Zebrafish (Danio rerio) were housed at the Zebrafish Core Facility at East Carolina University. The facility was kept at a temperature of at 28°C under a 14 h/10 h light/dark cycle. Fish were fed daily with a high protein commercial food (Otohime B2, Reed Mariculture, Campbell, CA, United States) and with newly hatched artemia (Brine Shrimp Direct, Ogden, UT, United States). Wildtype (AB) zebrafish were group-housed in 10 gallon mixed-sex tanks prior to isolation and pairing. All experiments were performed in accordance with the Institutional Animal Care and Use Committee at East Carolina University (AUP #D320a). The dopamine type 1 receptor knockout line [drd1b^(–/–)] was generously provided by the Nicolson’s lab (Oregon Health Sciences University). The line was originally constructed at the Sanger Institute with an AB genetic background (Busch-Nentwich et al., 2013) and later deposited at the Zebrafish International Resource Center (ZIRC).

Adult male fish (∼6–12 months old) were taken from their communal tanks and isolated in a tank for 1 week, separated spatially and visually from other fish to minimize pre-existing social experience (Miller et al., 2017). Subsequently, two animals of equal size were paired in a new tank for a 2-week period and their aggressive behavior was monitored daily for 5 min to assess dominance as described previously (Miller et al., 2017).

After the pairing phase was completed, fish were temporarily separated, and behavioral testing was performed on a single fish following the protocol described elsewhere (Issa et al., 2011). Each fish was placed in a testing chamber (dimensions: 11 × 4 × 3 cm). A pair of conductive electrodes placed on either side of the chamber recorded the electric field potentials. Bare electrodes were 1 mm in thickness with 3–5 mm metal exposure. Electrodes were connected to an AC differential amplifier (AM-Systems model 1700, Carlsborg, WA, United States), and signals were amplified 1,000-fold. Electrical signals were low-pass filtered at 300 Hz and high-pass filtered at 1 KHz. Electrical field potentials are generated by muscle contractions when the fish moves (Issa et al., 2011). These signals were digitized using a Digidata-1322A digitizer then stored using Axoscope software (Molecular Devices, Inc., Sunnyvale, CA, United States). The experimental animals were acclimatized for 30 min before behavioral testing was initiated. Swimming behavior was recorded immediately following acclimation. Immediately after, startle escape responses were recorded.

Auditory pulses consisting of phasic 4 ms sine waves were generated using Audacity open-source audio editor and recorder software¹. Sound intensity was measured and calibrated external to the tank using a decibel meter (Sinometer, MS6700). Sensitivity of the animal’s auditory startle escape response was determined by tracking startle escape probability as a function of sound intensity. Activation of the M-cell mediated escape has a short latency of 5–15 ms. Non-Mauthner mediated responses with a time onset ranging from 15 to 40 ms were not counted, as these are controlled by an independent set of neural circuit that is not the target of our investigation (Eaton et al., 2001). Pulse intensity ranged from 70 to 100 dB with 5 dB increments. Pulse intensities were randomized and presented with a minimum of 2-min intervals to prevent habituation of the startle reflex. Response probability for each intensity was tabulated, and these probabilities were averaged across animals.

Following the 30-min acclimation period, and before conducting startle escape experiments, the animal’s swimming behavior was recorded for 1 min. The same methods of data acquisition, amplification, digitization, and storage were used as previously stated. Swimming activity was measured by counting swim bursts with Clampfit software. The “Threshold” function was used for this purpose. A potential was marked as a swim burst if it was at least 8 mV in total amplitude and 30–200 ms in duration. This range was chosen based on the typical characteristics of rhythmic swimming potentials that we observed. The timing of each swim burst was saved into a Microsoft Excel spreadsheet in reference to the recording start time.

