Escape behaviors are transiently modulated after acutely induced epileptic seizures in larval zebrafish

Abstract

Epilepsy is a common neurological disorder giving rise to both recurrent and spontaneous seizures as well as additional neurological comorbidities. In approximately 30% of patients, seizures are not ameliorated by existing anti-seizure drugs. To bridge this treatment gap, the larval zebrafish has emerged as a powerful model for high-throughput drug-screening, based on changes in swimming behavior during seizures. However, such screening has low specificity and thus requires subsequent, low-throughput validation of drug candidates. We found that behavioral escape responses to startling stimuli are strongly reduced for a few hours after seizures, suggesting this behavioral change could serve as a useful assay to identify drugs ameliorating seizures with higher specificity. Moreover, we found that the extent of behavior change and the rate of recovery are dependent on convulsant concentration which determines seizure strength. Finally, we demonstrated that an anti-epileptic drug, Valproic acid, reduced swimming behavior during seizures as well as mitigated the reduction in post-seizure escape behaviors. Nevertheless, it also reduced escape behaviors in the absence of seizures. These data argue that monitoring escape behavior after drug treatment and after seizures provides an improved high throughput assay to identify new anti-epileptic drugs as well as reveal non-specific effects of drugs on neural circuit function, which may predict side-effects in humans.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-40684-3.

Introduction

Epilepsy is a common neurological disorder affecting ~ 1% of the global population^1–3. Epilepsy patients suffer from spontaneous and recurrent seizures—episodes of increased and highly synchronous neural activity giving rise to behavioral phenotypes such as jerking movements, muscle stiffening and falls. In addition, epilepsy syndromes are often associated with additional comorbidities, such as anxiety and depression as well as cognitive deficits including deficits in attention, learning and memory^4–9. Unfortunately, existing anti-epileptic drugs, more appropriately called anti-seizure drugs (ASDs), suppress seizures in only ~ 70% of patients and are often associated with adverse effects such as sedation and tremor, as well as cognitive and psychiatric symptoms, reducing tolerability^10–13. Many epilepsy animal models have been developed with the goal of capturing the heterogeneity of human epilepsies and thus facilitating the identification of effective treatments for drug-resistant epilepsies. Moreover, while existing animal models of epilepsy were designed to capture the phenotypes associated with seizures, there is an additional need for models that would also capture behavioral epilepsy comorbidities and the long-term effects of seizures. Such models are required to develop treatments that will effectively address more of the deficits associated with epilepsy and improve patients’ quality of life¹⁴.

The larval zebrafish recently emerged as a useful epilepsy model. In this model, application of Pentylenetetrazole (PTZ), a GABA A receptor antagonist used as a generalized seizure model¹⁵, causes excitation-inhibition imbalance giving rise to seizure-like behaviors and epileptiform neural activity, demonstrating face validity. Moreover, seizures induced by PTZ were ameliorated by application of commonly used ASDs, demonstrating predictive validity^16–27. All of these effects are consistent with findings in adult zebrafish and rodents^28–36. While these studies of seizures acutely induced by PTZ in larval zebrafish focused mostly on the behavioral and neural phenotypes presented during the seizure, one study used this model, together with other models of increased neural activity, to show that such activity can be followed by increased rebound sleep for the entire following light period³⁷. However, seizure effects on behavioral responses to strong stimuli during the hours immediately following the seizures, and their potential modulation in the presence or absence of ASD treatment, were not previously examined.

In addition to its validity, the larval zebrafish also presents a unique advantage in the context of anti-epileptic drug-screening, as it is amenable for high-throughput drug screening via quantification of swimming behavior in a 96-well plate assay^38–46. Application of this assay to genetic zebrafish epilepsy models, particularly of the severe and ASD-resistant Dravet syndrome, gave rise to multiple new drug candidates^47,48. However, the assay is limited by low-specificity, requiring secondary testing of candidates via much lower-throughput methods such as invasive electrophysiological measurements that further require expertise^39,40. Moreover, this assay only captures the behavioral correlates of seizures themselves, and not other comorbidities including long-term seizure effects and potential adverse effects of drugs. Examination of post-seizure effects can address this.

Here, we reveal that after undergoing seizures induced by PTZ, larval zebrafish present post-seizure behavioral deficits. In particular, the ability of larvae to perform a stereotypic sensory-motor transformation, an escape response to startling stimuli, dramatically decreases. This decrease was transient, lasting a few hours, and its extent and duration depended on the strength of the induced seizure. Moreover, we found that the measurement of escape responses was also useful for identifying potential drug side-effects, as responses were also reduced following treatment with a classic anti-epileptic drug, Valproic acid. Measuring startle responses in the absence of seizures and after seizures proved useful for both identifying the seizure-ameliorating property of valproic acid, as well as to reveal its undesired impact on circuit function. These findings point to the potential of extending the utility of the larval zebrafish epilepsy model to allow more specific and rapid drug screening, revealing drug side effects, and for studying the long-term impact of seizures on neural function and behavior.

Results

Seizing larvae present high swimming distances and velocities

We first applied an established protocol to acutely induce seizures in 6 days post-fertilization (dpf) larval zebrafish freely swimming in a 96 well plate via application of 15 mM Pentylenetetrazole (PTZ)^16,19 (Fig. 1A). To quantify seizure behavior we tracked each larva and extracted the kinematic parameters characterizing swimming movements. In particular, we examined the distance travelled and mean velocity during movements in 2.5 min intervals for 1 h. Consistently with previous observations¹⁹, we found that strongly seizing larvae transiently increased movement such that distances travelled were maximal ~ 20 min after convulsant application (Fig. 1B). Also consistently with previous reports²², movement velocities (i.e., velocities in frames in which the larvae were not static) were also increased in seizing larvae compared to controls, throughout the recording, and peaked slightly earlier than distance travelled (Fig. 1C). Finally, we also computed the mean distance travelled and mean movement velocities across all time intervals for each larva (Fig. 1D–E). As expected, mean distances travelled were higher in experimental larvae in 15 mM PTZ (mean and standard error of the mean: 117.51 ± 3.85 mm, N = 132 larvae), compared to controls (47.33 ± 4.58 mm, N = 66 larvae, Fig. 1D). This difference was statistically significant according to a two-tailed t-test with unequal variances (t = 11.737; df = 151.718; p = 4.828 * 10⁻²³). In addition, seizing larvae moved faster than controls (5.16 ± 0.08 mm/sec compared to 3.14 ± 0.07 mm/sec in corresponding controls, t = 18.228; df = 185.605; p = 3.206*10⁻⁴³, Fig. 1E). We note that since the distribution of velocities in experimental larvae deviated from normality, we further validated the significance with a Wilcoxon rank-sum test (z = 9.788, p = 1.264*10⁻²²).

