Ethics statement

This study utilized data from previous fish exposure experiments. All studies that contributed data to the current analysis were carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals by the National Institutes of Health. The exposure protocols for all studies included in this meta-analysis were approved by the Institutional Animal Care and Use Committees of St. Cloud State University. Fish were deeply anesthetized (complete loss of response to manual stimulation) in a neutral buffered solution of MS-222 prior to blood collection and were sacrificed by subsequent cervical dislocation. All efforts were made to minimize distress to the animals.

Study conditions

A meta-analysis integrating multiple previous studies (Table 1; S1 Table) was conducted to explore the heterogeneity of male fathead minnow populations subjected to estrogen or SSRI exposures. All studies were carried out under common exposure conditions and had similarly assessed biological endpoints including three morphometric indices: hepatosomatic index (HSI = liver weight/whole body weight), gonadosomatic index (GSI = gonadal weight/whole body weight), and body condition factor (BCF = (whole body weight/length^3) *100,000).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Studies contributing data to the current analysis.

https://doi.org/10.1371/journal.pone.0186807.t001

Exposure

The test organisms in all included studies were 6-month old mature male fathead minnows from the same laboratory fish supplier (Environmental Testing and Consulting laboratory, Superior, WI). Upon arrival at the St. Cloud State University Aquatic Toxicology Laboratory (St. Cloud, MN), all fish were subjected to a 21-day in-house flow-through exposure at constant environmental conditions; including similar photo-period (16:8 light:dark), temperature (20–23°C, S1 Fig), dissolved oxygen (5.0–7.5 mg/L), and pH (7.2–8.3) [16,18,31,34–39]. Exposure experiments followed published flow-through exposure protocols [16]. Briefly, male fathead minnows were randomly assigned to control or exposure aquaria at an approximate density of 10 fish per 16L tank. Fish were maintained according to U.S. Environmental Protection Agency guidelines [40] throughout the experiments, and fed frozen brine shrimp (Artemia fransiscana, San Francisco Bay Brand, Inc., Newark, CA) twice daily ad libitum.

Chemicals

Four treatment conditions were examined in the present study: an ethanol carrier control, E1, E2 and SSRIs. The SSRI treatment group contained the data from the exposures to sertraline, fluoxetine and venlafaxine (see Table 1 for additional treatment-specific information). As a condition for the meta-analysis, all concentrations for a treatment condition (for example, E1) were combined. Similarly, all SSRI pharmaceuticals were combined as one SSRI treatment. This approach is commonly applied in meta-analyses, and is justified as the current study attempted to assess whether social status modulates fish responses to a CEC exposure, not the biological effects of the exposure. All data used in this analysis were extracted from the original “raw” data collected at the time of the original studies, and are included in a supplemental file (S1 Table).

Biological endpoints

At the end of each experiment, a series of endpoints were documented, including whole body wet weight, length, liver wet weight, and testis wet weight. These values were used to calculate BCF, HSI, and GSI. The SSCs were evaluated based on the blind scoring system described by [4,41]. The prominence of nuptial tubercles, dorsal pad and the banding pattern were visually evaluated and scored on a 0–3 scale, where 0 indicates no expression, 1 minor presence, 2 clear presence, and 3 greatest prominence of the respective SSC. A SSC is represented as the sum of individual scores. For the purpose of this analysis, dominant males are considered as those with mean SSC scores of ≥ 2 and subordinate males < 2. Plasma VTG concentrations were measured via a competitive antibody-capture enzyme-linked immunosorbent assays. The VTG concentrations were log-transformed for statistical analyses.

Data organization

Data points from the studies were organized into tables based on treatment conditions (see S1 Table). Data were tested for the presence of outliers based on Mahalanobis distances (JMP Pro 11 for Macintosh). Outliers (Mahalanobis distance > 3.068) were removed from the subsequent analyses.

