained with haematoxylin to visualize cell nuclei. Imaging of slides was performed using a Zeiss Axioplan microscope equipped with DIC optics. Photomicrographs were collected with a Photometrics CoolSNAP camera and Improvision Openlab software.
Glyoxylic acid staining {#s2f}
-----------------------
A cytochemical method was used to assess the dopamine staining of tissue cyst-containing neural cells in infected mice brain. Glyoxylic acid reacts with catecholamines in a gaseous reaction to form fluorescent products. Dopamine reacts with glyoxylic acid to form a product that specifically emits at 478--480 nm [@pone.0023866-Lent1].
Dopamine accumulation and release from dopaminergic cells {#s2g}
---------------------------------------------------------
The dopaminergic cell line PC12 (ECACC) was infected with Prugniard strain of *T. gondii* tachyzoites that had been alkaline shocked to induce bradyzoite differentiation. Liberated tachyzoites were incubated in RPMI media at pH 8 with 1% FCS at 37°C and ambient CO~2~ for 16--18 h, then rinsed with DMEM and returned to standard PC12 cell culturing conditions. PC12 cultures were infected 2.5×10^5^--7.5×10^5^ parasites and cultured for five days prior to assay. The cultures infected with higher numbers of parasites had parasitemia of 40--50%. Prior to assay, samples were normalized to equivalent numbers of cells (2.5×10^6^) per assay. One set of cultures was harvested by centrifugation and lysed by sonication in 0.1 M perchlorate for total dopamine measurement by HPLC with electrochemical detection. A parallel set of cultures were equilibrated with wash buffer with low KCl containing buffer (140 mM NaCl, 4.7 mM KCl, 1.2 mM MgCl~2~, 2.5 mM CaCl~2~, 11 mM dextrose, 10 mM HEPES, pH 7.4) for 30 min followed by incubation with two volumes high KCl containing buffer (40 mM NaCl, 100 mM KCl, 1.2 mM MgCl~2~, 2.5 mM CaCl~2~, 11 mM dextrose, 10 mM HEPES, pH 7.4) for 2 min to induce dopamine release as previously described [@pone.0023866-Yamboliev1]. During the assay, samples were taken from the media, washing buffer, and high KCl containing buffer and immediately supplemented with 0.3 volumes 0.1 M perchlorate. Three independent experiments were performed with a representative experiment shown.
Following centrifugation, cell homogenates and media were assayed by HPLC-ED. Reverse phase chromatography, combined with electrochemical detection, was performed with a Dionex HPLC system consisting of a P580 Pump (Dionex) and Ultimate 3000 Autosampler Column Compartment with a C18 Acclaim 120 column (5 µm, 4.6×150 mm) and an ESA Coulochem III cell, equipped with a glassy carbon electrode used at 700 mV versus Ag/AgCl reference electrode for detection of monoamines. The mobile phase consisted of degassed 57 mM anhydrous citric acid (Fisher Scientific, Loughborough), 43 mM sodium acetate (Dionex, Sunnyvale) buffer containing 0.1 mM EDTA (Sigma Aldrich), 1 mM sodium octanesulphonate monohydrate, and 10% methanol. The pH was adjusted to 4. The mobile phase was delivered at a flow rate of 1.5 ml/min, and the column temperature was set at 40°C. Applied standards (dopamine, L-DOPA) were dissolved in 0.1 M perchlorate for chromatography. The concentration of compounds was determined using Chromeleon software.
Results {#s3}
=======
Dopamine metabolism in infected neural cells in brain tissue {#s3a}
------------------------------------------------------------
A previous study found that the global content of dopamine in the brains of mice chronically infected with *T. gondii* was increased by 14% (114% of uninfected (P\<0.01)), whereas other neurotransmitters were unchanged [@pone.0023866-Stibbs1]. The localized effects of *T. gondii* infection on dopamine metabolism in tissue cysts have not been examined. *T. gondii* forms intracellular tissue cysts in neurons with each tissue cyst containing hundreds of bradyzoites (slowly dividing stage) that may remain *in situ* through the host\'s lifetime [@pone.0023866-Dubey1]. Formaldehyde-fixed brain sections from mice chronically infected with *T. gondii* were probed with dopamine antibody (Abcam). Dopamine antibody staining was readily apparent in infected cells ([Fig. 1](#pone-0023866-g001){ref-type="fig"}). Surprisingly, the localization was primarily within the *T. gondii* tissue cysts containing the parasites visualized as intensely stained cysts ([Figs. 1](#pone-0023866-g001){ref-type="fig"}, [2](#pone-0023866-g002){ref-type="fig"}), rather than the host neural cell. The dopamine antibody staining in tissue cysts was punctate. Image rotation illuminated staining throughout the tissue cyst with most concentrated staining near the periphery. The antibody was raised against dopamine glutaraldehyde conjugated to bovine serum albumin (BSA). The antibody also labelled neurons in the amygdala and hippocampus in uninfected and infected mice (data not shown); areas with a high concentration of dopaminergic neurons. Intracellular tissue cysts were identified based on morphology (for immunohistochemistry) and by labeling the periphery of the tissue cysts with fluorescently-tagged lectin (for immunofluorescence) [@pone.0023866-Coppin1]. DAPI counterstaining of nuclei visualized individual parasites in the tissue cysts, highlighting the hundreds of bradyzoites within each tissue cyst.
{ref-type="fig"}) as the antibody stained neurons in the striatum, amygdala and hippocampus. (C) A 3D projection of a Z-stack reconstruction of serial images of a tissue cyst within a brain section stained with anti-dopamine antibody and lectin as described in B. Control without the primary anti-dopamine antibody is shown in the right panel.](pone.0023866.g001){#pone-0023866-g001}
![Specificity of dopamine *staining T. gondii* tissue cysts.\
(A) Histochemical (glyoxylic acid) staining of dopamine in brain sections from chronically-infected mice detected by fluorescence. Glyoxylate reacts with dopamine to fluoresce blue-white [@pone.0023866-Lent1]. Cells containing *T. gondii* cysts in brain tissue exhibited blue-white fluorescence. The tissue cysts stained darkly, similar to mouse cell nuclei, presumably due the high density of bradyzoites. (B) Brain tissue sections from chronically-infected mice were stained with indirect fluorescein staining as in [Fig. 1](#pone-0023866-g001){ref-type="fig"} except the anti-dopamine primary antibody was incubated in the presence of 50 µg/ml dopamine (top) and 50 µg/ml serotonin (botto). From left to right: bright field, fluorescein only channel (green), fluorescein and lectin-TRITC (green and red channels, respectively), and both channels plus bright field. Serotonin did not compete for dopamine staining.](pone.0023866.g002){#pone-0023866-g002}
The presence of dopamine in tissue cyst-containing neural cells was confirmed by cytochemical staining and competition assays. Glyoxylic acid staining, a classic method for detection of dopamine-containing cells by chemical reaction of glyoxylic acid with dopamine to produce a fluorescent product [@pone.0023866-Lent1], was applied. Interestingly, the tissue cyst infected cells fluoresced blue and white, with the entire cell body of the infected cell displaying fluorescent staining ([Fig. 2A](#pone-0023866-g002){ref-type="fig"}). Staining of the encysted parasites within cells and neural cell nuclei are black due to the presence of parasite and host nuclear chromatin. The lack of cytosolic staining in the immunofluorescent images with dopamine antibody are likely to be due to saturation of the image with the very intense cyst staining. The specificity of the dopamine antibody was confirmed by competition assays. Primary dopamine antibody staining of tissue sections was performed in the presence of exogenous dopamine followed by secondary staining with fluorescein labelled antibody. This eliminated staining as visualized by loss of fluorescence ([Fig. 2B](#pone-0023866-g002){ref-type="fig"}). In contrast, addition of exogenous serotonin (another catecholamine neurotransmitter) did not disrupt staining with the dopamine antibody. This verifies that the dopamine antibody is detecting dopamine. It remains possible that the dopamine antibody is also detecting the metabolic precursor to dopamine, L-DOPA, although manufacturer (Abcam) tests show a \>400-fold higher affinity for dopamine compared to L-DOPA using conjugates to BSA. Competition assays exhibited some decrease in staining with exogenous L-DOPA although this was not quantifiable (data not shown).
*T. gondii* infected cells release high amounts of dopamine {#s3b}
-----------------------------------------------------------
To assess whether the dopamine detected in *T. gondii* tissue cysts could affect neurotransmission, the effect of *T. gondii* infection on dopamine release from dopaminergic neural cells *in vitro* was determined. PC12 cells were utilized as this cell line is the most commonly used *in vitro* model of dopaminergic neurons. Dopaminergic PC12 cells were infected with *T. gondii* parasites incubated under conditions that induce differentiation, and dopamine content and release were monitored by HPLC-ED. Conditions were used (as described in the Materials and [Methods](#s2){ref-type="sec"}) for stage conversion of tachyzoites (the rapidly dividing stage of *T. gondii*) into the tissue cyst stages (ie. bradyzoites) with alkaline pH and decreased CO~2~ content as described by others [@pone.0023866-Dubey1].
The total dopamine in infected PC12 cultures was measured to determine whether infection increases the amount of dopamine synthesized in dopaminergic cells. Cultures were infected with different numbers of alkaline-induced *T. gondii* and total dopamine was quantitated by HPLC-ED following washes with low KCl buffer. We found that infected cultures accumulated significantly greater levels of dopamine and the increase correlated with infection rate ([Fig. 3](#pone-0023866-g003){ref-type="fig"}). Infection led to greater than three-fold increase in total dopamine content compared to mock-treated, uninfected cells.
{#pone-0023866-g003}
Dopamine release assays were performed with cultures of *T. gondii*-infected PC12 cells to assess effects of infection on dopamine signalling. The cultures infected with different numbers of alkaline-induced *T. gondii* were induced to release dopamine with potassium as K+ causes release of dopamine in vesicles following methods in other studies [@pone.0023866-Yamboliev1]. As a result of infection, dopamine release increased in infected cultures in a dose-dependent manner with the number of parasites in the culture correlating with the amount of dopamine released ([Fig. 3](#pone-0023866-g003){ref-type="fig"}). Dopamine release in infected cells was up to 350% greater compared to dopamine release in uninfected cells. Dopamine release was specific for high KCl induction since wash buffer ([Fig. 3B](#pone-0023866-g003){ref-type="fig"}) and media alone (data not shown) did not induce the release of detectable amounts of dopamine in infected or uninfected cultures. The low KCl wash ensures that the dopamine released is induced by potassium and not due to dopamine released by cell lysis of infected cells. The dopamine release reported here is the minimum amount increased by *T. gondii* infection as less than or equal to half of the cells in the cultures were infected. Normalizing for the infection rate results in a seven-fold increase in dopamine release in infected cells relative to uninfected PC12 cells.
Taken together, an increase in dopamine content and an increase in dopamine release were observed in neural cells as a direct response to *T. gondii* infection. The enhanced dopamine release observed in infected cells in this study is likely to be an underestimate of the effect on dopamine release *in vivo*, cultured parasites contain few bradyzoites per vacuole compared to brain tissue cysts that contain hundreds of bradyzoites.
Tyrosine hydroxylase is expressed in bradyzoites {#s3c}
------------------------------------------------
Tyrosine hydroxylase is the rate-limiting enzyme in dopamine biosynthesis. Tyrosine hydroxylase localization in the brain sections of mice chronically infected with *T. gondii* was determined to examine the expression of this crucial enzyme in infected neural cells. Significant levels of tyrosine hydroxylase were localized *within T. gondii* tissue cysts in the brain sections of infected mice ([Fig. 4](#pone-0023866-g004){ref-type="fig"}). As expected, tyrosine hydroxylase was also found in the cytosol of neurons in the expected areas of the brain in both infected and uninfected mice (data not shown). Staining was not apparent in control sections that were treated with only secondary antibody. It is intriguing that both tyrosine hydroxylase and dopamine staining were localized in the tissue cysts of infected mouse brains, displaying similar staining patterns ([Figs. 1](#pone-0023866-g001){ref-type="fig"}, [4](#pone-0023866-g004){ref-type="fig"}). Thus, the rate limiting enzyme for dopamine synthesis and the product itself were both found *in T. gondii* tissue cysts in the brain.
{#pone-0023866-g004}
It is possible that the tyrosine hydroxylase expression observed within the tissue cyst could be either the *T. gondii*-encoded tyrosine hydroxylase or neuronal tyrosine hydroxylase that has been imported from the host. *T. gondii* has complex interactions with its host cell and co-ops several host proteins (e.g. calpains), and hence could potentially import neuronal tyrosine hydroxylase into the tissue cyst [@pone.0023866-Chandramohanadas1].
Alternatively, *T. gondii* could provide an enzyme with tyrosine hydroxylase activity. We previously described a *T. gondii* encoded tyrosine hydroxylase that could be expressed in the brain tissue cysts [@pone.0023866-Gaskell1]. *T. gondii* has two copies of the tyrosine hydroxylase gene encoding nearly identical proteins (97.5%) with one gene induced in bradyzoite-stage parasites. The parasite tyrosine hydroxylase has a high degree of homology (53% identity) with mammalian tyrosine hydroxylases. Unique for tyrosine hydroxylases, the parasite orthologue enzyme contains a putative signal sequence that could permit the enzyme to be trafficked to an organelle or secreted by *T. gondii*. Additionally, it was observed that the commercial tyrosine hydroxylase antibody used in these studies recognizes the *T. gondii* encoded orthologue, as well as the mammalian tyrosine hydroxylases (unpublished observations). Indeed, *in vitro* cultivated parasites under alkaline conditions that induce formation of bradyzoites bind commercial tyrosine hydroxylase antibody ([Fig. 4B](#pone-0023866-g004){ref-type="fig"}). The tyrosine hydroxylase antibody stains the parasitophorous vacuole and also stains the periphery of parasites.
To specifically identify parasite-encoded tyrosine hydroxylase within brain tissue, a custom antibody for *T. gondii* tyrosine hydroxylase (TgTH) was developed. The target sequence of this custom antibody is located in the amino-terminal domain of TgTH, which is unique and divergent from mammalian tyrosine hydroxylases. *T. gondii* tyrosine hydroxylase was localized within tissue cysts in neural cells in chronically-infected mouse brains ([Fig. 5A](#pone-0023866-g005){ref-type="fig"}). Staining was only detectable in tissue cysts in infected neurons. Western analysis was performed to validate the specificity of the TgTH antibody. The antibody recognized recombinant TgTH but did not bind mouse brain proteins, confirming the specificity of this antibody for *T. gondii* tyrosine hydroxylases ([Fig. 5B](#pone-0023866-g005){ref-type="fig"}). Hence the intense dopamine antibody staining and TgTH are both found in *T. gondii* brain tissue cysts.
![Expression of a parasite-encoded tyrosine hydroxylase in brain tissue cysts.\
(A) 3D projections of serial images of *T. gondii* tissue cysts within brain sections were triple stained with *T. gondii* encoded tyrosine hydroxylase (TgTH) antibody (green), DAPI (blue), and lectin (red). The panels (from left to right) show all three channels, the lectin and antibody, and TgTH antibody alone (green). Staining was not apparent in control sections that received only secondary antibody (data not shown). DAPI identified neuronal cells and the individual bradyzoites within the tissue cyst and lectin stained the surface of the cyst. (B) Western analysis for specificity of the custom antibody for TgTH. Recombinant protein from Δ29TgAaaH2 [@pone.0023866-Gaskell1] and mouse brain were probed with TgTH antibody. No bands were detected in uninfected mouse brain. β-actin was used as a loading control.](pone.0023866.g005){#pone-0023866-g005}
Discussion {#s4}
==========
Changes in behavior of the intermediate host that could lead to increased transmission of a parasite to its definitive host are likely to be positively selected as these changes would provide a significant benefit in completion of the parasite\'s life cycle. *T. gondii* induces behavioral alterations in infected rodents that would facilitate the transmission of the parasite to its definitive feline host, however, the mechanism responsible for these changes remains unclear. Our study provides a mechanism for these changes. Previous studies showing that anti-dopaminergic drugs can prevent the development of the behavior changes in rodents suggest that dopamine regulation altered by *T. gondii* infection of mammals [@pone.0023866-Webster1].
The altered behavior may be a direct effect or an indirect effect of *T. gondii* infection. In this study, significant levels of dopamine was detected by immunohistochemistry in *T. gondii* tissue cysts in the brain ([Fig. 1](#pone-0023866-g001){ref-type="fig"}), as well as, increased dopamine release from dopaminergic cells infected with *T. gondii* ([Fig. 3](#pone-0023866-g003){ref-type="fig"}). Based on these novel findings, this is the first study to suggest that a parasite can directly alter dopamine signalling to mediate host behavior changes. These results provide a potential mechanism for *T. gondii*-induced host behavioural changes.
In our study, localizing the changes in dopamine metabolism during infection was crucial, as the location of dopamine metabolic changes in the brain is likely to be a critical factor for its effect on host behavior. Encysted *T. gondii* have been observed in functional neurons with intact synapses [@pone.0023866-Melzer1], [@pone.0023866-Ferguson1]. Tissue cysts have been detected throughout the brain, although higher percentages of cysts were reported in the amygdala and nucleus accumbens [@pone.0023866-Vyas1], [@pone.0023866-Gonzalez1]. These limbic brain regions are well known to contain dopamine that plays important functions in the control of movements (basal ganglia), reward to stimuli, pleasure, dependency (nucleus accumbens and hippocampus), motivation and cognition, and species and stimuli specific fear (amygdala). Altered dopamine levels induced by *T. gondii* in tissue cysts in these regions of the brain could have significant harmful consequences on a variety of brain functions, possibly leading to an array of behavioral changes and possible neurological malfunctions.
The observed intense dopamine staining within the *T. gondii* tissue cysts in brains was unexpected. Dopamine in neurons is synthesized in the cytosol, packaged into vesicles, and transported along axons [@pone.0023866-Cartier1]. Thus, dopamine staining in neurons is primarily detected within vesicles. Indeed, cytosolic dopamine can induce cell apoptosis if it is not properly packaged into vesicles [@pone.0023866-Ogawa1]. Packaged dopamine in neurons is rapidly transported away from the cell body to the axon terminal. In our brain sections, any dopamine released from the cyst into the cell body of the neuron would be packaged and transported by the efficient dopaminergic vesicle transport along axons. This may explain the apparent lower level of dopamine in the host cell body compared to the tissue cyst ([Figs. 1](#pone-0023866-g001){ref-type="fig"}, [2](#pone-0023866-g002){ref-type="fig"}). Alternatively, the observed staining by the dopamine antibody could be due to detection of L-DOPA within the *T. gondii* tissue cyst that escapes the cyst and is metabolised into dopamine in the host cytosol. This interpretation is coherent with the observed cytosolic staining of infected neurons using glycoxylic acid that yields a specific product with dopamine ([Fig. 2](#pone-0023866-g002){ref-type="fig"}). The parasite provides tyrosine hydroxylase, the rate-limiting enzyme in dopamine synthesis but the source of DOPA decarboxylase required for conversion of L-DOPA to dopamine needs further investigation. DOPA decarboxylase present in the cytosol of dopaminergic neurons could provide this enzyme. A hypothetical nutrient pore expressed in the parasitophorous vacuole membrane of *T. gondii* tachyzoites that permits the passage of metabolites (\<1300 Da) from the host cell cytosol into the parasitophorous vacuole could allow passage of small compounds from the vacuole into the host cytosol [@pone.0023866-Schwab1]. If the pore is expressed in bradyzoites then it could provide a means for dopaminergic metabolites (L-DOPA, dopamine) to exit the vacuole and enter the host cytosol where L-DOPA would be converted to dopamine by cytosolic DOPA decarboxylase and dopamine would be packaged into secretory vesicles. The generation of *T. gondii* mutants that will provide a conclusive dissection of the role of the parasite\'s tyrosine hydroxylase in the dopamine synthesis and release are in progress, but with either site of dopa decarboxylase action, the increased dopamine metabolism has important implications on the host neurochemistry.
In addition to dopamine, neurotransmitters such as serotonin and glutamate need to be considered in *T. gondii*-induced behavioral changes. Prior studies have proposed that the host immune response to *T. gondii* infection may lead to altered neurotransmitter levels [@pone.0023866-Webster2]. Immunocompetent hosts control chronic *T. gondii* infection with a T-lymphocyte--driven defense [@pone.0023866-Denkers1]. Infection of mice with *T. gondii* elicits a dominant Th1 response involving interferon-gamma (IFN-γ), interleukin-12 (IL-12), IL-18, and tumor necrosis factor alpha (TNF-α). TNF-α induction has a serious impact on *T. gondii* induced pathology at early stages of infection. Th2-associated cytokines, such as IL-4 and IL-10, appear relatively late after infection and may limit immune pathology. To resolve acute infection, IFN-γ induces indoleamine 2,3-dioxygenase (IDO) release, resulting in tryptophan degradation and kynurenic acid accumulation [@pone.0023866-Silva1]. Tryptophan depletion is thought to be responsible for suppression of the growth of the acute stage tachyzoites. Changes in serotonin levels were not observed in mice with *T. gondii* chronic infections although there may be localized undetected changes [@pone.0023866-Stibbs1]. Kynurenic acid accumulation in the CNS could potentially alter dopamine metabolism due to its NMDA antagonistic properties [@pone.0023866-Mortensen1]. Thus, the host immune response to *T. gondii* infection could contribute to alterations in neurotransmitter levels that could affect behaviour in conjunction with the increased dopamine mediated by the parasite. Further studies are essential to investigate these possibilities.
Behavioral changes associated with *T. gondii* infection may contribute to serious neurological disorders in humans. Several studies have observed an association between *T. gondii* seroprevalence with schizophrenia [@pone.0023866-Brown1], [@pone.0023866-Torrey1]. Since *T. gondii* infection has been found to last throughout the lifetime of the host, seroprevalence is likely to reflect chronic infection [@pone.0023866-Dubey1]. Dopamine dysregulation is proposed to play a central role in schizophrenia, potentially in combination with glutamate metabolism. How dopamine dysregulation plays a role in schizophrenia, however, is still unknown. The principal antipsychotic drug that has been used to treat schizophrenia, dopamine antagonist haloperidol, can also block the development of behavior changes in *T. gondii* infected rodents. It is possible that the increased dopamine accumulation and release observed during *T. gondii* infection may contribute to *T. gondii* associated schizophrenia. Dopamine metabolite concentrations have been inversely correlated with gray matter volume in schizophrenia patients, and recent MRI evidence found that the majority of volume reduction is in those patients seropositive for *T. gondii*, suggesting that *T. gondii* infection leads to an increase in dopamine metabolite concentrations [@pone.0023866-Breier1], [@pone.0023866-Horacek1]. It would be of interest to analyze the ability of other pathogens associated with schizophrenia, and other neurological disorders, to directly alter dopamine metabolism to see if other pathogens have this ability or if this phenomena is unique to *T. gondii*.
Malfunctions of dopamine metabolism have a serious impact on human behavior. Dopamine dysfunction has been associated with a variety of neurological disorders including schizophrenia, attention deficit hyperactivity disorder, tic disorders, Tourette\'s syndrome, and dyskinesias. The novel findings of this study, that demonstrate *T. gondii\'s* ability to directly alter dopamine levels will not only help to better understand the relationship between schizophrenia and *T. gondii* seroprevalence, but these findings may be critical for understanding the mechanism(s) involved in a variety of pathogen-associated neurological disorders [@pone.0023866-Brown1], [@pone.0023866-Torrey1]. Thus, it is crucial to determine if other pathogens associated with neurological disorders also have the ability to directly alter dopamine levels. It is also critical to determine the possible contributions of *T. gondii* infection to other dopamine-related diseases [@pone.0023866-Brynska1], [@pone.0023866-Miman1].
We would like to acknowledge Dr. Oliver Kwok and Gareth Howell for technical assistance and Prof. Elwyn Isaac and Drs. Sophie Bamps and Christopher D. O\'Donnell for helpful discussions and comments on the manuscript.
**Competing Interests:**The authors have declared that no competing interests exist.
**Funding:**This project was funded by the Stanley Medical Research Institute (to JPW, GAM) and the USDA CRIS 1265-32000-076 (to JPD). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
[^1]: Conceived and designed the experiments: GAM EG EP. Performed the experiments: EG EP HM JPD. Analyzed the data: EP GAM. Contributed reagents/materials/analysis tools: GAM JPD. Wrote the paper: GAM EP JPW JPD.
| {
"pile_set_name": "PubMed Central"
} |
Dear Editors,
This is with reference to my paper, '*John Locke on Personal Identity*', published in Mens Sana Monographs (2011) Vol. 9, No 1, Jan--Dec 2011 (pp. 268-75\[[@ref2]\]). I thank the editors for bringing it to my notice that the citations in the paper were inadequate. Kindly find the necessary citations which were earlier not mentioned in the paper.
"Personal identity theory is the philosophical confrontation with the ultimate \...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\... the identity of the person over time." (Korfmacher, 2006\[[@ref1]\]) \[p. 268 of my paper\]"Locke holds that personal identity is a matter of psychological continuity." (Korfmacher, 2006\[[@ref1]\]) \[p. 269 of my paper\]"Locke\'s answer to both of these questions is in the affirmative. Consciousness \...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\...\... affirmations amount to the claim that the same soul or thinking substance is neither necessary nor sufficient for personal identity over time." (Uzgalis, 2007\[[@ref3]\]) \[p. 269 of my paper\]"One answer is that the distinction solves the problem of the resurrection of the dead. What is this problem? The problem begins with Biblical texts asserting that we will have the same body at the resurrection as we did in this life." (Uzgalis, 2007\[[@ref3]\]) \[p. 270 of my paper\]"His account of personal identity is embedded in a general account of identity". (Uzgalis, 2007\[[@ref3]\]) \[p. 273 of my paper\].
Please accept my sincere regret and apologies for the inconvenience caused to the editors of Mens Sana Monograph and readers. I once again thank the editors for being patient and understanding towards me, in the entire ordeal.
CITATION: Nimbalkar N. Corrections to my paper: John Locke on Personal Identity. Mens Sana Monogr 2014;12:168-9.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Obesity contributes to the development of several chronic diseases and leads to important health-care costs \[[@B1]\]. Consequently, the management of obesity represents an important public health issue. Physical activity is a healthy behavior recommended for the prevention and treatment of obesity \[[@B2]\]. However, despite substantial investments in the promotion of physical activity in the past decade, an important proportion of the population is still inactive \[[@B3]\]. Thus, the promotion of regular participation in physical activity remains an important challenge for researchers and clinical practitioners. In a recent systematic review, it was shown that the effects of interventions aimed at modifying this behavior among obese individuals are quite modest \[[@B4]\]. In this review, a number of reasons were suggested to explain this situation and the authors stressed the importance of identifying efficient approaches to change physical activity behavior among obese individuals.
Recent developments in the domain of health psychology highlight the influence of measurement on the subsequent health-related behavior of individuals \[[@B5]\]. Indeed, a few researchers have observed that a rudimentary exercise such as completing a questionnaire on cognitions regarding a given behavior can actually change this behavior. In this respect, French and Sutton \[[@B6]\] have highlighted the importance of studying this effect for the prediction of habitual health behaviors. This phenomenon, known as the \"mere-measurement effect\" \[[@B7]\] or \"self-generated validity effect\", has been observed for a variety of health-related behaviors \[[@B8]-[@B13]\], including physical activity \[[@B14]\]. For instance, Williams, Block and Fitzsimons \[[@B14]\] observed that asking participants to answer a single intention question about participation in exercise led to a significant increase in exercise frequency (small-to-medium effect size; *d*= 0.26) at two-month follow-up \[[@B15]\]. This pattern of results has also been observed for commitment to health and fitness assessment \[[@B10],[@B11]\] and for self-reported walking \[[@B16]\], also leading to small-to-medium effect sizes on behavior (*d*= 0.20, 0.28 and η^2^= 0.07 respectively). However, these four studies were realized among undergraduate university students and were not based on clear theoretical frameworks, with the exception of the study by Spence et al.\[[@B16]\] which was inspired by the Self-efficacy Theory \[[@B17]\].
Although the mere-measurement effect is more likely to be detected for self-reported behaviors \[[@B5]\], this effect has also been observed for other objectively measured health-related behaviors. In a recent study, Godin et al. \[[@B9]\] reported significantly higher proportions of blood donations among blood donors after the completion of a questionnaire based on an extended version of the Theory of Planned Behavior (TPB). The number of blood donations was extracted from the electronic database of the local organization responsible for the blood drive and this effect was still significant at one-year follow-up. In a subsequent study, Sandberg and Conner \[[@B8]\] replicated this finding among a sample of women asked to complete a questionnaire on cervical screening attendance also based on the TPB \[[@B18]\]. Similarly, attendance was extracted from a database rather than being self-reported. Although the mechanisms by which the mere-measurement effect occurs are not fully understood \[[@B6],[@B19]\], it would appear that completing a TPB questionnaire could lead to a significant increase in future behavior. It must be mentioned that the TPB represents one of the most empirically supported theories of social psychology for the prediction of health-related behaviors, including physical activity \[[@B20]-[@B23]\] and has been recently identified as one of the most effective theories to inform an internet-based intervention in the context of physical activity \[[@B24]\].
Thus, to our knowledge, the present experimental study is the first to test the effectiveness of the mere measurement effect among overweight and obese individuals. As such, it offers the potential to identify a novel approach to behavior change.
Methods
=======
Design and sample
-----------------
The participants of this study were involved in a larger six-month longitudinal study on genetic \"susceptibility\" to obesity \[[@B25],[@B26]\]. The sample was drawn from a population of volunteer adults recruited in the Quebec City metropolitan area via local newspapers and radio advertisements between May 2004 and March 2007. Also, e-mails were sent to university students and employees as well as to hospital and government employees. The inclusion criterions for the study were to be aged between 18 and 55 years and to have a BMI ≥ 25 kg/m^2^. A trained research assistant conducted a 15-minute telephone interview with people who responded to the advertisements. They were then alternatively allocated to the experimental or control conditions depending on the order of the telephone interview. Both participants and interviewers were blind to the objective of the study that is testing a mere measurement intervention; the study was presented as a study on motivation. All participants signed the study consent form approved by the Ethics Committee of the local university.
Measures
--------
Both groups also completed the Minnesota leisure-time questionnaire to assess baseline levels of physical activity (i.e., daily energy expenditure) \[[@B27]\]. This questionnaire was administered by a trained interviewer who provided participants with detailed instructions and a list of clearly defined physical activities. Participants were asked to indicate whether they had performed or not each physical activity over the last year, when they performed these activities, at what frequency per month and for how long. Overall, the Minnesota leisure-time questionnaire contains 63 items related to sports, recreational, yard and household activities. This instrument presents adequate psychometric values; reliability (test-retest correlation coefficients of .92 and .98) and the correlation coefficients for convergent and concurrent validity were between .33 and .63 when compared to other physical activity questionnaires and .47 with peak oxygen consumption \[[@B28]\].
A second measure of physical activity was obtained at three-month follow-up by means of another previously validated self-administered questionnaire \[[@B29],[@B30]\]. Follow-up LTPA was assessed using the following question: \"Within the last three months, how often did you participate in one or more physical activities of moderate intensity, totaling at least 30 minutes in the same day during your leisure time?\" Responses were reported on a 7-point scale varying between (1) not at all to (7) four or more times a week. The LTPA psychosocial questionnaire assessed the constructs of an extended version of the TPB that included intention (3 items), perceived behavioural control (PBC) (3 items), attitude (6 items), subjective norm (3 items) and respective beliefs (8 and 5 items respectively). Moreover, additional variables from other theories known to contribute to the explanation of health related behaviors (i.e.: anticipated regret (3 items), moral (3 items) and descriptive (2 items) norms, self-efficacy (3 items), facilitating factors (5 items), and positive feelings (3 items)) were also assessed \[[@B31]\]. All 47 items were presented in reference to the studied behaviour defined as follow: \"to regularly participate in one or more physical activities during the next three months\". The LTPA psychosocial questionnaire was developed in accordance with guidelines provided by Ajzen \[[@B32]\] and Godin and Kok \[[@B20]\]. The majority of the theoretical constructs presented adequate internal consistency (α = 0.73 to 0.90 or Spearman\'s r = 0.33 to 0.52, *p*\< 0.001), except for positive feelings (α = 0.68) and facilitating factors (α = 0.61) which had moderate internal consistencies.
Results
=======
Sample Characteristics
----------------------
Of the 452 overweight or obese (all BMI ≥ 25 kg/m^2^) participants who completed the baseline questionnaires, 373 successfully completed the study at the three-month follow-up and were retained for data analysis (Figure [1](#F1){ref-type="fig"}). The overall attrition rate was 17.5% and did not differ significantly between the two groups. Baseline characteristics of the sample are presented in Table [1](#T1){ref-type="table"}. The two groups did not differ significantly in any of the variables assessed at baseline. On average, participants reported to spend 1750-1850 calories per week, suggesting that participants from both conditions were already physically active according to the recent American and Canadian physical activity recommendations (i.e., \> 1000-1500 kcal/week) (US guidelines; SCPE). Finally, participants who completed the study did not differ significantly from those who dropped out on all baseline variables, with the exception of gender; women were more likely to drop out than men, *χ*^2^(1, N = 452) = 10.6, *p =*0.001. Nonetheless, the results for the mere-measurement effect did not differ when controlled for gender (data not shown).
{#F1}
######
Baseline means and standard deviations of the socio-demographic characteristics of the sample by condition (N = 452)
Variables Control group (F&V) Experimental group (LTPA)
----------------------------- --------------------- ----------------------------- --------------
Age (y) 40.3 (10.7) 40.0 (10.7) 0.35 (0.73)
BMI (kg/m^2^) 30.8 (4.6) 31.1 (5.0) -0.73 (0.47)
Energy expenditure (kcal/d) 263.2 (199.3) 250.4 (211.7) 0.66 (0.51)
***N*(%)** **χ**^**2**^**(*p*-value)**
Gender
Male 110 (48.9) 128 (56.4) 2.55 (0.11)
Female 115 (51.1) 99 (43.6)
Education level
Primary/secondary 34 (15.1) 35 (15.4) 0.008 (0.93)
At least collegial 191 (84.9) 192 (84.6)
Annual income
\< 30,000 90 (40.9) 103 (45.6) 0.99 (0.32)
≥ 30,000 130 (59.1) 123 (54.4)
Mere-measurement Effect
-----------------------
An analysis of covariance with baseline level of physical activity as covariate was performed to evaluate the mere-measurement effect of completing a TPB questionnaire. Baseline level of physical activity was included as covariate given that the follow-up measure of physical activity was based on a different instrument. This analysis revealed a main effect for the study condition, *F*(1,370) = 6.85, *p*= .009. Post hoc t-test analysis revealed that the mean score of physical activity level at follow-up in the experimental group (*M*= 5.23, *SD*= 1.61) was significantly higher (*p*= 0.05) than the mean score in the control group (*M*= 4.90, *SD*= 1.63), leading to a significant small effect size (*d*= 0.20, CI~95%~= 0.00-0.41) \[[@B15]\].
Discussion
==========
Findings of the present study indicate that asking questions about cognitions regarding physical activity positively influence subsequent self-reported participation in this behavior over a three-month period. Moreover, these results compare favorably to a recent meta-analysis that observed short- and mid-term (i.e., six to twelve months of follow-up) small-to-medium effect sizes on self-reported physical activity (SMD = .28, CI~95%~= 0.15 to 0.41) in more intensive interventions among adults \[[@B33]\]. Thus, results from the present study suggest that having only overweight/obese individuals complete a TPB questionnaire that also included other theoretical constructs such as anticipated regret, moral norm, and positive feelings is an easy and inexpensive way to promote exercise, especially in clinical settings where time is scarce (e.g., in the physician\'s office).
In spite of the growing scientific evidence of the effect of the mere-measurement of cognitions on behavior, the review by French and Sutton \[[@B6]\] highlights the fact that this phenomenon is expected to be small for health-related behaviors and to be mostly observed for non-complex behaviors such as those requiring a single action (e.g., giving blood). However, our results do not support this view given that physical activity is considered to be a complex behavior defined along dimensions of frequency, intensity of practice and duration of exercise sessions. According to French and Sutton \[[@B6]\], this mere-measurement effect is also more likely to be detected when health-related behaviors are subjectively assessed. As such, it is noteworthy that our measure of physical activity was self-reported. However, mere-measurement effects were observed for physical activity assessed either by self-reported or accelerometer (in the corrected model) in a study by van sluijs et al. \[[@B5]\]. There are also few studies that showed this effect for objectively assessed behavior \[[@B8],[@B9],[@B34]\]. According to these latter authors, the question now is not about the existence of mere-measurement, but about what is causing this effect and under which conditions this phenomenon could be observed.
In the present study, the full constructs of the TPB as well as additional variables from other theories were evaluated in the psychosocial questionnaire on LTPA. This questionnaire measured ten theoretical constructs and three sets of beliefs that required about fifteen to twenty minutes to complete. Consequently, individuals were guided towards deliberation about their behavior throughout the questionnaire (i.e., weighting the desirability and feasibility of taking action) that could lead to greater introjections and behavioral resolutions to become more active. Although the design of the present study could not allow any conclusion about the mechanisms underlying this mere-measurement effect, one could surmise that such an effect is less likely to emerge after completing shorter questionnaires. However, previous studies have reported that assessing only intention by means of a few items changes subsequent performance of studied behaviors at follow-up \[[@B10],[@B12],[@B35]\], including physical activity \[[@B14]\]. Others might also infer that this effect might be the consequence of assessing specific cognitions such as those included in the questionnaire of the present study. In this line of thought, Sandberg and Conner \[[@B8]\] observed that measuring anticipated regret (i.e., the anticipation of regret not to perform the behavior) was an important variable responsible for mere-measurement effect on cervical screening attendance. Since anticipated regret was one of the variables assessed in our extended TPB questionnaire, it can be hypothesized that participants in the present study anticipated feelings of regret about not being physically active in response to the questionnaire and, consequently, increased their level of physical activity.
The precise mechanism behind the mere-measurement effect is still a matter of debate \[[@B6],[@B19]\]. At this time, the dominant explanation of this effect is that asking behavioral intention questions heightens the accessibility of the person attitude towards the behavior, which in turns increases the likelihood that the behavior will be performed \[[@B36]\]. Morwitz and Fitzsimons \[[@B36]\] showed that responding to a query about one\'s purchase intention increases the activation level of one\'s pre-existing brand attitude. When the respective brand attitude was both highly accessible and positively valenced then participants were likely to choose that brand. On the other hand, when the activated attitude was negatively valenced this led to a decrease in the choice of this brand. However, the mere-measurement effect may only operate among those whose thoughts and feelings are favorably disposed towards the behavior. In a laboratory setting Morwitz and Fitzsimons \[[@B36]\] showed that positive attitudes increased brand choice, while negative attitudes decreased brand choice. Obviously, future research in this direction is needed to elucidate the potential mechanisms of action of the mere-measurement effect on changes in physical activity behavior.
The relatively modest costs and simplicity of the intervention may add to the appeal of mere measurement as an important additional strategy for improving public health via health behavior change. Nevertheless the use of the mere-measurement effect as a mean to promote public health is likely to be limited by the compliance of individuals to complete adequately the questionnaire. Indeed, questionnaire completion appears to be a prerequisite for a mere-measurement effect to occur \[[@B8],[@B9],[@B34]\]. Consequently, an important implication for using the mere-measurement effect to promote health behaviors is that studies will need to maximize completion rates of the questionnaire in order to produce the greatest impacts on the targeted behavior. Further research could also examine the added value of combining the mere measurement effect with interventions that promote questionnaire completion and/or positive attitudes toward the target behavior. For example, Dillman \[[@B37]\] provides guidance on the content and type of cover letter, questionnaire format, token incentive and return envelop that could be used to improve response rates to mailed questionnaire.
Notwithstanding this positive and significant effect of mere-measurement of cognitions, it must be acknowledged that the effect remains small and might benefit from being used in combination with other techniques for behavior change (see Abraham and Michie \[[@B38]\] for a description of a set of theoretical techniques). Moreover, from a public health perspective, if asking questions regarding a given behavior reinforces or induces its adoption, caution should be exercised before asking questions about sedentary behavior. Indeed, Williams, Block and Fitzsimons \[[@B14]\] observed an increase in illegal drug use following the completion of one item about intention to use illegal drugs. Thus, researchers investigating cognitions regarding sedentary behaviors among overweight or obese individuals instead of those related to physical activity participation must be aware that they may cause more harm than good in asking such questions.
Some limitations of this study must be acknowledged. Firstly, the present study was completed among a group of volunteers representing individuals interested in the topic of the study. Thus, this effect might be more important in this kind of sample, given that the cognition scores were relatively high (data not shown). Sprott, Spangenberg and Fischer \[[@B11]\] also investigated the effect of mere-measurement (described as self-prophecy) when the behavior under investigation was considered by the participant as right or wrong. Their results indicated that this effect was more likely to occur among participants with stronger normative beliefs regarding low-fat snack consumption and health and fitness assessment. Secondly, although we used an experimental design, it was not possible to apply an intention-to-treat approach given that different measures of physical activity were used at baseline and follow-up. Thirdly, subjective measures of physical activity were used in the present study, although both tools have been validated. Also, given that in the present study the measure of physical activity at follow-up differed from the measure used at baseline, it is less likely that the size of the observed effect might be attributed to measurement habituation. Nonetheless, it will remain important in future studies to conduct randomized controlled trials with objective measures of physical activity. Finally, additional research should be conducted among other segments of the overweight/obese population, especially among sub-groups with low socio-economic status.
Conclusions
===========
To conclude, the present study is the first to report the effect of measurement of cognitions on physical activity among a sample of overweight and obese individuals. This adds to the growing evidence that asking questions influences subsequent physical activity behavior. It also paves the way to new approaches for changing behavior, since this method requires low investments in terms of time and money and could easily be integrated into more comprehensive prevention and health promotion programs, and health care practice.
Competing interests
===================
The authors declare that they have no competing interests.
Authors\' contributions
=======================
GG, MCV and LP designed the study and participated in its coordination. SA participated in data collection and performed data analysis, along with ABG and GG. ABG and GG wrote the draft version of the manuscript. SA, MCV and LP critically revised the manuscript. Finally, all the authors read and approved the final version of the manuscript.
Acknowledgements
================
ABG is supported by a doctoral fellowship grant of the Canadian Institute of Health Research (CIHR). This work was supported by a grant from the Canadian Institute of Health Research (CIHR)-New Emerging Teams Programs (NET) (\#OHN-63276).
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#sec1-2045894019883609}
============
Idiopathic pulmonary arterial hypertension (PAH) is considered as a rare, progressive and life-threatening disease with poor prognosis. Due to luminal obstruction of the distal pulmonary vasculature, pulmonary vascular resistance (PVR) and pulmonary arterial pressure are elevated, and finally leads to right ventricular (RV) failure and death.^[@bibr1-2045894019883609][@bibr2-2045894019883609]--[@bibr3-2045894019883609]^ As described in left heart failure, RV adverse remodeling response to the increased afterload may also destroy its mechanical synchronicity. There are differences in the timing of contraction between the different myocardial segments.^[@bibr4-2045894019883609]^ Recent study observed that the delayed contraction of basal and mid-RV free wall may be the main determinant of intraventricular dyssynchrony in patients with pulmonary hypertension.^[@bibr5-2045894019883609],[@bibr6-2045894019883609]^ RV dyssynchrony even can be observed in patients with narrow QRS, and is strongly correlates with the extent of RV function impairment. Patients with higher RV dyssynchrony showed a more advanced WHO functional class (WHO-FC) and worse exercise tolerance and would predict clinical worsening in patients with PAH.^[@bibr7-2045894019883609][@bibr8-2045894019883609]--[@bibr9-2045894019883609]^ However, the value of RV dyssynchrony to assess the mortality of idiopathic PAH has not been well defined. Therefore, we performed a study with a larger group of patients with idiopathic PAH, and extended the follow-up time to achieve their clinical outcomes to investigate correlations between RV dyssynchrony and the survival of idiopathic PAH.
Methods {#sec2-2045894019883609}
=======
Study population {#sec3-2045894019883609}
----------------
This study included patients who were diagnosed as idiopathic PAH for the first time in our hospital from January 2010 to December 2014, who had underwent comprehensive clinical evaluation. Idiopathic PAH was established according to the guidelines^[@bibr10-2045894019883609]^ by means of right heart catheterization (RHC). Other specific markers, including WHO-FC, 6-minute walking distance (6MWD) and standard transthoracic echocardiography parameters were sampled at the time of catheterization.^[@bibr11-2045894019883609]^ Patients with wide QRS (QRS ≥120 ms), which may exert an electromechanical delay, were excluded.
Hemodynamics evaluation {#sec4-2045894019883609}
-----------------------
RHC was performed using a standard protocol to measure the hemodynamic parameters, including the cardiac index (CI), pulmonary artery systolic pressure, pulmonary artery diastolic pressure (Pa dias), right ventricular systolic pressure, right ventricular end-diastolic pressure, mean right atrial pressure (MRAP), mixed venous oxygen saturation and pulmonary artery wedge pressure (PAWP). PVR was calculated as MPAP-PAWP, divided by cardiac output.
Standard echocardiographic evaluation {#sec5-2045894019883609}
-------------------------------------
Standard transthoracic echocardiography was obtained with the Philips iE33 system, and images were analyzed offline after the procedure according to the recommendation.^[@bibr12-2045894019883609]^ RV end-diastolic area (RVEDA) and RV end-systolic area (RVESA) were obtained from the apical four-chamber view. RV fractional area change (RVFAC) was defined as (RVEDA−RVESA)/RVEDA × 100%. Tricuspid annular plane systolic excursion (TAPSE) was acquired by M-mode image, and the cursor was placed through the lateral tricuspid annulus, the displacement of which from the end-diastole to the end-systole was measured. Right ventricular end-systolic volume (RVESV) and right ventricular end-diastolic volume (RVEDV) were detected by three-dimensional echocardiology, and right ventricular ejection fraction (RVEF) was calculated as (RVEDV−RVESV)/RVEDV × 100%.
A single experienced sonographer, who was blind of the clinical information of patients, performed image acquisition and analysis.
RV mechanical dyssynchrony analysis {#sec6-2045894019883609}
-----------------------------------
RV mechanical dyssynchrony was analyzed by conventional two-dimensional speckle-tracking echocardiography. RV longitudinal strain was assessed in the apical four-chamber images with a frame rate of 70--80 frames/s. Automated software (CMQ, Q-lab 10.5, Philips) divided the apical four-chamber image into six standard segments (apical, middle and basal of the RV free wall and interventricular septum). Longitudinal strain curves were obtained for six RV segments (shown in [Fig. 1](#fig1-2045894019883609){ref-type="fig"}) and time to peak-systolic strain (T~peak,~ from onset of QRS to the peak-systolic stain) of each segment was measured. As previous report and studies,^[@bibr7-2045894019883609],[@bibr9-2045894019883609]^ time to peak for RV apical segments were more variable, so that we also use the four-segment RV (RV-SD4) model to calculate the synchronicity index. RV-SD4, the standard deviation (SD) of the heart rate--corrected T~peak~ of 4 mid-basal RV segments, was used for qualify RV dyssynchrony. Fig. 1.Speckle-tracking strain imaging using the apical four-chamber view to assess RV dyssynchrony. The segmental (colored lines) strain curves represent the relative (percentage) shortening of the six regions of interest as a function of time (in milliseconds). The yellow dots in the segment strain curved represent the peak systolic strain of the segments. (a) A six-segment model of the right ventricle in a control subject; (b) An example of a patient with idiopathic PAH with RV-SD4 = 35 ms.
Endpoint and follow-up {#sec7-2045894019883609}
----------------------
All patients were followed up by outpatient clinic interview or telephone contact every six months from the date of referring to the hospital. The primary endpoint was all causes of mortality. Survival data were collected in a six-month interval.
Statistical analysis {#sec8-2045894019883609}
--------------------
All the analyses were performed using the Statistical Package for the Social Sciences software (SPSS, version 20.0). Continuous variables were presented as means ± SD. Student's *t*-test and analysis of variance (ANOVA) were applied to compare the mean values of continuous variables, and Mann--Whitney U test for ordered variables. A Chi-squared statistics test was used to assess the differences between proportions. Correlations between RV dyssynchrony and hemodynamic variables were explored using Spearman/Pearson's coefficient analysis. The prognostic value of RV dyssynchrony was tested by univariate and multivariate Cox proportional hazard regression analysis. Kaplan--Meier curves were used to illustrate the timing of endpoints during follow-up. The optimal cut-off value of RV dyssynchrony to predict mortality was determined by receiver operating characteristic (ROC) curve analysis. A *p*-value of ≤0.05 was considered statistically significant.
Results {#sec9-2045894019883609}
=======
Correlations of RV-SD4 with echocardiographic and hemodynamic parameters {#sec10-2045894019883609}
------------------------------------------------------------------------
Eleven patients were excluded because of poor image quality, and eventually 116 patients with idiopathic PAH were included in this study.
The mean age was 32 ± 10 years, and they are mainly female (75.8%). The baseline clinical, echocardiographic and hemodynamic characteristics of the study population are presented in [Table 1](#table1-2045894019883609){ref-type="table"}. Table 1.The baseline clinical, echocardiographic and hemodynamic characteristic of the study population.VariablesValuesAge (years)32 ± 10Female (n, %)88 (75.9)WHO-FC I, II (n, %)51 (44.0) III (n, %)62 (53.4) IV (n, %)3 (2.6)Heart rate (beats/minute)74 ± 12Systolic blood pressure (mmHg)114 ± 15Diastolic blood pressure (mmHg)74 ± 126MWD (m)422 ± 82NT-proBNP (fmol/ml)1306.0 ± 821.4Echocardiographic parameters RVFAC (%)29.1 ± 8.3 TAPSE (mm)16.2 ± 3.5 LVEF (%)65.0 ± 6.8 RVEDV (cm^[@bibr3-2045894019883609]^)125.4 ± 54.8 RVESV (cm^[@bibr3-2045894019883609]^)90.4 ± 46.9 RVEF (%)29.8 ± 9.8 T~peak~ of basal-RVFW (ms)429 ± 70 T~peak~ of mid-RVFW (ms)404 ± 72 T~peak~ of basal-IS (ms)414 ± 75 T~peak~ of mid-IS (ms)396 ± 68 RV-SD4 (ms)31 ± 23Hemodynamic parameters MRAP (mmHg)6 ± 5 MPAP (mmHg)59 ± 18 PVR (wood units)14.9 ± 7.5 CI (l/min/m^2^)2.7 ± 0.9Target therapy Sildenafil (n, %)70 (60.3) Bosentan (n, %)14 (12.1) Ambrisentan (n, %)6 (5.2) Inhaled iloprost (n, %)7 (6.1) Calcium channel blockers (n, %)15 (12.9) None (n, %)4 (3.4)[^2]
Correlations between RV-SD4 with echocardiographic and hemodynamic parameters are detailed in [Table 2](#table2-2045894019883609){ref-type="table"}. RV-SD4 had positive correlations with RVESV (r = 0.562, *p* \< 0.001), RVEDV (r = 0.538, *p* \< 0.001), PVR (r = 0.368, *p* \< 0.001), and negative correlations with TAPSE (r = −0.375, *p* \< 0.001), RVFAC (r = −0.411, *p* \< 0.001), RVEF (r = −0.349, *p* \< 0.001) and CI (r = −0.445, *p* \< 0.001). Table 2.Correlations between RV-SD4 and echocardiographic, hemodynamic parameters.VariablesRV-SD4R*p*-ValueEchocardiographic parameters TAPSE−0.375\<0.001 RVESV0.562\<0.001 RVEDV0.538\<0.001 RVEF−0.349\<0.001 LVEF−0.2300.018 RVFAC−0.411\<0.001Hemodynamic parameters MRAP0.2020.032 RVSP0.1540.099 RVEDP0.2130.022 MPAP0.1710.067 CI−0.445\<0.001 PVR0.368\<0.001[^3]
Clinical characteristics based on RV dyssynchrony {#sec11-2045894019883609}
-------------------------------------------------
All the patients were divided into three tertiles based on the RV-SD4. Characteristics of the patients in the three groups are shown in [Table 3](#table3-2045894019883609){ref-type="table"}. There was a significant gradient in severity of most variables among the three groups. Patients in the upper tertiles of RV dyssynchrony distribution experienced a more impaired WHO-FC, worse hemodynamic status and RV function. RV remodeling in the upper tertiles was also more significant, as RVEDV (165.2 ± 63.1 vs. 108.6 ± 40.8, 105.2 ± 37.0 cm^3^, *p* \< 0.001) and RVESV (124.9 ± 55.7 vs. 78.6 ± 32.5, 70.2 ± 30.3 cm^3^, *p* \< 0.001) were larger than the other groups. The plasma level of NT-proBNP of the intermediate and upper tertiles was higher than the lower tertile groups (1245.0 ± 734.6, 1628.4 ± 939.6 vs.1059.8 ± 691.1 fmol/ml, *p* = 0.008), which means worse RV function in these patients. Table 3.Comparison between patients with idiopathic PAH based on RV dyssynchrony.VariablesLower tertiles (n = 39)Intermediate tertiles (n = 39)Upper tertiles (n = 38)*p*-ValueAge (years)34 ± 1031 ± 1031 ± 110.402WHO-FC0.031 I, II (n, %)23 (59.0)15 (38.5)13 (34.2) III, IV (n,%)16 (41.0)24 (61.5)25 (65.8)6MWD (m)439 ± 77415 ± 82411 ± 870.306NT-proBNP (fmol/ml)1059.8 ± 691.11245.0 ± 734.61628.4 ± 939.60.008Echocardiographic parameters RVEDV (cm^[@bibr3-2045894019883609]^)105.2 ± 37.0108.6 ± 40.8165.2 ± 63.1\<0.001 RVESV (cm^[@bibr3-2045894019883609]^)70.2 ± 30.378.6 ± 32.5124.9 ± 55.7\<0.001 RVEF (%)34.6 ± 9.428.0 ± 8.526.3 ± 9.6\<0.001 RVFAC (%)33.2 ± 9.829.1 ± 7.525.0 ± 4.9\<0.001 TAPSE (mm)17 ± 316 ± 315 ± 30.002 LVEF (%)66.6 ± 6.363.9 ± 6.264.4 ± 7.70.202Hemodynamic parameters MRAP (mmHg)5 ± 46 ± 57 ± 50.191 CI (l/min.m^2^)3.2 ± 1.02.7 ± 0.62.2 ± 0.6\<0.001 MPAP (mmHg)56 ± 1858 ± 1764 ± 170.131 PVR (wood units)11.9 ± 6.114.1 ± 7.418.6 ± 7.5\<0.001[^4]
Univariate and multivariate Cox proportional hazard analysis for parameters to predict mortality {#sec12-2045894019883609}
------------------------------------------------------------------------------------------------
The mean duration of follow-up was 41 ± 15 months. During this period, eight patients (7%) were lost, and 93% patients were followed up for at least three years or until death. There were 19 deaths (all causes) and most were died of right heart failure. The baseline clinical and hemodynamic characteristics of the deceased patients are described in [Table 4](#table4-2045894019883609){ref-type="table"}. Among these patients, 18 were WHO-FC III at enrollment. Seventeen patients received PAH-specific drugs combined with conventional therapy as soon as the PAH diagnoses were confirmed. However, the other two patients only received conventional supportive therapy including diuretics, oxygen and digoxin because of financial issues. Compared with survivors, deceased patients had decreased 6MWD (386 ± 78 vs. 429 ± 82 m, *p* = 0.045), higher NT-proBNP (1728.0 ± 1148.6 vs. 1221.7 ± 717.8 fmol/ml, *p* = 0.014) and worse hemodynamics. Table 4.Baseline demographics, and clinical and hemodynamic characteristic of the survival and deceased patients.VariablesSurvivors (n = 97)Non-survivors (n = 19)*p*-ValueAge (years)32 ± 1031 ± 130.806Female (n, %)78 (80.4)10 (52.6)\<0.001WHO-FC\<0.001 I/II (n, %)47 (48.5)4 (21.1) III/ IV (n, %)50 (51.5)15 (78.9)Heart rate (beats/minute)74 ± 1279 ± 110.035Systolic blood pressure (mmHg)114 ± 14111 ± 190.469Diastolic blood pressure (mmHg)74 ± 1275 ± 130.6556MWD (m)429 ± 82386 ± 780.045NT-proBNP (fmol/ml)1221.7 ± 717.81728.0 ± 1148.60.014Echocardiographic parameters TAPSE (mm)16.5 ± 3.514.6 ± 2.80.031 RVEDV (cm^[@bibr3-2045894019883609]^)115.0 ± 47.5175.2 ± 61.1\<0.001 RVESV (cm^[@bibr3-2045894019883609]^)81.8 ± 41.8131.4 ± 49.3\<0.001 RVEF (%)30.7 ± 10.125.3 ± 6.10.027 LVEF (%)65.4 ± 6.962.9 ± 5.90.182 RV-SD4 (ms)25.9 ± 20.554.2 ± 22.9\<0.001Hemodynamic parameters MRAP (mmHg)5.0 ± 4.59.2 ± 4.2\<0.001 MPAP (mmHg)57.9 ± 17.466.3 ± 17.30.058 PVR (wood units)13.8 ± 6.820.2 ± 9.10.001 CI (l/min/m^2^)2.8 ± 0.92.2 ± 0.60.002Target therapy (yes)95 (97.9)17 (89.4)0.064 Sildenafil (n, %)62 (63.9)8 (42.1) Bosentan (n, %)9 (9.3)5 (26.3) Ambrisentan (n, %)6 (6.2)0 (0) Inhaled iloprost (n, %)3 (3.1)4 (21.1) Calcium channel blockers (n, %)15 (15.4)0 (0) None (n, %)2 (2.1)2 (10.5)[^5]
Univariate Cox regression analysis determined that TAPSE, RVFAC, RVEDV, RVEF, MRAP, CI, PVR and RV-SD4 were predictive factors for death. In multivariate Cox proportional hazard analysis, the results showed that RV-SD4 still had the ability to independently predict the mortality after adjusted by other factors (HR = 1.425, *p* \< 0.001) ([Table 5](#table5-2045894019883609){ref-type="table"}). Table 5.The results of univariate and multivariate Cox regression analysis for 116 patients with idiopathic PAH.VariablesUnivariate Cox regressionMultivariate Cox regressionHazard ratio (95% CI)*p*-ValueHazard ratio (95% CI)*p*-ValueEchocardiographic parameters TAPSE (mm)0.852 (0.740--0.982)0.0261.066 (0.857--1.328)0.565 RVFAC (%)0.932 (0.872--0.996)0.0370.994 (0.905--1.091)0.893 RVEDV (cm^[@bibr3-2045894019883609]^)1.013 (1.006--1.021)0.0011.005 (0.994--1.016)0.375 RVEF (%)0.946 (0.897--0.998)0.0431.025 (0.954--1.102)0.497 RV-SD4 (ms)1.473 (1.237--1.755)\<0.0011.425 (1.185--1.714)\<0.001Hemodynamic parameters MRAP (mmHg)1.115 (1.033--1.203)0.0051.166 (1.058--1.286)0.002 MPAP (mmHg)1.017 (0.996--1.039)0.111---- CI (l/min.m^2^)0.208 (0.077--0.567)0.0021.068 (0.326--3.502)0.913 PVR (wood units)1.075 (1.031--1.122)0.0011.005 (0.994--1.016)0.078[^6]
ROC curve analysis was performed to evaluate the utility of RV-SD4 as predictor for mortality. According to the ROC curve analysis, the cut-off value of RV-SD4 was 37.6 ms (AUC = 0.812, *p* \< 0.001), with a sensitivity of 84.2% and a specificity of 80.4%. The Kaplan--Meier survival analysis showed that patients with RV-SD4 \> 37.6 ms had worse prognosis (Log-rank, *p* \< 0.001; [Fig. 2](#fig2-2045894019883609){ref-type="fig"}). Fig. 2.Kaplan--Meier survival curve for the idiopathic PAH patients with RV-SD4 \> 37.6 ms or RV-SD4 \< 37.6 ms.
Reproducibility of RV-SD4 {#sec13-2045894019883609}
-------------------------
RV-SD4 was repeatedly measured in 20 randomly selected patients, and the interobserver and intraobserver variabilities are shown in [Fig. 3](#fig3-2045894019883609){ref-type="fig"}, which was assessed by the Bland--Altman method. As a result, the reasonable reproducibility of intraventricular dyssynchrony evaluated by speckle-tracking echocardiography in this study was demonstrated. Fig. 3.Interobserver (a) and intraobserver (b) reproducibility Bland--Altman plots for the right ventricular dyssynchrony (RV-SD4).
Discussion {#sec14-2045894019883609}
==========
Idiopathic PAH is characterized by pulmonary vascular remodeling, which leads to elevated pulmonary artery pressure and RV failure. RV remodeling responses to chronic pulmonary hypertension is a major cause of RV failure,^[@bibr1-2045894019883609],[@bibr2-2045894019883609],[@bibr13-2045894019883609][@bibr14-2045894019883609]--[@bibr15-2045894019883609]^ which often includes the adaptive stage and maladaptive stage. The former is associated with more concentric remodeling and preserved systolic function, whereas the latter is characterized by more eccentric hypertrophy and worse systolic function.^[@bibr16-2045894019883609],[@bibr17-2045894019883609]^ Extensive investigations concerning left ventricular dyssynchrony suggest that exploration synchronicity of the RV may be a productive approach to understand RV dysfunction in PAH. Previous studies^[@bibr18-2045894019883609][@bibr19-2045894019883609]--[@bibr20-2045894019883609]^ suggest that RV dyssynchrony is a marker of maladaptive remodeling and more severe dysfunction. In our study, we also found patients with greater RV dilatation had more significant RV dyssynchrony and poor pump function.
In contrast to the well-studied pathophysiology of LV dyssynchrony, the underlying mechanisms of dyssynchrony in the right ventricle remain largely unexplored. Recently, Gabrielli et al.^[@bibr21-2045894019883609]^ showed that in patients with PAH iloprost inhalation induced an acute reduction in RV peak systolic strain dyssynchrony index together with a better RV performance. Iloprost inhalation has been widely used in patients with PAH and is proved to be able to improve their hemodynamics. In this study, patients with QRS ≥120 ms were excluded, so an acute RV afterload reduction with inhaled iloprost may explain these changes in RV synchronicity. In concordance with recent study, we also showed patients with higher pulmonary arterial pressure and resistance demonstrated significant RV dyssynchrony. The differential effect of elevated pressure overload on RV regional contractility might be attributed to the uneven distribution of mechanical properties across each RV segment caused by the complex structure and the distortion imposed by RV remodeling.
Outcome prediction in patients with idiopathic PAH has been extensively studied. One consistent finding among studies is that survival in idiopathic PAH is closely related to the status of RV rather than the increased pressure overload.^[@bibr22-2045894019883609],[@bibr23-2045894019883609]^ Studies about hemodynamics had demonstrated the predictive value of MRAP and CI;^[@bibr24-2045894019883609][@bibr25-2045894019883609]--[@bibr26-2045894019883609]^ magnetic resonance imaging studies had proved the value of stroke volume and RVEF.^[@bibr23-2045894019883609],[@bibr27-2045894019883609],[@bibr28-2045894019883609]^ Echocardiographic studies found TAPSE and RVFAC had the predictive value in PAH.^[@bibr29-2045894019883609],[@bibr30-2045894019883609]^ Because hemodynamics assessed by RHC is invasive and magnetic resonance imaging is more expensive, echocardiography is the most common and convenient method used to evaluate RV function. RV dyssynchrony had been described by speckle-tracking echocardiography. Researches by Murata et al.^[@bibr31-2045894019883609]^ and Badagliacca et al.^[@bibr9-2045894019883609]^ had also discovered the prognosis value of RV dyssynchrony. However, the prior study included a heterogeneous population of PH patients in terms of etiology, which are different in the natural progress and prognosis and may affect the results. In addition, in the two studies, the endpoint was complicated. For example, hospitalization and worsening in exercise capacity were regarded as primary outcome. As we know, we reported for the first time the association between RV dyssynchrony and the mortality of idiopathic PAH. Our study emphasized that the RV dyssynchrony is a predictor of survival in patients with idiopathic PAH, and after adjusting by MRAP, CI, TAPSE and RVFAC, RV dyssynchrony could independently predict survival.
This study had several limitations. Firstly, the main limitation is the two-dimensional approach. We assessed RV synchronicity only in the apical four-chamber view, and RV free wall is not seen in its full extent. However, the apical four-chamber view mainly encompasses the lateral portion of the free wall and interventricular septum, which account for a significant part of RV function through its longitudinal shortening according to the earlier study.^[@bibr32-2045894019883609],[@bibr33-2045894019883609]^ Therefore, although far from conclusive, this approach might provide a relatively accurate model to evaluate RV synchronicity. Secondly, this was an observational study and was not able to provide information about the possible pathophysiologic mechanisms that underlie dyssynchrony in the right ventricle. Different from left heart failure, dilated right ventricle could also present with dyssynchrony in the absence of right bundle branch block. As in previous researches, RV dyssynchrony partners with adverse RV remodeling and dysfunction were seen in idiopathic PAH, but it is not clear whether RV dyssynchrony is the cause of RV dysfunction or just one of its marker. The mechanisms of RV dyssynchrony and its clinical use wait testing in the forge of additional research. At last, this study involved patients from a particular center. Therefore, we cannot exclude the possibility of biased estimates for the prevalence and degree of RV dyssynchrony in idiopathic PAH. Future studies with larger patient populations from multiple centers are necessary to verify the results.
In conclusion, RV dyssynchrony analyzed by speckle-tracking echocardiography provided added value to RHC and standard echocardiography in evaluating the survival of patients with idiopathic PAH.
Authors' contribution {#sec15-2045894019883609}
=====================
X-LC, B-YL: Study design, statistical analysis and manuscript editing; W-CW, HW: Echocardiographic image acquisition and analysis; WL, LH, TY: Literature research and follow-up of the patients; Z-HL: Manuscript revision; J-GH, C-MX: In charge of the entire study and manuscript final version approval.
Conflict of interest {#sec16-2045894019883609}
====================
The author(s) declare that there is no conflict of interest.
Funding {#sec17-2045894019883609}
=======
This study was supported by the Capital Health Development and Scientific Research Projects, China (project number: 2016-2-4036); CAMS Initiative for Innovation Medicine (CAMS-I2M), China (project number: 2016-I2M-3-006) and Biomedicine and Life Sciences Innovation Cultivation Research Project, China (Z161100000116052).
Ethical approval {#sec18-2045894019883609}
================
This study complied with the Declaration of Helsinki and was approved by the Institutional Review Board of Fuwai Hospital (ethical approval no. 2018-1063).
[^1]: These authors contributed equally to this work as co-first authors.
[^2]: Note: Data are presented as n (%) and mean ± SD. WHO-FC: World Health Organization functional class; 6MWD: 6-minute walking distance; LVEF: left ventricular ejection fraction; RVFAC: right ventricular fractional area change; TAPSE: tricuspid annular plane systolic excursion; RVEDV: right ventricular end-diastolic volume; RVESV: right ventricular end-systolic volume; RVEF: right ventricular ejection fraction; T~peak~: time to peak-systolic strain; RVFW: right ventricular free wall; IS: interventricular septum; MRAP: mean right atrial pressure; MPAP: mean pulmonary arterial pressure; PVR: pulmonary vascular resistance; CI: cardiac index.
[^3]: TAPSE: tricuspid annular plane systolic excursion; RVEDV: right ventricular end-diastolic volume; RVESV: right ventricular end-systolic volume; RVEF: right ventricular ejection fraction; LVEF: left ventricular ejection fraction; RVFAC: right ventricular fractional area change; MRAP: mean right atrial pressure; MPAP: mean pulmonary arterial pressure; PVR: pulmonary vascular resistance; CI: cardiac index; RVSP: right ventricular systolic pressure; RVEDP: right ventricular end-diastolic pressure.
[^4]: Note: Data are presented as n (%) and mean ± SD. WHO-FC: World Health Organization functional class; 6MWD: 6-minute walking distance; TAPSE: tricuspid annular plane systolic excursion; LVEF: left ventricular ejection fraction; RVFAC: right ventricular fractional area change; RVEDV: right ventricular end-diastolic volume; RVESV: right ventricular end-systolic volume; RVEF: right ventricular ejection fraction; MRAP: mean right atrial pressure; MPAP: mean pulmonary arterial pressure; PVR: pulmonary vascular resistance; CI: cardiac index.
[^5]: Note: Data are presented as n (%) and mean ± SD. WHO-FC: World Health Organization functional class; 6MWD: 6-minute walking distance; LVEF: left ventricular ejection fraction; RVFAC: right ventricular fractional area change; TAPSE: tricuspid annular plane systolic excursion; RVEDV: right ventricular end-diastolic volume; RVESV: right ventricular end-systolic volume; RVEF: right ventricular ejection fraction; MRAP: mean right atrial pressure; MPAP: mean pulmonary arterial pressure; PVR: pulmonary vascular resistance; CI: cardiac index.
[^6]: WHO-FC: World Health Organization functional class; 6MWD: 6-minute walking distance; MRAP: mean right atrial pressure; MPAP: mean pulmonary arterial pressure; PVR: pulmonary vascular resistance; CI: cardiac index. RVFAC: right ventricular fractional area change; TAPSE: tricuspid annular plane systolic excursion; RVESV: right ventricular end-systolic volume.
| {
"pile_set_name": "PubMed Central"
} |
This article is part of the Thematic Series \"Automated synthesis\".
Introduction
============
Enabling synthesis technologies such as flow chemistry are becoming commonplace in modern laboratories (for recent reviews of flow chemistry in synthesis see \[[@R1]\] and \[[@R2]\]). As more groups start to use this technology, there is an increasing demand to expand the capabilities of laboratory apparatus, in particular for the seamless integration of different types of apparatus from different manufacturers so that they can be used simultaneously and synergistically. Although manufacturers generally provide appropriate control software for use with their particular device ecosystem, these frequently have a limited scope and do not always integrate well with other equipment. For commercial reasons, the control software is rarely provided in a format that can be readily extended by the user to implement control of additional hardware.
Whilst the control of multiple devices and instruments is well-developed and standard practice for multi-step continuous processing on a large scale, these control systems tend to be custom built at a high cost and hence they are not usually appropriate for the research environment. However, even on a laboratory scale the ability to connect and share data between different devices is of critical importance for performing complex processes: recently reported examples include matching downstream flow rates to the concentration profile of a dispersed reaction slug \[[@R3]\], the linking of synthesis to purification apparatus \[[@R4]\] and the development of new work-up technologies \[[@R5]\]. Most manufacturers are willing to share control commands for their products, making these applications possible, but extensive work is often required to coordinate a new process into a smooth and simple operation. This may take the form of a significant programming effort to automate the apparatus, or involve the complex manual timing of events to ensure that the desired operational sequence takes place.
Ideally, we need the ability to create a control algorithm for any new process with minimal up-front effort. Furthermore, control software should be sufficiently flexible such that different items of apparatus can be swapped in and out of the integrated system without having to make significant changes to the automation protocol. Following a review of typical software packages and technologies, we chose to implement a framework for running scripted control algorithms that would allow interfaces for new instruments to be prototyped easily. We hoped that by defining a specification for each class of device (i.e., sensors, pumps, heaters, etc.), these unified interfaces would provide the desired flexibility between devices. In this way, more time can be spent on the engineering and chemistry challenges inherent in the synthesis process, rather than the logistics of control system interfaces.
Our group has experience using the Python programming language \[[@R6]\] to control laboratory devices \[[@R5]\]. This language claims to be ideal for rapid development and the use of free software fosters collaboration \[[@R7]\], enabling technology to be transferred without the large initial set-up costs typically involved with commercial control packages. Established technologies such as chemical intelligence \[[@R8]\], statistical analysis \[[@R9]\] and computer vision \[[@R10]--[@R11]\] are available as third-party libraries for easy integration. The simple text-based control scripts can be copied and pasted for simple re-use, and are compatible with version control systems \[[@R12]\]. Finally, these control programs tend to require low computational resources and will run on cheap, low-power computers such as the Raspberry Pi^®^ ([Fig. 1](#F1){ref-type="fig"}) \[[@R13]\].
{#F1}
In the work reported here, we describe the application of automation to performing routine research tasks such as the optimisation of experimental parameters for a particular transformation and describe how the application of remote monitoring can improve safety and efficiency within the research environment.
In order to test and demonstrate the development of simple inexpensive hardware and software solutions for the facile integration of laboratory hardware, we chose the goal of the efficient synthesis of small molecules which are essential for the generation of fragment-based libraries for medicinal chemistry research programmes. "3D Fragments" have become very attractive recently due to their potential to expand the available chemical space. The presence of nitrogen in a small 3D structure can be important for its biological activity \[[@R14]\]; this is particularly important when developing unnatural amino acid derivatives \[[@R15]--[@R16]\]. Such compounds represent an important contribution to a fragment database for medicinal chemistry. For example, piperazine-2-carboxamide (**1**, [Fig. 2](#F2){ref-type="fig"}) is an amino acid derivative with interesting biological properties \[[@R17]\]. At the time of writing, racemic **1** was identified as a notably expensive building block \[[@R18]\] and thus a good target for this transformation. We have explored the possibility of developing a machine-automated synthesis of this compound which might later be extended to analogous structures.
{#F2}
For the facile machine-assisted synthesis of **1** we devised and optimised the fully continuous sequential hydration of nitrile **3** to amide **2** and hydrogenation of pyrazine **2** to piperazine **1** ([Fig. 2](#F2){ref-type="fig"}). Both of these steps involve flowing through heterogeneous catalysts, a metal oxide for the hydration of the nitrile and a supported precious metal for the hydrogenation of the heteroaromatic ring. Furthermore, both steps involve the addition of a volatile small molecule to the substrate with no byproducts: the addition of water for the nitrile hydration, and the addition of hydrogen to the heteroarene for the reduction. Thus, this sequence is ideal for a fully continuous multi-step process.
Results and Discussion
======================
Nitrile hydration
-----------------
Primary amides can be prepared via a number of different approaches but the most environmentally friendly procedure is the hydration of nitriles \[[@R19]\]. Although this is generally considered to be a simple transformation, there are some inherent problems with the standard techniques of hydrolysis \[[@R20]\]. Further to our previous work which demonstrated the hydration of broad classes of nitriles by passing aqueous--organic solutions through a packed bed of manganese dioxide \[[@R21]\], we have found that heteroaromatic nitriles possessing a β-heteroatom can also be hydrolysed using hydrous zirconia \[[@R22]--[@R23]\] in a similar fashion ([Fig. 3](#F3){ref-type="fig"}). The directed activity of zirconia is very similar to that of ceria, a known nitrile hydration catalyst for batch reactions \[[@R24]\]. While zirconia has a lower turnover number than ceria, zirconia is less expensive than ceria and it has better physical properties for packed beds than ceria. Its use in packed beds facilitates recycling of the catalyst, providing a further cost advantage.
{#F3}
Interestingly, we have now found that no activation of the zirconia is needed (this is the same as for nitrile hydrations with manganese dioxide, but unlike the use of zirconia as a heterogeneous catalyst for Meerwein--Ponndorf--Verley reductions \[[@R22]\] and Oppenauer oxidations \[[@R23]\] where zirconia activation is required). In fact, it seems that the extent of hydration of pyrazine-2-carbonitrile is proportional to the initial water content of the zirconium catalyst. To confirm this hypothesis, we ran a control experiment in which no additional water was added to the organic solvent (absolute ethanol). A solution of **3** was continuously fed into a reactor column containing untreated zirconium hydroxide \[[@R25]\]. The reaction profile was followed with an in-line infrared (IR) spectrometer in order to determine the conversion. Notably, the hydrolysis process was constant for 3 hours and then the catalytic properties of the system disappeared ([Fig. 4](#F4){ref-type="fig"}), suggesting that the surface of the catalyst had been completely dehydrated by the nitrile. As soon as water was added to the reagent solution, quantitative hydration of the nitrile was again achieved for an extended period of time without any further drop in the catalytic activity.
{#F4}
In order to determine the optimal parameters for this reaction, a number of experiments were carried out. The reactor was configured as shown in [Fig. 3](#F3){ref-type="fig"}: using a Vapourtec R2+/R4 reactor unit, solutions of nitrile **3** were passed through heated column reactor **R1** (Omnifit^®^ glass column, 100 mm × 6.6 mm; a flow rate of 0.1 mL min^−1^ produced a residence time of 20 minutes) packed with 2.5 g of zirconium hydroxide. (Due to the practicalities involved with performing reactions using different solvents, these tests were run under manual control).
Quantitative transformation to the primary amide **2** was generally achieved within a 20 minute residence time at 100 °C ([Table 1](#T1){ref-type="table"}). In general, we observed good solvent compatibility; however, when using a water-immiscible solvent such as toluene or ethyl acetate the zirconium hydroxide functioned as a reagent rather than a catalyst. In this case, the reactivity of the metal oxide structure could be regenerated by feeding the reactor with an aqueous solution. As mentioned earlier, it seems to be the water present in the lattice, or fed into the column reactor, which is responsible for the hydration activity of the metal oxide. Consequently, the use of water-miscible solvents is preferred in order to have a continuous process instead of a plug flow protocol using water-immiscible solvents.
######
Optimisation for the hydration of the pyrazine-2-carbonitrile. Reactions were performed with dry solvents and a fresh batch of catalyst for each. The reactions were carried out on a 1 mmol plug of pyrazinecarbonitrile such that the zirconia was present in excess.
----------------- -------------------- --------------------------- ----------
Solvent^a^ Temperature \[°C\] Residence time^b^ \[min\] Yield^c^
isopropanol 100 10 49%
isopropanol 100 20 98%
ethanol 100 10 51%
ethanol 100 20 100%
methanol 100 20 100%
ethyl acetate 100 20 45%
toluene 100 20 92%
water 100 20 88%
tetrahydrofuran 100 20 95%
dioxane 100 20 94%
----------------- -------------------- --------------------------- ----------
^a^No water was added to the organic solution; ^b^residence time within the column reactor; ^c^the yield refers to that of the isolated product.
Under the optimised conditions, a solution of nitrile **3** in ethanol/H~2~O (0.6 M, 8:1 v/v) was passed through the column reactor **R1** heated at 100 °C, with a residence time of 20 minutes, to obtain a quantitative yield of the primary amide **2** after concentration of the reactor output.
To assist with the processing of a large amount of material, we applied an automated control and monitoring system being developed within our group \[[@R26]\], which can carry out a programmed sequence of operations written using the Python™ language. There is also a remote interface for observing the status of an ongoing reaction in real-time. In common with the industrial use of process analytical technologies (PAT), a number of parameters are read from devices or sensors, and the effluent is only collected when all of the parameters are stable, ensuring a high degree of purity in the collected material.
Larger zirconia columns were also used to improve the throughput. Employing 5 g of zirconium hydroxide within a 100 mm × 10 mm diameter Omnifit^®^ column (**R2**, [Fig. 5](#F5){ref-type="fig"}) gave a two-fold increase in throughput. Under these conditions, we could use a flow rate of 0.2 mL min^−1^ and still generate a quantitative hydration of the nitrile, generating an output equating to 0.45 g h^−1^ of amide product.
{#F5}
Using a Mettler--Toledo FlowIR™ fitted with a detector with a silicon window (required for visualising the nitrile region of the spectrum, which is blocked by the standard diamond window) the change from nitrile to amide can be observed, providing real-time feedback on the state of the reaction.
The monitoring software was connected to the Vapourtec R2+/R4 reactor via RS-232, and to the FlowIR™ spectrometer via the Auto-Export feature of the Mettler--Toledo iC IR control software running on a separate computer. The desired steady-state parameters were defined, and an output valve (**V1**) was controlled based on their states. Importantly, different responses were defined for each parameter: the IR absorbance represents the current reactor output and thus can command immediate responses from the valve. On the other hand, fluctuations in temperature or pressure could compromise an entire column volume, so a delay was added between the time that these parameters stabilised and the time that collection resumed ([Fig. 6](#F6){ref-type="fig"}). An alerting system was also included that could notify the operator if the reactor lost pressure, indicating an air bubble or a leak, or if it over-pressured -- situations that currently require manual intervention to rectify.
{ref-type="supplementary-material"} for the control program).](Beilstein_J_Org_Chem-10-641-g007){#F6}
All of the reaction parameters could be observed on remote computers, or wirelessly on a tablet computer when moving around the lab. The monitoring software generates an interface for each running experiment, which can be accessed through a web browser. The ability to access real-time experimental information from anywhere -- as opposed to only on computers situated next to the apparatus -- is very important, because it gives the chemist freedom to perform other tasks at the same time. This is particularly beneficial when data from multiple reactors and devices is combined into a single interface.
The sensors were interrogated approximately once a second; a Raspberry Pi^®^ microcomputer (as shown in [Fig. 5](#F5){ref-type="fig"}) has more than sufficient processing power to perform the required data collection, interpretation and control. We anticipate that much more complex systems than this one could be controlled using this miniature computer system.
An example of the error handling behaviour is shown in [Fig. 7](#F7){ref-type="fig"}. After approximately 40 minutes the product begins to elute and after it passes a threshold in the absorbance as detected by the FlowIR™ unit, valve **V1** is switched to collect the output. After approximately 1.5 h, a loss in pressure is detected corresponding to an air bubble in the input stream. Valve **V1** is switched to waste, and an SMS notification is sent to the operator, who re-primes the pump. After the pressure has returned to normal there is a delay calculated to be the dead volume of the column before the output is collected again.
{#F7}
Pleasingly, experiments on scale (50 mmol) gave very positive results. The catalystic activity of the system remained constant and the product **2** was recovered quantitatively and, more importantly, characterised by high purity as determined by elemental analysis (\>98%).
Pyrazine ring reduction
-----------------------
An initial investigation showed that this aromatic carboxamide could be efficiently reduced using an H-Cube^®^ reactor \[[@R27]--[@R28]\]. There has been recent interest in the use of automation to optimise reaction conditions \[[@R29]--[@R31]\], and thus we hoped to use a linear programming method \[[@R32]--[@R33]\] in a similar way to iteratively improve the hydrogenation settings. This turned out to be impractical for two reasons: the flow rate, dead volume, and stabilisation time required for the H-Cube to reach steady state meant that each iteration would take 30--60 minutes; and the discrete nature of the available parameters (for example, the column temperature can only be set in units of 10 °C) mean that simple linear optimisation methods are not suitable \[[@R34]\]. Therefore, we decided to use a Design of Experiments \[[@R35]--[@R38]\] method to determine which of the available parameters were important for the conversion and selectivity of this reaction. This process requires a lot of repetitive work to be done, and thus can greatly benefit from reaction automation.
A two-level factorial design with three parameters (temperature: 40 °C and 100 °C; H~2~ pressure: 20 bar and full hydrogen mode; and flow rate: 0.1 mL min^−1^ and 0.2 mL min^−1^) suggested 16 experiments -- two repeats each of eight sets of conditions. A single catalyst was used, a 10% Pd/C cartridge supplied by ThalesNano. With a large amount of material from the previous step in hand, we decided to use an automated system to perform these experiments in order to reduce the amount of operator's time that is required.
Combining the control of a Knauer HPLC pump, the H-Cube^®^ reactor and a multi-position valve (**V2**) ([Fig. 8](#F8){ref-type="fig"}, [Fig. 9](#F9){ref-type="fig"}) we could perform up to nine reactions in a row. A sample was taken at steady state for each set of conditions: we noticed two major reduction products, fully-reduced pyrazine-2-carboxamide (**1**) and 1,4,5,6-tetrahydropyrazine-2-carboxamide (**4**). Unfortunately, the carbonyl stretching frequencies of the different amides proved to be remarkably similar and the secondary frequencies had very low intensities, so the FlowIR™ was not suitable for analysis of this reaction mixture and these were subsequently analysed by NMR to quantify the results.
{#F8}
{ref-type="supplementary-material"} for control program).](Beilstein_J_Org_Chem-10-641-g010){#F9}
For each sample, four parameters were calculated: the degree of conversion, based on the residual **2** observed; the amount of the desired product **1** formed; and an estimated amount of undesired compounds, based on the integration of peaks visible in the ^1^H NMR spectra (a triplet at 3.2 ppm and a doublet at 3.8 ppm, the first corresponding to **4** and the other to a second unidentified compound) relative to the integration of **1** and **2** ([Fig. 10](#F10){ref-type="fig"}). As expected, the hydrogen pressure had the most significant effect on the conversion. The temperature and the flow rate had a lower effect on the conversion, although the combination of higher temperatures and lower flow rates gave a higher purity output as intermediates such as **4** are fully reduced. For maximum efficiency further hydrogenation procedures were carried out at the lower flow rate 0.1 mL min^−1^ and higher temperature 100 °C.
{#F10}
In situ generation of the intermediate
--------------------------------------
Performing an optimisation experiment in continuous flow -- such as described in the previous section -- has a notable disadvantage. After making a change to the reaction conditions there is some delay before the system settles to a steady state, at which point a new measurement can be taken. This is particularly relevant for reactions that take some time, as large quantities of material may be required to perform a number of different trials. In the previous case, we had a significant amount of the intermediate so this was not considered to be a problem, but we would like to avoid stockpiling of intermediates in this way. One major advantage of continuous processing is that material can be used directly from one step to another so that collecting large quantities of intermediates can be avoided.
Consequently, we started to investigate whether the two steps could be combined so that there is always enough feedstock to run the second step. Normally, this would dictate that the flow rate of the second step is always the same as that of the first step (or greater, using an additional pump). However, in some processes, such as the one described in the previous section, we would like to be able to vary the flow rate of a step to adjust the residence time. We envisaged the use of a reservoir to keep a small amount of intermediate ready for the second step. This required some means to measure the volume present in the reservoir so that the control software can decide whether there is enough material ready to perform an experiment.
In a recent review \[[@R11]\], we described the application of cameras and computer vision to synthesis procedures. Building on work in which a camera and a float were employed to measure the position of a biphasic (aqueous/organic) mixture within a settling column \[[@R5]\], we manufactured a float containing an air bubble \[[@R39]\] which would float on less dense solvents than the solid polyethylene version used previously. Using a pear-shaped flask as a reservoir, so that the inlet needle for the second pump could access as much of the solution as possible, a camera/float combination was able to measure the amount of intermediate in the reservoir ([Fig. 11](#F11){ref-type="fig"}). A digital camera was positioned so that it was observing the collection vessel. Regular snapshots were taken and analysed using computer vision software to locate the green float within the image. This information was used to estimate the height of the float in the reservoir, which enabled the control protocol to make decisions based on the amount of intermediate available. The Raspberry Pi^®^ computer was found to struggle when working with a USB webcam so the control protocol was instead run on a standard desktop computer (A dedicated camera module for the Raspberry Pi^®^ has recently been released which should solve this problem as this module requires minimal processor time to capture images).
{#F11}
By combining the control sequences for the two synthesis steps with the volume-measuring logic, the second step could be started when enough material of the intermediate had been collected to start pumping out. Furthermore, if the meniscus were to rise high enough to pose a risk of the reservoir overflowing, this can trigger an alert or cause valve **V1** to cease collecting the intermediate. If the liquid level were to fall too low, then the next iteration of the hydrogenation condition testing loop could pause until there was enough material to continue (see [Supporting Information File 3](#SD3){ref-type="supplementary-material"} for the control sequence, and [Supporting Information File 4](#SD4){ref-type="supplementary-material"} for the sequence diagram).
Two-step synthesis procedure
----------------------------
Finally, the previous procedure was modified to perform a two-step process using the optimised parameters for each step ([Fig. 12](#F12){ref-type="fig"}, [Fig. 13](#F13){ref-type="fig"}). The maximum flow rate to allow full conversion in the H-Cube^®^ was relatively low, which limited the throughput of the material in the reduction step. This could be mitigated if larger-scale hydrogenation apparatuses were available; but in this case the use of a reservoir meant that the flow rates did not necessarily have to be matched, resulting in a semi-continuous process.
{#F12}
{#F13}
Using the two different flow rates and having the hydrolysis step stop automatically when the reservoir filled up led to a total running time of about 10 hours ([Fig. 14](#F14){ref-type="fig"}). However, we can imagine that this could represent one cycle of a prolonged sequence where the first step is periodically stopped to allow time for the collected intermediate to be processed in the second step. With a large reservoir (for example, our 50 mL reservoir allows up to a 16 hour start/stop cycle) the proportion of material wasted during start up and shutdown of the first step can be reduced.
{#F14}
We anticipate that the potential to rapidly realise complex control sequences such as these will further broaden the scope for the use of flow chemistry reactors in both the research and scale-up environments. We hope that the increasing availability of free and open software to enable such processes will help to democratise the field of flow chemistry when applied as an advanced enabling technology in the laboratory.
Conclusion
==========
A machine assisted synthesis of pyrazine-2-carboxamide -- a component of Rifater^®^, used in the treatment of tuberculosis -- and its reduced derivative (*R,S*)-piperazine-2-carboxamide has been demonstrated, using a new open-source software platform for the simultaneous control of multiple devices. The protocol developed here represents a valid example of how these technologies can be used to implement chemistry processes for synthesis. Automated procedures can have a significant impact on productivity, not just for traditional applications such as library synthesis, but also for one-off protocols for which automation may previously have required a much greater time investment.
Supporting Information
======================
######
Experimental data.
######
Control sequence for extended period hydrolysis experiment with monitoring. Alternate web version: http://gist.github.com/richardingham/0a58a291bad2e3b9009f
######
Control sequence for performing DoE experiments. Alternate web version: http://gist.github.com/richardingham/83401127622036c6afd0
######
Control sequence for performing DoE experiments using intermediate from a reservoir. Alternate web version: http://gist.github.com/richardingham/f2117b9dc7504d6e1942
######
Flow chart representation of the control sequence for performing DoE experiments using intermediate from a reservoir.
######
Control sequence for performing two-step hydrogenation process with control and monitoring. Alternate web version: http://gist.github.com/richardingham/31f6f8efa47771c2ed02
We gratefully acknowledge Pfizer Worldwide Research and Development (C.B. and J.M.H.), the Ralph Raphael studentship (R.J.I.) and the BP endowment (S.V.L.) for financial support. We also thank Dr. Emiliana Dvininov (MEL Chemicals, Manchester) for the gift of the zirconia catalyst.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Gastric cancer is one of the most common malignant tumors worldwide with an occurance rate of 10.79% ([@b1-ol-0-0-11049]). Gastric cancer is the fifth leading cause of cancer-associated mortality in both the male and female population worldwide with a mortality rate of 8.8% ([@b2-ol-0-0-11049]). Despite the continuous development of comprehensive diagnosis and treatment technologies in recent years, the 5-year survival rate for patients with advanced gastric cancer is still \>30% ([@b3-ol-0-0-11049]--[@b5-ol-0-0-11049]). The reasons for this are complex, and one of the most important issues is that gastric cancer cells are prone to survival and migration/invasion ([@b6-ol-0-0-11049]--[@b8-ol-0-0-11049]). Therefore, it is worthwhile to explore the mechanism of gastric cancer cell survival and migration/invasion for early intervention, late treatment and improvement of treatment outcomes.
Endoplasmic reticulum (ER) is the primary site of protein folding, modification and assembly, as well as intracellular Ca^2+^ storage in eukaryotic cells ([@b9-ol-0-0-11049],[@b10-ol-0-0-11049]). Under stressed conditions, misfolded or unfolded protein aggregation and imbalances in Ca^2+^ levels in the ER lumen occur, and the cell enters a state termed ER stress ([@b11-ol-0-0-11049]--[@b13-ol-0-0-11049]). If the stress persists or the stress damage exceeds the ability of cell survival and protection, the ER stress-dependent apoptosis pathway is activated, leading to apoptosis ([@b14-ol-0-0-11049]--[@b16-ol-0-0-11049]). Recent studies have suggested that ER stress-mediated cell migration/invasion is closely associated with the occurrence and development of gastric cancer ([@b17-ol-0-0-11049]--[@b20-ol-0-0-11049]). However, the initiator of ER stress that regulates gastric cancer cell survival and migration/invasion remains unknown.
Yes-associated protein (YAP) is involved in the regulation of cell proliferation, organ development and the occurrence of tumors ([@b21-ol-0-0-11049]--[@b23-ol-0-0-11049]). Previous studies have demonstrated that YAP is abnormally expressed in breast, ovarian and other types of cancer, and its expression levels are associated with stage and prognosis of patients with tumors ([@b24-ol-0-0-11049]--[@b27-ol-0-0-11049]). Upregulation of YAP has been observed in gastric cancer and is associated with the clinicopathological characteristics of patients with gastric cancer ([@b28-ol-0-0-11049],[@b29-ol-0-0-11049]). In addition, YAP integrates ER stress to control liver size and tumorigenesis, suggesting a potential connection between YAP and ER stress ([@b29-ol-0-0-11049],[@b30-ol-0-0-11049]). Therefore, the present study hypothesized that YAP may reduce gastric cancer cell survival and migration through the activation of ER stress.
Materials and methods
=====================
### Cell culture and treatments
The gastric cancer MKN-28/74 cells and normal gastric GES-1 cells were purchased from the American Type Culture Collection. The MKN28 cell line has been reported as cross-contaminated with MKN74; thus, it is referred to as MKN-28/74 throughout the present study ([@b31-ol-0-0-11049]). MKN-28/74 cells were cultured in RPMI-1640 medium (Nacalai Tesque, Inc.) supplemented with 10% fetal bovine serum (FBS; HyClone; GE Healthcare Life Sciences) at 37°C in a 5% CO~2~ humidified incubator; GES-1 cells were cultured in DMEM (HyClone; GE Healthcare Life Sciences) containing 10% FBS (HyClone; GE Healthcare Life Sciences) at 37°C in a 5% CO~2~ humidified incubator ([@b32-ol-0-0-11049]). Tunicamycin (TM; 100 nM; Sigma-Aldrich; Merck KGaA) and 4-phenylbutyrate (10 mM; Sigma-Aldrich; Merck KGaA), the agonist and antagonist for ER stress, respectively, were added to the medium for 12 h. MKN-28/74 cell were pre-treated with PD98059 (10 µM) for 24 h at 37°C.
### Transfection
To evaluate the functional role of YAP, small interfering (si)RNA was used to knockdown its expression. siYAP (5′-GCGACATTCAGGGUGACUAUU−3′) and non-targeting sequences (siCtrl; 5′-UUCUCCGAACGUGUCACGU-3′) were purchased from GenePharma Co., Ltd. ([@b33-ol-0-0-11049]). A total of 20 nM siYAP or siCtrl was used to transfect MKN-28/74 cells (2×10^6^ cells/well) with Lipofectamine^®^ 2000 (Thermo Fisher Scientific, Inc.) for 48 h in 6-well plates, and the transfection efficiency was determined by western blotting.
### Reverse transcription-quantitative PCR (RT-qPCR)
Total RNA was extracted from the MKN-28/74 cells using an RNeasy kit (Beyotime Institute of Biotechnology) and reverse transcribed using One-step RT-PCR kit (cat. no., AE311-02; Beijing Transgen Biotech Co., Ltd.) at 37°C for 30 min according to the manufacturer\'s protocol ([@b34-ol-0-0-11049]). qPCR was performed using the SYBR Green RT-PCR kit (Takara Bio, Inc.) according to the manufacturer\'s protocol. The thermocycling conditions were as follows: 95°C for 5 min; followed by 40 cycles of 95°C for 40 sec, 60°C for 30 sec and 72°C for 30 sec. GAPDH was selected as an internal control. The following primers were used for PCR: YAP forward, 5′-AAGGCTTGACCCTCGTTT-3′ and reverse, 5′-CTGCTGCTGCTGGTTTGA-3′; and GAPDH forward, 5′-GTCAACGGATTTGGTCGTATTG-3′ and reverse, 5′-CATGGGTGGAATCATATTGGAA-3′. Fold-changes in mRNA expression were calculated using the 2^−ΔΔCq^ method ([@b35-ol-0-0-11049]).
### Western blotting
The MKN28/74 cells (5×10^6^) was homogenized and sonicated in a lysis buffer (Beyotime Institute of Biotechnology). Protein concentrations were detected using a BCA Protein Quantification kit, according to the manufacturer\'s protocol. The proteins (50 µg) were separated by 10% SDS-PAGE and then transferred onto polyvinylidene difluoride membranes. The membrane was blocked with 5% non-fat dry milk for 1 h at room temperature and incubated with specific primary antibodies overnight at 4°C. The primary antibodies used were as follows: YAP (1:1,000; Cell Signaling Technology, Inc.; cat. no. 14074), pro-caspase-3 (1:1,000; Abcam; cat. no. ab13847), cleaved caspase-3 (1:1,000; Abcam; cat. no. ab49822), glucose-regulated protein 78 kDa (GRP78; 1:1,000; Abcam; cat. no. ab21685), GADPH (1:1,000; Abcam; cat. no. ab8245), pro-caspase-12 (1:1,000; Abcam; cat. no. ab8117), cleaved caspase-12 (1:1,000; Cell Signaling Technology, Inc.; cat. no. 2202), C/EBP homologous protein (CHOP; 1:1,000; Abcam; cat. no. ab11419), ERK (1:1,000; Cell Signaling Technology, Inc.; cat. no. 4695), phosphorylated (p-)ERK (1:1,000; Cell Signaling Technology, Inc.; cat. no. 4370). The blots were detected with an enhanced chemiluminescence substrate kit (Thermo Fisher Scientific, Inc.), according to the manufacturer\'s protocol. The bands were scanned and quantified by ImageJ version 1.47 software (National Institutes of Health) ([@b36-ol-0-0-11049]).
### Immunofluorescence staining
Following transfection treatment, the MKN28/74 cell (0.5×10^6^ cells/well) were fixed with 3.7% paraformaldehyde for 10 min at room temperature and subsequently blocked with 5% bovine serum albumin (Sigma-Aldrich; Merck KGaA) in PBS for 1 h at room temperature. Cells were incubated with primary antibodies for 4 h at room temperature. The primary antibodies used were YAP (1:500; Cell Signaling Technology, Inc.; cat. no. 14074) and CHOP (1:500; Abcam; cat. no. ab11419). DAPI (5 mg/ml; Sigma-Aldrich; Merck KGaA) was used to stain the nuclei at room temperature for 3 min. A total of 5 randomly selected fields of view were used per smaple and images were captured with a laser confocal microscope (magnification, ×600; TcS SP5; Leica Microsystems, Inc.).
### Cell invasion and migration
Following transfection treatment, cell invasion was analyzed using a Transwell chamber assay as previously described ([@b37-ol-0-0-11049]). Briefly, cells (1×10^6^ cells/well) were suspended in RPMI-1640 medium containing 10% FBS and seeded into the upper chambers.
Cell migration was analyzed using a wound-healing assay and cells were cultured with RPMI-1640 medium in 12-well plates. Once cells reached \>80% confluency, a sterile pipette tip was used to evenly scratch the 12-well plate. Following cell attachment, a straight line was gently scratched in the cell layer with a 200 µl pipette tip, and the cells were washed with PBS (pH 7.4) three times. The relative wound closure was imaged under a light microscope (magnification, ×100; Leica Microsystems, Inc.) at 0 and 24 h. The wound was measured using ImageJ 1.74v software (National Institutes of Health).
### MTT assay and terminal deoxynucleotidyl transferase-mediated dUTP nick end labeling (TUNEL)
The MKN-28/74 cells were seeded into 96-well plates at 8×10^3^ cells/well and incubated overnight. Following transfection treatment, MTT (5 mg/ml) was added to each well and incubated for 4 h. The insoluble formazan was collected and dissolved in dimethylsulfoxide, and the optical density value was measured with a scanning spectrophotometer at a wavelength of 570 nm.
The TUNEL assay was used for the detection of apoptosis. A one-step TUNEL kit (Beyotime Institute of Biotechnology) was used for TUNEL staining. The MKN-28/74 cells (1×10^6^ cells) were incubated with fluorescein-dUTP (Invitrogen; Thermo Fisher Scientific, Inc.) to stain the apoptotic cell nuclei and with DAPI (5 mg/ml) to stain all cell nuclei at room temperature for 3 min. Images were captured with a laser confocal microscope (magnification, ×600; TcS SP5; Leica Microsystems, Inc.). The number of TUNEL-positive cells was calculated by counting at least five random fields of view as the ratio of the experimental samples to the control samples (untransfected cells).
### Statistical analysis
All analyses were performed with SPSS 20.0 software (IBM Corp.). Experiments were repeated three times and data are presented as the means ± standard error of the mean. Statistical analyses were performed using one-way analysis of variance with the Bonferroni test for post hoc comparisons. P\<0.05 was considered to indicate a statistically significant difference.
Results
=======
### YAP is upregulated in gastric cancer MKN-28/74 cells and promotes cell survival
The expression levels of YAP were detected by western blotting in MKN-28/74 gastric cancer cells and GES-1 normal gastric cells. The results demonstrated that YAP was significantly upregulated in gastric cancer MKN-28/74 cells compared with GES-1 cells ([Fig. 1A and B](#f1-ol-0-0-11049){ref-type="fig"}). To confirm the role of YAP in the progression gastric cancer, siYAP was transfected into MKN-28/74 cells to knockdown the expression of YAP. The transfection efficiency was detected by western blotting ([Fig. 1A and B](#f1-ol-0-0-11049){ref-type="fig"}), RT-qPCR ([Fig. 1C](#f1-ol-0-0-11049){ref-type="fig"}) and immunofluorescence ([Fig. 1D and E](#f1-ol-0-0-11049){ref-type="fig"}). The results demonstrated that siYAP, but not siCtrl, significantly inhibited the expression of YAP in gastric cancer MKN-28/74 cells compared with untransfected cells. The effect of YAP on MKN-28/74 cell viability was investigated. The results of the MTT assay demonstrated that YAP knockdown significantly reduced the viability of MKN-28/74 cells ([Fig. 1F](#f1-ol-0-0-11049){ref-type="fig"}). In addition, the inhibition of YAP expression increased the expression of cleaved caspase-3 ([Fig. 1G and H](#f1-ol-0-0-11049){ref-type="fig"}) and the number of TUNEL-positive cells ([Fig. 1I and J](#f1-ol-0-0-11049){ref-type="fig"}) in gastric cancer MKN-28/74 cells. These results suggested that YAP was upregulated in gastric cancer MKN-28/74 cells and promoted cell survival by inhibiting apoptosis.
### YAP is associated with MKN-28/74 cell migration and invasion
The role of YAP in MKN-28/74 cell migration and invasion was further investigated. Knockdown of YAP significantly reduced wound closure rates in the wound-healing assay ([Fig. 2A and B](#f2-ol-0-0-11049){ref-type="fig"}). In addition, compared with the control group, knockdown of YAP reduced the invasive ability of gastric cancer MKN-28/74 cells ([Fig. 2C and D](#f2-ol-0-0-11049){ref-type="fig"}). These results suggested that YAP promoted MKN-28/74 cell migration and invasion.
### YAP promotes MKN-28/74 cell survival and migration/invasion through the inhibition of ER stress
ER stress serves a critical role in the progression of cancer ([@b38-ol-0-0-11049],[@b39-ol-0-0-11049]). To determine the underlying mechanism by which YAP may regulate gastric cancer MKN-28/74 cell survival and metastasis, the present study focused on ER stress. TM, the activator of ER stress, was used to induce ER stress in MKN-28/74 cells transfected with siCtrl. 4-phenylbutyrate (4-PBA), the inhibitor of ER stress, was used to inhibit ER stress in YAP-knockdown MKN-28/74 cells. Western blotting ([Fig. 3A-C](#f3-ol-0-0-11049){ref-type="fig"}) and immunofluorescence ([Fig. 3D and E](#f3-ol-0-0-11049){ref-type="fig"}) were used to determine the changes in ER stress markers. Compared with the siCtrl group, knockdown of YAP contributed to the upregulation of GRP78, CHOP and cleaved caspase-12; similar results were observed following TM treatment in the siCtrl group. However, the upregulation of ER stress markers was partially reversed by 4-PBA ([Fig. 3A-E](#f3-ol-0-0-11049){ref-type="fig"}). These results suggested that YAP knockdown was associated with ER stress. In addition, ER stress activation was associated with apoptosis activation ([Fig. 3F-H](#f3-ol-0-0-11049){ref-type="fig"}) and the inhibition of migration/invasion ([Fig. 3I-L](#f3-ol-0-0-11049){ref-type="fig"}). By contrast, inhibiting ER stress with 4-PBA in YAP-knockdown cells promoted cell survival and invasion. These results indicated that YAP promoted gastric cancer MKN-28/74 cell survival and migration/invasion through the regulation of ER stress.
### YAP regulates ER stress via the ERK pathway
Finally, experiments were performed to determine how YAP inhibited ER stress. The ERK pathway has been reported to be involved in YAP-associated functions and ER stress inhibition ([@b40-ol-0-0-11049],[@b41-ol-0-0-11049]). PD98059, an inhibitor of the ERK pathway, was used to inhibit the ERK pathway in MKN-28/74 cells transfected with siCtrl. The activation of the ERK pathway was assessed by western blotting. Compared with the control group, YAP knockdown inhibited ERK phosphorylation, similar to PD98059 treatment ([Fig. 4A](#f4-ol-0-0-11049){ref-type="fig"}). The inhibition of the ERK pathway by siYAP promoted the activation of ER stress as indicated by the upregulation of CHOP and reduced cell invasion ([Fig. 4](#f4-ol-0-0-11049){ref-type="fig"}). These results suggested that the ERK pathway may contribute to the YAP-induced ER stress inhibition.
Discussion
==========
Previous studies have demonstrated that YAP is essential for gastric cancer cell survival and migration/invasion ([@b42-ol-0-0-11049]--[@b44-ol-0-0-11049]). However, the underlying mechanism remains unclear. The present study proposes a novel underlying mechanism by which YAP regulates gastric cancer MKN-28/74 cell survival and metastasis. The results of the present study demonstrated that: i) YAP was upregulated in gastric cancer MKN-28/74 cells compared with normal gastric GES-1 cells; ii) YAP promoted gastric cancer MKN-28/74 cell survival and migration/invasion by inhibiting ER stress; iii) YAP may regulate ER stress by activating the ERK pathway. The present study provides a new target for the treatment of gastric cancer that may affect cancer cell survival and metastasis. A limitation of the present study was that only one gastric cancer cell line was used. Additional cell lines will be used in our future study, to confirm the results.
In eukaryotic cells, the ER is responsible for protein synthesis and calcium storage; perturbations in the ER function, a process termed ER stress, have been reported to be involved in cancer initiation, growth and metastasis in the majority of solid tumors ([@b45-ol-0-0-11049],[@b46-ol-0-0-11049]). However, the role of ER stress in tumorigenesis and development is still controversial. Previous studies have demonstrated that ER stress is a tumor suppressor, and the activation of ER stress inhibits gastric cancer cell survival and migration ([@b19-ol-0-0-11049],[@b47-ol-0-0-11049],[@b48-ol-0-0-11049]). However, a number of studies have suggested that ER stress can promote tumor development ([@b49-ol-0-0-11049],[@b50-ol-0-0-11049]). Induction of ER stress protects gastric cancer cell apoptosis during cisplatin and doxorubicin treatment via the p38 MAPK pathway ([@b51-ol-0-0-11049]).
Recent studies have identified an association between YAP and ER stress. The activated Hippo-YAP signaling pathway promoted neuron survival in the TNFα-induced microenvironment by inhibiting ER stress ([@b52-ol-0-0-11049]). In addition, downregulation of YAP evoked ER stress and contributed to myocyte death in isoproterenol-induced myocardial infarction ([@b53-ol-0-0-11049]). The results of the present study are consistent with previous studies. However, the exact mechanism by which YAP controls ER stress remains unknown. The results of the present study suggested that YAP may inhibit ER stress via the ERK pathway. Thus, these results provide valuable information on the role of YAP and ER stress in tumorigenesis.
In the present study, the critical role of YAP in the progression of gastric cancer was identified. A recent study demonstrated that YAP regulates gastric cancer survival and migration through SIRT1/Mfn2/mitophagy ([@b42-ol-0-0-11049]). The results of the present study demonstrated that YAP may function via the ERK/ER stress pathway in gastric cancer survival and metastasis. To the best of our knowledge, this is the first identification of YAP functions involved in ER stress and the ERK pathway in the development of gastric cancer. However, *in vivo* experiments and clinical data are required to support these results.
In conclusion, the results of the present study identified the important role of YAP in gastric cancer cell migration and survival. YAP promoted gastric cancer MKN-28/74 cell survival and migration/invasion via the ERK/ER stress pathway. These results suggested that the YAP/ERK/ER stress pathway may be a potential target for the treatment of gastric cancer.
Not applicable.
Funding
=======
This work was supported in part by Inner Mongolia Autonomous Region Natural Science Foundation (grant no., 2016MS0847), Scientific Research Planning Project of Health and Family Planning Commission of Inner Mongolia Autonomous Region (grant no., 201701048) and Science and Technology Innovation Guidance Project of Inner Mongolia Autonomous Region (grant no., KCBJ2018021).
Availability of data and materials
==================================
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Authors\' contributions
=======================
HL and DM conceived and designed the study, performed the experiments, analyzed and interpreted the data and wrote the manuscript. PX, HW and YW were involved in data analysis and interpretation.
Ethics approval and consent to participate
==========================================
Not applicable.
Patient consent for publication
===============================
Not applicable.
Competing interests
===================
The authors declare that they have no competing interests.
{#f1-ol-0-0-11049}
{#f2-ol-0-0-11049}
{#f3-ol-0-0-11049}
{#f4-ol-0-0-11049}
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
Fungal infections are a major cause of morbidity and mortality despite the latest developments of diagnostic tools and therapeutic options. Early initiation of the correct antifungal therapy has been demonstrated to have a direct impact on the patient\'s outcome[@B46]. More severe infections affect mainly immunocompromised patients but other populations are also infected. New chronic lung infections have been described with a huge impact on the patient\'s quality of life, and a high cost of treatment and care. Besides, some skin fungal infections involving mucosa and subcutaneous tissues cause substantial morbidity[@B39]. *Cryptococcus, Candida, Aspergillus,* and *Pneumocystis* are the main etiologic agents of fungal infections[@B12]. The burden and mortality associated with these diseases depend on the region and the affected population. Thus, it has been estimated that cryptococcal meningitis affects nearly one million people per year. Despite treatment, mortality rates can reach 55 to 70% in AIDS patients in Latin America and sub-Saharan Africa, the estimated number of deaths per year being over 620,000[@B45].
*Cryptococcus, Candida, Aspergillus,* and *Pneumocystis* affect mainly immunocompromised individuals, however endemic dimorphic fungi such as *Histoplasma, Blastomyces, Coccidioides* and *Paracoccidioides* affect immunocompetent patients as well, and endemic areas include several regions of Latin America[@B24].
Nowadays, three main families of antifungals are used in the clinical setting to treat fungal infections: polyenes represented by amphotericin B (and its different formulations); azoles with several derivatives such as itraconazole, fluconazole, voriconazole, posaconazole, isavuconazole; and the echinocandins caspofungin, micafungin and anidulafungin. The availability of new antifungals in recent years has provided clinicians with more options, increasing the use of these compounds not just for treatment when the infection has been diagnosed, but also as prophylactic, empirical or preemptive treatment. The increased use of antifungals has induced a higher selective pressure on fungal strains and resistance has emerged in two main ways: several species have developed secondary resistance and susceptible species have been replaced by resistant ones, changing the epidemiology of fungal infections[@B41].
Antifungal susceptibility testing methods are available to detect antifungal resistance and to determine the best treatment for a specific fungus. Clinical microbiology relies on these methods to select the agent of choice for a fungal infection, and to know the local and the global epidemiology of antifungal resistance.
Microdilution methods are the gold standard or reference techniques. Two organizations, the European Committee on Antibiotic Susceptibility Testing (EUCAST) and the Clinical Laboratory Standards Institute (CLSI) have standardized methods to perform antifungal susceptibility testing. Differences between these two methods have been widely discussed in several reports and will be reviewed in the present manuscript, however their results have demonstrated to be comparable and are used worldwide. Both institutions have developed breakpoints (BPs) of some antifungals to *Candida* and *Aspergillus* species that are currently used to classify resistant strains.
Regardless of their advantages, the standardized broth microdilution methods of antifungal susceptibility testing are time-consuming and cumbersome for clinical laboratories. Some commercially available, including manual, semi-automated and automated methods, do not require complex handling and are cost-effective alternative methods to test antifungal agents *in vitro* against *Candida* isolates in routine usage, and *Cryptococcus* isolates and filamentous fungi for research purposes. The characteristics of these methods together with their comparison with the reference procedures and the available agar-based methods will also be reviewed in this manuscript.
REFERENCE METHODS
=================
Broth microdilution methods
---------------------------
Clinical Laboratory Standards Institute (CLSI)
----------------------------------------------
In 1985, the CLSI, formerly known as the National Committee for Clinical Laboratory Standards (NCCLS), formed a subcommittee on Antifungal Susceptibility Testing that published, in 1997, the document M27A \"Reference Method for Broth Dilution Antifungal Susceptibility Testing of Yeast; Approved Standard\"[@B19]. This document defined reference strains with ranges of Minimal Inhibitory Concentrations (MIC) and Break Points (BPs) for some antifungals and their action against yeasts such as *Candida* and *Cryptococcus*. Since then, several updates have been published, the current one having been approved in April 2008[@B20]. For filamentous fungi, the first document was published in 2002: M38A: \"Reference Method for Broth Dilution Antifungal Susceptibility Testing of Filamentous Fungi; Approved Standard\" with a second edition published in 2008, which is the currently accepted one[@B21].
European Committee on Antimicrobial Susceptibility Testing (EUCAST)
-------------------------------------------------------------------
The EUCAST is a standing committee jointly organized by the European Society of Clinical Microbiology and Infectious Diseases (ESCMID), the European Centre for Diseases Control (ECDC), and the European National Breakpoint Committees (www.eucast.org). The antifungal susceptibility testing subcommittee of the EUCAST (AFST-EUCAST) was formed in 1997, and in 2008 published a standard of susceptibility testing for yeasts (including *Cryptococcus*). This standard was updated in 2012[@B06]. Another standard for molds was published in 2008. All of these standards are available online and can be downloaded from the EUCAST website (www.eucast.org). Differences between both standards are mainly found in the inoculum size, incubation time and medium composition ([Table 1](#t01){ref-type="table"}). Despite these differences, the results obtained by both methods are comparable[@B15] ^,^ [@B55]. CLSI and EUCAST include in their standards to test yeast some modifications for *Cryptococcus.* Thus, in CLSI the recommendation is to read MICs for *Cryptococcus* after 70 to 74 hours of incubation (in contrast with 24-48h for *Candida*), while EUCAST recommends the incubation of the plates at 30 ºC when the growth control does not reach an optical density of 0.2 at 35 ºC. Neither CLSI nor EUCAST have published standards for endemic dimorphic fungi such as *Histoplasma* or *Paracoccidioides.*
Table 1CLSI vs EUCAST methodologies for antifungal susceptibility testing[^1]
Agar-based methods
------------------
Disk tests are inexpensive and easy to set up, and provide an ideal screening test. The disk diffusion method to test antifungals (CLSI M44 series) has been developed and validated only in the case of azoles and echinocandins for *Candida* spp. isolates[@B16]. It recommends the use of Mueller-Hinton agar supplemented with 2% glucose, providing a suitable growth for most yeasts, and 0.5 mg/L methylene blue dye medium (enhances the zone edge definition) minimizing the trailing effect. The pH of the medium needs to be between 7.2 and 7.4 after gelling and the agar should be 4 cm high. The inoculum is standardized to 0.5 McFarland using a densitometer and plates should be incubated at 35 ⁰C for 24 h; some strains show insufficient growth and may need 48 h of incubation. In addition, quality control parameters have been established following the CLSI standard procedures. The results of the susceptibility test according to the zone diameter interpretative criteria for caspofungin, fluconazole and voriconazole for *Candida* species allows to classify the isolate in one of the following categories: susceptible, resistant, susceptible dose dependent and non-susceptible, corresponding to MIC breakpoints[@B23].
The essential agreement between the disk diffusion and the CLSI microdilution method to test the susceptibility of azoles against *Candida* and *Cryptococcus* isolates is usually higher than 90% demonstrating that the disk diffusion is able to identify resistant isolates[@B11] ^,^ [@B35]. Regarding echinocandins and *Candida* species, the disk diffusion test appears to be able to differentiate caspofungin-susceptible between resistant mutant isolates. However, the disk diffusion test for micafungin appears to be less optimal due to a close overlapping of susceptible and resistant populations. In the case of *C. parapsilosis* and *C. glabrata,* there is a need for individual breakpoints. This behavior has been observed with either EUCAST or CLSI microdilution methods and thus appears to be drug-related rather than dependent on the choice of the *in vitro* susceptibility test format. Nevertheless, while susceptibility classification is improved by the application of recently revised breakpoints, further evaluation and refinement are needed[@B08] ^,^ [@B09].
The standard disk diffusion method to test antifungal drugs for non-dermatophyte filamentous fungi isolates (M51-A and supplement M51-S1) provides qualitative results in 8-24 h when caspofungin, triazoles, and amphotericin B are used, faster than the CLSI reference microdilution method[@B17]. Among *Aspergillus* species, a lower agreement of results produced by disk diffusion susceptibility tests was reported for *A. flavus* and amphotericin B or voriconazole. Amphotericin B to test *A. fumigatus* susceptibility also showed a lower agreement when compared to the reference method. Amphotericin B disks usually show the lowest correlation between MICs and inhibition zone diameters for filamentous fungi. The percentage of major errors is usually similar to that obtained with the itraconazole disk, but the percentage of minor errors is higher[@B34] ^,^ [@B38]. Although breakpoints for filamentous fungi have not been defined, epidemiological cut-off values can be proposed to identify non-wild-type isolates[@B18].
Although qualitative results provided by the disk diffusion method are useful in the clinical laboratory routine, quantitative MIC data is somewhat critical for the management of invasive infections.
Breakpoints
-----------
Even though the main goal of AFST is to select the best treatment for a given isolate, these methods are also very important to detect resistant strains, allowing the establishment of an epidemiology map of antifungals resistance that is an emerging problem in medical mycology. The two main factors are: the development of secondary resistance and the selection of species that are intrinsically resistant. Therefore, AFST has become critical for the choice of the best antifungal agent. Breakpoints have been developed for some fungal species and antifungals, in both CLSI and EUCAST methods. These BPs categorize fungal isolates into (i) susceptible (the drug is an appropriate treatment); (ii) resistant (the drug is not recommended as a treatment), and (iii) intermediate (the drug may be an appropriate treatment, depending on certain conditions; e.g. fluconazole to treat a urinary infection caused by an intermediate strain).
BPs definition is a complex process based on the critical review of several aspects and data. CLSI evaluates MIC distributions, the relationship between MICs and clinical outcome, pharmacokinetics and pharmacodynamics. CLSI proposed a single interpretative BPs for fluconazole, itraconazole[@B19], voriconazole[@B52], and echinocandins[@B50] for all *Candida* species. Latter, CLSI BPs was revised including a number of clinical studies and cases reporting strains classified as susceptible but associated with treatment failure, and as a consequence, species-specific BPs were proposed, as had been previously established by EUCAST[@B22] ^,^ [@B49].
EUCAST evaluates five aspects to develop BPs: (i) The most common dosage used in each European country; (ii) the definition of the wild type population for each target microorganism at the species level, and the determination of epidemiological cut-offs; (iii) the pharmacokinetics of the drug; (iv) the pharmacodynamics including Monte Carlo simulations; and (v) the correlation of MICs with patients\' clinical outcome treated with this drug. Clinical BPs have been established for several antifungals for *Candida* spp. and *Aspergillus* spp. These BPs are freely available online at: www.eucast.org/clinical_breakpoints.
As stated before, although some differences have been recognized for several years, currently CLSI and EUCAST breakpoints are in agreement. [Tables 2](#t02){ref-type="table"} and [3](#t03){ref-type="table"} represent the established BPs for several antifungal agents for *Candida* and *Aspergillus* in both standards.
Resistance
----------
Antifungal resistance is becoming an emerging problem. On the one hand, there is the intrinsic resistance, and on the other hand the development of secondary resistance, that should be detected because resistant strains are associated with poorer outcomes. To illustrate this problem, intrinsic resistance of *C. glabrata* and *C. krusei* to fluconazole is well known. In these cases, appropriate treatment can be decided on the basis of species identification[@B07]. This intrinsic resistance has justified the use of echinocandins as primary treatment, instead of fluconazole, in the empirical treatment of candidemia and invasive candidiasis in recently published guidelines[@B25] ^,^ [@B58]. In addition, intrinsic resistance to echinocandins has been described in *C. parapsilosis,* and *Cryptococcus neoformans* isolates[@B59]. Although it is less common, during antifungal therapy acquired resistance in *Candida* spp. infections has also been reported. Most cases involve *C. glabrata* resistance to echinocandin although other species such as *C. albicans, C. tropicalis* and *C. krusei*, have also proven able of developing secondary resistance[@B31] ^,^ [@B37]. Alterations on genes encoding the target enzymes of these drugs (beta 1-3 D-glucan synthase for echinocandins (FKS) and 14 alpha sterol demethylase for azoles (ERG11) or up regulation of multidrug efflux transporters also for azoles (ABC \[ATP-binding cassette\]/MFS \[major facilitator superfamily\]) have been blamed for the *Candida* spp. resistance to antifungal agent. Point mutations located at two hot spot regions within the FKS genes of *Candida* spp. have been described and associated with echinocandins resistance[@B53].
Secondary resistance to amphotericin B has been described in *C. tropicalis*, *C. parapsilosis*, *C. lusitaniae*, and *C. haemulonii* [@B33]. In addition, several emerging pathogens such as *A. terreus*, *Fusarium* spp, and *Lomentospora prolificans* (syn. *Scedosporiumprolificans*) are intrinsically resistant to amphotericin B[@B02] ^,^ [@B30]. The mechanism of resistance to amphotericin B has been associated with a decrease of ergosterol content in fungal membranes, mainly due to alterations in the ergosterol biosynthesis pathway. It has also been suggested that resistance to amphotericin B could be related to disruption of the fungal mitochondria[@B44].
The azole resistance of *Aspergillus* isolates has been rigorously investigated in the last years. Alterations in the coding region of the *cyp*51A gene (positions G54, G138, M220, G448) or an insertion of a 34 to 36 base pair tandem repeat in the promoter region of the gene, together with point mutations (positions L98, Y121 and T289) have been associated with azole resistance. Mechanisms of azole resistance have been described both prior to triazole exposure and acquired during therapy[@B04]. The use of azoles in agriculture has been described as a cause of the emergence of triazole resistant in *Aspergillus fumigatus* isolates, particularly in Europe and Asia[@B56].
Table 2EUCAST and CLSI antifungal breakpoints for *Candida*[^2]
Table 3EUCAST antifungal breakpoints for *Aspergillus*
Other filamentous fungi are intrinsically resistant to some antifungals. The order mucorales comprises several pathogenic species that are resistant to voriconazole[@B01]. Species of the genera *Fusarium* and *Scedosporium* also show elevated MICs of several antifungals[@B03] ^,^ [@B26]. In addition, multi-resistant species are also present as human pathogens. *Lomentospora prolificans* (syn. *Scedosporiumprolificans*) is resistant to all azoles, echinocandins and amphotericin B, and has been associated with poorer outcomes[@B54].
Commercial methods
------------------
Clinical laboratories can determine susceptibility to antifungals through a series of commercially available systems, including the Sensititre YeastOne(r) panel (TREK Diagnostic Systems, Cleveland, USA) and the Vitek 2 system, both based on microdilution methods, or agar-based assays, e.g. test strips (E-Test(r), bioMérieux; MIC(r), Oxoid) and discs impregnated with a single antifungal agent.
In order to choose a commercial method, first of all, the laboratory should be aware of the commercial techniques results considering the susceptibility of each drug to a particular fungus, comparing the CLSI and the EUCAST reference procedures. In general, the correlation is based on the essential agreement (EA), defined as the discrepancies between MIC results of no more than ± 2 twofold dilutions, and the categorical agreements (CA). The latter depends on the existence of interpretative break points[@B27] ^,^ [@B48]. Of note, the laboratory should perform tests strictly as instructed in the commercial guidelines to get reliable results. Additionally, quality control strains, such as *C. krusei* ATCC 6258 and *C. parapsilosis* ATCC 22019 must be included in each commercial system batch, and be ascertain that all MIC values are within the expected ranges.
Commercial broth microdilution methods
--------------------------------------
Sensititre YeastOne(r) is a well-described colorimetric microdilution panel that contains dried serial twofold dilutions of up to ten antimycotics in individual wells. YeastOne(r) provides customizable dual-isolate five antifungal format for *Candida* spp. (clinical use), and single-isolate, nine antifungal, research-use-only format including anidulafungin and micafungin to be tested for yeast and filamentous fungi (not for use in diagnostic procedures). The susceptibility of isolates to antimycotics is assessed on the basis of growth or inhibition of the isolate in the culture media containing antimycotic agents. The system incorporates Alamar Blue(r), a colorimetric indicator of an oxidation-reduction reaction (fungal growth changes media color from blue to pink). Endpoint determination was based on visual reading or software-facilitated visual reading (Vizion(r) system) after 24-25 h (*Candida* spp.) or 48-72h (*Cryptococcus* spp.) of incubation at 35 °C. The panel has the advantage of being ready to use, easy to perform, quick and timely results, and individual packaging allows the test of one plate at a time. The Sensititre Yeast One(r) method showed good results in terms of reproducibility and agreement with reference methods considering fluconazole and *Candida* spp. (EA ³ 95%) although a lower agreement (EA 79-92%) was found to *C. neoformans* isolates. YeastOne(r) panel was reported to yield higher MICs, in comparison with the CLSI method, for all drugs except for caspofungin and flucytosine[@B05]. With respect to filamentous fungi, a strong correlation with the M38-A2 (CLSI) method was found for itraconazole and voriconazole. The method showed a strong correlation with CLSI to detect resistant isolates and may help to monitor the emergence of isolates with decreased susceptibility to antifungal agents.
Another commercially available system, called SensiQuattro *Candida* EU(r) (bestbiondx, Germany), correlates well with the antifungal clinical break points established by EUCAST. This 32-well colorimetric microdilution panel includes four doubling serial concentrations of amphotericin B, fluconazole, voriconazole, posaconazole, caspofungin, anidulafungin, micafungin, and flucytosine. The resulting colors are interpreted as follows: a yellow/orange color indicates yeast growth; a red color indicates yeast growth inhibition. When compared to the EUCAST reference, the broth microdilution method showed a good correlation for amphotericin B and azoles, but poor for echinocandins[@B40].
Vitek 2(r) yeast susceptibility test (bioMérieux, Inc.) is an automated method of yeast species identification and antifungal susceptibility testing through the analysis of yeast growth. The system provides 64-well cards containing aliquots of amphotericin B, fluconazole, flucytosine, and voriconazole in a miniaturized version of the broth dilution method. The system integrates a software program which validates and interprets susceptibility test results according to CLSI clinical breakpoints based on the drug MIC values**.** The high level of standardization achieved by this automated system was demonstrated in several studies[@B05] ^,^ [@B13] ^,^ [@B27] **.** The number of hours to deliver an MIC result was reported to vary from 9.1h to 15h for *Candida* species (minimum 7.5 h to maximum 18 h) and 12.1 h for *C. neoformans* [@B05] ^,^ [@B13] ^,^ [@B27]. In general, the MICs obtained by the Vitek 2(r) system are slightly higher than those generated by the CLSI methodology for both *Candida* and *Cryptococcus* **.** However, Vitek 2(r) results are reproducible, accurate, present a strong correlation with those obtained with the CLSI and the AFST-EUCAST reference methods for fluconazole, amphotericin B, flucytosine, and voriconazole. The correlation with the reference methods was also very good when resistance to antifungals was studied[@B43] ^,^ [@B47] ^,^ [@B51].
Commercial agar-based methods
-----------------------------
Commercially prepared strips are available from bioMérieux (Etest(r)) and Liofilchem Diagnostici (MIC Test Strip(r)). The method consists of a predefined gradient of antifungal drug concentrations on a plastic strip that is used to determine the MIC. When the strip is applied on an inoculated agar surface, the antifungal agent is immediately transferred to the agar matrix and after an incubation time, an inhibition ellipse centered along the strip is formed. The recommended agar is RPMI 1640 supplemented with 2% glucose, prepared with MOPS in a 1.5% agar base. The Etest(r) provides strips with fluconazole, itraconazole, amphotericin B, flucytosine, voriconazole, posaconazole, and caspofungin. The MIC Test Strips(r) contains the same antifungal drugs plus anidulafungin, micafungin and ketoconazole. The incubation time range from 24-48 h for *Candida* species, from 48-72 h fo*r C. neoformans*, and for filamentous fungi it lasts 16 h or longer depending on the fungus\' genus. The MIC is read directly from the scale at the point where the edge of the ellipse intersects the strip. However, it is important to consider that, as for any test evaluating the antimicrobial susceptibility, the medium formulation and, in this case, the depth of the agar can strongly influence MIC results. Therefore, the manufacturer\'s recommendations should be strictly followed to obtain MICs using strips[@B16] ^,^ [@B17] ^,^ [@B21].
Results obtained by the E-test method shows a \> 71% correlation with those obtained by the AFST-EUCAST method. In both methods, the CLSI and EUCAST AFST, the agar-based E-test has been proposed as a more sensitive technology to discriminate strains of *Candida* species with *fks* mutations from wild-type (WT) strains by virtue of much higher MIC results observed in mutant strain. Considering *Cryptococcus*, the overall agreement level using the E-test MICs and the EUCAST AFST-MICs seems to be higher for voriconazole, fluconazole, itraconazole and flucytosine, than for amphotericin B, which has the lowest level of agreement. Regarding filamentous fungi, the agreement is higher for itraconazole than for amphotericin B, and the E-test method showed a good correlation with the CLSI M38-AFST one to detect *Aspergillus* resistance. Systematic comparisons between MIC results from reference laboratories and routine results obtained using commercially available methods could be more representative than the current practice to perform quality control with a specific set of reagents using a limited number of isolates[@B14].
CONCLUSIONS
===========
The role of microdilution methods seems to be restricted to reference laboratories because they are laborious. In addition, the microbroth format is not commonly used in clinical laboratories. Several automated or semi-automated commercial methods based on agar diffusion or the use of colorimetric indicators in Etest, Sensititre YeastOne, Fungitest or Vitek have been designed for routine daily practice. Disk and strip diffusion methodologies are simple, rapid, cost-effective and produce similar results to the reference methods for yeasts. Automated systems significantly reduce the biologist hands-on time, turnaround time, and variability due to the standardized format. Evaluation of these methodologies requires the determination of break point category agreements with reference methods. It is noteworthy that interpretative break points are only available for a few species of *Candida*, *Cryptococcus* and *Aspergillus.* Most of these tests have been able to detect *in vitro* resistance of *Candida* isolates, however some discrepancies have also been described. Reference procedures are irreplaceable nowadays to test and validate new antifungal agents, new methods and techniques, and the susceptibility profile of rare species which have not been evaluated by other methods[@B27] ^,^ [@B29] ^,^ [@B42]. Also, the increase of resistant strains associated with treatment failure highlights the need of antifungal resistance surveillance, which should ideally be made in reference laboratories using reference procedures.
[^1]: Differences between two methods are in bold. AMB = amphotericin B; FCZ = fluconazole; Candins = anidulafungin, caspofungin, micafungin.
[^2]: SDD = susceptible dose dependant; IE = insufficient evidence.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Hypertensive patients are at risk for cardiovascular complications related to endothelial dysfunction and angiogenesis ([@B1]). Angiogenesis is the process of formation of new vasculature and expansion of the capillary network from preexisting vessels which is essential for regeneration and tissue repair ([@B2]). Vascular endothelial function is well correlated with angiogenesis and the endothelium promotes vasodilation, inflammation and vascular smooth cell proliferation by releasing nitric oxide (NO) ([@B3]). NO as a biological messenger is known to be involved in diverse physiological and pathophysiological processes in various organ systems. NO is a potential regulator for angiogenesis and many studies have shown that NO is closely involved in the regulation of systemic blood pressure ([@B4]).
In the other hand, morphine is commonly used to control severe pain and also in addicted persons. Whereas the mechanisms of opioids have been mostly characterized in nervous system, little is known about their systemic effects. Some studies have shown the presence of specific opioid receptors, including MOR on the endothelial cells ([@B5],[@B6]). Moreover, morphine induces NO in the endothelium and some other tissues and leads to vasodilation ([@B7]). Opioids also promote cell proliferation in non endothelial cells ([@B9],[@B10]). These findings show NO- dependent angiogenesis is induced by low-dose of morphine, but there are some evidence which shows that endothelial cells express a local circuit regulatory pathway driven by endogenous opiates and constitutive ([@B11]). However, very few studies are available investigating the effect of morphine on angiogenesis in hypertensive animals and results are controversial ([@B12]-[@B15]). For example, high-dose morphine has cytotoxic effects in macrophages ([@B16]), vascular endothelia ([@B17]), mesangial, and epithelial cells ([@B18]).
Because the effects of morphine on these processed remains unknown, we examined the effects of prolonged low-dose morphine on induction of hypertension and angiogenesis in two-kidneys one clip reno vascular hypertensive (2K1C) rats.
Materials and Methods
=====================
Thirty two male Wistar rats with initial weight of 200±20 gram were enrolled in this study and housed at a controlled temperature with free access to food and water. The animals were randomly divided into 2 groups: sham normotensive and 2K1C hypertensive rats. Each group was divided into saline and morphine receiving groups (n= 8). Eight weeks immediately after 2K1C surgery 3 mg/kg of Morphine sulphate was injected i.p. and the saline group received the same volume of saline in according to a similar protocol. Surgeries and protocols were performed as follow:
***Preparation of hypertensive rats by 2K1C gold blatt method***
The rats were anesthetized with ketamine hydrochloride (60 mg/kg) and xylazine (7.5 mg/kg) intraperitonealy. Left kidneys were exposed via flank incision and a silver clip with internal gap of 0.2 mm was put around the renal artery. In the sham group, the same procedure was done without using silver clip. Penicillin G (25000 U IM) was injected after surgery. Rats were fed a commercial rat chow (Razi Institute, Iran) and had free access to tap water. A few days after placement of the clip, the systolic blood pressure (SBP) was measured twice a week with the tail-cuff method (AD instrument Australia). After 8 weeks, the animals were anesthetized and direct blood pressure was measured by a catheter (PE50) inserted into femoral artery. Blood samples were taken for subsequent determination of plasma renin activity (PRA).
***Plasma renin activity assay***
PRA was measured with a kit from Diasorin Inc. using ^125^-I Angiotensin I generation. Angiotensin I coated-tube radioimmunoassay (RIA) was performed in two aliquots of the same sample, one incubated at 37 ºC for generation and one non-incubated; PRA was calculated as ng angiotensin I generated/ml/h (Renctk P2721, Sorin-Biomedica Diagnostic Division RIA kit, Italy). The PRA assay sensitivity was 0.13 ng/ml; intra-and interassay coefficients of variation were 7.5 and 7.7%, respectively.
***Protocol for determining serum nitric oxide concentration***
From all animals blood samples were taken before and after the study. Serum NO concentration was measured by Gris reagent system (Promega Corporation, Madison, USA) and using available reagents. Serums were added into wells (96-well flat-button enzymatic assay plate). A sulfanilamide solution was added to all collected samples and then *N*-1-naphtylethylenediamine dihydrochloride (NED) was added under acidic conditions. The absorbances were detected in 520-550 nm wavelengths by a microreader ([@B19]). NO concentration in the samples was determined by comparison to nitrite standard curve. The limitation of detection was 2.5 µM nitrite.
***Murine matrigel angiogenesis assay***
Angiogenesis was assessed *in vivo* using 500 μl of Matrigel (BD Bioscience, San Jose, CA) containing fibroblast growth factor (10 ng/ml, R&D System, Minneapolis, MN) and heparin (60 U/ml, Braun Melsungen AG, Melsungen, Germany) which was injected subcutaneously into the abdominal wall of rats 8 weeks after the commencement of morphine or saline administration (as described earlier). Ten days later, 25 mg/ml FITC-dextran was injected systemically, and blood samples were collected. After sacrificing the rats, Matrigel implants were excised and photographed under a fluorescent microscope and then homogenized with 5 units/ml dispase (Life Technologies, Inc., Grand Island, NY). Angiogenic response was expressed as the fluorescence ratio of Matrigel implant: plasma, obtained using a Fluorescence Multi Plate Reader (Applied Biosystems, Foster City, CA) ([@B20]).
***Statistical analysis***
The results are presented as Mean±SEM. Data were compared by an unpaired t-test or ANOVA and Tukey as a post test as appropriate. Statistical significance was accepted at a level of *P*\< 0.05.
Results
=======
***Blood pressure and heart rate***
Systolic blood pressure (SBP), diastolic blood pressure (DBP), mean arterial pressure (MAP), and heart rate (HR) are shown in [Figure 1](#F1){ref-type="fig"}-[3](#F3){ref-type="fig"} in all groups. In 2K1C group SBP and DBP increased significantly (*P*\< 0.01) and HR decreased (*P*\< 0.01) compared with sham animals. Morphine had no effects on blood pressures and HR in sham normotensive rats but attenuates DBP (*P*\< 0.01) and MAP (*P*\< 0.01) in 2K1C compared with saline receiving group ([Figure1](#F1){ref-type="fig"} to [3](#F3){ref-type="fig"}).
{#F1}
{#F2}
***Plasma renin activity***
Plasma renin activity (PRA) level was significantly higher in 2K1C (*P*\< 0.01) in comparison with sham groups ([Figure 4](#F4){ref-type="fig"}). Morphine decreased PRA in 2K1C compared with saline receiving group (*P*\< 0.01) but PRA in 2K1C morphine group was still higher than Sham (*P*\< 0.01)
***Serum NO concentration***
[Figure 5](#F5){ref-type="fig"} illustrates serum NO concentration in normotensive sham and 2K1C hypertensive rats. Morphine had no effect on serum NO concentration in sham-clip operated animals (*P*\> 0.05). After clipping, serum NO concentration in 2K1C hypertensive rats was decreased and it was significantly lower than sham-clipped group (*P*\< 0.01). Morphine improved serum NO level in 2K1C group (*P*\< 0.01) but NO was still lower than 2K1C-saline group (*P*\< 0.0001).