Fig. 1.

Rapid swimming is induced during PTZ application. (A) Example 4 out of 9 imaged wells during seizure behavior tracking. Yellow circles—identified well edges; Red diamonds—identified larval head positions. (B). The distance travelled in 2.5 min intervals following the application of 15 mM PTZ (blue curve) or E3 (black curve) at time 0. Shaded regions around lines indicate the standard error of the mean at each time point, lines are means of all larvae. (C) Same as (B) but showing the mean velocity of the larvae during movements. (D) Box plots of mean distance travelled during the entire experiment, following application of 15 mM (blue boxes) PTZ and in corresponding controls in E3 (black boxes). In all box plot figures: Red line—median; bottom and top edges—25^th and 75^th percentiles, respectively; whiskers extend to the most extreme data points not considered as outliers. Red ‘ + ’ marks indicate outliers. (E) Same as (D) but for mean velocity during movements rather than distance.

Startle responses are reduced following acutely-induced seizures

We next asked whether long-lasting post-seizure effects on behavior can be detected after the convulsant is removed. In particular, we focused on a key sensory-motor transformation that is critical for survival—the escape response. We used an established protocol for triggering startle responses via mechanical tapping⁴⁹ and validated that startle responses, their tracking and classification, are consistent with previous reports. Indeed, we identified C-start responses (Fig. 2A), with both short and long latencies (SLCs and LLCs, respectively; Fig. 2B–C) having distinct kinematics as previously reported. We examined escape responses during the first 3 post-seizure hours (Fig. 2D–F). Since handling of larvae during this two-stage experiment that includes the characterization of both swimming behavior in a 96 well plate and startle behavior in PDMS wells can affect startle responses, we included two controls – 1. Larvae that were not placed in a 96 well plate prior to examination of startle responses, and thus their response rate reflects the intact baseline; and 2. Larvae whose swimming behavior in a 96 well plate was examined, but that did not undergo induced seizures. Interestingly, post-seizure startle response rates, the fraction of trials in which responses of any type were observed, (0.08 ± 0.02, N = 48) were reduced compared to both control groups (0.71 ± 0.05, N = 23; and 0.59 ± 0.06, N = 23; Fig. 2D). The reduced responsivity was significant according to a Kruskal–Wallis test comparing the response rate distributions (χ² = 62.4; df = 2,91; p = 2.814*10⁻¹⁴); and post-hoc tests using the Dunn-Šidák method to compare the experiment group to each of the controls (z = 6.904; p = 1.521*10⁻¹¹ and z = 5.869; p = 1.315*10⁻⁸, respectively). Moreover, the rates of short-latency C-start responses (SLCs), mediated by M cell activity, were also significantly reduced following seizures, similarly to total response rates, from 0.48 ± 0.06 in the baseline group and 0.37 ± 0.05 in controls to 0.07 ± 0.02 in the experiment group (χ² = 50.7; df = 2,91; p = 9.776*10⁻¹² for Kruskal–Wallis test and z = 6.251, p = 1.225*10⁻⁹; z = 5.251, p = 4.534*10⁻⁷ for post-hoc comparisons to the control groups; Fig. 2E).

Fig. 2.

Startle response rates transiently decrease after seizures induced by PTZ. (A) Example of startle behavior tracking, for a control larva performing an SLC. Different panels show key time-points of the response. Red star: center point between the eyes. Green star: middle point of the swim bladder. Blue star: tip of the tail. Cyan line: head direction, for head angle extraction. Magenta line: tail direction, for tail angle extraction. (B–C) Histograms of latencies (B) and maximal angular velocities (C) characteristic of identified SLC (purple) and LLC (orange) responses. (D–E) Box plots of startle response rates (for all response types in (D) and for SLC only in (E)) in baseline larvae, not treated with PTZ nor tracked in a 96 well plate, control larvae whose swimming behavior was tracked in a 96 well plate similarly to the experimental larvae, and experimental larvae after treatment with 15 mM PTZ. (F–G) Box plots of startle response rates, for all response types, for the same groups as in (D–E) but divided by the post-seizure time (in hours), in short-term (F) and long-term (G) experiments.

The reduction in startle responses after acutely-induced seizures is transient

In these first experiments, startle responses in different larvae were examined at slightly distinct post-seizure time-points. Thus, we next asked whether the altered startle responsivity after seizures depended on the time elapsed since the seizure ended. To address this question, we separated our control and experimental groups according to the post-seizure time (Fig. 2F and Supp. Figure 1). We found that the rate at which larvae previously treated with 15 mM PTZ responded to startling stimuli was lower than the rate of controls at all times tested (Fig. 2F). In particular, using a Wilcoxon rank sum test to compare response rates at distinct post-seizure times between experimental larvae and controls, we found a significant difference at all times tested (1, 2 and 3 h post seizure; z = 3.676, p = 2.370*10⁻⁴; z = 3.824, p = 1.314*10⁻⁴; z = 3.15, p = 1.632*10⁻³, respectively). We note that these differences are significant also when applying a Bonferroni correction to control the familywise error rate (i.e., comparing to a significance level of 0.05/3). We also noticed that in experimental larvae response rates showed a small increase from 0.06 ± 0.03 one hour after seizure to 0.12 ± 0.05 three hours after seizure (N = 16 in all groups), in contrast to a small decrease in control rates from 0.59 ± 0.12 to 0.46 ± 0.08 (N = 7–8 in different groups), yet these differences were not statistically significant (A Kruskal–Wallis test showed no significant effect of time on response rates—χ² = 2.61; df = 2,45; p = 2.713*10⁻¹ for experimental larvae and χ² = 0.05; df = 2,20; p = 9.747*10⁻¹ for controls). Examination of SLC response rates over-time yielded similar results (Supp. Figure 1A), with rates following application of 15 mM PTZ being lower in a statistically significant manner in experimental with respect to control larvae at 1–3 h post-seizure (z = 3.028, p = 2.460*10⁻³; z = 3.647, p = 2.655*10⁻⁴; z = − 2.430, p = 1.512*10⁻²; respectively). Moreover, Kruskal–Wallis tests showed no significant effect of time on SLC rates in either experimental larvae or controls (χ² = 1.88; df = 2,45; p = 3.902*10⁻¹ and χ² = 3.09; df = 2,20; p = 2.133*10⁻¹; respectively).

