A total of 216 wild-type zebrafish (Danio rerio), with a heterogeneous genetic background and an average size of 3.5 cm, were used in this study. The fish were purchased from Carolina Biological Supply Co. (Burlington, NC, USA). Following standard recommendations (Avdesh et al., 2012; Aleström et al., 2020), we housed the fish in a 615 L vivarium [180 cm (length) × 60 cm (height) × 87 cm (width)] at the stocking density of ~3 fish/L. The vivarium was equipped with two canister filters; one Fluval FX6 (Fluval, Hagen Inc, Montreal, Quebec, Canada) and another one Penn-Plax Cascade (Penn-Plax, Hauppauge, NY, USA). Water parameters of the holding tank were regularly checked, and temperature and pH were maintained at 27°C and 7.2, respectively. Regular tap water was used with the addition of a stress coat (AquaSafe plus, Tetra, Spectrum Brands Inc., Sulzbach, Germany) to remove chlorine and chloramines. We also added API Quick Start freshwater and saltwater aquarium Nitrifying bacteria (API, Mars Inc., McLean, Virginia, USA) to enhance the initial cycling of the tank. We kept male and female subjects separated using a transparent perforated separator placed in the middle of the tank. Such a separation allowed the identification of male and female subjects to be used in the experiments. This procedure was aimed at maintaining the experimental population balanced across sex in all experimental conditions. Fish were maintained on 12 h light/12 h dark photo-period and they were fed with commercial flake food once a day at 7 PM. Prior to the beginning of the experiments, fish were acclimatized in the holding facility for approximately 1 month.

The experimental set-up consisted of a rectangular tank, a video camera (Logitech C910 HD Pro Webcam, Logitech, Switzerland) located in front of the tank, an array of lights, and black curtains surrounding the test area to minimize potential disturbance from cues outside the experimental tank. An experimental test section of 75 cm × 10 cm × 35 cm (length, width, and height) was arranged inside the tank using corrugated white plastic. The tank width and depth were selected to promote swimming along the water column. Such decision rested upon the hypothesis that SSRIs generally result in an overt alteration of geotaxis (Sackerman et al., 2010; Macrì et al., 2019; Clèment et al., 2020); this strategy has been borrowed from available literature (Cachat et al., 2011; Mathuru et al., 2017).

Experimental subjects were administered a water solution of Citalopram (Citalopram Hydrobromyde, Sigma-Aldrich, Burlington, USA) of 0 mg/L (control), 30 mg/L, and 100 mg/L. Based on available literature (Sackerman et al., 2010; Macrì et al., 2019), we administer Citalopram through immersion of the experimental subject in a 500 ml beaker with the desired concentration for 5 min. Adopting this procedure, Sackerman et al. (2010) observed the presence of a direct correlation between Citalopram concentrations in the beaker and in zebrafish brain following a 3–4 min exposure. Citalopram concentrations were selected based on our previous work (Macrì et al., 2019; Clèment et al., 2020) wherein we observed an alteration in anxiety-related behaviors in response to Citalopram administered at the chosen concentrations.

To investigate the extent to which Citalopram-induced behavioral alterations translated to untreated group members, we tested experimental subjects under two main conditions: Singles (subjects tested in isolation), and Groups (a treated subject tested in the presence of four untreated conspecifics). The experiments were performed between November 20, 2020 and December 10, 2020. Overall, our study consisted of two testing conditions (Singles and Groups) and three treatments [0 mg/L (control), 30 mg/L, and 100 mg/L]. A schematic of the experimental procedure is shown in (Figure 1A). Twelve trials were conducted for each treatment/condition yielding a total of 72 trials counterbalanced across sexes (Singles and Groups). Thus, the experimental population consisted of 36 male and 36 female zebrafish. The 72 trials were evenly distributed over 6 days, keeping always the same number of tests per day (six in the morning and six in the afternoon). All fish were experimentally naïve, and were used only once.

For condition Singles, at the beginning of the trial, a single subject was transferred from the housing tank to a 500 mL beaker containing the desired concentration of Citalopram. The beaker was placed inside the experimental tank and the subject was kept there for 5 min following our previous work (Macrì et al., 2019). Then, the subject was transferred to the experimental tank and its behavior was video recorded for 10 min. Similarly, for condition Groups, at the beginning of the trial, four subjects were transferred from the housing tank into the experimental tank and left therein for 5 min. Towards the end of this phase, the treated subject was transferred from the housing tank to a 500 mL beaker containing the desired Citalopram concentration. The beaker was immediately placed inside the experimental tank and the subject was kept there for additional 5 min. Finally, the subject was transferred to the experimental tank and the group behavior was video recorded for 10 min.

