The adaptor protein 2 (AP2) complex modulates habituation and behavioral selection across multiple pathways and time windows

Summary

Animals constantly integrate sensory information with prior experience to select behavioral responses appropriate to the current situation. Genetic factors supporting this behavioral flexibility are often disrupted in neuropsychiatric conditions, such as the autism-linked ap2s1 gene which supports acoustically evoked habituation learning. ap2s1 encodes an AP2 endocytosis adaptor complex subunit, although its behavioral mechanisms and importance have been unclear. Here, we show that multiple AP2 subunits regulate acoustically evoked behavior selection and habituation learning in zebrafish. Furthermore, ap2s1 biases escape behavior choice in sensory modality-specific manners, and broadly regulates action selection across sensory contexts. We demonstrate that the AP2 complex functions acutely in the nervous system to modulate acoustically evoked habituation, suggesting several spatially and/or temporally distinct mechanisms through which AP2 regulates escape behavior selection and performance. Altogether, we show the AP2 complex coordinates action selection across diverse contexts, providing a vertebrate model for ap2s1’s role in human conditions including autism spectrum disorder.

Graphical abstract

Highlights

• •Autism-linked ap2s1 promotes non-associative learning in zebrafish
• •ap2s1 biases behavioral choice across diverse sensory contexts
• •ap2s1 modulates escape behavior at multiple distinct times and/or locations

Molecular biology; Neuroscience; Behavioral neuroscience; Molecular neuroscience

Introduction

Animals must constantly integrate a variety of sensory stimuli to select the behavior best suited for survival. When threatening stimuli such as a looming shadow or an abrupt loud noise are perceived, genetically encoded mechanisms are essential for assessing the situation and initiating an appropriate response.^1,2,3 This decision-making process allows individuals to select the most appropriate behavior from their repertoire depending on the current environmental context, internal state, and prior learning.^2,4,5,6 Simple non-associative habituation learning provides experience-dependent flexibility in behavior by allowing animals to shift or reduce their response behavior following repeated inconsequential stimuli.⁷ Deficits in decision-making and habituation learning are often associated with autism spectrum disorder (ASD), attention deficit hyperactivity disorder (ADHD), anxiety disorders, and schizophrenia.^2,8,9,10 These conditions all have complex genetic underpinnings, so understanding the genetic mechanisms modulating behavior selection is critical to understanding functional neurodiversity.^11,12,13,14

Many clathrin interactome components have been connected to neuropsychiatric conditions including schizophrenia, bipolar disorder, intellectual disability, and ASD, suggesting that clathrin-dependent processes play central roles in complex behavior regulation.^15,16,17 The Adapter Protein 2 (AP2) complex targets specific membrane-associated proteins for clathrin-mediated endocytosis via its α, β, σ, and μ subunits (Figure 1A),¹⁸ and these subunits have been linked to diverse aspects of clinically relevant behavioral control.^17,19,20,21 Expression levels of the AP2β subunit (AP2B1) are altered in the brains of individuals with schizophrenia and major depressive disorder, and AP2 subunit expression is reduced in the brains of untreated individuals with bipolar disorder.^19,22,23,24 Mutations disrupting the AP2μ subunit (AP2M1) are associated with severe disruptions in cognition and behavior in humans,²¹ while mutations in the AP2σ subunit (AP2S1) are associated with behavior changes characteristic of ASD, ADHD, learning disorders, and cognitive dysfunction.^{20,25,26,27,28} We previously found that the zebrafish AP2σ homolog (ap2s1) regulates acoustically evoked behavior selection and habituation learning, indicating the conserved importance of this gene in modulating behavior across vertebrates.^29,30 AP2 is also behaviorally important for invertebrates, where glial AP2σ knockdown in D. melanogaster disrupts circadian behavior³¹ and mutations disrupting AP2α result in locomotor and coordination defects.³² In C. elegans, mutations in the AP2S1 ortholog also impair habituation and enhance sensitivity to mechanosensory stimuli.³³ Overall, the AP2 complex plays common conserved roles in learning and action selection across species.

Figure 1.

Diverse ap2s1 alleles disrupt acoustically evoked habituation learning, responsiveness, and behavior selection bias

(A) Diagram of AP2 complex, annotated with the zebrafish genes encoding each subunit. The heterotetrametric complex is composed of AP2α (ap2a1, green), AP2β (ap2b1, dark gray), AP2σ (ap2s1, red), and AP2μ (ap2m1a, ap2m1b, light gray).

(B) Diagram of ap2s1 transcript and alleles, regions contributing to AP2 target binding indicated in orange with mutant disruptions indicated in red.

(C‒H) Acoustic SLC habituation of wild type sibling (WT, blue), heterozygote sibling (Het, gray) and ap2s1 mutants (red), presented with 10 intense (23.8 dB) stimuli at 20s ISI, followed by 30 more identical stimuli at 1s ISI. Individual habituation scores were normalized to their baseline SLC response rate at 20s ISI (gray shading), and ap2s1 mutants were compared to siblings with a 2-way repeated measures ANOVA (Genotype: p < 0.0001, Time: p < 0.0001, Individual: p < 0.0001), and significant pairwise comparisons (Dunnett’s test) vs. wild type are indicated, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001. Average SLC habituation scores across 300 identical stimuli at 1s ISI with hyperbolic fitted curves (F–H).

(I‒K) Average responsiveness of individual larvae to 10 low intensity (−0.6 dB) acoustic stimuli presented at 20 s ISI, compared with two-tailed Mann-Whitney test with Bonferroni correction for multiple comparisons.

(L‒N) Relative behavioral bias of individual larvae to 10 low intensity (−0.6 dB) acoustic stimuli presented at 20 s ISI, compared with two-tailed Mann-Whitney test with Bonferroni correction for multiple comparisons. Mutants (red) and heterozygotes (gray) were always compared to their corresponding wild type (blue) siblings, using ap2s1^p172 (C, F, I, L), ap2s1^hv1 (D, G, J, M), and ap2s1^p199 (E, H, K, N) alleles. Data are represented as mean ± SEM in all panels and n indicates individual larvae. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

The AP2 complex impacts a broad array of targets and cellular processes, providing several potential mechanisms to affect behavior, for instance through developmental circuit assembly, acute neuronal function, or a combination thereof. The AP2 complex has been implicated in neurodevelopment through its regulation of cell polarity during differentiation and development,³⁴ developmental Wnt signaling,³⁵ cell migration,^36,37 neuronal survival,³⁸ axonal growth cone outgrowth and guidance,³⁹ and dendrite formation and extension.^38,40 The four AP2 subunits recognize different sets of targets and cofactors, and multiple additional vertebrate clathrin adaptor complexes allow for AP2 subunit-specific functions during these diverse processes.¹⁸ Additionally, AP2 activity can regulate acute neuronal function presynaptically through synaptic vesicle recycling^41,42,43,44 and postsynaptically by regulating internalization, recycling, and localization of neurotransmitter receptors including GABA_A receptors and ionotropic glutamate receptors in vertebrates and invertebrates.^45,46,47,48 AP2 subunits are broadly expressed in neurons, glia, and several endocrine tissues, allowing autonomous and non-autonomous nervous system regulation.^31,49,50,51 Thus, the locations and degrees to which AP2 regulates behavior developmentally in constructing neural pathways critical for behavioral selection processes, or acutely by regulating synaptic function, remain uncertain.

Here, we use a zebrafish model to determine AP2 complex requirements for behavioral selection and plasticity, leveraging their performance of discrete motor patterns selected from a relatively simple bout repertoire.^52,53 We use a set of AP2 mutants to probe escape behavior selection and habituation, showing multiple subunits of the AP2 complex are required to regulate acoustically evoked escapes. Moreover, we show that ap2s1 regulates behavior selection across diverse visual, vibrational, and spontaneous contexts, indicating a pervasive impact of the AP2 complex on behavior flexibility. Finally, through tissue- and temporally specific expression experiments we demonstrate that ap2s1 acts in neurons to modulate acoustically evoked habituation through an acute mechanism.

Results

ap2s1 modulates habituation, sensitivity, behavioral selection, and behavioral performance of acoustically evoked escape behaviors

We previously identified ap2s1 as a critical gene in modulating zebrafish acoustically evoked escape behavior and habituation.²⁹ Sudden acoustic stimuli largely evoke one of two neuronally and kinematically distinct escape maneuvers in larval zebrafish: a high-intensity stereotyped response initiating within 15 ms called the Short Latency C-Start (SLC) or a more flexible Long Latency C-Start (LLC) initiating 16–50 ms after the stimulus with kinematic intensity varying to correspond to the acoustic stimulus intensity and visual system input.^30,54,55 The original ap2s1^p172 allele identified from our genetic screen (formerly ignorance is bliss^p172)³⁰ disrupts the splice donor site of the third intron, producing an in-frame deletion removing amino acids 52–89 of the AP2σ protein (Figure 1B). We generated two additional ap2s1 alleles using the CRISPR/Cas9 system (ap2s1^hv1 and ap2s1^p199, Figure 1B) to determine if ap2s1^p172 retains some function and if other portions of the protein are also critical for regulating acoustically evoked behavior. The ap2s1^hv1 allele deletes amino acids 17 & 18 of the protein while remaining in frame, creating a disruption in close proximity to the critical Arg15 target binding residue.⁵⁶ The ap2s1^p199 allele contains a frameshift mutation after amino acid 16 that leads to an early stop codon and a severe truncation of nearly 90% of the normal AP2σ protein, likely rendering it functionally null (Figure 1B).³⁰

To test whether the new ap2s1 alleles also disrupted acoustically evoked habituation learning, we exposed larvae to 10 intense stimuli at 20s interstimulus intervals (ISI) to establish their baseline SLC escape response rate, followed by a series of 30 equally intense stimuli at 1s ISI during which SLC escape behavior robustly habituates in wild type animals.⁵⁷ Compared to their wild type and heterozygous siblings, ap2s1^p172 mutants showed reduced habituation across the 1s ISI stimuli, consistent with previous findings (Genotype: p < 0.0001, Time: p < 0.0001, Individual: p < 0.0001, 2-way repeated measures ANOVA, Figure 1C).²⁹ Similarly, both ap2s1^hv1 (Figure 1D) and ap2s1^p199 (Figure 1E) significantly impaired acoustic habituation in homozygous mutants, indicating that all three alleles disrupt the ability of AP2σ to regulate acoustic habituation learning. To determine if these habituation defects reflect a reduced rate of habituation and/or a decreased total capacity for habituation, we extended the habituation assay to include 300 identical intense stimuli at 1s ISI, calculated the degree of habituation for each individual in blocks of 10 stimuli across the assay, then fit their habituation to a hyperbolic curve (Figures 1F–1H). To assess habituation rate of the mutants, we compared the number of stimuli required to reach half-maximal habituation from the fitted habituation curves. Each of the 3 alleles produced significant increases in the number of stimuli required to reach half maximal habituation (ap2s1^p172: 10.0 vs. 3.5 for siblings, p < 0.0001; ap2s1^hv1: 90.1 vs. 5.3 for siblings, p < 0.0001; ap2s1^p199: 31.1 vs. 3.9 for siblings, p < 0.0001), indicating that all 3 ap2s1 lesions reduced acoustic habituation rate. To test total habituation capacity, we compared the maximal habituation asymptotes for sibling and mutant habituation curves. While siblings quickly plateaued to 100% habituation in each case, all three ap2s1 mutations impaired the projected maximal level of habituation reached, indicating that ap2s1 also regulates the total acoustic habituation capacity (ap2s1^p172: 95.8%, p < 0.0001; ap2s1^hv1: 93.4%, p < 0.0001; ap2s1^p199: 90.2%, p < 0.0001).

