Sex-specific neurological dysregulation may underpin distinctly different male and female behavioural phenotypes in a zebrafish model of autism spectrum disorder

Abstract

Background

Although distinctly different phenotypes of male and female autism spectrum disorder (ASD) have been long proposed, mechanistic insights are relatively lacking. Added complexities are that human ASD neurological data has been predominantly generated in males, while behavioural observations are influenced by social norms. Using a preclinical in vivo model of idiopathic ASD, the current study may contribute to our understanding of the differences between the male and female ASD phenotype.

Methods

Briefly, ASD-like phenotype was induced in zebrafish embryos via valproate immersion. After confirmation of the larval phenotype, zebrafish were raised to adulthood to allow for sex-specific assessments. Adult behaviour was assessed in terms of anxiety (novel tank test), social interaction (social preference test) and aggression (mirror biting test). Behavioural data was interpreted in the context of whole brain proteome profiles obtained by untargeted proteomics. Differential protein expression analyses were performed using the Benjamini-Hochberg false discovery rate (significance at < 5%). In addition, main effects of ASD-like phenotype and sex were evaluated for specific neurotransmitter proteins.

Results

Behavioural data generally illustrated convergence of sexes in the ASD-like groups. For example, ASD-like males demonstrated a similar but exaggerated anxiety-like outcome, while ASD-like females exhibited behavioural responses more like control and ASD-like males within the social preference assay. Current data do not support aggressive behaviour as a hallmark of ASD in this model. In terms of neurotransmission profiles, significant sex-specific dysregulation was observed within the glutamatergic, GABAergic and dopaminergic systems. For example, greater excitation/inhibition imbalance was exhibited in ASD-like males vs. ASD-like females. Due to low abundance, data on the serotonergic system is less conclusive.

Conclusion

Despite similar behavioural profiles, distinct neurotransmitter mechanisms elucidated may potentially warrant the consideration of sex-specific therapeutic targets in the ASD context.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12993-026-00353-x.

Introduction

Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder with increasing worldwide prevalence – recently estimated to be approximately 1% in the global population [82], but up to 3% in children [70]. Characterized by impairments in social interaction and communication, alongside repetitive patterns of behaviours [46], ASD is associated with dysregulated sensory stimulation reactivity and varying degrees of cognitive impairment. Despite substantial research efforts, the fact that no ASD-specific pharmaceuticals have been developed to specifically target core ASD symptoms is testament to the large gaps that still exist in our understanding of this multifaceted disorder.

A striking feature of ASD epidemiology is the marked male predominance, with males being diagnosed 3–4 times more frequently than females across a range of case ascertainment methods [49]. Although increasing recognition of underdiagnosis and delayed diagnosis in females has led to some re-evaluation of this sex bias, the consistent male predominance suggests that some biological factors may contribute to sex-specific ASD vulnerability. For example, sex hormones, are known to influence brain development [43] and may contribute to sexually distinct ASD neurobiology. Although definitive evidence in support of already decades old behavioural theories of sex differences in ASD [6, 41, 74] remains limited, some evidence suggest that males and females may differ both in ASD presentation and underlying biology. For example, in humans with ASD, sex-specific neuroanatomical differences have been described, albeit in the absence of links to ASD symptomology [44]. Furthermore, the fact that approximately 92% of the gene mutations associated with ASD risk are autosomal (https://gene.sfari.org/) and that 90% of ASD cases are estimated to be idiopathic (i.e. not linked-to specific gene mutations) [29] suggest that sex differences in ASD are unlikely to be explained at genetic level alone. Rather, it is more likely that males and females exhibit differential vulnerability to distinct biological mechanisms of ASD [12, 79]. In support of this possibility, distinct sex differences in ASD-related social skills, e.g. female with ASD manifesting with less overt repetitive behaviours and more subtle social communication difficulties than males [47], underscore the critical need to understand the molecular basis for these observed differences.