All videos were recorded at 30 frames per second yielding a total of N_s = 18,000 images of 1,280 × 720 pixels. Images were input to a Matlab-based multi-target tracking algorithm Peregrine (Butail et al., 2013; Bartolini et al., 2015). The software fitted a parabola on the fish blob and returned time series of fish centroid coordinates {x(t)}t=1Nsand{y(t)}t=1Ns, along with their velocity components {vx(t)}t=1Nsand{vy(t)}t=1Ns, with t representing the time step. Here, x(t) and y(t) correspond to the position across the horizontal plane and along the water column, respectively. Each time series consists of N_s samples corresponding to the total experimental time of 600 s (10 min). Exemplary tracked trajectories for conditions Groups and Singles are shown in Figures 1B,C, respectively.

The output of the tracking system was used to investigate fish behavior through the scoring of salient metrics. In particular, geotaxis was quantified by computing the time average of the distance from the bottom of the tank. Group cohesion and coordination were assessed through the average nearest neighbor distance and alignment of the swimming direction (Theodorakis, 1989; Parrish et al., 2002; Miller and Gerlai, 2012). In particular, we computed the average nearest neighbor distance between the treated subject and the untreated ones, and between the untreated subjects only (excluding the treated individual; Rosa et al., 2020). The former was computed as the average of the minimum distance between the treated subject and each of the four untreated fish over the total experimental time of 10 min. The latter was computed as the average of the minimum distance between the four untreated subjects over the total experimental time of 10 min. To quantify the alignment between the four untreated subjects, we computed the time average of the instantaneous polarization given by Aureli et al. (2012)

where the components of the vectors are explicitly indicated, N = 5, and the treated subject is assigned the index i = 1. The instantaneous polarization ranges from 0 to 1, where 1 indicates full alignment of swimming directions between the untreated fish and zero absence of alignment.

To study causal interactions within the group, we implemented a transfer entropy analysis following an ample body of literature on the subject (Pilkiewicz et al., 2020). In particular, transfer entropy quantifies the cause-and-effect relationship between two time series of length N_d, {s1(k)}k=1Nd (effect) and {s2(k)}k=1Nd (cause), and it is defined as the extent by which knowledge of the presence of the cause reduces uncertainty in the prediction of the future of the effect from its present. Because the computation of transfer entropy entails the use of probabilities, we represent the time series s₁(k) and s₂(k) as two stationary stochastic processes S₁(k) and S₂(k), respectively (for further details, see Bossomaier et al., 2016). Then, transfer entropy (measured in bits) from S₂(k) to S₁(k) is given by

Here, p (S₁ (k + 1 ), | S₁ (k),S₂ (k)) is a joint probability indicating how likely the realizations of these three variables S₁(k+1), S₁(k), and S₂(k) occur. p (S₁ (k + 1) | S₁ (k), S₂ (k)) is a conditional probability indicating how likely future realizations of S₁ (at time (k+1)) happen, given a particular value for the present realizations of S₁ and S₂ (at time k). Similarly, p (S₁ (k + 1 | S₁(k)) denotes the likelihood of future realizations of S₁, given its present value at time k. Transfer entropy TE_S2→S1 is zero if S₂(k) does not cause S₁(k), while a nonzero value of TE_S2→S1 indicates the existence of a cause-and-effect relationship.

We computed transfer entropy among all fish in condition Groups for all the three Citalopram treatments; control, 30 (mg/L) and 100 (mg/L). Given our interest in detailing emotional contagion related to diving and rising patterns, we used the time series of velocity along the water column to compute transfer entropy. Based on our previous work (Porfiri and Marín, 2017) and the need to ensure Markovianity of the stochastic processes, we downsampled the time series of the velocity using a resolution of six time steps (time period of 0.2 s or 5 frames/s) yielding a total of N_d = 3,000; this time period was informed by our previous studies (Butail et al., 2016; Mwaffo et al., 2017). To compute transfer entropy, we first binned the time series of the velocity of each individual using two bins, that is, v˜yi(k) = “−” if v_yi(k) < 0, corresponding to motion towards the bottom of the tank and v˜yi(k) = “+” if v_yi(k) ≥ 0, corresponding to motion towards the top. Using the time series of v˜yi(k) of each fish in the group, we calculated 20 values of transfer entropy aggregated in a five by five matrix TE. The diagonal elements of the matrix are zero, while the generic element off the diagonal, TE_ij, corresponds to transfer entropy from individual i to individual j.