We next examined the acoustic responsiveness and acoustically evoked behavior selection bias of each ap2s1 allele to determine which lesions disrupt these behaviors, independent of habituation learning. We analyzed responses of homozygous ap2s1 mutant larvae to weak acoustic stimuli, spaced at 20s ISI to avoid habituation, and compared their response rates to their wild type and heterozygous siblings (Figures 1I–1K). Both ap2s1^p172 and ap2s1^hv1 mutants were significantly hyper-responsive to weak stimuli compared to their wild type siblings (p = 0.0136 for ap2s1^p172, p < 0.0001 for ap2s1^hv1, two-tailed Mann-Whitney test, Figures 1I and 1J), while we did not observe any responsiveness change in ap2s1^p199 mutants relative to their wild type siblings (p = 0.7344, two-tailed Mann-Whitney test, Figure 1K). We also observed an intermediate responsiveness phenotype in ap2s1^hv1/+ heterozygotes (p = 0.0432 for ap2s1^hv1/+, two-tailed Mann-Whitney test with Bonferroni correction, Figure 1J). We then examined the acoustically evoked behavior selection biases of all three ap2s1 alleles. Larvae select between their 2 major escape responses to acoustic stimuli, the SLC initiated by activating the hindbrain Mauthner neurons vs. the LLC initiated by activating hindbrain prepontine neurons, reflecting the assessed threat level of the stimulus and recent stimulus history.^30,54,55 All three ap2s1 alleles shifted the mutant response bias toward SLCs for all stimuli tested (p < 0.0001 for ap2s1^p172, p = 0.0054 for ap2s1^hv1, p = 0.002 for ap2s1^p199, two-tailed Mann-Whitney test, Figures 1L and 1N). Response bias also moderately shifted toward SLC selection in ap2s1^p172/+ and ap2s1^p199/+ heterozygotes (p = 0.003 for ap2s1^p172/+, p = 0.0138 for ap2s1^p199/+, two-tailed Mann-Whitney test with Bonferroni correction, Figures 1L and 1N). Together these results indicate that each lesion disrupts key functions of ap2s1 in modulating acoustically evoked responsiveness, behavior selection, and habituation learning.

To determine if ap2s1 impacts the kinematic performance of the acoustically evoked escape behaviors selected, we examined the response latency and several key parameters of the initial “C-bend” turns that begin all escape responses: initial turn angle, initial turn duration, and maximum angular velocity (Figure 2). We observed shortened SLC response latency (Figures 2A–2E), mildly reduced SLC turn angles (Figures 2G–2I), and mildly reduced SLC maximal angular velocity (Figures 2M–2O) in all three ap2s1 mutants compared to their siblings. In contrast, SLC turn duration was not significantly altered for any allele (Figures 2J–2L). LLC responses were less perturbed in ap2s1 mutants, though we observed increased LLC turn duration for ap2s1^p172 (Figure 2J), increased LLC maximal angular velocity for ap2s1^hv1 and ap2s1^p199 (Figures 2N and 2O), and shortened LLC latency for ap2s1^hv1 and ap2s1^p199 (Figures 2D and 2F). Together, these results reveal that all three mutations disrupt ap2s1’s function in regulating escape response kinematics, and that ap2s1 regulates SLC and LLC response kinematics in different ways. Importantly, we still observed clear and significant kinematic distinctions between SLC and LLC responses in both siblings and ap2s1 mutants. This indicates that our behavioral classification is robust to these mild kinematic changes and demonstrates that the observed effects of ap2s1 on habituation learning and behavioral bias are not through altering response kinematics, but rather due to behavioral selection.

Figure 2.

ap2s1 modulates distinct kinematic features of acoustically evoked SLC and LLC escape responses

Average movement parameters of acoustically evoked behavioral responses of sibling (blue) and ap2s1 mutant (red) larvae across 40 strong (23.8 dB) acoustic stimuli, separated by SLC (left) and LLC (right) responses.

(A‒F) Average acoustic response latencies for SLC (A,C,E) and LLC (B,D,F) behavior of ap2s1 mutants and their siblings.

(G‒I) Average maximum turning angle of the initial C-bend of ap2s1 mutant and sibling responses.

(J‒L) Average duration of the initial C-bend of ap2s1 mutant and sibling responses.

(M‒O) Average maximal angular velocity of SLC and LLC responses of ap2s1 mutants and siblings. The ap2s1^p172 allele was used in panels A, B, G, J, M (n = 260 siblings, n = 88 mutants), ap2s1^hv1 was used in panels C, D, H, K, N (n = 123 siblings, n = 40 mutants), ap2s1^p199 was used in panels E, F, I, L, O (n = 100 siblings, n = 28 mutants). Data are represented as mean ± SEM in all panels. ∗p < 0.05, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-tailed t-test with Welch’s correction.

Multiple AP2 complex subunits are critical for acoustically evoked behavior modulation

Since the four AP2 subunits can perform different functions within and possibly independent of the full AP2 heterotetramer,⁵⁸ we tested whether ap2s1’s impact on acoustic habituation and behavior selection was subunit-specific, or reflective of the function of the entire AP2 complex. We used a point mutation in ap2a1 that disrupts the splice donor site of the 13^th intron, predicted to disrupt the C-terminal Appendage Domain of AP2α, which interacts with many adaptor and accessory proteins during endocytosis (ap2a1^sa1907, Figure 3A).⁵⁹ Similar to the habituation deficit of ap2s1 mutants, acoustic habituation was compromised in ap2a1^sa1907 mutants compared to their wild type and heterozygous siblings (Genotype: p < 0.0001, Time: p < 0.0001, Individual: p < 0.0001, 2-way repeated measures ANOVA, Figure 3B). In response to weak acoustic stimuli, ap2a1^sa1907 mutants were both more responsive (p = 0.0014, two-tailed Mann-Whitney test, Figure 3C) and more strongly biased toward selecting SLC responses than their wild type siblings (p < 0.0001, two-tailed Mann-Whitney test, Figure 3D). To see if ap2a1 also influences escape response kinematics, we examined the SLC and LLC latencies, turn angles, turn durations, and maximal angular velocities of ap2a1 mutants vs. their siblings. The ap2a1^sa1907 mutation resulted in mild kinematic changes to the SLC and LLC behaviors that resembled ap2s1 disruption (Figures 3E–3I). Compared to their siblings, ap2a1^sa1907 mutants initiated SLC behavior with shortened response latencies (p < 0.0001, two-tailed t-test with Welch’s correction, Figure 3E) and a mild decrease in the initial turning angle (p < 0.0001, two-tailed t-test with Welch’s correction, Figure 3G), though the initial turn duration (p = 0.3031, two-tailed t-test with Welch’s correction, Figure 3H) and maximum angular velocity of SLC responses were not significantly altered (p = 0.2472, two-tailed t-test with Welch’s correction, Figure 3I). We also observed a shortening of the initiation latency of LLC behaviors in ap2a1^sa1907 mutants (p = 0.0002, two-tailed t-test with Welch’s correction, Figure 3F), though the maximum turning angle, turn duration, and maximal angular velocity of LLC behaviors were not significantly impacted in mutants (p = 0.9332, p = 0.2655, p = 0.1971, respectively, two-tailed t-tests with Welch’s correction, Figures 3G–3I). Nevertheless, SLC and LLC behaviors of ap2a1^sa1907 mutants remained kinematically distinguishable from each other, demonstrating that the major acoustically evoked behavior defects were due to a shift in behavior selection rather than abnormal movement patterns. Taken together, these data show that ap2a1 modulates acoustic habituation learning and behavior selection bias similarly to ap2s1, and thus these likely reflect critical functions of the AP2 complex as a whole.

Figure 3.

The AP2α subunit regulates acoustically evoked escape behavior

(A) Diagram of ap2a1 transcript and disrupted splice donor site of ap2a1^sa1907 (green). Core domain represented in orange and flexible appendage domain represented in blue.

(B‒D) Acoustically evoked response modulation of wild types (blue, n = 36), heterozygotes (gray, n = 99) and ap2a1^sa1907 mutants (green, n = 38). Average SLC habituation scores across 30 stimuli at 1s ISI (B), normalized to each individual’s baseline SLC response levels to 10 strong (23.8 dB) stimuli at 20s ISI. ap2a1^sa1907 mutants were compared to siblings with a 2-way repeated measures ANOVA (Genotype: p < 0.0001, Time: p < 0.0001, Individual: p < 0.0001), and significant pairwise comparisons (Dunnett’s test) vs. wild type are indicated, ∗∗∗∗p < 0.0001. Average responsiveness (C) and relative behavioral bias (D) of individuals to 10 weak (−6.7 dB) and 10 strong (23.8 dB) stimuli at 20 s ISI. ∗p < 0.05, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-tailed Mann-Whitney test with Bonferroni correction for multiple comparisons.