Addressing this knowledge gap required a double pronged approach to inform our understanding of sex-related susceptibility: firstly, by linking sex-specific behavioural outcomes to neurobiological profiles and secondly by comparing non-ASD vs. ASD male/female differences comprehensively using an -omics approach. The use of preclinical models (including those established in zebrafish, rodents and non-human primates) are essential tools with which to elucidate disease mechanisms associated with ASD-like phenotypes [14, 81]. Unlike clinical studies, where behavioural outcomes are often influenced by complex environmental and social factors, preclinical models allow researchers to more readily distinguish primary pathological processes from secondary maladaptive consequences. Furthermore, these models facilitate direct integration of behavioural and molecular endpoints within controlled experimental settings and comparable genetic backgrounds, thereby improving mechanistic interpretation.

The current study employed the well-established valproic acid (VPA)-exposure zebrafish model of ASD. In contrast to genetic mutation models of ASD based on specific-ASD associated mutations (e.g. SHANK3, CNTNAP2, FMR1/MECP2 models), the VPA model reproduces a broad range of ASD-like phenotypes and is considered representative of idiopathic ASD-associated changes across multiple species [14, 50, 81]. As such, insights gained from this model may have broader translational relevance than those resulting from single gene mutation models. The aim of this study was to integrate behavioural phenotyping and whole-brain proteomic profiling, in a VPA zebrafish model of ASD, to elucidate potential sex-specific neurobiological alterations and determine how these relate to ASD-like behavioural outcomes.

Methods and materials

Ethical considerations

All experimental protocols received ethical approval from the Research Ethics Committee for Animal Care and Use at Stellenbosch University (Ref: ACU-2025-29984) and aligned with the South African National Standard for the care and use of animals for scientific purposes (SANS 10386:2021). All experiments were conducted in the Zebrafish Husbandry and Research Unit (Department of Medicine, Stellenbosch University; South African Veterinary Council (SAVC) registration number RF23/16441) by or under supervision of South African Veterinary Counsel authorised para-veterinarians (SAVC authorisation numbers AR23/22357 and AR21/19296).

Zebrafish husbandry and induction of ASD-like phenotype

Following standard breeding protocols, commercial wild-type (PET) zebrafish embryos were collected and randomly distributed in petri dishes (n = 50 in 30 mL embryo media; E3) and maintained in an incubator at 28.5 °C with a light/dark cycle of 14 h/10 h as previously described [61].

The VPA exposure model [84] was employed to induce an ASD-like phenotype. Briefly, embryos from three independent spawning events were randomly divided into experimental groups (control and ASD-like). ASD-like groups were exposed to 37.5 µM VPA (P4543, Sigma-Aldrich) in E3 via immersion during early development (4 to 52 h post fertilisation, refreshed after 24 h).

To facilitate assessments in ASD-like adult zebrafish, control and VPA-exposed larvae were reared to adulthood following standard husbandry procedures. Briefly, at 5 days post fertilisation (dpf), larvae were transferred to stand-alone tanks at low density (10 fish/L) where they were maintained in conditioned reverse-osmosis water that mimicked system water conditions. Once the larvae were of adequate size (30 dpf), they were transferred to an iWaki LAbREED^® Benchtop system where they were housed at a density of 1 fish/L for the duration of the study. All behavioural assessments on adult zebrafish were conducted at 6 months post fertilisation (mpf), after which the same fish were utilised for proteomic analysis at 7 mpf.

Larval behavioural assessments

In line with a previous report of the use of the light-dark transition test (LDTT) for confirmation of successful induction of an ASD-like phenotype in the VPA exposure model [7], a 40 min LDTT protocol was performed as previously described [30], to assess changes in locomotion at 5 dpf. Briefly, total locomotive activities during different lighting condition phases, as well as the magnitude of response to dark-to-light and light-to-dark transitions were quantified (Fig. 1).

Fig. 1.

Representative locomotion activity track during the light-dark transition test (LDTT). The 40-minute LDTT protocol consists of four 10-minute events with instantaneous changes in lightning conditions. These included: (i) an acclimation phase (lights off), (ii) basal activity phase (lights off), (iii) bright light-phase (lights on at 100% of on-board bulb capacity), and finally (iv) a dark phase (lights off). Locomotive responses were quantified as the total distance moved during a particular phase, as well as the responsiveness to the light stimulus – startle response – (minute 20 to 21) and light-to-dark transition – visual motor response – (minute 30 to 31)

Adult behavioural assessments

The current study aimed to assess exploratory/anxiety-like, social interaction and aggressive-like behaviours, and included the novel tank test (NTT), social preference test (SPT) and mirror biting test (MBT) respectively [19, 26, 42, 56, 57]. For all tests, adult zebrafish were habituated to the experimental room for 1 h prior to analysis, and swimming behaviours were recorded for 5 min. The tank(s) set up and dimensions for each test, as well as the respective zone allocations for quantification are presented in Fig. 2.