We reasoned that even if no recovery of response rates occurs within 3 h after seizure, it may still occur over longer time-scales. We therefore conducted longer duration experiments, in which startle behavior was examined over 3–6 h following 15 mM PTZ application for 1 h and washout. Indeed, we observed recovery in these longer experiments (Fig. 2G): The total response rate in the experiment group increased over this time interval from 0.16 ± 0.05 to 0.51 ± 0.09 (N = 10–28 in different groups) in contrast to a slight decrease observed in controls (from 0.42 ± 0.08 to 0.35 ± 0.12, N = 6–15 in different groups). A Kruskal–Wallis test performed on response rates in experimental larvae at these 4 distinct post-seizure time points showed significant differences in group means (χ² = 15.37, df = 3,80, p = 1.529*10⁻³) and post-hoc tests showed that the high response rate observed 6 h after the seizure was significantly different than the response rates 3 and 4 h after the seizures (z = 3.676, p = 1.419*10⁻³; z = 2.862, p = 2.503*10⁻²; respectively). Moreover, a Wilcoxon rank sum test revealed no difference between response rates in the control and experimental larvae 6 h post-seizure (z = 1.0879, p = 2.767*10⁻¹), while a difference existed 3 h post-seizure (z = 2.921, p = 3.493*10⁻³ < 0.05/2 for Bonferroni correction). In addition, the same trend of increasing response rates in experimental larvae between 3 and 6 h was also observed when limiting the analysis to SLC responses only (a time-dependent effect in experimental larvae was identified with a Kruskal–Wallis test: χ² = 9.17, df = 3,80, p = 2.710*10⁻²; Supp. Figure 1B). As with total response rates, post-hoc tests showed a significant difference in SLC response rates of experimental larvae at 6 h post-seizure (0.28 ± 0.07) compared to 3 h post-seizure (0.1 ± 0.03; z = 2.955, p = 1.860*10⁻²), but there was no significant difference between 4 and 6 h post-seizure (0.10 ± 0.02; z = 2.507, p = 7.096*10⁻²). Similarly, a Wilcoxon rank sum test reveals no difference between SLC rates in the control and experimental larvae 6 h post-seizure (z = 1.088, p = 2.767*10⁻²), while a difference exists 3 h post-seizure (z = 2.921, p = 3.493*10⁻³). Put together, this demonstrates that the post-seizure effect on a stimulus-triggered behavior is transient, in that the behavioral responsivity recovers over 6 h following PTZ treatment.

Weakly induced seizures give rise to weaker and more transient post-seizure effects

Having observed a post-seizure effect for strong seizures, we next asked whether more weakly induced seizures also give rise to milder post-seizure effects. To test this, we reduced the PTZ concentration from 15 to 5 mM and repeated the experiments described above. First, comparing the behavior during the induced seizures (Fig. 3A–D), we found that, in weakly seizing larvae (in 5 mM PTZ), movement increased more gradually following convulsant application and did not decrease almost throughout the hour long induced seizure (Fig. 3A). This is in contrast with the gradual reduction in distances travelled observed in more strongly seizing larvae after 20 min of convulsant application. These observations are consistent with previous studies showing that in weakly induced seizures, seizure phenotypes develop more gradually, and in particular, the transitions between initial seizure stages, associated with increased movement, and advanced seizure stages, associated with loss of posture and reduced movements, are slower^16,19,20. In addition, the movement velocities of weakly seizing larvae were higher than control velocities and nearly constant throughout the recording (Fig. 3B). Examining the mean distances travelled over the recording time, we found that they were higher in the 5 mM PTZ group (104.8 ± 6.6 mm, N = 78) than in controls (43.65 ± 5.42 mm, , N = 39, Fig. 3C) in a statistically significant manner (t = 7.163, df = 112.315, p = 8.757 * 10⁻¹¹).

Fig. 3.

Startle response rates recover more quickly after weak seizures. (A) The distance travelled in 2.5 min intervals following the application of 15 mM PTZ (blue curve), 5 mM PTZ (green curve) or E3 (black curves, continuous – 15 mM PTZ controls, dashed – 5 mM PTZ controls) at time 0. Shaded regions around lines indicate the standard error of the mean at each time point, lines are means of all larvae. (B) Same as (A) but showing the mean velocity of the larvae during movements. (C) Box plots of mean distance travelled during the entire experiment, following application of 5 mM (green boxes) or 15 mM (blue boxes) PTZ and by corresponding controls in E3 (gray and black boxes, respectively). (D) Same as (C) but for mean velocity during movements rather than distance. (E) Box plots of startle response rates, for all response types, after treatment with either 5 or 15 mM PTZ (two right-most columns, green and blue boxes, respectively) for one hour, in baseline larvae (two left-most columns) that were not treated nor tracked in 96 well plates and control larvae whose swimming behavior was tracked in a 96 well plate similarly to the experiment larvae (two middle columns). Box plots for 5 mM baseline and controls are gray, for 15 mM black. (F) Box plots of startle response rates, for all response types, for the same groups as in (E) but divided by the post-seizure time (in hours), in short-term experiments.