Startle escape and swimming behavioral data was analyzed using Prism (GraphPad software Inc., San Diego, United States) and IBM-SPSS (RRID:SCR_002865). Unless specified otherwise, all comparisons were first subjected to one-way ANOVA or mixed design (a mixture of between-group and repeated-measures variables) ANOVA (between factor as group; within-factors as treatment and decibel) followed by the least significant difference (LSD) or paired two-sided t-test post hoc test for all multiple comparisons. Before using mixed-design ANOVA, sphericity was tested by using Mauchly’s test. When the assumption of sphericity was violated, the degree of freedom in Greenhouse-Geisser correction was used. For startle escape data, nonlinear regressions were performed using the Boltzmann sigmoidal equation:

A day after initial behavioral testing, fish were treated with either AM-251 or JZL184 and re-tested according to the previously stated protocol. Paired fish were separated with a divider during the injection and post-testing phase. The acclimation period was initiated 2 h post-injection. Fish were treated with a drug injected intraperitoneally following the protocol of Song et al. (2015). Intraperitoneal injections are preferred over direct brain injections because there is less risk of altering behavior with the physical injection and because both drugs can effectively cross the blood-brain barrier (Song et al., 2015). The drugs AM-251 and JZL184 were dissolved in DMSO to produce a 40 mM stock solution. For injection, capillary tubing was used, having the dimensions 1.0 mm OD × 0.5 mm ID × 100 mm in length. These were pulled using Flaming/Brown Micropipette Puller – Model P-87 from Sutter Instrument Co. The 40 mM stock solution was diluted in saline to 400 μM AM-251 and 400 μM JZL184. The tip of the micropipette was broken off with a razor blade, before loading with the drug solution. Loaded micropipettes were placed in Pneumatic PicoPump PV 820 for drug administration. A 0.3 % tricaine solution was used to anesthetize the animal prior to injection. Zebrafish were determined to have an average weight of 100 mg, therefore 2 μL of drug was injected to achieve a concentration of 4 mg/kg AM-251 and 4 mg/kg JZL184. To control for injury from injection and possible effects from solvents, separate dominant-subordinate pairs were injected with 10% DMSO in saline. To control for social status, communal fish were injected with either AM-251 or JZL184.

For western blot analysis four separate western blot trials were conducted with 10 brains per trial for each social phenotype. Zebrafish were anesthetized with 0.02% MS-222 (1 min) then placed in iced water (10 min). Brains were dissected out, placed in a 1.5 mL microcentrifuge tube, and stored at -20°C until use. Then they were prepared using the total membrane isolation protocol. Brains were homogenized in 1 ml of resuspension buffer: 2.5 ml 2M Sucrose, 2 ml 10 × 10 mM Tris–HCl, 400 μL 0.25M EDTA, 40 μl 20x protease cocktail inhibitor. The homogenized brains were centrifuged (2,000 rpm, 4°C, 10 min) followed by ultra-centrifugation (37,000 rpm, 4°C, 60 min). Protein sample concentrations were determined with a Lowery protein assay. For Western blots, 10 μg of each protein sample was denatured using 4X buffer containing 10X reducing reagent at 70°C for 10 min and loaded onto a Mini-PROTEAN^® TGX^TM Precast Protein Gels (Bio-Rad, Hercules, CA, United States) and run ∼120 min at 60V. The proteins were transferred to a 0.2 μm Nitrocellulose membrane using Trans-Blot^® Turbo^TM RTA Mini Transfer Kit (Bio-Rad, Hercules, CA, United States). The membrane was blocked with PBS containing 5% (w/v) nonfat dry milk and 0.1% (v/v) Tween 20 for 1 h at room temperature and then incubated with primary antibodies overnight at 4°C. After three washes with PBS containing 0.1% (v/v) Tween 20, the membrane was incubated with HRP-conjugated goat anti-mouse or goat anti-rabbit IgG (Cell Signaling) secondary antibody for 1 h. After washing, protein detection was performed using Horseradish peroxidase (HRP) Detection kit (SuperSignal^TM West Pico, Thermo Fisher Scientific) and visualized using the ChemiDoc Imaging System (Bio-Rad). The intensity of bands was quantified with ImageLab software. Band intensity was normalized by calculating protein/β-actin ratio. Data from dominants and subordinates were then normalized to protein/β-actin values from communal fish (Bio-Rad, Hercules, CA, United States). Primary antibodies used were as follows: mouse anti-CB₁R (1:500 Novus Biologicals), Rabbit anti-DAGL (1:500 Bioss Antibodies), β-actin (1:1000 Cell Signaling).