Fig. 2.

Set up of adult behavioural assessments. Purple planes indicate virtual zoning grids implemented for quantification of behavioural responses. The A novel tank test (NTT) was performed in a 14 × 14 × 14 cm tank with defined top (upper half) and bottom (lower half) zones. The B mirror biting test (MBT) was performed in a 28 × 14 × 14 cm tank, with an interactive zone indicated near the mirror. The C social preference test (SPT) was performed in a 28 × 14 × 14 cm tank flanked by two 14 × 14 × 14 cm tanks (separated by removable screens). Asocial and social zones were delineated as indicated. All tanks were filled to the 10 cm mark with system water, which was refreshed after every fourth assessment

The NTT was performed as previously described [33] with slight modifications in tank size. Swimming behaviour recording was initiated as soon as the fish entered the tank. Parameters assessed included the total time spent in both top and bottom zones, the latency to first enter the top zone and the number of transitions between zones.

The SPT was performed as previously described [3] with slight modifications in tank size. Three age and sized matched conspecifics (used consistently throughout all assessments) were placed into the conspecific tank, while the other tank was left empty. The test fish was individually placed into the central tank and allowed to acclimatise for 5 min before the dividers were removed (revealing the conspecific and empty tanks) and swimming behaviour recorded for 5 min. Side bias was controlled for by randomising the side containing the conspecific fish. Parameters assessed included the total time spent (TTS) in both social and asocial zones, as well as the frequency of transition into these zones.

The MBT was performed as previously described [60]. The test fish was placed into the assessment tank and allowed to acclimatise for 5 min before the divider was removed (revealing the mirror) and swimming behaviour recorded for 5 min. Parameters assessed included the TTS in the interactive zone, the number of zone entries, the latency to first entry into the interactive zone and the number of mirror bites.

Sex determination for all behavioural assays was based on morphology. Each behavioural assay was completed on the same day across all experimental groups. All fish were subjected to every behavioural assay; however, consecutive tests were separated by one week to minimise the potential effects of cumulative stress.

In terms of data processing, video files were analysed with EthoVision XT analysis tracking software, using a virtual grid to divide tanks into zones required for each test. Prior to analysis, detection settings were optimised to ensure accurate recognition of the fish in each video, and all tracks were visually inspected to confirm that the software reliably followed the fish’s movement. All behavioural analyses were conducted under blinded conditions, with video identifiers and treatment allocations unblinded only after completion of the analysis.

Proteomic analysis

Sample collection

Adult zebrafish were euthanised via the rapid cooling method [78] and subsequently decapitated. Sex assignments made prior to behavioural analysis were confirmed by gonadal inspection at the time of tissue collection. Whole brains were dissected out [32], flash-frozen in liquid nitrogen and immediately stored at -80 °C. Whole-brain tissue (rather than isolated brain regions) was selected for proteomic analysis since sufficient protein could be obtained from a single whole brain, eliminating the need to pool samples across individuals. Avoiding sample pooling was considered important, as it preserved individual biological variability. Subsequent sample processing and proteomic analyses were completed by D-CYPHR (https://www.d-cyphr.org.za/).