Similarly, mean movement velocities were also higher, in a statistically significant manner (t = 15.475, df = 114.698, p = 7.456 * 10⁻³⁰) in the 5 mM PTZ group (5.39 ± 0.11 mm/sec) compared to corresponding controls (3.25 ± 0.08 mm/sec). Nevertheless, there was no significant differences between the 15 mM PTZ and 5 mM PTZ groups in mean distances (t = 1.665, df = 129.519, p = 9.841*10⁻²) nor in mean velocities (t = 1.675, df = 157.482, p = 9.589*10⁻²). These results were also confirmed with a Kruskal–Wallis test followed by a multiple comparisons procedure using the Dunn-Šidák method (accounting both for multiple comparisons and lack of normality in the distribution of velocities during seizures induced by 15 mM PTZ; χ² = 117.13; df = 3,275; 3.195*10⁻²⁵; 5mM PTZ experimental larvae to controls: z = 5.587, p = 1.383*10⁻⁷; 15mM PTZ experimental larvae to controls: z = 5.587, p = 1.383*10⁻⁷; z = 9.1; p = 0; 5 mM to 15 mM larvae: z = 1.552, p = 0.538). i.e., while both weakly and strongly seizing larvae behaved more than controls, the dynamic changes of behavior over time, rather than mean distances and velocities, must be examined to determine which seizure is stronger.

We next compared post-seizure effects between the 15 mM and 5 mM PTZ groups, during the first 3 post-seizure hours (Fig. 3E–F). Averaging over responses throughout the measurement time, we found that the weakly induced seizure was sufficient to impact escape responses after the seizure. In particular, the total response rate, of startle responses of all kinds, in the weakly seizing larvae (N = 72) was 0.33 ± 0.04 compared to 0.67 ± 0.04 in the baseline group (N = 39) and 0.5 ± 0.05 in the control group (N = 38; Fig. 3E). A Kruskal–Wallis test showed a significant effect to the condition (χ² = 28.49; df = 2,146; p = 6.500*10⁻⁷), and a significant difference between experimental larvae and both control groups (baseline controls: z = 5.251, p = 4.535*10⁻⁷; controls that have undergone behavior tracking: z = 2.728, p = 1.902*10⁻²). Moreover, the effect on post-seizure responses in weakly seizing larvae was weaker than that of strongly induced seizures, where the post-seizure total response rate was lower (0.08 ± 0.02), in a statistically significant manner (Wilcoxon ranksum test: z = 4.764, p = 1.900*10⁻⁶). We also examined the rates of SLCs (Supp. Figure 1C), and similarly found a significant effect to the condition (χ² = 29.11; df = 2,146; p = 4.777*10⁻⁷) and an SLC rate in the 5 mM PTZ group (0.2 ± 0.03) lower than the baseline (0.48 ± 0.04) in a statistically significant manner (z = 5.382, p = 2.210*10⁻⁷) although the difference from the SLC rates in behavior-tracked controls (0.31 ± 0.04) was only close to significance (z = 2.229, p = 7.532*10⁻²). In addition, SLC rates during the first 3 post-seizure hours in strongly seizing larvae (0.07 ± 0.02) were lower, in a statistically significant manner (Wilcoxon ranksum test, z = 3.803, p = 1.431*10⁻⁴) than in the weakly seizing larvae. Put together, this shows that weakly induced seizures also affect post-seizure behavior, but to a lesser extent than strongly induced seizures. Thus, quantifying post-seizure behavior can distinguish between strong and weak seizures.

Finally, in further examining the temporal post-seizure trend of weakly seizing larvae (Fig. 3F), we found that the total response rate of these larvae varied between 0.24 ± 0.07 in the 1^st hour to 0.42 ± 0.06 in the last hour (N = 18–30 in different groups), while control rates in the corresponding time points were 0.55 ± 0.12 and 0.49 ± 0.08, respectively (N = 9–16 in different groups). Moreover, while a Kruskal–Wallis analysis of response rates in weakly seizing larvae over time did not find a significant effect (χ² = 0.63; df = 2,69; p = 7.281*10⁻¹), Wilcoxon rank-sum tests showed that the response rate 1 h post-seizure was significantly lower in the experiment group compared to the control (z = 2.355, p = 1.85*10⁻²), while the rates were not different 3 h post-seizure (z = 0.734, p = 0.463). Finally, analysis of SLCs rather than total response rates lead to the same conclusion (Supp. Figure 1D). Put together, this shows that in weakly induced seizures, the impact on post-seizure startle responses is weaker, as well as more transient, with recovery occurring within the first 3 post-seizure hours.

Valproic acid has a non-specific effect on startle responses, in addition to mitigating seizure effects both during and after the seizure

Since we found startle responses to be a useful measure of the impact of seizures on the function of the nervous system, we decided to test whether startle responses themselves could identify non-specific effects of anti-epileptic drugs on brain function, in the absence of seizures. To do this, we used a previously established protocol for pre-seizure anti-epileptic drug treatment^19,20,50, treated larval zebrafish for 24 h with 1mM Valproic Acid (VPA), and examined whether startle responses were affected. Indeed, we found that larvae treated by VPA responded to startling stimuli less frequently than controls (Fig. 4A; total response rate, experiment group: 0.4 ± 0.03, N = 52; control group: 0.59 ± 0.04, N = 43; Wilcoxon rank-sum test: z = 3.552, p = 3.825*10⁻⁴). A similar reduction was observed in SLCs (experiment rate: 0.33 ± 0.03; control rate: 0.45 ± 0.04, z = 2.308, p = 2.099*10⁻²; Fig. 4B).

Fig. 4.

VPA treatment causes a reduction in startle behaviors. (A) Box plots of startle response rates, for all response types, in controls not treated with VPA (black boxes) and larvae treated with 1 mM VPA for 24 h (light red boxes). (B) Same as (A) but for SLC rates.