Ultimately, to quantify the overall influence of fish i on these of the group, we computed net transfer entropy for each individual (Mwaffo et al., 2017)

For each trial, we computed five values of net transfer entropy, one for the treated subject and four for the untreated shoal members. We averaged the latter four values so that each trial would yield two measures of net transfer entropy quantifying the role of Citalopram administration on the influence of any focal subject (be it treated or not) on the other shoal members.

To investigate the role of the social context on individual response to Citalopram administration, we analyzed the distance from the bottom in fish in the condition Singles and in Citalopram-treated fish in the condition Groups. We conducted an analysis of the distance from the bottom using a two-way ANOVA with Citalopram concentration and social context as independent variables. To address the extent to which an individual exposed to Citalopram altered group behavior in terms of geotaxis, we compared the distance from the bottom of all the animals in condition Groups as a function of exposure to Citalopram. To account for the possibility that the group average was influenced by the treated subject, we also investigated the distance from the bottom response of the untreated subjects in the condition Groups. For these analyses, we utilized a one-way ANOVA with Citalopram concentration as the independent variable. Similarly, we used a one-way ANOVA to analyze variables associated with group cohesion and coordination; namely, the nearest neighbor distance to the treated subject, the nearest neighbor distance between untreated subjects, and polarization. Outlier values were identified as those above or below two standard deviations from the median of the distance from the bottom for condition Singles and Groups at each Citalopram concentration. One outlier was identified for condition Groups at 0 mg/L and discarded from the analysis. Net transfer entropy was analyzed using a two-way ANOVA where Citalopram concentration (0, 30, and 100 mg/L) and treatment administration (treated fish vs. untreated shoal members) were selected as independent variables. Post hoc comparisons were performed using the Tukey test.

Comparisons of nearest neighbor distances and polarization with chance were conducted using one-tailed t-tests with Holm corrections (Holm, 1979). To establish chance values, we follow these steps: (i) randomly choose five individuals among all 216 fish used in the experiment; (ii) calculate the average values of the nearest neighbor distance to a single fish, the nearest neighbor distance between four fish, and polarization; (iii) repeat steps (i)-(ii) 10,000 times; and (iv) calculate the average values of the 10,000 iterations. Net transfer entropy values were compared with chance using a two-tailed t-test. All statistical analyses were performed with the statistics software R (version 3.6.1) with a significance level of 0.050. Specifically, we used the following functions in R; aov for conducting all ANOVAs, emmeans with Tukey adjustment was used for post hoc analyses, and finally, the t-tests were conducted using t_test.

When administered to zebrafish swimming in isolation (condition Singles), Citalopram did not affect the distance from the bottom of the tank. In particular, we did not register a statistically significant difference between Citalopram treatments (F_(2,32) = 1.882; p = 0.169; Figure 2).

Interestingly, we found that the social context, that is, the presence of four (untreated) subjects, altered the response of the focal individual. Specifically, while Citalopram-treated subjects and control individuals exhibited an indistinguishable distance from the bottom in condition Singles, Citalopram-treated subjects showed a greater distance from the bottom of the tank compared to vehicle-treated individuals in the condition Groups (Figure 2). Thus, for the focal fish in condition Groups, we registered a significantly different distance from the bottom between 0 mg/L and 30 mg/L (p < 0.050) and between 0 mg/L and 100 mg/L (p < 0.050). Additionally, when we compared the effect of the social context on the behavior of the treated subject, we observed that swimming alone or in a group remarkably altered the average distance from the bottom of vehicle-treated individuals but not of Citalopram-treated ones as indicated by the significant social context x Citalopram administration interaction (F_(2,64) = 5.535; p < 0.050) and further confirmed by post hoc analyses. Specifically, post hoc analyses revealed that vehicle-treated subjects swam closer to the bottom when they were in groups rather than alone (p < 0.001; Figure 2). This effect however was absent for 30 mg/L (p = 0.931) indicating that the presence of a shoal did not alter the behavior of the focal fish when treated with Citalopram for this concentration. For the higher Citalopram dose of 100 mg/L, we registered a trend (p = 0.065).