Sample extraction

Briefly, triethylammonium bicarbonate (TEAB) was added to each sample (50 µL of 50 mM/0.01 g sample) prior to homogenisation with a PIXUL multi-sample sonicator (Acrive motif, USA). Extracted proteins were quantified (BCA assay) in the resultant supernatant (cellular debris removed via centrifugation). From each sample, 20 µg of protein was transferred to a LoBind plate (0030504.100, Merck), before undergoing reduction and alkylation via the addition of 20mM dithiothreitol (D9779, Sigma-Aldrich) and 30 mM iodoacetamide (I6125, Sigma-Aldrich) respectively. For this step, equal volumes as that of the sample volume were added and samples were incubated at 95 °C for 10 min. The HILIC magnetic bead workflow was performed according to manufacturer’s instructions (HLC010, ReSyn Biosciences). Briefly, once the beads were prepared, they were added to the samples (5:1 v/v total protein) and placed on a shaker for 30 min to promote protein binding. Following this, the beads were washed 4x with 550 µL of 95% acetonitrile (ACN). Protein digestion was facilitated via the addition of trypsin (PRV5111, Promega) in TEAB at a ratio of 1:20 total protein, as well as LycS (90307, Pierce) at a ratio of 1:250 total protein. To facilitate digestion, the samples were incubated at 45 °C for 2 h on a shaker. Resultant peptides were removed and acidified (via addition of 5% formic acid (FA; 85178, Thermo Fisher Scientific) to a final concentration of 0.5%). Approximately 150 ng of each sample was loaded on Evotips (EV2013, Evosep) according to manufacturer’s instructions.

Liquid chromatography mass spectrometry (LC-MS)

A timsTOF Pro 2 mass spectrometer (Bruker, USA), integrating trapped ion mobility separations and coupled online via a Captivespray electrospray source (Bruker) to an Evosep One LC system (Evosep Biosystems, Denmark), was utilised for LC-MS analysis. Software for both systems were: timsControl Setup V: 5.0.9_e3816ef_3 and Bruker Compass HyStar 6.2 V6.2.1.13. Peptides were separated on a PepSep C18 column (150 μm x 8 cm, 1.5 μm; 1893470, Bruker) maintained at 40 °C. A pre-formed linear gradient of solvents A (0.1% FA) and B (100% ACN/0.1% FA) was utilised to achieve separation over 21 min (standard 60SPD Evosep One method). The MS was operated in positive ion mode and data acquired using the standard DIA PASEF-short gradient acquisition method (25 Da mass windows, 21 mass steps per cycle, 0.85 to 1.30 V-s/cm2 CCS window, 100 ms TIMS window).

Data processing

Initial data analysis was conducted using the directDIA™ (deep) analysis workflow in Spectronaut (version 19.1.240806.62635). Searches were conducted against Danio rerio reference proteome database obtained from UniprotKB, downloaded on 12/04/2024. All processing parameters were kept as standard unless otherwise stated. The enzymes for digest were set as trypsin/P and LysC/P. The variable modifications were set as acetylation of the N-terminus of the protein, deamidation of glutamine and asparagine and oxidation of methionine. A maximum number of 5 variable modifications was allowed. A fixed modification (carbamidomethyl) was added for alkylation of cysteine (C) by iodoacetamide (composition C(2)H(3)NO with a mass of 57.021464 Da.

In terms of data-preprocessing and quality control, raw protein intensity matrices were imported into R (V 4.5.2) and filtered. Proteins with high missingness (retained if  ≥  50% of values were non-missing per group or globally) or near-zero variance (≤ 1 × 10^− 3) were removed. Protein-level outliers were identified using IQR-based thresholds (multiplier = 3). Remaining missing values were imputed using a hybrid strategy to account for both missing-not-at-random (MNAR) and missing-at-random (MAR) mechanisms. MNAR values were imputed using left shift imputation (Perseus defaults utilised), while MAR values were imputed using k-nearest neighbour imputation (k = 10). Protein values were log₂ transformed following the addition of a pseudo count of 1 and samples were median centred for normalisation.

Data analysis

Principal component analysis (PCA) was performed to visualise data clusters and identify outliers. Mahalanobis distances were calculated from the first five principal components and assessed using chi-squared distribution. Samples with p < 0.01 were classified as outliers and removed from downstream analyses. PCA plots were generated and 95% confidence ellipses overlaid.

Differential protein expression analyses were performed for the following groups: control male vs. control female, ASD-like male vs. ASD-like female, control male vs. ASD-like male, and control female vs. ASD-like female. For each comparison, protein-wise differential expression was assessed using two-sample student t-tests. Raw p-values were adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR), with proteins considered significantly differentially expressed if FDR < 5%. Log₂ fold changes were calculated as the difference in mean abundance between groups. Volcano plots were generated for each comparison.