Statistical analyses

Differences in water temperature can have extensive physiological implications for ectothermic vertebrates [42,43]. To assure a high degree of similarity among the environmental conditions evaluated in all studies for which data were available, we tested for similarities in water temperature using one-way ANOVA. Ethanol carrier control data were used for model construction. The distribution profile, a representation of distribution of the single continuous variable using histograms [44], was first run for each variable separately to explore patterns (JMP Pro 11 for Macintosh). The normality test was then performed on SSC and log(VTG) variables to confirm the observed distribution patterns. Since raw data from all studies were available for analyses, weighted averages, a common approach of meta-analysis when original raw data are not available, were not considered necessary. The use of raw data strengthens the overall power of an analysis [33]. Next, a principle component analysis (PCA) was conducted on all continuous variables to reduce data dimensionality and minimize information loss (JMP Pro 11 for Mac software) [45,46]. Given the number of biological variables measured, PCA provides a systematic approach reduce the data dimensions and avoid co-linearity while retaining most of the variability in the data matrix. PCA converts all quantitative variables into uncorrelated weighted linear combinations or principal components (PCs) to reduce the data dimensions by using the most representative PCs. PCA on correlation matrix instead of covariance matrix was used for this study because not all variables are of the same scale and/or shape. The number of retained principle components (PCs) was determined based on the following criteria: eigenvalues > 1 and/or overall variation explained > 80% [45]. Five PCs were chosen for K-Means clustering to assess the natural separation of data points (JMP Pro 11 for Mac software) [46,47]. K-Means clustering technique is useful when working with large data sets. The optimal number of clusters is represented by the cubic cluster criterion (CCC) value; the more positive the CCC, the greater the cluster separation. The number of clusters to retain was determined based on either highest positive or lowest negative CCC. A MANOVA was chosen to evaluate the relationship between clusters and the variables of interest as there were more than one of each dependent and independent variable present in the data matrix. To evaluate the obtained clusters, the MANOVA was run as a follow-up test, where retained PCs (Y1, Y2, Y3, Y4, Y5) represent independent variables, and PC-based clusters the dependent variables (X1 + X2) (JMP Pro 11 for Mac software) [48]. A MANOVA was first run on clusters and PCs, then on clusters and the variables of interest: SSC and log(VTG). The same steps were then repeated for the other three treatments: E1, E2 and SSRIs. For all analyses significance level was set at α = 0.05.

Social hierarchy under control conditions

The PCA yielded five PCs (first five eigenvalues > 1, which together explain 81% of variability). The K-Means cluster analysis run on five PCs yielded two clusters (CCC = -3.2424) as an optimal solution, which supports our hypothesis of the presence of more than one distinct population of male fish within a treatment group (Fig 5A). A MANOVA of the two clusters (dependent variables) versus five PCs (independent variables) showed that clusters are significantly different from each other while already accounting for 81% of variability (p-value < 0.0001). Another MANOVA output resulted in two clusters being significantly different relative to SSC and logVTG variables (p-value < 0.0001). The cluster separation relative to SSC was from 5.71 to 5.76, whereas that relative to logVTG ranged from 0.86 to 2.14 (Fig 5A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Data separation.

Naturally occurring data separation from single sex male exposures was observed with an aid of K-Means cluster analysis performed following PCA. Five PCs were retained based on the number of eigenvalues >1 and minimum percentage of variability explained by these PCs and used for clustering. The most representative variables with contribution percentage driving the PCs are shown on 3D biplots. The treatment groups are: (A) Control, (B) E1, (C) E2 and (D) SSRI.

https://doi.org/10.1371/journal.pone.0186807.g005

Social hierarchy under E1 treatment

As with the control condition, five PCs were retained for the E1 treatment (five eigenvalues > 1 and 85% of variability explained). The retention of five PCs again resulted in data separation into two groups (Fig 5B). While two clusters were significantly different from each other (p-value < 0.0001, MANOVA), clusters were not significantly different relative to SSC and logVTG variables (p-value = 0.1748). The mean SSC values decreased to 4.33 and 4.86 relative to those of control, whereas logVTG values remained within 1.76 and 1.81 range (Fig 5B). We observed an increase in VTG synthesis, but to a lesser degree as compared to the E2 treatment.