(E‒I) Average kinematic parameters of individual sibling (blue, n = 142) and ap2a1^sa1907 mutant (green, n = 37) larvae in response to 40 strong (23.8 dB) acoustic stimuli, separated by SLC (left) and LLC (right) responses. Parameters measured were the initial response latency (E-F), maximum turning angle of the initial C-bend (G), initial C-bend duration (H), and maximum angular velocity of the initial C-bend (I). ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001, two-tailed t-tests with Welch’s correction. Data are represented as mean ± SEM in all panels.

ap2s1 regulates escape behavior selection in a modality-specific manner

To determine whether ap2s1 modulates escape behavior selection specifically in acoustic contexts or more broadly across modalities, we examined the escape behavior selection of ap2s1^p172 mutants in response to lateral line dependent mechanosensory and visual stimuli. First, we exposed larvae to sudden, localized water vibrations which are sensed largely through the lateral line hair cells and can evoke both SLC and LLC escape behaviors (Figure 4A).⁶⁰ Overall, ap2s1^p172 mutants were less responsive to these stimuli than their siblings (Figure 4B), though they still performed both SLC and LLC behaviors (Figures 4C and 4D). In contrast to their behavioral bias following acoustic stimuli, disrupting ap2s1 reduced the likelihood of selecting SLC responses (p < 0.0001, two-tailed t-test with Welch’s correction, Figure 4C), while overall the rate of LLC response performance was unperturbed (p = 0.5424, two-tailed t-test with Welch’s correction, Figure 4D). As a result, the relative behavioral bias of ap2s1^p172 mutants showed a significant shift toward LLC escapes in response to local water vibrations (p < 0.0001, two-tailed Mann-Whitney test, Figure 4E). To test whether this behavioral change could be due to a change in the number or distribution of sensory hair cell neuromasts of the lateral line, we stained anterior and posterior lateral line neuromast hair cells in ap2s1^p172 mutants and siblings with FM1-43X (Figures 4F–4I), a vital dye that requires functional endocytosis and mechanotransduction to label these hair cells.⁶¹ We observed no significant difference in number or distribution of labeled anterior lateral line neuromasts (p = 0.1751, two-tailed Mann-Whitney test) or posterior lateral line neuromasts (p = 0.3914, two-tailed Mann-Whitney test) in ap2s1^p172 mutants compared to their non-mutant siblings (Figure 4J), suggesting the observed behavioral differences are due to changes downstream of lateral line mechanosensation.

Figure 4.

ap2s1 differentially regulates SLC and LLC escape behavior across varied stimulus modalities

(A and B) C-start response probability of ap2s1^p172 mutants (red, n = 26) to vibrating piezo stimuli compared to non-mutant siblings (blue, n = 56). ∗∗∗∗p < 0.0001, two-tailed t-test with Welch’s correction.

(C‒E) Within the lateral line dependent mechanosensory-induced C-start responses, the probabilities of SLC responses (C) and LLC responses (D) for siblings (blue) and ap2s1^p172 mutants (red), ∗∗∗∗p < 0.0001, ns = not significant, two-tailed t-test with Welch’s correction. Relative behavioral bias of lateral line dependent mechanosensory-evoked escape behaviors (E) for siblings and ap2s1^p172 mutants, ∗∗∗∗p < 0.0001, two-tailed Mann-Whitney test.

(F‒J) Distribution of sensory neuromasts of the anterior lateral line (F, H) and posterior lateral line (G, I) labeled with FM1-43X in 3 dpf ap2s1^p172 mutants (red, n = 11) compared to non-mutant siblings (blue, n = 38), ns = not significant, two-tailed Mann-Whitney test. Scale bar is 200μm.

(K‒M) Maximum angular velocity (L) and response latency (M) of individual behavioral responses of siblings (blue, n = 5) and ap2s1^p172 mutants (red, n = 5) to looming visual stimuli. ∗∗∗∗p < 0.0001, two-tailed Student’s t test with Welch’s correction.

(N‒O) Average SLC frequency of non-mutant siblings (blue, n = 53) and ap2s1^p172 mutants (red, n = 33) in response to visual “chasing dot stimuli.” ∗∗∗∗p < 0.0001, two-tailed Mann-Whitney test. Data are represented as mean ± SEM in all panels.

We next examined if ap2s1 regulated visually-evoked escape behavior selection, using a looming visual stimulus paradigm which, like acoustic stimuli, can evoke both Mauthner-dependent and -independent escape behaviors.^62,63 Larvae were presented with expanding black spots (“looming” visual stimuli), projected from below and locked to the instantaneous position and orientation of the fish throughout the assay (Figure 4K). Surprisingly, while these looming stimuli reliably evoked high-velocity short-latency escape responses from siblings, ap2s1^p172 mutants failed to perform any high-velocity short-latency escape responses to these stimuli (Figures 4L and 4M). To determine if this lack of any escape responses might simply indicate that ap2s1 mutant fish are blind to visual stimuli, we tested a second visual context which can evoke escape behavior: a dark spot approaching at constant speed (Figure 4N).⁶⁰ This “chasing dot” stimulus reliably evoked escape behavior in sibling larvae (63.46 ± 3.18%), while overall escape behavior of ap2s1^p172 larvae was reduced though not abolished (5.62 ± 1.96%), indicating that visual spot stimuli can indeed be detected by ap2s1 mutant larvae. Whereas siblings show a strong bias toward responding to the chasing dot stimuli with SLCs, ap2s1 mutants were much less likely to perform SLC escape behavior to these chasing dot visual stimuli (p < 0.0001, two-tailed Mann-Whitney test, Figure 4O). Taken together, these data indicate that ap2s1 plays a critical role in regulating escape behavior selection in each modality tested, though the direction of its effect on behavior bias is specific to stimulus modality.

ap2s1 broadly regulates larval behavior selection across environmental contexts

Given the modality-specific shifts in escape behavior of ap2s1 mutants, we hypothesized that ap2s1 may play a broad role in behavioral selection, beyond escape behaviors. Zebrafish larvae select between discrete movement bouts from a simple motor repertoire to navigate their environment.^53,64 We first examined the role of ap2s1 in selecting responses to whole-field illumination changes. Larvae typically respond to sudden whole field light reductions (Dark Flashes, DFs) with characteristic large-angle reorientation bouts which are independent of the acoustic escape circuitry, including O-bends and Spot Avoidance Turns (SATs).^53,65,66 ap2s1^p199 mutant and sibling larvae showed similar response rates to DFs presented at 150s ISI, indicating that mutants can detect this visual stimulus (93.15 ± 1.13% in siblings vs. 92.85 ± 2.07% in mutants, p = 0.89, two-tailed Mann-Whitney test, Figure 5A). In contrast to the decreased latency of acoustically evoked startles in ap2s1 mutants, ap2s1 mutants-initiated DF responses at significantly longer latencies than their siblings (Figure 5B). These mutant response latencies were bimodally distributed (p < 0.0001, Hartigan’s dip test for unimodality/multimodality, D = 0.114), suggesting that ap2s1 mutants shifted their DF-evoked responses to kinematically distinct longer-latency turning bouts. To distinguish between these responses, which are more kinematically continuous than acoustically evoked behavior,⁶⁷ we classified sibling and ap2s1^p199 DF responses into 13 kinematically distinct swim and reorientation bouts using a previously described approach⁵³ that we adapted for offline tracking and differences in frame rate and spatial resolution (Figure 5C). The most common turn bouts selected by siblings in response to DF stimuli were SATs (41.1%, Figure 5D), LLCs (19.4%, Figure 5E), Routine Turns (9.2% RTs, Figure 5F) and O-bends (9.0%, Figure 5G). In contrast, ap2s1^p199 mutants shifted their response profile to select significantly fewer SATs (p < 0.0001, two-tailed Mann-Whitney test, Figure 5D) and more RTs at longer latency instead (p < 0.0001, two-tailed Mann-Whitney test, Figure 5F), with no significant change in LLC or O-bend response frequency (LLC p = 0.3671, O-bend p = 0.9208, two-tailed Mann-Whitney tests, Figures 5E and 5G). To confirm that the observed change in bout selection was not caused by misclassifying mutant movements due to a general shift in movement kinematics, we compared the distributions of bouts in kinematic space of both ap2s1^p199 mutants and siblings to previous data used to assign bout categories,⁵³ finding consistent overlap (Figure S1A). Furthermore, in the same PCA space the ap2s1^p199 mutant and sibling distributions showed the specific increase in RTs and decrease in SATs affecting the full kinematic range of each bout type, outside of the region of overlap (Figure S1B). Finally, to determine if ap2s1 regulates visual habituation in this context, we presented the larvae with an additional 42 Dark Flash stimuli at 15s ISI to provoke short-term habituation of the O-bend response, which manifests as a reduction in total response rate to DFs as well as a weakening of kinematic parameters including latency.^57,66 Since we had observed differences in bout selection between ap2s1 mutants and siblings following non-habituating stimuli, we grouped all large-angle turns together and examined their habituation to repeated DF stimuli. Compared to their siblings, ap2s1^p199 mutants showed enhanced habituation of large angle turns (two-way repeated measures ANOVA, Genotype p = 0.0043, Time p < 0.0001, Individual p < 0.0001, Figure 5H) indicating that ap2s1 may normally act to decrease this aspect of visually-evoked habituation, in contrast to its function increasing acoustically evoked habituation, and thus ap2s1’s impact on short term habituation is modality-specific.

Figure 5.

ap2s1 modulates visually evoked behavior selection and habituation learning

(A) Total responsiveness (A) to 6 dark flash (DF) stimuli presented at 150 s ISI by sibling (blue, n = 234) and ap2s1^p199 mutant larvae (red, n = 79), two-tailed Mann-Whitney test.

(B) Response latency of large angle turns (>95°) by siblings (blue) and ap2s1^p199 mutants (red) following DF stimuli in A, two-tailed Mann-Whitney test, ∗∗∗∗p < 0.0001.

(C) Relative frequencies of 13 behavioral bout types in response to DFs between siblings (blue) and ap2s1^p199 mutants (red). Forward swims and turning bouts are placed roughly in order of increasing vigor, as classified by Marques et al.,⁵³ including Approach Swim (AS), Slow Swim 1 (Slow 1), Slow Swim 2 (Slow 2), Short Capture Swim (SCS), Long Capture Swim (LCS), Burst Swim (BS), J-turn (J-turn), High Angle Turn (HAT), Routine Turn (RT), Spot Avoidance Turn (SAT), O-Bend (O-Bend), Long Latency C-start (LLC), and Short Latency C-start (SLC).

(D‒G) Absolute probabilities of DF response behaviors by siblings (blue) and ap2s1^p199 mutants (red), focusing on SATs (D, Mann-Whitney U = 7842, two-tailed Mann-Whitney test, ∗∗∗p = 0.0001), LLCs (E, Mann-Whitney U = 10110, two-tailed Mann-Whitney test, p = 0.367), RTs (F, Mann-Whitney U = 6565, two-tailed Mann-Whitney test, ∗∗∗∗p < 0.0001), and O-bends (G, Mann-Whitney U = 570.5, two-tailed Mann-Whitney test, p = 0.921). See also Figure S1 for details.

(H) Habituation of large angle turn responses of siblings (blue, n = 214) and ap2s1^p199 mutants (red, n = 69) to eight blocks of six identical DF stimuli, presented at 150s ISI for the first block and 15s ISI for subsequent blocks. Habituation assessed by two-way repeated measures ANOVA (Genotype ∗∗p = 0.0043, Time ∗∗∗∗p < 0.0001, Individual ∗∗∗∗p < 0.0001). Data are represented as mean ± SEM in all panels.