Sample size estimations and statistical analysis

Sample size estimations for larval behavioural analyses were based on previous reports in literature [7, 84]. Specifically, we assessed n = 48/group for each of the three independent spawning events. Sample size estimations for adult analyses were based on data obtained from a preliminary pilot study which yielded relatively large effect sizes (Cohen’s d = 1.02–2.00). Using an average effect size (1.5) and a statistical power of 80% (α = 0.05), an a priori power analysis (G*Power) indicated that a minimum sample size of n = 9/group would be sufficient to detect significant differences. To account for potential sample loss and the influence of technical and biological variability, which has been reported in similar studies, the sample size was increased to n = 12/group. Importantly, the same fish were used for both behavioural and proteomic analyses, allowing direct comparison between behavioural phenotypes and underlying molecular signatures.

Unless elsewhere described, statistical analyses were conducted using GraphPad Prism v10.6.1 (http://www.graphpad.com). Data was assessed for normality using the Shapiro-Wilks test. For all larval and adult behavioural data, outliers were identified using the ROUT test (Q = 1%). No outliers were removed from the larval analysis, and three outliers were removed from the adult behavioural analyses. For comparisons between two groups (larval studies), parametric data were analysed with unpaired t tests (similar variances) or Welch’s t tests (dissimilar variances) and presented as mean ± standard deviation (SD), while non-parametric data were analysed with Mann Whitney U tests and presented as medians ± 95% confidence intervals (CI). To evaluate main effects of the ASD-like phenotype and sex (adult studies), 2-way ANOVAs were performed, with Fisher’s LSD post hoc tests to indicate differences between groups. Results were considered statistically significant at p ≤ 0.05, but to remain conservative with our interpretation of the proteomics data, only findings with p < 0.01 were interpreted.

Results

No change in 24 h survival following low dose VPA-exposure was evident (median survival of 94% in non-exposed vs. 93% in VPA-exposed groups). However, a slight delay in hatching (57% vs. 36% hatched on 2 dpf) was observed in the VPA-exposed group, suggestive of developmental deferment. In terms of effect on locomotor activity in early larval stages (5 dpf), VPA-exposure significantly (i) reduced basal activity levels, (ii) intensified the startle response to a bright light stimulus, and (iii) reduced the ability to recover in the presence of a bright light stressor, but did not change the visual motor response to light-to-dark transition or hyperlocomotion response during the dark phase (Fig. 3).

Fig. 3.

Locomotive responses during the light-dark transition test comparing non-exposed (control) and VPA-exposed (ASD-like) larvae at 5 dpf. A Qualitative changes in activity levels over the 40-minute protocol are presented as the average distance moved per minute per group. Subsequent quantification of the total distance moved (TDM) in the B basal, D light, and F dark phases, as well as the magnitude of the C startle response, and the E visual motor response following lighting condition transitions is presented. Data are presented individual scatter plots with median ± 95% CI, n = 144/group. Statistical analysis: Mann Whitney test, ** p < 0.01 and **** p < 0.0001

VPA-exposure did not cause observable alterations in adult standard length or total body mass, and the ASD-like phenotype did not alter the magnitude of male vs. female differences in body mass (Fig. 4).

Fig. 4.

Sex-specific adult metadata comparing control and ASD-like zebrafish at 7 mpf. Comparisons for A standard length and B mass are presented as mean ± SD, n = 13–34/group. Statistical analysis: 2-way ANOVA with Tukey’s multiple comparison test, **** p < 0.0001

Behavioural outcomes in the NTT aligns with a relatively more anxious phenotype in control females vs. control males (Fig. 5A-D) and reflects an increase in anxiety as main effect of ASD (Fig. 5A, B, D). Notably, males showed a significantly higher effect size in this measure and reflected a relative female phenotype, which abolished the sex difference observed in controls. An exception to this is the latency to first movement into the top zone (Fig. 5C), where sex-differences were maintained across groups, likely owing to the relatively insignificant effect of the ASD phenotype on this parameter.

Fig. 5.