***Effects of morphine in angiogenesis***
Low-dose morphine stimulated angiogenesis on the Matrigel plugs in sham--clip operated groups compared with saline ([Figure 6](#F6){ref-type="fig"}). 2K1C hypertension significantly impairs angiogenesis compared with sham operated rats (*P*\< 0.01) but morphine promoted angiogenesis in both sham-clip (*P*\< 0.01) and 2K1C (*P*\< 0.0001) groups in comparison with saline.
{#F3}
{#F4}
{#F5}
{#F6}
Discussion
==========
Our results showed that prolonged low dose morphine did not change blood pressure and heart rate in sham-clipped animals. Blood pressure and PRA increased and heart rate decreased in 2K1C compared with sham-clipped group. Prolonged low dose morphine prevented the promoting of PRA and hypertension in 2K1C compared with saline. Serum NO concentration was significantly decreased after clipping in 2K1C saline receiving group but morphine induced NO in 2K1C group compared with saline.
Studies indicated that in early phase of 2K1C hypertensive animals, increased activity of PRA and renin-angiotensin-aldosterone is responsible for increasing blood pressure ([@B21],[@B22]), however, after 8 week of clipping, change of endothelial function is importance regarding hypertension. Reduced blood pressure after prolonged use of low-dose morphine in 2K1C animals may be due to decreased PRA level ([@B23]).
We also found that serum NO level was reduced in hypertensive group but prolonged use of low-dose morphine reversed blood pressure, PRA and NO close to normotensive levels. Abnormality in endothelium function which is characterized by decreased in NO concentration, is an important risk factor in hypertension. Endothelial dysfunction in releasing endothelium-derived relaxation factors such as NO and impaired endothelium--dependent relaxation have been demonstrated in several animal models of hypertension and clinical studies ([@B24]-[@B27]). It suggests that abnormality of NO pathway in hypertension may be due to lower NO production and/or higher NO degradation. Moreover, an increase in reactive oxygen species generation and lower level of antioxidants were detected in hypertensive subjects ([@B28]). In addition, lower NO production may be due to reduced L-argenine or endothelium nitric oxide synthase expression in hypertensive rats ([@B29]). Thus, reduced serum NO concentration in 2K1C hypertensive rats may be due to cardiovascular effects of hypertension.
Furthermore, 2K1C model induces renin-angiotensin-aldosterone dependent hypertension and another explanation for reduced NO bioavailability in this model of hypertension is that, high angiotensin II level decreases NO level by promoting oxidative stress ([@B30]). We also found that prolonged low dose morphine reversed blood pressure and serum NO concentration close to normal level.
A predominantly antihypertensive role has been reported for endogenous morphine- NO signaling events. Therefore morphine modulates endothelial function and vascular endothelial cells functionality and expresses a local paracrine-autocrine regulatory pathway by endogenously expressed authentic morphine, and constitutive NO. Moreover, considerable evidence shows that NO signaling pathway plays essential role in opioid receptor-mediated responses in the neurocardiovascular system ([@B31]-[@B33]).
We also examined the effect of low-dose morphine on angiogenesis using *in vivo* Matrigel assays. Our result shows that 2K1C hypertension impaired angiogenesis and low-dose morphine promote it. Some studies showed that hypertension is associated with several vascular abnormalities including vascular rarefaction and endothelial dysfunction ([@B34]-[@B36]).
The literatures are mixed concerning the mechanism of morphine effects on angiogenesis. Gupta *et al* reported that low dose morphine stimulates angiogenesis especially in tumor ([@B37]). They showed that morphine acts via NO to induce cell proliferation and they believe that morphine signaling and angiogenesis are similar to vascular endothelial growth factor (VEGF). Other studies have shown that the morphine induced mitogenic and survival signaling is comparable with the effect of VEGF on the endothelium ([@B38]-[@B40]).
Changes of angiogenesis during consumption of morphine have been demonstrated in several experimental studies and the results are controversial. Some studies have shown that high-dose morphine reduces blood vessel proliferation ([@B41]) and increased production of superoxide anions in endothelial cells ([@B42]). Roy *et al* demonstrated that morphine inhibits VEGF during hypoxic condition in a dose-dependent fashion ([@B43]). Chen-Fuh Lam *et al* demonstrated that high-dose morphine was associated with impaired angiogenesis and mobilization of progenitor endothelial cells ([@B12]). We suggest that morphine is cytotoxic to endothelial cells at high concentrations. Therefore, the inhibition of angiogenesis, which observer by others may be due to high concentration of morphine and its cytotoxic effect.
Conclusion
==========
Previously, little was known about the effect of low-dose morphine on the blood pressure and angiogenesis in hypertensive subjects. Based on our data, low-dose morphine prevents induction of hypertension and stimulates angiogenesis in two-kidney one clip hypertensive rats probably via NO pathways. The cardioprotective and proangiogenic activities of low-dose morphine that are shown here might have implications for its therapeutic application in cardiovascular medicine.
This work has been done by the grant (no: 9/20/390) of Rafsanjan University of Medical Sciences. The authors declare that they have no conflict of interest.
| {
"pile_set_name": "PubMed Central"
} |
Sir,
Nontuberculous mycobacterial (NTM) infections involve the musculoskeletal system in approximately 5--10% of the patients.\[[@ref1]\] The most common manifestation of NTM infections is osteomyelitis. However, swelling and lytic lesions of the affected bone are rarely seen.\[[@ref2]\]
A 19-year-old male presented with pain and swelling of the right great toe since three months. No punctured wound was reported. Past history and family history were not contributory. A plain roentgenogram of the right foot revealed \[[Figure 1](#F1){ref-type="fig"}\] a lytic lesion at the base of the first metatarsal bone, with arthritic changes of the first metatarsophalangeal joint, surrounding soft tissue swelling, and erosion of the base of the first phalanx. The patient was empirically started with a combination of ceftriaxone 1 g/day and netilmicin 150 mg twice/day, administered intravenously for 15 days, and subsequently an oral therapy of ciprofloxacin 250 mg twice/day for one month with regular follow-up. However, the symptoms continued to aggravate and did not show any improvement. His erythrocyte sedimentation rate was 30 mm/hour. The chest roentgenogram was normal. An Enzyme-Linked Immunosorbent Assay test for human immunodeficiency virus (HIV) detection was negative. A surgical incision and drainage procedure was performed. The debrided synovial tissue was submitted for histopathology and microbiology examinations. The microscopic examination revealed multiple well-formed, epithelioid granulomas, without caseous necrosis \[[Figure 2](#F2){ref-type="fig"}\]. The Ziehl Neelsen stain was negative for acid fast bacilli. The culture for mycobacterium tuberculosis was negative. A tuberculin skin test was negative. The Mycobacterium COMBO test for detection of IgM antibodies against two highly purified antigens, derived from the Mycobacterium, namely, the cell wall and 38kDa antigens, was elevated (1.03 index value). A repeat culture from the lesion was performed, which identified the mycobacterium avium complex (MAC) species. A final diagnosis of NTM arthritis of the right first metatarsophalangeal joint, caused by MAC species, was made based on the correlation of the clinical features, and the radiology, histopathology, and microbiology findings. The patient received a one-year course of Azithromycin 1000 mg/day, Ethambutol 15 mg/kg/day, rifabutin 150 mg/day, along with indomethacin, which resulted in resolution of the lesions in the patient, with follow-up.
{#F1}
{#F2}
Unlike *M. tuberculosis*, NTM is not transmitted from person to person.\[[@ref1][@ref2]\] Most osseous infections are caused by *Mycobacterium kansasii* and *Mycobacterium scrofulaceum*.\[[@ref3]\] With NTM infection, the onset of nonspecific symptoms is indolent and usually includes local pain and swelling, joint stiffness, low-grade fever, sweats, chills, anorexia, malaise, and weight loss. The triad of Phemister, consisting of osteoporosis, peripheral marginal erosions, and slowly progressing destruction of the articular cartilage, characterizes mycobacterial arthritis.\[[@ref4]\] On account of their infrequent occurrence and difficulty in identification, with a lack of specificity of imaging findings, a heightened clinical suspicion of slow-growing nonchromogenic mycobacterial species in cases of arthritis is needed, when a routine bacterial culture or histopathological findings do not readily identify an organism.\[[@ref5]\] The prescribed management in the setting of NTM infection is the combination of different antituberculous drugs and antibiotics, along with surgical drainage, which was administered to our patient, resulting in a favorable outcome.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-sensors-19-05241}
===============
As one of the information acquisition means, aerial optoelectronic platform has the advantages of flexibility, real-time, and detailed information. It plays an important role not only in civil surveying and mapping, but also in military applications, thus is the all around the world research focus. Since the temperature environment on the ground is quite different from the working environment in the air, changes in the external environment cause the aerial remote sensors to suffer from adverse conditions, resulting in inoperability or degraded performance.
The purpose of thermal control design is to determine which thermal control measures are taken in specific part of the aerial optoelectronic platform, how to allocate active thermal control resources, and how to control the internal and external heat exchange processes of the system, so that the temperature level and temperature gradient can meet the temperature index requirements in order to guarantee image quality. At present, many scholars have carried out a large number of theoretical studies and thermal equilibrium tests on the thermal analysis and thermal design of space optical systems \[[@B1-sensors-19-05241],[@B2-sensors-19-05241],[@B3-sensors-19-05241],[@B4-sensors-19-05241],[@B5-sensors-19-05241],[@B6-sensors-19-05241],[@B7-sensors-19-05241],[@B8-sensors-19-05241]\], and the technology is relatively mature. However, there is not much research on the thermal control design of aerial optoelectronic platforms, and most of them focus on individual camera. Jongeun Choi et al. \[[@B9-sensors-19-05241]\] presented the design and control of a thermal stabilizing system for an optomechanical uncooled infrared imaging camera. Yang Wengang et al. \[[@B10-sensors-19-05241]\] proposed the thermal control index of a high-resolution space camera based on the wave phase difference distribution theory. On this basis, the thermal control technology was studied, and special thermal control measures were taken for the charged-coupled device (CCD). Fan Yue et al. \[[@B11-sensors-19-05241]\] established the heat transfer model and thermal resistance network model of the lens and window components to optimize the heating power. Liu Weiyi et al. \[[@B12-sensors-19-05241]\] studied the effects of temperature uniformity and temperature gradient on the imaging quality of aerial cameras, and an improved thermal control scheme was proposed based on the structural characteristics of the aerial camera, which effectively reduced the temperature gradient and improved the imaging quality. In addition, in Ref. \[[@B13-sensors-19-05241]\], Liu Weiyi et al. focused on a thermal control strategy considering the contradiction between a low power consumption and high modulation transfer function (MTF) requirement. Liu Fuhe et al. \[[@B14-sensors-19-05241]\] analyze the different impacts of temperature changes on the imaging resolution, and proposed an active thermal control method to improve the imaging quality of the sensor in a low-temperature environment.
Different from the previous thermal control research on aerial loads, the research object of this paper is not a single camera, but the whole optoelectronic platform, involving more components and more complicated thermal control methods are needed. In addition, due to the usage of a Cassegrain-reflecting objective lens, using active thermal control method with specific thermal control strategy to ensure the imaging quality of primary and secondary mirror is the focus. In this paper, on the basis of an aerial optoelectronic platform, the thermal control index was analyzed to ensure high-resolution image quality. Subsequently, the implementation of the active thermal control strategy and the combined thermal control scheme were elaborated. Next, thermal analysis was conducted under the extreme conditions by using thermal analysis software, and the effectiveness and correctness of the proposed thermal control technology were verified by simulations and experiments.
2. Analysis of Heat Flow Outside Working Environment of Aerial Optoelectronic Platform {#sec2-sensors-19-05241}
======================================================================================
Generally, the environment in which the aerial optoelectronic platform works is mainly divided into two situations: one is the ambient temperature range while parking on the ground, which is −40 °C\~+60 °C; the other one is the air temperature (heat sink) range at aerial photography stage, which is −55 °C\~0 °C. High-resolution camera thermal control index is very demanding which must be carefully analyzed for the entire heat flow environment outside the aerial optoelectronic platform. In fact, the external heat flow environment is complex and changeable, not only being affected by alternating direct sunlight, earth reflection, and infrared radiation of the earth, but also by factors such as different seasons, in and out of the earth shadow, and degradation of the coating. Here, the analysis of solar radiation, infrared radiation of the earth, and the external heat flow reflected by the earth to sunlight is illustrated as follows.
2.1. Solar Radiation {#sec2dot1-sensors-19-05241}
--------------------
The temperature of the sun is about 6000 K, and the energy it radiates into space is about 3.94 × 10^24^ W. 99.99% of the radiant energy is in the range of wavelength 0.18\~40 μm, where the radiant energy of the band 0.15\~0.3 μm corresponds to the radiation by black body of 4500 K while the radiation of the band 0.3\~2.5 μm corresponds to that by the black body of 6000 K \[[@B15-sensors-19-05241]\]. For the solar system planets, the earth-to-sun distance changes within a year frequently. The solar constants (the radiation density of the sun projected onto the earth) are shown in [Table 1](#sensors-19-05241-t001){ref-type="table"} for the spring equinox, summer solstice, autumnal, and winter solstice.
At an altitude of 8 km, as the atmosphere becomes thinner, the actual solar radiation intensity over time varies from 400 to 1376 W/m^2^. Generally, it is considered that there is no direct solar radiation at night.
2.2. Earth Reflection {#sec2dot2-sensors-19-05241}
---------------------
After solar radiation enters the earth atmosphere system, most of the energy is absorbed and reflected in the troposphere and the surface of the earth. The part that is reflected is called the Earth reflection, which includes the scattering of atmospheric molecules, the diffuse reflection of clouds, diffuse and specular reflections of soil, rocks, vegetation, waters, ice, and snow on the earth's surface. According to the propagation characteristics of solar radiation in the atmosphere, Equation (1) is used to calculate the Earth reflection heat flow:$$q_{f} = \rho I_{0}\sin\theta$$ where $\rho$ is reflectivity of the earth, the value of which is 0.20\~0.46, $I_{0}$ is the total energy radiated by the sun to the earth, and $\theta$ is the angle of the earth reflection.
2.3. Infrared Radiation of Earth {#sec2dot3-sensors-19-05241}
--------------------------------
When the earth receives radiation from the sun, it continually radiates heat to the space simultaneously. The solar radiation that falls on the earth is mostly absorbed by the earth and the atmosphere. The energy converted into heat is radiated in the form of long waves, that is, the infrared heat radiation of the earth. It is generally assumed that the spatial distribution of the infrared radiation of the earth is diffuse reflection, which can be equivalent to blackbody radiation. Therefore, ground infrared radiant heat flux density may be calculated by Equation (2) $$q_{IR} = \varepsilon\sigma T_{s}^{4}$$ where *ε* is the emissivity of the earth's surface, *σ* is the Boltzmann constant, and $T$ is the surface temperature of the earth. Different surfaces have different emissivity, and the surface temperature will fluctuate due to the change of day and night thermal environment. Considering the attenuation effect of atmospheric absorption when the surface infrared radiation is emitted into the atmosphere, the infrared radiation heat flux density at different heights is:$$q_{h} = \tau_{IR}q_{IR}$$ where $\tau_{IR}$ is the infrared radiation atmospheric transmittance, and $q_{IR}$ is ground infrared radiant heat flux density calculated by Equation (2).
Because it is difficult to determine the parameters involved in Equation (3), it is generally hard to calculate the earth's thermal radiation accurately. The actual engineering problem could be calculated by the following equation roughly \[[@B16-sensors-19-05241]\]:$$q_{k} = \left( {1 - \rho} \right)S_{0}/4$$ where $S_{0}$ is e solar constant.
2.4. Convection {#sec2dot4-sensors-19-05241}
---------------
The convective heat flux density can be calculated using the following equation:$$q_{c} = h\cdot\mathsf{\Delta}T$$ where, $h$ is the convective heat transfer coefficient; $\mathsf{\Delta}T$ is the temperature difference between the outer surface and the atmosphere. Before calculating the average surface heat transfer coefficient of the fluid sweep ball, Nusselt number, $Nu$ needs to be calculated firstly using the following equation:$$Nu = 2 + \left( {0.4Re^{\frac{1}{2}} + 0.06Re^{\frac{2}{3}}} \right)Pr^{0.4}\left( \frac{\eta_{\infty}}{\eta_{w}} \right)^{1/4}$$ where $Nu$ is a dimensionless number, $Re$ is Reynolds number, $Pr$ is the Planck constant; and $\eta_{\infty}/\eta_{w}$ is the ratio of the aerodynamic viscosity to the aerodynamic viscosity at the spherical shell temperature. The speed of the optoelectronic platform with the aircraft is about 180 m/s, if the air temperature in the low temperature condition is −55 °C. After calculation, $Nu$ is 1915.
According to the definition of $Nu$, $$Nu = hl/\lambda$$ $h$ is the average convective heat transfer coefficient, $l$ is the characteristic length, which is the diameter of the shell, and $\lambda$ is the air thermal conductivity. $$h = \left( {Nu\cdot\lambda} \right)/l$$
Finally, the convective heat transfer coefficient of the outer surface of the spherical shell is *h* = 62.9 W/(m^2^·°C).
Since the spherical shell has good sealing performance, it can be considered that the convection heat transfer inside the system is completed in the form of natural convection heat transfer. There are many internal components inside, and the shape is complex. It may take a lot of work to solve the convective heat transfer coefficient of each component. Thus, it is generally believed that the natural convective heat transfer coefficient is in the range of 2 W/(m^2^·°C) to 0 W/(m^2^·°C). Here, it is estimated that the conctive heat transfer coefficient between the internal components of the spherical shell and the air inside the sphere is 5 W/(m^2^·°C).
3. Thermal Control Design of Aerial Optoelectronic Platform {#sec3-sensors-19-05241}
===========================================================
3.1. Structure of Aerial Optoelectronic Platform {#sec3dot1-sensors-19-05241}
------------------------------------------------
The entire optoelectronic platform system is 580 mm wide, 780 mm high, and weighs 110 kg. The structural material is aluminum alloy, and each component is electrically oxidized. The external structure of an aerial optoelectronic platform is shown in [Figure 1](#sensors-19-05241-f001){ref-type="fig"} and the optical components are shown in [Figure 2](#sensors-19-05241-f002){ref-type="fig"}, including the primary mirror, the secondary mirror, the primary mirror barrel, and the main frame. The photoelectric loads share a Cassegrain-reflecting objective lens, where the main mirror is parabolic, and the secondary mirror is hyperboloid. The spectroscopic plate material on the inner side of the primary mirror is Crystallite, and a transmissive sub-image system is connected behind each imaging surface to realize long focal length, large aperture, and dual band imaging. The main frame is the structure that carries photoelectric loads, thus must have sufficient stiffness and strength.
3.2. Thermal Control Index {#sec3dot2-sensors-19-05241}
--------------------------
Aiming at the special working environment of the aerial optoelectronic platform, in view of the initial temperature uncertainty, the special characteristics of the rapid change of the working environment temperature and the thermal stability required for the long focal length loads, effective thermal control measures are needed. Liu Weiyi \[[@B12-sensors-19-05241]\] analyzed the influence of temperature level and temperature gradient on the imaging ability of long focal length camera. The research showed that the system appeared large spherical aberration and coma when the temperature gradient changed. When the temperature level changed, the system had a large spherical aberration and relatively small coma. The effect of the spherical aberration may be partially compensated by the camera's focusing function; however, the coma cannot be compensated. Therefore, it is considered that the effect of the temperature gradient on the image quality was higher than the temperature level. The wave phase difference is an index for judging the imaging quality of the optical system. The smaller the wave phase difference, the better the imaging quality of the system. In this paper, root mean square value of the system-wide wave difference assigned is not greater than 0.11*λ* in the optical camera design. The thermal tolerance is assigned to half of the total wave aberration, that is 0.055*λ*. Subsequently, through the thermo-optical analysis, the influence curve of temperature gradient between primary and secondary mirrors on wave phase difference is obtained, which is shown in [Figure 3](#sensors-19-05241-f003){ref-type="fig"}. When the wave phase difference is smaller than 0.055*λ*, the axial temperature gradient between the primary and secondary mirrors should be less than 5 °C. In other words, only an axial temperature gradient is less than 5 °C to meet the wave phase difference that the system can accept, thus ensuring the image quality. Generally, the aerial optoelectronic platform system's working environment temperature ranges from −30 °C to 50 °C Increasing temperature of the loads helps to ensure a good working condition for the optical system. If the temperature of the working environment is too low, optomechanical structure may not work smoothly. At the same time, due to the limitation of the capacity of the heaters, taking into account the power that can be utilized by the thermal control unit, the temperature level of the primary mirror is set to −20 °C\~+50 °C.
In summary, the thermal control index of an aerial optoelectronic platform system is set as:(a)The temperature level of the primary mirror is expressed by the average temperature of three temperature measuring points $T_{avg} = {\left( {T_{ZJ1} + T_{ZJ2} + T_{ZJ3}} \right)/3}$ is within the range of −20 °C\~+50 °C;(b)The temperature gradient between the primary and secondary mirrors is lower than 5 °C.
Based on the analysis of the thermal control index requirements and the internal and external thermal environment of the system, an active-passive combined control technology is studied.
3.3. Passive Thermal Insulation Measures {#sec3dot3-sensors-19-05241}
----------------------------------------
### 3.3.1. Covering Heat Insulation Layer {#sec3dot3dot1-sensors-19-05241}
In order to reduce the influence of external environment changes on the temperature inside the system, the sensors and the shell are insulated. The heat insulating layer covering area is mainly divided into two parts: the inner surface of the spherical shell and the back surface of the secondary mirror heating cover (facing the secondary mirror).
The inner surface of the spherical shell is covered with a 6 mm polyurethane foam insulation layer (having good heat preservation performance, good waterproof performance, good bonding performance, and good aging resistance), except for the position of optical entrance and the position that inconvenient to be covered by the heat insulation layer. The insulation layer is attached to the inner side of the system with a double-sided aluminized polyester film, as shown in [Figure 4](#sensors-19-05241-f004){ref-type="fig"}. In order to achieve active thermal control, a heating cover with a thickness of 1 mm that has the same shape with the shell is added to facilitate the adhesion of the heating pieces, seen in [Figure 5](#sensors-19-05241-f005){ref-type="fig"}. It can be discovered in the figures that there is a plurality of 6 mm high cylindrical bosses on the shell with threaded holes. The shell heating cover has holes corresponding to the bosses. When installing, the shell heating cover is fixed by screws to the shell.
Similarly, in order to achieve active thermal control of the secondary mirror, a heating cover wrapped around the secondary mirror is designed. The back surface of the secondary mirror heating cover is covered with 6 mm thick polyurethane foam, and attached with a double-sided aluminized polyester film, as shown in [Figure 6](#sensors-19-05241-f006){ref-type="fig"}.
### 3.3.2. Blackening {#sec3dot3dot2-sensors-19-05241}
In order to ensure the uniform temperature of each component inside the system, blackening is performed at key parts. It is blackened by spraying black paint or black anodizing blackening treatment, and its emissivity is not less than 0.8. The blackened components are the main frame, optical element support components, the inner and outer surface of the primary mirror barrel and heating covers.
3.4. Active Thermal Control Measures {#sec3dot4-sensors-19-05241}
------------------------------------
### 3.4.1. Thermal Control Strategy {#sec3dot4dot1-sensors-19-05241}
The idea of active thermal control strategy is to measure the temperature of control points by temperature sensors, and then adjust the working state of the heater adhered on the heating cover, according to the temperature difference by a comparing with the target temperature, which makes the temperature of the control points reaches the target temperature. Here, the setting of the target temperature is divided into two cases depending on different areas: for the secondary mirror heating cover, in order to reduce the temperature gradient of the primary and secondary mirrors, the target temperature is set to the current average temperature of the control points on the primary mirror, as shown in [Figure 7](#sensors-19-05241-f007){ref-type="fig"} (The primary mirror diameter is 260 mm, if only one point is used to represent the temperature level of the primary mirror itself, the radial temperature uniformity of the primary mirror cannot be taken into account. Considering the area of the primary mirror involved in this paper, three points on the radial center index circle within the range of the aperture diameter are selected, which are evenly distributed on the circumference. Taking the average temperature of these three points, the temperature of the primary mirror is characterized.); for other areas, the target temperature is the current average temperature of the primary mirror, but the target temperature range has to be controlled within −10 °C\~30 °C. That means if the average temperature of the primary mirror is less than −10 °C, the target temperature is set to −10 °C. Similarly, if it is greater than 30 °C, the target temperature is set to 30 °C. This setting mode ensures the temperature inside the entire system is uniform and the working environment temperature for all loads more comfortable.
Switch type is adopted as temperature control mode for all the heating areas. The sensor used is DS18B20, a kind of digital sensor, which has the advantages of small size, strong anti-interference ability, and high precision. The heating piece is made of electric heating piece of 125 type polyimide film attached to the outer surface of the heating cover. The heating areas are controlled according to the switch type, of which control threshold is the target temperature ±0.1 °C. If the measure temperature is higher than the target temperature of 0.1 °C, the heating area is powered off. In contrast, if below the target temperature 0.1 °C, the heating area is heated. The control period of heating area is no more than 1 s.
### 3.4.2. Implementation of Active Thermal Control Measures {#sec3dot4dot2-sensors-19-05241}
In order to realize the thermal control strategy, heating cover is added. In addition to the shell heating cover and the secondary mirror heating cover mentioned above, the primary mirror heating cover is also designed on the back side of the primary mirror. The shell heating cover has a total of 18 heating areas, 4 on front, 4 on top, 4 on bottom and 6 on rear. The black dotted frames in [Figure 8](#sensors-19-05241-f008){ref-type="fig"} denote heating areas on one side of the heating cover, and red strips denote the heating pieces. Heating pieces on the other side are symmetrically arranged. The purpose of this arrangement is to reduce heat leakage of the front and rear shell buckle lines. One heating area is provided on the back of the secondary mirror, as shown in [Figure 9](#sensors-19-05241-f009){ref-type="fig"}a, which may reduce the temperature gradient between the primary and secondary mirrors and increase the adaptability of the optical system to the environment. A total of three heating areas are provided on the primary mirror heating cover, as shown in [Figure 9](#sensors-19-05241-f009){ref-type="fig"}b, to ensure temperature uniformity. One heating area is arranged at each end of the main frame, where one-side heating area is shown in [Figure 10](#sensors-19-05241-f010){ref-type="fig"}. These two heating areas may reduce the leakage caused by strong convective heat transfer between the external low temperature environment and the shafts at both ends of the main frame. In summary, the entire system has 24 heating areas totally, and each heating area consists of a heating circuit. Thermal design power for different heating areas is shown in [Table 2](#sensors-19-05241-t002){ref-type="table"}, where numbers in the first column denote heating areas in [Figure 8](#sensors-19-05241-f008){ref-type="fig"}, [Figure 9](#sensors-19-05241-f009){ref-type="fig"} and [Figure 10](#sensors-19-05241-f010){ref-type="fig"}. It is worth to be noticed that the design power of symmetrical heating areas are equal.
4. Thermal Control Analysis and Results {#sec4-sensors-19-05241}
=======================================
Thermal analysis modeling and solving are done using the NX SST module. The thermal analysis model of the aerial optoelectronic platform system is shown in [Figure 11](#sensors-19-05241-f011){ref-type="fig"}. There are 9459 grid cells, 10,693 nodes, and 136 thermally coupled heat transfer channels being used in total. The main structural surface treatment and surface properties involved in the simulation process are shown in [Table 3](#sensors-19-05241-t003){ref-type="table"}.
4.1. Thermal Analysis Conditions {#sec4dot1-sensors-19-05241}
--------------------------------
According to the external heat flow, coating properties, sensor working mode, etc., the transient thermal analysis under extreme low temperature conditions is conducted, and the flight time is set at night, at which time the external heat flow has minimal influence, and the outer surface temperature of the shell is regarded to be equal to the ambient temperature in simulation. Combined with the heat flow analysis described in the second section, the specific operating conditions are shown in [Table 4](#sensors-19-05241-t004){ref-type="table"}. The thermal analysis is conducted under the same operating conditions with active thermal control measure on and off, respectively.
4.2. Thermal Analysis Results {#sec4dot2-sensors-19-05241}
-----------------------------
[Figure 12](#sensors-19-05241-f012){ref-type="fig"} shows thermal analysis results of the primary mirror temperature under above mentioned operating condition with and without active thermal control measures respectively. It can be seen from the figure that when there is no active thermal control measures, the primary mirror temperature is lower than −30 °C, which doesn't satisfy the temperature index that the temperature should be kept within −20 °C\~+50 °C. In the case with active thermal control measures, the primary mirror temperature to −12.4 °C, meeting the temperature index requirements. Compared to uncontrolled conditions, the primary mirror temperature is increased by approximately 25 °C, making the primary mirror working environment more comfortable.
[Figure 13](#sensors-19-05241-f013){ref-type="fig"} shows the temperature results of the primary and secondary mirrors. It can be found that the colors of the two mirrors are in the same range, respectively, indicating that the temperature uniformity is pretty good. In the case of no active thermal control, the temperature gradient of the primary and secondary mirrors is less than 2 °C. However, the temperature gradient is greater than 5 °C when there are active control measures. This is because it is unable to realize the temperature control strategy that the target temperature of the secondary mirror changes with the temperature of the primary mirror using the thermal analysis software (the thermal heating software sets the target temperature of the secondary mirror heating cover to a fixed temperature of 10 °C). In order to further assess whether the temperature gradient of the primary and secondary mirrors meets the thermal control index requirements with active thermal control measures, the test is necessary.
5. Thermal Control Test and Results {#sec5-sensors-19-05241}
===================================
In order to further verify whether the thermal control method proposed makes the temperature of the primary and secondary mirrors meet the index requirements, the tests are carried out under low temperature conditions with and without active thermal control measures, respectively. The atmosphere pressure is 30 kPa at the actual flight altitude of the aerial optoelectronic platform system. Due to the limitations of the test conditions, the pressure of chamber of high-low temperature and pressure cannot be reduced to 30 kPa. Considering that the sealing condition of the optoelectronic platform system is good enough, the test under normal pressure will enhance the influence of convective heat transfer and reduce the ability of the thermal control system to control the temperature through radiation heat transfer. Thus, the temperature control capability of the thermal control system under 30 kPa pressure conditions is better than that under normal pressure conditions. In other words, if the results from the test carried out under normal pressure conditions meet the temperature control index requirements, it can be considered that the test results at the pressure of 30 kPa must meet the index requirements as well. [Figure 14](#sensors-19-05241-f014){ref-type="fig"} is the photograph of the test scenario.
5.1. Test Results {#sec5dot1-sensors-19-05241}
-----------------
The temperature curves of the main measure points at low temperature test without active thermal control measurements are shown as [Figure 15](#sensors-19-05241-f015){ref-type="fig"}. Since the entire optoelectronic platform has no heat source, the temperature tends to decrease when the environment temperature is −55 °C. During the temperature drop process, there is a maximum temperature difference between the primary mirror and the secondary mirror. The temperature difference gradually decreases with time, and the whole system tends to be a low temperature steady state. Finally, the temperature of the primary mirror is stable at around −34 °C, which doesn't meet the thermal control index requirements is the temperature level of the primary mirror should be between −20 °C\~+50 °C. It can be determined from the curves that the maximum temperature gradient of the optical system is 15.7 °C. Although the temperature gradient decreases gradually over time, it is still higher than the 5 °C required by the thermal control index.
[Figure 16](#sensors-19-05241-f016){ref-type="fig"} shows the temperature curves of the main measure points at low temperature test with active thermal control measurements. It can be seen from the figure that after the active thermal control is turned on, although the environment temperature is at a low temperature of −55 °C, the temperature of the primary mirror is stable at about 20 °C, meeting thermal control index requirements. Further, the temperature gradient of the primary and secondary mirrors is less than 5 °C, which also meet the index requirements. And the range of temperature changes is significantly reduced, basically in equilibrium.
The test results are compared with the simulation results obtained in [Section 4](#sec4-sensors-19-05241){ref-type="sec"}, as shown in [Table 5](#sensors-19-05241-t005){ref-type="table"}. It is obvious that there is a certain gap between these two results. The reason is that the input conditions of the working conditions during the test and simulation are not completely consistent due to the limited test conditions, and the simulation software cannot realize the active control strategy accurately. The comparison results demonstrate that the verification of the proposed technology still relies mainly on the experimental means. Although the simulation analysis may obtain the temperature change trend qualitatively, and play an auxiliary role, it needs to be further improved in accuracy.
In combination with the simulation and test results, it is necessary to adopt active thermal control to ensure that the optical system meets the thermal control index requirements. Active thermal control makes the aerial optoelectronic platform system more adaptable to the effects of wide initial temperature and rapid changes of ambient temperature when working at different regions and different time periods.
5.2. Actual Imaging Quality {#sec5dot2-sensors-19-05241}
---------------------------
[Figure 17](#sensors-19-05241-f017){ref-type="fig"} shows the images of the target in the simulated high altitude environment, without active thermal control and with active thermal control. The imaging quality is much better when there are active thermal control measurements, which shows active thermal control effective.
6. Conclusions {#sec6-sensors-19-05241}
==============
Based on the complex and variable working environment of the aerial optoelectronic platform, considering the structure itself, an active-passive combined thermal control scheme is proposed. Passive thermal control measures mainly use insulation materials. Based on effective control strategy, active thermal control measures utilize heating pieces adhered on the heating cover to achieve active electric heating control technology, especially local elaborate heating for the primary and secondary mirrors. After thermal analysis and experiment in extreme low temperature conditions, it is verified that the proposed combined method can improve the temperature level of the primary mirror, reduce the temperature gradient of the primary and secondary mirrors, and improve imaging quality effectively. It has certain guidance and references for the precision thermal control of aerial optoelectronic platform systems. In the future work, the active thermal control algorithm should be studied further to achieve precise thermal control.
The authors are thankful for the support from the Key Laboratory of Airborne Optical Imaging and Measurement, Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences.
Conceptualization: Z.C.; data curation: L.S.; formal analysis: F.L.; funding acquisition: L.L.; investigation: Z.C.; methodology: X.L. and Q.L.; project administration: Z.C.; resources: X.L.; software: Z.C. and R.H.; supervision: Z.C.; validation: L.S.; visualization: F.L.; writing---original draft: Z.C.; writing---review & editing: Z.C.
This research was funded by the National Natural Science Foundation of China, grant number 61905240, and the National Natural Science Foundation of China, grant number 61675202.
The authors declare no conflict of interest.
{#sensors-19-05241-f001}
{#sensors-19-05241-f002}
{#sensors-19-05241-f003}
{#sensors-19-05241-f004}
{#sensors-19-05241-f005}
{#sensors-19-05241-f006}
{#sensors-19-05241-f007}
{#sensors-19-05241-f008}
{#sensors-19-05241-f009}
{#sensors-19-05241-f010}
{#sensors-19-05241-f011}
{#sensors-19-05241-f012}
{#sensors-19-05241-f013}
{#sensors-19-05241-f014}
{#sensors-19-05241-f015}
{#sensors-19-05241-f016}
{#sensors-19-05241-f017}
sensors-19-05241-t001_Table 1
######
Solar constant.
Time Spring Equinox Summer Solstice Autumnal Winter Solstice
----------------------- ---------------- ----------------- ---------- -----------------
Solar constant W/m^2^ 1378 1323 1357 1412
sensors-19-05241-t002_Table 2
######
Thermal design power for different heating areas.
Nr. Location Thermal Design Power/W
----- -------------------------------- ------------------------
1 Front heating area 10
2 Front heating area 14
3 Top heating area 16
4 Top heating area 40
5 Bottom heating area 20
6 Bottom heating area 40
7 Rear heating area 30
8 Rear heating area 30
9 Rear heating area 15
10 Secondary mirror heating cover 10
11 Primary mirror heating cover 5
12 Primary mirror heating cover 5
13 Primary mirror heating cover 5
14 Main frame 130
sensors-19-05241-t003_Table 3
######
Main structural surface treatment and surface properties.
Structural Components Surface Treatment Status Infrared Emissivity Solar Absorption Rate
---------------------------------- ---------------------------------------------- --------------------- -----------------------
Inner surface of spherical shell Paste double-sided aluminized polyester film 0.1 0.1
outer surface of heating cover Paste double-sided aluminized polyester film 0.1 0.1
inner surface of heating cover blacking 0.8 0.8
Primary mirror surface coating 0.1 0.1
Secondary mirror surface coating 0.1 0.1
Main frame blacking 0.8 0.8
Main working loads blacking 0.8 0.8
sensors-19-05241-t004_Table 4
######
Operating conditions.
Environment Conditions Low Temperature without Active Thermal Control
------------------------------------------ ------------------------------------------------
Flight altitude (m) 8000
Atmospheric pressure (kPa) 30
Atmospheric density (kg/m^3^) 0.5
Solar radiation (W/m^2^) 0 Night flight
Infrared Radiation of the earth (W/m^2^) 270
Earth Reflection (W/m^2^) 0 Night flight
Initial temperature (°C) −40
Environment temperature (°C) −4\~−55
Working duration 3 h
sensors-19-05241-t005_Table 5
######
Comparison of test results and simulation results.
Operation Condition Low Temperature Without Active Thermal Control Low Temperature with Active Thermal Control
------------------------------------------------------------ ------------------------------------------------ --------------------------------------------- ------ ------------
Method test simulation test simulation
Temperature of primary mirror (°C) −34 −37.8 20 −12.4
Temperature gradient of primary and secondary mirrors (°C) 6.9 0.4 1.8 14.5
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#s0005}
===============
A driving force to maintain and enhance the competitiveness of food industry is technological innovation. The development of innovations capable of improving wine quality and processing sustainability are increasingly attracting for wineries. Pulsed electric field technology (PEF) is an innovative processing technology with potential applications in wineries to accelerate and/or increase the extraction of phenolic compounds during the maceration-fermentation step of red winemaking ([@bb0085]). Red wine quality is strongly affected by the phenolic compounds which are responsible for the sensory characteristics such as colour and taste, aging properties, and antioxidant properties which may play a positive role in human health ([@bb0095]).
Application of PEF technology for improving extraction of polyphenols results particularly interesting in those vintages in which the concentration of these compounds in the grape skins is poor or the polyphenol extraction is difficult. Another advantage of PEF treatments is that it permits reducing the time of contact of the grape pomace with the fermenting must to obtain a given polyphenol concentration in the wine, thus increasing the productivity of the winery ([@bb0110]).
During PEF processing, a mix of must, skin and seeds obtained after de-stemming and lightly mechanically crushing the whole bunches is subjected to short pulses (μs) of high voltage (kV). When exposed to a sufficiently strong electric field, the cell membrane of grape skins undergoes electroporation, which renders it permeable to molecules such as polyphenols that are otherwise unable to cross it ([@bb0015]).
PEF application to improve extraction of phenolic compounds during red winemaking has been deeply investigated in different grape varieties, using moderate electric fields (0.5--1 kV/cm) and treatment times in the range of 40--100 ms ([@bb0025], [@bb0030], [@bb0040]) or higher electric fields (1--10 kV/cm) and treatment times in the range of 100 μs ([@bb0075], [@bb0080], [@bb0035]). These studies have demonstrated that electroporation of the cells of the grape skins by PEF depends on the grape varieties but also that the physicochemical composition of the grape, that may differ for different vintages or even during the harvesting period, could influence the PEF effect.
The release of polyphenols during the maceration-fermentation step and the conversion of sugars of the must into ethanol is a slow process that takes several days. Therefore, the effectivity of the PEF treatment in the improvement of the polyphenol content of the wine only can be observed some days after the treatment. On the other hand, if the electroporation of the grape skins is not intense enough, effects observed during the first days of maceration may disappear at the end of the maceration-fermentation step. Therefore, evaluating in a short period of time the electroporation degree of the grape skins in terms of improvement in the polyphenol release could result of interest to define the required PEF treatment conditions to maximize polyphenol extraction minimizing energetic costs. Different methods such as microscopic observations of cell integrity, measurement of the liquid release, evaluation of the conductivity of the exuded liquid, analysis of textural parameters of treated tissues or impedance measurement have been proposed to assess electroporation of plant cells ([@bb0035], [@bb0115]). However it has not been demonstrated that these methods were effective estimating the effect of PEF when subsequent processing steps are required.
The aim of this study was to develop a procedure to establish the PEF parameters that cause enough permeabilization in the skin cells of different grape varieties to obtain a significant improvement in the vinification process in terms of increment on the polyphenol content or reduction of maceration time. The influence of the electric field intensity and pulse width on the improvement of polyphenol extraction has been investigated in three grape varieties, Syrah, Tempranillo and Grenache.
2. Material and methods {#s0010}
=======================
2.1. Grape samples {#s0015}
------------------
Grapes from *Vitis vinifera* L. var. Syrah, Tempranillo and Grenache, from the certified origin Campo de Borja (Aragon, north-east Spain), were harvested from the 2015 vintage. The grapes were manually harvested in good sanitary conditions during their optimal ripening stage. In the case of Grenache grapes, they were harvested during their optimal ripening stage (Grenache 1) and two weeks later (Grenache 2).
Total acidity, pH, °Brix, and total phenols were analyzed in the must ([@bb0100]). Phenols at pH 3.2 and 1.0 and extractability of phenols were obtained by macerating a grape homogenate (Ika labortechnik A10, Staufen, Germany) for 4 h at two different pH values (3.2 and 1.0), according to the method described by [@bb0125].
The grapes were transported in 20-kg boxes from the field to the laboratory for the subsequent experiments. Then, the grapes were destemmed and crushed, and the grape juice was separated for treating the skins by PEF. The proportion of grape juice and pomace was measured in order to maintain this proportion during the vinifications.
2.2. PEF equipment {#s0020}
------------------
The PEF unit used in this investigation (EPULSUS®-PM1-10, Energy Pulse System, Lisbon, Portugal) is a Marx generator that can apply monopolar square waveform pulses with a frequency up to 200 Hz. The maximum output voltage and current were 10 kV and 180 A, respectively. The pulse width can be modified, ranging from 5 to 100 μs. It is a compact PEF generator with 800 × 600 × 400 mm as dimension and only 80 kg of weight that can be used both at lab and pilot plant scale, and can be controlled directly from its touchscreen. The actual voltage, current and pulse duration were measured using a high voltage probe (Tektronix, P6015A, Wilsonville, OR, USA) and a current probe (Stangenes Industries Inc. Palo Alto, CA, USA), respectively, connected to an oscilloscope (Tektronix, TDS 220,Wilsonville, OR, USA).
The PEF treatments were applied to a parallel plate electrodes treatment chamber of 19.6 cm^2^ of electrode area and 2 cm of gap. Batches of 50 g of grapes were treated at 1, 3 and 5 kV/cm for 100 μs of total treatment time, and the corresponding number of pulses of 5, 20, 50 and 100 μs of pulse width were applied. The specific energy applied was 0.14, 1.26 and 3.5 kJ/kg at electric fields of 1, 3 and 5 kV/cm respectively.
2.3. Extraction in ethanol solution {#s0025}
-----------------------------------
50 g of grapes of untreated and PEF treated grapes were placed in 250 mL Erlenmeyer flasks containing 100 mL of a 30% v/v ethanol solution, at room temperature without agitation. Previous results indicated that a 30% v/v of ethanol solution showed the highest difference of absorbance at 280, 420, 520 and 620 nm after 2 h of extraction at room temperature between the untreated grapes and PEF treated grapes (data not shown). Samples of 1 mL were taken during 120 min and centrifuged at 8640*g* for 90 s (Minispin®plus, Eppendorf, Hamburg, Germany). The absorbance was measured at 280 nm for the total polyphenol index (Eq. [(1)](#fo0005){ref-type="disp-formula"}) and 420, 520 and 620 for the colour index (Eq. [(2)](#fo0010){ref-type="disp-formula"}) in a spectrophotometer (Unicam UV500, Unicam Limited, Cambridge, UK) according to Glories\' methods ([@bb0050], [@bb0055]):$$\mathit{TPI} = \mathit{Abs}_{280} \times {DF}$$$$\mathit{CI} = \left( {\mathit{Abs}_{420} + {Abs}_{520} + {Abs}_{620}} \right) \times {DF}$$where *TPI* is the total polyphenol index, *CI* is the colour index, *Abs*~*λ*~ is the absorbance at the corresponding length wave (*λ*) of 280, 420, 520 and 620 nm and *DF* is the dilution factor.
Each experiment was carried out by triplicate.
2.4. Vinifications {#s0030}
------------------
Laboratory fermentations were performed in 500 mL Erlenmeyer flasks that were opened to the atmosphere by a small orifice of 1 mm of diameter. 100 g of grapes were mixed with the corresponding proportion of grape juice. The fermentation was carried out at by a commercial preparation of yeast of *Saccharomyces cerevisiae* (Lalvin, Ontario, Canada) at an initial concentration of 10^6^ CFU/mL. The fermentation temperature was controlled at 25 °C and the weight of the samples was monitored once a day, until the measured weight was constant at least two consecutive days. The weight loss is related to the loss of CO~2~ during the fermentation. The duration of the fermentation-maceration step was 10 days for all the grapes. Every day a sample of 1 mL was taken and the TPI and CI were measured as described previously.
Each vinification was carried out by triplicate.
2.5. Statistical data treatment {#s0035}
-------------------------------
The results represent the mean ± standard error of the mean, of the analysis performed on the three flasks containing samples receiving the same treatment. A *t*-test was conducted to assess significant differences between vinifications conducted with untreated and PEF-treated gapes along maceration--fermentation time. The differences were considered significant at p \< 0.05.
3. Results and discussion {#s0040}
=========================
3.1. Grape characterization {#s0045}
---------------------------
Experiments to define PEF processing parameters that cause the required electroporation in the grape skin cell to obtain a significant improvement polyphenol release during vinification were conducted with three different grape varieties and, in the case of Grenache, with grapes harvested in two different moments. It is well known that the release of polyphenols during fermentation-maceration is influenced not only by the total polyphenol content in the grapes, that may depend on factors affecting the berry development such as soil, geographical location and weather conditions, but also on the cell wall structure and morphology of the skin cells that are intrinsic characteristics determined by the grape variety ([@bb0065], [@bb0070]).
[Table 1](#t0005){ref-type="table"} shows the results of the analysis carried out on the grapes used in this investigation. Acidity was higher in Syrah than Tempranillo and Grenache and the total sugar content expressed as °Brix and the pH was similar in Syrah, Tempranillo and Grenache 1. However the °Brix of the Grenache 2 grapes was much higher because they were harvested two weeks later. In order to evaluate the total content of polyphenols the grapes were macerated for 4 h at pH 1.0 and 3.2. At pH 1.0 a complete degradation of the membranes of cell skin and vacuoles occurs facilitating the complete release of phenolic compounds while at pH 3.2 the extraction was conducted in similar conditions as those occurring during maceration in vinification process. The TPI determined at pH 1.0 was higher for Tempranillo than for Syrah and Grenache 1 grapes while this index was higher for the Grenache 2 grapes, harvested later. Similar results were obtained for the TPI determined at pH 3.2. Tempranillo grapes had the highest TPI value and the TPI for the Grenache 2 grapes was higher than Syrah and Grenache 1 grapes, harvested before. No significant differences were observed between the conductivity of the most of the different varieties or of the same variety harvested in different moments (1.5 ± 0.1 mS/cm). Conductivity of the must affects the performance of PEF modulators by influencing the electrical resistance of the treatment chamber. Energy requirements to generate a given electric field strength are lower as lower is the resistance of the treatment chamber. The conductivity of the grape must as compared with other food is quite low indicating that specific energetic requirements for this application may be lower than for others ([@bb0060])
3.2. Influence of PEF treatments of different electric field strength and pulse width on the extraction of polyphenols from different grapes varieties in ethanol solution {#s0050}
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------
[Fig. 1](#f0005){ref-type="fig"}A and B shows the influence of the electric field strength and pulse width on the polyphenol index and the colour intensity of the solution of ethanol (30%) containing PEF pre-treated grapes of the three varieties after 2 h of extraction. The main procedures to measure total polyphenols in wineries are the Folin-Ciocalteu index (FCI) and the Total Polyphenol Index (TPI). Both are spectrophotometric methods, however, whereas the FCI requires the use of reagents and the incubation of the sample for 30 min, the determination of the TPI simply consist in a direct measure of the absorbance of the sample at 280 nm - wavelength that corresponds to the maximum of absorbance of the benzene ring. As the ultimate goal of this paper is to set up a procedure to evaluate grape electroporation the faster procedure to measure polyphenols was chosen. On the other hand, the determination of the colour intensity of wine and must is other direct spectrophotometric measured conducted in the wineries consisting in the sum of 420, 520, 620 nm absorbance.
For the four grape varieties, it was observed that the application of a PEF treatment of 1 kV/cm of intensity did not increase de polyphenol extraction at any pulse width investigated. No differences were observed in the TPI for the treatments applied at different pulse width as compared with the control. This lack of effect of the PEF could be a consequence of the low total specific energy of the treatments applied at 1 kV/cm (0.14 kJ/kg). Other authors have observed improvement in polyphenol extraction during red winemaking at lower electric fields but applying pulses in the range of milliseconds with much higher specific energies (50 kJ/kg) ([@bb0040]). The electroporation of the grape skin cell by PEF treatments of 3 kV/cm increased the TPI as compared with the untreated grapes for the treatments applied at higher pulse widths. The TPI of the treated samples as compared with the control increased 30.1%, 13.8%, 19.1% and 33.3% for the treatments applied at 100 μs in the Syrah Tempranillo, Grenache 1 and Grenache 2 varieties respectively. For the three grape varieties, the pre-treatment of the grapes increased the TPI when the PEF treatment was applied using pulses higher than 5 μs at 5 kV/cm. For Syrah these treatments increased the polyphenol extraction between a 46% (5 pulses of 20 μs) to 70% (1 pulse of 100 μs), for Grenache 1 between a 38% (5 pulses of 20 μs) to 63% (1 pulse of 100 μs), Grenache 2 between a 37% (5 pulses of 20 μs) to 46% (1 pulse of 100 μs), and for Tempranillo variety between a 8% (5 pulses of 20 μs) to 24% (1 pulse of 100 μs). It was observed that although the total duration (100 μs) and as consequence the total specific energy (3.5 kJ/kg) of the different treatments was the same the electroporation was more effective by decreasing the number of pulses and increasing the pulse width.
The influence of the pulse width on electroporation caused by PEF of microbial and eukaryote cells at a fixed total treatment time has not been widely investigated. Some authors observed that at a constant quantity of the applied specific energy pulse width did not have a significant effect on microbial inactivation by PEF ([@bb0120], [@bb0135]) but others reporter higher microbial inactivation when pulses applied were wider ([@bb0090], [@bb0005]). However, in these studies higher inactivation could be consequence of the increment of the temperature of the treatment medium when wider pulses were applied ([@bb0130]). Microbial inactivation requires the application of higher specific energy (high electric field strengths and longer treatments) than electroporation of eukaryotes so higher increment of the temperature when wider pulses were applied could cause the higher efficacy of PEF. In the case of plant cells results obtained in this investigation are in contradiction with those reported by [@bb0045] that observed that the ion leakage from onion tissue decreased with decreasing pulse number and increasing pulse widths. However the higher polyphenol release from grape tissues when longer pulses were applied confirm observation of [@bb0020] that reported that samples of sugar beet and apple tissues exposed to the same PEF treatment time showed higher electroporation in terms of cell disintegration index when the pulse duration was wider. The higher efficacy of longer pulses has been related to the fact that an efficient electroporation of the cell membranes requires pulses of longer duration, as compared to the membrane charging time, in order to reach the maximum transmembrane voltage ([@bb0010]).
[Fig. 1](#f0005){ref-type="fig"}B shows that influence of the PEF treatment of different electric field strength and pulse width on the colour intensity (CI) of the extraction medium containing grapes of the three varieties follows a pattern similar to the TPI. No significant differences were observed in the CI for the treatments applied at 1 kV/cm as compared with the control. At 3 kV/cm the treatment increased the CI for the Syrah variety. Significant differences were observed between the control and the samples treated by pulses longer than 5 μs. However for the other two varieties, 3 kV/cm only increased significantly the CI using a pulse width of 100 μs for the Tempranillo variety. Similarly to the effect observed for the TPI, for the three varieties the treatment at 5 kV/cm significantly increased the CI especially when the treatments were applied using the wider pulses. For Syrah these treatments increased the CI between a 93%(5 pulses of 20 μs) to 121% (1 pulse of 100 μs), for Grenache 1 variety between a 40% (5 pulses of 20 μs) to 59% (1 pulse of 100 μs), for Grenache 2 variety between a 34% (5 pulses of 20 μs) to 40% (1 pulse of 100 μs) and for Tempranillo variety between a 24% (5 pulses of 20 μs) to 36% (1 pulse of 100 μs).
The correspondence observed between the TPI and CI could be related with the fact that the main compounds of the grape skins responsible of the CI are anthocyanins that are the pigments responsible of the colour of red grapes and represents one of the main phenolic compounds present into the cells of the grape skins.
3.3. Influence of PEF treatments of different electric field strength and pulse width on the extraction of polyphenols from different grapes varieties during vinification {#s0055}
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------
The evolution of the TPI of the fermenting must containing untreated grapes and PEF treated grapes under the same treatment conditions that in the extraction experiment in the ethanol solution (30%) for Syrah, Grenache1, Grenache 2 and Tempranillo varieties are shown in [Fig. 2](#f0010){ref-type="fig"}, [Fig. 3](#f0015){ref-type="fig"}, [Fig. 4](#f0020){ref-type="fig"}, [Fig. 5](#f0025){ref-type="fig"} respectively. In all cases, the maceration-fermentation process was extended for 10 days. After this period of time, the weight of the flask containing the fermenting must was constant indicating the end of the fermentation.
The evolution of the TPI along the maceration-fermentation process depended on the grape variety and on the intensity of the PEF treatment applied. For the Syrah variety, the maximum value for the TPI was achieved after 6 days of maceration fermentation for the untreated and PEF treated samples at 1 and 3 kV/cm. After this maximum, TPI slightly decreased in all cases. This fall of polyphenols at the end of the maceration-fermentation process is attributed to their attachment to other compounds and to their precipitation causing the decrease of free polyphenols in dissolution. While no significant differences were observed between the control and PEF treated samples at 1 kV/cm along the fermentation-maceration time, the TPI after 4 days of maceration was higher for the PEF treated samples at 3 kV/cm than for the control. However, differences in TPI between untreated and treated samples disappeared after 6 days of maceration-fermentation. In the case of the samples treated at 5 kV/cm, the TPI of the PEF treated samples was higher than for the control even after 6 days of maceration-fermentation. On the other hand, the TPI of the PEF samples after 4 days of maceration fermentation was similar or slightly higher than for the control after 6 days of maceration-fermentation. These results indicate that the PEF treatment could reduce the maceration time to obtain the highest phenolic concentration in the wine for 2 days. [Fig. 3](#f0015){ref-type="fig"} shows that the maximum value for TPI was achieved after 5 days of maceration-fermentation for the untreated and PEF treated samples at 1 and 3 kV/cm of Grenache 1 samples and then the value of TPI in the fermenting must was maintained practically constant until the end of the fermentation. Similarly to Syrah grapes, the treatment at 1 kV/cm was ineffective to significantly increase the TPI along maceration-fermentation. On the other hand the differences observed between untreated and treated samples at 3 kV/cm during the 3 first days of maceration fermentation disappeared after 5 days. In the case of the samples treated at 5 kV/cm, the PEF treatment was very effective for improving TPI of fermenting must during the first 3 days of maceration-fermentation but the differences between the control and PEF treated samples disappear after 5 days of maceration-fermentation. However, the maximum value for the TPI of the control that was obtained after 5 days of maceration fermentation was achieved after 2 days for the samples treated with pulses of 50 and 100 μs and after 3 days for the rest of the PEF treatments. Results obtained indicate that the pretreatment of the grapes of Grenache 1 by PEF could reduce maceration time between 2 and 3 days depending of the width of the pulses applied. Concerning the influence of PEF intensity and pulse width on polyphenol extraction from the Grenache grapes harvested later ([Fig. 4](#f0020){ref-type="fig"}), it was observed a slightly higher polyphenol release for the samples treated by PEF at 3 and 5 kV/cm until the first 4 days of maceration-fermentation. However, the differences in TPI observed between treated and untreated samples would not allow reducing maceration time for the samples PEF treated without affecting the TPI in the final wine. Finally, [Fig. 5](#f0025){ref-type="fig"} shows that for the Tempranillo variety, the PEF treatments applied to the grapes at different electric field strengths and pulse widths did not significantly increase TPI at any moment of maceration-fermentation process.
As it happened during the extraction experiments conducted in a 30% ethanol solution, the evolution of the colour index (CI) followed a similar trend than the evolution of the TPI in the must during the vinifications for all the grape varieties studied (data not shown).
3.4. Relationship between the PEF effect on the polyphenol release from different varieties of grapes in the ethanol solution and during fermentation-maceration {#s0060}
----------------------------------------------------------------------------------------------------------------------------------------------------------------
Results obtained regarding the extraction of polyphenols in the alcoholic solution for 2 h showed that the effect of PEF treatments in the improvement of the polyphenol release was much higher in Syrah and Grenache 1 grapes than in Grenache 2 and Tempranillo grapes. In all cases, the most significant effect of PEF was observed at the highest electric field strength and when longer pulses were applied. In Syrah and Grenache 1 grapes a treatment of 5 pulses of 20 μs at 5 kV/cm increased the TPI around 40--45% while when 1 pulse of 100 μs was applied the observed increment was around 60--70%. In the case of Grenache 2 and Tempranillo the highest increment in the TPI of the extraction medium was 46 and 25% respectively after applying 1 pulse of 100 μs at 5 kV/cm. When these results were compared with the total phenols measured in the grapes it was observed that the PEF treatment was more effective in that grapes in which the total polyphenol content obtained by macerating the grapes at pH 3.2 and 1.0 was lower. These results confirm previous observations that indicated that the PEF treatment might be more useful when the extraction of the phenolic compounds from the grape skins is hampered or its total polyphenol content is lower ([@bb0105]).
When results obtained in ethanol solution were compared with the polyphenol release during maceration-fermentation it was observed that the treatment was ineffective in those grapes (Grenache 2 and Tempranillo) in which the extraction in the alcohol solution was lower but effective in those grapes (Syrah and Grenache 1) and under those PEF treatment conditions in which the highest effect on polyphenol extraction was observed in the ethanol solution for 2 h. From a practical point of view the key advantage of the application of the PEF treatment in these grapes was the reduction of the duration of the maceration during vinification between 2 and 3 days. The efficiency of the transfer of polyphenols from grape skins to the fermenting must depends on the extent of cell degradation. In the first days of maceration-fermentation the extraction was more effective in the PEF treated samples that in the control samples because the electroporation of the cells facilitates the polyphenol release. However differences in the TPI in the fermenting must containing untreated and PEF treated grapes becomes smaller along maceration-fermentation because the alcohol produced during fermentation contributes to the disorganization of the cell envelopes of the grape skins facilitating the polyphenol release from the untreated grapes. Currently wineries are interested in reducing the duration of the time of contact of the grape skins with the fermenting must during maceration-fermentation. Grape skins represent a huge volume in the fermentation tanks, therefore, the removal of the grape skins increases the amount of fermenting must that can be stored in the tanks increasing the production capacity of the wineries without increasing the number of fermentation tanks.
As conclusion this investigation demonstrates that an extraction of polyphenols in a solution of ethanol (30%) for 2 h could be a procedure to know if the PEF technology is effective for improving extraction of polyphenols from the grapes that are arriving to the winery during vinification and to determine the most suitable PEF treatment conditions to obtain this objective. Other interesting observation from this research is the highest efficacy of PEF when treatments of the same duration are applied using longer pulses. Longer pulses permit reducing the number of pulses to be applied for a given treatment time. Therefore in a continuous process, where the flow processed is determined by the frequency applied by the PEF generator, it is possible to increase the processing capacity of the PEF installation.
This research was supported by the European Commission (635632-FieldFOOD-2020).
{#f0005}
{#f0010}
{#f0015}
{#f0020}
{#f0025}
######
Physicochemical characteristics of the grapes at harvesting time.
Table 1.
pH Total acidity (g/L) °Brix Phenols at pH 1 (OD 280 nm) Phenols at pH 3.2 (OD 280 nm) Extractability (%)
------------- ----- --------------------- ------- ----------------------------- ------------------------------- --------------------
Syrah 3.2 6.5 23.4 23.4 16.6 29.3
Tempranillo 3.2 5.1 22.7 32.9 28.0 15.0
Garnacha 1 3.2 5.1 23.8 20.6 17.7 13.8
Garnacha 2 3.5 4.8 31.3 28.0 22.3 20.5
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#sec1-1}
============
The serotonin is a well-known brain neurotransmitter but during the last years, a great interest has been noticed in its actions over the bone. The dynamic of understanding the serotonin signaling has changed since 5-hydroxytryptamine with gut origin was found to regulate the bone loss via LRP-5 \[**[@R1]**,**[@R2]**\]. In vitro studies revealed that human osteoblasts and osteoclasts express tryptophan hydroxylase type 1, serotonin transporter and serotonin receptors (type 2A only in osteoblasts, type 1B in both osteoclasts and osteoblasts, and type 2B in precursors and mature osteoclasts), while selective serotonin reuptake inhibitors (SSRI), induce apoptosis of both types of cells \[**[@R3]**\]. Moreover, studies in female mice pointed bone microarchitecture changes of the distal femur as characterized by X-ray micro computed tomography analysis under the effect of antidepressants, probably by interfering with serotonin metabolism \[**[@R4]**\]. The clinical studies in literature found an increased risk of fracture based on high bone turnover markers and low bone mineral density in patients with depression and (SSRI) antidepressants \[**[@R5]**\]. The most important effect was due to the activation of 5-hydroxytriptamin receptors on bone (mainly on osteoclasts and osteoblasts) by using different pathways as endocrine or neural pathways \[**[@R6]**\]. The other observations on patients with depression pointed an increase of serum osteocalcin and decrease of serum resorption β-CTX marker after depression therapy with SSRI drugs \[**[@R7]**\]. The evidence between serotonin actions on bone raised the still unanswered question, as which is the exact place of serotonin assessment and if SSRI should be listed among the many causes of bone loss \[**[@R8]**\]. Some reports express a twofold fracture risk in SSRI users versus non-SSRI users but the risk is different with regard to the type of drugs, to the timing of therapy or the discontinuing of the medication \[**[@R9]**\].
The serotonin studies at different levels and the association with metabolism complications involve various observations. One study in 264 Japanese women found a correlation between fasting blood glucose and polymorphisms of the serotonin transporter-linked polymorphic region (5-HTTLPR) which is the main regulator of the transcriptional activity of serotonin \[**[@R10]**\]. In a report on 252 Greek subjects with type 2 diabetes, the S allele of 5-HTTLPR was associated with this glucose pathology \[**[@R11]**\]. The same type of connections was found on 234 type 2 diabetic patients with an increased risk of anxiety/depression in cases with 5-HTTLPR/rs25531 genotype \[**[@R12]**\]. The observations from Kansai Medical University refer to the plates that excessively release serotonin parallel to the renal function damage in diabetic subjects \[**[@R13]**\]. Another mechanism that involves serotonin in diabetes is, as proved in a rat model, the possible disturbances of insulin communication in the hypothalamus \[**[@R14]**\]. The anomalies were also found in adipocytes where their long-term exposure to high levels of serotonin induces insulin resistance \[**[@R15]**\]. The metabolic complications pathways are closely connected to the bone status via serotonin signaling. One relationship is established via leptin in serotoninergic brain signaling acting both on food regulation and on bone mass \[**[@R16]**\].
Our aim was to correlate the bone turnover markers or Dual Energy X-ray Absorptiometry (DXA) assessment with the levels of serum serotonin in postmenopausal women without a previous specific bone disease.
Material and Methods {#sec1-2}
====================
This is a cross-sectional pilot original research study. We included Caucasian women in menopause who were not previously diagnosed with bone diseases as osteoporosis or Paget disease, etc. They were 40 years and older. We excluded the subjects previously treated for osteoporosis or for fragility fracture risk prevention as bisphosphonates, and also the patients previously or currently treated for bone metastases. Moreover, the women known with carcinoid disease or neuroendocrine tumors were not enrolled.
Anamnesis was performed; the weight (in kilograms or kg) and the height (in meters or m) were measured in order to calculate the Body Mass Index (BMI) in kg/m2. The serum levels of calcium and phosphorus were assessed. The bone turnover markers were evaluated: the bone formation markers were serum alkaline phosphatase or AP (colorimetric assay) in Units/Liter (U/L), serum osteocalcin or OC (fotochemiluminescence assay) in nanogram/millimeter (ng/mL); the bone resorption marker was serum CrossLaps or CL in nanogram/millimeter (ng/mL). The serum serotonin was performed (high-pressure liquid chromatography). The normal values of the bone turnover markers were AP between 35 and 129 U/L, CL between 0.166 and 0.476 ng/mL, OC between 4.9 and 30.5 ng/mL. The normal serotonin values were between 80 and 450 ng/mL. All the subjects had a central DXA at least at two central sites (with a GE Prodigy device). This analysis was performed by using the data provided by lumbar spine DXA: bone mineral density (BMD) in g/cm2. The WHO criteria of osteoporosis were applied to the diagnosis of osteoporosis (T-score ≤ 2.5SD), osteopenia (T-score \>-2.5SD and ≤-1), and normal DXA (T-score \>-1) \[17\].
**Statistical analysis**
The studied parameters were expressed as mean, standard deviation, ranges. SPSS 21 (IBM C) was used to calculate bivariate and partial correlations (to adjust the effect of age and BMI) between serotonin and lumbar BMD. Linear regressions and bivariate correlations were calculated for serotonin - osteocalcin, serotonin - CrossLaps and serotonin - alkaline phosphatase and the results were the same up to the second decimal. Partial correlations for the adjustment of the effect of age and BMI were also calculated for these three relationships. A two sided alpha value of under 0.05 was considered statically significant (p\<0.05).
Results {#sec1-3}
=======
191 postmenopausal women were enrolled. The mean age at evaluation was 57.109 years. The mean BMI was 29.088 kg/m2. The values of the bone turnover markers were calculated (**[Table 1](#T1){ref-type="table"}**).
######
The baseline characteristic of the entire cohort (number of subjects: N=191)
------------------- --------- --------- -------- ----------------
Parameters Minimum Maximum Mean Std. Deviation
age (years) 41 78 57.109 7.683
BMI (kg/m2) 18 64 29.088 6.205
serotonin (ng/mL) 23 393 159.98 69.019
CL(ng/ml) 0.03 1.6600 0.452 0.269
OC(ng/ml) 4.061 69.990 22.262 11.027
------------------- --------- --------- -------- ----------------
63 subjects had a normal DXA with a mean age of 53.269 years (**[Table 2](#T2){ref-type="table"}**).
######
The baseline characteristic of the subjects with normal DXA (number of subjects: N=63)
------------------- --------- --------- --------- ----------------
Parameters Minimum Maximum Mean Std. Deviation
age (years) 41 66 53.269 5.355
BMI (kg/m2) 19 50 30.134 6.316
serotonin (ng/mL) 25 323 154.349 67.783
CL(ng/ml) 0.14 1.43 0.445 0.288
OC(ng/ml) 6.6 67.95 20.627 10.538
AP (U/L) 39 238 78.667 29.999
------------------- --------- --------- --------- ----------------
88 women had osteopenia (**[Table 3](#T3){ref-type="table"}**).
######
The baseline characteristic of the subjects with osteopenia based on central DXA (number of subjects: N=88)
------------------- --------- --------- --------- ----------------
Parameters Minimum Maximum Mean Std. Deviation
age (years) 42 78 57.943 7.558
BMI (kg/m2) 18.5 64 29.228 6.334
serotonin (ng/mL) 23 393 166.465 74.625
CL(ng/ml) 0.03 1.66 0.464 0.281
OC(ng/ml) 6.84 69.99 23.004 11.352
AP (U/L) 28.4 153.67 78.841 23.586
------------------- --------- --------- --------- ----------------
40 patients were diagnosed with osteoporosis based on DXA and applying the WHO criteria (**[Table 4](#T4){ref-type="table"}**).
######
The baseline characteristic of the subjects with osteoporosis (number of subjects: N=40)
------------------- --------- --------- -------- ----------------
Parameters Minimum Maximum Mean Std. Deviation
age (years) 44 78 61.325 8.422
BMI (kg/m2) 18 37 27.131 5.379
serotonin (ng/mL) 41 319 154.6 57.486
CL(ng/ml) 0.13 1.02 0.439 0.201
OC(ng/ml) 4.061 58.86 23.543 11.109
AP (U/L) 46 153\. 79.512 21.974
------------------- --------- --------- -------- ----------------
The mean values of the serum serotonin were within the normal ranges: for the entire cohort (159.98 ng/mL), for subjects with normal DXA (154.349 ng/mL), for osteopenia group (166.465 ng/mL), and osteoporosis group (154.6 ng/mL). The higher value was registered in the women with osteopenia with no statistical significant difference between the three groups. The linear regression analysis between serum serotonin levels and the bone formation marker serum osteocalcin pointed a positive r-value for the entire studied population and for each of the three DXA groups (DXA normal, osteopenia, and osteoporosis). None of these results was statistically significant (**[Table 5](#T5){ref-type="table"}**).
######
The linear regression between serotonin and osteocalcin (OC), CrossLaps (CL), and alkaline phosphatase (AP). The partial correlation between serotonin and lumbar BMD (DXA)
----------------------------------------- ---------------- ------ ---------------- ------ ---------------- ------ ----------------- ------
correlation Serotonin - OC Serotonin - CL Serotonin - AP Serotonin - BMD
r p r p r p r p
all 0.07 0.4 0.05 0.53 0.07 0.35 0.02 0.77
all (adjusted for age and BMI) 0.06 0.43 0.05 0.52 0.08 0.31 0.03 0.97
normal DXA 0.08 0.56 -0.07 0.62 -0.17 0.19 -0.13 0.3
normal DXA (adjusted for age and BMI) 0.04 0.77 -0.01 0.96 -0.14 0.29 -0.14 0.3
osteopenia 0 0.99 0.05 0.66 0.24 0.03 0.14 0.,2
osteopenia (adjusted for age and BMI) 0 0.99 0.06 0.63 0.24 0.03 0.14 0.2
osteoporosis 0.24 0.19 0.4 0.03 0.18 0.29 0.15 0.34
osteoporosis (adjusted for age and BMI) 0.24 0.21 0.4 0.03 0.18 0.29 0.16 0.33
----------------------------------------- ---------------- ------ ---------------- ------ ---------------- ------ ----------------- ------
The linear regression between serotonin and the resorption marker serum CrossLaps was positive, except for the normal DXA group. Statistically significant results were found in the subjects with osteoporosis (N=40), meaning r=0.4, p=0.03, with similar results when adjusting for age and BMI. The linear regression between serotonin, on one hand, and serum alkaline phosphatase, on the other hand, was positive, except for the women with normal DXA evaluation. The only statistically significant values were in the patients with osteopenia: r=0.24, p=0.03, with no changes when adjusting for age and BMI (**[Fig. 1](#F1){ref-type="fig"}**).
{#F1}
The partial correlation between serotonin levels and lumbar BMD (DXA) was positive for all the patients, and for each of the groups with osteopenia, and osteoporosis, but no results had statistical relevance (**[Table 5](#T5){ref-type="table"}**).
Discussions {#sec1-4}
===========
This study represents an attempt to point out the place of the serum serotonin as a possible bone turnover marker. There are very limited similar data in literature, in this particular field of clinical practice involving the current bone evaluation in apparently normal subjects, meaning with no particular pathology related to the serotonin metabolism.
Based on our observations, the partial correlation between serotonin and lumbar DXA was not statistically significant. From another point of view, the present analysis regarding the bone turnover markers showed statistically significant results between the levels of serotonin and CrossLaps in osteoporotic women, and between the levels of serotonin and alkaline phosphatase in osteopenic subjects. It seems that serum serotonin is a possible resorption marker, more useful in postmenopausal women with abnormal DXA results than in those with normal DXA. The clinical use of performing serotonin in order to obtain more information about the bone is still unclear. For example, studies in untreated patients with carcinoid syndrome could not find significant changes in bone turnover markers despite high levels of serum serotonin, and consecutively increased urinary 5-hydroxy indole acetic acid \[**[@R18]**\].
The present study has some limits. One of them is the limited number of patients with osteoporosis (N=40) but the total number of 191 subjects to whom both serotonin and bone evaluation were performed is relatively large compared to preexistent data in literature. Another is the fact that we did not focus on the subjects' history regarding different types of medication, especially from the psychiatric area because we considered the peripheral levels of serotonin as the most useful marker to assess the complex serotonin metabolism in current clinical practice, regardless the interferences of its metabolism pathways. Moreover, our aim was to evaluate the use of serotonin assessment independently of depression and antidepressant drugs. Generally, it is known that in short term, SSRI administration increases the 5-hydroxy-tryptophan levels, but on long term, its levels decrease more than a half \[**[@R19]**\]. Another aspect is related to the fact that a depressed individual displays a lower BMD and higher bone resorption markers than non-depressed people, but the direct serotonin underlying mechanism is still unclear \[**[@R20]**\]. Some data in literature supported the idea of type 2 diabetes mellitus and obesity, linked to the serotonin metabolism, but in our study, the BMI influence was adjusted, with no significant changes of the results \[**[@R11]**\].
Conclusion {#sec1-5}
==========
This pilot study in a field with very few similarities in the current clinical non-psychiatric practice revealed some correlations between the levels of serum serotonin and the bone turnover markers, but none between the levels of serotonin and the bone mass density as provided by lumbar DXA in postmenopausal women. The exact place of serotonin in the skeletal health assessment is still a matter of debate.
**Conflict of interest**
The authors have nothing to disclose.
| {
"pile_set_name": "PubMed Central"
} |
Japanese encephalitis virus (JEV) is an arthropod borne virus of family *Flaviviridae*. It is one of the most important causes of viral encephalitis worldwide leading to an estimated 35,000-50,000 encephalitis cases and 10,000-15,000 deaths annually in Asia[@ref1][@ref2]. In case of central nervous system (CNS) infections, viruses localize to specific regions of the brain and cause neuronal damage. Neurologic invasion can develop, possibly by growth of the virus across vascular endothelial cells, leading to involvement of large areas of the brain, including the thalamus, basal ganglia, brain stem, cerebellum, hippocampus, and cerebral cortex. The capacity of viruses to selectively infect specific tissues depends on an interaction between viral gene, proteins and host factors. There are over 120 viruses which cause encephalitis. These viruses have affinity for different species and in different parts of the brain, *e.g.* Herpes simplex virus I (HSVI) has affinity for frontotemporal area because of neurochemical and immunological properties[@ref3], rabies virus has affinity for acetylcholine receptors, reovirus for beta adrenalin receptors, HIV for CD~4~[@ref4] and poliovirus for hPVR and CD155 receptors which belong to the immunoglobulin superfamily[@ref5]. JEV does not produce encephalitis in pigs and birds, suggesting a genetic resistance[@ref6]. Mice model has been used since 1960s for the study of pathophysiology and possible treatment of JEV infection. Following the intracerebral inoculation of JEV in mice, 100 per cent mortality and 4.8 days mean survival have been reported[@ref7]. In JEV infection, cytopathic effect may have temporal sequence and may be influenced by a number of variables such as virus load and genetic susceptibility. The study of such changes is possible in an experimental model with a longer survival. Majority of the studies have used mouse model for reporting histological and immunohistopathological changes in JE[@ref8][@ref9][@ref10]. Ogata and colleagues[@ref11] developed a rat model to study the parkinsonian features in JE and emphasized age related neurotropism. They studied the changes up to 12 wk after JEV inoculation. In our previous radiological study on JE patients, maximum involvement of thalamus was noted[@ref12]. In the cerebral cortex, tropic and non-tropic areas have also been identified in a mouse model of JE[@ref8]. However, no effort has been made to evaluate affinity of JEV to different regions of brain and temporal changes in viral load. The limitation of conventional real time quantitative-PCR in the diagnosis of JE has been reported[@ref13]. Real time PCR assay has emerged as a promising technique because of its high sensitivity, specificity and rapidity[@ref14]. The usefulness of this rapid diagnostic assay in other viral diseases has been suggested[@ref15]. Based on the common involvement of thalamus, basal ganglia and midbrain on magnetic resonance imaging (MRI) in JE patients[@ref12], it may be worthwhile to evaluate the temporal changes of JEV RNA copies in different regions of the brain in the experimental model of JEV infection. The present study was aimed to document the distribution and quantitation of JEV RNA copies in different regions of the brain of rat infected with JEV and changes over a period of time using real-time PCR assay.
Material & Methods {#sec1-1}
==================
*Virus*: GP 78668A strain of JEV (a kind gift from Dr S. Vrati, National Institute of Immunology, New Delhi), a neurovirulent strain was used in this study. Virus was propagated in 3-4 days old suckling mouse brain. After 4 days of infection[@ref4] mice were sacrificed. Brain was removed aseptically, homogenized in sterile phosphate buffer saline (PBS) and centrifuged at 15,000 g for 30 min at 80°C. The supernatant was collected, aliquoted and stored at -70°C till further use. Virus titre was determined by the standard plaque assay[@ref16]. The titre of the virus was determined by the following formula:
*Animal*: Suckling pups of Wistar strain rats (12 days old rats with mother) purchased from Central Drug Research Institute, Lucknow were used in the study. The rats were maintained in the animal facility at Sanjay Gandhi Postgraduate Institute of Medical Sciences (SGPGIMS), Lucknow on alternating 12 h light and dark cycle. The study was approved by the ethics committee of SGPGIMS, and experiments were carried out in accordance with the institutional guidelines on the care and use of experimental animals. Rats were divided into two groups: JEV infected group (n=24) and mock infected controls (n=6). In JEV infected group, the rats were inoculated intracerebrally with 3 × 10^6^ pfu/ml of JEV[@ref11]. Control rats were inoculated with sterile PBS 1X (Sigma, USA). The study period was 20 days. The rats were monitored daily and six rats each were sacrificed on days 3, 6, 10 and 20 post-inoculation. The brains were excised aseptically and regions of the brain (cortex, striatum, thalamus and mid brain) were dissected out and were homogenised in chilled lysis buffer (1 x PBS with 10 μg/ml phenyl nethyl sulphonyl fluoride (PMSF).
*Rota rod test*: Motor deficit was assessed by the rota rod test one day prior to 3, 6, 10 and 20 days post-inoculation. Six rats from each group were tested on a motorized rota rod consisting of a grooved metal roller. The acceleration rate was set at 0.15 rpm/sec. Rats were placed on the roller, and the time they remained on the roller during rotation, was measured. A maximum of 120 sec was allowed per animal for fixed speed tests.
*RNA extraction*: Total RNA was extracted from tissue samples using a QIAmp viral RNA kit (QIAGEN, Inc., USA) according to the manufacturer\'s instructions. RNA was eluted in 50 μl of diethylpyrocarbonate (DEPC) treated distilled water and stored at -80°C till further use. The amount of RNA in each sample was quantified[@ref17] by measuring the absorbance at 260 nm using a spectrophotometer (Hitachi, Japan) with the following formula:
RNA concentration (μg/ml) = OD~260~ × DF × 40
Where, OD~260~ is the absorbance of the diluted samples at 260 nm and DF is the dilution factor.
The purity of RNA was determined by calculating the ratio of OD 260 to OD 280.
*Real-time PCR (RT-PCR)*: The real-time quantitative PCR assay was performed using Geno-Sen\'s JEV Real-time PCR Kit (Corbett Research, Australia). The forward and reverse primers hybridize to a specific sequence product. JEV-specific forward primer 391-GCAGAAAGCAAAACAAAAGAG and reverse primer 757-ACGGATCTCCTGCTTCGCTTG were designed from the C-prM region by aligning the available sequences in GenBank accession no. AF075723. An internal control gene was added to the reaction mix which was provided in the kit. RNA was reverse-transcribed using reverse primer at 50°C for 15 min using superscript II reverse transcriptase (Invitrogen, India). The cDNA was PCR amplified using the forward (391) and reverse (757) primers for detection of the JEV. PCR amplification was carried out by denaturing the DNA at 95°C for 10 min, followed by 45 cycles of 95°C for 30 sec, 55°C for 20 sec, and 72°C for 15 sec. An internal control gene was added to the reaction mix which has been provided in the kit. For each step, the temperature transition rate was 20°C/sec. After the thermal cycle a standard curve with the dynamic range of detection in copies/ml was constructed by preparing 10-fold serial dilutions of standard JEV.
*Statistical analysis*: The comparison among groups were made using one-way analysis of variance (ANOVA) with a *post-hoc* comparison (Newman Keuls multiple) test. Differences between means were considered significant at *P*\<0.05. All the statistical analysis was done using GraphPad Prism (3.03) software, USA.
Results {#sec1-2}
=======
There were 24 rats in JEV infected group and 6 in controls. All rats were subjected to daily clinical observation. The animals in JEV group started showing clinical symptoms from day 4 post-inoculation which manifested with huddling and slight hind limb weakness, pelvic elevation, somnolence, sluggishness, and lethargy.
No motor deficit was observed in any rat before JEV infection. Durations of stay on the accelerating rota rod for JEV infected rats were 115.5 ± 3.7, 98.7 ± 3.1, 101.4 ± 2.9 and 104.3 ± 3.2 sec, respectively on days 3, 6, 10 and 20 post-inoculation. There was a significant increase in motor deficit on day 6 compared to day 3 post-inoculation (*P*\<0.001) as well as with mock infected control on day 6. The duration of stay on the rota rod was longer in control rats compared to JEV inoculated rats on day 6 (*P*\<0.001), 10 (*P*\<0.0001) and 20 post-inoculation (*P*\<0.0001).
JEV RNA copies were present in all brain regions studied on days 3, 6 and 10 post-inoculation, however, no viral RNA copies were detected on day 20 post- inoculation in any studied brain region. Maximum numbers of JEV RNA copies were present in mid brain on days 3 and 10 post-inoculation as compared to cortex, striatum and thalamus. However, on day 6 post-inoculation, maximum numbers of JEV RNA copies were present in the thalamus ([Table](#T1){ref-type="table"}). There was a significant increase in JEV RNA copies on day 6 post-inoculation compared to day 3 in all the regions of brain studied (*P*\<0.001); however, JEV RNA copies significantly decreased on day 10 post-inoculation compared to day 6 (*P*\<0.001). No JEV RNA copies were detected in controls.
######
Japanese encephalitis virus (RNA copies/ml) in different brain region of rat at various time points
Discussion {#sec1-3}
==========
The present study revealed that JEV localizes in thalamus, striatum, mid brain and cortex. JEV showed maximum affinity to mid brain and thalamus followed by striatum and cortex. There was a decrease in JEV RNA copies in later phase of the disease and an improvement in motor deficit.
Binding of viral attachment proteins to the host cell receptors is a well-established phenomenon of viral tissue tropism. Using immunohistochemistry, Kim *et al*[@ref8], demonstrated that JEV has neuronal tropism in certain regions of the mice brain such as brain stem, thalamus, striatum and cortex. They have also reported perivascular cuffing in the hippocampus which is considered as non-tropic region, whereas in the tropic cerebral cortex there was disruption of cortical layer and moderate inflammatory changes.
Local replication of JEV occurs at the site of mosquito bite, which is followed by lymphatic or haematogenous spread[@ref18]. Replication of JEV beyond the primary site leads to secondary viraemia which can result in CNS involvement. In most patients, primary viraemia is terminated by a macrophage response and subsequently by the development of antibodies. JEV infects the endothelial cells of the capillaries of the brain and crosses the blood brain barrier[@ref19]. Following intracerebral inoculation, the virus may propagate through intracellular or extracellular spaces to lodge into the different areas of central nervous system depending on their tropism. Ogata *et al*[@ref11] demonstrated age dependent neurotropism in a rat model of JE. They reported that JEV antigen started declining in 14 days old rats and was undetectable in any region in 17 days old rats. In thalamus it was undetectable in 14 days old; striatum and brain stem in 17 days old and cerebral cortex in 14 days old rats[@ref11]. Several studies have detected latent JEV in the mosquitoes using real-time PCR[@ref20] and in the serum of humans, pigs and mice[@ref21][@ref22]. Studies have been conducted for detection of JEV in human samples by real-time RT-PCR assay[@ref23], and for detection of WNV and other flaviviruses of the JEV antigenic complex in different wild bird species[@ref24]. Differential expression of various receptors in different regions of the brain with reference to age could be responsible for difference in JEV RNA copies in different brain regions; however, this has not been investigated in earlier studies. Using immunohistochemistry in autopsy specimen, the highest concentration of antigen was demonstrated in thalamus and brain stem[@ref25]. These observations are in agreement with the present study. In a study on 56 JE patients, MRI revealed thalamic involvement in 83 per cent, basal ganglia in 46 per cent, mid brain in 35 per cent and cerebral cortex in 23 per cent[@ref12]. Similar distributions have also been reported in Eastern equine encephalitis, revealing thalamic and basal ganglia involvement in 71 per cent, brain stem in 43 per cent and cortex in 36 per cent patients[@ref26].
In this study, the JEV RNA copies were detected on day 3 post-inoculation, highest concentration recorded on day 6, and thereafter progressively declined and were undetectable on day 20. In another study on 12 days old rats, the distribution of JEV was shown in caudate, putamen, substantia nigra, thalamus and amygdaloid nuclei on day 3 post-inoculation using immunohistochemistry[@ref11]. We observed similar findings using real-time PCR assay. The decline in JEV RNA copies may be due to associated immune response and neutralizing antibodies. Early host defence against JEV infections is mediated by phagocytic cells, followed by a complex mechanism involving B and T effector cells[@ref27][@ref28]. Motor deficit was more marked on day 6 post-inoculation compared to day 3 in JEV infected rats; however, no significant improvement was observed over the time. In spite of significant reduction in viral load, lack of significant improvement in motor deficit after day 6 post-inoculation may be due to JEV induced neuronal damage. We have earlier reported free radicals generation by neurons in JE rat model. Overproduction of free radicals was observed in acute phase of disease which leads to neuronal damage and abnormal postural reflexes[@ref29]. Neuronal damage, inflammation, perivascular cuffing and glial nodule have also been reported in histopathological studies in animal and autopsy studies in humans[@ref30]. Reduction in JEV RNA copies in later phase of the disease corresponds well with our previous findings showing significant decrease in free radicals generation at the later stage of the disease[@ref29]. We have earlier reported high affinity of JEV to thalamus, mid brain, striatum and cerebral cortex using immunohistochemical and histological techniques[@ref31]. Cellular infiltration, perivascular cuffing, meningeal disruption, neuronal damage, neuronal shrinkage, and plaque formation were observed in rats from 10 post-inoculation.
In conclusion, thalamus and mid brain were found to be the most affected brain areas after JEV inoculation, indicating region specific affinity of JEV to the brain.
We acknowledge the Indian Council of Medical Research, New Delhi for financial support to Ruchi Srivastava as Senior Research Fellow and Council of Science and Technology, Uttar Pradesh for the project on encephalitis research.
| {
"pile_set_name": "PubMed Central"
} |
1. INTRODUCTION
===============
Spondyloarthropathies (Spondyloarthritis)(SpA) are a collection of chronic inflammatory rheumatic conditions that share multiple clinical features including axial and/or peripheral arthritis, enthesitis, absence of the serum rheumatoid factor and presence of common extra articular manifestations \[[@r1]\]. The spondyloarthropathy family comprises of Ankylosing Spondylitis(AS), psoriatic arthritis, reactive arthritis, inflammatory bowel disease associated SpA, juvenile SpA and undifferentiated SpA(uSpA) \[[@r2]\]. These diseases are strongly associated with the genes of the Major Histocompatibility Complex (MHC), in particular, the Human Leucocyte Antigen(HLA) B27 \[[@r3], [@r4]\].
Research in recent years has documented the clinical features of Ankylosing Spondylitis in the Middle East. The data highlighted the wide variation in HLA-B27 positivity across the region \[[@r5]-[@r10]\] There is an even greater paucity of documentation of non-AS SpA in the Middle East, which may affect the management of these patients \[[@r11]\]. Uppal et al. found several interesting differences between AS and undifferentiated SpA patients amongst the South Asian and Middle Eastern patients in Kuwait. However there remains a lack of information on whether similar inter-ethnic SpA profiles exist amongst other Middle Eastern countries. With the exception of a small case series on AS patients in the UAE, there has been no other comparison of inter-ethnic profiles of any SpA subtype in the Middle East \[[@r8]\].
In recent years, UAE has seen significant reforms in demographic and health care organization \[[@r12]\]. As an emerging market in medical tourism, Dubai in particular boasts a very diverse patient population. With only 11% of the UAE population constituting the native Emiratis, there remains a respectable opening for studying inter-ethnic profiles in the country \[[@r13]\]. The inter-ethnic profiling may be of clinical benefit if there is an awareness of the prevalence of certain features in specific patient groups \[[@r14]\].
This case series is the first of its kind in studying all the subtypes of SpA in the UAE. The primary aim of our study was to demonstrate the interethnic variations and clinical features of Spondyloarthritis (SpA) patients from a specialized arthritis center and a Rheumatology department at a multi-specialty center in Dubai, United Arab Emirates.
2. MATERIALS AND METHODS
========================
We conducted a retrospective study on patients with a diagnosis of SpA at a specialized arthritis centre and a rheumatology department of a multi-specialty medical centre in Dubai, UAE. Payment modalities for the patients at these private centres range from self-funding to comprehensive insurance cover. Patients were identified by examining electronic records for ICD 9 codes during the month of August 2012. Reviews of medical records were carried out to obtain the required data. The study was approved by an internal ethics committee at the specialized arthritis centre and was conducted in accordance with the recommendations of the Declaration of Helsinki.
The socio-demographic data collected included age, gender, smoking status, ethnicity and family history of SpA. Clinical data included the age of symptom onset, age of diagnosis, characteristic of back pain (inflammatory back pain was defined as low back pain and stiffness for more than three months that improves with exercise, but not relieved by rest), presence of peripheral arthritis, extra-articular manifestations (enthesitis, uveitis, psoriasis and inflammatory bowel disease), Human Leucocyte Antigen-27(HLA-B27) status, and treatment were recorded. The relevant information was obtained by accessing the last clinic consultation notes, when appropriate. HLA-B27 status was based on the result of the standard assay on peripheral blood as conducted by the local pathology laboratory. HLA-B27 subtype data was not recorded in our cohort due to inconsistent testing. The Bath AS Disease Activity Index(BASDAI) and Schoebers test are routinely conducted by the physician at every consultation for AS patients, but only the last documented entry was used for our analysis. Radiological evaluation was used to assess the evidence of unilateral or bilateral sacroiliitis on X-ray. A Magnetic Resonance Imaging (MRI) scan was used when the diagnosis was unclear.
Both centres adopted the same diagnostic criteria for the management of SpA. Patients were categorised as Ankylosing Spondylitis (AS) based on the modified New York criteria. Patients who fulfilled the European Spondyloarthropathy Study Group(ESSG) criteria but who did not meet the modified New York criteria were classed in the 'other SpA' group. In recent years, the terminology has evolved to classify as radiographic and non-radiographic SpA, however they were not classified in this manner at the time of diagnosis.
The analysis was conducted on the statistical package SPSS v20. The mean values were calculated for all continuous variables. The Unpaired Student's t test was utilised to compare the means of continuous variables, whilst the Fischer's exact test was used to evaluate percentages. A One-way ANOVA test was performed to study differences in means of more than two groups. A p-value of \<0.05 was deemed to be of statistical significance for this analysis.
3. RESULTS
==========
3.1. Socio-demographic Features
-------------------------------
A total of 141 patients were identified, of whom 88 held a diagnosis of AS and 53 of other SpA. Table **[1](#T1){ref-type="table"}** summarises the socio-demographic characteristics of the patient populations. The 'other SpA' group constituted 2 IBD related SpA, 5 psoriasis related SpA, 2 reactive arthritis related SpA and 44 undifferentiated SpA. The difference in the mean age of symptom onset(p\<0.01) and diagnosis(p\<0.001) between AS and 'other SpA' patients was statistically significant. Nine AS patients were known to have a family history of AS, but their HLA-B27 status were unknown. No 'other SpA' patients were known to have any family history of SpA. HLA-B27 was positive in 76% of our AS patients but only 36% in the 'other SpA' cohort.
3.2. Clinical Features
----------------------
The clinical characteristics of the patients are summarised in Table **[2](#T2){ref-type="table"}**.
Vitamin D insufficiency was observed in 42 (48%) AS and 26 (49%) 'other SpA' patients at some point during their follow-up care. The knees were the most commonly affected joint in 'other SpA' patients (49%) with 58% of these patients having bilateral knee involvement. Twenty three (43%) and five (9%) 'other SpA' patients showed back and neck involvement respectively. Eight AS patients had a characteristic bamboo spine whilst this feature was not observed in the 'other SpA' patients.
Achilles enthesitis was the most common form of enthesitis and was found in 28% and 13% of 'other SpA' and AS patients respectively. Plantar Fascitis was noted in 9(17%) 'other SpA' and 2(2%) AS patients Seventeen (32%) of 'other SpA' patients were on anti-TNF therapy at the last follow-up, whilst four(8%) other patients had previously utilised it. Anti-TNF was prescribed to 29 (33%) AS patients at the last follow up, whilst 14 (16%) others previously used this treatment modality. Nine(10%) AS and four(8%) 'other SpA' patients were on Methotrexate at the last follow-up; but seventeen(19%) AS and twelve (23%) 'other SpA' patients were on Sulfasalazine at the last follow-up.
3.3. Ethnic Variations
----------------------
The inter-ethnic variations in the features of SpA were studied in the Arab, Caucasian and Indian Subcontinent patients since they contributed to the large majority of the cohort(93%). Table **[3](#T3){ref-type="table"}** summarises some of the features.
Uveitis occurred in 5%, 23% and 12% of Arab, Indian Subcontinent and Caucasian AS patients, respectively. In 'other SpA' patients, uveitis occurred in 9% of Caucasians, 18% of Indian Subcontinent and none in Arab patients. The inter-ethnic differences in uveitis were not statistically significant in the patient populations. Enthesitis was observed in 10% of Arab, 15% of Indian Subcontinent and 24% of Caucasian AS patients. It was seen in 75% of Arab, 44% of Indian Subcontinent and 45% of Caucasian 'other SpA' patients. Of the eight AS patients that had a bamboo spine, six(15%) were from the Indian subcontinent, one(6%) patient was Caucasian and another(5%) was Arab. The mean(SE) BASDAI for the AS patients was 2.7(0.49) for Caucasian, 3.3(0.68) for Arab, And 3.6(0.41) for the Indian Subcontinent patients.
In patients with AS, HLA-B27 was positive in 53% of Arabs, 80% of Indian subcontinent and 93% of Caucasians. This contrasted to the low HLA-B27 positivity in the 'other SpA' population with 50% in Arabs, 25% in Indian Subcontinent and 33% in Caucasian patients. The difference in HLA positivity between the ethnic groups was statistically significant in the AS patients but not in the 'other SpA' patients.
4. DISCUSSION
=============
The presence of variations in clinical features of spondyloarthropathies across various ethnicities and geographical demographics is well recognized \[[@r15], [@r16]\]. The UAE offers a great insight into studying such conditions owing to its diverse ethnic population. This ethnic diversity in our cohort reflects to a certain degree the general demographics of the UAE, with Indian Subcontinent expats and Arab nationals comprising an estimated 54.7% and 27.5% of the UAE population \[[@r17], [@r18]\]. Our results, based on a relatively large patient cohort, will provide a stepping stone for setting up local guidelines for better standardised management.
4.1. Clinical Features
----------------------
AS and uSpA form the largest subgroups of spondyloarthropathies \[[@r19], [@r20]\]. USpA constituted 83% of the 'other SpA' group, and together with AS, they formed 94% of the total cohort. Due to lack of published data on 'other SpA' and since uSpA amounted to the large majority of this group, we decided to make comparisons with other documented uSpA data, mentioned in the discussion section. 6% of the AS patients did not have inflammatory back pain but were diagnosed on the basis that they still fulfilled the Modified New York criteria. Inflammatory back pain is one of the main clinical features of uSpA \[[@r20]\], but only 53% of the 'other SpA' cohort exhibited it. This prevalence is similar to those documented in other uSpa studies \[[@r21], [@r22]\].
4.2. Gender Distribution
------------------------
The male predominance in the Caucasian populations of our AS and 'other SpA' groups was similar to those AS and uSpA data published in other literature \[[@r20], [@r23], [@r24]\]. The patients from our Indian subcontinent cohort however, exhibited a much poorer male to female ratio than the previously published data on Indian AS and uSpA groups, which ranged from 5:1 -- 16:1 \[[@r22], [@r25]-[@r27]\]. The Indian Subcontinent category included countries such as Pakistan and Sri Lanka which had not been featured in these reported studies. The inclusion of spondyloarthropathy patients from these other countries perhaps explains the distinct differences in gender ratio. The Arab AS cohort displayed a very high male prevalence(95%) than that documented in similar Middle Eastern patient groups \[[@r7], [@r14], [@r28]\] including the Arab group of the previous UAE study \[[@r8]\]. Our series is the first to list the male to female sex ratio for any non-AS SpA patients of Arab ethnicity at 2:1.
4.3. Onset & Diagnosis
----------------------
The mean age at the onset of symptoms in AS Arab and Indian subcontinent patients was similar to those in the Arab and Indian populations of the Kuwaiti and UAE study \[[@r8], [@r11]\]. Interestingly, the onset of symptoms in 'Other SpA' patients occurred 8.2 and 9.8 years later in Arabs and Indian Subcontinent(South Asian) patients, respectively, than the Kuwaiti uSpA cohort \[[@r11]\]. The age at the onset of symptoms in Caucasian populations reflected those observed in a larger Caucasian dominant series \[[@r23], [@r24], [@r29]\].
Spondyloarthropathies usually take 5-6 years to be diagnosed after symptom onset, particularly in the presence of early limited features \[[@r30]\]. This delay may even stretch to over 10 years depending on circumstances \[[@r31]\]. Most patients in our study had a very short delay to diagnosis. These results are much better than those published in other Middle Eastern \[[@r11], [@r14], [@r28]\] countries and elsewhere \[[@r24], [@r32]-[@r34]\]. The
Arab subset of our 'Other SpA' group was an exception, with long delays to diagnosis compared to the other Arab uSpA data. We speculate that the Arab 'other SpA' cohort possibly initially presented with slowly progressive non-specific or peripheral joint symptoms which further exacerbated by delays in seeking medical attention.
4.4. Extra-articular Features
-----------------------------
Enthesitis was seen twice in as many 'other SpA' patients as AS patients and the difference was strongly significant. The occurrences of enthesitis amongst the ethnicities were similar to the large range of 20-60% occurrence reported in the literature which varied with the study population \[[@r8], [@r11], [@r22], [@r23], [@r26], [@r28], [@r35]\]. The Arab AS patients in our study had lower occurrences of enthesitis than those reported in other Arab studies \[[@r8], [@r9], [@r11]\].
Uveitis rates in our ethnic cohorts were also similar to the overall ranges(9-22%) reported in other studies with similar ethnicities and SpA subsets \[[@r8], [@r20], [@r23], [@r25], [@r27], [@r35]\] \[. Al Attia reported a much higher occurrence in Arab and lower occurrence in South Asian(Indian Subcontinent) population of uveitis in AS patients \[[@r8]\]. Our findings suggested the prevalence of uveitis in Indian subcontinent and Arab patients similar to those reported in other Indian and Arab dominant studies \[[@r9], [@r26], [@r27], [@r35]\]. The absence of uveitis amongst the Arab 'other Spa' patients could be associated with the low predominance of uveitis amongst Arab SpA patients or a consequence of a small Arab cohort.
4.5. Smoking
------------
A significant proportion of the patients had a history of smoking. Smoking is associated with poor functional outcomes in spondyloarthropathy patients \[[@r36]\]. Furthermore, spondyloarthropathy patients are at an independently increased risk of cardiovascular disease \[[@r37]\]. It is therefore vital for patient education to form an integral component of management.
4.6. Anti TNF
-------------
Anti TNF usage has been poorly reported in the Arab countries. The usage in our cohort is much greater than 9% reported in an Egyptian spondyloarthropathy registry \[[@r35]\]. This may be because of the tertiary nature of the clinics which tend to deal with the more severe cases. The long term follow up seen in this database may explain the increasing prevalence of anti-TNF usage with time as patients deteriorate justifying anti-TNF usage.
4.7. HLA-B27 Positivity
-----------------------
The association between HLA-B27 and SpA has led to significant research into HLA-B27 and its subtypes. There appears to be a rough correlation between the incidences of HLA-B27 in the general population with the incidences in the same ethnic SpA population \[[@r38], [@r39]\].
The prevalence of HLA-B27 is known to vary only slightly amongst the resident population of the Arab countries. It has been found to be 4% in Kuwait \[[@r5]\], 1.4% in Lebanon \[[@r6]\], 2.4% in Jordan \[[@r7]\], 6.4% in UAE \[[@r40]\] and 0.3% in Oman \[[@r41]\]. Arab and Indian Subcontinent people in the UAE were found to have a background HLA-B27 prevalence of 5.7% and 7.4%, however no data is available on Caucasians in this cohort. The HLA prevalence in the general Indian population was found to be varying between 26% \[[@r22], [@r42], [@r44]\] with relatively lower positivity seen in the South Indian population \[[@r45]\]. However specific geographical populations in India have been found to have HLAB27 as high as 19.6-29% \[[@r46]\]. In Caucasian, the background prevalence of HLA-B27 was found to be approximately 8-10% \[[@r3], [@r47]\]. We would therefore expect to find higher HLA-B27 prevalence in our Indian subcontinent and Caucasian population, which is the case.
The HLAB27 prevalence in Caucasians amongst our cohorts was observed to be similar to 90% reported in Caucasian AS patients but was much lower than the 70% in Caucasian uSpA patients \[[@r3]\].
There is a marked variation between HLA-B27 prevalence of AS patients in Arab countries, but is known to be generally lower than the global figures \[[@r10]\]. A review derived the prevalence of HLA-B27 to be 64% in Arab AS patients, by pooling together several studies \[[@r10]\]. HLA-B27 prevalence was found to vary between 76 -- 94% in Indian AS patients \[[@r29], [@r30], [@r48]\]. Our results in fact are very similar to those seen in the study on UAE AS population where Arab and Indian Subcontinent patients had HLA-B27 prevalence of 56% and 81%, respectively \[[@r8]\]. It is however slightly different to the Qatar resident Arab(74%), Asian(61%), Kuwait resident Arab(86.7%) and Indian subcontinent(75%) AS patients \[[@r10], [@r11]\].
Our Indian Subcontinent 'other SpA' demographic had a much lower HLA-B27 prevalence than the previously reported 45% and 84% amongst Indian uSpA patients \[[@r25], [@r43]\]. Our 'other SpA' patient profile had a markedly lower HLA-B27 prevalence than the Kuwaiti uSpA population both in Arab(66.7%) and Indian subcontinent(71.4%) patients \[[@r11]\].
4.8. Vitamin D Insufficiency
----------------------------
Vitamin D insufficiency characterized as \<30ng/ml was found to be nearly 50% in our SpA cohort during testing performed at any random point in their follow up. Its significance was found in its association with higher SpA disease activity and severity \[[@r49], [@r50]\].
5. LIMITATIONS
==============
Not everyone had HLA-B27 testing which may explain the low prevalence, particularly in the 'other SpA' group(55% were tested). This may be due to the fact that the results of the test were perhaps lost or the clinician felt that the result did not add enough diagnostic evidence to justify the cost of the test. Insurance cover was not recorded for patients, which may have been a possible source of bias in view of the high usage of Anti-TNF therapy in our cohort. Despite 83% of 'other SpA' group being constituted by uSpA, the 'other SpA' might have markedly affected the results, perhaps accounting for the differences in the results between other uSpA study groups.
CONCLUSION
==========
Our study demonstrates the largest database of clinical features of Spondyloarthropathy patients in the UAE. It also appears to be a noteworthy research work analysing inter-ethnic variations in SpA patients in the region. The study highlights that HLA-B27 prevalence is relatively poorer amongst Arab AS patients and 'other SpA' patients. Spondyloarthropathy patients in the UAE generally have a very short delay in diagnosis. The study also highlights the relatively higher prevalence of Anti-TNF usage in the UAE. Greater emphasis should be placed on patient and clinician awareness regarding uveitis in Indian Subcontinent SpA patients in view of its high prevalence. Improvements need to be made for awareness among primary care physicians and rheumatologists regarding the varied presentations of 'other SpA' Arab patients in order to minimise the remarkably long delay in diagnosis.
The authors would like to acknowledge the support offered by the staff members of the study centres.
ETHICS APPROVAL AND CONSENT TO PARTICIPATE
==========================================
The study was approved by an internal ethics committee at the specialized arthritis centre and was conducted in accordance with the recommendations of the Declaration of Helsink.
HUMAN AND ANIMAL RIGHTS
=======================
No Animals were used in this research. All human research procedures followed were in accordance with the ethical standards of the committee responsible for human experimentation (institutional and national), and with the Helsinki Declaration of 1975, as revised in 2008.
CONSENT FOR PUBLICATION
=======================
A written informed consent was obtained from all patients when they were enrolled.
CONFLICT OF INTEREST
====================
The authors declare no conflict of interest, financial or otherwise.
######
Clinical and laboratory features seen in our Spondyloarthropathy cohort.
**Feature** **AS** **Other SpA**
------------------------------- --------------- ---------------
Cohort Size 88 53
Male Proportion 71 (80.7%) 29 (54.7%)
Mean age at Symptom onset(SD) 28.4 (9.8) 33.88 (11.8)
Mean age at Diagnosis(SD) 31.9 (9.7) 37.85 (10.1)
Smoking Status
• Never 52/76 (68.4%) 42/49 (85.7%)
• Former 4/76 (5.3%) 1/49 (2.04%)
• Current 20/76 (26.3%) 6/49 (12.3%)
HLA-B27 Positivity 58/77 (75.3%) 10/28 (35.7%)
Ethnicity
• Indian Subcontinent 40(46.6%) 18 (34%)
• Arab 21 (23.9%) 22 (41.5%)
• Caucasian 17 (19.3%) 12 (22.6%)
• Afro Caribbean 1 (1.1%) 1 (1.9%)
• Eastern European 4 (4.5%) 0 (0%)
• East Asian 3(3.4%) 0 (0%)
AS, Ankylosing Spondylitis; Other SpA, Other Spondyloarthropathies; SD, Standard Deviation; HLA, Human Leucocyte Antigen
######
Clinical features of the Ankylosing spondylitis and other Spondyloarthropathy patients.
-------------------------------------------------------------------------------
**Clinical Characteristics AS**\ **Other SpA**\ **P value †**
**Frequency(%)** **Frequency(%)**
---------------------------------- ------------------ --------------- ---------
Cohort Size 88 53 \-
Inflammatory back pain 80(94) 28(53) \<0.001
Peripheral arthritis 54(61) 53(100) \<0.001
Enthesitis 14(16) 28(53) \<0.001
Uveitis 16(18) 6(11) 0.017
Psoriasis 3(3) 6(11) 0.08
IBD 2(2) 2(4) \>0.1
Last BASDAI (out of 10)**\*** 3.37 \- \-
Last Schoebers (cm)**\*** 4.98 \- \-
Sacroiliitis bilateral **∆** 70/80(88) 14/33(42) \<0.001
Sacroiliitis unilateral **∆** 7/80(9) 4/33(12) \>0.1
Bamboo spine 9(17) 0(0) 0.014
-------------------------------------------------------------------------------
†Comparisons were performed using Fischer's exact test **\***Mean Last clinic consultation BASDAI and Shoebers analysed **∆** Sacroiliitis on MRI prevalent in 93% and 36% of AS and SpA patients respectively. AS, Ankylosing Spondylitis; Other SpA, Other Spondyloarthropathies; IBD, Inflammatory Bowel Disease; BASDAI, Bath Ankylosing Spondylitis Disease Activity Index
######
Clinical and laboratory features of Ankylosing Spondylitis and other Spondyloarthropathy patients of the three major ethnicities.
-----------------------------------------------------------------------------------------------------------------------------------
-- AS Patients Other SpA Patients
------------------------------------ ------------------------ -------------------- ------- ------------------- ------------- ------
Cohort size 21 40 17 \- 12 18 22 \-
Male (%) 20(95) 32(80) 12(71) \- 8(67) 9(50) 12(55) \-
Age in years at symptom onset (SD) 29.7 (9.4) 28.6 (10.2) 26.6 (8.7) 0.64 30.0 39.2 (11.7)\ 31.1 (10.6) 0.05
(11.8)
Delay to diagnosis in years(SD) 2.89 (3.4) 3.85 (6.0) 1.87 (2.5) 0.39 6.92 1.22 (2.6)\ 3.38 (7.0) 0.06
(8.7)
HLAB27\ 10/19(53%) 28/35(80%) 14/15(93%) 0.023 2/4(50%) 2/8(25%) 5/15(33%) 0.45
Positivity
-----------------------------------------------------------------------------------------------------------------------------------
†Comparisons were performed using Fisher's exact test or a One-way ANOVA test
AS, Ankylosing Spondylitis; Other SpA, Other Spondyloarthropathies; SD, Standard Deviation; HLA, Human Leucocyte Antigen
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
In developing countries, where there are poor regulations for food hygiene, food handlers are appointed in food and drinking establishments without investigating their health status for the common intestinal parasite.[@b1-rrtm-10-025] The majority of asymptomatic individuals for parasitic infections can be considered dangerous to society, because such food handlers routinely practice their jobs without giving due attention to the transmission of infections. Intestinal parasites can be transmitted to consumers directly or indirectly through food, water, nails, and fingers from food handlers.[@b2-rrtm-10-025],[@b3-rrtm-10-025]
Studies have been done on food handlers to investigate the prevalence of intestinal parasites and associated risk factors. A study in west Iran indicated that among food handlers, about 9% stool specimens were positive for different intestinal parasites, which included *G. lamblia* 2.9%, *Entamoeba coli* 4.3%, *Blastocystis* spp. 1.4%, and *Hymenolepis nana* 0.5%. A valid health card, awareness of transmission of intestinal parasites, and participation in training courses in environmental health with intestinal parasite were negatively associated factors.[@b4-rrtm-10-025]
Also, in a study done in Gambia, about 250 (46.3%) food handlers were intestinal parasite carriers. From the identified parasites, the majority were *Entamoeba histolytica*/*dispar* (150 \[46%\]), followed by *G. lamblia* (52 \[16%\]), *E. coli* (40 \[12.3%\]), *Entamoeba hartmanii* 20 (\[6.1%\]), *Strongyloides* spp. (18 \[5.5%\]), *Ascaris lumbricoides* (14 \[4.3%\]), *Iodamoeba bütschlii* (nine \[2.8%\]), and *Taenia* spp. (six \[1.8%\]). Among the risk factors, living with domestic animals, lack of training in food handling, and handwashing practices was associated risk factors of intestinal parasite infections.[@b5-rrtm-10-025]
In a study done at Jimma University Specialized Hospital cafeteria food handlers, among 148 samples, 33% were positive for one or more intestinal parasites, of which the most prevalent identified was *A. lumbricoides* (16%), followed by *E. histolytica*/*dispar* (4.3%). Parasitic infection and food handlers who did not practice handwashing after defecation or before serving food were positively associated.[@b6-rrtm-10-025] According to studies done in southern Ethiopia, about 36% of food handlers were found to be positive for different intestinal parasites, with the most abundant being *E. histolytica*/*dispar* (14%), followed by *A. lumbricoides* (9.27%).[@b7-rrtm-10-025]
Therefore, the carrier states of humans are of concern to food-manufacturing and food-service institutions, because of the perceived risk of contamination of food by infected food handlers and the risk of food-borne disease outbreaks. The prevalence of these intestinal parasites is different in different areas of the country. In addition to geographical conditions, participation in different awareness programs might be another factor that needs investigation in this study area. This study aimed to assess the prevalence of intestinal parasites and associated risk factors among food handlers in Nekemte town from April to May, 2016.
Methods
=======
The study was conducted in Nekemte town, East Wollega zone, west Oromia, Ethiopia from April to May, 2016. Nekemte is located about 331 km from the capital city, Addis Ababa, the specific location 90°14' north latitude and 36°30' east longitude, and has an altitude range of 1,960--2,170 m above sea level. Nekemte is classified into six sub-administrative cities, with a total population of 104,806 and an area of 1,962 ha. In Nekemte, there were 55 hotels (Kasso sub-city 13, Darge sub-city 1, Bakkanisa Kasse subcity 25, Chelaleki sub-city 12, Burka Jato sub-city four) and 51 bars and restaurants (Darge 5, Bakkanisa Kasse 11, Chelaleki 12, Burka Jato 18, Bakka Jamaa 5) during the study period (data lists obtained from each subcity administration office). A total of 527 food handlers were working in these food-service establishments (obtained from Nekemte Town Health Office, Regulatory Department).
A cross-sectional study design was used. Source populations were all food handlers who prepare and serve food in the kitchens of different hotels, bars and restaurants found in Nekemte. All food handlers preparing and/or serving food for consumers in hotels, bars and restaurants found in Nekemte during the study period were included. A total of 240 food handlers were included in the study, selected using a simple random-sampling technique. To determine sample size from each food-service establishment (55 hotels and 51 bar and restaurants), proportional random allocation was used.
To collect data, structured questions on background information and risk-factor assessment were developed to a standardized level by translation to the local language (Afan Oromo). To check the correctness of the questions and need for modification, 24 (10%) of the required sample size were checked in the nearest place Diga town which is nearest to Nekemte. With pretested questions, study participants were interviewed to collect data on sociodemographic and related risks factors: certification of food training (it is food service--management certification that helps to ensure that food-service operators take proper precautions to minimize the spread of food-borne illness) and medical checkup (diagnosis of health status of food handlers in health institutions at least every 3 months).
After the questionnaire had been completed, appropriate instructions were given to the food handlers on how to collect stool specimens. Each food-handler specimen-collection container was labeled with a specific code for the purpose of identification. Stool samples provided by participants were processed for direct microscopy and concentration. Using an applicator stick, a fresh specimen was emulsified with normal saline and iodine solution on the slides and became ready for microscopic examination. Then, cysts, trophozoites, and eggs of intestinal parasites were examined directly under microscopy. Stool samples were emulsified in formol water, the suspension discarded to remove large fecal particles, ether or ethyl acetate added, and the mixed suspension centrifuged. Cysts and eggs were fixed and concentrated for microscopic examination.
In addition, a small portion of fresh stool samples were processed for detection of opportunistic parasites using the Ziehl--Neelsen method. Thin smears were prepared directly from sediments of concentrated stools and allowed to air-dry. Slides were then fixed with methanol for 5 minutes and stained with carbol fuchsin for 30 minutes. After slides had been washed in tap water, they were decolorized with acid alcohol for 1--3 minutes and stained with methylene blue for 1 minute. Slides were then washed in tap water and observed under light microscopy with a magnification of 100×.[@b8-rrtm-10-025]
For data quality, training was given to two senior laboratory professionals on how to complete the structured questionnaire with accuracy. Data quality was ascertained through review of all data-collection formats and follow-up of all stages of quality control in data collection. All stages of laboratory-test processes were managed according to available job aids. The accuracy of test results in identification of intestinal parasites was increased by conducting two smear examinations using physiological saline and using iodine solution. Additionally, one formalin-concentrated smear examination and one stained slide by modified acid-fast stain using formalin-concentrated samples was examined. Data analysis was done using SPSS version 20. Association between intestinal parasites with risk factors was determined usinglogistic regression and *P*\<0.05 was considered significant.
The study was approved by the Wollega University Research and Ethics Review Board in accordance with the Declaration of Helsinki. Then, an official permission letter was obtained from Wollega University and Nekemte municipality. Data were collected after written informed consent had been obtained from all study participants. For those aged \<18 years, their parents were contacted for their assent to participate in the study and provide written informed consent. An ethical issue that may have arisen from the results of this study was further circumvented by ensuring that the names of establishments and individuals were not mentioned. Being in collaboration with the town health office, food handlers who were positive for helminthes infections were treated free of charge. Those positive for protozoa were given prescriptions to be treated.
Results
=======
From the total food handlers, about 159 (66.3%) were in the age group of 20--40 years. Mean age of food handlers was 24 years(range 12--59 years). Of 240 participants, 109 (45.4%) were male and 131 (54.6%) were female. One hundred and twenty four (51.7%) of food handlers had educational status of grade 7--12. Only 20 (8.3%) food handlers had not had any formal education. The percentage of parasite infectivity decreased proportionally with age category among the food handlers: \<20 years 54.2%, 20--40 years 51.6%, and \>40 years 44.4% intestinal parasites were detected. In sum, 54.1% of males and 50.4% females were infected with intestinal parasites. For the education level, \>12 years 52.9%, 7--12 years 50.0%, 1--6 years 58.1%, and of the illiterate 45.0% were infected with intestinal parasites. Sociodemographic factors, such as age, sex, and education level, had no statistically significant association with intestinal parasitic infection (*P*\>0.05, [Table 1](#t1-rrtm-10-025){ref-type="table"}).
The prevalence of intestinal parasites in this study was 125 of 240 (52.1%). Among those positive individuals, 104 of 125 (83.2%) had a single infection, 19 of 125 (15.2%) double infection, and two of 125 (1.6%) triple infection ([Figure 1](#f1-rrtm-10-025){ref-type="fig"}). *E. histolytica*/*dispar* was the most predominant parasite (71 of 125 \[56.8%\]), followed by *A. lumbricoides* (33 of 125 \[26.4%\]), hookworm (21 of 125 \[16.8%\]) and *Taenia saginata* (20 of 125 \[16%\]). *G. lamblia* (two of 125 \[1.6%\]) and *S. mansoni* (one of 125 \[0.8%\]) were found less commonly ([Table 2](#t2-rrtm-10-025){ref-type="table"}).
The wet-mount method is more effective for protozoan infection (100%) than helmenthic infection (81%). In contrast, the concentration method is more effective for helmenthic infection (100%) than protozoan infection (18%), due to the destruction of trophozoites during centrifugation. From a total of 240 food handlers examined, the consistency of stool specimens showed statistically significant association (*P*=0.015) with intestinal parasite infection. Among those taking medical checkups, 45.2% were positive for intestinal parasites, and of those not taking medical checkups, 53.5% were positive for intestinal parasites. Of those who had been certified and those who had not, 63.2% and 51.1% were infected with intestinal parasites, respectively. Of those with \>2 service year, 1--2 service year, and \<1 service year, 46.2%, 44.0%, and 60.8% were infected with intestinal parasites, respectively ([Figure 2](#f2-rrtm-10-025){ref-type="fig"}).
There was statistical significance between personal hygienic conditions of food handlers and intestinal parasitic infection (all *P*\<0.05) ([Table 3](#t3-rrtm-10-025){ref-type="table"}). Most of those who did not use hygienic practices were infected by intestinal parasites greatly: those who did not wash their hands after touching different body parts 100%, after touching dirty materials 100%, and after using the toilet 94.7%, and those who did not wear proper working clothes 90%. Most hygienic practices, such as handwashing with water after using the toilet by, handwashing with water and soap after using the toilet, trimming finger nails, and wearing proper working clothes and shoes were significantly associated with intestinal parasitic infection if not regularly performed (*P*\<0.05, [Table 3](#t3-rrtm-10-025){ref-type="table"}).
Discussion
==========
The prevalence of intestinal parasites in this study was 52.1% (125 of 240). This prevalence was higher than a study done in Nigeria.[@b9-rrtm-10-025] This may be due to differences in sociodemographic factors. This study was somewhat closer to a study done in Bahir Dar town, northwest Ethiopia, in which the prevalence of intestinal parasites among food handlers was 41.1%.[@b10-rrtm-10-025] The prevalence of intestinal parasites in this study is comparable to a study conducted in the Mekelle University student cafeteria (49.4%).[@b11-rrtm-10-025] The present prevalence shown in the current study was higher than studies conducted in Amritsar, India (12.9%),[@b12-rrtm-10-025] Eastern Province, Saudi Arabia (9.3%),[@b13-rrtm-10-025] and Sai-Yok District, Kanchanaburi Province, Thailand (10.3%).[@b14-rrtm-10-025]
Among positive cases, 104 of 125 (83.2%) were a single infection, 19 of 125 (15.2%) double infection, and two of 125 (1.6%) triple infection among food handlers examined. The results of this study are similar to a study conducted in Jubail, Saudi Arabia, in which single infection was 83.94%, double infection 11.51%, and triple infection 1.55%.[@b15-rrtm-10-025]
In the present study*, E. histolytica*/*dispar* was the most predominant parasite (71 of 125 \[56.8%\]), followed by *A. lumbricoides* (33 of 125 \[26.4%\]), hookworm (21 of 125 \[16.8%\]), and *T. saginata* (20 of 125 \[16%\]). *G. lamblia* (two of 125 \[1.6%\]) and *S. mansoni* (one of 125 \[0.8%\]) were less prevalently found. In contrast to this, in a study conducted at a tertiary-care hospital in Nigeria, *E. histolytica* (4.5%) (next to *G. lamblia* \[9%\]) was the second-most predominant parasite detected.[@b9-rrtm-10-025] This difference might be due to geographic and sociodemographic differences among the food handlers. Our results are also in agreement with the study conducted in Bahir Dar town, northwest Ethiopia, in which *E. histolytica* (12.76%) and *A. lumbricoid* (11.7%) were the predominant parasites[@b10-rrtm-10-025] and also similar to the study conducted at Gondar University, northwest Ethiopia, in which *A. lumbricoid* (6.5%) and *E. histolytica* (6.0%) were the predominant parasites.[@b16-rrtm-10-025]
This research has revealed that most hygienic practices, such as handwashing with water after toilet use, handwashing with water and soap after toilet use, handwashing after touching dirty materials, trimming of finger nails, wearing of footwear, wearing proper working clothes, and handwashing after touching different body parts had statistically significant values (*P*\<0.05) as a determinant factor for intestinal parasite infection. This study result shows similarity with other studies[@b11-rrtm-10-025],[@b17-rrtm-10-025],[@b18-rrtm-10-025] that identified personal hygiene like poorly kept nails, dirty working clothes, lack of footwear, and handwashing practices as the determinant factors for intestinal parasite infection.
The scope of this study was limited to hotels, bars, and restaurants. Food handlers working in universities, hospitals, and butchers and those who sell coffee and tea were not included, which needs further study. There was no technique used to differentiate *E. histolytica* and *E. dispar*.
Conclusion
==========
The prevalence of intestinal parasites in this study was 52.1%. Among these positive cases, 82.4% were a single infection, 16% double infection, and 1.6% triple infection. *E histolytica*/*dispar* was the most predominant parasite (56.8%), followed by *A. lumbricoides* (26.4%), *T. saginata* (16.8%), and hookworm (16.8%). This research has revealed that hygienic practices, such as handwashing, trimming fingernails, wearing footwear, and wearing proper working clothes, were determinant factors for reduction of intestinal parasite infection. Food handlers should practice safe handling of food in preparation and service.
We would like to acknowledge all of the food-handler participants, data collectors, and health extension workers in the study area, and also Wollega University for funding this study.
**Disclosure**
The authors report no conflicts of interest in this work.
{#f1-rrtm-10-025}
{#f2-rrtm-10-025}
######
Sociodemographic factors of food handlers in food services in relation to parasite positivity in Nekemte town from April to May, 2016 (n=240)
Socio demographic variables Parasite infection *P*-value
----------------------------- -------------------- ----------- ------ -------
**Age, years**
\>40 5 4 44.4 0.839
20--40 77 82 51.6
\<20 33 39 54.2
**Sex**
Female 50 59 54.1 0.563
Male 65 66 50.4
**Education, grade**
\>12 16 18 52.9 0.680
7--12 62 62 50.0
1--6 26 36 58.1
Illiterate 11 9 45.0
######
Types of parasite recovered from food handlers in food services in Nekemte town from April to May, 2016 (n=240)
n \%
-------------------------------------------------------- ----- ------
**Single infection**
**Protozoa**
*Entamoeba histolytica*/*dispar* 52 21.7
Trophozoites 44 18.3
Cysts 8 3.4
*Giardia lamblia* (trophozoite) 1 0.4
**Helminthes**
*Ascaris lumbricoides* 20 8.3
*Taenia saginata* 17 7.1
Hookworm 13 5.4
*Schistosoma mansoni* 1 0.4
**Double infection**
*E. histolytica*/*dispar*--*A. lumbricoides* 9 3.8
*E. histolytica*/*dispar*--hookworm 4 1.7
*E. histolytica*/*dispar*--*T. saginata* 3 1.3
*A. lumbricoides*--hookworm 2 0.8
*E. histolytica*/*dispar*--*G. lamblia* 1 0.4
**Triple infection**
*E. histolytica*/*dispar*--*A. lumbricoides*--hookworm 2 0.8
Positive (single 104, double 19, triple 2) 125 52.1
Negative for any parasite 115 47.9
Total 240 100
######
Hygiene practices of food handlers in food services in relation to parasite positivity in Nekemte town from April to May, 2016 (n=240)
Response Parasite infection AOR 95% CI
------------------------------------------------- ---------- -------------------- ----- -------- -------- --------
**Hygienic practices** No 18 1 25.76 24.24 27.37
Yes 107 114
Handwashing with water after toilet No 26 3 29.76 21.92 68.11
Yes 97 111
Handwashing with water and soap after toilet No 20 0 28.24 16.65 40.00
Yes 104 115
Handwashing after touching dirty materials No 34 14 24.89 6.20 99.31
Yes 91 99
Trimming of fingernails No 24 6 62.30 6.20 99.31
Yes 100 109
Wearing footwear No 18 2 13.59 12.81 30.10
Yes 90 100
Wearing proper working clothes No 15 0 21.82 26.119 53.783
Yes 107 114
Handwashing after touching different body parts No 24 6 24.82 19.38 45.21
Yes 100 109
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Esophageal carcinoma is one of the leading causes of cancer-related death worldwide, especially in Asia.[@b1-ott-10-3965] Most Asian patients are diagnosed as having esophageal squamous cell carcinoma (ESCC), and the histology is somewhat different from non-Asian populations.[@b2-ott-10-3965],[@b3-ott-10-3965] Despite timely surgical interventions at an early stage, many cases tend to recur during the follow-ups.[@b4-ott-10-3965],[@b5-ott-10-3965] Currently, platinum-based regimens are a standard first-line treatment for advanced ESCC with a median progression-free survival (PFS) under 6 months.[@b6-ott-10-3965] No definitive chemotherapeutic regimen has been properly established for those who have failed prior first-line chemotherapy.
Vascular endothelial growth factor (VEGF) could stimulate the growth of new blood vessels, regulate vascular permeability and exert anti-apoptotic effects in endothelial cells. It frequently becomes overexpressed in esophageal cancers.[@b7-ott-10-3965],[@b8-ott-10-3965] In addition, its overexpression was identified as a poor prognostic predictor for advanced ESCC.[@b9-ott-10-3965]
Previous studies have indicated that apatinib, a VEGFR-2 inhibitor, was potentially efficacious for solid carcinomas.[@b10-ott-10-3965] As a small-molecule, VEGFR tyrosine kinase inhibitor improved PFS and overall survival (OS) in pretreated patients with advanced gastric cancer.[@b11-ott-10-3965],[@b12-ott-10-3965] However, no clinical studies have examined the efficacy and safety of apatinib treatment for advanced ESCC.
A retrospective study was conducted to evaluate the efficacy and safety of apatinib for advanced ESCC after failed prior first-/further-line treatment.
Patients and methods
====================
Patient eligibility
-------------------
Patients with advanced ESCC receiving apatinib as second/further-line treatment between March 2014 and June 2016 were included. All histological diagnoses of ESCC were made according to the histopathological criteria of WHO 2015 version. No local radiotherapy or interventional therapy was offered during apatinib dosing. The study protocol was approved by our institutional review board of Zhejiang Cancer Hospital. All participants provided informed consent prior to treatment.
Treatment regimen
-----------------
Apatinib was administered at a daily dose of 500 mg, and one treatment cycle lasted 28 days. In addition, one dose reduction (500--250 mg) was allowed for drug-related toxicity.
Responses and toxicities
------------------------
Tumor efficacy was evaluated by the Response Evaluation Criteria in Solid Tumors (RECIST 1.1). Objective tumor responses included complete response (CR), partial response (PR), stable disease (SD) and progressive disease (PD). In addition, toxicities were assessed by the National Cancer Institute Common Toxicity Criteria version 4.0 (CTC 4.0). Tumor responses were evaluated for every two cycles when no noticeable sign of progression was present.
Follow-ups and statistical analyses
-----------------------------------
PFS denoted the time from the first dosing day of apatinib to documented progression or mortality from any cause. In addition, OS was defined as the time from the first dosing day to mortality or the last follow-up. Survival analysis was conducted using the Kaplan--Meier method and compared using log-rank test. The survival curves were plotted according to the Kaplan--Meier method. Statistical analysis was performed using SPSS version 17.0 (SPSS Inc., Chicago, IL, USA). The median follow-up period was 10.2 (2.0--22) months. Follow-ups were conducted up to October 30, 2016.
Results
=======
Patient characteristics
-----------------------
A total of 62 patients diagnosed with ESCC were included in the current study. Among them, 54 were male and eight were female with a median age of 60.5 years. In addition, 46 of them were previous or current smokers and 16 belonged to never smoker category. All of them received platinum-based first-line chemotherapy. Apatinib was prescribed as second-line (n=21) and further-line (n=41) treatments. Performance status (PS) was 0--1 in 52 patients and 2 in 10 patients. Patient characteristics are summarized in [Table 1](#t1-ott-10-3965){ref-type="table"}.
Clinical efficacies
-------------------
The clinical responses were as follows: CR (n=0), PR (n=15), SD (n=31) and PD (n=16). The values of objective response rate (ORR) and disease control rate (DCR) were 24.2% and 74.2%, respectively. The median PFS was 115 days (95% CI, 97--133; [Figure 1](#f1-ott-10-3965){ref-type="fig"}), and the median OS was 209 days (95% CI, 165--253; [Figure 2](#f2-ott-10-3965){ref-type="fig"}).
No significant correlation existed in PFS among gender (*P*=0.51), age (*P*=0.43), line of therapy (*P*=0.43), smoking history (*P*=0.23), location of tumor (*P*=0.44) and PS (*P*=0.06). Univariate analysis is detailed in [Table 2](#t2-ott-10-3965){ref-type="table"}. Patients with grade 3/4 toxicities showed a longer PFS than those without grade 3/4 toxicity (136 vs 63 days, *P*=0.044; [Figure 3](#f3-ott-10-3965){ref-type="fig"}). Interestingly, PFS in individuals with grade 3/4 hypertension and hand-foot syndrome was longer than that in other patients (153 vs 112 days, *P*=0.037).
Among 46 SD patients, 13 had a PFS of \>6 months, and the patient characteristics are summarized in [Table 3](#t3-ott-10-3965){ref-type="table"}.
Toxicity evaluations
--------------------
The median dose of apatinib was 500 (250--500) mg. Ten dosage reductions were available. The rate of grade 3/4 toxicities was 59.7% (37/62). Five patients presented with grade 4 toxicity, including worsening proteinuria (n=2), hypertension (n=2) and hand-foot syndrome (n=1). The most common grade 3/4 adverse events were as follows: hand-foot syndrome (n=10), hypertension (n=7), proteinuria (n=7), hepatic injury (n=5), fatigue (n=3), esophagitis (n=3) and nausea/vomiting (n=2; [Table 4](#t4-ott-10-3965){ref-type="table"}).
Discussion
==========
To sum up, apatinib had some potential efficacy as a salvage treatment for advanced ESCC therapy. To the best of our knowledge, it represented the first-ever attempt of examining the efficacy and safety of apatinib for advanced ESCC.
Platinum-based agents are currently a standard first-line treatment for advanced ESCC, and the median PFS has a range of 4--6 months.[@b13-ott-10-3965]--[@b15-ott-10-3965] Half of the patients unresponsive to first-line treatment might receive a second-line therapy. Yet, the median PFS remains at a range of 2--4 months.[@b16-ott-10-3965],[@b17-ott-10-3965] For patients who have failed second-line chemotherapy, no definitive chemotherapeutic regimen has been recommended. New treatment strategy is urgently needed for achieving a better PS.
Several studies have identified the blockage of VEGFR-2 as a promising therapy for inhibiting angiogenesis.[@b18-ott-10-3965],[@b19-ott-10-3965] Apatinib, the first oral VEGFR-2 inhibitor, has previously demonstrated survival benefits for metastatic gastric cancer.[@b12-ott-10-3965] Although approved domestically for gastric cancer treatment, apatinib was also effective for patients with advanced breast carcinoma and lung cancer who are unresponsive to standard pretreatment.[@b20-ott-10-3965],[@b21-ott-10-3965] In the current study, the values of DCR and ORR were 74.2% and 24.2%, respectively. There was a trend of better efficacy compared with second-line chemotherapy for advanced ESCC.[@b5-ott-10-3965] Interestingly, patients with grade 3/4 toxicities had a longer PFS than those without grade 3/4 toxicities. Patients with hypertension and hand-foot syndrome benefited more than those with other adverse events. Together with previous study,[@b12-ott-10-3965] our results indicated that some toxicities would be predictive factors for the efficacy of apatinib treatment.
Hand-foot skin reaction, proteinuria and hypertension were the most common adverse events in apatinib treatment, with grade 3/4 adverse events occurring in over 60% of patients with gastric carcinoma.[@b11-ott-10-3965],[@b12-ott-10-3965] Over 20% of patients experienced dose modifications with a recommended daily dose of apatinib (850 mg) treatment in a Phase III trial.[@b12-ott-10-3965] In another trial, the recommended daily dose was 500 mg and grade 3/4 toxicities significantly decreased, and the efficacy was similar to those of high dose for breast carcinoma.[@b20-ott-10-3965] In the current study, a recommended dose of 500 mg was used. The results showed that grade 3/4 toxicities occurred in over half of the patients. Although different daily apatinib doses were used for gastric carcinoma (850 mg) and breast carcinoma (500 mg), similar toxicities were observed. It was considered that several patients with gastric carcinoma underwent previous gastrectomy, and the absorption ability of apatinib might be lowered.
Retrospective nature and a small sample size were two major limitations of the current study. In addition, the dose of 500 mg apatinib adopted in this study was not widely recommended. Hence, this dose must be confirmed by further prospective studies. However, without prospective clinical studies in the literature, our study may be deemed as meaningful.
Conclusion
==========
Our results support that apatinib is efficacious for advanced ESCC as salvage treatment. However, further prospective studies are required to fully elucidate its efficacy and toxicity.
**Disclosure**
The authors report no conflicts of interest in this work.
{#f1-ott-10-3965}
{#f2-ott-10-3965}
{#f3-ott-10-3965}
######
Clinical characteristics of 62 patients
Variables N (%)
----------------------------------------- ---------------
Gender
Male 54 (87.1)
Female 8 (12.9)
Age (years)
Median (range) 60.5 (40--72)
\>60 32 (51.6)
≤60 30 (48.4)
PS
0--1 52 (83.9)
2 10 (16.1)
Smoking history
Yes 46 (74.2)
No 16 (25.8)
Alcohol use
Yes 49 (79.0)
No 13 (21.0)
Location of tumor
Upper third 9 (14.5)
Middle third 24 (38.7)
Lower third 29 (46.8)
Line of apatinib therapy
Second 21 (33.9)
Further 41 (66.1)
Prior therapies in advanced stage
Chemotherapy 44 (71.0)
Chemoradiotherapy 18 (29.0)
Post-progression therapy after apatinib
Chemotherapy 16 (25.8)
Palliative treatment 46 (74.2)
**Abbreviation:** PS, performance status.
######
Univariate analysis of the current population (n=62)
Characteristics PFS 95% CI *P*-value OS 95% CI *P*-value
------------------------- ----- ---------- ----------- ----- ---------- -----------
Gender 0.51 0.24
Male 95 72--125 176 165--256
Female 122 87--142 234 199--277
Age (years) 0.43 0.34
\>60 110 107--133 231 168--266
≤60 117 87--121 201 176--254
PS 0.06 0.04
0--1 132 106--157 254 226--312
2 87 55--125 156 117--204
Line of therapy 0.43 0.45
Second 126 75--131 234 176--265
Further 111 87--129 199 167--254
Smoking history 0.23 0.55
Yes 98 67--119 204 187--255
No 124 89--132 255 211--269
Location of tumor 0.44 0.25
Upper and middle third 101 78--125 187 167--254
Lower third 117 111--135 231 207--288
**Abbreviations:** OS, overall survival; PFS, progression-free survival; PS, performance status.
######
Clinical profile of patients with PFS \>6 months
Case Gender Age (years) Smoking Line of therapy Grade 3/4 toxicity PFS (days) OS (days)
------ -------- ------------- --------- ----------------- -------------------- ------------ -----------
1 Male 56 Yes Second Yes 188 252
2 Male 63 Yes Third Yes 181 254
3 Male 64 No Second Yes 195 357+
4 Female 54 No Fifth No 251 661
5 Male 65 Yes Second Yes 304 564
6 Female 48 No Fifth No 189 216
7 Male 63 Yes Third Yes 242 458
8 Male 64 No Second Yes 511 583
9 Female 62 No Second Yes 251 279
10 Female 54 No Third No 274 348
11 Male 57 Yes Third Yes 191 303
12 Male 54 Yes Second No 182 191
13 Male 50 Yes Fourth No 211 351
**Abbreviations:** OS, overall survival; PFS, progression-free survival.
######
Major toxicities of apatinib dosing
Toxicity Total (%) Grades 3/4 (%) Dosage reduction (%) Discontinuation (%)
-------------------- ----------- ---------------- ---------------------- ---------------------
Hand-foot syndrome 32 (51.6) 10 (16.1) 2 (20.0) 1 (33.3)
Hypertension 13 (21.0) 7 (11.3) 2 (20.0) 1 (33.3)
Proteinuria 15 (24.2) 7 (11.3) 2 (20.0) 1 (33.3)
Hepatic injury 12 (19.4) 5 (8.1) 0 (0.0) 0 (0.0)
Fatigue 9 (14.5) 3 (4.8) 2 (20.0) 0 (0.0)
Esophagitis 4 (6.5) 3 (4.8) 2 (20.0) 0 (0.0)
Nausea/vomiting 9 (14.5) 2 (4.8) 0 (0.0) 0 (0.0)
| {
"pile_set_name": "PubMed Central"
} |
Allgemeines {#Sec1}
===========
Die allogene Transplantation beinhaltet im Gegensatz zur autologen Transplantation die Übertragung von Knochenmark oder Blutstammzellen eines anderen Spenders. Die Stammzellen werden dabei entweder direkt aus dem Knochenmark oder aus dem Blut dieses Spenders gewonnen.