To examine if ap2s1 regulates more complex behavioral scenarios where larvae must coordinate sequential patterns of movement bouts, we tested if disrupting ap2s1 impacted the optomotor response (OMR) and hunting behavior. To assess the role of ap2s1 in optomotor behavior, we tracked mutant and sibling motion when reacting to directionally moving gratings projected below. ap2s1^p172 mutants and their non-mutant siblings both move in the correct direction along with the stimuli significantly more than they would swim that direction on average (p < 0.0001 and p = 0.020, respectively, one sample t-test, Figure 6A), indicating that they can both perceive and respond to OMR stimuli. Mutants choose to move in the correct direction more often than siblings during this assay (p = 0.0216, Mann-Whitney U = 180, two-tailed Mann Whitney test, Figures 6A and 6B). To assess if ap2s1 influences goal-directed hunting behaviors we examined ap2s1^p172 mutants in a rotifer consumption assay, in which larvae must execute a complex sequence of largely visually guided behaviors to successfully track and capture prey.^68,69,70,71 While siblings consumed 15.0% of available rotifers during the 135-min hunting period, ap2s1^p172 mutants did not appreciably consume rotifers during this period, indicating deficits in visually guided hunting ability (p = 0.00012, unpaired t-test with Welch’s correction, Figure 6C). Notably, we have successfully raised homozygous ap2s1^p172 mutants past 90 days old on a rotifer and brine shrimp diet, indicating that ap2s1 mutants do retain the capacity to hunt and consume prey over longer time periods than this assay. Combined, these data reveal specific abnormalities across a diverse array of behavior selection in simple and complex visually guided contexts.

Figure 6.

ap2s1 regulates behavioral performance and choice in goal-directed and spontaneous contexts

(A) OMR score of siblings (blue, n = 16) and ap2s1^p172 mutants (red, n = 11) to moving optomotor visual stimuli, where positive scores indicate overall movement in the same direction as the visual stimuli. ∗p < 0.05, two-tailed Mann-Whitney test.

(B) Relative fish movement trajectories following the onset of each leftward (green) or rightward (orange) optomotor stimulus, depicted starting from the black cross at the center of each graph. Presented paths of siblings (left graph) and mutants (right graph) were truncated upon exiting a square region of interest during the trial (see STAR Methods).

(C) Relative consumption of rotifers by non-mutant siblings (blue, n = 23), ap2s1^p172 mutants (red, n = 11), and control arenas with no fish (black, n = 3) over a 135-min time period (∗∗∗p ≤ 0.001, two-tailed t-test with Welch’s correction, at conclusion of assay). Rotifer counts for each individual were normalized to 100% at time 0.

(D and E) Sibling and ap2s1^p172 mutant spontaneous behavior in the light (n = 52 siblings, n = 33 mutants) and the dark (n = 72 siblings, n = 18 mutants). ∗∗∗∗p < 0.0001, two-tailed Mann-Whitney test. See also Figure S2 for details. Data are presented as mean ± SEM in all panels.

Finally, we assessed whether ap2s1 impacts behavior selection during spontaneous activity in the absence of acute stimuli by recording and classifying the behavioral bout choice of spontaneously active larvae in light and dark environments, following the classification scheme of Marques et al.⁵³ In both environments, ap2s1^p172 mutants significantly shifted their selection of behavioral bouts, tending to select lower amplitude forward swims and turns (Figures 6D and S2). In both illumination contexts, ap2s1^p172 mutants showed a significant enrichment for J-turn behavior (dark: p < 0.000001, light: p < 0.000001, two-tailed Mann-Whitney test, Figure S2G) and were significantly less likely to select spontaneous routine turns (RT) than their siblings (dark: p < 0.000001, light: p < 0.000001, two-tailed Mann-Whitney test, Figure S2I). Overall, mutants performed significantly fewer bouts per minute than their siblings in both dark and light conditions (dark: p < 0.0001, light: p < 0.0001, two-tailed Mann-Whitney test, Figure 6E). Together, these results indicate that ap2s1 regulates the baseline bout preference of larvae even in a spontaneous context.

Neuronal expression of ap2s1 is sufficient for acoustic habituation

Given the broad expression of ap2s1 across neural and non-neural tissues,⁵⁰ we sought to determine the cell types where ap2s1 is important for behavioral regulation, focusing on acoustic response habituation, behavior selection, and kinematic performance. To determine if ap2s1 regulates acoustically evoked behavior through a direct neuronal function, we generated the Tg(NBT:ap2s1-gfp) transgenic line to pan-neuronally express GFP-tagged AP2σ in larvae. Transgene expression produced sustained green fluorescence broadly throughout the brain and spinal cord as early as 2 dpf (Figures 7A and 7B), allowing us to assess the behavioral impact of this neuronal tissue-specific expression.

Figure 7.

Neuronal expression of ap2s1 is sufficient for acoustic habituation and response kinematics

(A and B) 3 dpf non-transgenic larva (A) and larva carrying Tg(NBT:ap2s1-gfp) (B) expressing green fluorescence in the brain and spinal cord. Scale bar is 200μm.

(C) Acoustic habituation percentage for sibling (blue) and ap2s1^p172 (red) larvae in the final 10 of 40 intense (23.8 dB) stimuli at 1s ISI (stimuli 31–40), with or without Tg(NBT:ap2s1-gfp) as indicated. ∗∗p < 0.01, ∗∗∗∗p < 0.0001, Bonferroni corrected one-tailed Mann-Whitney test.

(D) Behavior selection bias of sibling (blue) and ap2s1^p172 (red) larvae in response to weak stimuli (−8.2 dB). ∗p < 0.05, ns = not significant, Bonferroni corrected one-tailed Mann-Whitney test.

(E‒G) Average kinematic parameters of SLC startle responses for sibling (blue) and ap2s1^p172 (red) larvae: latency (E), maximal turning angle (F), maximal angular velocity (G). ∗∗∗∗p < 0.0001, ns = not significant, one-tailed t-test with Bonferroni correction. Data are represented as mean ± SEM in all panels.

We first tested if neuronal ap2s1 expression was sufficient to rescue the habituation deficits of ap2s1 mutants. While acoustic habituation is compromised in non-transgenic ap2s1^p172 mutants compared to non-transgenic siblings (42.4% vs. 95.3%, p < 0.0001, Mann-Whitney test, Figure 7C), neuronal expression of ap2s1-gfp in ap2s1^p172 mutants significantly restored the ability to habituate (79.6%, p = 0.0029, one-tailed Mann-Whitney test, Figure 7C). These data indicate that restoring neuronal expression of ap2s1 is sufficient to modulate acoustic habituation. We next tested whether neuronal expression of ap2s1 was also sufficient to regulate acoustically evoked behavior selection. Compared to their siblings, ap2s1^p172 mutants show a significantly stronger bias toward selecting SLCs in response to low intensity stimuli (p = 0.0282, Mann-Whitney test, Figure 7D). Expressing Tg(NBT:ap2s1-gfp) did not significantly rescue the response bias of ap2s1^p172 mutants back toward LLC’s, compared to non-transgenic mutants (p = 0.1201, one-tailed Mann-Whitney test, Figure 7D), in contrast to the rescue observed for acoustic habituation. Finally, we examined whether neuronal ap2s1 was sufficient to modulate the kinematic performance of acoustically evoked responses, focusing on SLC latency, turning angle, and maximal angular velocity. Compared to their siblings, ap2s1^p172 mutant larvae initiate SLC bouts with a shorter latency (p < 0.0001, two-tailed t-test with Welch’s correction, Figure 7E), achieving a weaker SLC turning angle (p < 0.0001, two-tailed t-test with Welch’s correction, Figure 7F) and slower maximal angular velocity than their siblings (p < 0.0001, two-tailed t-test with Welch’s correction, Figure 7G). While neuronal expression of Tg(NBT:ap2s1-gfp) was not sufficient to restore the normal SLC latency in ap2s1^p172 mutants (p = 0.397, one-tailed t-test with Welch’s correction, Figure 7E), this neuronal ap2s1 expression did significantly restore SLC turn angle (p < 0.0001, one-tailed t-test with Welch’s correction, Figure 7F) and maximal angular velocity (p < 0.0001, one-tailed t-test with Welch’s correction, Figure 7G), indicating that ap2s1 regulates these kinematic aspects of acoustically evoked behavior performance through neural-specific functions. Taken together, these data indicate that neuronal ap2s1 activity is critical for regulating acoustic habituation learning and some response kinematics, though ap2s1 likely modulates acoustically evoked behavior selection bias and response latency through a different mechanism.

Acute ap2s1 activity is sufficient to regulate acoustic habituation

Since the AP2 complex has the potential to regulate neuronal function and behavior through developmental or acute activities, we tested the temporal requirements for ap2s1 to understand the mechanisms through which this complex regulates acoustically evoked behavior. We generated the Tg(hsp70:ap2s1-gfp) transgenic line to induce exogenous ap2s1-gfp expression in mutant larvae, using the heat-shock-inducible hsp70 promoter for temporal control over AP2 complex activity.⁷² A single 38°C heat shock for 15 min induced ubiquitous green fluorescence, indicating robust ap2s1-gfp expression that persisted for >24 h (Figures 8A–8C). To determine whether the AP2 complex regulates acute circuit function to control acoustically evoked behavior, we allowed mutant and sibling larvae to mature through 6 dpf, allowing development of the escape circuitry in the presence or absence of ap2s1, before inducing ap2s1-gfp expression via heat shock at 6 dpf. We then tested the acoustic habituation, behavior selection, and kinematic performance of larvae with and without the Tg(hsp70:ap2s1-gfp) transgene 12–18 h later.

Figure 8.

Acute expression of ap2s1 is sufficient for acoustic habituation

(A) Timeline for acute induction of ap2s1-gfp expression in larvae.

(B and C) 6 dpf larva carrying Tg(hsp70:ap2s1-gfp) without induction (B) and 5 h after a single 15-min 38°C heat shock treatment (C). The transgene is marked with an independent GFP marker in the heart (asterisk). Scale bar is 200μm.

(D) Acoustic habituation percentage for sibling (blue) and ap2s1^p172 (red) larvae in the final 10 of 40 intense (23.8 dB) stimuli at 1s ISI (stimuli 31–40), with or without Tg(hsp70:ap2s1-gfp) as indicated. All larvae were treated identically with a heat shock. ∗∗∗∗p < 0.0001, ∗∗p < 0.01, Bonferroni corrected one-tailed Mann-Whitney test.

(E) Behavior selection bias of sibling (blue) and ap2s1^p172 (red) larvae in response to weak stimuli (−8.2 dB). ∗∗p < 0.01, ns = not significant, Bonferroni corrected one-tailed Mann-Whitney test.

(F‒H) Average kinematic parameters of SLC startle responses for sibling (blue) and ap2s1^p172 (red) larvae: latency (F), maximal turning angle (G), and maximal angular velocity (H). ∗∗∗∗p < 0.0001, ns = not significant, Bonferroni-corrected one-tailed t-test with Welch’s correction. Data are represented as mean ± SEM in all panels.