We constructed a neurocomputational model network that is composed of one excitatory cell, one inhibitory cell, and one M-cell. In the model, auditory inputs are initially delivered to the excitatory cell, which then excites the M-cell and the inhibitory cell. The inhibitory cell inhibits the M-cell only. In other words, auditory inputs affect the M-cell via two paths: a direct path via the excitatory cell and an indirect path via the excitatory cell and then the inhibitory cell. All model neurons were modeled as a conductance-based modified Morris–Lecar model with additional calcium-dependent potassium current (Morris and Lecar, 1981; Izhikevich, 2007; Ermentrout and Terman, 2010; Miller et al., 2017; Park et al., 2018). The membrane potential of each cell obeys the following current balance equation:

where IK=gKn(v-vK),ICa=gCam∞(v)(v-vCa),IKCa=gKCa{[Ca][Ca]+k1}(v-vK), I_L = g_L(v−v_L) represent the potassium, calcium, calcium-dependent potassium, and leak currents, respectively. [Ca] represents intracellular calcium concentration. For all neurons, g_K = 8, v_K = −84,g_L = 2, v_L = −60,g_Ca = 4, v_Ca = 120, and k₁ = 10. In the M-cell, g_KCa = 0.3 and C=1. In other neurons, g_KCa = 0.25 and C=20.

m_∞ is an instantaneous voltage-dependent gating variable for the calcium current where,

with v₁ = −1.2 and v₂ = 18.

The concentration of intracellular Ca^{2 +} is governed by the calcium balance equation:

where ε = 0.005,μ = 0.19 for all neurons. k_Ca = 0.9 in the M-cell and k_Ca = 1 in other neurons.

n is a gating variable for the potassium current obeying,

where ϕ = 0.23, v₃ = 12, and v₄ = 17 for all neurons.

In an excitatory cell and an inhibitory cell, the synaptic variable, s, is modeled by an equation for the fraction of activated channels,

where s∞(v)=1/(1+exp⁡(-v+θsσs)) with θ_s = 0 and σ_s = 4. The parameters α = 15 and β = 0.3 in an excitatory cell, and α = 8.5 and β = 0.046 in an inhibitory cell.

I_Tot represents the total input that a cell receives and is composed of a fixed constant (I₀), the synaptic current (I_syn) which represents the sum of synaptic inputs from other cells, and an applied current [I_app(t)]. The synaptic current is given by:

where the summation is over s variables from all neurons projecting to a given neuron.

In the current neuronal network, an excitatory cell does not receive any synaptic input but receives an external stimulus to simulate the effect of an external stimulus from the sensory input. Thus, in an excitatory cell, I_Tot = I_E0 + I_syn + I_app(t) where I_E0 = 43.9 is a fixed constant, I_syn = 0, and I_app(t) = W_EI(τ). Here, W_E is the stimulus strength, and I(τ) is the stimulus which resembles the square unit pulse with height 1 with duration of 2 ms starting at time τ.

In the M-cell, which receives synaptic inputs from an excitatory cell and an inhibitory cell in the network, the synaptic input is,

where s_M is the synaptic variable from another M-cell, which is assumed to be a constant in the current study.