Vor der Übertragung erfolgt in der Regel eine myeloablative Konditionierung durch Hochdosis-Chemotherapie, ggf. plus Ganzkörperbestrahlung. Ein wesentlicher Vorteil der allogenen Transplantation ist, dass die letzten Reste des blutbildenden Knochenmarks von den Spenderzellen zerstört werden ( Graft-versus-Tumor-Aktivität, GvT). Indirekt zeigt sich dieses Phänomen an den häufigeren Rezidiven bei Spenden von eineiigen Zwillingen sowie von T-Zell-depletierten Transplantaten; direkte Evidenz ergibt sich aus der Beobachtung, dass durch Spenderlymphozyten z. B. ein Rezidiv einer chronischen myeloischen Leukämie in Remission gebracht werden kann. Die allogene Transplantation besteht also nicht nur in einem Ersatz blutbildender Zellen, sondern auch in einer Art Immuntherapie.
Dieses Geschehen leitet auch ein alternatives, noch in der Evaluation befindliches Verfahren der Stammzelltransplantation mit reduzierter Konditionierung; hier erfolgt im Wesentlichen eine Immunsuppression (keine Ablation), die Spenderzellen sollen eine hinreichende GvT-Aktivität bewirken. Die Neutropeniephase ist wesentlich kürzer. Im Ergebnis ist die Letalitätsrate außerhalb von Rezidiven offenbar geringer.
HLA-Gewebemerkmale von Empfänger und Spender sollten möglichst gut übereinstimmen. Dies ist jedoch nur begrenzt realisierbar. Mit jedem HLA-Mismatch sinkt allerdings die Chance auf eine erfolgreiche Transplantation. Die Beherrschung der Transplantat-gegen-Wirt-Reaktion (Graft versus Host Disease, GvHD) ist eine zentrale Herausforderung der allogenen Transplantation. Sie manifestiert sich hauptsächlich in Haut, Leber und Darm.
Phasen der Immunsuppression {#Sec2}
===========================
Im Rahmen der Stammzelltransplantation kommt es zu differenten Phasen der Immunsuppression (Abb. [1](#Fig1){ref-type="fig"}). Diese Phaseneinteilung ist klinisch hilfreich, muss jedoch stets im individuellen Kontext gesehen werden.
Frühe Phase {#FPar1}
-----------
Die frühe Phase beinhaltet die myeloablative Konditionierung durch Chemotherapie bzw. Ganzkörperbestrahlung (prä-engraftment). In dieser Phase steht das Risiko der Neutropenie im Vordergrund. Zusammen mit der Dauer der Konditionierung (3--6 Tage) und ca. drei Wochen bis zum Engraftment dauert diese Phase ca. 4 Wochen
Mittlere Phase {#FPar2}
--------------
Die mittlere Phase ist bestimmt durch die Geschwindigkeit der Rekonstruktion des zellulären Immunsystems (die wiederrum durch die Art der Konditionierung mitbestimmt wird) und die erforderliche Intensität der medikamentösen Immunsuppression bzw. die GvHD. Infektionen sind überwiegend Folge der zellulären Immunsuppression. Diese Phase dauert in der Regel zwei bis drei Monate, zuweilen (bei schwerer akuter bzw. chronischer GvH-Reaktion) aber auch deutlich länger.
Das lymphozytäre System benötigt eine erheblich längere Zeit bis zur Rekonstitution als die Neutrophilen. NK-Zellen erholen sich aus Progenitorzellen als erste, gefolgt von B-Zellen, zuletzt die CD4-T-Zellen, sodass für längere Zeit ein erniedrigter CD4/CD8-Quotient besteht. Die zelluläre Immunsuppression kann viele Monate anhalten, zuweilen auch Jahre.
B-Zellen benötigen zur Regeneration ein „Bursa-Äquivalent", d. h. eine spezifische Knochenmarksumgebung; diese wird durch eine medikamentöse Immunsuppression im Rahmen einer GvHD empfindlich gestört. Auch bei zeitgerechter Rekonstitution der B-Zellen und ohne GvHD besteht bis zu einem Jahr noch kein ausgebildeter B-Memoryzellpool mit der Folge einer defizitären Ausbildung von neutraliserenden Antikörpern; daraus begründet sich die lange anhaltende Empfindlichkeit gegen Viren und bekapselte Erreger.
T-Zellen erholen sich durch Expansion bei Lymphopenie des Wirts, allerdings wesentlich langsamer als CD8-Zellen. Die Geschwindigkeit ist stark vom Lebensalter abhängig und erfolgt mit zunehmendem Alter langsamer.
Plasmazellen und dendritische Zellen sind relativ robuster gegenüber der Konditionierungsbehandlung (Storek 2008).
Späte Phase {#FPar3}
-----------
Die medikamentöse Immunsuppression der späten Phase (ca. nach 100 Tagen) richtet sich ebenfalls nach der Intensität der chronischen GvHD bzw. der dadurch erforderlichen medikamentösen Immunsuppression.
Die Reihe der Faktoren, die das Infektionsrisiko zusätzlich determinieren, sind in Tab. [1](#Tab1){ref-type="table"} zusammengefasst.FaktorRisiko erhöhtZeit seit der TransplantationJe kürzer Transplantation zurückliegtMaligne ErkrankungBei fortgeschrittener ErkrankungPrä-TransplantationsphaseBei intensiverer Immunsuppression, prolongierter Neutropenie, Infektionen vor TransplantationHLA-MatchBei HLA-Mismatches sowie haploidentischen SpendernKonditionierungJe intensiver das ProtokollTransplantatBei Knochenmark (vs. Stammzellen)Bei T-Zell-depletierten TransplantatenEngraftment der NeutrophilenBei verzögertem EngraftmentGvHDBei akuter GvHD Grad II--IVImmunsuppressivaBei Steroiden, Antithymozyten-Globulin, Alemtuzumab
Im Falle einer Pneumonie reflektiert die Phase der Immunsuppression das zu erwartende Erregerspektrum. Im Unterschied zur autologen Stammzelltransplantation umfasst das Risiko für bestimmte Erreger je nach Phase nicht nur das der Neutropenie, sondern auch der zellulären und humoralen Immunsuppression durch die immunsuppressive Medikation bzw. die akute und chronische GvHD.
Zudem besteht das Risiko für nichtinfektiöse pulmonale Komplikationen.
Erreger {#Sec3}
=======
Allgemeines {#Sec4}
-----------
Das Erregerspektrum in Abhängigkeit von der Häufigkeit findet sich in Tab. [2](#Tab2){ref-type="table"} zusammengefasst.Häufig (bis 20 %)Bakterielle PneumonienAspergillus spp. (selten unter Posaconazol-Prophylaxe)RS-VirusRespiratorische Viren (in der Saison)Selten (\<5 %)Mykobakterien (TB, NTM)Nocardia spp.Candida spp.Pneumocystis (unter Prophylaxe)Cytomegalie (unter präventiver Therapie)AdenovirusToxoplasma gondiiZunehmendAspergillus spp. (non-fumigatus)ZygomyzetenFusarium spp.Scedosporium spp.
Bakterien {#Sec5}
---------
Das Risiko für bakterielle Pneumonien ist am höchsten in der frühen Phase bis zum Engraftment sowie in der späten Phase bei chronischer GvHD und bestehender obliterativer Bronchiolitis.
Die Diagnose einer bakteriellen Pneumonie erfolgt häufig empirisch. Die wichtigsten Erreger sind S. pneumoniae, S. viridans, S. aureus, H. influenzae, Enterobakterien und P. aeruginosa. S. viridans kann im Rahmen einer Bakteriämie zu embolischen pulmonalen Foci führen. Im Rahmen eines septischen Schocks kann sich ein akutes Lungenversagen mit bilateralen Verschattungen entwickeln.
Mykobakterien {#Sec6}
-------------
Mykobakterielle pulmonale Infektionen ( Tuberkulose, nichttuberkulöse Mykobakteriosen) sind sehr selten, treten aber häufiger als in der Allgemeinbevölkerung auf.
Pilze {#Sec7}
-----
Pneumonien durch Aspergillus sind unverändert häufig. Sie treten bevorzugt in der Zeit bis zum Engraftment, zwischen dem zweiten und dritten Monat nach akuter GvHD und in der späten Phase bei Patienten auf, die für längere Zeit eine intensive Immunsupression, vor allem Steroide, benötigen.
Die Patienten können eine Aspergillose zu Hause oder innerhalb des Krankenhauses erwerben. Letztere Übertragung ist im Falle bestehender HEPA-Filtration nur noch sehr selten.
Candida spp. sind dagegen nur sehr selten Erreger einer Pneumonie. Schimmelpilze durch nicht-fumigatus Aspergillen, Zygomyzeten und Fusarium nehmen zu.
Eine CMV-Seropositivität ist ein zusätzlicher unabhängiger Risikofaktor für Pilzinfektionen.
Pneumocystis {#Sec8}
------------
Das Risiko für Pneumocystis-Pneumonien ist hoch, insbesondere bei Patienten, die länger als vier Wochen Steroide erhalten.
Viren {#Sec9}
-----
Das Zytomegalievirus ist einer der wichtigsten Erreger von schweren Infektionen bzw. Pneumonien. Nach Etablierung der präventiven Therapien sind CMV-Pneumonien bis Tag 100 eher selten geworden; ein Problem bleiben weiterhin die späten CMV-Pneumonien durch ihre hohe Letalität (Nguyen et al. 1999).
Risikofaktoren umfassen:CMV-seropositive Empfänger,CMV-seronegative Empfänger und CMV-seropositive Spender,nicht verwandte Spender.
Umgekehrt haben seronegative Empfänger seronegativer Spender ein sehr geringes Risiko, vorausgesetzt, sie erhalten CMV-negative oder leukozytendepletierte Blutprodukte.
Hochrisiko-Patienten sind v. a. Patienten nach Nabelschnurtransplantation, haplo-identischer Transplantation und Transplantation von HLA-Mismatches; zudem solche, die behandelt werden mit:hohen Dosen Steroiden (≥1 mg/kgKG),Mycophenolat Mofetil,T-Zell-depletierten Produkten (über CD34-Selektion),spezifischen Anti-T-Zell-Medikamenten (z. B. Antithymozyten-Globulin).
HLA-Mismatches und eine akute und chronische GvHD sind zusätzliche Risikofaktoren speziell für die CMV-Pneumonie.
Über CMV hinaus kommen die Herpesviren Herpes simplex, Varizella Zoster (Styczynski et al. 2009), HHV-6 (Buchbinder et al. 2000) und selten Epstein-Barr-Virus vor (Stycynski et al. 2011); sie verursachen jedoch nur sehr selten Pneumonien. Respiratorische Viren wie Influenzavirus, Parainfluenzavirus (Wendt et al. 1992; Lewis et al. 1996) und RS-Virus häufen sich in den Herbst-/Wintermonaten (Ljungman et al. 2001). Epidemien wie durch H1N1-Influenzavirus führen nachweislich zu einer erhöhten Letalität (Ljungman et al. 2011). Adenoviren können neu erworben werden oder -- häufiger -- reaktivieren.
Parasiten {#Sec10}
---------
Pneumonien durch Toxoplasma gondii entstehen meist durch Reaktivierung; entsprechend hängt ihre Häufigkeit von der Durchseuchungsrate ab. Mehr als 95 % der Patienten mit pulmonaler Toxoplasmose waren vor Transplantation seropositiv. Sie treten meist innerhalb der ersten beiden Phasen der Transplantation auf. Eine akute GvHD scheint diese zu begünstigen.
Toxoplasmen werden nur selten intra vitam diagnostiziert. Entsprechend schlecht ist die Prognose (Martino et al. 2000a, b).
Diagnostik {#Sec11}
==========
Bronchoskopie {#Sec12}
-------------
Eine Reihe von Arbeiten haben die Ätiologien von pulmonalen Komplikationen bei Patienten nach Stammzelltransplantation beschrieben, allerdings ohne autologe und allogene Transplantationen bzw. nichtmyeloablative Protokolle zu trennen.
Ein direkter Vergleich der Ergebnisse ist aufgrund der Heterogenität der Settings, der Studiendesigns und der untersuchten Populationen nicht möglich. Dennoch lassen sich einige Schlussfolgerungen aus den verschiedenen Studien ableiten (Sirithanakul et al. 2005):Der primäre diagnostische Zugang über die Bronchoskopie erzielt eine hohe Ausbeute, wenn er einem systematischen Protokoll folgt, das sowohl infektiöse als auch nichtinfektiöse Ätiologien untersucht (Dunagan et al. 1997; Huaringa et al. 2000; Patel et al. 2005; Bissinger et al. 2005; Gilbert et al. 2013; Lucena et al. 2014).Die bronchoalveoläre Lavage ist die wichtigste Untersuchungstechnik, die angepasst an die individuelle Situation durch bioptische Techniken ergänzt werden kann (Patel et al. 2005; Lucena et al. 2014).Die bronchoskopische Untersuchung sollte umgehend innerhalb von vier Tagen erfolgen, weil in diesem Zeitrahmen die höhere Ausbeute zu erwarten ist (Shannon et al. 2010; Lucena et al. 2014).Obwohl die diagnostischen Ergebnisse Konsequenzen für die Therapie hatten, konnte ein Einfluss der bronchoskopischen Untersuchung auf das Überleben bislang nicht gezeigt werden (Patel et al. 2005).Die Komplikationsrate ist bei Patienten mit Stammzelltransplantation nicht höher als in anderen Populationen.
Vor diesem Hintergrund besteht unverändert kein allgemeiner Konsens über den Stellenwert der Bronchoskopie in der Diagnostik pulmonaler Komplikationen (Wahla et al. 2014).
### Empfehlung {#FPar4}
Die Indikation zur Bronchoskopie sollte den allgemeinen Prinzipien der Diagnostik von pulmonalen Komplikationen unter Immunsuppression folgen. Unilaterale flächige Verschattungen können demnach zunächst kalkuliert antimikrobiell behandelt werden. Bilaterale Verschattungen, nicht-flächige Verschattungen und Therapieversagen nach kalkulierter Therapie wären demnach Indikationen zur Bronchoskopie. Biopsien können zusätzlich gewonnen werden.
Feinnadelaspiration {#Sec13}
-------------------
Über die Bronchoskopie hinaus sind -- vor allem bei bis dahin negativen Befunden -- transthorakale Feinnadelaspirationen zu erwägen; diesen scheint insbesondere bei möglicher Pilzinfektion eine hohe Ausbeute zuzukommen (Jantunen et al. 2002).
Chirurgische Biopsie {#Sec14}
--------------------
Schließlich ist auch eine chirurgische Biopsie (vorzugsweise über VATS) eine Option. In einer größeren Serie konnte in drei Viertel der Fälle eine spezifische Diagnose gestellt werden. Die Aspergillose war die häufigste Diagnose in ca. 20 %. Eine Änderung der Therapie nach Biopsie erfolgte bei 40 % der Patienten. Unspezifische Befunde wurden insbesondere bei Patienten unter invasiver Beatmung gefunden (Zihlif et al. 2005).
Therapie der Zytomegalovirus-Pneumonie {#Sec15}
======================================
Standardtherapie {#Sec16}
----------------
Erstmals 1988 wurde die Wirksamkeit der antiviralen Therapie der CMV-Pneumonie nach Knochenmarkstransplantation in zwei Studien etabliert. Sie bestand aus Ganciclovir plus intravenösen IgG-Immunglobulinen bzw. CMV-Hyperimmunglobulinen (Emanuel et al. 1988; Reed 1 et al. 988).
Die Dosis bestand in der ersten Studie aus Ganciclovir 3 × 2,5 mg/kgKG über 20 Tage plus Immunglobuline 500 mg/kgKG jeden zweiten Tag, insgesamt 10 Dosen. Dazu kam eine Erhaltungstherapie mit Ganciclovir 3--5 × 5 mg/kgKG pro Woche, insgesamt 20 Dosen plus Immunglobuline 2 × 500 mg/kgKG pro Woche, insgesamt 8 Dosen (Emanuel et al. 1988).
In der zweiten Studie bestand die Induktionsphase über 14 Tage mit derselben Ganciclovir-Dosis. Das CMV-Hyperimmunglobulin wurde in einer Dosierung von 400 mg/kgKG an den Tagen 1, 2 und 7 sowie 200 mg/kgKG an Tag 14 gegeben. Eine Erhaltungstherapie erhielten nur Patienten, die noch symptomatisch waren, dies über 14 weitere Tage mit derselben Dosis für Ganciclovir und CMV-Hyperimmunglobulin 200 mg/kgKG an Tag 21. Patienten, die sich unter Therapie verschlechterten, wurden in den Dosierungen der Induktionsphase weiterbehandelt (Reed et al. 1988).
Die heute gebräuchliche Dosis des IgG-Immunglobulins beträgt 0,5 mg/kgKG (Maffini 2016). Auch aktuelle Daten belegen keinen Vorteil der zusätzlichen Gabe von Immunglobulin-Präparaten auch von CMV-Hyperimmunglobulin.
Foscarnet ist nicht in vergleichbarer Qualität systematisch untersucht worden, wird jedoch alternativ erfolgreich eingesetzt. Cidofovir ist ebenfalls wirksam (Ljungman et al. 2001). Neuere Ansätze bestehen in der Gabe CMV-spezifischer T-Zellen (Boeckh und Ljungman 2009; Boeckh 2011).
Therapie bei Vorliegen von Resistenzen {#Sec17}
--------------------------------------
Der weite Einsatz antiviral wirksamer Substanzen innerhalb präventiver Therapien erhöht das Risiko für die Entstehung resistenter Stämme. Zusätzliche Risikofaktoren bestehen in inadäquaten Dosierungen, mangelnder Absorption bzw. Bioverfügbarkeit der oralen Valganciclovir-Präparation. Auch das Ausmaß der Immunsuppression erhöht das Risiko einer Resistenz.
UL97-Mutationen führen zu einer Ganciclovir- bzw. Valganciclovir-Resistenz, eine UL54-Mutation auch zu einer Resistenz gegen Foscarnet.
Eine Resistenz sollte klinisch vermutet werden bei CMV-Virämie (belegt durch Antigenämie oder CMV-DNA im Blut), die nach zwei Wochen adäquater antiviraler Therapie nicht verschwindet bzw. sogar weiter ansteigt. Die Resistenz sollte dann genotypisch belegt und aufgeschlüsselt sein.
Für die Therapie resistenter CMV-Pneumonien bieten sich eine Reihe verschiedener Optionen an: Umstellung auf Foscarnet (wenn sensibel); Prüfung der Empfindlichkeit von Cidofovir bzw. Brincidofovir; Reduktion der Intensität oder Umstellung der Immunsupression (falls möglich); höhere Dosierungen von Ganciclovir (2 × 15 mg/kgKG plus G-CSF); Kombinationstherapien von Ganciclovir mit Foscarnet; IgG-Immunglobulintherapie; CMV-spezifische T-Zell-Therapie, die allerdings noch nicht etabliert ist. Neuere antivirale Substanzen wie Maribavir und Letermovir sind ebenfalls noch keine gesicherten Alternativen (Boeckh 2011; El Chaer et al. 2016).
Präventive Strategien: Prophylaxen und präemptive Therapie {#Sec18}
==========================================================
Die hier dargestellten Empfehlungen folgen denen eines gemeinsamen Statements verschiedener Fachgesellschaften (Tomblyn et al. 2009) sowie der DGHO (Ullmann et al. 2016).
Bakterien {#Sec19}
---------
Empfohlen wird eine antibakterielle Prophylaxe mit einem Fluorchinolon, z. B. Levofloxacin (1 × 500 mg oral). Diese sollte mit der Stammzell-Infusion beginnen und nach Erreichen einer normalen Granulozytenzahl beendet werden. Bei schwerer Hypogammaglobulinämie (IgG \< 400 mg/dl) sollte eine Substitution mit Immunglobulinen erfolgen (500 mg/kgKG/Woche).
Zusätzlich zum Impfschema wird eine langdauernde Prophylaxe gegen Pneumonien durch Streptococcus pneumoniae empfohlen für Patienten mit chronischer GvHD und/oder Hypogammaglobulinämie. Das Regime besteht aus 1 × 500--1000 mg Penicillin pro Tag.
Mykobakterien {#Sec20}
-------------
Eine Indikation zur Chemoprophylaxe besteht in folgenden Konstellationen:Patienten mit positivem IGRA, die keine antituberkulöse Therapie erhalten haben und keinen Anhalt für eine aktive Tuberkulose habenKontakt zu Patienten mit aktiver Tuberkulose
Die Chemoprophylaxe erfolgt mit Isoniazid 300 mg/Tag über 9 Monate, alternativ mit Rifamipcin 600 mg/Tag über 4 Monate.
Viren {#Sec21}
-----
### Zytomegalievirus {#Sec22}
Vor Einführung der präventiven Therapie war die Inzidenz der CMV-Pneumonie mit ca. 25 % sehr hoch. Die präemptive Therapie wurde 1991 begründet durch eine Studie, die zeigen konnte, dass bei asymptomatischen Patienten eine Detektion von CMV in der BALF an Tag 35 nach Transplantation hochgradig prädiktiv für eine spätere CMV-Pneumonie war (Schmidt et al. 1991; Rubin 1991). Aktuell erfolgt die Detektion einer Virusaktivierung über den Nachweis einer Antigenämie (pp65), die quantitative CMV-DNA oder CMV-mRNA (Boeckh und Ljungman 2009). Neuere Ansätze arbeiten mit CMV-DNA-Verdopplungszeiten (Solano et al. 2016).
Kandidaten für eine präventive Therapie sind:CMV-seropositive EmpfängerCMV-seronegative Empfänger mit CMV-positivem Spender
In Frage kommen eine prophylaktische oder eine präemptive Therapie. Eine Prophylaxe umfasst eine antivirale Behandlung aller Risikopatienten; eine präemptive Therapie erhalten nur derjenigen, die im noch asymptomatischen Stadium Anzeichen einer hohen Virusreplikation aufweisen.
Die Prophylaxe impliziert somit eine häufigere unnötige Gabe von antiviral wirksamen Substanzen und damit das Risiko der Resistenzentwicklung. Dieses Risiko ist aber auch bei präemptiven Therapien noch gegeben.
Die spezifischen Empfehlungen sind in Tab. [3](#Tab3){ref-type="table"} und [4](#Tab4){ref-type="table"} zusammengefasst.IndikationSubstanzenAlternativen30--100 Tage nach TransplantationRisikopatientenInitialGanciclovir2 × 5 mg/kgKG i.v. 5--7 TageErhaltung1 × 5 mg/kgKG i.v.InitialFoscarnet 2 × 60 mg/kgKG i.v.7 TageErhaltung 1 × 90--120 mg/kgKG i.v.oderInitialValganciclovir 2 × 900 mgErhaltung 1 × 900 mgoderInitialCidofovir 2 × 5 mg/kgKGErhaltung 1 × 5 mg/kgKG alle 14 Tage IndikationSubstanzenAlternativen\<100 Tage nach TransplantationAllogene SCT mit aktiver CMV-Infektion, gemessen an einem der Tests:• Antigenämie (pp65) (≥5 Zellen positiv)• quantitativer CMV-DNA• CMV-mRNAInitialGanciclovir2 × 5 mg/kgKG i.v. 7--14 Tage(autologe SCT: 7 Tage)mindestens 1 WocheErhaltung(bei persistierend positiven Tests)1 × 5 mg/kgKG i.v.Gesamtdauer Initialtherapie und Erhaltung mindestens 2 WochenAbsetzen nur bei negativierten TestsInitialFoscarnet 2 × 60 mg/kgKG i.v.Erhaltung 1 × 90 mg/kgKG i.v.oderInitialValganciclovir 2 × 900 mgErhaltung 1 × 900 mgoderInitialCidofovir 2 × 5 mg/kgKGErhaltung 1 × 5 mg/kgKG alle 14 TageGesamtdauer Initialtherapie und Erhaltung mindestens 2 WochenAbsetzen nur bei negativierten TestsAutologe SCT, seropositiv,mit hohem Risiko durch:• Ganzkörperbestrahlung (TBI)• Fludarabin• 2-Chloro-Deoxyadenosinund• Antigenämie (pp65) (≥5 Zellen positiv)• oder CD34 grafts\>=100 Tage nach TransplantationAllogene SCTBestehende Steroidtherapie für GvHDVorausgegangene CMV-Infektion in den ersten 100 TagenJeweils wenn positiv in einem der Tests:• Antigenämie (pp65) (≥5 Zellen positiv)• Virämie• Zweifach-Nachweis PCRInitialGanciclovir2 × 5 mg/kgKG i.v. 7--14 Tagemindestens 1 WocheErhaltung(bei persistierend positiven Tests)1 × 5 mg/kgKG i.v.oderInitialValganciclovir 2 × 900 mgErhaltung1 × 900 mgGesamtdauer Initialtherapie und Erhaltung mindestens 2 WochenAbsetzen nur bei negativierten TestsInitialFoscarnet 2 × 60 mg/kgKG i.v.Erhaltung1 × 90 mg/kgKG i.v.Gesamtdauer Initialtherapie und Erhaltung mindestens 2 WochenAbsetzen nur bei negativierten TestsSCT = Stammzelltransplantation
Maribavir (MBV), Letermovir (LMV) und Brincidofovir (BDF) sind neue Kandidaten für eine präventive Therapie, die zur Zeit noch evaluiert werden (Boeckh et al. 2015).
Maribavir ist ein UL97-Proteinkinase-Hemmer mit Wirksamkeit gegen CMV. Es ist oral verfügbar. Die Substanz wurde in einer randomisierten, Placebo-kontrollierten Phase 3 Multicenter Studie als Prophylaxe der CMV-Infektion untersucht. In einer Dosierung von 2 × 100 mg konnte die Inzidenz der CMV-Erkrankungen nicht reduziert werden. Allerdings war die Inzidenz in der Kontrollgruppe mit in den ersten 100 Tagen \<2,5 % sehr niedrig. Zudem könnte die Wahl der Dosis zu gering ausgefallen sein (Marty et al. 2011). Weitere Studien sind noch nicht abgeschlossen.
Letermovir ist ein CMV-Terminaseinhibitor (UL56) und selektiv gegen CMV wirksam, auch gegen ansonsten resistente Virusstämme. Es kann oral oder intravenös gegeben werden und ist gut verträglich. In einer Dosis von 240 mg zeigte es sich in der Prophylaxe gegenüber Placebo als wirksam (Chemaly et al. 2014). Letermovir steht kurz vor der Markteinführung.
Brincidofovir (Propyl-Cidofodir) ist die lipophile Form von Cidofovir, die auch oral eingesetzt werden kann. Es zeigte sich in einer Dosierung von 200 mg zweimal wöchentlich als wirksam in der Prophylaxe gegenüber Placebo. Diarrhoen waren dosislimitierend (Marty et al. 2013).
### Andere Herpesviren und Adenovirus {#Sec23}
Risikopatienten sind HSV- bzw. VZV-seropositive Patienten. Die Empfehlungen zur Prophylaxe gehen aus Tab. [5](#Tab5){ref-type="table"} hervor. Bei Varizella zoster besteht nach Exposition zu einem Erkrankten auch die Indikation zur passiven Immunisierung.IndikationSubstanzenAlternativenHerpes-simplex-VirusStart mit Beginn der Konditionierung, bis Engraftment bzw. bis Mukositis abklingtRisikopatientenAciclovir 2 × 400--800 mg/Tag oraloder 2 × 250 mg/m^2^ i.v.Valacyclovir 1--2 × 500 mg/Tag oralVarizella-zoster-VirusStart mit Beginn der Konditionierung und Gabe über 1 Jahr; bei chronischer GvHD und systemischer Immunsuppression auch längerRisikopatientenAciclovir 2 × 800 mg/Tag oral für ein JahrValacyclovir 2 × 500 mg/Tag oralAdenovirusFür 2--4 Wochen oder bis zur ImmunrekonstitutionRisikopatientenCidofovir 5 mg/kgKG i.v. einmal wöchentlichoder 1 mg/kgKG dreimal wöchentlichRibavirin3 × 15 mg/kgKG/Tag über 4 Tage, dann 3 × 8 mg/kgKG über bis zu 10 Tage
Das Risiko für Adenovirus-Infektionen wird in vier Stufen eingeteilt. Kandidaten für eine präemptive Therapie sind seropositive Patienten mit hohem Risiko. Dazu zählen v. a. Patienten nach Nabelschnurtransplantation, haplo-identische Transplantation und Transplantation von HLA-Mismatch-Spendern mit chronischer GvHD, T-Zell-depletierte Transplantate, T-Zell-Antikörper-Therapien (Antithymozyten-Globulin, Alemtuzumab) (Tab. [5](#Tab5){ref-type="table"}).
### Respiratorische Viren {#Sec24}
Präventive Strategien sind nur für Influenzaviren definiert. Eine lebenslange Influenza-Vakzinierung ist indiziert. Zusätzlich ist eine Prophylaxe bzw. präemptive Therapie bei Ausbrüchen für alle Patienten bis zu zwei Jahre nach Transplantation bzw. für Patienten mit chronischer GvHD und systemischer Immunsuppression auch darüber hinaus indiziert. Mittel der Wahl ist Oseltamivir, alternativ Zanamivir.
Die Gabe von Oseltamivir ist ebenfalls bei bereits an einer Influenzainfektion der oberen Atemwege erkrankten Patienten präventiv hinsichtlich einer Ausbreitung hin zur Pneumonie und verkürzt die Virusausscheidung (Nichols et al. 2004).
Pilze {#Sec25}
-----
Ein Risiko besteht vor allem von der Konditionierung bis zum Engraftment sowie in der späten Phase bei GvHD bzw. systemischer Immunsuppression. Daher erhalten alle Patienten mit allogener Stammzelltransplantation eine Prophylaxe in Form von Fluconazol 400 mg/Tag oder Posaconazol 3 × 200 mg oral. Letztere ist in Behandlungszentren mit hohen Raten Fluconazol-resistenter Candida-Stämme und bei Patienten mit prolongierter Neutropenie bzw. GvHD zu bevorzugen (Tacke et al. 2014).
Pneumocystis jirovecii {#Sec26}
----------------------
Alle Patienten mit allogener Stammzelltransplantation erhalten eine Pneumocystis- Prophylaxe. Über sechs Monate hinaus besteht eine Indikation bei Patienten mit chronischer GvHD und systemischer Immunsuppression. Mittel der Wahl ist Cotrimoxazol; zu weiteren Optionen siehe 10.1007/978-3-662-55741-9_61\#Sec81 (Vasconceles et al. 2000; Marras et al. 2002). Wird Cotimoxazol gegeben, ist eine zusätzliche antibakterielle Prophylaxe mit Ciprofloxacin nicht indiziert.
Toxoplasma gondii {#Sec27}
-----------------
Seropositive Patienten haben ein Risiko und eine Indikation für eine Prophylaxe. Diese beginnt mit dem Engraftment und wird für mindestens sechs Monate fortgesetzt, bei fortgesetzter Immunsuppression auch länger. Cotrimoxazol ist auch gegen Toxoplasma wirksam. Alternativ kommt Clindamycin/Pyrimethamin in Frage.
Nichtinfektiöse pulmonale Komplikationen {#Sec28}
========================================
Differentialdiagnostisch müssen nichtinfektiöse Komplikationen in Betracht gezogen werden. Der radiologischen Bildgebung kommt dabei eine hohe Bedeutung zu (Peña et al. 2014) (10.1007/978-3-662-55741-9_58). Eine CT sollte immer in In- und Exspiration angefertigt werden, um nach dem differentialdiagnostisch wichtigen Muster der Mosaikperfusion zu fahnden.
Das idiopathische Pneumonie-Syndrom (IPS) {#Sec29}
-----------------------------------------
Das IPS ist nach ATS definiert als ein schwerer Alveolarschaden, ohne dass eine pulmonale Infektion und ohne dass eine akute Herz- oder Niereninsuffizienz bestehen. Die Definition schließt sehr konkret ein, nach welchen Methoden eine Infektion ausgeschlossen sein muss. Unter diesem Oberbegriff folgen dann Kategorien des IPS entsprechend einem angenommenen Schaden auf Ebene des Parenchyms, Epi- oder Endothels (Panoskaltsis-Mortari et al. 2011) (Tab. [6](#Tab6){ref-type="table"}).LungenparenchymVaskuläres EndothelAtemwegsepithelAkute interstitielle Pneumonitis (AIP)Peri-engraftment respiratory distress syndrome (PERDS)Kryptogene organisierende Pneumonie (COP)Acute respiratory distress syndrome (ARDS)Nichtkardiogenes Kapillarlecksyndrom (CLS)Bronchiolitis obliterans (BO)BCNU-PneumonitisDiffuse alveoläre Hämorrhagie (DAH)StrahlenpneumonitisPulmonale veno-okklusive Erkrankung (PVOD)Delayed pulmonary toxicity syndrome (DPTS)Transfusion-related acute lung injury (TRALI)Post-Transplant lymphoproliferative Erkrankung (PTLD)Pulmonale zytolytische Thromben (PCT)Eosinophile Pneumonie (EP)Pulmonale arterielle Hypertonie (PAH)Alveolarproteinose (PAP)Pulmonale Thromboembolie (PTE)
Das IPS tritt innerhalb der ersten 120 Tage in 3--15 % der Fälle auf. Nach myeloablativer Konditionierung ist es deutlich häufiger als nach einer nichtmyeloablativen (8,4 vs. 2,2 %). Bestrahlung und alloreaktive T-Zellen stellen somit für die Lunge ein erhebliches Risiko dar.
Das IPS tritt im Median nach 19 Tagen auf (Spanne 4--106), ist also eine überwiegend frühe Komplikation. Die Prognose ist mit 60--80 % Letalität sehr schlecht, bei Patienten unter invasiver Beatmung mit mehr als 95 % nahezu infaust.
Als Risikofaktoren wurden identifiziert: Ganzkörperbestrahlung, akute GvHD, höheres Lebensalter sowie Leukämien und MDS als Grunderkrankungen.
Therapeutisch werden Steroide gegeben, jedoch sind die Ansprechraten gering.
Das IPS kommt auch bei autologer Transplantation vor, weist jedoch deutliche Unterschiede auf: Die Inzidenz ist deutlich geringer, der Zeitpunkt des Auftretens überwiegend später (im Median nach 63 Tagen; Spane 7--336 Tage), und spricht auf eine Steroidtherapie sehr gut an.
Alveoläre Hämorrhagie {#Sec30}
---------------------
Diese ist eine frühe Komplikation und tritt meist innerhalb der ersten 14 Tage nach Transplantation auf. Klinisch äußert sich die alveoläre Hämorrhagie mit Dyspnoe, Fieber, Husten und (selten) Hämoptysen. Die Letalität beträgt bis ca. 50 %.
Die Hämorrhagie kann Ausdruck einer pulmonalen Infektion sein (Gupta et al. 2007).
Die Diagnose setzt sich aus folgenden Kriterien zusammen:CT mit diffusen bilateralen Verdichtungen,zunehmend blutiger imponierende rückgewonnene Portionen der BAL sowie\>20 % Hämosiderin-beladene Makrophagen in der BALF.
Tritt sie innerhalb der ersten 30 Tage auf, ist dies günstiger. Die Therapie besteht in der Gabe von Steroiden (Afessa et al. 2002a, b).
COP und Bronchiolitis obliterans {#Sec31}
--------------------------------
Die COP (früher: Bronchiolitis obliterans mit organisierender Pneumonie, BOOP) weist unter Steroidtherapie eine gute Prognose auf. Ihr funktionelles Schädigungsmuster ist restriktiv. Die Bronchiolitis obliterans hingegen weist ein obstruktives Schädigungsmuster auf, ist meist irreversibel und mit einer hohen Letalität assoziiert (Soubani und Uberti 2007).
Andere {#Sec32}
------
Der Begriff IPS schließt zusätzlich nahezu alle möglichen nichtinfektiösen Lungenkomplikationen ein. Zudem müssen lediglich noch die Überwässerung durch Herz- und Niereninsuffizienz in Betracht gezogen werden. Eine Überwässerung ist die häufigste Komplikation in der zweiten bis dritten Woche nach Transplantation.
Weiterführende Literatur {#Sec33}
========================
Umfassendes Dokument zur Prävention infektiöser Komplikationen bei Patienten mit Stammzelltransplantation. Enthält ebenso detaillierte Ausführungen zur Immunsuppression und Dynamik der Immunrekonstitution:Tomblyn M, Chiller T, Einsele H, Gress R, Sepkowitz K, Storek J, Wingard JR, Young JA, Boeckh MJ, Center for International Blood and Marrow Research, National Marrow Donor program, European Blood and MarrowTransplant Group, American Society of Blood and Marrow Transplantation, Canadian Blood and Marrow Transplant Group, Infectious Diseases Society of America, Society for Healthcare Epidemiology of America, Association of Medical Microbiology and Infectious Disease Canada, Centers for Disease Control and Prevention (2009) Guidelines for preventing infectious complications among hematopoietic cell transplantation recipients: a global perspective. Biol Blood Marrow Transplant 15:1143--1238
Zeittafel des Infektionsrisikos nach allogener Stammzelltranplantation:Küpeli E, Eyüboğlu FÖ, Haberal M (2012) Pulmonary infections in transplant recipients. Curr Opin Pulm Med 18:202--212
Dynamik der Immunrekonstitution:Storek J (2008) Immunological reconstitution after hematopoietic cell transplantation -- its relation to the contents of the graft. Expert Opin Biol Ther 8:583--597
Erreger:Nguyen Q, Champlin R, Giralt S, Rolston K, Raad I, Jacobson K, Ippoliti C, Hecht D, Tarrand J, Luna M, Whimbey E (1999) Late cytomegalovirus pneumonia in adult allogeneic blood and marrow transplant recipients. Clin Infect Dis 28:618--623Styczynski J, Reusser P, Einsele H, de la Camara R, Cordonnier C, Ward KN, Ljungman P, Engelhard D, Second European Conference on Infections in Leukemia (2009) Management of HSV, VZV and EBV infections in patients with hematological malignancies and after SCT: guidelines from the second European conference on infections in leukemia. Bone Marrow Transplant 43:757--770Buchbinder S, Elmaagacli AH, Schaefer UW, Roggendorf M (2000) Human herpesvirus 6 is an important pathogen in infectious lung disease after allogeneic bone marrow transplantation. Bone Marrow Transplant 26:639--644Ljungman P, Ward KN, Crooks BN, Parker A, Martino R, Shaw PJ, Brinch L, Brune M, De La Camara R, Dekker A, Pauksen K, Russell N, Schwarer AP, Cordonnier C (2001) Respiratory virus infections after stem cell transplantation: a prospective study from the infectious diseases working party of the European group for blood and marrow transplantation. Bone Marrow Transplant 28:479--484Ljungman P, de la Camara R, Perez-Bercoff L, Abecasis M, Nieto Campuzano JB, Cannata-Ortiz MJ, Cordonnier C, Einsele H, Gonzalez-Vicent M, Espigado I, Halter J, Martino R, Mohty B, Sucak G, Ullmann AJ, Vázquez L, Ward KN, Engelhard D, Infectious Diseases Working Party, European Group for Blood and Marrow Transplantation, Infectious Complications Subcommittee, Spanish Group of Haematopoietic Stem-cell Transplantation (2011) Outcome of pandemic H1N1 infections in hematopoietic stem cell transplant recipients. Haematologica 96:1231--1235Wendt CH, Weisdorf DJ, Jordan MC, Balfour HH Jr, Hertz MI (1992) Parainfluenza virus respiratory infection after bone marrow transplantation. N Engl J Med 326:921--926Lewis VA, Champlin R, Englund J, Couch R, Goodrich JM, Rolston K, Przepiorka D, Mirza NQ, Yousuf HM, Luna M, Bodey GP, Whimbey E (1996) Respiratory disease due to parainfluenza virus in adult bone marrow transplant recipients. Clin Infect Dis 23:1033--1037Bruno B, Gooley T, Hackman RC, Davis C, Corey L, Boeckh M (2003) Adenovirus infection in hematopoietic stem cell transplantation: effect of ganciclovir and impact on survival. Biol Blood Marrow Transplant 9:341--352Martino R, Maertens J, Bretagne S, Rovira M, Deconinck E, Ullmann AJ, Held T, Cordonnier C (2000a) Toxoplasmosis after hematopoietic stem cell transplantation. Clin Infect Dis 31:1188--1195Martino R, Bretagne S, Rovira M, Ullmann AJ, Maertens J, Held T, Deconinck E, Cordonnier C (2000b) Toxoplasmosis after hematopoietic stem transplantation. Report of a 5-year survey from the infectious diseases working party of the European group for blood and marrow transplantation. Bone Marrow Transplant 25:1111--1114
Diagnostische Ausbeute der Bronchoskopie:Sirithanakul K, Salloum A, Klein JL, Soubani AO (2005) Pulmonary complications following hematopoietic stem cell transplantation: diagnostic approaches. Am J Hematol 80:137--465Dunagan DP, Baker AM, Hurd DD, Haponik EF (1997) Bronchoscopic evaluation of pulmonary infiltrates following bone marrow transplantation. Chest 111:135--141Huaringa AJ, Leyva FJ, Signes-Costa J, Morice RC, Raad I, Darwish AA et al (2000) Bronchoalveolar lavage in the diagnosis of pulmonary complications of bone marrow transplant patients. Bone Marrow Transplant 25:975--979Patel NR, Lee PS, Kim JH, Weinhouse GL, Koziel H (2005) The influence of diagnostic bronchoscopy on clinical outcomes comparing adult autologous and allogeneic bone marrow transplant patients. Chest 127:1388--1396Bissinger AL, Einsele H, Hamprecht K, Schumacher U, Kandolf R, Loeffler J, Aepinus C, Bock T, Jahn G, Hebart H (2005) Infectious pulmonary complications after stem cell transplantation or chemotherapy: diagnostic yield of bronchoalveolar lavage. Diagn Microbiol Infect Dis 52:275--280Shannon VR, Andersson BS, Lei X, Champlin RE, Kontoyiannis DP (2010) Utility of early versus late fiberoptic bronchoscopy in the evaluation of new pulmonary infiltrates following hematopoietic stem cell transplantation. Bone Marrow Transplant 45:647--655Gilbert CR, Lerner A, Baram M, Awsare BK (2013) Utility of flexible bronchoscopy in the evaluation of pulmonary infiltrates in the hematopoietic stem cell transplant population -- a single center fourteen year experience. Arch Bronconeumol 49:189--195Lucena CM, Torres A, Rovira M, Marcos MA, de la Bellacasa JP, Sánchez M, Domingo R, Gabarrus A, Mensa J, Agustí C (2014) Pulmonary complications in hematopoietic SCT: a prospective study. Bone Marrow Transplant 49:1293--1299Wahla AS, Chatterjee A, Khan II, Conforti JF, Haponik E (2014) Survey of academic pulmonologists, oncologists, and infectious disease physicians on the role of bronchoscopy in managing hematopoietic stem cell transplantation patients with pulmonary infiltrates. J Bronchology Interv Pulmonol 21:32--39
Feinnadelaspirationen:Jantunen E, Piilonen A, Volin L, Ruutu P, Parkkali T, Koukila-Kähkölä P, Ruutu T (2002) Radiologically guided fine needle lung biopsies in the evaluation of focal pulmonary lesions in allogeneic stem cell transplant recipients. Bone Marrow Transplant 29:353--356
Chirurgische BiopsienZihlif M, Khanchandani G, Ahmed HP, Soubani AO (2005) Surgical lung biopsy in patients with hematological malignancy or hematopoietic stem cell transplantation and unexplained pulmonary infiltrates: improved outcome with specific diagnosis. Am J Hematol 78:94--99
Präventive und antivirale Therapie der CMV:Boeckh M, Ljungman P (2009) How we treat cytomegalovirus in hematopoietic cell transplant recipients. Blood 113:5711--5779Boeckh M (2011) Complications, diagnosis, management, and prevention of CMV infections: current and future. Hematology Am Soc Hematol Educ Program 2011:305--309Solano C, Giménez E, Piñana JL, Vinuesa V, Poujois S, Zaragoza S, Calabuig M, Navarro D (2016) Preemptive antiviral therapy for CMV infection in allogeneic stem cell transplant recipients guided by the viral doubling time in the blood. Bone Marrow Transplant 51:718--721Ullmann AJ, Schmidt-Hieber M, Bertz H, Heinz WJ, Kiehl M, Krüger W, Mousset S, Neuburger S, Neumann S, Penack O, Silling G, Vehreschild JJ, Einsele H, Maschmeyer G, Infectious Diseases Working Party of the German Society for Hematology and Medical Oncology (AGIHO/DGHO) and the DAG-KBT (German Working Group for Blood and Marrow Transplantation) (2016) Infectious diseases in allogeneic haematopoietic stem cell transplantation: prevention and prophylaxis strategy guidelines 2016. Ann Hematol 95:1435--1455Schmidt GM, Horak DA, Niland JC, Duncan SR, Forman SJ, Zaia JA (1991) A randomized, controlled trial of prophylactic ganciclovir for cytomegalovirus pulmonary infection in recipients of allogeneic bone marrow transplants; The City of Hope-Stanford-Syntex CMV Study Group. N Engl J Med 324:1005--1011Rubin RH (1991) Preemptive therapy in immunocompromised hosts. N Engl J Med 324:1057--1059Emanuel D, Cunningham I, Jules-Elysee K, Brochstein JA, Kernan NA, Laver J, Stover D, White DA, Fels A, Polsky B (1988) Cytomegalovirus pneumonia after bone marrow transplantation successfully treated with the combination of ganciclovir and high-dose intravenous immune globulin. Ann Intern Med 109:777--782Reed EC, Bowden RA, Dandliker PS, Lilleby KE, Meyers JD (1988) Treatment of cytomegalovirus pneumonia with ganciclovir and intravenous cytomegalovirus immunoglobulin in patients with bone marrow transplants. Ann Intern Med 109:783--788Maffini E, Giaccone L, Festuccia M, Brunello L, Busca A, Bruno B (2016) Treatment of CMV infection after allogeneic hematopoietic stem cell transplantation. Expert Rev Hematol 9:585--596El Chaer F, Shah DP, Chemaly RF (2016) How I treat resistant cytomegalovirus infection in hematopoietic cell transplantation recipients. BloodLjungman P, Deliliers GL, Platzbecker U, Matthes-Martin S, Bacigalupo A, Einsele H, Ullmann J, Musso M, Trenschel R, Ribaud P, Bornhäuser M, Cesaro S, Crooks B, Dekker A, Gratecos N, Klingebiel T, Tagliaferri E, Ullmann AJ, Wacker P, Cordonnier C (2001) Cidofovir for cytomegalovirus infection and disease in allogeneic stem cell transplant recipients. The infectious diseases working party of the European group for blood and marrow transplantation. Blood 97:388--392Boeckh M, Murphy WJ, Peggs KS (2015) Recent advances in cytomegalovirus: an update on pharmacologic and cellular therapies. Biol Blood Marrow Transplant 21:24--29Marty FM, Ljungman P, Papanicolaou GA, Winston DJ, Chemaly RF, Strasfeld L, Young JA, Rodriguez T, Maertens J, Schmitt M, Einsele H, Ferrant A, Lipton JH, Villano SA, Chen H, Boeckh M, Maribavir 1263--300 Clinical Study Group (2011) Maribavir prophylaxis for prevention of cytomegalovirus disease in recipients of allogeneic stem-cell transplants: a phase 3, double-blind, placebo-controlled, randomised trial. Lancet Infect Dis 11:284--292Chemaly RF, Ullmann AJ, Stoelben S, Richard MP, Bornhäuser M, Groth C, Einsele H, Silverman M, Mullane KM, Brown J, Nowak H, Kölling K, Stobernack HP, Lischka P, Zimmermann H, Rübsamen-Schaeff H, Champlin RE, Ehninger G, AIC246 Study Team (2014) Letermovir for cytomegalovirus prophylaxis in hematopoietic-cell transplantation. N Engl J Med 370:1781--1789Marty FM, Winston DJ, Rowley SD, Vance E, Papanicolaou GA, Mullane KM, Brundage TM, Robertson AT, Godkin S, Momméja-Marin H, Boeckh M, CMX001-201 Clinical Study Group (2013) CMX001 to prevent cytomegalovirus disease in hematopoietic-cell transplantation. N Engl J Med 369:1227--1136
Influenza:Nichols WG, Guthrie KA, Corey L, Boeckh M (2004) Influenza infections after hematopoietic stem cell transplantation: risk factors, mortality, and the effect of antiviral therapy. Clin Infect Dis 39:1300--1306
Pilze:Tacke D, Buchheidt D, Karthaus M, Krause SW, Maschmeyer G, Neumann S, Ostermann H, Penack O, Rieger C, Ruhnke M, Sandherr M, Schweer KE, Ullmann AJ, Cornely OA (2014) Primary prophylaxis of invasive fungal infections in patients with haematologic malignancies. 2014 update of the recommendations of the Infectious Diseases Working Party of the German Society for Haematology and Oncology. Ann Hematol 93:1449--1456
Pneumocystis jirovecii (Prophlyaxe mit Pentamidin-Aerosol):Vasconcelles MJ, Bernardo MV, King C, Weller EA, Antin JH (2000) Aerosolized pentamidine as pneumocystis prophylaxis after bone marrow transplantation is inferior to other regimens and is associated with decreased survival and an increased risk of other infections. Biol Blood Marrow Transplant 6:35--43Marras TK, Sanders K, Lipton JH, Messner HA, Conly J, Chan CK (2002) Aerosolized pentamidine prophylaxis for Pneumocystis carinii pneumonia after allogeneic marrow transplantation. Transpl Infect Dis 4:66--74
Forschungsbericht zum Idiopathischen Pneumonie Syndrom (IPS):Panoskaltsis-Mortari A, Griese M, Madtes DK, Belperio JA, Haddad IY, Folz RJ, Cooke KR, American Thoracic Society Committee on Idiopathic Pneumonia Syndrome (2011) An official American thoracic society research statement: noninfectious lung injury after hematopoietic stem cell transplantation: idiopathic pneumonia syndrome. Am J Respir Crit Care Med183:1262--1279
Exzellente Übersicht über die Rolle der Bildgebung in der Diagnostik nicht-infektiöser Komplikationen:Peña E, Souza CA, Escuissato DL, Gomes MM, Allan D, Tay J, Dennie CJ (2014) Noninfectious pulmonary complications after hematopoietic stem cell transplantation: practical approach to imaging diagnosis. Radiographics 34:663--683
Zur Komplikation der alveolären Hämorrhagie:Gupta S, Jain A, Warneke CL, Gupta A, Shannon VR, Morice RC, Onn A, Jimenez CA, Bashoura L, Giralt SA, Dickey BF, Eapen GA (2007) Outcome of alveolar hemorrhage in hematopoietic stem cell transplant recipients. Bone Marrow Transplant 40:71--78Afessa B, Tefferi A, Litzow MR, Peters SG (2002a) Outcome of diffuse alveolar hemorrhage in hematopoietic stem cell transplant recipients. Am J Respir Crit Care Med 166:1364--1368Afessa B, Tefferi A, Litzow MR, Krowka MJ, Wylam ME, Peters SG (2002b) Diffuse alveolar hemorrhage in hematopoietic stem cell transplant recipients. Am J Respir Crit Care Med 166:641--645
Bronchiolitis obliterans als Spätkomplikation:Soubani AO, Uberti JP (2007) Bronchiolitis obliterans following haematopoietic stem cell transplantation. Eur Respir J 29:1007--1019
| {
"pile_set_name": "PubMed Central"
} |
{#f8}
Introduction {#s1}
============
Nonalcoholic fatty liver disease (NAFLD), a result of the pandemic of obesity and diabetes, has become a leading cause of liver disease and emerged a major challenge in modern society. It affects 6--45% of the general population worldwide with highly prevalence and incidence ([@B5]). NAFLD is one of the most common chronic liver diseases, ranging from simple steatosis to nonalcoholic steatohepatitis (NASH), and eventually to hepatocellular carcinoma (HCC) ([@B29]). Hepatic steatosis occurs in the early stages of NAFLD, which is defined as abnormal lipid deposition in the liver. When the rate of hepatic lipid uptake exceeds the rate of lipid disposal, lipid accumulates in the liver and results in steatosis ([@B11]). Very low-density lipoprotein (VLDL) metabolism is considered as a beneficial factor in overcoming the excess formation of triglyceride (TG) and regulating intrahepatic and plasma lipid homeostasis. A dysregulation in VLDL uptake, export, or synthesis is one of the major causes of hepatic steatosis ([@B1]). Current studies have demonstrated that endoplasmic reticulum (ER) stress is implicated in the inhibition of VLDL metabolism. ER stress interferes with VLDL metabolism in several manners, including enhancing VLDL delivery to hepatocytes and inhibiting VLDL synthesis and export from these cells. Free fatty acids (FFAs) from diet are thought to be crucial for the onset of hepatic steatosis. Elevated level of FFAs results in disruption of ER homeostasis and VLDL metabolism, bringing about TG accumulation in the liver and finally leading to hepatic steatosis ([@B25]; [@B6]).
To date, there are no effective medical intervention strategies for NAFLD. Natural compounds are deemed as viable treatment regimens to inhibit the progress of NAFLD because of the beneficial effects they have shown ([@B15]; [@B32]). *Pogostemon cablin* (Blanco) Benth. (Labiatae) is a widely used traditional healthy food and medicinal herb in Asian countries such as China, Malaysia, and India. Its fresh leaves and dried powder are used in the form of food flavour supplements, tea, beverages, candy, baked products, and common botanical ingredients in functional foods and dietary supplements ([@B8]; [@B19]). It has been reported to display excellent anti-inflammatory, anti-oxidative, and multiple-organ protective activities ([@B31]; [@B3]). Our previous study also confirmed the protective effect of patchouli oil (the extractive from the dry leaves of *Pogostemon cablin*) against lipid accumulation in a rat model of alcoholic liver injury (ALI) ([@B10]). Patchouli alcohol (PA, [**Figure 1A**](#f1){ref-type="fig"}), as the phytochemical marker determining the quality of *Pogostemon cablin* and patchouli oil, has been demonstrated to possess various medicinal activities ([@B28]; [@B16]). However, the mechanism of PA action in NAFLD by reducing hepatic steatosis still remains uncertain. Therefore, this study aimed to evaluate if PA supplementation to a HFD would reduce hepatic steatosis by alleviating ER stress and regulating VLDL metabolism in rats.
{#f1}
Materials and Methods {#s2}
=====================
Drugs and Chemicals {#s2_1}
-------------------
PA was isolated from patchouli oil according to published article at purity of 99.0% ([@B20]). Vitamin E (VE; purity ≥ 98%) was purchased from Dalian Meilun Biological Technology Co. Ltd (Dalian, Liaoning, China). Normal diet (ND; ≥ 4% energy as fat) was purchased from the Medical Experiment Animal Center of Guangzhou University of Chinese Medicine, and high fat diet (HFD, D12492, 60% energy as fat) was purchased from Guangdong Medical Lab Animal Center (Guangzhou, Guangdong, China). The kits for biochemical analysis of aspartate transaminase (AST), alanine aminotransferase (ALT), triglyceride (TG), total cholesterol (TC), superoxide dismutase (SOD), glutathione (GSH) catalase (CAT), and malondialdehyde (MDA) were obtained from Nanjing Jiancheng Bioengineering Institute (Nanjing, Jiangsu, China). ELISA kits for free fatty acid (FFA), reactive oxygen species (ROS), apolipoprotein B100 (apoB 100), and VLDL measurement were purchased from Shanghai Enzyme-linked Biotechnology Co. Ltd (Shanghai, China). Primers for determination of mRNA expressions of glucose-regulated protein 78 kDa (GRP78), protein kinase-like ER kinase (PERK), inositol-requiring transmembrane kinase/endoribonuclease 1 (IRE1), eukaryotic translation initiation factor 2α (eIF2α), activating transcription factor 4 (ATF4), very low-density lipoprotein receptor (VLDLR), X box binding protein 1 (XBP1), protein disulfide isomerase (PDI), microsomal triglyceride-transfer protein (MTP), activating transcription factor 6 (ATF6), apoB 100, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were provided by Sangon Biotech Co., Ltd (Shanghai, China). Primary antibodies against GRP78 (AF5366), PERK (AF5304), p-PERK (DF7576), eIF2α (AF6087), p-eIF2α (AF3087), IRE1 (DF7709), p-IRE1 (DF8322), ATF4 (DF6008), XBP1 (AF5110), MTP (DF6591), ATF6 (DF6009), β-actin (AF7018), and goat anti-rabbit IgG (H+L) HRP (S0001) secondary antibody were obtained from Affinity Biosciences Inc. (USA). Primary antibody against VLDLR (19493-1-AP) was purchased from Proteintech Group Inc. (USA). Primary antibody against PDI (CY6636) was purchased from Abways Technology Inc. Primary antibody against apoB 100 (A1330) was purchased from ABclonal Technology Inc. (USA). All reagents were of analytical grade.
Animals and Treatments {#s2_2}
----------------------
Animal experiment procedures were approved by the Ethics Committee for the Welfare of Experimental Animals of Guangzhou University of Chinese Medicine (No. 20181015002). Male Sprague Dawley rats were purchased from the Medical Experiment Animal Center of Guangzhou University of Chinese Medicine (SYXK (YUE) 2018-0085). After 7 days acclimation, rats were randomly divided into six experimental groups (*n* = 8): ND, HFD, HFD supplemented with VE (100 mg/kg), and HFD supplemented with different doses of PA (10, 20, and 40 mg/kg), respectively. VE was used as positive control. PA and VE were dissolved in 0.5% tween 80 solution. Except the ND group (fed with normal diet), all rats were fed with HFD and water *ad libitum* for a period of 4 weeks to induce NAFLD. Meanwhile, rats received intragastric administration once daily. The body weights of rats were recorded daily, and food intake were recorded every week throughout the study. At the last day, all rats were weighted and anesthetized with sodium pentobarbital after fasting overnight. Blood samples from each rat were collected for biochemical analysis and their livers were rapidly dissected for histological evaluation and further analysis.
Serum Biochemistry {#s2_3}
------------------
Blood samples were centrifuged at 3000 rpm at 4°C for 10 min, and the supernatants were collected for serum biochemistry examination. The serum levels of ALT, AST, TG, and TC were measured with commercial assay kits according to manufacturer's instructions by microplate reader.
Histopathological Analysis {#s2_4}
--------------------------
After fixed with 4% paraformaldehyde, liver segments were dehydrated, cleaned, and embedded in paraffin. Then, the liver slices of 5 μm thickness were stained with hematoxylin and eosin (H&E) and 5-μm-thick frozen sections were stained with Oil red O. Analysis were performed under a light microscope according to the method previously reported ([@B4]; [@B9]).
Hepatic Biochemical Analysis {#s2_5}
----------------------------
The liver tissues were homogenized in ice-physiological saline or absolute ethanol, and centrifuged at 4000 rpm and 4°C for 10 min to obtain supernatant for further analysis. Liver SOD, GSH, CAT, MDA, TG, and TC levels were determined using commercial assay kits according to corresponding product specifications and analyzed by microplate reader.
Enzyme Linked Immunosorbent Assay (ELISA) {#s2_6}
-----------------------------------------
Liver tissues were homogenized in phosphate buffer saline (pH 7.2--7.4) and then centrifuged at 4000 rpm at 4°C for 10 min to obtain supernatants for hepatic FFA, ROS, and VLDL examination. Blood samples were centrifuged at 3000 rpm at 4°C for 10 min, and the supernatants were collected for serum VLDL and apoB 100 examination. Hepatic levels of FFA, ROS, and VLDL as well as serum levels of VLDL and apoB 100 were measured by ELISA kits using a microplate reader.
Western Blot Analysis {#s2_7}
---------------------
Hepatic proteins were extracted using a commercial protein extraction kit (Servicebio, Wuhan, Hubei, China). After denaturation, proteins were separated by electrophoresis on SDS-PAGE gels and transferred to PVDF membranes. Membranes were blocked for 1 h with 5% (w/v) skim milk in Tris-buffered saline-tween 20 (TBST) and incubated with 1: 1000 dilution of primary antibodies and 1:3000 dilution of HRP-conjugated secondary antibody. Protein bands were visualized with ECL reagents (AmershamBiosciences, Buckinghamshire, UK), and densitometry analysis was performed using Quantity One 4.6.2 software. All blots were quantified and normalized against β-actin to adjust for the amount of proteins loaded.
Reverse Transcription-Quantitative Polymerase Chain Reaction (RT-qPCR) {#s2_8}
----------------------------------------------------------------------
Total RNA was extracted from liver tissue using TRI-zol^®^ reagent and further synthesized as cDNA. The cDNA was used as the template for real-time PCR. The mRNA expressions of GRP78, PERK, eIF2α, IRE1, ATF4, VLDLR, XBP1, PDI, MTP, ATF6, and apoB 100 (the primers of PCR are listed in [**Table 1**](#T1){ref-type="table"}) were measured by HiScript^®^ II Q RT SuperMix (+gDNA wiper) and ChamQ^™^ SYBR^®^ qPCR Master Mix Kit according to manufacturer's instruction (Vazyme Biotech, China). Each reaction was conducted in triplicate under the following cycling conditions: 10 min at 95°C, followed by 40 cycles at 95°C for 15 s and then 60°C for 1 min. The relative expression levels of target genes were normalized by GAPDH. The relative quantification of gene expression was calculated using the 2^−ΔΔCt^ method.
######
Primer sequences.
Gene Forward primer (5'-3') Reverse primer (5'-3')
---------- -------------------------- --------------------------
GRP78 CGGAGGAGGAGGACAAGAAGGAG ATACGACGGTGTGATGCGGTTG
PERK CGCTGCTGCTGCTGTTCCTG GCAATGCCTCGGCGTCTTCC
IRE1 GACGAGCATCCGAATGTGATCCG GAGGTGGTCTGATGAAGCAAGGTG
ATF6 GGCTTCCTCCAGTTGTTCTGTCTC GCTTCTCTTCCTTCAGTGGCTCTG
eIF2α GCCGATAAGGTTACGATGCTGTGG GTAGGAAGCGCCTGTCTTGTCAAC
ATF4 GACCGAGATGAGCTTCCTGAACAG CCGCCTTGTCGCTGGAGAAC
VLDLR GACGCAGACTGTTCCGACCAATC GCAGGTTCGAGAAGGACAGTTGAC
apoB 100 TCTGACTGGTGGACTCTGACTGC TCTTGGAGAGCGTGGAGACTGAC
XBP1 AGGTCTCAGAGGCAGAGTCCAAG AAGAGGCAACAGCGTCAGAATCC
PDI CAACGTCCTGGTGCTGAAGAAGAG TGCTAGTCGGATCTCAGAGCCTTC
MTP TTCATTCAGCACCTCCGCACTTC AGTCCAGGATGGCTTCCAGTGAG
GAPDH ACGGCAAGTTCAACGGCACAG CGACATACTCAGCACCAGCATCAC
Statistical Analyses {#s2_9}
--------------------
Data were presented as mean ± SD. Statistical analyses were performed using Statistical Product and Service Solutions (SPSS) software (version 20.0). Group differences were assessed by one-way analysis of variance (ANOVA) followed by an LSD test for multiple comparisons. *P* \< 0.05 was considered statistical significant.
Results {#s3}
=======
Effect of PA on Body Weight, Liver Weight, Liver Index, and Lipid Accumulation in HFD-Fed Rats {#s3_1}
----------------------------------------------------------------------------------------------
Chronic exposure to HFD disturbed lipid homeostasis in a time-dependent manner leading to development of NAFLD. Data in [**Figures 1B--D**](#f1){ref-type="fig"} showed that a profound increase in body weight, liver weight, and liver index was observed in the HFD group. VE treatment did not alter the increase of body weight and liver weight, but significantly decreased liver index in rats. Furthermore, PA treatment dramatically reduced body weight, liver weight, and liver index in HFD-fed rats. Data in [**Figure 1E**](#f1){ref-type="fig"} showed that there was no significant difference in the food intake among all groups. As shown in [**Figures 1F--K**](#f1){ref-type="fig"}, compared to the rats only fed with ND, the serum levels of AST, ALT, TG, and TC along with hepatic levels of TG and TC were significantly increased in the HFD-fed rats. These lipid parameters in VE and PA treated groups were lower than the HFD group. Moreover, rats with 40 mg/kg PA treatment exhibited superior effect in the reduction of serum ASL, ALT, TG, and TC among all treated groups. Histological assessments of liver tissue in [**Figure 2**](#f2){ref-type="fig"} showed presence of NAFLD, characterized by increased vesicular lipid droplets, hepatic vacuoles, and slight inflammatory infiltrate in HFD-fed rats. VE and PA supplementation for 4 weeks markedly decreased vacuoles, lipid droplets area, and inflammation in the liver of HFD-fed rats. These results indicated that PA was effective in reducing HFD-induced body weight gain and preventing hepatic steatosis in rats.
{#f2}
Effect of PA on FFA and Oxidative Stress in HFD-Fed Rats {#s3_2}
--------------------------------------------------------
ER stress is often associated with FFA and oxidative stress. To explore the role of PA against elevated FFA and oxidative stress, we examined the hepatic levels of FFA and the oxidative stress indicators: ROS and MDA. As shown in [**Figures 3A--C**](#f3){ref-type="fig"}, higher levels of FFA, ROS, and MDA were presented in the HFD-fed rats when compared to the ND-fed rats. However, the PA or VE groups showed similar hepatic levels of FFA, ROS, and MDA to ND group when compared to the HFD group. Data in [**Figures 3D--F**](#f3){ref-type="fig"} showed that the hepatic levels of GSH, SOD, and CAT were lower in the HFD-fed rats than ND-fed rats. Conversely, VE or PA supplementation dramatically increased the activities of SOD and CAT in HFD-fed rats. Moreover, the level of GSH was normalized in VE or PA treated rats, whereas there was no significant difference between HFD-fed rats and VE treated rats. Collectively, these results indicated that PA exerted a protective effect against elevated FFA and oxidative stress in HFD-fed rats.
{#f3}
Effect of PA on ER Stress and VLDL Secretion in HFD-Fed Rats {#s3_3}
------------------------------------------------------------
The enhanced FFA and oxidative stress induced by HFD-fed resulted in aggravating ER stress and decreasing VLDL secretion. As shown in [**Figures 3G and H**](#f3){ref-type="fig"}, HFD feeding decreased the hepatic and serum levels of VLDL; however, VE or PA treated rats exhibited higher hepatic and serum levels of VLDL than HFD-fed rats. Data in [**Figure 4**](#f4){ref-type="fig"} indicated that HFD-induced ER stress markers, including GRP78, PERK, IRE1, and ATF6, were inhibited by VE or PA supplementation. Data in [**Figures 4A--E**](#f4){ref-type="fig"} showed that HFD-fed rats significantly increased hepatic protein expressions of GRP78 and ATF6. These ER markers were down regulated by VE or PA treatment, whereas VE-treated rats had no significant change on GRP78 expression. In addition, VE or PA treatment decreased the ratios of p-PERK/PERK and p-IRE1α/IRE1α in HFD-fed rats, whereas there was no significant difference in the ratio of p-IRE1α/IRE1α between HFD and the treated groups. Data in [**Figures 4F--I**](#f4){ref-type="fig"} showed that the hepatic mRNA expressions of GRP78, PERK, IRE1, and ATF6 were dramatically decreased after administration with PA and VE. These results indicated that PA was a regulator for ER homeostasis and VLDL secretion.
{#f4}
Effect of PA on VLDLR Expression in HFD-Fed Rats {#s3_4}
------------------------------------------------
VLDLR plays a vital role in modulating VLDL-TG metabolism. To investigate the impact of PA on VLDLR expression, we measured the VLDLR and VLDLR-related indicators. Data in [**Figures 5A--D**](#f5){ref-type="fig"} showed that the hepatic protein expressions of ATF4, VLDLR, and the ratio of p-eIF2α/eIF2α were promoted in HFD-fed rats. In contrast, VE and PA treatment dramatically demoted the expressions of these proteins, while there was no significant difference in VLDLR and ATF4 expression between HFD and VE treated groups. Data in [**Figures 5E--G**](#f5){ref-type="fig"} showed that the hepatic mRNA expressions of ATF4, VLDLR, and eIF2α were significantly increased in HFD-fed rats, whereas these trends were completely inhibited by VE and PA treatment. These results suggested that PA may prevent hepatic steatosis by decreasing VLDLR expression and regulating VLDL metabolism.
{#f5}
Effect of PA on ApoB100 Secretion in HFD-Fed Rats {#s3_5}
-------------------------------------------------
Serum and hepatic levels of apoB 100 were assessed to evaluate the effect of PA on apoB secretion. Data in [**Figure 6**](#f6){ref-type="fig"} showed that the serum level as well as the hepatic mRNA and protein expressions of apoB 100 were markedly reduced after exposure to HFD. When rats were supplemented with VE and PA, the reduced serum level of apoB 100 as well as the protein and mRNA expressions in liver tissue of apoB 100 were enhanced. In addition, 40 mg/kg PA exhibited prominent effect comparable to other treated groups. No significant difference of the protein and mRNA expressions of apoB 100 was observed between the HFD and VE-treated groups. This suggested that under condition of PA treatment, serum and hepatic apoB 100 secretion was normalized and further beneficial for VLDL secretion.
{#f6}
Effect of PA on MTP Level in HFD-Fed Rats {#s3_6}
-----------------------------------------
To further understand the role of PA on MTP expression, we analyzed the MTP and related genes and proteins expressions in this study. Data in [**Figure 7**](#f7){ref-type="fig"} showed that the chronic stimulation with HFD caused decreases of hepatic protein and mRNA expressions of XBP1, PDI, and MTP. However, these trends were dramatically attenuated by VE and PA administration. These results indicated that PA supplementation increased MTP level and restored VLDL secretion in HFD-fed rats.
{#f7}
Discussion {#s4}
==========
The increasing intake of dietary lipid makes the occurrence of NAFLD widespread. *Pogostemon cablin*, a traditional healthy food and medicinal herb, was proved to possess hepatoprotective activity against lipid accumulation in rats. PA is a main effective component of *Pogostemon cablin*. However, its protective effect for the treatment of lipid deposition still remains elusive. In this study, it is firstly provided evidence that PA could alleviate HFD-induced hepatic steatosis by inhibiting lipid droplet formation and lipid accumulation in liver accompanied with reduced levels of TG, TC, FFA, AST, and ALT. Its key mechanism may be involved in suppressing ER stress and regulating VLDL metabolism.
ER is served as a main site of lipid synthesis and VLDL assembly. Previous studies have demonstrated hepatic ER stress in several animal models of steatosis, suggesting that ER stress may contribute to the induction of NAFLD ([@B6]). Elevation of FFA concentration and oxidative stress are common feature of NAFLD and proved to be tightly associated with ER stress. Excessive FFA not only activates cellular ER stress, but also causes oxidative stress by enhancing mitochondria-associated membranes (MAM) and increasing ROS production ([@B7]; [@B6]). Oxidative stress could perturb the redox status of ER lumen and inhibit protein folding, then acts as a trigger to ER stress ([@B2]). In response to ER stress, normal ER function in maintaining protein homeostasis becomes compromised, resulting in accumulation of unfolded or misfolded proteins and triggering unfolded protein response (UPR). During the response to ER stress, all three main brunches of UPR including PERK, IRE1, and ATF6 pathways are activated and mediates hepatic steatosis. GRP78 is an ER stress marker in liver. Under unstressed conditions, IRE1, PERK, and ATF6 are associated with GRP78 and remain inactive. Upon ER stress, GRP78 dissociates from these sensor proteins, and further actives PERK and IRE1, and regulates intramembrane proteolysis of ATF6 ([@B1]). In this study, after 4 weeks of PA administration, the increased levels of ROS and MDA were markedly decreased in HFD-fed rats. PA also increased the GSH, SOD, CAT, and VLDL levels in HFD-fed rats. Moreover, PA treatment decreased the protein and mRNA expressions of ER stress markers including GRP78, IRE1, PERK, and ATF6. This indicated that PA may restore VLDL secretion and attenuate hepatic steatosis by alleviating ER stress in HFD-induced NAFLD rats.
Recent data has revealed that the activated PERK--eIF2α--ATF4 pathway during ER stress induces hepatic steatosis *via* increase VLDLR by enhancing intracellular TG accumulation with VLDL uptake ([@B12]). VLDLR is a member of low-density lipoprotein receptor (LDLR) superfamily. It binds APOE-containing VLDL, which then converts into TG, leading to decrease lipid secretion and increase lipid accumulation ([@B13]). Previous study has demonstrated that inhibition of VLDLR upregulation can protect mice against hepatic steatosis induced by HFD feeding ([@B30]). Upon dissociation from GRP78, PERK is activated by dimerization and autophosphorylation, triggering phosphorylation of the eIF2α. Furthermore, promotion of eIF2α halts global protein translation and selectively translates ATF4 mRNA ([@B14]). ATF4 is a well-known transcription factor that mediates PERK downstream pathway and functions to increase hepatic VLDLR expression ([@B12]). In our work, PA treatment significantly reduced the mRNA and protein expressions of eIF2α, ATF4, and VLDLR. These results demonstrated that PA is able to lower VLDL uptake by down-regulating VLDLR.
ER stress not only impacts the hepatocytes uptake of VLDL but also impairs VLDL secretion. VLDL synthesis is a two-stage process. The first step in VLDL assembly is the apoB synthesis within the ER lumen following by its lipidation by MTP and the inclusion of TGs into a lipid droplet. In the second step, bulk neutral lipid, especially TGs, are added to the VLDL precursors and form lipid-rich VLDL. ApoB100 is a major protein component of VLDL, accounting for approximately one third of total lipoproteins present in VLDL. Impaired apoB 100 synthesis results in reduced VLDL synthesis, which inhibits the transport of endogenous TG from the liver to the extrahepatic, leading to TG deposition in hepatocytes ([@B24]). Hepatic apoB100 synthesis and secretion is a complex process involving ER stress. Under conditions of ER stress, hepatic lipid synthesis and secretion are affected, making a significant proportion of newly synthesized apoB100 degraded *via* the ubiquitin-proteasome-dependent degradative pathway ([@B21]). Evidences showed that ER chaperone protein such as GRP78 increased accompanied by decreasing apoB100 secretion, suggesting that there is an inverse relationship between ER stress and apoB100 secretion ([@B17]). In addition, apoB100 secretion appears to be regulated by PERK and ATF6 pathways. Activated PERK pathway is found to impair apoB100 synthesis in glucosamine-treated cells ([@B18]). Recent study has pointed out that ATF6α-knockout mice show enhanced hepatic steatosis caused by impaired formation of VLDL due to destabilized apoB100, whereas the exact mechanism of ATF6 in regulating apoB100 formation still remains unclear ([@B26]). Following exposed to HFD, the serum level as well as the protein and mRNA expressions of apoB100 were decreased in rats. However, PA treatment altered apoB100 secretion in HFD-fed rats. This alteration of apoB100 secretion was proved to beneficial for VLDL assembly.
MTP is an ER-localized lipid transfer protein, plays a crucial role in lipoprotein assembly, and acts as a cofactor to apoB100 at both stages of VLDL synthesis. It is responsible for the lipidation of the nascent apoB protein and the transfer of neutral lipids between vesicles ([@B27]). PDI is a subunit of MTP necessary for normal MTP activity. Previous studies demonstrated that MTP activity is highly dependent on PDI expression and related to IRE1α- XBP1-PDI pathway ([@B23]). IRE1α-XBP1 induce PDI expression to increase MTP activity for VLDL assembly and secretion. IRE1 appears in mammal with two isoforms: IRE1α and IRE1β. IRE1α is a transmembrane protein that possesses endoribonuclease (RNase) activity, which is responsible for production spliced XBP1 (XBP1s). After activated by ER stress, IRE1α initiates unconventional splicing of XBP1 mRNA and translates it into a potent transcription factor (XBP1s) ([@B22]). In turn, XBP1s drives a large transcriptional program to adjust the ER's protein-folding capacity according to the protein folding load in the ER lumen ([@B15]). Our data showed that HFD-induced ER stress led to defective XBP1, MTP, and PDI expressions in rats. PA administration not only improved XBP1, PDI, and MTP expressions but also decreased IRE1α expression from ER stress. These results indicated that PA may be able to reduce steatosis by attenuating MTP down-regulation and restoring VLDL secretion.
In conclusion, this study provides compelling evidence to support that PA is effective in ameliorating hepatic steatosis caused by HFD through suppressing ER stress and regulating VLDL uptake, assembly, and secretion, which is associated with the regulation of VLDLR, apoB100, and MTP expression. Given the promising preclinical findings presented in this study, we suggest that PA might play a protective role as possible therapeutic agents acting on hepatic steatosis.
Data Availability Statement {#s5}
===========================
The datasets analyzed in this manuscript are not publicly available. Requests to access the datasets should be directed to <[email protected]>.
Ethics Statement {#s6}
================
The animal study was reviewed and approved by Animal experiment procedures were approved by the Ethics Committee for the Welfare of Experimental Animals of Guangzhou University of Chinese Medicine (No. 20181015002).
Author Contributions {#s7}
====================
XW and YhL drafted and prepared the article. YhL and ZS conceived and designed the experiments; XW, NX, ML, QH, and JW performed experiments; YG, HL, and LC analyzed the data; YcL and XH prepared figures and tables.
Funding {#s8}
=======
This work was supported by grants from Science and Technology Planning Project of Guangdong Province, China (2017A050506044), Guangdong Provincial Department of Education Feature Innovation Project (2016KTSCX018), Key Disciplines Construction Projects of High-level University of Guangdong Province, Key Program for Subject Research of Guangzhou University of Chinese Medicine (XK2018016 & XK2019002), and Characteristic Cultivation Program for Subject Research of Guangzhou University of Chinese Medicine (XKP2019007).
Conflict of Interest {#s9}
====================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Abbreviations {#s10}
=============
apoB 100, apolipoprotein B100; ALT, alanine aminotransferase; AST, aspartate transaminase; ATF4, activating transcription factor 4; ATF6, activating transcription factor 6; CAT, catalase; eIF2α, eukaryotic translation initiation factor 2α; ER, endoplasmic reticulum; FFA, free fatty acid; GRP78, glucose-regulated protein 78 kDa; GSH, glutathione; HFD, high fat diet; IRE1α, inositol-requiring transmembrane kinase/endoribonuclease 1α; MDA, malondialdehyde; MTP, microsomal triglyceride-transfer protein; NAFLD, nonalcoholic fatty liver disease; ND, normal diet; PA, patchouli alcohol; PDI, protein disulfide isomerase; p-eIF2α, phospho-eIF2α; PERK, protein kinase-like ER kinase; p-IRE1α, phospho-IRE1α; p-PERK, phosphor-PERK; ROS, reactive oxygen species; SOD, superoxide dismutase; TC, Total Cholesterol; TG, triglyceride; VE, vitamin E; VLDL, very low-density lipoprotein; VLDLR, very low-density lipoprotein receptor; XBP1, X box binding protein 1.
[^1]: Edited by: Ping Liu, Shanghai University of Traditional Chinese Medicine, China
[^2]: Reviewed by: Jinghua Peng, Shanghai University of Traditional Chinese Medicine, China; Xiaoling Wang, Shanghai University of Traditional Chinese Medicine, China; Xianbo Wang, Capital Medical University, China
[^3]: This article was submitted to Ethnopharmacology, a section of the journal Frontiers in Pharmacology
| {
"pile_set_name": "PubMed Central"
} |
*Disclosure:* Authors have nothing to disclose with regard to commercial support.
Introduction {#s0005}
============
Hydatid disease is a parasitic infection caused by *Echinococcus granulosus*. It remains endemic in regions such as the Mediterranean, the Middle East, South Africa, America, and Australia. It affects various organs; the most common ones being the liver and lungs [@b0005]. Multivisceral echinococcus with cardiac involvement is exceptional [@b0010]. Cardiac location is an uncommon presentation of hydatid disease and constitutes 0.5--2% of different hydatid disease locations. The most affected areas of cardiac involvement are the left ventricle (60--70%) and right ventricle (10%), while interventricular septum, pericardium, and atria are the least affected [@b0015]. Approximately 10% of patients with cardiac hydatid cysts are symptomatic [@b0020]. Clinical presentation is polymorphic with nonspecific symptoms such as atypical chest pain. Cardiac cysts can rupture and cause anaphylactic reaction, systemic or pulmonary embolization, or pericardial tamponade [@b0020]. We report the case of a young woman with multivisceral hydatidosis revealed by a giant cardiac hydatid cyst treated surgically with success.
Case presentation {#s0010}
=================
A 36-year-old woman with a history of prolonged close proximity to dogs and sheep, reported atypical chest pain and a lack of appetite over three months. Physical examination was unremarkable with normal heart sounds and a normal pulmonary auscultation. Laboratory tests were normal. Her electrocardiogram showed sinus rhythm with negative T-waves in the inferior leads (DII, DIII and aVF). However, trans-thoracic echocardiography (TTE) showed a giant rounded cystic mass with echo-negative contents close to the posterior mitral valve covering the half of the posterior left ventricular wall and measuring 50 × 48 mm without hemodynamic consequences ([Fig. 1](#f0005 f0010){ref-type="fig"}). Coronary angiography was normal (normal coronary arteries origin, course, and termination). A thoraco-abdominal CT scan revealed a large cystic homogeneous mass related to the atrio-ventricular groove, extending to the left cardiac chambers and measuring 50 × 40 mm, a left pulmonary cyst measuring 15 × 12 mm, two bilateral breast cysts, and a right hepatic cyst ([Fig. 2](#f0015 f0020 f0025){ref-type="fig"}). Mammography confirmed the presence of well-defined right and left heterogeneous cysts with partially calcified edges ([Fig. 3](#f0030){ref-type="fig"}). Brain scan was normal. Serological tests performed with enzyme-linked immunosorbent assay (ELISA) were positive for *E. granulosus*. Ultimately, we retained a diagnosis of multivisceral hydatidosis with cardiac location. After one week of albendazole treatment, uncomplicated excision of the cardiac cyst was performed under cardiopulmonary bypass surgery. Myocardial protection was achieved through intermittent anterograde warm blood cardioplegia. The epicardial cyst was approached directly from the bottom of the heart without opening any cardiac chambers. The cyst was isolated by sponges soaked with hypertonic saline serum in order to prevent local invasion by the parasite. The germinative membrane of the hydatid cyst was completely ablated and cyst contents were totally aspirated. Hypertonic saline solution was injected into the residual cavity to kill any viable daughter vesicles. A capitonnage of the cavity wall was then made by *U*-shaped interrupted sutures ([Fig. 4](#f0035){ref-type="fig"}). Finally, an excision and a capitonnage of the left pulmonary cyst as well as an excision of the right and left breast hydatid cysts were performed. A histopathological exam of the resected tissues was positive for scolices of *E. granulosus*. Control TTE showed a residual cavity of 2 cm^2^ ([Fig. 5](#f0040){ref-type="fig"}). The postoperative course was uneventful and medical treatment with albendazole was continued over a period of six months.
Discussion {#s0015}
==========
Hydatid disease, commonly known as echinococcus or hydatidosis, remains endemic in some areas of the world. This parasitic disease is a significant public health problem in these countries. It is a tissue infestation frequently caused by the larva of *E. granulosus* [@b0020]. Humans are accidental hosts in the cycle of *E. granulosus* and are infected by handling dogs or ingesting cyst-containing meat from an intermediate host [@b0010]. The most common localizations of hydatid cysts are the liver (in 50--70% of cases) and lungs (in 5--30% of cases). But other parts of the body can also be affected [@b0005]. Multivisceral hydatidosis inside thoracic and abdominal compartments with cardiac hydatid cyst as first presentation is exceptional [@b0010]. A literature review did not reveal reports of multivisceral hydatidosis involving liver, lung, both breasts, or heart with chest pain as a revealing symptom. In our case, this was diagnosed using imaging techniques. Cardiac involvement is an uncommon presentation of hydatid cyst disease, accounting for approximately 0.5--2% of all hydatidosis cases, and mainly occurring as part of a systemic infection [@b0015]. Areas of cardiac involvement in hydatid disease include the left ventricle (60% of cases), the right ventricle (10%), the pericardium (7%), the pulmonary artery (6%), the left atrial appendage (6%), and the interventricular septum (4%) [@b0015]. Echinococcus larvae necessarily get through two filters (liver and lung) and then reach the heart mainly through the coronary circulation [@b0020; @b0025]. The second route of infestation is the pulmonary vein due to the rupture of pulmonary echinococcal cysts in the vein. The heart can also be secondarily affected by direct contact with hydatid cysts originating from the liver or the lungs [@b0025]. The left ventricle is the most frequently involved site of cardiac hydatid cysts due to the rich coronary blood supply and the good perfusion of the left ventricular myocardial mass [@b0020; @b0025]. Cardiac cysts grow toward the weaker side of the ventricular wall; either the epicardium or endocardium [@b0030]. Clinical manifestations vary according to the cyst site, size, and number and are due to related complications. Symptoms are mostly nonspecific and include atypical chest pain, breath shortness, asthenia, and palpitations. Left ventricular hydatid cysts are usually located subepicardially and may compress the small coronary arteries. Hence, chest pain can be the symptom revealing hydatid cysts and mimicking the coronary artery disease [@b0035]. Cardiac hydatid cysts may result in serious consequences such as rupture into the pericardial cavity or the cardiac chambers. Rupture into the pericardial cavity is a rare complication of the subepicardial hydatid cysts. It may be silent or cause an acute tamponade, constrictive pericarditis, or pericardial cysts. Rupture of subendocardial cysts into the circulation can cause anaphylactic reaction and may be fatal [@b0030]. The diagnosis of a subepicardial hydatid cyst revealed through chest pain was made in our case. However, there were no complications, and the angiography did not reveal compression of the coronary arteries.
Various serological tests are available for the diagnosis and postoperative follow-up of hydatid cyst recurrence, such as immunoelectrophoresis, ELISA, latex agglutination, and the indirect hemagglutination (IHA) test [@b0040]. However, previous reports have shown that serologic tests can have false-negative results, and therefore imaging modalities such as ultrasonography, CT scan, and magnetic resonance imaging (MRI) have been the methods of choice [@b0045]. For cardiac involvement, TTE is the exam of choice thanks to its availability, high sensitivity, good resolution, and the ability to detect hemodynamic repercussion. CT scan and MRI can help localize the lesion, and detect multiple lesions and multi-organ involvement [@b0025]. In our case, the diagnosis of cardiac hydatid cyst was made by TTE and ELISA tests which were positive for hydatid disease. Other multiple lesions including lung, bilateral breast and liver involvement were detected by CT scan.
Due to their localization in the myocardium or pericardium and the risk of life-threatening complications, hydatid cardiac cysts should be operated on as soon as diagnosis is made [@b0050]. Surgical treatment depends on the size, location, and number of the cysts. The main principle of surgical treatment is to empty the cyst, remove daughter cysts and the germinative membrane, excise the pericyst, and then obliterate the residual cavity with sutures (capitonnage). The use of local scolicidal solution such as hypertonic saline solution is obligatory after cysto-pericystectomy in order to minimize the risk of dispersion of cystic content [@b0050]. Supplemental medical therapy with albendazole is recommended, which has better results in preventing recurrence of hydatid disease. The duration of anti-parasitic postoperative treatment depends on intra-operative findings and the presence of complications.
Conclusion {#s0020}
==========
Multivisceral hydatidosis with cardiac involvement is an uncommon entity and has nonspecific clinical presentation. Chest pain may be a revealing symptom. Diagnosis of hydatid disease of the heart depends on a series of tests including hydatid serology, echocardiography, MRI, and CT scan. Cardiac hydatidosis should be considered in the differential diagnosis of tumoral and cystic masses. The treatment of choice is surgical excision, even in asymptomatic patients.
Peer review under responsibility of King Saud University.
{#f0005}
{#f0010}
{#f0015}
{#f0020}
{#f0025}
{#f0030}
######
Surgical treatment of the cardiac hydatid cyst: a--e. (a) Peroperative view showing a huge epicardial hydatid cyst covering the posterior left ventricular wall. (b) Total ablation of the germinative membrane of the hydatid cyst. (c) Uncompleted excision of the hydatid cyst showing the residual cavity. (d) Capitonnage of the cavity wall made by U-shaped interrupted sutures. (e) The germinative membrane and the content of the cardiac hydatid cyst.
{#f0040}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
It is becoming increasingly clear that many Lewy body disorders can be characterized by a prodromal phase during which nonmotor symptoms occur before a clinical diagnosis can be made.^[@CR1]^ The strongest prodromal symptom associated with future risk of a Lewy body disease is idiopathic REM sleep behavior disorder (iRBD), a parasomnia associated with unpleasant dreams and vigorous behaviors during REM sleep.^[@CR2]^ Several studies have shown that individuals with iRBD may be ideal candidates for neuroprotective trials since they have a near universal risk of developing a Lewy body disorder,^[@CR1],[@CR3]^ namely Parkinson's disease (PD), dementia with Lewy bodies (DLB), and multiple system atrophy (MSA). However, in order to plan such trials, it is essential to estimate when iRBD patients will convert to one of these diseases, preferably with a marker that could potentially be used across different studies.^[@CR4]^
Cortical thickness is a sensitive marker of brain atrophy that measures the distance between the gray/white and pial surfaces with submillimeter accuracy.^[@CR5]^ There is consistent evidence showing cortical thinning in frontal, temporal, occipital, or parietal areas in patients with PD, DLB, or MSA, in association with motor, nonmotor and cognitive abnormalities.^[@CR6]--[@CR11]^ However, it is still unclear whether these changes are already present in the prodromal stages of these diseases and whether they could be used to identify iRBD individuals with a higher risk of converting to one of them. Previous studies using structural magnetic resonance imaging (MRI) have shown cortical thinning in medial frontal, lateral frontal, postcentral, temporal, and occipital regions.^[@CR12]--[@CR14]^ However, these studies did not have information regarding conversion to a Lewy body disorder, which can occur at highly variable time intervals in iRBD.^[@CR15]^ In addition, they did not compare the atrophy patterns in iRBD with those observed in PD.
To address these questions, we measured cortical thickness and subcortical volumes in patients with iRBD, patients with newly diagnosed PD and healthy controls. Specifically, our aims were to: (1) compare neuroanatomical and clinical markers between iRBD patients and the other groups; (2) explore the association between brain atrophy and clinical features in iRBD; and (3) test the ability of baseline atrophy patterns to predict which iRBD individuals will progress to a Lewy body disorder over the short term (i.e., up to a 3-year period). We hypothesized that cortical and subcortical changes would be detectable in iRBD patients at baseline and that these changes would correlate with clinical deficits and predict conversion to a Lewy body disorder.
Results {#Sec2}
=======
Clinical differences between groups {#Sec3}
-----------------------------------
The clinical characteristics of our sample can be found in Table [1](#Tab1){ref-type="table"}. In total, 27 patients with iRBD, 151 patients with newly diagnosed PD and 31 healthy controls were included.Table 1Clinical characteristics of iRBD patients, PD patients and controlsiRBD (*n* = 27)PD (*n* = 151)CTR (*n* = 31)iRBD vs. CTR (*p* value)PD vs. CTR (*p* value)iRBD vs. PD (*p* value)Age (mean, SD)68.9 (5.5)60.6 (9.6)58.5 (11.0)**\<0.001**0.335**\<0.001**Sex (M/F)22/594/5720/110.1490.8120.053Education (mean, SD)12.7 (5.2)15.4 (2.9)16.5 (3.1)**0.006**0.0670.022MDS-UPDRS III (mean, SD, range)4.2 (3.6; 0--15)20.5 (9.2; 0--51)0.32 (0.9; 0--4)**\<0.001\<0.001\<0.001**Hoehn and Yahr (mean, range)0 (0--0)1.6 (1--3)0 (0--0)1.000**\<0.001\<0.001**UPSIT (mean, SD, range)17.6 (6.2; 9--35)21.9 (8.4; 1--39)36.7 (1.6; 34--40)**\<0.001\<0.001**0.436RBDSQ (mean, SD, range)9.3 (2.9; 1--13)3.9 (2.6; 0--12)2.1 (1.4; 0--4)**\<0.0010.002\<0.001**ESS (mean, SD, range)8.4 (4.5; 0--20)5.4 (3.2; 0--15)4.8 (3.1; 0--12)**0.007**0.465**0.002**GDS (mean, SD, range)6.0 (2.1; 3--11)5.3 (1.5; 1--11)5.2 (1.0; 2--7)0.1680.9520.104MoCA (mean, SD, range)25.3 (4.5; 11--30)27.3 (2.3; 19--30)28.3 (1.2; 27--30)0.2790.2240.587Immediate recall (HVLT-R) (mean, SD, range)20.7 (5.4; 9--33)25.2 (5.4; 11--36)26.7 (4.7; 16--35)0.0200.5030.102Delayed recall (HVLT-R) (mean, SD, range)6.8 (3.0; 0--12)8.6 (2.7; 0--12)10.0 (1.9; 6--12)0.2100.0570.908Recognition (HVLT-R) (mean, SD, range)10.4 (1.5; 7--12)11.4 (1.0; 8--12)11.7 (0.6; 10--12)0.1020.6180.125Benton Judgment Line Orientation (mean, SD, range)11.5 (1.9; 8--15)12.9 (2.0; 7--15)13.3 (1.8; 9--15)0.0370.9340.027Letter and Number Sequencing (mean, SD, range)8.6 (3.1; 4--17)10.9 (2.9; 2--20)11.9 (3.0; 8--20)0.0650.6390.117Semantic fluency (mean, SD, range)44.3 (9.2; 27--65)49.8 (12.1; 20--103)55.2 (9.5; 39--74)0.1010.0290.507Symbol and Digit Modalities Test (mean, SD, range)31.4 (9.3; 15--56)41.6 (9.8; 7--70)49.0 (11.5; 30--76)**0.0030.001**0.037Values correspond to means followed by standard deviation or standard deviation and range. Comparisons between groups were performed using *X*^2^, Mann--Whitney *U* tests, or ANOVA. Age and sex were included as covariates in the analyses of motor and nonmotor variables, whereas education was included as an additional covariate in the analyses of cognitive variables. Values in bold correspond to significant group differences after adjusting for multiple comparisons with false-discovery rate corrections (FDR) (*q* \< 0.05)
Compared to controls, iRBD patients were significantly older (Cohen's *d* = 1.196, *p* \< 0.001), less educated (Cohen's *d* = 0.888, *p* = 0.006) and had higher MDS-UPDRS III scores (Cohen's *d* = 1.264, *p* \< 0.001). In addition, they also had more daytime sleepiness (Cohen's *d* = 0.773, ESS, *p* = 0.007) and performed significantly worse in cognitive tests measuring attention (Cohen's *d* = 0.879, SDMT, *p* = 0.003). PD patients showed greater motor impairment (Cohen's *d* = 3.142, MDS-UPDRS III, *p* \< 0.001) and worse attention (Cohen's *d* = 0.539, SDMT, *p* = 0.001) compared to controls, after FDR corrections.
When iRBD and PD patients were compared to each other, we found that iRBD patients were older (Cohen's *d* = 1.023, *p* \< 0.001) and more impaired in tests evaluating RBD (Cohen's *d* = 0.319, RBDSQ, *p* \< 0.001) and sleepiness (Cohen's *d* = 0.668, ESS, *p* = 0.002), after FDR corrections. On the other hand, as expected, PD patients showed greater motor impairment (Cohen's *d* = 2.419, MDS-UPDRS III, *p* \< 0.001) compared to iRBD.
MRI differences between groups {#Sec4}
------------------------------
In patients with iRBD, cortical thinning was found in the left lateral occipital (Cohen's *d* = 0.586; *p* = 0.03) and postcentral (Cohen's *d* = 0.786; *p* = 0.002) gyri, which extended to left inferior parietal and supramarginal areas (Fig. [1](#Fig1){ref-type="fig"}, Table [2](#Tab2){ref-type="table"}) compared to controls.Fig. 1Cortical thinning in patients with iRBD and patients with newly diagnosed PD compared to controls. Vertex-wise comparisons of cortical thickness between: **a** controls and patients with idiopathic REM sleep behavior disorder (iRBD); and **b** CTR and patients at early stages of Parkinson's disease (PD). The color scale bar shows the logarithmic scale of *p* values (−log~10~). All results were adjusted for multiple comparisons (cluster-wise threshold *p* \< 0.05 with Monte Carlo simulations) and corrected for age, sex and education. Lh left hemisphere, Rh right hemisphereTable 2Regions that showed cortical thinning in iRBD patients and PD patients compared to controlsCortical areaEffect size (Cohen's *d*)Cluster size (mm^3^)Cluster-wise *p* valueTalairach coordinates*xyziRBD vs. controls*Lh lateral occipital G0.5861999.090.03370−29.5−84.916.2Lh postcentral G0.7863064.100.00200−60.1−8.819.8*PD vs. controls*Lh inferior temporal G0.5522971.710.00410−53.5−26.1−24.7Lh superior frontal G0.4074625.030.00020−7.6−4.951.5Rh inferior temporal G0.4512815.450.0043047.5−18.6−26.3*Lh* left hemisphere, *Rh* right hemisphere, *G* gyrus. All results were corrected for multiple comparisons using a cluster-wise threshold of *p* \< 0.05 with Monte Carlo simulations. In addition they were also adjusted for age, sex, and education
In patients with PD, there was significant cortical thinning in the bilateral inferior temporal (left: Cohen's *d* = 0.552; *p* = 0.004; right: Cohen's *d* = 0.451; *p* = 0.004) and left superior frontal (Cohen's *d* = 0.407; *p* \< 0.001) gyri compared to controls (Fig. [1](#Fig1){ref-type="fig"}, Table [2](#Tab2){ref-type="table"}).
There were no significant differences in subcortical volumes between groups (Supplementary Table [1](#MOESM1){ref-type="media"}) or differences in cortical thickness between iRBD and PD patients.
MRI measures correlate with clinical impairment in iRBD {#Sec5}
-------------------------------------------------------
There were several significant correlations between cortical thickness and clinical measures in iRBD (Fig. [2](#Fig2){ref-type="fig"}, Supplementary Table [2](#MOESM1){ref-type="media"}). Increasing motor disease severity (MDS-UPDRS III) was associated with cortical thinning in the left superior frontal (*r* = −0.584; *p* = 0.021), left fusiform (*r* = −0.489; *p* = 0.011) and right precentral gyri (*r* = −0.447; *p* = 0.002). Moreover, worse olfaction (UPSIT) correlated with left medial orbitofrontal (*r* = 0.478; *p* \< 0.001), left precentral (*r* = 0.468; *p* = 0.003), and right medial orbitofrontal (*r* = 0.468; *p* = 0.02) thinning, whereas RBD (RBDSQ) correlated with right superior frontal (*r* = 0.466; *p* \< 0.001) thinning.Fig. 2Associations between cortical thinning and motor, nonmotor and cognitive deficits in patients with iRBD. Significant vertex-wise correlation between cortical thinning and **a** Movement Disorders Society Unified Parkinson's Disease Rating Scale (MDS-UPDRS) III motor scores; **b** University of Pennsylvania Smell Identification Test scores; **c** Rapid Eye Movement Behavior Disorder Sleepiness Questionnaire (RBDSQ) scores; **d** Immediate recall scores of the Hopkins Verbal Learning Test---Revised (HVLT-R); and **e** Benton Judgment of Line Orientation (BJLO) test visuospatial scores. The color scale bar shows the logarithmic scale of *p* values (−log~10~). All results were adjusted for multiple comparisons (cluster-wise threshold *p* \< 0.05 with Monte Carlo simulations) and corrected for age, sex (all correlations) in addition to education (correlations with cognition). Lh left hemisphere, Rh right hemisphere
Regarding cognitive tests, we found that worse memory performance (HVLT-R Immediate recall) correlated with cortical thinning in the left superior temporal (*r* = 0.453; *p* = 0.02), left caudal middle frontal (*r* = 0.419; *p* \< 0.001), right superior frontal (*r* = 0.447; *p* \< 0.001) and right lateral occipital (*r* = 0.449; *p* = 0.03) gyri; whereas visuospatial impairment (BJLO) correlated with left fusiform (*r* = 0.400; *p* = 0.001) and right supramarginal (*r* = 0.390; *p* = 0.004) thinning. No significant correlations were found with subcortical regions.
Differences between converters and nonconverters {#Sec6}
------------------------------------------------
All 27 iRBD patients were clinically assessed after a few years. As a result of this clinical assessment, six iRBD patients (22.2%) were clinically diagnosed with a Lewy body disorder: three patients were diagnosed with PD, one patient with DLB, one patient with MSA, and one patient with nonspecific parkinsonism. The remaining 21 iRBD patients did not convert to another disorder at follow-up. The clinical characteristics for both converters and nonconverters can be found in Table [3](#Tab3){ref-type="table"}. The last available follow-up for the converters was 2.5 years (3 patients), 3 years (1 patient), and 3.5 years (2 patients), whereas the last available follow-up for the nonconverters was 1.5 years (1 patient), 2 years (4 patients), 2.5 years (2 patients), 3 years (10 patients), and 3.5 years (4 patients) (Table [3](#Tab3){ref-type="table"}).Table 3Clinical characteristics of iRBD patients that converted to as Lewy body disorder and nonconvertersConverters (*n* = 6)Non-converters (*n* = 21)Converters vs. nonconverters (*p* value)Age (mean, SD)67.8 (4.1)69.2 (5.9)0.593Sex (M/F)4/218/30.289Education (mean, SD)11.8 (5.0)13.0 (5.3)0.649Years between baseline and follow-up (mean, SD, range)2.9 (0.5, 2.5--3.5)2.8 (0.5, 2.0--3.5)0.798Years to PD conversion (mean, SD, range)2.3 (0.3, 2.0--2.5)----Last available follow-up (number of patients assessed after 1.5/2/2.5/3/3.5 years)0/0/3/1/21/4/2/10/40.798MDS-UPDRS III Baseline (mean, SD, range)5.8 (4.8; 2--15)3.7 (3.2; 0--12)0.409Follow-up (mean, SD, range)22.8 (14.8; 9--49)4.4 (3.9; 0--14)**\<0.001**UPSIT baseline (mean, SD, range)16 (9.6; 9--35)18.1 (5.1; 9--31)0.440RBDSQ baseline (mean, SD, range)11.0 (1.8; 8--13)8.9 (3.1; 1--12)0.151Follow-up (mean, SD, range)8.5 (4.7; 0--13)7.6 (3.9; 0--13)0.270ESS baseline (mean, SD, range)11.0 (5.9; 6--20)7.7 (3.9; 0--16)0.255Follow-up (mean, SD, range)12.0 (5.1; 6--20)5.6 (3.5; 1--13)0.040GDS baseline (mean, SD, range)7.5 (2.4; 5--11)5.6 (1.8; 3--9)0.205Follow-up (mean, SD, range)7.0 (0.9; 6--8)5.5 (1.6; 2--8)0.232MoCA baseline (mean, SD, range)23.2 (6.1; 11--27)25.9 (4.0; 14--30)0.422Follow-up (mean, SD, range)22.3 (4.6; 14--27)25.6 (3.2; 18--30)0.223Immediate recall (HVLT-R) baseline (mean, SD, range)16.3 (4.3; 9--22)21.9 (5.1; 13--33)0.031Follow-up (mean, SD, range)16.0 (4.9; 11--22)22.3 (6.7; 4--34)0.028Delayed recall (HVLT-R) baseline (mean, SD, range)5.0 (3.8; 0--11)7.3 (2.6; 3--12)0.058Follow-up (mean, SD, range)5.7 (4.0; 1--12)7.5 (2.4; 1--11)0.100Recognition (HVLT-R) baseline (mean, SD, range)9.3 (1.5; 8--12)10.7 (1.3; 7--12)0.041Follow-up (mean, SD, range)8.7 (3.3; 3--12)10.7 (1.4; 7--12)0.047Benton judgment line orientation baseline (mean, SD, range)10.2 (2.4; 8--14)11.8 (1.7; 8--15)0.083Follow-up (mean, SD, range)11.6 (2.1; 9--14)10.4 (2.4; 5--15)0.404Letter and number sequencing baseline (mean, SD, range)7.2 (2.1; 4--9)9.0 (3.3; 4--17)0.699Follow-up (mean, SD, range)6.2 (3.3; 3--11)8.2 (2.9; 4--15)0.596Semantic fluency baseline (mean, SD, range)41.5 (8.6; 35--56)45.1 (9.4; 27--65)0.429Follow-up (mean, SD, range)41.0 (7.7; 34--53)44.3 (10.8; 26--64)0.465Symbol and digit modalities test baseline (mean, SD, range)32.0 (6.1; 24--41)31.2 (10.0; 15--56)0.798Follow-up (mean, SD, range)32.2 (7.3; 23--40)29.5 (11.9; 0--54)0.732Values correspond to means followed by standard deviation or standard deviation and range. Comparisons between groups were performed using *X*^2^, Mann--Whitney *U* tests, or ANOVA. Age and sex were included as covariates in the analyses of motor and nonmotor variables, whereas education was included as an additional covariate in the analyses of cognitive variables. Values in bold correspond to significant group differences after adjusting for multiple comparisons with false-discovery rate corrections (FDR) (*q* \< 0.05)
At baseline, there were no significant differences in clinical variables between patients that converted to a Lewy body disorder and those that remained disease free, after adjusting for multiple comparisons. After approximately 3 years, these patients showed greater motor disease severity (MDS-UPDRS III, *p* \< 0.001) in line with their conversion to a Lewy body disorder, compared to nonconverters.
The analyses of cortical thickness using the baseline MRI images revealed widespread thinning in the left superior frontal (Cohen's *d* = 1.191, *p* \< 0.001), right precentral (Cohen's *d* = 1.128, *p* \< 0.001) and right lateral occipital gyri (Cohen's *d* = 1.130, *p* \< 0.001) in converters compared to nonconverters (Fig. [3](#Fig3){ref-type="fig"}, Supplementary Table [3](#MOESM1){ref-type="media"}). No differences were found in subcortical regions between these groups.Fig. 3Cortical thinning in iRBD patients that converted to a Lewy body disorder compared to non-converters. Vertex-wise comparisons of cortical thickness between patients with idiopathic REM sleep behavior disorder (iRBD) that progressed to a Lewy body disorder at follow-up (converters) compared to iRBD patients that remained disease free (nonconverters). The color scale bar shows the logarithmic scale of *p* values (−log~10~). All results were adjusted for multiple comparisons (cluster-wise threshold *p* \< 0.05 with Monte Carlo simulations) and corrected for age, sex, education, baseline MDS-UPDRS III motor scores and time interval between baseline and last follow-up assessment. Lh left hemisphere, Rh right hemisphere
The comparisons between healthy controls and converters, and healthy controls and non-converters have been included in Supplementary Fig. [1](#MOESM1){ref-type="media"} and Supplementary Table [4](#MOESM1){ref-type="media"}. Compared to controls, converters showed even more widespread cortical thinning in similar frontal, precentral and occipital areas. Although no significant differences were found in nonconverters compared to controls after correcting for multiple comparisons, at an uncorrected level there was cortical thinning in brain regions that were similar to the ones observed in the whole iRBD group (Fig. [1](#Fig1){ref-type="fig"}). No significant differences were found in subcortical volumes between groups, after FDR corrections (Supplementary Table [5](#MOESM1){ref-type="media"}).
Cortical thinning as a risk factor for developing a Lewy body disorder {#Sec7}
----------------------------------------------------------------------
We built a Cox univariate model using the mean cortical thickness extracted from the brain vertices showing significant group differences in Fig. [3](#Fig3){ref-type="fig"} to assess whether it could predict conversion to a Lewy body disorder. The results showed that the mean cortical thickness significantly predicted conversion to a Lewy body disease (HR = 0.784; 95% CI: 0.640--0.960; *p* = 0.019), with a sensitivity of 95.2%, specificity of 100% and an AUC of 0.984 (CI 95%: 0.945--1.000; *p* \< 0.001).
Influence of probable RBD in PD patients {#Sec8}
----------------------------------------
Although PD patients did not undergo polysomnography to confirm the presence of RBD, to assess the potential influence of RBD symptoms in the clinical and MRI profiles of these patients, we used the scores from the rapid eye movement behavior disorder (RBD) Questionnaire---RBDSQ (cut-off ≥ 5) to divide them into two groups with probable presence (*n* = 56) or absence (*n* = 94) of RBD, similarly to a previous study.^[@CR16]^
Compared to controls, PD patients without probable RBD had greater motor impairment (Cohen's *d* = 3.074, *p* \< 0.001) and olfactory dysfunction (Cohen's *d* = 2.107, \<0.001), whereas PD patients with probable RBD, in addition to motor and olfactory deficits, presented greater sleepiness (Cohen's *d* = 0.210, *p* = 0.001), memory impairment (delayed recall, Cohen's *d* = 0.741, *p* = 0.001) and semantic fluency impairment (Cohen's *d* = 0.910, *p* \< 0.001) (Supplementary Table [6](#MOESM1){ref-type="media"}).
Compared to iRBD patients, both PD patients with and without probable RBD were more educated, had greater motor impairment, less sleepiness and more RBD symptoms (*p* range: 0.016--\<0.001). In addition, PD patients without RBD were less impaired in memory (Cohen's *d* = 0.462, recognition, *p* = 0.008) and visuospatial (Cohen's *d* = 0.646, BJLO, *p* = 0.005) tests, whereas PD patients with probable RBD were less impaired in semantic fluency (Cohen's *d* = 0.643, *p* = 0.011) compared to iRBD.
The cortical thickness analyses showed that PD patients without probable RBD had cortical thinning only in the left superior frontal gyrus (Cohen's *d* = 0.407, *p* = 0.002), whereas PD patients with RBD had more widespread cortical thinning in bilateral superior frontal (left: Cohen's *d* = 0.737, *p* \< 0.001; right: Cohen's *d* = 0.508, *p* \< 0.001), bilateral inferior temporal (left: Cohen's *d* = 0.540, *p* = 0.003; right: Cohen's *d* = 0.548, *p* = 0.0013) and left rostral middle frontal (left: Cohen's *d* = 0.508, *p* \< 0.001) regions (Supplementary Fig. [2](#MOESM1){ref-type="media"}; Supplementary Table [7](#MOESM1){ref-type="media"}), compared to controls. These results suggest that the presence of RBD symptoms in PD is associated with greater cortical changes. There were no significant differences in cortical thickness between the PD groups and iRBD patients.
Discussion {#Sec9}
==========
In this longitudinal study we explored the potential of structural neuroimaging to identify brain abnormalities in individuals with iRBD enriched for incipient parkinsonism, their relationship with clinical impairment and value as risk markers to develop an imminent neurodegenerative disease. Our main findings showed a pattern of cortical thinning in iRBD compared with controls, which was different from the one observed in PD patients. There were also several significant correlations between cortical thickness and motor, nonmotor and cognitive measures in iRBD. Finally, we found that cortical thinning in frontal, occipital and parietal areas predicted a substantially increased risk of progression to a clinically defined Lewy body disorder in iRBD after a relatively short period of 3 years. These findings suggest that cortical thickness could potentially be used to identify iRBD individuals who will convert faster to a neurodegenerative disease.
The prodromal stage of many Lewy body disorders is a period where motor, nonmotor and cognitive clinical manifestations occur but motor features are too subtle to allow a formal diagnosis of disease.^[@CR1]^ Our findings are in line with this as we found worse motor, sleep, and attention symptoms or functions in iRBD patients. These clinical changes were accompanied by cortical thinning in left motor, parietal, occipital areas. This asymmetric pattern of atrophy was in line with the cortical thinning pattern observed in newly diagnosed PD patients in the current study, which was also more prominent in the left hemisphere. In addition, this atrophy pattern partially overlapped with a metabolic brain network that has previously shown to be implicated in iRBD.^[@CR17]^ This network has been described in multiple studies^[@CR18]^ and includes lateral occipital and parietal regions, among other areas. These regions have shown to present hypometabolism in iRBD and they also overlap with the areas of a metabolic network that is affected in PD.^[@CR18],[@CR19]^ Together with our findings, these results suggest that occipital and parietal areas might be especially vulnerable to iRBD and could potentially represent an early manifestation of preclinical PD.
Despite the low number of phenoconverters in our study, we found that the most frequent disorder to which iRBD patients converted to was PD. This finding is in line with previous evidence showing that 50% of iRBD cases convert to PD within 5 years.^[@CR1]^ In addition, we did not find any significant differences in cortical thickness between iRBD and PD patients, despite the fact that their baseline atrophy patterns were quite different with respect to controls. The absence of significant differences between these two patient groups could be due to the potentially high prevalence of future PD converters in our iRBD group. It is possible that there is already ongoing subtle atrophy in iRBD in similar areas as the ones affected in PD, although this effect is not strong enough to be detected at baseline and after controlling for multiple comparisons.
The presence of RBD symptoms has been associated with a malignant PD subtype characterized by rapid progression in cognitive, motor and nonmotor symptoms over time.^[@CR20],[@CR21]^ Our results agree well with these findings as we found that PD patients with probable RBD presented greater memory and executive impairment in addition to more widespread cortical thinning than PD patients without RBD. This suggests that the presence of RBD symptoms may be responsible for a worse prognosis in PD and greater cortical changes. However, future studies are needed to replicate these findings in PD patients with confirmed RBD based on polysomnography results, which unfortunately was not available for the PD patients included in the current study.
In addition, we found that almost all clinical tests that were impaired in iRBD were associated with cortical thinning in relevant brain areas. For instance, motor impairment correlated with thinning in motor areas such as the precentral gyrus and olfaction correlated with thinning in the medial orbitofrontal gyrus, which is adjacent to the olfactory bulb.^[@CR22]--[@CR24]^ Regarding cognition, we found that visuospatial functions correlated with right parietal regions and other areas that are important for visual perception,^[@CR24]^ whereas memory correlated with left temporal regions, which are important for memory consolidation.^[@CR24]^ Together, these findings suggest that cortical thinning might contribute to some of the clinical deficits observed in iRBD.
To our knowledge, the current study is the first in assessing the value of a structural neuroimaging biomarker in predicting short-term progression to a parkinsonian syndrome in patients with iRBD. Our findings suggest that cortical thinning in frontal, occipital and parietal areas is a significant predictor for early development of a Lewy body disorder. Due to the high variability in time intervals between iRBD diagnosis and phenoconversion, it is important to find biomarkers that are able to identify iRBD patients at high risk for early conversion into clinically defined synucleinopathies.^[@CR15]^ Our findings suggest that cortical thickness could be one of these biomarkers.
Some limitations should be recognized in this study such as the potential bias of including very healthy controls in the Parkinson's Progression Markers Initiative (PPMI) cohort with MoCA scores ≥27, and the fact that only 31 of these controls did not present RBD (≥5 on RBDSQ) or olfactory dysfunction. In addition, the sample size of the converters group was very small so our findings should be interpreted with caution and replicated in larger, separate samples of iRBD patients that convert either to PD, DLB, or MSA, which may have different brain atrophy patterns.
Another limitation is the fact that patients with iRBD were older, less educated and more cognitively impaired compared to controls. The presence of cognitive deficits has been associated with RBD symptoms in PD, suggesting that cognition and sleep disturbances are not independent from each other.^[@CR21]^ Regarding age and education, we included these variables as covariates of no interest in the comparisons between controls and iRBD patients. In addition, we also observed cortical thinning in iRBD converters compared to nonconverters, who did not differ in age or education. Hence, our MRI findings are most likely not related to age or education differences between the groups. One additional limitation is the fact that one of the iRBD converters was diagnosed with non-specific parkinsonism at follow-up; hence, we do not yet know which specific disease this patient had. Finally, the last available follow-up varied between iRBD patients, with some having a longitudinal assessment after 1.5 or 2 years and others having an assessment after 3 and 3.5 years. Although there were no significant differences in the last available follow-up between converters and non-converters, it would have better that all patients had been followed for the same number of years. Future studies with more homogeneous longitudinal evaluations are needed to assess whether this variable has any effect on the findings.
In summary, we found that cortical thinning is a useful marker to detect iRBD patients with an increased risk of short-term conversion to a Lewy body disorder, suggesting it could potentially be used in future neuroprotective trials aimed at preventing or delaying the onset of motor disease.
Methods {#Sec10}
=======
Participants {#Sec11}
------------
Data used in this article were obtained from the PPMI database (www.ppmi-info.org/data),^[@CR25]^ accessed on May 20, 2017. For up-to-date information on the study, visit www.ppmi-info.org. For the purposes of this study, only iRBD patients, PD patients and healthy controls with a T1-weighted scan, that passed quality control before and after image preprocessing, were included.
Healthy controls were required not to have significant neurological dysfunctions, first-degree family members with PD, or cognitive impairment (Montreal Cognitive Assessment^[@CR26]^ (MoCA; score ≥27). In addition, in order to maximize the likelihood of being free from a subclinical neurodegenerative disorder, controls who screened positive for RBD (≥5 on the Rapid Eye Movement Behavior Disorder (RBD) Questionnaire---RBDSQ)^[@CR27]^ or had significant olfactory dysfunction for their age and sex^[@CR28]^ were excluded from this study.
iRBD patients were required to have a diagnosis of iRBD based on clinical history and polysomnography results. To enrich this cohort with individuals presumed to have an incipient parkinsonian syndrome, most iRBD patients from PPMI had a dopamine transporter imaging (DaTscan) deficit. At the time of scanning, all iRBD patients were free of neurological diseases and did not present with significant parkinsonism. iRBD patients were followed for approximately 3 years. At each visit, a neurological examination was performed to apply clinical criteria for the diagnosis of neurodegenerative disorders, including PD, DLB, and MSA.
PD patients were required at baseline to meet standard diagnostic criteria for PD, have been diagnosed within 2 years, be untreated for PD and present a significant DaTscan deficit.
Clinical evaluations {#Sec12}
--------------------
All subjects underwent a comprehensive assessment of motor, nonmotor and cognitive functions. Motor severity and disease stage were assessed using the Movement Disorders Society Unified Parkinson's Disease Rating Scale (MDS-UPDRS) part III scores^[@CR29]^ and the Hoehn & Yahr scale.^[@CR30]^ Nonmotor functions were assessed using the University of Pennsylvania Smell Identification Test (UPSIT)^[@CR28]^ (olfaction), the Epworth Sleepiness Scale (ESS)^[@CR31]^ (daytime sleepiness), the RBDSQ^[@CR27]^ (RBD) and the Geriatric Depression Scale-15 (GDS-15)^[@CR32]^ (depression). Cognitive assessments included the MoCA^[@CR26]^ (global cognition); the immediate, delayed and recognition recall scores of the Hopkins Verbal Learning Test--Revised (HVLT-R)^[@CR33]^ (memory); the Benton Judgment of Line Orientation test (visuospatial functions)^[@CR34]^; the Letter Number Sequencing (LNS)^[@CR35]^ test and semantic fluency tests (executive functions); and the Symbol Digit Modalities Test^[@CR36]^ (attention).
Ethical approval {#Sec13}
----------------
Each participating PPMI site received approval from an ethical standards committee on human experimentation before study initiation, and obtained written informed consent for research from all individuals participating in the study.
MRI acquisition and preprocessing {#Sec14}
---------------------------------
All subjects were scanned on a 3T Siemens Tim Trio scanner using a high-resolution T1-weighted scan, acquired with a magnetization-prepared rapid acquisition gradient echo sequence (176 slices; repetition time = 1900--2300 ms; echo time = 2.27--2.98 ms; inversion time = 900 ms; flip angle = 9°; voxel size = 1 mm^3^ isotropic). T1-weighted images were preprocessed using FreeSurfer (version 6.0; http://freesurfer.net/) as published elsewhere.^[@CR37],[@CR38]^ For every subject, a cortical surface model was generated, providing a measure of cortical thickness at each vertex. The final cortical maps were smoothed using a 15-mm full width at half maximum kernel. The volumes of subcortical gray matter structures (hippocampus, amygdala, thalamus, caudate, putamen, pallidum, accumbens) were also obtained from Freesurfer^[@CR39]^ in addition to the estimated total intracranial volume (TIV).^[@CR40]^
Statistical analyses {#Sec15}
--------------------
### Clinical group comparisons at baseline {#Sec16}
Differences between groups in demographic and clinical variables were analyzed using chi-squared tests (*X*^2^), Mann--Whitney *U* tests or analysis of variance (ANOVA) in SPSS 24.0 (IBM Corp., Armonk, NY), while controlling for age and sex (motor and nonmotor variables) and additionally education (cognitive variables). To adjust the results for multiple comparisons, false-discovery rate (FDR) corrections^[@CR41]^ were applied at *q* \> 0.05.
### MRI group comparisons at baseline {#Sec17}
To assess cortical thickness differences between groups, a general linear model was estimated at each vertex using FreeSurfer. In this general linear model, cortical thickness was included as the dependent variable; group as a factor; and age, sex and education as nuisance variables. To adjust the results for multiple comparisons, Monte Carlo simulations with 10,000 iterations were applied (cluster-wise threshold *p* \< 0.05). We calculated the Cohen's effect size for all significant group comparisons.
Differences between groups in subcortical gray matter volumes were assessed using an ANOVA in SPSS, while controlling for the previous covariates in addition to TIV. To adjust the subcortical results for multiple comparisons, FDR corrections (*q* \< 0.05) were applied.
### Association between clinical impairment and MRI in iRBD {#Sec18}
To assess the relationship between cortical thickness and the clinical tests that showed deficits in iRBD patients compared to controls, we estimated a general linear model that included cortical thickness as the dependent variable; clinical test scores as predictors; and age, sex, and education as nuisance variables. The correlation coefficients of the significant associations between thickness and clinical variables were calculated.
Partial correlation analyses were also carried out between subcortical volumes and clinical impairment, while adjusting for the previous covariates, TIV and FDR corrections.
### Follow-up group comparisons {#Sec19}
Patients with iRBD were followed up for approximately 3 years (mean = 2.8, range: 2.0--3.5). The diagnosis and clinical test scores were recorded at the last visit for all patients. We divided iRBD patients into two groups: (i) those who converted to a Lewy body disorder (converters) and (ii) those who did not convert to any neurodegenerative disorder (nonconverters). Differences between these groups in baseline clinical and imaging measures were assessed using an ANOVA or general linear models, similarly to previous analyses. In addition, to assess differences in all clinical variables across time we used repeated-measures ANOVAs, including clinical test scores at baseline and follow-up as dependent variables and group as a factor. All analyses were adjusted for age, sex, education, baseline MDS-UPDRS III scores, and FDR corrections.
### Risk factor analysis {#Sec20}
To assess the predictive ability of baseline brain atrophy for conversion to a Lewy body disorder in patients with iRBD, Cox regression analyses were performed using incident Lewy body disorder over 3 years as the outcome. We calculated hazard ratios, while adjusting for age, sex and education. Finally, we generated receiver operating characteristic curves for the significant predictors and calculated the area under the curve, sensitivity, and specificity.
Reporting Summary {#Sec21}
-----------------
Further information on experimental design is available in the [Nature Research Reporting Summary](#MOESM2){ref-type="media"} linked to this article.
Supplementary information
=========================
{#Sec22}
Supplementary material Reporting summary
**Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
=========================
**Supplementary information** accompanies the paper on the *npj Parkinson's Disease* website (10.1038/s41531-019-0079-3).
PPMI (a public--private partnership) is funded by the Michael J. Fox Foundation for Parkinson's Research and funding partners, including AbbVie, Avid, Biogen, Bristol-Myers Squibb, Covance, GE Healthcare, Genentech, GlaxoSmithKline, Lilly, Lundbeck, Merck, Meso Scale Discovery, Pfizer, Piramal, Roche, Servier, Teva, and UCB. We would like to thank the Swedish Foundation for Strategic Research (SSF), the Strategic Research Program in Neuroscience at Karolinska Institutet (StratNeuro), Hjärnfonden, and Birgitta och Sten Westerberg for financial support.
Research project design and execution: J.B.P., D.W., L.C., and D.A. Statistical and data analysis: J.B.P., D.W., and L.C. Drafting of manuscript: J.B.P. Revision of manuscript: J.B.P., D.W., L.C., D.A., O.H., and E.W.
The data that support the findings of this study are available from the corresponding authors upon reasonable request.
As previously mentioned in the MRI subsection, all T1-weighted images were preprocessed using FreeSurfer (version 6.0).
Competing interests {#FPar1}
===================
The authors declare no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
Mechanical forces play an important role in the development, homeostasis and repair of tissues. This is mainly the result of the mechanosensitivity of many biological remodeling processes at the cellular level such as proliferation, migration, differentiation, apoptosis and extracellular matrix synthesis (Ingber [@CR10]). Cell stretching devices have demonstrated their potential to contribute to our fundamental understanding of pathways in cellular mechanotransduction and mechanosensitivity (Huh et al. [@CR9]; Huang and Nguyen [@CR7]; Kim et al. [@CR14]; Moraes et al. [@CR15], [@CR16]). When integrated on a microfluidic platform, these devices offer significant improvements over their macroscale counterparts (Wang et al. [@CR20]; Huang et al. [@CR8]), mainly given their potential for high throughput processing as well as their ability to be combined with other on-chip functions. Recently, organ-on-a-chip systems have attracted attention by highlighting the ability to better understand the effects of mechanical forces at the cellular level for different organs (Huh et al. [@CR9]; Kim et al. [@CR14]). In vivo, tissue-embedded cells undergo mechanical strains that often vary spatially and temporally. It is the case in vascular tissues where the combination of the local hemodynamic forces (Frydrychowicz et al. [@CR5]) with the anisotropic mechanical properties of vascular tissues (Duprey et al. [@CR3]; Tremblay et al. [@CR17]) exposed endothelial and smooth muscle cells to complex multi-axial and cyclical deformations. Moreover, these strain fields can induce significant sub-cellular, cellular- and multi-cellular remodeling responses in a frequency and magnitude dependent manner (Balachandran et al. [@CR1]; Goldyn et al. [@CR6]; Jungbauer et al. [@CR12]).
Microfluidic stretching devices have been developed to study single cell response to mechanical deformation or to observe multi-culture cell system mimicking organ-level functions under mechanical stimuli. The elegant work by Huh et al. ([@CR9]) demonstrated the ability to mimic organ-level functions in a microfabricated stretching device. They were able to uniaxially stretch a co-culture of alveolar epithelial cells and endothelial cells to examine cellular responses to mechanical deformation in a model of the lung. Using a similar device Kim et al. ([@CR14]) demonstrated that human intestinal epithelial cells exhibit changes in cell morphology and increased aminopeptidase activity under cyclic uniaxial stretching. Several groups have now integrated microfabricated stretching devices into microfluidic networks in order to allow for high throughput screening. Huang and Nguyen ([@CR7]) have integrated microfabricated uniaxial devices in a high throughput platform allowing the investigation of the effect of various uniaxial stretching conditions on cell response within the same experiment. Other systems have used piston-like structures to deform a membrane on which cells are firmly attached to perform high throughput screening. Kamotani et al. ([@CR13]) employed microwells with flexible bottom membranes placed over computer-controlled, piezoelectrically actuated pins inducing a broad range of biaxial strain fields in the same microwells. Similar high throughput devices have also been used to monitor the influence of mechanical substrate strain on β-catenin accumulation in the nucleus or myofibroblast differentiation (Moraes et al. [@CR15], [@CR16]). Taken together, these devices have strongly contributed to the development of a new class of microfabricated devices capable of studying cellular biological processes under mechanical strain.
Although existing devices have clear utility, an area of improvement would be the integration of full and independent biaxial control of the strain field. While idealized strain fields have provided important insights into strain-induced cellular remodelling processes, imposing more complex strain fields in the future would better mimic in vivo cellular systems. In this study, we build upon existing microfluidic stretcher designs and present a complementary device capable of imposing dynamic anisotropic biaxial strains on cells. In addition, our device can also maintain microfluidic control over the introduction of samples and allowing simultaneous imaging by optical microscopy. This device allows the independent and dynamic control of the strain magnitude and waveform frequency (milliseconds to days) in two orthogonal directions during the same stretching experiment, leading to better replication of complex multi-axial cyclic strains common to in vivo systems. We chose human foreskin fibroblast (HFF) cells as a model system for this study as fibroblasts are well known to sense and respond to strain (Wang et al. [@CR19]). We show that the device can maintain cell viability over several days and allows the study of the same group of cells in response to a changing biaxial strain field.
Materials and methods {#Sec2}
=====================
Working principle of the device {#Sec3}
-------------------------------
We present a microfabricated biaxial stretcher which draws upon the designs presented by Huh et al. ([@CR9]) and Huang and Nguyen ([@CR7]). Our device is fabricated using poly(dimethylsiloxane) (PDMS; Sylgard184) by multi-layer soft lithography (Fig. [1](#Fig1){ref-type="fig"}). Figure [1](#Fig1){ref-type="fig"}a shows an exploded cross-section view of the multilayer device with the low pressure and fluidic channels. The 10 μm thick, suspended membrane on which cells adhere and proliferate makes a liquid tight seal between the top fluidic channel (purple) and the bottom fluidic channel (blue). This configuration ensures that no pressure differential is established across the suspended membrane, which prevents any upward or downward displacement of the membrane causing it to stick on the upper or lower surface of the stretching chamber. During the assembly process, the membrane was carefully punctured with a sharp needle to provide access to the channels of the bottom section, while maintaining cleanliness. Also, it was important for the fluidic channels of the top part to be open to the air during the alignment process to equilibrate pressures between the top and bottom fluidic channels, thus avoiding membrane collapse. A detailed fabrication process is included as Supplementary data Fig. 1b shows a cross-section of the device with cells in the stretching chamber. Lateral deformation of the vertical walls occurs when a low pressure is applied (red chambers), which pulls on the attached suspended membrane and induces deformation, as depicted Fig. [1](#Fig1){ref-type="fig"}c--f and in Supplementary data: videos A and B. The microfabricated device is maintained on an inverted microscope at 37 °C in a humid atmosphere of 5 % CO~2~/95 % air using a custom incubation chamber in order to perform time-lapse live cell imaging, as described in Supplementary Fig. 1.Fig. 1**a** Exploded cross-section of the multi-layer PDMS-based cell stretching device. Low pressure is applied to the low pressure channels (*red*) to induce a deformation in the walls located at each of the four sides of the cell stretching chamber (800 × 800 μm; 10 μm thick membrane). The *top* and *bottom* fluidic channels (*purple* and *blue*) are isolated from each other by a suspended membrane. The *bottom* fluidic channel (*blue*) serves to equilibrate pressures when seeding cells. *Bottom left* of **a**: Photographic image of the assembled device with the *bottom* and *top* fluidic channels (*blue* and *purple* channels) connected and the four low pressure channel inlets (see *arrows*). **b** Detailed view of the assembled device cross-section showing the cell stretching chamber along with the low pressure chambers on both sides (circled "L" indicates low pressure). **c**--**d** Schematic cross-section of the device with cells attached on the membrane and the low pressure chambers under atmospheric pressure conditions (**c**) and low pressure conditions (**d**). **e**--**f** Phase-contrast images of the device viewed from the *top*; two of the four low pressure chambers are visible under atmospheric pressure conditions (**e**) and low pressure conditions (**f**)
Cell seeding {#Sec4}
------------
Before introducing cells, the device's top and bottom fluidic channels are first wetted and sterilized with 95 % ethanol for 5 min prior to being flushed with autoclaved deionized water for another 5 min. Water is then replaced by a fibronectin solution at 10 μg/ml of HEPES-buffered salt solution (HBSS; 20 mM HEPES at pH 7.4, 120 mM NaCl, 5.3 mM KCl, 0.8 mM MgSO~4~, 1.8 mM CaCl~2~ and 11.1 mM glucose). Once the microfluidic channels are filled with the fibronectin solution, the ends of all tubing leading to the device are placed in a single solution-filled vial. This equilibrates all pressures and completely stops flow within the device, promoting fibronectin functionalization of the membrane. Fibronectin is incubated for 2 h at 5 % (v/v) CO~2~ and 37 °C. Subsequently, the fibronectin solution is replaced with culture medium (DMEM) supplemented with 10 % (v/v) fetal bovine serum and 1 % penicillin/streptomycin at a flowrate of \~10 μl/min. In the mean time, cells cultured in a standard incubator (5 % v/v CO~2~ and 37 °C) are trypsinized and resuspended in culture medium at 2 × 10^6^ cells/ml. The top microchannel is then filled with the culture medium supplemented with cells, whereas the bottom channel is further flushed with fresh culture medium. Individual cells quickly adhere to the fibronectin-coated membrane surface within 10 s under no flow conditions. After 10 s, more cells were carried in the device's chamber while the cells already present in the chamber remained attached to the membrane. Cells are thus immobilized to the membrane, one by one, until about 70 cells are present in the stretching chamber. Once the cells are adhered to the membrane, flow is again completely stopped by placing all tubing in the same media-filled vial. The cells are left to firmly attach to the fibronectin-coated PDMS membrane overnight. Supplementary Fig. 2 shows the speed at which cells attach to the fibronectin-coated PDMS membrane. The deposited cells are initially somewhat lined up with the fluid flow direction. However, HFFs are very motile and quickly cover the entire surface of the membrane after overnight incubation.
Image analysis and cell orientation {#Sec5}
-----------------------------------
Cell orientations were quantified using filtered and thresholded phase-contrast images of the cells. A FFT band-pass filter was first applied on the phase-contrast images using ImageJ (<http://rsbweb.nih.gov/ij/>) to smooth background and isolate cell features. Thresholding was applied to create binary images of the cell features. The orientation of each of the features was computed and record to produce a histogram for each of the stretching conditions.