Non-transgenic ap2s1^p172 mutants displayed habituation deficits relative to their non-mutant siblings following acute heat shock treatment (p < 0.0001, one-tailed Mann-Whitney test, Figure 8D). In larvae carrying the Tg(hsp70:ap2s1-gfp) transgene however, acute ap2s1-gfp induction at 6 dpf significantly improved the acoustic habituation of ap2s1^p172 mutants relative to non-transgenic mutants (p = 0.0058, one-tailed Mann-Whitney test, Figure 8D), demonstrating an acute impact of ap2s1 on acoustic habituation. In contrast to the rescue of habituation, after acutely inducing ap2s1-gfp we did not observe any appreciable rescue of the acoustically evoked behavior selection bias back toward LLCs (Figure 8E). Finally, to determine if ap2s1 regulates SLC kinematic performance through an acute mechanism, we tested for temporal-specific rescue of the ap2s1-dependent regulation of SLC initiation latency, turning angle, and maximal angular velocity. In the absence of Tg(hsp70:ap2s1-gfp), heat shock treated ap2s1 mutant larvae displayed shortened initiation latency, reduced SLC turning angle, and a trend toward reduced maximal angular velocity (Figures 8F–8H). Unlike for habituation, acute induction of ap2s1-gfp at 6 dpf was not sufficient to significantly improve SLC latency (p = 0.2274, one-tailed t-test with Welch’s correction, Figure 8F), turn angle (p = 0.3779, one-tailed t-test with Welch’s correction, Figure 8G), or maximal angular velocity (p = 0.3014, one-tailed t-test with Welch’s correction, Figure 8H) in transgenic ap2s1^p172 mutants, suggesting that ap2s1 regulates these kinematic aspects of acoustically evoked behavior performance at an earlier timepoint, rather than through acute functions. Together, these data demonstrate that ap2s1 acutely modulates acoustic habituation, while it may function through a distinct temporal mechanism to impact behavior selection and kinematic performance.

Discussion

Flexibility to select and perform contextually appropriate behaviors is critical for animals to survive and thrive. Here we demonstrate that the AP2 complex broadly regulates behavioral choice and learning across environmental contexts (summarized in Table S1). Human alleles of AP2S1 produce complex behavioral effects,^20,25,26,27 and while the AP2 complex plays highly conserved cell biological roles, connecting its cellular activity with resulting behavioral function has been challenging. ap2s1 mutants are viable in zebrafish, unlike in rodents and Drosophila,^73,74 providing the opportunity to directly probe the behavioral impacts of AP2 complex function through both neuron-autonomous and non-autonomous roles, distinguishing developmental and acute roles for AP2 in regulating behavior.

The AP2 complex modulates multiple aspects of acoustically evoked escape behavior

All three ap2s1 alleles impact acoustic responsiveness, habituation, behavioral selection, and kinematic performance of escape behaviors (Figure 1), highlighting the varied critical roles for the AP2 complex in modulating when, where, and how escape behavior is carried out. Regulation of behavioral responsiveness, habituation, and selection are all conceptually similar, requiring the nervous system to process whether a given escape behavior occurs. However, genetic and molecular dissociation between these different modulations of escape behavior indicate that each is subject to separate regulatory mechanisms.^{29,30,57,75,76}

Increased acoustic responsiveness is frequently observed when acoustic habituation is impaired in zebrafish, both through diverse genetic disruptions including pappaa, hip14, kcna1a, and cacna2d3,^29,77,78 and through pharmacological disruption of the NMDA-type glutamate receptor signaling with MK-801, ketamine, or 12-MDA.⁵⁷ We demonstrate that ap2s1 similarly co-regulates acoustic responsiveness and habituation in zebrafish (Figure 1). This co-regulation appears broadly conserved as orthologs of ap2s1 and many other ASD-associated genes regulate both responsiveness and habituation to mechanosensory stimuli in C. elegans.³³ The AP2 complex also critically regulates selection between alternative escape behaviors, as ap2s1 and ap2a1 mutations all shift SLC vs. LLC selection bias to favor acoustically evoked SLC responses (Figures 1 and 3). Serotonergic and dopaminergic signaling appear critical for biasing this behavioral choice, suggesting possible neurotransmitter pathways that might be subject to AP2 regulation.³⁰

The shifts in movement kinematics of ap2s1 and ap2a1 mutants reveal the AP2 complex regulates behavioral performance of escape maneuvers in addition to behavior selection. The kinematic effects on SLC and LLC escape behavior performance dissociate (Figures 2A, 2B, and 2G‒2J) and shift in opposing directions (Figures 2M‒2O), underlining the behavioral specificity of ap2s1’s functions. For example, while ap2s1 mutants show a significant delay to initiate responses to dark flashes and shift to performing lower-energy turning behaviors (Figure 5), all ap2s1 and ap2a1 mutant lines bias their acoustically evoked responses toward higher energy SLC behaviors (Figures 1L‒1N) that were consistently initiated with shorter latency (Figures 2 and 3). SLC latency shortens after ablating feedforward glycinergic inputs to the Mauthner neuron, pharmacological inhibition of glycine receptors, and cacna2d3 mutation, which all also reduce SLC habituation.^78,79,80,81 In contrast, the NMDA antagonist l-7,01,324 disrupts habituation through the feedforward excitatory spiral fiber pathway with no impact on SLC latency,⁸² consistent with our transgenic dissociation of SLC habituation and latency. Together these suggest the AP2 complex could both enhance the activity of the feedforward inhibitory glycinergic pathway to regulate acoustically evoked SLC latency and regulate the feedforward excitatory pathway to modulate habituation learning. Impaired feedforward inhibition is a proposed mechanism underlying sensory processing and response shifts in individuals with autism,⁸³ so detailing how ap2s1 impacts acoustic feedforward pathways will help model these effects in humans.

The AP2 complex uses different binding pockets and subunits to recognize and target a wide cohort of proteins for endocytic regulation, allowing for subunit specific roles.⁸⁴ Indeed, single amino acid variants in human AP2S1 can result in highly specific dysregulation of target proteins such as CaSR without disrupting the function of the overall AP2 complex.⁵⁶ However, since ap2s1 and ap2a1 mutants show quite similar phenotypes in habituation, behavior selection, and kinematic performance, we argue that this reflects disrupting the overall AP2 tetramer complex rather than independent subunit roles. We do observe some phenotypic intensity differences across the ap2s1 allele series. The extended habituation assay revealed that the ap2s1^p172 splice site mutation, which generates an in-frame protein deletion from the alternately spliced transcripts, has a milder effect on habituation rate and capacity than the ap2s1^hv1 deletion and ap2s1^p199 early truncation (Figure 1), suggesting the ap2s1^p172 protein product retains some function. An alternatively spliced ap2s1 transcript producing an identical protein change to ap2s1^p172 has been detected endogenously in humans,⁸⁵ suggesting the ap2s1^p172 splice variant may be functionally relevant. Milder substitution mutations of AP2 subunits, rather than deletions, may allow future dissection of the specific molecular targets regulating these diverse behavioral functions.

ap2s1 modulates acoustically evoked behavior through multiple distinct circuit and temporal mechanisms

Our genetic rescue approaches demonstrate that even just in the context of a “simple” behavior such as the SLC escape, ap2s1 modulates behavior through multiple mechanisms. By selectively restoring ap2s1 function pan-neuronally through development (Figure 7), or acutely following escape circuit wiring and maturation (Figure 8), we demonstrate spatially and temporally distinct roles for the AP2 complex in learning, behavior selection, and behavioral performance. We propose there are at least three temporally and/or spatially distinct mechanisms through which ap2s1 regulates acoustically evoked escape behavior: (1) acute neuronal activity, modulating acoustic habituation of SLC behavior, (2) developmental neuronal activity, regulating SLC kinematic performance, and (3) developmental non-neuronal activity, regulating escape behavior selection and response latency.

First, we propose that neuronal ap2s1 functions after developmental wiring of escape circuitry to acutely modulate the rate and maximal acoustic habituation of SLC behavior. Pan-neuronally expressing ap2s1-gfp in ap2s1 mutants was sufficient to rescue acoustic habituation learning (Figure 7B). The neuronal circuitry allowing acoustic habituation learning has already been established by 5 dpf in larvae,⁵⁷ so inducing ap2s1-gfp expression at 5 dpf dissociates any potential developmental effects of ap2s1 from acute roles in habituation. Since expressing ap2s1 across 6–7 dpf is sufficient to rescue acoustic habituation, we conclude that ap2s1 is critical for habituation after 5 dpf, and its activity is dispensable during circuit wiring for habituation. These results suggest the AP2 complex plays an active acute role in habituation learning, perhaps by promoting robust synaptic vesicle recycling to maintain behavioral inhibition or through postsynaptic receptor internalization to temporarily reduce excitatory signaling.

Second, we postulate that neuronal ap2s1 functions prior to 6 dpf to regulate aspects of SLC kinematic performance, likely during the developmental wiring of the escape and/or motor circuitry. As we observed with habituation, pan-neuronal ap2s1 expression was sufficient to rescue the weakened SLC turning angle and velocity of mutants. However, the Tg(hsp70:ap2s1-gfp) experiments reveal a functional dissociation between habituation and the kinematic performance of escape behavior, as acute ap2s1-gfp expression significantly restored acoustic habituation learning without any corresponding effect on response latency or turning angle (Figures 8D–8G). Since ap2s1-gfp was induced ubiquitously at 6 dpf, the critical window when ap2s1 regulates SLC kinematic performance is at an earlier timepoint, likely during the developmental wiring of escape circuitry. While we cannot exclude the possibility that acute ap2s1-gfp expression at 6 dpf was at an insufficient level to rescue response kinematics, this level of acute expression restored normal habituation in the same individuals with unrescued SLC kinematic performance, thus our experiments still demonstrate a clear dissociation between the functional levels and mechanisms of ap2s1’s regulation of SLC habituation and kinematics.

Third, we propose that ap2s1 functions developmentally in a non-neuronal cell type to regulate acoustically evoked behavior selection and response latency, as neither pan-neuronal, nor ubiquitous acute expression of ap2s1 were sufficient to rescue escape behavior selection bias or appropriate SLC latency. For example, ap2s1 is expressed in glial populations which could regulate formation of relevant synaptic connections,³¹ and also regulates parathyroid hormone secretion and serum ion balance by endocrine cells,⁸⁶ allowing non-autonomous effects on circuits through altered hormone or ion levels during development. While we cannot exclude the possibility that transgene expression levels were too low to rescue behavior selection or latency despite being in the right location or time, these same individuals demonstrated robust rescue of SLC habituation so the data nonetheless dissociate the mechanism of SLC habituation from behavior selection and SLC latency modulation. Examining additional candidate cell populations and inducing ap2s1-gfp expression during alternate developmental windows will be important to determine where and when ap2s1 regulates escape behavior selection and SLC latency.