Now, calcium is known to modulate the presynaptic neurotransmitter release via retrograde signaling (Diana and Bregestovski, 2005). In the model M-cell we assumed that intracellular calcium level reciprocally modulates the presynaptic input to the M-cell and I_syn is updated as follows:

where g_I obeys the following equation:

where g_Imax is the maximal g_I value, ρ is the time constant of g_I, and [Ca] is the intracellular calcium concentration of the M-cell. The total input to the M-cell is I_Tot = I_M0 + I_syn where I_M0 = 31 is a fixed constant and the synaptic input I_syn is as in Eq. (10). Other parameter values are given as follows: g_E→M = 0.15, v_E→M = 30, g_I→M = 0.5, v_I→M = −50, g_M→M = 0.5, v_M→M = −50, s_M = 0.029, g_Imax = 20, k₂ = 10, and ρ = 10000.

An inhibitory cell does not receive any direct external stimulus [I_app(t) = 0] but receives a synaptic input from the excitatory cell. Thus, in the inhibitory cell, I_syn = g_I*g_E→I(v−v_E→I)s_E where s_E is the synaptic variable from the excitatory cell. Thus, the total input to an inhibitory cell is I_Tot = I_I0 + I_syn where I_I0 = 36, v_E→I = 30, and g_E→I = 0.75 for a dominant-like model and g_E→I = 0.7 for a subordinate-like model.

Some parameters were modified to reflect different firing properties of each cell based on experiments (Eaton et al., 2001; Korn and Faber, 2005; Liao and Fetcho, 2008; Song et al., 2015). We note that all three cells are excitable cells so that they do not fire action potentials unless they receive enough excitatory inputs from other active cells or external stimulus. The main parameter that we controlled to implement these firing properties is the baseline level of a fixed constant I₀. Note that only the excitatory cell will receive the external stimulus which mimics the sensory input.

2-AG released from the M-cell modulates the release of the neurotransmitters in the pre-synaptic cells through CB₁R. To explore how 2-AG modulates the observed social status dependent escape responses to the external stimulus, we implemented the 2-AG modulation of synaptic inputs in the cells as follows.

In an inhibitory cell,

The main parameter CB1R_EI depends on the social status and we let CB1R_EI = 0.32 for a dominant-like model and CB1R_EI = 0.3 for a subordinate-like model.

In the M-cell,

The main parameters CB1R_EM and CB1R_IM depend on the social status. We let CB1R_EM = 0.27 for a dominant-like model and 0.3 for a subordinate-like model. Similarly, CB1R_IM = 0.2 for a dominant-like model and 0.25 for a subordinate-like model.

Simulations were performed on a personal computer using the software XPP (Ermentrout, 2002). The numerical method used was an adaptive-step fourth order Runge-Kutta method with a step size 0.01 ms. The neurocomputational model is available online in ModelDB². This website offers one of the largest opensource selections of neurocomputational models for various brain regions.

Previously, we demonstrated that social status regulates the activation of the startle escape and swim behaviors. The sensitivity of the startle escape response significantly increases in subordinates relative to dominants and group-housed fish; while swimming frequency significantly increases in dominants and decreases in subordinates (Miller et al., 2017). Given that the escape and swim circuits receive descending and local neuromodulatory inputs, we hypothesized that the differences in excitability and behavioral selection are likely due to a rebalance in the strength of excitatory and inhibitory neuromodulatory inputs. One potential mechanism is the previously described retrograde release of 2-AG from the post-synaptic M-cell shown to regulate the M-cell’s excitability by potentiating release from pre-synaptic dopaminergic inputs (Cachope et al., 2007). Moreover, the ECS is known to modulate other brain and spinal circuits involved in regulating motivated behavior (El Manira et al., 2008; Wenzel and Cheer, 2018). This lends credence to the notion that the ECS plays a key regulatory role in the molecular mechanism by which social status impacts circuit excitability. Therefore, we measured whole brain gene expression patterns of diacylglycerol lipase (DAGL), the primary enzyme that synthesizes 2-AG, monoacylglycerol lipase (MGL), the primary enzyme to degrade 2-AG, and Cannabinoid receptor type 1 (CB₁R) (Zou and Kumar, 2018). We found that the RNA gene expression of DAGL and CB₁R are socially regulated (Figure 2A). Dominant animals showed a significant increase in DAGL expression relative to subordinates, and subordinates showed a significant decrease in DAGL expression relative to controls [Kruskal–Wallis test, p < 0.05; Figure 2A]. We also observed that the expression of CB₁R was significantly increased in subordinates relative to controls [Kruskal–Wallis test, p < 0.05; Figure 2A], but CB₁R expression was not significantly different between dominants and subordinates. Western blot analysis of DAGL and CB₁R showed similar protein expression patterns (Figure 2B). DAGL western blots consistently showed multiple protein bands. Several causes for this are possible, including: denaturation of protein structure leading to increased antibody cross-reactivity with similar amino acid residues; differences in phosphorylation states of the DAGL proteins; and reduced antibody specificity for DAGL. Regardless, the largest bands were reliably ∼110 kDa, consistent with the size of DAGL. Such variations were not observed with CB₁R western blots that consistently showed single bands (Figure 2B).