Social hierarchy under E2 treatment

Similar to the other treatments, five PCs were retained for the E2 treatment (5^th eigenvalue = 0.8924 and 87% of variation is explained). The inclusion of five PCs into a K-Means cluster analysis yielded two distinct populations, similar to the E1 treatment (Fig 5C). A MANOVA on two clusters versus five PCs showed that clusters were significantly different (p-value < 0.0264) (Fig 5C). However, they were again not significantly different relative to SSC and logVTG variables (p-value = 0.1236). The cluster mean SSC values decreased to 4.8 and 5.37 respectively relative to the control, while logVTG values increased to 3.15 and 2.88.

Social hierarchy under SSRI treatment

Five PCs from the SSRI treatment explained 87% of variation while fifth eigenvalue was at 0.92. Two clusters were identified from clustering of retained PCs (Fig 5D). Clusters by PCs (p-value < 0.0001) and clusters by SSC and logVTG were significantly different according to MANOVA (p-value = 0.0009) (Fig 5D). The separation of cluster mean SSC increased to 5.21 and 6.5 relative to those of control, while logVTG values of 0.70 and 2.17 remained within the range of the Control treatment.

Impact of estrogenic exposure on social hierarchy

Based on the findings from the cluster analysis and MANOVA, exposure to estrogenic compounds E1 and E2 reduced the variability within the male fathead minnow population by constraining the socially distinct groups to a tighter range relative to the control group, as evidenced by SSC cluster means of 4.33 and 4.86 for E1 treatment; and 4.8 and 5.37 for E2 treatment. In other words, the dominance in male fish was suppressed, as indicated by the expression of SSC, shifting males towards the subordinate spectrum. Such a linkage between estrogenic exposures and social status changes may involve a variety of molecular pathways with complex interactions and regulatory controls (Fig 1). Estrogens could initiate their effects through interactions with estrogen receptors (ERs) in the nucleus [49] or through alternative signaling pathways involving a rapid, non-genomic route initiated by membrane-bound ERs [50]. For example, transcriptional activation of the ER gene at low estrogen concentrations produce more ERs, which initiate the transcription of the VTG gene [51]. The primary location of ER up-regulation upon exposure to exogenous estrogenic compounds is the liver, where VTG synthesis takes place [29,52]. Consequently, increased VTG biosynthesis is correlated with reduced SSCs (Fig 1).

The constraint of social hierarchy among males exposed to both E1 and E2 can be, in part, explained by previous observations that male fish exposed to exogenous estrogens suppress 11-Ketotestosterone production, a primary androgen in teleost fish associated with dominance [23,28]. It is plausible that the reduction in androgen synthesis in male fish results in the downgrading of social status [23,28].

Alternatively, the shift towards subordinance may involve perturbations in the HPG axis. It is known that estrogens modulate the activity of the HPG axis via positive/negative feedback [9,53], which continuously informs the brain of the physiological status of distantly located organs [54]. Exposure to exogenous estrogenic compounds interferes with the stimulatory effects of γ-aminobutyric acid (GABA) that either acts on the hypothalamus to induce the release of GnRH [55], or acts directly on the pituitary gland to induce the release of luteinizing hormone (LH) and follicle stimulating hormone (FSH) [6,54]. Previous studies showed that high levels of estrogens, but not testosterone [56], block the stimulatory effect of GABA and ultimately lead to a reduction of androgens for males [52]. In addition, estrogens inhibit the expression of the ar gene that codes for androgen receptors in testis [57], leading to the reduced production of male sex steroids [28]. The reduction in male sex hormone production in either case affects the reproductive status of males [29,52], as manifested in the forms of suppressed aggressive behavior, impaired ability to acquire a nest site under competitive pressure [30], and less prominent SSCs [9]; all consistent with observations in the current study.

Impact of SSRI exposure on social hierarchy

In contrast to the constraining effects of estrogenic exposure, the exposure to SSRIs relaxed social hierarchy within the male fish population by pushing two socially distinct groups of males further apart (Fig 5C) as evidenced by SSC cluster means of 5.2 and 6.5 for subordinate and dominant groups, respectively. It has previously been established that social stress from aggressive interactions is expressed differently in the brain regions of dominant and subordinate males [17]. The effects of SSRIs are propagated through their complex interactions with serotonin, which interacts with the HPG and HPA axes [18,20,21]. The HPA axis is involved in mediating stressful conditions while altering the serotonin system that is conserved among vertebrates [20]. Studies on teleost fish have also shown an increase in serotonin activity in socially subordinate individuals while inducing the release of cortisol [20,21]. Persistently high levels of cortisol are associated with reduced abilities of an individual to access food, leading to decreased body mass and condition factor, and a chronically activated HPA axis which will further raise cortisol level [7]. Thus chronic exposure to SSRIs elevates serotonin concentrations in the synaptic cleft and prolongs the effects of serotonin, which mimics either a stressful event or a social subordination.