Next, we tracked the spontaneous swimming behavior of VPA treated and untreated larvae, with or without inducing seizures by application of 15 mM PTZ. As expected, we found VPA to suppress the increase in swimming behavior observed in larvae in which a seizure was induced by PTZ (Fig. 5A), consistent with previous studies^16,19,21. ANOVA analysis revealed a significant effect to the condition on distances averaged over the tracking window of one hour (Fig. 5B), in these experiments (F = 16.56; df = 3,140; p = 2.907*10⁻⁹). In particular, the mean distance traveled in larvae not treated with VPA, to which PTZ was applied (143.97 ± 11.14, N = 48), was higher than in all other groups (control larvae: 70.47 ± 15.08, N = 24, q = 6.086, df = 140, p = 9.904*10⁻⁵; VPA treated larvae to which PTZ was not applied: 35.62 ± 9.92, N = 27, q = 9.3241, df = 140, p = 4.015*10⁻⁹; VPA treated larvae to which PTZ was applied: 81.22 ± 9.51, N = 45, q = 6.26, df = 140, p = 5.660*10⁻⁵). Moreover, the mean distances in VPA treated larvae, with or without PTZ application, were no different than in controls (q = 2.572, p = 2.644*10⁻¹ and q = 0.881, p = 9.249*10⁻¹, respectively). This indicates that the VPA pre-treatment suppressed the seizure behavior, yet a non-specific effect of VPA on behavior was not captured by this analysis. As seizures also affected movement velocities (Fig. 1), we also compared movement velocities between the same groups (Supp. Figures 2A–B). Thus, we found velocities to also differ between conditions (F = 24.75; df = 3,135; p = 8.022*10⁻¹³). Specifically, VPA treatment slightly increased movement velocities in the absence of PTZ (4.28 ± 0.26 in treated larvae without PTZ application versus 3.07 ± 0.11 in untreated control larvae; q = 5.305; p = 1.012*10⁻³), indicating a non-specific effect of VPA. Velocities were also increased in VPA treated larvae to which PTZ was applied (4.61 ± 0.18) with respect to controls (q = 7.603; p = 4.591*10⁻⁷), but not with respect to VPA treated larvae without PTZ application (q = 1.715; p = 6.187*10⁻¹). Moreover, the movement velocity in the PTZ only group (in the absence of VPA treatment), 5.46 ± 0.17, was significantly higher than in all other groups (q = 11.98, p = 3.768*10⁻⁹ compared to untreated controls; q = 6.1, p = 9.501*10⁻⁵ compared to VPA treated larvae without PTZ application and q = 5.11, p = 1.720*10⁻³ compared to VPA treated larvae with PTZ application), again demonstrating the seizure-ameliorating effect of the VPA treatment.

Fig. 5.

Mitigation of PTZ induced seizures by VPA is apparent in behavior both during and after the seizure. (A) The distance travelled in 2.5 min intervals in untreated controls (black), controls previously treated with 1 mM VPA (light red), following the application of 15 mM PTZ at time 0 to untreated larvae (blue), or to larvae previously treated with 1 mM VPA (magenta). Shaded regions around lines indicate the standard error of the mean at each time point, lines are means of all larvae. (B) Box plots of mean distance traveled during the entire experiment, under the same conditions as in (C). (C) Box plots of response rates—for responses of all types, to startling stimuli applied within the first 3 h post-seizure, under the same conditions as in A and B.

Finally, having observed that VPA ameliorates seizure-like behavior in PTZ, we next asked whether it also ameliorates the post-seizure impact on startle behaviors, even though a non-specific effect of VPA on these behaviors was already detected. We thus examined startle behaviors after PTZ application for 1 h in the same experiment groups from above (Fig. 5C), limiting analysis to larvae whose behaviors were measured less than 3 h after convulsant washout, so as to not be affected by the initiation of recovery (Fig. 2). Comparing total response rates in all groups to the rate in controls (0.52 ± 0.08, N = 17) using a Kruskal–Wallis test (that showed a significant effect to the condition, χ² = 20.55; df = 3,86; p = 1.305*10⁻⁴) followed by multiple comparisons using the Dunn-Šidák method, only the response rate in VPA untreated larvae in which seizures were induced with PTZ (0.15 ± 0.05, N = 29) was significantly lower (z = 4.185, p = 1.708*10⁻⁴). The total response rate, of responses of all types, in larvae treated with VPA, was similar to the response rate in controls, regardless of whether a seizure was induced by PTZ (0.36 ± 0.06, N = 28, z = 1.55, p = 5.403*10⁻¹) or not (0.34 ± 0.06, N = 16, z = 1.11, p = 8.445*10⁻¹). This implies that the VPA treatment successfully ameliorated the seizures and thus the reduction in post-seizure startle responsivity was also mitigated. Limiting the analysis to SLC response rates (Supp. Figure 2C), there was again a significant effect to the condition (χ² = 15.37; df = 3,86; p = 1.526*10⁻³; and a significant difference between the VPA treated groups, with PTZ (0.35 ± 0.06) or without PTZ (0.26 ± 0.04), and the PTZ group not treated with VPA (0.14 ± 0.05; z = 3.54, p = 2.395*10⁻³ and z = 2.818, p = 2.868*10⁻²; respectively)—but no significant differences between the control rates (0.26 ± 0.05) and the rates in any other group.

Since VPA both reduced seizure behavior (in particular, travel distances) as well as ameliorated the reduction in post-seizure startle behaviors, we decided to test whether the two effects co-occur in the same larvae, giving rise to a negative correlation between excessive behavior (representing excessive neural activity) during the seizure and response rates post-seizure. This revealed only weak correlations between metrics capturing seizure behavior and post-seizure response rates, which were nonetheless consistent with stronger seizure mitigation similarly affecting seizure and post-seizure behaviors (Supp. Figure 3). In particular, post-seizure response rates were positively correlated with both the time to peak distance travelled during the seizure—which increases with a weaker and thus more gradually developing seizure (a Pearson correlation coefficient of ρ = 0.452; Supp. Figure 3A) and the time-point in which movement distances exceeded a threshold representing the mean and two standard deviations in the distribution of distances in controls (ρ = 0.169; Supp. Figure 3B), which is similarly expected to be higher in weaker seizures. Moreover, post-seizure response rates were weakly negatively correlated with the movement velocity during the seizure (ρ = − 0.167; Supp. Figure 3C), indicative of strong seizures. Put together, this shows that examination of startle responses after spontaneous swimming behavior tracking for capturing seizures, is an effective tool for identifying seizure mitigation, likely to reveal complementary information with respect to tracking spontaneous behavior during seizures alone.