These results led us to postulate that differences in the ECS may account for the social status-dependent differences in locomotor behavior. To test this hypothesis, we augmented 2-AG levels by injecting the animals with JZL184, an irreversible inhibitor of MGL (Long et al., 2009; Figures 3A–C). To compare differences in the startle escape response probabilities among the three animal groups and treatment (JZL184 injection), we performed a mixed-design ANOVA (between-subject factor as Group; within-subject factors as Treatment and Decibel). There was a significant main effect of Decibel [F(3.35,107.06) = 338.337, p < 1.0e-16; Figures 3A–C] and marginal main effect of Group [F(2,32) = 3.023, p = 6.30e-2; Figures 3A–C], but there was no effect of Treatment [F(1,32) = 0.080, p > 0.05; Figures 3A–C]. There were significant interactions of Group^∗Treatment [F(2,32) = 15.69, p = 1.80e-5; Figures 3A–C], Group^∗Decibel [F(6.69,107.06) = 2.26, p = 3.75e-2; Figures 3A–C], Treatment^∗Decibel [F(3.75,120.06) = 2.18, p = 7.99e-2; Figures 3A–C], and Group^∗Treatment^∗Decibel [F(7.50,120.06) = 3.61, p = 1.12e-3; Figures 3A–C]. We performed the post hoc test to determine which animal groups had higher escape response probability. We observed that the response probability for dominants was significantly lower compared to subordinates (LSD, p = 3.95e-2; Figures 3A–C), but there was no difference compared to communals (LSD, p > 0.05; Figures 3A,B). Moreover, the response probability for subordinates was also significantly higher compared to communals (LSD, p = 4.41e-2; Figures 3A,C).

We then performed further analysis to determine whether JZL184 injection affected the escape response in each animal group. In communals, there was significant main effect of Decibel [F(3.13,31.27) = 107.87, p < 1.0e-16; Figure 3A], but no effect of Treatment [F(1,10) = 2.63, p > 0.05; Figure 3A]. There was also no Treatment^∗Decibel interaction [F(3.47,34.72) = 1.72, p > 0.05; Figure 3A]. We also observed a significant difference of the startle escape responses due to JZL184 at 85 dB for communals [paired one-sample two-sided t-test, t(10) = 3.08, p = 1.16e-2; Figure 3A]. In dominants, there were significant main effects of Treatment [F(1,11) = 7.57, p = 1.88e-2; Figure 3B] and Decibel [F(2.40,26.43) = 106.12, p < 1.0e-16; Figure 3B]. There was no effect of Treatment^∗Decibel interaction [F(2.90,31.89) = 1.43, p > 0.05; Figure 3B]. We did not observe a significant difference after the treatment at a particular decibel although we observed that JZL184 injection increased the overall startle escape response over the wide range of decibel. In subordinates, there were significant main effects of Treatment [F(1,11) = 30.08, p = 1.91e-4; Figure 3C], Decibel [F(2.15,23.65) = 134.28, p < 1.0e-16; Figure 3C], and Treatment^∗Decibel [F(3.14,34.59) = 5.67, p = 8.50e-5; Figure 3C]. We observed that JZL184 injection significantly decreased the overall startle escape response over the wide range of decibels. In particular, we observed significant differences in the startle escape responses at 75 dB [paired one sample two-sided t-test; t(11) = 3.45, p = 5.46e-3; Figure 3C], at 80 dB [t(11) = 3.82, p = 2.86e-3; Figure 3C], and at 85 dB [t(11) = 2.72, p = 1.99e-2; Figure 3C]. In summary, the results show that blocking 2-AG degradation increased the startle escape response sensitivity in communals and moderately increased it in dominants, but significantly decreased the startle escape sensitivity in subordinates.