Results {#Sec6}
=======
Device performance {#Sec7}
------------------
Prior to performing each stretching experiment, calibration was performed by relating the pressure in the low pressure chambers and the strain field in the flexible membrane. A MATLAB script allowed us to compute the Green strain tensor in the plane of the PDMS membrane by tracking the position of embedded fluorescent beads during stretching. Figure [2](#Fig2){ref-type="fig"}a--c illustrates the strain field in the membrane along two orthogonal directions as four embedded particles are tracked (white lines). A strain map can be generated based on the beads tracking computation. Figure [2](#Fig2){ref-type="fig"}d, e highlights the agreement between the experimental results and the finite-element simulation of the stretching device in action. While the configuration of the stretching device allowed us to precisely control the strain along the two orthogonal axes, it also leads to a non-uniform strain magnitude over the entire extent of the membrane surface, as depicted in Fig. [2](#Fig2){ref-type="fig"}d, f. Representing the iso-deformation field of the membrane (white dashed lines) during deformation allows the better appreciation of the presence of deformation gradients, as shown in Fig. [2](#Fig2){ref-type="fig"}f. By carefully characterizing the spatial variation of the strain magnitude in the membrane, we found that the deformation in the central region of the membrane (266 × 266 μm^2^) was relatively constant (±0.4 % variation in strain magnitude) and compares to other microfabricated stretching devices (Moraes et al. [@CR15], [@CR16]; Kamotani et al. [@CR13]). Typically, a pressure of 0.1 atm in the low pressure chambers induced a deformation of about 20 % in the center part of the membrane and is highly consistent between devices. Six devices have been used to quantify the repeatability of the fabrication process. At most, we observed a variation in the strain magnitude of ±2.6 % at 22.5 % deformation between devices, as depicted in Fig. [2](#Fig2){ref-type="fig"}g, h. We also investigated the repeatability of the strain field over time and found very little change in the magnitude of the deformation over 20 h under constant low pressure conditions, as depicted in Supplementary Fig. 3. Exploiting the ability to independently control the deformation along each orthogonal axis allowed us to expose cells to horizontal or vertical uniaxial strain fields. The simple relationship between pressure in the low pressure chambers and membrane strain allowed us to easily interpolate and precisely induce the desired strain magnitude along both axes.Fig. 2**a**--**c** Fluorescent images showing the fluorescent beads embedded in the membrane, used to monitor membrane deformation. Low pressure chambers are independently activated to induce deformation in the membrane along two *orthogonal* directions. **d** Typical deformation field calculated from the displacements of the embedded beads during uniaxial stretching along the *vertical* direction. **e**--**f** Strain map and contour map of the magnitude of the deformation in the membrane using COMSOL (Burlington, USA). The *white dashed lines* in **f** follows the general alignment of the cells when stretched vertically. **g**--**h** Typical calibration curves illustrating the relationship between pressure and membrane deformation. The symmetry of the devices result in producing very similar calibration curves along the *horizontal* (**g**) and *vertical* direction (**h**)
Cellular responses to dynamic and complex strain fields {#Sec8}
-------------------------------------------------------
Figure [3](#Fig3){ref-type="fig"} demonstrates the device's ability to apply complex strain fields, by inducing a deformation along two orthogonal directions. Figure [3](#Fig3){ref-type="fig"}a shows a phase-contrast image of the cells prior to deformation. HFF cells, immobilized on the suspended membrane, were then stretched, subject to a uniaxial strain of a magnitude of 20 % along the horizontal and vertical directions, as highlighted in Fig. [3](#Fig3){ref-type="fig"}b, c, respectively. Figure [3](#Fig3){ref-type="fig"}d, e are insets showing the instantaneous change in cell morphology during substrate stretching for the selected group of cells.Fig. 3**a**--**c** Phase-contrast images of the same group of cells immobilized to the suspended membrane exposed at first to no deformation (**a**) and then exposed to a *horizontal* (**b**) and *vertical* deformation (c). *Arrows* indicate stretching directions. **d**--**e** *Insets* showing a particular group of cells exposed to the corresponding strain fields
The HFF cells of Fig. [4](#Fig4){ref-type="fig"}a are first exposed to a cyclic uniaxial strain field (20 % in magnitude; 0.5 Hz) along the horizontal direction, inducing a collective alignment of the cells along the vertical direction after 8 h, as highlighted in Fig. [4](#Fig4){ref-type="fig"}b. Then the orientation of the strain field is suddenly changed to mechanically stimulate the same cells along the vertical direction with the same magnitude and frequency as before. This induces a collective re-alignment of the cells along the horizontal direction after 16 h. As revealed in Fig. [4](#Fig4){ref-type="fig"}c, the cells have completely reoriented themselves horizontally as they align perpendicularly to the stretching direction, in agreement with previous work (Wang et al. [@CR18]; Jungbauer et al. [@CR12]). Cellular orientation under different conditions was quantified as shown in Fig. [4](#Fig4){ref-type="fig"}d--f. These histograms show the absolute value of the angle the cells assume with respect to the horizontal. Cells are randomly oriented before imposing deformation, as depicted in Fig. [4](#Fig4){ref-type="fig"}d. Reorientation occurs as the number of features orientated along the vertical (Fig. [4](#Fig4){ref-type="fig"}e) and horizontal (Fig. [4](#Fig4){ref-type="fig"}f) axes increases following a cyclic uniaxial mechanical deformation of the cells along the horizontal and vertical axes respectively.Fig. 4**a** Cells cultured for 24 h in the device prior to perform the cyclic stretching experiment. **b** Cells exposed to a sinusoidal cyclic deformation along the *horizontal* direction with an amplitude of 20 % and a frequency of 0.5 Hz for 8 h. **c** Same group of cells exposed to the same strain field but this time along the *vertical* direction for 16 h. *Insets* in **b** and **c** reveal the contour map of the magnitude of the membrane deformation (finite element simulation; see Online Resource 1), and the *dotted white lines* highlight the transversal contours. The cells align to follow these lines as well. **d** Cells are randomly orientated before inducing deformation. **e** Cells are mostly aligned along the *vertical* direction after 8 h of uniaxial stretching along the *horizontal* direction. **f** Cells are mostly aligned along the *horizontal* direction after 16 h of stretching along the *vertical* directions
When stretching the membrane in one direction, the suspended membrane contracts in the orthogonal direction, as expected. The data shown in Fig. [2](#Fig2){ref-type="fig"}g, h reveal this orthogonal compression. However, this effect can be minimized by compensating the compression by simultaneously stretching the membrane in the direction orthogonal to the main axis of stretching. This is illustrated in Fig. [5](#Fig5){ref-type="fig"} where a uniaxial strain field is applied in the x-direction, while the compression in the y-direction is suppressed by simultaneously stretching in the perpendicular direction. The ability to induce strains using four independent low pressure chambers is unique in that it gives more control over the membrane's strain field.Fig. 5**a** Deformation-pressure relationship for a standard uniaxial strain field where the principal deformation occurs along the *horizontal* direction (scale and low pressure chambers colored in *red*) with the presence of a compressive strain along the *vertical* direction. Note that the low pressure chambers, along the *vertical* direction, are *left* at atmospheric pressure (scale and low pressure chambers colored in *blue*). **b** Deformation-pressure relationship for a pure uniaxial strain field where the principal deformation occurs along the *horizontal* direction while applying a stretch along the *vertical* direction to eliminate any compressive strains
Discussion {#Sec9}
==========
The recent development of microscale stretching devices has provided numerous insights into the kinetics of cellular responses to mechanical strain, at various time scales (milliseconds to hours) (Huh et al. [@CR9]; Huang and Nguyen [@CR7]; Kim et al. [@CR14]; Moraes et al. [@CR15]; Jungbauer et al. [@CR12]; Moraes et al. [@CR16]; Wang et al. [@CR18]; Chen et al. [@CR2]). Here, we build upon existing designs and present a microfabricated device that allows cells to be exposed to a strain field that can be controlled in two orthogonal directions independently. Existing approaches typically employ only uniaxial strain and do not possess the ability to dynamically change its direction. This provides the ability to change strain directions on the fly and also to create dynamic, complex and anisotropic strain fields. This approach provides a method for studying cellular reorientation resulting from complex and dynamic strains that better mimic what happens in vivo. Similarly, the work of Moraes et al. ([@CR15], [@CR16]) demonstrated a device that could independently change the radial and circumferential strain components, albeit with a maximum strain magnitude of 6 %. Our device is able to independently change both of the strain-field components dynamically with a maximum strain magnitude of 20 %. Importantly, the device allows the investigation of the effects of pure uniaxial or standard uniaxial stretching on cellular responses. Indeed, the effect of deforming cells perpendicularly to their orientation axis can induce severe disruption of microarchitecture of valve endothelial cells (Balachandran et al. [@CR1]). Consequently, precise control of the strain field (pure uniaxial, standard uniaxial, biaxial, equibiaxial) as well as its direction and magnitude, will facilitate a systematic understanding of how cells respond to the complex, anisotropic and time-varying strain fields they encounter in vivo.
HFF cells respond to cyclic substrate deformations by changing their morphology and orientation. Indeed, cells undergo morphological changes under uniaxial stretching by orienting themselves almost perpendicularly to the stretching direction. The orientation of the cells reflects the slight non-uniformity of the applied strain field, as evident from the inset in Fig. [4](#Fig4){ref-type="fig"}b, c. As revealed by the strain maps obtained experimentally and from finite element simulations (Fig. [2](#Fig2){ref-type="fig"}d, f), the magnitude of the vertical deformation of the membrane is non-uniform and follows a curved shape, the gradient of which is estimated by the white dashed lines. This arrangement suggests that individual cells are sensitive to local strain variation. It is still not clear what mechanisms are responsible for this behavior found in many cell types, but it is hypothesized that cells position themselves to experience the least amount of deformation (Wang et al. [@CR18]; Faust et al. [@CR4]). To our knowledge, cell response to non-uniform strain fields has never been investigated before. Given the microscale dimensions of our device, it is possible to investigate the effect of strain field gradients across the same cell while monitoring cellular remodeling and migration. In other applications, up-sizing the device dimensions would provide for larger areas with uniform strain, where a greater number of cells could be exposed to similar deformations.
The integration of independent biaxial stretching capabilities on a microfluidic device provides precise control over the biochemical and mechanical environments experienced by cells. We have demonstrated that cells are able to proliferate in the device and reorient themselves in response to applied strain. The ability to induce deformation along two orthogonal directions allows the investigation of how anisotropic strain modulates the mechanisms governing cellular proliferation, organization and cytoskeletal remodeling in response to cyclic stretch (Goldyn et al. [@CR6]; Chen et al. [@CR2]; Jaalouk and Lammerding [@CR11]). This may contribute to our understanding of how complex and anisotropic mechanical forces and strain originating in the extra-cellular matrix couple to the cytoarchitecture. Building upon the designs of previous microfluidic or macroscale stretching devices, we present an approach that provides the user with a unique ability to generate changing, anisotropic and time-varying strain fields in order to more closely mimic the complexities of strains occurring in vivo.
Electronic supplementary material
=================================
{#Sec10}
######
Supplementary material 1 (PDF 70 kb)
######
Supplementary material 2 (MP4 428 kb)
######
Supplementary material 3 (MP4 422 kb)
######
Supplementary material 4 (PDF 1603 kb)
DT thanks the Fond de recherche du Québec: Nature et Technologie (FQRNT) and Mitacs Elevate Program. AEP. acknowledges generous support from a Province of Ontario Early Research Award, a Canada Research Chair (CRC), a NSERC Discovery Grant and a NSERC Discovery Accelerator Supplement. MG acknowledges a NSERC Discovery Grant and a CFI grant.
| {
"pile_set_name": "PubMed Central"
} |
The content published in Cureus is the result of clinical experience and/or research by independent individuals or organizations. Cureus is not responsible for the scientific accuracy or reliability of data or conclusions published herein. All content published within Cureus is intended only for educational, research and reference purposes. Additionally, articles published within Cureus should not be deemed a suitable substitute for the advice of a qualified health care professional. Do not disregard or avoid professional medical advice due to content published within Cureus.
Introduction
============
According to the global anemia prevalence report published by the World health Organization (WHO) in 2011, Pakistani women of reproductive age (15-49 years) have a mean serum hemoglobin (Hb) level of 11.7 g/dL \[[@REF1]\]. According to 2016 reports, 52.1% of non-pregnant Pakistani women have iron deficiency anemia (IDA) \[[@REF2]\].
The two major causes of IDA in non-pregnant women of reproductive age are irregular menstrual bleeding (9-14%) \[[@REF3]\] and a gastrointestinal (GI) source of bleed such as mucosal erosions (6-30%) \[[@REF4]\]. Iron deficiency anemia is morbid as well as mortal \[[@REF5]\]. In order to prevent its complications, supplementation of iron becomes essential. In non-pregnant women, mild, moderate, and severe IDA is classified as serum Hb level 11.9-10.0 g/dL for mild, 9.9-7.0 g/dL for moderate, and less than 7.0 g/dL for severe IDA \[[@REF6]\].
The most appropriate form of iron supplementation is oral, unless it cannot be tolerated or absorbed. Worsening GI symptoms and major bowel surgeries which reduce iron absorption may require parenteral replacement \[[@REF7]\]. Oral elemental iron of 30-60 mg/day is required for treating IDA in adults. The total duration of treatment is three months. Adequate response is gauged by a rise in serum Hb levels of 1 g/dL within one month of compliant therapy \[[@REF8]\].
Although all of the available forms of oral iron supplementation may replenish the therapeutically required dose of elemental iron, the chief differences are in adherence to therapy. Inadequate treatment adherence due to non-availability, fear of side effects, particularly GI symptoms - nausea, vomiting, constipation, metallic taste \[[@REF9], [@REF10]\] - has been reported in the literature. Conventional iron supplements include ferrous sulphate with or without mucoproteose, ferrous sulphate (glycine), iron protein succinylate, ferrous gluconate and ferrous fumarate. Most of these salts injure the mucosal lining of the GI tract and cause other side effects such as constipation \[[@REF11]\]. Along with poor absorption of conventional iron, side effect profile hampers treatment compliance \[[@REF12]\].
Micronization and microencapsulation of iron into liposomes, in sachet form, is the most advanced approach to combat both issues with iron supplementation - absorption and tolerance. Micronization is the phenomenon of reducing individual particle size, thereby, increasing solubility, and hence, bioavailability, due to increased surface area to drug ratio. Microencapsulation is when the micronized iron is encapsulated inside a lipid bilayer, similar to the biological lipid bilayer. The outer phospholipid bilayer protects the inner iron core from enzymatic degradation in the mouth or stomach, and also prevents iron oxidation and degradation. Liposomes of iron, which are nanosized, have the advantage of quicker and better absorption with minimum oxidative damage and lower incidence of side effects. The lipid bilayer of liposomal iron gives the molecule stability and ability to release the contents gradually. Gradual release helps in better absorption. This sophisticated technology of liposomal encapsulation also prevents iron to come in direct contact with intestinal mucosal lining, hence improved tolerance \[[@REF12]-[@REF14]\].
With a shift of paradigm from prescription and over-the-counter drugs to nutraceuticals, a need was established to extract real world data regarding the safety and efficacy of various nutraceuticals. Iron pills are also important nutritional supplement and nutraceutical. Therefore, this study aims to assess the efficacy of microencapsulated iron pyrophosphate sachets in non-pregnant otherwise healthy women with iron deficiency anemia. Efficacy was measured in terms of increase in serum hemoglobin levels during the treatment period. This study also evaluated taste tolerability and palatability with this newer form of iron supplementation.
Materials and methods
=====================
It was a multicenter, open label clinical trial conducted after approval from ethics review committee. It was registered at [www.clinicaltrials.gov](www.clinicaltrials.gov) (NCT03112187). Otherwise healthy, non-pregnant women, of age 15 till 49 years, diagnosed with iron deficiency anemia who had their hemoglobin \<8 to \>5 g/dl were included after attaining written informed consent. All participants were given the right to refuse to participate in the study. Participants could also withdraw from the study at any time after informing the prescribing doctor. Furthermore, women who had a history of allergic reaction to iron supplementation, iron intolerance, hypersensitivity to vitamin C and vitamin B12, and individuals with causes of anemia other than iron deficiency were not included in the study.
A brief sociodemographic profile of all participants was recorded. It included their age, co-morbidities, and whether or not they have taken any iron supplements before. A brief gynecological assessment including menstrual frequency, painful menstruation and any history of anemia was also recorded.
The food supplement under consideration in this trial is microencapsulated iron pyrophosphate in liposomal form (Ferfer®: manufactured by PharmEvo Pvt. Ltd, Karachi, Pakistan). It is a water dispersible micronized source of iron that has been microencapsulated to enhance iron absorption and to lessen both GI aspect results and unwanted organoleptic attributes. Ferfer® is a 1.5-gm sachet which contains 14 mg iron in liposomal form, 80 mg vitamin C, and 2.5 mcg vitamin B12 with orodispersible granulate which immediately dissolves in the mouth without the want for water and therefore appropriate additionally for those who\'ve problem swallowing.
Each patient completed four-doctor visits during the study period. On the first visit, women with serum Hb \<8 to \>5 g/dl were inducted and prescribed Ferfer®, twice daily for 12 weeks. Patients who were anemic from other causes were excluded by detailed history and relevant laboratory tests like B12 or folate deficiency. Complete blood count (CBC) with peripheral films and in some cases serum ferritin was used to diagnose patients. All subsequent visits were four weeks apart. On each of the four visits, serum hemoglobin level was measured. On second, third, and fourth visit, women were asked to document adverse effects and taste tolerability of Ferfer®. Taste tolerability was scored at a scale of 1-5, with 1 being the least tolerable and 5 being the most tolerable. Side effects including nausea, vomiting, bloating, abdominal cramps, early satiety, acid eructation/heartburn, sickness, loss of appetite, retrosternal discomfort, epigastric or upper abdominal pain, constipation were to be reported. Participants could also report any side effect other than these. The participants were also provided with a contact number to communicate in case any adverse effects occurred in between the monthly scheduled visits.
With an estimated anemia prevalence of 50%, the sample size calculated was 384 with 95% significance level. After consideration of dropout throughout the study period (12 weeks), 600 patients were enrolled in the trial. All data, including sociodemographic profile, weekly lab investigations, and adverse effects were entered and analyzed using SPSS version 24 (IBM Corp, Armonk, NY, USA). Frequency and percentages were computed for categorical variables including sociodemographic profile. Mean ± Standard Deviation (SD) was computed for numerical variables such as age, lab results, and tolerability symptoms. Paired T-test was used to compare means of haemoglobin at base line and after 12 weeks of therapy. P value of less than 0.05 was considered statistical significance. For the purpose of analysis, "all-treated population" was considered which included any subject who received at least one day therapy with agent.
Results
=======
At the start of the trial, there were 558 women who agreed to participate and fulfilled the inclusion criteria. Their sociodemographic and clinical profile is shown in Table [1](#TAB1){ref-type="table"}.
###### Baseline demographics and clinical characteristics of patients (n = 558).
----------------------------------------------- -------------------------
Sociodemographic and clinical characteristics Frequency (%) (N = 558)
Age in years (Mean ± SD) 33.2 ± 8.83
History of iron supplementation 92 (16.4)
Comorbidities
Diabetes mellitus 16 (2.9)
Hypertension 53 (9.4)
Chronic heart disease 27 (4.8)
Angina 34 (6.1)
Previous history of iron deficiency anemia 180 (32.1)
Regular monthly menstruation for 5-7 days 212 (37.8)
Painful menstruation 212 (37.8)
----------------------------------------------- -------------------------
By the end of the first month, 15 women were lost to follow up. By the end of the second month, 105 more women dropped out. With one more woman dropping out by the end of the trial period, there were 437 women who completed the entire trial. The mean serum hemoglobin level increased from 8.71 ± 2.24 g/dL at the start of the study to 10.47 ± 1.69 g/dL at the end of the trial. The mean taste tolerability also improved throughout the study period. The mean change in hemoglobin levels at each visit and the taste tolerability status is shown in Table [2](#TAB2){ref-type="table"}.
###### Change in hemoglobin level after therapy with microencapsulated iron in female iron deficiency anemia.
------------------ ------------------------ ------------------------------- --------------------------------
Scheduled visit Number of subjects (N) Hemoglobin (g/dL) (Mean ± SD) Taste tolerability (Mean ± SD)
Start of therapy 558 8.71 ± 2.24 3.93 ± 0.93
Week 4 543 9.49 ± 1.81 3.94 ± 0.89
Week 8 438 9.80 ± 1.70 4.03 ± 0.82
Week 12 437 10.47 ± 1.69 4.05 ± 0.88
------------------ ------------------------ ------------------------------- --------------------------------
The mean change in serum hemoglobin level from the start of therapy till 12 weeks was statistically significant as shown in Table [3](#TAB3){ref-type="table"}.
###### Change in hemoglobin level at week 12 after therapy with microencapsulated iron in female iron deficiency anemia (n = 437).
---------------------------- ------------------ --------- ------------------ -----------
Hemoglobin level Mean ± SD (g/dL) t value 95% CI P value\*
Start of therapy (n = 558) 8.71 ± 2.24 -25.6 -2.02 -- (-1.74) \<0.001
At Week 12 (n = 437) 10.47 ± 1.69
---------------------------- ------------------ --------- ------------------ -----------
Discussion
==========
There is a significant increase in mean hemoglobin levels after 12-week supplementation with microencapsulated iron pyrophosphate sachets. Overall taste acceptability and palatability for this novel compound has already been studied \[[@REF14]\].
Over the years, it has been established that the key component of successful iron replenishment is treatment adherence and compliance. Oral iron salts are absorbed via divalent metal transporter 1 (DMT-1). Conventional forms of oral iron salts have this advantage of cost effectiveness and extensive availability, nonetheless, the constraint of GI intolerance, especially metallic after taste, stands still \[[@REF10], [@REF12]\]. In an Ugandian study, only 12% pregnant women were compliant to their iron supplementation \[[@REF9]\]. For Pakistan, reports indicate that only 38% women took iron and folic acid during their pregnancy \[[@REF15]\].
Not many studies have been conducted to evaluate iron supplementation in non-pregnant women of reproductive age. In a 16-week long randomized double-blind placebo-controlled trial, non-pregnant iron deficient women were randomized to fruit juice fortified with placebo or microencapsulated iron pyrophosphate (18 mg/day elemental iron). There was significant improvement in total erythrocyte count, hematocrit count, red cell distribution width, serum ferritin and soluble transferrin receptor \[[@REF16]\].
In another interesting study with post-menopausal iron deficient women, eight weeks of microencapsulated iron pyrophosphate (liposomal) was supplemented. There was significant rise in mean serum Hb, and hematocrit. They also reported higher tolerability with improved adverse effects as compared to previous conventional regimes of iron supplements taken by these women \[[@REF17]\]. In a study with chronic kidney disease patients, parenteral iron supplementation was compared with oral liposomal iron and it was seen that with eight weeks of therapy, liposomal iron group had significant increase in serum Hb from baseline while the other group did not have significant rise in serum Hb \[[@REF18]\]. Similarly, in this study, there was a significant increase in serum Hb levels with liposomal iron supplementation.
In another study conducted with microencapsulated iron pyrophosphate in liposomal form (Ferfer®), the mean taste score on Visual Analogue Scale (VAS) was 2.92 ± 2.44 with other forms of iron supplementation. VAS score increased to 7.66 ± 1.32 immediately after taking Ferfer® and to 7.96 ± 1.37 after five minutes \[[@REF14]\].
Microencapsulated iron pyrophosphate in liposomal form is a novel advancement in management of iron deficiency anemia. This salt is "generally recognized as safe (GRAS)" by United States Food and Drugs Administration (USFDA) Code of Federal Regulation. Furthermore, European Food Safety Authority (EFSA) has also declared iron pyrophosphate to be a safe food additive \[[@REF12]\]. Comparatively to conventional oral iron salts, microencapsulated liposomal iron has the highest bioavailability. It leads to quicker increase in serum hemoglobin levels, its taste has better palatability, and it doesn't have unwanted effects such as heartburn, GI upset, and constipation.
Microencapsulated liposomal iron is an effective and efficacious means of iron replenishment in deficient populations. Further studies are recommended to evaluate the safety profile and adherence to therapy with microencapsulated liposomal iron as compared to other conventional oral iron salts. Although results of the current study are very favorable, however, there is a need for comparative studies with other forms of iron available. There is a need for much larger studies to generalize the findings.
Conclusions
===========
Iron deficiency anemia is an easily manageable yet highly prevalent condition. The population at risk includes children, adolescents, and women of reproductive age in underdeveloped and low economy countries. The key factor to efficacious treatment of IDA is adherence to therapy. Adherence is governed by higher palatability and lesser side effects. Microencapsulated liposomal iron pyrophosphate sachets come with enhanced palatability, higher bioavailability, and consequently increased adherence among people with IDA.
The authors have declared that no competing interests exist.
Consent was obtained by all participants in this study
**Animal subjects:** All authors have confirmed that this study did not involve animal subjects or tissue.
| {
"pile_set_name": "PubMed Central"
} |
1. Introduction {#sec1-jcm-08-00774}
===============
Carbamoylation is a post-translational modification, playing a role in chronic kidney disease (CKD), comparable to the role of glycation in diabetes mellitus \[[@B1-jcm-08-00774],[@B2-jcm-08-00774]\]. This non-enzymatic reaction is characterized by a covalent binding of isocyanic acid to the amino group (either the epsilon amino group of lysine residus or the N-terminal amino group) of amino acids, polypetides, and (lipo)proteins. This post-translational molecular modification contributes to the molecular ageing of proteins. Isocyanic acid is formed continuously in an equilibrium reaction with urea or by myeloperoxidase (MPO). MPO is an enzyme (present in e.g., neutrophils, monocytes, and some tissue macrophages) that catalyzes the oxidation of thiocyanate in the presence of hydrogen peroxide, producing isocyanate at inflammation sites (e.g., atherosclerotic plaques) \[[@B3-jcm-08-00774]\]. Plasma isocyanic acid concentrations increase with a declining kidney function \[[@B4-jcm-08-00774]\].
An increased cardiovascular morbidity and mortality (10--30 times higher than in the general population) has been reported in the end-stage renal disease (ESRD) population \[[@B5-jcm-08-00774]\], which can be partly attributed to the influence of carbamoylated lipoproteins \[[@B6-jcm-08-00774]\]. In renal failure, dyslipidemia contributes to a worsening kidney function. Proteinuria is accompanied by a marked elevation of low-density lipoproteins (LDL) \[[@B7-jcm-08-00774]\]. Due to its carbamoylation, LDL can exert its prothrombotic \[[@B8-jcm-08-00774]\] and atherosclerosis-prone effects by stimulating an increased adhesion of monocytes to endothelial cells \[[@B9-jcm-08-00774]\], by inducing endothelial dysfunction \[[@B6-jcm-08-00774]\] and endothelial mitotic cell death \[[@B10-jcm-08-00774]\], and by promoting smooth muscle proliferation \[[@B11-jcm-08-00774]\]. Carbamoylated LDL (cLDL) has been identified as the most abundant modified LDL isoform in human blood, which is also present in healthy individuals \[[@B12-jcm-08-00774]\]. It is generated by carbamoylation of apolipoprotein B, the protein component of the LDL particle \[[@B13-jcm-08-00774]\]. CKD is also associated with decreased serum high-density lipoprotein (HDL) concentrations. Carbamoylation of HDL leads to a loss of the atheroprotective function of HDL, illustrated by an impaired ability to promote cholesterol efflux from macrophages \[[@B14-jcm-08-00774]\].
As carbamoylation is of major clinical importance, practical biomarkers for assessing carbamoylation and lipoprotein carbamoylation, in particular, are needed. In the present study, we explored the possibilities of infrared (IR) spectroscopy to assess non-HDL carbamoylation. The advantages of IR spectroscopy to determine lipid profiles were already applied in the past \[[@B15-jcm-08-00774]\]. In this paper, the effects of in vitro carbamoylation on spectral changes of non-HDL were studied. Furthermore, non-HDL carbamoylation was investigated in healthy subjects, in non-dialysis (CKD stage 3--5) as well as in hemodialysis patients (CKD stage 5d).
2. Materials and Methods {#sec2-jcm-08-00774}
========================
2.1. Study Participants {#sec2dot1-jcm-08-00774}
-----------------------
The control group consisted of 45 healthy subjects (median age: 28 years, interquartile range (IQR): 24--33 years), whereas the patient group consisted of 84 CKD patients (CKD stage 3--5: *n* = 37, median age: 70 years, IQR: 56--75 years, and CKD stage 5d (hemodialysis): *n* = 47, median age: 67 years, IQR: 56--75 years) of the Department of Nephrology, Ghent University Hospital. The approval of this study was granted by the Ethical committee of the Ghent University Hospital (EC/2015/0932).
2.2. In Vitro Carbamoylation of Lipids {#sec2dot2-jcm-08-00774}
--------------------------------------
In vitro carbamoylation of lipids was achieved by adding increasing volumes of 0.5 mol/L potassium cyanate (KOCN) solution (Sigma--Aldrich, St. Louis, MO, USA) in a phosphate buffered salt (PBS) solution (0.1 mol/L, pH 8.0) (Sigma--Aldrich, MO, USA) to 1000 µL serum of healthy subjects. Serum samples were carbamoylated using increasing concentrations of KOCN: 0 mmol/L, 20 mmol/L, 50 mmol/L, 80 mmol/L and 100 mmol/L. In vitro carbamoylation was carried out for 48 h at 37 °C (these reaction conditions warrant a completeness of the reaction). Proof of carbamoylation was obtained by verifying the electrophoretic mobility of lipoprotein fractions on a lipoprotein agarose electrophosis (Hydragel 7, Sebia, Lisses, France) using a semi-automated HYDRASYS instrument (Sebia, Lisses, France). The separated lipoproteins were stained with a lipid-specific Sudan black stain. The excess of stain was removed with an alcoholic solution. The resulting electropherogram was evaluated visually.
2.3. In Vitro Oxidation {#sec2dot3-jcm-08-00774}
-----------------------
Oxidative stress is involved in the exacerbation of disease burden in CKD patients. In vitro oxidation of serum was performed to reveal potential influences on the infrared spectrum. Ten samples from the serum pool at the laboratory of clinical biology of the University Hospital in Ghent were randomly selected. One milliliter of each sample was pooled. Before the oxidation process, non-HDL were precipitated 5 times. According to a modified version of the method used by Coffey et al. \[[@B16-jcm-08-00774]\], oxidized non-HDLs were prepared by dialyzing 4 mL of the serum pool against 400 mL isotonic saline (0.15 mol/L NaCl (VWR International, Haasrode, Belgium) dissolved in distilled water) containing 60 µmol/L CuSO4 (copper(II)sulphate pentahydrate, Merck Eurolab, Leuven, Belgium) during one hour. The dialysate was then changed to isotonic saline containing 0.5 mmol/L EDTA (BDH Chemicals, Poole, England) and dialysis continued during one hour with changes of dialysate every 15 min. After the oxidation process, non-HDL were precipitated 5 times from the oxidized serum pool.
2.4. Precipitation Procedure {#sec2dot4-jcm-08-00774}
----------------------------
All serum samples were centrifugated during 10 min at 3000× *g*. A precipitation reaction was performed, in which non-HDL fats very-low-density lipoprotein (VLDL), intermediate-density lipoprotein (IDL), lipoprotein(a), LDL and chylomicrons) were precipitated. 20 µL of a 13 mmol/L sodium phosphotungstate hydrate solution (Sigma Aldrich St Louis, MO, USA) and 5 µL of 2 mol/L MgCl~2~ (E. Merck KG, Darmstadt, Germany) were added to 200 µL serum. After vortexing, the samples were centrifuged (10 min, 6000× *g*) (centrifuge 54515 D, Eppendorf, Hamburg, Germany) \[[@B17-jcm-08-00774]\]. The precipitate was subsequently dried for 48 h in an incubator. Completeness of the precipitation reaction was assessed by lipid electrophoresis of the serum pre- and post-precipitation (5 samples were precipitated in triplicate). The formed pellet was ground prior to analysis with attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy.
2.5. ATR-FTIR Analysis {#sec2dot5-jcm-08-00774}
----------------------
All spectra were obtained by a Perkin Elmer Two ATR-FTIR spectrometer with the ATR accesory and spectrum 10 software (Perkin Elmer, Waltham, MA, USA). Before and after each analysis, the 50 mm ZnSE crystal was thoroughly cleaned with an alcoholic solution (Dax alcoliquid, Dialex biomedica, Sweden). A background scan was taken after the complete evaporation of the alcoholic solution. The lipoprotein powder was placed in contact with the surface of the crystal until complete covering. The pressure applied to the sample was standardised at 100 gauche to obtain a good contact between the sample and the crystal.
Three peaks were investigated: the carbonyl peak, the peak of the amide I band and the peak of the amide II band. Within-run coefficients of variation (CV) and between-run CV were calculated. The area under the curves (AUC) of the amide I and amide II bands were obtained by auto-labeling of the peaks in the Perkin Elmer 10 software.
Spectra were analyzed using the software program SIMCA version 14.1 (Umetrics, Sartorius Stedim Biotech, Umeå, Sweden). SIMCA (soft independent modeling of class analogy) was used to identify the spectral changes due to carbamoylation of lipoproteins. By applying various spectral filters, the noise was eliminated and the region of interest was selected. Data were normalized using the standard normal variate (SNV) method and were converted to their second derivative. The Savitsky-Golay algorithm allowed smoothing of the spectrum.
2.6. Routine Laboratory Measurements {#sec2dot6-jcm-08-00774}
------------------------------------
After overnight fasting, blood samples were collected and centrifuged (10 min, 3000× *g*). Urea, creatinine, albumin, triglycerides, total and HDL-cholesterol concentrations were assayed using commercial reagents on a Cobas 8000 analyzer (Roche, Mannheim, Germany) \[[@B18-jcm-08-00774]\]. The serum concentration of apolipoprotein B was determined by immunonephelometry on a Behring BN II nephelometer (Siemens, Marburg, Germany) \[[@B19-jcm-08-00774]\]. The LDL-cholesterol concentration was estimated using the Friedewald-formula \[[@B20-jcm-08-00774]\]. The estimated glomerular filtration rate (eGFR) was calculated with the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula \[[@B21-jcm-08-00774]\].
2.7. Statistics {#sec2dot7-jcm-08-00774}
---------------
Statistical analyses were performed using MedCalc (MedCalc, Mariakerke, Belgium). Normality of distributions was tested using the D'Agostino Pearson test. Data are expressed as median ± IQR or mean ± standard deviation (SD). Differences between patient groups were assessed using the Student's *t*-test and the Kruskall--Wallis test. The effect of the biological parameters on the spectrum was evaluated using a multiple linear regression model. A *p*-value \< 0.05 was considered a priori to be statistically significant.
3. Results {#sec3-jcm-08-00774}
==========
3.1. In Vitro Carbamoylation {#sec3dot1-jcm-08-00774}
----------------------------
In vitro carbamoylation of lipoproteins in serum of healthy subjects was demonstrated by agarose gel electrophoresis, which showed a progressive increase in electrophoretic mobility of lipoproteins with increasing KOCN concentrations. The effectiveness of the precipitation reaction was illustrated by the disappearance of the LDL and VLDL fraction on agarose gel electrophoresis. Serum samples with the highest concentrations of KOCN showed a less efficient precipitation, probably due to an altered protein structure, which could interfere with the precipitation process.
Carbamoylation was further investigated with ATR-FTIR spectroscopy. [Figure 1](#jcm-08-00774-f001){ref-type="fig"} presents the different IR spectra of the pellet of precipitated lipoproteins, urea, and KOCN. The visual spectrum of urea and KOCN did not interfere with the IR spectra of the precipitated lipoproteins. The between-run and within-run CVs for the detection of the carbonyl band (4.5% and 4.9%), the amide I band (4.9% and 8.4%) and the amide II band (5.8% and 8.1%) were low.
Using the software package SIMCA 14.1, it was possible to differentiate non-carbamoylated from in vitro carbamoylated non-HDL. The data set was centered, normalized and fitted, and a loading line was formed from the cleaned data. We focused on the fingerprint region (1500--600 cm^−1^) and the amide I and amide II region (1700--1500 cm^−1^). Carbamoylation resulted in a small increase in the amide I band (1714--1589 cm^−1^) of the spectrum ([Figure 2](#jcm-08-00774-f002){ref-type="fig"}). The data set was reduced to the amide I band, normalized and the second derivative was taken before fitting. Moreover, the in vitro experiments showed a diminishing amide II band/amide I area under the curve (AUC) ratio with increasing KOCN concentrations.
In vitro oxidation revealed an increased absorption in the amide I and amide II band (*p* \< 0.01). However, the amide II/amide I ratio remained the same before and after in vitro oxidation (ratio 0.55).
3.2. In Vivo Samples {#sec3dot2-jcm-08-00774}
--------------------
[Table 1](#jcm-08-00774-t001){ref-type="table"} describes the general characteristics of the healthy subjects and the CKD patients. In the in vivo part of the study, the findings of the in vitro study were compared with the CKD patients' samples. The same spectral filters were applied and the amide I band was selected. There was a clear distinction between the various groups (healthy subjects, patients with CKD stage 3--5 and CKD 5d patients).
In addition, a significant difference in the amide II/amide I AUC ratio was observed between the healthy subjects and the CKD groups (*p* \< 0.0001) ([Figure 3](#jcm-08-00774-f003){ref-type="fig"}), as well as between the two CKD groups (non-dialysis versus hemodialysis patients). A negative correlation was observed between the amide II/amide I AUC ratio and the serum urea concentration (*r* = −0.63, *p* \< 0.0001).
Multiple regression analysis with the amide II/amide I AUC ratio as a dependent variable revealed that the serum urea concentration and the serum apolipoprotein B concentration were the main predictors ([Table 2](#jcm-08-00774-t002){ref-type="table"}).
4. Discussion {#sec4-jcm-08-00774}
=============
In the present study, we have demonstrated for the first time the detection of carbamoylated non-HDL using ATR-FTIR spectroscopy. More specifically, in vitro carbamoylation of non-HDL induced structural changes, which were clearly visible in the mid-IR spectrum of the lipid pellets. The amide I band, containing mainly C=O stretching vibrations of protein peptide bonds was identified as the relevant region. In the clinical part of this study, significant differences at the amide I band were observed between healthy subjects, patients with CKD stage 3--5 and hemodialysis patients (CKD stage 5d). The amide I band depends on the secondary structure of the backbone and is, therefore, the amide vibration, which is most commonly used for secondary structure analysis \[[@B22-jcm-08-00774]\]. The amide II mode is the out-of-phase combination of the N−H in plane bend and the C≡N stretching vibration with smaller contributions from the C=O in plane bend and the C≡C and N≡C stretching vibrations. Although the protein secondary structure and frequency correlate less straightforward than for the amide I vibration, the amide II band provides valuable structural information \[[@B23-jcm-08-00774]\]. Previous studies have attributed the amide I band to apolipoproteins \[[@B24-jcm-08-00774],[@B25-jcm-08-00774]\]. The change in the amide I band can be expected as the carbamoylation process alters the protein component of LDL, namely apolipoprotein B \[[@B13-jcm-08-00774]\].
Building on the in vitro model of carbamoylation, we showed that increasing serum KOCN concentrations resulted in a reduced amide II/amide I AUC ratio. These amide bands in the IR spectrum take part in the adding of the carbamoyl group on the amino group of the epsilon-amino group of lysine and the terminal amino groups. The amide II/amide I AUC ratio reflects the observed spectral changes due to carbamoylation. Significant differences of the amide II/amide I AUC ratio were observed between healthy subjects, patients with advanced stages of CKD and hemodialysis patients. As expected, this ratio showed a negative correlation with the serum urea concentration. This result is supported by earlier findings, showing a similar regression coefficient between % carbamoylated albumin and blood urea concentrations in ESRD subjects \[[@B26-jcm-08-00774]\].
As demonstrated in the multiple regression model, age was identified as a minor predictor of the amide II/amide I AUC ratio in comparison with apolipoprotein B and urea. Tissue accumulation of carbamoylated proteins may be considered as a general hallmark of ageing, linking cumulative metabolic alterations and age-related complications. In addition to the association with carbamoylation, many other nonenzymatic posttranslational modifications occur during the biological life of proteins, leading to protein molecular ageing \[[@B27-jcm-08-00774]\].
A limitation of the present study is the fact that we did not perform liquid chromatography tandem-mass spectrometry (LC-MS/MS) to objectify the amount of carbamoylation. This technique has already been used for the detection of carbamoylated albumin \[[@B26-jcm-08-00774]\], but not for carbamoylated non-HDL. However, the relationship between the amide II/amide I AUC ratio and the KOCN concentration, as well as its relationship with the serum urea concentration are very suggestive for the carbamoylation process of non-HDL. A potential confounder could be the effect of diabetes mellitus, as the lysine residues are susceptible to both carbamoylation and glycation by glucose \[[@B28-jcm-08-00774]\]. However, previous work of our research group showed no significant changes in the amide I and amide II band after in vitro glycation of keratins in nail powder. After incubation of nail powders with respectively 1 mL of 0.9% sodium chloride solution, 5% glucose solution and 10% glucose solution, a clear difference in the area under the infrared peak was observed at wavenumber 1047 cm^−2^, a region characterized by a characteristic carbohydrate absorption. This band was used for measuring the degree of keratin glycation. No influence of glycation products was observed on the amide I and amide II bands \[[@B29-jcm-08-00774]\]. In the CKD stage 3-5 group and in the CKD 5d group, there were respectively 29.7% and 31.3% diabetics as compared to 0% in the group of healthy subjects. In addition, also oxidation may modify lipoproteins in a similar way as it has been demonstrated by amino acid analysis that modification of lysine residues occurs during LDL oxidation \[[@B30-jcm-08-00774]\]. However as demonstrated in our in vitro experiments, oxidation had no important influence on the reported results.
5. Conclusions {#sec5-jcm-08-00774}
==============
Carbamoylation is involved in the pathogenesis of various diseases (atherosclerosis, kidney diseases, autoimmune diseases, infections, and thrombus formation) and has been identified as an important risk factor for mortality in dialysis patients or in those with accelerated atherogenesis. The development of novel tools to determine the degree of post-translational modification-derived products is a demanding task. At this moment, the majority of potential carbamoylation biomarkers can only be assessed by rather complex analytical methods, hampering their use in clinical practice \[[@B2-jcm-08-00774]\].
In the present study, we have demonstrated that ATR-FTIR spectroscopy is an easy-to-use, reagent-free, and cost-effective method. It is a non-destructive technique, consuming only a small amount of sample \[[@B31-jcm-08-00774]\]. Spectral changes of non-HDL were observed depending on a declining kidney function. So, ATR-FTIR can be regarded as a new method for identification of carbamoylated non-HDL in CKD patients.
Conceptualization, J.R.D.; analysis, S.E.D., S.D.B. and L.D.B.; writing, review and editing, S.E.D., W.V.B. and M.M.S.
This research was funded by an Assistant Academic Personnel Grant from the Ghent University (S.E.D.) and by a Senior Clinical Researcher Grant from the Research Foundation Flanders (FWO) (M.M.S.). The APC was funded by the Ghent University Hospital.
The authors declare no conflict of interest.
{#jcm-08-00774-f001}
{#jcm-08-00774-f002}
{#jcm-08-00774-f003}
jcm-08-00774-t001_Table 1
######
General characteristics of the healthy subjects and the chronic kidney disease patients.
Healthy Subjects CKD Stage 3, 4 or 5 CKD Stage 5d *p*
---------------------------- ------------------- ---------------------- ---------------------- ----------
N 45 37 47
Median age (years) 28 (24--33) 70 (56--75) 67 (56--75) \<0.0001
\% diabetes mellitus 0 30 32
Urea (mmol/L) 8.9 (7.8--9.9) 25.0 (18.7--37.5) 32.8 (28.5--40.1) \<0.0001
Creatinine (µmol/L) 72.5 (62.8--80.0) 160.9 (138.6--243.8) 627.6 (474.7--774.4) \<0.0001
eGFR (mL/min/1.73 m^2^) \>90 31.0 ± 13.6 \<15 \<0.0001
Albumin (g/L) 47.3 (45.0--50.0) 42.2 (40.3--44.5) 40.0 (36.2--42.9) \<0.0001
Total cholesterol (mmol/L) 4.8 (4.3--5.5) 4.7 (3.7--5.5) 4.3 (3.7--5.8) NS
HDL cholesterol (mmol/L) 1.6 (1.3--2.0) 1.3 (1.0--1.6) 1.1 (0.9--1.5) \<0.0001
LDL cholesterol (mmol/L) 2.7 (2.3--3.1) 2.3 (1.8--3.1) 2.4 (1.7--3.2) NS
Triglycerides (mmol/L) 1.0 (0.8--1.3) 1.7 (1.1--2.3) 1.4 (1.1--2.4) =0.0002
Apolipoprotein B (g/L) 0.8 (0.7--1.0) 0.8 (0.7--1.0) 0.8 (0.7--0.9) NS
NS = not significant.
jcm-08-00774-t002_Table 2
######
Multiple regression model with the amide II/amide I area under the curves (AUC) ratio as dependent variable.
Variable β (Standard Error) *p*
--------------------------------------------------------- --------------------------- ---------------------- --------
Amide II/amide I AUC ratio, *r*^2^ = 0.54, *p* \< 0.001 Triglycerides (mmol/L) −0.001819 (0.001367) 0.1858
Apolipoprotein B (g/L) −0.0306 (0.006277) \<0.0001
Creatinine (µmol/L) −0.00001067 (0.000006989) 0.1293
Urea (mmol/L) −0.0006622 (0.0001436) \<0.0001
Age (years) −0.0002128 (0.00008219) 0.0108
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#s1}
============
Chronic pain is a major public health epidemic, debilitating more than 10% of adults globally with afflictions like persisting headaches, lower back pain, and rheumatoid arthritis ([@bib16]; [@bib17]). While some patients are responsive to commonly used analgesics, many are resistant to treatment ([@bib12]; [@bib15]). The P2X7 receptor, an extracellular ATP-gated ion channel predominantly expressed in immune cells of the blood and the brain ([@bib44]; [@bib19]), is an emerging target for treating refractory pain ([@bib9]; [@bib18]; [@bib6]; [@bib36]; [@bib41]). Over the last decade, enormous effort has been made to develop a number of structurally distinct P2X7 specific antagonists, some of which have been demonstrated to alleviate chronic pain in animal models ([@bib22]; [@bib32]; [@bib3]). However, mechanisms of action for these drugs remain poorly understood, hampering the development of effective therapeutic compounds for human patients ([@bib30]; [@bib43]; [@bib4]).
P2X receptors are trimeric ligand-gated ion channels that facilitate extracellular-ATP mediated signaling along with the G protein-coupled P2Y receptors ([@bib37]; [@bib7]). The P2X7 receptor belongs to the P2X receptor family, however, it was originally identified as a unique ATP-receptor named \"the P2Z receptor\", as it harbors many characteristics distinct from P2X and P2Y receptors ([@bib10]). For instance, the P2X7 receptor requires an unusually high concentration of ATP (EC50 ≥ 1 mM under physiological ion concentrations) for initial activation ([@bib47]), its channel activity is facilitated by prolonged or repeated ATP applications ([@bib44]), and it opens a membrane pore large enough for molecules up to \~900 Da to permeate ([@bib42]; [@bib38]; [@bib48]). While crystal structures of the P2X3 and P2X4 receptors have uncovered common mechanisms such as ATP-binding and gating for the P2X receptor family ([@bib28]; [@bib20]; [@bib31]), many questions continue to exist regarding subtype specific mechanisms, especially for the enigmatic P2X7 receptor subtype. How do structurally-unrelated drugs antagonize only the P2X7 receptor but not the other P2X subtypes? Do they target an activation mechanism unique to the P2X7 subtype? Here we identified and mapped the binding site for the P2X7 specific inhibitors for the first time using X-ray crystallography, and demonstrated by electrophysiological experiments that those inhibitors allosterically abrogate conformational changes associated with P2X7 receptor activation.
Results {#s2}
=======
Architecture of a mammalian P2X7 receptor {#s2-1}
-----------------------------------------
To define the structural basis for drug binding, we first sought to determine the crystal structures of a mammalian P2X7 receptor in the presence of five structurally distinct antagonists (A740003 ([@bib22]), A804598 ([@bib11]), AZ10606120 ([@bib33]), GW791343 ([@bib34], [@bib35]), and JNJ47965567 ([@bib5])). Using fluorescence detection size exclusion chromatography (FSEC) ([@bib27]), we screened full-length and a series of C-terminally truncated versions of nine human P2X7 (hP2X7) orthologues (72--89% identical), and found that an artificially truncated version of the panda (*Ailuropoda melanoleuca*) P2X7 receptor (pdP2X7) not only expressed better than other orthologues in insect cells but also remained trimeric and monodisperse in detergents commonly used for crystallography ([Figure 1A](#fig1){ref-type="fig"}). Whole cell patch clamp recordings confirmed that pdP2X7 presents comparable characteristics to hP2X7 ([Figure 1B--G](#fig1){ref-type="fig"}). The truncated pdP2X7 was further optimized to obtain a construct termed pdP2X7cryst(Δ1-21/Δ360-600/N241S/N284S/V35A/R125A/E174K) that we used to solve the crystal structures at \~3.5 Å resolution. pdP2X7~cryst~ exhibited slower deactivation and no obvious current facilitation (run-up) after repeated ATP applications ([Figure 1C and F](#fig1){ref-type="fig"}).10.7554/eLife.22153.002Figure 1.Characterization of pdP2X7.(**A**) FSEC traces (Ex: 488 nm and Em: 509 nm) for the full length (blue) and the truncated (green) pdP2X7. (**B**) and (**C**) Whole cell patch clamp recordings from the wildtype pdP2X7 (**B**) and pdP2X7~cryst~ (**C**) triggered by 1 mM ATP. (**D**) ATP dose responses of pdP2X7 (square) and pdP2X7~cryst~ (circle) determined by the whole cell patch clamp experiments. The plots were made using the normalized means of five independent experiments and the error bars represent SEM. Dose response curves were fit with the Hill equation. EC50 values of pdP2X7 and pdP2X7~cryst~ were 122 μM and 40 μM, respectively. (**E**) and (**F**) Whole cell patch clamp recordings from pdP2X7 (**E**) and pdP2X7~cryst~ (**F**) expressed in HEK293 cells. Each trace represents the pdP2X7 mediated current triggered by 1 s applications of 100 μM ATP from the same patch. The numbers below the traces indicate the number of repeated ATP applications. (**G**) Whole cell patch clamp recordings of pdP2X7 triggered by 1 mM ATP in the presence of different P2X7 specific antagonists. Drugs were applied for 1 min in the presence of ATP. Membrane was held at −60 mV. Concentrations of the drugs were; A740003: 600 nM; A804598: 180 nM; AZ10606120: 2.3 μM; GW791343: 50 μM; JNJ47965567: 136 nM.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.002](10.7554/eLife.22153.002)
Overall, a single subunit of the P2X7 receptor resembles the 'dolphin-like' shape of zebrafish P2X4 (zfP2X4; 45% identical to pdP2X7) ([@bib28]; [@bib20]) and human P2X3 (hP2X3; 38% identical to pdP2X7) ([@bib31]), the other P2X receptor subtypes with known architecture ([Figure 2A](#fig2){ref-type="fig"} and [Figure 2---figure supplement 1A](#fig2s1){ref-type="fig"}). The P2X7 structure obtained in the absence of antagonists (apo-form) likely represents a closed conformation, as the transmembrane helices constrict the channel gate at residues G338, S339, and S342 ([Figure 2B](#fig2){ref-type="fig"} and [Figure 2---figure supplement 1B](#fig2s1){ref-type="fig"}). The structural resemblance of the transmembrane helices between our current P2X7 and the zfP2X4 in the closed state (root-mean-square displacement (RMSD) of the protomers is 2.8 Å) also supports that our apo-structure represents a closed conformation ([Figure 2---figure supplement 1C](#fig2s1){ref-type="fig"}). Likewise, antagonist-bound P2X7 structures represent the same closed conformation with a RMSD of less than 0.5 Å to the apo structure, indicating that these drugs likely stabilize a resting closed state.10.7554/eLife.22153.003Figure 2.Drug-binding pocket of the P2X7 receptor.(**A**) Cartoon representation of a \'dolphin-like\' single subunit of the apo pdP2X7 structure. Fourteen beta strands are labeled as β1-14. Each domain is colored consistent with the previous studies for better comparison ([@bib28]; [@bib20]). (**B**) Cartoon representation of the trimeric pdP2X7 structure viewed from the side. The black box indicates an approximate location of the upper body domains shown in (**C**) and (**D**). (**C**) Side view of the upper body domains exhibiting A804598 binding sites with respect to the ATP-binding pockets (orange dashed lines). A804598 is shown as CPK spheres. (**D**) Top view of the apo pdP2X7 structure with respect to the ATP-binding pockets (orange dashed lines) and one of the drug-binding pockets (green dashed line).**DOI:** [http://dx.doi.org/10.7554/eLife.22153.003](10.7554/eLife.22153.003)10.7554/eLife.22153.004Figure 2---figure supplement 1.Structural comparison between pdP2X7 and zfP2X4.(**A**) Superposition of the apo closed pdP2X7 monomer (green) onto the apo closed zfP2X4 structure (4DW0; gray). The two structures were superposed against each other at the body domain using Lsqkab software ([@bib24]). Notable differences between these two structures are highlighted in orange. (**B**) Stick representations of the pdP2X7 gate residues. Three TM2 helices are shown as cartoon in green. (**C**) Superposition of the transmembrane helices of the apo closed pdP2X7 monomer (green) onto the apo closed zfP2X4 structure (gray) Root-mean-square displacement of the Cα positions between monomeric pdP2X7 and zfP2X4 is 2.8 Å.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.004](10.7554/eLife.22153.004)10.7554/eLife.22153.005Figure 2---figure supplement 2.Electron density around the P2X7 specific antagonists.(**A**) 2Fo-Fc map (contoured at σ = 2.0) around the drug-binding pocket (orange) in the apo closed pdP2X7. (**B**) Fo-Fc map (contoured at σ = 3.0) around the P2X7 specific antagonists. Each drug is depicted as a stick representation. (**C**) Chemical structure of each P2X7 antagonist is shown.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.005](10.7554/eLife.22153.005)
A novel drug binding pocket {#s2-2}
---------------------------
Surprisingly, all five structurally-unrelated compounds bind in the same pocket formed between neighboring subunits, which is juxtaposed with the ATP-binding pocket ([Figure 2C--D](#fig2){ref-type="fig"} and [Figure 2---figure supplement 2](#fig2s2){ref-type="fig"}). This drug-binding pocket is surrounded by thirteen residues projecting mainly from β-strands (β4, β13 and β14) in the upper body domains of the neighboring subunits ([Figure 3A](#fig3){ref-type="fig"}). While the precise distances and angles between the side-chains and the drugs cannot be determined at the current resolutions (\~3.2--3.6 Å), electron density was clear enough to localize and orient the side chains of the drug-coordinating residues ([Figure 3---figure supplement 1](#fig3s1){ref-type="fig"} and [Video 1](#media1){ref-type="other"}). Drug binding seems to be mediated mainly by hydrophobic interactions, especially at positions deep within the cavity, involving F95, F103, M105, F293, and V312. Despite structural diversity, all five P2X7 antagonists fit within the drug-binding pocket ([Figure 3B](#fig3){ref-type="fig"}), highlighting that the size and the shape of the drug-binding pocket play major roles in determining the affinity and specificity of the drugs. Indeed, the equivalent pocket in the P2X4 receptor is too narrow to accommodate the smallest P2X7 antagonist, A804598, even though it is similarly hydrophobic to that of the P2X7 receptor ([Figure 3---figure supplements 2](#fig3s2){ref-type="fig"} and [3](#fig3s3){ref-type="fig"}). Thus, we suggest that the differences in the size of the inter-subunit hydrophobic pocket is the major factor that confers P2X7 specific binding of the inhibitors.10.7554/eLife.22153.006Figure 3.Coordination of the P2X7 specific antagonists.(**A**) Stick representations of the indicated P2X7 specific antagonists (green) and the binding residues in the drug-binding pocket. Structural frame of each subunit is depicted as a cartoon representation in a different color. Oxygen atoms are shown in red and nitrogen atoms are shown in blue. (**B**) Surface representation of the drug-binding pocket with CPK sphere representation of each drug (green). Yellow and light blue surfaces represent two of the three subunits. The third subunit was omitted for clear representation. (**C**) Fold changes in IC50 values of the P2X7 specific inhibitors against the YO-PRO-1 uptake activity on alanine mutants of the drug coordinating residues. Increased IC50 values in some of the alanine mutants indicate that these residues bind to the P2X7 antagonists. (**D**) Schild plots against drug concentrations over a range of three orders of magnitude. Plots for all five of the different P2X7 specific antagonists are non-linear, supporting that these drugs work non-competitively. (**E**) Whole cell patch clamp recordings of the wildtype pdP2X7 triggered by 10 μM ATP (black) or 10 μM ATP-Alexa (red). The holding potential was −60 mV. (**F**) Fluorescence anisotropy derived from 10 μM ATP-Alexa without protein (white), with pdP2X7 (black), or with pdP2X7 in the presence of 1 mM ATP (grey). The concentration of pdP2X7 was 100 μM. Concentrations of the drugs were: A740003: 600 nM; A804598: 180 nM; AZ10606120: 2.3 μM; GW791343: 50 μM; JNJ47965567: 136 nM. The dots and the bars represent the means of five independent experiments and the error bars represent SEM. Asterisks indicate significant differences from wildtype or the no protein control (p\<0.01) determined by one way ANOVA followed by Dunnett\'s test.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.006](10.7554/eLife.22153.006)10.7554/eLife.22153.007Figure 3---figure supplement 1.Side-chain electron density near the drug-binding pocket.(**A--C**) 2Fo-Fc electron density map of the JNJ47965567-bound P2X7 structure contoured at σ = 1.0. Electron density of β13 (**A**) and β14 (**B**) in the upper body domain, and the α-helix near the drug-binding pocket (**C**) are shown in red. The amino acid residues are depicted as stick representations. Red represents oxygen and blue represents nitrogen.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.007](10.7554/eLife.22153.007)10.7554/eLife.22153.008Figure 3---figure supplement 2.The equivalent position of the drug-binding pocket in the P2X4 receptor is too narrow for P2X7 antagonist to fit in.(**A**) Surface representation of the apo pdP2X7 receptor viewed from the side. Light blue, yellow, and light pink represent each subunit. Green residues are the A804598 binding residues (**F88, F95, F103, M105, F108 and Y295**). Light pink subunit is presented as semi-transparent surface for visualizing the yellow and blue subunits. (**B**) Zoomed in drug-binding pocket of the pdP2X7 shown in (**A**). (**C**) CPK sphere representation of A804598 at the same magnification and orientation as the drug binding site shown in (**B**). (**D**) Equivalent pocket in apo zfP2X4. The green residues represent counterparts of the A804598 binding residues (W87, I94, F103, L105, M108, and F289).**DOI:** [http://dx.doi.org/10.7554/eLife.22153.008](10.7554/eLife.22153.008)10.7554/eLife.22153.009Figure 3---figure supplement 3.Sequence alignment of P2X7 and P2X4 receptors.Amino acid sequences of the panda P2X7 (pdP2X7: XP_002913164.1), the human P2X7 (hP2X7: Q99572.4), the mouse P2X4 (mP2X4: NM_011026.3), and the zebrafish P2X4 (zfP2X4: AF317643.1). Starting and ending points of the pdP2X7~cryst~ are indicated by arrows. The drug surrounding residues are highlighted in green. Mutations at N-linked glycosylation sites (N241S and N284S) and at crystal contact sites in the P4~2~ crystal form (V35A, R125A, and E174K) are indicated. The C-terminal sequences beyond 359 AA are omitted for simplicity.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.009](10.7554/eLife.22153.009)10.7554/eLife.22153.010Figure 3---figure supplement 4.YO-PRO-1 uptake assays on the alanine mutants of the drug coordinating residues.(**A**) pdP2X7 mediated YO-PRO-1 uptake triggered by 1 mM ATP (blue). Vector control (red) shows little accumulation of YO-PRO-1. (**B**) Dose responses of the wildtype and each alanine mutant for the five P2X7 specific antagonists are shown. YO-PRO-1 uptake was initiated with 1 mM ATP application. IC50 values are obtained by fitting the dose responses with the Hill equation. The circles represent the mean values of five independent experiments and the error bars represent SEM.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.010](10.7554/eLife.22153.010)10.7554/eLife.22153.011Figure 3---figure supplement 5.The P2X7 specific antagonists are allosteric non-competitive inhibitors.(**A**) and (**B**) BzATP dose response of the wildtype pdP2X7 in the presence of different concentrations of antagonists. Normalized initial rates of YO-PRO-1 uptake are plotted and the dose response curves are fitted with a non-competitive (**A**) or competitive (**B**) inhibition model. The circles represent the means of five independent experiments and the error bars represent SEM. (**C**) Fitting statistics for the non-competitive and the competitive inhibition models. SS indicates the sum of squared difference. Both R-squared values and the p-values from the F-test support that the drugs used in this study are allosteric non-competitive inhibitors.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.011](10.7554/eLife.22153.011)10.7554/eLife.22153.012Figure 3---figure supplement 6.Fluorescent anisotropy experiments.(**A**) Schematic representation of the fluorescence anisotropy experiment. (**B**) Fluorescence emission spectra for 10 μM ATP-Alexa (red) and 100 μM GFP-tagged pdP2X7~cryst~ (green). Excitation wavelength was 590 nm. (**C**) An SDS-PAGE gel demonstrating that the GFP-tagged pdP2X7~cryst~ is pure. (**D**) Fluorescence anisotropy plotted against concentrations of the pdP2X7. Polarized emission from 10 μM ATP-Alexa in response to binding to P2X7 was recorded and fitted with the Hill equation with the slope (**n**) of 0.90. (**E**) Fluorescent anisotropy decreases as ATP concentration increases, indicating that ATP and ATP-Alexa competes for binding to pdP2X7. Competitive binding of ATP supports that ATP-Alexa binds specifically at the ATP-binding pocket. The circles represent the mean values of five independent experiments and the error bars represent SEM.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.012](10.7554/eLife.22153.012)Video 1.Electron density of the drug-coordinating residues.2Fo-Fc electron density map of the JNJ47965567-bound P2X7 structure contoured at σ = 1.0. Electron density is shown as mesh in firebrick red. JNJ47965567 (green) and amino acid residues (white) are depicted as stick representations. Red represents oxygen and blue represents nitrogen.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.013](10.7554/eLife.22153.013)10.7554/eLife.22153.013
We further validated the drug-receptor interactions by site-directed mutagenesis on the drug-binding residues. To facilitate robust and systematic data collection, we monitored cellular uptake of the fluorescent dye, YO-PRO-1, as a proxy for receptor activity ([@bib44]) ([Figure 3---figure supplement 4A](#fig3s4){ref-type="fig"}). Consistent with the crystal structures, the mutants, F88A, F95A, F103A, M105A, F108A, and V312A, showed increased IC50 values (see [Table 1](#tbl1){ref-type="table"} for IC50 values on the wildtype), supporting that these residues play important roles in drug binding ([Figure 3C](#fig3){ref-type="fig"} and [Figure 3---figure supplement 4B](#fig3s4){ref-type="fig"}). In particular, interaction with F103 is crucial for the inhibitory action of all five drugs.10.7554/eLife.22153.014Table 1.IC50 value of each P2X7 inhibitor from the YO-PRO-1 uptake assay.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.014](10.7554/eLife.22153.014)InhibitorA740003A804598AZ10606120GW791343JNJ47965567IC5069.3 nM21.7 nM231 nM8.9 μM11.9 nM
Allosteric non-competitive inhibition {#s2-3}
-------------------------------------
Binding of all five drugs to a site distinct from the ligand-binding pocket suggests that these compounds act as non-competitive inhibitors. However, previous studies using cell-based Ca^2+^ influx and IL-1β release assays proposed that three of those compounds, A740003, A804598, and JNJ47965567, compete against ATP-binding to inhibit the P2X7 receptor ([@bib22]; [@bib11]; [@bib5]). To clarify the working mechanism for each drug, we measured the dose responses of the P2X7 mediated YO-PRO-1 uptake in the presence of each drug at multiple concentrations. We used BzATP---an artificial but potent agonist of P2X7 receptors---for these experiments to achieve saturating responses in the presence of concentrated antagonist. For each drug, dose response curves fit well with a non-competitive inhibition model ([@bib29]), but poorly with a competitive model ([Figure 3---figure supplement 5](#fig3s5){ref-type="fig"}). Furthermore, the Schild plots displayed non-linear relationships, especially at higher concentrations, consistent with a non-competitive mechanism ([Figure 3D](#fig3){ref-type="fig"}) ([@bib39]).
To confirm the non-competitive mode of inhibition, we performed an ATP-binding assay on purified P2X7 receptor pretreated with each drug. We exploited fluorescence anisotropy ([Figure 3---figure supplement 6A--C](#fig3s6){ref-type="fig"}) using a fluorescently-labeled ATP analogue (Alexa-ATP), which is as potent as ATP and is capable of triggering P2X7 channel opening at 10 μM in the absence of divalent cations ([Figure 3E](#fig3){ref-type="fig"}). At this concentration, Alexa-ATP gave rise to fluorescence anisotropy when incubated with P2X7 in a dose-dependent manner ([Figure 3---figure supplement 6D](#fig3s6){ref-type="fig"}). Notably, drug-treated P2X7 receptor exhibited similar levels of fluorescence anisotropy, which drastically decreased in the presence of unlabeled-ATP ([Figure 3F](#fig3){ref-type="fig"} and [Figure 3---figure supplement 6E](#fig3s6){ref-type="fig"}). Together, these experiments strongly support that all five P2X7 antagonists are allosteric non-competitive inhibitors.
P2X7 specific conformational change during channel activation {#s2-4}
-------------------------------------------------------------
We propose that the five studied drugs antagonize the P2X7 receptor through a common mechanism and that the unique inter-subunit cavity may be a critical locus for functional regulation. Interestingly, the inter-subunit cavity formed by β13 and β14 in the upper body domain is much wider in P2X7 than in zfP2X4 or in hP2X3 ([Figure 4A--C](#fig4){ref-type="fig"}) ([@bib31]). Furthermore, this \'turret-like\' structure and the cleft corresponding to the P2X7 drug-binding pocket remain relatively occluded in zfP2X4 and in hP2X3 after activation by ATP ([Figure 4A--C](#fig4){ref-type="fig"} and [Figure 4---figure supplement 1](#fig4s1){ref-type="fig"}) ([@bib31]). Do the drug-binding pocket and the turret in P2X7 become narrower during channel activation? To explore the involvement of the inter-subunit cavity in P2X7 receptor activation, we monitored the movement of the cavity residues with and without ATP. We first created a series of single cysteine mutants in the drug-binding pocket ([Figure 4D](#fig4){ref-type="fig"}) and measured ATP-gated channel activity after applying a large cysteine-reactive agent (MTS-TPAE; Mw: 447 Da). We reasoned that modification of a cysteine residue with a bulky moiety should interfere with the conformational changes necessary for channel opening, thereby resulting in diminished channel activity. When MTS-TPAE was applied in the absence of ATP, four cysteine mutants (F103C, K110C, T308C, and I310C) showed irreversible current reduction ([Figure 4E and F](#fig4){ref-type="fig"}), consistent with the idea that the covalently bound MTS-TPAE at these positions hinders the conformational changes required for channel opening. When MTS-TPAE was applied in the presence of ATP, on the other hand, none of the cysteine mutants presented significant current reduction ([Figure 4E and F](#fig4){ref-type="fig"}). These results indicate that at least four residues in the drug-binding pocket are more accessible to MTS-TPAE in the closed state than in the open state. Our data therefore suggest that the drug-binding pocket narrows upon ATP binding and that such a conformational rearrangement is crucial for P2X7 channel opening.10.7554/eLife.22153.015Figure 4.Drug-binding pocket narrows during P2X7 activation.(**A**) Cartoon representation of the turret formed by β13 and β14. P2X7 (apo), zfP2X4 (apo), and ATP-bound zfP2X4 are shown as the top (left) and the side (right) views. (**B**) Dot representations of the internal-space along the molecular threefold axis running through the center of the apo closed pdP2X7 (top), the apo closed zfP2X4 (4DW0; middle), and the ATP-bound zfP2X4 (4DW1; bottom). The dot plots are made using the HOLE program ([@bib40]). Purple: \> 2.3 Å; green: 1.15--2.3 Å; red: \< 1.15 Å. (**C**) Central pore radii of the three P2X crystal structures shown in (**B**). (**D**) Surface representation of the residues in the drug-binding pocket tested for accessibility. (**E**) Whole cell patch clamp recordings of the wildtype pdP2X7 and the I310C mutant triggered by 1 mM ATP. MTS-TPAE (1 mM) was applied in the absence (left) or presence (right) of ATP for 10 s to probe the accessibility in the closed or open state, respectively. The membrane was held at −60 mV. (**F**) Normalized channel activities after MTS-TPAE treatment during the closed (grey) or open (black) states. The bars represent the means of five or more independent experiments (numbers above the bars indicate the n value) and the error bars represent SEM. Asterisks indicate significant differences from the widltype control (p\<0.01) determined by one way ANOVA followed by Dunnett\'s test.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.015](10.7554/eLife.22153.015)10.7554/eLife.22153.016Figure 4---figure supplement 1.Volume of the P2X7 drug-binding pocket is larger than the corresponding clefts in P2X3 or P2X4.(**A--C**) Volumes of the drug-binding pocket in P2X7 (**A**), and equivalent regions in P2X4 (**B**) and P2X3 (**C**) are depicted as mesh representation based on the filled water molecules using HOLLOW ([@bib21]). Green mesh represents the volume in apo, closed state and the red mesh represents the volume in ATP-bound, open state. Side view (upper panel) and the top view (lower panel) include cartoon representation of the corresponding P2X structures to locate the cleft.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.016](10.7554/eLife.22153.016)
To examine the movement of the turret during P2X7 activation, we took advantage of a cysteine mutation at Y295, whose side chain faces the center of the turret ([Figure 5A](#fig5){ref-type="fig"}). Because two of the three introduced cysteines may form a disulfide bond, we pretreated the cells with a reducing agent prior to cysteine accessibility experiments. Under these conditions, MTS-TPAE diminished the Y295C channel activity by \~80% in the absence of ATP but to a lesser extent (\~40%) in the presence of ATP ([Figure 5B and C](#fig5){ref-type="fig"}). These data support that the turret also narrows during P2X7 activation. Widening of the turret upon ATP-binding is unlikely, as 1) MTS-TPAE had no effect on the Y295C mutant prior to treatment with a reducing agent ([Figure 5D](#fig5){ref-type="fig"}), indicating the formation of a disulfide bond that would bring the two neighboring subunits closer (Cβ-Cβ distance would change from \~14 Å to \~4 Å) and 2) Y295C mutant exhibited comparable current density with the wildtype in the absence of a reducing agent ([Figure 5---figure supplement 1A](#fig5s1){ref-type="fig"}). Notably, narrowing of the inter-subunit space seems unique to the P2X7 receptor, as suggested by the crystal structures of P2X3 and P2X4 solved in the presence of ATP. Indeed, MTS-TPAE application either in the presence or absence of ATP did not exhibit current reduction for cysteine mutants of the counterpart residues in the P2X4 receptor ([Figure 5B and E](#fig5){ref-type="fig"}, [Figure 5---figure supplement 1B and C](#fig5s1){ref-type="fig"}).10.7554/eLife.22153.017Figure 5.The turret closes during P2X7 activation.(**A**) Surface representation of the P2X7 receptor viewed from the side highlighting the location of Y295 (red) facing the center of the turret (blue). Only two of the three subunits are shown for clarity. (**B**) Summary of MTS-TPAE accessibility. Normalized channel activities after MTS-TPAE treatment indicate that Y295C in P2X7 is more accessible in the closed state than in the open state. The bars represent the means of more than five independent experiments (numbers above the bars indicate the n value) and the error bars represent SEM. Asterisk indicates significant difference in normalized channel activity between the cells treated with MTS-TPAE in the closed and in the open states (p\<0.01) determined by Dunnett\'s test. N.S. indicates not significant. (**C**) and (**D**) Whole cell patch clamp recordings of the P2X7/Y295C mutant triggered by 1 mM ATP. MTS-TPAE (100 μM) was applied in the absence (left) or presence (right) of ATP for probing the accessibility in the closed or open state, respectively. Cells were used before (**D**) or after treating with 10 mM DTT for 5 min (**C**). Membrane was held at −60 mV. (**E**) Whole cell patch clamp recordings of the P2X4/F296C mutant triggered by 10 μM ATP. Cells were treated with 10 mM DTT for 5 min before recordings.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.017](10.7554/eLife.22153.017)10.7554/eLife.22153.018Figure 5---figure supplement 1.The equivalent position of the drug-binding pocket in the P2X4 receptor is not accessible.(**A**) Current densities of the pdP2X7 wildtype (left) and the Y295C mutant (right) in the absence (gray) or presence (black) of 10 mM DTT. Currents were obtained by whole cell patch clamp recordings triggered by 1 mM ATP. Membrane potential was held at −60 mV. The bars represent the means (numbers above the bars indicate the n value) and the error bars represent SEM. N.S. indicates there is no significant difference between the wildtype and Y295C (p=0.216 without DTT; p=0.389 with DTT) by the t-test. (**B**) Whole cell patch clamp recordings of the wildtype mouse P2X4 (mP2X4) and the T312C mutant triggered by 10 μM ATP (closed state) or 5 μM ATP (open state). MTS-TPAE (1 mM) was applied in the absence (left) or presence (right) of ATP to probe the accessibility in the closed or open state. The membrane was held at −60 mV. (**C**) Normalized channel activities after MTS-TPAE treatment in the closed (grey) or open (black) states. Residues in parenthesis indicate the counterpart residues in pdP2X7, whose cysteine mutants exhibit current reduction by MTS-TPAE application in the closed state. The bars represent the means of five or more independent experiments (numbers above the bars indicate the n value) and the error bars represent SEM. There was no statistically significant differences from the widltype control (p\>0.05) judged by one way ANOVA followed by Dunnett\'s test.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.018](10.7554/eLife.22153.018)
To confirm that allosterically-bound P2X7 inhibitors prevent the turret from narrowing upon ATP-binding, we obtained the crystal structure of the P2X7 receptor in the presence of both ATP and A804598 ([Figure 6A--D](#fig6){ref-type="fig"} and [Figure 6---figure supplement 1](#fig6s1){ref-type="fig"}). Consistent with the activation mechanisms proposed for P2X3 and P2X4, ATP-binding brings the dorsal fin domain towards the head domain and pushes the left flipper domain away from the ATP-binding pocket ([Figure 6C](#fig6){ref-type="fig"}). These movements are coupled with widening of the lower body domain ([Figure 6D](#fig6){ref-type="fig"}), though to an extent in which the transmembrane helices remain closed. In contrast, little conformational change was observed for the upper body domain including the turret and the drug-binding pockets ([Figure 6A](#fig6){ref-type="fig"}), supporting that the turret closure is essential for P2X7 channel opening. Altogether, our data suggest that P2X7 receptors undergo unique conformational rearrangements where both the drug-binding pocket and the turret in the P2X7 receptor narrow upon ATP-binding. Because such conformational changes are required for channel opening, binding of the P2X7 antagonists preclude these constrictions, thereby efficiently blocking receptor activation ([Figure 7](#fig7){ref-type="fig"} and [Video 2](#media2){ref-type="other"}).10.7554/eLife.22153.019Figure 6.A804598 prevents conformational changes in the upper body domain triggered by ATP-binding.(**A**) Cartoon representations indicate little conformational change upon ATP-binding in the upper body domain. Each domain of the ATP/A804598-bound P2X7 is colored according to [Figure 2](#fig2){ref-type="fig"} and the overall structure is shown in (**B**). Apo P2X7 structure is presented in gray for comparison (**A, C** and **D**). (**C**) and (**D**) ATP-binding evokes conformational rearrangement in the ATP-binding pocket (**C**) and in the lower body domain (**D**). The arrows highlight the direction of the domain movement upon ATP binding.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.019](10.7554/eLife.22153.019)10.7554/eLife.22153.020Figure 6---figure supplement 1.P2X7 structure in the ATP/A804598-bound state.(**A**) and (**B**) Cartoon representations of the two protomers in the asymmetric unit. Clear ATP density was observed in chain B (**B**) but not in chain A (**A**). (**C**) Fo-Fc map (contoured at σ = 3.0) around the ATP molecule in chain B.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.020](10.7554/eLife.22153.020)10.7554/eLife.22153.021Figure 7.Mechanisms of P2X7 activation and inhibition.Schematic representations of the P2X7 receptor viewed from the top. Each color represents a different subunit. The drug-binding pocket and the turret narrow during channel activation. P2X7 specific antagonists stabilize its closed conformation by preventing the movement of these inter-subunit cavities.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.021](10.7554/eLife.22153.021)Video 2.**Conformational differences between the apo and ATP/A804598-bound P2X7 structures**.Each domain of the ATP/A804598-bound P2X7 is colored according to [Figure 2](#fig2){ref-type="fig"}. ATP and A804598 molecules are presented as CPK spheres.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.022](10.7554/eLife.22153.022)10.7554/eLife.22153.022
Discussion {#s3}
==========
The presented crystal structures uncover the unique inter-subunit cavity in the upper body domain of the P2X7 receptor. Our data suggest that this cavity shrinks during activation, which allows the lower body domain to widen further. Indeed, the ATP-binding left flipper domain is connected with the pore-lining transmembrane helix through the turret composed of two β-strands (β13 and β14), which supports that movements of the upper body domain are tightly coupled with the channel opening ([Figure 2A](#fig2){ref-type="fig"}). Although the lower body domain in our ATP/A804598-bound structure does seem to widen to some extent upon ATP-binding, the turret diameter remains locked due to A804598 binding, rendering the accompanying movement of the pore-lining transmembrane helices insufficient for channel opening ([Figure 6](#fig6){ref-type="fig"} and [Video 2](#media2){ref-type="other"}). Closure of the turret, therefore, seems to be a prerequisite for full-widening of the lower body domain, which is necessary for channel opening. Interestingly, the turrets of both P2X3 and P2X4 are narrower even in the closed state, which may facilitate subsequent conformational changes in the lower body domain. Obviously, the crystal structure of an ATP-bound, open-state P2X7 would strengthen our hypothesis. However, it is technically challenging to obtain such a crystal structure, as ATP-bound P2X7 receptors are prone to aggregation in a detergent containing solution. Future efforts to uncover molecular details about the open state of P2X7 may yield further insight into subtype-specific activation mechanism.
Our data demonstrate that all five tested drugs are allosteric and non-competitive inhibitors, which is in contrast to the previous studies showing that A740003, A804598, and JNJ47965567 are competitive inhibitors ([@bib23]; [@bib11]; [@bib5]). This discrepancy may be attributed to the high agonist concentrations (i.e. up to 3 mM ATP or 1 mM BzATP) required for obtaining reliable EC50 values in the presence of antagonists. With such a high concentration of agonists, we noticed that the YO-PRO-1 uptake activity starts to decline \~10 min after agonist application. To minimize potential experimental artifacts, we took advantage of the initial rates instead of total counts at a certain time point used by the previous studies. Nevertheless, the Schild plots of these three compounds appear to be linear at lower concentrations ([Figure 3D](#fig3){ref-type="fig"}), making it difficult to rely solely on the Schild regression analysis for determining the mode of drug action. We therefore provided extra evidence of non-competitive allosteric drug action using 1) curve fitting with known models of competitive and non-competitive inhibition, 2) ligand-binding assays in the presence of antagonists, and 3) the crystal structure of P2X7 obtained in the presence of both an agonist and an antagonist.
Combined with the crystal structures, information about the drug-receptor specific interactions provide an excellent platform for rationally improving physicochemical properties of currently available P2X7 antagonists. For example, replacing the phenyl group in JNJ47965567 with a slightly bulkier group may increase the binding affinity of this compound, as that would fill the void around residue F95. Likewise, low potency of AZ10606120 and GW791343 ([Table 1](#tbl1){ref-type="table"}) may be overcome by replacing chemical moieties with those fit better into the drug-binding pocket. Our data also suggest that it may be possible to attach an extra chemical group to the site away from the residue interacting with F103 without affecting binding of the P2X7 antagonists. This leads to an interesting possibility that these drugs could be conjugated with cell penetrating peptides or specific ligands expressed on endothelial cells, which may improve the delivery of drugs across the blood-brain barrier ([@bib8]). In addition, identification of an unexpected binding pocket within P2X7, that is absent in the closely related P2X4, should allow for engineering of isoform-specific drugs. Development of a new generation of P2X7 antagonists with increased clinical efficacy has the potential to have a significant impact on the treatment of chronic pain and inflammation.
Materials and methods {#s4}
=====================
Construct design {#s4-1}
----------------
Ten mammalian P2X7 genes (human: Q99572.4; giant panda: XP_002913164.1; horse: XP_001495622.1; cattle: NP_001193445.1; dog: NP_001106927.1; rabbit: XP_002719791.1; rat: Q64663.1; mouse: Q9Z1M0.2; opossum: XP_001373213.1; guinea pig: NP_001166578.1) were synthesized based on the protein sequences (GenScript, Piscataway, NJ) and subcloned into a modified pFastBac baculovirus expression vector (Thermo Fisher Scientific, Waltham, MA) that harbors a Strep-tag and EGFP at the N-terminus (pNGFP-FB3). These P2X7 constructs were expressed in Sf9 insect cells and screened using the FSEC strategy ([@bib27]). Briefly, Sf9 cells at a cell density of 1.0 × 10^6^ cells/mL in six well plates were infected with 20 μl/well of P1 virus (10^6^--10^7^ PFU/mL), harvested after 48 hr, and solubilized in 150 μl of S1 buffer (1% n-Dodecyl-β-D-Maltopyranoside (DDM; Anatrace, Maumee, OH) and Halt protease inhibitor cocktail (Thermo Fisher Scientific) in phosphate-buffered saline; (pH 7.4)) for 30 min at 4°C. The soluble fractions were collected by ultra-centrifugation at 87,000 × g for 30 min and analyzed by FSEC. Because the full-length P2X7 receptors were poorly expressed in Sf9 cells and did not crystallize in our initial trials, C-terminally truncated versions of the nine P2X7 receptor orthologues (\~240 residues removed from the C-terminus) were rescreened by FSEC. Based on the sharp and symmetrical peak profile, the C-terminally truncated panda P2X7 (pdP2X7) was selected for crystallization trials. A series of terminal deletions and glycosylation mutants of pdP2X7 were screened for improving the expression level, stability, and monodispersity. The best behaved construct (Δ1-21/Δ360-600/N241S/N284S) formed crystals in the P4~2~ space group that diffracted to \~7 Å, which was sufficient for obtaining an initial electron density map by molecular replacement using the crystal structure of the zebrafish P2X4. To improve crystal packing, we screened combinations of point mutants at the predicted crystal contact sites for crystallization/diffraction behavior. Finally, we obtained a construct termed pdP2X7~cryst~ (Δ1-21/Δ360-600/N241S/N284S/V35A/R125A/E174K), which formed crystals in a different space group (I2~1~3) that diffracted to \~3.5 Å.
Expression and purification {#s4-2}
---------------------------
A detailed protocol for expression and purification of pdP2X7 is available at Bio-protocol ([@bib26]). pdP2X7~cryst~ was expressed as an EGFP-fusion protein using a baculovirus-insect cell system. Sf9 cells were infected at 3.5--4.0×10^6^ cell/mL with 25--30 mL/L P2 virus (10^7^--10^8^ PFU/mL) for one day at 27°C and for another two days at 18°C. Cells were harvested by centrifugation at 2040 ×g, washed with PBS, and resuspended with PBS containing leupeptin (0.5 μg/mL), aprotinin (2 μg/mL), pepstatin A (0.5 μg/mL), and phenylmethylsulfonyl fluoride (0.5 mM). The cells were broken by nitrogen cavitation at 600 psi using a 4635 cell disruption vessel (Parr Instrument, Moline, IL). After removing unbroken cells and debris by centrifugation at 12,000 × g for 10 min, the membrane fraction was collected by centrifugation at 185,000 × g for one hour and solubilized in S2 buffer (2% Triton X-100 (Anatrace) in PBS (pH 7.4)) for one hour. The soluble fraction was collected after centrifugation at 185,000 × g for one hour and incubated with StrepTactin Sepharose High Performance resin (GE Healthcare, Marlborough, MA) for 30 min using a batch procedure. The resin was transferred into a gravity column (Bio-rad, Hercules, CA) and washed with 10 column volumes of W buffer (100 mM Tris-HCl; pH 8.0, 150 mM NaCl, 1 mM EDTA, and 0.5 mM DDM), and pdP2X7cryst-EGFP was eluted with E buffer (100 mM Tris-HCl; pH 8.0, 150 mM NaCl, 1 mM EDTA, 2.5 mM desthiobiotin, 15% glycerol, and 0.5 mM DDM). The N-terminal EGFP and the strep tag was removed by incubating the P2X7cryst-EGFP with human thrombin (HTI; 1/30 w/w) overnight. P2X7~cryst~ was isolated by size exclusion chromatography using Superdex 200 (GE Healthcare) in C buffer (50 mM Tris-HCl; pH 7.4, 150 mM NaCl, 15% Glycerol, and 0.5 mM DDM). The peak fractions were pooled, concentrated to 10 mg/mL, and used for crystallization after ultracentrifugation at 265,000 × g for 20 min. All purification steps were carried out at 4°C or on ice.
Crystallization {#s4-3}
---------------
P2X7~cryst~ was crystalized using the hanging drop vapor diffusion method by mixing 1:1 (v/v) ratio of protein and reservoir solutions at 4°C. The apo crystals appeared in two weeks and were fully grown in 3--4 months using a reservoir solution containing 100 mM HEPES (pH 7.0), 100 mM NaCl, 4% ethylene glycol, 15% glycerol, 29% PEG-400, 0.1 mg/mL lipid mixture (60% POPE, 20% POPG, and 20% cholesterol). The P2X7~cryst~ crystals with antagonists reached their maximum sizes in about two months using reservoir solutions containing 100 mM HEPES or Tris (pH 6.0--7.5), 100 mM NaCl, 4% ethylene glycol, 15% glycerol, 27--32% PEG-400 or 31--36% PEG-300, 0.1 mg/mL lipid mixture (60% POPE, 20% POPG, and 20% cholesterol), and 1 mM P2X7 antagonists (AZ10606120, JNJ-47965567, A804598, GW791343 (Tocris Biosciences, UK), and A740003 (Sigma Aldrich, St. Louis, MO)). For heavy atom derivatization, A740003-bound P2X7~cryst~ crystals were soaked into a reservoir solution containing 1 mM K2IrCl6 and 0.5 mM DDM overnight at 4°C. For determination of the ATP/A804598-bound P2X7 structure, crystals of P2X7~cryst~ were grown in 33% PEG-400, 100 mM MES (pH 6.5), 100 mM NaCl, 5% Glycerol, and 1 mM A804598 for 1 month at 4°C. Cubic-shaped crystals were soaked with 1 mM ATP overnight. Crystals were flash-frozen in liquid nitrogen for data collection.
Structure determination {#s4-4}
-----------------------
X-ray diffraction data were collected using synchrotron radiation at the Cornell High Energy Synchrotron Source (beamline F1) and Advanced Photon Source at Argonne National Laboratory (beamlines 24ID-C and E). The following X-ray wave length was used for each data set: Apo: 1.1051 Å (24ID-C); A740003: 0.9792 Å (24ID-E); A804598: 0.9774 Å (F1); AZ10606120: 0.9782 Å (F1); GW791343: 0.9782 Å (F1); JNJ47965567: 0.9774 Å (F1); A740003/K~2~IrCl~6~: 1.1051 Å (24ID-C); ATP/A804598: 0.9791 Å (24ID-C). Diffraction data were indexed, integrated, and scaled using XDS ([@bib25]), and merged using AIMLESS ([@bib14]) in the CCP4 suite ([@bib45]). Low resolution electron density map (\~7 Å) of the P4~2~ crystal was obtained by molecular replacement (RFZ = 7.9, TFZ = 9.3, and LLG = 257) using the zebrafish P2X4 model (PDB code: 4DW0) with the program PHENIX ([@bib2]). Initial phase information of the A740003-bound pdP2X7~cryst~ (I2~1~3 space group) was obtained by MR-SAD with PHENIX using the datasets from Ir-derivatized crystals and the zebrafish P2X4 structure (PDB code: 4DW0) as a searching model. Model building was carried out manually using COOT ([@bib13]) and the structure was refined using PHENIX. The final model was used as a searching template in molecular replacement for solving the structures of the apo state as well as the antagonist bound forms of pdP2X7~cryst~ with AZ10606120, JNJ47965567, A804598, GW791343. For ATP/A804598-bound form, A804598-bound structure was used as a searching model. Model quality was assessed using PHENIX where stereochemistry and R-values are satisfactory ([Table 2](#tbl2){ref-type="table"}). Molecular models presented in the figures and videos are created using the PyMOL Molecular Graphics System, Version 1.8 Schrödinger, LLC ([@bib1]). The atomic coordinate files have been deposited in the Protein Data Bank under the accession codes 5U1L (apo), 5U1U (A740003-bound), 5U1V (A804598-bound), 5U1W (AZ10606120-bound), 5U1X (JNJ47965567-bound), 5U1Y (GW791343-bound), and 5U2H (ATP/A804598-bound).10.7554/eLife.22153.023Table 2.Data collection and refinement statistics.**DOI:** [http://dx.doi.org/10.7554/eLife.22153.023](10.7554/eLife.22153.023)pdP2X7~cryst~pdP2X7~cryst~-JNJ47965567pdP2X7~cryst~-A740003pdP2X7~cryst~-A804598pdP2X7~cryst~-AZ10606120pdP2X7~cryst~-GW791343pdP2X7~cryst~-ATP/A804598**Data collection**Space group*I2~1~3I2~1~3I2~1~3I2~1~3I2~1~3I2~1~3P2~1~3*Cell dimensions*a*, *b*, *c* (Å)169.1, 169.1, 169.1169.3, 169.3, 169.3169.6, 169.6, 169.6170.4, 170.4, 170.4170.7, 170.7, 170.7169.7, 169.7, 169.7167.6, 167.6, 167.6*α, β, γ* (°)90, 90, 9090, 90, 9090, 90, 9090, 90, 9090, 90, 9090, 90, 9090, 90, 90Resolution (Å)45.2--3.40 (3.52--3.40)\*45.3--3.20 (3.31--3.20)\*42.4--3.61 (3.73--3.61)\*42.6--3.40 (3.52--3.40)\*42.7--3.50 (3.63--3.50)\*45.4--3.30 (3.42--3.30)\*46.5--3.90\
(4.04--3.90)\**R*~merge~0.099 (1.80)0.12 (0.95)0.23 (1.71)0.15 (1.49)0.17 (2.21)0.12 (1.72)0.12 (2.17)*I*/σ17.9 (1.30)13.4 (2.90)13.2 (1.65)11.0 (1.71)11.01 (1.16)12.55 (1.34)15.7 (1.25)Completeness (%)99.8 (98.8)100 (100)100 (100)100 (100)100 (100)99.8 (99.4)99.8 (99.2)Redundancy9.8 (9.9)10.1 (10.4)11.0 (11.0)10.1 (10.0)10.1 (9.9)10.1 (10.2)10.0 (10.3)**Refinement**Resolution (Å)45.2--3.4045.3--3.2042.4--3.6142.6--3.4042.7--3.5045.4--3.3046.5--3.90No. reflections11,19713,4829,51011,47410,59812,38314,510*R*~work/~*R*~free~ (%)24.2/26.322.3/26.723.2/26.025.0/27.324.5/26.824.0/27.133.4/38.6No. atoms2304247323842439245423824116 Protein2276238223072359236723133967 Ligand/ion289177808769149B-factors124.4103.697.897.5114.9119.0149.3 Protein123.8102.997.196.8114.5118.5149.1 Ligand/ion171.1124.1120.6118.3128.0137.4153.2R.M.S deviations Bond lengths (Å)0.0020.0040.0030.0030.0040.0040.001 Bond angles (°)0.490.690.620.640.680680.45[^1]
YO-PRO-1 uptake assay {#s4-5}
---------------------
Human embryonic kidney (HEK293) cells were maintained in DMEM medium (Thermo Fisher Scientific) supplemented with 10% FBS (Atlanta Biologicals, Flowery Branch, GA), and 10 μg/mL gentamicin (Thermo Fisher Scientific). HEK293 cells were split into six well plates at a cell density of 2.0 × 10^6^ cells/well and incubated overnight. Three hours after transfection with 2 μg of the full length pdP2X7 in pIM2 vector (IRES-mCherry; modified from pIE2 vector) using jetPRIME reagent (Polyplus-transfection, France), cells were trypsinized, transferred into poly-D-Lysine coated black-walled 96 well plates (Corning, Corning, NY) at 7.5 × 10^4^ cells/well, and incubated for 24 hr. Cells were washed with assay buffer (2 mM KCl, 0.1 mM CaCl~2~, 13 mM Glucose, 147 mM NaCl, 10 mM HEPES (pH 7.3)) and were incubated with 5 μM YO-PRO-1 Iodide (Thermo Fisher Scientific), in the presence or absence of antagonists at multiple concentrations at 37°C for 10 min. Upon application of 1 mM ATP (final), uptake of YO-PRO-1 was recorded with 1 min intervals by following the fluorescence change using a Synergy HT multi-detection microplate reader (Bio-Tek, Winooski, VT; Ex: 485 nm/20, Em:528 nm/20, sensitivity 60). To obtain the dose dependent uptake inhibition by antagonists, the initial rates of YO-PRO-1 uptake were plotted against multiple concentrations of the P2X7 specific drugs. The inhibition curves from five independent measurements were fitted to the Hill equation using Origin software (Originlab, Northampton, MA) to determine the IC~50~ values.
Cell line generation {#s4-6}
--------------------
HEK293 (CRL-1573) cell lines were purchased from the American Type Culture Collection (ATCC, Manassas, VA), and therefore were not further authenticated. The mycoplasma contamination test was confirmed to be negative at ATCC.
Data analysis {#s4-7}
-------------
Dose response curves of the YO-PRO-1 uptake experiments were fitted with either competitive or non-competitive inhibition models using Origin 6.0 software (OriginLab) as previously described ([@bib29]). For the competitive antagonism model, we used the equation:$$Response = \frac{\left\lbrack A \right\rbrack/KA}{\left\lbrack A \right\rbrack KA\left( {1 + \tau} \right) + \left\lbrack B \right\rbrack/KB + 1}$$
where $\left\lbrack A \right\rbrack$, $\left\lbrack B \right\rbrack$ are the concentrations of BzATP and antagonists, respectively; $K_{A}$ and $K_{B}$ are the equilibrium dissociation-constant of BzATP and antagonists, respectively. Dose response curves without antagonist were fitted with this equation, which gives the values K~A~ = 28 μM, and τ = 0.031. K~B~ was then determined using the dose response curves in the presence of antagonists. The resulting K~B~ value for each antagonist was; JNJ: 1.7 nM; A80: 15 nM; A74: 24 nM; AZ10: 56 nM; GW: 3.0 μM. For the non-competitive inhibition model, we used the equation:$$\mathit{R}\mathit{e}\mathit{s}\mathit{p}\mathit{o}\mathit{n}\mathit{s}\mathit{e} = (\lbrack\mathit{A}\rbrack^{\mathit{n}}\tau^{\mathit{n}}\mathit{E}\mathit{m}\mathit{a}\mathit{x})/(\lbrack\mathit{A}\rbrack^{\mathit{n}}\tau^{\mathit{n}} + (\lbrack\mathit{A}\rbrack(1 + \alpha\lbrack\mathit{B}\rbrack/\mathit{K}_{\mathit{B}}) + \mathit{K}_{\mathit{A}}\lbrack\mathit{B}\rbrack/\mathit{K}_{\mathit{B}} + \mathit{K}_{\mathit{A}})^{\mathit{n}}$$
where n is the Hill coefficient, and Emax is the maximum initial rate. K~A~, τ, n values were determined using the data without antagonists (K~A~ = 220 μM, τ = 10, and n = 2.6) and the rest of the parameters were determined by curve fitting of data with different concentrations of antagonists (α = 1 and K~B~ value for each antagonist was: JNJ: 0.14 μM; A80: 0.57 μM; A74: 1.2 μM; AZ10: 3.4 μM; GW: 266 μM). Schild regression analysis was performed as described previously ([@bib46]). Briefly, concentrations of BzATP at half maximal initial rates of YO-PRO-1 uptake in the presence (EC~50~') or absence (EC~50~) of antagonists were determined by fitting the dose responses with the Hill equation. Dose ratio *r=*EC~50~'/EC~50~ was calculated for various antagonist concentrations and *r-1* values were plotted against the antagonist concentrations in log scale to obtain Schild plots.
Ligand-binding experiment {#s4-8}
-------------------------
GFP fused pdP2X7~cryst~ (P2X7 GFP) was purified in a buffer containing 150 mM NaCl, 50 mM Tris-HCl (pH 7.4), 15% glycerol, and 0.5 mM DDM as described in \"Expression and purification.\" GFP-tagged pdP2X7~cryst~, which is substantially more stable than pdP2X7~cryst~, was used in this experiment as it does not interfere with the fluorescence properties of Alexa-ATP ([Figure 3---figure supplement 5B](#fig3s5){ref-type="fig"}). P2X7-GFP (100 μM) was pre-incubated with each P2X7 specific antagonist (100 μM) for 30 min at room temperature. P2X7 GFP was then incubated with 10 μM ATP-Alexa 647 (Thermo Fisher Scientific) at 30°C for 10 min, which was required to obtain a stable background prior to the fluorescence measurement. Fluorescence anisotropy was measured at 30°C using FluoroMax four fluorimeter (Horiba,Edison, NJ) with excitation and emission wavelengths of 590 nm and 670 nm, respectively. For binding competition experiments, various concentrations of ATP ranging from 10 μM to 10 mM (pH was adjusted to 7.0 with NaOH) were added from 100X solutions. Fluorescence anisotropy $\left\langle r \right\rangle$ was defined as:$$\left\langle r \right\rangle = \frac{IVV - G\, \ast \, IVH}{IVV + 2\, \ast \, G\, \ast \, IVH}$$
where $I_{VV}$ and $I_{VH}$ are the fluorescence intensities with the excitation polarizer mounted vertically and the emission polarizer mounted vertically or horizontally, respectively. *G* is defined as:$$G = \frac{IHV}{IHH}$$
where *$I_{HV}$* and $I_{VV}$ are the fluorescence intensities with the excitation polarizer mounted horizontally and the emission polarizer mounted vertically or horizontally, respectively.
Electrophysiology {#s4-9}
-----------------
HEK293 cells were split onto glass coverslips in six well plates at 1 × 10^5^ cells/well and incubated at 37°C overnight. Cells were transfected with 1 μg of the full length pdP2X7 (wildtype or mutants) or the full length mP2X4 (wildtype or mutants) in pIE2 vector using FuGENE6 (Promega, Madison, WI). Cells were used 18--32 hr after transfection for measuring the P2X receptor activities using the whole cell patch clamp configuration. Membrane voltage was clamped at −60 mV with an Axopatch 200B amplifier (Molecular Devices, Sunnyvale, CA), currents were filtered at 2 kHz (eight-pole Bessel; model 900BT; Frequency Devices, Ottawa, IL) and sampled at 10 kHz using a Digidata 1440A and pCLAMP 10.5 software (Molecular Devices). The extracellular solution contained 147 mM NaCl, 10 mM HEPES, 13 mM Glucose, 2 mM KCl, 0.1 mM CaCl~2~, (pH 7.3). The pipette solution contained 147 mM NaCl, 10 mM HEPES, 10 mM EGTA, which was adjusted to pH 7.0 using NaOH. Whole cell configuration was made in an extracellular solution supplemented with 2 mM CaCl~2~ and 1 mM MgCl~2~ and the extracellular solutions were rapidly exchanged to the solutions containing desired concentrations of ATP using a computer-controlled perfusion system (RSC-200; Bio-Logic, France). Because pdP2X7 substantially runs up ([Figure 1B and E](#fig1){ref-type="fig"}), we measured the channel activity after treating the cells with 1 mM ATP for at least 20 s. For testing the effects of P2X7 specific antagonists on pdP2X7, these drugs were incubated with ATP (1 mM) for 1 min. Concentrations of the drugs were: A740003: 600 nM; A804598: 180 nM; AZ10606120: 2.3 μM; GW791343: 50 μM; JNJ47965567: 136 nM. For the cysteine accessibility studies on pdP2X7, 0.1 mM MTS-TPAE (Toronto Research Chemicals, Canada) was perfused for 10 s either in the absence or presence of 1 mM ATP. For probing mP2X4 accessibility in the closed state, 0.1 mM MTS-TPAE was applied for 10 s and application of 10 μM ATP for 1 s was used to measure channel activity. For mP2X4 accessibility in the open state, 5 μM ATP was applied for 9 s and 0.1 mM MTS-TPAE was applied for 3 s. For measuring cysteine accessibility of pdP2X7/Y295C or mP2X4/F296C mutants, cells were treated with 10 mM dithiothreitol (DTT) for 5 min prior to recording. To normalize the channel activities from multiple experiments, the ratio between channel activity before and after MTS-TPAE application was calculated for each cell. Since mP2X4 rapidly runs down, these ratios were further normalized to the control channel activities measured on the same cell in the absence of MTS-TPAE. Stock solutions of 100 mM MTS-TPAE were prepared freshly every 5 hr in water and stored on ice and were diluted to the desired concentrations in the extracellular solution immediately before each recording. All recordings were performed using more than five cells.
Funding Information
===================
This paper was supported by the following grants:
- http://dx.doi.org/10.13039/100000002National Institutes of Health NS072869 to Toshimitsu Kawate.
- http://dx.doi.org/10.13039/100000002National Institutes of Health GM114379 to Toshimitsu Kawate.
We thank the personnel at beamlines 24-ID of the Advanced Photon Source, X-25 of National Synchrotron Light Source, and F1 of the Cornell High Energy Synchrotron Source. We also thank H Lin, G Hollopeter, R Cerione, C Sevier, and K Michalski for discussion.
Additional information {#s5}
======================
The authors declare that no competing interests exist.
AK, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article, Contributed unpublished essential data or reagents.
TK, Conception and design, Analysis and interpretation of data, Drafting or revising the article, Contributed unpublished essential data or reagents.
10.7554/eLife.22153.024
Decision letter
Swartz
Kenton J
Reviewing editor
National Institutes of Health
,
United States
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting your article \"Structural basis for subtype-specific inhibition of the P2X7 receptor\" for consideration by *eLife*. Your article has been favorably evaluated by Gary Westbrook as the Senior Editor and three reviewers, one of whom, Kenton J Swartz (Reviewer \#1), is a member of our Board of Reviewing Editors. The following individuals involved in review of your submission have agreed to reveal their identity: Hiro Furukawa (Reviewer \#2) and Mark L Mayer (Reviewer \#3).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
Summary:
This is an exciting manuscript describing the first X-ray structure of the P2X7 receptor, a subtype of ATP-activated P2X receptors implicated in a range of interesting biological processes, including pain and apoptosis. Structures were solved in the presence of 5 distinct inhibitors, and for one of them in the combined presence of ATP. These structures show that the drugs bind in a completely unanticipated region of the receptor, in a cleft originating from the external surface of the receptor that resides behind the ATP binding pocket. In the one case when ATP was bound to the receptor along with antagonist, closure of the lateral ATP binding domain occurs similar to what is seen in the original structure of the ATP-bound zfP2X4 receptor, as well as the very recently published ATP-bound P2X3 receptor, but opening the pore is prevented by the presence of the drugs in the cleft, suggesting that it narrows as the channel opens. Interestingly, the drug-binding cleft is essentially absent from P2X3 and 4 receptors, explaining why the drugs studied here are selective for P2X7. The new structures presented are accompanied by extensive functional experiments, including 1) mutagenesis of the drug-binding cleft to confirm the unexpected binding site, 2) Schild analysis demonstrating that the drugs do not work through a competitive mechanism as suggested for 3 of the compounds by earlier work, and 3) accessibility studies with bulky MTS reagents that are consistent with the cleft being wide in the closed state and narrowing upon ATP binding. Overall this is a beautiful study that arrives at interesting and important new conclusions concerning the gating mechanism and mechanism of drug action for the P2X7 receptor. The work could be published anywhere and *eLife* is fortunate to have this one. Revision should be straightforward.
Essential revisions:
The following two concerns could be addressed by careful revision of the manuscript without performing additional experiments, but it is important that they take these issues seriously.
1\) Although the location of the bound drug molecules is convincingly demonstrated by Fo-Fc maps, it is far from clear that the current resolution of any structure reported in the paper (3.2 -- 3.9 Å) is high enough to support the detailed models in presented in [Figure 3A and B](#fig3){ref-type="fig"} in which side chain interactions with bound ligands are reported. Indeed, the 2Fo-Fc map shown in [Figure 2A](#fig2){ref-type="fig"} appears to show tubes of electron density for α-helices, without any side chains. How robust are the chosen side chain rotamers during refinement; is the helical register correct; how do these issues impact the models shown in [Figure 3](#fig3){ref-type="fig"}? Also, how dependent on side chain conformations are the HOLE plots and graph presented in [Figure 4](#fig4){ref-type="fig"}?
2\) The conclusion that the antagonists \"allosterically prevent narrowing of the drug-binding pocket and the turret-like architecture during channel opening\" is a reasonable hypothesis derived from a combination of functional experiments with MTS reagents, and a comparison with previously reported P2X3 and P2X4 crystal structures, but there is no direct proof of this. In our opinion the authors overstate what their data actually shows. What is lacking is an agonist (ATP) bound P2X7 structure in which the turret has adopted an active compact conformation. They should carefully revise the manuscript to address this, and perhaps explain why this was not done, and an indirect approach of functional assays with Cys mutants and MTS-TPAE adopted.
10.7554/eLife.22153.025
Author response
*Essential revisions:*
*The following two concerns could be addressed by careful revision of the manuscript without performing additional experiments, but it is important that they take these issues seriously.*
1\) Although the location of the bound drug molecules is convincingly demonstrated by Fo-Fc maps, it is far from clear that the current resolution of any structure reported in the paper (3.2 -- 3.9 Å) is high enough to support the detailed models in presented in [Figure 3A and B](#fig3){ref-type="fig"} in which side chain interactions with bound ligands are reported. Indeed, the 2Fo-Fc map shown in [Figure 2A](#fig2){ref-type="fig"} appears to show tubes of electron density for α-helices, without any side chains. How robust are the chosen side chain rotamers during refinement; is the helical register correct; how do these issues impact the models shown in [Figure 3](#fig3){ref-type="fig"}? Also, how dependent on side chain conformations are the HOLE plots and graph presented in [Figure 4](#fig4){ref-type="fig"}?
We agree with the reviewers that the resolution of the reported crystal structure is not high enough to determine the accurate distances/angles between the drug-binding residues and the drugs. We therefore avoided describing potential specific-interactions, such as hydrogen bonding. We also performed site-directed mutagenesis for the drug-surrounding residues, where we validated the involvement of these residues in drug-binding. Nevertheless, locations and orientations of the side chains around the drug-binding pocket are well-supported by the reasonable electron density in these regions. To clarify these points, we added 1) a sentence in the main text describing the limitation of our current structures, 2) a new video ([Video 1](#media1){ref-type="other"}) showing the electron density of the drug-binding residues around one of the antagonists (JNJ47965567), and 3) a new figure ([Figure 3---figure supplement 1](#fig3s1){ref-type="fig"}) that shows the side chain electron density for the residues comprising the turret (β13 and β14) and the α-helix whose helical-register was questioned. We believe that these extra video and figure support the models presented in [Figure 3](#fig3){ref-type="fig"} as well as the HOLE plots and graph presented in [Figure 4](#fig4){ref-type="fig"}.
The 2Fo-Fc map shown in [Figure 2---figure supplement 2A](#fig2s2){ref-type="fig"} includes only one short helix (D89-T90-A91-D92-Y93) whose side chain densities are relatively well-defined. Unfortunately, the angle of this figure prevents readers from appreciating how good the side chain densities are, as the point of this figure was to show no obvious density in the drug-binding pocket. The newly added figure ([Figure 3---figure supplement 1](#fig3s1){ref-type="fig"}) should solve this issue. In addition, combining the fact that the helical register in this region is consistent with those of P2X3 and P2X4, miss-assignment of residues in this region is highly unlikely.
We selected the side chain rotamers that fit best in the electron density, which is a standard way of modeling a low resolution structure. While there is a chance that the side chain rotamer is incorrect, a wrongly assigned rotamer should not have major impact on the models shown in [Figure 3](#fig3){ref-type="fig"}, as the approximate orientation of each side chain should remain the same. Likewise, the HOLE plots would be only minutely different, if different side chain conformations are used. Indeed, positions of the peptide backbone account for the major differences in the HOLE plots.
*2) The conclusion that the antagonists \"allosterically prevent narrowing of the drug-binding pocket and the turret-like architecture during channel opening\" is a reasonable hypothesis derived from a combination of functional experiments with MTS reagents, and a comparison with previously reported P2X3 and P2X4 crystal structures, but there is no direct proof of this. In our opinion the authors overstate what their data actually shows. What is lacking is an agonist (ATP) bound P2X7 structure in which the turret has adopted an active compact conformation. They should carefully revise the manuscript to address this, and perhaps explain why this was not done, and an indirect approach of functional assays with Cys mutants and MTS-TPAE adopted.*
We agree with the reviewers and rephrased the stronger statement about the conformational changes. We also added a description about why we are short in obtaining an ATP-bound P2X7 structure in the Discussion of the revised manuscript.
[^1]: \*Highest resolution shell is shown in parenthesis.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Melanoma is a malignancy that arises from uncontrolled proliferation and metastasis of neoplastic melanocytes. Normally, melanocytes are located in the most basal epidermal layer, attached to a collagen-IV rich basement membrane, which separates epidermal and dermal compartments of the skin \[[@CR1], [@CR2]\]. Malignant melanoma is the most lethal form of skin cancer \[[@CR3]\]. Its incidence continues to increase each year and is currently responsible for more than 80% of deaths from skin cancer \[[@CR4]\]. The majority of melanoma mutations are C \> T transitions caused by ultraviolet light irradiation \[[@CR5]\], which mostly affect the mitogen-activated protein kinase pathway \[[@CR6]--[@CR9]\]. When diagnosed in its early 'non-tumorigenic' stages, resection of the lesion results in very high survival rates \[[@CR10]\]. In this period, which is also termed as radial growth phase \[[@CR11]\], pigmented patches of skin (nevi) increase laterally in size and become palpable, but melanoma cells typically still reside within the epidermis and are not metastasis competent \[[@CR12]\]. Nonetheless, already at this point, they affect cellular behavior in their local environment. For example, neoplastic melanocytes induce hyperproliferation and impair differentiation of keratinocytes \[[@CR13]\]. Once further mutations have mediated metastasis competence, the lesion becomes 'tumorigenic' and enters the vertical growth phase, during which the morphology of nevi often switches from plaque to balloon-like \[[@CR12]\]. Melanoma cells lead to breakdown of the basement membrane, massively invade the dermal and hypodermal compartments and metastasize to distant organs. Surgery is then no longer sufficient and the disease becomes much more challenging to treat \[[@CR3], [@CR14], [@CR15]\]. Treatment options for late stage melanomas include kinase inhibitors and immunotherapies like the BRAF inhibitor vemurafenib and the anti-cytotoxic T-lymphocyte antigen-4 antibody ipilimumab \[[@CR16]--[@CR18]\]. However, monotherapy is unlikely to yield a long-term benefit due to multi-drug resistance and, therefore, combination therapies with different targeted and immunotherapies as well as standard chemotherapeutics are being evaluated \[[@CR19]--[@CR22]\]. In melanoma cells, ATP-binding cassette (ABC) transporters, in particular of type ABCB5, were found to mediate resistance to the chemotherapeutics doxorubicin and temozolomide \[[@CR23], [@CR24]\]. Although ABCB5 is present in several human tissues, it is highly abundant in melanocyte progenitors, melanoma cell lines, and melanoma biopsies \[[@CR23], [@CR25]--[@CR28]\]. Furthermore, its expression correlates with tumor progression and metastasis competence \[[@CR29]\].
For the reasons of simplicity, convenience, and cost, in vitro studies on melanoma are often performed in 2D-cell culture assays. However, gene expression is significantly different between 2D and 3D melanoma cultures, likely affecting the signaling exerted by and the sensitivity to drugs of melanoma cells \[[@CR30]\]. Furthermore, the interactions between different cell types of tumor and stroma are difficult to model in 2D. Thus, to better mimic the in vivo situation, different 3D-cell culture approaches with several degrees of complexity have been developed, including spheroids, tumorospheres, human skin equivalents, and melanoma-on-chips assays \[[@CR31]\] as well as xenografts of human melanoma spheroids in rodent recipients \[[@CR32]\]. Such formats are very useful for basic and applied melanoma research, but the currently existing models are either composed of only melanoma cells or they are so complex that the behavior of individual cell types is difficult to understand, and often they are then hard to establish and expensive. In the present work, we describe a novel, simple spheroid-based melanoma model composed of fibroblasts, keratinocytes, and melanoma cells. It allows to track cellular behavior in a cell-type specific manner and recapitulates different characteristics of early melanoma stages. The different cell types arranged into a collagen-IV rich fibroblast core, a ring of keratinocytes, and groups of highly proliferating melanoma cells on the outside. Some melanoma cells were also regularly found to invade the fibroblast core. While in the absence of melanoma cells the keratinocyte ring stratified into central basal-like and peripheral, more differentiated cells, addition of melanoma cells clearly reduced keratinocyte differentiation. Treatment with the cytostatic drug, docetaxel, which has been primarily tested for combination therapy of melanoma \[[@CR33]--[@CR35]\], restored keratinocyte differentiation and ablated external melanoma cells. The few remaining external melanoma cells, however, showed a significantly increased amount of ABCB5-immunoreactivity.
Methods {#Sec2}
=======
Cell culture {#Sec3}
------------
The human fibroblast cell line CCD-1137Sk (ATCC® CRL-2703™) was cultured in Iscove's Modified Dulbecco's Medium (IMDM), with L-Glutamine, supplemented with 10% fetal bovine serum (Sigma), and 1% Penicillin Streptomycin (Capricorn). The human keratinocyte cell line HaCaT (CLS order no. 300493) and the human malignant melanoma cell line SK-MEL-28 (CLS order no. 300337) were cultured in Dulbecco's Modified Eagle Medium (DMEM) High Glucose (4.5 g/l), with L-Glutamine, with Sodium Pyruvate (Capricorn) supplemented with 10% fetal bovine serum, and 1% Penicillin Streptomycin. Cells were maintained at 37 °C in 5% CO~2~. Cell lines were obtained in 2016 and repeatedly authenticated by phenotypic analysis, including expression of collagen IV for CCD-1137Sk, establishment of a CK10/CK14 gradient in 3D for HaCaT, and high proliferation rate for SK-MEL-28. Mycoplasma tests using the MycoAlert™ Mycoplasma Detection Kit (Lonza) were routinely performed to ensure mycoplasma-free cell cultures.
3D spheroid cultures and docetaxel treatment {#Sec4}
--------------------------------------------
Spheroids were prepared using 96- and 384-well cell repellent plates (Greiner). For mono-culture spheroids, fibroblasts (10,000 cells/well) and HaCaT cells (20,000 cells/well) were seeded. For skin bi-cultures, 10,000 cells of each, fibroblasts and keratinocytes, were used per well, and HaCaT cells were added three days after formation of the fibroblast core. Mono- and bi-cultures were cultured for seven days. For tri-culture spheroids, fibroblasts (10,000 cells/well) were seeded. After three days, HaCaT (10,000 cells/well) and SK-MEL-28 cells (2500 cells/well) were added simultaneously. To distinguish between the different cell lines, CellTracker Fluorescent Probes (Life Technologies) were used. Before adding cells to the 3D co-culture, HaCaT cells were labeled with CellTracker Red CMPTX dye (Life Technologies, C34552) and SK-MEL-28 cells were labeled with CellTracker Green CMFDA (Life Technologies, C2925), each for a time period of 45 min according to the CellTracker manuals. Another two days later, tri-culture spheroids were treated with 100 nM docetaxel or 0.01 ‰ of DMSO as control for 15, 24, 48, and 72 h, respectively. Stock solutions (10 mM) of docetaxel (Sigma) were prepared in dimethylsulfoxide (DMSO). After treatments, spheroids were normally fixed and immunostained as described below. For some experiments, spheroids were transferred to 3D agarose molds (Sigma, Z764051) on day five after seeding in cell repellent plates. Treatment with DMSO or 100 nM docetaxel for 72 h, as well as fixation and cryosectioning were then carried out in the molds.
Immunofluorescence {#Sec5}
------------------
Immunostaining of spheroids used the following steps. Spheroids were collected in an Eppendorf tube, washed once with PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na~2~HPO~4~ × 2 H~2~O, 2 mM KH~2~PO~4~, pH 7.4), and fixed with 4% wt/vol paraformaldehyde in PBS at room temperature for 30 min. Then, spheroids were incubated overnight at 4 °C in 15% sucrose in PBS, followed by an incubation overnight at 4 °C in 25% sucrose (Roth, 4621.1) in PBS, before they were embedded in OCT (Leica Biosystems). A CM-1950 cryostat (Leica Biosystems) was used for preparing 10-μm-thick sections. 3D molds were washed once with PBS and fixed with 4% wt/vol paraformaldehyde in PBS at room temperature for 30 min. Next, molds were embedded in OCT and cut with a cryostat into 20-μm-thick sections. All sections were permeabilized with 0.1% Triton X-100 (Roth, 3051.4) in PBS, blocked with 3% BSA (Roth, 8076.3) in PBS, and stained with rabbit anti-Ki67 (Merck, AB9260), rabbit anti-cleaved caspase 3 (CAS3)(Cell Signaling, 9661), rabbit anti-cytokeratin 10 (CK10)(Thermo Fisher Scientific, PA5--32459), rabbit anti-collagen IV (Rockland, 600--401-106S), mouse anti-cytokeratin 14 (CK14)(Merck, MAB3232), or mouse anti-ABCB5 (3C2-1D12 \[[@CR29]\]; and Thermo Fisher Scientific, MA5--17026) antibodies, followed by goat anti-rabbit Alexa Fluor 647 (Invitrogen, A21246), goat anti-mouse Alexa Fluor 555 (Invitrogen, A21424), or donkey anti-mouse Alexa Fluor 647 (Invitrogen, A31571) secondary antibody labeling. Nuclei were stained with Dapi (Sigma, 10,236,276,001). Finally, sections were washed with PBS and mounted with Mowiol (Roth, 0713.2) for confocal microscopy (SP8, Leica).
Statistical analysis {#Sec6}
--------------------
Images were composed using Adobe Illustrator (Adobe Systems Software) and ImageJ. All numeric data were handled using Microsoft Excel 2013 and were subsequently incorporated into the Adobe Illustrator composite. Quantitative analysis of Ki67-, CAS3-, and CK10-positive cells was performed using ImageJ. Graphs are presented as mean ± SEM and statistically analyzed using one-way ANOVA with post-hoc Tukey HSD Calculator or Student t-test. *P*-values are indicated as \* \< 0.05, \*\* \< 0.01.
Results {#Sec7}
=======
Characterization of spheroid keratinocyte and fibroblast mono- and bi-cultures {#Sec8}
------------------------------------------------------------------------------
To set up a simple, robust and multiplexable melanoma test system, we first tested the growth and differentiation behavior of major components of the stroma-like environment of melanoma, i.e. human keratinocytes and fibroblast cells, in 3D spheroids and also analyzed potential effects of co-culturing. Thus, both cell types were either cultured as mono- or co-cultures using a cell-repellent culturing system in 96-well format with a culture time of seven days. Then, spheroids of all types were cryosectioned into 10-μm thick slices and immunostained for the proliferation marker Ki67 or the apoptosis marker CAS3. Cytokeratins CK10 and CK14 were immunostained to detect more differentiated and basal keratinocytes, respectively. Nuclei were labeled with Dapi. Figure [1a-c](#Fig1){ref-type="fig"} show representative confocal sections of these samples as indicated. While proliferation was limited to few cells in the periphery of spheroids (Fig. [1a](#Fig1){ref-type="fig"}), apoptotic cells were found throughout the whole spheroid diameter (Fig. [1b](#Fig1){ref-type="fig"}). In both, mono-cultures and bi-cultures, HaCaT keratinocytes showed a clear stratification with basal-like CK14-positive and more differentiated CK10-positive cells in the center and on the periphery of the spheroids (Fig. [1c](#Fig1){ref-type="fig"}), respectively. In bi-cultures, fibroblasts formed a central core, while keratinocytes were located in a ring-like fashion around this fibroblast core. Quantitative analysis showed that co-culturing significantly reduced the number of proliferating and increased the amount of CK10-positive peripheral keratinocytes (Fig. [1d](#Fig1){ref-type="fig"}). These results suggest that the 3D spheroid skin model reflects some stratification and differentiation features of skin even without the use of a pH or Ca^2+^ gradient or an air-liquid interface to induce differentiation.Fig. 1Proliferation, apoptosis, and differentiation of fibroblasts and HaCaT cells in mono- and bi-culture spheroids. Spheroids were cultured as mono- and bi-cultures for seven days, cryosectioned into 10-μm thick slices, and then stained for markers for cell proliferation (**a**, Ki67, green), apoptosis (**b**, CAS3, green), differentiated (**c**, CK10, green) and basal keratinocytes (**c**, CK14, red). In **a** and **b**, nuclei were stained with Dapi (blue). **a-c** Representative confocal images. Scale bars: 100 μm. **d** Quantification of Ki67- and CAS3-positive cells (percentage of total) and CK10-positive cells (percentage of peripheral nuclei). Given is mean ± SEM (n ≥ 3 independent experiments; \*\* *P* \< 0.01)
Melanoma cells invade the fibroblast core and decrease keratinocyte differentiation in tri-cultures {#Sec9}
---------------------------------------------------------------------------------------------------
After these initial characterizations, the spheroid skin model was complemented by the addition of SK-MEL-28 melanoma cells. Therefore, fibroblasts were seeded and cultivated in 3D. Three days later, HaCaT keratinocytes and SK-MEL-28 melanoma cells were added simultaneously. To distinguish between the different cell types, HaCaT and SK-MEL-28 cells were labeled with CellTrackers Red CMPTX and Green CMFDA, respectively. After another four days, tri-culture spheroids were harvested, cryosectioned into 10-μm thick slices and stained for Ki67, CAS3, CK10 and CK14, or the basement membrane marker collagen IV. As shown in Fig. [2a](#Fig2){ref-type="fig"}, fibroblasts remained in the central core of these tri-cultures, followed by a few layers of keratinocytes. While most melanoma cells were grouped in clusters of several dozens of cells on the shell of the cultures, a few melanoma cells were very regularly also found in the fibroblast core, but almost never in the keratinocyte layers. In the following, for simplicity, SK-MEL-28 cells in the outer rim of the tri-cultures will be termed 'external', those in the fibroblast core as 'internal' melanoma cells. A schematic representation of the tri-culture composition is depicted in Additional file [1](#MOESM1){ref-type="media"}: Figure S1A. The qualitative analysis further showed that numerous external melanoma cells were proliferating, while internal melanoma cells, keratinocytes, and fibroblasts were rarely doing so. Interestingly, CK10 expression as an indicator of keratinocyte differentiation was strongly reduced at the contact sites with melanoma cells.Fig. 2Characteristics of a melanoma tri-culture spheroid model. Tri-culture spheroids were made by 3D cultivation of CCD-1137Sk fibroblasts for three days, followed by simultaneous addition of HaCaT keratinocytes and SK-MEL-28 melanoma cells, and then further culturing for another four days. HaCaT and SK-MEL-28 cells were labeled with CellTrackerRed CMPTX and CellTrackerGreen CMFDA dyes, respectively. As indicated, spheroids were incubated on day five after seeding either with 0.01 ‰ of DMSO as control (**a**) or 100 nM docetaxel in DMSO (**b**) for 48 h, then cryosectioned into 10-μm-thick slices and stained for Ki67, CAS3, CK10, CK14, and the basal membrane marker collagen-IV. Nuclei were labeled with Dapi. Images show representative confocal sections through these samples. In overlay panels, all immunostainings except for CK14 are depicted in red, SK-MEL28 cells in green, HaCaT cells or CK14 in yellow, and nuclei in blue. Scale bars: 100 μm
Docetaxel treatment affects external but not internal melanoma cells {#Sec10}
--------------------------------------------------------------------
To evaluate the tri-culture as a model system for testing drug candidates, we here chose to try a cytostatic, docetaxel, which is being explored in particular for combination treatments of malignant melanoma \[[@CR33]--[@CR35]\]. To find a useful concentration for in vitro tests, a dose-response curve was prepared. After their proper formation, tri-cultures were incubated for 48 h with different concentrations of docetaxel, i.e. 0 nM, 10 nM, 50 nM, 100 nM, 500 nM, and 1000 nM (Additional file [2](#MOESM2){ref-type="media"}: Figure S2). As exemplified in Fig. [2b](#Fig2){ref-type="fig"}, incubation with 100 nM of docetaxel strongly affected external melanoma cells and led to their nearly quantitative loss. Conversely, neither internal melanoma cells nor fibroblasts nor keratinocytes showed obvious defects. Quantification of remaining external melanoma cells after 48 h as a primary surrogate end point showed that already the lowest docetaxel concentrations slightly reduced external melanoma cells, but the effects were statistically significant in this setting only at concentrations ≥100 nM of docetaxel (Additional file [2](#MOESM2){ref-type="media"}: Figure S2). Therefore, in all following experiments this drug concentration was used.
Docetaxel treatment of tri-culture spheroids decreases melanoma cell proliferation {#Sec11}
----------------------------------------------------------------------------------
To get an insight into the kinetics of docetaxel effects on the proliferation of the tri-culture spheroids, these were harvested after 15, 24, 48, and 72 h of treatment with 100 nM of docetaxel. Samples were cryosectioned and slices were stained for the proliferation marker Ki67. Figure [3a and b](#Fig3){ref-type="fig"} depict representative fields of view. DMSO controls showed a continuous increase in the number of external SK-MEL-28 cells over time (Fig. [3a](#Fig3){ref-type="fig"}, Table [1](#Tab1){ref-type="table"}) and between 82.4% ± 2.4% (mean ± SEM, at 15 h) and 79.1% ± 3.2% (mean ± SEM, at 72 h) of those cells were proliferating. Conversely, docetaxel-treated spheroids led to increasing ablation of external melanoma cells (Fig. [3b](#Fig3){ref-type="fig"}, Table [1](#Tab1){ref-type="table"}). Notably, survival of internal SK-MEL-28 cells in the fibroblast core was apparently not affected by docetaxel. Since external melanoma cells constituted the major source of proliferating cells in untreated tri-cultures, their selective loss upon docetaxel treatment reduced the fraction of Ki67-positive cells in the entire tri-culture from 20.9% ± 3.4% (mean ± SEM, *n* = 5 independent experiments, Fig. [3c](#Fig3){ref-type="fig"}) to 10.2% ± 3.0% (mean ± SEM, n = 5 independent experiments, Fig. [3c](#Fig3){ref-type="fig"}) after 48 h of treatment. 72 h after start of treatment, the difference was even higher with 9.2% ± 2.1% (mean ± SEM, *n* = 4 independent experiments, Fig. [3c](#Fig3){ref-type="fig"}) compared to 27.8% ± 3.6% (mean ± SEM, *n* = 3 independent experiments, Fig. [3c](#Fig3){ref-type="fig"}) in the presence and absence of docetaxel, respectively. In summary, these data demonstrate that docetaxel affects proliferating cells, which are in this model primarily external melanoma cells.Fig. 3Docetaxel treatment reduces the amounts of proliferating cells in melanoma 3D tri-cultures. Tri-culture spheroids were produced by 3D cultivation of fibroblasts for three days, followed by simultaneous addition of keratinocytes and melanoma cells. HaCaT cells and SK-MEL-28 cells were labeled with CellTrackerRed CMPTX dye and CellTrackerGreen CMFDA, respectively. After another cultivation period of two days, tri-culture spheroids were treated with 0.01 ‰ of DMSO as control or 100 nM docetaxel for 15, 24, 48, and 72 h. Spheroids were cryosectioned into 10-μm thick slices and stained for Ki67. **a** and **b** Representative confocal images of control (**a**) and docetaxel-treated cultures (**b**). In overlay images, Ki67 immunostaining signals, SK-MEL-28, HaCaT, and nuclei are depicted in red, green, yellow, and blue, respectively. Scale bars: 100 μm. **c** Quantification of Ki67-positive cells. Graph displays the amounts of Ki67-positive cells as mean ± SEM (n ≥ 3 independent experiments; \*\* P \< 0.01) in percent of the whole cell count per slice. For each experiment and time point, ≥ 3 spheroids were analyzedTable 1Docetaxel progressively reduces the numbers of external SK-MEL-28. Tri-culture spheroids were harvested after 15, 24, 48, and 72 h of treatment with 100 nM of docetaxel or 0.01 ‰ of DMSO. HaCaT and SK-MEL-28 cells were labeled with CellTrackerRed CMPTX and CellTrackerGreen CMFDA dyes, respectively. Upon cryosectioning into 10-μm thick slices, labeling of nuclei with Dapi, and confocal imaging, external SK-MEL-28 cells were counted using ImageJ. Shown is mean ± SEM of the number of external SK-MEL-28 cells over time (*n* ≥ 3 independent experiments)DMSO control100 nM docetaxel15 h80.4 ± 6.681.1 ± 4.024 h77.6 ± 6.958.4 ± 6.2 \*\*48 h142.9 ± 20.533.1 ± 3.1 \*\*72 h135.2 ± 1.919.7 ± 3.2 \*\*
Docetaxel treatment does apparently not affect apoptosis in tri-culture spheroids {#Sec12}
---------------------------------------------------------------------------------
We next investigated the effect of docetaxel on the apoptosis of tri-culture spheroids over time. Therefore, tri-cultures were incubated with solvent control or docetaxel as before and harvested after 15, 24, 48, and 72 h of treatment. Cryosections were made and stained for CAS3. Figure [4](#Fig4){ref-type="fig"} a-b shows representative confocal microscopy images of spheroids treated with solvent control (Fig. [4a](#Fig4){ref-type="fig"}) or docetaxel (Fig. [4b](#Fig4){ref-type="fig"}). The general morphology of tri-cultures with increasing and decreasing numbers of external melanoma cells in the control and docetaxel-treated samples, respectively, was as observed before. Yet, quantitative analysis showed that the fraction of CAS3-positive cells in the entire tri-culture was apparently unaltered by docetaxel. Only at 24 h of treatment, a significant difference was found with 68.1% ± 6.7% (mean ± SEM, *n* = 3 independent experiments, Fig. [4](#Fig4){ref-type="fig"}c) compared to 46.9% ± 10.5% (mean ± SEM, *n* = 4 independent experiments, Fig. [4](#Fig4){ref-type="fig"}c) of CAS3-positive cells in the absence versus presence of docetaxel. This result suggested that apoptosis was either not involved in the drug response or that technical constraints of the model might have blurred this information.Fig. 4Loss of external melanoma cells upon docetaxel treatment hampers mechanistic explanation of drug effects. Tri-culture spheroids were generated by 3D cultivation of fibroblasts for three days, followed by simultaneous addition of keratinocytes and melanoma cells. HaCaT and SK-MEL-28 cells were labeled with CellTrackerRed CMPTX and CellTrackerGreen CMFDA dyes, respectively. After another two days, tri-culture spheroids were treated with 0.01 ‰ of DMSO as control or 100 nM docetaxel for 15, 24, 48, and 72 h. Spheroids were cryosectioned into 10-μm thick slices and stained for CAS3. **a** and **b** Representative confocal images of control (**a**) and docetaxel-treated cultures (**b**). In overlay images, CAS3 immunostaining signals, SK-MEL-28, HaCaT, and nuclei are depicted in red, green, yellow, and blue, respectively. Scale bars: 100 μm. (**C**) Quantification of CAS3-positive cells. Graph displays the amounts of CAS3-positive cells as mean ± SEM (n ≥ 3 independent experiments; \* *P* \< 0.05) in percent of the whole cell count per slice. For each experiment and time point, ≥ 3 spheroids were analyzed
A modified preparation avoids docetaxel-induced loss of external SK-MEL-28 cells {#Sec13}
--------------------------------------------------------------------------------
Apparently, treatment of tri-culture spheroids with docetaxel led to an ablation of external SK-MEL-28 cells. To understand the fate of those cells, we performed experiments using a 3D-mold technique. Reasoning that docetaxel might weaken cell-cell interactions and that many of the affected cells might have been lost in the previous assays where tri-cultures were transferred by pipetting from the culture to a staining/washing station after docetaxel treatment (for a scheme, see Additional file [1](#MOESM1){ref-type="media"}: Figure S1B), we here tried to avoid any post-treatment stress to the samples. Thus, tri-culture spheroids were first grown as before in cell-repellent plates, but were then transferred on day five into an agarose 3D mold (Additional file [1](#MOESM1){ref-type="media"}: Figure S1C). Treatment with docetaxel and all following processing steps were carried out in these molds. Indeed, the entire molds with the treated spheroids inside were embedded in OCT, cryosectioned, and slices were then stained for nuclei and Ki67 or CAS3. Figure [5a](#Fig5){ref-type="fig"} shows a comparison of representative confocal images of the tri-culture spheroids processed with both techniques, i.e. agarose mold ('with mold', left panels) and standard transfer washing/staining station technique ('without mold', right panel). As shown in Fig. [5a](#Fig5){ref-type="fig"}, the majority of external melanoma cells were lost or still present upon docetaxel treatment when processed without or within molds, respectively. This was confirmed by quantitative analysis of external melanoma cells, which revealed significant differences in the numbers of external melanoma cells between the two techniques (Fig. [5b](#Fig5){ref-type="fig"}). Upon processing without the molds, docetaxel treatment resulted in a significant decrease of the number of external SK-MEL-28 cells from 135 ± 2 to 20 ± 3 (mean ± SEM, *n* ≥ 3 independent experiments, Fig. [5](#Fig5){ref-type="fig"}b). Conversely, when processed within the molds, the number of external melanoma cells remained unchanged after docetaxel treatment (Fig. [5](#Fig5){ref-type="fig"}b). With respect to the number of proliferating cells, the processing technique had less impact. Indeed, in both, with and without the mold, docetaxel treatment led to a significant drop of the Ki67-positive numbers of external melanoma cells (Fig. [5](#Fig5){ref-type="fig"}c). However, the processing was relevant when addressing the fraction of apoptotic cells. Using the processing without mold, the number of CAS3-positive external melanoma cells decreased from 59 ± 12 cells (mean ± SEM, *n* = 3 independent experiments) to 6 ± 3 cells (mean ± SEM, *n* = 4 independent experiments) in the absence versus presence of docetaxel (Fig. [5](#Fig5){ref-type="fig"}d). Conversely, when processed within the molds, the number of CAS3-positive external SK-MEL-28 cells significantly increased from 25 ± 6 (mean ± SEM, n = 4 independent experiments) with DMSO to 56 ± 5 (mean ± SEM, n = 4 independent experiments) with docetaxel (Fig. [5d](#Fig5){ref-type="fig"}). In summary, these data suggest that proper post-treatment processing is essential for the interpretation of the behavior of external melanoma cells and that a combination of reduced proliferation and increased apoptosis is induced by docetaxel treatment. Furthermore, docetaxel-induced apoptosis can, at least partially, explain the loss of external melanoma cells in the tri-culture model.Fig. 5Processing of tri-cultures in special agarose molds reveals docetaxel-induced increase of apoptosis and reduction of proliferation of external melanoma cells. Tri-culture spheroids were generated by 3D cultivation of fibroblasts for 3 days, followed by simultaneous addition of keratinocytes and melanoma cells. HaCaT cells were labeled with CellTrackerRed CMPTX dye and SK-MEL-28 cells with CellTrackerGreen CMFDA dye. For the mold technique, spheroids were transferred on day five into 3D-agarose molds and then treated with 0.01 ‰ of DMSO as control or 100 nM docetaxel in DMSO for 72 h. Washing and embedding for cryosectioning occurred in the molds, as well. For the samples without mold, spheroids handled as in Figs. [3](#Fig3){ref-type="fig"} and [4](#Fig4){ref-type="fig"}, i.e. they were treated in the cell repellent plate and then transferred to an Eppendorf tube for washing and embedding. Subsequently, all spheroids were cryosectioned into 20-μm thick slices and stained for either Ki67 or CAS3 as indicated (**a**). Scale bars: 100 μm. **b-d** Quantification of total numbers of external SK-MEL-28 cells (**b**), as well as amounts of Ki67- (**c**) and CAS3-positive external SK-MEL-28 cells (**d**) of tri-culture spheroids processed with or without mold. Graph displays mean ± SEM (n ≥ 3 independent experiments; \*\* P \< 0.01). For each experiment, ≥ 3 spheroids were analyzed
Docetaxel treatment of tri-culture spheroids restores keratinocyte differentiation {#Sec14}
----------------------------------------------------------------------------------
Previously, it was observed in human malignant melanoma biopsies that neoplastic cells hamper keratinocyte differentiation \[[@CR13]\]. To address this finding in our 3D-melanoma model, we stained cryosections of tri-culture spheroids with an antibody against CK10 in the absence and presence of docetaxel. Figure [6a-b](#Fig6){ref-type="fig"} shows representative confocal images of these samples. In control-treated spheroids, an increase in the number of external SK-MEL-28 cells as well as a low level of the keratinocyte differentiation marker CK10 were observed throughout the experiment time course (Fig. [6a](#Fig6){ref-type="fig"}). In contrast, treatment with docetaxel resulted in a restoration of CK10 expression that occurred concomitant to the ablation/apoptosis of external melanoma cells (Fig. [6b](#Fig6){ref-type="fig"}). Quantitative analysis revealed that the number of peripheral CK10-positive cells was significantly higher in docetaxel treated spheroids as compared to controls, beginning at 24 h of treatment until the end of the observation period (Fig. [6c](#Fig6){ref-type="fig"}). Treated tri-culture spheroids reached a number of CK10-positive cells of 36% ± 5% (mean ± SEM, *n* = 4 independent experiments, Fig. [6c](#Fig6){ref-type="fig"}). In summary, this suggests that the tri-culture model is able to reflect effects of melanoma cells on keratinocyte differentiation as observed in human disease and that such loss of differentiation can be restored by treatment with docetaxel.Fig. 6Melanoma cells impair expression of keratinocyte differentiation marker CK10. Tri-culture spheroids were generated by 3D cultivation of fibroblasts for three days, followed by simultaneous addition of keratinocytes and melanoma cells. HaCaT and SK-MEL-28 cells were labeled with CellTrackerRed CMPTX dye and CellTrackerGreen CMFDA, respectively. After another two days, tri-culture spheroids were treated with 0.01 ‰ of DMSO as control or 100 nM docetaxel for 15, 24, 48, and 72 h. Spheroids were cryosectioned into 10-μm thick slices and stained for CK10. (**A and B**) Representative confocal images of control (**A**) and docetaxel-treated cultures (**B**). In overlay images, CK10 immunostaining signals, SK-MEL-28, HaCaT, and nuclei are depicted in red, green, yellow, and blue, respectively. Scale bars: 100 μm. (**C**) Quantification of CK10-positive cells. Graph displays the amounts of CK10-positive cells as mean ± SEM (n ≥ 3 independent experiments; \*\* P \< 0.01) in percent of the peripheral nuclei per slice. For each experiment and time point, ≥ 3 spheroids were analyzed
Docetaxel treatment leads to augmented ABCB5-signals in external melanoma cells {#Sec15}
-------------------------------------------------------------------------------
Considering the multidrug resistance capabilities of ABCB5 for doxorubicin and temozolomide in melanoma cells \[[@CR23], [@CR24]\], we wondered whether there would be a correlation between ABCB5 expression and melanoma cell survival to drug treatment also in the tri-culture model. Therefore, tri-cultures were exposed to 100 nM of docetaxel or DMSO for 48 h, fixed, sliced, and then stained for ABCB5 with m3C2-1D12 \[[@CR23]\] primary antibody. Surface expression of ABCB5 and specificity of the antibody on SK-MEL-28 cells were proven using flow cytometry and competitive peptide analyses in immunofluorescence (Additional file [3](#MOESM3){ref-type="media"}: Figure S3). As depicted in Fig. [7a-f](#Fig7){ref-type="fig"}, mostly melanoma cells as well as keratinocytes showed ABCB5 immunoreactivity. Quantitative analysis confirmed an increase in the number of external melanoma cells with high ABCB5 immunofluorescence intensity upon drug treatment (Fig. [7g](#Fig7){ref-type="fig"}), while internal melanoma cells were apparently unaffected in that respect (Fig. [7h](#Fig7){ref-type="fig"}). Interestingly, also keratinocytes displayed an increase in ABCB5 immunofluorescence upon docetaxel treatment (compare Fig. [7b and e](#Fig7){ref-type="fig"}). The findings regarding the drug-induced enrichment of strongly ABCB5-positive external SK-MEL-28 and HaCaT cells as well as the lack of effect on internal melanoma cells were confirmed by another anti-ABCB5 antibody (Additional file [4](#MOESM4){ref-type="media"}: Figure S4). In summary, these data are consistent either with docetaxel-induced enhancement of ABCB5 expression in weakly expressing cells or selection of cells with high ABCB5 levels.Fig. 7Cytostatic treatment leads to enhanced ABCB5-signals in keratinocytes and external melanoma cells. Tri-culture spheroids were generated by 3D cultivation of CCD-1137Sk cells for three days, followed by the combined addition of HaCaT and SK-MEL-28 cells. HaCaT cells were labeled with CellTrackerRed CMPTX dye and SK-MEL-28 cells with CellTrackerGreen CMFDA dye. After another two days, tri-culture spheroids were treated with 0.01 ‰ of DMSO as control (**a-c**) or 100 nM docetaxel in DMSO (**d-f**) for 48 h. Spheroids were cryosectioned into 10-μm thick slices and immunostained for mouse anti-ABCB5 from TICEBA. **a** and **d** Overlay images of the confocal sections shown in **b** and **e**. In overlays, ABCB5 signals, melanoma cells, keratinocytes, and nuclei are depicted in red, green, yellow, and blue, respectively. Scale bars: 100 μm. **c** and **f** Detail images of ABCB5 signals from boxed regions in **b** and **e**. **g-h** Quantification of the relative intensity of ABCB5-positive external (**g**) and internal (**h**) SK-MEL-28 cells (percentage of total). Given is mean ± SEM (*n* = 4 independent experiments; \* P \< 0.05, \*\* P \< 0.01). For each experiment, ≥ 3 spheroids were analyzed
Discussion {#Sec16}
==========
3D in vitro models of melanoma are increasingly used to study drug efficacy and mode of action as well as drug combinations. Compared to classical two-dimensional cell cultures, 3D models are thought to better represent a series of parameters that are critical for cancer cell behavior, including substrate stiffness, cell-cell interactions, distribution of oxygen and waste products, as well as drug diffusion \[[@CR36]\]. Currently existing 3D models are mostly tuned for either simplicity and high-throughput, complexity and similarity to the in vivo situation, or personalized medicine \[[@CR31], [@CR32], [@CR36]--[@CR43]\]. Here, we aimed to set up an early-stage 3D melanoma model that would allow to investigate several relevant drug-induced processes in a quantitative and cell-type specific manner. Yet, it should be also fast, easy, and robust in performance. The solution presented in this study is an easy to handle spheroid-based model composed of melanoma cells and major cellular components of a stroma-like environment, i.e. human fibroblasts and keratinocytes. To avoid batch-to-batch variability and to render the system cost effective, we opted for the established cell lines SK-MEL-28, HaCaT, and CCD-1137Sk. Within these constraints, the model was found to reliably mimic melanoma cell invasion into the dermal compartment, drug-induced selection of ABCB5-expressing melanoma cells, and loss of melanoma-induced keratinocyte differentiation.