Our results indicate that the presence of a treated fish in the shoal decreases the anxiety-related behavior of the entire group in the form of increased distance from the bottom of the tank. When investigating this parameter in all shoal members, we observed that the intermediate Citalopram concentration (30 mg/L) resulted in reduced geotaxis compared to vehicle (Citalopram treatment: F_(2,32) = 4.702; p <0.050; p <0.050 in post hoc tests; Figure 3A). While experimental subjects exposed to 30 mg/L and 100 mg/L did not differ (p = 0.832), we observed a trend towards increased distance from the bottom of the tank 100 mg/L compared to 0 mg/L (p = 0.064).

Previous data may be partly explained by the possibility that the treated individual skewed the overall distribution. To account for this potential confound, we conducted an analysis only on the untreated shoal members in the three experimental groups. In accordance with the hypothesis that untreated subjects were influenced by the treated individual, we observed that untreated fish swimming together with a 30 mg/L Citalopram-treated individual displayed a higher distance from the bottom compared to untreated shoal members swimming with a vehicle-treated subject (Citalopram treatment: F_(2,32) = 3.384; p <0.050; p <0.050 in post hoc tests; Figure 3B). We did not register an equivalent effect in untreated fish swimming with a 100 mg/L Citalopram-treated individual (p = 0.155).

We first addressed whether Citalopram administration altered cohesion by analyzing the nearest neighbor distance to the treated fish, and afterward the nearest neighbor distance only between untreated subjects. Neither did Citalopram treatment modulate the nearest neighbor distance to the treated subject (F_(2,32) = 1.214; p = 0.310), nor did it alter the nearest neighbor distance of untreated subjects (F_(2,32) = 0.905; p = 0.415; Figures 4A,B). Similarly, experimental groups did not differ in terms of polarization (F_(2,32) = 0.037; p = 0.963; see Figure 4C).

We further compared cohesion and coordination variables with chance values. The neighbor distance to the treated subject was significantly lower than chance, for all three Citalopram treatments 0 mg/L (t₍₁₀₎ = −8.030; p < 0.050), 30 mg/L (t₍₁₁₎ = −3.060; p < 0.050), and 100 mg/L (t₍₁₁₎ = −2.720; p < 0.050). We also observed a significant difference from chance with respect to the nearest neighbor distance of untreated subjects at each Citalopram concentration (p < 0.050). Likewise, we found that polarization differed from the chance for each Citalopram concentration (p < 0.050).

We investigated whether the treated (focal) individual influenced untreated shoal members by studying the corresponding net transfer entropies and investigating the potential modulatory role of Citalopram. Our results show that the influence exerted by the Citalopram-treated fish on the group of untreated fish varied as a function of Citalopram concentration (interaction between Citalopram concentration (0, 30, and 100 mg/L) and treatment administration (treated vs. untreated; F_(2,64) = 3.284; p < 0.050) in net transfer entropy). The main effect of treatment administration was significant (F_(1,64) = 6.368; p < 0.050), while the main effect of Citalopram concentration did not attain statistical significance (F_(2,64) = 1.182; p = 0.313). Conducting post hoc analyses, we observed that net transfer entropy from the Citalopram-treated subjects to the untreated members was higher than net transfer entropy from untreated to treated for Citalopram concentrations of 30 mg/L (p < 0.050) and 100 mg/L (p < 0.050); this difference was absent in vehicle-treated fish (p = 0.492; Figure 5). Additionally, while untreated subjects displayed an equivalent net transfer entropy across Citalopram concentrations (0, 30, and 100 mg/L), focal fish treated with Citalopram displayed higher net transfer entropy compared to vehicle-treated individuals. Thus, we observed a different net transfer entropy between 0 mg/L and 30 mg/L (p < 0.050), along with a trend in the comparison between 0 mg/L and 100 mg/L (p = 0.058). Ultimately, we found that net transfer entropy is significantly higher than zero for the treated fish at 30 mg/L (t₍₁₁₎ = 3.560; p < 0.050), but not for vehicle (t₍₁₀₎ = −1.130; p = 0.857) and 100 mg/L (t₍₁₁₎ = 1.440; p = 0.088).