Activation of the startle escape response suppresses swimming activity by inhibiting the slow motor neurons that drive swimming behavior (Svoboda and Fetcho, 1996; Satou et al., 2009). Given that the ECS is implicated in promoting motivated behavior, we hypothesized that increasing the availability of 2-AG is likely to promote a behavioral switch in the activation pattern that would favor motivated behavior (i.e., swimming) over submissive behavior (i.e., escape). The notion is that augmenting 2-AG would be sufficient to reverse the activation pattern of the swim circuit in a socially dependent manner as was observed with the escape response. Indeed, we found that injection of JZL184 had the opposite effects on dominants and subordinates in that subordinates significantly increased their swimming activity while dominants significantly decreased their swimming and communal animals showed only moderate change (Figures 3D–G).

We compared the difference of swim bursts among three animal groups and treatment (JZL184 injection) and performed a mixed-design ANOVA (between-subject factor as Group; within-subject factor as Treatment). We found no main effect of Group [F(2,31) = 0.44, p > 0.05; Figures 3D–G] and no effect of Treatment [F(1,31) = 1.50, p > 0.05; Figures 3D–G]. But there was a significant effect of Group^∗Treatment interaction [F(2,31) = 10.89, p = 2.62e-4; Figures 3D–G]. We performed further analysis to determine whether JZL184 injection affects the swim bursts in each animal group. In communals, we found a marginal main effect of Treatment [F(1,11) = 3.65, p = 8.25e-2; Figures 3D–G]. JZL184 injection slightly increased the swim bursts in communals. In dominants, there was a significant main effect of Treatment [F(1,10) = 6.98, p = 2.46e-2; Figures 3D–G]. JZL184 injection significantly decreased the swim bursts for dominants. In subordinates, there was a significant main effect of Treatment [F(1,10) = 9.41, p = 1.19e-2; Figures 3D–G]. JZL184 injection significantly increased the swim bursts for subordinates.

To determine whether these status-dependent differences are due to changes in CB₁R activity, we tested the startle escape response in the presence of AM-251, a specific CB₁R antagonist (Seely et al., 2012; Figures 4A–C). To compare the difference of the startle escape response probabilities to auditory pulses among three animal groups and treatment (AM-251 injection), we performed a mixed-design ANOVA (between-subject factor as Group; within-subject factors as Treatment and Decibel). There were significant main effects of Treatment [F(1,27) = 10.32, p = 3.39e-3; Figures 4A–C] and Decibel [F(2.99,80.62) = 218.56, p < 1.0e-16; Figures 4A–C]. But there was no main effect of Group [F(2,27) = 2.22, p > 0.05; Figures 4A–C]. We also observed significant interaction effects of Group^∗Decibel [F(5.97,80.62) = 2.27, p = 4.55e-2; Figures 4A–C] and Treatment^∗Decibel [F(2.90,78.16) = 4.36, p = 7.41e-3; Figures 4A–C]. But there were no effects of Group^∗Treatment interaction [F(2,27) = 0.633, p > 0.05; Figures 4A–C] and Group^∗Treatment^∗Decibel interaction [F(5.79,78.16) = 0.76, p > 0.05; Figures 4A–C]. We performed the post hoc test to determine which animal groups had higher probability. We observed that the response probability for subordinates was marginally higher compared to dominants (LSD, p = 5.44e-2; Figures 4B,C). But there were no differences between subordinates and communals (LSD, p > 0.05; Figures 4A,C) and between dominants and communals (LSD, p > 0.05; Figures 4A,B).