Discussion

To identify possible long-term consequences of epileptic seizures, we examined behavioral escape responses to startling stimuli in larval zebrafish after PTZ induced seizures and discovered a dramatic reduction in responsivity. The reduction was transient, and full recovery was observed by 6 h post-seizure. Moreover, when the convulsant concentration was lower, and thus a weaker seizure was induced, the post-seizure effect was milder and recovery was faster. This shows that acutely induced seizures have short-term effects on the basic computations underlying a sensory-motor transformation. Furthermore, we found that treatment with a first generation anti-epileptic drug, VPA, reduced induced seizure behavior as well as post-seizure escape responses, consistent with its efficacy as an anti-epileptic drug, but also reduced responses in the absence of induced seizures. These results reveal a limitation of this drug given its seizure-independent effects, and underscore the value of the assay for improving future drug-testing in this model.

Our results, demonstrating a post-seizure effect on behavior lasting a few hours, suggest that the PTZ-induced larval zebrafish animal model presents a post-ictal state akin to that described in humans, where sensory, motor, psychological and cognitive deficits may occur during the first post-seizure hours or days^51,52. The behavioral symptoms of the post-ictal state were suggested to relate to the seizure-focus localization as well as to the neurological status of the animal prior to the seizure. The post-ictal state was also suggested to be mediated by reduced perfusion and hypoxia in brain areas affected by seizures^53–55. Our identification of a behavioral modulation at this period in the larval zebrafish model will now enable additional mechanistic studies taking advantage of the approaches uniquely available in this system, including non-invasive whole-brain activity imaging^18,23,25,56. Such studies could, e.g., dissect the roles of neural network or synaptic alterations or reveal a potential involvement of glia in this phenomenon. We further note that one mechanism that could be contributing to the phenotypes we observed is a rebound-sleep condition which was previously found to follow periods of globally increased neural activity broadly, and activity induced by PTZ treatment specifically³⁷. Such a state is consistent with epilepsy patients presenting fatigue after seizures⁵⁷. However, it cannot fully account for our observations as sleep is reversible and thus responses to salient stimuli such as the ones we presented in this study are expected to be intact. Moreover, the previously described rebound sleep condition lasted for 14 h, while we found effects to recover within 3 or 6 h, depending on the seizure strength. We believe our results of a short-duration effect of 5 mM PTZ application on escape responses, where recovery occurred within 3 h, are consistent with a short term effect on responses to dark pulses observed in the previous study following application of 10 mM PTZ, which recovered at the same time-line and was not explained by rebound sleep.

One limitation of our study is that examining post-seizure behavior in a different setup than the one in which seizure behavior was examined, with intermediate washing and incubation steps, required multiple handling steps that reduced startle response rates even in controls. Nevertheless, this can be addressed by the future design of a system for this purpose, and the post-seizure effect was strong enough to be significant beyond this limitation, even for weakly induced seizures. In addition, we were able to identify a more significant effect and slower recovery from reduced behavior after stronger seizures, compared to weaker ones, and this is consistent with studies in both humans and rodents which associated longer and more intense seizures with a slower recovery from post-ictal activity depression^6,58.

Long-term functional effects of PTZ induced seizures observable as behavior modulation were also previously studied in adult zebrafish. In particular, seizures were found to cause subsequent impairments in learning, memory acquisition and consolidation, and shoaling behavior, as well as to induce anxiety and aggression^59–63. Consistent with our findings, some of these deficits occurred immediately after seizures, and were followed by recovery. However, the timeline was generally slower, with some effects still present 24 and 48 h after seizures. Similarly, in rodents, PTZ-induced seizures, including brief seizures and status epilepticus (SE), were found to impair learning and memory (particularly fear-related and spatial memory) at a broad range of time-scales, from shortly after administration to days, weeks and months later^64–67. In one case, recovery was observed, but after 6 months⁶⁴. Similar post-seizure effects on behavior, reflecting learning and memory deficits, were identified with other models of SE, where the seizure is induced via electrical stimulation or by pilocarpine, or kainic acid application^68–70. SE was also associated with neural damage, cell death and synaptic reorganization, at the same timeline as behavioral deficits, which can thus constitute mechanisms mediating these deficits^71–73. Moreover, as we found, some studies showed seizure intensity to correlate with subsequent behavioral deficits, with SE having the most significant effects^6,74,75. Our observation of behavioral deficits post-seizure in larval zebrafish, and recovery, now opens the door for future mechanistic studies to explain both these phases.

While we found that the effect of a PTZ induced seizure on the brain is transient, lasting only a few hours, one limitation of this study is that the recovery can be attributed to either the mechanism of seizure induction, or the age of the animal in which the seizure was induced, or both—it is impossible to tell. In particular, unlike pilocarpine and kainic-acid induced SE, PTZ induced SE does not give rise to the subsequent development of spontaneous seizures and thus does not constitute a model of epileptogenesis^76–78. Thus, it is possible that the recovery of the behavior after the induced seizure is due to the transient impact of PTZ on the brain, that does not give rise to further long-lasting effects. Another consideration is that the larval zebrafish brain is immature, and thus potentially more resistant to seizure-induced injury. Nevertheless, seizure activity can induce damage in immature animals as well, and such damage, particularly when it occurs at critical developmental windows, can also have functional impacts in adulthood^79,80. Thus, it would be of interest to establish, in the future, whether the impact we observed and its transient nature is a property of the mechanism of seizure induction, comparing to seizures induced by other means, and whether a similar impact is observed in adult animals.