As for the latter, optimal differentiation of keratinocytes in vitro leading to human skin equivalents (HSE), typically requires the use of primary cells and multifactorial external control, including pH and Ca^2+^ gradients as well as air lift \[[@CR38], [@CR41], [@CR42]\]. Given that these operations are time consuming and difficult to generate in high numbers, we avoided such complex maneuvers and allowed HaCaT cells to automatically stratify on top of a fibroblast core. Compared to HSE models, we observed a partial differentiation pattern that included stratification into lower and upper strata expressing either CK14 or CK10, respectively, but lacked a cornified layer. However, while HSE models often come with generation times of several weeks \[[@CR44]\], the spheroid-based tri-culture was ready for use after only seven days. It is arguable, whether the observed stratification really reflects differentiation or if pre-differentiated cells migrated to the outer regions of the spheroid. Yet, we observed an interesting related feature that is also known from human melanoma. Indeed, melanoma cells were reported to influence the differentiation pattern of human epidermal keratinocytes in vivo, i.e. that it leads to a loss of CK10 in hyperplastic regions \[[@CR13]\]. Consistent with this, we found that melanoma cells also decreased CK10 expression by HaCaT cells in our tri-culture spheroids (Figs. [2](#Fig2){ref-type="fig"} and [6](#Fig6){ref-type="fig"}). Notably, such loss of CK10 expression mostly occurred in direct vicinity of melanoma cells and keratinocyte differentiation was restored upon treatment with docetaxel which led to apoptosis of external melanoma cells (Fig. [6](#Fig6){ref-type="fig"}).
Another interesting feature of the present model was the division of melanoma cells into two populations, i.e. external and internal. The finding of internal SK-MEL-28 cells suggested their invasion into the fibroblast core and this would fit to the fact that this cell line is from the metastatic phase of melanoma \[[@CR37]\] and known to rapidly migrate downwards through the skin \[[@CR45]\]. In general, it would be interesting to further explore these cultures as a simple test system for antimigratory effects of diverse drugs. Apart from these future prospects, the identification of two melanoma cell pools was also interesting, because both pools showed differential behavior in at least three characteristics. First, external melanoma cells, which were located on the outside of the spheroids and thus in direct contact to keratinocytes, tended to form growing aggregates. Conversely, internal melanoma cells, which were found in the fibroblast core, were typically solitary and did not coalesce (Fig. [2](#Fig2){ref-type="fig"}). Apart from the first day after adding the keratinocyte-melanoma cell mixture to the fibroblast core, melanoma cells were hardly ever found in the HaCaT ring but always in the fibroblast core (Additional file [5](#MOESM5){ref-type="media"}: Figure S5). The second clear difference between internal and external melanoma cells was their response to docetaxel. While external cells massively went into apoptosis and became loose, internal melanoma cells remained apparently unaffected. Their numbers were stable even after 72 h of treatment and the relative amounts of apoptotic and proliferating cells was unaltered. It would be interesting to know whether such differential behavior was due to limited access of the drug to the spheroid core or rather due to cell-specific differences. For example, it could be that only drug-resistant cell subpopulations, which are frequently observed in malignant melanoma \[[@CR46]\], were able to invade the fibroblast core or whether some cellular signaling within the core would have led to drug insensitivity. In any case, it was intriguing to observe that docetaxel changed proliferation and apoptosis apparently only in melanoma cells but not in keratinocytes or fibroblasts. A third difference between external and internal melanoma cells was related to their expression of the ATP-dependent transporter protein, ABCB5. Based on ABCB5 immunofluorescence signal intensity profiles, docetaxel led to higher ABCB5 immunofluorescence signals in the external but not the internal melanoma cells (Fig. [7](#Fig7){ref-type="fig"} and Additional file [4](#MOESM4){ref-type="media"}: Figure S4). We do not think that internal melanoma cells would have been unable to increase ABCB5 expression, because -- contrary to our expectation -- they typically showed lower ABCB5 signals than external ones before treatment. Thus, it could again be that internal melanoma cells either represented a special subpopulation of cells, or they were not exposed to sufficient amounts of the drug, or their local microenvironment impaired such drug-induced changes in gene expression. However, the observed effect of docetaxel on ABCB5 signals in external melanoma cells is compatible with either, an up-regulation of ABCB5 in weakly expressing cells or selection of strongly expressing cells. In general, our results fit nicely to previous studies, which reported that ABCB5 expression is enhanced in malignant melanoma \[[@CR47]\], that it has a functional role in tumor growth \[[@CR48]\], and that chemotherapy leads to the selection of ABCB5-expressing cells \[[@CR24]\].
For the present study, docetaxel was used as a test substance. Although mitogen-activated protein kinase pathway inhibitors and immunotherapies against the immune checkpoints cytotoxic T lymphocyte-associated antigen and programmed death 1 have largely replaced classical alkylating and cytostatic chemotherapeutics as first-line treatment \[[@CR20], [@CR22]\], the mitotic inhibitor paclitaxel and its derivative docetaxel \[[@CR49]\] are being considered as adjuvant treatments \[[@CR19], [@CR21], [@CR22]\] and explored for use in novel formulations (see e.g. \[[@CR34], [@CR35], [@CR50], [@CR51]\]). Given that our model in its current version lacks immune cells, immunotherapies were not in the focus of this study. While the addition of T-cells and other immune cell components to future adaptations of the present 3D tri-culture might be valuable ideas to follow, we here concentrated on the effects of a classical agent on melanoma cells and on their chemoresistance features. With respect to effective drug concentrations, significant effects on external SK-MEL-28 survival were observed after 48 h at 100 nM of docetaxel. In comparison, SK-MEL-28 cells cultured in 2D appeared much more susceptible to docetaxel treatment (Additional file [2](#MOESM2){ref-type="media"}: Figure S2). This is in agreement with previous studies, which found a maximal effect of docetaxel at around 10--20 nM on different 2D melanoma cell cultures \[[@CR52]\] and a generally altered sensitivity of cells grown in 2D versus 3D \[[@CR53]--[@CR55]\].
In comparison with the present study, other three-dimensional melanoma spheroid models using, for example, the liquid overlay method \[[@CR56]\] are only composed of one cell type, the melanoma cells. Thus, they do not aim to represent the stromal environment of a tumor. On the other hand, HSE models are often generated by seeding primary fibroblasts in collagen type I followed by simultaneous seeding of primary keratinocytes together with melanoma cells \[[@CR57]\] or by seeding melanoma cells with primary fibroblasts to embed both cell types in the collagen type I matrix \[[@CR44]\]. This method spontaneously forms melanoma nests. Therefore, numbers and sizes of such nests might vary between individual skin reconstructs. As a consequence, it is often difficult to quantitatively validate these models and to predict therapeutic impacts. Conversely, the present tri-culture spheroid model always formed very similar sizes of spheroids with a highly reproducible arrangement of the different cell types allowing reliable quantification of cellular drug effects. Next, skin-on-a-chip models can be performed under a controlled perfusion of growth factors or nutrients \[[@CR58]\]. This cannot be realized in a static spheroid-based system as presented here. Using a skin-on-a-chip platform, Abaci and co-workers demonstrated that the cancer drug, doxorubicin, may have direct toxic effects on keratinocyte proliferation and differentiation \[[@CR59]\]. However, this platform is not suitable for high-throughput screening. For this purpose, simple spheroid models might be more appropriate. Given that the tri-culture spheroid model is composed of both, stroma and tumor cells, it is also possible to test general cell toxicity of a drug by evaluating the effect on surrounding non-transformed cells \[[@CR60]\].
On a technical note, the docetaxel treatment of tricultures led to a consistent loss of external melanoma cells. However, this was only true when the cultures were transferred after treatment from the spheroid formation plate into another container for washing (Fig. [5](#Fig5){ref-type="fig"}). If docetaxel treatment, washing, and embedding were carried out altogether without any transfer, presumably all -- dead and alive -- cells were still present in the immediate vicinity of the spheroids. Although it cannot be completely excluded that the observed difference in cell numbers was due to a distinct effect of the drug in the agarose mold versus the plastic plate, the most straight forward explanation appears to be that many of the external melanoma cells became loose upon drug treatment and were lost during the transfer from one container to the next due to mechanical shear force (see Additional file [1](#MOESM1){ref-type="media"}: Figure S1B for schematic illustration). This finding might be of general interest, because similar mechanisms of treatment-induced cell loss could possibly also occur in other spheroid or organoid models. If phenotypic quantitative analysis of either culture size, cell number, or fraction of apoptotic or proliferating cells is used, this effect could easily lead to erroneous data interpretation. Clearly, further investigation in that direction would be useful.
Conclusions {#Sec17}
===========
In the present study, a convenient spheroid-based tri-culture melanoma model was established. This model is composed of fibroblasts, keratinocytes, and melanoma cells that arrange in a highly reproducible and quantifiable manner in 3D. Melanoma cell invasion into the fibroblast core, melanoma-induced loss of keratinocyte differentiation, and cell-type specific drug responses to docetaxel were described as major hallmarks of the model. Future applications might consider addition of further cell types, including immune or primary cells, to further expand the applicability of such a system for the screening of drug candidates and their modes of action.
Additional files
================
{#Sec18}
Additional file 1:**Figure S1.** Transfer of docetaxel-treated tri-cultures leads to massive loss of external melanoma cells. Drawings schematically depicting the general composition of the tri-culture model (**A**) and the proposed mechanism for the loss of external SK-MEL-28 cells upon docetaxel treatment (**B**). (**A**) A core of CCD-Sk1137 fibroblasts (grey) is surrounded by a ring of CK14-positive HaCaT keratinocytes (yellow), and this by CK10-positive HaCaT keratinocytes (red). SK-MEL-28 melanoma cells (green) can be divided in individual 'internal' melanoma cells found largely in the fibroblast core, and clustered 'external' melanoma cells located on the outer rim of the tri-cultures. (**B**) In all experiments, spheroid formation was performed in cell repellent plates. In mold experiments (left part), spheroids were then transferred to an agarose mold, where docetaxel treatment was followed by washing and embedding for cryosectioning. Subsequently, cryosections were immunostained. In experiments without mold, docetaxel treatment was also done in the cell repellent plate. Then, treated spheroids were transferred to another standard plastic well for washing and embedding. Presumably, external melanoma cells got loose upon docetaxel treatment and were largely lost upon transfer in the experiments without mold. This is schematically shown by the loosened cells in the pipette on the right side of the scheme. (**C**) Micrograph of a tri-culture spheroid in the agarose mold. Note, that the agarose does not cover the spheroid, thus, docetaxel can freely access the spheroid as in the standard plastic well. The advantage of the mold is, that it can be directly cryosectioned avoiding further steps of pipetting. (JPG 1420 kb) Additional file 2:**Figure S2.** Comparison of SK-MEL-28 response to docetaxel in 2D versus 3D**.** 2D cultures of SK-MEL-28 cells were grown up to 50% of confluency. Tri-culture spheroids were produced by 3D cultivation of fibroblasts for 3 days, followed by the combined addition of keratinocytes and melanoma cells, and another 2 days without treatment. Then, all cultures were treated with different concentrations of docetaxel for 24 h (2D) or 48 h (spheroids). Spheroids were cryosectioned into 10-μm-thick slices, 2D cultures were directly fixed. Subsequently, all samples were labeled with Dapi and then imaged by confocal microscopy. The numbers of remaining SK-MEL-28 cells (2D cultures) or of external SK-MEL-28 cells (spheroids) were determined. The graph shows the amounts of SK-MEL-28 cells as a function of docetaxel concentration and normalized to the control condition without docetaxel. Given is mean ± SEM (*n* ≥ 3; \* *P* \< 0.05, \*\* *P* \< 0.01). (JPG 173 kb) Additional file 3:**Figure S3.** Specificity of m3C2 anti-ABCB5 antibody on SK-MEL-28 cells is proven by FACS and immunofluorescence methods. (**A-D**) SK-MEL-28 cells were analyzed for surface expression of ABCB5 by incubation of 2.5 × 10^5^ cells for 30 min at 4 °C with m3C2-1D12 anti-ABCB5 antibody or MOPC-31C mouse isotype control antibody (10 μg/ml). This was followed by incubation with FITC-conjugated goat anti-mouse secondary antibody (PharMingen) and single-color flow cytometry. Panels depict cytometry-scatter plots of unstained (**A**), only secondary-antibody stained (**B**), isotype plus secondary-antibody stained (**C**), or anti-ABCB5 plus secondary-antibody stained samples (**D**). Gate C was used to count ABCB5-positive cells. This contained 0.34% ± 0.15% (mean ± SD) and 6.64% ± 1.46% (mean ± SD) of cells in C and D, respectively. (**E-H**) Specificity of m3C2-1D12 anti-ABCB5 antibody on immunofluorescence of SK-MEL-28 cells was tested using standard protocols in the presence of FITC-conjugated secondary antibody only (**E**) or of m3C2-1D12 plus FITC-conjugated secondary antibody (**F-H**). In addition, primary antibody binding was competed by incubation of 2 μM ABCB5 epitope peptide (**F**) or scrambled peptide (**G**). Scale bar: 20 μm. (JPG 962 kb) Additional file 4:**Figure S4.** Enhancement of ABCB5-signals in keratinocytes and external melanoma cells upon docetaxel treatment is confirmed by a second anti-ABCB5 antibody. Tri-culture spheroids were generated by 3D cultivation of CCD-1137Sk cells for 3 days, followed by the combined addition of HaCaT and SK-MEL-28 cells. HaCaT and SK-MEL-28 cells were labeled with CellTrackerRed CMPTX and CellTrackerGreen CMFDA dye, respectively. After another 2 days, tri-culture spheroids were treated with 0.01 ‰ of DMSO as control (**A-C**) or 100 nM docetaxel in DMSO (**D-F**) for 48 h. Spheroids were cryosectioned into 10-μm thick slices and immunostained with mouse anti-ABCB5 antibody MA5--17026. (**A and D**) Overlay images of the confocal sections shown in B and E. In overlays, ABCB5 signals, melanoma cells, keratinocytes, and nuclei are depicted in red, green, yellow, and blue, respectively. Scale bars: 100 μm. (C and F) Detail images of ABCB5 signals from boxed regions in B and E. (G-H) Quantification of the relative intensity of ABCB5-positive external (G) and internal (H) SK-MEL-28 cells (percentage of total). Given is mean ± SEM (*n* = 4 independent experiments; \* *P* \< 0.05, \*\* *P* \< 0.01). For each experiment, ≥ 3 spheroids were analyzed. (JPG 1327 kb) Additional file 5:**Figure S5.** Accumulation of external melanoma cells in tri-cultures is an active separation process. Tri-culture spheroids were generated by 3D cultivation of fibroblasts for 3 days, followed by simultaneous addition of keratinocytes and melanoma cells. HaCaT and SK-MEL-28 cells were pre-labeled with CellTrackerRed CMPTX and CellTrackerGreen CMFDA dyes, respectively. After another one ('day 4', upper row) or 2 days ('day 5', lower panels), tri-culture spheroids were cryosectioned into 10-μm thick slices and stained with Dapi. Representative confocal images are shown. While most melanoma cells were embedded in the keratinocyte ring on day four, they segregated from keratinocytes on day five and either accumulated in the periphery of the culture ('external' melanoma cells) or within the fibroblast core ('internal' melanoma cells). The fibroblast core is located in the center of the tri-culture and identified as Dapi-positive plus CellTracker-negative. Scale bars: 100 μm. (JPG 467 kb)
ABCB5
: ATP-binding cassette transporter type B5
ANOVA
: analysis of variance
BRAF
: gene encoding proto-oncogene B-Raf
CAS3
: cleaved caspase 3
CK10
: cytokeratin 10
CK14
: cytokeratin 14
DMSO
: dimethylsulfoxide
HSE
: human skin equivalent
Ki67
: Antigen Ki-67
We thank Prof. Frank for providing anti-ABCB5 antibodies and Prof. Gretz for providing SK-MEL-28 and HaCaT cells.
Funding {#FPar1}
=======
This work was funded by the German Federal Ministry of Research (BMBF) as part of the Innovation Partnership M^2^Aind, project SM^2^all (03FH8I01IA) within the framework "Starke Fachhochschulen -- Impuls für die Region" (FH-Impuls). This research project is part of the Forschungscampus M^2^OLIE and funded by the German Federal Ministry of Education and Research (BMBF) within the "Framework Forschungscampus: public-private partnership for Innovations". At no point in the study did the funding agencies influence the design of the study, the collection, analysis, or interpretation of data, nor the writing of the manuscript.
Availability of data and materials {#FPar2}
==================================
The data used during the current study are available from the corresponding author on reasonable request.
JK, CM conducted the cell biology and molecular biology experiments. JK, CM, RR analyzed results. JK performed statistical analysis. JK, CM, AK, MH, RR designed the whole experiments. JK, RR wrote the paper. CM, AK, MH edited the manuscript. All authors have read and approved the manuscript, and ensure that this is the case.
Ethics approval and consent to participate {#FPar3}
==========================================
Not applicable.
Consent for publication {#FPar4}
=======================
Not applicable.
Competing interests {#FPar5}
===================
The authors declare that they have no competing interests.
Publisher's Note {#FPar6}
================
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
| {
"pile_set_name": "PubMed Central"
} |
Review
======
Introduction
------------
Conduct problems are common and disabling. Based on a survey by the Office of National Statistics (UK) from 1999\[[@B1]\], 5.3% of all children and adolescents between the ages of 5--15 had clinically significant conduct problems, the commonest reason for referral for psychological and psychiatric treatment in childhood \[[@B2]\]. The prognosis for children with conduct problems is poor, with outcomes in adulthood including criminal behaviour, alcoholism, drug abuse, domestic violence, child abuse and a range of psychiatric disorders \[[@B3]-[@B6]\].
Conduct problems are costly\[[@B7]\] due to the trauma and psychological problems caused to others who are victims of crime, aggression or bullying, together with the financial costs of services for treatment of both the condition and its long-term sequelae. Services include community youth justice services, prison services, social services, psychiatric, general practice and A&E services, and the costs of unemployment and other benefits. A recent UK study\[[@B8]\] covering a limited selection of these costs suggested that by age 28, costs for individuals with a clinical diagnosis of conduct disorder were 10.0 times higher than for those with no problems (CI: 3.6 to 20.9) and costs for those with conduct problems not meeting diagnostic criteria were 3.5 times higher (CI: 1.7 to 6.2).
Treatment for conduct problems
------------------------------
Various interventions have been used to treat conduct disorder including behaviour therapy, residential treatment, drugs, family therapy, multisystemic therapy and programmes which aim to improve parenting. The latter are unique in that they are structured, short-term interventions (average of two-hourly weekly sessions over 10--12 weeks) provided in a variety of settings (hospital, community, clinic/office or home) with a group or with individual parents (face-to-face or via telephone). They are directed at parents and reflect an increasing recognition that aspects of parenting such as boundary setting, positive discipline and warm and affectionate relationships are key in the prevention of behaviour problems \[[@B9]\].
A range of professionals can deliver the programmes, including psychologists, therapists/counsellors, social or community workers. In self-administered courses parents are encouraged to view videotapes or read training materials (books and leaflets). In some programmes the index child attends as well as the parents allowing parents to rehearse new skills or therapists to coach parent-child interaction. Some parenting programmes cover additional components such as stress or anger management.
There has been a rapid expansion of group based parent-training programmes over the past 10 years \[[@B10]\] and the provision of parenting programmes is central to the UK governments\' social inclusion agenda. A systematic review of existing reviews of the effectiveness of parent training for conduct disorder that were judged to be of high quality using a recognised checklist \[[@B11]\] suggested that parenting programmes are an effective intervention for children with behaviour problems.
Two of these reviews produce summary measures suggesting parent training programmes have a significant positive effect in crime prevention \[[@B12]\] and for non-compliant children \[[@B13]\] although this latter review does not provide any indication of the uncertainty of the effect estimate. One review reports a summary measure suggesting a non significant trend favouring parent training in children 0--3 years \[[@B14]\]. Two reviews do not report summary measures of effectiveness but suggest that parent training has a positive effect on children\'s behaviour problems, parental well-being and social outcomes \[[@B15]\] and a positive effect for young children with conduct disorder \[[@B16]\].
In addition two recent reviews have investigated moderators of effectiveness of parenting programmes on disruptive child behaviour \[[@B17]\] and on child externalizing behaviour problems \[[@B18]\]. Variables such as socioeconomic status, the inclusion of children in the parenting programme, maternal mental health and individual versus group approaches to delivery moderated effectiveness although these effects tended to be modest.
However these existing reviews have limitations, such as the inclusion of non-randomised studies, the absence of a test for heterogeneity prior to the conduct of a meta-analysis and failure to report confidence intervals. The two reviews investigating moderators of effectiveness both suffer from statistical limitations such as use of small data sets and underestimation of heterogeneity. In addition these existing reviews have largely been restricted to the impact of parenting programmes on specific population sub-groups and have not endeavoured to estimate the overall impact of parenting programmes on children with conduct problems. Further no existing reviews have attempted to compare the relative effectiveness of different types of programmes.
The objective of the current study was therefore to systematically review randomised controlled trials (RCTs) of parenting programmes for the treatment of children (≤ 18 yrs) with conduct problems to investigate i) the overall effectiveness of parenting programmes, and; ii) the relative effectiveness of different approaches to delivery.
Methods
=======
Search strategy
---------------
Twenty electronic databases (including PsycInfo, MEDLINE, EMBASE and the Cochrane Library) from the fields of medicine, social science and education, and the National Research Register Issue 1 (2006) were searched up to February 2006. There were no language restrictions. In addition citations from previous reviews and included studies were searched and information was requested from manufacturers and experts.
Inclusion and exclusion
-----------------------
Studies were included if: (a) they were RCTs, (b) the population comprised parents/carers of children up to the age of 18 where at least 50% had a conduct problem (defined using objective clinical criteria, the clinical cut-off point on a well validated behaviour scale or informal diagnostic criteria), (c) the intervention was a structured, repeatable (manualised) parenting programme (any theoretical basis, setting or mode of delivery) and (d) there was at least one standardised outcome measuring child behaviour. Studies where children accompanied their parents to all or some of the sessions were included providing the main focus of treatment was on the parents (i.e. children were present for parental skill rehearsal or assessment). Inclusion of studies was not restricted by child or parental co-morbidity or by type of comparator (e.g. wait list control, different parenting programme or other treatment).
Studies were excluded where the intervention (a) was aimed at prevention rather than treatment; (b) was aimed specifically at children, the whole family as a unit or at teachers; or (c) was non-structured, such as an informal support group or unstructured home visits.
Quality assessment and data extraction
--------------------------------------
Potential threats to internal study validity (selection bias, detection bias, performance bias, attrition bias) were assessed using Cochrane Collaboration \[[@B19]\] criteria. Appropriateness of statistical analyses was critically appraised by statisticians. Inclusion and exclusion of studies, data extraction and quality assessment were undertaken in duplicate, with discrepancies being resolved by a third reviewer.
Data analysis and synthesis
---------------------------
Studies that had used a child-behaviour measure (reported in at least 20% of all studies) and where there was sufficient statistical information were synthesised quantitatively (n = 24 studies). All meta-analyses were undertaken in Stata™ 7.0. Standardised mean differences were derived to take account of the variety of behavioural outcome measures included and random effect models adopted in view of variability of the intervention and target populations across studies. Tests for publication bias (Egger and Begg tests) were also undertaken.
Planned subgroup analyses involved comparisons between different approaches to delivery for four key characteristics: group or individual or self-administered, length of programme (same or different), index child involvement or adjunctive treatment.
In order to look at the evidence from all relevant studies a vote-counting exercise was undertaken to assess the results of included studies that had not used one of the predominant child-behaviour measures or had not provided enough statistical information to be included in the meta-analysis. For the vote-counting exercise a statistically significant (p ≤ 0.05) difference in favour of the intervention was considered a positive outcome, a statistically significant difference in favour of control was considered a negative outcome and no statistically significant difference was considered a neutral outcome. Thirty eight studies reporting 170 child-behaviour outcome measures were included in the vote-counting exercise.
Ethics approval
---------------
Ethics approval was not required.
Results
=======
Figure [1](#F1){ref-type="fig"} shows the inclusion and exclusion process. Fifty seven studies were included of which 40 included a control comparison group (no treatment). Twenty eight studies compared parent training with an alternative form of parent training: 17 of these compared parent training with an alternative form of parent training only (no control comparison group) and 11 studies compared parent training with alternative parent training and a control comparison group.
{#F1}
Intervention characteristics (57 included studies)
--------------------------------------------------
The majority of interventions (n = 37) focussed on the parents alone. In 20 studies the intervention(s) involved the child at various levels of intensity, from attendance at all sessions (e.g. Barrett *et al*., 2000\[[@B20]\]), attendance at some sessions for parental skills rehearsal (e.g. 3/8 sessions Pfiffner *et al*., 1990\[[@B21]\]) or observation of children in another setting with feedback to parents during home visits (Sanders & McFarland 2000\[[@B22]\]). Most studies (n = 24) investigated group programmes, of these 23 focussed on parents only. Twenty studies investigated individual based programmes, 15 of which involved index children at some level. The remaining studies investigated self-administered programmes (n = 5) or combinations of group, individual and self-administered programmes (n = 8). Adjunctive treatment such as partner support training, friendship liaison or treatment of depression, was included in the intervention in 8/28 studies comparing two or more parenting programmes. In 3 studies, children were receiving medication for ADHD, all other studies either specifically excluded children receiving concurrent treatment or did not give details of concurrent treatment. No studies comparing parenting programmes with a control group evaluated outcomes past 6 months and only a minority (n = 5) compared 2 alternative interventions between 1 and 3 years.
Concerning 102 parent training programmes and within study variations of these programmes across 57 studies. The majority of programmes (51) were conducted over 10 sessions or less; 17 programmes were 11--20 sessions in length and 10 programmes were greater than 20 sessions in length. For 24 programmes the number of sessions was unclear or not stated. Interventions that were not self-administered (93) were delivered by a variety of professionals: 40 programmes were delivered by psychologists, 1 each delivered by a teacher and a psychiatric nurse and in 51 programmes the professional background of the person delivering the programme was unclear. Social workers were jointly involved in 7 programmes. The great majority of programmes (86) were based on behavioural approaches, 8 on relationship approaches and 4 on both approaches. For 4 programmes the underlying principle was not clear or not stated.
Population characteristics
--------------------------
Recruitment of populations was via self-referral, media advertisement or fliers in 44 studies; through health professionals or organisations in 10 studies and in 3 studies there was no information on recruitment.
Index children were aged 12 and under or had a mean age \< 12 in 49/57 studies and 68% of the agregated study population were male.
Diagnostic criteria (DSM \[[@B23]\]or clinical cut-off on a behavioural scale such as the Eyeberg Child Behaviour Inventory \[[@B24]\]) were used to recruit populations in 48 studies and in 9 studies parent or professional description of child behaviour was used.
In 10 studies some or all children had a diagnosis of Attention Deficit Hyperactivity Disorder (ADHD).
Of 22 studies reporting ethnicity \> 70% of study populations were white Caucasian families.
Of the 26 studies reporting family structure more than 30% of index children were in single parent households.
Quality of research
-------------------
Few studies reported sufficient information to assess all aspects of quality, and in particular lacked detail about methods of randomisation and allocation concealment. Further detail is provided in \[Additional file [1](#S1){ref-type="supplementary-material"}\]. No studies were completely bias free, but 4 studies were considered to be of good quality on the basis of only one threat to validity out of a total possible of five \[[@B25]-[@B28]\].
No evidence of publication bias was found.
Effectiveness results
---------------------
### Parent-report of outcome
A total of 24 studies contributed a parent-report measure of outcome \[[@B25],[@B26],[@B29]-[@B50]\]. Details of these studies can be found in \[Additional file [1](#S1){ref-type="supplementary-material"}\]. Two instruments were used -- Eyberg Child Behaviour Inventory (ECBI): Intensity (n = 20) and the Child Behaviour Checklist (CBCL) (n = 4). The ECBI is a parental report of conduct behavioural problems in children and adolescents that measures the number of difficult behaviour problems (intensity) and the frequency with which they occur \[[@B24]\].
The CBCL is a device by which parents or other individuals who know the child well, rate a child\'s problem behaviours and competencies \[[@B51]\].
The results were combined using a random effects model, and the combined results (see Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}) show a significant standardised mean difference favouring the intervention group of -0.67 (95% CI: -0.91, -0.42). The results were similar (SMD -0.62 95% CI: -0.85, -0.40) where the frequency scale (i.e. as opposed to the Intensity scale) of the Eyberg Child Behaviour Inventory was used as the main outcome.
{#F2}
{#F3}
### Independent assessment of outcome
Only 7 studies provided independent assessments of outcomes all of which were undertaken using the Dyadic Parent Interactive Child Scale (DPICS). DPICS is designed for use in assessing the quality of parent-child social interaction. Interaction between parent and child in three standard situations that vary in the degree to which parental control is required is observed and coded by an independent observer behind a two-way mirror \[[@B52]\]. DPICS scores were combined using a random effects model and the combined data (see Figure [4](#F4){ref-type="fig"}) show a significant standardised mean difference favouring the intervention group of SMD -0.44 (95% CI: -0.66, -0.23).
{#F4}
### Vote Counting
The results of the vote-counting supported the results of the meta-analysis.
Of 170 child behaviour outcomes measured across 36 studies, 59% were statistically significant and favoured parenting programme over control, with the remaining outcomes showing no statistically significant difference (a neutral outcome). No study demonstrated a less favourable outcome for parent-training compared to control.
### Relative effectiveness of different approaches to delivery
28 included RCTs compared one parenting programme with another. Most studies were small and none of the studies reported a power calculation to estimate the number of individuals required in order to detect a significant difference in effect for the outcomes measured. Only 10 studies directly compared programmes that differed in only one of the four key characteristics: delivery approach (group, individual or self-administered), length of programme, child involvement and adjunctive treatment (or none)\[[@B21],[@B22],[@B39],[@B47],[@B48],[@B53]-[@B57]\]. Comparisons possible were: 3 studies with treatment arms differing only in the approach (group, individual or self-administered), 2 studies differing only in number of sessions and 5 studies differing only in adjunctive treatment. Of 26 behavioural measure comparisons used across these 10 studies only 4 were reported as significantly different. These are detailed in \[Additional file [2](#S2){ref-type="supplementary-material"}\].
Discussion
==========
These results show that using both parent-report and independent observations of outcome, parenting programmes are effective in improving conduct problems. Independent observations of change were on the whole smaller than parent-report (SMD of 0.4 compared with 0.7), and very few (7/25) of the included studies had provided an independent assessment of outcome.
There was insufficient evidence to show clear superiority of any one approach to delivery. Many of the comparisons that were undertaken were invalidated by the fact that more than one of the four key characteristics (i.e. group versus one to one, length; child involvement; adjunctive treatment) was varied. Of the ten studies that compared programmes, which varied in only one of the key characteristics, few differences were identified. This is most likely to be due to inadequate power in this analysis.
There may be some restrictions in terms of the generalisability of these findings, due to the involvement in many studies of parents who had self-referred. Similarly, due to the case-mix in many trials there is also some uncertainty regarding the families that would most benefit from this form of treatment.
Our review was restricted to a limited number of behavioural outcomes and we were unable to exploit the full range of behavioural outcome measures used across included studies and for some studies reporting of multiple measures of child behaviour in the meta-analysis. Other reviews have suggested that parenting programmes can have a significant impact on parent psychosocial well-being including stress and self-esteem\[[@B58]\], and that there may be some benefit of such programmes irrespective of ethnic group\[[@B59]\].
Further RCTs comparing different approaches are still needed, focusing in particular on those features that are likely to influence cost as well as effect, such as group versus individual programmes. There is also a need to compare the effectiveness of different programmes in primary studies.
Uncertainty remains regarding the importance of the improvements in child behaviour scores and how these improvements translate into clinically meaningful outcomes. Those who remain sceptical that the demonstrated changes in conduct problems translate into important gains in health and quality of life will point to the need for research quantifying the relationship between change in child behaviour scores and health utility in the index child as well as parents, siblings and peers. Research addressing the long-term impact of parenting programmes is also required.
Work on cost-effectiveness carried out as part of the previous HTA report on this topic\[[@B60]\] and by the Decision Support Unit at the National Institute for Health & Clinical Excellence (NICE) \[[@B61]\] suggests that group-clinic based parenting programmes are likely to be cost-effective or may lead to cost-savings through avoidance of alternative treatment.
Limitations of the review
-------------------------
While we conducted the review using established criteria \[[@B62]\] it is impossible to exclude certain sources of bias, particularly the possibility of having overlooked eligible studies. Furthermore, as a result of the data available it was not possible to incorporate the findings from all of the studies into the meta-analyses. As noted above, there was also a lack of independent assessments of the presence and size of improvements in conduct problems. Our application of strict inclusion criteria with respect to the structured and repeatable nature of the parenting programme interventions included in this review aimed to ensure that included interventions were similar enough in nature to be pooled in a meta-analysis. In addition the sub-group analysis did not demonstrate any measurable difference in effectiveness according to some aspects of intervention delivery. Nevertheless we cannot rule out the possibility that variation in effectiveness of individual programmes has not been detected.
Conclusion
==========
We conclude that on balance, parenting programmes are an effective treatment for children with conduct problems. The relative effectiveness of different parenting programmes requires further research.
Summary points
--------------
• Conduct problems among children and adolescents are associated with high psychological and financial costs and with poor prognosis if left untreated
• Parenting programmes are short-term, structured interventions, which have in previous reviews been shown to be effective in treating conduct problems in certain groups of children
• Our systematic review identified 57 randomised controlled trials, which compared parenting programmes to a wait list control or to an alternative form of parenting programme or other treatment
• There was a consistent trend across all studies showing a benefit from parenting programmes; meta-analysis of the most commonly reported child behaviour outcomes showing statistically significant improvements
• There was insufficient evidence to directly determine the relative effectiveness of one type of parenting programme delivery approach over another
• Parenting programmes are an effective treatment for children with conduct problems
Abbreviations
=============
RCT: randomised controlled trial; SMD: standardised mean difference.
Competing interests
===================
The authors declare that they have no competing interests.
Authors\' contributions
=======================
All authors contributed to protocol development. SB contributed to the development and running of search strategies. JD, CD, EF, CH, JB contributed to inclusion and exclusion of studies. JD, CD, EF, JB, RT, JS contributed to data extraction. JD, CD, RT, CH, JS, JB, SS-B contributed to clinical effectiveness analysis. JD, CD, CH, JB, EF, SS-B contributed to interpretation of effectiveness data and discussion. RT, JS gave statistical advice. JB, SS-B gave clinical advice. CH is the guarantor.
Supplementary Material
======================
###### Additional file 1
Characteristics of 24 RCTs included in the meta-analysis. The table provides information about study population characteristics; details of intervention and control groups; main results; quality assessment of studies and the outcome measure contributing to the meta-analysis.
######
Click here for file
###### Additional file 2
Relative effectiveness of parenting programmes. The table provides information about 10 studies directly comparing parenting programmes differing in only one of 4 key characteristics (delivery approach; programme length; child involvement and adjunctive treatment). Information includes type of comparison; child behaviour outcome measures demonstrating a significant difference between comparison groups; numbers of children in each comparison group.
######
Click here for file
Acknowledgements
================
This report was commissioned by the NHS R&D HTA programme. It was one component, which fed into the National Institute of Health and Clinical Excellence\'s (NICE) appraisal process on this topic. The author\'s work was independent of the funders.
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Ensuring skilled care at birth, with the right person in an enabled environment, can prevent mortality and morbidity in women and newborns. In high-burden and resource-scarce settings, such as countries of sub-Saharan Africa, the use of skilled care at birth is still far from universal \[[@CR1]\]. A wide range of different social, woman, birth-related, and macro-level barriers to using skilled care at birth have been identified in the literature \[[@CR2]--[@CR4]\]. Low household wealth/socioeconomic status (SES) and problematic physical accessibility to an adequate provider are amongst the most persistent barriers. A number of studies have shown that wealthier women consistently report higher use of skilled care at childbirth than their poorer counterparts \[[@CR5]--[@CR7]\]. For the poor, the direct (e.g. medical bills) and indirect (e.g. transportation, lost earnings) costs associated with seeking and using skilled childbirth care may be unaffordable \[[@CR8], [@CR9]\].
In addition to financial affordability, lack of physical accessibility to health services also imposes barriers to using skilled care at birth. Physical accessibility is determined by one's geographic location, and is captured by factors such as the distribution of facilities, travel time or distance from home to facility, availability of transportation, and the condition of roads. It shapes people's options for care-seeking and their decision making \[[@CR10]\], and can cause delays in reaching an adequate provider when needs arise \[[@CR3]\]. The negative effect of poor physical accessibility on the use of skilled care at birth was first reviewed by Thaddeus and Maine in 1994 \[[@CR4]\], and reaffirmed in systematic reviews, including Gabrysch and Campbell 2009 \[[@CR3]\], Moyer and Mustafa 2013 \[[@CR2]\], Wong et al. 2017 \[[@CR11]\] and Tegegne et al. 2018 \[[@CR12]\].
Removing financial and accessibility barriers may be complicated by the correlation between them \[[@CR13]\], since resource and infrastructure often concentrate in wealthier urban places, and are scant in poorer and remote areas. Higher availability and better accessibility to healthcare in wealthier urban places may exacerbate the inequity gap in health service uptake between people living in such places and their counterparts in poorer and remote areas. A recent study of wealth inequalities in travel time to the nearest hospital in Kenya, Malawi, Nigeria and Tanzania found dramatic differences between wealth subgroups. Average travel time to the nearest hospital for the wealthiest decile was \< 15 min -- 4-14 times shorter compared to the poorest deciles in these countries \[[@CR14]\]. Such gap in travel time raises questions regarding the potential overlap of the negative effects of poverty and travel time on use of skilled care at birth, in other words -- are women too poor or too far to use skilled care at birth? This question exposes a gap in the current literature about the separate and combined contributions of these two barriers.
To address this question, we propose to examine the variability in the proportion of births occurring in hospitals (rather than in any health facility), since the full range of live-saving "skilled" childbirth services, such as caesarean section and blood transfusion, are typically only available in hospital settings if at all \[[@CR15]\]; and equipment and staffing at lower-level, primary facilities (e.g., health centres/posts/huts and dispensaries) are often inadequate for the basic functions that they are expected to provide \[[@CR16]--[@CR18]\]. In this study, we quantify the relative contribution of poverty and travel time on rates of hospital birth in sub-Saharan African countries. We also aim to test if poverty and travel time interact. Our results generate insights that can be used for health policy making to ensure that the most left behind expectant mothers receive skilled and adequate care for childbirth.
Data and methods {#Sec2}
================
Study settings {#Sec3}
--------------
We studied four LMICs in sub-Saharan Africa -- Kenya, Malawi (excluding Likoma Island), Nigeria and Tanzania (excluding Zanzibar). These countries were selected over others in the sub-Saharan African region because they had a recent complete list of hospitals with geographic coordinates, and represented different contexts in terms of demography, geography, travel time to the nearest emergency care and facility-based childbirth. National statistics according to the World Bank \[[@CR19]\], the Demographic and Health Surveys Program \[[@CR20]\], and the 2015 geocoded inventory of emergency hospitals in sub-Saharan Africa by Ouma and colleagues \[[@CR21]\] are presented in Table [1](#Tab1){ref-type="table"}. Table 1Country data and statisticsKenyaMalawiNigeriaTanzaniaTotal area (km^2^) \[[@CR19]\]580,367118,484923,768947,300National population in 2015 (million) \[[@CR19]\]471818154% urban population in 2015 \[[@CR19]\]26164832% of all births in health facilities^a^ \[[@CR20]\],61.291.435.862.6% population \> 2 h travel time to public emergency hospital care \[[@CR21]\]77825^a^ The most recent Demographic and Health Survey as of January 2019 for each country -- Kenya 2014, Malawi 2015/16, Nigeria 2013 and Tanzania 2015/16
Data and measurement {#Sec4}
--------------------
We used four data sources: (a) Demographic and Health Surveys (DHS) to determine place of childbirth, household location, household wealth and other potential confounders, (b) a master list of all health facilities with geographic coordinates for each country, (c) the Global Friction Surface 2015 by the Malaria Atlas Project (MAP) is used in conjunction with (a) and (b) to determine travel time from household to hospital, and (d) country administrative boundary files (version 2.5, July 2015) downloaded from the GADM database on gadm.org \[[@CR22]\].
First, we used the most recent DHS as of January 2019 for each study country -- Kenya 2014, Malawi 2015/16, Nigeria 2013 and Tanzania 2015/16. The DHS collect nationally representative data on population health and sociodemographic characteristics using a multi-stage cluster sampling design with enumeration area as the cluster, or primary sampling unit. As part of the DHS sampling procedure, a list of established households in each sampled cluster is obtained and used as the sampling frame for household selection \[[@CR23]\]. All women aged 15--49 in selected households were interviewed with a standardized questionnaire with questions on all their livebirths in the 5 years before the survey. All these births were considered in the current analysis.
In each survey, a household wealth index was constructed by the DHS using household asset data via a principal component analysis \[[@CR24]\]. Each livebirth is assigned its household's wealth index. The outcome of interest is hospital-based childbirth. For each livebirth, place of childbirth was based on women's answer to: "Where did you give birth to \[name of child\]?" in the Women's Questionnaire. The major categories of response options were domestic environments (home of respondent, family member, or traditional birth assistant (TBA)), public/government sector health facilities and private/non-government sector health facilities. The DHS conflated clinics and hospitals as one response option for health facilities in the non-government sector for Kenya, Malawi and Nigeria. In line with the approach taken by Hanson and colleagues \[[@CR25]\], the categorisation of facility delivery locations into hospital was done in consideration of the local context and health system in each country, and the response options on the survey. Data on other potential predictors of hospital birth, including maternal education, maternal age at birth and birth order, were also sourced from the DHS. We captured the context-specific barriers associated with the lived environment beyond the predictor variables described here by including a random effect at the level of survey cluster.
The DHS include the longitude and latitude coordinates of the population centroids of sampled clusters. All individuals residing in the same cluster have the same geo-referenced location. For anonymity reasons, urban clusters are displaced up to 2 km and rural clusters up to 5 km \[[@CR26]\]. We excluded nine clusters in Kenya and seven clusters in Nigeria with missing coordinates from our analysis.
Second, master lists of health facilities were obtained online \[[@CR27]--[@CR31]\]. These lists are inventories of all government and non-government health facilities in the country, with data on facility type -- hospital vs. others -- and geographic coordinates. These lists contain facility data from 2015 (Kenya), 2013 (Malawi), 2010--2014 (Nigeria) and 2016 (Tanzania).
Third, we quantified physical access as the travel time required to travel from the displaced cluster centroid to the nearest hospital using the MAP Global Friction Surface (the friction surface below) 2015. The friction value represents the generalized difficulty to cross a pixel depending on land surface condition, such as the type of road, water bodies, and terrain with slope. Travel time to the nearest hospital was computed for every 1 × 1 km^2^ pixels covering the study region using an algorithm devised by Weiss and colleagues \[[@CR32]\]. This algorithm identifies the path that requires the least time through the friction surface between two points \[[@CR32]\], and has been used to construct accessibility maps enumerating travel time to the nearest hospital in previous studies \[[@CR14], [@CR33]\]. DHS suggests generating average values using neighbourhood buffers to moderate the potential impact of point displacements \[[@CR34]\]. In this study, we extracted travel time values for each DHS cluster as the average of the four nearest pixels.
Statistical analysis {#Sec5}
--------------------
We tested travel time estimated from the MAP friction surface by comparing 20% of DHS clusters (selected at random) against travel time estimates obtained using data from the OpenStreetMap (OSM) project \[[@CR35]\]. We used Pearson correlation coefficient to assess the linear correlation between the two sets of values.
Generalized additive models (GAMs) were used to assess the effects of wealth, travel time to the nearest hospital and other predictor variables on hospital birth \[[@CR36]\]. The "mgcv" package for the R statistical package \[[@CR37]\] was used to construct mixed-effects GAM models with the application of survey sampling weights. A different GAM was constructed for each country. A GAM model is expressed as $$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$ {\displaystyle \begin{array}{c} logit\left( hospital\ birth\right)={f}_1\left( wealth\ index, travel\ time\right)+{f}_2\left( maternal\ age\ at\ birth\right)+\\ {} maternal\ education+ birth\ order\end{array}} $$\end{document}$$
We used the logit link logit(*.)* to relate the predictors with the expected value of the response. Smoothing functions *f*~*i*~ are found for the different predictor variables. We tested whether the effect of travel time varied by wealth using an interaction term specified as a scale invariant tensor product smooth. For this term, we tested two different numbers of knots for smoothing -- 5 and 10. A penalized thin plate regression spline was fitted to maternal age at birth, as very young and very old women may use hospital childbirth care differently \[[@CR38]\]. A truncated eigen-decomposition is used to achieve the rank reduction \[[@CR37]\]. Linear terms were used for maternal education and birth order. We applied survey-specific weighting to account for the sampling procedures used in the surveys.
We present the marginal effects of all predictors from the fully-adjusted mixed-effects GAMs. For each model predictor, we calculated the predicted probabilities of hospital birth for every standard deviation (SD) change from mean -- μ ± 1SD -- whilst holding other predictors at the respective sample mean. These predictions showed the effect that varying each predictor variable within a country's population would result in. For normally-distributed data, with a mean and median being the same and 68% of the data falling within 1SD from the mean value, the comparison between μ-1SD, μ, μ + 1SD is equivalent to comparing the 16th, 50th and 84th percentiles. The marginal effect of the survey cluster random effect was obtained from the distribution of predicted values with all model predictor variables set to the sample mean. Again, we calculated the predicted probabilities of 1SD around the model mean predicted probabilities of hospital birth.
We further used a response surface to show the additive effect of DHS wealth index and travel time on hospital birth. The predicted probabilities were represented by a colour gradient. Model residuals were plotted as heat maps to show the locations at which the variability of hospital birth was well explained by the fully-adjusted mixed-effects GAM models.
Ethics approval {#Sec6}
---------------
The DHS receive government permission and follow ethical practices including informed consent and assurance of confidentiality. The authors requested and received approval to download and use the data from the DHS websites as detailed under the data sharing page. Master facility lists were publicly available \[[@CR23]\]. The Research Ethics Committee of the London School of Hygiene and Tropical Medicine approved our secondary-data analysis.
Results {#Sec7}
=======
Descriptive {#Sec8}
-----------
Across the study countries, the numbers of DHS clusters identified were 1565 (Kenya), 828 (Malawi), 889 (Nigeria), and 527 (Tanzania). Travel time estimated from the MAP friction surface and that obtained using OSM data showed good alignment (Pearson correlation coefficients over 0.75 in all countries, see Additional file [1](#MOESM1){ref-type="media"}), apart from a few clusters with long travel time of ≥5 h estimated using the MAP friction surface. For this reason, we excluded 12 and 6 clusters from Kenya and Tanzania from the final analysis (Fig. [1](#Fig1){ref-type="fig"}). Fig. 1Map of the study region, hospitals and DHS clusters. {#d29e722}Hospital; {#d29e725} DHS clusters in the study region; {#d29e728}DHS clusters excluded from the final analysis due to high estimated travel time
The numbers of DHS clusters, livebirths and hospitals used in our final analysis are shown in Table [2](#Tab2){ref-type="table"}, together with summary statistics of travel time to the nearest hospital and the percentage of births in hospitals by country. Overall, Kenya and Nigeria had the shortest mean travel time from clusters to the nearest hospital (about 25 min), and Tanzania the longest (62 min). Travel time was highly right-skewed, and a cube-root transformation was used in subsequent analyses. The percentage of births in hospitals ranged between 27% in Nigeria to 39% in Kenya. Majority of hospital births occurred in government hospitals, except in Nigeria, where the shares of government hospital births and non-government hospital births were similar (Table [2](#Tab2){ref-type="table"}). Table 2Summary statistics in study countriesKenyaMalawiNigeriaTanzaniaDHS survey year20142015/1620132015/16Number of DHS clusters1585828889527Number of DHS clusters^a^\<5 h from a hospital1573828889521Number of livebirths included in the final analysis^b^19,46317,38431,8288317Year of master facility list data201520132010--20142016Number of hospitals in the master facility list4851163787265Number of geo-referenced hospitals4801153787265Travel time to the nearest hospital in minutes Mean (standard deviation)26.6 (40.5)30.9 (28.5)25.2 (33.5)61.7 (58.4) Median (interquartile range)12.7 (4.1--29.8)24.9 (10.7--40.7)14.2 (3.7--34.1)45.1 (16.9--87.9) Maximum291.2268.3293.9296.0Percentage distribution of place of childbirth among livebirths included in the final analysis^b^HospitalGovernment sector30.327.414.123.0Non-government sector9.17.913.08.3Other health facilitiesGovernment sector15.851.48.527.1Non-government sector6.14.80.23.6Not in a health facility (own/TBA/other home)37.27.163.237.9Unknown/missing1.51.51.00.0Total percentage of hospital childbirth39.435.327.131.4Total percentage of facility childbirth61.391.435.862.1*TBA* Traditional birth attendant^a^ Excluding Likoma Island in Malawi (22 DHS clusters) and Zanzibar in Tanzania (81 DHS clusters), and DHS clusters without geographic coordinates (9 in Kenya and 7 in Nigeria)^b^ The final analysis comprised livebirths from geo-referenced survey clusters \< 5 h from a hospital, and with the same residence at the time of survey and birth (where data was available)
The association of wealth, travel time, and other covariates with hospital birth {#Sec9}
--------------------------------------------------------------------------------
The deviances explained by the fully-adjusted mixed-effects GAMs were similar using both 5 and 10 knots for smoothing on the interaction term between travel time and wealth (Additional file [1](#MOESM1){ref-type="media"}). We present results from the simpler models with 5 knots. Results of the fully-adjusted mixed-effects GAMs are shown in Table [3](#Tab3){ref-type="table"}. All predictor variables were significant. The mean predicted probabilities of hospital birth obtained from these models were 33.2% (Kenya), 32.7% (Malawi), 26.6% (Nigeria) and 29.6% (Tanzania). Table 3Results of generalized additive models of hospital-based childbirth by countryKenyaMalawiNigeriaTanzaniaApproximate significance of smooth termsEDFREF DF*p*-valueEDFREF DF*p*-valueEDFREF DF*p*-valueEDFREF DF*p*-valueWealth index × travel time ($\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$ \sqrt[3]{\mathrm{hours}} $$\end{document}$)6.487.31\< 0.00110.7124.00\< 0.00111.7724.00\< 0.0018.3724.00\< 0.001Maternal age at birth (years)2.362.96\< 0.0012.899.00\< 0.0012.549.00\< 0.0013.796.00\< 0.001Parametric coefficients of linear termsESTSE*p*-valueESTSE*p*-valueESTSE*p*-valueESTSE*p*-valueMaternal education (years)0.060.01\< 0.0010.030.01\< 0.0010.090.00\< 0.001−0.050.01\< 0.001Birth order− 0.280.02\< 0.001− 0.120.02\< 0.001−0.100.01\< 0.001−0.160.03\< 0.001Random effectsEDFREF DF*p*-valueEDFREF DF*p*-valueEDFREF DF*p*-valueEDFREF DF*p*-valueSurvey cluster5151052\< 0.001482609\< 0.001575701\< 0.001319481\< 0.001Mean of predicted probabilityof hospital birth (%)33.232.726.629.6*EST* Estimate, *SD* Standard error, *EDF* Estimated degrees of freedom, *REF DF* Reference degrees of freedom
Figure [2](#Fig2){ref-type="fig"} shows the marginal effect of 1 SD change from mean for each predictor variable whilst holding other model covariates at sample mean. In Kenya, compared to the average model-predicted value of 33.2%, a decrease in wealth index by 1SD from the mean reduced the predicted probability of hospital birth to 16.1%, and an 1SD increase from mean brought the predicted probability of hospital birth to 49.3% -- a difference of 33.2 percentage points between the 16th and 84th percentiles. The marginal effect of μ ± 1SD change for travel time was weaker than that of wealth index (16.6 percentage points). The overall additive effect between wealth index and travel time by 1SD around the mean was 43.8 percentage points. The marginal effect of μ ± 1SD change for maternal age at birth, maternal education and birth order were 10.8, 9.9 and 25.0 percentage points, respectively. Lastly, the survey cluster random effect for 1SD change from mean was obtained from the distribution of predicted probabilities of hospital birth, whilst holding all other predictor variables at the sample mean. Comparing survey clusters 1SD below and above the mean led to a change of 21.0 percentage points in the predicted probability of hospital birth. Fig. 2Marginal effects of one standard deviation (SD) change from mean (μ) of the predictor variables on the predicted probabilities of hospital birth
In Malawi, the marginal effect of 1SD change in wealth was weaker than that of travel time (13.1 versus 34.0 percentage points), and additive effect between wealth and travel time was not notably stronger (36.0 percentage points) than individual effect of travel time alone. In Nigeria, the marginal effects of wealth and travel time was similar (22.3 and 24.8 percentage points), and their additive effect was considerably stronger (44.6 percentage points). In Tanzania, the marginal effect of wealth was weaker than that of travel time (20.4 versus 33.7 percentage points), and their additive effect was stronger (50.4 percentage points). In all three countries, the marginal effects of maternal education, maternal age at birth and birth order were weaker than that of wealth and travel time. Survey clusters 1SD below and above the mean led to a change of approximately 30 percentage points in the predicted probability of hospital birth.
The additive effect of wealth and travel time {#Sec10}
---------------------------------------------
We then plotted the additive effects between wealth and travel time as response surfaces, with the other model predictors held at the sample mean (Fig. [3](#Fig3){ref-type="fig"}). The response surfaces show the predicted probabilities as a function of travel time and wealth. In all four countries, livebirths to women who lived closer to a hospital and were from the least poor (lower right corner of the graph) had the greatest predicted probability of hospital birth; whilst the poorest who lived furthest away (top left corner) had the lowest. In Kenya, however, the predicted probability of hospital birth was low for the poorest, regardless of travel time. In addition, the increase in predicted probability of hospital birth with wealth index levelled off for the least poor. On average, in Malawi the predicted probability of hospital birth was high only for those living close to a hospital, regardless of wealth. In Nigeria, the predicted probability of hospital birth was low for those with either a long travel time or a low wealth index. Fig. 3Predicted probability of hospital birth by travel time to the nearest hospital and household wealth index ^\^^. ^\^^ Model covariates -- maternal education, maternal age at birth and birth order -- were set to sample mean. Random effect at the survey cluster level was applied. All the observed combinations of values between travel time and wealth index were contained within the border. The colour gradient represents the value of the predicted probability of hospital birth (red: highest probabilities; blue: lowest probabilities). Contour lines are drawn to connect points that have the same predicted values. We drew contour lines for each 2.5% point increment in the predicted probabilities of hospital birth
The angle of the contour lines represents the responsiveness of predicted probabilities of hospital birth to changes in the two predictor variables. Contour lines angled close to being vertical in Kenya show that the predicted probabilities of hospital birth were more responsive to changes in wealth, and the effect of travel time was relatively weaker -- in line with results shown in Fig. [2](#Fig2){ref-type="fig"}. In Malawi, contour lines were angled more horizontally, indicating responsiveness of hospital birth to changes in travel time. In Nigeria, hospital birth was most responsive to changes in travel time among those who were far and poor, and less so for those who were far but less poor. The predicted probabilities of hospital birth were more responsive to changes in travel time for those living very far away in Tanzania.
The spaces between contour lines are widest among those who have the lowest predicted probability of hospital birth in Kenya, Nigeria and Tanzania, thus for them a fixed unit decrease in travel time and a fixed unit increase in wealth would have the smallest effect on the outcome. In Malawi, on the other hand, the widest gaps between contour lines were among those who have the highest predicted probability of hospital birth, for whom decreasing travel time or improving wealth would have the smallest increase in the likelihood of such births.
GAMs residuals {#Sec11}
--------------
Model residuals can show the extent of the variance in the data not explained by the model, with higher values indicating worse model fit. Model residuals were generally smallest when the predicted probability of hospital birth was low (Fig. [4](#Fig4){ref-type="fig"}), estimated travel time was short and wealth index was low to medium (Additional file [1](#MOESM1){ref-type="media"}). But there are exceptions; some groups of DHS clusters with low-to-medium predicted values stand out with large residuals, such as in Elwak, Bella Wagberi and Zubak in Kenya, Lilongwe in Malawi, and Kano and Gombe in Nigeria. In Nigeria, both high proportion of predicted hospital birth and high model residuals were mostly in the south, except for some costal clusters in southern Delta and Bayelsa States along the Gulf of Guinea. Fig. 4Model predicted probabilities of hospital birth and model residuals
Discussion {#Sec12}
==========
Summary of study results {#Sec13}
------------------------
Poverty and long travel time to health services are important barriers of maternity care-seeking in LMICs. They are commonly treated as collinear, and their separate effects have not been studied extensively. To our knowledge, this is the first study to partition their effects on hospital-based childbirth. We confirmed the substantial barriers posed by poverty and long travel time in Kenya, Malawi, Nigeria and Tanzania. By separating the effects of poverty and travel time, we found that the situation differed by country. The marginal effect of wealth on hospital birth was stronger than that of travel time in Kenya; the opposite was observed in Malawi and Tanzania. In Nigeria, the two were similar but their additive effect was twice as influential as their separate effects. Also, in Nigeria, hospital birth was generally most responsive to changes in travel time for women who were poor and lived the furthest away from a hospital. In most cases, women who were already least likely to give birth in a hospital would benefit the least from changes in wealth and travel time. Although both poverty and travel time were important, the random effects of survey clusters explained a substantial extent of between-cluster variability in hospital birth in all countries, indicating other unobserved local factors were at play.
Interpretation of results {#Sec14}
-------------------------
The differences in the relative contribution of poverty and long travel time on giving birth in a hospital within and across countries identified in our results require a context-specific interpretation. In Kenya, we found that wealth index was the predominant determinant of hospital birth for those from low- and middle-SES households. The Kenyan governments has implemented various pro-poor interventions to support the use of maternal health services since the early 2000 -- including childbirth fees abolishment in 2007 in government dispensaries and health centres (with the replacement of a registration fee of 10--20 Kenyan Shillings, ≈ 0.1--0.2 US dollars) \[[@CR39], [@CR40]\], and from 2006 to 2016 a reproductive health voucher programme under which poor women could purchase subsidized vouchers for 200 Kenyan Shillings to cover the cost of antenatal care, facility childbirth and postnatal care \[[@CR41], [@CR42]\]. In 2013, the government extended the abolishment of maternity services (including childbirth) fees in all levels of government health facilities under the Free Maternity Services (FMS) policy \[[@CR43]\]. Data used in our analysis primarily included childbirth prior to this change; other studies conducted afterwards have shown positive overall results -- including sustained increase in hospital-based childbirth (1--2 years post implementation) \[[@CR44], [@CR45]\], higher rates of childbirth in hospitals than in lower-level facilities \[[@CR46]\], greater increase of childbirth than antenatal care in hospitals \[[@CR47]\], and a mild decline in the use of low-cost private hospital for childbirth \[[@CR47]\] -- but a 2019 study found small gains in the wealth-inequality of skilled childbirth services following the announcement of the FMS policy due to a relatively small increase in service uptake among low SES women to catch up with existing inequality gap \[[@CR48]\].
In Tanzania, where both the number of hospitals by land area and average travel time to the nearest hospital were the least optimal among countries studied here \[[@CR14], [@CR21]\], we found that the effect of travel time was greater than that of wealth. Hospitals in Tanzania are primarily located in the southern and northern regions, with lower-level facilities serving rural areas in the central region. The Tanzanian government is committed to expanding service coverage so that people "don't have to travel long distance to access the services in distant facilities", putting forward projects to adding and renovating government health facilities in recent health policy plans \[[@CR49], [@CR50]\]. Both the Kenyan and Tanzanian governments have shown commendable attempts to support the use of maternal healthcare (including for childbirth) by removing user fees in public health facilities (Kenya and Tanzania) and making services geographically closer to the population (Tanzania) \[[@CR49], [@CR50]\]. The implementation of these different strategies, however, seems to face similar challenges. In Kenya, limited pre-existing health infrastructure and other supply-side capacity to match the increased workload following fee removal, insufficient referral and emergency obstetric care capacities contribute to persisting poor maternal (and newborn) health and its inequalities \[[@CR51], [@CR52]\]. Indeed, decline in maternal/neonatal mortality and stillbirths does not appear to have followed as a result of increase in facility utilization for childbirth \[[@CR44], [@CR53]\]. FMS in Kenyan government facilities may also have limited impact on increasing hospital birth for the poorest and the most remote women/families (among whom mortality and morbidity are typically the highest) due to the small number of hospitals that are within their reach \[[@CR52]\]. For Tanzania, some studies suggest that policies aiming to reduce distance or travel time, by expanding service provision, deteriorate service quality when scarce resources are diluted. This may put the poorest people who cannot pay the cost of bypassing their nearest facility at higher risk of receiving suboptimal care \[[@CR54], [@CR55]\]. To ensure access to adequate care for all, concerted effort and innovative targeting are required. In a setting of high facility density and limited resources, it has been shown that concentrating available resources in fewer, but strategically selected, facilities/sites may promote geographic accessibility for all. In Tanzania and other LMICs, positive outcomes in physical accessibility and quality of care were achieved when interventions were supported by the right tools and approaches \[[@CR55]--[@CR58]\].
The government of Malawi promotes childbirth at primary health facilities, with referral to hospitals for women known to be at high risk \[[@CR59], [@CR60]\]. As part of the Banda era legacy, Malawi had a reasonably strong health centre system, and in a relatively well populated small rural country this meant that most women were not geographically too far from one of these facilities. Health services in the government sector are free-of-charge at the point of use in the country \[[@CR59]\]. Since 2006, the government has also been progressively exempting childbirth fees for catchment populations of Christian Health Association of Malawi (CHAM) health facilities (often located in remote areas; approximately 40 and 25% of hospitals and health centres in the country are CHAM facilities, respectively \[[@CR61]\]). Malawi has attained a near universal level of facility birth -- 91% of livebirths in the 5 years before the 2015/16 DHS were delivered in a health facility \[[@CR62]\] -- yet only an estimated 25% of obstetric complications occurred in facilities with the capacity to provide the level of obstetric and newborn care required (such as in a hospital) \[[@CR63], [@CR64]\]. In pre-hospital settings, the median distance to the nearest point of obstetric surgical care is over 30 km. In *The Lancet*'s Maternal Health series in 2016, Campbell and colleagues called for all women to give birth in health facilities that can guarantee at least basic emergency obstetric care standard and timely referral for women with complications to higher-level care to ensure safe motherhood \[[@CR1]\]. Our results suggested that the overall effect of travel time on hospital birth was greater than that of wealth, and their additive effect did not substantially explain further variability. Measures should be put in place to improve physical accessibility to EmONC services, including strengthening the capacity of health centres (to which some solutions are available to strategically select locations for facility upgrading that balances travel time across the whole population and equity as defined by wealth subgroups \[[@CR14]\]); and expanding the provision of free maternal healthcare at more CHAM hospitals, especially those that are in very remote locales. However, recent reduction of development partners' contribution to the Malawian total health budget has impaired the fee exemption mechanism with CHAM, resulting in certain facilities re-introducing user fees to cope with the financial setback. Such reduction is speculated to be related to internal political instability, scandals and poor governances \[[@CR59], [@CR65]\]. Strategies that include fee-based, non-profitable health providers working in rural areas mitigates financial barriers to use of care and expands the options for higher-level health providers that poor remote dwellers are otherwise unable to use, thus shortening the travel time required to obtain and receive adequate care \[[@CR66], [@CR67]\]. Long-term implementation of these strategies should not be hampered by unfavourable policy environment and government challenges.
In Nigeria, women who either had to travel for long or were poor were very unlikely to give birth in a hospital. These women were concentrated in specific geographic settings, with the poorest being largely in the north, and especially in Yobe State, while women travelling for long were mostly in the southern coastal areas in Delta and Bayelsa States. For those in Yobe State, the effect of travel time appeared to be very strong. The state has one of the lowest levels of skilled care for childbirth in the country \[[@CR68]\], and while several studies have found ethnicity, social norm and religion as fundamental reasons for homebirths, there were also very few health facilities in the region \[[@CR69]\]. Lembani and colleagues further posited that the Boko Haram Insurgency in the area since 2011 has resulted in the destruction and closing of many health facilities, with health personnel preferring to relocate in other areas \[[@CR68]\]. The general lack of service provision in the area may have strongly affected the population's ability to access health services. On the other hand, for those in the south who are approximately equally far but are relatively less poor, wealth played a relatively stronger role. Difficult riverine terrains in Bayelsa State pose additional impediments to overcoming travel-related barriers \[[@CR70]\]. Although the area's energy sector has generated interest among multi-national companies \[[@CR71]\], most Bayelsans remain poor, while the state's public infrastructure is underdeveloped \[[@CR72], [@CR73]\]. The proportion of women in Bayelsa who cited financial reasons for homebirth is higher than the national average \[[@CR74]\]. Under such special economic and environment conditions, wealth may be additionally helpful for overcoming cost of transport, as well as trade-offs in time and financial loss from daily/productive activities.
In the context of health equity, horizontal equity refers to the principle that people with the same needs should have a similar level of access to the required health services; this contrasts to vertical equity which denotes unequal access to healthcare for people with different needs \[[@CR75]--[@CR77]\]. Assuming the need for skilled and adequate care for childbirth is universal or somewhat even across all population subgroups by sociodemographic characteristics (e.g., wealth and place of residence), the principle of horizontal equity is met if service uptake is also similarly distributed. In many LMICs, however, this is not the case. Wealth and physical accessibility to care continue to act as drivers of inequitable uptake of health services. Understanding variability in hospital births by poverty and travel time is useful for the design of policies to reduce inequity and strategies to reach populations that are likely to be left behind in terms of access to services. In addition, we note that our analysis revealed substantial survey cluster random effects, demonstrating local factors other than wealth and travel time are at play, and may limit the impact of strategies that are aimed at removing financial and accessibility barriers. Future studies are required to identify such local factors and how they can be overcome.
Study limitations {#Sec15}
-----------------
Our results have important implications but should be interpreted with a few limitations in mind. First, the estimation of travel time from DHS cluster centroids to the nearest hospital using the MAP friction surface assumes a generalized travel speed by the type of land surface, and does not account for variabilities in temporality, seasonality and transportation used by the individuals. In particular, in rural areas characterized by a high level of poverty, walking and non-motorized vehicles remain the major means of transportation, while adoption of motorized transportation is limited by affordability issues \[[@CR78]--[@CR81]\]. In contrast, there is a wider range of transportation in urban settings. Of these, private and privately-owned vehicles -- such as matatus in Kenya -- have become very common. In poorer urban areas, however, many people still struggle to afford the fees to take these private vehicles and walk, whilst others who can afford them face challenges due to poor road networks \[[@CR82]--[@CR84]\]. The additional cost, time and difficulty of movement likely mean that we may have underestimated travel time for urban poor households, and the true negative effect of long travel time on hospital-based childbirth may be stronger than the effect estimated. Second, the accuracy of our estimates of the effect of travel time may be influenced by the displacement of DHS cluster. Applying Karra and Canning's proposed method to correct the biased estimator with the expected minimum distance \[[@CR85]\], Sato and colleagues found larger corrected effects than the uncorrected effects for distance on facility-based childbirth and attendance by doctor in Tanzania, although the differences were small (\< 2 percentage points) \[[@CR86]\]. Third, we excluded DHS clusters for which the estimated travel time from the MAP friction surface was over 5 h. In checking our travel time estimates against those obtained from OSM Routing Services, larger discrepancies tended to come from long travel time estimates using the MAP friction surface. This only affected a small number of data points (12 in Kenya, 6 in Tanzania and none in Malawi and Nigeria), but more detailed validity assessment of travel time estimates might be relevant in future work where manual checking becomes a feasible task. Fourth, this analysis employed data on livebirths in the 5 years preceding survey interviews and hospital data at given timespans. Although their occurrences are rare, we may have missed a very small number hospitals that may have been opened, closed, upgraded or downgraded between the time of survey and listing of facilities. Fifth, the use of wealth index as a measure of poverty may not accurately identify the very poor \[[@CR7]\]; this may be particularly true for Malawi where the data appears to be considerably right-skewed. Sixth, we used one standard deviation around the mean as a consistent unit of change in our comparison of marginal effects of the model predictors. Other choices of unit (e.g. 5- or 10-year increment in maternal age at birth and maternal education, 60-min change in travel time) may vary the comparison and lead to different results. Last, our definition for hospital was based on data on the type of health facility as given in the master facility lists; the hospitals may vary in capacity, quality of care, and the range of health services that they provide. Such unmeasured covariates may confound the exposure to the outcome of our study.
Conclusion {#Sec16}
==========
By assessing the relative contribution of poverty and long travel time, we found that these two factors determine whether women give birth in hospitals to a varying extent within and across the four study countries. For the poor and those living in remote areas who do not give birth in hospitals, the effect of poverty was stronger in some cases, while the effect of long travel time was stronger in others. Given the focus of "leaving no one behind" in the Universal Health Coverage agenda, more precise identification of population subgroups who are more likely to be left behind in terms of access to health services warrants further research. Such additional understanding can help inform the financial and geographic barriers that people face, devise tailor-made system-wide strategies to bring skilled care to meet health needs, and ultimately contribute to attaining the desired improvements in maternal and newborn health in resource-limited settings.
Supplementary information
=========================
{#Sec17}
**Additional file 1.** Supplementary A: Additional information on the travel time estimates. Supplementary B: Model ouput. Supplementary C: Model predictions, model residuals, travel time and wealth index.
**Publisher's Note**
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
=========================
**Supplementary information** accompanies this paper at 10.1186/s12939-020-1123-y.
We wish to thank Dr. Mardieh Dennis, Professor Andrea Pembe and Dr. Hannah Blencowe for useful discussions.
KLMW, OJB, OMRC and LB conceptualized the study. KLMW and OJB designed the analytical approach. KLMW performed the data analysis, with supervision provided by OJB, OMRC and LB. OJB and LB provided scientific advice and support throughout the study. LB provided overall management. KLMW wrote the first draft of the manuscript. All authors gave feedback to the drafts and approved the final manuscript for publication.
OJB is supported by a Sir Henry Wellcome Fellowship funded by the Wellcome Trust (grant number 206471/Z/17/Z).
The datasets generated analysed during the current study are available in the following repositories:
1\. [dhsprogram.com](http://dhsprogram.com)
2\. <https://explorer.earthengine.google.com/#detail/Oxford%2FMAP%2Ffriction_surface_2015_v1_0>
3\. [http://downloads.afyaresearch.org/mfl/AbridgedeHealth Kenya Facilities Sept 2015.xls](http://downloads.afyaresearch.org/mfl/AbridgedeHealth%20Kenya%20Facilities%20Sept%202015.xls)
4\. [kmfhl.health.go.ke](http://kmfhl.health.go.ke)
5\. <https://databox.worldbank.org/en/dataset/nigeria-nmis-health-facility-data-2014>
6\. <http://moh.go.tz/hfrportal/index.php?r=site/index>
The DHS receive government permission and follow ethical practices including informed consent and assurance of confidentiality. The authors requested and received approval to download and use the data from the DHS websites as detailed under the data sharing page. Master facility lists were publicly available23. The Research Ethics Committee of the London School of Hygiene and Tropical Medicine approved our secondary-data analysis.
Not applicable.
The authors declare that they have no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
Notice of Republication {#sec001}
=======================
This article was republished on July 23, 2015, to replace the declaration of Competing interests, which was incorrect. While this article was in production, in-house editorial staff noted that the declaration of competing interests supplied by the authors required revision in order to ensure compliance with the journal's editorial policies. The editorial team followed up with the authors and the declaration was updated; unfortunately this updated declaration of competing interests was not reflected in the published article.
The republished article includes the correct declaration of Competing interests, as below:
**Competing interests:** The authors have received funding for this and earlier research from CRIIGEN, the Foundation Lea Nature and Malongo, the JMG Foundation and Foundations Charles Léopold Mayer for the Progress of Humankind, Nature Vivante, Denis Guichard, Institute Bio Forschung Austria, and the Sustainable Food Alliance. The laboratory received funding from Sevene Pharma in the last five years to study the detoxifying capacity of plant extracts on Roundup residues, bisphenol A and atrazin. Prof Seralini participated and received payment for a lecture organized by Sevene Pharma.
Supporting Information {#sec002}
======================
###### Originally published, uncorrected article.
(PDF)
######
Click here for additional data file.
###### Republished, corrected article.
(PDF)
######
Click here for additional data file.
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
Since the beginning of the 1990s, the evolution of information technology (IT) has spawned concerns about technological addiction \[[@b1-pi-2017-09-27-2]\], particularly concerning internet addiction \[[@b2-pi-2017-09-27-2]\]. However, the operational definition of internet addiction remains contentious. Whether internet addiction is even real is still debatable. While Young \[[@b2-pi-2017-09-27-2]\] described the concept of internet addiction, Griffiths \[[@b3-pi-2017-09-27-2]\] argued that the internet is just the place where people engage in specific behavior. That is, excessive users use the internet as a medium to fuel other addictions while not being addicted to the internet itself \[[@b4-pi-2017-09-27-2],[@b5-pi-2017-09-27-2]\]. A Second point of contention concerns the diagnostic criteria of internet addiction, that is what conditions should be satisfied for internet addiction diagnosis \[[@b6-pi-2017-09-27-2],[@b7-pi-2017-09-27-2]\]. For example, although Tao et al. \[[@b8-pi-2017-09-27-2]\] and Griffiths \[[@b9-pi-2017-09-27-2]\] indicated that tolerance is needed to identify addictive behavior, some researchers have argued that it is difficult to objectively define or measure tolerance for internet addiction, and there is a lack of grounds of tolerance even for substance use disorders \[[@b10-pi-2017-09-27-2]\]. Also, it is still unclear whether internet addiction is an isolated disease entity or whether it is a manifestation/subset of other underlying mental disease such as depressive disorder \[[@b11-pi-2017-09-27-2],[@b12-pi-2017-09-27-2]\], anxiety disorder \[[@b13-pi-2017-09-27-2]\], social phobia \[[@b14-pi-2017-09-27-2]\], attention-deficit hyperactivity disorder \[[@b15-pi-2017-09-27-2]\], or impulse control disorder \[[@b16-pi-2017-09-27-2],[@b17-pi-2017-09-27-2]\].
To clarify concepts and diagnoses of internet addiction, there have been attempts to explore the causes of the phenomenon. Some researchers argued that internet addiction can be resulted as a response or coping strategy to stressful event \[[@b18-pi-2017-09-27-2],[@b19-pi-2017-09-27-2]\], since the addictive behavior tends to be triggered when one cannot achieve a sufficient satisfaction from natural rewards \[[@b20-pi-2017-09-27-2]\]. Another study focused on individual sensation seeking tendency or impulsivity based on the unique feature of the internet that enables immediate satisfaction with minimal delay \[[@b19-pi-2017-09-27-2]\]. Other studies suggested that demographic factors like gender \[[@b6-pi-2017-09-27-2]\], education \[[@b21-pi-2017-09-27-2]\], or socioeconomic status could be risk factors for internet addiction \[[@b22-pi-2017-09-27-2],[@b23-pi-2017-09-27-2]\]. Other underlying mental disorders, personality traits, or low self-esteem might also lead to internet addiction \[[@b24-pi-2017-09-27-2],[@b25-pi-2017-09-27-2]\]. Recent studies have suggested that similar to other substance use disorders, internet addiction is likely to be related to neurobiological abnormalities \[[@b10-pi-2017-09-27-2]\] or dysfunctions of dopaminergic brain systems \[[@b26-pi-2017-09-27-2]\]. However, these studies seem to offer explanations only about vulnerabilities or the pathway to the disease state.
The reason why internet addiction studies are complicated is because internet addiction presents unique characteristics and results. For internet addiction, direct physical effects of substances do not affect the brain. However, sleep patterns are often disrupted by the extended nocturnal overuse of the internet. Snoring, teeth grinding, and sleep apnea can be prevalent in individuals considered to be internet addicts \[[@b27-pi-2017-09-27-2]\]. This leads to fatigue during daytime, reduces academic and occupational functioning, and negatively affects physical homeostasis and immune system function \[[@b2-pi-2017-09-27-2],[@b28-pi-2017-09-27-2]\]. The case of a South Korean male who died due to excessive internet gaming after a few sleepless nights has caused awareness about the serious results of internet addiction \[[@b29-pi-2017-09-27-2]\]. Also, internet addiction is likely to cause problems with family or friends, which can lead to secondary psychological and emotional problems \[[@b2-pi-2017-09-27-2]\].
With concerns over the negative consequences of excessive internet use, many studies have tried to explore the current status of internet addiction by developing appropriate screening or diagnostic tools. However, the tools devised so far have failed to reach agreement. Criticisms of the tools include differences in the criteria between tools and lack of appropriate standardization process \[[@b17-pi-2017-09-27-2],[@b30-pi-2017-09-27-2]\]. Moreover, research on screening tools for at-risk internet users is insufficient with a few exceptions including the Internet Addition Test (IAT) \[[@b2-pi-2017-09-27-2]\] or Chinese Internet Addiction Scale (CIAS) \[[@b31-pi-2017-09-27-2]\]. Even though the IAT and CIAS are widely used, these tools also have limitations. Since the IAT was intended for adults, it may not be appropriate to measure the problem of youth. The revised Korean version for adolescent was suggested \[[@b32-pi-2017-09-27-2]\], but the translation and modification process was not presented precisely. Also, the study did not provide diagnostic information such as cut-off point because the study was conducted with community-dwelling students without diagnostic interview. Although Young (1998) suggested cut-off point, it would be inappropriate to use it without considering the cultural differences. For CIAS, the cut-off point was well investigated \[[@b29-pi-2017-09-27-2]\]. The study, however, did not suggest factor structure or concurrent validity of the instrument. Also, CIAS has largely been used in the Taiwan and was not standardized in Korea. Although the Korean Scale for Internet Addiction (K-scale) also has been widely used in Korea \[[@b33-pi-2017-09-27-2]\], it contains some ambiguous expressions that make it difficult to interpret results. For example, item 2 for adolescents (There are more people who recognize me online than offline) or adults (I become more confident during using the internet) do not seem to reflect internet addiction properly.
Since the internet is essential for academic or occupational functioning in modern society, it is difficult to determine between essential-use and overuse with only a few brief questions. In addition, the problem is easily concealed if the subject denies the symptoms. Therefore, questions that reflect various aspects of lifestyle and fulfill an unmet need of the filed are needed. Of all internet users, an estimated 3% to 15% might be considered addicted, with those at high-risk of becoming addicted estimated as 10% to 40% \[[@b7-pi-2017-09-27-2],[@b17-pi-2017-09-27-2],[@b18-pi-2017-09-27-2],[@b34-pi-2017-09-27-2]\]. Internet addiction is likely to become a serious public health issue in the near future, considering that even young children use the internet these days and it can easily be accessed anytime and anywhere through smartphones. Although internet addiction was not included in the 5th edition of the Diagnostic and Statistical Manual of Mental Disorders \[[@b35-pi-2017-09-27-2]\], internet gaming disorder was included in the section 'condition for further study'. In this situation, a systematically developed and validated screening tool could assist in identifying current status, preventative intervention, and countermeasures.
METHODS
=======
Scale development
-----------------
A preliminary questionnaire was compiled through a comprehensive review of published research papers about internet addiction, interview material with addiction center visitors, and diagnostic criteria for both pathological gambling and substance use disorders. Next, an expert group consisting of a psychiatrist, psychologist, social worker, and sociologist determined 36 preliminary items after a series of discussions to refine the contexts. For the questions, 50 members of an addiction society rated the adequacy and importance of each item using a 5-point (1--5) Likert scale.
The average score rated by 50 experts was 3.94 for adequacy and 3.90 for importance. Based on the results of the assessment, an item was removed if any of the adequacy or importance scores were below 3.6 (items 7, 12, 18, 25, 28, and 33). Next, if adequacy or importance were below the mean, the expert panel discussed whether the item needed to be included in the final questionnaire considering its clinical value. Items 1 and 4 were deleted because they were similar to items 2 and 5, respectively. Items 3 and 8 were included because they were considered to have high validity for preoccupation to internet. Items 6, 9, 13, and 15 were included because these were the only items representing academic neglect, tolerance, emotional relief, and attempt to conceal a problem respectively. In addition, items 27, 29, and 32 were included because these questions suggested interpersonal problems caused by excessive internet usage. Item 35, which indicates willingness to change, was considered to have predictive value for therapeutic prognosis. Item 17 was included because it reflects evaluation from others. Finally, 28 questions were confirmed ([Table 1](#t1-pi-2017-09-27-2){ref-type="table"}).
Participants
------------
The subjects were recruited through six 'I-will' centers located in Seoul, South Korea. The centers are operated for the management and intervention for internet, game, or smartphone addiction under the management of Seoul city. Most visitors are teenagers who attended at the request of family members or schools, although some visit voluntarily. The visitors of the center were given the full explanation of the study, and 158 people voluntarily signed the consent form. They included 88 men and 70 women, with an average age of 22.12 (SD=7.56) years. For all participants, the psychologist at each center collected information on the reason for referral or voluntary visit, and explored internet usage habits in a face-to-face interview. Based on the interview material, a psychiatrist and two clinical psychologists evaluated whether a person could be classified as an problematic internet user or not, through overall assessment of internet usage time, dependence, tolerance, withdrawal symptoms, perceived controllability, subjective discomfort, and functional impairment. The Samsung Medical Center Institutional Review Board approved the study, and all participants were paid \$30 for their participation (IRB No. 2014-08-114-016).
Measures
--------
### Internet addiction screening-Questionnaire (IOS-Q)
Through the aforementioned procedure, the IOS-Q was finalized. The questionnaire was designed to help clinicians to explore the internet usage habits of respondents and identify at-risk internet addiction individuals. Each respondent was asked to evaluate the frequency of the statements on a four-point scale (Not at all, Sometimes, Often, or Always).
### Young's internet addiction scale (IAT)
To verify the concurrent validity of the SOS-Q, the IAT was used. The IAT is the most frequently used measure of internet addiction. We used a Korean translated version contained in a mental health screening survey scale book published by Seoul Child, Adolescent Mental Health Center \[[@b36-pi-2017-09-27-2]\]. Subjects rated their internet usage habits on a 5-point Likert scale. The psychometric properties of the IAT have been verified in South Korea \[[@b32-pi-2017-09-27-2]\]. Cronbach's alpha for this study sample was 0.95.
### Korean Scale for internet addiction (K-Scale)
We used the short-form, 15-item K-Scale for children, adolescents, and adults \[[@b33-pi-2017-09-27-2]\]. The first K-Scale was developed by the Korea National Information Society Agency and Seoul National University in 2002 to measure internet addiction in adolescents \[[@b37-pi-2017-09-27-2]\]. It is composed of 40 questions revised from Young's 20 item Scale. The K-Scale for adults was published in 2005 \[[@b38-pi-2017-09-27-2]\]. The measure requires individuals to rate on a 4-point scale about their internet usage habits. Based on a total score, users are classified as high-risk, potential risk, or general user groups. Cronbach's alpha coefficient was 0.78 in this study.
### Smartphone scale for smartphone addiction (S-Scale)
Based on the K-Scale, the S-Scale was developed to measure smartphone addiction. The term 'internet' was changed to 'smartphones'. The S-Scale consists of 15 items, on a 4-point scale like the K-Scale. Respondents are categorized according to the total score. It has adequate reliability and validity \[[@b39-pi-2017-09-27-2]\]. Cronbach's alpha coefficient was 0.81 in our sample.
### Statistical analyses
The Statistical Package for Social Sciences version 21 (SPSS ver. 21; IBM Corp., Armonk, NY, USA) was used for analyzing group statistics, internal consistency, inter-item correlation, test-retest reliability, and concurrent validity. T-test and chi-squared test were used to compare the results between addicts and non-addicts. To examine the factor structure of IOS-Q, we conducted Exploratory Factor Analysis (EFA) using Comprehensive Exploratory Factor Analysis (CEFA) Version 3.04 \[[@b40-pi-2017-09-27-2]\]. The factor analysis method used maximum likelihood extraction with oblique direct Quartimin rotation. Oblique rotation was used because correlations between factors were expected. Also, diagnostic ability of the IOS-Q was assessed by investigating sensitivity, specificity of each cut-off score using Receiver Operating Curve (ROC) analysis. If the value of Area Under the Curve (AUC) is \<0.5, it is considered random guess, 0.7--0.8 is acceptable, and ≥0.8 is considered excellent \[[@b41-pi-2017-09-27-2]\].
RESULTS
=======
Demographics
------------
[Table 2](#t2-pi-2017-09-27-2){ref-type="table"} presents the basic demographic information and internet usage habits by the addiction/non-addiction groups. Differences in age, gender, and internet usage habits among addicts and non-addicts were significant. In addition, the internet addiction group highly assessed the internet's influence on academic/occupational and household function. For internet addicts, the main purpose of the internet usage was social media, internet games, web-surfing, watching broadcasting, comics, shopping, and blogging in descending order of frequency. On the other hand, the non-addiction group used internet mainly for web-surfing, social networking, watching broadcasting events, and to play internet games.
Construct validity
------------------
The significance of Bartlett's test of sphericity \[χ^2^ (df=378)=2453.23, p\<0.001\] and the Kaiser-Meyer-Olkin (KMO) results (KMO=0.094) suggested that factor analysis was appropriate for these data. The Kaiser criterion \[[@b42-pi-2017-09-27-2]\], scree test \[[@b43-pi-2017-09-27-2]\], and the root mean square of approximation (RMSEA) \[[@b44-pi-2017-09-27-2]\] was used to determine the appropriate number of factors. The Kaiser criterion, which assesses the number of factors with eigenvalues \>1.0, suggested 6-factors. The drop rate of the scree plot suggested a 5- or 6-factor structure. According to these results, 4- to 6-factor models were examined. The RMSEA of the 4-, 5-, and 6-factor model was 0.080 \[90% confidence interval (CI): 0.070--0.089\], 0.073 (90% CI: 0.063--0.084), and 0.062 (90% CI: 0.050--0.073), respectively, which demonstrated that the 5- or 6-factor model would be reasonable (i.e. RMSEA \<0.08) \[[@b44-pi-2017-09-27-2]\]. Although the RMSEA value of the 6-factor model was greater, the 5-factor structure was selected since 4 items of 6-factor structure showed under-factoring, in which the factor loading was \<0.30.
After accepting the 5-factor solution, each item's factor loading was examined ([Table 3](#t3-pi-2017-09-27-2){ref-type="table"}). In order to clarify the factor structure, items that had factor loading less than 0.3 (item 3) and items that showed cross loading (factor loading \>0.3 to more than 2 factors; items 2, 4, 6, 11, 12, 14, 17, 18, 24, and 26) were deleted. As a result, total of 17 items remained, and factor 3 was removed because there were no items that remained. Items that belonged to each factors are: Factor 1=Items 1, 13, 20, and 23, Factor 2=Items 16 and 25, Factor 4=5, 7, 8, 9, 10, 15, 27, and 28, and Factor 5=19, 21, and 22). Each factor was named loss of control, preoccupation, craving, and neglect of other areas, respectively. All the sub-factor scores were higher in the addiction group compared to the non-addiction group at the p\<0.001 level.
Internal consistency and test-retest reliability
------------------------------------------------
Coefficient alphas and corrected item-total correlations were computed for the IOS-Q. The Cronbach's alpha was 0.91 for total 17 items and the corrected item-total correlations ranged from 0.34 to 0.78. Cronbach's alpha if item deleted were 0.90 for all items. The Cronbach's alpha for sub-factors are: Loss of control=0.79, Preoccupation=0.54, Craving=0.86, and Neglect of other areas: 0.76. The correlations between items of the IOS-Q ranged from 0.07 to 0.71.
The test-retest correlation was calculated with the exception of 30 cases which missed the date of second visit. The test-retest reliability for an average time lapse of 10.29 days was 0.72 (p\<0.001).
Concurrent validity
-------------------
Inter-correlations between IOS-Q and other self-reported measures are presented in [Table 4](#t4-pi-2017-09-27-2){ref-type="table"}. Although correlations between all the measures were significant, the correlation between IOS-Q and the internet addiction related measures, IAT and K-Scale, were higher compared to correlations between the IOS-Q and S-Scale. Also, all the sub-factors of IOS-Q showed higher correlations with IAT and K-Scale compared to S-Scale. These results support convergent and discriminant validity of the IOS-Q.
ROC analysis
------------
Two clinical psychologists and one psychiatrist independently reviewed the interview material of all participants of I-will center and decided normal or problematic user. The diagnostic concordance rate was 94.94% among raters, and the discrepancy occurred when functional impairment, self-controllability, or excessive use was unclear. In these cases, the individual was assigned a group in which two or more experts agreed after case discussion. As a result, 28 out of 158 people were classified as internet addicts. [Figure 1](#f1-pi-2017-09-27-2){ref-type="fig"} shows the results of the ROC analysis of the final 17 items. The AUC value was 0.87 (95% CI: 0.80--0.95), and the optimal cut-off score was 25.5 considering the sensitivity and the specificity determined by Youden's J statistic. The point determined by the Youden's index represents the furthest point from the diagonal line, where the sum of the sensitivity and specificity can be maximized \[[@b45-pi-2017-09-27-2]\]. At the cut-off point of 25.5, the sensitivity and the specificity was 0.93 and 0.76, respectively. Results of ROC analysis of sub-factors are presented in [Table 5](#t5-pi-2017-09-27-2){ref-type="table"}.
DISCUSSION
==========
This study identified the developmental process of the IOS-Q and explored the psychological properties of the scale. Internal reliability was superior, except for one sub-factor (preoccupation). Inter-item correlation and test-retest reliability were adequate. Moreover, the correlations between the IOS-Q and the internet-related measures supported concurrent validity of the scale. EFA revealed a 5-factor structure, with all sub-factor scores being higher in the addiction group compared to the non-addiction group, supporting content validity. ROC analysis revealed that addiction and non-addiction group can be effectively distinguished based on a cut-off score of 25.5. Regarding the result of ROC analysis, it should be noted that the cut-off score and the total mean score were almost similar. Given that it is usually considered abnormal from 1.5 standard deviation above the mean based on normal distribution, it is likely that the cut-off score of this study was somewhat lower. This could be due the under-reporting of problems by the affected individuals. Since internet addicts often underreport their symptoms, this cut-off score is plausible for screening purposes, assuring adequate sensitivity of the instrument.
The average daily internet usage time was \<4 hours even in those classified as internet addicted. Considering that initial study of internet addiction defined internet addiction as 'use of internet more than 38 hours per week' \[[@b19-pi-2017-09-27-2]\], this seems less than expected. Similar results have been reported in several previous studies. For example, in a review study by Dowling and Quirk \[[@b46-pi-2017-09-27-2]\], there were no significant differences in internet usage time and psychological distress between internet addicts and non-addicts. The authors discussed the possibility of limited internet use due to management of parents or schools. Although the influence of the internet to academic/occupational and household function was highly estimated by internet addiction group compared to non-addiction group in our sample, it was normal range on a Likert scale of 1 to 10. Internet over-users might have thought that there is no problem with that amount of usage time. Also, there were cases in the sample complaining of craving symptoms even though actual usage time was average, or who cannot remember their own usage time exactly. Given this, the intent of the IOS-Q to identify problematic internet users through various internet usage habits seems reasonable.
There are several limitations in this study. Although environmental factors such as social economic status might be related to internet addiction \[[@b18-pi-2017-09-27-2],[@b21-pi-2017-09-27-2]\], we could not examine these variables since most of our samples were single students. In addition, the question whether internet addiction is an isolated disease entity or related to other comorbid mental disorders cannot be answered because we did not carry out a structured clinical interview. There are other limitations related to methodological issues. First of all, it is likely that the respondents were disingenuous or insincere because all the measures were self-report. Also, due to the fact that the participants were recruited through the I-will center, there is a possibility that the sample have been biased. Finally, given the number of items included in the factor analysis (n=28), a sample of 158 participants might not be able to provide sufficient power to support the analysis. The 5-factor structure revealed in this study should be re-examined with confirmatory factor analysis in the future study.
Despite these limitations, the IOS-Q underwent a systematical development and standardization process: i.e. item rating by 50 addiction experts, and group classification and case discussion by independent clinicians. The overall psychological properties of the questionnaire turned out to be favorable. Research on internet addiction is still in development, and this instrument would be worthwhile to identify the status and characteristics of the phenomenon. However, in the absence of consensus on the definition of internet addiction, it should be noted that we cannot solely rely on this instrument for screening or diagnosis purposes. The IOS-Q should serve as the basis for in-depth discussions.
This study was supported by a grant of the Korean Mental Health Technology R&D Project, Ministry of Health & Welfare, Republic of Korea (No. HM14C2567; PI, HJJ). This work was partly supported by Institute for Information & communications Technology Promotion (IITP) grant funded by the Korea government (MSIP) (No. B0132-15-1003, the development of skin adhesive patches for the monitoring and prediction of mental disorders), and by the Original Technology Research Program for Brain Science through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (No. NRF-2016M3C7A1947307; PI HJJ), and the Bio & Medical Technology Development Program of the NRF funded by the Korean government, MSIP (No. NRF-2017M3A9F1027323; PI HJJ).
{#f1-pi-2017-09-27-2}
######
Preliminary item development, and adequacy and importance rated by 50 addiction experts
Pilot item Included item Content A I Pilot item Included item Content A I
------------ --------------- -------------------------------------------------------------------------------------------------------------------------------------------- ----------- ----------- ------------ --------------- ---------------------------------------------------------------------------------------------------------------- ----------- -----------
1 I keep thinking about using the internet. 3.96 3.92 19 14 I skip out of school or work to go somewhere I can use the internet. 4.33 4.40
2 1 I often think about the internet even while doing other work. 4.02 3.96 20 15 I don\'t think I can reduce my internet usage without help from others. 3.98 3.96
3 2 I exceedingly wait to access the internet again. 3.94 3.85 21 16 Despite I am sick, I continue the internet usage (e.g. lack of sleep, eye fatigue, hand or neck pain, etc.) 4.15 4.21
4 Even though it is overtime, I don\'t sleep and keep surf the internet. 3.88 3.90 22 17 I spend most of the allowance or salary for the internet usage. 4.02 3.98
5 3 I stay up all night using internet. 4.23 4.21 23 18 I am late for the school, work, appointment, etc. due to the internet usage. 4.46 4.48
6 7 4 I am tired and sleepy during class or work due to the internet usage. If I have no replies, no recommendations, or comments, I am nervous. 3.83 3.35 3.88 3.29 24 25 19 I skip meals or eat while using the internet. Even if I want to go to the bathroom, I keep using the internet. 4.02 3.46 4.00 3.38
8 5 I check the internet excessively by habit. 3.71 3.63 26 20 I use the internet instead of doing things I need to do. 4.40 4.33
9 6 I need to use the internet for a longer period of time to become as satisfied as before. 3.92 3.92 27 21 I like using the internet more than socializing with friends. 3.96 3.85
10 7 I become irritated or angry without the internet. 4.23 4.23 28 I feel that I am a better person in the internet space than in real life. 3.58 3.54
11 8 I become anxious or nervous without the internet. 4.27 4.29 29 22 I like using the internet more than spending time with family. 3.63 3.52
12 If I can\'t use the internet, I feel alienated from people around me. 3.60 3.58 30 23 I quarrel with family due to the internet usage. 4.10 4.06
13 9 When I am in bad mood, internet use makes me feel better. 3.81 3.83 31 24 I am not interested in anything except the internet usage. 3.94 4.00
14 10 I tried to reduce the internet usage, but it is difficult. 4.33 4.40 32 25 I became distant from friends and colleagues since the internet usage. 3.65 3.63
15 11 I underreport the amount of time I spend on the internet. 3.96 3.88 33 I do internet because it is difficult to hang out with friends or co-workers. 3.40 3.27
16 12 When I use the internet, I lose track of how much time has passed. 4.08 4.06 34 26 I think that I am addicted to the internet. 4.02 3.98
17 13 People around me point out that I spend a lot of time on the internet. 3.92 3.75 35 27 I want to change my current internet usage habits. 3.94 3.88
18 My family keeps me from using the internet. 3.56 3.40 36 28 I think I use the internet excessively. 4.04 4.00
A: Adequacy, I: Importance
######
Descriptive statistics of basic characteristics and internet usage habits of participants
Addicts (N=28) Non-addicts (N=130) p-value Total (N=158)
---------------------------------- ---------------- --------------------- --------- ---------------
Age (M, SD) 17.61 (3.81) 23.10 (7.82) \<0.001 22.12 (7.56)
Female, N (%) 18 (64.29) 52 (40.00) 0.019 70 (44.30)
Internet habits, N (%)
Average daily use time 3.68 (2.20) 2.04 (1.47) 0.001 2.33 (1.73)
Maximum usage time per use 5.29 (4.61) 3.21 (3.37) 0.031 3.58 (3.69)
Academic/occupational influence 5.14 (2.45) 3.42 (2.33) 0.001 3.73 (2.44)
Household influence 4.14 (2.65) 2.95 (2.27) 0.016 3.16 (2.38)
Interpersonal influence 2.96 (2.66) 2.92 (2.31) 0.921 2.92 (2.37)
Main usage, N (%) (double count)
Internet games 15 (19.74) 36 (15.25) 51 (16.35)
Watching broadcasting or porn 14 (18.42) 40 (16.95) 54 (17.31)
Social medias 16 (21.05) 46 (19.49) 62 (19.87)
Web surfing 14 (18.42) 56 (23.73) 70 (22.44)
Blogging or web cafes 3 (3.95) 20 (8.47) 23 (7.37)
Web-cartoon 10 (13.16) 22 (9.32) 32 (10.26)
Shopping 4 (5.26) 16 (6.78) 20 (6.41)
Data were expressed as numbers (percent), mean±standard deviation. p value was derived from independent two sample t-test
######
Quartimin rotated factor loadings of preliminary 28 items
Item Factor 1 Factor 2 Factor 3 Factor 4 Factor 5
------ -------------------------------------------------------- -------------------------------------------------------- -------------------------------------------------------- -------------------------------------------------------- --------------------------------------------------------
1 0.46^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.15 0.09 0.23 0.00
2 0.41^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.02 0.05 0.46^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.06
3 0.11 0.06 -0.17 0.00 0.02
4 0.34^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.48^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.29 0.05 0.02
5 0.18 0.13 -0.29 0.35^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.22
6 0.30^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.22 -0.17 0.37^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.09
7 -0.02 -0.05 0.13 0.71^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.06
8 0.02 0.03 0.05 0.68^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.09
9 -0.06 0.13 0.08 0.50^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.10
10 -0.01 0.08 -0.11 0.73^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.15
11 0.44^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.05 0.09 0.50^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.17
12 -0.10 0.34^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.01 0.47^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.18
13 0.68^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.01 -0.02 0.03 0.17
14 0.07 0.09 0.52^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.40^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.10
15 -0.01 0.09 0.13 0.52^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.23
16 0.18 0.36^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.09 0.05 0.23
17 0.31^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.12 0.56^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.10 0.22
18 0.42^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.35^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.14 0.00 -0.01
19 0.19 0.06 0.02 0.15 0.38^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^
20 0.30^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.23 -0.13 0.17 0.27
21 -0.02 0.13 -0.03 0.01 0.79^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^
22 0.05 -0.04 0.11 0.15 0.74^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^
23 0.61^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.04 0.18 0.05 0.13
24 0.40^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.08 0.21 0.00 0.32^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^
25 -0.10 0.71^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.14 0.02 0.05
26 0.31^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ -0.18 -0.19 0.44^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.35^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^
27 -0.01 0.07 -0.02 0.70^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.09
28 0.21 -0.02 -0.11 0.61^[\*](#tfn1-pi-2017-09-27-2){ref-type="table-fn"}^ 0.18
factor loadings \>0.3
######
Correlations between other related measures and IOS-Q total and sub-factor scores
IOS-Q IAT K-scale S-scale M SD
------------------------ --------------------------------------------------------- --------------------------------------------------------- --------------------------------------------------------- --------------------------------------------------------- ------- -------
IOS-Q 1 25.54 8.03
IAT 0.745^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 1 36.58 15.94
K-Scale 0.551^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.562^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 1 28.99 6.56
S-Scale 0.504^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.513^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.522^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 1 29.57 6.84
Loss of control 0.862^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.642^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.411^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.360^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 6.20 2.40
Preoccupation 0.591^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.460^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.320^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.273^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 2.54 0.83
Craving 0.792^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.516^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.489^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.487^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 6.47 2.16
Neglect of other areas 0.750^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.644^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.404^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 0.324^[\*](#tfn2-pi-2017-09-27-2){ref-type="table-fn"}^ 4.06 1.60
all correlations were p\<0.001. IOS-Q: Internet Overuse Screening-Questionnaire, IAT: Young Internet Addiction Test
######
Results of Receiver Operating Curve Analysis of IOS-Q
M (SD) AUC Cut-off score Sensitivity Specificity
------------------------ -------------- -------------- --------------- ------------- ------------- ------
IOS-Q total score 34.21 (8.57) 23.33 (6.24) 0.87 25.5 0.93 0.76
Loss of control 7.38 (2.88) 5.63 (1.90) 0.81 6.5 0.79 0.78
Preoccupation 3.17 (1.20) 2.39 (0.63) 0.67 2.5 0.61 0.65
Craving 7.76 (2.29) 6.13 (2.00) 0.78 6.5 0.71 0.71
Neglect of other areas 5.79 (2.34) 3.63 (1.01) 0.77 4.5 0.70 0.80
IOS-Q: Internet Overuse Screening-Questionnaire, AUC: Area Under the Curve
| {
"pile_set_name": "PubMed Central"
} |
Aims and Scope
==============
*Brain and Behavior* is a peer-reviewed, interdisciplinary journal, providing rapid publication of high-quality research across neurology, neuroscience, psychology and psychiatry. The journal will publish quality research reports that enhance understanding of the brain and behavior.
*Brain and Behavior* will give rapid consideration to papers in all areas of clinical and basic research. Molecular, cellular, systems and population-level research in humans and animal models are welcome. The journal will consider empirical and theoretical submissions in areas including but not limited to mechanisms and treatment of all neurological diseases and psychiatric disorders, behavioral and cognitive science, neuronal and glial cell biology, neurosurgery, neurophysiology, neuropharmacology, neuroimmunology, neuropathology, computational neuroscience, functional and structural neuroimaging, neurogenetics and psychiatric genetics, child and adolescent psychiatry, psychiatry of affective and cognitive disorders, epidemiology, emerging technologies and new research methods, translational research, neuropsychology, clinical psychology, and developmental, cognitive, and social psychology.
*Brain and Behavior* features original research articles, reviews, methods papers, editorials, and commentaries. Original research papers must report well-conducted research with conclusions supported by the data presented in the paper.
*Brain and Behavior* publishes papers submitted directly to the journal and those referred from a select group of prestigious journals published by Wiley-Blackwell. *Brain and Behavior* is a Wiley Open Access journal, one of a new series of peer-reviewed titles publishing quality research with speed and effi ciency. For further information visit the Wiley Open Access website at <http://www.wileyopenaccess.com>.
Open Access and Copyright
=========================
All articles published by *Brain and Behavior* are fully open access: immediately freely available to read, download and share. All articles accepted from 14 August 2012 are published under the terms of the Creative Commons Attribution License. All articles accepted before this date were published under a Creative Commons Attribution Non-Commercial License. The Creative Commons Attribution License permits use, distribution and reproduction in any medium, provided the original work is properly cited and allows the commercial use of published articles.
Copyright on any research article in a journal published by *Brain and Behavior* is retained by the author(s). Authors grant Wiley a license to publish the article and identify itself as the original publisher. Authors also grant any third party the right to use the article freely as long as its integrity is maintained and its original authors, citation details and publisher are identifi ed. Further information about open access license and copyright can be found at <http://www.wileyopenaccess.com/details/content/12f25db4c87/Copyright--License.html>.
Purchasing Print Reprints
=========================
Print reprints of Wiley Open Access articles can be purchased from <[email protected]>.
Disclaimer
==========
The Publisher and Editors cannot be held responsible for errors or any consequences arising from the use of information contained in this journal; the views and opinions expressed do not necessarily refl ect those of the Publisher and Editors, neither does the publication of advertisements constitute any endorsements by the Publisher and Editors of the products advertised.
Wiley Open Access articles posted to repositories or websites are without warranty from Wiley of any kind, either express or implied, including, but not limited to, warranties of merchantability, fi tness for a particular purpose, or non-infringement. To the fullest extent permitted by law Wiley disclaims all liability for any loss or damage arising out of, or in connection with, the use of or inability to use the content.
Editors-in-Chief
================
Andrei V. Alexandrov, MD
Maryann E. Martone, PhD
Address correspondence to the Editorial Offi ce: <[email protected]>
Editorial Board
===============
Associate Editors:
------------------
Dona Lee Wong *Harvard Medical School,* *McLean Hospital, USA*
Editorial Board Members:
------------------------
Hubert Amrein *Texas A&M, USA*
Isabelle Aubert *University of Toronto, Canada*
David Belin *University of Poitiers, France*
Sandra Black *University of Toronto, Canada*
Sandra Bosacki *Brock University, Canada*
Robert Bryan *Baylor College of Medicine, USA*
Hans Crombag *University of Sussex, United Kingdom*
Mirella Dapretto *University of California, Los Angeles, USA*
David Diamond *University of South Florida, USA*
Scott Grafton *University of California, Santa Barbara, USA*
Ilona Grunwald-Kadow *Max Planck Institute of Neurobiology, Germany*
Martin Hagger *Curtin University, Australia* Jozsef Haller *Hungarian Academy of Sciences, Hungary*
Ahmad Hariri *Duke University, USA*
Shu-Leong Ho *University of Hong Kong, HongKong*
Andrea Huber Brösamle *Helmholtz Center Munich, Germany*
Kazuhiro Ikenaka *National Institute for Physiological Sciences, Japan*
Stephen Jackson *University of Nottingham, UK and Korea University, South Korea*
Takeshi Kaneko *Kyoto University, Japan*
Kenneth Kosik *University of California, Santa Barbara, USA*
Klas Kullander *Uppsala University, Sweden*
Jonathan Lee *University of Birmingham, United Kingdom*
Edward Levin *Duke University, USA*
Bai Lu *Tsinghua University and GlaxoSmithKline, China*
Randolph Marshall *Columbia University Medical Center, USA*
Cheryl McCormick *Brock University, Canada*
Douglas Mennin *Hunter College, USA*
Martin M. Mortazavi *University of Alabama, Birmingham, USA*
Declan Murphy *Kings College London, United Kingdom*
Heinz Reichmann *University of Dresden, Germany*
Richard Rende *Brown University, USA*
Giacomo Rizzolatti *University of Parma, Italy*
Fred Sabb *University of California, Los Angeles, USA*
Veronika Skvortsova *Russian State Medical University, Russian Federation*
Kai-Christian Sonntag *Harvard Medical School, McLean Hospital, USA*
Takashi Suzuki *Tokyo Institute of Technology, Japan*
Wai Kwong Tang *Chinese University of Hong Kong, Hong Kong*
Tanya N. Turan *Medical University of South Carolina, USA*
Hartmut Wekerle *Max Planck Institute of Neurobiology, Germany*
Jianhong Zhu *Fudan University, China*
Consulting Editors
------------------
Alexander Bystritsky *University of California, Los Angeles, USA*
Wesley Thompson *University of California, San Diego, USA*
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#S1}
============
A growing body of research has provided evidence for hope as a psychological strength, particularly for children and adolescents confronted with adverse conditions ([@B44]; [@B35]). Empirical research oeuvres has established that hope is associated with subjective well-being, life satisfaction and overall quality of life among children and adolescents (e.g., [@B12]; [@B36]; [@B26]; [@B30]; [@B28]; [@B33]). Within the social sciences, hope is conceptualized as a cognitive-motivational construct focusing on goal-directed behavior, the "future self," and is encompassed in a typology of concepts, inclusive of coping, faith, resilience, and empowerment ([@B18]).
Snyder's theory on hope, developed across a period of more than 30-years, is the seminal theory in the field ([@B35]). In describing Hope theory, [@B38] delineated a cognitive model consisting of the "trilogy" of goals, pathways, and agency. "Goals" are the foundation of the theory and represents the cognitive component that grounds the theory. The theory works from the premise that people are goal-directed, with hope conceptualized as people's perceptions regarding their capacities to (1) formulate clear goals, (2) develop the specific strategies or "workable routes" to reach those goals (pathways thinking), and (3) "self-related beliefs about initiating and sustaining movement toward those goals" (agentic thinking; [@B40], [@B39], p. 401).
Both pathways and agency are fundamental in "hopeful thinking" and encompass relatively stable subjective evaluations of goal-oriented competencies ([@B37], [@B38]). The two components are positively correlated, additive, iterative, and reciprocal; however, they are not synonymous, and neither define hope separately ([@B38]). In contrast to emotion-based theories of hope (see [@B11]), [@B38] theoretical supposition of hope is decidedly cognitive and purports that individuals' perceptions of goal pursuits are antecedent to, and drive emotions (see also [@B16]; [@B7], [@B8]). Those with high levels of hope are usually efficacious in their pursuit of goals, and typically, experience increased positive emotions ([@B38]). In contrast, those with low levels of hope face additional challenges in attaining goals given various hindrances to goal attainment, and are likely to experience negative emotions ([@B38]; [@B24]).
In children, the theory portends that goal-directed hopeful thinking develops in the first few years of life and is crucial for the child's development and survival ([@B38]). How children make sense of and appraise their capacity to respond to challenges and barriers are important considerations for hope in children. [@B39] developed the Children's Hope Scale (CHS) to address an identified gap in the literature in evaluating children's hope and the key aspects that contribute to this construct. Diverging from earlier conceptualizations focusing on negative aspects of hope (see [@B21]), the scale emphasizes "positive expectancies." The CHS is a six-item dispositional self-report scale developed for 8 to 16-year olds that evaluates the two key constituents of pathways and agency.
Along with the considerable reflection and engagement concerning conceptual definitions of hope, there has been an increasing investment in empirical studies exploring children's hope using the CHS ([@B39], [@B41]) across various contexts. Within low and middle-income contexts, the work of [@B1] in Turkey; [@B35], and [@B13] in South Africa; [@B14] in Burundi, Indonesia, and Nepal; [@B45] in Malaysia; [@B19] in Serbia; and [@B22] in China, have made substantial contributions to the literature on children's hope. In high-income contexts, the empirical work of [@B7], [@B8] in Australia; [@B10]; [@B12]; [@B9], and [@B43], [@B44] in the United States of America; [@B27], [@B25] in Portugal; [@B32] in Spain; and [@B47] in Singapore, are noteworthy. Taken together, all of the aforementioned studies delineate the CHS as a reliable and valid measure with samples of children and adolescents. These validation studies have revealed two different conceptualizations of the hope model. For example, while a two-factor model, as specified by the initial scale authors, was supported in some validation studies (see e.g., [@B43]; [@B32]), others have found a better fit with a one-factor model (see e.g., [@B2]; [@B9]; [@B35]).
While the focus of early research was on the validation of the CHS in various contexts, more recently the focus has shifted toward group comparisons and measurement invariance testing. For example, [@B22] and [@B35] tested the measurement invariance of the CHS across socio-economic status groups, while [@B14] conducted group comparisons on samples of children in Burundi, Indonesia, and Nepal. Another recent trend in research is the relation of children's hope with other psychological constructs such as quality of life ([@B28]), life satisfaction ([@B30]; [@B33]), and subjective well-being ([@B20]). However, no studies have investigated hope using national representative or population-based samples of children, and standardized scores on the CHS have not been proposed. The goal of the present study is to explore hope in children using a nationally representative sample and to present a standardized mean score of the CHS.
Aim of the Study {#S2}
================
Using data from Wave 3 of the Children's Worlds International Survey on Children's Well-Being, the study aimed to explore hope amongst a random population-based sample of children in South Africa. The study further aimed to explore children's level of hope across the nine provincial regions of South Africa.
Method {#S3}
======
Research Design {#S3.SS1}
---------------
The study forms part of Wave 3 of the Children's Worlds International Survey on Children's Well-Being^[1](#footnote1){ref-type="fn"}^. Conducted across 35 countries, the survey is the largest multinational study to assess children's subjective perceptions of their well-being across different contexts and domains. The study in South Africa employed a cross-sectional survey design and used a nationally representative proportionate stratified random sample of 10- and 12-year-olds, across the country's nine provincial regions. A central management committee consisting of a range of experts in comparative international surveys was tasked with overseeing the sampling protocol, instrument development and data analytic plan of each participating country. It has been found that the central management of multinational collaborative studies leads to improved quality and integrity of the data ([@B4]).
Participants and Sampling {#S3.SS2}
-------------------------
The study included a nationally representative proportionate stratified random sample of children selected from public primary schools across the nine provincial regions of South Africa, namely: Western Cape; Eastern Cape; Northern Cape; North West; Mpumalanga; Gauteng; KwaZulu-Natal; Free State; and Limpopo. The stratification for the study was based on school grade (4 or 6), geographical context (urban or rural), and provincial region.
Instrumentation {#S3.SS3}
---------------
### Children's Hope Scale {#S3.SS3.SSS1}
The CHS developed by [@B39], measures goal-directed and hopeful thinking in children and adolescents between the ages of 8--16 years old. The scale consists of six items, with three questions evaluating "pathways thinking" (items 2, 4, and 6) and three evaluating "agency thinking" (items 1, 3, and 5). Response options are on a six-point verbal response format ranging from "None of the time" = 1 to "All of the time" = 6. A composite score is calculated by summing the raw scores on each item. [@B2] suggested mean cut-scores of low (\<3.0), medium (3.0--4.67), and high (\>4.67) hope categories, where higher scores are indicative of high levels of hope and lower scores indicate low levels of hope.
The CHS has shown good psychometric properties across a range of contexts (see the reliability analysis conducted by [@B15]). Studies in South Africa conducted by [@B35] and [@B13] have reported Cronbach's alpha coefficients of 0.82 and 0.73, respectively.
Data Analysis {#S3.SS4}
-------------
Data were analyzed by means of confirmatory factor analysis (CFA; maximum likelihood estimation) using the Analysis of Moment Structures (AMOS, version 25) software. Following recommendations by [@B17], the Comparative Fit Index (CFI), Root Mean Squared Error of Approximation (RMSEA), and Standardized Root Mean Residual (SRMR) were used as fit indexes. Scores higher than 0.950 for the CFI and scores below 0.05 for the RMSEA and SRMR were regarded as a good fit. Improvement of model fit was achieved by the consideration of modification indices, standardized residual covariances, and the expected parameter change ([@B34]). Measurement invariance was used to compare the results across provincial regions and was tested using multi-group confirmatory factor analysis (MGCFA) to ensure meaningful, reliable, and unambiguous group comparisons ([@B29]; [@B31]). This process comprised three sequential steps wherein restrictive constraints were incrementally applied. In the first step, configural invariance was tested in a multi-group model by allowing the loadings and intercepts to be freely estimated; this represents the baseline model against which other models are tested. In the second step, metric invariance was tested by constraining the factor loadings. Finally, scalar invariance was tested by constraining the factor loadings and intercepts. Each subsequent constrained model was regarded as tenable if the model fit did not worsen by more than 0.01 on the CFI ([@B6]) and by 0.015 on the RMSEA and SRMR ([@B5]). The tenability of scalar measurement invariance suggests that meaningful comparisons across groups (provincial regions) can be conducted by correlations, regression coefficients and mean scores. A means comparative analysis across the provincial regions was achieved through a one-way Analysis of Variance (ANOVA) using Stata (version 14).
Procedure and Ethics {#S3.SS5}
--------------------
The South African study obtained ethics clearance from the Humanities and Social Sciences Research Ethics Committee of the University of the Western Cape, and the nine provincial education departments. Potential participants at each participating school attended a briefing session with the research team who explained the nature and details of the study. The research team also discussed the participants' rights, the ethics principles of informed consent, confidentiality, the right to withdraw, privacy and the scientific use of the data. The final step in obtaining consent involved providing information sheets and consent forms to participants and seeking active consent from the children and their parents. The data collection process followed a researcher-administered protocol wherein the research team, led by the principal investigators, administered the questionnaire in a group setting to the participants by reading each question and explaining the response options. This took place during an administration period at the beginning of the school day with an average administration time of 30 minutes.
Data Analytic Plan {#S3.SS6}
------------------
The South African research team captured the data and submitted the database to the aforementioned central management committee, which oversaw the data management process. The initial database consisted of 7428 participants. The cleaning and depuration of the dataset followed a range of processes that included deleting cases with more than a third of missing data, the assessment of systematic response sets and attending to clustering due to survey design effects. The final dataset was weighted based on the proportion of children per provincial region. For the current study, analysis revealed missing items to be "missing completely at random." Cases with two or less missing values on the CHS were substituted by regression imputation, as per the recommendations of [@B3]. The final dataset consisted of 7067 participants (males = 45.6%, girls = 54.4%) between the ages of 9 to 12-years (*M*~age~ = 10.79, *SD* = 1.28), in Grades 4 (*n* = 3383), and 6 (*n* = 3684), attending 61 primary schools across the nine provincial regions of South Africa.
Results {#S4}
=======
Skewness of the scores on the items of the CHS ranged from −1.222 to −0.916, with kurtosis ranging from −0.392 to 0.442. Given that these scores were outside the range of acceptable deviation, the bootstrap method (500 samples) in AMOS 25 was used as a resolution. A reliability analysis showed a Cronbach alpha of 0.80.
Confirmatory Factor Analysis {#S4.SS1}
----------------------------
In line with the original scale authors' supposition, a two-factor model was initially tested. However, while this model presented with an adequate fit, it showed an unacceptably high correlation between the latent constructs (0.92). This suggests that the two constructs are indistinguishable and would likely lead to convergent validity issues. Thereafter, a single factor model was tested; this model did not meet the criteria for an adequate fit. However, through the consideration of the modification indices, an error covariance was included between Item 1 (I think I am doing well) and Item 3 (I am doing just as well as other children my age). This resulted in an excellent fit (see Model 3 in [Table 1](#T1){ref-type="table"} and [Figure 1](#F1){ref-type="fig"}). Standardized regression weights ranged from 0.59 to 0.68 and were all significant at the 0.001 level (see [Figure 1](#F1){ref-type="fig"}).
######
Fit indexes for the confirmatory factor models.
Model Bootstrap, ML, 95% Confidence Intervals, Resamples = 500 Chi-Square *df* *p*-value CFI RMSEA SRMR
---------------------------------------------------------------- ------------ ------ ----------- ------- ---------------------- --------
1\. Initial two-factor model 123.566 8 0.000 0.989 0.045 (0.038--0.052) 0.0182
2\. Initial one-factor model 200.387 9 0.000 0.982 0.055 (0.048--.062) 0.0223
3\. Modified one-factor model with one error covariance 72.776 8 0.000 0.994 0.034 (0.027--.041) 0.0138
4\. Configural one-factor model (across provincial region) 186.548 72 0.000 0.989 0.015 (0.012--.018) 0.0185
5\. Metric one-factor model (across provincial region) 295.001 112 0.000 0.986 0.015 (0.013--.017) 0.0263
6\. Scalar one-factor model (across provincial region) 377.232 152 0.000 0.979 0.014 (0.013--.016) 0.0363
{#F1}
Multi-Group Confirmatory Factor Analysis {#S4.SS2}
----------------------------------------
As previously mentioned, measurement invariance across provincial regions was tested using MGCFA through a sequential process of applying increasingly restrictive constraints. Given that each subsequent constrained model did not worsen by more than 0.01 on the CFI, nor by 0.015 on the RMSEA and SRMR ([@B6]; [@B5]) configural, metric, and scalar measurement invariance was attained (see Models 4--6). This means that across provincial regions, children's scores on the hope scale can be compared by correlations, regression coefficients, and means. The standardized regression weights (constrained loadings and intercepts) which represents the scalar measurement model, are presented in [Table 2](#T2){ref-type="table"}.
######
Standardized regression weights: (Provincial regions -- constrained loadings and intercepts).
Parameter Eastern Cape North West Western Cape
-------------------------------------- -------------- ------------------- ---------------- ------------------- -------------- ----------- ----------- -------------- ----------- ----------- -------
Hope doing well \<-- Hope 0.616 0.579 0.656 0.565 0.516 0.615 0.571 0.524 0.613
Hope get things important \<-- Hope 0.704 0.664 0.742 0.651 0.602 0.702 0.674 0.625 0.722
Hope doing as well as other children \<-- Hope 0.651 0.610 0.690 0.598 0.548 0.649 0.641 0.598 0.680
Hope solve problems \<-- Hope 0.645 0.603 0.682 0.577 0.528 0.627 0.662 0.615 0.708
Hope past will help in future \<-- Hope 0.625 0.585 0.660 0.550 0.504 0.606 0.571 0.522 0.616
Hope others quit i can solve problem \<-- Hope 0.616 0.585 0.646 0.605 0.551 0.659 0.611 0.564 0.657
**Northern Cape** **Free State** **Mpumalanga**
**Estimate** **Lower** **Upper** **Estimate** **Lower** **Upper** **Estimate** **Lower** **Upper**
Hope doing well \<-- Hope 0.545 0.477 0.612 0.552 0.492 0.612 0.599 0.553 0.658
Hope get things important \<-- Hope 0.704 0.620 0.782 0.684 0.621 0.752 0.636 0.575 0.691
Hope doing as well as other children \<-- Hope 0.701 0.626 0.775 0.634 0.574 0.698 0.632 0.576 0.686
Hope solve problems \<-- Hope 0.699 0.625 0.781 0.644 0.580 0.713 0.677 0.627 0.726
Hope past will help in future \<-- Hope 0.582 0.513 0.657 0.524 0.462 0.586 0.577 0.519 0.630
Hope others quit i can solve problem \<-- Hope 0.616 0.539 0.689 0.619 0.562 0.683 0.594 0.541 0.645
**Limpopo** **Gauteng** **KwaZulu Natal**
**Estimate** **Lower** **Upper** **Estimate** **Lower** **Upper** **Estimate** **Lower** **Upper**
Hope doing well \<-- Hope 0.608 0.574 0.644 0.622 0.585 0.656 0.565 0.535 0.600
Hope get things important \<-- Hope 0.693 0.653 0.730 0.689 0.650 724 0.648 0.616 0.680
Hope doing as well as other children \<-- Hope 0.631 0.590 0.671 0.660 0.622 0.694 0.604 0.572 0.633
Hope solve problems \<-- Hope 0.676 0.635 0.715 0.691 0.651 0.724 0.633 0.601 0.666
Hope past will help in future \<-- Hope 0.623 0.587 0.659 0.609 0.567 0.652 0.561 0.530 0.590
Hope others quit i can solve problem \<-- Hope 0.631 0.594 0.668 0.653 0.614 0.691 0.581 0.548 0.613
All values are significant at \<.001.
Comparisons Across Mean Scores {#S4.SS3}
------------------------------
The mean score of the pooled sample using the weighted data was *M* = 4.781 (*SD* = 1.082). Given the tenability of scalar measurement invariance across provincial regions, a comparative means analysis was apposite. A one-way ANOVA demonstrated significant mean differences between the provincial regions \[*F* = 22.981; *df* = 8; *p* \< 0.001; η^2^ = 0.025; and 95% CI = (0.0178, 0.0320)\]. Mean scores ranged from 4.511 (*SD* = 1.163) for the Northern Cape to 4.982. (*SD* = 0.974) for the Western Cape (see [Table 3](#T3){ref-type="table"}). Five provinces (Northern Cape, Eastern Cape, Mpumalanga, KwaZulu Natal, and the Free State) scored below the national mean, while four provinces (North West, Western Cape, Limpopo, and Gauteng) scored above. Using the threshold cut-scores proposed by [@B2], two provinces scored within the category of "medium-hope," while seven provinces scored within the "high-hope" category (see [Table 3](#T3){ref-type="table"}). [Table 3](#T3){ref-type="table"} also presents the percentage of the overall and provincial samples scoring below the overall country mean. For the overall country sample, 41.9% scored below the mean, with percentages ranging from 33.2 % for the North-West Province to 52.7% for the Northern Cape.
######
Mean scores and percentage of the sample scoring below the overall mean.
Provincial Region *N* Mean *SD* \% scoring below overall country Mean
------------------- ------ ------- ------- ---------------------------------------
Eastern Cape 988 4.723 1.077 44.4
North West 627 4.972 0.952 33.2
Western Cape 734 4.982 0.974 35.1
\*Northern Cape 201 4.511 1.163 52.7
Free State 279 4.708 1.123 44.8
Mpumalanga 486 4.702 1.001 48.4
Limpopo 1195 4.958 1.086 34.2
Gauteng 1145 4.841 1.126 48.9
\*Kwa-Zulu Natal 1412 4.520 1.096 51.9
Overall 7067 4.781 1.082 41.9
\*These provinces score within the medium-hope category. All other provinces score within the high-hope category.
Discussion {#S5}
==========
Using data from Wave 3 of the Children's Worlds: International Survey on Children's Well-Being, this study aimed to explore hope amongst a random population-based sample of children in South Africa. The study further aimed to explore children's level of hope across the nine provincial regions. CFA demonstrated an appropriate fit structure for the CHS using a population-based sample of children, while MGCFA confirmed the tenability of scalar measurement invariance.
Confirmatory factor analysis found that while a two-factor model of the CHS presented with an appropriate fit, an unacceptably high correlation was observed between the latent constructs (pathways and agency). This calls into question the distinctiveness of the latent constructs and could result in convergent validity issues. This finding resonates with previous research conducted by [@B35] who found similarly high correlations between the latent constructs in a South African sample. A one-factor model presented with an appropriate fit structure with the addition of one error co-variance. In later commentary, one of the original scale authors acknowledged the supposition of a one-factor model and recommended further exploration of the unidimensional structure ([@B23]). The inclusion of the correlated error was justified based on probable method bias and is likely due to the semantic overlap of the content of the items. Standardized regression weights were all acceptable, ranging from 0.59 to 0.68. The findings ultimately indicate an appropriate fit structure for the overall pooled sample. The mean score for the overall pooled sample (*M* = 4.781, *SD* = 1.082) can be categorized as "high hope" according to the cut-scores proposed by [@B2]. Given the use of a randomized population-based sample, this mean score represents a standardized score and is generalizable to the population of 10- to 12-year-old children attending public schools in South Africa.
Multi-group confirmatory factor analysis was used to test the comparability of the CHS across provincial regions. Given that scalar invariance was tenable, the scores on the CHS across provincial regions can be meaningfully compared by correlations, regression coefficients, and means. The results show significant mean differences between provincial regions, with the Western Cape, North West, Limpopo, and Gauteng scoring above the national mean; while the Northern Cape, Eastern Cape, Mpumalanga, KwaZulu Natal, and the Free State scored below the national mean. Furthermore, the Northern Cape and KwaZulu Natal were the only two provinces that scored in the "medium hope" category, while the other provinces all scored within the "high hope" category. Interestingly, the Western Cape, the province with the highest human development index in South Africa ([@B42]; [@B46]) presented with the highest mean score on the CHS (*M* = 4.982, *SD* = 0.974).
Conclusion {#S6}
==========
This is the first study to use a nationally representative sample to measure hope in children in South Africa. Given the use of a randomized sample, and the attainment of an appropriate fit structure for the CHS, the overall mean score represents a standardized score of 10- to 12-year-old children attending public schools in South Africa. The results from the study suggest that children in South Africa present with high levels of hope. This finding is important given the often-maligned social conditions and constrained socio-economic contextual realities associated with growing up in South Africa. With the construct of hope being closely related to subjective well-being and quality of life ([@B33]) the study highlights the need for further research to understand the factors related to hope. To this end, longitudinal research could provide a more complete understanding of the mechanisms through which these factors function. Future research should also endeavor to explore the measurement invariance of hope across other groups or cohorts of the child population. Finally, given the role of hope as causal to emotions, self-esteem and other psychological constructs and behaviors, social service and educational practitioners should focus their efforts on developing goal-directed thinking in children.
Data Availability Statement {#S7}
===========================
The datasets generated for this study are available on request to the corresponding author.
Ethics Statement {#S8}
================
The studies involving human participants were reviewed and approved by Humanities and Social Sciences Research Ethics Committee of the University of the Western Cape. Written informed consent to participate in this study was provided by the participants and their parents/legal guardians.
Author Contributions {#S9}
====================
SS conceptualized and executed the study.
Conflict of Interest {#conf1}
====================
The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Funding.** The study was supported by the Jacobs Foundation and the National Research Foundation of South Africa (Grant No: 111770).
[www.isciweb.org](http://www.isciweb.org)
[^1]: Edited by: Fabian Gander, University of Zurich, Switzerland
[^2]: Reviewed by: Maria Cristina Ginevra, University of Padua, Italy; Elizabeth Dowling, Tufts University, United States
[^3]: This article was submitted to Quantitative Psychology and Measurement, a section of the journal Frontiers in Psychology
| {
"pile_set_name": "PubMed Central"
} |
Celiac disease (CD) is caused by an abnormal intestinal immune response to proline and glutamine-rich wheat gluten proteins and to similar proteins in barley and rye (Green and Cellier [@CR17]). Oat is generally considered safe for consumption by CD patients (Garsed and Scott [@CR16]), although some patients appear to be sensitive to oat as well (Lundin et al. [@CR24]). The only established treatment for the disease is a lifelong gluten exclusion diet.
CD has a strong HLA association. The most prominent association is with HLA-DQ2.5 (DQA1\*05:01, DQB1\*02:01) (Table [1](#Tab1){ref-type="table"}). In individuals who carry the DR3DQ2 haplotype, this molecule is encoded by DQA1 and DQB1 alleles located on the same chromosome (in *cis* configuration), whereas in individuals who are DR5DQ7/DR7DQ2 heterozygous it is encoded by alleles located on opposite chromosomes (*trans* configuration) (Sollid et al. [@CR35]). Most of the remaining patients carry DR4DQ8 haplotypes, and in these patients it is the DQ8 molecule encoded by DQA1\*03, DQB1\*03:02 that is involved in the pathogenesis (Lundin et al. [@CR23]). In the small remaining population of CD patients that are neither DQ2.5 or DQ8, the patients typically express HLA-DQ molecules that contain 'half' of DQ2.5 molecule as they are either DQ2.2 (DQA1\*02:01, DQB1\*02:01) or DQ7.5 (DQA1\*05, DQB1\*03:01) (Karell et al. [@CR19]). As both the DQA1 and DQB1 loci are polymorphic, unique HLA-DQ molecules can be encoded in *trans* configuration. Examples of such molecules are DQ2.3 (DQA1\*03, DQB1\*02:01) and DQ8.5 (DQA1\*05, DQB1\*03:02) found in DR3DQ2/DR4DQ8 heterozygous individuals (Table [1](#Tab1){ref-type="table"}).Table 1Description and naming of HLA-DQ molecules that are associated with celiac disease and which are used as antigen presenting elements for CD4^+^ T cells of celiac disease patientsEncoded byRisk for celiac diseaseExpression in *cis* or *trans* positionPart of common *cis* haplotypeHLA-DQ moleculeDQA1\*DQB1\*HLA-DQ2.50502High*cis, trans*DR3DQ2HLA-DQ2.20202Low*cis, (trans)*DR7DQ2HLA-DQ2.30302Likely low^a^*trans, (cis)*^b^HLA-DQ7.50503:01Very low*cis, (trans)*DR5DQ7HLA-DQ80303:02Low*cis*DR4DQ8HLA-DQ8.50503:02Likely low^a^*trans, (cis)*^ba^Risk for celiac disease has not been established in population studies^b^Molecule can also be encoded in *cis* on some rare haplotypes
CD4^+^ T cells of CD patients (Lundin et al. [@CR22]), but not healthy subjects (Molberg et al. [@CR25]), recognize gluten peptides when presented by disease associated HLA-DQ molecules. This was first shown for DQ2.5 and DQ8 (Lundin et al. [@CR23]), and recently it was demonstrated that DQ2.2 patients (Bodd et al. [@CR7]) and a patient who carries DQ8.5 in a rare *cis* configuration (Kooy-Winkelaar et al. [@CR21]) also have gluten-reactive T cells in the intestinal mucosa. Gluten-reactive T cells can readily be established from intestinal biopsies cultured in vitro (Camarca et al. [@CR8]; Lundin et al. [@CR22], [@CR23]; Tye-Din et al. [@CR39]; Vader et al. [@CR41]; van de Wal et al. [@CR45]). T cells recognizing the same gluten epitopes are normally not detected in the peripheral blood (Anderson et al. [@CR1]), but can be found in the blood of treated CD patients on day 6 after a 3-day oral gluten challenge (Anderson et al. [@CR1], [@CR2]; Ráki et al. [@CR30]).
HLA is the most important genetic factor in CD, and carriage of certain HLA alleles is a necessary, but not sufficient, factor for disease development (Sollid [@CR34]). The other factors required for disease development are non-HLA genes, of which 39 loci have been identified so far (Trynka et al. [@CR38]), and possibly environmental factors other than gluten. To note, mice transgenic for HLA-DQ2.5 and gluten specific T-cell receptors do not develop a CD-like enteropathy (de Kauwe et al. [@CR10]; Du Pre et al. [@CR13]). The reason why these mice do not develop enteropathy may relate to fundamental differences in the gut physiology between mouse and man, and to the lack of appropriate non-MHC genes in the mouse strains tested that parallel the non-HLA susceptibility genes of CD patients.
The differential risk of DQ2.5 and DQ2.2 is linked with the T-cell response to gluten. It has been demonstrated that DQ2.5 binds a larger gluten peptide repertoire compared to DQ2.2 (Vader et al. [@CR43]). Further, gluten T-cell epitopes form stable complexes with DQ2.5, and the increased risk of DQ2.5 over DQ2.2 correlates with a different ability of the two HLA molecules to form stable complexes with many gluten peptides (Fallang et al. [@CR14]).
Characteristically, gluten-reactive T cells of CD patients recognize their antigenic peptides much better when specific glutamine residues are converted to glutamate by the enzyme transglutaminase 2 (TG2) (Molberg et al. [@CR26]; van de Wal et al. [@CR44]). Deamidated gluten peptides bind with increased affinity to DQ2.5 and DQ8 (Arentz-Hansen et al. [@CR3]; Camarca et al. [@CR8]; Henderson et al. [@CR18]; Kim et al. [@CR20]; Moustakas et al. [@CR27]; Quarsten et al. [@CR29]), and the rate of dissociation of deamidated gluten peptides from DQ2.5 has been shown to be substantially slower than for their native counterparts (Xia et al. [@CR47]). The ability to form stable peptide--MHC complexes again seems to be a key factor for the initiation of the anti-gluten T-cell response.