AP2 broadly regulates behavior selection and learning across diverse contexts

By assessing behavioral responses across stimulus modalities and complexity, we demonstrate that ap2s1 is a broad regulator of behavioral selection and learning. ap2s1 modulates behavioral choice between distinct behavioral bouts in response to acoustic, visual, and lateral line dependent mechanosensory stimuli, as well as in spontaneous contexts. In several contexts, ap2s1 mutants shift their bias toward less vigorous movement bouts, suggesting that ap2s1 promotes selection of higher intensity behaviors in these contexts (Figures 5 and 6). Importantly, in acoustic contexts ap2s1 mutants shifted toward the shorter latency and faster SLC responses, arguing against more trivial explanations of general sickness or low energy causing the observed behavioral changes. AP2 also regulates total bout performance frequency in a context-specific way: while AP2 has little effect on DF-evoked response rate, AP2 activity enhances spontaneous and loom-evoked bout frequency, while reducing acoustically evoked response frequency. Visually evoked behaviors of DF, OMR, and prey capture assays are independent of the escape response circuitry,^65,87 thus ap2s1 likely regulates these behaviors through distinct circuit mechanisms. Weak spontaneous turns and stronger visually-evoked turns are thought to be controlled by a shared subset of spinal projection neurons of the brainstem whose activity is critical for selecting turn vs. swim behaviors,⁶⁷ making these neurons candidate loci for ap2s1’s regulation of visual and spontaneous behavior selection.

The modality-specific impacts of ap2s1 on escape behavior selection highlight multiple critical roles for the AP2 complex in information processing in the brain at diverse circuit locations. Although acoustic, visual, and lateral line dependent mechanosensory stimuli all engage the SLC and LLC escape circuitries, our data suggest that AP2 function promotes LLC selection in response to acoustic stimuli while promoting SLC selection in response to visual and mechanosensory modalities. Mechanistically, this suggests that ap2s1 regulates escape behavior selection at the level of information processing and/or integration of sensory information, rather than simply regulating the excitability of shared command and motor neurons involved in escape behavior execution. Determining how ap2s1’s temporal and spatial requirements for visual or lateral line dependent mechanosensory escape behaviors compare to those of acoustic escape behaviors will shed light on whether ap2s1 has independent roles in modality-specific processing or a more centralized role in multisensory integration.

Habituation reflects experience-driven shifts in behavior selection biases and we show the AP2 complex regulates this as well. Disrupting ap2s1 results in a dramatic deficit in habituation to repeated acoustic stimuli (Figure 1), though disrupting ap2s1 enhances the habituation of visually-evoked turns (Figure 5G), suggesting AP2 plays multiple diverse roles in short-term habituation across sensory modalities. DF habituation is mechanistically more complex than SLC habituation, also involving kinematic weakening and a progressive delay in response latency through parallel molecular pathways.⁶⁶ Since ap2s1 mutants already show a significant shift in DF response latency under non-habituating conditions, possible roles for ap2s1 in regulating these other aspects of visual habituation remain unclear. Surprisingly, the OMR directionality of ap2s1 mutants was also significantly enhanced over their siblings during the optomotor assays (Figures 6A and 6B). We speculate ap2s1 could regulate habituation to these visual stimuli and mutants then fail to reduce their responses across the assay, though more extended OMR assays with finer temporal resolution would be required to test this possibility.⁸⁸ Alternatively, biasing bout selection toward less intense bouts as observed during spontaneous swimming (Figure 6D) might help mutants maintain the correct heading.

In humans, function-disrupting AP2S1 alleles can result in learning disabilities, cognitive deficits, and/or behavioral disturbances. These alleles have been linked to ASD and ADHD in children, though the specific behavioral changes associated with these AP2S1 cohorts are just beginning to be elucidated.^25,26,28 The diverse behavioral impacts of ap2s1 disruption in zebrafish echo many behavioral phenotypes of adolescents with ASD. Increased sensitivity and enhanced behavioral responses to acoustic or mechanosensory stimuli are some of the most common sensory phenotypes of ASD^83,89,90,91 and children with ASD show reduced habituation of behavioral responses to repeated sounds.^92,93 Visual sensitivity and visually-evoked responses are often altered in populations with ASD,^94,95 where some repeated visual stimuli evoke less neural activity habituation.⁹⁶ Notably, observed behavioral shifts vary considerably across cohorts and populations with ASD, reflecting the genetic diversity underlying the autism spectrum.⁸³ Thus, the zebrafish ap2s1 model suggests behavioral phenotypes to explore specifically in AP2S1-linked ASD, helping to illuminate the breadth of its conserved roles in more complex learning and behavior selection.

Limitations of the study

In this study, we explored how, when, and where ap2s1 regulates vertebrate behavior, developing a zebrafish model to identify the roles and mechanisms through which ap2s1 modulates behavior. While the three ap2s1 alleles detailed here show very similar acoustically evoked behavioral profiles, we have not tested all ap2s1 alleles for each additional behavior assay. Thus, the different lesions might produce distinct behavioral changes in response to other stimulus modalities or in other untested contexts. We note that the zebrafish ap2s1 alleles do not precisely mimic human variants which alter the Arg-15 residue of the protein, though the ap2s1^hv1 in-frame lesion removes two amino acids in close proximity to this Arg-15 target-binding residue leaving the rest of the protein intact and we speculate this lesion might be less generally disruptive and might more closely mimic the human variants. While human phenotypes have been reported in heterozygous individuals, most zebrafish phenotypes reported here were only observed in homozygotes. Furthermore, homozygous ap2s1 mutant zebrafish are viable, unlike in rodents and Drosophila,^73,75 so it is possible that an unknown mechanism partially compensates for loss of ap2s1 function in zebrafish. However, heterozygous zebrafish do display mild intermediate acoustically evoked response rate and escape selection bias phenotypes in some ap2s1 alleles (Figures 1J, 1L, and 1N), and we cannot rule out subtle heterozygous or haploinsufficient effects of ap2s1 disruption in other behavioral contexts and sensory modalities. Thus, the zebrafish model can detect haploinsufficient behavioral phenotypes and clearly demonstrates critical roles for ap2s1 in these behavioral processes, opening the door for future studies introducing human ap2s1 missense variants into zebrafish. Finally, our transgenic rescue experiments have several caveats. Since transgenic ap2s1-gfp expression is unlikely to precisely match endogenous ap2s1 expression levels, we tested for significant improvement of behavioral parameters in ap2s1 mutants expressing the transgene vs. mutants without the transgene, rather than complete rescue. While our transgenic rescue experiments are all consistent with the three hypothesized mechanisms for ap2s1 regulation of acoustically evoked escape behavior, we cannot exclude the possibility that the failure to rescue certain behaviors might reflect a technical limitation of the approach, such as insufficient expression levels of ap2s1-gfp at a critical time or location. However, both rescue transgenes were sufficient to restore normal habituation in the same individuals where kinematic performance, behavior selection, or latency remained unrescued. Therefore, our data still dissociate the mechanisms of ap2s1’s regulation of SLC habituation from behavior selection and SLC latency modulation into multiple distinct mechanistic paths for future study.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
One-Shot TOP10 Chemically Competent Cells	Invitrogen	Cat# C404010
Chemicals, peptides, and recombinant proteins
BsmAI Restriction Endonuclease	NEB	Cat# R0529S
MscI Restriction Endonuclease	NEB	Cat# R0534S
HinfI Restriction Endonuclease	NEB	Cat# R0155S
I-SceI Restriction Endonuclease	NEB	Cat# R0694S
Cas9 Protein with NLS	PNA Bio	Cat# CP01-50
Instant Ocean Sea Salt	Instant Ocean	Product No: SS15-10
DMSO	Sigma-Aldrich	D8418
FM1-43X	ThermoFisher	F35355
Methyl Cellulose (1500 cP viscosity)	Sigma-Aldrich	M0387
Critical commercial assays
GoTaq Mastermix	Promega	Cat# M7132
MEGAshortscript T7 Transcription kit	Ambion	Cat# AM1354
MEGAclear Transcription Clean-up kit	Ambion	Cat# AM1908
In-Fusion HD Cloning Kit	Clonetech	Cat# NC1449982
Experimental models: Organisms/strains
Zebrafish: TLF : Tüpfel long fin wild type strain	ZIRC	ZFIN: ZDB-GENO-990623-2
Zebrafish: ap2s1p172	Jain et al.30	ZFIN: ZDB-ALT-150701-2
Zebrafish: ap2s1p199	Jain et al.30	ZFIN: ZDB-ALT-180129-11
Zebrafish: ap2s1hv1	This paper	ZFIN: ZDB-ALT-240305-1
Zebrafish: ap2a1sa1907	ZIRC	ZFIN: ZDB-ALT-120727-264
Zebrafish: Tg(Xla.Tubb2b:ap2s1-gfp)hv2Tg	This paper	ZFIN: ZDB-ALT-240305-2
Zebrafish: Tg(hsp70l:ap2s1-gfp)hv3Tg	This paper	ZFIN: ZDB-ALT-240305-3
Oligonucleotides
ap2s1p172 Genotyping, Forward: GTACTTCT GCATCTGTGTGG	Jain et al.30	N/A
ap2s1p172 Genotyping, Reverse: TTAAAGCTG GCTTTTGTCAAATTTG	Jain et al.30	N/A
ap2s1p199 & ap2s1hv1 Genotyping, Forward: CCAACTTTACCCCAATAAGGTTT	Jain et al.30	N/A
ap2s1p199 & ap2s1hv1 Genotyping, Reverse: CAGCATGCACCTCTTCAATTAG	Jain et al.30	N/A
gfp Genotyping, Forward: GTACTTCT GCATCTGTGTGG	This paper	N/A
gfp Genotyping, Forward: GTACTTCTG CATCTGTGTGG	This paper	N/A
ap2a1sa1907 Genotyping, Forward: CTGGTGTCAATGATGATGGC	This paper	N/A
ap2s1sa1907 Genotyping, Reverse: GAAGAGCAGAAAGCCTGACG	This paper	N/A
Recombinant DNA
pENTR/D-TOPO Vector	ThermoFisher	Cat# K240020
pDestTol2CG2	Kwan et al.97	Provided by Kristen Kwan
pCS2FA-transposase	Kwan et al.97	Provided by Kristen Kwan
Tg(Xla.Tubb2b:Has.MAPT-GFP)	Peri and Nüsslein- Volhard72	ZFIN: ZDB-TGCONSTRCT-070117-135
Software and algorithms
FasMotion	Fastec Imaging, Inc	https://www.fastecimaging.com/firmware-software/
DAQTimer Software	Burgess & Granato65	https://www.nichd.nih.gov/research/atNICHD/Investigators/burgess/software
FLOTE v2.1	Burgess & Granato 65	https://www.nichd.nih.gov/research/atNICHD/Investigators/burgess/software
Graphpad Prism 9 & 10	Graphpad Software	https://www.graphpad.com/
DeepLabCut	Mathis et al.98	https://github.com/DeepLabCut/DeepLabCut
FIJI/ImageJ	Schindelin et al.99	https://imagej.net/software/fiji/
Bout Map	Marques et al.53	https://data.mendeley.com/datasets/r9vn7x287r/1
Clusterdv	Marques and Orger100	https://github.com/jcbmarques/clusterdv
MATLAB	MathWorks	https://www.mathworks.com/products/matlab.html
RStudio	RStudio	github.com/rstudio/rstudio

Resource availability

Lead contact

Further information and requests for resources, code, and reagents should be directed to and will be fulfilled by the lead contact, Roshan Jain (rjain1@haverford.edu).