Consistently with our results showing VPA amelioration of seizure and post-seizure phenotypes for PTZ induced seizures, VPA is used for epilepsy treatment and seizure amelioration in humans and was previously shown to ameliorate PTZ induced seizures in larval zebrafish and rodents^{19,24,27,46,81}. Moreover, also consistently with our results, VPA pre-treatment was previously shown to ameliorate memory and learning deficits occurring after PTZ-induced seizures^67,82–84. Nevertheless, effects on more basic functions such as a sensory-motor transformation and at the larval stages were not previously examined. Moreover, our finding that VPA, in addition to reducing seizure phenotypes, causes a functional deficit in the absence of seizures, is consistent with a previous demonstration of a working memory impairment following treatment with other ASDs in adult zebrafish, which was akin to the post-seizure effect of PTZ⁶¹. VPA was also previously found to be teratogenic, causing malformations when applied during pregnancy in humans and mammals and in zebrafish embryos that were raised in media containing the compound^85–88. In adult humans, VPA was also found to sometimes cause psychiatric and behavioral side effects, albeit at low rates relative to other AEDs^89,90. Since VPA exerts its anti-epileptic effects via multiple mechanisms, including increased GABAergic neurotransmission, and inhibition of voltage-gated Na⁺ channels, Ca²⁺ channels and glutamatergic neurotransmission, this drug has multiple neuroprotective advantages but also demonstrated neurotoxic effects giving rise to tremor, drowsiness, incoordination, ataxia and aggression⁹¹. In preclinical studies, VPA treatment was shown to give rise to a spatial working memory deficit in adult rats⁹² and to cognitive and motor deficits in neonatal mice, where it was suggested to constitute an autism model⁹³. Our novel finding of an acute effect of 24 h of VPA treatment on sensory-motor processing in 6 dpf larval zebrafish, could represent either a behavioral side-effect, or an effect on development. Future studies using this assay could shed light on the type of drugs and treatment conditions under which sensory motor transformations are altered and thus facilitate the interpretation of this finding.

The larval zebrafish epilepsy model is often used for high-throughput anti-seizure drug screening. However, the main disadvantage of this assay is low specificity, requiring subsequent validation of results with low-throughput activity monitoring^39,40. Our findings suggest an intermediate step that can be incorporated in a high-throughput setup and that can further decrease the number of drug candidates requiring additional validation—the examination of startle responses induced by an acoustic (tap) stimulus. Using this approach, only anti-seizure drugs that not only reduce behavior during seizures, but also eliminate the post-seizure reduction in stimulus-triggered behavioral responses, would be selected by the screen, and these drugs will more likely constitute true positive hits. Nevertheless, we note that modulation of sensory-motor transformations may not be the only post-seizure effects to be mitigated or prevented and there could be additional cognitive impacts. Thus, this assay may still miss drugs that do not restore such transformations but otherwise positively affect the post-seizure condition, and thus future development of drug screening assays to assess additional effects would still be useful. Nevertheless, the addition of the suggested assay is expected to reduce the number of drug-candidates for low-throughput experiments and increase the information obtained, at high-throughput, regarding the effects of the putative anti-seizure drugs on the brain. Moreover, the drugs selected would be ones that do not themselves affect the performance of sensory-motor transformations—reflecting sensory or motor deficits that are not revealed by selecting drugs that reduce during-seizure swimming behavior. Thus, the assay may predict some adverse effects of drugs in humans—which reduce patients’ quality of life and may result in drug intolerability^89,94. Finally, this assay could be used to find drugs mitigating impairments induced by seizures, if not the seizures themselves, that would be useful for approximately a third of epilepsy patients suffering from drug-resistant seizures^10–13,95.

Methods

Zebrafish husbandry

Adult zebrafish were raised and maintained according to standard protocols and procedures. All experiments were performed on 6 dpf larval zebrafish from the Tübingen long-fin strain, according to protocols approved by the Technion – Israel institute of Technology animal care and use committee (IACUC, protocol number IL0820524), in accordance with relevant guidelines and regulations. Moreover, all methods are reported and described in accordance with ARRIVE guidelines 2.0⁹⁶. Larvae used in experiments were raised in an incubator at 28.5 °C with a 14-h light, 10-h dark cycle.

Behavior monitoring

Behavior was captured with an Olympus MVX10 wide-field macro-zoom microscope, using a 0.63 × magnification factor to maximize the field of view. A fast and sensitive sCMOS camera (BSI Express, Photometrics) for image capture and an Arduino board for stimulus application were controlled with micro-manager software⁹⁷.

Seizure behavior monitoring

In these experiments as well as in all experiments described below, larvae incubated together were randomly divided into control and experiment groups. The seizure behavior tracking setup is shown in Supp. Figure 4A. To acutely induce seizures and track the resulting behavior, each of 9 larvae was placed in an individual well within a 96 well-plate. In experiments where 15 mM (5 mM) PTZ was applied, the larvae were first placed with 125 μL (175 μL) of embryo medium (E3), and then 75 μL (25 μL) of a 40 mM PTZ stock made from powder (Sigma-Aldrich, P6500) were added. In controls, E3 was added instead of PTZ. Then, after 2.5 min of acclimation, behavior was monitored for 60 min. Subsequently, larvae were thoroughly washed twice in E3 and moved to a 28.5 °C lit incubator until the time of startle behavior experiments.

Startle behavior monitoring

The seizure behavior tracking setup is shown in Supp. Figure 4B. To capture responses to startling auditory/vibration stimuli, each of 3 larvae were placed in custom-made Polydimethylsiloxane (PDMS) wells (Sylgard kit, Barnaor BN4019862) of size 3.1 × 8.1 × 0.8 mm³, glued to a glass surface. The well design enabled limiting the ROI of imaging and thus obtaining a frame rate of 650–750 Hz required for capturing rapid escape movements⁴⁹. The larvae acclimated to the setup for 10 min and then were exposed to 20 startling stimuli applied at 1 min intervals. The stimuli consisted of taps, applied with an Arduino-controlled solenoid to a petri dish in which the PDMS container was placed, similarly to⁴⁹. Each behavior time-series movie consisted of 250 frames where a stimulus was triggered 50 frames after imaging started. As the frame rate was ~ 700 Hz, and the delay from stimulus triggering to application was ~ 21.5 ms (estimated by capturing the water movement induced by the tap), behavior was captured for ~ 93 ms prior to the stimulus and ~ 264 ms after the stimulus. These experiments were first conducted in 3 larvae that did not undergo seizure behavior tracking, constituting a ‘Full Control’ that the setup and larvae escape behaviors were intact in this experiment day. Then, startling stimuli were applied to larvae that underwent seizure behavior tracking—2 experiment larvae and 1 control in each experiment. Each group of 3 larvae was examined at a distinct post-seizure time, starting from ~ 45 min after convulsant washout. In experiments were post-seizure behavior was examined for up to 6 h after convulsant application, seizure behavior was captured twice consecutively in 2 groups of 9 larvae to allow subsequently examining startle behaviors throughout the time interval from 3 to 6 h post-seizure.