Gluten is defined as the cohesive mass that remains when dough is washed to remove starch (Shewry et al. [@CR32]). Traditionally and strictly speaking, gluten is a name of wheat proteins only, but gluten is now increasingly used as a term to denote proline- and glutamine-rich proteins of wheat, barley, rye and oat. In wheat, gluten consists of the gliadin and glutenin subcomponents. The gliadin proteins can be subdivided into α-, γ- and ω-gliadins, while the glutenin proteins can be subdivided into high molecular weight (HMW) and low molecular weight (LMW) subunits. Common bread wheat is a hexaploid species, and in addition some of the gluten protein encoding genes originate from duplicated loci. Thus, in a single wheat variety there exits up to several hundred different gluten proteins, many of which only differ by a few amino acids. The proline- and glutamine-rich proteins of barley, rye and oat are termed hordeins, secalins and avenins, respectively.
Given the heterogeneity of the wheat gluten proteins, it is no surprise that many distinct gliadin and glutenin derived T-cell epitopes exist (Table [2](#Tab2){ref-type="table"}). T-cell epitopes derived from either α-, γ-, and ω-gliadins as well as from HMW and LMW glutenins have been reported (Arentz-Hansen et al. [@CR3]; Sjöström et al. [@CR33]; Vader et al. [@CR41]; van de Wal et al. [@CR45]). T-cell epitopes in both hordeins and secalins have been described and they are highly homologous to those found in wheat (Tye-Din et al. [@CR39]; Vader et al. [@CR42]). The avenins of oat are more distinct, and although oat is considered safe for CD patients (Garsed and Scott [@CR16]), some CD patients are clinically sensitive to oat (Lundin et al. [@CR24]). Avenin specific as well as cross-reactive responses have been described (Arentz-Hansen et al. [@CR5]; Vader et al. [@CR42]).Table 2List of celiac disease relevant T-cell epitopes recognized by CD4^+^ T cellsEpitope^a^Previous namesPeptide-binding register^b^Reference123456789DQ2.5 restricted epitopesDQ2.5-glia-α1aDQ2-α-I, α9PFPQP**E**LPY(Arentz-Hansen et al. [@CR3])DQ2.5-glia-α1bDQ2-α-IIIPYPQP**E**LPY(Arentz-Hansen et al. [@CR4])DQ2.5-glia-α2DQ2-α-II, α2PQP**E**LPYPQ(Arentz-Hansen et al. [@CR3])DQ2.5-glia-α3glia-α20FRP**E**QPYPQ(Vader et al. [@CR41])DQ2.5-glia-γ1DQ2-γ-IPQQSFP**E**Q[Q]{.ul}(Sjöström et al. [@CR33])DQ2.5-glia-γ2DQ2-γ-II, γ30IQP**E**QPAQL(Qiao et al. [@CR28]; Vader et al. [@CR41])DQ2.5-glia-γ3DQ2-γ-III[Q]{.ul}QP**E**QPYP[Q]{.ul}(Arentz-Hansen et al. [@CR4])DQ2.5-glia-γ4aDQ2-γ-IVSQP**E**Q**E**FPQ(Arentz-Hansen et al. [@CR4])DQ2.5-glia-γ4bDQ2-γ-VIIcPQP**E**Q**E**FPQ(Qiao et al. [@CR28])DQ2.5-glia-γ4cDQ2-γ-VIIa[Q]{.ul}QP**E**QPFPQ(Arentz-Hansen et al. [@CR4])DQ2.5-glia-γ4dDQ2-γ-VIIbPQP**E**QPFC[Q]{.ul}(Qiao, unpublished)DQ2.5-glia-γ5DQ2-γ-VIQQPFP**E**QPQ(Arentz-Hansen et al. [@CR4])DQ2.5-glia-ω1DQ2-ω-IPFPQP**E**QPF(Tye-Din et al. [@CR39])DQ2.5-glia-ω2DQ2-ω-IIPQP**E**QPFPW(Tye-Din et al. [@CR39])DQ2.5-glut-L1glutenin-17PFS**E**Q**E**QPV(Vader et al. [@CR41])DQ2.5-glut-L2glutenin-156FS[Q]{.ul}QQ**E**SPF(Stepniak et al. [@CR36]; Vader et al. [@CR41])DQ2.5-hor-1Hor-α9, Hα9PFPQP**E**QPF(Tye-Din et al. [@CR39]; Vader et al. [@CR42])DQ2.5-hor-2Hor-α2, Hα2PQP**E**QPFPQ(Vader et al. [@CR42])DQ2.5-hor-3hor-I-DQ2PIP**E**QPQPY(Tye-Din et al. [@CR39])DQ2.5-sec-1Sec-α9, Sα9PFPQP**E**QPF(Tye-Din et al. [@CR39]; Vader et al. [@CR42])DQ2.5-sec-2Sec-α2, Sα2PQP**E**QPFPQ(Vader et al. [@CR42])DQ2.5-ave-1aAv-α9APYPEQ**E**EPF(Arentz-Hansen et al. [@CR5]; Vader et al. [@CR42])DQ2.5-ave-1bAv-α9B, 1490PYPEQ**E**QPF(Arentz-Hansen et al. [@CR5]; Vader et al. [@CR42])DQ2.2 restricted epitopesDQ2.2-glut-L1glutenin-17PFS**E**Q**E**QPV(Bodd et al. [@CR7])DQ8 restricted epitopesDQ8-glia-α1DQ8-α-I**E**GSFQPSQ**E**(van de Wal et al. [@CR45])DQ8-glia-γ1aDQ8-γ-Ia**E**QP[Q]{.ul}QPFPQ(Tollefsen et al. [@CR37])DQ8-glia-γ1bDQ8-γ-Ib**E**QP[Q]{.ul}QPYP**E**(Tollefsen et al. [@CR37])DQ8-glut-H1HMW-glutenin[Q]{.ul}GYYPTSP[Q]{.ul}(van de Wal et al. [@CR46])DQ8.5 restricted epitopesDQ8.5-glia-α1DQ8-α-I**E**GSFQPSQ**E**(Kooy-Winkelaar et al. [@CR21])DQ8.5-glia-γ1PQQSFP**E**Q**E**(Kooy-Winkelaar et al. [@CR21])DQ8.5-glut-H1HMW-glutenin[Q]{.ul}GYYPTSP[Q]{.ul}(Kooy-Winkelaar et al. [@CR21])^a^In the epitope names, these short terms are used to denote the type of proteins that the epitopes derive from: 'glia-α' denotes α-gliadin, 'glia-γ' denotes γ-gliadin, 'glia-ω' denotes ω-gliadin, 'glut-L' denotes low molecular weight glutenin, 'glut-H' denotes high molecular weight glutenin, 'hor' denotes hordein, 'sec' denotes secalin and 'ave' denotes avenin^b^Glutamate residues (E) formed by TG2-medited deamidation which are important for recognition by T cells are shown in bold. Additional glutamine residues also targeted by TG2 are underlined
There is at present no standard nomenclature for CD-relevant gluten epitopes. Here, we propose such a nomenclature based on the following three criteria:Reactivity against the epitope must have been defined by at least one specific T-cell clone.The HLA-restriction element involved must have been unequivocally defined.The nine-amino acid core of the epitope must have been defined either by an analysis with truncated peptides and/or HLA-binding with lysine scan of the epitope or comparable approach.
Searching the literature using these criteria, we have compiled a list of epitopes (Table [2](#Tab2){ref-type="table"}). This list includes sequences from α-gliadin, γ-gliadin, ω-gliadin, LMW- and HMW-glutenins, hordeins, secalins and avenins. To note, most gluten-reactive T cells have minimal epitopes longer than those listed in Table [2](#Tab2){ref-type="table"}. This is because MHC class II restricted T-cell receptors usually are sensitive to a few residues flanking the nine-amino acid core region of the epitopes.
In Table [2](#Tab2){ref-type="table"}, some sequences that previously were defined as individual epitopes have been grouped together as members of the same family. This pertains to the DQ2.5-glia-α1a and DQ2.5-glia-α1b as well as the DQ2.5-glia-γ4a, DQ2.5-glia-γ4b, DQ2.5-glia-γ4c and DQ2.5-glia-γ4d epitopes. The reason is that the sequences defining these epitopes are very similar, although there are occasionally T-cell clones that can distinguish between members of the same epitope family (Arentz-Hansen et al. [@CR4]; Qiao et al. [@CR28]). Most T-cell clones appear to cross-react between peptide sequences of the same family.
The nine-amino acid core sequences of some of the DQ2.5 restricted epitopes are identical, but because these epitopes derive from different cereal species they are still listed as unique epitopes. This applies to the DQ2.5-glia-ω1, DQ2.5-hor-1 and DQ2.5-sec-1 epitopes, as well as DQ2.5-hor-2 and DQ2.5-sec-2 epitopes. T-cell cross-reactivity between proteins of different species hence does often occur, but T cells reactive with these epitopes can also be species-specific as the T-cell receptors may be sensitive to unique residues flanking the nine-amino acid core region of the epitopes. In addition, there are epitopes, like DQ2.5-hor-3 (Tye-Din et al. [@CR39]), that have sequences which are hordein or secalin specific and which elicit species specific T-cell responses.
The majority of gluten-reactive T cells generated from DQ2.5 positive CD patients can recognize their epitopes when confronted in vitro with antigen presenting cells expressing the closely related HLA-DQ2.2 molecule (DQA1\*02:01, DQB1\*02:01). Yet, DQ2.2 positive CD patients do not mount a T-cell response to the same gluten epitopes, but rather have responses to gluten peptides that bind stably to DQ2.2 (Bodd et al. [@CR7]). When defining an epitope, it is thus important to assess and categorize the epitope only in the context of the HLA molecules that are expressed by the T-cell donor.
The selection of gluten T-cell epitopes is best understood in HLA-DQ2.5 positive CD patients, and is influenced by at least three factors: (a) resistance of the polypeptide sequence to proteolytic degradation, (b) specificity of TG2 and (c) HLA binding specificity. The proline-rich nature of gluten makes the gluten proteins resistant to proteolytic degradation in the gastrointestinal lumen, and long gluten peptide fragments ranging from 15 to 50 residues therefore survive in the small intestine (Shan et al. [@CR31]). T-cell epitopes are often localized within such long fragments (Arentz-Hansen et al. [@CR4]). The resulting proline and glutamine rich peptides are often good substrates for TG2 (Dørum et al. [@CR11], [@CR12]; Fleckenstein et al. [@CR15]; Vader et al. [@CR40]). Proline is influencing the specificity of TG2 as the enzyme typically recognizes glutamine residues in glutamine-X-proline sequences (Fleckenstein et al. [@CR15]; Vader et al. [@CR40]). T-cell epitopes in their native form are usually very good substrates for TG2. TG2, being an important factor in the selection of T-cell epitopes, is demonstrated by the fact that TG2 is selectively targeting peptides which harbor T-cell epitopes from a digest of gluten with extreme complexity (Dørum et al. [@CR12]). Finally, determinant selection by MHC is influencing repertoire of T-cell epitopes. In general, gluten peptides are poor binders to HLA class II molecules with the exception of HLA-DQ molecules associated with CD (Bergseng et al. [@CR6]). Some gluten peptides also bind HLA-DR53 (Clot et al. [@CR9]), although so far no T cells of celiac lesions recognizing these peptides have been described. The selective force of HLA is illustrated by the observation that the DQ2.5 and DQ8 epitopes generally come from non-overlapping sequences of gluten proteins (Tollefsen et al. [@CR37]).
The glutamate introduced by TG2 is usually in position 4 (P4), P6 or P7 in HLA-DQ2.5 restricted epitopes and at position P1 and/or P9 in HLA-DQ8 restricted epitopes (Table [2](#Tab2){ref-type="table"}). These glutamate residues serve as anchor residues important for binding of the peptides. Both HLA-DQ2.5 and DQ8 prefer negatively charged residues at these anchors sites. This positioning of deamidated glutamine residues is strongly related to the positioning of proline residues, which is particularly strict in the case of DQ2.5 epitopes, as DQ2.5 only accepts proline at certain position in the peptide binding groove (Kim et al. [@CR20]). This results in a dominant presence of proline at P1, P6 and P8 and leads to the modification by TG2 of the glutamine residues at P4 and P6, respectively. Such positioning of proline residues is less strict in the case of the DQ8 epitopes. Although polyclonal T-cell responses to multiple T-cell epitopes are almost invariably found in CD patients, responses to the DQ2.5-glia-α1a, DQ2.5-glia-α1b, DQ2.5-glia-α2 epitopes, DQ2.5-glia-ω1, DQ2.5-glia-ω2, DQ2.5-hor-1 and DQ2-sec-1 are dominant in DQ2.5 positive patients (Arentz-Hansen et al. [@CR3]; Camarca et al. [@CR8]; Tollefsen et al. [@CR37]; Tye-Din et al. [@CR39]). In DQ8-positive patients, responses to the DQ8-glia-α1 epitope are most frequently found (Tollefsen et al. [@CR37]; van de Wal et al. [@CR45]).
The list of gluten epitopes recognized by T cells of CD patients presented in Table [2](#Tab2){ref-type="table"} will be extended as new epitopes become known in the future. A dedicated website (<http://www.isscd.org/EpitopeNomenclature>) will update this list as more epitopes are identified.
Open Access {#d29e2170}
===========
This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.
| {
"pile_set_name": "PubMed Central"
} |
In the article cited above, Fig. 3 was updated: the number 5 in the *x*-axis of panels *A* and *B* were deleted. Also, two consecutive sentences on page 1675 were revised to read: ... NGR was greatest in the quartile at lowest risk of progressing to DM (35 vs. 17% and difference of 18%). In those at highest risk of progressing to DM, the corresponding rates were 24 and 11% (difference of 13%).
The online version (<https://doi.org/10.2337/dc17-1116>) has been corrected to reflect these changes.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Microbial electrochemical technologies (METs) have been heavily investigated for over a decade, specifically for applications in energy generation, waste reuse and resource recovery. Essentially, these are bioreactors with an anode and cathode where either one or both is biotic, allowing for oxidation (anode) and reduction (cathode) reactions. Electrode-assisted methanogenesis, or electromethanogenesis, refers to microbial electrochemical CO~2~ reduction to methane at the biocathode (biotic cathode). This process has been mainly investigated in microbial electrolysis cells (MECs) and, more recently, in microbial electrosynthesis (MES) ([@B13]). Applications for electromethanogenesis include bioelectrochemical power-to-gas, biogas upgrading, and wastewater treatment ([@B29]; [@B13]).
In electromethanogenesis, CO~2~ is converted to methane using reducing equivalents generated from the cathode, either through direct uptake of electrons from the cathode surface ([@B54]) or indirectly via H~2~. H~2~ is considered to be the main electron donor and can be produced abiotically through the hydrogen evolution reaction (HER) at the electrode surface at low cathode potentials (\<-0.6 V vs. Ag/AgCl) or biotically by proton-reducers in mixed cultures such as sulfate-reducing bacteria (SRB) ([@B52]; [@B62]; [@B10]; [@B13]; [@B1]). Formate, which can arise as an intermediate during CO~2~ reduction, can also serve as an indirect electron mediator ([@B62]).
The microbial community at the biocathode of MES is less studied compared to MEC. In recent years, many MES studies have been done, focusing on methane production, as well as acetate and other volatile fatty acids (VFAs). These studies generally provide basic descriptions of the cathodic microbial community but with little analysis of the ecology and community assembly ([@B38]; [@B37]). There is generally a lack of in-depth analysis into why certain communities are present, their distribution and interactive networks for electromethanogenesis, except for some notable exceptions ([@B15]; [@B25]). Understanding community assembly in MES is important as these systems rely entirely on their microbiome to function; a deeper understanding allows for microbial ecology-based engineering of efficient systems ([@B37]). Further, understanding community assembly dynamics in these systems enables the development of accurate models to predict reactor performance ([@B28]) and ensure predictable communities that function to achieve reproducibility and reliability for large scale applications ([@B35]).
Community assembly refers to the species present in a community at a given space and time ([@B11]). There are two main theories on what drives microbial community assembly. The niche theory follows the assumption that certain microorganisms with specialized fitness are better suited to survive in certain environments or niches, and thus community assembly in these niches is driven by deterministic factors such as substrate availability and competition. On the other hand, the neutral theory assumes equal fitness amongst different microbial communities, with community assembly differences arising from stochastic factors such as birth/death and immigration ([@B32]). Although electroactive microbes do not belong to a unique ecological niche ([@B36]), METs create a highly selective niche environment for electroactive microbes. Electroactive microbes have an added fitness due to their ability to perform extracellular electron transfer to donate electrons to anodes in METs, which serves as the main deterministic driver of anodic community assembly, although stochastic assembly has been reported ([@B69]; [@B22]). However, the cathode environment in MES creates two main selective factors: the ability to accept extracellular electrons using the cathode as the sole electron donor (through direct electron uptake or H~2~ due to proton reduction catalyzed by the cathode surface at lower potentials) and the capability of autotrophic growth as CO~2~ is the only externally added carbon source. Extracellular electron transfer capabilities (whether transfer to anode or uptake from cathode) have been demonstrated across a range of phyla, while autotrophic growth in the conditions set at the cathode in MES is thus far limited to mainly three phyla (Euryarchaeota, Firmicutes, and Proteobacteria) ([@B36]; [@B39]). Therefore, while there are studies into community assembly and spatial variability of bioanodes, those conclusions may not necessarily be applicable to MES biocathode community assembly.
No studies to date have investigated cathode spatial variability in methanogenic MES reactors. Spatial variability can arise due to a number of factors that affect microbial community assembly. While there is a bulk environment in these reactors, local microenvironments arise across the cathodes due to H~2~ and CO~2~ mass transfer limitations from the bulk solution into the biofilm, charge limitations across the biofilm as shown in other electrode-associated biofilms ([@B6]). Within the biofilm, further variations can occur in terms of differences in H~2~ gas bubble evolution along the cathode surface as a function of cathode roughness and H~2~ saturation ([@B61]; [@B56]). The pH gradients due to HER ([@B17]) affects CO~2~ solubility, and thus its availability, due to the shifts in the bicarbonate -- carbonate equilibrium in response to pH change ([@B9]). Collectively, these can potentially lead to spatial variability across the cathode biofilm, where certain sections may have higher or lower amounts of biomass and different types of microorganisms aggregating.
Understanding community spatial distribution and heterogeneity is important to ensure appropriate sampling strategies are undertaken. If there is significant spatial variability, wrong conclusions can be drawn depending on the number and position of samples. In larger scale electrodes, this spatial variability could be quite significant. Additionally, understanding spatial distribution can give insights on syntrophic or competitive relationships that occur. Mixed communities offer more robust and functionally redundant systems that are more suited to industrial applications due to their resiliency to possible operational fluctuations; understanding spatial variability of intact mature biofilms is important in predicting how these biofilms will behave under large scale applications. Therefore, the objective of this study was to evaluate the spatial distribution and variability across cathodic biofilms in an electromethanogenic MES system, as well as the reproducibility of biocathode community between biological replicates. We hypothesized that if deterministic factors are the dominant driving factor, then no significant difference in community composition (i.e., beta diversity) is expected between replicate reactors. In contrast, if neutral or stochastic factors prevail, then significant beta diversity would be observed.
Materials and Methods {#s1}
=====================
MEC Set Up and Methanogenic Biocathode Enrichment
-------------------------------------------------
Triplicate single-chamber MEC reactors were prepared using 300 ml screw-capped borosilicate glass bottles, with a working volume of 280 ml. The caps and bottles were modified with appropriate ports to place the electrodes, gas collection bag (Calibrate, Inc., United States) and gas sampling port. The anodes, made of carbon fiber brush with a titanium core (4 cm × 2.5 cm, The Mill-Rose Company, United States), were cleaned by soaking in acetone overnight, washing with sterile deionized water and heat-treated at 450°C for 15 min. Carbon cloth cathodes were prepared with 160 cm^2^ (8 cm length × 10 cm width) geometric surface area, with titanium wire woven through as the current collector. The cathodes were cleaned by soaking in acetone overnight, washed with sterile deionized water and dried at room temperature. The anode and cathode were positioned vertically within the reactor, approximately 2 cm apart. Teflon tape and epoxy were applied on all the connections to ensure a proper seal. Reactors were inoculated with sludge from an anaerobic membrane bioreactor (10% v/v), mixed with a synthetic influent medium containing 10 mM sodium acetate as the carbon source and electron donor. The influent medium was prepared using a modified DSMZ Medium 826 (DSMZ, Leibniz, Germany) with the following composition (g/L): NH~4~Cl, 1.5; Na~2~HPO~4~, 0.6; KCl, 0.1; Na~2~HCO~3~, 2.5; CH~3~COONa, 0.82, and 10 ml trace minerals and vitamin solution each. To maintain anaerobic conditions, the media was sparged for 1 h using a N~2~:CO~2~ (80:20) gas mixture and then autoclaved. The sodium bicarbonate was sterile filtered into the media after autoclaving to maintain a pH of ∼7.5. The reactors were operated in fed-batch mode with an applied voltage of 0.7 V using an external power source (3645 A; Circuit Specialists, Inc., United States). A data logger (ADC 24, PicoLog, United Kingdom) was used to measure the voltage across an external resistor (R~ex~ = 10 Ω). A 10% decrease in voltage from the peak reading signaled the end of each batch for media replacement and sampling. This was approximately every 48 h. The reactors were enriched for a total of 5 months, after which the enriched methanogenic biocathodes were transferred to sterile triplicate double-chambered three-electrode MES reactors.
MES Set Up and Operation
------------------------
For the double-chambered MES reactors, the anodes were titanium sheets using titanium wires as the current collectors. A Nafion^®^ 117 cation exchange membrane (Sigma, United States) was used to separate the double chambers. Gas bags were attached to gas outlet ports to collect biogas produced during each batch for analysis. The same MEC enrichment media composition was used for the MES operation with the omission of sodium acetate, and continuous stirring. Therefore, the only carbon source was CO~2~ in the form of dissolved sodium bicarbonate for pH adjustment and 100% CO~2~ gas, which was continuously bubbled into the reactors at the beginning of each batch for 5 min and acted as a CO~2~ reservoir through passive gas diffusion from the gasbags into the reactor headspace. An Ag/AgCl reference electrode (BASi, United States) was inserted in the cathode chamber to maintain the set potential control. A VMP3 potentiostat (BioLogic, United States) was connected to the three-electrode system to chronoamperometrically maintain a cathode set potential of -1.0 V vs. Ag/AgCl. The reactors were batch-fed with each batch lasting 140 h. Once stable methane production was observed for three batches, the biocathodes were removed for microbial community analyses. To minimize disturbance to the biofilm, biocathode sampling was done only at the end of the experiment. Three 2 cm^2^ samples were cut using sterilized scissors from each biocathode at the top, center, and bottom positions ([Figure 1](#F1){ref-type="fig"}) and suspended in 6 ml sterile media. These were vortexed for 1 min to detach the microbial cells from the cathode and stored at -80°C for subsequent protein analysis, DNA extraction and amplicon sequencing.
{#F1}
Measurement and Analyses
------------------------
Liquid and gas (H~2~, CH~4~, and CO~2~) samples were measured at the end of each batch cycle using chromatographic methods. Volatile fatty acids were detected at 210 nm using an Aminex HP-87H column (Bio-Rad, Hercules, CA, United States) with a UV-detector high performance liquid chromatography (HPLC; Shimadzu, Japan). The mobile phase was 0.005M H~2~SO~4~ at a flow rate of 0.55 ml/min. Samples were filtered through 0.2 um filters prior to analysis. The gas compositions in the reactor headspace and gas bag were analyzed using a gas chromatograph (Model\# 8610C, SRI Instruments, United States) as previously described ([@B31]).
### MEC Calculations
Current density *j* (mA/cm^2^) was calculated as:
I
=
V
R
where I was the current (mA) calculated from the recorded voltage (mV) across the resistor (1000 mΩ, R), divided by the geometric surface area of the cathode (160 cm^2^). Coulombic efficiency (CE%) was calculated as:
CE
=
C
t
C
th
×
100
where C~t~ is the total coulombs calculated by integrating the current over time (C~t~ = Σ *I* Δt, Δt is the cycle duration), C~th~ is the theoretical amount of coulombs available based on the acetate removed over the same amount of time, calculated as C~th~ = \[*F b* (*C*~in~ -- *C*~out~)\]/*M*, where *F* is Faraday's constant (96485 C/mol), *b* = 8 is the number of electrons produced per mole of acetate, *C*~in~ and *C*~out~ are the influent and effluent acetate concentrations and *M* = 82 is the molecular weight of acetate ([@B66]).
### MES Calculations
The current density was calculated as the recorded current (mA) divided by the geometric surface area of the cathode (160 cm^2^). Coulombic efficiency (CE%) was calculated as the actual total coulombs C~t~ recovered as H~2~, CH~4~, formate, and acetate divided by the theoretical total coulombs C~th~ as recorded by the potentiostat software. Cathodic hydrogen and methane recoveries (r~catH2~ and r~catCH4~) were calculated by:
r
catCH2
=
n
H2
n
CE
r
catCH4
=
n
CH4
n
CE
where n~H2~ and n~CH4~ are the moles of the respective gas. n~CE~ is the total moles of gas possible based on the total coulombs C~t~, and is calculated by:
n
CE
=
C
t
bF
where b is the number of moles of electrons required for hydrogen production (2 mol e^-^) or methane production (8 mol e^-^) and F is Faraday's constant (96,485 C/mol e^-^).
Microbial Community Analyses
----------------------------
### Protein Analysis
The total protein was measured based on the Lowry method ([@B41]). The suspended samples described above (see section "MES Set Up and Operation") were thawed at room temperature and the total protein was determined using the DC-protein assay kit (BIO-RAD Laboratories, Inc., United States) following the manufacturer's instructions after being re-suspended in deionized (DI) water, with a series of graded Bovine Serum Albumin (BSA, Sigma Aldrich, United States) solutions as standards (0--0.5 μg/μl, *R*^2^\> 0.97) ([@B15]; [@B24]; [@B12]).
### DNA Extraction, Library Preparation, Amplicon Sequencing, and Bioinformatic Processing
Genomic DNA was co-extracted with RNA from the carbon cloth and 150 μl of the cell suspension in which it was stored in using the PowerBiofilm RNA Isolation Kit (Qiagen, Germany) with a modified protocol using phenol:chloroform:isoamyl alcohol pH 6.5--8.0 (AMRESCO, Inc., United States) and bead beating lysing matrix E tubes (MP Biomedicals, New Zealand) instead of the original bead beating tubes. The extracted DNA concentration was measured using Qubit^®^dsDNA HS Assay Kit (Thermo Scientific, United States), according to the manufacturer's instructions.
Amplicon libraries were prepared for the archaeal and bacterial 16S rRNA gene V3--V4 region using up to 10 ng of the extracted DNA, the forward primer Pro341F (5′-CCTACGGGNBGCASCAG-3′) and the reverse primer Pro805R (5′-GACTACNVGGGTATCTAATCC-3′) ([@B30]). Each PCR reaction (25 μL) contained dNTPs (100 μM of each), MgSO~4~ (1.5 mM), Platinum Taq DNA polymerase HF (0.5 U/reaction), Platinum High Fidelity buffer (1x) (Thermo Fisher Scientific, United States) and tailed primer mix (400 nM of each forward and reverse primer). The PCR amplification was conducted by an initial denaturation step at 95°C for 2 min, 35 cycles of amplification (95°C for 20 s, 50°C for 30 s, 72°C for 60 s) and a final elongation at 72°C for 5 min ([@B56]). Duplicate PCR reactions were performed for each sample and the duplicates were pooled after PCR. The resulting amplicon libraries were purified using the standard protocol for Agencourt Ampure XP Beads (Beckman Coulter, United States) with a bead to sample ratio of 4:5. DNA concentrations were measured using the Qubit^®^dsDNA HS Assay Kit, followed by product size and purity validation with gel electrophoresis using Tapestation 2200 and D1000/High sensitivity D1000 screentapes (Agilent, United States).
Sequencing libraries were prepared from the purified amplicon libraries using a second PCR. Each PCR reaction (25 μL) contained PCRBIO HiFi buffer (1x), PCRBIO HiFi Polymerase (1 U/reaction) (PCRBiosystems, United Kingdom), adaptor mix (400 nM of each forward and reverse) and up to 10 ng of amplicon library template. The PCR amplification was conducted by an initial denaturation step at 95°C for 2 min, 8 cycles of amplification (95°C for 20 s, 55°C for 30 s, 72°C for 60 s) and a final elongation at 72°C for 5 min. The resulting sequencing libraries were purified as mentioned above using the Agencourt Ampure XP Beads. DNA concentration, product size and purity were measured as mentioned above. The purified sequencing libraries were pooled in equimolar concentrations and diluted to 2 nM. The samples were paired-end sequenced (2 bp × 300 bp) on a MiSeq using a MiSeq Reagent kit v3 (Illumina, United States) following the standard guidelines for preparing and loading samples on the MiSeq. \> 10% PhiX control library was spiked in to overcome low complexity issues often observed with amplicon samples.
Forward and reverse reads were trimmed for quality using Trimmomatic v. 0.32 ([@B14]) with the settings SLIDINGWINDOW:5:3 and MINLEN: 275. The trimmed forward and reverse reads were merged using FLASH v. 1.2.7 ([@B42]) with the settings -m 10 -M 250. The trimmed reads were dereplicated and formatted for use in the UPARSE workflow ([@B26]). The dereplicated reads were clustered, using the usearch v. 7.0.1090 -cluster_otus command with default settings. Operational taxonomic unit (OTU) abundances were estimated using the usearch v. 7.0.1090 -usearch_global command with -id 0.97 -maxaccepts 0 -maxrejects 0. Taxonomy was assigned using the RDP classifier ([@B64]) as implemented in the parallel_assign_taxonomy_rdp.py script in QIIME ([@B18]), using --confidence 0.8 and the MiDAS database v. 1.23 ([@B43]), which is a curated database based on the SILVA database, release 123 ([@B51]). The results were analyzed in R v. 3.5.0 ([@B21]) through the RStudio using the ampvis2 package, which was also used to visualize the relative read abundance as a heatmap ([@B2]). The log abundance ratio was calculated by taking the log (base 10) of the ratio of the OTU relative abundance of organism X (OTU_X) to organism Y (OTU_Y) from within the same sample:
log
abundance
ratio
=
log
\[
relative
abundance
OTU
\_
X
/
relative
abundance
OTU
\_
Y
\]
Statistical Analyses
--------------------
Statistical analyses were performed in RStudio using the base R, the ampvis2 package for alpha diversity and rank abundance of the core community ([@B3]), and QIIME 1.9.1 for beta diversity analysis ([@B18]). The normality of data distribution was examined by the Shapiro--Wilk test. The two-tailed (independent) Student's *t*-test was used to compare means between unpaired groups with an assumption of unequal variance between sample sets. The Mann--Whitney *U*-test was used to compare non-parametric variables between two groups. One-way analysis of variance (ANOVA) was used to compare parametric variables among three or more groups, and the Kruskal--Wallis test was used for non-parametric variables. Quantitative variables were expressed as the mean ± standard deviation or median and interquartile range, according to the sample distribution. *p*-Values less than 0.05 were considered to indicate statistical significance against the null hypothesis of no variance.
The beta-diversity dissimilarity analyses were done using the Bray--Curtis and Weighted UniFrac metrics with the beta_diversity.py script in QIIME. The Bray--Curtis dissimilarity is based on abundance, while Weighted UniFrac distance matrix calculates dissimilarity based on abundance and phylogeny. These results were visualized using non-metric multidimensional scaling (nMDS) with the ampvis2 R package. To assess the significance of the calculated beta-diversity dissimilarities, pairwise analyses of similarities (ANOSIM) based on 999 permutations and Adonis/permutational multivariate analysis of variance (PERMANOVA) based on 719 permutations were performed to compare each reactor and each sampling position groups using compare_categories.py for ANOSIM and Adonis, wrapping in the vegan R package ([@B47]). The ANOSIM R statistic is based on the difference of mean ranks between groups and within groups and ranges from 0 (no separation) to 1 (complete separation). The ANOSIM *R* statistic is based on the difference of mean ranks between groups and within groups and ranges from 0 (no separation) to 1 (complete separation). The Adonis/PERMANOVA pseudo-*F* statistic operates on ranked dissimilarity, comparing the total sum of squared dissimilarities between groups to the squared dissimilarity within groups. Larger *F*-ratios indicate greater group dissimilarity. Statistical significance is determined comparing the statistic values (*R* or *F* depending on the test) retrieved from multiple permutations of the test ([@B16]). The beta-diversity analyses were performed using raw reads, reads rarefied to 41,955, reads normalized by cumulative sum scaling (CSS), and reads normalized by DESeq2 to assess if different normalization methods would affect the statistical significance of the calculated dissimilarities. Rarefaction involves subsampling all samples to an even depth without replacement. CSS involves scaling only the relatively invariant counts across samples, to reduce the influence of larger count values in the same matrix column ([@B48]). DESeq2 calculates a scaling factor for each OTU in each sample based on the median of the scaling factors of the mean across all samples. It assumes a Negative Binomial distribution and minimizes the influence of large count values on the values of other OTUs allowing for increased sensitivity for smaller datasets ([@B40]). CSS and DESeq2 were implemented with the normalize_table.py script in QIIME.
Nucleotide Sequence Accession Numbers
-------------------------------------
The 16S rRNA gene sequencing reads have been deposited in the National Center for Biotechnology Information (NCBI) under BioProject ID [PRJNA541055](PRJNA541055) with Accession Nos. SAMN11571464--SAMN11571473.
Results
=======
MES Performance
---------------
The reactors were operated for 5 months in MEC mode to enrich for methanogens, indicated by the detection of methane. The enriched methanogenic biocathodes were transferred to double-chambered MES reactors and operated in batch-fed mode for six batches (140 h batch-length) until stable methane production was observed in the last three consecutive batches (ANOVA, *F* = 4.2, *p* \> 0.05, [Supplementary Table S1](#SM1){ref-type="supplementary-material"}). The MES cathodic current density (0.02--0.04 mA/cm^2^) was similar to that observed during the MEC operation (0.02--0.07 mA/cm^2^), with an overall increase in current consumption over time ([Figure 2A](#F2){ref-type="fig"} and [Supplementary Table S1](#SM1){ref-type="supplementary-material"}). While there was little variability between reactors in each batch, there was a significant difference between batches (ANOVA, *F* = 5.1, *p* = 0.001), with the exception of methane production, which consistently averaged between 8 and 13 μmol/cm^2^ ([Figure 2A](#F2){ref-type="fig"}). H~2~ production was more variable (Kruskal--Wallis, χ^2^ = 5.1, *p* = 0.02), reaching as high as 14 μmol/cm^2^ (Batch 5) and as low as 0.08 μmol/cm^2^ (Batch 6). The average HER rate was 0.04 μmol/cm^2^/h, compared to an abiotic HER rate of 6.3 μmol/cm^2^/h. The r~catH2~ (113 ± 7.5%, [Supplementary Table S1](#SM1){ref-type="supplementary-material"}) in the abiotic control reactor was much higher than for the biotic r~catH2~ (average of 4.9 ± 6.4%). Both electrode-assisted methanogenesis and acidogenesis occurred, as evidenced by the products detected at the end of each batch (methane, formate, and acetate) ([Figure 2B](#F2){ref-type="fig"}). The VFA production was more significantly variable than gas production. Formate was detected in minimal amounts compared to the other products. Acetate levels varied significantly between batches (Kruskal--Wallis, χ^2^ = 21.9, *p* \< 0.001), reaching a maximum of 10 μmol/cm^2^ to undetectable concentrations in Batch 6.
{#F2}
Biomass Analysis and Alpha Diversity
------------------------------------
Triplicate samples (top, center, and bottom) were taken from each of the replicate biocathodes at the end of the experiment once stable performance was achieved (as determined by methane production) to quantify biomass and characterize microbial community diversity through amplicon sequencing. Total biomass did not vary significantly by position (Kruskal--Wallis, χ^2^ = 0.047, *p* \> 0.05), although it varied significantly between the biological replicates (Kruskal--Wallis, χ^2^ = 15.16, *p* = 0.0005), with the highest total biomass measured in Reactor 1 ([Figure 3](#F3){ref-type="fig"}).
{#F3}
Sequence reads after quality filtering ranged between 41,944 and 75,499, for a total of 473,048 reads which were resolved into 263 total observed OTUs. The sampling depth was sufficient to capture most of the species in the samples, as seen in the rarefaction curves ([Supplementary Figure S1](#SM1){ref-type="supplementary-material"}). Diversity in each sample was calculated based on the number of observed OTUs, Shannon--Weaver, Simpson's Diversity and Chao1 richness estimator after rarefying to 41,944 reads. Shannon and Simpson diversity indices place more emphasis on abundant OTUs, whereas Chao1 takes into consideration rare OTUs. The results for the enriched biocathodes are presented based on reactor and position in [Figure 4](#F4){ref-type="fig"}. Species richness (observed OTUs) was the highest in Reactor 1, with a median slightly greater than 130 observed OTUs, followed by Reactor 2 (median centered between 127 and 130 observed OTUs) and Reactor 3 (median centered between 123 and 125 OTUs). However, when considering Shannon--Weaver and Simpson diversity indices, Reactor 1 and 3 were most similar (Student's *t*-test*, p* \> 0.05), whilst Reactor 2 had the lowest alpha diversity, and was less evenly distributed as reflected in the rank abundance for the three reactors ([Supplementary Figure S2](#SM1){ref-type="supplementary-material"}), in which only four OTUs comprised more than 80% of the cumulative read abundance for Reactor 2 as compared to the other two reactors (11 OTUs for Reactor 1, 8 OTUs for Reactor 3). While alpha diversity was higher in Reactor 1 and 3 compared to Reactor 2, based on the abundance-based indices (i.e., Shannon--Weaver and Simpson diversity indices), there was less difference observed when considering the Chao1 richness and evenness index. No statistically significant difference was observed except when comparing the Shannon--Weaver diversity index between reactors (ANOVA, *F* = 7.34, *p* = 0.0244).
{#F4}
When comparing diversity based on position, the top samples had a higher diversity in terms of observed OTUs and Chao1 richness compared to bottom samples. This suggests some spatial heterogeneity across the cathode surface, with more richness observed in the top part of the cathode compared to the bottom regardless of the individual differences between reactors. However, when considering the dominant OTUs, no statistically significant difference was apparent in Shannon--Weaver and Simpson indices between the different sampling positions.
Core Dominant OTUs
------------------
In this study, the core dominant OTUs were defined as the OTUs present in all samples with a relative abundance ≥ 0.1%. The biocathodes were enriched with a core dominant OTUs representing 8% of total OTUs (21/263 total OTUs) and \> 97% of total reads in all samples ([Supplementary Figure S2](#SM1){ref-type="supplementary-material"}). Of the 63 OTUs present at a relative abundance ≥ 0.1% in the initial anaerobic sludge inoculum ([Supplementary Figure S3](#SM1){ref-type="supplementary-material"}), only 10 were still represented at ≥ 0.1% in the final biocathode community for all samples. These 21 OTUs represented 17 core genera (or lowest taxonomic classification) belonging to the phyla Euryarchaeota (domain Archaea) and to Synergistetes, Bacteroidetes, Firmicutes, Proteobacteria, and Chloroflexi (domain Bacteria), as presented in a heatmap of relative abundance ([Figure 5](#F5){ref-type="fig"}). The communities were dominated by the methanogenic archaeal communities and, to a lesser degree, a diverse group of bacteria. The biocathodes were highly enriched with hydrogenotrophic methanogens belonging to the family *Methanobacteriaceae* (five OTUs in total with no less than 41%, up to almost 70% relative abundance), and, to a lesser degree, *Methanosarcina* (one OTU) and *Methanomassiliicoccus* sp. (one OTU). The SRB, *Desulfovibrio* and *Desulfuromonas* sp., were relatively equally distributed across the cathode, with a relative abundance between 0.3 up to 4.6%.
{#F5}
There appeared to be a preferential localization for some communities at the different positions on the cathode based on the relative abundance. This is more easily visualized by comparing log abundance ratios for the methanogenic and sulfate-reducing communities within each sample ([Figure 6](#F6){ref-type="fig"}). With log ratios, every 0.33 represents a doubling in ratio, or every 1.0 represents a 10-fold increase. This means that for two OTUs with the same relative abundance, the log ratio would be zero, while if one OTU is two times more abundant, the log ratio would equal 0.33 and so on. While *Methanobacterium* sp. were the most relatively abundant community and were enriched across the cathode, they were more localized in the upper part of the cathode relative to the unclassified *Methanobacteriaceae* sp. (log ratio of 1) and *Methanosarcina* sp. (log ratio of 1.1), and almost two times less in the bottom part of the cathode compared to the *Methanobacteriaceae* sp. (log ratio -0.3). This seems to indicate a preferential localization for this community at the top of the cathode that decreased in the lower part of the cathode, in an inverse relationship to *Methanobacteriaceae* and *Methanosarcina* sp. While the log ratio decreased between *Methanobacterium* sp. and the two SRBs from the top to bottom, this appeared to be due to the decrease in relative abundance of *Methanobacterium* rather than a decrease in the SRB abundance, as evidenced by the log ratio comparing the two SRBs showing little difference regardless of cathode position (0.1--0.2). The other less abundant members of the core community also appeared to demonstrate less spatial variation; the results of their log ratio distributions are not shown in [Figure 6](#F6){ref-type="fig"} for simplicity.
{#F6}
Beta Diversity
--------------
Beta-diversity statistical analyses were done to determine the statistical significance of the observed spatial heterogeneity in relative abundance. The nMDS plots of the core community revealed that samples clustered more by reactor rather than by position regardless of the dissimilarity matrix used (Bray--Curtis or weighted UniFrac) ([Figure 7](#F7){ref-type="fig"}). The data were normalized differently to see if beta-diversity results would be affected by normalization methods ([@B44]; [@B65]). Pairwise comparisons using ANOSIM or Adonis/PERMANOVA with raw OTU abundance data, data rarefied to 41,944 reads, data normalized by DESeq2 and CSS methods for the core OTUs and all the retrieved OTUs are shown in [Supplementary Table S2](#SM1){ref-type="supplementary-material"}. Pairwise comparisons revealed no significant differences in beta-diversity regardless of the normalization methods and whether core or all retrieved OTUs were used. In no case was there a significant difference from the null distribution ([Supplementary Table S2](#SM1){ref-type="supplementary-material"}, *p* \> 0.05).
{#F7}
Discussion
==========
Variations in Reactor Performance
---------------------------------
Enrichment of the methanogenic biocathodes was done in single-chambered MECs prior to the dual-chambered MES operation to facilitate the growth of a mature methanogenic biocathode. HER rates are lower in dual-chambered reactors due to mass transfer limitations of protons across the cation exchange membrane separating the anode from the cathode. Several MES studies report an initial enrichment step prior to MES reactor transfer ([@B27]; [@B8]; [@B68]). While there was little variability between reactors within each batch during MES operation, there was a significant difference between batches in line with reported variability in methanogenic biocathode performance ([@B59]; [@B15]; [@B8]; [@B13]). Methane production was relatively consistent between batches compared to H~2~ production. It is important to note that r~catH2~ values ([Supplementary Table S1](#SM1){ref-type="supplementary-material"}) only reflect the average H~2~ detected at the end of each batch, and not the total H~2~ produced in the system. H~2~ is difficult to accurately quantify in such systems due to its role as an electron donor (biotic reaction) in the case of indirect electron transfer and because it can easily diffuse through the membrane and leak out from reactors and tubing ([@B55]; [@B24]; [@B22]). This was further supported by the r~catH2~ (113 ± 7.5%, [Supplementary Table S1](#SM1){ref-type="supplementary-material"}) for the abiotic control reactor, which was significantly higher than for the biocathode (r~catH2~ of 4.9%). While most of this difference was due to H~2~-mediated methanogenesis, it is not clear how much of the evolved H~2~ was lost from the system. In either case, the H~2~ evolving from the cathode would not have been enough to account for the methane detected, since CO~2~ reduction to methane requires 4 moles of H~2~ for every mole of methane. The average amount of methane detected during the three MES batches was 1.6 mmol; an equivalent of ∼ 6.5 mmol H~2~ would be required to maintain the 1:4 stoichiometric ratio. However, the average abiotic H~2~ was only 0.89 mmol, indicating that abiotic HER was not the only source for reducing equivalents for methanogenesis, as has been previously reported ([@B24]) which is discussed further in Section "Methanogen-Dominated Core Community."
Methanogen-Dominated Core Community
-----------------------------------
Absolute abundance values are not possible with just amplicon sequencing, and thus the data were only reported in terms of relative abundance. The relative abundance values may not fully capture the abundance of organisms due to the known limitations associated with amplicon sequencing, namely primer specificity, PCR, and 16S rRNA gene copy number variations between different species ([@B45]). Nevertheless, the methanogenic biocathodes were clearly highly enriched with a core community (\>97% similarity across all the samples) that was made up of 60--80% hydrogenotrophic methanogens from the family *Methanobacteriaceae* (*Methanobacterium* sp. and an unclassified genus of *Methanobacteriaceae*) and *Methanosarcina* sp. (∼10%). Hydrogenotrophic methanogen communities are frequently reported to dominate methanogenic biocathodes, whether in MECs or MES, especially *Methanobacterium* sp. and *Methanobrevibacter* sp. ([@B60]; [@B13]; [@B24]). Members of the family *Methanobacteriaceae* are capable of reducing CO~2~ to CH~4~ in the presence of H~2~ (or formate) as an electron donor. H~2~-mediated methanogenesis is generally the most dominant pathway for methane generation in these systems as hydrogenotrophic methanogens do not contain cytochromes for direct electron uptake from the cathode ([@B62]; [@B10]; [@B13]), although *Methanospirillum hungatei* was recently reported to be capable of producing electrically conductive filaments ([@B63]). *Methanosarcina* spp. are metabolically versatile with mixotrophic growth; different species can produce CH~4~ through the three methanogenesis pathways (hydrogenotrophic, aceticlastic, and methylotrophic) ([@B34]; [@B67]). While they contain cytochromes and are known to be involved in electroactivity ([@B53]; [@B54]), they were present in lower abundance than the hydrogenotrophic methanogens, which is consistent with previous reports of methanogenic MES systems where *Methanobacteriaceae* spp. dominate the cathodic community ([@B13]).
Roughly 10% of the core community was represented by Proteobacteria. Members of the phylum Proteobacteria have been described as important members of methanogenic biocathodes, especially SRB like *Desulfovibrio* sp. and *Desulfuromonas* sp. which were enriched across the cathodes. *Desulfovibrio* spp. require an organic carbon source along with CO~2~ for growth due to their incomplete Krebs cycle ([@B50]), consuming carbohydrates and VFAs with H~2~ as an electron donor ([@B5]) in the presence of sulfates. While SRB can outcompete hydrogenotrophic methanogens for H~2~ in the presence of sulfates due to their lower K~s~ and higher growth rates, under sulfate-limited conditions they instead establish a syntrophic relationship where SRB act as H~2~ producers and hydrogenotrophic methanogens as H~2~ consumers to maintain the thermodynamic favorability of the reaction ([@B46]; [@B66]). This was indeed the case in our reactors due to the presence of only trace amounts of sulfate in the media. Additionally, electrotrophy in SRB has been demonstrated, where they can directly accept electrons from the cathode and reduce protons to H~2~ since they contain hydrogenases ([@B7]; [@B1]). Thus, the H~2~ evolution coupled with limited sulfate and acetate favored the growth of hydrogenotrophic methanogens, especially the observed *Methanobacterium* sp. and the other unclassified member of the *Methanobacteriaceae* family.
The remainder of the core community was made up of a diverse group of fermenters of the phyla Bacteroidetes, Synergistetes, Firmicutes, and Chloroflexi. Since no external organic carbon source was added, their presence was probably due to endogenous decay of the biofilm and amino acid fermentation, as has been previously reported ([@B25]). These fermenters can produce acetate, H~2~, and CO~2~ as end products of their fermentation, and they are discussed further in the [Supplementary Discussion](#SM1){ref-type="supplementary-material"}. Although not part of the core dominant community (not present at a relative abundance of 0.1% or higher in all samples), a variety of genera capable of aerobic growth were also enriched at relative abundances \> 0.1%. These included *Aquamicrobium* sp., *Thiobacillus* and the family *Comamonadaceae*. Aerobic microorganisms have been previously reported in other anaerobic bioreactors, including microbial fuel cells ([@B58]), where it is expected that they persist by consuming any intruding oxygen in the system, thus aiding in maintaining an anaerobic environment. There may have been oxygen intrusion through the cation-exchange membrane separating the two chambers from the water-splitting abiotic anode ([@B19]; [@B57]). The diversity of the microbiomes in the replicate reactors can help with stability and adaptability in the face of such destabilizing/unfavorable conditions.
Log Ratio Abundance Highlights Preferential Spatial Localization of Hydrogenotrophic Methanogens
------------------------------------------------------------------------------------------------
Although the hydrogenotrophic methanogens were relatively evenly distributed across the cathode, some spatial segregation was apparent for *Methanobacterium* sp. (four OTUs) and the unclassified *Methanobacteriaceae* sp. (one OTU) ([Figure 5](#F5){ref-type="fig"}). *Methanobacterium* sp. were more highly abundant in the top samples versus the bottom, in a somewhat inverse relation to the *Methanobacteriaceae* sp.; this is more evident when comparing the log ratio abundance between the two communities ([Figure 6](#F6){ref-type="fig"}).
Considering the broad range of species that belong to the family *Methanobacteriaceae*, the results suggested that these were two different spatially segregated hydrogenotrophic methanogen groups, possibly due to differences in their H~2~ utilization and growth kinetics. Different local microenvironments or niches may have developed at the top compared to the bottom of the cathode, resulting in spatial segregation of hydrogenotrophic methanogen communities. The H~2~ that evolves at the cathode does not reach an equilibrium state between the headspace and dissolved H~2~ due to its low solubility and density ([@B66]), and the dynamics between H~2~ production and microbial consumption rates across the cathode ([@B24]). In this study, the rate of H~2~ production was a function of the abiotic HER and biotic H~2~ evolution by SRB and endogenous decay. The rate of H~2~ consumption is a function of the maximum H~2~ utilization rates (v~max~) and maximum specific growth rates (μ~max~) of the hydrogenotrophic methanogens ([@B20]). The physical proximity of the top part of the cathode to the headspace, which has an abundance of H~2~ relative to the solution, may have resulted in relatively higher H~2~ availability in the compared to the bottom part of the cathode. Since the *Methanobacterium* sp. preferentially aggregated in the top, it may be inferred that this genus had a lower affinity to H~2~ (higher half-saturation constant, K~s~). The unclassified *Methanobacteriaceae* sp. may have had higher affinity to H~2~ (lower K~s~) allowing it to be more competitive in an environment with lower H~2~ availability, i.e., the bottom of the cathode. Higher versus lower HER rates have been shown to result in the dominance of different *Methanobacteriaceae* spp. in methanogenic biocathodes ([@B66]). In the case of similar H~2~ affinity, competition would be based on μ~max~ ([@B4]). Therefore, it is probable that H~2~ affinity and maximum specific growth rates differed between the two communities. It is not possible to determine exactly the species-level taxonomic classification with 16S rRNA amplicon sequencing, so no comparison of exact growth kinetics (i.e., K~s~ and μ~max~) can be made for the different species.
The distribution in relative abundance of the *Methanosarcina* sp. was more uniform across the cathode. As previously stated, they are capable of methanogenesis by using acetate, cathode, or H~2~ as an electron donor ([@B53]; [@B54]). *Methanosarcina* sp. have a reported acetate threshold between 0.2 and 1.2 mM ([@B33]). The maximum acetate concentration measured in the reactors was 1.6 mM, which is above the minimum threshold, although it varied between batches to below 0.2 mM. It should be noted that the acetate concentration was measured at the end of the batch and it is possible that the concentrations of acetate were higher during the batch and decreased with time. H~2~-driven methanogenesis would have led to direct competition between *Methanosarcina* sp. and other hydrogenotrophic methanogens, which have a lower K~s~ (H~2~) and thus a higher affinity to H~2~. A direct electron transfer mechanism by *Methanosarcina* sp. would not have involved competition for anything other than physical space to enable direct interaction with the cathode (along with H^+^ and CO~2~). It is possible that *Methanosarcina* sp. grew using a combination of acetate or direct electron transfer relatively independent of the H~2~ availability, leading to their uniform distribution across the cathode.
A similar uniformity of distribution was observed when comparing the abundances of the sulfate-reducing *Desulfovibrio* sp. and *Desulfuromonas* sp. As with the *Methanosarcina* sp., direct electron transfer for H^+^ reduction to H~2~ by the SRB would lead to uniformity across the cathode since no concentration gradients occur in terms of physical location. While the SRB abundance was relatively stable across the cathode in relation to each other (log ratio of 0.1--0.2, [Figure 6](#F6){ref-type="fig"}), they were generally more highly abundant in the bottom part of the cathode as compared to the top of the cathode in relation to the *Methanobacterium* sp. This was due to the decrease in *Methanobacterium* sp. abundance in the bottom of the cathode. While *Methanobacterium* sp. abundance was overall lower in the bottom of the cathode, higher abundance was observed (\>21%) in Reactor 1 and 3 samples with concurrently higher abundances of the SRB (\>4%) compared to Reactor 2 (8.5% *Methanobacterium*), which had a lower abundance of SRB (0.6%) ([Figure 5](#F5){ref-type="fig"}). This may suggest that *Methanobacterium* sp. had a stronger reliance on their syntrophic relationship with the SRB due to the lower H~2~ availability at the bottom.
[Figure 8](#F8){ref-type="fig"} presents a hypothetical schematic describing these apparent spatial distribution trends of the core community members that are central to the functional performance of methanogenic MES in terms of current consumption, hydrogen production and methane production. Overall, it seemed the differences in the local segregation of the hydrogenotrophic methanogens could be mainly due to the difference in micro-scale H~2~ and CO~2~ availability as well as their growth kinetics (i.e., K~s~ and μ~max~).
.](fmicb-10-01747-g008){#F8}
No Significant Beta-Diversity Within and Between Replicate Biocathodes
----------------------------------------------------------------------
Despite the observed differences in relative abundance distribution, no statistically significant variance in beta-diversity was observed within and amongst the triplicate biocathodes, suggesting a deterministic-driven assembly of the cathodic microbial community. Seeded with the same inoculum, the microbial community at the cathode of replicate reactors converged to the same core community. This convergence to a core community of 21 OTUs supports a deterministic community assembly as has been shown in bioanodes and in anaerobic digestion ([@B23]; [@B49]). Acting as the sole electron donor, the cathode creates a highly selective stress for chemolithoautotrophs capable of growth via direct electron transfer mechanisms or with H~2~ as an electron donor that clearly shapes the cathodic community, driving it toward a core community dominated by hydrogenotrophic methanogens (five OTUs). These results are promising, as they support the reproducibility of methanogenic MES biocathodic communities and their functional redundancy (i.e., different species that can perform the same function) which is important when considering larger scale applications subjected to operational fluctuations. Functional redundancy can help maintain the overall performance of the cathode, since differences in the cathode local micro-environments can arise in terms of HER variability that can occur due to cathode materials and pH gradients, H~2~ availability throughout the thickness of the biofilm, and syntrophic relationships that contribute to substrate (H~2~ and CO~2~) availability. The results highlight the importance of sufficient and appropriate sampling for microbial community analyses. Local variabilities in abundance can affect the conclusions drawn regarding factors shaping the community and the dominance of certain communities. Triplicate samples from multiple points across the cathode are the minimum needed for statistical analyses to determine whether observed variabilities are significant. However, many MEC and MES do not report their results in the framework of statistically relevant differences. It should be noted that amplicon DNA sequencing technique does not differentiate active from non-active members in the microbial community. Methods such as reverse-transcribed rRNA, can be applied in future studies, for identifying active populations to gain a deeper understanding of the functionally relevant interactions between communities.
Conclusion
==========
This study presents insights into the microbial community assembly, spatial distribution and homogeneity of electromethanogenic biocathodes. Our data showed that while the functional performance of these bioreactors may vary, it is unlikely to be due to differences in the overall communities present as deterministic assembly led to the development of a specific core community responsible for the majority of CO~2~ conversion to CH~4~ via different syntrophic relationships. Even though local community segregation may occur due to the differences in H~2~ utilization and competitive relationships, this did not result in any statistically significant overall beta diversity within the cathodes or between reactors. This information is relevant to understanding cathode community assembly in METs, especially those conducted with CO~2~ as the sole electron donor. For METs applied to wastewater treatment, where more complex organic substrates are available, obviously there would be differences in the core community assembly with a higher abundance of fermenters and heterotrophic growth due to the presence of larger amounts of fermentable substances, and stochastic community assembly may be stronger with continuous-operation reactors due to a regular immigrant influx. Additionally, this study highlights the importance of sufficient sampling for statistical analyses purposes that allow for more in-depth and meaningful investigations of sequencing data generated from the many MET studies carried out.
Data Availability
=================
The datasets generated for this study can be found in the 16S rRNA gene sequencing reads have been deposited in the National Center for Biotechnology Information (NCBI) under BioProject ID [PRJNA541055](PRJNA541055) with Accession Nos. SRR9017425--SRR9017429.
Author Contributions
====================
AR, KK, and PS designed the experiments. AR performed the experiments, analyzed the data, and wrote the manuscript. MA provided guidance on the statistical analyses. AR, KK, MA, and PS contributed to the discussion and editing of the manuscript.
Conflict of Interest Statement
==============================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Funding.** This work was funded by Competitive Research Grant (URF/1/2985-01-01) to PS from King Abdullah University of Science and Technology.
Supplementary Material
======================
The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fmicb.2019.01747/full#supplementary-material>
######
Click here for additional data file.
[^1]: Edited by: Jorge Rodríguez, Khalifa University, United Arab Emirates
[^2]: Reviewed by: Anthony P. Malanoski, United States Naval Research Laboratory, United States; Guangli Liu, Sun Yat-sen University, China
[^3]: This article was submitted to Microbiotechnology, Ecotoxicology and Bioremediation, a section of the journal Frontiers in Microbiology
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
Individuals with lower limb amputation face challenges in maintaining their balance when navigating uneven terrains or encountering perturbations during walking^[@CR1]--[@CR3]^. The fear of falling and decreased balance confidence are prevalent among lower limb amputees (LLAs)^[@CR2],[@CR4]^, which are important factors in their mobility and participation in social activities^[@CR4]--[@CR8]^. Compared to individuals without lower limb loss, LLAs have slower walking speeds, possibly because of decreased gait stability and the need for increased conscious attention while walking on uneven or changing terrains^[@CR1]^. In a survey of community-dwelling LLAs, more than 50% reported that they had fallen at least once in the past year^[@CR4],[@CR9]^. Amputees typically place more trust in the intact limb, which results in overuse and destructive long-term consequences, such as osteoarthritis of the intact knee and/or hip^[@CR10]^. Decreased loading on the affected limb can also lead to osteopenia and subsequent osteoporosis. With the growing number of people who lose limbs due to vascular diseases or trauma, it is important to develop assistive technologies that improve standing stability in this population.
Three main sensory systems, the visual, vestibular, and somatosensory, contribute to stable posture during stance^[@CR11],[@CR12]^. Theses inputs are integrated and processed in the central nervous system which generates appropriate movement strategies and motor commands to maintain postural stability^[@CR13],[@CR14]^. However, when any of the sensory inputs are absent or inaccurate, the CNS adjusts the gains for each input to control the stability^[@CR13]^. Such adjustments are often demonstrated in increased body sway, and if not successful can result in loss of balance and falls.
The absent sensory feedback from the missing foot in LLAs plays a crucial role in the degradation of their balance^[@CR15]--[@CR17]^. LLAs mainly rely on other sensory inputs, such as vision or proprioception from the intact and residual lower limbs, to compensate for compromised sensory information^[@CR3],[@CR17]^. When vision is blocked, LLAs have significantly more postural sway and are less stable compared to able-bodied controls^[@CR18],[@CR19]^, indicating that the lack of somatosensory feedback from the missing limb contributes to the marked differences in stability^[@CR17]^. Moreover, unilateral amputees use sensory feedback from their intact ankle and foot to compensate for the somatosensory information lost with the missing limb. Studies show unilateral amputees rely more on their intact limb to make balance adjustments and reduce the risk of fall^[@CR20]^. When LLAs have trouble maintaining balance with their intact leg, they are more likely to have poor functional outcomes related to personal care, household activities, and recreational activities^[@CR21]^.
Electrical stimulation of the remaining nerves in the residual limb of LLAs via various neural interface technologies can elicit somatosensory percepts referred to the missing limb^[@CR22],[@CR23]^. The modality and the intensity of the reported sensations can be modulated by tuning the stimulation parameters^[@CR23]^. The sensations evoked by non-penetrating multi-contact cuff electrodes implanted on the peripheral nerves above the knee in the residual limbs of LLAs have been robust and consistent for more than two years. Furthermore, the perceived sensations generated by neural stimulation have central processing times and temporal sensitivities similar to natural tactile sensation^[@CR24]^. Although LLAs report improvements in self-reported confidence with the sensory feedback elicited by neural stimulation^[@CR22],[@CR25]^, the effects of such feedback on objective measures of balance has not previously been determined.
The Sensory Organization Test (SOT) has been utilized as a clinical and research tool to objectively and quantitatively examine the contribution of different sensory systems to standing balance^[@CR12]^. In the SOT, visual and somatosensory inputs are selectively perturbed or missing (Fig. [1](#Fig1){ref-type="fig"}) and the results on postural control are examined individually and in combination^[@CR8]^. Outcomes of the test are correlated to overall balance performance during ambulation and activities of daily living^[@CR2],[@CR26]^. The SOT has been administered on different patient populations with standing stability deficits, including stroke survivors^[@CR27]^, individuals with Parkinson's Disease^[@CR28],[@CR29]^, LLAs^[@CR2],[@CR3]^, and elderly people^[@CR14],[@CR26]^.Figure 1Conditions of SOT in which controlled perturbation to visual and somatosensory inputs could be applied. Red boxes denote perturbation of the corresponding sensory input. Participant's eyes were closed in conditions 2 and 5.
Standing balance is one of the most basic tasks in amputee rehabilitation and plays an essential role in most functional activities^[@CR16],[@CR30]^. In this study, we examined whether the sensory feedback provided by chronically implanted non-penetrating, epineural nerve cuff electrodes could improve balance stability in transtibial amputees. Our hypotheses were: (1) somatosensory feedback elicited by direct neural stimulation will reduce the sway exhibited by LLAs when other sensory inputs are perturbed, and (2) electrically elicited sensations related to the missing foot will improve weight distribution symmetry between the intact and prosthetic limbs. The results of this study may have implications to the development of new prosthetic technologies intended to reduce the risk and fear of falls, improve standing balance and balance confidence, encourage engagement in unstructured community environments, or accelerate the rehabilitation process following lower limb amputation.
Methods {#Sec2}
=======
Research participants {#Sec3}
---------------------
Two individuals with unilateral transtibial amputations (LL01 & LL02) volunteered and enrolled in this study. A summary of their characteristics at the time of enrollment is presented in Table [1](#Tab1){ref-type="table"}. Both participants were regular prosthesis users with no medical history of peripheral neuropathy, dysvascular disease, phantom pain, or uncontrolled diabetes. Participants had no fall history for at least nine months prior to the beginning of the study, and were therefore both classified as non-fallers^[@CR2]^. The experiments described in this work were conducted at least a year after their enrollment. However, during the first year post nerve cuff implantation, both participants regularly visited the laboratory where they received neural stimulation and performed other tests including impedance measurements, sensory threshold determination, sensory mapping, and psychometric experiments described elsewhere^[@CR23],[@CR31],[@CR32]^. The Louis Stokes Cleveland Veterans Affairs Medical Center Institutional Review Board and Department of the Navy Human Research Protection Program approved all study procedures, which were conducted under an Investigational Device Exemption obtained from the United States Food and Drug Administration. The study was designed in accordance with relevant guidelines and regulations, and both individuals gave their written informed consent to participate.Table 1Summary of participant characteristics enrolled in the study.ParticipantSexAge (year)Height (cm)Weight(kg)Amputated sideEtiologyTime since amputationLL01M67173106LeftTraumatic48LL02M5416867RightTraumatic11
Neural interface technology {#Sec4}
---------------------------
The details of neural interface technology and implantation technique have been described previously^[@CR23]^. Both participants had 16-contact Composite Flat Interface Nerve Electrodes (C-FINEs) installed around their sciatic, tibial and/or common peroneal nerves during an outpatient surgical procedure. All C-FINE contacts were connected to percutaneous leads via industry-standard 8-contact in-line connectors (Medtronic Inc.). The percutaneous leads exited the skin on the upper anterior thigh. To deliver stimulating currents during laboratory visits, the percutaneous leads from C-FINEs were connected to a custom-designed external stimulator^[@CR23],[@CR32]^. Figure [2a](#Fig2){ref-type="fig"} depicts schematically the implanted and external components of the system.Figure 2Neural interface technology and the sensory neuroprosthesis. (**a**) The cuff electrodes were implanted on sciatic and/or tibial and peroneal nerves. The access to individual contacts within each cuff electrode was through percutaneous leads, which connected to an external stimulator. (**b**) Pressure distribution under the prosthetic foot is sensed via an array of FSRs integrated into an in-shoe insole, electrical stimulation to specific cuff contacts is determined based on the FSR readings, and sensations that match in location and perceived intensity of the pressure profile under prosthetic foot are elicited. (**c**) Reported percept locations from LL01 and LL02. Stimulation delivered selectively to contacts generating perceived sensations referred to the missing toes and heels in response to pressures applied to the insole FSR array.
Electrical stimulation {#Sec5}
----------------------
The pulse amplitude range for the external stimulator was 0--5.6 mA with the resolution of 0.1 and 0.2 mA for values below and above 2 mA, respectively. The pulse width (PW) could be modulated between 0--255 μs with a resolution of 1 μs^[@CR23],[@CR33]^. Stimulating currents were delivered to the nerves in a series of asymmetric, charge-balanced, cathodic-first pulses with return to a common anode placed on the skin above the iliac crest. Stimulation parameters were set through a custom-made routine in Simulink (MathWorks Inc.) and then compiled and downloaded into a dedicated computer running xPC Target real-time kernel (MathWorks Inc.) for real-time operation during standing experiments. An optical isolator between the xPC target computer and the stimulator ensured electrical isolation between the participant and other AC-powered electrical equipment. Stimulation charge density was kept below 60 uC/cm^2^ to avoid any potential of damage to the neural tissue and/or platinum contacts^[@CR23],[@CR31]^.
Sensory neuroprosthesis {#Sec6}
-----------------------
Able-bodied individuals sense their center of pressure partly through cutaneous sensation from the plantar surface of the foot. The pressure distribution under the feet changes as they sway in an anterior-posterior or medial-lateral direction, which provides feedback utilized by the central nervous system to maintain balance. Similarly, we implemented a mechanism of sensory feedback in which the perceived intensity of elicited sensations was proportional to pressure underneath the prosthetic foot (Fig. [2b](#Fig2){ref-type="fig"}). For each participant, stimulating currents were delivered through a subgroup of C-FINE contacts to elicit sensations corresponding to pressures applied to either the heel or forefoot (Fig. [2c](#Fig2){ref-type="fig"}). This selection was based on prior mapping experiments^[@CR23]^.
The pressure distribution underneath the prosthetic foot was measured using dynamic force-sensing resistors (FSRs) incorporated into a shoe insole (IEE S.A.). The resistance for the FSRs was more than 1 MΩ when the insole was unloaded and decreased to 2 KΩ with increasing loads up to 70 N/cm^2^ pressure. Each insole contained eight individual FSR cells, and readings from cells were collected using a data acquisition board (NI PCI-6071E, National Instruments) with a sampling rate of 1000 Hz. The readings from two FSR cells at the heel were averaged together to estimate an overall value for rearfoot load. Similarly, readings from the first metatarsal and the big toe FSR cells were averaged together to provide an estimate of the overall load on the forefoot.
To modulate the perceived intensity of the elicited sensation, stimulation PW varied proportionally in response to pressure readings from the FSR insoles. A calibration process was performed to determine the minimum, reference, and maximum FSR values and their corresponding PWs. Minimum values were recorded when no load was applied to the insole by the prosthetic foot (i.e., either while subjects sat or stood with their prosthetic foot off the ground), which were associated with sub-sensory threshold PWs. Reference values were obtained by having participants stand with equal weight on both legs, and the corresponding PWs were set such that the perceived intensities matched the pressures reported for the intact foot. Isolated maximal pressures were then applied to the forefoot and rearfoot by shifting body weight to the prosthetic toe and heel, respectively, and recording the maximum signal values from the corresponding regions of the FSR insole. In these postures, PW values were set at levels that participants verbally confirmed were higher in perceived intensity than the sensations elicited with the prosthetic foot flat on the ground in the reference position. Having established the minimum, reference, and maximum values for the FSR and PW values, the input-output relationship between pressure and PW was defined by a piecewise linear function (Eq. [1](#Equ1){ref-type=""}):$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$pw(v)=\{\begin{array}{c}\,0,\,v < {V}_{min}\\ \left(\frac{P{W}_{ref}-P{W}_{min}}{{V}_{ref}-{V}_{min}}\right)v,\,{V}_{min}\le v\le {V}_{ref}\\ \left(\frac{P{W}_{max}-P{W}_{ref}}{{V}_{max}-{V}_{ref}}\right)v,\,{V}_{ref}\le v\le {V}_{max}\\ P{W}_{max},\,{V}_{max}\le v\end{array}$$\end{document}$$where the *v* is the voltage readings from FSR.
Experimental design {#Sec7}
-------------------
The SOT was administered using a SMART Balance Master (Natus Medical Inc.). The device was equipped with a controllable platform with two embedded dynamic force plates capable of anterior-posterior translation or rotating about the ankle, and a visual surround capable of rotating about the subject. Movements of the platform and visual surroundings were controlled by the NeuroCom Balance Manager Software Suite (Natus Medical Inc.). Participants were tested under six sensory conditions while they were secured in a loosely fitting safety harness attached to an overhead bar. The conditions for the SOT, as listed in Table [2](#Tab2){ref-type="table"} and illustrated in Fig. [1](#Fig1){ref-type="fig"}, involve visual and/or somatosensory perturbations. Rotations of the platform and/or the visual surroundings in the fore-aft direction was proportionally matched with a gain to the sway of each participant during the test, such that higher postural sway resulted in greater perturbations in the platform or visual surroundings. The gain was selected after test trials in which participants found it difficult to maintain their balance during the most challenging condition (\#6 -- inaccurate visual and compromised somatosensory inputs), yet not to a degree that it would result in a fall. Prior work with lower-limb amputees performing the SOT either excluded trials with falls in data analysis^[@CR34]^ or allowed participants to repeat the trial^[@CR8]^, which could skew the sway-related outcomes. Therefore, we decided to set the gain such that the test would be maximally challenging without compromising the validity of the analysis. For LL01 and LL02 the gains were set to 1 and 2, respectively.Table 2Summary of conditions in SOT.ConditionPlatformEyesSurrounding1StationaryOpenStationary2StationaryClosedStationary3StationaryOpenMoving4MovingOpenStationary5MovingClosedStationary6MovingOpenMoving
Each SOT condition lasted for 20 s and participants were instructed to maintain as little postural sway as possible, to keep their feet in the same position throughout the test, and to keep their arms at their sides. One test block consisted of all six SOT conditions, with each condition tested two times. Each block was performed under one sensory stimulation mode: closed-loop sensory neuroprosthesis active (stimulation "on") or inactive (stimulation "off"). The order of conditions was randomized within each block, and the order of stimulation modes was randomized between blocks. Six blocks were collected for each sensory stimulation mode in total, i.e. 12 trials for each SOT condition and sensory stimulation mode.
For every trial, the time series of ground reaction forces, Center of Pressure (COP), and estimates of Center of Gravity (COG) were extracted using the clinical module in the NeuroCom Balance Manager Software Suite. The raw force plate data were sampled at 100 Hz and saved on a local hard drive for offline processing.
Data analysis and outcome measures {#Sec8}
----------------------------------
Equilibrium Score (ES), a clinically known measure to quantify sway amplitude during SOT conditions^[@CR3]^, was calculated for every trial based on Eqs. [2](#Equ2){ref-type=""} and [3](#Equ3){ref-type=""}, consistent with the built-in equations used in NeuroCom Software Suite clinical module^[@CR35],[@CR36]^.$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$ES=100\times \frac{{12.5}^{o}-(Max({\theta }_{A})-Max({\theta }_{P}))}{{12.5}^{o}}$$\end{document}$$
In Eq. [2](#Equ2){ref-type=""}, *Max*(*θ*~*A*~) and *Max*(*θ*~*P*~) are the maximum COG angular sways in the anterior and posterior directions, respectively. 12.5^o^ is an accepted range of anterior to posterior sway before an able-bodied individual loses balance during stance^[@CR37],[@CR38]^. An ES approaching 100 denotes minimal sway, whereas scores around zero indicate that balance is approaching the limits of stability. The *Max*(*θ*~*A\ or\ P*~) was calculated using Eq. [3](#Equ3){ref-type=""}, in which *h* is the participant's height^[@CR36]^:$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$Max({\theta }_{AorP})(deg)={\tan }^{-1}\left(\frac{Max(CO{G}_{AorP})(cm)}{0.55\times h(cm)}\right)\times \frac{{180}^{o}}{\pi }$$\end{document}$$
Because ES only considers extreme limits of sway angle, it cannot capture the complete sway history during a trial. Therefore, we calculated two additional sway-related outcomes, Root Mean Square (RMS) distance of the COP, and elliptic area approximation of COP. In summary, the higher ES indicates better balance. Conversely, higher values for COP-related measures indicate less stability.
The RMS distance of the COP (*DIST*~*RMS*~) is an indicator of variability in COP movement. It has been shown to be a reliable measure of postural equilibrium^[@CR39]--[@CR41]^ and is sensitive to altered sensory inputs^[@CR42],[@CR43]^. *DIST*~*RMS*~ was calculated using Eq. [4](#Equ4){ref-type=""}:$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$DIS{T}_{RMS}=\sqrt{\frac{1}{N}\mathop{\sum }\limits_{1}^{N}RD{[n]}^{2}}$$\end{document}$$where N is the total number of samples during a trial, N = 2000 and *RD*\[*n*\] is the resultant distance (RD) vector of the COP as given below (Eq. [5](#Equ5){ref-type=""}):$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$RD[n]=\sqrt{CO{P}_{AP}{[n]}^{2}+CO{P}_{ML}{[n]}^{2}}$$\end{document}$$
In Eq. [5](#Equ5){ref-type=""}, COP~AP~ and COP~ML~ are the COP components in the Anterior-Posterior (AP) and Medial-Lateral (ML) directions, respectively. The lower-case 'n' in Eqs. [4](#Equ4){ref-type=""} and [5](#Equ5){ref-type=""} indicates a discrete-time sample. The mean values of the AP and ML components were subtracted from the COP vectors in every trial to eliminate any inconsistency due to foot placement across trials. The mean was calculated over the 20 s, the period of the trial.
Additionally, an elliptic area approximation of the COP path was computed for each trial. This measure captures the changes in COP path during standing and has been utilized as an indicator of overall postural performance^[@CR44]^. Following the method described in Schubert *et al*.^45^, we calculated a 95% prediction ellipse based on the assumption that points in the COP scatter follow a Chi-square distribution^[@CR45]^. The area of the ellipse was calculated using Eq. [6](#Equ6){ref-type=""}:$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$Are{a}_{PE}=\pi ab$$\end{document}$$where a and b were semimajor and semiminor axes of the confidence ellipse and they were estimated according to Equation 7:$$\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$a=\sqrt{{\chi }_{2}^{2}.{\lambda }_{1}},b=\sqrt{{\chi }_{2}^{2}.{\lambda }_{2}}$$\end{document}$$
In Eq. [7](#Equ7){ref-type=""}, λ~1~ and λ~2~ are eigen values of the COP covariance matrix. $\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$${\chi }_{2}^{2}$$\end{document}$ is the value of the Chi-square cumulative distribution with two degrees of freedom at probability level = 0.95.
Lastly, any changes in weight symmetry were ascertained by calculating the percentage of the body weight placed on the prosthesis. The ground reaction forces from the force plate underneath the prosthetic foot were normalized to the sum of ground reaction forces from both feet, i.e., body weight.
Statistical analyses {#Sec9}
--------------------
A two-way ANOVA was conducted to examine the effects of stimulation mode (i.e., sensory neuroprosthesis active or inactive) and SOT condition on the means ± standard deviations of the outcome measures. Extreme outliers, defined as data points more than three interquartile ranges away from either the lower quartile or upper quartile, were removed from the analysis. Normality was assessed using Kolmogrov-Smirnov normality test for each cell of the design. Any statistically significant interactions between stimulation mode and SOT conditions were followed up by analysis of simple main effects to determine the impact of stimulation under specific SOT conditions. For the analysis of simple main effects, the statistical significance received a Bonferroni adjustment for the two stimulation modes and was accepted at the p \< 0.025 level. If no interaction effects were found, we tested for the main effect of stimulation on the measured outcome. Because there were only two stimulation modes, no post hoc analyses were deemed necessary. For all other comparisons, we used two-tailed t-tests followed by Bonferroni adjustments if multiple paired comparisons took place. All the statistical analysis were performed using IBM SPSS Statistics Ver. 22 (IBM Corp.).
Results {#Sec10}
=======
Sensations elicited by the sensory neuroprosthesis {#Sec11}
--------------------------------------------------
For both participants, stimulation was delivered via different contacts in the cuff electrodes implanted on the sciatic nerve. For LL01, when pressure was applied to the FSRs at a location corresponding to the first metatarsal and big toe of the prosthetic foot, electrical stimulation was delivered to evoke sensations perceived as arising from the missing forefoot (Fig. [2c](#Fig2){ref-type="fig"}). The readings from the first metatarsal and the big toe FSR cells were averaged together to provide an estimate of the overall load on the forefoot. In response to pressure on the FSRs underneath the prosthetic heel, the neuroprosthesis elicited sensation perceived as originating in the missing heel. The readings from two FSR cells at the heel were averaged together to estimate an overall value for the rearfoot load.
Similarly, for LL02, pressure to the first metatarsal and toe FSRs triggered electrical stimulation, which elicited sensation related to the missing first to fifth metatarsal areas. In response to pressure on the heel FSRs, the neuroprosthesis elicited sensation perceived as arising from the missing heel and lateral ankle.
Effects of sensory stimulation on ES {#Sec12}
------------------------------------
A significant interaction between stimulation mode and SOT condition for ES was found in both participants (LL01: p = 0.029, LL02: p \< 0.001). This suggests that the effect of stimulation on ES depended on SOT condition (Fig. [3](#Fig3){ref-type="fig"}). For both participants, ES was significantly lower in condition six (visual and somatosensory inputs compromised) during trials with the sensory neuroprosthesis active. For participant LL02, sensory feedback also led to an improvement in ES for conditions four (somatosensation compromised) and five (vision and somatosensation compromised).Figure 3Effects of sensory stimulation on Equilibrium Score for LL01 (left) and LL02 (right). Age-matched normative means are shown in red. There was a significant interaction between stimulation mode and SOT condition. For LL02, the ES was improved with sensory feedback in conditions four, five, and six. For LL01, this improvement was observed in condition six. \* and \*\* denote p \< 0.05 and p \< 0.001, respectively.
In SOT condition six, the ES values were 74.8 ± 9.6 and 65.2 ± 7.9 for the sensory neuroprosthesis active and inactive, respectively, for LL01. This represents a statistically significant mean improvement of 9.2 with 95% Confidence Interval (CI) between 5.4 to 12.9 (p \< 0.001). For LL02, the ES significantly improved in conditions four, five, and six. Additionally, for LL02, the effect of electrically elicited sensory feedback on ES grew bigger from condition four to six, and approached able-bodied norms. In condition four, the ES values for LL02 were 79.6 ± 4.0 and 74.3 ± 7.5 for sensory stimulation on and off, respectively, a statistically significant mean improvement of 5.2 (95% CI 0.2 to 10.3, p = 0.042). In condition five, the ES values for LL02 were 96.7 ± 5.7 and 63.4 ± 7.1 for the sensory neuroprosthesis active and inactive, respectively, a statistically significant mean improvement of 6.3 (95% CI 1.3 to 11.3, p = 0.015). In condition six, the ES values for LL02 were 65.3 ± 4.0 and 49.7 ± 13.7 for sensory stimulation on and off, respectively, a statistically significant mean improvement of 15.6 (95% CI 10.6 to 20.7, p \< 0.001).
There was a statistical difference in baseline ES values without electrically elicited sensory feedback between LL01 and LL02 in conditions four, five, and six. In condition four, without sensory stimulation, the ES values were 84.0 ± 6.8 and 74.4 ± 7.1 for LL01 and LL02, respectively (p = 0.006). In condition five, they were 72.9 ± 4.1 and 63.4 ± 6.8 for LL01 and LL02, respectively (p = 0.001). In condition six, they were 65.1 ± 7.5 and 49.7 ± 13.1 for LL01 and LL02, respectively (p = 0.004). Such differences between the two participants suggest that without the sensory neuroprosthesis active, LL01 had higher postural stability than LL02 in the last three conditions of the SOT. Lastly, the ES from these two participants were compared to previously reported aged-matched normative ES values^[@CR46]^. For conditions four, five, and six, LL02 had significantly lower ES without sensory stimulation (p \< 0.05) whereas LL01 either scored equal or higher than normative values (Fig. [3](#Fig3){ref-type="fig"}).
Effects of sensory stimulation on RMS distance of COP {#Sec13}
-----------------------------------------------------
Sensory stimulation affected the *DIST*~*RMS *~in both participants, as shown in Fig. [4](#Fig4){ref-type="fig"}. For both LL01 and LL02, there was a statistically significant interaction between stimulation mode and SOT condition on *DIST*~*RMS*~ (LL01: p \< 0.001, LL02: p = 0.001). For both participants, *DIST*~*RMS *~was significantly lower in condition six during trials with electrically elicited sensory feedback. For LL02 only, *DIST*~*RMS *~also improved with sensory stimulation in condition four.Figure 4Effects of sensory stimulation on RMS distance of COP for LL01 (left) and LL02 (right). There was a significant interaction between stimulation mode and SOT condition. For LL01, the RMS distance of the COP was reduced with sensory feedback in condition six, indicating improved balance. For LL02, the reduction in RMS distance of COP was observed in conditions four and six. \* and \*\* denote p \< 0.05 and p \< 0.001, respectively.
For LL01 in SOT condition six, the *DIST*~*RMS *~were 1.3 ± 0.4 cm and 2.1 ± 0.4 cm for the sensory neuroprosthesis active and inactive, respectively, a statistically significant mean difference of 0.8 cm (95% CI 0.6 cm to 1.1 cm, p \< 0.001). Similarly, for LL02, the *DIST*~*RMS *~during condition six were 2.0 ± 0.4 cm and 2.9 ± 0.7 cm for sensory stimulation on and off, respectively, a statistically significant mean difference of 0.8 cm (95% CI 0.5 cm to 1.1 cm, p \< 0.001). Additionally, for LL02, the *DIST*~*RMS *~in condition four was 1.1 ± 0.2 cm and 1.4 ± 0.4 cm for electrically elicited sensory feedback on and off, respectively, a statistically significant mean difference of 0.3 cm (95% CI 0.1 cm to 0.6 cm, p = 0.018). In condition four, the RMS distances without sensory feedback were 0.9 ± 0.4 cm and 1.4 ± 0.4 cm for LL01 and LL02, respectively. This suggests that without sensory stimulation, LL02 had higher sway compared to LL01 (p=0.008) which might have contributed to the sensory stimulation effect seen for LL02 in condition four.
Effects of sensory stimulation on area of prediction ellipse {#Sec14}
------------------------------------------------------------
We also found statistically significant interactions between stimulation mode and SOT condition for the area of prediction ellipse in both participants (LL01: p = 0.003; LL02: p \< 0.001) that paralleled those for *DIST*~*RMS*~ (Fig. [5](#Fig5){ref-type="fig"}). For both participants, the area was significantly lower in condition six during trials with the sensory neuroprosthesis active. For participant LL02, electrically elicited sensory feedback also led to an improvement in condition four. Representative COPs and corresponding prediction ellipses are shown in Fig. [5](#Fig5){ref-type="fig"}.Figure 5Effects of sensory stimulation on area of prediction ellipse for LL01 (top left) and LL02 (bottom left). Representative COPs and corresponding prediction ellipses from SOT condition four are shown on the right (data from LL02). There was a significant interaction between stimulation mode and SOT condition. For LL01, the area of prediction ellipse was reduced with sensory feedback in condition six, suggesting an improvement in balance. For LL02, the reduction in the area of prediction ellipse was observed in conditions four and six. \* and \*\* denote p \< 0.05 and p \< 0.001, respectively.
In SOT condition six for LL01, the areas of the prediction ellipses were 9.3 ± 5.4 cm^2^ and 18.5 ± 2.0 cm^2^ for the sensory neuroprosthesis active and inactive, respectively. This represents a statistically significant mean difference of 9.2 cm^2^ (95% CI 5.4 cm^2^ to 12.9 cm^2^, p \< 0.001). Similarly, for LL02 in condition six, the mean areas of the prediction ellipses were 18.1 ± 6.4 cm^2^ and 38.6 ± 18.2 cm^2^ for electrically elicited sensory feedback on and off, respectively. This represents a statistically significant mean difference of 20.5cm^2^ (95% CI 14.6 cm^2^ to 26.4 cm^2^, p \< 0.001). In addition to condition six, LL02 exhibited a significant difference in prediction ellipse area for condition four. In this condition the mean areas of the prediction ellipses were 5.6 ± 2.2 cm^2^ and 12.5 ± 8.4 cm^2^ for sensory stimulation on and off modes, respectively, with a statistically significant mean difference of 6.93 cm^2^ (95% CI 1.03 cm^2^ to 12.84 cm^2^, p = 0.027). Without the sensory neuroprosthesis active, the area of prediction ellipse under SOT condition four was 6.1 ± 4.3 cm^2^ and 1.4 ± 0.4 cm^2^ for LL01 and LL02, respectively. This suggests LL02 had much higher fluctuations in his sway compared to LL01 (p = 0.01) without sensory stimulation.
Effects of sensory stimulation on weight symmetry {#Sec15}
-------------------------------------------------
There was no statistically significant interaction between stimulation mode and SOT condition on body weight percentage on the prosthesis (LL01: p = 0.809; LL02: p = 0.571). However, the follow up analysis of the main effect for stimulation revealed a statistically significant effect of stimulation across all conditions. During trials with the sensory neuroprosthesis active, participant LL02 increased the percentage of his body weight on the prosthesis by 2% (95% CI 1.1% to 2.8%, p \< 0.001) (Fig. [6](#Fig6){ref-type="fig"}). These results suggest that LL02 shifted more weight onto his prosthesis when he received sensory stimulation regardless of SOT condition. The follow up analysis of the main effects did not show any changes in body weight distribution between sensory stimulation modes for LL01 (p = 0.22). Compared to previously reported results with transtibial amputees aged 50 and older, our participants had a similar weight distribution on their prosthetic limbs^[@CR19]^.Figure 6Overall effects of sensory stimulation on weight symmetry across all SOT conditions. No significant interaction between stimulation mode and SOT condition were found on weight symmetry. However, there was a statistically significant effect of stimulation on weight symmetry regardless of SOT condition for LL02. \*\* denote p \< 0.001. In a study with 22 unilateral transtibial amputees aged 50 years or older, weight distribution during quiet stance was reported to be 44.4 ± 7.7% on the prosthetic limb^[@CR19]^.
Discussion {#Sec16}
==========
In this study, we demonstrated that sensations elicited in the missing foot of two transtibial amputees could decrease sway and improve balance when visual and vestibular inputs were incongruent and somatosensation in the intact foot was compromised. The sensations in the missing foot were elicited using a sensory neuroprosthesis that electrically activated nerves in the residual limb via implanted non-penetrating nerve cuff electrodes. The location and intensity of perceived sensations were determined and modulated according to prosthetic foot-floor contact pressure. Using this approach, we were able to examine the role of plantar somatosensory feedback from the missing foot during standing balance under challenging, dynamic conditions.
In both participants, we observed that the information from the sensory neuroprosthesis was most useful during condition six of the SOT, during which vestibular and visual inputs were incongruent and somatosensation in the intact leg was simultaneously perturbed. This improvement was seen in all three balance measures (ES, *DIST*~*RMS*~, and area of predicted ellipse), which demonstrates that not only were the maximum boundaries of sway reduced, but participants also remained steadier throughout the entire trial period with the neuroprosthesis active. Consistent with previous reports of naturally occurring sensory inputs, our findings show that participants utilized the most reliable sources of sensory information when others were compromised^[@CR47]^, including the perceptions of plantar sensation elicited by neural stimulation.
Our results confirm that LLAs adapt to lack of sensory input from their missing limb in part by relying on sensation from the intact leg. For LL02, we found the ES decreased during all three conditions (\#4--6) that perturbed somatosensation in the intact foot. A prior study showed that poor perception of vibration and pressure in the intact foot and ankle was associated with poor static and dynamic balance in dysvascular transtibial amputees^[@CR16]^. Moreover, it has been reported that LLAs use their intact limb to obtain sufficient sensory information for function^[@CR16],[@CR20]^. In a study by Miller *et al*., the number of reported falls per year for bilateral amputees was more than double that of unilateral amputees, suggesting that the loss of sensory input from both legs drastically increases fall risk^[@CR4]^. These observations suggest that sensory neuroprostheses may be the most beneficial for LLAs with poor intact limb sensation.
The differences in the effect of the sensory neuroprosthesis on outcome measures between the two participants can be explained mainly by how they prioritized other sensory inputs. LL01 also had equal or better ES compared to age-matched able-bodied controls, an indicator of good balance stability among traumatic transtibial amputees^[@CR2],[@CR48]^. Furthermore, LL01 was more stable without sensory stimulation in conditions four and five compared to LL02, which signifies that he may not have needed the additional sensory feedback as much and therefore did not utilize it in those conditions. However, LL02 found himself in a less stable situation; therefore, the electrically elicited sensory feedback resulted in an improvement in balance during the same conditions. Other factors such as residuum length^[@CR49]^ and choice of prosthetic foot^[@CR6]^ could have contributed to differences seen between two participants. Additionally, the amplitude for the surround and the platform movements during the SOT was chosen based on the confidence of each individual in controlling their balance. The difference in balance confidence between participants may also explain better sway measures for LL01 without the sensory neuroprosthesis active.
Maintaining balance is a complex sensorimotor function, which requires central processing of multiple sensory inputs at the vestibular nuclei^[@CR50]^. The CNS compares the sensory inputs against an internal model and attributes relative weights to them to generate appropriate motor responses^[@CR51]^. With reduced or conflicting sensory information, the motor performance is directly affected, and balance stability may subsequently become compromised^[@CR52]^. In LLAs, not only is the sensory information from the missing foot absent, but also the internal body model has changed as a result of the altered neuromuscular and sensorimotor systems following amputation^[@CR53]^. For example, it has been shown that plantar pressure sensations are used to update internal estimates of center of mass location, which is a key factor in balance stability^[@CR54],[@CR55]^. It is possible that the internal models of the participants in this study were updated after the first use of the sensory neuroprosthesis. Future studies should consider baseline measurements of balance prior to providing any electrically evoked somatosensations to LLAs to investigate if updates to internal model contribute to observed improvements in balance. Alterations in the internal model by prior exposure to sensory stimulation would further support the implications that the intact neuromuscular balance control apparatus interprets the electrically elicited sensations in a similar manner to naturally occurring sensory inputs, and utilized them effectively to help maintain standing balance and stability.
The sensory neuroprosthesis appeared to improve body weight symmetry in LL02 but it did not have any significant effects on weight distribution in LL01. This finding confirms that weight symmetry in LLAs could be affected by loss of sensation, however other variables such as prosthetic alignment, prosthetic foot design, socket fit, and even poor hip abductor muscle strength could play a role in this outcome measure^[@CR19],[@CR56],[@CR57]^. Moreover, several studies have shown that weight symmetry in LLAs is regained within eight weeks after first prosthesis use, and in many cases there is not much improvement beyond this period^[@CR6],[@CR58]^. Since the participants were long-term prosthesis users, their no-stimulation baseline symmetry values should have stabilized. Similarly, they were both exposed to sensory stimulation in the laboratory for a year prior to these experiments, thus the symmetries exhibited with the neuroprosthesis should have also plateaued. The time course of changes in symmetry due to the sensory neuroprosthesis can be the topic of future exploration. Lastly, sensory feedback affected sway measures differently than weight symmetry, suggesting that improvements in balance are not always correlated with a more symmetrical weight distribution^[@CR6]^.
It is likely that participants used the pressure exerted by the prosthetic socket on the residual limb to obtain information regarding movements of the support platform, and their own sway behaviors. However, the feedback through the socket and residuum is often not refined enough to compensate for the missing plantar sensation^[@CR16]^. Additionally, sensory feedback through the socket can vary based on changes in skin sensitivity^[@CR59]^, residual limb volume^[@CR60]^, liner material^[@CR61]^, and alignment^[@CR62]^. Furthermore, in dysvascular amputees, sensation through the socket could be limited due to diminished sensation in the residual limb due to the primary disease process^[@CR48]^.
In contrast to our approach that interfaces with remaining nerves in the residual limb to generate somatosensations directly referred to the missing foot, methods that utilize electro- or vibro-cutaneous input have attempted to provide indirect feedback regarding the status of the missing lower limbs^[@CR34],[@CR63]--[@CR70]^. However, only a few studies have examined the functional outcomes of such sensory substitution techniques with LLAs^[@CR34],[@CR67],[@CR71]^. Rusaw *et al*.^[@CR34]^ investigated the effects of vibratory feedback on static and dynamic balance in transtibial amputees by performing four out of the six conditions of SOT (Conditions 1--2 & 4--5). Four pressure sensors under the prosthetic foot were linked to four tactors located around the circumference of the thigh on the affected side. No improvements in any measures of sway were reported, suggesting that amputees were not able to effectively utilize the feedback functionally or integrate it into their balance control^[@CR34]^. Sabolich *et al*.^[@CR66]^ applied electrical stimulation to the skin of the residual limb based on the anterior and posterior loading conditions on the prosthetic foot. They reported improvement in weight distribution of transtibial amputees during static stance. However, the feedback did not result in any significant improvements in single-leg standing time, body weight symmetry, or step length symmetry during walking.
Although non-invasive approaches could be considered as preliminary tools to examine benefits of sensory feedback after limb loss, they impose limitations such as slow response time, inconsistencies based on changes in the skin-prosthesis interface, cumbersome donning and doffing, extended training times, and poor psychological acceptability and embodiment^[@CR69],[@CR71]--[@CR73]^. In addition to the apparent mismatch between the original and substituted sensory modality, a major limitation with sensory substitution is an abnormally long temporal delay between stimulus onset and conscious perception^[@CR73]^. The response times for vibrotactile devices mounted on residual limbs of lower-limb amputees are as long as a typical gait cycle^[@CR73]^, which makes this mechanism of sensory feedback impractical for balance and gait tasks that require rapid adjustments (i.e. responding to external perturbations). In addition, the detection threshold for vibrotactile or electrotactile stimuli could be greatly affected by factors such as the material of the prosthesis liner, mechanical properties of prosthetic components, skin condition, or movement of an electrode or actuator inside the socket^[@CR69],[@CR72],[@CR73]^. Furthermore, because sensory substitutive approaches do not result in sensations perceived as originating in the lost limb, users must be trained to associate the external stimulus with the applied load^[@CR71]^. Finally, the long-term functionality and acceptability of such systems have yet to be determined.
Our participants reported proprioception around the ankle during threshold and mapping experiments with our sensory neuroprostheses^[@CR23]^, however, they do not report proprioception when postural expectations are incongruent, i.e. when standing upright with a fixed prosthetic ankle. In this scenario, the participants are consciously aware that the ankle is locked; therefore, the elicited sensations are reported as muscle tightening around the ankle or perceived as contractions of the calf muscles. Future effort will focus on integrating the sensory neuroprosthesis with volitionally controlled prosthetic ankles, so that the ankle joint is a part of the sensory neuroprosthesis and participants can benefit from elicited proprioception in addition to plantar pressure sensation.
In able-bodied individuals, three main motor strategies are utilized to maintain balance during static and dynamic conditions^[@CR3]^. Movements at the ankle (i.e., the ankle strategy) are in response to small perturbations. Movements at the hip (i.e., the hip strategy) are often used to compensate for large perturbations. If there is a sudden change in the base of support in relation to the COG, then a stepping strategy is utilized to maintain balance^[@CR74]^. Because transtibial amputees are missing an ankle joint, they often use the hip joint to stabilize their COG in response to small perturbations of balance^[@CR75],[@CR76]^. In this case, accurate sensory feedback is still required to activate proper trunk rotation around the hip joint to maintain stability. However, if LLAs could control their prosthetic ankle joint to generate sufficient moment in response to sensory input, even greater improvements in balance could be expected from integrating a sensory neuroprosthesis with an active ankle.
Unilateral LLAs depend on visual feedback, the intact leg, and/or their upper bodies to control their posture during the early stages of rehabilitation post-amputation^[@CR53],[@CR77]^. Such dependency reduces over time as they learn to capitalize on remaining sensory inputs, but amputees still primarily depend on their intact limb as well as their vision to maintain balance control during static and dynamic tasks^[@CR2],[@CR3],[@CR78]--[@CR80]^. Sensory neuroprostheses may have the potential to reduce dependency on these resources and accelerate progress through post-amputation rehabilitation. The results of the experiments described here were based on limited use of a sensory neuroprosthesis in the laboratory. It is possible that with continuous use of the system at home and in the community, amputees could learn to rely on the new somatosensory input and use it even more effectively in controlling balance.
Although the time since amputation for the second participant was more recent (11 years for LL02 compared to 48 years in LL01), it was unlikely that differences in SOT outcomes between participants were due to the time of amputation. A prior study conducted with 15 unilateral transtibial amputees between 2--44 years post-amputation reported no effect of the time post-amputation on SOT outcomes^[@CR8]^. Both participants also received equivalent exposure to the sensory neuroprosthesis and had similar amounts and types of experiences with the system in the laboratory on a weekly basis for approximately 1.5 years prior to testing. Therefore, differences in prior exposure and practice are also unlikely to be the cause of the observed inter-subject variability in the outcomes.
The duration of the test was set based on clinical guidelines for the SOT. Although we did not investigate possible adaptation effects in this study, the time-course and magnitude of adaptation for touch elicited via electrical stimulation of the nerve through cuff electrodes are equivalent to natural, mechanically-induced sensations^[@CR81]^. This suggests that the underlying neural mechanisms for adaptation are similar between mechanically- and electrically-induced sensations. Furthermore, we implemented sensory stimulation that was proportional to pressure underneath the prosthetic foot (i.e., the applied electrical stimulation was modulated based on changes in plantar pressure). Therefore, participants did not receive a constant stimulus, which reduces the likelihood of adaptation due to the dynamic nature of the inputs to the nervous system^[@CR82]^.
Although participants in this study were transtibial amputees, other populations such as transfemoral amputees and elderly people exhibit comparable sensorimotor characteristics, which predisposes them to an increased risk of fall^[@CR2]^. It has been reported that when any two sensory inputs are simultaneously compromised in elderly people, a significant increase in sway occurs^[@CR52],[@CR83]^. As such, providing neural sensory stimulation to those who have compromised sensory perception in their lower limbs could be an effective way to improve standing stability in multiple user populations.
Conclusions {#Sec17}
===========
The functional benefits of a sensory neuroprosthesis for improving standing balance were documented by computerized dynamic posturography in two individuals with transtibial limb loss. Appropriately localized and modulated sensations of plantar pressures under the prosthetic foot were elicited by delivering stimulating currents directly to the nerves in the residuum via multi-contact non-penetrating cuff electrodes. We demonstrated these elicited sensations were integrated into the intact neuromuscular control system to reduce sway and increase stability in terms of variations in the Center of Pressure and Equilibrium Scores during perturbed standing. Symmetry of loads applied to the intact and prosthetic legs was also significantly improved with the information provided by the sensory neuroprosthesis. The sensory neuroprosthesis had the strongest impact on maintaining balance when other resources, such as vision, vestibular, or somatosensory inputs from the intact leg, were compromised. These findings indicate that the information provided by a closed-loop sensory neuroprosthesis employing implanted neural stimulation technology was processed by the central and peripheral nervous systems as if they arose from the missing limb to positively impact standing balance. The generalizability of these results on a larger sample of LLAs, and their implications on daily function in uncontrolled home and community environments, their impact on the incidence and risk of falls and losses of balance, and the potential benefits of integrating sensory stimulation with active or semi-active microprocessor controlled prosthetic ankle or knee joints remain to be determined.
**Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This material is based upon the work supported by the Defense Advanced Research Projects Agency (DARPA) and Space and Naval Warfare Systems Center Pacific (SSC Pacific) under Contract No. N66001-15-C-4038. In addition, this material is the result of work supported with resources and the use of facilities at the Louis Stokes Cleveland VA Medical Center. The authors would like to thank Clay Kelly, MD, for referring the participants, Gilles Pinault, MD for performing the implant surgery, Melissa Schmitt, RN, for coordinating the clinical aspects of the study, and Jennifer Kerbo for elegant illustrations of sensory neuroprosthesis. We also thank our participants for their time, patience, and dedication.
H.C. analyzed data and wrote the manuscript. H.C. and B.P.C. conducted the experiments and collected data. R.J.T. supervised the project. All authors conceived the experiments, discussed the results, and commented on the manuscript.
The data that support the findings of this study are available upon request from the corresponding author, H.C.
The authors declare no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
INTRODUCTION
============
Inflammation, defined as the immune system response to injury or infection, is a protective response that leads to the removal of initiating noxious stimuli or offending factors and the restoration of tissue structures and physiological functions. However, excessive chronic inflammation represents the basis of severe diseases including inflammatory bowel disease, arthritis, asthma, cancer, atherosclerosis, Alzheimer\'s disease, and Parkinson\'s disease ([@B1][@B2][@B3][@B4][@B5]). Therefore, controlling the inflammatory responses is important to prevent and treat many diseases. Nitric oxide (NO), a gaseous free radical, and prostaglandin E2 (PGE~2~), an eicosanoid derived from arachidonic acid, have been identified as important biological molecules involved in the immune responses and inflammation ([@B6][@B7]). Upon inflammatory stimulation, large amounts of NO and PGE~2~ produced by inducible NO synthases (iNOS) and cyclooxygenase 2 (COX-2), respectively, induce substantial inflammatory response and related processes ([@B6][@B8]). Therefore, the regulation of NO production and PGE~2~ generation is the \"gold standard\" for the prevention and treatment of many inflammatory diseases.
*Ixeris dentata* Nakai (IDN) has been used as a traditional herbal medicine in East Asian countries to treat indigestion, pneumonia, hepatitis, contusions, and tumors ([@B9]). Previous studies reported that IDN exhibits multiple pharmacological activities including antidiabetic, anticolitic, antiallergic, and neuroprotective properties ([@B10][@B11][@B12][@B13][@B14][@B15]). Furthermore, it has been reported that the IDN water extract possesses an anti-inflammatory activity against 2,4-dinitrofluorobenzene- induced atopic dermatitis-like skin inflammation ([@B16]). Previous studies demonstrated that several biologically active compounds including sesquiterpene lactones and flavonoids isolated from Ixeris dentate possesses multiple pharmacological activities on cancer ([@B9][@B17]), acylcoenzyme A: cholesterol acyltransferase (ACAT) activity ([@B9]), and inflammation ([@B13][@B18][@B19]). While the anti-inflammatory activities of IDN and its bio-active compounds have been reported previously, the effect of IDN on the production of pro-inflammatory mediators including PGE~2~ and NO has not yet been fully studied. Thus, in the present study, we evaluated the inhibitory properties of IDN methanol extract (MeOH-ex) and its solvent fractions on NO production and PGE~2~ generation in a murine macrophagelike cell line RAW264.7, which can be activated with LPS to induce pro-inflammatory mediators.
MATERIALS AND METHODS
=====================
Plant material, extraction, and fractionation
---------------------------------------------
The IDN was purchased from the local herbal market (Asan, Chungnam, Korea) and verified at the Department of Plant Resources, Soonchunhyang University (Asan, Chungnam, Korea). The air-dried IDN (2 kg) was powdered and exhaustively extracted with 99.8% MeOH (3×1.5 L) at room temperature for 48 h. The solution was filtered and concentrated under reduced pressure on a rotatory evaporator at 45℃, resulting in 205.8 g of crude MeOH-ex. The entire MeOH-ex (200 g) was suspended in 0.5 L of water and then partitioned sequentially with equal volumes of hexane (*n*-hexane), chloroform (CHCl~3~), and ethyl acetate (EtOAc). Each fraction was evaporated *in vacuo* to yield the residues of *n*-hexane, CHCl~3~, and EtOAc fractions. The MeOH-ex and the three fractions were tested for inhibitory effects against NO and PGE~2~ generation by using a murine macrophage cell line, RAW264.7.
Cell culture
------------
RAW264.7 cells, obtained from American Type Culture Collection (Manassas, VA, USA), were cultured in DMEM (Gibco-BRL, Grand Island, NY, USA) supplemented with 10% heat-inactivated FBS, 100 U/mL of penicillin, and 100 µg/mL streptomycin at 37℃ under humidified air containing 5% CO~2~ inside an incubator. Cells were plated in 35-mm culture dishes at a density of 6×10^5^ cells/dish for the following experiments.
NO assay
--------
RAW264.7 cells were plated at a density of 2×10^4^ cells/well in 96-well plates, and then stimulated with or without LPS (1 µg/mL) in the absence or presence of various concentrations of testing samples for 24 h. The nitrite accumulation in the supernatant was assessed by the Griess reaction ([@B20]). Each 50 µL of culture supernatant was mixed with an equal volume of Griess reagent (0.1% *N*-(1-naphthyl)- ethylenediamine and 1% sulfanilamide in 5% phosphoric acid) and incubated at room temperature for 10 min. The absorbance at 550 nm was measured using an automated microplate reader and a series of known concentrations of sodium nitrite was used as the standard.
Measurement of PGE~2~ generation
--------------------------------
RAW264.7 cells were plated at a density of 2×10^4^ cells/well in 96-well plates, and then stimulated with or without LPS (1 µg/mL) in the absence or presence of various concentrations of testing samples for 24 h. Vehicle (DMSO) only cultures served as controls. Cell culture supernatants were harvested and the generation of PGE~2~ in cell culture was measured using commercial ELISA kit (R&D Systems, Minneapolis, MN, USA) according to the manufacturer\'s instructions.
Cell viability assay
--------------------
The cytotoxic effect of the various samples after 24 h of incubation was evaluated in the cells using the MTT assay to determine the appropriate concentration that is not cytotoxic to the cells. Briefly, RAW264.7 cells were plated at a density of 2×10^4^ cells/well in 96-well plates and then treated with various concentrations of testing sample. Following incubation for 24 h, the culture medium was removed and replaced with 100 µL of fresh medium, and then 20 µL of 0.5 mg/mL MTT solution was added to each well. After incubation for 1.5 h, the medium with MTT was removed and 200 µL of DMSO was added to each well. The plates were then gently agitated until the color reaction was uniform and the colorimetric evaluation was performed with a microplate reader at 540 nm.
Western blot analysis
---------------------
Whole cell lysates were obtained as previously described ([@B21]). SDS-PAGE and western blot analysis were performed as described previously ([@B21][@B22]). Briefly, the cells were cultured in 6-well culture plates at a density of 5×10^5^ cells/well for 24 h. The cells were then stimulated with or without LPS (1 µg/mL) in the absence or presence of various concentrations of testing samples for 24 h. The cells were washed twice with PBS, lysed, and the proteins were separated on 10% to 15% SDS-PAGE. The proteins were transferred to polyvinylidene fluoride (PVDF) membrane, and membranes were blocked with 5% skim milk in TBST-buffer for 2.5 h at room temperature. The protein-transfer membranes were probed with the following primary antibodies: mouse monoclonal antibodies directed against rabbit polyclonal antibodies directed against iNOS (1:500, Santa Cruz Biotechnology, Dallas, TX, USA) and COX-2 (1:500, Santa Cruz Biotechnology). Protein expression was visualized using a chemiluminescence reagent (Amersham Pharmacia Biotech, Inc., Buckinghamshire, UK), and detected using a digital chemiluminescence imaging system equipped with a charge coupled device (CCD) camera (Fusion-FX, Fisher BioTec Ltd., Wembley, Australia).
Statistical analysis
--------------------
Statistical analyses were carried out using the GraphPad Prism 5.0 software (GraphPad Software Inc., La Jolla, CA, USA). Pairwise comparisons were performed using oneway ANOVA Dunnett\'s tests. Data are presented as mean±SEM in the indicated number of experiments. The differences between groups were considered significant at p-value below 0.05.
RESULTS AND DISCUSSION
======================
IDN EtOAc fraction potently inhibits NO production and PGE~2~ generation in LPS-stimulated RAW264.7 cells
---------------------------------------------------------------------------------------------------------
One hallmark of chronic inflammation is the continuous recruitment and activation of macrophages to the sites of inflammation. Macrophages play a central role in inflammatory responses through phagocytosis, antigen presentation, and immunomodulation ([@B23]). Activated macrophages produce a wide variety of pro-inflammatory mediators, such as NO and PGE~2~ which are generated by iNOS and COX-2, respectively ([@B24]). Excessive release of these pro-inflammatory mediators from activated macrophages has long been recognized as a risk factor for inflammatory diseases ([@B25]). To evaluate the anti-inflammatory property of IDN, we initially evaluated the effect of IDN MeOH-ex and its solvent fractions on LPS-induced NO production in murine macrophage-like cell line, RAW264.7. In contrast to the LPS control group, IDN MeOH-ex treatment significantly inhibited the LPS-induced NO production in a concentration-dependent manner ([Fig. 1A](#F1){ref-type="fig"}). We further found that treatment with *n*-hexane fraction (n-hexane-fr, [Fig. 1B](#F1){ref-type="fig"}), CHCl~3~ fraction (CHCl~3~-fr, [Fig. 1C](#F1){ref-type="fig"}), and EtOAc fraction (EtOAc-fr, [Fig. 1D](#F1){ref-type="fig"}) effectively inhibited NO production in a concentration-dependent manner with IC~50~ values of 102.6, 87.8, and 16.4 µg/mL, respectively. In addition, we observed that the LPS-induced PGE~2~ generation in RAW264.7 cells was also significantly decreased by n-hexane-fr ([Fig. 2B](#F2){ref-type="fig"}), CHCl~3~-fr ([Fig. 2C](#F2){ref-type="fig"}), and EtOAcfr ([Fig. 2D](#F2){ref-type="fig"}) with IC~50~ values of 94.4, 123.7, and 85.8 µg/mL, respectively, while MeOH-ex showed a non-dose-dependent inhibitory activity ([Fig. 2A](#F2){ref-type="fig"}). Our results indicated that the IDN EtOAc-fr has a potent inhibitory activity on LPS-stimulated inflammatory responses in macrophages. Previous studies have also reported that TNF-α, IL-6, and IL-1β levels were significantly lowered by IDN water extract treatment in acute colitis and dermatitis ([@B12][@B16]). Taken together, these results suggested that the IDN extract possessed an effective anti-inflammatory property against multiple inflammatory responses.
IDN EtOAc fraction effectively suppresses iNOS and COX-2 expression in LPS-stimulated RAW264.7 cells
----------------------------------------------------------------------------------------------------
Since EtOAc-fr was identified to have the most potent anti-inflammatory property against LPS-induced inflammation, we further evaluated its anti-inflammatory mechanism. Previously, it has been demonstrated that treatment with IDN water extract effectively reduced the expression of pro-inflammatory proteins, such as NF-κB and MAPKs ([@B12][@B16][@B18]), but the effects of IDN MeOH-ex and its solvent fractions on the expression of pro-inflammatory proteins regulating NO production and PGE~2~ generation have not been extensively investigated. To investigate the possible mode of action, we tested the effect of IDN EtOAc-fr on iNOS and COX-2 expression in LPS-stimulated RAW264.7 cells. Stimulation with LPS upregulated the expression of iNOS and COX-2 in cells compared with the unstimulated controls. We found that treatment of cells with IDN EtOAc-fr significantly suppressed LPS-induced iNOS expression in a dose-dependent manner ([Fig. 3A](#F3){ref-type="fig"}). This result suggests that the IDN EtOAc-fr inhibits LPSinduced NO production and PGE2 generation through suppression of iNOS and COX-2 expression. To examine whether the anti-inflammatory activity of IDN EtOAc-fr in RAW264.7 cells is attributable to its cytotoxicity, we examined the effect of IDN EtOAc-fr treatment on cell viability. Importantly, an MTT assay using RAW264.7 cells treated with IDN EtOAc-fr alone (up to 200 µg/mL) showed that the cell morphology ([Fig. 3D](#F3){ref-type="fig"}) and cell viability ([Fig. 3E](#F3){ref-type="fig"}) were not significantly altered. These results demonstrated that IDN EtOAc-fr exhibited anti-inflammatory activity without affecting cell viability.
In conclusion, we identified the EtOAc-fr of IDN MeOH-ex as an anti-inflammatory substance that effectively inhibited LPS-induced NO production and PGE~2~ generation by suppressing iNOS and COX-2 expressions in RAW264.7 cells. Therefore, the bioactive components of IDN EtOAc-fr may be promising natural anti-inflammatory compounds for the prevention and treatment of multiple inflammatory diseases. Further studies are required to identify the bioactive compounds of IDN.
This work was supported by the Soonchunhyang University Research Fund.
**CONFLICTS OF INTEREST:** The authors have no financial conflict of interest.
IDN
: *Ixeris dentata* Nakai
MeOH-ex
: methanol extract
EtOAc-fr
: ethyl acetate fraction
PGE~2~
: prostaglandin E~2~
iNOS
: NO synthases
COX-2
: cyclooxygenase 2
{#F1}
{#F2}
{#F3}
[^1]: ^\#^These authors contributed equally to this work.
| {
"pile_set_name": "PubMed Central"
} |
Abstract {#s1}
========
Incidents of professional failure and the necessity to improve efficiency and quality of care in the health service have led to increasing demand for quality assurance and audits of medical institutions.[@R01]-[@R05] This has allowed quality appraisal and optimal targeting of resources to areas of need. These processes have led to significant improvements in health outcomes; however, variation in hospital performance remains.[@R05],[@R06]
A widely used and acceptable method to control variation in health outcomes is based on case mix adjustment.[@R07]-[@R09] However, failure to adjust appropriately for differences in case mix may result in unfairly targeting hospitals admitting high-risk patients. Indeed, the identification of hospitals having unusual performance depends on the variables used in the risk-adjustment model.[@R07],[@R08] Furthermore, comparing hospitals on the basis of a risk-adjustment model could be erroneous, as the risk model may be wrong, or suffer from incorrect inclusion of prognostic factors.[@R04]
More importantly, the disparity in risk-adjusted outcomes may result from a variety of factors including definitions, data quality, structural and institutional management factors, and resource characteristics that have a direct effect on clinical processes.[@R04]-[@R06] To this end, differences in case mix should be accounted for in a suitable risk-adjustment model and differences in definitions and data quality kept to a minimum. Any residual variation in outcome between hospitals would therefore reflect hospital quality of care, the basis for medical institutional profiling methods.[@R07]-[@R15] However the extent to which these hopes are satisfied remains uncertain.
There is a large literature base on statistical methodology for health provider profiling.[@R10]-[@R13] Simple methods use ratios of the observed to the expected outcomes (indirect standardisation) or odds ratios from a logistic regression analysis.[@R08],[@R15] A number of studies have shown disagreements between different frequentist or Bayesian methods for profiling hospital performance (Marshall and Spiegelhalter,[@R11] Austin,[@R12] Ohlssen *et al.*,[@R15] Delong *et al.*[@R16] and Leyland and Boddy[@R17]). In particular, random-effects models are found to be more conservative in classifying institutions as performance outliers.[@R11] There is therefore a need for research to identify statistical models and ways that robustly differentiate between hospitals and remain meaningful to the medical practitioner.[@R12]
Normand *et al.*,[@R10] Marshall and Spiegelhalter,[@R11] Austin[@R12] and Ohlssen *et al.*[@R15] advocated using hierarchical Bayesian random-effects methods for provider profiling. These methods easily permit data pooling across institutions; thus overcoming uncertainty associated with small institutions, which might be outliers by chance alone.[@R12] Estimated performances are stabilised and shrunk towards the population average; the degree of shrinkage being larger for small hospitals than for large hospitals.
Bayesian methods provide complete probabilistic information in determining the probability that a hospital-specific risk-adjusted rate exceeds a specified threshold.[@R11] Furthermore, a researcher is able to place credible intervals on the derived ranks to quantify the uncertainty associated with institutional ranking before relative performance can be assessed.[@R11],[@R18]
In the current study, rather than calibrate the methods, we concentrated on comparing the performance of four methods and assessing how well they agreed with one another, using Marshall and Spiegelhalter,[@R11] and Austin's approaches.[@R12] The methods were applied to data on short-term mortality in acute coronary syndrome (ACS) patients. The data are part of the Myocardial Infarction National Audit Project (MINAP), which currently reports percentage attainment of standards on five clinical process variables, namely door-to-needle and call-to-needle thrombolysis times, and the use of aspirin, beta-blockers and statins post-acute myocardial infarction (AMI).[@R19],[@R20] A use of the MINAP data for hospital comparison was presented in Gale *et al*.,[@R05] using funnel plots on the same five process variables.
To the best of our knowledge, the present study is the first to use an outcome measure and to control for any variation, specifically for case mix, with contemporary data on ACS. We did not duplicate MINAP tabulations or the Gale *et al*.[@R05] funnel plot methodology. Instead, we determined (a) whether or not a hospital's risk-adjusted mortality rate exceeded a specified threshold, and (b) the hospital's rank, based on its risk-adjusted mortality rate using two statistical models: fixed and hierarchical models, on the number of deaths among patients admitted by the hospital. While this article does not add sufficient new methodological questions on profiling methods, the topic of healthcare performance is timely, important and interesting within the medical and health services domain.
Methods {#s2a}
=======
MINAP was established in 1998. It is reported to be the largest and most comprehensive clinical database of ACS care in the world and is a valuable resource for monitoring coronary heart disease audit standards for patients presenting with AMI in England and Wales.[@R20] All hospitals in England and Wales that treat patients with acute AMI submit data to MINAP. The project collects information on the quality of care and outcome of patients. Each patient entry offers details of the patient's journey, including the method and timing of admission, in-patient investigations, results and treatment, and, if applicable, dates of death from linkage to the Office of National Statistics, United Kingdom.
Prospective data are collected locally, electronically encrypted and transferred to a central database. The database may be used for identifying performance indicators to identify examples of good practice. With such data, it is feasible to evaluate contemporary care practices consistent with national guidelines for the management of ACS, investigate whether hospital performance varies between hospitals, identify hospital characteristics predictive of adherence to guidelines, and assess whether adherence to guidelines is associated with mortality rates.[@R07]
We examined all 187 069 ACS events entered into the MINAP database from 1 January 2003 to 31 March 2005. We selected first (index) admissions reported to MINAP and therefore excluded re-admissions. We then analysed all patients who were aged between 18 and 100 years, who had an admission systolic blood pressure between 49 and 250 mmHg, and an admission heart rate between 20 and 200 beats/min.
In total there were 134 hospitals, six of which were discarded from the analyses because of sparse data, i.e. not sufficiently varied (two with one admission, three with fewer than five deaths, one with excessive missing codes on death status). For the remaining 100 686 patients, the overall in-hospital mortality rate was 8.1%, the total mortality rate was 17.8%, and the 30-day mortality rate was 10.2%. Hospital-specific 30-day mortality rates ranged from 5 to 21%, with a median rate of 8.3%.
Statistical models {#s2b}
==================
We assumed that *O*~i~ is the observed number of 30-day deaths in patients admitted to hospital *i* (*i* = 1, ..., 128) and *E*~i~ is the expected number of deaths, given the case mix of its patients. The number of deaths in the period 1 January 2003 to 31 March 2005 can be assumed to follow a Poisson distribution with unknown mean *λ*~i~. Therefore *O*~i~ \~ Poisson (*λ*~i~) taking log *λ*~i~ = log *E*~i~ + θ~i~ where log *E*~i~ is an offset that adjusts for the patient effects and θ~i~ is a residual representing hospital-specific effect of interest.
The expected number *E*~i~ is obtained from a logistic regression on the pooled data, adjusting for relevant risk factors, to determine each patient's predicted probability of 30-day mortality. These probabilities are then summed within a hospital to give the expected number of deaths at that hospital, given its case mix. The hospital-specific effect θ~i~ is the log-relative risk or logarithm of the hospital's standardised mortality ratio (log SMR).
Other than to compare hospitals using their SMRs, we used the hospital risk-adjusted 30-day mortality rate (RAMR),[@R07] defined as RAMR = µ~30~ exp (θ~i~), where µ~30~ (= 10.2%) is the overall 30-day mortality rate. The RAMR can be thought of as the estimated 30-day mortality rate for a hospital admitting a population of patients identical to the overall case mix.[@R11] We adopted Bayesian methods in estimating a hospital-specific random effect θ~i~ to obtain its specific risk-adjusted mortality rate using 10.2 × exp (θ~i~), which we used in this study for institutional profiling.
In order to estimate the hospital-specific effect, we firstly assumed that it has a prior normal distribution with mean 0 and variance 1 000. This is the fixed-effects model, and the prior distribution implies that the hospital-specific standardised mortality rate has a prior mean of 1.
Secondly, as an alternative, we considered a Bayesian random-effects model, which, by using hierarchical modelling, pools data across hospitals. This approach produces more reliable estimates of hospital performance, in that genuinely low or high hospital outliers are identified. It reduces the chance of a small hospital being classified as an outlier by chance alone.[@R11]
Under the latter modelling approach, it was assumed that the hospital-specific random effects θ~i~ were drawn from a normal distribution with an unknown mean µ~0~ and variance σ~0~^2^. Therefore, θ~i~ \~ Normal (µ~0~, σ~0~^2^), where the hyper-parameters µ~0~ and σ~0~^2^ were the underlying overall log-standardised mortality ratio and between-hospital variance, respectively.
In order to complete the Bayesian implementation of the model, we also needed to specify prior probability distributions for the hyper-parameters µ~0~ and σ~0~^2^ for the hospital-specific random effects, θ~i~ distribution. The hyper-mean, µ~0~ was assigned a normal distribution with mean of 0 and variance 1 000. The hyper-precision, σ~0~^-2^ (inverse of the hyper-variance, σ~0~^2^) was given a gamma distribution with shape and scale parameters both equal to 0.001; implying that the hyper-precision had a mean of 1 and variance 1 000. This prior translates into a locally uniform distribution on the logarithm of the hyper-variance.
We used two ways of identifying outliers; one based on the hospital's RAMR, and the other based on the rank of RAMR among all the hospitals' RAMR. Assessments of agreement were initially based on point estimates between each hospital's ranks, and between risk-adjusted mortality rates. These pairwise agreements could be assessed using Bland--Altman plots.[@R21] However, we used simple two-way scatter plots, where agreement was judged against the line of equality.
We concentrated on categorising the different classification outcome measures into low, normal or high mortality risk, and then assessing agreement across the categories. In categorising a hospital's RAMR, we examined the probability of it exceeding a specified threshold. The overall 30-day mortality rate was 10.2% for our patient cohort. A hospital *i* is classified as a high outlier if Prob \[RAMR~i~ \> (1 + σ) 10.2\] ≥ 0.75 and, similarly, it is classified as a low outlier if Prob \[RAMR~i~ \< (1 -- σ) 10.2\] ≥ 0.75, otherwise the hospital is classified as normal.
The threshold value δ can take any value, but values of 10, 15 and 20% are commonly used.[@R18] We conservatively chose δ to be 20%, which has the effect of minimising the number of outlying hospitals, therefore hospital *i* is a high outlier if Prob (RAMR~i~ \> 12.24) ≥ 0.75, and a low outlier if Prob (RAMR~i~ \< 8.16) ≥ 0.75.
For ranks, we calculated Bayesian point estimates and 95% credible intervals of each hospital's rank. Hospitals whose 95% intervals fell entirely in the bottom or upper quartile of ranks (i.e. upper limit is ≤ 32.75 or lower limit is ≥ 96.25) were classified as low or high outliers, respectively; otherwise they were normally performing hospitals.
With two modelling approaches (the fixed- and random-effects models) plus two ways of classifying hospital performance, we had four different methods for profiling hospitals. In all, there were six possible pair-wise comparisons.
For each comparison, we used the kappa (κ) statistic to assess the amount of agreement between the methods. The statistic measures the proportion of observed-to-expected agreement, and we adopted the convention that κ \> 0.75 indicates excellent agreement, κ = 0.4--0.75 indicates good agreement, and κ \< 0.4 indicates marginal agreement,[@R22] even though κ has been criticised for its limitations. In order to allow for different levels of agreement, we used a weighted κ statistic.
Implementation {#s2c}
==============
The computation of the models was done using Markov Chain Monte Carlo methods (MCMC); specifically we used Gibbs sampling as implemented in WinBUGS.[@R23] For each method considered, three parallel Gibbs sampler chains from independent starting positions were run for 50 000 iterations. We monitored 10 randomly chosen random effects, and for hierarchical models also hyper-parameters for convergence.
Trace plots of sample values of each of these parameters showed that they were converging to the same distribution. We formally assessed convergence of the three chains using Gelman--Rubin reduction factors,[@R24] and all were estimated near 1.0 by 15 000 iterations. We therefore took 15 000 iterations to be in the burn-in period. For posterior inference, we used the remaining 35 000 iterations to give a combined sample size of 105 000.
Results {#s3}
=======
Existing ACS risk scores include a multitude of factors. Patient age, systolic blood pressure (SBP), heart rate (HR) at admission and ECG findings are systematically included in most of the risk-scoring systems.[@R25]-[@R27] In a large sample of European patients with ACS, age was found to impact on most of the clinical presentations and on hospital mortality.[@R28] Therefore inclusion of age in a risk model would account for many of the baseline, prior and clinical risk factors.
The risk variables that we used in the case mix logistic regression model for the risk adjustment are presented in [Table 1](#T1){ref-type="table"}, where age cut-off points were based on Resengren *et al*.,[@R28] and SBP and HR on their fifths. The fitted model had an estimated *c*-statistic (area under the ROC curve) of 0.798, with a 95% confidence interval of 0.794 to 0.803. The inclusion of co-morbidities (e.g. diabetes and chronic renal failure) resulted in loss of data and minor improvement on the *c*-statistics. Using only age, SBP and HR, whether continuous or categorised, resulted in a similar value of the c-statistic of 0.777 (0.772--0.781).
###### The Risk-Adjustment Model Of 30-Day Mortality Using Baseline Risk Factors, Discharge ECG Findings And Biochemical Markers
*Risk factor* *Number of patients* *Number of deaths (%)* *Odds ratio (95% CI)*
------------------------ ---------------------- ------------------------ -----------------------
Age group (years)
\< 55 14 116 233 (1.7) 1.00
55--64 16 396 549 (3.4) 2.02 (1.72--2.37)
65--74 21 442 1 703 (7.9) 5.06 (4.38--5.84)
75--84 23 006 3 656 (15.9) 10.73 (9.33--12.34)
≥ 84 9 249 2 259 (24.4) 18.03 (15.61--20.83)
SBP (mmHg)
\< 117 16 609 3 082 (18.6) 1.00
117--132 16 745 1 716 (10.3) 0.56 (0.52--0.60)
133--146 16 458 1 354 (8.2) 0.43 (0.40--0.46)
147--164 17 072 1 161 (6.8) 0.33 (0.31--0.36)
≥ 165 17 325 1 087 (6.3) 0.27 (0.25--0.29)
Heart rate (beats/min)
\< 62 18 135 1213 (6.7) 1.00
62--72 15 538 991 (6.4) 1.10 (0.99--1.20)
73--83 16 836 1 373 (8.2) 1.38 (0.27--1.51)
84--98 16 600 1 905 (11.5) 1.84 (1.70--2.00)
≥ 99 17 100 2 918 (17.1) 2.55 (2.36--2.75)
Discharge diagnosis
ST elevation 29 389 3 612 (12.3) 8.59 (6.09--12.11)
Non-ST elevation 29 462 3 379 (11.5) 5.29 (3.75--7.47)
Tropin positive 6 719 368 (5.5) 2.59 (1.81--3.71)
Tropin negative 6 326 58 (0.9) 0.67 (0.43--1.02)
Chest pain 3 136 34 (1.1) 1.00
Other
Total 84 209 8 400 (9.98) 4.68 (3.29--6.67)
Using this predictive model of 30-day mortality shown in [Table 1](#T1){ref-type="table"}, we evaluated the expected number of deaths, *E*~i~ in hospital *i* to obtain its standardised mortality ratio, SMR~i~ = *O*~i~ *E*~i~ and risk-adjusted mortality rate, RAMR~i~ = 10.2 × SMR~i~, which ranged from 4.54 to 19.44% with a median of 9.91%.
[Table 2](#T2){ref-type="table"} shows the top and bottom five ranked hospitals according to their risk-adjusted 30-day mortality rate. The top or bottom ranked 10 hospitals were more or less the same using only age, SBP and HR but with a slightly longer range, 4.14 to 23.32%.
###### Observed, Expected And Risk-Adjusted 30-Day Mortality Rate After ACS Admission, 2003--2005, England And Wales
*Hospital* *Number of admissions\** *Observed deaths* *Expected deaths* *RAMR (95% CI)*
------------- -------------------------- ------------------- ------------------- ----------------------
Top five
1 737 39 89.65 4.54 (3.32--6.21)
2 167 5 10.58 4.82 (2.01--11.58)
3 232 9 18.99 4.83 (2.52--9.29)
4 209 10 20.10 5.07 (2.73--9.43)
5 2 158 71 123.56 5.86 (4.64--7.40)
Bottom five
124 289 42 27.43 15.62 (11.54--21.13)
125 24 5 3.21 15.90 (6.62--38.19)
126 21 4 2.50 16.31 (6.12--43.44)
127 348 63 37.45 17.16 (13.40--21.96)
128 97 19 9.97 19.44 (12.40--30.48)
\*With a valid 30-day status.
Comparisons of agreement between a hospital's risk-adjusted mortality rates and between ranks of the risk-adjusted mortality rates from fitting the fixed- and random-effects models are shown in [Fig. 1A, B](#F1){ref-type="fig"}. For each plot, lines of equality are shown, and comparisons are based on posterior medians. The observed agreement appears to be very poor between the risk-adjusted mortality rates. On the other hand, for the ranks, the points lie evenly around the line of unity, showing very good agreement.
{#F1}
In both plots, agreement is very poor between outcome measures for either low or high outlying hospitals. Furthermore, the plots show that estimated outcome measures are more variable under the fixed-effects model.
The problems observed from using point estimates for assessing agreement can be partially nullified by categorising the hospitals into low, normal and high performing. Comparisons based on categories of risk between different methods are shown in [Table 3](#T3){ref-type="table"}. All methods were able to classify hospitals as lowand high-outcome outliers; however, only seven and 11 from 128 were classified as such under the hierarchical rank and RAMR methods, respectively, while 31 and 33 were outliers under the fixed-effects rank and RAMR methods, respectively. As expected, profiling methods using hierarchical models were more conservative in classifying hospitals as performance outliers than were the non-hierarchical models.
###### Classification Of Hospitals Under The Fixed And Hierarchal Models
*Fixed RAMR* *Fixed rank* *Hierarchical RAMR*
------------------- -------------- -------------- --------------------- ---- -------- ---- --- -------- ---
Fixed RAMR
Low -- -- -- 20 0 0 6 14 0
Normal -- -- -- 7 88 0 0 95 0
High -- -- -- 0 9 4 0 8 5
= 0.71 = 0.46
Fixed rank
Low -- -- -- -- -- -- 6 21 0
Normal -- -- -- -- -- -- 0 96 1
High -- -- -- -- -- -- 0 0 4
= 0.44
Hierarchical rank
Low 2 0 0 2 0 0 2 0 0
Normal 18 95 8 25 96 0 4 117 0
High 0 0 5 0 1 4 0 0 5
= 0.32 = 0.29 = 0.77
The observed agreement in the methods' classification of hospitals ranged from 90 to 98% of the time, the highest being between the two hierarchical methods. In only one of the six comparisons was agreement excellent, as reflected by the κ statistic of 0.77. In three cases, the agreement was moderate (0.40 \< κ \< 0.75). In the remaining two cases, the agreement was only marginal (κ = 0.29--0.32), and these involved comparisons of the random-effects rank and fixed-effects methods. The cross tabulations in [Table 3](#T3){ref-type="table"} are in close agreement with those obtained when using only age, SBP and HR in the risk-adjustment model, an indication that our results are insensitive to which factors are included in the risk-adjustment model.
The results presented here are based on arbitrary choices. In particular, the prior for the between-hospital variation is critical as it dictates how much shrinkage is assumed in the individual hospital estimates.[@R29] However, there is no standard solution to the problem of choosing a prior on the random-effects variance in hierarchical models. In standard Bayesian analyses, the inverse-gamma prior family is preferred because of its conditional conjugate properties, which allows ease of mathematical derivations. But this prior has been shown to give biased results.[@R30]
On the other hand, the threshold values for RAMR have an influence of the number of hospitals classified as outliers. We performed a limited-sensitivity analysis to find out the extent to which the choices impact on the results. We used a uniform (0, 100) prior on the random-effects standard deviation σ~o~ and 15% for the threshold value δ. The uniform prior produced exactly the same classifications of the hospitals as the inverse-gamma prior on the random-effects variance. Using a threshold of 15% affected only the 117 hospitals that were previously classified as normal, and now two were classified as low outliers and five as high outliers. Our results were therefore not affected by changes in random-effects variances but slightly so when the threshold value was changed.
Discussion {#s4}
==========
This study compared the performance of four methods for profiling hospitals and assessed their agreement. The methods included combinations of two Bayesian methods, fixed and hierarchical, and two ways of identifying outliers, rank and exceeding some threshold using a hospital's risk-adjusted mortality rate; two were based on a hospital's rank for its risk-adjusted mortality rate, obtained from fitting both fixed- and random-effects models. The agreements between the different methods were empirically examined using an extensive dataset of ACS patients.
Even though all the methods were able to classify hospitals as low- and high-outcome outliers, profiling methods using random-effects models were more conservative than fixed-effects models in classifying hospitals as having better- or worse-than-expected mortality. These findings were expected on theoretical grounds and support the results from a multitude of prior studies, showing that random-effects models identify fewer performance outliers.[@R08],[@R11]
In the present study, the observed agreement in the methods' classification of hospitals ranged from 90 to 98%, the highest being between the methods within each effects model. The agreement was excellent (κ = 0.77) in only one of the six comparisons. Otherwise, in all the remaining five scenarios, the agreement was, at best moderate (κ \< 0.75).
Our findings relied on routinely collected clinical data. These types of data suffer from incompleteness and inaccuracy of the variables entered.[@R31] In our preliminary investigation, 11% of the total patients had missing codes on survival status. We did not have full data for admission age, SBP, HR, ECG findings and biochemical markers of the patients. Other risk variables that may have been used also demonstrated missing data, thus limiting the number of risk factors in the case mix adjustment model on this occasion. However, our findings were shown to be robust to which factors were included in the risk-adjustment model. Indeed, difficult-to-obtain key clinical variables add little to the predictive power of ACS risk scores.[@R27]
It may well be that the hospital performance variation exhibited in this study was substantially contributed to by the variation in definitions and data quality, as alluded to by Lilford *et al*.[@R04] However, it is unlikely that these issues alone could be attributed to the outcome variation found across the four analytical strategies examined.
We did not impute for missing data since other researchers have shown that this does not affect the prediction model or mortality.[@R32] A more elaborate assessment of MINAP data quality and validity on the resulting classification of hospitals is the subject of a British Heart Foundation-funded project within our group undertaken by Gale *et al*.[@R33] For the present study, it suffices to say that the number of patients analysed and the data used were of sufficient quality to enable a comparison of different methods to assess the hospitals' performance for 30-day mortality among ACS patients. However, we remain cautious regarding the exact inference made for some hospitals, given their data quality.
We performed a limited-sensitivity analysis to different prior specifications of the hospital random-effects variation and threshold values. We found classification of outlying hospitals was not affected by changes in the random-effects variations, but it was slightly affected when the thresholds were changed.
A more elaborate sensitivity analysis would alter specification of the hospital random-effects distribution as the assumed normal distribution is not robust and flexible enough to account for outlying hospital effects. Therefore it may be necessary in future research to model the hospital effects more flexibly, for example by heavy-tailed *t* distributions to investigate both sensitivity and robustness of the results, as in Manda,[@R34] or mixtures or non-parametric Dirichlet distributions, as in Ohlssen.[@R35]
The threshold level chosen and the required probability of exceeding this threshold to classify a hospital using the risk-adjusted mortality rate as an outlier were subjective and completely arbitrary. We could have used other thresholds and probabilities, as in Austin,[@R12] which may have generated stronger or weaker levels of agreement between the methods. Furthermore, the requirement that intervals of the ranks must lie entirely in the bottom or top quarters of ranks for the hospital to be classified as an outlier was also arbitrary but has been used before.[@R11],[@R12]
Results from any study on profiling hospitals' performance are predictably used to produce league tables of performance. We are aware of the many criticisms surrounding the statistics used in measuring performance and the subsequent ranking of hospitals. We did not intend to contribute to this controversy. Our aim was to describe and compare the performance of four different Bayesian methods for institutional profiling. In using ranks to compare hospitals, caution should be exercised since most hospitals had considerably overlapping intervals, which made it difficult to obtain reliable ranking, especially for hospitals admitting fewer patients.