Materials availability

Zebrafish lines and plasmids developed in this study will be available on re quest from the lead contact.

Data and code availability

• •Data are available on request from the lead contact.
• •Code is available on request from the lead contact and Michael Orger (Michael.orger@neuro.fchampalimaud.org).
• •Any additional information required to reanalyze the data reported in the paper is available from the lead contact upon request.

Experimental model and study participant details

Zebrafish maintenance

All zebrafish (Danio rerio) were maintained on a 14h:10h light:dark cycle. Zebrafish larvae and embryos were raised and maintained in 1X E3 solution at 29°C unless otherwise noted. All mutant and transgenic lines were crossed into and maintained in the Tüpfel Long Fin (TLF) strain.³⁰ Sex is not determined in zebrafish until 25-60 dpf so behavioral analyses of larvae were performed on all individuals without consideration of sex. All experiments with zebrafish were approved by the Haverford College IACUC and/or the French Ministère of Research and the Ethic Committee of the Sorbonne Université and Institut Curie. Animal handling and experimental procedures conducted at the Champalimaud Centre were approved by the Champalimaud Centre for the Unknown Bioethics Committee and the Portuguese Veterinary General Board (Direcção Geral Veterinaria), in accordance with the European Union Directive 2010/63/EU.

Method details

Generation of mutants and transgenic lines

The ap2s1^hv1 mutation was generated using CRISPR as previously described for ap2s1^p199,³⁰ using a sgRNA targeting the genomic sequence 5'-GGAAGACCAGACTGGCCAAGTGG-3'. The ap2a1^sa1907 allele was a TILLING mutant generated by the Zebrafish Mutation Project¹⁰¹ and obtained from ZIRC, crossed one generation into the TLF genetic background to generate adult carriers. To generate the ap2s1-gfp transgenic constructs, full length wild type ap2s1 cDNA was amplified from TLF zebrafish larvae and cloned into pENTR/D-TOPO (ThermoFisher), inserting an in-frame 15bp flexible linker sequence (encoding the amino acids GGGGA) followed by EGFP at the C-terminus using Infusion cloning (Clontech). For the Tg(Xla.Tubb2b:ap2s1-gfp)^hv2Tg construct, abbreviated Tg(NBT:ap2s1-gfp), ap2s1-gfp was cloned into a destination vector under control of the Xenopus NBT promoter for pan-neuronal expression⁷² and followed by the SV40 polyadenylation sequence. The full transgene was flanked by I-SceI meganuclease cleavage sites, and transgenic lines were generated by coinjecting with I-SceI (NEB) as previously described.¹⁰² For the Tg(hsp70l:ap2s1-gfp)^hv3Tg construct, abbreviated Tg(hsp70:ap2s1-gfp), ap2s1-gfp was cloned into the pDestTol2CG2 vector between the heat-activated hsp70 promoter to drive inducible ap2s1-gfp expression in all tissues and the SV40 polyadenylation sequence. This transgene also carries an independent myl7:gfp marker expressing GFP in the heart, and we generated transgenic lines by coinjection with tol2 transposase mRNA.⁹⁷

DNA preparation and genotyping

Mutant ap2s1 and ap2a1 alleles were detected as previously described by Jain et al.,³⁰ where larval genomic DNA was extracted,¹⁰³ then PCR amplified with GoTaq MasterMix (Promega) and subsequently digested with a restriction enzyme to distinguish wild type and mutant alleles (see key resources table). PCR reactions were supplemented with 5% DMSO for ap2s1^p172. ap2s1^p172 alleles were distinguished by gain of a BsmAI (NEB) cutting site, ap2s1^p199 and ap2s1^hv1 alleles were distinguished by loss of a MscI (NEB) cutting site, and ap2a1^sa1907 alleles were distinguished by gain of a HinfI (NEB) cutting site. PCR amplification used T_anneal 48°C for ap2s1^p172 and 54°C for ap2s1^p199 and ap2s1^hv1 with the following program: 2 min - 94°C, 40x[30s - 94°C, 30s - T_anneal, 45s - 72°C],10 min - 72°C. Amplification of ap2a1^sa1907 used the following program: 2 min - 94°C, 35x[30s - 94°C, 30s - 55°C, 45s - 72°C], 10 min - 72°C. Transgenic larvae were molecularly verified by PCR using primers to detect gfp sequence (see key resources table) and the following program: 2 min - 94°C, 30x[30s - 94°C, 60s - 59°C, 60s - 72°C], 5 min - 72°C.

Behavioral testing and analysis

All larval behavior was assessed at 5-7 dpf, and any larvae with gross morphological abnormalities at testing time were excluded. Since ap2s1 mutants sometimes show variable delays in swim bladder inflation, fish with and without swim bladders were both included in experiments to avoid selecting against mutants. All behavioral experiments were carried out blind to genotype with subsequent genotyping of all individuals, always comparing mutants to their siblings (grouping together all genotyped wild type and heterozygous individuals, unless otherwise noted) which were identically tested and handled in the same experiment on the same days and testing times. Behavioral experiments were performed at least twice and figures represent data from single representative experiments. In most cases multiple clutches were combined together at 0 dpf to reduce the effect of any potential background effects from a single clutch. In all cases comparisons were performed within the same population and data collection conditions to avoid any confounding impacts of genetic background. For convenience, the behavioral results of this study are summarized in Table S1.

Acoustically-evoked behavior

After acclimating to testing conditions on a light box for 45min, healthy fish were placed in individual wells of a custom plexiglass testing arena, either 9mm diameter circular wells in a 36-well arena or 9mm x 9mm square wells of a 16-well arena, illuminated from below with an IR array (CM-IR200-850, CMVision), attached to a vibrational exciter (4810, Brüel & Kjær) that administers acoustic stimuli, as previously described.^30,104 Acoustic stimuli were 2ms tones (1000 Hz), varying intensity as needed for each assay. Stimuli were generated and triggered using DAQTimer software,⁶⁵ and videos were recorded in 120 ms bursts initiating 30 ms prior to each acoustic stimulus with a high-speed camera (TS4, Fastec) at 1000 fps. To test larvae for bias and sensitivity phenotypes, acoustic stimuli of varying intensities (-8.2 dB to 25.9 dB) were delivered interleaved at 20 sec ISI for 10 stimuli at each intensity. Larvae were tested for acoustic habituation by presenting 10 pre-habituation stimuli at 20s ISI (23.8 dB) followed by 30 or 300 more identical stimuli at 1sec ISI. Video files were tracked and analyzed using FLOTE,⁶⁵ using a latency threshold of 15 ms in separating SLC from LLC behavior. Larvae responding to fewer than 40% of pre-habituation stimuli were excluded from habituation analysis. SLC Habituation scores for the larvae were calculated in blocks of 10 stimuli [SLC habituation score = 100∗(current block – pre- habituation)/pre-habituation]. For the extended habituation assay, a least squares fit hyperbola was generated for each data set using Graphpad Prism 9, with "Bmax" representing the maximal habituation asymptote and "Kd" representing the half-maximal habituation.

Recording of mechanosensory, chasing dot, and spontaneous behavior

The behavioral setups used for the mechanosensory startle assay, chasing dot assay, and spontaneous swimming in the light and dark have been previously described.⁵³ The fish were recorded from above at 700 frames per second (shutter time 1423 μs) using a high-speed infrared sensitive camera (MC1362, Mikrotron-GmbH, Germany) equipped with a Schneider apo-Xenoplan 2.0/35 lens. The larvae were illuminated from bellow with a 10 x 10 cm LED-based diffusive backlight (850 nm) and the camera was fitted with a 780 nm long pass filter that blocked visible light. Fish tracking and segmentation of the tail were performed online using a custom-made algorithm in C#, as previously described.¹⁰⁵ To avoid tracking errors and kinematic changes that might result from a lack of swim bladders, only larvae with inflated swim bladders were tested in these assays.

Mechanosensory startle assay and spontaneous (dark) behavior

The mechanosensory startle assay was adapted from Groneberg et al.⁶⁰ A 2 cm long capillary rod, made of the tip of a loading-pipette (outer diameter: 0.3 mm), was attached to a piezoelectric bender actuator (Thorlabs; voltage range: 150 V; maximum displacement: 450 μm). The capillary rod was immersed 2 mm into the arena’s water and mounted at a 45° angle. Single fish were placed in circular acrylic arenas with 2.5 cm of diameter and 3 mm depth for 1h30min. The piezo deflections were composed of five 2 ms pulses at a fixed 80 V voltage with a 2 ms inter-pulse interval. To avoid habituation, stimuli were triggered with a minimum inter-stimulus interval of 2 min. To ensure that the larva was always facing away from the rod the piezo was triggered only when the larva was at a distance of 4 mm of the rod’s tip and swam to at least a 5 mm distance from it. After, there was a 400 ms delay to activate the piezo to guarantee that the fish was not performing a bout when the stimulus was triggered. To ensure that fish swam to the vicinity of the capillary rod and were able to activate it, circular sine wave converging gratings (spatial period of 10 mm, velocity 5 mm/s) centered at the tip of the piezo were presented underneath the fish by a DLP projector (BenQ). Mechanosensory-induced startle bias was calculated using the formula: 100% x (SLC frequency – LLC frequency)/(total SLC + LLC response frequency), where score of 100% results from only SLCs and a score of -100% from LLCs. The spontaneous swimming in the dark assay was performed in the same behavioral set up and arenas as the mechanosensory startle assay, but the projector was turned off ensuring that the fish was in total darkness (0 lux) and the epochs where the fish activated the piezo were excluded from the analysis.