Sex-specific behavioural responses in control vs. ASD-like zebrafish at 6 mpf. Outcomes following the novel tank test (A-D), social preference test (E-H), and mirror biting test (I-L) are presented. Data are presented as scatter dot plots with mean ± SD (parametric) or median ± 95% CI (non-parametric), n = 12/sex/group. Statistical analysis: 2-way ANOVA – main ANOVA effect of sex (δ) and ASD-like phenotype (ψ) indicated – with Fisher’s LSD multiple comparison test, * p < 0.05, ** p < 0.01, *** p < 0.001, and **** p < 0.0001

As in the NTT, a main effect of sex in the SPT (Fig. 5E-H) confirms the naturally affiliative behaviour in females, with control females spending significantly more time in the social zone and fewer transitions between zones than control males. This difference was abolished in the ASD phenotype, where subtle but similar shifts towards social avoidance in both sexes was observed. Of further interest, ASD-like females seemed to exhibit a behavioural profile that is more similar to control males in this test.

Finally, the outcomes from the MBT aligns with a relatively more aggressive stance in males vs. females (Fig. 5I) that was not affected by the ASD phenotype. However, data does not reflect execution of more aggressive acts in males or in ASD-like phenotypes (Fig. 5L). Also noteworthy, main effects of ASD-like phenotype to increase both the number of zone entries and the time taken to bite the mirror for the first time were also observed across sexes (Fig. 5J, K). The latter data, when interpreted with the transition data from the SPT seems to suggest a desire for social interaction coupled with a failure to follow through in the ASD-like groups. Given the observed high anxiety in both ASD-like sexes this may suggest social anxiety or irritability rather than overt asocial or aggressive tendencies.

Proteomic analysis illustrated that all brain samples demonstrated similar clustering patterns across experimental groups (Fig. 6). Differential protein expression analyses revealed significant differences only between sexes of the same experimental (control or ASD-like) group. Although this analysis did not point to direct effects of the ASD-like phenotype, there were considerably more differentially upregulated proteins in males and females in the ASD-like group (33 and 11 respectively) than in the control group (1 and 8 in males and females respectively).

Fig. 6.

Sample clustering and differential protein expression of the adult zebrafish brain proteome. Principal component analysis plots indicating A all data (including outliers) and B clean data sets. Volcano plots illustrating differential protein expression in detected the whole brain proteome is represented for selected comparisons: C control male vs. ASD-like male, D control female vs. ASD-like female, E control male vs. control female, and F ASD-like male vs. ASD-like female, n = 12/group. The number of significantly upregulated proteins are indicated. Statistical analysis: two-sample student t-tests p-values were adjusted using the Benjamini-Hochberg false discovery rate of 5%. Abbreviations: PC, principal component; FDR, false discovery rate; FC, fold change

To identify signalling pathologies in the brain compartment specifically, and perform additional analysis on parameters with more homogenous (low) abundance, all detected proteins involved in neurotransmission were grouped. The number of the proteins identified in each neurotransmission class are indicated in Table 1. Expectedly, the percentage of total representation aligned with the average abundance of each neurotransmitter system in the dataset.

Table 1.

Number of proteins (including paralogs, expressed as a fraction of total possible number of proteins) involved in neurotransmission that were detected in the zebrafish whole brain proteome

Neurotransmitter system	Enzymes (synthesis and/or metabolism)	Transporters	Receptors	Relative system representation (%)	Relative total abundance (%)
Glutamate	6/6	EAAT (5/13) VGLUT (4/5)	AMPA (8/8) NMDA (6/13) KA (4/6) GRM (11/12)	70	1.29
GABA	5/5	Co-transporters (2/2) VGAT (1/1) GAT (3/5)	GABAA (10/21) GABAB (3/3)	65	0.24
Catecholamine	5/6	VMAT2 (1/1) DAT/NET (1/2)	Dopaminergic (0/10) Adrenergic (0/15)	21	0.028
Serotonin	3/4	VMAT2 (1/1) SERT (1/2)	Serotonergic (1/15)	27	0.019
Acetylcholine	2/3	High affinity transporter (1/2) VAChT (1/2)	Nicotinic (1/12) Muscarinic (1/10)	21	0.0099
Histamine	2/3	n/a	Histaminergic (0/4)	29	0.0028
Glycine	3/3	3/3	Alpha (2/5) Beta (2/2)	77	0.040