We then performed further analysis to determine whether AM-251 injection affects the escape response in each animal group. In communals, there was significant effect of Decibel [F(2.70,23.30) = 46.88, p < 1.0e-16; Figure 4A]. But there were no effects of Treatment [F(1,9) = 3.32, p > 0.05; Figure 4A] and Treatment^∗Decibel interaction [F(3.04,27.31) = 0.41, p > 0.05; Figure 4A]. In dominants, there was a significant main effect of Decibel [F(2.16,19.41) = 77.29, p < 1.0e-16; Figure 4B]. But there were no effects of Treatment [F(1,9) = 1.74, p > 0.05; Figure 4B] and Treatment^∗Decibel interaction [F(1.74,15.66) = 1.26, p > 0.05; Figure 4B]. In subordinates, there were significant main effects of Treatment [F(1,9) = 13.50, p = 5.12e-3; Figure 4C] and Decibel [F(1.75, 15.72) = 139.13, p < 1.0e-16; Figure 4C]. There was also a significant effect of Treatment^∗Decibel interaction [F(2.84,25.60) = 5.65, p = 4.63e-3; Figure 4C]. In particular, we observed significant differences in the startle escape responses at 80 dB [paired one sample two-sided t-test; t(9) = 3.69, p = 5.02e-3; Figure 4C], at 85 dB [t(9) = 2.35, p = 4.34e-2; Figure 4C], and at 90 dB [t(9) = 2.51, p = 3.33e-2; Figure 4C]. In summary, irrespective of social rank, AM-251 decreased startle escape sensitivity particularly so in subordinates. These findings supports Cachope et al. (2007) results in that 2-AG potentiates the sensitivity of the M-cell. We extend on their finding and show that the supply of 2-AG is socially regulated, and it modulates M-cell excitability in a status-dependent manner.

To determine whether CB₁R activity influences swimming patterns differently among the three social groups, we compared swim bursts among three animal groups and treatment (AM-251 injection) by performing a mixed-design ANOVA (between-subject factor as Group; within-subject factor as Treatment). There were main effects of Group [F(2,27) = 4.25, p = 2.48e-2; Figures 4D–G] and Treatment [F(1,27) = 5.77, p = 2.34e-2; Figures 4D–G]. But there was no effect of Group^∗Treatment interaction [F(2,27) = 1.35, p > 0.05; Figures 4D–G]. Post hoc tests showed that the swim bursts for dominants were significantly higher compared to communals (LSD, p = 4.61e-2; Figures 4D–G) and subordinates (LSD, p = 9.19e-3; Figures 4D–G). But there was no difference of swim bursts between communals and subordinates (LSD, p > 0.05; Figures 4D–G). We performed the further analysis to determine whether AM-251 injection affects the swim bursts in each animal group. In communals, there was a significant main effect of Treatment [F(1,9) = 8.05, p = 1.95e-2; Figures 4D–G]. AM-251 injection significantly decreased the swim bursts in communals. In dominants, there was no main effect of Treatment [F(1,9) = 2.48, p > 0.05; Figures 4D–G]. In subordinates, there was no main effect of Treatment [F(1,9) = 0.004, p > 0.05; Figures 4D–G]. Thus, blockage of CB₁R significantly decreased swimming in communals while no changes in swimming were observed in dominants or subordinates.

Collectively, the results of supplementing 2-AG and blockage of CB₁R indicate that 2-AG’s regulation of the swim and escape behaviors is socially regulated. More importantly, 2-AG serves as a molecular switch in shifting the activation pattern between competing circuits by modulating the balance of excitatory and inhibitory inputs onto the two motor circuits in a social status-dependent manner.