We follow Normand *et al*.,[@R10] Marshall and Spiegelhalter,[@R11] Austin[@R12] and Ohlssen *et al*.[@R18] in advocating the use of Bayesian methods, which when pooling data across hospitals, handles the problem of small hospitals better than frequentist methods, for which a minimum number of patients is required before a hospital can be included.[@R12] However, if we are willing to accept wide confidence intervals, the exact probabilistic methods can be used within a frequentist framework to handle small hospitals (see Luft and Brown[@R36]). Furthermore, it is much easier within Bayesian methods to determine uncertainty associated with the ranks, which are very sensitive to sampling variations (see Marshall and Spiegelhalter[@R11] and Greenwood[@R18]).
The main interest of this work was not to find the best model for hospital profiling, but to investigate whether or not the methods agree. In order to inform which method gives a better fit would require other model-checks statistics, such as posterior predictive checks.
Conclusion {#s5}
==========
The main overall finding from our example is that the choice of ways to classify a hospital is less critical than the statistical method used. We suggest profiling hospitals using a hierarchical model and RAMR with an appropriate threshold, which seems to offer more reliable results. However, these methods warrant further investigation, possibly of simulated data sets in which the impact of underlying assumptions (and derivation thereof) may be evaluated. There is a need for robust systems of 'regulation' or 'performance monitoring', which, with more rigorous work, we hope to achieve in the future.
We thank Dr JS Birkhead, clinical director of MINAP, the National Audit of Myocardial Infarction, National Institute for Clinical Outcomes Research, and the Heart Hospital, London, for providing the extract from MINAP. We acknowledge all hospitals in England and Wales for their contribution of data to MINAP. We also thank Darren Greenwood for helpful comments.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#Sec1}
============
Perrault syndrome (PRLTS) is a rare autosomal recessive disorder that is characterised by ovarian dysgenesis in females and sensorineural hearing loss (SNHL) in both genders^[@CR1]^. The disease is clinically heterogeneous and in severe cases, additional symptoms may include ataxia, neuropathies and intellectual disability. To date, mutations in six different genes have been linked to the disease: PRLTS1 is caused by compound heterozygous mutations in *HSD17B4* (17-beta hydroxysteroid dehydrogenase); PRLTS2 results from compound heterozygous mutations in *HARS2* (mitochondrial histidyl-tRNA synthetase); PRLTS3 is caused by homozygous or compound heterozygous mutations in *CLPP* (mitochondrial protease); PRLTS4 results from homozygous or compound heterozygous mutations in *LARS2* (mitochondrial leucyl-tRNA synthetase); PRLTS5 is caused by homozygous or compound heterozygous *C10orf2* (mitochondrial DNA helicase Twinkle) and PRLTS6 has been recently linked to a missense mutation in *ERAL1* (mitochondrial chaperone required for mitochondrial ribosome assembly). Most of the causative genes for PRLTS (including *CLPP*) are implicated in the maintenance of proteostasis in mitochondria, in particular mitochondrial protein translation^[@CR1]--[@CR9]^. Consistent with an important role for CLPP in Perrault syndrome, *CLPP* null mice (*CLPP*^−/−^) also display infertility, deafness and growth retardation^[@CR10]^.
In humans, CLPP is a nuclear encoded mitochondrial protease that is directed to the mitochondrial matrix through an N-terminal targeting sequence^[@CR11]^. Following import and processing, mature CLPP is proposed to assemble, initially into a ring-shaped heptamer^[@CR12]^ and finally, in the presence of its cognate ATP-dependent unfoldase -- CLPX, into an active barrel-shaped tetradecamer^[@CR13]^. Based on studies with bacterial ClpXP homologs, formation of the human CLPXP complex is mediated by two sets of interactions, a set of *static* interactions between a loop on ClpX, containing the conserved tripeptide motif (\[L/I/V\]-G-\[F/L\]), and the hydrophobic pocket (Hp) on ClpP^[@CR14]--[@CR16]^, as well as a set of *dynamic* interactions between the flexible N-terminal loop of ClpP and the pore-2 loop of ClpX^[@CR14],[@CR16]--[@CR18]^.
Perrault syndrome type 3 (PRLTS3) is caused by mutations in *CLPP*, and to date seven different mutations have been identified in patients. Structurally, these mutations can be broadly classified into two regions. The first group of mutations (P142L, C144R, T145P, C147S and G162S) is located in, or around, the Hp of CLPP^[@CR2],[@CR4],[@CR19]^. As this region is responsible for interaction with the ATPase component^[@CR2],[@CR4],[@CR16]^, these mutations are proposed to impair CLPX interaction and as a consequence, CLPX-mediated proteolytic activity, but not CLPP peptidase activity^[@CR2],[@CR4]^. The second group of mutations (Y229D and I208M) is located in close proximity to the catalytic triad of CLPP, and as such these mutations are proposed to alter the peptidase activity of CLPP, but not its docking to CLPX^[@CR20],[@CR21]^. To date, however, neither the structural nor functional consequences of these mutations in CLPP have been extensively examined. To understand the effect of these different types of mutations, we chose to study two mutations from the first group (namely T145P and C147S, both of which were identified in the initial study by Newman and colleagues^[@CR4]^) and one mutation from the second group (namely Y229D). Here we show, that each of the different CLPP mutations exhibit a specific defect, from protein assembly to peptidase activity, or their interaction with CLPX. Interestingly, of the two Hp mutants tested, only one (T145P) exhibited severely compromised activity. Surprisingly, the other mutation (C147S) had no measurable defect on CLPP function *in vitro*. In contrast to C147S, the catalytic-region mutant (Y229D) exhibited a number of significant defects in CLPP function.
Materials and Methods {#Sec2}
=====================
Cloning {#Sec3}
-------
Constructs encoding mutant CLPP (T145P, C147S, Y229D, ∆C, RA or ECRC) were generated by site-directed mutagenesis^[@CR22]^ using either *pHUE/CLPP* or *pOTB7/CLPP*^[@CR23]^ as template DNA. Refer to supplementary material for primer sequences and the final plasmid constructs generated.
Cell culture, preparation of mitochondria and *in vitro* import of radiolabelled preproteins into mammalian mitochondria {#Sec4}
------------------------------------------------------------------------------------------------------------------------
HeLa (ATCC® CCL2™) cells were grown and maintained at 70--90% confluency at 37 °C with 5% (v/v) CO~2~ in Dulbecco's Modified Eagle Medium (DMEM; ThermoFisher Scientific) supplemented with 10% (v/v) Fetal Bovine Serum (FBS) for a maximum of one month. Mitochondria were isolated from HeLa cells as described^[@CR24]^. *In vitro* import reactions were performed essentially as described^[@CR25]^. Wild type and mutant precursor CLPP (preCLPP) proteins were radiolabelled in the presence of 11 µCi \[^35^S\]Met/Cys EXPRE^35^S ^35^S Protein Labelling Mix (specific activity \>1000 Ci/mmol) (Perkin Elmer, Waltham, MA, USA), using the TnT® SP6 Coupled Reticulocyte Lysate System (Promega, Australia), according to the manufacturer's instructions. Following import, the proteins were separated by SDS-PAGE (see below), gels were then dried, and the radiolabelled proteins visualised by digital autoradiography using a Typhoon Trio Molecular Imager (GE Healthcare).
Protein expression and purification {#Sec5}
-----------------------------------
With the exception of CLPP^T145P^, all recombinant proteins were expressed in BL21-Codon Plus (DE3)-RIL *Escherichia coli* cells. CLPP^T145P^ was expressed in ∆*dnaK* (EN2) *E*. *coli* cells^[@CR26]^. Wild type and mutant, mature human CLPP (m-CLPP) was expressed as a His~6~Ubiquitin (H~6~Ub) fusion protein and subsequently purified by immobilised metal ion affinity chromatography (IMAC) using Ni-NTA agarose beads (QIAGEN). The H~6~Ub moiety was cleaved from the fusion protein using the deubiquitinating (DUB) enzyme, Usp2cc^[@CR27]^ and the untagged protein was purified, essentially as described^[@CR28]^. As required, recombinant proteins were applied to a Superdex 200 HiLoad 16/60 pg column (GE Healthcare) pre-equilibrated in GF buffer (50 mM Tris-HCl \[pH 7.5\], 100 mM NaCl, 200 mM KCl, 5% (v/v) glycerol, 0.025% (v/v) Triton X-100 and 20 mM MgCl~2~) in the presence or absence of 0.5 mM TCEP, connected to a NGC^TM^ Quest Plus Chromatography System (Bio-Rad) with Chromlab software v3.1.0.06. The elution profile was monitored by absorbance at 280 nm (A~280~) and the column was calibrated according to the manufacturer's instructions using HMW Gel Filtration Calibration Kit (GE Healthcare). Mature human CLPX was expressed as a C-terminal His~10~ fusion protein and purified as described previously^[@CR23]^. *E*. *coli* ClpP (EcClpP) and His~6~GFP-SsrA were expressed and purified as described^[@CR29]^, while *E*. *coli* ClpX (EcClpX) was purified as described^[@CR30]^. Fluorescein isothiocyanate (FITC)-casein and N-Suc-Leu-Tyr-7-amino-4-methylcoumarin (Suc-LY-amc) were purchased from Sigma-Aldrich.
*In vitro* degradation assays {#Sec6}
-----------------------------
Human CLPX(P)-mediated degradation assays were performed in hXP buffer (50 mM Tris-HCl \[pH 7.5\], 100 mM KCl, 100 mM NaCl, 20 mM MgCl~2~, 10% (v/v) glycerol, 0.025% (v/v) Triton X-100), in the presence or absence of 1 mM DTT (as indicated). All degradation assays were performed at 30 °C in black 96 well plates (Corning flat bottom) using a SpectraMax M5e plate reader (Molecular Devices). For the degradation of FITC-casein, the final protein concentrations were as follows, 0.3 µM FITC-casein, 2.4 µM CLPX and 5.6 µM CLPP (wild type or mutant). FITC fluorescence was excited at 490 nm and the emission was monitored at 520 nm. The degradation reaction was initiated with the addition of ATP (5 mM). To determine the rate of FITC-casein degradation, the fluorescence intensity of FITC-casein (alone) was subtracted from the fluorescence intensity of FITC-casein (in the presence of the CLPX). FITC-casein protein turnover was also monitored by SDS-PAGE and visualised using a Typhoon Trio Molecular Imager as described previously^[@CR31]^. For the degradation of GFP-SsrA the final concentrations were 1.2 µM ecClpX, 2.8 µM CLPP (wild type or mutant) and 1 µM GFP-SsrA. The degradation reaction was initiated with the addition of ATP (5 mM). GFP fluorescence was excited at 410 nm and the emission was monitored at 500 nm. For the CLPP-mediated degradation of Suc-LY-amc the final concentrations were as follows, 1 mM Suc-LY-amc and 5.6 µM CLPP (wild type or mutant). In this case, substrate turnover was initiated with the addition of Suc-LY-amc. Fluorescence was monitored at 460 nm (excitation at 380 nm).
Protein separation and analysis {#Sec7}
-------------------------------
To identify the N-terminus of mature human CLPP, preCLPP~FLAG~ was expressed in HeLa cells, mitochondria were isolated as described^[@CR24]^, then solubilised in pre-chilled mitochondrial immunoprecipitation (mito IP) buffer (50 mM Tris-HCl \[pH 7.5\], 100 mM KCl, 10 mM Mg-acetate, 5% (v/v) glycerol, 1% (v/v) Triton X-100) supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF). Following incubation on ice (30 min), the insoluble material was separated by centrifugation (16060 *g*, 10 min, 4 °C) and CLPP~FLAG~ recovered from the clarified lysate by immunoprecipitation. Briefly, the clarified lysate was incubated end-over-end (1 h, 4 °C) with 50 μl (settled volume) of anti-FLAG M2 affinity beads (Sigma-Aldrich) pre-equilibrated with Tris-buffered saline (TBS; 50 mM Tris-HCl \[pH 7.5\], 150 mM NaCl). The beads were washed (20 BV of TBS) prior to elution of the bound protein using 2 BV of 50 mM glycine-HCl \[pH 3.0\]. Following neutralisation (using 1 M Tris-HCl \[pH 8.0\]), the protein sample was separated by SDS-PAGE and then transferred to PVDF membrane using CAPS buffer (10 mM CAPS-KOH \[pH 11.0\], 10% (v/v) methanol) and the band of interest was excised before being subjected to seven cycles of automated Edman degradation using an Applied Biosystems 494 Procise Protein sequencing system. Sequencing and analysis was performed by the Australian Proteome Analysis Facility.
To monitor the assembly of human CLPP, recombinant wild type or mutant CLPP (4 µg) was prepared in Native-PAGE buffer (10 mM Tris-HCl \[pH 7.0\], 10 mM NaCl, 5 mM MgCl~2~, 10 mM KCl, 5% (v/v) glycerol), supplemented with 50 mM Bis-Tris, 50 mM NaCl, 10% (v/v) glycerol, 0.001% (w/v) PonceauS then separated using 4--16% Native-PAGE Novex Bis-Tris gels (Invitrogen) according to the manufacturer's instructions and visualised by staining with CBB. Native HMW calibration kit (GE Healthcare) was used as molecular weight markers. To monitor the import of radiolabelled CLPP (in HeLa mitochondria), mitochondrial proteins were separated using 12.5% SDS-PAGE. Following electrophoresis, the gels were dried and then scanned by digital autoradiography using a Typhoon Trio Molecular Imager (GE Healthcare). Images were analysed using the ImageQuant 5.1 software (GE Healthcare).
Analysis of human CLPX-CLPP complexes by co-immunoprecipitation (coIP) {#Sec8}
----------------------------------------------------------------------
To assess human CLPX-CLPP complex formation, wild type or mutant CLPP (0.5 µM) was incubated in the absence or presence of human CLPX (1 µM) in IP buffer (50 mM Tris-HCl \[pH 7.5\], 100 mM NaCl, 100 mM KCl, 40 mM Mg-Acetate, 10% (v/v) glycerol and 0.1% (v/v) Triton X-100), supplemented with 2 mM ATPγS (as indicated). Following a short preincubation of the proteins at room temperature, the samples were then incubated with Protein A-Sepharose (Sigma-Aldrich) containing pre-bound anti-CLPX antibodies and mixed end-over-end for 60 min at 4 °C. Following removal of the unbound proteins, the Protein A-Sepharose beads were washed, essentially as described^[@CR32]^, using ice cold IP buffer supplemented with 10 mM ATP. Finally, the bound proteins were eluted with 50 mM glycine-HCl \[pH 2.5\]. All proteins were separated by 16.5% Tricine SDS-PAGE^[@CR33]^, transferred to PVDF membrane and detected by immunodecoration with specific antisera (either anti-CLPP (Origene OTI1D3) or anti-CLPX, as described^[@CR23]^).
Results {#Sec9}
=======
All PRLTS3-causing CLPP mutants are imported into mitochondria {#Sec10}
--------------------------------------------------------------
To better understand the molecular basis of the PRLTS3-causing mutations in CLPP, we introduced specific point mutations (T145P, C147S and Y229D) into human CLPP and compared their import and processing. Initially, we established the conditions for *in vitro* import of wild type CLPP into mitochondria isolated from HeLa cells (Fig. [1](#Fig1){ref-type="fig"}). As expected for a matrix located protein, the radiolabelled preprotein (preCLPP) was imported into mammalian mitochondria as determined by the time and membrane potential dependent accumulation of the processed protein (Fig. [1b](#Fig1){ref-type="fig"}, compare lanes 19 and 20). Interestingly, processing of CLPP appeared to occur in two steps, via an intermediate (here referred to as i-CLPP, Fig. [1a](#Fig1){ref-type="fig"}). We propose that the first processing step results in removal of the N-terminal presequence (likely via the matrix processing protease (MPP)) to generate the intermediate (i-CLPP), which is further processed into the mature protein (m-CLPP) via the autocatalytic removal of the CLPP propeptide. Although the final processing step is consistent with the autocatalytic processing of many ClpP proteins, including *E*. *coli* ClpP^[@CR34]^ we cannot exclude that processing of i-CLPP to m-CLPP is mediated by another mitochondrial peptidase. Next, we examined the import and processing of the different PRLTS3-causing CLPP mutants. Each mutant was imported into isolated mitochondria with similar kinetics to wild type CLPP, as can be seen by the appearance of i-CLPP by \~5 min (Fig. [1b](#Fig1){ref-type="fig"}, compare lane 17 with lanes 2, 7 and 12). This was followed by the appearance of mature CLPP (m-CLPP) (Fig. [1b](#Fig1){ref-type="fig"}, compare lane 18 with lanes 3, 8 and 13). Surprisingly, although CLPP^Y229D^ was imported into mitochondria, as demonstrated by the requirement of polarised mitochondria (Fig. [1b](#Fig1){ref-type="fig"}, compare lanes 14 and 15), the processing of this mutant was different to the other proteins (wild type and the other mutants) as judged by the altered molecular weight of i-CLPP^Y229D^, suggesting that CLPP^Y229D^ is processed via an alternate mechanism (Fig. [1b](#Fig1){ref-type="fig"}, compare lanes 14 and 19). Importantly, the molecular weight of m-CLPP is seemingly unchanged for all forms of CLPP (wild type and mutant). Irrespective of whether CLPP processing occurs via slightly different pathways, we wanted to ensure that all subsequent *in vitro* experiments were performed using the fully processed form of the protein. Thus, we determined the N-terminus of wild type human CLPP via N-terminal sequencing (Edman degradation). For this purpose, we overexpressed human preCLPP bearing a C-terminal FLAG-tag (CLPP~FLAG~) in HeLa cells, isolated the tagged processed protein via immunoprecipitation (IP) from purified HeLa mitochondria, and determined the N-terminal sequence by performing seven rounds of Edman degradation. From the data obtained, we identified Thr53 as the most abundant N-terminal residue of mature human CLPP~FLAG~ (Fig. [1a](#Fig1){ref-type="fig"}). Significantly, cleavage at this site (Thr53) is not consistent with the R-2 rule of processing by the mitochondrial processing peptidase^[@CR35],[@CR36]^, however it is consistent with autocatalytic propeptide cleavage of human CLPP proposed by Maurizi and colleagues^[@CR37]^. It is also consistent with propeptide processing sites identified in several bacterial ClpP homologs^[@CR34],[@CR38]^. Consistent with our identification of Thr53 as the N-terminal residue of mature CLPP, radiolabelled m-CLPP expressed *in organello* co-migrated with purified untagged wild type recombinant CLPP (CLPP~53--277~) used in this study (Fig. [S1](#MOESM1){ref-type="media"}).Figure 1Import and processing of PRLTS3-causing mutations in CLPP. (**a**) Cartoon of Human precursor CLPP (preCLPP), illustrating the relative location of intermediate CLPP (i-CLPP) and mature CLPP (m-CLPP), including the identity of the N-terminus of m-CLPP as identified by N-terminal sequencing. (**b**) Radiolabelled precursor protein of CLPP^T145P^ (lanes 1--5), CLPP^C147S^ (lanes 6--10), CLPP^Y229D^ (lanes 11--15) and wild type CLPP (lane 16--20) was imported into mitochondria isolated from HeLa cells, in the presence or absence of a membrane potential (∆ψ) as indicated. The precursor (pre) protein was processed into an intermediate (i-) and finally mature (m-) CLPP. All radiolabelled proteins were separated by 12.5% SDS-PAGE and visualised by digital autoradiography. The full-length images are presented in Supplementary Fig. [S7](#MOESM1){ref-type="media"}.
Human CLPP forms a double-ringed complex in the absence of CLPX {#Sec11}
---------------------------------------------------------------
Given each mutant protein was imported into mitochondria, we wanted to examine the ability of each protein to form a functional oligomer. Initially, we used Native-PAGE to examine the oligomeric assembly of wild type human CLPP *in vitro* (Fig. [2](#Fig2){ref-type="fig"}). This technique was recently used by Osiewacz and colleagues^[@CR39]^ to monitor the oligomeric state of ClpP, and consistent with their findings, we observed two oligomeric species for human CLPP (Fig. [2b](#Fig2){ref-type="fig"}, lane 2). Based on the apparent molecular weight of these species (\~220 kDa and \~460 kDa), they are expected to represent the heptamer and tetradecamer of CLPP respectively. However, given that human CLPP was proposed to only form a tetradecamer in the presence of its partner protein CLPX^[@CR13]^, we wanted to confirm the identity of the oligomeric species observed in Native-PAGE before examining the effect of the different PRLTS3-causing mutations on CLPP assembly. To do so, we designed three mutations in human CLPP. The first two mutations were based on the identification of an arginine sensor (Arg171) in *Staphylococcus aureus* ClpP (*Sa*ClpP), which forms a salt-bridge with Asp170 across the ring-ring interface^[@CR40]^, facilitating tetradecamer formation (Fig. [2a](#Fig2){ref-type="fig"}). The first mutation in human CLPP (namely Arg226Ala herein referred to as CLPP^RA^; equivalent to the Arg171Ala mutation in *Sa*ClpP) was designed to impair tetradecamer formation. The second mutant protein (herein referred to as CLPP^ECRC^) contained two complementary point mutations at the inter-ring interface (namely Glu225Cys \[equivalent to Asp170 in *Sa*ClpP\] and Arg226Cys \[equivalent to Arg171 in *Sa*ClpP\], see Fig. [2a](#Fig2){ref-type="fig"}). This mutant was designed to reversibly stabilise the tetradecamer through the formation of disulphide bonds between adjacent heptameric rings. The final mutant, which lacks the last 28 residues of CLPP (here termed CLPP^∆C^), was equivalent to the mutant described by Maurizi and colleagues^[@CR13]^. This mutation creates a more compact and symmetrical heptamer, and hence is proposed to exhibit conventional behaviour on gel filtration^[@CR13]^. Next, each mutant was generated by site-directed mutagenesis of cDNA encoding mature human CLPP. The different proteins were expressed in *E*. *coli* as Ubiquitin (Ub)-fusion proteins and purified. Following purification of the fusion protein, the H~6~-Ub moiety was cleaved, and the relative purity of the untagged recombinant protein was examined by SDS-PAGE (Fig. [S2a](#MOESM1){ref-type="media"}). Initially we compared the behaviour of wild type CLPP with each mutant protein using Native-PAGE (Fig. [2b](#Fig2){ref-type="fig"}). As described above, wild type mature CLPP formed two distinct species of relatively equal intensity -- a heptamer at \~220 kDa and a tetradecamer at \~460 kDa (Fig. [2b](#Fig2){ref-type="fig"}, lane 2); both of which were consistent with their theoretical MW (172 kDa and 344 kDa, respectively). In contrast to wild type CLPP, replacement of Arg226 with Ala (which was designed to preclude the formation of a putative salt-bridge across the ring-ring interface) destabilized the tetradecamer significantly, as CLPP^RA^ migrated almost exclusively as the lower species -- a heptamer (Fig. [2b](#Fig2){ref-type="fig"}, lane 3). Consistent with the importance of this site in the formation of the tetradecamer, the CLPP^ECRC^ mutant migrated solely as a tetradecamer under oxidising conditions (Fig. [2b](#Fig2){ref-type="fig"}, lane 4), whilst under reducing conditions (in the presence of TCEP), there was a significant (although incomplete) increase in the amount of heptamer observed (Fig. [2b](#Fig2){ref-type="fig"}, lane 5, lower band). Unexpectedly, for reasons that are currently not understood, neither the heptamer nor the tetradecamer of CLPP^∆C^ was resolved by Native-PAGE (Fig. [2b](#Fig2){ref-type="fig"}, lane 1). Therefore, we also analysed the mutant proteins using size exclusion chromatography (SEC) (Fig. [2c](#Fig2){ref-type="fig"}). Initially, we monitored the elution profile of wild type CLPP (Fig. [2c](#Fig2){ref-type="fig"}, top panel). Interestingly, in comparison to the Native-PAGE only a single species was observed for wild type CLPP using SEC. This species however, eluted at \~54 ml, which is equivalent to the elution profile of a \~500 kDa globular protein and is similar to the estimated MW (\~460 kDa) of the tetradecamer observed in Native-PAGE. One explanation for the variation in the type of oligomers observed by the two techniques, is that the native electrophoresis conditions promote dissociation of CLPP tetradecamers into heptamers. Nevertheless, we next monitored the elution profile of CLPP^∆C^ (Fig. [2c](#Fig2){ref-type="fig"}, second panel). In contrast to our Native-PAGE analysis of CLPP^∆C^, in which we were unable to detect the protein, CLPP^∆C^ migrated as a single symmetrical peak in SEC, with an apparent molecular weight of \~240 kDa (Fig. [2c](#Fig2){ref-type="fig"}). The molecular weight of CLPP^∆C^ was not only consistent with published findings^[@CR13]^ and similar to the theoretical MW of a CLPP heptamer (\~172 kDa) but it was also comparable to the observed MW of the heptamer of wild type CLPP in Native-PAGE (Fig. [2b](#Fig2){ref-type="fig"}, lower band). Having established the behaviour of wild type CLPP and CLPP^∆C^, we next examined the elution profile of the different CLPP mutants (designed to alter tetradecamer formation). Consistent with our Native-PAGE analysis, CLPP^ECRC^ (under oxidising conditions) eluted from the column in a single symmetrical peak at \~54 ml (equivalent to the tetradecamer), while CLPP^RA^ eluted in a single symmetrical peak at \~61 ml, slightly earlier than the CLPP^∆C^ heptamer peak (\~64 ml). This difference in elution volume of the CLPP^RA^ heptamer and the CLPP^∆C^ heptamer is expected to be due to the presence of the exposed C-termini, which increases the hydrodynamic radius of CLPP proposed by Maurizi and colleagues^[@CR13]^. Consistent with the classification of these peaks, CLPP^ECRC^ (under reducing conditions) eluted in two peaks, one at \~54 ml (equivalent to the tetradecamer) and the other at \~61 ml (equivalent to the heptamer). Collectively, these data demonstrate that reduction of CLPP^ECRC^ (with TCEP) destabilized the tetradecamer, resulting in the formation of the heptamer. The data also confirmed that we can use gel filtration and Native-PAGE to monitor the assembly of CLPP *in vitro*. Significantly, these data suggest that in the absence of CLPX, human CLPP is able to assemble into tetradecamers, under the conditions examined.Figure 2Wild type CLPP forms heptamers and tetradecamers *in vitro* in the absence of human CLPX. (**a**) Model of *Staphylococcus aureus* ClpP~14~ (SaClpP~14~) PDB: 5C90 highlighting the position of D170 and R171, equivalent to E225 and R226 in human CLPP (here termed hE225 and hR226 respectively). (**b**) Assembly of wild type or mutant CLPP under native conditions. Recombinant, untagged wild type or mutant human CLPP (4 µg) was separated by Native-PAGE and visualised by staining with Coomassie Brilliant Blue (CBB) R250. The tetradecamer (CLPP~14~) and heptamer (CLPP~7~) are indicated. The oligomeric composition of wild type human CLPP (lane 2) was compared to either CLPP^ΔC^ (lane 1), CLPP^RA^ (lane 3) or CLPP^ECRC^ either in the absence (lane 4) or presence (lane 5) of TCEP. (**c**) The tetradecamer (14-mer) and heptamer (7-mer) of recombinant, untagged wild type or mutant CLPP was separated by size exclusion chromatography (SEC) using a Superdex 200 HiLoad 16/60 pg column. Elution profiles of wild type CLPP (top panel), CLPP^∆C^ (second panel), CLPP^RA^ (middle panel), CLPP^ECRC^ in the absence (fourth panel) or presence of TCEP (bottom panel) were measured at 280 nm (A~280~). Lines indicate the peak elution volume of thyroglobulin (669 kDa), ferritin (440 kDa), aldolase (158 kDa) and conalbumin (75 kDa).
CLPP^Y229D^ displays impaired tetradecamer assembly {#Sec12}
---------------------------------------------------
Next, having established that we could use size exclusion chromatography and Native-PAGE to monitor the oligomeric state of mature human CLPP *in vitro*, we examined the oligomeric state of the different Perrault mutant proteins (Fig. [3a](#Fig3){ref-type="fig"}, highlighted in yellow). To do so, we first introduced the appropriate mutation into cDNA coding for mature human CLPP, then overexpressed each protein in *E*. *coli*. Following affinity isolation, cleavage and recovery of the untagged proteins, we examined the relative purity of each mutant by SDS-PAGE (Fig. [S2b](#MOESM1){ref-type="media"}). Although two of the mutant proteins (CLPP^C147S^ and CLPP^Y229D^) could be purified to homogeneity (\>95% purity), the third mutant (CLPP^T145P^) could only be recovered to \~85% purity (Fig. [S2b](#MOESM1){ref-type="media"}, lane 3). While expression of CLPP^T145P^ in ∆*dnaK* (EN2) *E*. *coli* cells was successful in removing one impurity -- DnaK, despite numerous attempts using a variety of different approaches, two impurities remained. These impurities are anticipated (based on their apparent molecular weights of \~60 and 40 kDa, respectively) to be GroEL and DnaJ and hence human CLPP^T145P^ likely has a folding defect.Figure 3CLPP^Y229D^ displays altered oligomerisation *in vitro*. (**a**) Model showing the surface of Human ClpP~7~ (PDB: 1TG6) highlighting two adjacent subunits (subunit A in blue and subunit B in pink) showing the main chain in "spaghetti", indicating the position of the Hp residues L104 (blue) and Y138 (blue) on subunit A and Y118 (pink) and W146 (pink) on subunit B. The relative position of the residues that are mutated in Perrault syndrome (T145, C147 and Y229) and analysed in this study are indicated in yellow. (**b**) Assembly of wild type CLPP and CLPP Perrault mutants under native conditions. Recombinant, untagged wild type or mutant CLPP (4 µg) was separated by Native-PAGE and visualised by staining with CBB. The oligomeric composition of wild type human CLPP (lane 2) was compared to CLPP^T145P^ (lane 3), CLPP^C147S^ (lane 4) and CLPP^Y229D^ (lane 5). The tetradecamer (CLPP~14~) and heptamer (CLPP~7~) are indicated, as is the monomer (CLPP) and dimer (CLPP~2~) of CLPP^Y229D^. (**c**) The high order oligomeric complexes of recombinant, untagged wild type or mutant CLPP were separated by size exclusion chromatography (SEC) using a Superdex 200 HiLoad 16/60 pg column (GE Healthcare). Elution profiles of wild type CLPP (black line and top panel), CLPP^T145P^ (green line), CLPP^C147S^ (blue line and middle panel) and CLPP^Y229D^ (red line and bottom panel) were measured at 280 nm (A~280~). Proteins from the indicated fractions were separated by SDS-PAGE and visualised by staining with CBB. Lines indicate the peak elution volume of thyroglobulin (669 kDa), ferritin (440 kDa), aldolase (158 kDa) and conalbumin (75 kDa). The full-length gels for SEC are presented in Supplementary Fig. [S8](#MOESM1){ref-type="media"}.
Initially, to test the structural integrity of each CLPP mutant, we monitored the ability of wild type CLPP to form an oligomer and compared its assembly to the different mutant proteins. To do so, we performed non-specific chemical crosslinking using glutaraldehyde (GA) as described previously^[@CR13]^. As expected, wild type CLPP formed a ladder of oligomeric species in the presence of GA (Fig. [S3](#MOESM1){ref-type="media"}, lanes 2--4). Consistent with wild type CLPP, both CLPP^C147S^ (Fig. [S3](#MOESM1){ref-type="media"}, lanes 8--10) and CLPP^Y229D^ (Fig. [S3](#MOESM1){ref-type="media"}, lanes 11--13) formed a ladder of seven discrete crosslinked protein bands. In contrast, these discrete oligomeric species were not observed with CLPP^T145P^ (Fig. [S3](#MOESM1){ref-type="media"}, lanes 5--7). Collectively, these data suggest that in the absence of CLPX, CLPP^T145P^ is unable to assemble into a heptamer. Next, we examined the ability of these proteins (in the absence of crosslinker) to form heptamers and higher order complexes. Consistent with our previous findings (Fig. [2](#Fig2){ref-type="fig"}), wild type CLPP migrated as two discrete bands in Native-PAGE, a heptamer at \~220 kDa and a tetradecamer at \~460 kDa (Fig. [3b](#Fig3){ref-type="fig"}, lane 2). Similar to wild type CLPP, both bands were observed for CLPP^C147S^ (Fig. [3b](#Fig3){ref-type="fig"}, lane 4). In contrast, only the lower MW species (the heptamer) was observed for CLPP^Y229D^ (Fig. [3b](#Fig3){ref-type="fig"}, lane 5). This suggests that the ring-ring interface of CLPP^Y229D^ was compromised and given that a small amount of CLPP^Y229D^ monomers and dimers were also observed in the gel (Fig. [3b](#Fig3){ref-type="fig"}, lane 5), it appears that this mutation (in contrast to the other mutants tested) may also affect intra-ring stability. Notably, in contrast to CLPP^C147S^ and CLPP^Y229D^, the majority of CLPP^T145P^ failed to enter the separating gel, suggesting that CLPP^T145P^ has a propensity to aggregate *in vitro* (Fig. [3b](#Fig3){ref-type="fig"}, lane 3). Finally, we examined that ability of each mutant to oligomerise in solution, using gel filtration (Fig. [3c](#Fig3){ref-type="fig"}). Consistent with Native-PAGE analysis of the different mutants, CLPP^C147S^ (Fig. [3c](#Fig3){ref-type="fig"}, blue) eluted in a symmetrical peak at \~53 ml (equivalent to the elution volume of a CLPP tetradecamer; Fig. [3c](#Fig3){ref-type="fig"}, black), while CLPP^Y229D^ (Fig. [3c](#Fig3){ref-type="fig"}, red) eluted in a peak at \~61 ml (equivalent to the elution volume of the CLPP heptamer). In contrast, CLPP^T145P^ (Fig. [3c](#Fig3){ref-type="fig"}, green) eluted in multiple peaks, with much of the protein eluting near the void volume of the column or as a monomer. Collectively, these data demonstrate that CLPP^Y229D^ exhibits a mild defect in CLPP assembly, while mutations in the Hp have quite different effects on protein assembly (Fig. [3](#Fig3){ref-type="fig"}). Interestingly, although one Hp mutant (CLPP^T145P^) was unable to assemble into its native oligomer (likely due to a folding defect), the assembly of the other Hp mutant (CLPP^C147S^) was unaffected (Fig. [3](#Fig3){ref-type="fig"}). As such, PRLTS3 mutations, even different mutations within the same region, are likely to have broadly different effects on proteolytic activity.
PRLTS3 CLPP mutants demonstrate varied peptidase activities {#Sec13}
-----------------------------------------------------------
Next, we examined the peptidase activity of the different PRLTS3 mutant proteins. Initially we screened the peptidase activity of wild type CLPP, in the absence of its cognate unfoldase CLPX, using fluorescently labelled peptides. Consistent with the recent findings of Sieber and colleagues^[@CR41]^, we discovered that human CLPP displayed weak but measurable peptidase activity using the fluorescently labelled peptide Suc-LY-amc (Fig. [4](#Fig4){ref-type="fig"}, black circles). Therefore, we tested the peptidase activity of each CLPP mutant using this substrate. As predicted, due to proximity of the mutation to the active site, the peptidase activity of CLPP^Y229D^ was completely abolished (Fig. [4](#Fig4){ref-type="fig"}, red triangles). Importantly, this loss of activity, by CLPP^Y229D^, validates Suc-LY-amc as a CLPP substrate. In contrast to CLPP^Y229D^, both Hp mutants (CLPP^T145P^ and CLPP^C147S^) displayed peptidase activity towards the peptide substrate, even though they each showed substantial differences in assembly (Fig. [3](#Fig3){ref-type="fig"}). One mutant, CLPP^C147S^, exhibited a similar peptidase activity to wild type CLPP (Fig. [4](#Fig4){ref-type="fig"}, blue squares), while the peptidase activity of the other Hp mutant (CLPP^T145P^) was stimulated by \~20-fold (Fig. [4](#Fig4){ref-type="fig"}, green diamonds). These data demonstrate that PRLTS3 mutations in CLPP exhibit broadly different effects on peptidase activity, from complete loss of function to enhanced activity. Although the "hyper-activation" of CLPP^T145P^ was quite unexpected, it does share some similarities with a "gain-of-function" mutant recently described for *Sa*ClpP, in which Y63 (in the Hp) was replaced with alanine^[@CR42]^. Interestingly, based on the structure of *Sa*ClpP, Y63 (found in β strand 3) is located only \~3.5 Å from the highly conserved residue T90 (located in the adjacent strand, β strand 5), which is equivalent to T145 in human CLPP. Hence, we propose that structural rearrangement of β strand 5 (caused by the T145P mutation) may affect the adjacent strand (β strand 3) in a similar fashion to that observed for the *Sa*ClpP^Y63A^ mutation^[@CR42]^. Collectively, our data suggest that mutations within, or around, the Hp of CLPP can have distinct effects on peptide substrate recognition and/or cleavage. Therefore, different point mutations within this region are likely to exhibit different phenotypes. Indeed, based on our data, in addition to the predicted dysregulation of CLPX-mediated degradation by CLPP, the T145P mutation in CLPP may result in a toxic gain-of-function in mammalian cells. Consistent with this idea, patients with the T145P mutation exhibit profound deafness and severe neurological symptoms^[@CR4],[@CR43]^. As such, analysis of the different pathogenic mutations in CLPP might provide evidence for a link to the severity of the disease observed in different Perrault syndrome patients.Figure 4CLPP^T145P^ displays enhanced peptidase activity. The turnover of the fluorogenic peptide substrate Suc-LY-amc (1 mM) was monitored in the presence of either wild type CLPP (black circles), CLPP^T145P^ (green diamonds), CLPP^C147S^ (blue squares) or CLPP^Y229D^ (red triangles). The cleavage of Suc-LY-amc was monitored by fluorescence (excitation = 380 nm, emission = 460 nm). Error bars represent the standard error of the mean (SEM) of three independent experiments.
CLPP^Y229D^ and CLPP^T145P^ exhibit decreased interaction with human CLPX {#Sec14}
-------------------------------------------------------------------------
Finally, we examined the ability of each mutant protein to interact with its cognate unfoldase (Fig. [5](#Fig5){ref-type="fig"}). Initially this was performed using an indirect assay, in which we examined the turnover of a CLPX-dependent substrate, the model unfolded protein casein. In this instance we used fluorescently labelled casein (FITC-casein) and monitored the change in FITC fluorescence as a measure of the rate of CLPXP-mediated turnover (Fig. [5a](#Fig5){ref-type="fig"}). As expected, the CLPXP-mediated turnover of FITC-casein (Fig. [5a](#Fig5){ref-type="fig"}, black bar and Fig. [5b](#Fig5){ref-type="fig"}, lanes 1--6) was dependent on the addition of ATP (Fig. [S4](#MOESM1){ref-type="media"}). Consistent with peptidase activity of CLPP^C147S^ (Fig. [4a](#Fig4){ref-type="fig"}), the C147S mutant also facilitated the CLPX-mediated degradation of FITC-casein at a similar rate to that of wild type CLPP (Fig. [5a](#Fig5){ref-type="fig"}, blue bar and Fig. [5b](#Fig5){ref-type="fig"}, lanes 7--12). In contrast to wild type CLPP and CLPP^C147S^, neither CLPP^T145P^ (Fig. [5a](#Fig5){ref-type="fig"}, green bar) nor CLPP^Y229D^ (Fig. [5a](#Fig5){ref-type="fig"}, red bar) were able to facilitate the turnover of the model unfolded protein. In the case of CLPP^T145P^ (which exhibits increased peptidase activity), this loss of CLPX-mediated turnover appears to be due to a compromised interaction with the unfoldase component. However, in the case of CLPP^Y229D^, it remained unclear if the CLPP mutant had lost or retained the ability to interact with CLPX. To determine which mutant retained the ability to interact with human CLPX, we performed a series of immunoprecipitation experiments. In this case, we immobilised CLPX antisera to Protein A Sepharose (PAS) and then incubated the immobilised antibody with wild type or mutant CLPP in the absence or presence of CLPX. Following incubation of the recombinant proteins with the beads and removal of any non-specifically bound proteins from the PAS through extensive washing, the interacting proteins were eluted, and the samples separated by SDS-PAGE before being transferred to PVDF membrane for analysis via immunoblotting with specific antisera (Fig. [5c](#Fig5){ref-type="fig"}). Despite some weak non-specific binding of wild type CLPP to the beads (Fig. [5c](#Fig5){ref-type="fig"}, lane 3 lower panel), there was a significant increase in the recovery of CLPP in the presence of CLPX (Fig. [5c](#Fig5){ref-type="fig"}, lane 7), demonstrating a specific interaction between CLPX and CLPP. Consistent with the interaction observed for wild type CLPP, CLPP^C147S^ also exhibited some weak non-specific binding to the beads (Fig. [5c](#Fig5){ref-type="fig"}, lane 5), with a significant increase in its recovery in the presence of CLPX (Fig. [5c](#Fig5){ref-type="fig"}, lane 9). In contrast to CLPP^C147S^, there was little recovery of either CLPP^T145P^ (Fig. [5c](#Fig5){ref-type="fig"}, lane 8) or CLPP^Y229D^ (Fig. [5c](#Fig5){ref-type="fig"}, lane 10) in the presence of CLPX. To determine the relative level of specific interaction between CLPX and each of the mutants, we quantitated the recovery of CLPP (wild type and mutant) in the presence and absence of CLPX, subtracting the amount of non-specific binding (i.e. in the absence of CLPX) (Fig. [5d](#Fig5){ref-type="fig"}). These data demonstrate that CLPP^C147S^ binds to CLPX with a similar affinity to wild type CLPP, while in contrast neither CLPP^T145P^ (Fig. [5d](#Fig5){ref-type="fig"}, green bar) nor CLPP^Y229D^ (Fig. [5d](#Fig5){ref-type="fig"}, red bar) showed any significant interaction with CLPX. Taken together, the data demonstrate that the lack of CLPX-mediated degradation of FITC-casein by CLPP^T145P^ is due to a loss of interaction with CLPX, while in contrast the lack of CLPX-mediated degradation of FITC-casein by CLPP^Y229D^ is likely due to a combined effect of a loss of peptidase activity as well as a loss of interaction with CLPX. Interestingly, despite the loss of CLPP^T145P^ interaction with human CLPX, we also the examined (indirectly) the ability of CLPP^T145P^ to interact with ecClpX. This was performed by monitoring the ecClpX-mediated turnover the model "folded" substrate (GFP-ssrA). Incredibly, CLPP^T145P^ retained the ability to facilitate the ecClpX-mediated the turnover of GFP-ssrA (Fig. [S5](#MOESM1){ref-type="media"}, green diamonds) at a similar rate to both wild type CLPP (Fig. [S5](#MOESM1){ref-type="media"}, black circles) and CLPP^C147S^ (Fig. [S5](#MOESM1){ref-type="media"}, blue squares). In contrast, CLPP^Y229D^ (which lacks peptidase activity) was unable to facilitate the ecClpX-mediated turnover of GFP-ssrA (Fig. [S5](#MOESM1){ref-type="media"}, red circles). Consistent with these data, Sieber and colleagues^[@CR41]^ recently showed that CLPP^T145P^ was able to facilitate the ecClpX-dependent turnover of GFP-ssrA at a similar rate to wild type CLPP. Despite our validation of this activity (for CLPP^T145P^), the molecular mechanism of GFP-ssrA turnover (mediated by ecClpX) remains unclear. Significantly, these data demonstrate that although CLPP^T145P^ can interact with ecClpX, it is unable to interact with human CLPX. This unexpected result reinforces the importance of examining homogeneous systems to study the effect of pathogenic mutations.Figure 5CLPP^T145P^ and CLPP^Y229D^ abolish functional association with human CLPX. (**a**) Human CLPX-dependent turnover of FITC-casein was monitored by a change in fluorescence at 520 nm \[excitation = 490 nm\] in the presence of either wild type CLPP (black bar), CLPP^T145P^ (green bar), CLPP^C147S^ (blue bar) or CLPP^Y229D^ (red bar). The change in fluorescence at 520 nm (ΔF~520\ nm~) of FITC-casein was calculated relative to the initial fluorescence of the substrate from three independent experiments. Error bars represent SEM. (**b**) To monitor the turnover of FITC-casein directly, samples as described in (**a**) containing either wild type CLPP (lanes 1--6) or CLPP^C147S^ (lanes 7--12) were separated by SDS-PAGE and monitored by fluorescence (upper panel). As a loading control, the levels of CLPX and CLPP (in each reaction) were monitored by staining with CBB (lower panels). (**c**) To monitor the interaction between human CLPX and CLPP, human CLPX was immunoprecipitated (IP) using a specific anti-CLPX antisera, in the absence (lanes 2) or presence (lanes 7--10) of wild type or mutant human CLPP. To ensure the specificity of the co-immunoprecipitation (co-IP) the recovery of wild type and mutant human CLPP was also monitored in the absence of human CLPX (lanes 3--6). Following co-IP of human CLPP, the input (1.25%) and the eluted (33%) proteins were separated by SDS-PAGE and transferred to PVDF, before being immunodecorated with specific antisera, as indicated. (**d**) Quantitation of human CLPP recovered in CLPX coIP (as described in (**c**) in which non-specific interaction (lanes 3--6, respectively) was subtracted from the specific interaction (lanes 7--10, respectively)) of three independent experiments. Error bars represent SEM. The full-length gels for (**b**) are presented in Supplementary Fig. [S9](#MOESM1){ref-type="media"} and the full-length western blot for (**c**) are presented in Supplementary Fig. [S10](#MOESM1){ref-type="media"}.
Conclusion {#Sec15}
==========
Although the physiological role of mitochondrial CLPP (and CLPX) remains poorly understood, a handful of studies suggest that mammalian CLPP similar to its bacterial homologs, plays a crucial role^[@CR10],[@CR44]--[@CR46]^. Consistently, a number of recent studies have linked mutations in human CLPP with PRLTS3^[@CR2],[@CR4],[@CR19]--[@CR21]^. Importantly, PRLTS3 patients share a number of phenotypes (such as deafness and infertility in both sexes) with *CLPP* null mice (*CLPP*^−/−^) which provides strong support for a link between the mutations in *CLPP* and PRLTS3. Interestingly, in addition to the shared phenotypes (between PRLTS3 patients and *CLPP*^−/−^ mice), PRLTS3 patients also exhibit some major neurological impairments such as epilepsy, microcephaly or learning difficulties, which suggests that some of the pathogenic mutations in CLPP might trigger a toxic gain-of-function in the mitochondrion.
In this study, we have dissected the biogenesis (import, processing and assembly) and activity (with and without its cognate unfoldase, CLPX) of three different PRLTS3 associated mutants of CLPP *in vitro*. Although two mutants (CLPP^T145P^ and CLPP^Y229D^) displayed significant defects in CLPP assembly and/or activity, one mutant (CLPP^C147S^) quite unexpectedly failed to display a single defect in any of the activities examined. Significantly, although this mutation is located within the Hp of CLPP (required for interaction with CLPX), a defect in CLPX-interaction or CLPX-mediated degradation of either a folded or unfolded substrate was not observed. One possible explanation for this surprising result is that CLPP^C147S^ may exhibit a specific defect in the recognition and/or translocation (into CLPX) of a unique substrate such as ERAL1 that was recently linked to the PRLTS^[@CR9],[@CR47]^, or a currently unknown substrate that may be associated with PRLTS. An alternate explanation is that the mutant protein is unstable *in vivo*, and hence the levels of CLPP drop below a minimum threshold concentration required for normal cellular function. Consistent with this idea, a splice mutation in *CLPP* (which reduces CLPP expression) is known to cause PRLTS3^[@CR4]^. Significantly, the C147S mutation was previously observed to display a decreased melting temperature^[@CR41]^, hence in contrast to other Hp mutations, this mutation may affect the physiological levels of CLPP. Collectively, these findings are consistent with relatively mild symptoms observed in patients with the C147S mutation, although the extent of gonadal dysgenesis was not completely examined in these patients^[@CR4]^. The final possibility is that PRLTS3 patients (with the C147S mutation) is caused not by the mutation in CLPP, but rather by an unidentified mutation within a novel gene. In support of this idea, C147 (in contrast to T145 and Y229) is only moderately conserved across ClpP sequences (see Fig. [S6](#MOESM1){ref-type="media"}). Indeed, at this position (i.e. the equivalent of position 147 in CLPP) the pathogenic residue -- Ser -- is permissive in several ClpP homologs, including *P*. *falciparum* ClpP (see Fig. [S6](#MOESM1){ref-type="media"}). Therefore, in order to better understand the molecular basis of PRLTS3 caused by CLPP^C147S^, the physiological levels (and/or half-life) of CLPP in these patients will need to be examined.
In contrast to C147S, the two remaining mutations (T145P and Y229D) are located at highly conserved positions within the protein -- T145 is almost completely conserved amongst ClpP sequences, while Y229 is restricted to large hydrophobic residues (see Fig. [S6](#MOESM1){ref-type="media"}). Consistent with the highly conserved nature of these residues, PRLTS3 patients with mutations in these residues display acute developmental defects (e.g. microcephaly, learning difficulties and muscle spasticity) in addition to SNHL and infertility. Not surprisingly, both proteins (CLPP^T145P^ and CLPP^Y229D^) exhibit severe defects in a range of biochemical activities. In addition to the loss of all human CLPX-mediated activities by these mutants, the peptidase activity of CLPP^Y229D^ was also abolished, while in contrast the peptidase activity of CLPP^T145P^ was enhanced, at least for some substrates (Fig. [4](#Fig4){ref-type="fig"}). Interestingly, although we observed an enhanced activity for the "poor" substrate (Suc-LY-amc), a slight reduction in the peptidase activity of CLPP^T145P^ was reported by Sieber and colleagues^[@CR41]^ for an optimised substrate (Ac-Phe(3,4-Cl~2~)-hArg-Leu-ACC). In contrast, neither group observed a change in the proteolytic activity of CLPP^T145P^ (in the presence of ecClpX) using the substrate GFP-ssrA (Fig. [S5](#MOESM1){ref-type="media"})^[@CR41]^. Collectively, these data demonstrate the importance of studying homogeneous systems when studying the effect of a pathogenic mutation. Moreover, they also appear to suggest that CLPP^T145P^ exhibits an altered substrate specificity. Nevertheless, in addition to the altered specificity of CLPP^T145P^, both mutants (CLPP^T145P^ and CLPP^Y229D^) display defects in folding and/or assembly. Hence, the additional phenotypes observed in patients carrying these mutations might be caused by the accumulation of incompletely folded or unassembled mutant proteins. This could be caused either directly, through the accumulation of misfolded or aggregated protein in the mitochondrion or indirectly through the sequestration of important molecular chaperones. Indeed, protein misfolding and aggregation is an underlying cause of several neurodegenerative diseases^[@CR48],[@CR49]^. Likewise, reduced chaperone activity of HSP60 has been associated with an autosomal dominant form of spastic paraplegia (SPG13)^[@CR50]^. Therefore, we propose that PRLTS3 patients carrying mutations in CLPP that only affect the cellular levels of CLPP (and not CLPP function) are likely to manifest in mild forms of the disease. In contrast, mutations in CLPP (such as the T145P and Y229D), that affect the folding and/or assembly of CLPP, may exhibit a toxic gain-of-function as a result of the reduced proteostatic capacity of the organelle. While the introduction of different pathogenic mutations into *CLPP*^−/−^ mice should help to determine if gain-of-function mutations contribute to the pathogenesis of the disease, the identification of validated human CLPP substrates could be very helpful in identifying a critical loss-of-function. Significantly, the recent identification of ERAL1 (an assembly chaperone for the mitochondrial ribosome associated with PRLT6) as a putative CLPP substrate suggests that PRLTS3 arises from a defect in mitochondrial protein translation. Consistent with this idea, several other forms of PRLTS have also been associated with mutations in genes involved in mitochondrial translation. Hence, in the future it will be interesting to examine if the PRLTS3 associated CLPP mutants affect ribosome assembly in the mitochondrion.
Electronic supplementary material
=================================
{#Sec16}
Supplementary Information
**Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
=================================
**Supplementary information** accompanies this paper at 10.1038/s41598-018-30311-1.
This work was supported by Australian Research Council Fellowships to K.N.T. (FT0992033) and D.A.D. (DP110103936), and an Australian Postgraduate Award to E.J.B. We thank Ravi Gogata for generating mutations, coding for T145P, C147S and ΔC, in pHUE/*CLPP*.
Experiments were conceived and designed by E.J.B., K.N.T. and D.A.D. Experimental work was performed by E.J.B., H.Z. and T.S. The manuscript was written by E.J.B., K.N.T. and D.A.D. All authors reviewed the manuscript.
Competing Interests {#FPar1}
===================
The authors declare no competing interests.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
In the focus on the interaction between university students and their education environment, as well as development of ways of conceptualizing and measuring this process in different disciplinary contexts, student engagement has become an important perspective ([@B40]; [@B33]; [@B51]; [@B54]; [@B38]). In descriptions of the complex relationship between individual students and their educational context, as well as understanding questions related to progression, retention or dropping out of school, engagement is considered the primary construct ([@B85]; [@B11]). Strong student engagement has been shown to be linked to smooth progression of studies, positive learning experiences, a deep approach to learning, general satisfaction, well-being and persistence, as well as better learning outcomes, such as quality of knowledge, higher-order thinking, ethical qualities, career readiness and intentions, professional identity and grades ([@B91]; [@B77]; [@B34]; [@B86]; [@B38]; [@B63]).
The first year of higher education has been identified as a crucial phase from the viewpoint of successful engagement ([@B47]) and influences a student's educational career. First-year students familiarize themselves with their domain and the practices of their scholarly learning community ([@B53]; [@B57]). Students' previous experiences, motives for studying and abilities to adapt themselves to new practices, as well as the atmosphere and participatory qualities of the community, affect the successful ongoing transition to and engagement in the community ([@B82]; [@B91]). The consequences of disengagement are serious, because they might lead to student attrition, unplanned changes in the study program, withdrawal and even failure to complete one's education ([@B35]). Attrition rates are significantly higher during the first year of higher education, and thus, student engagement with the scholarly learning community is necessary ([@B48]). Such understandings respond to the multifaceted and complex needs of diversifying student populations in present-day universities ([@B34]).
Engagement emerges in the interaction between the student and the institution ([@B1], [@B2]; [@B69]; [@B49]). In this dynamic, individual objectives and starting points for engagement, as well as empowering contextual elements, are essential. Previous studies have shown that internal factors (like motivation, expectations for higher education and emotions) and formal and informal external contextual factors (like systemic structures, curricular issues, and pedagogical practices) contribute either positively or negatively to student engagement ([@B81]; [@B69]; [@B52]; [@B34]; [@B61]; [@B72]; [@B78]; [@B38]).
Previous research on student engagement covered multiple different perspectives, from formal and informal aspects of student experience ([@B13]) to the intrinsic or extrinsic dimensions of experiences of engagement ([@B27]; [@B85]; [@B38]). However, studies focused simultaneously on the internal and external factors that contribute to student engagement that take into account the dynamics between individuals and the context are scarce. The primary key factors in first-year student engagement have not been systematically identified. Moreover, individual differences and variation in students' engagement since the beginning of their education have not been widely explored. Therefore, this study explores the structure of student engagement and the dynamics of intrinsic and extrinsic factors related to student engagement and utilizes network analysis.
Theoretical Conceptualization of the Engagement Process
-------------------------------------------------------
Most of the established conceptualizations of student engagement are based on the conception that engagement emerges in the interaction between the student and the educational context ([@B2]; [@B69]; [@B49]; [@B38]). A growing body of evidence suggests that participating in learning communities contributes positively to student engagement, which, in turn, may affect educational attainment and smoother progression of education ([@B70], [@B71]). The effect can be seen in the purposeful activities that strengthen the engagement ([@B50]; [@B49]), especially community-based practices, such as learning communities through curricula and courses, common assignments and projects, and students working together with experienced scholars ([@B50]; [@B49]). [@B91] found that the relationship between membership in learning communities and student engagement is significant especially for first-year students. Strongly engaged students tend to emphasize the meaning of social relations and cooperation for their education (peer communities and the academic teaching--learning community; [@B13]).
From the viewpoint of [@B89], [@B90] situational and sociocultural theorization of community of practice, strong student engagement means emerging bonds between the student and the closest discipline-related communities. For successful engagement, two intertwined processes must be realized: student's self-motivated active agency (subjectivity) and developed and deepened participation in discipline-related communities (collectivity; [@B89]). Engaging experiences may occur during the participation processes for educationally effective and inclusive practices ([@B13]; [@B34]; [@B54]). Following Wenger's thinking, we defined four core areas for the engagement process (meaning, participation, sense of belonging and sense of identity) and complemented the overall picture with two more dimensions: academic skills and social practices (see [Figure 1](#F1){ref-type="fig"}). Academic and social integration in studies play a fundamental role, for example, in widely cited [@B80], [@B82] model. The *individual process* consists of experiencing one's education meaningfully together with mastering certain central academic skills, whereas the *collaborative process* consists of participating in academic teaching--learning communities and adopting certain social practices. In our view, student engagement is constructed based on these processes. The fundamental features of the engagement process, "sense of belonging" and an evolving "identity," emerge from the interaction of individual and collective processes.
{#F1}
The dimensions of engagement were first outlined in a conceptual article ([@B41]) and next a theory-driven qualitative study with a sample of university students ([@B72]) was conducted, where the dimensions were found appropriate for describing students' experiences of engagement. The term "meaning of studies" is used here to refer to the personal significance of the recently started education program and the perceived opportunities higher education offers to the student. This dimension covers students' self-concepts, values, attitudes or beliefs regarding their education ([@B39]). Students bring their own meanings and preferences to the educational experience, and this has been shown to affect academic motivation and values ([@B36]). "Academic skills" refer to the skills that are necessary for participating in academic teaching--learning practices. [@B10] used the similar concept "academic competence" which involves knowledge acquisition, increased intellectual sophistication and development of higher-order cognitive skills.
Turning to collective aspects of engagement, by "participation" we refer to taking part in different study-related communities, such as student peer communities or academic teaching--learning communities. [@B89] defines participation as a process of being in relationships with others. It suggests action and connections in local communities of practice. The term "social practices" is used to refer to the various disciplinary practices through which students become socialized in a disciplinary culture ([@B3]; [@B84]). Students do not simply learn "about" something; instead, they also learn "to be" and "to do" something. The social practices perspective takes learning as an aspect of participation in socially situated and locally constructed practices ([@B41]). Scholars have shown that engagement varies considerably among students in different disciplines ([@B8]; [@B38]). Therefore, locally constructed practices play an important role in triggering engagement.
From the viewpoint of engagement, "identity" refers to how learners interpret their experiences, perceive their actions and function as active agents in an academic environment ([@B9]; [@B7]). The developing identity refers to the students' personal insight into themselves and their abilities as learners, as well as the ways in which they position themselves in various communities related to their education ([@B9]). When students experience their education as meaningful for the goals they have set for themselves, the feeling of belonging is strengthened ([@B81]). An appropriate definition for the sense of belonging consists of a student being accepted, valued, included and encouraged by others, such as teachers and peer learners ([@B79]; [@B83]). Students' positive experiences and actual patterns of participation naturally affect the students' developing sense of belonging ([@B55]; [@B60]). Together with a student's developing professional identity as a member of the academic community, the sense of belonging makes up the overall process of engagement. Engaging experiences create a stronger sense of belonging that further expands identity in multiple different ways ([@B89]).
Factors Intertwined With Student Engagement
-------------------------------------------
From the viewpoint of the integrative theorization of student engagement in this study, it is interesting to investigate how students' motivations to study, approaches to learning, chosen field of study and possible intention to drop out are intertwined. These are relatively established theoretical constructs and fields of research, and they at least partly overlap with academic and social, as well as individual and collective, aspects of student engagement---or disengagement. The motives for attending higher education are related to the skills necessary for education ([@B16], [@B17]). According to [@B16], [@B17], these motives are categorized into five basic dimensions. *Personal-intellectual development* means interest in intellectual and cultural self-development and a striving for understanding the complexities of life. *Humanitarian* motivation means an internal interest in improving the world, changing the system and helping others. *Expectation-driven* motivation refers to a student's efforts to meet the expectations of family and friends to attend university and obtain a degree. *Careerism--materialism* means seeing the degree as a tool for achieving a certain social and economic status in life. *Default motivation* refers to a situation in which students do not really know why they are attending higher education, just that they consider it a better option than the alternatives. Previous studies with Canadian and Finnish university students showed that these study motives are connected to the nature of students' engagement and progression in education ([@B16], [@B17]; [@B44]; [@B75]). Specifically, Coté and Levine (1997) showed that personal-intellectual motivation predicted the development of good self-management and self-motivation skills, while default motivation was related to a poorer prognosis for skill development and academic achievement.
University students adopt different approaches to learning depending on the task at hand, their skills and strategies, as well as the characteristics of the learning environment ([@B59]; [@B20]). Students with a *surface approach to learning* tend to memorize facts and reproduce information, and as a result, have fragmented knowledge ([@B20]; [@B56]). Students who apply the *deep approach* have intention to analyze and understand and thus, utilize multiple strategies in their learning to evaluate and relate the contents to be studied ([@B20]). The *strategic approach* focuses on students' intention to achieve the highest grades and especially, on their method for regulating their studying effectively ([@B19]). A previous study viewed the deep approach to learning as a component of cognitive engagement, and it is at least partly a matter of definition whether approaches to learning and engagement are different or overlapping phenomena ([@B42]).
Intention to drop out has been found to be related to challenges experienced in education and to weak engagement, which can further be seen to be connected to slow progress in education ([@B45]). Specifically, [@B44] found expectation-driven motivation and default motivation predict problems in adapting to one's program and motivating oneself to study or the lack thereof, in the case of personal-intellectual motivation. Intention to drop out can be related either to social or academic aspects of studies, or reasons can be found outside the department, for example, due to changes in life and work situations. Intention to drop out can develop for many different reasons, and the intention tends to evolve slowly. They are serious indications of problems in the students' well-being and engagement, and they are associated with students' self-regulation skills, interactions in the study-related communities and academic experiences in their program ([@B37]; [@B64]). In other words, intention to drop out shows students' overall disengagement with their education and the communities in which the educational programs take place.
Network Analysis as a Methodological Basis for Studying Engagement
------------------------------------------------------------------
We perceive engagement as a dynamic phenomenon that emerges from the complex interactions of the components. For instance, interactions among people are classic examples of systems that can be modeled as networks where nodes correspond to people, and edges connecting nodes correspond to the nature of the relationship. Recently, the idea of networks was applied to descriptions of psychological phenomena, such as depression ([@B28]; [@B29]), posttraumatic stress disorder (PTSD; [@B62]), intelligence ([@B87]) and health-related quality of life ([@B46]). In psychopathological network models, individual symptoms figure as the nodes. This is in marked contrast to the traditionally employed latent variable models in which the components of psychological characteristics are seen as passive reflections of the underlying constructs. Further, although latent variable models are premised on the idea that the components are interchangeable, in network models the centrality or importance of the components can be assessed ([@B62]). Therefore, we think that network analysis is well suited for describing how the phenomenon of engagement is formed when its components influence each other in complex ways.
In psychometric network models, the term "component" refers to a part of the network that bears unique causal relations to the rest of the network ([@B18]; [@B6]). For instance, perceiving one's education as meaningful has a cognitive component (M1, perceiving one's education as supporting self-development) and an emotional component (M2, being enthusiastic about education) that are differentially related to the remaining components of engagement. Accordingly, we interpret the study findings as causal hypotheses, while bearing in mind that (1) conditioning on a common effect of two variables may introduce a spurious edge into the graph ([@B24]; [@B74]), (2) a clique in the network graph may indicate the presence of an unmodeled latent variable ([@B24]) and (3) the direction of the potential causal effect cannot be inferred from the network graph ([@B24]).
Background variables or outcome variables can be included in a network model ([@B62]). We adopt the idea and assess the relations between the components of engagement and motives for attending university, students' approaches to learning, their intention to drop out of their program and students' certainty about their chosen field of study. These factors have been proven to be central for successful engagement, or the opposite, for disruptive engagement in the first year of higher education. For instance, problems in self-regulation and management of one's own learning are obvious among disengaged and slowly progressing higher education students in Finland ([@B44]). Furthermore, uncertainty about chosen field of study and intention to drop out are typical of this disengaged group. When examining connections to motives for attending university, problems managing learning correlate in particular with the default motivation and somewhat with the expectation-driven motivation ([@B44]). These factors related to study motives, students' approaches to learning, and their intention to drop out form a complex set of intertwined factors. It has also been observed that dropping out of school is most common during the first year of higher education ([@B73]; [@B45]). Therefore, this phase is crucial for the education career as a whole. Further, from the perspective of beginning higher education, and continuing, it is crucial that students feel that their chosen field of study is right for them. In contrast, the unstructured default study motivation seems to lead to problems in managing one's learning and to intention to drop out ([@B44]).
Aim and Research Questions
--------------------------
This study used network analysis to gain a better understanding of the intrinsic (individual/psychological) and extrinsic (collective/contextual) components of student engagement and the complex associations between these aspects. Based on previous research, we developed a student-centered method of measuring engagement in different disciplinary contexts. To identify the key components and structure of engagement, the following research questions are addressed:
1. How do the core components of engagement interact to give rise to the phenomenon of engagement?
2. How are the components of engagement related to
1. students' motives (especially personal-intellectual and default motives) for attending the university?
2. students' approaches to learning?
3. whether students intend to drop out?
Materials and Methods {#s1}
=====================
Participants, Design, and Data Collection
-----------------------------------------
The target population of the study were the all first-year students in different disciplines at 13 Finnish universities. The aim was to collect a nationally representative sample of the Finnish first-year university students. The sampling procedure was designed in co-operation with the Finnish Research Foundation of Studying and Education (OTUS), which also conducted the data collection in connection with the Finnish Student Barometer survey ([@B76]). The final target population was 16,972 (where men 42.5% and women 57.5%). This describes the gender distribution of Finnish university students where majority are women, like in studied year in target population (43.0% men and 57.0% women: [@B65]). From the target population, about one-third were randomly chosen to the sample population for the study. Because of a small sampling error, medical students were excluded from the final sample and the remaining sample population size is 6,040 (where men 42.5% and women 57.5%), as mentioned in [Table 1](#T1){ref-type="table"}.
######
The target population, the sample population and the actual participants by discipline.
Academic field New students 2012 Sample population \% of new students Participants Response %
---------------------------- ------------------- ------------------- -------------------- -------------- ------------
Agricultural sciences 362 233 64% 96 41.2%
Arts 413 139 34% 56 40.3%
Business and management 2216 495 22% 139 28.1%
Educational sciences 2141 815 38% 327 40.1%
Engineering and technology 2539 589 23% 173 29.4%
Pharmacy 366 161 44% 73 45.3%
Health sciences 436 148 34% 72 48.6%
Humanities 2408 940 39% 454 48.3%
Law 561 295 53% 117 39.7%
Natural sciences 3118 1200 38% 431 35.9%
Psychology 218 94 43% 59 62.8%
Social sciences 1846 697 38% 316 45.3%
Theology 279 179 64% 72 40.2%
Veterinary medicine 69 55 80% 37 67.3%
Total 16972 6040 36% 2422 40.1%
The University of Helsinki was weighted in the sample population because there were aims to use collected data in their own development work, but otherwise, each first-year student from each university and each discipline had the same probability of being included in the sample population. The survey was targeted at Finnish students and was implemented in Finnish; therefore, respondents from the majority population were emphasized. The share of international students in Finnish higher education at the time of the survey was 9.7% of all students ([@B12]). Because the largest university in Finland, the University of Helsinki, was emphasized in the sample, less common academic fields, such as veterinary medicine, theology and agriculture and forestry, were slightly overrepresented in the sample population.
The final sample, those who participated in the survey during the academic year 2012--2013, comprised altogether 2,422 first-year students \[men: 574 (23.7%); women: 1,848 (76.3%)\]. They ranged in age from 19 to 67 years (mean: 24.1 years, median: 22.0 years; [Table 1](#T1){ref-type="table"}). Almost all participants (98.1%) were Finnish citizens. The final response rate was satisfactory 40.1%.
Data for this study were collected with an extensive online questionnaire ([@B76]), which included background information, previous education, application motives, education progress, first-year experiences, values, attitudes, well-being, subsistence and employment. The authors of this study suggested three additional specific areas for the online questionnaire: student engagement, motives for attending university and learning approaches. These were utilized in this study. The students' participation in the study was voluntary, and the participants' informed consents were guaranteed. Informed consent was inferred from the participants' returning of the questionnaires. The study did not include any threat to physical integrity, children under the age of 15, strong stimuli, mental harm or the risk of safety for participants (cf. ethical principles of research in the humanities and social and behavioral sciences and proposals for ethical review, prepared by the [@B25]). According to the principles, this study did not require ethical review and approval in Finland.
Measures
--------
For this study, we utilized the following measures and scales included in the 2013 Finnish Student Barometer survey \[[@B76]: the Engagement Evaluation Questionnaire (EEQ; [@B43], [@B42]), the Student Motivations for Attending University (SMAU) questionnaire ([@B16], [@B17]) and Learning Strategy scales from the HowULearn questionnaire ([@B67])\]. In addition, questions about intention to drop out and chosen field of study were included.
### Engagement Evaluation Questionnaire
The EEQ was used to measure the different dimensions of the student engagement process ([@B43], [@B42]) and was utilized as the main measure to understand the multidimensionality of student engagement during the first year of higher education. The EEQ was developed to measure the theory-based dimensions of student engagement and aimed at getting an overall picture of the three overlapping engagement processes: individual, collaborative, and engaging ([@B41]). These processes are operationalized into six subscales: meaning (M), academic skills (Sk), participation (Pa), social practices (Pr), sense of belonging (B), and identity (I). All items on these subscales were measured with a seven-point Likert scale (1 = strongly disagree, 7 = strongly agree). The EEQ was applied in a 12-item version form in this study (six subscales, two items on each subscale). The psychometric properties of the EEQ measurement from the point of view of classical test theory has been demonstrated in our previous study concerning transitions between first and second year in university education ([@B42]), where Cronbach alpha test values for subscales in both years were for meaning (M) 0.82--0.83, academic skills (Sk) 0.70--0.73, participation (Pa) 0.75--0.75, social practices (Pr) 0.50--0.64, sense of belonging (B) 0.69--0.76 and identity (I) 0.74--0.80.
### Student Motives for Attending University
Student motives for attending university and their possible connections to the engagement dimensions were an area of interest in this study. SMAU is a questionnaire that has five scales measuring different motives for attending university: personal-intellectual development, humanitarian, expectation-driven, careerism--materialism and default ([@B16], [@B17]). The original SMAU questionnaire consists of 23 items (Coté and Levine, 1997), and it resembles other widely used student typologies (i.e., [@B2]). Coté and Levine (1997) stated that their typology better reflects attitudes and motivations formed before university participation. We utilized a shortened 15-item version of the SMAU questionnaire, where the three top-selective items were chosen for each of the five subscales. The subscales describing different motives for attending university are personal-intellectual development, humanitarian, careerism--materialism, expectation-driven and default motivation. All items in these scales were measured in this study with a seven-point Likert scale (1 = strongly disagree, 7 = strongly agree). The SMAU questionnaire has been tested to be a reliable instrument in previous Canadian ([@B16], [@B17]) and Finnish studies ([@B44]; [@B75]) concerning university students.
### Approaches to Learning
We also observed the use of learning strategies and their relationship to the different areas of student engagement in the first year of higher education and adopted three appropriate scales (deep, systematic and surface) from the HowULearn questionnaire used previously with Finnish university students ([@B67]). We measured the approaches to learning on a short two-item form in each of the three learning strategy scales. The items on the scales were measured with a seven-point Likert scale (1 = strongly disagree, 7 = strongly agree). The validity and reliability of these scales were demonstrated in previous studies in different learning contexts (see [@B68]).
Analysis Methods
----------------
### Network Analysis
When network models are used as psychometric models, the key difference from the traditional application areas of network analysis is that the network weights are parameters whose values are estimated from the data ([@B21]). After the estimates have been calculated, traditional methods of characterizing networks, such as different centrality indices, can be calculated the same way as for other network models. Network weights in psychometric models have been estimated by calculating either correlations or partial correlations ([@B21]), resulting in both cases in a signed and weighted network model. However, it has become customary to analyze data in the form of partial correlations ([@B21]), as pairwise associations, when controlling for the effects of the other variables, are the phenomenon of central interest in network models. A non-zero edge may indicate potential causal connections ([@B23]), logical relationships among the nodes ([@B46]), while the possibility remains that a cluster of nodes is formed because of an unmodeled latent variable influencing all the nodes in the cluster ([@B32]).
Partial correlations are often estimated using a method known as the graphical lasso, which constrains small correlations to zeroes ([@B21]), thus avoiding capitalizing on sampling variation and results in what are known as sparse models. In this contribution, the level of sparsity was determined by the tuning parameter λ, the value of which was chosen based on the extended Bayesian information criterion (EBIC; [@B26]). The EBIC model selection is governed by the hyperparameter γ, which was set to the recommended default value of 0.5 ([@B26]). The locations of the nodes were determined using a modified version of the [@B31] algorithm for weighted networks ([@B22]), which places strongly connected nodes that have many edges in common close to one another. All network analyses were performed in R using the packages qgraph ([@B22]) and bootnet ([@B21]).
### Centrality Indices in Network Models
We used three centrality indices to characterize the networks: strength, betweenness and closeness centrality ([@B66]). The first is defined simply as the sum of the absolute values of the weights of the edges connected to the focal node, and thus, this index describes the extent that a node is connected to other nodes. The two other centrality indices are defined using the concept of distance between the nodes in a network. In a weighted network, the distance between two nodes is defined as the inverse of their connection weight. Betweenness centrality is the distance of the focal node to other nodes, and it quantifies the importance of a node in connecting other nodes of the network. Closeness centrality is the inverse of the average distance of the focal to node to other nodes in the network, and it quantifies the degree to which the node is indirectly connected to other nodes in the network. For a succinct introduction to the centrality indices, please see [@B15].
### Accuracy and Stability of the Results
To assess the potential replicability (the stability and accuracy) of the results, we performed bootstrap analyses in which we calculated the confidence intervals for the edge weights and the centrality indices. The bootstrap analyses help assess the degree to which the results are affected by sampling variability. This procedure is described by [@B21].
In assessing the *accuracy* of the edge weight estimates, we used non-parametric bootstrapping (resampling the data with replacements) because most of the input variables are ordinals. Further, we assessed the *stability* of the centrality indices using the so-called *case-dropping subset bootstrap*, which involves repeatedly calculating the values of the centrality indices based on different subsets of data. If the results depend on carrying out the analyses specifically on the original sample, then they cannot be considered stable. Stability is assessed using the correlation stability coefficient (CS coefficient), which indicates the maximum proportion of cases that can be dropped such that with 95% probability the correlation between the original centrality indices and those calculated based on the subset exceeds 0.7 (Cohen's suggested value indicates a very large effect) (cf. [@B14]; [@B21]). The values of the CS coefficient should, at a minimum, exceed 0.25, and preferably be larger than 0.5 ([@B21]).
We assessed the differences between the values of the edge weights and the centrality indices by calculating bootstrapped confidence intervals for the difference scores of all pairs of edge weight or centrality index values. If the bootstrapped confidence interval for the difference score does not cover the value of zero, the edge weights or the values of the centrality indices can be said to differ from one another ([@B21]). No correction for multiple comparisons was made (for a discussion of the problematicity of performing such corrections in this context, see [@B21]). This procedure is called the bootstrapped difference test.
Results
=======
Engagement Phenomenon in the Total Sample of First-Year Students
----------------------------------------------------------------
Descriptive statistics related to the engagement items on the EEQ are displayed in [Table 2](#T2){ref-type="table"}, which shows that network analysis is a feasible option for these data: The item means are not extremely high or extremely low, and all variables exhibit roughly similar amounts of variability. As described, the engagement items and dimensions were operationalized into six subscales on the EEQ, including two items in each subscale: meaning (M), academic skills (Sk), participation (Pa), social practices (Pr), sense of belonging (B), and identity (I).
######
Descriptive statistics of the engagement items and the covariates.
Item Mean SD Skewness Kurtosis
--------------------------------------------- ------ ------ ---------- ----------
M1 (self-development) 5.75 1.18 --1.27 2.37
M2 (enthusiasm) 5.53 1.36 --1.00 0.86
Pa1 (not knowing others) 2.17 1.63 1.43 1.14
Pa2 (contacts with others) 5.04 1.62 --0.76 --0.13
Pr1 (working in small groups) 3.54 1.78 0.26 --0.97
Pr2 (education as a solitary enterprise) 4.55 1.59 --0.35 --0.67
Sk1 (scheduling) 3.69 1.70 0.19 --0.92
Sk2 (regular studying) 4.75 1.55 --0.57 --0.34
I1 (fit in well as a university student) 5.48 1.29 --0.85 0.53
I2 (have found an appropriate study method) 4.75 1.41 --0.41 --0.31
B1 (belongingness) 5.40 1.41 --0.93 0.64
B2 (alienation) 2.27 1.48 1.19 0.76
Personal-intellectual motivation 5.70 1.06 --1.08 1.52
Default motivation 2.45 1.35 0.83 --0.05
Certainty about field of study 1.74 0.44 --1.11 --0.76
Intention to drop out 1.26 0.44 1.07 --0.87
Deep approach 5.15 1.02 --0.41 0.19
Strategic approach 4.20 1.38 --0.03 --0.62
Surface approach 3.61 1.29 0.17 --0.30
Meaning (M) and academic skills (Sk) represent the individual process of engagement. The two items related to experiencing one's education as meaningful are M1 (self-development) and M2 (enthusiasm), while the two items related to academic skills (Sk) are Sk1 (scheduling) and Sk2 (regular studying). Participation (Pa) and social practices (Pr) represent the collaborative process of engagement. The items related to participation are Pa1 (not knowing others) and Pa2 (contacts with others), while the two items related to social practices are Pr1 (working in small groups) and Pr2 (education as a solitary enterprise). Finally, sense of belonging (B) and identity (I) represent the overarching properties of the engagement process. The items related to sense of belonging (B) are B1 (belongingness) and B2 (alienation), while the items related to identity are I1 (fit in well as a university student) and I2 (have found an appropriate study method). From Suppelementary Tables [S1](#SM1){ref-type="supplementary-material"} and [S2](#SM1){ref-type="supplementary-material"} more information on inter-item relations.
The lasso-estimated partial correlation network of the 12 engagement items is shown in [Figure 2](#F2){ref-type="fig"} with the associated centrality indices. The colors of the nodes were chosen to reflect the composition of the phenomenon of engagement. The strength of the partial correlations among the components of engagement is reflected in the width and saturation of the edges connecting the nodes with blue edges corresponding to positive associations and red to negative ones. For instance, an extremely strong connection between the two meaning items (M1 and M2) remains when controlling for the other connections. The edge weights were estimated accurately as evidenced by the narrow confidence intervals in [Supplementary Figure S1](#SM1){ref-type="supplementary-material"}. This enables us to interpret differences among the edge weights. The values of the centrality indices were similarly stable for changes in the composition of the sample ([Supplementary Figure S2](#SM1){ref-type="supplementary-material"}). The centrality index values remained extremely stable even when up to 70% of the original cases were dropped. The CS index values were similarly high for all three centrality indices, CS (cor = 0.7) = 0.75 for all three indices. These results show that all three centrality indices are interpretable as they stand.
{#F2}
The centrality indices in the lower part of [Figure 2](#F2){ref-type="fig"} indicate unequivocally that the most central components of engagement were the experiences of belonging and alienation (nodes B1 and B2) and assuming the role of a student (nodes I1 and I2). The bootstrapped difference tests ([Supplementary Figure S3](#SM1){ref-type="supplementary-material"}) showed that the centralities of these nodes differed for the most part from the rest of the network but were very similar to each other. The high closeness centralities indicate that all other nodes can be easily reached from these four nodes via direct or indirect paths. Changes in the closeness centrality nodes had a marked effect on other nodes in the network. Nodes B2 (alienation) and B1 (belonging) had the highest betweenness centralities in the network. In this model, the alienation node links the nodes related to participation (Pa1 and Pa2) and social practices (Pr2) to the rest of the network through belongingness (B1) and enthusiasm (M2). Finally, being enthusiastic about one's education (M2) had the highest strength centrality in the network ([Supplementary Figure S3](#SM1){ref-type="supplementary-material"}), and thus, was related to many other parts of the engagement phenomenon whereas there are only very weak links from M1 (self-development) to the rest of the network.
Analyses With Covariates
------------------------
In addition to the network analysis of student engagement, we investigated the ways that the students' motives for education, approaches to learning and intention to drop out were related to students' engagement in education. The analysis was exploratory: Based on the evidence presented in the Introduction, we assumed that the students' motives to study would be related to engagement, although we could not derive exact hypotheses from previous research. Therefore, we first performed an analysis that included all five motives and approaches to learning, intention to drop out and students' certainty about their chosen field of study. These preliminary analyses are reported in the [Supplementary Figures S4](#SM1){ref-type="supplementary-material"}--[S7](#SM1){ref-type="supplementary-material"}. The analyses showed that of the five motives, the default motivation and personal-intellectual motivation were linked to various components of engagement. Thus, and for the sake of simplicity, we report the network model incorporating these motives in [Figure 3](#F3){ref-type="fig"}.
{#F3}
The edge weight estimates were very accurate as indicated by the narrow confidence intervals in [Supplementary Figure S8](#SM1){ref-type="supplementary-material"}. The estimates of the values of the centrality indices were similarly stable ([Supplementary Figure S9](#SM1){ref-type="supplementary-material"}), with the CS index was 0.75 for all three indices.
Personal-intellectual motivation was positively associated with the components of engagement that were deemed central in the model shown in [Figure 2](#F2){ref-type="fig"} (B1, feeling like belonging to the university, and I1, fitting in well as a university student). Similarly, having a deep approach to learning and perceiving education as supporting self-development shared positive links with personal-intellectual motivation. Interestingly, the more certain the students were about their field of study, the lower personal-intellectual motivation they had. Default motivation, however, was related to considering dropping out of school and feeling alienated from the university (B2) and shared a negative link with being enthusiastic about one's education.
When the associations that the approaches to learning had with the rest of the network were examined, the deep approach was positively associated with personal-intellectual motivation as observed above. Personal-intellectual motivation also had quite a strong positive association with the strategic approach, which, for its part, was most strongly connected with the nodes related to study skills (Sk1, Sk2, and I2). The surface approach was related to intention to drop out and not having found an appropriate study method.
Last, intending to drop out from one's university was related, through a negative edge, to certainty about one's field of study and being enthusiastic about one's education. Further, this node shared a positive edge with default motivation, feeling alienated from the university and the surface approach to learning. Interestingly, intention to drop out emerged as the third most strength central node in the network, signifying that this node had strong connections to the rest of the network. Intention to drop out had similarly high closeness centrality. The statistical significance of the differences between the centrality index values is shown in [Supplementary Figure S10](#SM1){ref-type="supplementary-material"}. Intention to drop out had closeness and strength centrality values that did not differ statistically from those of B1-2 and I1-2, although this node was more central than most of the other nodes in the network. The betweenness centrality of intention to drop out, however, was affected by sampling variability: This index did not differ from the rest of the nodes as clearly as the other centrality indices. This pattern of results can be interpreted such that intention to drop out was easily affected by the other nodes in the network and affected them in return, and that its connections to the other nodes of the network were quite strong. Nevertheless, whether the shortest paths between other pairs of nodes go through intention to drop out remains an open question.
Discussion
==========
The present study examined student engagement in education as a network of interlinked components, building on the theoretical conceptualization of engagement presented in [@B41] and [@B42]. The structure of the network model of engagement was remarkably similar to that of the theoretical conceptualization. Research question one about core components of engagement was confirmed in the results both theoretically and through the network model. According to the theoretical assumptions of the process model of engagement, successful long-term engagement builds on an emerging sense of belonging and an evolving identity as a university student ([@B41]; [@B42]). Theoretically, these overarching components of engagement bind together the remaining components of engagement: meaningfulness, participating in social practices and study skills. This is what we found in the network model: The nodes related to sense of belonging (B1, B2) and identity (I1, I2) figured as the most central ones in the network, connecting the remaining components. Interestingly, the centrality of the nodes in the network bore no obvious relationship to how commonly the corresponding statements were endorsed. For instance, the most commonly endorsed item M1 ("my education supports my self-development") was among the least central nodes in the network, and the extremely infrequently endorsed item B2 ("I feel alienated from the university") was among the most central nodes.
The present contribution can be considered substantive-methodological synergy ([@B5]; [@B58]), in which a novel analysis method was used to address a substantively important research question based on a recently developed theory. In particular, student engagement is a dynamic phenomenon that emerges from the interaction of its components ([@B44], [@B45]), and network analysis is a method that is naturally suited to analyzing the dynamics in such systems ([@B4]). Further, network models can assess the centrality of the components of a phenomenon (for the definition of the term "component," see the section "Introduction"), unlike factor analysis that treats all observed variables as equally good indicators of the latent traits. Assessing the centrality of the items allows us to draw conclusions concerning the nature of the phenomenon of engagement that would not be possible based on factor analysis of the same data. For instance, when looking at the nodes related to the meaning of studies, node M1 (self-development) represents the cognitive aspect of meaning and is not particularly central in the network, whereas node M2 (enthusiasm) represents the emotional aspect of meaning and is a much more central node in the network. In the social practices scale item Pr1 (working in small groups) seemed to be positioned less centrally than compared to its counterpart item Pr2 (solitary enterprise), which raises questions about the sparsity of collaborative engaging practices in general.
In response to the research question two, a model including certain covariates of engagement, identified in previous studies ([@B44]; [@B38]; [@B63]), was presented. In the network model with covariates, personal-intellectual motivation shared a positive edge with enthusiasm (M2), which, in turn, shared a negative edge with intention to drop out. Thus, it can be hypothesized, based on the network model, that the relationship between personal-intellectual motivation and intention to drop out observed here and in [@B44] study is mediated through enthusiasm (but see the "Limitations" section for a warning about over-interpreting the present results). Further, when student approaches to learning are investigated in the network model, the deep approach has a strong connection to personal-intellectual development (Pr-), which appears to mediate the relationship of the deep approach to nodes related to meaning, sense of belonging and identity. In addition, the direct connections between the deep approach and the cognitive component of meaningfulness (M1) and having found an appropriate study method (I2) are reminiscent of the way the deep approach is occasionally seen as part of cognitive engagement ([@B27]). The strategic approach is strongly connected to components of academic skills (Sk1, Sk2) and the other component of identity (having found an appropriate study method, I2).
The intention to drop out became one of the most central nodes in the network, with the strength and closeness centrality values exceeding those of the other nodes of the network. As the edges in psychometric network models can be interpreted as causal hypotheses ([@B6]; [@B24]), it is of interest to examine the connections between intention to drop out and the rest of the network in more detail. The negative edges from intention to drop out (Drp) to feelings of enthusiasm (meaning, M2) and certainty concerning the chosen field of study (Crt) suggest that enthusiasm and certainty protect students against intention to drop out, whereas feelings of alienation (belongingness, B2), the surface approach to learning (Srf) and default motivation (Dfl) function as predisposing factors. The result is in line with that of [@B44] who found the personal-intellectual motivation and default motivation are similarly related to intention to drop out.
Limitations of the Study
------------------------
As noted above, it is a central assumption in psychometric network models that the nodes of the network correspond to components of the phenomenon under investigation. This may not be the case for all the nodes in network models. For instance, the strong edges connecting the strategic approach to learning with the items related to academic skills (Sk1 and Sk2) may be related to the fact that these phenomena were operationalized using similarly formulated questions. It may be that adding the node "strategic approach" to the network provides little unique information, over and above that included in the two skills-related nodes. Further, would the models have been robust to the replacement of the two practices-related items with ones that tap into other components of practices than the individual and collective aspects of engagement? Similar considerations apply to the choice of covariates, which may affect the relationships among the rest of the nodes. However, the relationships between the engagement-related nodes were quite robust to the addition of the covariates (the nodes did not essentially change with the addition of the covariates), which shows that the results reported in the first network model were not artifacts caused by the covariates added to the second model.
One obvious shortcoming of the present study is that we were not able to test the causal assumptions and hypotheses that were formed based on the two network models. This is because the present study was based on a cross-sectional sample instead of time-series data. Intensive longitudinal data, perhaps collected using the experience sampling method (ESM), might prove useful in making more educated guesses concerning the presence and direction of causal links among the components of engagement. This would enable us to think about causal effects at least in terms of Granger causality, which is based on the fulfilling the temporal requirement of causality, that is, the cause preceding the effect ([@B24]).
Finally, an important potential limitation of the network models must be discussed. It remains possible that some of the non-zero edges in the network models were artifacts due to conditioning on a collider variable. A collider is defined for a pair of variables as a third variable that is causally influenced by both. When one statistically or through experimental design controls for a collider variable, a spurious association may appear between the two variables. These ideas are introduced clearly and using intuitive examples in [@B74] in an article that considers problems inherent in drawing causal inferences based on correlational data. Although it is possible that in the future novel statistical methods might allow researchers to diagnose such situations, for now the advice given to applied researchers is to use their own judgment in choosing the relevant variables to include in a network model and to not over-interpret the results as unproblematically representing reality ([@B30]). In the present covariate model, the negative edge between personal-intellectual motivation and certainty concerning one's field of study may be a candidate for a spurious edge that is due to conditioning on a collider. It is possible that the two would function as causes of M1 (education supports self-development); That is, the more one emphasizes personal growth as a motivation for studying and the more certain one is about one's chosen field of study, the more one comes to think that one's education supports one's self-development. When we inspect the pattern of zero-order correlations ([Supplementary Table S3](#SM1){ref-type="supplementary-material"}) against the network model, we notice that the correlation between personal-intellectual motivation and certainty is practically zero. The triangle formed between Pa1, Pa2, and Pr2 is another candidate for potential spurious associations, but when we examine the zero-order correlations ([Supplementary Table S3](#SM1){ref-type="supplementary-material"}) among these variables against the network model (or its adjacency matrix in [Supplementary Table S4](#SM1){ref-type="supplementary-material"}), we notice that they are both of an equal sign, with the zero-order correlations just slightly stronger. To conclude, it cannot be ruled out that the former of these examples could be due to conditioning on a collider, whereas we do not believe the latter is an example of a spurious result.
Implications for Higher Education and Suggestions for Further Studies
---------------------------------------------------------------------
The unique benefit of this analytical approach was that by calculating different centrality indices we quantified the importance of the individual components of engagement. This is in marked contrast to previous studies of student engagement (for a comprehensive review, see [@B38]), which conceptualized the dimensions of engagement as latent variables and thus, perceived the observed variables as equally good indicators of the latent dimensions, differing only in the amount of error variance of each observed variable.
Thus, based on the present findings, we can make certain novel practical recommendations concerning interventions for students who struggle with engagement during their first year of higher education. First, if a student feels alienated from the university, this should be an immediate warning sign of potential problems in other areas of engagement, such as skills, practices, and participation. In practice, feelings of alienation could be targeted by, for example, providing low-threshold access to education psychologists and encouraging students to contact the professionals immediately if the students experience feelings of alienation. Second, it is important to encourage first-year students to consider from the very beginning of their education an appropriate study method. For instance, planning systematically interactive first-year courses and intensive peer-group activities and including discussion groups that focus on different study methods might encourage first-year students to think about the aspects of engagement that are the best for them. In addition, by early identification of drop-out intentions or shortcomings in academic skills, it can be influenced the students to correct wrong study field choices or identify not being capable of studying in higher education. This can have a very positive meaning for the student personally.
In addition, using the model with covariates, we formulated causal hypotheses and tentative recommendations concerning preventing students from dropping out of education. It has been observed that most dropouts leave their study program or university during their first year ([@B73]; [@B45]). This phenomenon seems to be stable in Finnish higher education, despite the many reforms implemented over the years ([@B45]). As noted above, in the present study alienation (B2) seems to be a central mediating node between intention to drop out (Drp) and the nodes related to participation (Pa1, Pa2), as well as between dropping out and "belongingness" (B1). This may indicate that if one wants to prevent experiences of alienation and support engagement and feelings of belonging, it is necessary to create such educational environments for teaching and learning where it is natural to participate and work together. Further, we propose that an increase in alienation, a particularly central experience in the engagement network, will have a larger effect on the other components of engagement than an increase in a more peripheral experience, such as knowing other students (Pr1). In particular, studies focusing on individual-level networks of engagement in a time-series design (i.e., separate network models for individual students) would enable us to examine phase transitions from the engaged to the disengaged state. This process might share essential similarities with the proposed phase transition from the non-depressed to depressed state as examined in network models of psychopathology ([@B6]; [@B29]).
In addition, we hypothesize, based on the covariate model, that an intervention that increases students' enthusiasm (M2) for education will decrease their intention to drop out (other things being equal). However, an increase in the perception that education supports one's self-development (M1), may exert influence intention to drop out unless they also become more enthusiastic about their education (M2) and more certain that their field of study is the correct choice for them (i.e., the effect of the cognitive component of meaningfulness may exert influence intention to drop out through other phenomena). In short, network analysis provides interesting hypotheses to test in future studies.
A further research topic for the application of network analysis and models of engagement could be to consider the time dimension and in what ways the engagement phenomenon develops over a certain time period. For instance, when on a macro-level (institutionally or cross-institutionally) is monitored the same student cohort and their engagement with periodical measurements ([@B42]). Alternatively, in the micro-level case (selected sample of students), when the ESM is used as a structured diary method ([@B88]) and subjective experiences are assessed daily during the follow-up period. It would be interesting to more closely follow whether central nodes in the network work well for predictors for strong engagement or intention to drop out. All these options assume diverse follow-up study designs.
Data Availability
=================
The datasets generated for this study are available on request to the corresponding author.
Ethics Statement
================
This study was carried out according to the ethical guidelines of the Finnish National Board of Research Integrity (<https://www.tenk.fi/en>) for this kind of research in the Finnish research context. All participants received written informed consent in accordance with the guidelines.
Author Contributions
====================
VK, AT, MI, and MM: conceptualization. VK, AT, and MI: investigation and funding. VK and AT: project management. AT: resources. MM: methodology, software, analysis, validation, and visualization. VK, MM, and AT: original draft preparation and final editing.
Conflict of Interest Statement
==============================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
**Funding.** Grant from University of Helsinki, Finland received for the data collection and publication fee.
The authors would like to thank the Finnish Research Foundation of Studying and Education (OTUS) for the valuable co-operation in data-collection phases of this study that made this research possible.
Supplementary Material
======================
The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpsyg.2019.01056/full#supplementary-material>
######
Click here for additional data file.
[^1]: Edited by: Lynne D. Roberts, Curtin University, Australia
[^2]: Reviewed by: Stephen Larmar, Griffith University, Australia; Külli Kori, Tallinn University, Estonia
[^3]: This article was submitted to Educational Psychology, a section of the journal Frontiers in Psychology
| {
"pile_set_name": "PubMed Central"
} |
{#sp1 .136}
{#sp2 .137}
{#sp3 .138}
{#sp4 .139}
{#sp5 .140}
{#sp6 .141}
{#sp7 .142}
{#sp8 .143}
{#sp9 .144}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#sec1-1}
============
Chronic obstructive pulmonary disease (COPD), a chronic inflammatory disease, is a major worldwide health problem. According to the World Health Organization (WHO), 65 million people have moderate-to-severe COPD.\[[@ref1]\] Crude estimates suggest that there are 30 million COPD patients in India, and it contributes to a significant and growing percentage of COPD mortality.\[[@ref2]\]
Smoking and biomass exposure, along with genetic predisposition, are the major risk factors for developing COPD. Comorbidities are important events in the natural history of the disease and have a negative effect on the morbidity and mortality of COPD patients. Diabetes, hypertension, cardiovascular diseases, lung cancer, osteoporosis, and depression are common comorbidities seen in COPD patients.
Anemia is seen to be present as comorbidity in various chronic disease states, and therefore, understanding its pathogenesis is important. In recent years, anemia is also seen as a common comorbidity in COPD patients and associated with reduced functional capacity, impaired quality of life, greater likelihood of hospitalization, and early mortality.\[[@ref3]\] Thus, developing new tools for its treatment should be our priority.
Anemia of chronic disease (ACD) is an immune-driven abnormality associated with chronically very low levels of circulating hemoglobin that has been seen to occur in many inflammatory diseases. The systemic inflammation that is now recognized as a feature of COPD makes it a possible cause of ACD. If present in COPD, anemia could worsen dyspnea and limit exercise tolerance.\[[@ref4]\]
Hemoglobin is the principal oxygen transport molecule. As per the WHO, men with hemoglobin levels \<13 g/dl and women with hemoglobin levels \<12 g/dl are defined as anemic.\[[@ref5]\] Any decrease in hemoglobin levels results in a corresponding reduction in the oxygen-carrying capacity of the blood. Impairment of this mechanism exerts a negative impact on clinical status.\[[@ref6]\] The prevalence of anemia in patients with COPD varies from 7.5% to 33%,\[[@ref3][@ref7]\] and this variability might be due to various methods of studies, selection of patient group, and various definitions of anemia.\[[@ref8]\] We conducted a study in North Indian COPD population to evaluate the prevalence of anemia and to study its association with various parameters.
Materials and Methods {#sec1-2}
=====================
Study population and selection of subjects {#sec2-1}
------------------------------------------
The present study was carried out in the Department of Respiratory Medicine, King George Medical University, Lucknow. The study was approved by the Institutional Ethical Committee. One hundred and fifty stable COPD patients were enrolled from the Outpatient Department of Respiratory Medicine from October 2015 to January 2017 after obtaining written informed consent from all the patients.
The diagnosis of COPD was based on pulmonary function test which was done in all patients. According to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria, COPD was defined on the basis of the postbronchodilator forced expiratory volume in 1 s (FEV1)/forced vital capacity (FVC) ratio of \<0.70 and reversibility to an inhaled bronchodilator in FEV1 \<12% or \<200 ml after administration of 200 μg salbutamol (2 puffs) using a pressurized metered-dose inhaler with a spacer. The GOLD system categorizes airflow limitation into four stages in patients with FEV1/FVC \<0.70.\[[@ref8]\]
GOLD stage 1 -- mild: FEV1 ≥80% predictedGOLD stage 2 -- moderate: 50% ≤FEV1 \<80% predictedGOLD stage 3 -- severe: 30% ≤FEV1 \<50% predictedGOLD stage 4 -- very severe: FEV1 \<30% predicted.
Anemia was defined according to the WHO criteria as hemoglobin level \<13 g/dl in men and 12 g/dl in women. COPD patients were divided into two groups: anemic (Group 1) and nonanemic (Group 2) on the basis of this definition.
Patients reporting with a history of pulmonary tuberculosis, cardiac diseases, interstitial lung disease, pregnancy, diabetes, and cancer were excluded from the study. Patients with any other systemic disease other than COPD and having Vitamin B12 or folic acid deficiency were also excluded from the study. A detailed clinical history of respiratory symptoms was also obtained. Chest X-ray, spirometry, and routine blood investigations were done in all patients. Dyspnea was measured by the modified Medical Research Council (mMRC) Dyspnea scale in both the groups for assessing health status of a patient and grading the degree of a patient\'s breathlessness and disability caused.
Statistical analysis {#sec2-2}
--------------------
GraphPad PRISM version 6.01 (GraphPad software Inc.; La, Jolla, CA, USA). was used for the analysis of data. All demographic and clinical data were expressed as a mean ± standard deviation or percentage. The Chi-square test was used for categorical data and groups were compared by unpaired *t*-test or one-way analysis of variance. *P* \< 0.05 was considered statistically significant.
Results {#sec1-3}
=======
In this study, 150 COPD patients representing all stages of disease severity as defined by GOLD were recruited. The baseline characteristics of the study groups are shown in [Table 1](#T1){ref-type="table"}. As per the WHO criteria for patients to be anemic in our study, we found 47 patients in anemic group while 103 patients in nonanemic group. The prevalence of anemia in COPD patients in the present study was 31.6%. Majority of patients were male in both the groups and proportions of males were slightly higher in nonanemic group as compared to the anemic group. The mean hemoglobin levels in anemic group were 11.04 ± 1.1 g/dl, whereas in nonanemic group, it was 13.9 ± 0.8 g/dl. Difference in body mass index (BMI) of anemic and nonanemic group was not found to be significant although mean BMI of anemic patients was lower than nonanemic. The age of the patients ranged from 35 to 75 years. Mean age of anemic COPD patients was slightly higher than nonanemic COPD patients.
######
Demographic profile and spirometry variables of anemic and nonanemic chronic obstructive pulmonary disease patients
There was a positive correlation of hemoglobin with FEV1% predicted (FEV1% pred, *r* = 0.36) and negative correlation with age (*r* = −0.02, *P* = 0.87) and BMI (*r* = 0.03, *P* = 0.8).
Proportions of anemic patients were higher in age group of 51--60 and 61--70 years in comparison to nonanemic patients which were higher in age group of 41--50 years \[[Figure 1](#F1){ref-type="fig"}\]. In anemic group, large number of patients was in age group of 61--70 years followed by 51--60 years.
{#F1}
According to the GOLD criteria, COPD patients were grouped into four stages based on their severity in both the COPD groups \[[Figure 2](#F2){ref-type="fig"}\]. Proportion of anemic patients was higher in Stage 3 and 4 (82%) while proportions of nonanemic patients were (65%) (*P* = 0.03). Mean values of spirometric variables such as FEV1% pred, FVC %, and FEV1/FVC were lower in anemic patients in comparison to nonanemic COPD patients. Patients with anemia had severer COPD.
{#F2}
Proportion of anemic patients was found to be higher (45%) than nonanemic of mMRC Grade 3 and 4, while proportion of nonanemic patients was higher of mMRC Grade 1 and 2 (77%) \[[Figure 3](#F3){ref-type="fig"}\]. Difference in mMRC grade of patients of anemic and nonanemic group was found to be statistically significant (*P* = 0.04).
{#F3}
Discussion {#sec1-4}
==========
Anemia is one of the extrapulmonary manifestations of COPD. The prevalence of anemia has been reported between 7.5% and 33%.\[[@ref3][@ref7]\] John *et al.*\[[@ref9]\] in 2006 reported the prevalence of anemia in COPD patients to be 23.1%, while another study observed anemia in 13% of 101 COPD patients and they pathogenetically related it to the presence of inflammation.\[[@ref10]\] Halpern *et al.*\[[@ref11]\] found anemia in 21% of total patients with a COPD diagnosis while Parveen *et al.*\[[@ref12]\] in a hospital-based, cross-sectional study in 200 COPD patients reported anemia in 18% of the patients. Of 107 consecutive patients hospitalized with an acute exacerbation of COPD patients (AECOPD), 47 (43.9%) were found to be anemic on admission in a study by Silverberg *et al.*\[[@ref13]\] One recent study by El-Korashy *et al.*\[[@ref14]\] showed that almost half of patients have anemia; however, one earlier study reported a very low prevalence of anemia in COPD patients that is 6%.\[[@ref15]\]
This difference in the prevalence rate can be due to various factors, such as type of studies; COPD patients in the study (either stable or exacerbated or admitted patients); the use of different cutoff levels of hemoglobin to define anemia; or the existence of different confounding factors such as the presence of other known causes of anemia, such as heart failure and renal failure. Shorr *et al.* in a retrospective data analysis of 2404 COPD patients from the USA also reported a very high frequency of anemia in COPD patients of 33%,\[[@ref16]\] which was comparable to our study.
We found the prevalence of anemia in COPD patients to be 31.6%. Pulmonary function variables (FEV1, FVC, and FEV1/FVC) were lower in anemic patients in comparison to nonanemic COPD patients although difference was not significant, but we do find that patients with anemia had more severe COPD in terms of the postbronchodilator FEV1%, which was in accordance to the study by Casanova *et al.*\[[@ref15]\] In a previous cross-sectional study by Zavarreh *et al.*,\[[@ref17]\] in 760 COPD patients, they found no correlation between severity of COPD and anemia, but they found anemic patients (71.1 ± 8.5) to be significantly older than nonanemic patients (65.4 ± 12.8) (*P* = 0.03).
Likewise, in our study, the anemic patients were older than nonanemic and proportion of anemic patient was more in higher age group (61--70). Therefore, we suggest that hemoglobin levels estimation should necessarily be done in older age group. Proportion of anemic patients was higher in Stage 3 and 4 which were also observed in a study by Parveen *et al.*\[[@ref10]\] We also found anemia to be associated with increased morbidity in the form of number of exacerbation and hospital admissions. Ferrari *et al.* associated anemia with increase of dyspnea and deterioration in a health-related quality of life.\[[@ref18]\] In patients with COPD and chronic respiratory failure, higher hemoglobin level was associated with longer survival.\[[@ref19]\] In a cohort study of 5683 stable COPD outpatients, anemia was present in 116 (17%) patients, and these patients showed a significantly higher mMRC score, lower 6-min walk distance, and shorter median survival (49 vs. 74 months) than nonanemic patients.\[[@ref20]\] Anemia was significantly associated with increased dyspnea in our study which was assessed by mMRC grade (*P* = 0.04). We also found association between worsening of dyspnea (calculated as per mMRC score) with anemia as 45% of anemic patients having mMRC dyspnea Grade 3 and 4. In nonanemic patients, only 27% were of mMRC dyspnea Grade 3 and 4.
Multiple independent factors have been associated with increased risk of readmission in persons with COPD, and among them, anemia was one of the greatest risks; anemic patients had a 25% higher risk of readmission than nonanemic patients.\[[@ref21]\] We also found in our study that exacerbations leading to hospitalization were more in anemic patients in comparison to nonanemic in the previous year. A study on 117 AECOPD also showed that anemia (Hb \<13 g/dl) and previous exacerbations (3.5 exacerbations) were independent predictors of mortality after 1 year in patients hospitalized for AECOPD.\[[@ref22]\] Dyspnea and fatigue are important symptoms in COPD, and both symptoms have a negative impact on the quality of life of COPD patients. Acase series of five ventilator-dependent COPD patients with anemia showed that whole-blood transfusion and raising the hemoglobin levels to \>12 g/dl resulted in successfully weaning all the patients.\[[@ref23]\] All these studies show that anemia when present in COPD patients have a definitive impact on the quality of life of COPD patients. Therefore, correction of anemia in COPD patients may improve their clinical outcome. Anemia is increasingly recognized as an important comorbidity in COPD, affecting a relevant number of patients with high indices of mortality, increased health-care costs, and having a negative impact on quality of life. Previous studies have shown that anemia is associated with an increased risk of hospitalization and mortality in patients with COPD. Therefore, it is important to study both conditions together to improve patient management and survival. Hence, efforts should be made to prevent, diagnose, and treat anemia as early as possible in COPD patients as a means of improving their overall prognosis, reducing the increase in hospitalizations, reducing respective length of stay, and thereby improving their health-related quality of life in patients with COPD.
Conclusion {#sec1-5}
==========
COPD is a multicomponent disease which affects the physiological conditions and social life of patients. The prevalence of anemia in COPD patients was 31.6%, and we found that proportion of anemic patients were higher in Stage 3 and 4 and had more dyspnea. Anemia is present as comorbidity in COPD patients and is associated with poor quality of life and increased morbidity in the form of number of exacerbation and hospital admission. During severe COPD exacerbations, anemia can be a risk factor for mortality. The clinical impact of anemia is significant, so correction of anemia should be an important goal in the management of COPD.
Financial support and sponsorship {#sec2-3}
---------------------------------
Nil.
Conflicts of interest {#sec2-4}
---------------------
There are no conflicts of interest.
We are greatly thankful to the Department of Respiratory Medicine for carrying out the study and appreciate patients who participated in the study.
| {
"pile_set_name": "PubMed Central"
} |
Introduction
============
Menopause is the permanent cessation of menses, resulting from the reduced secretion of ovarian hormones namely estrogen and progesterone. This takes place as the finite store of ovarian follicles are depleted (Al-Safi et al., 2015). Thai women typically begin menopause at an age range of 48-52 years, with a median age of 51 years (Ongsupharn et al., 2018).
Cervical cancer screening is a simple tool that could be used to detect precancerous and cancerous cervical lesions. In many studies, it has been found that the most common cervical cytology is the atypical squamous cells of undetermined significance (ASC-US). This was shown in many studies in Thailand and other Southeast Asian countries (Laiwejpithaya et al., 2008; Tanabodee et al., 2015; Kingnate et al., 2016; Mu-Mu-Shwe et al., 2014; Hav et al., 2016). It is known that the cervical cytology of the ASC-US type is an atypical report that indicated either a low-risk or high grade cervical cytology abnormality. Therefore, treatment was still not a definite conclusion (Massad et al., 2013).
In 2012, the American Society for Colposcopy and Cervical Pathology (ASCCP) had revised the consensus guidelines for the management of abnormal cervical cancer screening tests and cancer precursors. According to ASCCP 2012, the acceptable management in women with ASC-US on cytology was either a repeat cytology at one year or an immediate human papillomavirus (HPV) testing (reflex test). Colposcopy was recommended in ASC-US cases with positive result of high risk HPV test (Massad et al., 2013).
Literatures showed that the prevalence of ASC-US cytology in menopausal women was 1.8 percent (Keating et al., 2001; Cakmak et al., 2014). Incidence of silent cervical intraepithelial neoplasia (CIN) 2/3 in menopausal women ranged from 1.8 to 6.1 percent (Tokmak et al., 2014; Goksedef et al., 2011).
The objective of this study was to evaluate the effect of obesity and BMI on remission, persistence and progression of the disease in pre and postmenopausal women with a cytology of ASC-US after initial management and a 2 year follow up. The prevalence of silent high grade cervical intraepithelial neoplasia (CIN 2/3 and cancer) in ASC-US cytology results in both groups of patients were studied.
Materials and Methods
=====================
This was a retrospective study, approved by the institutional review board, Bhumibol Adulyadej Hospital (BAH), Bangkok, Thailand (IRB: 18/62). Medical records stored in computerized systems were reviewed. Data from all patients with cervical cancer screening data in a 6 year period between January 2013 and October 2016 were enrolled. Inclusion criteria were women who had a cervical cytology result with ASC-US cytology. Exclusion criteria included pregnancy, hysterectomy and known histories of pre-invasive or invasive cervical lesions or other gynecologic cancers. Papanicolaou smears were performed using conventional (CPP) or liquid based cervical cytology (LBP) and evaluated according to the 2001 Bethesda. Subjects were then divided into pre and postmenopausal women groups. Demographic characteristics consisted of age, menstrual status, parity, screening proposal, history of sexually transmitted diseases (STD) including human immunodeficiency virus (HIV), number of sexual partners, smoking, alcohol consumption and education. Management of ASC-US cytology was followed by the previous ASCCP guideline by either a repeated cervical cytology testing in 6 months or a colposcopic directed biopsy (immediate colposcopy or positive high risk HPV test). Outcomes were classified as remission, persistent disease, completed and loss to follow-up after the initial of ASC-US management at a two year follow up appointment. Remission was defined as normal cervical cytology results at follow-up within two years. Persistent disease was defined as abnormal cervical cytology and colposcopic directed biopsy results at follow-up within two years. Completed and loss to follow-up were defined as a completion of cervical cytology or an absence of follow-up within two years.
The data was analyzed by using the SPSS statistical software version 18 (IBM, Armonk, NY, USA) for analysis. Descriptive statistics were used to analyze patient demographic data. Continuous variables were presented as mean and standard deviation (SD). Categorized data were expressed as number and percentage. Pearson Chi-square and Fisher's exact test were used in data analysis when appropriate. The p-value of less than or equal to 0.05 was considered to be statistical significance. Multivariable analysis of patient risk factors was calculated using binary logistic regression analysis and described as an odds ratio.
Results
=======
This study was conducted in the main hospital of the Royal Thai Air Force. In this study, around ninety percent of the premenopausal women were air force officers and employees while fifty percent of postmenopausal women were housewives. Participants were people who worked at and lived in the surrounding area of the hospital.
A total of 506 patients with ASC-US cytology were enrolled. Subjects underwent either immediate colposcopy or follow up cervical cytology after thorough counseling, at an incidence of 412 and 50 cases, respectively ([Figure 1](#F1){ref-type="fig"}).
The mean age of women in premenopausal versus postmenopausal women groups were 37.4 and 61.1 years, respectively. The participants in postmenopausal women had higher mean age of coitarche than premenopausal women with statistically significance (19.8 vs 17.9, p\< 0.001). Ninety-seven percent of cases in the postmenopausal women group were still sexually active.
Demographic data of subjects with ASC-US reports are presented in [Table 1](#T1){ref-type="table"}. There were 372 and 134 cases in pre and postmenopausal women group, respectively.
Half of the postmenopausal group was classified as unsatisfactory colposcopy (Type 3). It was more than those of the premenopausal women group with statistical significance as shown in [Table 2](#T2){ref-type="table"}. The histopathological results showed the prevalence of less than or equal to CIN 1 and CIN 2/3 in ASC-US cytology at 90.1/92.8 and 9.9/7.2 percent in pre and postmenopausal women groups, respectively. Women with conclusive results for CIN 2/3 underwent Cone biopsy either by either Loop Electrosurgical Excision Procedure (LEEP) or Cold Knife Conization (CKC) performed in those who had cervical biopsies of CIN 2/3. Neither micro invasive or invasive cervical cancer was found in both groups.
At two year follow-up periods, in immediate colposcopy arm, 233 patients (56.6%) had completed follow-up. Among non-obese cases, women in postmenopausal group had remission rate lower than premenopausal group with statically significant (91 vs 99 percent). Among obese women, remission rate of pre and postmenopausal women were equal. Among patients who chose follow up cervical cytology, remission rate of pre and postmenopausal women of the obese and non-obese categories were comparable. No progressive disease was shown in both groups including both management groups as presented in [Table 3](#T3){ref-type="table"}.
The multivariate analysis of correlation of the remission disease and clinical factors (including menopausal status, body mass index (BMI), sexual activity, parity, smoking and hormone replacement therapy) was performed as shown in [Table 4](#T4){ref-type="table"}. There was no significant correlation between clinical factors and the progression or regression of disease.
{#F1}
######
Baseline Characteristics of Women Participating in This Study
Characteristics Pre Post *P-value*
------------------------- ------------ ------------ ----------------
Age (years) \* 37.4 + 9.4 61.1 + 7.5 \< 0.001\*\*\*
BMI (kg/m^2^)\* 22.7 + 4.1 25.5 + 4.6 \< 0.001\*\*\*
Coitarche (years) \* 17.9 + 7.7 19.8 + 4.6 0.001\*\*\*
Sexual intercourse\*\* 322 (86.6) 130 (97.0) \<0.001\*\*\*
Monogamy\*\* 340 (91.4) 119 (88.8) 0.375
Parity\*\* \<0.001\*\*\*
0-1 258 (69.4) 52 (38.8)
≥ 2 114 (30.6) 82 (61.2)
Screening proposal\*\* 0.027
Check up 306 (82.3) 121 (90.3)
Others 66(17.7) 13 (9.7)
Non smoking\*\* 368 (98.9) 132 (98.5) 0.658
Alcohol consumption\*\* 6 (1.6) 2 (1.5) 1
HIV positive\*\* 4 (1.1) 0 (0) 0.577
Education\*\* 0.001\*\*\*
Below bachelor 108 (29.0) 60 (44.8)
Bachelor or higher 264 (71.0) 74 (55.2)
LBP type\*\* 104 (28.0) 24 (17.9) 0.022\*\*\*
Specimen adequacy\*\* 190 (51.1) 56 (41.8) 0.065
BMI level \< 0.001\*\*\*
Normal 222 (59.7) 46 (34.3)
Overweight 61 (16.4) 18 (13.4)
Obese 89 (23.9) 70 (52.3)
Medical disease \< 0.001\*\*\*
No 352 (94.6) 81 (60.4)
Yes 20 (5.4) 53 (39.6)
\*: mean + SD (standard deviation), \*\*: n (%), \*\*\*: The result is significant at p \<0.05, Pre: premenopausal women, Post: postmenopausal women, HIV: human immunodeficiency virus, LBP: liquid based cervical cytology, BMI: body mass index, Normal:
######
Results of Pre- and Postmenopausal Women who had Cytology of ASC-US
---------------------------------------------------------------------------
Results Premenopause\ Postmenopause\ *p-value*
(n=314) (n=98)
----------------------------- ---------------- ---------------- -----------
Colposcopic findings\*\* \< 0.001\*\*\*
Type 1/2 232 (73.9) 46 (46.9)
Type 3 82 (26.1) 52 (53.1)
Histopathologic results\*\* 0.4
≤ CIN 1 283 (90.1) 91 (92.8)
CIN 2/3 31 (9.9) 7 (7.2)
---------------------------------------------------------------------------
\*\*: n (%), \*\*\*: The result is significant at p\<0.05, ASC-US: atypical squamous cells of undetermined significance, ≤ CIN 1: cervical intraepithelial neoplasia grade 1 or less, CIN 2/3: cervical intraepithelial neoplasia grade 2 and 3.
######
Results of Pre- and Postmenopausal Women who had Completed a 2-Year F/U
Characteristics Non-obese *p-value* Obese *p-value*
----------------- ------------ ----------- ------------- ----------- ---------- -------
IC Arm 0.010\*\*\* 0.369
Remission 115 (99.1) 32 (91.4) 47 (95.9) 33 (100)
Persisted 1 (0.9) 3 (8.6) 2 (4.1) 0 (0)
F/U Arm 0.505 0.472
Remission 15 (100) 11 (100) 11 (100) 9 (90)
Persisted 0 (0) 0 (0) 0 (0) 1 (10)
Pre: premenopausal women, Post: postmenopausal women, IC: immediate colposcopy, F/U: follow up, Remission: remission of disease, Persisted: persistent of disease, Non-obese: BMI 18.5-29.9 kg/m^2^, Obese, BMI \> 30 kg/m^2^; \*\*\*: The result is significant at p\<.05
######
Correlation of the Remission of Disease and Patient Factors by Multivariate Analysis
Factors Remission Persisted OR (95% CI) *P-value*
--------------------------- ----------- ----------- ------------------- -----------
Menopausal status
Premenopause 188 3 3.12 (0.57-17.09) 0.189
Postmenopause 75 3 0.31(0.05-1.64) 0.169
BMI
Normal 128 3 1.01 (0.18-5.61) 0.983
Overweight 40 0 3.03 0.998
Obese 95 3 0.92 (0.17-4.92) 0.922
Sexual intercourse
No 31 0 2.24 0.998
Yes 232 6 0 0.998
Number of sexual partners
0-1 237 6 0 0.998
≥ 2 26 0 5.36 0.998
Parity
0-1 171 3 1.97 (0.34-11.26) 0.445
≥ 2 92 3 0.51 (0.08-2.89) 0.445
Smoking
No 260 6 2.78 1
Yes 3 0 0.23 1
HRT used
No 259 6 0 0.999
Yes 4 0 1.16 0.999
BMI, body mass index; Normal, BMI 18.5-22.9 kg/m^2^; Overweight, BMI 23-29.9 kg/m^2^; Obese, BMI \> 30 kg/m^2^; HRT, hormone replacement therapy; Remission, remission of disease; Persisted, persistent of disease; OR, odds ratio; CI, confidence interval; \*\*, The result is significant at p \< 0.05
Discussion
==========
ASC-US is the most common cytological abnormality in cervical cytology. It carried the lowest risk of CIN 3 or cancer because one third to two third was not HPV associated (Katki et al., 2013).
Age of coitarche in the premenopausal group was lower than that of the postmenopausal group (17 and 19 years, respectively). These findings might be the result of different occupations and life styles among both groups in this study. Half of the postmenopausal group were housewives while eighty percent of premenopausal group were employees in either government or the private sector.
Most of both groups were monogamous, non-smokers and non-consumers of alcohol (in line with demographic trends of Thai women). However, the effect of the second hand cigarette smoke could not be evaluated.
In the current study, mean BMI of the postmenopausal women was higher than that of the premenopausal women (25.5 vs 22.7 kg/m^2^). Abnormal BMI level indicated overweight and obese status. Two third of postmenopausal women in the present study had abnormal BMI. Fifty percent of this group had BMI in obese ranges. The findings were comparable to the results from Chen KL et al that reported increasing prevalence of overweight and obesity in 70 percent of postmenopausal women (Chen et al., 2018).
From the present study, postmenopausal women were more likely to be obese and have underlying diseases including diabetes mellitus, hypertension and cardiovascular disease, similar to the findings reported in previous literature (Krychman et al., 2015). According to 2003-2012 National Health and Nutrition Examination Survey (NHANES) data, metabolic syndrome presented in about 34 percent of the population and the rates of metabolic syndrome were increasing among postmenopausal women (Chen et al., 2018). Among postmenopausal women in this study, there was a forty percent incidence of metabolic disease. The rate of findings was comparable to the previous literature.
The rate of inadequate colposcopy in postmenopausal women in this investigation was significantly higher than that of the premenopausal women (53.1 and 26.1 %, respectively). This finding was a consequence of physiologic migration of the squamocolumnar junction (SCJ) to the cervical opening. Local estrogen therapy to the cervicovaginal region improved the adequacy of colposcopic examination (Goksedef et al., 2011; Richards et al., 2015).
In the current study, the prevalence of silent CIN2/3 in ASC-US cytology were 9.9 and 7.2 percent in pre and postmenopausal groups, respectively. Both groups had high prevalence of high grade CIN. Thus we recommended that all women with ASC-US cytology results should undergo colposcopic directed biopsy. Our recommendation differed from the American Society for Colposcopy and Cervical Pathology (ASCCP) guidelines that recommended colposcopy in persistent ASC-US cytology or concomitant with positive high risk HPV (Massad et al., 2013).
Among non-obese cases in both pre and postmenopausal groups, persistence rates of abnormal cervical cytology of the postmenopausal group was higher than that of the premenopausal group (3/35 and 1/116, respectively). Using binary logistic regression analysis, no statistically significant links were found between remission rates of abnormal cervical cytology and patient factors (including menopausal status, BMI, sexual activity, parity, smoking and HRT). Among postmenopausal women with obese BMIs, there was a slightly lower rate of disease regression, but this was found to not be statistically significant. This was explained by the alteration of host immunity (Liu et al., 2013). Age was an important risk factor for a persistent HPV infection due to decreasing immune function in the elderly. HPV was hard to be clear after its acquisition (Sui et al., 2018).
Obesity was a result of disruption of energy balance that led to weight gain and metabolic disturbances that causes tissue stress and dysfunction. Metabolic disturbances lead to immune activation in tissue such as adipose tissue, liver, pancreas and vasculature. These effects may increase the risk for other infectious and chronic diseases (Andersen et al., 2016). However menopause was often accompanied by redistribution of adipose tissue from the periphery to the abdomen, hip and thigh. Central obesity has been associated with a pro-inflammatory state in postmenopausal women. Adiposity had been linked to increased susceptibility to viral pathogens, such as herpes simplex virus-1 and -2 and adenovirus-36 (Liu et al., 2013). In particular, Baker et al (Baker et al., 2011) showed increased levels of adipokines in older women with persistent HPV infection.
The limitations of this study were its nature as a retrospective study. Demographic data and further follow-ups of cervical cytology were not completed.
There was no data on blood hormonal, adipokine and immunological factor levels in our retrospective data. It would cause of remission or persistence of disease in women with BMI status in obese level between pre and postmenopausal women with no difference. Therefore, this topic should be of concern and merits further study.
In conclusion, in postmenopausal women, obesity might have a link on the remission rate. Cervical cytology follow-up should be prescribed in this population. However, obesity was not correlated with progression or regression of CIN, and further studies should be done to clarify the relationship. The prevalence of silent high grade CIN in ASC-US cases in pre and postmenopausal women were higher than the data of ASCCP guideline. Based on our patient population, we recommended immediate colposcopy in all women who had cervical cytology report of ASC-US without HPV testing result. The suggestion might be applied to the referral centers with high incidence of silent high grade CIN.
| {
"pile_set_name": "PubMed Central"
} |
Exposure to a large number of potentially toxic compounds renders the liver, in particular, susceptible to injury. Xenobiotic-induced hepatocellular damage is a much studied and clinically relevant phenomenon. The toxicity of most xenobiotics is associated with their biotransformation or metabolism that is frequently coupled with irregularities in cellular oxidant/antioxidant balance.^[@bib1]^ As impairment of hepatocellular functions may be a prelude to hepatic failure, an understanding of the mechanisms by which toxic compounds inflict irreversible damage to cells is crucial for alleviation of liver injury.
Nrf2 (nuclear factor erythroid 2-related factor 2), a redox sensitive transcription factor, coordinates the controlled expression of antioxidant genes so as to reinstate redox homeostasis in an event of oxidative prevalence. Nonetheless, studies indicate that a number of pathological circumstances involving oxidative imbalances are correlated with perturbed activity or stability of Nrf2 itself.^[@bib2],\ [@bib3],\ [@bib4],\ [@bib5],\ [@bib6]^ Considerable research has been conducted to delineate the mechanisms responsible for regulating Nrf2 responses within the cell. The phosphatidylinositol 3′-kinase (PI3K)/Akt pathway forms an important component of cell survival,^[@bib7]^ which is activated in response to oxidative stress.^[@bib8]^ Previous studies have reported functional interactions between the PI3K/Akt pathway and Nrf2 activation,^[@bib9],\ [@bib10],\ [@bib11],\ [@bib12]^ but no direct relationship has yet been confirmed. Fyn kinase, a member of Src family of tyrosine kinases, is believed to influence stability and nuclear accumulation of Nrf2 by promoting its degradation.^[@bib13]^ Fyn kinase has been demonstrated to localize to the nucleus during stress and shares an inverse relationship with nuclear Nrf2 density.^[@bib14],\ [@bib15]^
The PHLPP isoforms (PH domain and leucine-rich repeat protein phosphatases) regulate phosphorylation of Akt kinases at Ser473 residue. Recently, PHLPP knockout has been reported to protect the brain against ischemic insult.^[@bib16]^ Studies indicate that the cellular levels of PHLPP2, the isoform that specifically targets Akt1,^[@bib17]^ affect proliferative potential of tumorigenic cells,^[@bib18],\ [@bib19],\ [@bib20]^ highlighting its role in promoting apoptotic events. However, the importance of PHLPP2 in regulation of Akt activity and its subsequent effect on cell survival mechanisms has not yet been addressed in relation to Nrf2 responses in oxidatively stressed cells.
Targeting the regulatory processes that act in the development of oxidative stress-induced toxicity may be a proper strategy to restore a disturbed balance. Thus, a clear understanding of the molecular events that influence the cell\'s potential to thwart damaging effects of reactive oxygen species (ROS) is indispensible. In this study, we have explored the novel relationship shared between PHLPP2, Akt1 and Fyn kinase in determining Nrf2 stability and hence its activity during oxidative stress-evoked hepatocellular toxicity. The study suggests that inhibition of PHLPP2 could be a promising approach to reinstate cell defense mechanism, which has been compromised owing to dysregulation in Nrf2 signaling during excessive ROS generation.
Results
=======
Inhibition of Akt diminishes cellular antioxidant defense triggering oxidative stress-mediated death of hepatocytes
-------------------------------------------------------------------------------------------------------------------
To validate the involvement of Akt in regulating redox balance through Nrf2 signaling mechanism, we measured the activities of some of the key antioxidant and detoxification enzymes that are downstream Nrf2 targets in primary rat hepatocytes exposed to varying concentrations of LY294002, a PI3K inhibitor. As expected, inhibition of Akt activation led to significant decline in the activities of key antioxidant (thioredoxin reductase (TrxRed), glutathione reductase (GR), glutathione peroxidase (GPx)) and detoxification enzymes (glutathione *S*-transferase (GST) and NAD(P)H quinone oxidoreductase 1 (NQO1)) in a concentration-dependent manner ([Figure 1a](#fig1){ref-type="fig"}). LY294002 treatment also diminished the nuclear as well as cytoplasmic levels of reduced glutathione (GSH), indicating the important role of Akt in maintaining balanced redox environment in sub-cellular compartments ([Figure 1b](#fig1){ref-type="fig"}). Henceforth, the disturbed defensive mechanism of the cell warranted enhanced oxidative stress as indicated by estimation of DCF fluorescence and Ethidium/DHE ratio ([Figures 1c and d](#fig1){ref-type="fig"}; [Supplementary Figure S1a](#sup1){ref-type="supplementary-material"}). Increased cellular oxidative burden not only perturbed the mitochondrial membrane potential ([Figure 1e](#fig1){ref-type="fig"}) but also compromised the viability of hepatocytes, which was observed to decrease significantly with increasing LY294002 concentration ([Supplementary Figure S1b](#sup1){ref-type="supplementary-material"}). In accordance with the role of Akt in cell survival pathway, the data suggest that inhibition of Akt undermines antioxidant defense mechanisms, which thus exacerbate free radical accumulation and related functional insufficiencies culminating in cell toxicity.
Inhibition of Fyn kinase elevates cellular antioxidant defense against intra-cellular free-radical generation
-------------------------------------------------------------------------------------------------------------
Fyn kinase has been reported to destabilize Nrf2,^[@bib13]^ which suggests its inverse relevance to cellular oxidative burden. The activities of redox enzymes (TrxRed, GR, GPx, GST and NQO1) were observed to increase with increasing concentration of (4-amino-5-(methylphenyl)-7-(t-butyl)pyrazolo-(3,4-d)pyrimidine (PP1), inhibitor of Fyn kinase activation, reaching their maximum at 15 *μ*M concentration as compared with control ([Figure 2a](#fig2){ref-type="fig"}). Thereafter, a consistent drop in the activity of the studied enzymes was observed, reaching values comparable to that of control at 25 *μ*M PP1 concentration. Moreover, PP1 exposure also enhanced the nuclear as well as overall levels of GSH ([Figure 2b](#fig2){ref-type="fig"}). Accordingly, treatment with PP1 exhibited considerable decrease in endogenous ROS and superoxide levels as compared with control ([Figures 2c and d](#fig2){ref-type="fig"}; [Supplementary Figure S2a](#sup1){ref-type="supplementary-material"}). Inhibiting Fyn kinase did not indicate any significant loss of cell viability ([Supplementary Figure S2b](#sup1){ref-type="supplementary-material"}) and also did not appear to modulate mitochondrial membrane polarity ([Figure 2e](#fig2){ref-type="fig"}). Taken together, the data confirms that intervention in Fyn kinase activation augments cell\'s resistance to free radical generation and/or accumulation.
Akt regulates Nrf2 signaling mechanism by repressing Fyn kinase phosphorylation and its nuclear localization
------------------------------------------------------------------------------------------------------------
As the antioxidant and detoxification enzymes are under the direct regulation of Nrf2, we further evaluated the level of Nrf2 protein expressed in the hepatocytes upon inactivation of Akt and Fyn kinase enzymes. Significant decline in Nrf2 and its target proteins\' levels first appeared around 30 *μ*M concentration of LY294002 as indicated in [Figure 3a](#fig3){ref-type="fig"} (left panel). On the other hand, a consistent increase in the levels of Nrf2, NQO1 and HO1 (heme oxygenase 1) protein levels could be observed as the concentration of PP1 (Fyn kinase inhibitor) increased. In addition, inhibiting Akt pathway decreased the levels of Nrf2 within the nuclear compartment while Fyn kinase inhibition followed the opposite trend ([Figure 3b](#fig3){ref-type="fig"}). The immunofluorescent detection of sub-cellular Nrf2 localization further supported the above conclusion that Fyn kinase inactivation promotes nuclear retention of Nrf2 ([Figure 3e](#fig3){ref-type="fig"}). Not only this, intervention of the Akt and Fyn kinase pathway also affected the functional activity of Nrf2. LY294002 at 30 *μ*M caused upto 40% decrease in antioxidant redox element (ARE)-binding affinity of Nrf2, while Fyn kinase inhibition with 15 *μ*M PP1 exhibited upto 38% increase in its functional capacity ([Figure 3c](#fig3){ref-type="fig"}). The results clearly indicate the opposing roles of Akt and Fyn kinase in terms of mechanistic regulation imposed on Nrf2 signaling.
We observed that at 30 *μ*M LY294002 concentration reduction in phosphorylated levels of both Akt Thr308 and Ser473 residues was accompanied by a significant decrease in GSK3*β*(Ser9) phosphorylation ([Figure 3a](#fig3){ref-type="fig"}). GSK3*β* is the immediate downstream effector molecule of Akt that is deactivated when phosphorylated by Akt at its Ser9 residue. Studies have also proven that GSK3*β* is the upstream activator of Fyn kinase phosphorylation.^[@bib21]^ Thus, accordingly, we also observed a significant increase in phosphorylation status of Fyn kinase upon LY294002 exposure. Phosphorylation of Fyn kinase has been found to be associated with its nuclear localization. Western blotting analysis ([Figure 3b](#fig3){ref-type="fig"}) together with immunofluorescent imaging ([Figure 3f](#fig3){ref-type="fig"}) confirmed that inactivation of Akt pathway evokes activation and nuclear localization of Fyn kinase. Inhibition of Fyn kinase using PP1 did not result in any change in phosphorylation status of both Akt Ser473 residue and GSK3*β*(Ser9), suggesting that Fyn kinase functions downstream of Akt pathway. However, significant reduction in phosphorylation of Akt at Thr308 residue could be observed ([Figure 3a](#fig3){ref-type="fig"}). This might be indicative of feedback regulation imposed by Fyn kinase on Thr308 site of Akt. Further, in accordance with the role of Fyn kinase in promoting Nrf2 degradation, treatment with PP1 subsided the levels of ubiquitinated Nrf2 as compared with control ([Figure 3d](#fig3){ref-type="fig"}, right panel), but LY294002 exposure promoted Nrf2 ubiquitination ([Figure 3d](#fig3){ref-type="fig"}, left panel). The data collectively demonstrate that PI3K/Akt pathway imposes its regulation on Nrf2 signaling by checking Fyn kinase activation.
*Tert*-butyl hydroperoxide (tbhp)-induced increase in PHLPP2 (PH-domain and leucine-rich repeat protein phosphatase 2) causes site-specific Akt deactivation resulting in impairment of Nrf2 signaling
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Nrf2 is a key cellular transcription factor regulating the expression of proteins involved in the maintenance of redox homeostasis. Reports suggest that toxicity arising due to oxidative damage is a result of impairment of redox balance. In order to ascertain whether an event of oxidative toxicity implies any dysregulation in Nrf2 signaling due to intervention of pathway relating Akt and Fyn kinase, we treated primary hepatocytes with tBHP, a commonly used oxidative stress inducer. We observed that a concentration of 250 *μ*M tBHP was sufficient to elicit significant cell death of hepatocytes ([Supplementary Figure S3](#sup1){ref-type="supplementary-material"}), which corresponded to increased free radical generation and loss of mitochondrial membrane potential (data not shown). Western blotting analysis demonstrated that tBHP exposure significantly decreased total Nrf2 levels at 120 and 180 min (0.7- and 0.8-fold, respectively), but significant reduction in its target proteins HO1 and NQO1 became apparent as early as 60 min ([Figure 4a](#fig4){ref-type="fig"}). This could be explained by the reason that nuclear retention of Nrf2 started to diminish from 60 min time period of tBHP exposure ([Figure 4d](#fig4){ref-type="fig"}). Western blotting analysis of key components of Akt signaling pathway revealed that tBHP stress did not affect the total Akt1 levels as well as phosphorylation of Akt at Thr308 residue (except for the initial 1.5-fold increase at 15-min exposure period, [Figures 4a and b](#fig4){ref-type="fig"}); however, consistent time-dependent reduction with respect to phosphorylation of Akt at Ser473 residue could be observed ([Figures 4a and b](#fig4){ref-type="fig"}). Accordingly, PDK1, which is responsible for phosphorylating Akt at its Thr308 residue, showed no change with respect to its phosphorylation. Further, while phosphorylation of PTEN(Ser380) decreased (which implies enhanced PTEN activity), a remarkable decline in GSK3*β* phosphorylation was detected. As earlier reports and our data here ([Figure 3](#fig3){ref-type="fig"}) confirm that Fyn kinase is associated with suppression of Nrf2 activity, we assessed the levels of phosphorylated Fyn kinase as well as its nuclear density. tBHP treatment led to significant enhancement in levels of phosphorylated Fyn kinase ([Figures 4a and b](#fig4){ref-type="fig"}), with concomitant increase in its nuclear localization in tBHP-treated hepatocytes ([Figures 4d and e](#fig4){ref-type="fig"}). This suggests that oxidative toxicity of tBHP is due to subdued Nrf2 functionality brought about by lifting the repression imposed by Akt(Ser473)-GSK3*β*(Ser9) wing on Fyn kinase activation.
As our data revealed specific downmodulation of Akt phosphorylation at Ser473 residue, we hypothesized that either decreased phosphorylation or enhanced selective dephosphorylation of Ser473 residue may be involved. To testify the two possibilities, we checked the levels of phosphorylated mTORC2 (kinase) as well as PHLPP2 (phosphatase), both of which specifically target Ser473 residue of Akt1.^[@bib17],\ [@bib22]^ Though no significant alteration in levels of phospho-mTORC2 were revealed except for a 0.7-fold dip at 180 min, we did observe pronounced increase in levels of PHLPP2 as early as 30 min time period of incubation with 250 *μ*M tBHP ([Figure 4c](#fig4){ref-type="fig"} and [Supplementary Figure S4a](#sup1){ref-type="supplementary-material"}). This highlights the involvement of phosphatase PHLPP2 in downregulating Akt activity. Apart from this, we also observed increased nuclear levels of PHLPP2 that corresponded to decreasing phosphorylation of Akt Ser473 within the nucleus ([Figure 4d](#fig4){ref-type="fig"}). Immunofluorescent detection also supported the increased sub-cellular levels of PHLPP2 ([Supplementary Figure S4b](#sup1){ref-type="supplementary-material"}). The findings clearly demonstrate the involvement of PHLPP2 in influencing Nrf2-regulated survival pathway through site-selective Akt deactivation.