The response in a particular trial was excluded from analysis if the larvae was moving when the stimulus was applied making it impossible to infer whether additional movement was caused by the stimulus. In addition, all data related to specific larvae was excluded from analysis if the larvae were found to not be sufficiently developed (did not have a swim bladder) or died prior to startle behavior tracking; a handling step that can inadvertently cause injury and death. The exact numbers of larvae and trials excluded in each experiments are detailed in Supp. Table 1. Overall, exclusion rates were low and according to expectations: Exclusion rates in controls were lower than 5% and in experimental larvae (where handling is combined with PTZ application, increasing risk of death due to injuries) exclusion rates reached ~ 10%. A higher death rate occurred in experiments combining VPA treatment with PTZ application, where handling is more extensive and thus also the injury and death risk.

Anti-epileptic drug treatment

In experiments described in Fig. 4, some groups of larvae were treated with the anti-epileptic drug valproic acid (VPA). This was done by moving larvae with 135 μl into a 96 well plate, then adding 15 μl from a stock of 10 mM to obtain a final concentration of 1 mM. Larvae were then incubated for 24 h and washed twice in E3 prior to conducting subsequent seizure induction or startle behavior experiments, or both. In experiments were both seizure behavior and subsequent startle responses were examined, out of 9 larvae tested in each experiment round, 3 were controls incubated for 24 h in E3 and 6 were VPA treated larvae.

Behavior tracking and escape response classification

We developed and implemented analysis code in Matlab to extract, from seizure behavior datasets: The total distance travelled and velocity during movements in each of 25 movies capturing behavior for 2.5 min-long intervals. Similarly, from startle behavior datasets, we extracted the same kinematic parameters as well as head and tail angles and angular velocities. Moreover, we identified the time intervals corresponding to the first two stages of the C-start escape response, the C-start and the counter bend⁹⁸ (or the first two bouts of a different response), and extracted kinematic features characteristic of these stages—the response latency, the eccentricity of the larval body in the 1^st stage of the response, maximal head angular velocities, total head angle change and duration of each stage. Based on these kinematic parameters, we classified the responses to SLCs, LLCs, other or no response. The implemented algorithms are described in more detail in the supplemental methods provided as part of the supplemental information.

Statistical analysis

Statistical analysis as all other data analysis described in this paper was performed using Matlab code. The number of larvae per experiment was set such that even when dividing post-seizure responses to time bins there will be at least 10 larvae in each group for each bin. Accordingly, larval numbers for comparisons across multiple time points are higher, representing at least 3 distinct clutches and experiment days. In addition, using the results of the first set of experiments in 15 mM PTZ, we conducted a post-hoc power analysis to validate that these group sizes will suffice to reliably detect changes in responsivity. Thus, we found that with a group size of 10 larvae we will be able to detect a decrease of 0.25 in the response rate at a significance level of 0.05 and power of 0.8; and with 20 larvae a decrease of 0.17 or more would be detectable. We note that these are conservative estimates, as we assumed equal variances in the control and experiment group, even though the variance tends to decrease with response rate, increasing detectability. The effects here observed were larger than these limits and thus reached significance.

Comparisons of mean values of distances travelled and movement velocities between experiment and control groups, as well as between strongly and weakly seizing larvae, were conducted using two-tailed t-tests with unequal variances when comparing two groups and with ANOVA when comparing four groups. These parametric tests were appropriate for these data, since sample sizes were large (N = 132 and 78 strongly and weakly seizing larvae, and N = 66 and 39 corresponding controls; N = 24, 48, 27 and 45 for experiments with and without VPA treatment and PTZ application), and a normality hypothesis for the compared samples was not rejected by a one-sample Kolmogorov–Smirnov test for most groups. Exceptions included mean velocities in 15 mM PTZ which presented a heavy-tailed distribution; mean distances and velocities in the VPA-treated group to which PTZ was not applied, and distances in the PTZ group not treated with VPA. For these cases, we verified the result of parametric tests applied with a non-parametric Wilcoxon rank-sum test or a Kruskal–Wallis test followed by multiple comparisons using the Dunn-Šidák method. Since total response rates and SLC rates often presented significant deviations from normality (particularly in experimental groups where response rates were low), a non-parametric approach was applied for statistical test involving these metrics. Thus, comparisons between 2 parameters of interest were conducted using a Wilcoxon rank-sum test, applying a Bonferroni correction to control the familywise error rate when conducting multiple comparisons. Moreover, comparisons between multiple groups were conducted using a non-parametric Kruskal–Wallis test followed by multiple comparisons using the Dunn-Šidák method.

Author contributions

YE developed code for tracking, classification and quantification of seizure and startle behaviors and conducted behavior experiments where siezures were induced by PTZ. EBS conducted behavior experiments of all types and analyzed data. Noa Lavy and Lotem Bijaoui conducted behavior experiments where larvae were treated by VPA. DM developed the startle stimulation and behavior monitoring system. LF designed the research, experiments and analyses, analyzed data and wrote the manuscript.

Funding

This research was supported by the Zuckerman STEM Leadership Program.

Data availability

The data presented in this paper and code used for its analysis are available from the corresponding author upon request.

Declarations

Competing interests

The authors declare no competing interests.