Serotonin, associated with the display of aggressive behavior [17] in male fish, also influences the critical stages of their reproductive development [18]. It has a close association with the HPG axis by inducing LH and FSH release upon stimulation of the pituitary gland by serotonin [17]. Thus, exposures to a SSRI can indirectly affect spermatogenesis in fish by stimulating the release of LH from the pituitary gland [53] through serotonin₂ receptor [19]. Although serotonin has a stimulatory effect on LH hormones in vertebrates that in turn influence the production of testosterone and consequently spermatogenesis [18], it shows mixed effects between vertebrates on intraspecific aggression. Serotonin exhibits an inverse relationship with aggression in some lizard species [58] whereas studies on invertebrates, crustaceans, resulted in an increase in aggression (Pyle, pers. com.). Johnson et al. [59] confirmed that the serotonin binding receptor mediates aggressive behavior. These authors found that 5-HT₂ receptor activation reduces the aggression in male fish [59], the same receptor that induces the release of LH from the pituitary gland [6,18]. In contrast, the binding of serotonin to the 5-HT_1A receptor induced aggression [59]. Moreover, the increased metabolism and release of serotonin in teleost fish typically occur in response to social stress, the degrees of which differ in extent and timing between the individuals of different social statuses [17,22].

Consequences of social hierarchy for toxicological studies

The establishment of socially complex communities has been documented for many fishes, including the fathead minnow [3], and is driven by changes in the HPG axis. The dominance of males is determined by the concentration of circulating male sex steroids that ultimately lead to enhanced SSC expression [9,60]. This is consistent with observations that some fish became dominant while others remained subordinate in all of the pooled studies, as indicated by the varying expression of SSCs. It is important to note that, according to our current knowledge, social status is not genetically determined, but rather regulated at the level of the endocrine state of a male. The genetic contributions to social status, if any, remain to be identified. Thus, dominant males are not limited to those genetically predisposed to achieve the highest level in a social organization [61].

Toxicologists designing exposure experiments with single-sex test organisms should carefully evaluate the implications of social hierarchy. In experiments in which exposure compounds are likely to homogenize the population by suppressing dominance, social hierarchy may have little effect on the statistical power of the study design. This appears to be the case for the estrogenic CEC assessed in the current study and likely for other CECs that reduce the overall social hierarchy. In contrast, studies examining the effects of CECs that liberate social constraints, such as the SSRIs examined in the current study, may undermine the statistical power of the experimental design. If possible within the constraints of the hypothesis being tested, experiments that separate individuals or create breeding pairs may be advantageous to single-sex exposure experiments as they avoid the development of a social hierarchy during the experimentation. However, tests subjects at the onset of the study may already represent varying dominance based on their prior housing in larger breeding groups, a common holding approach in toxicological testing facilities. In those instances, recording of the social status at the beginning of each experiment and equal distribution of dominant and subordinate test subjects among treatments may reduce the effect of existing social hierarchy. The current study focused on males of a species exhibiting sexual dimorphism that is known to establish social hierarchies [3,4]. However, female fish exposed to androgenic CECs may also develop social hierarchies as some females become masculinized. It is therefore prudent to assess any experimental design using hierarchy-forming species in light of the impacts of social hierarchies.

In summary, we demonstrated in the current study that male fathead minnows separate into physiologically distinct subpopulations within treatment groups. We showed that while social hierarchy is present in all social animals, the dominance of individuals can be affected depending on their initial status and the class of contaminants they are exposed to. These findings suggest that treating a fish population as homogeneous may lead to an increased chance of Type II error (false negative), which implies that the effects of some endocrine-disrupting compounds may have been underestimated.