GABA, -aminobutyric acid; EAAT, excitatory amino acid transporter; VGLUT, vesicular glutamate transporter; AMPA, -amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor; NMDA, N-methyl-D-aspartate receptor; KA, kainate receptor; GRM, glutamate metabotropic receptor; VGAT, vesicular GABA transporter; GAT, GABA transporter; VMAT2, vesicular monoamine transporter 2; DAT, dopamine transporter; NET, norepinephrine transporter; VAChT, vesicular acetylcholine transporter

Analysis of this cluster of proteins demonstrated differential abundance of various role players in glutamatergic and GABAergic signalling – aligning with known excitatory/inhibitory (E/I) imbalance associated with human ASD. However, dopamine and glycine neurotransmitter systems were also affected. Most differences observed implicated enzymes which are responsible for the synthesis of these neurotransmitters, although not all changes were homogenous. Full data is presented as supplementary data (SM Fig. 1), but for clarity only parameters exhibiting significant differences between groups are presented here (Figs. 7, 8 and 9).

Fig. 7.

Differential glutamatergic-specific brain proteome results comparing control male and female and ASD-like male and female zebrafish at 7 mpf. Data are presented as scattered dot plots with mean ± SD (parametric) or median ± 95% CI (non-parametric), n = 12/sex/group. Statistical analysis: 2-way ANOVA – main ANOVA effects of sex (δ), ASD induction (ψ) and/or their interaction (ψ:δ) are indicated – with Fisher’s LSD multiple comparison test, * p < 0.05, ** p < 0.01 and *** p < 0.001

Fig. 8.

Differential GABAergic-specific brain proteome results comparing control male and female and ASD-like male and female zebrafish at 7 mpf. Data are presented as scattered dot plots with mean ± SD (parametric) or median ± 95% CI (non-parametric), n = 12/sex/group. Statistical analysis: 2-way ANOVA – main ANOVA effects of sex (δ), ASD induction (ψ) and/or their interaction (ψ:δ) are indicated – with Fisher’s LSD multiple comparison test, * p < 0.05, ** p < 0.01 and *** p < 0.001

Fig. 9.

Differential dopaminergic- and glycinergic-specific brain proteome results comparing control male and female and ASD-like male and female zebrafish at 7 mpf. Data are presented as scattered dot plots with mean ± SD (parametric) or median ± 95% CI (non-parametric), n = 12/sex/group. Statistical analysis: 2-way ANOVA – main ANOVA effects of sex (δ), ASD induction (ψ) and/or their interaction (ψ:δ) are indicated – with Fisher’s LSD multiple comparison test, * p < 0.05 and ** p < 0.01

Considering the glutamatergic system first, in terms of glutamate synthesis, in contrast to similar levels in control males and control females, ASD-like males exhibited significantly increased aspartate aminotransferase levels of an extent severe enough to result in a sex difference (Fig. 7A). In terms of transporters, subtle differences in the ASD-like phenotype strengthened a sex difference for VGLUT2, which transports neuronal glutamate (lower in females, Fig. 7B), but losing it for EAAT1, which facilitates synaptic glutamate uptake into glial cells (Fig. 7C). In terms of metabotropic receptors, GRM4 was decreased in ASD-like males, abolishing the sex difference seen in controls (Fig. 7D), while GRM5 was increased in only in ASD-like females, establishing a sex difference (Fig. 7E).

Moving on to GABAergic signalling, overall females were not affected as result of ASD. However, males exhibited significant dysregulation in this context. Specifically, 4-aminobutyrate transaminase, the first enzyme in the GABA shunt, was higher in ASD-like males than in ASD-like females (Fig. 8A). In contrast, both glutamate decarboxylases, the second enzyme(s) in the GABA shunt, which under control circumstances were more abundant in males than females, exhibited significantly reduced expression in the ASD-like male group only, effectively abolishing the sex-difference (Fig. 8B, C). A similar picture was apparent at receptor level, in that two GABA_A subunits demonstrated decreased abundance in ASD-like males only, to levels similar to that of females (Fig. 8D, E).