PHLPP2 induction imposes negative regulation on Nrf2 signaling during tBHP-induced hepatocellular toxicity
----------------------------------------------------------------------------------------------------------
We reasoned that if PHLPP2 is responsible for blocking Nrf2 activation via modulation of Akt(Ser473)--GSK3*β*--Fyn kinase axis, then knockdown of PHLPP2 should restore this activation. A 15-nM concentration of small interfering RNA (siRNA) targeting PHLPP2, which produced nearly 60--70% knockdown after 48 h incubation ([Supplementary Figure S5b](#sup1){ref-type="supplementary-material"}), was used for silencing experiments. Stimulations with 250 *μ*M tBHP (for 90 min) followed 48 h after transfection. Our data reveal that PHLPP2-silencing not only significantly enhanced phosphorylation of Akt(Ser473) and GSK3*β*(Ser9) in tBHP-treated hepatocytes but also suppressed Fyn kinase activation, as indicated by decline in Fyn kinase phosphorylation ([Figures 5a and b](#fig5){ref-type="fig"}). Knockdown of PHLPP2 also led to a substantial increase in nuclear phospho-Akt(Ser473), which was accompanied by decreased nuclear retention of Fyn kinase ([Figure 5c](#fig5){ref-type="fig"}). Consequently, blocking PHLPP2 expression restored Nrf2 activation as indicated by enhanced NQO1, HO1 levels ([Figure 5a](#fig5){ref-type="fig"}), increased nuclear retention of Nrf2 ([Figures 5c and d](#fig5){ref-type="fig"}), increased Nrf2 stability ([Figure 5e](#fig5){ref-type="fig"}) and Nrf2-ARE-binding affinity ([Figure 5f](#fig5){ref-type="fig"}) as compared with tBHP-treated normal hepatocytes. In all, the data confirm that PHLPP2 imposes negative regulation on Nrf2 survival mechanism through suppression of Akt-induced Fyn kinase deactivation.
PHLPP2 knockdown checks tBHP-induced oxidative stress
-----------------------------------------------------
As we speculated that during tBHP exposure hindered Nrf2 responses result in oxidative overload leading to hepatocellular death, PHLPP2 knockdown should therefore prevent tBHP-mediated free radical generation through potentiation of Nrf2 signaling. Conforming to the positive outcome of PHLPP2-silencing on Nrf2 activation, we observed a significant reduction in ROS and superoxide generation ([Figure 6a](#fig6){ref-type="fig"}) as well as mitochondrial depolarization ([Figure 6b](#fig6){ref-type="fig"}) induced due to tBHP exposure. In addition, considerable enhancement in sub-cellular GSH levels could also be observed ([Figure 6c](#fig6){ref-type="fig"}) by blocking PHLPP2 expression. The data collectively manifest plausible relationship between PHLPP2 and Nrf2-regulated redox homeostasis ([Figure 7](#fig7){ref-type="fig"}) and the ensuing cell survival/death mechanism.
Discussion
==========
The normal balance between activated oxygen generation and cellular antioxidative processes, which are predominantly under Nrf2 regulation, has been observed to be disturbed under many pathological circumstances.^[@bib2],\ [@bib3],\ [@bib4],\ [@bib5],\ [@bib6]^ Though considerable literature exists highlighting the mechanistic aspects of Nrf2 activation in response to oxidative stress, any explanation regarding Nrf2 insufficiencies observed during certain physiological and pathological conditions is still lacking. Here, we have shown that oxidative stress, which initially activates pro-survival antioxidant defenses, may be aggravated owing to signaling cues that suppress Nrf2-mediated transcriptional induction. Hence, the ultimate result following an oxidative insult, that is, survival or cell death, may depend upon the prevailing stress levels and the subsequent signaling arrays activated by them. Our data indicate that: (a) pathway regulating Nrf2 stability involves Akt--Fyn kinase crosstalk; (b) hepatocellular toxicity arising from tBHP-mediated oxidative stress is associated with Nrf2-suppression due to increased Fyn kinase activation; (c) tBHP-induced oxidative stress selectively downmodulates Akt activation at Ser473 residue; and (d) the selective site-specific deactivation of Akt, and hence Nrf2 stability, is controlled by PHLPP2. Thus, the study highlights a novel aspect in Nrf2 regulation with respect to PHLPP2 induction.
The PI3K/Akt pathway is implicated in a number of biological responses, such as apoptosis, cell growth, differentiation, calcium signaling and insulin signaling. Our study using LY294002, a PI3K inhibitor, reveals that the role of Akt in cell survival is linked to the promotion of Nrf2/ARE transcriptional regulation, which is brought to effect by checking GSK3*β*-mediated Fyn kinase activation. Fyn kinase, which is said to have key role in promoting Nrf2 degradation,^[@bib13],\ [@bib15],\ [@bib23]^ has been implicated in oxidative stress-associated functional insufficiencies^[@bib24],\ [@bib25]^ as well as apoptosis.^[@bib26],\ [@bib27],\ [@bib28]^ We observed that Fyn kinase inhibition using PP1 suppressed the endogenous ROS levels along with enhancement in stability and functional capacity of Nrf2. Not only this but also the dysregulation in Nrf2 signaling observed during hepatocytes death following tBHP treatment was seen as a consequence of Fyn kinase activation resulting from modulation of Akt pathway.
An important finding of this study is the selective downmodulation of Akt phosphorylation at Ser473 residue during tBHP-induced hepatocyte death. Contrary to the common belief, necessitating phosphorylation of both Thr308 and Ser473 for Akt activation, increasing numbers of studies now indicate that selective phosphorylation of Akt, at either of its two residues, may activate Akt and generate distinct physiological response depending upon the specificity for its downstream substrates.^[@bib29]^ There are instances where distinct stimulants have been found to result in site-specific modulation of Akt phosphorylation.^[@bib30],\ [@bib31]^ Studies show that alteration in Akt activity through downmodulation of phosphorylation at Ser473 (independent of phosphorylation at Thr308 residue) renders the cell vulnerable to apoptosis.^[@bib32],\ [@bib33]^ Study by Kuo *et* *al*.^[@bib34]^ has demonstrated that selective dephosphorylation of Thr308, without an alteration in Akt(Ser473) phosphorylation, inhibited cell proliferation, but the effect was attributed to growth arrest and not increased cell death. Hence, while selective activation of Thr308 is observed to have a key role in promoting cell survival or proliferation,^[@bib29],\ [@bib34],\ [@bib35]^ a reduction in Ser473 phosphorylation seems to be necessary for establishment of apoptotic response. Moreover, GSK3*β*, a downstream target of Akt which is said to be involved in oxidative stress-mediated cell death,^[@bib36]^ is observed to be uniquely dependent on Akt phosphorylation at Ser473.^[@bib33],\ [@bib37]^ Although we detected a decline in phospho-Akt(Ser473) levels, no change was observed with respect to phosphorylation at Thr308 during tBHP stress ([Figure 4a](#fig4){ref-type="fig"}). As in the experiment using Fyn kinase inhibitor PP1, we particularly observed decrease in phosphorylation at Thr308 of Akt ([Figure 3a](#fig3){ref-type="fig"}), we may assume that the enhanced Fyn kinase levels during oxidant attack may exert some feedback regulation at Thr308 site of Akt. Interestingly, a previous report illustrates that phosphorylation of Akt at Ser473 somehow inhibits that at Thr308 residue.^[@bib38]^
The pattern of site-specific Akt activation conjures the involvement of upstream regulators of this kinase. PHLPP is a relatively recent addition to the milieu of signaling moieties regulating cellular homeostasis. Akt is one of the prime substrate of PHLPP, which, being a phosphatase, selectively dephosphorylates Akt at its Ser473 residue.^[@bib17]^ Owing to its decreased expression in several types of tumors, PHLPP has been defined as a tumor-suppressor gene. In a study by Liu *et al.*,^[@bib39]^ PHLPP2 was not only observed to be significantly lost in colorectal cancer patient specimens but an overexpression of the same also inhibited tumorigenic potential of colon cancer cells in nude mice. Incidence of polymorphism in the phosphatase domain of PHLPP2 in breast cancer cells^[@bib40]^ and repressed expression of PHLPP2 in Bcr-Abl expressing CML cells (chronic myelogenous leukemia cells)^[@bib41]^ as well as in non-small cell lung cancers^[@bib42]^ further highlight the relevance of PHLPP2 in modulating proliferative potential of the cell. In other words, the index of PHLPP2 expression within the cell may also be a measure of its susceptibility to death. A recent study reported that Atorvastatin (a drug used for the treatment of cardiovascular diseases) induced apoptosis in cardiac myxomas, a common primary tumor of the heart, by specifically enhancing phosphatase activity of PHLPP2 isoform of PHLPP.^[@bib19]^ Similar other reports^[@bib20],\ [@bib43]^ point toward pro-apoptotic functions of PHLPP2.
The role of PHLPP in regulating Akt activity in the liver is not much explored. PHLPP2 is not observed to be sufficiently expressed in un-induced rat liver;^[@bib44]^ however, DNA microarray analysis of liver tissue from lanthanum nitrate-treated rats revealed 23.4-fold induction of PHLPP2 transcript.^[@bib45]^ This is indicative of a possible role played by PHLPP2 in the progression of compromised liver pathophysiology. We are well aware that oxidative stress is a cardinal player in the inception, exacerbation or progression of many diseases, may they be of proliferative or of degenerative nature. Liver is a major site of xenobiotic transformation and, as a consequence, is often challenged with hepatocellular injury arising due to oxidant/antioxidant imbalances. We have already discussed the crucial role of PI3K/Akt pathway in containing ROS-mediated cellular damage via promotion of Nrf2-driven antioxidant defense. Given the ability of PHLPP2 to suppress Akt signaling, with this study we have also unveiled a previously unexamined role of PHLPP2 in regulating Nrf2 responses in the liver cells. We observed pronounced increase in both total and nuclear PHLPP2 levels together with significant decline in Akt(Ser473) phosphorylation upon tBHP exposure. PHLPP2-silencing not only restored normal Nrf2 signaling but also prevented mitochondrial depolarization and ROS generation in tBHP-exposed hepatocytes. Thus, our data confer convincing evidence that PHLPP2 may have a key role in determining the fate of cell by favoring apoptosis through selective suppression of Nrf2-regulated cellular defenses.
Increased basal activation of Nrf2, as has been observed in genetic analyses of many human tumors, allows the cancer cell to avoid the adverse effects of high levels of ROS and hence evade apoptosis.^[@bib46]^ In view of the loss of PHLPP2 function in many cancers, our investigation, which unveils a salient aspect of survival network describing PHLPP2--Akt--GSK3*β*--Fyn kinase--Nrf2 signaling axis, may explain the reason behind unhindered Nrf2 activation during cancer progression.
In summary, we show that Nrf2 insufficiency arising during oxidant attack may, at least in part, be a result of perturbed upstream signaling pathway regulating Nrf2 stability. We propose a mechanism by which PHLPP2 specifically dephosphorylates Akt at Ser473 residue, thereby lifting the regulation imposed by it on GSK3*β*, which in turn activates Fyn kinase that promotes Nrf2 degradation ([Figure 7](#fig7){ref-type="fig"}). The study identifies for the first time the critical function of PHLPP2 in regulating redox-sensitive Nrf2 signaling pathway, which could serve as a new target for developing strategies to manage pathological conditions exacerbated owing to oxidative stress. However, further insight into the multitude of signaling avenues in terms of mechanistic regulation imposed by PHLPP2 on cell survival need to be explored in order to minimize off-target effects.
Materials and Methods
=====================
Materials and reagents
----------------------
All reagents are listed in [Supplementary Materials and Methods](#sup1){ref-type="supplementary-material"}.
Primary rat hepatocytes isolation, culture and treatment
--------------------------------------------------------
Hepatocytes were isolated from 6- to 8-week-old Wistar rat through portal vein collagenase perfusion of liver as per the method of Seglen.^[@bib47]^ For attachment to collagen-coated surface, cells were cultured for 4 h in William\'s medium E supplemented with 50 nmol/l dexamethasone and 5% fetal bovine serum (FBS) in addition to 2 mmol/l glutamine and 1 × anti-mycotic and anti-biotic solution. Thereafter, the cells were cultured in the same medium but without dexamethasone and FBS. Hepatocytes were treated with LY294002 (10--50 *μ*M concentration range) to inhibit PI3K and PP1(5--25 *μ*M concentration range) to inhibit Fyn kinase activity, both for a period of 30 min. For inducing oxidative stress in hepatocytes, exposure to 250 *μ*M of standard oxidant tBHP was given for time periods ranging from 15 min to 3 h.
ROS detection
-------------
To measure intracellular ROS, cells were stained with 10 *μ*M DCFH-DA (2′,7′-dichlorofluorescein diacetate) for 30 min before treatment. Fluorescence-activated cell sorting (FACS) was performed using a FACSCalibur flow cytometer (BD Biosciences, San Jose, CA, USA). For fluorescent microscopic detection of ROS, hepatocytes were stained with 10 *μ*M DCFH-DA and 5 *μ*M DHE. Hoechst 33258 was used to stain nuclei and observed under Leica DMLB Fluorescence Microscope (Wetzlar, Germany).
JC-1 (5,5\',6,6\'-tetrachloro-1,1\',3,3\'-tetraethylbenzimidazolyl carbocyanine iodide) staining
------------------------------------------------------------------------------------------------
In order to assess alterations in mitochondrial membrane potential, hepatocytes were incubated with JC-1 at a final concentration of 2 *μ*M at 37 °C through the 30 min time period of the experiment involving Akt and Fyn kinase inhibition. Nuclei were counterstained with Hoechst 33258 and observed under Leica DMLB Fluorescence Microscope.
Intracellular GSH estimation and localization
---------------------------------------------
For GSH estimation, CellTracker Green CMFDA dye (5-chloromethylfluorescein diacetate; Invitrogen, Life Technologies Corp., Carlsbad, CA, USA) was used. Cells were incubated with 2 *μ*M CMFDA at 37 °C for 30 min, following which fluorescent microscopic detection was performed under Leica DMLB Fluorescence Microscope. Nuclei were stained using Hoechst 33258.
Enzyme activities
-----------------
For methods followed to measure enzyme activities, refer Supplementary Materials and Methods.
Sub-cellular fractionation
--------------------------
Nuclear and cytoplasmic fractions were obtained following the NE-PER nuclear and cytoplasmic extraction protocol (Pierce Biotechnology, Rockford, IL, USA). Concentration of protein was determined using the bicinchoninic acid method.^[@bib48]^
Immunoprecipitation and western blotting analysis
-------------------------------------------------
*In-vivo* ubiquitinated Nrf2 levels were estimated as described earlier.^[@bib49]^ To minimize non-specific precipitation, cell lysates were incubated with normal goat serum for 1 h before precipitation with Nrf2 antibody (developed in goat). For western blot transfer, BioTrace PVDF (polyvinylidene fluoride) membranes (Pall German Laboratory, MI, USA, USA) were used and visualized using the Immobilon Western Chemiluminescent Horseradish Peroxidase Substrate Kit (Millipore Corporation, Billerica, MA, USA) on ImageQuant LAS 500 detection system (GE Healthcare, Upsala, Sweden). All western blotting images are representative of three independent experiments. The bands from western blotting were quantified by the ImageJ 1.47v software (National Institutes of Health, Bethesda, MD, USA).
Immunofluorescence
------------------
Following treatment, cells were washed with cold 0.01 M PBS (pH 7.2) and fixed in 4% paraformaldehyde for 10 min. The cells were then washed with 0.05% glycine in PBS and then permeabilized with 1% Triton X-100 (v/v in PBS) for 15 min followed by overnight incubation with primary antibody at a dilution of 1:200 in PBS. The cells were rinsed thrice with PBS for 5 min each. This was followed by 1-h incubation in fluorescence-tagged secondary antibody at 1 : 500 dilution. Nuclei were counter-stained with Hoechst 33258 (5 *μ*g/ml) for 5 min. Microscopic detection was performed under Leica DMLB Fluorescence Microscope.
siRNA transfection
------------------
Hepatocytes were cultured overnight in collagen pre-coated 24-well plate and transfected with 10 nM Silencer Cy3-labeled negative control siRNA using Lipofectamine RNAiMAX Transfection Reagent (Invitrogen, Life Technologies Corp.) in agreement with the manufacturer\'s instruction to monitor transfection efficiency of isolated hepatocytes. Examination of hepatocytes 24 h after transfection with Cy3-labeled siRNAs revealed a near 70--80% transfection efficiency ([Supplementary Figure S5a](#sup1){ref-type="supplementary-material"}). Silencer Select predesigned siRNA against PHLPP2 was obtained from Ambion (Austin, TX, USA). Transfection was performed 24 h after plating using Lipofectamine RNAiMAX Transfection Reagent in agreement with the manufacturer\'s instruction in Opti-MEM Reduced Serum Medium (Invitrogen, Life Technologies Corp.). After 4-h incubation, the transfection medium was changed with serum supplemented William\'s medium E, and the hepatocytes were further cultured for an additional 48 h, following which stimulations by tBHP treatment (250 *μ*M) were given for additional 90 min where required. Western blotting analysis was performed to confirm knockdown of PHLPP2 compared with negative control siRNA.
TransAM Nrf2-ARE binding assay
------------------------------
The Nrf2 DNA binding activity was measured using ELISA-based assay (TransAM kits, Active Motif, Carlsbad, CA, USA) following the manufacturer\'s instructions.
Statistical analysis
--------------------
All computational calculations of quantitative data were performed using the Microsoft Excel program (Microsoft Office 2007, Microsoft Corporation (India) Pvt. Ltd., Gurgaon, India). Each experiment was repeated at least three times. The quantitative variables represented in histograms are expressed as mean±S.E. Statistical comparisons between means of different groups were conducted by one-way analysis of variance followed by Tukey\'s *post hoc* test using the SPSS 14.0 statistical package (SPSS Inc., Chicago, IL, USA). Differences were considered statistically significant when *P*\<0.05.
Financial support was received from CSIR Network Project-BSC 0111. FRSS acknowledge the Council of Scientific and Industrial Research (CSIR), India for award of fellowship. We are grateful to the Institutional Manuscript Review Committee of CSIR-IITR for allotting communication number 3197 for the manuscript.
ARE
: antioxidant redox element
CMFDA
: 5-chloromethylfluorescein diacetate
DCFH-DA
: 2′,7′-dichlorofluorescein diacetate
FBS
: fetal bovine serum
GPx
: glutathione peroxidase
GR
: glutathione reductase
GSH
: reduced glutathione
GST
: glutathione *S*-transferase
HO1
: heme oxygenase 1
JC-1
: 5,5\',6,6\'-tetrachloro-1,1\',3,3\'-tetraethylbenzimidazolyl carbocyanine iodide
NQO1
: NAD(P)H quinone oxidoreductase 1
Nrf2
: nuclear factor erythroid 2-related factor 2
PHLPP2
: PH domain and leucine-rich repeat protein phosphatase 2
PP1
: 4-amino-5-(methylphenyl)-7-(t-butyl)pyrazolo-(3,4-d)pyrimidine
ROS
: reactive oxygen species
siRNA
: small interfering RNA
tBHP
: *tert*-butyl hydroperoxide
TrxRed
: thioredoxin reductase
[Supplementary Information](#sup1){ref-type="supplementary-material"} accompanies this paper on Cell Death and Disease website (http://www.nature.com/cddis)
Edited by A Stephanou
The authors declare no conflict of interest.
Supplementary Material {#sup1}
======================
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
######
Click here for additional data file.
{#fig1}
{#fig2}
{#fig3}
{#fig4}
{#fig5}
{#fig6}
{#fig7}
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#S1}
============
The functional consequences of mossy fiber sprouting in the epileptic tissue have been interpreted as either contributing to ([@B1]--[@B3]) or counteracting epileptic seizures ([@B4]--[@B6]). The controversial nature of this synaptic reorganization has generated at least three hypotheses: the mossy fiber sprouting hypothesis ([@B7]) that holds sprouting as a major factor underlying hippocampal hyperexcitability; the dormant basket cell hypothesis ([@B6], [@B8]) that emphasizes the importance of changes in inhibitory activity; and the irritable mossy cell hypothesis ([@B9]) that focuses on the hyperexcitability of mossy cells. The apparent inconsistency of the data derives partly from the limited perspective of most anatomical studies, which focus primarily on a single-synaptic input (e.g., sprouted mossy fibers as opposed to data from field neuronal recordings).
Dentate granule cells receive predominant input from the medial septum, entorhinal cortex, and hilus \[for review, see Ref. ([@B10])\]. In this context, while most studies in this area have concentrated on mossy fiber sprouting, it is clear that synaptic changes from sources other than granule cells might also contribute to the development of epileptogenesis. These, however, remain largely unknown ([@B11]) due the lack of staining methods to specifically characterize such terminals (in contrast to the Timm's staining technique used to study mossy fiber terminals). One approach to address this issue is to study synaptic terminals that innervate the dentate molecular layer at the ultrastructural level.
We have previously demonstrated that induction of *status epilepticus (SE)* in the presence of cycloheximide (CHX) is associated with marked reduction of hilar mossy cell loss in mice and rats ([@B13]). Indeed, there is little or no sprouting of mossy fibers (granule cell axons) in animals subject to *SE* under the presence of CHX ([@B14]--[@B16]). Here, we investigated whether mossy fiber sprouting, and additional synaptic reorganization of the dentate molecular layer, would be associated with an increase in the number of asymmetric synaptic profiles, or simply with synaptic replacement.
Materials and Methods {#S2}
=====================
Animals and Protocol for Pilocarpine Induction of Chronic Seizures {#S2-1}
------------------------------------------------------------------
All experimental protocols were approved by the Animal Care and Use Ethics Committee of UNIFESP and were performed in accordance with the Society for Neuroscience guidelines for animal research. Male Wistar rats (*n* = 30, 200--250 g) were kept on standard light/dark cycle (12/12 h) with lights on at 7:00 a.m. Animals had free access to rat chow pellets (Nuvilab) and tap water. Seizures were induced by injections of pilocarpine hydrochloride (Pilo, 320 mg/kg, i.p. Merck). Scopolamine methyl bromide (1 mg/kg, i.p., Sigma) was administered 30 min prior to Pilo to reduce its peripheral effects. In addition, one group of animals also received CHX (1 mg/kg, i.p., Sigma) 30 min prior to Pilo (CHX + Pilo). All animals developing *SE* received thionembutal (25 mg/kg, i.p., Cristalia, Brazil) 90 min later, as previously described ([@B17]). Four months after *SE*, animals were sacrificed and had their brains processed, as described below.
Tissue Processing {#S2-2}
-----------------
Two different protocols were performed 120 days after *SE* induction (*Experiment 1* and *Experiment 2*); for each protocol, we analyzed five animals per group. For *Experiment 1*, five animals from each group (Pilo, CHX + Pilo, and control) were transcardially perfused with 500 mL sulfide solution (4% glutaraldehyde, 0.1% Na~2~S, 0.002% CaCl~2~, in 0.12M Millonig's phosphate buffer, pH7.3) ([@B18]). One hour later, brains were processed according to the Timm's staining method for the ultrastructural evaluation of silver grains in synaptic terminals of the hippocampal dentate molecular layer. After removal from the skull, brains were placed in the same fixative solution for 24 h at 4°C. Coronal sections (100 μm thick) were cut on a vibratome (Vibratome Series 1000 Sectioning System) and transferred to a fresh developing solution (60 mL gum Arabic 50%, 10 mL of a 2M citrate buffer; 15 mL hydroquinone 5.67%, and 15 mL silver lactate 0.73%) for 1.5 h in the dark, under constant agitation, and subsequently processed for electron microscopy (EM).
Another set of five animals for each group (Pilo, CHX + Pilo, and control) was used to evaluate synaptic profiles in the dentate gyrus molecular layer (*Experiment 2*). Under deep anesthesia with thionembutal (50 mg/kg, i.p.), rats were transcardially perfused with 500 mL of modified Karnovsky solution at 4°C (2.5% glutaraldehyde, 2% formaldehyde in 0.1M phosphate buffer, pH 7.4), for at least 1 h. One hour later, brains were removed from the skull, immersed in the same fixative solution for at ([@B19]) least 24 h at 4°C, and subsequently processed for EM.
Electron Microscopy Procedures {#S2-3}
------------------------------
Tissue specimens were obtained from the right dorsal hippocampus (corresponding approximately to levels 28--31 of Swanson's rat brain atlas) ([@B20]). Samples remained overnight in a 0.1M pH 7.4 cacodylate buffer solution \[Na\[CH~3~\]^2^.AsO~2~.3H~2~O\], at 4°C. After rinsing, specimens were postfixed with 1% OsO~4~ sodium cacodylate buffer, washed in sodium cacodylate buffer, and kept overnight in uranyl acetate. The specimens were then dehydrated in a series of ethanol baths, placed in propylene oxide, transferred to pure Epon resin, and placed in vacuum for 4 h. The block polymerization took place at 60°C for at least 2 days. Seventy-nonometer-thick (silver interference color) and 90-nm-thick (gold interference color) sections were cut (Ultracut S/FC S, Reichert) for *experiments* 2 and 1, respectively. These sections were then stained with 2% uranyl acetate and lead citrate.
### Methodological Considerations {#S2-3-1}
The criteria used to identify ultrastructural synaptic profiles have been previously described ([@B21], [@B22]). Briefly, synaptic profiles were identified by cleft material between parallel membranes of a presynaptic element using at least two spherical vesicles and a postsynaptic element with a postsynaptic density (PSD). Active zones were distinguished from puncta adhaerentia by the lack of a pronounced presynaptic thickening and the usual presence of apposed presynaptic vesicles. We restricted our evaluation to asymmetric synaptic profiles (excitatory synapses) of the dentate molecular layer. The asymmetric synaptic profiles were classified as perforated and non-perforated based on shape of the PSD ([@B23]--[@B25]). These were categorized as: PSD1 (non-perforated type 1 synapses), synaptic profiles with a single-synaptic bouton associated to a continuous disk shape; PSD2 or PSD3 (perforated types 2 or 3 synapses), synaptic profiles with two or more PSDs, respectively. We have also evaluated the number of synaptic profiles located on dendritic spines versus those located on dendritic shafts. Dendritic spines were discerned from dendritic shafts by morphological features of spines. As an example, the dendritic shaft cytoplasm contains microtubules, mitochondria, and a multivesicular body, while the cytoplasm of the spine consists of stacks of smooth endoplasmic reticulum interdigitated by electron-dense plates ([@B26], [@B27]) (Figure [1](#F1){ref-type="fig"}).
![**Electron micrograph of the inner molecular layer showing asymmetric synaptic profiles and their localization in dendritic spines or shafts (A)**. **(B--D)** are higher magnification views of different synaptic contacts: PSD1 \[**(B)** -- non-perforated type 1, synaptic profiles with a single-synaptic bouton\]; PSD2 \[**(C)** -- perforated type 2, synaptic profiles with two postsynaptic densities\]; and PSD3 \[**(D)** -- perforated type 3, synaptic profiles with more than two postsynaptic densities\]. Note the spherical presynaptic vesicles and mitochondrion (\*) in the axon terminal (a) contacting a dendritic spine (e) and a dendritic shaft (s) with a mitochondria (^\#^). Scale bars, 1.25 μm.](fpsyt-06-00157-g001){#F1}
The number of non-perforated and perforated synapses is not shown in absolute values because a quantification of this order would require the analysis of serial sections. The basic assumption for counting synaptic profiles at various single sections is the probability that different planes of section would be equally distributed across groups.
The area of silver grain electron-dense deposits from the Timm's reaction was evaluated by quantitative densitometric stereological analysis, through a cross test system, in approximately the same area of that used for counting the asymmetric synaptic profiles (approximately 1,713 μm^2^ per layer per animal). Although one could consider desirable to obtain quantitative estimates of the relative number of silver grain-containing sprouted fibers, technical limitations in applying the Timm's method at an ultrastructural level make such estimates potentially unreliable. Moreover, the size and density of silver grains is not linearly related to the concentration of zinc within a terminal. It is important to emphasize that data obtained with Timm's staining for EM did not allow a clear definition of synaptic membranes. In some of the sections, however, we could observe silver grain clusters around dendritic shafts (putative mossy fiber sprouting terminals) with a non-perforated asymmetric synaptic profile (Figure [1](#F1){ref-type="fig"}). This is in agreement with a previous study using Timm's staining in kainate-treated rats ([@B28]).
### Analysis of Synaptic Profiles {#S2-3-2}
In each experiment, six randomly selected photographs were taken from every examined tissue field \[in the inner molecular layer (IML) and outer molecular layer (OML)\]. Each 285 μm^2^ field was photographed at 5000× and amplified to 15,000×. Quantification of synaptic profiles was performed only in non-overlapping areas (free of large blood vessels, tissue tears, or folds), by counting all asymmetric synaptic profiles in an area of 1,712.88 μm^2^ per dentate molecular layer/animal. Synaptic profiles on the exclusion lines were not counted. The IML and OML boundaries were determined for each section. The rat dentate molecular layer has a 250 μm thickness on average ([@B10]). The innermost 50--70 μm are often considered the IML, whereas the outermost 150--200 μm comprise the OML. While this subdivision tends to ignore the intermediate molecular layer, it provides a well-defined distinction between the IML and OML with no chance of sampling overlapping fields. In addition, the characterization of the connectivity of the intermediate molecular layer has always been demonstrated as being similar to that of the OML. Bearing this in mind, we chose to describe our data as a fraction of the afferent connections, dividing the dentate gyrus molecular layer in inner and outer counterparts.
Comparative width analysis of the dentate molecular layer (from dentate granule cell layer to hippocampal fissure) using 100-μm-thick coronal sections revealed that this layer had a similar width in all experimental groups, indicating that there was no differential tissue shrinkage in controls, Pilo-, and CHX + Pilo-treated animals.
The estimation of the area of silver grains-impregnation and that of the density of asymmetric synaptic profiles were determined with a stereological test system method and an unbiased counting frame, respectively ([@B29]). The test system was applied as a mask over the final enlarged electron micrograph prints for estimating the area of silver grains. The space between the points of the test system was 11 mm, corresponding to a tissue area of 0.85 μm^2^ (0.92 μm × 0.92 μm).
Statistical Analysis {#S2-4}
--------------------
Results are presented as mean ± SEM. Comparisons between parameters were carried out by one-way analysis of variance (ANOVA) followed by Newman--Keuls *post hoc* test, using the Statistica 7 software. Significance was set at *P* \< 0.05.
Results {#S3}
=======
Spontaneous recurrent epileptic seizures were first observed at approximately 15--21 days after *SE* in all animals. However, given that we did not perform a complete 24/7 seizure assessment, it is possible that spontaneous seizures may have emerged earlier.
The comparative width analysis of the dentate molecular layer (from dentate granule cell layer to hippocampal fissure) using 100-μm-thick coronal sections revealed that this layer had a similar width in all experimental groups, indicating that there was no differential tissue shrinkage in control, Pilo-, and CHX + Pilo-treated animals.
CHX + Pilo Treatment Reduced Deposits of Silver Grains in the IML {#S3-1}
-----------------------------------------------------------------
At the EM level, the Timm's sulfide silver method did not reveal any silver grains outside the hilus in the dentate gyrus of control animals (Figures [2](#F2){ref-type="fig"}A and [3](#F3){ref-type="fig"}A). This finding is in agreement with previous light ([@B13], [@B14], [@B30]--[@B32]) and EM studies ([@B28], [@B33], [@B34]). By contrast, all Pilo-treated animals had silver grains in the IML but not in the OML (Figures [2](#F2){ref-type="fig"}C,D, respectively). A detailed examination of the density of silver grains indicated that the higher staining score mainly stemmed from those IML areas closer to the granule cell layer (up to 10 μm apart from the granule cells, Figure [3](#F3){ref-type="fig"}B), where it was about twice as high as the one recorded in more distant IML regions (50--70 μm from the granule cell layer, Figure [3](#F3){ref-type="fig"}B). In the IML of Pilo group, silver grains were mostly deposited on terminal axons where asymmetric synaptic profiles could be identified (Figure [3](#F3){ref-type="fig"}B). In CHX + Pilo-treated animals, the density of silver grains occupying the IML was only 7% of that observed in Pilo-treated animals and similar to that found in the control group (Figures [2](#F2){ref-type="fig"}A and [2](#F2){ref-type="fig"}E respectively). In the hilus, the intensity of silver grains labeling did not differ among the three groups of animals. No animal, irrespective of group, showed silver grains in the OML (Figures [2](#F2){ref-type="fig"}B,D,F).
{#F2}
{#F3}
Reduced Density of Asymmetric Synaptic Profiles in Epileptic Rats {#S3-2}
-----------------------------------------------------------------
The analysis of asymmetric synaptic profiles in the dentate molecular layer (IML + OML) of the control group revealed a density of 23.64 ± 0.59/100 μm^2^. This was significantly reduced by 8 and 20% in Pilo and CHX + Pilo groups, respectively. It is noteworthy that when considering only the IML, Pilo-treated animals had a synaptic density similar to that of controls (−4.5%). By contrast, animals in the CHX + Pilo group had significant loss of synaptic profiles when compared to control (−14%) and Pilo (--9.5%) groups. Figure [4](#F4){ref-type="fig"} summarizes results from Table [1](#T1){ref-type="table"}. In the OML, a significant loss in the density of asymmetric synaptic profiles was found in both Pilo (--11%) and CHX + Pilo (--26%) treated rats, as compared with controls. These data suggest that the influence of CHX to inhibit the growth and/or formation of new asymmetric synaptic contacts after Pilo treatment occurs particularly in the OML, as the loss of asymmetric synaptic profiles in this region was twice as high as that observed in the IML.
{#F4}
######
**Mean number of asymmetric synaptic profiles per 100 **μ**m^2^ in different layers of the dentate gyrus**.
Type of synapses Groups Molecular layer
------------------ ------------------ --------------------------- --------------
PSD 1 Control 21.43 ± 0.66 21.29 ± 0.98
Pilo 20.38 ± 0.79 19.52 ± 0.66
CHX + Pilo 18.54 ± 0.62\*\* 15.16 ± 0.86^\*\*\*,\#\#^
PSD 2 Control 1.67 ± 0.19 2.17 ± 0.21
Pilo 1.74 ± 0.17 1.32 ± 0.19\*\*
CHX + Pilo 1.40 ± 0.15 2.18 ± 0.26^\#^
PSD 3 Control 0.36 ± 0.07 0.37 ± 0.07
Pilo 0.29 ± 0.06 0.37 ± 0.09
CHX + Pilo 0.25 ± 0.04 0.35 ± 0.08
*Data expressed as mean ± SEM. ANOVA followed by Newman--Keuls*.
*\*\**P* \< 0.01, and \*\*\**P* \< 0.001 as compared to controls*.
*^\#^*P* \< 0.05 and ^\#\#^*P* \< 0.01 as compared to Pilo group*.
*Each group was comprised of five animals; for each animal, six slices were analyzed*.
Distribution of PSD1, PSD2, and PSD3 Asymmetric Synaptic Profiles in the Dentate Molecular Layer {#S3-3}
------------------------------------------------------------------------------------------------
In all groups (Pilo, Pilo + CHX, and controls), PSD1 was the most abundant contact type, followed by PSD2 and PSD3. The control group had a mean of 21.36 ± 0.58 PSD1; 1.92 ± 0.14 PSD2, and 0.37 ± 0.05 PSD3 synaptic profiles/100 μm^2^, respectively. This corresponded to 90, 8, and 2% of the asymmetric synaptic profiles found in the dentate molecular layer (Table [1](#T1){ref-type="table"}). A similar distribution of synaptic profiles was also seen in Pilo- and CHX + Pilo-treated animals, though with lower absolute values. Significant reductions in the number of PSD1 (21%, *P* \< 0.001) were recorded in the CHX + Pilo group as compared to control group. In the same group, we found that the densities of PSD1 were diminished in the IML (*P* \< 0.01, as compared to controls) and in the OML (*P* \< 0.001 and *P* \< 0.01, as compared to control and Pilo groups, respectively). Significant reductions in the number of PSD2 were found in the Pilo group (20%, *P* \< 0.05 compared to controls), particularly in the OML (*P* \< 0.01, as compared to both controls or CHX + Pilo-treated rats). By contrast, no differences were found in the PSD3 counts across groups (see Table [1](#T1){ref-type="table"}). In summary, Pilo-treated animals had a lower density of PSD2 profiles in the OML, whereas the CHX + Pilo group had less PSD1 in both the IML and OML as compared to control group.
Synaptic Reorganization Preferentially Involved Spine Synapses Rather than Reorganizations on Shaft Synapses {#S3-4}
------------------------------------------------------------------------------------------------------------
Irrespective of the treatment (control, Pilo-, or CHX + Pilo-treated animals), synapses in the dentate molecular layer were largely located on dendritic spines rather than shafts (Table [2](#T2){ref-type="table"}), as previously reported ([@B35], [@B36]). In control animals, 87, 7, and 1% of the synapses in the molecular layer were PSD1, PSD2, and PSD3, respectively. The remaining 5% of synapses were located on dendritic shafts. While a similar frequency was recorded in the Pilo-treated group, this was not the case for rats given CHX + Pilo, which had a relatively lower frequency of PSD1 and a higher number of PSD2 profiles as compared to the other groups (Table [2](#T2){ref-type="table"}).
######
**Frequency of dendritic spines and dendritic shafts (PSD1, 2, and 3) in the dentate molecular layer**.
Groups Type of Synapses IML OML Total of Asymmetric Synaptic Profiles (to each group)
------------ ------------------ ----------------- ------------------------ ------------------------------------------------------- ------------------- -------------------
Control PSD1 618 (44%) 26 (1.8 606 (43%) 34 (2.4%) 1284/1421 (90.4%)
Pilo 565 (43%) 48 (3.7%)\*\*\* 545 (42%) 42 (3.2%)\* 1200/1312 (91.5%)
CHX + Pilo 519 (46%) 38 (3.3%)\*\*\* 428 (38%)^\*\*,\#\#^ 27 (2.4%)^\#^ 1012/1138 (88.9%)
Control PSD2 47 (3.3%) 3 (0.2%) 58 (4.1%) 7 (0.5%) 115/1421 (8.1%)
Pilo 48 (3.7%) 5 (0.4%) 36 (2.7%)\*\* 4 (0.3%) 93/1312 (7.1%)
CHX + Pilo 39 (3.4%) 3 (0.3%) 62 (5.5%)^\*\*,\#\#\#^ 4 (0.4%) 108/1138 (9.5%)
Control PSD3 11 (0.8%) 0 (0%) 8 (0.6%) 3 (0.2%) 22/1421 (1.6%)
Pilo 7 (0.5%) 1 (0.1%)\* 10 (0.8%) 1 (0.1%) 19/1312 (1.5%)
CHX + Pilo 6 (0.5%) 2 (0.2%)\* 8 (0.7%) 2 (0.2%) 18/1138 (1.6%)
*Data expressed as frequency of synaptic profiles. Chi-square*.
*\**P* \< 0.05, \*\**P* \< 0.01, and \*\*\**P* \< 0.001 as compared to controls*.
*^\#^*P* \< 0.05, ^\#\#^*P* \< 0.01, and ^\#\#\#^*P* \< 0.001 as compared to Pilo group*.
*Each group was comprised of five animals; for each animal, six slices were analyzed*.
In the IML of control animals, 96% of the asymmetric synaptic profiles contacted dendritic spines, while corresponding values for Pilo and CHX + Pilo animals were 92 and 93%, respectively. In the OML, 94, 93, and 94% of the asymmetric synaptic profiles occurred on dendritic spines of control, Pilo, and CHX + Pilo animals, respectively (Table [2](#T2){ref-type="table"}). In the IML, the frequency of PSD1 and PSD3 asymmetric synaptic profiles on dendritic shafts of epileptic animals (Pilo and CHX--Pilo groups) was greater than controls (*P* \< 0.001 and *P* \< 0.05, respectively). In the OML, the frequency of PSD2 asymmetric synaptic profiles in Pilo group was significantly lower than in controls (*P* \< 0.01). In CHX + PILO group, while the frequency of PSD1 asymmetric synaptic profiles was significantly reduced (*P* \< 0.01), PDS2 synaptic profiles were significantly increased (*P* \< 0.01) as compared to controls.
Thus, while control animals lacked PSD3 profiles on dendritic shafts in the IML, these could be seen in both epileptic groups (Pilo and CHX + Pilo) (Table [2](#T2){ref-type="table"}). Moreover, for Pilo-treated animals, PSD1 profiles apposing dendritic shafts in the OML were more numerous (*P* \< 0.05) than in controls. By contrast, however, CHX--Pilo-treated animals had less PSD1 (*P* \< 0.05) in the OML than Pilo-treated animals (Table [1](#T1){ref-type="table"}).
Figure [5](#F5){ref-type="fig"} summarizes the significant results of synaptic profiles and dendritic location. In general, the most conspicuous result of our study was that the density of all types of asymmetric synaptic profiles in the epileptic groups was remarkably similar to that registered in controls. There were, however, a few noticeable differences. In Pilo animals, PSD1 contacts were distributed in both IML and OML dendritic shafts, whereas PSD3 contacts were only observed in the IML dendritic shafts. In CHX-treated animals, PSD1 and PSD3 contacts were distributed in IML dendritic shafts, whereas PSD2 contacts were only observed in OML dendritic spines. Finally, comparison of the Pilo and CHX--Pilo groups showed that the latter had a decrease in PSD1 (in both spines and shafts) and an increase in PSD2 (in spines only) in the OML.
{#F5}
Discussion {#S4}
==========
The main findings of our study are the following: (1) the density of silver grains in the IML of animals receiving CHX + Pilo was much reduced when compared to that recorded in rats given Pilo alone. (2) CHX + Pilo treatment led to a significant reduction in the density of asymmetric synaptic profiles in the IML and OML (14 and 26%, respectively), whereas animals treated with Pilo did not differ from controls (4.5% for IML and 11% for OML). (3) Both Pilo and CHX + Pilo had an altered distribution of asymmetric synaptic profiles types (e.g., PSD1, PSD2, PSD3) in the dentate molecular layer as compared to controls. The current estimate of 96% of asymmetric synaptic profiles apposing dendritic spines in the IML of control animals is in agreement with previous findings from Buckmaster and colleagues ([@B37]). Overall, these findings support and expand our previous observations, suggesting that CHX blocks *SE*-induced supragranular mossy fiber sprouting ([@B14], [@B15], [@B38]). Indeed, here we not only showed a dramatic reduction of putative Timm-stained mossy fiber terminals in the IML of CHX + Pilo-treated animals but also a reduction in the number of asymmetric synaptic profiles in the IML and OML, and a shift in the type of synaptic terminals present in the same area. The assembly of synaptic profiles with different synaptic efficacies in Pilo- and CHX + Pilo-treated animals may significantly affect information processing in the dentate gyrus.
Silver Grain Deposits and Asymmetric Synaptic Profiles in the Dentate Molecular Layer {#S4-1}
-------------------------------------------------------------------------------------
The *SE*-related hilar cell loss in the Pilo model is intense ([@B17], [@B39]) and has been considered to be critical for the development of subsequent mossy fiber sprouting given that hilar neurons represent 36% of all inputs to the dentate IML ([@B40]). Our current demonstration of a similar number of IML asymmetric synaptic profiles in both control and Pilo-treated animals further indicates that the synaptic reorganization observed in mossy fiber sprouting represents a tendency to replace lost synaptic contacts rather than the establishment of additional synaptic contacts. In the CHX + Pilo group, we did not find a correspondent loss of asymmetric synaptic profiles, despite the 93% reduction in Timm's IML labeling. Considering that CHX inhibited the mossy fiber sprouting from hilus, the similarity in the number of asymmetric synaptic profiles between control and CHX + Pilo animals could be an indication of synaptic plasticity from other sources, such as entorhinal cortex, given that this structure is a major source of afferent projections to the dentate gyrus ([@B41]). On the other hand, CHX may have protected the hilus from damage, given that the loss of hilar (as usually suggested) and entorhinal neurons are often a pre-requisite for mossy fibers to sprout in animal model of epilepsy, including Pilo ([@B39], [@B42], [@B43]). In fact, neuronal loss in the hilus ([@B44]) and entorhinal cortex (unpublished data) is less intense in CHX + Pilo than in Pilo animals. However, the present study is not sufficient to specifically elucidate whether the effects stem from CHX-related cell protection in the hilus, entorhinal cortex, or both. Our own previous findings in Pilo-treated animals provided evidence indicative of a protective role of CHX over hilar mossy cells in mice ([@B12]) and rats ([@B13]).
Synaptic Profile Morphology {#S4-2}
---------------------------
Ganeshina and collaborators (2004) demonstrated that perforated synapses (PSD2 and PSD3) have an invariably higher concentration of AMPA receptors than non-perforated synapses (\~660% more) and have 80% more immunoreactive NMDA receptors than non-perforated synapses (PSD1) in the hippocampal CA1 stratum radiatum. Moreover, \~35% of non-perforated synapses do not show any immunoreactivity for AMPA receptors ([@B45]), and may thus be considered "silent" synapses ([@B46], [@B47]). Therefore, perforated synapses may evoke synaptic responses with AMPA and NMDA receptors-mediated components of an exceptional magnitude and thereby contribute to an enhancement of synaptic transmission.
It is widely accepted that perforated (PSD2 and PSD3) synapses are much more efficient in impulse transmission than non-perforated (PSD1) synapses ([@B45], [@B48]--[@B55]). In the present work, the decreased number of PSD1 in the IML and OML of the CHX + Pilo group may suggest a reduced excitability. To the same extent, however, only Pilo-treated animals had a significant loss of the more effective PSD2 synaptic contacts (39% less compared to control group) in the OML (Table [1](#T1){ref-type="table"}). Thus, the reduction in the number of perforated synapses (PSD2) in Pilo animals might have a greater impact in reducing the excitability of the OML than the reduction of non-perforated synapses in CHX + Pilo animals.
Interpreting changes in synaptic morphology is not an easy task. The total number of PSD1 in our study was one order of magnitude greater than that of PSD2 and two- to fivefold greater than that of PSD3. Therefore, a small change of 10% in the frequency of PSD1 synapses could indicate a decrease or increase of 150 synapses. By contrast, changes in only 15 PSD2 synapses may result in a similar 10% increase or decrease in the number of synapses. Of course, given that we did not perform any functional evaluation in the current study, these are merely assumptions based on anatomical data.
The reduction of silver grain profiles seen at the ultrastructural level in the IML of CHX + Pilo-treated animals, as compared to Pilo-treated animals, represents important additional evidence of CHX ability to block mossy fiber sprouting, confirming our previous findings ([@B14], [@B16]) despite contrasting data ([@B56], [@B57]). Our results suggest that the dentate molecular layer synaptic reorganization that follows *SE* is a fine tuned process, which might be more suitable to restore dentate function than increasing excitation.
Author Contributions {#S5}
====================
SB manipulated animals and performed the EM procedures, discussed results, and wrote the first draft of the manuscript; FF and EF helped in the EM procedures, CH, BL, and OO discussed results and commented on discussion. LC and LM conceived the experimental design, participated in discussion of results and the final writing of the manuscript. All authors read and approved the final manuscript.
Conflict of Interest Statement {#S6}
==============================
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
This work was supported by CNPq and FAPESP/CInAPCe.
[^1]: Edited by: Alberto A. Rasia-Filho, Federal University of Health Sciences, Brazil
[^2]: Reviewed by: Robert S. Sloviter, Morehouse School of Medicine, USA; Ronald Sebastian Petralia, National Institutes of Health, USA
[^3]: Specialty section: This article was submitted to Systems Biology, a section of the journal Frontiers in Psychiatry
| {
"pile_set_name": "PubMed Central"
} |
Background {#Sec1}
==========
Diseases caused by hantaviruses differ enormously in severity and clinical course. Host- and virus-specific determinants are discussed as reasons for the broad range of clinical pictures \[[@CR1]--[@CR5]\]. The most obvious differences exist between the clinical picture of hantaviral cardiopulmonary syndrome (HCPS) and HFRS caused by New and Old World hantaviruses, respectively \[[@CR6]\]. Whereas HCPS manifests predominantly in the lung, HFRS is mostly characterized by renal failure. However, there is also a broad variety of symptoms in hantavirus disease caused by Old World hantaviruses. In contrast to Hantaan virus (HTNV), infection with PUUV is associated in most cases with a mild form of HFRS. Various genotypes exist within the species Dobrava-Belgrade virus and they cause diseases of different severity \[[@CR7]\]. In addition, hantavirus infection exhibits individual differences ranging from subclinical to fatal outcome.
The reasons for the variation of severity between virus species/genotypes and in individual patients are not yet known. Diverse determinants concerning virus- and patient-specific characteristics may play a role in the pathogenesis. Differences in the use of entry receptors, in the regulation of cytokine response and in viral replication were described to be associated with pathogenicity \[[@CR8]--[@CR11]\]. Studies with genetic reassortants in vitro and in animal models suggest molecular determinants to be responsible for virulence \[[@CR5], [@CR12]\]. However, the species-specific factors of hantaviruses that are responsible for pathogenicity and clinical picture are not identified so far. Interestingly, the pathogenicity of related viruses of DOBV genotypes differs enormously with case fatality rates (CFRs) between 0.3%-0.9% for DOBV genotype Kurkino and 14.5% for DOBV genotype Sochi \[[@CR13]\]. In addition to severe courses that are linked to specific virus species or genotypes, several serious cases were reported for infection with PUUV that usually causes a milder form of hantavirus disease \[[@CR14], [@CR15]\]. These infections often involve extrarenal manifestations \[[@CR16], [@CR17]\]. Severe cases caused by various hantavirus species are not well characterized with regard to their differences and similarities in symptoms, organ involvement, laboratory parameters, and clinical course.
The comparison of the clinical picture and course of severe disease caused by different hantavirus species may provide useful insights into their pathogenicity. Therefore, we analyzed the course of two severe hantavirus cases caused by infection with PUUV and DOBV-Sochi.
Case presentation {#Sec2}
=================
We report on hantavirus disease of a 25-year old German and a 20-year-old Russian woman infected with hantavirus PUUV and DOBV, respectively. Patients infected with hantavirus PUUV and DOBV were hospitalized in the Department of Nephrology, University of Heidelberg, Germany, in 2012 and 2014, respectively. The infection with DOBV occurred in the district of Krasnodar, South Russia, and the one with PUUV in Heidelberg, Germany. Infection was diagnosed by positive IgG and IgM hantaviral serology (*recom*Line HantaPlus assay, Mikrogen Diagnostik). Admission was on day four and on day six after onset of symptoms for the patient with PUUV and DOBV infection, respectively. To analyze the genotype of DOBV and to obtain partial nucleotide sequences of genomic segments, RT-PCR of serum and urine samples with subsequent sequencing was performed as described previously \[[@CR18], [@CR19]\]. Hantaviral RNA was detected in serum but not in urine. Partial nucleotide sequences of S, M, and L segments amplified from serum derived from the DOBV-Sochi patient were deposited in GenBank (accession numbers KU529946, KU529944 and KU529945). The sequences showed high similarity to DOBV-Sochi sequences obtained from Black Sea field mice (*Apodemus ponticus*) and from a fatal case of hantavirus disease reported in a 47-year-old woman in the district of Krasnodar in southern European Russia (Table [1](#Tab1){ref-type="table"}) \[[@CR13], [@CR20], [@CR21]\]. No pre-existing conditions, such as renal, pulmonary or cardiovascular disease, diabetes mellitus, hypertension or obesity, were found. Body weight and height were similar between the patients. Both patients were non-smokers. Symptoms during the early phase of both cases were very similar (Table [2](#Tab2){ref-type="table"}). Both cases showed the typical initial signs of hantavirus infection: Sudden onset of fever and flu-like symptoms. However, the course of DOBV infection resulted in a rapid decline of general condition and required admission to intensive care unit on day eight after onset of symptoms because of progressive respiratory problems with beginning hypoxia. The maximal and minimal levels of laboratory parameters differed between PUUV and DOBV-Sochi disease (Table [3](#Tab3){ref-type="table"}). It is to note that the absolute peak and nadir levels probably occurred before admission. Thereby, the impairment of laboratory parameter levels may be underestimated particularly with regard to DOBV-Sochi infection because the admission occurred two days later compared to the PUUV-infected patient. We observed elevated levels of lipase and P-amylase in the patient infected with DOBV-Sochi, indicating a possible hantavirus-related acute pancreatitis. The association of HFRS with acute pancreatitis was described for several cases of infection with Dobrava-Belgrade and Hantaan virus \[[@CR22]--[@CR24]\], but not for infections with Puumala virus \[[@CR25]\].Table 1Nucleotide (nt) and amino acid (aa) sequence identities (%) of partial DOBV-Sochi sequencesS segmentM segmentL segmentVirus isolate^a^ntaantaantaaSochi/hu98.698.497.498.998.6100.0Sochi/Ap98.898.997.498.9n.a.^b^n.a.10645/Ap98.698.9n.a.n.a.99.4100.0^a^Sequences were amplified from serum sample of our DOBV-Sochi (2014) infected patient and compared to published sequences of DOBV-Sochi strains isolated from human (hu) and *Apodemus ponticus* (Ap). Accession numbers for nucleotide and amino acid sequences: Sochi/hu (S, M, L segment): JF920150, JF920149, JF920148 and AES92929, AES92928, AES92927; Sochi/Ap (S, M segment): EU188449, EU188450 and ABY64966, ABY64967; 10645/Ap (S, L segment): KP878312, KP878309 and ALP44173, ALP44170^b^ *n.a* not available Table 2Characteristics and symptoms of two patients infected with PUUV and DOBV-SochiPUUVDOBV-SochiAge (years)2520Body weight change (kg)0.810BMI^a^ at discharge18.418.5Hospitalization (days)918Intensive care unit stay (days)02Maximum temperature (°C)39.640.0Headachen.d.^b^+Abdominal pain++Back-/side pain+-Myalgian.d.+Pain in the limbsn.d.+Nausea++Vomiting++Diarrhea-+Obstipation+-Night sweats\--Dyspnea-+Cough+-Pleural effusion-+Pulmonary congestion-+Pulmonary edema\--Infiltrates\--Vertigo+-Petechiae-+Edema-+Ascites+-Hyperkalemia++Dialysis (number)+ (1)+ (6)^a^ *BMI* body mass index, ^b^ *n.d* not determined Table 3Maximum and minimum levels of laboratory parameters of two patients with hantavirus infectionPUUVDOBV-SochiReference valuesSerum creatinine (mg/dL)10.899.340.1--1.3Urea (mg/dL)120231\<45Uric acid (mg/dL)6.511\<6Serum albumin (g/L)31.127.230--50CRP (mg/L)^a^61.2101.5\<5LDH (U/L)^b^422553\<248Lipase (U/L)24860\<51P-amylase (U/L)255208--53Bilirubin total (mg/dL)0.90.9\<1.0GPT (U/L)^c^5955\<35GOT (U/L)^d^6687\<35γ-GT (U/L)^e^7781\<40Alkaline phosphatase (U/L)9012955-105Hemoglobin (g/dL)118.212--15Hematocrit (L/L)0.310.250.36--0.47Platelets (10^9^/L)5153150--440Leukocytes (10^9^/L)7.5713.074--10^a^ *CRP* C-reactive protein, ^b^ *LDH* lactate dehydrogenase, ^c^ *GPT* glutamate pyruvate transaminase, ^d^ *GOT* glutamate oxalacetate transaminase, ^e^ *γ-GT* γ -glutamyl transferase
Urine analysis revealed proteinuria and the presence of erythrocytes and leukocytes in the urine with higher cell counts for erythrocytes (43 cells/μl versus 561 cells/μl) and leukocytes (6 cells/μl versus 34 cells/μl) in the patient with DOBV-Sochi. Apart from these characteristic urine pathologies, both patients developed uremia and oliguria. Glucosuria, pollakiuria, nycturia or dysuria were not observed. Lastly, they suffered from anuria in the further clinical course. As a consequence, renal replacement therapies were applied. The reasons for dialysis were uremia and severe fluid overload for DOBV-Sochi patient and uremia for PUUV patient. The patient infected with PUUV infection was dialyzed once on day seven after onset of symptoms, whereas the patient with DOBV-Sochi infection underwent dialysis six times between day nine and day 18 after onset of symptoms (Fig. [1](#Fig1){ref-type="fig"}). With exception of scleral bleeding and petechiae in the patient with DOBV-Sochi infection, no bleedings, such as epistaxis, hematoma, melena or hematochezia, were observed in the two patients. Symptoms of involvement of the respiratory tract were cough in the case of PUUV infection, pleural effusion and pulmonary congestion in the DOBV-Sochi patient (Fig. [2](#Fig2){ref-type="fig"}). The patient with DOBV-Sochi presented with tachycardia. No other cardiovascular or other extrarenal organ manifestations were observed. Patients did neither exhibit ophthalmological symptoms nor complications of the CNS.Fig. 1Course of laboratory parameters in patients infected with DOBV-Sochi and PUUV. *Black* and *gray* arrowheads indicate dialysis in PUUV and DOBV-Sochi patient, respectively. dpo, days post onset Fig. 2Chest x-ray of patients infected with PUUV (**a**, admission) and DOBV-Sochi (**b**, admission, bedside chest x-ray; **c**, after renal replacement therapy, 12 dpo)
The analysis of the course of laboratory parameters in DOBV-Sochi infection demonstrated a prolonged phase with elevated levels of leukocytes and serum creatinine and decreased levels of thrombocytes and serum albumin compared to infection with PUUV (Fig. [1](#Fig1){ref-type="fig"}). Several parameters, e.g. thrombocytopenia, have been described to be associated and predictive for severe courses of hantavirus disease \[[@CR26]--[@CR28]\]. A low platelet count (\<60 × 10^9^/L) indicates a subsequent acute renal failure with a rise in serum creatinine levels in Puumala virus infection \[[@CR27], [@CR29]\]. Corresponding to this definition for severe cases of PUUV infection, we observed platelet level of 51 × 10^9^/L for the patient with PUUV infection. For the DOBV-Sochi patient the level (53 × 10^9^/L) was also below 60 × 10^9^/L on admission.
The hospitalization of the patient with PUUV infection lasted nine days, whereas the patient with DOBV-Sochi infection was hospitalized for 18 days. The outcome of the hantavirus infection of both patients was complete recovery of renal function.
Our previous studies revealed the role of circulating endothelial progenitor cells (cEPCs) and cEPCs-mobilizing cytokines in the clinical course of patients infected with PUUV \[[@CR30]\]. As the normalization of laboratory parameters is paralleled to the mobilization of cEPCs, we analyzed the levels of cEPCs and of cEPC-mobilizing cytokines in the patients (Fig. [3](#Fig3){ref-type="fig"}). Quantification of levels of cEPCs by flow cytometry and of cytokines by Quantikine enzyme-linked immunosorbent assay (ELISA; R&D Systems) of patients and of 23 healthy persons was performed as described previously \[[@CR30]\].Fig. 3Course of cEPC numbers and plasma cytokine levels during hantavirus infection with DOBV-Sochi and PUUV. Horizontal dashed lines indicate the mean levels of 23 healthy control persons. EPO levels of some patient samples were below the limit of detection of the assay (\<2.5 mIU/ml, horizontal line)
Both patients demonstrated an increase in levels of cEPCs, Ang-2, VEGF, and SDF-1α compared to levels observed in healthy controls. Erythropoietin (EPO) levels were decreased during the disease indicating damage to the EPO-producing renal cells. All four samples of the PUUV patient and the samples of day 16 and 21 of the DOBV-Sochi patient were below detection limit of the EPO assay (\<2.5 mIU/ml). Besides the varying extent of cytokine level elevation, differences existed in the course of cEPC and cytokine level changes between both infections. A prolonged elevation of cEPC levels with a slow normalization in the patient with DOBV-Sochi infection was observed. The duration of the increase of Ang-2 and SDF-1α levels was also extended and much higher in DOBV-Sochi infection than in infection with PUUV. Furthermore, levels of VEGF in DOBV-Sochi infection increased later than in PUUV infection. The same delay was observed for the decrease of EPO levels. Taken together, both infections are characterized by mobilization of cEPCs and cytokine level elevation, but the temporal course and the extent of increase of cytokine levels differ enormously between infection with PUUV and DOBV-Sochi.
Conclusions {#Sec3}
===========
Among Old World hantaviruses, DOBV genotype Sochi is characterized by severe clinical course and a high CFR of 14.5% \[[@CR13]\]. In contrast, PUUV disease exhibits a low CFR of less than 1% \[[@CR31]\]. However, diseases caused by DOBV-Sochi and PUUV infection may also differ individually and cases of severe hantavirus disease due to PUUV infection were reported \[[@CR14], [@CR32], [@CR33]\]. We compared two severe hantavirus infections caused by these two different virus species, PUUV and DOBV-Sochi. The two cases did not differ significantly with regard to symptoms and organ involvement. However, the DOBV-Sochi infection presented with an enhanced impairment of laboratory parameters and a prolonged renal phase. The more severe clinical picture of infection with DOBV compared to PUUV corresponds to the observations made in other studies that compared infections with PUUV and different DOBV genotypes. Patients with DOBV infections were more often hypotensive, exhibited higher levels of serum creatinine, displayed more severe thrombocytopenia and required dialysis more often compared to patients infected with PUUV \[[@CR34]--[@CR36]\]. Our analysis of the clinical course of the two infections revealed further differences between the two infections. Impairment of laboratory parameters and upregulation of cytokines were prolonged in DOBV-Sochi infection. The mechanisms that are responsible for the more severe and protracted course of DOBV-Sochi infection are not completely understood. Previous studies have demonstrated a role of endothelial activation and repair in the clinical course of PUUV infection \[[@CR30], [@CR37]\]. The normalization of clinical parameters has been paralleled to the mobilization of endothelial progenitor cells. We also observed a mobilization of cEPCs in these two hantavirus infected patients. Similar to the observations for PUUV infection, levels of cEPCs and mobilizing cytokines were elevated in DOBV-Sochi infections. However, the increase of cytokines started later after onset of symptoms and was higher in infection with DOBV-Sochi than in PUUV infection.
Different cytokines were discussed to be responsible for hantavirus pathogenesis \[[@CR9], [@CR38]--[@CR41]\]. Several studies analyzed the role of VEGF in hantavirus disease \[[@CR30], [@CR42]--[@CR45]\]. The effect of VEGF seems to be temporally regulated. Early and localized upregulation of VEGF may be responsible for the clinical symptoms such as capillary leakage during hantavirus infection. In contrast, late and systemic elevation of VEGF may contribute to endothelial repair. As shown for VEGF, Ang-2 may also contribute to the pathogenesis of hantavirus disease. Altered ratios between angiopoietin-1 and −2 impair the barrier function of the endothelial monolayer during Dengue virus infection \[[@CR46]\]. The levels of Ang-2 were much higher in the patient with DOBV-Sochi infection than in the one with PUUV.
A reason for the altered clinical course and cytokine deregulation of DOBV-Sochi infection may be the enhanced replication of the virus. Viral load and antibody response influence the severity and the clinical course of hantavirus infection \[[@CR47]--[@CR49]\]. Unfortunately, we could neither measure the titer of hantaviral genomes nor of hantavirus-specific antibodies to explore a possible association between clinical course and viral titer or antibody response in our patients. It would be of interest to analyze if the differences observed in these two cases are specific for DOBV-Sochi infections compared to PUUV infections in a larger cohort of patients. The comparison of the clinical course of hantavirus genotypes with different pathogenicity may help to explore the underlying mechanisms. It seems that the infections were very similar in symptoms and induce the same pathways of cytokine signaling and endothelial damage and repair. However, research should further focus on the observed differences in the kinetics of cytokine mobilization. As observed for VEGF in hantavirus infection, cytokines may have detrimental as well as beneficial effects during the clinical course. Therefore, the knowledge about the role of cytokines in the clinical course and its temporospatial regulation in infections with different pathogenic hantavirus is crucial for the development of therapeutic strategies interfering with cytokine signaling.
The comparison of hantavirus disease caused by infection with PUUV and DOBV-Sochi revealed a more severe course for DOBV-Sochi. Initial symptoms and organ involvement did not vary noticeably. The two infections differed especially in the course and levels of cytokine upregulation. These results may indicate that temporal control and high level upregulation of certain cytokines contribute to the severity of the clinical course of hantavirus disease.
Ang-2
: angiopoietin-2
AV
: atrioventricular
BMI
: body mass index
cEPC
: circulating endothelial progenitor cells
CFRs
: case fatality rates
CRP
: C-reactive protein
DOBV
: Dobrava-Belgrade virus
dpo
: days post onset
EPO
: erythropoietin
GOT
: glutamate oxalacetate transaminase
GPT
: glutamate pyruvate transaminase
HCPS
: hantaviral cardiopulmonary syndrome
HFRS
: hemorrhagic fever with renal syndrome
HTNV
: Hantaan virus
LDH
: lactate dehydrogenase
PUUV
: Puumala virus
SDF-1α
: stromal derived factor-1α
VEGF
: vascular endothelial growth factor
γ-GT
: γ -glutamyl transferase
We thank Vanessa Bollinger for performing flow cytometry analysis and ELISAs, as well as Brita Auste for technical assistance with PCR and sequencing.
Funding {#FPar1}
=======
None.
Availability of data and materials {#FPar2}
==================================
Datasets generated and analyzed during this study were deposited in GenBank. Accession numbers KU529946, KU529944 and KU529945.
Authors' contributions {#FPar3}
======================
EK, CN, and AB prepared the manuscript. AB, CN, and JS were responsible for data acquisition. JH, PS, and BK were responsible for serological testing and sequencing. PTW and DHK substantially contributed to the interpretation of data and revised the initial manuscript. MZ and EK designed the study. All authors read and approved the final manuscript.
Competing interests {#FPar4}
===================
The authors declare that they have no competing interests.
Consent for publication {#FPar5}
=======================
Not applicable.
Ethics approval and consent to participate {#FPar6}
==========================================
This study was approved by the Ethics Committee of the University Hospital of Heidelberg, Germany, and it adhered to the Declaration of Helsinki. Written informed consent was obtained from the participants.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
*myo*-Inositol is an essential component in the biosynthesis of an array of derivatives ranging from simple inositol phosphates to complex membrane-associated products with important cellular functions. It can be isomerized and (or) methylated to form a variety of species-specific epimers and methyl ethers. A number of these, including Ins, have been noted to accumulate mostly in osmotically challenged plants and have since been recognized as osmoprotectant metabolites \[[@B1]\].
Ins is also central to the biosynthesis of a number of antinutritional components such as sucrose galactosides (e.g. RFO) and inositol polyphosphates such as PhA (also known as InsP~6~). The synthesis of PhA predominates in developing seeds, and constitutes the major storage form of seed phosphorus. Both Ins and its bound phosphates are released by hydrolysis upon germination.
The antinutritional properties of PhA reside in its strong binding affinity for positively charged species such as essential minerals (e.g. iron and zinc) and proteins, significantly lowering their bioavailability to humans and animals. As a consequence, the presence of high levels of PhA in canola seed hinders the full exploitation of the pure meal and underrates its potential as a major crop worldwide.
In contrast, PhA has been accredited as an effective antioxidant with antitumor properties and risk reduction of certain types of cancer \[[@B2]\]. Besides decreasing uncontrolled cellular proliferation, PhA is also thought to cause differentiation of malignant cells resulting in reversion to the normal phenotype \[[@B3]\]. Further, PhA has been shown to play a critical role in many cellular events such as signaling \[[@B4]\], apoptosis \[[@B5],[@B6]\], neuroprotection \[[@B7]\], as well as functioning as enzyme cofactor \[[@B8]\]. The antioxidant properties of PhA have also been shown to inhibit free radical formation and lower lipid peroxidation, making it a very efficient natural food and feed preservative \[[@B9]\]. Moreover, PhA has been shown recently to protect developing seeds against oxidative stress \[[@B10]\]. Because of the seemingly paradoxical and unique roles of PhA, and in view of the agronomic value placed on low phytate-containing seeds and (or) meals, total elimination of PhA was not our intended goal.
Phytic acid biosynthesis constitutes a uniquely complex process, consisting of a primary substrate, Ins, and a number of interjecting secondary Ins polyphosphate substrates from other sources and pathways \[[@B11],[@B12]\]. In the primary pathway, the *de novo* synthesis of Ins involves the oxidative cyclization of glucose 6-phosphate (G-6-P) to L*-myo*-inositol-1-phosphate (L-Ins-1-P) by the action of a single enzyme, L*-myo*-inositol-1-phosphate synthase (MIPS). It has been shown that by affecting the production of this enzyme directly or indirectly through mutagenesis in maize \[[@B13],[@B14]\] and soybean \[[@B15]\] or by genetic engineering methods in rice \[[@B16],[@B17]\], soybean \[[@B18]\] and canola \[[@B19]\] PhA accumulation can be reduced by 20--94.5% with a concomitant increase in inorganic phosphate (P~i~). Of the various transgenic approaches reported, *MIPS-*RNAi \[[@B18]\] and *MIPS-*cosuppression \[[@B19]\] transgenics yielded the lowest levels of PhA in the corresponding mature seeds (94.5 and 44%, respectively). However, the RNAi approach reportedly has resulted in hindered seed development \[[@B18]\]. Thus, in view of the fact that Ins is a key substrate in the biosynthesis of many essential cell components, we decided to assess the consequences of its metabolic diversion versus its complete elimination through *MIPS* down-regulation, and compare the effects associated with each approach. We reasoned that while metabolic diversion of Ins may not interfere with its biosynthesis, it could preferably limit its participation in PhA production.
The first MIPS product, L-Ins-1-P, appears to be in quasi equilibrium with free Ins due to the activity of a two-enzyme system, in which L*-myo-*inositol-1-phosphate monophosphatase converts L-Ins-1-P to free Ins, while *myo*-inositol-1-kinase regenerates it (Scheme [1](#C1){ref-type="fig"}). Depending on the rigidity of the requirements for free Ins *vs.* L-Ins-1-P in the developing seed, the relative abundance of each substrate is ultimately determined by the differences in the kinetics of the two opposing reactions. We hypothesized that continual exclusion of free Ins from this quasi equilibrium by metabolic diversion through methylation, could drive the quasi equilibrium in the direction of free Ins. This would subsequently limit the participation of both substrates (Ins and L-Ins-1-P) in PhA biosynthesis. Towards this end, we studied the effect of over-expressing the gene encoding IMT (EC 2.1.1.129) from *M. crystallinum*\[[@B20]\] on seed PhA accumulation in transgenic *B. napus* under two different seed-specific promoters, napin and phaseolin \[[@B21]\]. We also examined the effect of enhancing the translational context of *IMT* on its gene product accumulation and PhA reduction during seed development.
{#C1}
Although the *M. crystallinum IMT* (*McIMT*) has been used to study the osmoprotective properties of methylated cyclitols in transgenic plants \[[@B22]-[@B25]\], the current investigation is the first example of seed-specific McIMT-mediated metabolic diversion to reduce phytic acid biosynthesis in seed crops through *in vivo* methylation of Ins.
Results
=======
^3^H-*myo*-inositol metabolism in developing seeds of *B. napus*
----------------------------------------------------------------
*In vivo* labeling of developing *B. napus* seeds with ^3^H-*myo*-inositol, and subsequent fractionation of different cell components (acid-soluble, hexane-soluble, trifluoroacetic acid \[TFA\]-soluble) and cell debris revealed the relative incorporation of ^3^H-*myo*-inositol in each fraction (Figure [1](#F1){ref-type="fig"}). The acid-soluble fraction contains free Ins, Ins monophosphates and Ins polyphosphates. The hexane-soluble fraction consists mainly of Ins-containing phospholipids. The TFA-soluble fraction and cell debris mainly include tightly bound membrane components such as glycosyl-phosphatidyl inositol (GPI) protein anchors \[[@B26]\]. Between 15 and 20 DAP, most of the label was recovered in the acid soluble fraction, which contains PhA. After 20 DAP, a decrease in the relative content of the acid-soluble ^3^H-*myo*-inositol-labeled fraction occurred until at least 30 DAP. A simultaneous increase in the relative amount of label incorporation appeared in the corresponding hexane-soluble fraction. After 30 DAP, the relative rates of incorporation in the acid-soluble fraction increased again and remained high until at least 40 DAP.
{#F1}
Phytic acid accumulation in developing seeds of *Brassica napus*
----------------------------------------------------------------
HPLC analysis profiles indicate that PhA started to accumulate in detectable amounts during very early stages of seed development (12 DAP). Its accumulation became more pronounced at 20 DAP and continued progressively, reaching maximum levels at about 35 DAP (Figure [2](#F2){ref-type="fig"}). This time window (12--35 DAP) is, therefore, important in temporal targeting of molecular strategies for phytic acid reduction in canola seeds. In spite of the fact that Ins was shown to be present in both endosperm and seed coat at all stages examined (Figure [3](#F3){ref-type="fig"}B), no phytate accumulation was found in either tissue (Figure [2](#F2){ref-type="fig"}).
{#F2}
{#F3}
Variations in *myo*-inositol levels during seed development
-----------------------------------------------------------
At early stages of seed development, a steady decline in Ins levels occurred from 5 DAP up to 25 DAP (Figure [3](#F3){ref-type="fig"}A). These levels continued to decline further at 30 DAP, reaching their lowest point at 45 DAP through maturity. The sharp decline in the levels of Ins during the period 5--20 DAP coincided with the initial gradual accumulation of PhA (Figure [2](#F2){ref-type="fig"}) as well as the increase in the levels of the non-polar derivatives (Figure [1](#F1){ref-type="fig"} at 15--30 DAP). At 30 DAP the rapid decline in Ins levels resumed with a concomitant rise in PhA levels (Figure [2](#F2){ref-type="fig"}) at the expense of the non-polar components (Figure [1](#F1){ref-type="fig"}).
Meanwhile, as the embryo continued to expand through the different stages (Figure [3](#F3){ref-type="fig"}C), the level of Ins declined from its highest point in seed coat and endosperm and gradually increased in embryo tissues starting at 15--20 DAP (Figure [3](#F3){ref-type="fig"}B). At this point the rate of PhA biosynthesis in the embryo began to increase (Figure [2](#F2){ref-type="fig"}). No PhA synthesis was observed in either seed coat or endosperm.
Generation of transgenic lines of *B. napus* carrying the *myo*-inositol methyltransferase gene
-----------------------------------------------------------------------------------------------
Transgenic lines were generated from constructs pN*IMT* (IMT under napin promoter) and pPh*IMT* (IMT under phaseolin promoter). In both napin and phaseolin groups of transgenics, 80% of the lines showed reduced levels of PhA. Three transgenic lines with the highest PhA reduction were chosen from each promoter group for subsequent experiments. These were selfed to homozygosity and were shown to be consistently stable in terms of the *IMT* gene integrity and phytate reduction. One line from each group was chosen for further studies, namely N-11 (napin) and Ph1-18 (phaseolin). Lines Ph2-15 and Ph3-19 originated from two more transformation events and were chosen in the same way. These lines differed from Ph1-18 in that they harbored changes in the translational context of the *IMT* gene (Table [1](#T1){ref-type="table"}). In all transgenic lines germination rate was 100% for fresh seed, which did not differ from Westar controls. Additionally, over several generations the *IMT* transgenics did not exhibit changes in seed yield.
######
**Modified translational contexts for*IMT*gene driven by the phaseolin promoter**
**Name of construct** **Line** **Sequence**
----------------------------------- ---------- --------------------------
pPhIMT1 (parent-transgenic *IMT*) Ph1-18 A^-3^A^-2^A^-1^**ATG** A
pPhIMT2 Ph2-15 G^-3^C^-2^C^-1^**ATG** A
pPhIMT3 Ph3-19 A^-3^C^-2^C^-1^**ATG** A
Sequence column shows DNA triplets 5\' upstream from the translation initiation codon: line Ph1-18 (parent *IMT*) with dA nucleotides in positions -1 to -3; line Ph2-15 with dC nucleotides in positions -1 and -2 and dG nucleotide in position -3; line Ph3-19 with dC nucleotides in positions -1 and -2 and dA nucleotide in position -3.
***IMT*** expression occurs progressively in developing seeds of transgenic *B. napus*
--------------------------------------------------------------------------------------
In both lines of transformants (using napin and phaseolin promoters) production of IMT was verified by Western blot analysis, which also revealed absence of any native equivalents of IMT in developing seeds of non-transformed *B. napus* (Figure [4](#F4){ref-type="fig"}A and [4](#F4){ref-type="fig"}B). Although the *IMT* transcript started to appear at 20 DAP in developing transgenic seeds as exemplified by the phaseolin lines (Figure [4](#F4){ref-type="fig"}A), Western-blot analysis showed the corresponding protein to be produced in detectable amounts only after 25 DAP. The accumulation pattern of both the *IMT* transcript and its corresponding protein were similar in that they increased progressively to 40 DAP. Likewise, the napin-*IMT* line exhibited a similar manner of protein expression at 25 DAP (Figure [4](#F4){ref-type="fig"}B).
{#F4}
Transgenic seeds produce enzymatically active IMT and D-ononitol
----------------------------------------------------------------
HPLC analysis of mature transgenic seeds revealed the presence and accumulation of a new compound, which co-eluted with authentic D-ononitol standard (Figure [5](#F5){ref-type="fig"}A and [5](#F5){ref-type="fig"}B). The newly introduced IMT activity in transgenic lines was further confirmed by the ability of total soluble protein extracts from 40-DAP transgenic seeds to convert Ins to D-ononitol, in an *in vitro* IMT enzymatic assay (Figure [5](#F5){ref-type="fig"}C). Similar extracts from wild-type seeds as well as from transgenic leaves failed to produce this product. In addition, an unidentified compound eluted immediately prior to the Ins peak in the transgenic mature seed samples (not shown), which was absent in the wild-type samples. This compound did not co-elute with an authentic sample of pinitol.
{#F5}
Phytic acid content is reduced and inorganic phosphate content is enhanced in the seeds of transgenic plants
------------------------------------------------------------------------------------------------------------
HPLC analysis for phytate content in mature napin-*IMT* transgenic seeds showed that a 35% reduction in PhA level was achieved. HPLC analysis also showed a reduction of 19-29% in PhA in mature phaseolin-*IMT* seeds despite the translational context modification (Figure [6](#F6){ref-type="fig"}). Additionally, P~i~ levels increased from 10 to 31% which is consistent with earlier observations (Figure [7](#F7){ref-type="fig"}).
{#F6}
{#F7}
Carbohydrate analysis
---------------------
Changes in the carbohydrate species associated with Ins metabolism (e.g. galactinol, RFO and sucrose) were observed in mature transgenic seeds. In the three lines examined, the changes were consistent with the competition for available Ins among three pathways (PhA, galactinol and ononitol production, Scheme [1](#C1){ref-type="fig"}). Such competition appears to have resulted in the reduction of galactinol biosynthesis with consequential downstream effects, which are reflected by increases in galactose and sucrose levels with a concomitant decrease in raffinose level. However, the decrease in raffinose level appears to be accompanied by a proportionate rise in stachyose (Figure [8](#F8){ref-type="fig"}).
{#F8}
Discussion
==========
The relative incorporation of ^3^H-Ins in the different fractions of developing *B. napus* seeds, presented in Figure [1](#F1){ref-type="fig"}, depicts an image of the dynamics of Ins participation in phospholipid and PhA biosynthesis at early to mid stages of development. The data suggests the occurrence of horizontal interconversions between ^3^H-Ins and its polar and non-polar phosphorylated variants. Evidence of this is seen at 20 and 30 DAP where substantial shifts seem to occur across the hexane- and acid-soluble fractions.
The observed decrease in the relative contents of the acid-soluble fraction in the period 20--25 DAP reflects the utilization of *myo*-inositol in the biosynthesis of non-polar compounds (e.g. phospholipids). This is supported by the observation that a parallel increase in the relative amounts of label incorporation appeared in the hexane-soluble fraction during the same period. After 30 DAP, the relative rates of incorporation in the acid-soluble fraction increased again at the expense of the non-polar fraction, reflecting the rapid accumulation of PhA during this period (Figure [2](#F2){ref-type="fig"}). This is likely, since hydrolysis of Ins-containing phospholipids is known to lead to increased phytate accumulation in seeds of *B. napus*\[[@B11]\].
Typically, the newly forming seed coat is made up of four layers of cells, the outermost epidermal and palisade cell layers (which develop from the outer integument of the ovule) and parenchyma and endothelial cell layers (which are derived from the inner ovule integument) \[[@B27]\]. At 5 DAP the majority of seed inner space is filled with differentiating integument and a small amount of endosperm. That the majority of Ins at 15--20 DAP is localized in the seed coat of dissected seeds (Figure [3](#F3){ref-type="fig"}B) and that the bulk of the seed at 5--15 DAP is primarily made up of this tissue may be construed as evidence that the seed coat is the primary site of Ins biosynthesis in the *B. napus* seed. In support of this is the abundant supply of sucrose, which is transported through the phloem, apoplastic region and the seed coat during this early stage \[[@B28],[@B29]\] and which, through hydrolysis, acts as source of G-6-P, the precursor for Ins.
As the embryo expands through its initial globular, heart, torpedo and bent cotyledon stages, the endosperm proportionally shrinks (Figure [3](#F3){ref-type="fig"}C), and Ins becomes localized mainly in the embryo and seed coat (Figure [3](#F3){ref-type="fig"}B). That the seed coat is the tissue where Ins is most abundant at the early stages of seed development is of special importance since it plays a major role in the biosynthesis of mucilage, other seed coat polysaccharides and sugar acids through its participation in the oxidative pathway \[[@B30]\]. PhA synthesis started at very early stages (\<10DAP) and reached its maximum levels at about 35 DAP (Figure [2](#F2){ref-type="fig"}). Distribution analysis of PhA accumulation between cotyledons and embryo axes revealed that from 20 to 75 DAP more than 80% of PhA is accumulated in cotyledons (80% at 20--25 DAP, increasing up to 90% after 30 DAP) (data not shown). At 25--30 DAP, as the *Brassica* embryo expanded, higher accumulation of the non-polar derivatives (needed for membrane biogenesis) occurred (Figure [1](#F1){ref-type="fig"}). This period (25--30 DAP) appears to be marked by a slower decline in Ins levels, nearly reaching a plateau (Figure [3](#F3){ref-type="fig"}A), while, in parallel, PhA levels continued to rise at a relatively slower rate. After 25 DAP and through Ins appears to have reached a stable level in embryo, and more so in seed coat which could account for its apparent slower decline between 25 and 30 DAP (Figure [3](#F3){ref-type="fig"}B). At this stage the decline in Ins levels was accompanied by a concomitant rise in PhA levels (Figure [2](#F2){ref-type="fig"}) at the expense of the non-polar components (Figure [1](#F1){ref-type="fig"}) as the seed approached desiccation through the onset of RFO biosynthesis (30-35DAP) (unpublished results).
Since the initial phosphorylation steps of free Ins commence either with the reconstitution of L-Ins-1-P or through other positional phosphate esters, our strategy was to investigate the effect of competitive metabolic shunt of Ins on its phosphorylation and subsequent PhA and RFO accumulation in canola seeds. To accomplish this, we have explored the conversion of Ins into ononitol (1-D-4-*O*-methyl-*myo*-inositol), by methylation at the D-4 position, through the action of the IMT enzyme.
The observed variations in PhA reduction levels do not necessarily reflect different *IMT* expression levels since all selected lines displayed almost uniform levels of the IMT protein at the time of sampling as shown by Western blot analysis (Figure [4](#F4){ref-type="fig"}A and [4](#F4){ref-type="fig"}B). That the recorded ranges of PhA reduction appear to be similar with either promoter (Figure [6](#F6){ref-type="fig"}) suggests one of two possible scenarios: a) A certain threshold may exist at which a steady balance between supply and removal of Ins is reached, which confines PhA reduction levels within the observed limits; b) The temporal appearance of IMT activity under these promoters may not be in synchrony with the highest point of Ins accumulation, which is presumably reached in less than 10 DAP (Figure [3](#F3){ref-type="fig"}A), leading to the early onset of PhA synthesis. The latter scenario is likely, since the IMT protein was not detected until 25 DAP in lines Ph1-18 and Ph2-15 (Figure [4](#F4){ref-type="fig"}A) in contrast to PhA accumulation which is shown to be in progress at 15--20 DAP (Figure [2](#F2){ref-type="fig"}). A similar result was obtained when a *MIPS* antisense transcript was expressed in transgenic rice driven by the glutelin *GluB-1* promoter \[[@B17]\].
The phaseolin promoter was chosen based on its reported early transcriptional activation in transgenic systems such as tobacco (15--16 DAP) \[[@B31],[@B32]\]. The napin promoter which is native to *Brassica* was chosen with a view to comparing the effect of temporal expression differences of the two promoters on PhA accumulation. However, because *IMT* transcription from the phaseolin promoter did not commence as early as in the case of transgenic tobacco, it was not possible to assess such differences. The complex architecture of the phaseolin promoter has been shown to play a major role in spatial regulation of this promoter in transgenic systems \[[@B33]\]. Further, seed-specific transcriptional regulatory regions in the same promoter have been identified which affect its activation in a temporally dependent manner \[[@B34],[@B35]\]. Therefore, it is possible that the phaseolin promoter, when expressed in different systems, could be affected by elements that may impose temporal expression variations. This is suggested by the fact that, in our hands, the transcriptional activation of phaseolin-*IMT* was triggered in canola at a later time point (20 DAP, Figure [4](#F4){ref-type="fig"}A) than in tobacco, indicating that this promoter could be influenced by developmentally regulated programs in a host-specific manner.
Attempts at enhancing the *IMT* translation efficiency through the modification of its translational context \[[@B36],[@B37]\] resulted in somewhat improved translation levels at 20 DAP under the phaseolin promoter when a dA nucleotide was positioned at the -3 position and a dC nucleotide at each of positions -2 and -1 of the initiation codon (transgenic line Ph3-19, Table [1](#T1){ref-type="table"} and Figure [4](#F4){ref-type="fig"}A). Translation levels at subsequent stages in the same line also appear to have been relatively enhanced. However, in spite of the protein level enrichment at 20 DAP in line Ph3-19, this did not substantially improve the overall PhA reduction, confirming the need for a more temporally and (or) spatially synchronized *IMT* expression and Ins production, as opposed to early enhancement of *IMT* translation. Nevertheless, while the apparent compartmentalization of Ins in the seed coat may shield it from IMT action in the embryo, seed-specific promoters are known to drive gene expression in the inner layer of seed coats and there is active transport of Ins from seed coat to embryo \[[@B38]\].
In mutant lines of other crops (e.g. maize \[[@B13],[@B39]\] and soy bean \[[@B15]\]), the decrease in PhA phosphorous in mature seeds is generally accompanied by a parallel, albeit variable, increase in P~i~\[[@B15]\]. However, in some mutant lines (e.g. maize *lpa*2-1), the rise in P~i~ can be accompanied by an accumulation of other Ins phosphates (Ins(1,2,4,5,6)P~5~; Ins(1,4,5,6)P~4~; and Ins(1,2,6)P~3~) \[[@B39]\]. This makes the decrease in PhA futile in such mutants since these highly phosphorylated Ins species retain many of the adverse PhA properties. In the present study, although we observed a similar inverse relationship between PhA and P~i~, there was no detectable accumulation of any of the partially phosphorylated Ins intermediates such as InsP~3~, ~4or\ 5~. Furthermore, over-expression of *IMT* did not affect the seed viability or the germination efficiency of transgenic canola seeds. In addition, there were no measurable yield penalties (Table [2](#T2){ref-type="table"}).
######
**Seed yield of*IMT*transgenic lines**
**Line** **Seed yield (g/plant)**
---------- --------------------------
Westar 10.56 ± 0.76
N-11 11.20 ± 0.80
Ph1-18 10.11 ± 0.62
Ph2-15 9.74 ± 0.69
Ph3-19 9.76 ± 0.69
Mature seeds of Westar, napin-*IMT* (N-11) and phaseolin-*IMT* (Ph1-18, Ph2-15 and Ph3-19) transgenic lines were harvested and weighed. Each data point represents mean value of five biological replicates ±SE (Standard error).
Absence of ononitol in wild-type HPLC chromatograms and failure of the wild-type protein extracts to produce ononitol, together with the Western analysis results suggest possible absence of native IMT-like activity in *B. napus*.
In mature transgenic seeds, the level of Ins was not changed significantly (Figure [8](#F8){ref-type="fig"}) but presumably during maturation was partitioned among PhA, galactinol and ononitol biosyntheses (Figure [5](#F5){ref-type="fig"}B). Accordingly, PhA formation was decreased in concert with ononitol accumulation. This pathway perturbation (Scheme [1](#C1){ref-type="fig"}) would have lowered the production of galactinol as well, and consequently, RFO accumulation \[[@B40]\] with the notable increase in sucrose levels (Figure [8](#F8){ref-type="fig"}). Curiously, the decrease in raffinose levels was accompanied by an almost proportionate rise in stachyose. Since sucrose levels remained high in transgenic seeds, the inverse modulation in the levels of raffinose and stachyose could indicate raffinose as a possible galactosyl donor for the subsequent chain elongation in a raffinose:raffinose galactosyltransfer manner, in which a molecule of raffinose would release its sucrose moiety after each galactosyl-residue transfer to another raffinose molecule (Figure [9](#F9){ref-type="fig"}B) \[[@B41],[@B42]\]. This is conceivable in view of the limited availability of galactinol (Figure [8](#F8){ref-type="fig"}). Alternatively, since methylated derivatives of galactinol are known to take part in RFO synthesis in some plants \[[@B43]\], and although not *hitherto* proven in *B. napus*, we propose that ononitol, the methylated product of IMT, may be utilized to form the corresponding galactosylononitol (methylated galactinol), which could still participate, in part, in RFO-chain elongation to the higher oligomer, stachyose (Figure [9](#F9){ref-type="fig"}A). Since this does not account for the observed increase in sucrose levels, we postulate that both routes may be working concertedly. The differential extent of stachyose and sucrose accumulation (Figure [8](#F8){ref-type="fig"}) could be indicative of the different kinetics of the two routes (Figure [9](#F9){ref-type="fig"}A and [9](#F9){ref-type="fig"}B).
{#F9}
Conclusions
===========
Early stages of *Brassica* seed development appear to be dominated by reciprocal interconversions of polar Ins phosphates and non-polar species. Although Ins is shown to be produced in significant amounts in the endosperm as well as seed coat during the early stages, no PhA accumulation occurs in those tissues. Instead, PhA accumulation appears to be mainly restricted to the embryo throughout seed development and maturation.
In addition to lowering PhA levels in the developing seeds, the competitive methylation of Ins resulted in changes in the distribution and accumulation patterns of seed carbohydrates, leading to enhancement of the digestible and metabolizable energy profile of the meal as demonstrated by the higher content of the nutritionally useful sugar, sucrose. While the ratio of raffinose to stachyose was altered, the overall balance of RFO appears to be unaffected. No deleterious effects were encountered as a result of Ins methylation in developing *Brassica* seed.
It is evident from this study as well as previous ones that one of the answers to reducing phytate in crop seeds to a level which allows an appreciable beneficial effect on phosphorous and microelements bioavailability without adversely affecting yield and phenotype may be found in a model which combines the additive effects of more than one mechanism acting concertedly with separate independent contributions. Such an approach would allow selective inhibition of Ins phosphorylation at different steps for optimum phytate reduction. Nonetheless, the current strategy, if adopted, could potentially raise the canola market value as well as that of other crops.
Methods
=======
Plant material and chemicals
----------------------------
Ice plant (*Mesembryanthemum crystallinum*) seeds were germinated, grown, salt treated and sampled as described previously \[[@B20],[@B44]\]. Seeds of *B. napus* (Westar) plants, grown under growth chamber conditions (16 hour day at 20°C/8 hour night at 15°C photoperiod) were harvested at maturation. Seeds at different developing stages were also collected. Developing seeds were separated into seed coats and embryos on ice under a binocular dissecting microscope. Fresh seeds were cut half open with a scalpel. Endosperm of 15 and 20 DAP was collected by pipette. The seed coat and embryos were washed with ultra pure water three times. The excess water was absorbed with filter paper. The developing seeds and dissected tissue were frozen immediately in liquid N~2~. These were freeze dried for 24 hours and then extracted for PhA and Ins analysis.
Ononitol standards were purchased from GlycoSyn Technologies, Lower Hutt, New Zealand.
*In Vivo* labeling of developing seeds with ^*3*^*H-myo-*inositol
-----------------------------------------------------------------
Siliques at different developmental stages (15--40 DAP) were cut from plants and the cut end immediately put into 10 ml sterile culture medium in 50 ml tubes supplemented with 5μCi ^3^H-*myo*-inositol and incubated in a growth chamber for two days. Seeds were harvested and crushed in liquid nitrogen then extracted consecutively with hexanes (lipid fraction), 0.5N HCl (phytate-containing acid-soluble fraction) and trifluoroacetic acid (TFA-soluble cell debris fraction). The radioactivity in each fraction was assessed in a scintillation counter.
Phytic acid extraction
----------------------
Mature canola seeds (350 mg) were homogenized in 4 ml of ammonia in methanol (10% w/w) \[[@B45]\], then vortexed. After a 10-minute incubation, 3 ml of hexanes was added and the samples were vortexed and centrifuged (RCF 2500). The liquid phases were discarded and the seed pellets re-extracted with 3 ml of hexanes, centrifuged and the supernatants discarded. Pellets were washed three times with 5 ml of absolute methanol, resuspended in 6 ml of 0.5 N HCl and kept at room temperature for 15 minutes. The slurry was centrifuged and the supernatant was filtered through a 0.45 μm GHP Acrodisk filter (Gelman Science) prior to HPLC analysis. At this stage samples were stored at -20°C. PhA was extracted from 50-100 mg of seeds at different developmental stages. For dissected material, seed coat (20 mg), endosperm (20 mg), embryo: 15 DAP (10 mg), 20 DAP (50 mg) and older (100 mg) was used. The extraction volume of HCl was adjusted proportionately.
HPLC conditions for phytic acid analysis
----------------------------------------
Phytic acid analysis was performed on a Waters 660E multi-solvent delivery system equipped with in-line degasser AF, 717plus Autosampler and a Sedex 55 (S.E.D.E.R.E.) evaporative light scattering detector at 50°C, 2 bar, gain 7. The HPLC system was controlled and data processed by the Waters Millennium™ 2010 Chromatography Manager, version 2.15.01. Samples were chromatographed on an IC-Pak Anion HC, 150 mm × 4.6 mm column (WAT026770, Waters), at 22°C, with 100 mM nitric acid at a flow rate of 1.0 ml/min. A Waters IC-Pak Anion Guard-Pak (WAT010551) was used as the pre-column. A fritted filter guard (A-103X Rep Frit (BLK) 0.94 × 0.25 from Upchurch Scientific) was placed in front of the pre-column. Injections, typically 75 μl of undiluted sample, were in duplicate with a separate result generated for each injection. A calibration curve was prepared for each run with the levels of standards at 25, 50, 75 and 100 μg. Standards were prepared with phytic acid dodecasodium salt C~6~H~6~O~24~P~6~Na~12~·9H~2~O (Sigma), which was dissolved in the same concentration HCl as that used to extract the samples (0.5 N) to a concentration of 10 μg/μl. The concentrated solution was diluted to 1.0 μg/μl in 0.5 N HCl. Volumes of 25, 50, 75 and 100 μl of the diluted standard were injected in duplicate and phytic acid was detected at an average retention time of 3.9 minutes.
HPLC conditions for carbohydrate analysis
-----------------------------------------
Sugars were extracted from seeds, and analyzed by HPLC as described \[[@B40]\]. Briefly, duplicates of defatted tissue of ten seeds, unless otherwise indicated, were extracted with 80:20 v/v ethanol-water at 70°C for 30 min, followed by centrifugation and evaporation of the supernatant to dryness. Samples were reconstituted in 18-MΩwater, and filtered through 0.45 μm nylon filters prior to HPLC analysis. To accommodate analysis of early seed stages, extraction and reconstitution volumes were proportional to sample weight.
Galactinol, *myo*-inositol and D-ononitol were separated on a CarboPac™ MA1 column (4 mm × 250 mm) preceded by a CarboPac™ MA1 guard column (4 mm × 50 mm) with 500 mmol L^-1^ isocratic NaOH at 0.40 mL min^-1^ as eluent and detected by high performance anion exchange-pulsed amperometric detection (HPAE-PAD) using a Dionex ICS-3000 system (Dionex Corp., Sunnyvale, Calif.). Standards were galactinol, *myo*-inositol (Sigma-Aldrich, St. Louis, MO) and D-ononitol (GlycoSyn Technologies, Lower Hutt, New Zealand).
Glucose, galactose, fructose, sucrose, stachyose and raffinose were separated on a CarboPac™ PA1 column (2 mm × 250 mm) preceded by a CarboPac™ PA1 guard column (2 mm × 50 mm) and then an Amino-Trap™ guard column (2 mm × 50 mm) with 25 mmol L^-1^ isocratic NaOH at 0.25 mL min^-1^ as eluent and detected by HPAE-PAD. Standards were galactose, fructose, sucrose, raffinose, stachyose (Sigma-Aldrich, St. Louis, MO), and glucose (Fisher Scientific, Hampton, NH).
Inorganic phosphate analysis
----------------------------
Inorganic phosphate (P~i~) in mature transgenic seeds was assayed using published protocols \[[@B46]\].
Cloning of *IMT* and plant transformation
-----------------------------------------
Fresh salt-stressed ice plant leaf tissue was frozen in liquid nitrogen and crushed to fine powder. Total RNA was extracted with TRIzol® Reagent (Invitrogen). Poly (A)^+^ RNA was isolated using published methods \[[@B47]\]. The first strand cDNA was generated using the Roche reverse transcription kit, and amplified with the sequence specific primers (forward, 5'-TTTTTGGATCCAAGAGAA AAAAAAATGACTACTTACACC-3\' and reverse, 5'-TTTTTGCGGCCGCATAAAGGCAAATCATACACTG-3') by PCR based on the published sequence (Accession No. M87340). The reaction was initiated by heating at 94°C for 2 min followed by 35 cycles of heating at 94°C for 1 min, annealing at 52°C for 1 min and extension for 3 min at 72°C. The PCR product (1418bp) was purified using a PCR purification kit (Promega) followed by digestion with *Bam*HI and *Not*I, whose sites were incorporated into the forward and reverse primers, respectively. The digested DNA fragment was subcloned into pSPORT1 (Invitrogen) and sequenced. The *IMT* gene was subcloned into pRD400 \[[@B48]\], which contains a napin promoter to produce plasmid pNIMT. Additional versions of the IMT gene, with modified translational context were produced \[[@B36],[@B37]\] (Table [1](#T1){ref-type="table"}). These as well as the parent-transgene were cloned into pRD 400 under the phaseolin promoter to generate pPhIMT1-3. All constructs were transferred into *Agrobacterium tumefaciens* strain GV3101 containing the helper plasmid pMP90 by triparental mating, followed by *Agrobacterium*-mediated transformation of *Brassica napus* (cv. Westar) \[[@B49]\].
Over-expression of *IMT* in *E coli* and production of antibodies
-----------------------------------------------------------------
The IMT cDNA fragment was cloned into the bacterial expression plasmid, pPROEXHTb (Invitrogen). Protein expression was induced by adding IPTG to the culture medium to a final concentration of 1 mM. The *E. coli* culture was harvested after a 3-hour incubation at 37°C. The His-tagged protein was purified with Ni-NTA Agarose (Qiagen) under denaturing conditions. The purified protein (43 Kda) was used to raise polyclonal antibodies against the IMT enzyme \[[@B50]\], which were subsequently used in Western analyses of transgenic lines.
Southern and Northern blot analyses
-----------------------------------
Genomic DNA was extracted from leaf tissue using Wizard® Genomic DNA Purification Kit (Promega). RNA extraction from developing seeds of both transgenic and wild-type plants was conducted by using RNeasy plant total RNA kit (Qiagen). Southern and Northern analyses using Hybond-N^+^ membrane (Amersham) were performed essentially as described \[[@B51]\].
IMT Enzyme assay and Western analysis in transgenic plants
----------------------------------------------------------
Total protein extracts from developing seeds (approximately 40 DAP) as well as leaves of transgenic lines, and wild-type plants were assayed for IMT-enzyme activity exactly as described previously \[[@B20]\].
For Western analysis, total soluble proteins were extracted from developing seeds as described \[[@B44]\]. TND buffer, 90 mM Tris--HCl (pH 8.3 at 4°C), 9mM DTT and 2 mM Leupeptin (100 μl) was added to 50 mg of frozen seeds crushed in liquid nitrogen. Soluble protein samples were prepared by collecting the supernatants. Protein concentration in each sample was determined using the Bradford assay with BSA as the standard. Protein samples (15 μg each) were used in Western blot analysis. The samples were separated by PAGE and blotted on a Nitrocellulose membrane (Bio-Rad) \[[@B51]\]. The immune-reactions were conducted using Bio-Rad Immun-Blot® Assay Kit.
Abbreviations
=============
DAP: Days after pollination; Ptd-CMP: Phosphatidylcytosine monophosphate; PtdIns: Phosphatidylinositol; PI-K: Phosphatidylinositol kinase; PtdInsP: Phosphatidylinositol monophosphate; GolS: Galactinol synthase; RFO: Raffinose oligosaccharides.
Competing interests
===================
The authors declare that they have no competing interests.
Authors' contribution
=====================
FG, project PI, designed the concept, supervised all the experiments, contributed to and edited the manuscript. JD, WY, CB, KN performed the experiments and participated in the interpretation of results. WK supervised Brassica transformation experiments. All authors read and approved the final manuscript.
Acknowledgements
================
This study was supported through an NRCC-DASCI Strategic Alliance Agreement. Ice plant seeds were a gift from Professor Hans Bohnert. The phaseolin promoter was provided by Dow AgroSciences as pAGM219 plasmid. The contributions of Atta Hussain, supported by grants from the Canola Council of Canada and Dow AgroSciences, are hereby acknowledged. We also thank Sandra Polvi for technical support. This is NRCC publication number 54672.
| {
"pile_set_name": "PubMed Central"
} |
Introduction {#sec1-0300060517700299}
============
Increased glucocorticoid (GC) levels are the most common nontraumatic cause of osteonecrosis of the femoral head (ONFH).^[@bibr1-0300060517700299],[@bibr2-0300060517700299]^ GC-induced ONFH in young adults usually requires hip replacement^[@bibr3-0300060517700299],[@bibr4-0300060517700299]^ However, several studies have shown poor prosthetic durability in patients with ONFH.^[@bibr5-0300060517700299][@bibr6-0300060517700299]--[@bibr7-0300060517700299]^ A previous study showed that the mean daily GC dose was strongly associated with osteonecrosis (ON).^[@bibr8-0300060517700299]^ Most cross-study analyses demonstrate that a sustained large dose of GC can induce symptomatic ON.^[@bibr9-0300060517700299],[@bibr10-0300060517700299]^
There is no widely held consensus on the pathogenesis of GC-induced ON. Several mechanisms of GC-induced ON have been proposed ([Figure 1](#fig1-0300060517700299){ref-type="fig"}). A novel mechanism of GC-induced ON is apoptosis in osteoblasts and osteocytes, thus compromising bone formation and integrity.^[@bibr11-0300060517700299][@bibr12-0300060517700299][@bibr13-0300060517700299]--[@bibr14-0300060517700299]^ However, the traditional concept of GC-induced ON implicates ischaemia as the main aetiological factor. GCs are thought to interrupt blood supply to the bone and eventually cause ONFH in a variety of ways.^[@bibr15-0300060517700299][@bibr16-0300060517700299][@bibr17-0300060517700299]--[@bibr18-0300060517700299]^ The most common causes of interruption of the blood supply include fat embolism and coagulation disorders.^[@bibr19-0300060517700299][@bibr20-0300060517700299][@bibr21-0300060517700299]--[@bibr22-0300060517700299]^ This article summarizes existing knowledge on coagulation disorders in the context of GC-induced ON. We review the literature and highlight controversies, with emphasis on the questions of how GC-induced coagulation disorders, directly or indirectly, relate to ischaemia in GC-induced osteonecrosis. Figure 1.Plausible mechanisms for steroid-induced development of ONFH
Hypofibrinolysis and thrombophilia {#sec2-0300060517700299}
==================================
Previous studies showed that high doses of dexamethasone administered to rats inhibited fibrinolytic activity by decreasing tissue plasminogen activator (t-PA) activity and increasing plasma plasminogen activator inhibitor-1 (PAI-1) antigen levels.^[@bibr23-0300060517700299][@bibr24-0300060517700299]--[@bibr25-0300060517700299]^ PAI-1 plays a role in fibrinolysis by forming complexes with t-PA. The t-PA/PAI-1 complexes do not have the ability to activate plasminogen to plasmin. GCs increase the activity of PAI-1, leading to hypofibrinolysis and a relatively hypercoagulable state.^[@bibr26-0300060517700299]^ Subsequent research showed decreased fibrinolytic activity, as a consequence of increased PAI-1, and decreased t-PA, by GCs in animals and patients with ON.^[@bibr18-0300060517700299],[@bibr27-0300060517700299][@bibr28-0300060517700299]--[@bibr29-0300060517700299]^. Furthermore, as important factors of hypofibrinolysis, plasma fibrinogen and lipoprotein (a) (Lp(a)) are also abnormalities found in GC-induced or idiopathic ON.^[@bibr30-0300060517700299][@bibr31-0300060517700299][@bibr32-0300060517700299][@bibr33-0300060517700299]--[@bibr34-0300060517700299]^ In an ON animal model, Drescher et al.^[@bibr30-0300060517700299]^ showed that plasma fibrinogen was significantly elevated in ON following mega-dose GC treatment, which suggested a hypercoagulable condition in GC-induced ON. In a clinical study, Pósán et al.^[@bibr34-0300060517700299]^ found that Lp(a) levels were elevated in patients with primary and secondary ONFH. Other studies have investigated the association between thrombophilia and development of ON following GC treatment.^[@bibr17-0300060517700299],[@bibr32-0300060517700299],[@bibr35-0300060517700299][@bibr36-0300060517700299]--[@bibr37-0300060517700299]^ Guan et al.^[@bibr35-0300060517700299]^ showed that, at 24 hours after prednisolone injection, abnormal hypercoagulability occurred in a rabbit model. Glueck et al.^[@bibr32-0300060517700299]^ compared 36 patients with primary and secondary ONFH with healthy volunteers. They found that these patients were more likely to have thrombophilic disorders, heterozygosity or homozygosity for platelet glycoprotein IIIa P1A1/A2 polymorphism, anticardiolipin antibodies, lupus anticoagulant, or both, and deficiency in proteins C and S, or antithrombin III.
However, the association between hypofibrinolysis or thrombophilia with primary or secondary ON is unclear. Seguin et al.^[@bibr38-0300060517700299]^ showed that there was no association between thrombophilia with ON and considered that GC-induced regional endothelial dysfunction was a more likely reason. Asano et al.^[@bibr39-0300060517700299]^ found that genotypes of PAI-1 4G/5G and MTHFR C677T or plasma concentrations of PAI-1 Ag and tHcy had no effect on the incidence of ONFH in Japanese subjects, and suspected that this may differ according to race.
Endothelial cell dysfunction and damage {#sec3-0300060517700299}
=======================================
Endothelial dysfunction may present early in GC-induced ONFH. Yu et al.^[@bibr40-0300060517700299]^ found that GC significantly affected the transcriptome of vascular endothelial cells of the human femoral head. Chen et al.^[@bibr41-0300060517700299]^ showed circulating endothelial progenitor cell damage in patients with GC-induced ONFH. In a histopathological study, Nishimura et al.^[@bibr42-0300060517700299]^ found endothelial cell damage by electron microscopy in steroid-treated rabbits. Li et al.^[@bibr27-0300060517700299]^ also showed endothelial cell damage with a high coagulant and low fibrinolytic milieu in an experimental study on GC-induced ON. In patients with dysbaric osteonecrosis, Slichter et al.^[@bibr43-0300060517700299]^ found platelet thrombus formation, which was secondary to endothelial cell damage in the femoral head.
The pathogenesis of GC-induced endothelial cell dysfunction and damage is multiple, and oxidative stress may play an important role.^[@bibr44-0300060517700299][@bibr45-0300060517700299][@bibr46-0300060517700299]--[@bibr47-0300060517700299]^ After initial damage of endothelial cells triggered by GCs or other factors, a hypercoagulable state is produced. This is followed by vascular problems (thrombosis, poor blood flow, and ischaemia), and this in turn results in endothelial cell damage, which may be cyclic.^[@bibr48-0300060517700299][@bibr49-0300060517700299][@bibr50-0300060517700299]--[@bibr51-0300060517700299]^
Endothelial cell apoptosis {#sec4-0300060517700299}
==========================
GCs can induce endothelial cell apoptosis by a different signalling pathway.^[@bibr52-0300060517700299][@bibr53-0300060517700299][@bibr54-0300060517700299]--[@bibr55-0300060517700299]^ Endothelial cell apoptosis consequently promotes thrombus formation and ON by two major mechanisms. First, apoptotic bodies can indirectly lead to coagulopathic changes by endothelial dysfunction. Second, apoptotic endothelial cells can stimulate adhesion of platelets to endothelial cells and activate platelets, eventually leading to thrombus formation.^[@bibr50-0300060517700299]^
However, GCs can induce endothelial cell apoptosis and lead to a hypercoagulable state. Cessation or a reduction in blood flow along capillaries could also play an aetiological role in endothelial cell apoptosis.^[@bibr59-0300060517700299][@bibr60-0300060517700299]--[@bibr61-0300060517700299]^
Lipid metabolism {#sec5-0300060517700299}
================
There is abundant evidence that excessive GCs can induce hyperlipidaemia, fat hypertrophy, fat deposition within the femoral head intramedullary tissue, and fat embolism. These factors may cause ischaemia by elevating intraosseous pressure and decreasing blood flow, eventually leading to ONFH.^[@bibr62-0300060517700299][@bibr63-0300060517700299][@bibr64-0300060517700299][@bibr65-0300060517700299][@bibr66-0300060517700299][@bibr67-0300060517700299][@bibr68-0300060517700299][@bibr69-0300060517700299]--[@bibr70-0300060517700299]^
However, beyond the above-mentioned changes, dyslipidaemia can also lead to a hypercoagulable state and aggravate ischaemia.^[@bibr20-0300060517700299][@bibr21-0300060517700299]-[@bibr22-0300060517700299],[@bibr50-0300060517700299],[@bibr71-0300060517700299]^ Jones et al.^[@bibr22-0300060517700299]^ found intraosseous fibrin thromboses after induction of experimental fat emboli and speculated that fat emboli could trigger intravascular coagulation. Additionally, some vasoactive substances that are released from injured marrow adipocytes can affect endothelial cells that line blood vessels and produce a hypercoagulable state.^[@bibr50-0300060517700299]^
Platelet activation {#sec6-0300060517700299}
===================
High doses of GCs induce platelet aggregation.^[@bibr72-0300060517700299],[@bibr73-0300060517700299]^ There is evidence that platelet activation is involved in GC-induced ON. Masuhara et al.^[@bibr74-0300060517700299]^ found that platelet activation may play an important role in experimental ON in rabbits. In patients with ONFH, Pósán showed that platelet activation (measured by beta triglyceride) was significantly higher compared with that in healthy controls.^[@bibr34-0300060517700299]^ Similarly, in some animal studies on GC-induced femoral head necrosis, blood platelet levels were decreased in the early stage.^[@bibr35-0300060517700299],[@bibr75-0300060517700299]^ This finding indicates the occurrence of consumption coagulopathy caused by activation not only of endothelial cells, but also of platelets. Additionally, platelet thrombus formation has been detected in arterioles adjacent to the necrotic area by histopathological observation.^[@bibr43-0300060517700299],[@bibr71-0300060517700299],[@bibr75-0300060517700299]^
In summary, platelet activation is involved in progression of GC-induced ON and the effect may be secondary to endothelial cell damage by GC.
Anticoagulant treatment {#sec7-0300060517700299}
=======================
Hypofibrinolysis (decreased ability to lyse clots) and thrombophilia (increased likelihood of forming thrombi) appear to play important roles in ON. If coagulation abnormalities cause ON, then anticoagulation therapy might ameliorate it. Wada et al.^[@bibr76-0300060517700299]^ found that warfarin decreased the incidence of ON in spontaneously hypertensive rats. Glueck et al.^[@bibr77-0300060517700299]^ studied patients whose ON was caused by heritable thrombophilia or hypofibrinolysis. They showed that 12 weeks of therapy with enoxaparin before femoral head collapse may slow progression or stabilize ON, as determined by X-ray and MRI, while providing pain relief. Motomura et al.^[@bibr78-0300060517700299]^ demonstrated that the combined use of warfarin and probucol helps prevent steroid-induced ON in rabbits. Kang et al.^[@bibr79-0300060517700299]^ also found that combination treatment with enoxaparin and lovastatin reduced the incidence of GC-induced ON in the rabbit.
In summary, coagulation abnormalities may play an important role in GC-induced ON. Additionally, anticoagulation therapy can significantly decrease the incidence of ON in GC-treated rabbits.
Conclusion {#sec8-0300060517700299}
==========
This article provides an overview of the role of coagulopathy in GC-induced ON. GCs can directly lead to hypofibrinolysis and thrombophilia or indirectly lead to endothelial cell dysfunction and damage. Endothelial cell apoptosis, lipid metabolism, and platelet activation lead to a hypercoagulable state, followed by poor blood flow, ischaemia, and eventually ONFH. Experimental studies have shown that use of an anticoagulant alone or combined with a lipid-lowering agent is beneficial in preventing GC-induced ON. Better understanding of the pathogenesis of GC-induced ON can generate better treatment options.
Declaration of conflicting interest {#sec9-0300060517700299}
===================================
The Authors declare that there are no conflicts of interest.
Funding {#sec10-0300060517700299}
=======
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
[^1]: These authors contributed equally to this work.
| {
"pile_set_name": "PubMed Central"
} |
Background
==========
Obesity is an important risk factor for many chronic conditions, such as cardiovascular diseases, cancers and diabetes mellitus \[[@B1]-[@B3]\]. The worldwide prevalence of obesity has doubled since 1980 \[[@B4]\]. In the UK it has trebled in the past thirty years and further increases are predicted \[[@B5],[@B6]\]. A range of approaches will be needed to halt the increase in obesity and to reduce its prevalence. These will include primary prevention by engineering public places to promote physical activity and encouraging the avoidance of unhealthy foods \[[@B7]\]. Effective treatment of overweight and established obesity, however, will also be needed \[[@B8]\]. Several models for providing obesity management in primary care have been reported, including the use of community pharmacies.
It is estimated that 95% of the population visit a community pharmacy during the year \[[@B9],[@B10]\]. Because of this, pharmacies in Scotland have been encouraged to provide a variety of health related services - these have included health assessments for diabetes, cardiovascular diseases, asthma, and smoking cessation \[[@B9]\]. Pharmacies also offer over-the-counter weight loss products \[[@B11]\]. Where pharmacy has been used as one treatment arm in a randomised controlled trial comparing a range of weight reduction programmes, 14% of pharmacy participants lost at least 5% of their initial weight although there was no significant difference in their mean weights at 12 months \[[@B12]\]. A recent systematic review of the effectiveness and cost effectiveness of community pharmacy-based weight management identified 10 initiatives involving 582 pharmacies in 5 countries. The authors of the review found that community pharmacy weight management could produce modest weight loss at 12 months of between 1.1-4.1 kg, but concluded that there was insufficient evidence of effectiveness and cost effectiveness because of limitations in how the studies were carried out and reported \[[@B13]\]. For example, only 3 studies reported long-term (12 month) weight change; only one study employed evidence-based guidelines in its programme; and few reported any cost information. The authors of the review recommended that body weight should be objectively measured and that proportions of patients achieving clinically significant weight loss should be reported alongside mean weight loss to enable comparison of effectiveness from a clinical point of view \[[@B13]\]. The health benefits of clinically significant weight loss, defined as loss of ≥5% baseline weight, include reduced blood pressure, improved glycaemic control, reduction in risk of type 2 diabetes, improved lipid profiles, and reduced osteoarthritis-related disability \[[@B14]\].
Counterweight is a weight management programme that has been evaluated for use in routine National Health Service primary care \[[@B15],[@B16]\]. A case series study showed that it achieved clinically significant long-term weight loss in 14% of all patients at 12 months \[[@B16]\]. The programme was introduced into community pharmacies in the Fife region of Scotland in 2009. Our aim was to evaluate the effectiveness of the Counterweight Programme delivered within community pharmacies, using a primary outcome of clinically significant weight change at 12 months.
Methods
=======
Counterweight intervention
--------------------------
The Counterweight weight management programme was provided in the Fife region (population 365,000) as part of the Keep Well project. The Keep Well project encourages 40 to 64 year olds, who live in geographical areas which have been identified as having greatest need, to improve their health. The program targets those individuals at high risk of cardiovascular disease, and offers medical advice and support through enhanced Primary Care services. Between 2008 and 2010, 50 general practices in Fife were engaged in the delivery of the Keep Well project. Patients registered with participating practices who had a BMI ≥ 30 kg/m^2^ or a BMI ≥ 28 kg/m^2^ with a co-morbidity and who were assessed as being motivated to lose weight were referred to the Counterweight Programme.
Eighty community pharmacies were approached to determine interest in delivering the Counterweight Programme. Twenty three pharmacies expressed an interest, and 18 were invited because they were located in geographical areas where local general practices did not deliver Counterweight. Sixteen pharmacies subsequently agreed to deliver the programme and received training. Twelve of the participating pharmacies where situated in small urban settlements \[[@B17]\] with between 10,000 to 125,000 inhabitants. The remaining four were located in small towns of 3,000--10,000 people. Participating pharmacies were required to have a private consultation room and time to deliver the intervention. All pharmacies had extended opening hours and offered appointments in the evening and at weekends. Marsden High Capacity Portable Scales (Class III) and Seca Leicester Portable Height Measures were supplied by the NHS Fife Keep Well project to measure weight and height. Scales were calibrated on an annual basis \[[@B18]\]. Counterweight resources (training manuals, desk top flip charts and patient information booklets) were initially funded through the core Counterweight Scottish Government funding. Pharmacies were paid a single commitment fee of £100 for taking part, plus a payment per patient and payments for the provision of replacement staff while staff were trained . Between March 2009 and May 2010, the payment per patient was £54, which comprised £30 for patients attending 1-3 appointments and a further £24 for patients attending 4 or more appointments. From May 2010 to date these payments rose to £64 and £40, respectively.
Specialist dietitians competent in Counterweight Programme delivery conducted two four-hour training sessions and a further 3 hour session after 6 months to consolidate the initial training. Most trained staff were pharmacy assistants rather than pharmacists. It was agreed that pharmacy staff would not sell over-the-counter weight loss medications to patients enrolled in the programme. The specialist dietitians also provided mentoring to all pharmacies.
The Counterweight approach to weight management has been described in detail elsewhere \[[@B16]\]. In brief, pharmacy staff delivered patient education by discussing weight management, and communicating information on behaviour change strategies. The initial interventions involved a prescribed eating plan or a goal-setting approach. The aim was to achieve an energy deficit of 500-600 kcal/day. As patients progressed through the programme, emphasis was increasingly directed to weight loss maintenance and the prevention of weight regain. Patients were asked to commit to nine appointments in 12 months following the initial screening visit. This included six initial appointments (10--30 minutes each) with follow-up visits at 6, 9 and 12 months. The total time for one patient to be taken through the full programme was estimated at 130 minutes.
The data collected at each visit were recorded using paper based patient forms. Anonymised patient forms were collated centrally and entered into a bespoke Microsoft Access database. The data were checked for incomplete or inaccurate information. The central database was sent to an independent team at Glasgow University at set time points.
Ethics
------
Formal ethical approval was unnecessary as this was an audit of a planned delivery of an existing intervention, no new or untested treatment was being offered, and there was no experimentation. No personally identifiable data were collected and written consent was not required.
Data definitions
----------------
Baseline weight and height were collected when patients attended the first session of the weight loss programme. During the first appointment patients were asked whether they smoked and whether they had diabetes. Their response was recorded as a binary variable (yes/no). Smoking and diabetes status were recorded because they are relevant factors associated with weight, attendance and weight loss. Weight was measured at each subsequent visit. Weight change was evaluated at 3, 6 and 12 months. The weight measurements used in the evaluation were recorded in kilograms at dates closest to 3, 6 and 12 calendar months from enrolment (within time frames of 6--15 weeks, 15 weeks--9 months, and 9 -- 18 months respectively). The primary outcome of the study was weight loss of at least 5% of baseline weight at 12 months.
Statistical methods
-------------------
Descriptive statistical methods were used to present change in weight at 3, 6 and 12 months. Values are presented as means and 95% confidence intervals (95% CI), if not indicated otherwise. We present weight change as absolute weight change and the percentage achieving at least 5% weight loss. The analyses were carried out separately for patients who attended at each time point, and for all patients assuming that participants for whom weight at follow-up was not available retained their baseline weight (baseline-observation-carried-forward BOCF), and assuming participants retained their last observed weight (last-observation-carried-forward LOCF). BOCF and LOCF were included to enable comparison with other studies, but they are biased methods of imputing missing data \[[@B19]\].
Kruskal-Wallis one way analysis of variance, the chi-square test for differences in proportions and logistic regression were used to examine the association of age, sex, and starting BMI with weight loss and attendance. Age and BMI were employed both as continuous and categorical variables. Age was categorised into 3 groups (\<50, 50-59, and 60+ years). BMI was categorised as follows \<30, 30- \< 35, 35- \< 40, and 40+ kg/m^2^. The conventional statistical significance threshold of 5% was used (p \< 0.05). All analyses were conducted with STATA version 11 (StataCorp, CollegeStation, TX, USA).
Results
=======
Between March 2009 and July 2012, 458 patients were enrolled by 16 community pharmacies. The baseline characteristics of patients are shown in Table [1](#T1){ref-type="table"}. Sex, age and BMI were not recorded for 2 (0.4%), 12 (2.6%) and 6 (1.3%) patients respectively. Three-quarters of patients were women, mean age was 54 years and mean BMI was 36.0 kg/m^2^. One-fifth of patients had a BMI of 40 kg/m^2^ or over. Smoking was reported among 14% of patients. Patients who smoked had a slightly lower mean BMI 34.3 kg/m^2^ (95% CI 33.0, 35.6) than those who reported that they did not smoke 36.6 kg/m^2^ (35.9, 37.3) (p = 0.005). One in ten patients reported diabetes. Patients with diabetes had a higher baseline BMI 39.1 kg/m^2^ (37.0, 41.2) than patients who did not have diabetes 35.8 kg/m^2^ (35.2, 36.4) (p = 0.002).
######
Baseline characteristics of 458 Counterweight patients enrolled in community pharmacies
**Patients (n = 458)**
------------------------------------ -------------- ------ ---------
**Men**% 24.9 \(114\)
**Women**% 74.7 \(342\)
**Age** mean years (SD) 54.0 (7.4)
**Weight** mean kg (SD) 96.4 (18.3)
**Starting BMI** mean kg/m^2^ (SD) 36.0 (5.9)
**Starting BMI** kg/m^2^% \<30 9.8 \(45\)
30-34 43.9 \(201\)
35-39 23.8 \(109\)
≥40 21.2 \(97\)
not recorded 1.3 \(6\)
Smoking status% smoker 14.4 \(66\)
not recorded 18.8 \(86\)
Diabetes Status% diabetic 11.6 \(53\)
not recorded 15.7 \(72\)
Attendance and weight change at 3, 6 and 12 months are shown in Table [2](#T2){ref-type="table"}. Of the 458 patients who started Counterweight within the enrolment period, progressively fewer had been in the programme long enough to be eligible to attend at later time points. Attendance declined over time from 56.0% at 3 months to 24.5% at 12 months. Of 314 patients enrolled for at least 12 months, 32 (10.2%) had achieved the target weight loss of ≥5%. This represents 41.6% of those who attended at 12 months.
######
Weight loss among patients at 3, 6, and 12 months since starting programme
**Time since starting programme**
------------------------------------ -------------------- ----------------------------------- ------------------- -------------------
**Attendance**% (n) 56.0 (241) 33.7 (133) 24.5 (77)
**Weight loss** mean kg (95% CI\*) Attending patients 2.4 (2.02, 2.70) 3.5 (2.66, 4.25) 4.1 (2.83, 5.41)
BOCF\*\* 1.3 (1.10, 1.54) 1.2 (0.85, 1.58) 1.0 (0.64, 1.38)
LOCF\*\*\* 1.3 (1.10, 1.54) 1.6 (1.25, 1.89) 1.7 (1.31, 2.14)
**≥5% weight loss**% (95% CI\*) Attending patients 17.0 (12.5, 22.4) 34.6 (26.6, 43.3) 41.6 (30.4, 53.4)
BOCF\*\* 9.5 (6.9, 12.7) 11.6 (8.7, 15.2) 10.2 (7.1, 14.1)
LOCF\*\*\* 9.5 (6.9, 12.7) 13.9 (10.7, 17.7) 15.9 (12.1, 20.4)
\*95% confidence interval.
\*\*BOFC - baseline observation carried forward.
\*\*\*LOFC - last observation carried forward.
The distribution of weight change at 3, 6 and 12 months is illustrated in Figure [1](#F1){ref-type="fig"}. Weight change was approximately symmetrically distributed around the mean at 3 and 6 months. At 12 months there was stronger evidence of skew in the distribution; for example 2 patients lost more than 20 kgs. There were progressively larger mean weight losses over time, from 2.4 kg at 3 months to 4.1 kg at 12 months. At 12 months, 57 patients (74% of patients who attended, 18% of all patients) had lost some weight, 15 patients (19% of patients who attended, 5% of all patients) had gained weight and 5 (6% of patients who attended, 2% of all patients) had no appreciable change in weight since baseline (absolute change ≤ 250 g). The maximum weight loss was 27 kg and the maximum weight gain 4.6 kg at 12 months. Weight change, expressed as a percentage of baseline weight, was similar to absolute weight change because the mean baseline weight was close to 100 kg (data not shown).
{#F1}
A higher percentage of men than women attended at 12 months (Table [3](#T3){ref-type="table"}). Attendance increased with age and decreased with increasing BMI but these trends were not statistically significant. Men appeared to lose more weight than women (5.8 kg vs. 3.4 kg; p = 0.66 - Table [3](#T3){ref-type="table"}) but there was no difference in clinically significant weight loss (39% vs. 43%; p = 0.78). A relationship between weight loss and age or BMI was not apparent. Table [3](#T3){ref-type="table"} shows that patients aged 40-49 lost more weight than other age groups while those with a BMI 30-34 kg/m^2^ lost the least amount of weight. Statistically significant differences were not found when weight loss was modelled by sex, age and BMI individually (sex p = 0.66; age p = 0.66; BMI p = 0.21 - Table [3](#T3){ref-type="table"}) or in combination. The percentage achieving ≥5% weight loss similarly did not show statistically significant associations with sex (p = 0.78), age (p = 0.86) or BMI (p = 0.86). Table [3](#T3){ref-type="table"} shows that the imputed measures of mean weight loss (BOCF and LOCF) produced substantially lower estimates of mean weight loss than estimates based only on patients who attended. Both BOCF and LOFC estimates similarly showed non-significant patterns by age and BMI.
######
Weight change and percent of patients achieving \>5% weight loss at 12 months by sex, age and Body Mass Index
**Number of patients** **Percentage (95% CI\*) of patients who attended** **Mean (95% CI\*) weight loss (kg)** **Percentage (95% CI\*) of patients losing ≥5% of baseline weight**
------------------------------------ ------------------------ ---------------------------------------------------- -------------------------------------- --------------------------------------------------------------------- ------------------- ------------------- ------------------ -------------------
**Sex**
Men 79 29.1 (19.4, 40.4) 5.79 (2.47, 9.10) 1.69 (0.58, 2.79) 2.53 (1.41, 3.56) 39.1 (19.7, 61.5) 9.8 (6.3, 14.4) 17.7 (10.0, 27.9)
Women 234 23.1 (1.78, 29.0) 3.41 (2.25, 4.57) 0.79 (0.47, 1.11) 1.46 (1.05, 1.86) 42.6 (29.2, 56.8) 11.4 (5.3, 20.5) 15.4 (11.0, 20.7)
*p value* *0.28* *0.66* *0.62* *0.21* *0.78* *0.69* *0.62*
**Age group (years)**
40-49 61 19.7 (10.6, 31.8) 4.78 (1.50, 8.06) 0.94 (0.16, 1.72) 1.49 (0.56, 2.43) 50.0 (21.1, 78.9) 9.8 (3.7, 20.2) 16.4 (8.2, 28.1)
50-59 134 24.6 (17.6, 32.8) 3.50 (1.40, 5.59) 0.86 (0.30, 1.43) 1.66 (1.01, 2.32) 39.4 (22.9, 57.9) 9.7 (5.3, 16.0) 15.7 (10.0, 23.0)
60+ 111 27.9 (19.8, 37.2) 4.29 (2.36, 6.22) 1.20 (0.56, 1.84) 1.91 (1.22, 2.61) 38.7 (21.8, 57.8) 10.8 (5.7, 18.1) 16.2 (9.9, 24.4)
*p value* *0.49* *0.66* *0.77* *0.57* *0.78* *0.96* *0.99*
**BMI grouping (kg/m**^**2**^**)**
\<30 30 26.7 (12.3, 45.9) 5.37 (2.35, 8.38) 1.43 (0.24, 2.63) 2.02 (0.78, 3.25) 75.0 (34.9, 96.8) 20.0 (7.7, 38.6) 23.3 (9.9, 42.3)
30- \< 35 136 25.7 (18.6, 33.9) 2.61 (1.35, 3.88) 0.67 (0.30, 1.05) 1.40 (0.89, 1.91) 37.1 (21.5, 55.1) 9.6 (5.2, 15.8) 16.2 (10.4, 23.5)
35- \< 40 80 23.8 (14.9, 34.6) 3.82 (1.09, 6.55) 0.91 (0.17, 1.64) 1.66 (0.84, 2.48) 31.6 (12.6, 56.6) 7.5 (2.8, 15.6) 13.8 (7.1, 23.3)
40+ 64 23.4 (13.8, 35.7) 7.35 (3.08, 11.63) 1.72 (0.47, 2.98) 2.39 (1.08, 3.70) 46.7 (21.3, 73.4) 10.9 (4.5, 21.2) 15.6 (7.8, 26.9)
*p value* *0.97* *0.21* *0.74* *0.91* *0.18* *0.28* *0.68*
\*95% confidence interval. \*\*BOFC - baseline observation carried forward. \*\*\*LOFC - last observation carried forward.
Patients who smoked had similar weight loss as patients who did not smoke (Table [4](#T4){ref-type="table"}). Patients with diabetes who attended at 12 months appeared to lose less weight compared to patients without diabetes (1.8 kg vs. 4.6 kg; p = 0.24 - Table [4](#T4){ref-type="table"}), but a difference was not apparent when the proportions (LOCF) with clinically significant weight loss were compared (15% vs. 17%; p = 0.78). Patients who did not report smoking or diabetes status appeared to be less likely to attend at 12 months (p = 0.07 & p = 0.08 respectively - Table [4](#T4){ref-type="table"}).
######
Weight change and percent of patients achieving \>5% weight loss at 12 months by smoking and diabetes status
**Number of patients** **% attending** **Mean 12 month weight loss (kg) (95% CI\*)** **% losing 5% weight (BOCF\*\*) (95% CI\*)** **% losing 5% weight (LOCF\*\*\*) (95% CI\*)**
--------------------- -------------- ------------------------ ----------------- ----------------------------------------------- ---------------------------------------------- ------------------------------------------------
**Smoking status** smokers 45 28.9 4.47 (0.60, 8.35) 11 (4, 24) 13 (5, 27)
non smokers 210 26.2 4.19 (2.60, 5.79) 11 (7, 16) 17 (12, 22)
not recorded 59 15.3 3.16 (0.69, 5.63) 7 (2, 16) 15 (7, 27)
*p value* *0.17* *0.99* *0.63* *0.85*
**Diabetes status** diabetes 33 21.2 1.78 (-0.80, 4.34) 3 (0, 16) 15 (5, 32)
no diabetes 228 27.2 4.57 (3.02, 6.13) 12 (8, 17) 17 (12, 23)
not recorded 53 15.1 2.66 (0.12, 5.12) 6 (1, 16) 11 (4, 23)
*p value* *0.16* *0.36* *0.13* *0.58*
\*95% confidence interval.
\*\*BOCF - baseline observation carried forward.
\*\*\*LOFC - last observation carried forward.
Discussion
==========
Our study reports on the largest prospective evaluation of a pharmacy based weight management programme in the UK over a twelve month follow-up period. We found that 10% of patients enrolled in a community pharmacy-based weight management programme had lost ≥5% of baseline weight after 12 months. Measurement of successful weight loss maintenance at 12 months requires both initial weight loss and continued attendance and we found that these did not differ significantly by age, sex or baseline BMI. Attendance was highest among men, and appeared to increase with age and decrease with increasing BMI. Weight loss -- measurable only among attendees -- was greatest in men and patients under 50 but showed no clear relationship to BMI.
Jolly's randomised controlled trial, Lighten Up, reported mean weight losses of 1.19 kg (95% CI -0.7 to 3.1), corresponding to 14.3% of the cohort (using BOCF) losing ≥5% at one year among users of a pharmacy-based weight management programme \[[@B12]\]. Weight losses in our pharmacy trial were similar, with mean loss at 12 months of 1.01 kg (0.64, 1.38) (BOCF) but the proportion achieving a ≥5% loss was lower, at 10.2%. Lighten Up pharmacy patients had lower baseline BMIs than our pharmacy group (96% vs. 78% respectively with BMI \< 40), they were younger (mean age 49 vs. 54 years) but a similar proportion were men (27% vs. 25%). As we did not find significant effects of either BMI or age on weight loss, it seems unlikely that these explain the observed difference in outcomes. However, 20% of weights reported in the Lighten Up pharmacy group were self-reported and weights were available for 57% of patients, compared to our 25%. It would seem that the community pharmacies delivering the Counterweight Programme were poorer at retaining patients but more effective in achieving clinically meaningful weight losses among those who attended. Assessment of readiness to change, for example the stages of change model by Prochaska \[[@B20]\], is widely used in health promotion interventions and is part of the Counterweight Programme. Such information was not collected as part of routine data recording in our study. Improved screening with regular review of motivation may improve retention and efficacy of the programme. Among the three pharmacy studies identified by Gordon and others that reported 12-month outcomes \[[@B13]\], one reported mean weight losses of up to 2.4 kg (2.7%) with the addition of high risk counselling \[[@B21]\]; another reported 1.9 kg mean weight loss \[[@B22]\]; and a third reported mean weight loss of 4.1 kg \[[@B23],[@B24]\]. Toubro and others' study \[[@B23]\], however, used baseline and subsequent self-reported weights only, and is possibly affected by reporting bias.
Evidence-based weight loss programmes for adults in the UK report 12 month mean weight loss ranging from 1.1 to 6.6 kg, and they achieve 5% weight loss for between 14% to 46% of patients \[[@B25]\]. Commercial community-based organizations such as Weight-Watchers appear to be more effective \[[@B26]\], but a direct comparisons between programmes is difficult because of differences in the case-mix of patients they serve and the context in which the programmes are delivered. All programmes suffer from high attrition which limits the ability to compare outcome data. Effective ways to increase retention and attendance are needed, and this may improve weight loss outcomes by increasing the time spent in programme participation. Weight loss programmes based within community pharmacies have the attraction of being widely accessible which may increase participation. Our study, however, found that when the Counterweight Programme was based in pharmacies, attendance at 12 months was comparable to that achieved when the programme was delivered in general practice (25% vs. 28%) \[[@B27]\].
Our study has several strengths and weaknesses. Its strengths are that it reports weight loss using a clinically-based threshold of ≥5% rather than a mean weight change in a patient group; weights were objectively measured and not self-reported; and that it describes long-term weight loss rather than end-of-programme results. Weaknesses of the study include possible unrepresentativeness of the patients or pharmacies and a lack of detailed information about other social and clinical factors that may have influenced patients' attendance and weight loss. The Keep Well population, for example, is composed of patients from disadvantaged areas, identified at being at high risk of ill health, and who are not fully engaged with primary care services. The present study was not an RCT and we did not employ a comparison group. It is difficult to recruit (particularly to a control group) and conduct such a study when pharmacies were the main delivery point of the weight management programme in localities where the study population was composed mainly of patients from disadvantaged backgrounds.
In established practice the Counterweight Programme delivered in primary care achieved ≥5% weight loss at 12 months in 10% of patients \[[@B27]\] and this Counterweight Programme delivered in pharmacies also achieved the same weight loss in 10% of patients. Our study indicates the effectiveness of a programme delivered in areas where GPs would not provide Counterweight services and we therefore suggest that pharmacy-delivered weight management remains an option that should be considered where alternatives are not available.
Conclusions
===========
In conclusion, we have demonstrated that a pharmacy-based weight management programme achieves clinically significant, objectively-measured weight losses at 12 months in 10% of patients who enrol. There are few other evaluations of long-term weight loss outcomes in community pharmacies and several include self-reported weights, which may be subject to significant reporting biases. The Counterweight Programme delivered in pharmacies should be considered as part of a range of services available to a community to manage overweight and obesity.
Competing interests
===================
LM and NB are employees and shareholders of Counterweight Ltd. The other authors have declared no competing interests.
Authors\' contributions
=======================
DM & PM were responsible for the statistical analyses and drafting and writing the manuscript. AS, JG, LM and NB arranged and coordinated pharmacy involvement, data acquisition and contributed to the drafting of the paper. All authors read and approved the final manuscript.
Pre-publication history
=======================
The pre-publication history for this paper can be accessed here:
<http://www.biomedcentral.com/1471-2458/13/282/prepub>
Acknowledgements
================
The intervention was conducted during the Scottish Government Health Department funding of the Counterweight weight management programme in primary care. The pharmacy delivery of the Counterweight Programme was funded through the NHS Fife Keep Well project. The manuscript was produced on behalf of the Counterweight Research Group: J Iain Broom, Nick Finer, Gary S Frost, David W Haslam, Sudhesh Kumar, Michael EJ Lean, E Louise McCombie, Philip McLoone, David S Morrison, John PD Reckless, Hazel M Ross and Billy Sloan who contributed to the study design and commented on drafts of the manuscript
| {
"pile_set_name": "PubMed Central"
} |
Subsets and Splits