Chasing dot and spontaneous (Light) behavior

Single larvae were placed in an acrylic circular arena with 5 cm of diameter and 4.5 mm depth. The arena is completely transparent so that visual stimuli could be displayed even when the larva was swimming in its borders and had a round bottom as animals spend less time in its borders compared with arenas with a flat bottom and vertical edges. The visual stimuli were projected by an Optoma ML 750e projector onto a diffusing screen placed 1.5mm below the larva. In the spontaneous swimming assay, a homogenous white square with 1000 lux was displayed for 20 min. In the chasing dot assay, a dark spot of 1 mm radius was projected 2 cm away from the larva and approached it with a speed of 0.5 cm/s and was updated in real time to stay locked to the fish’s position and orientation. When the dot reached the larva, it stayed underneath it for 1 s. The dot approached the larva from the left or the right, and both directions of stimuli were presented with an inter-stimulus interval of 5 min. The background during stimulus and inter-stimulus periods was the same homogenous white square displayed for the spontaneous swimming in the light assay. Chasing dot stimuli were presented 7 times from each direction per animal, 14 stimuli total.

Visual loom behavior

6 dpf larvae were adapted to the behavior arena conditions for at least 2 hours prior to the experiment, and individually tested in 35mm petri dishes in E3 media. Video was acquired at 20.2 pixels/mm at 750 Hz using a high-speed camera (MC4082, Mikrotron-GmbH, Germany) and a Kowa LM35HC 1" 35mm F1.4 lens, using an Active Silicon FireBird frame grabber board (P/N: MP-FBD-4XCXP6-2PE8). The behavioral arena was illuminated from below with an infrared LED array, and an 850 nm infrared bandpass filter (BP850-35.5, Midwest Optical Systems, Inc.) was used on the lens to block all the visible light. A custom-written C# software extracted the fish position and orientation, and updated stimulus position in real-time in a closed loop (adapted from Dunn et al.⁶²). Stimuli were projected on a plexiglass screen underneath the fish using a cold mirror and a DLP projector (Optoma). Stimuli were presented to either the left or right side of the fish, centered 0.5 cm from the center of mass of the fish. The stimulus expanded with a velocity of 250 mm/s and disappeared 6 s after the expansion began or after a high-velocity escape response (calculated based on a moving average) was performed by the larva. The analysis of the escape response induced by the looming stimulus was carried out using a custom-written MATLAB script, available on request.

Optomotor behavior

6 dpf larvae were individually tested in a 9cm clear petri dish resting on a white plexiglass sheet, using the same imaging system as Visual Loom experiments. Larvae were presented with a black/white grating of 5 mm period projected from below. The gratings moved perpendicular to the stripe direction at 5 mm/s for 30s, followed by 10s of the non-moving grating. Stimuli were presented 10x, alternating between left- and rightward stimulus movement, and larval behavior was captured at 20 Hz from above. Head position was tracked offline using DeepLabCut.⁹⁸ Analysis pipeline and path plotting was done with a custom MATLAB script, available on request. The OMR score was calculated with the formula: 100∗(distance traveled in stimulus direction – distance traveled opposite stimulus direction)/total distance traveled. Only coordinates within a cropped region of the arena were included for directionality and path plotting analysis, to exclude data points where fish encountered the edge of the dish. Fish paths that reentered the cropped arena during the trial were not depicted in Figure 6B for clarity, though they were included in OMR score. Velocities were smoothened by excluding data points exceeding a 9 pixel/frame threshold.

Dark flash behavior

Larvae were acclimated to constant illumination for at least 3 hours prior to testing, then transferred to the array of 36 individual 9mm diameter testing arenas used for acoustically-evoked behavior, lit obliquely above with a dimmable white LED (MCWHL5, Thor Labs). Behavior was recorded at 500 fps, triggered to start recording 30ms before each "Dark Flash" where the white LED was turned off for 2 seconds. All components were surrounded by a custom black vinyl enclosure to shield from external lights. 6 dpf larvae were presented with 6x “pre-habituation” dark flashes at 150s ISI, followed immediately by 42 “habituating” dark flashes at 15s ISI. Testing took place between the hours of 10:00am and 5:00pm. Videos were tracked and analyzed for response rates and response latency using Flote v2.1 ⁶⁵, and subsequent bout classification was performed through an offline bout-based tracking approach (see “Bout-based Tracking” section below).

Rotifer consumption

Imaging set-up was the same as employed in the looming and OMR experiments but without the infrared bandpass filter as a dark field illumination was created using a low-angle white-light LED ring (CCS, Japan; model: LDR2-100SW2-LA) to allow visualization of both the larva and rotifers. Image acquisition frequency was 20 Hz in 10s bursts. Sets of 60-80 6 dpf larvae were pre-fed with >1000 rotifers in 90mm petri dishes for 3 hours, then rotifers were washed out with fresh filtered E3. Individual larvae were transferred to 60mm dishes containing 5ml of 5g/L sea salt (Instant Ocean Salt) and starved for 24 hours. Starved 7 dpf larvae were given approximately 100-120 rotifers per fish per dish to hunt. Starting immediately upon addition of rotifers (t= 0 min), 10s bursts were recorded from each dish at five time points across 135 minutes of the experiment. Each individual larva was saved for subsequent genotyping. Videos were analyzed using “Find Maxima” in FIJI,⁹⁹ counting the total number of visible rotifers in each frame of each video. Since some rotifers could move in and out of video detection at the edge of the chamber, the frame with the largest number of particles was used to measure a single rotifer count for each time point from, subtracting the number of detected particles that represented parts of the fish larva rather than rotifers. Rotifer counts were normalized to 100% against the initial prey count at t= 0 min for each fish to assess rotifers consumed by each larva across the assay.

Bout-based tracking

The swim bout type classification has been described in detail by Marques et al.⁵³ Bouts and tail half-beats were detected and this information was used to compute, for each swim bout, 73 kinematic parameters as described previously.⁵³ This information was used to embed the data into a previously computed principal component analysis (PCA) space that was obtained from a data set of 3 million bouts that fish executed over a wide range of behaviors and stimuli.⁵³ To categorize the bouts into types, we used a dataset composed of equal number examples of each bout type (Bout Map),⁵³ that were categorized using density valley clustering (clusterdv),¹⁰⁰ and used k nearest neighbors (k = 50) over the first 20 principal components to assign every new bout to one of the possible 13 categories. To apply the same analysis pipeline to the dark flash data acquired using a different set-up and frame rate, an offline version of the tracking algorithm from Marques et al.⁵³ was applied to the recorded movies, and the data was interpolated to 700 frames per second. To visually compare the kinematic distributions in mutant and sibling groups, scatter plots were made using the first two PCA dimensions from Marques et al.⁵³ Changes in the distribution of kinematics between groups were visualized by taking the difference between kernel density estimates based on 2D Gaussian kernels (Matlab ksdensity function) in the first two PCA dimensions.

Neuromast staining

The fluorescent dye FM1-43X (ThermoFisher), a fixable analog of FM1-43, was used to stain functional larval lateral line neuromast hair cells.⁶¹ A fresh FM1-43X stock solution dissolved in distilled water (817μM) was diluted to 3μM final concentration in E3 embryo medium. 3dpf larvae were stained in 3μM FM1-43X for 20 seconds, then immediately anesthetized in Syncaine (MS-222, Syndel) diluted in E3 embryo medium for 10 minutes and mounted laterally onto a glass slide in 4.0% methyl cellulose (M0387, Sigma Aldrich). Fish were immediately imaged using a Nikon dissecting microscope (Nikon SMZ1500) and camera (Nikon DS-Ri1) equipped with a fluorescent lamp illuminator (X-Cite Series 120Q) fitted with an appropriate emission filter (Ex 450-490 nm, Em >500 nm). Anterior Lateral Line neuromasts of the head were counted on only one side of each larva, in the stereotyped SO, O, OC, D, M, IO, OP, and MI regions of the head (as described by Harris et al.¹⁰⁶). Posterior Lateral Line neuromasts of the trunk and tail (P1-P9, as described by Harris et al.¹⁰⁶) were scored along the horizontal myoseptum, including neuromasts on both sides of the tail as these were all clearly visible with a minor adjustment in focus. All larvae were scored blind to genotype, and scores were subsequently matched to genotypes post-scoring.

Transgenic rescue

Pan-neuronal rescue of ap2s1 was carried out using an incross of Tg(NBT:ap2s1-gfp); ap2s1^p172/+ adult zebrafish. Larvae were genotyped for both ap2s1 and the transgene after behavioral testing. Acute rescue of ap2s1 was carried out using an incross of Tg(hsp70:ap2s1-gfp); ap2s1^p172/+ adult zebrafish. Larvae were pre-screened for transgene presence through heart fluorescence from the myl7:GFP marker on the transgene, and ap2s1 genotype was molecularly determined after behavior testing. The hsp70 promoter is ubiquitously inducible by abrupt heat shock, though as previously described the promoter also drives targeted expression in the lens without heat shock induction.¹⁰⁷ To induce ap2s1-gfp expression, unanesthetized larvae were transferred to individual wells of a PCR plate in 100μl E3 to receive a 38°C heat shock for 15 min. Following heat shock, larvae were returned to 6cm dishes in fresh E3 at 29°C. Transgenic fish were always compared with non-transgenic siblings that also received the same heat shock regimen to control for background and handling.

Quantification and statistical analysis

Statistical analyses

2-tailed Mann-Whitney tests were used to compare end-point habituation, bias, and responsiveness scores, where floor or ceiling values produced non-Gaussian distributions (GraphPad Prism 9). 2-tailed unpaired t-tests with Welch’s correction for unequal variance were used to compare kinematic parameters (GraphPad Prism 9). 2-tailed Mann-Whitney tests were used to compare lateral line neuromast counts between genotypes, using Dunn’s test to correct for multiple comparisons (GraphPad Prism 10). For experiments where bout types were categorized (mechanosensory startle assay, chasing dot assay, spontaneous swimming in the light and dark) we applied a two-tailed Wilcoxon signed-rank test (MATLAB: ranksum), and adjusted p-values were corrected for multiple comparisons using the Holm-Bonferroni method. Two-way repeated measures ANOVA analyses were used to assess the significance of differences across time and genotype for response likelihood measures of the acoustically- and Dark Flash-evoked habituation assays. R Studio was used to compute Hartigan’s dip test for multimodality on Dark Flash latency distributions. 1-tailed statistical tests were used in assessing transgenic rescue, as these tested for a phenotypic shift in a single direction. The Bonferroni correction was used to conservatively account for multiple comparisons with single datasets. Graphs display mean ± SEM, using the statistical abbreviations ns: p>0.05, ∗p<0.05, ∗∗p<0.01, ∗∗∗p<0.001, ∗∗∗∗p<0.0001, unless otherwise indicated. Statistical details are noted in each figure’s respective legend.

Published: March 8, 2024