Dopamine is known to be involved in social regulation (Watanabe and Yamamoto, 2015), motivation (Hamid et al., 2016), and aggression (Filby et al., 2010), and its function is highly interdependent with the ECS signaling pathway (Wenzel and Cheer, 2018). Specifically, ECS signaling can be mediated via dopamine receptor type 1 (DRD1) (Zenko et al., 2011). In zebrafish, the most compelling evidence is the capacity of 2-AG to potentiate mixed synaptic transmission to the M-cell that requires activation of DRD1 (Cachope et al., 2007). Moreover, evidence by Pereda et al. (1992) shows direct dopaminergic innervation of the M-cell, DA release directly potentiates M-cell excitability, and this potentiation can be blocked by antagonizing DRD1. We hypothesized that ECS signaling underlying status-dependent differences in escape sensitivity is mediated through DRD1. To test this hypothesis, we repeated our experiment of augmenting 2-AG levels but in DRD1 knockout zebrafish [drd1b^(–/–)] (Busch-Nentwich et al., 2013). We found that startle escape sensitivity was unaffected in drd1b^(–/–) fish following JZL184 injection contrasting with the results of WT animals, in which an increase of 2-AG levels significantly increased startle escape sensitivity (Figure 5A compare with Figure 3A). We compared differences in the startle escape response probabilities to auditory pulses for DRD1 knockout zebrafish before and after the treatment of JZL 184 injection and performed repeated measures ANOVA (within-subject factors as Treatment and Decibel). We found a significant main effect of Decibel [F(2.03,18.30) = 64.03, p < 1.0e-16; Figure 5A]. But there was no main effect of Treatment [F(1,9) = 0.12, p > 0.05; Figure 5A] and no Treatment^∗Decibel interaction [F(2.77,24.92) = 0.75, p > 0.05; Figure 5A]. Similarly, we found no change in swimming frequency in the drd1b^(–/–) animals following JZL184 injection [F(1,9) = 0.41, p > 0.05; Figures 5B,C]. These results show that the potentiating effects of 2-AG on startle escape sensitivity are mediated through DRD1b.

Our empirical results show that the ECS is affected by social experience to regulate the activation of the startle escape and swim behaviors. This complex interaction between social factors, neuromodulatory systems and motor circuits necessitated the development of a neurocomputational model to better understand how social status influences the ECS to modulate the excitability and pattern of motor activity. Toward this end, we developed a computational model whereby we simulated the M-cell along with two pre-synaptic cells, an excitatory cell (representing glutamatergic and/or dopaminergic neurons) and an inhibitory cell (representing GABAergic and/or glycinergic neurons) (see section “Materials and Methods,” Figure 6A, inset).

Using this model, we tested the hypothesis that differences in the amount of 2-AG release due to differences in DAG lipase expression would account for how 2-AG sets the gain of the pre-synaptic excitatory and inhibitory inputs onto the M-cell. More specifically, we tested whether the behavioral differences in the escape circuit are modulated by changing network properties (synaptic strengths, g_E→I) and three activity levels of CB₁Rs (CB1R_EM, CB1R_EI, and CB1R_IM) in the model to reflect different levels of 2-AG and social status. Here, g_A→B and CB1R_AB represent the synaptic strength and the activity level of CB₁R from a neuron A to a neuron B where E, I, and M represent for the excitatory, inhibitory, and M-cell, respectively. We assumed that dominants and subordinates have different levels of a synaptic strength and activity levels of CB₁R. In particular, we assumed that low values of CB1R_EM and CB1R_IM, high values of CB1R_EI and g_E→I for a dominant-like model while high values of CB1R_EM and CB1R_IM, low values of CB1R_EI and g_E→I for a subordinate-like model. That is, for a dominant-like model, the inhibitory pathways were enhanced while a subordinate-like model had strong excitatory pathways.