Turning attention to relatively poorer represented pathways, the only clear ASD-like difference observed along the dopaminergic pathway was lower tyrosine hydroxylase, which reached statistical significance only in ASD-like females (Fig. 9A). Likely due to low representation, no insights on sex differences in ASD was gained in terms of serotonergic, acetylcholinergic or histaminergic pathways. Lastly, although not directly pointing to a sex difference, glycine decarboxylase exhibited a significantly higher levels in ASD in both sexes (Fig. 9B). Given that glycine acts as an inhibitory neurotransmitter, this result aligns with known dysregulated E/I signalling in ASD [55].

Discussion

This study demonstrated that the VPA zebrafish model of ASD reproduced key behavioural and neurochemical features that persist into adulthood and exhibit distinct sex differences. Behavioural outcomes seemed to indicate a convergence of behavioural profiles of males and females in ASD-like groups. In contrast, detailed characterisation of the whole brain proteome revealed heterogenous, sex-specific changes in brain proteins involved in neurotransmission. Although the proteomic data do not support the identification of definitive male- or female-specific biomarkers, the observed signalling differences seemed to suggest that ASD-like males exhibit dysregulation consistent with increased excitation and decreased inhibition, whereas ASD-like females seemed to present with alterations indicative of dopaminergic-glutamatergic dysregulation. Given that behavioural profiles appeared similar across sexes, but neurotransmitter profiles remain distinct, these findings support theories proposing distinct female vs. male phenotypes of ASD.

Early developmental exposure of zebrafish to VPA is a well-established preclinical model of ASD [14]. In the current study, we confirmed the induction of a mild ASD-like phenotype following VPA exposure, as well as the retention of this phenotype into adulthood, aligning with several adult zebrafish studies demonstrating enduring molecular and behavioural anomalies in this model [7, 48, 54, 65, 84]. In addition, we identified sexually dimorphic alterations in several neurotransmission profiles. The specific neurotransmitter systems exhibiting the most prominent sex-specific differences are illustrated in Fig. 10. Together this may suggest differential dysregulation of glutamatergic, GABAergic and dopaminergic signalling pathways, highlighting sex-specific vulnerabilities that may contribute to ASD-associated behavioural outcomes.

Fig. 10.

Major neuronal pathways highlighting sex-specificity of neurotransmitter dysregulation in an ASD-like phenotype. Signalling pathways implicated include A glutamatergic, B GABAergic and C dopaminergic systems. Coloured stars indicate differentially expressed proteins. Glu, glutamate; Gln, glutamine; VGLUT1/2, vesicular glutamate transporter 1/2; EAATs, excitatory amino acid transporters; AMPA, -amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor; NMDA, N-methyl-D-aspartate receptor; KA, kainate receptor; GRM, glutamate metabotropic receptor; GAD1/2, glutamate decarboxylase 1/2; GABA, -aminobutyric acid; VGAT, vesicular GABA transporter; GAT1/3, GABA transporter 1/3; NKCC1, Na-K-Cl cotransporter 1; KCC2, K-Cl cotransporter 2; L-DOPA, levodopa; DDC, dopamine decarboxylase; DA, dopamine; COMT, catechol-O-methyltransferase; 3MT, 3-methoxytyramine; MAO, monoamine oxidase; HVA, homovanillic acid; VMAT2, vesicular monoamine transporter 2; DOPAC, 3,4-dihydroxyphenylacetic acid; DAT, dopamine transporter; DR, dopamine receptor

Beginning with ASD-associated anxiety, several monoaminergic neurotransmitters (serotonin, dopamine, norepinephrine and histamine) have been implicated in both the pathogenesis and clinical manifestations of anxiety symptoms. In the current study, the NTT provides compelling evidence of heightened anxiety-like behaviours in both ASD-like and female groups. This observation is consistent with sex differences in zebrafish [25] and humans [20, 52], as well as previous VPA-based zebrafish models of ASD [48, 84]. Importantly, these findings also align with clinical observations demonstrating a high prevalence of anxiety among individuals with ASD [11, 39, 75, 76]. Higher rates of anxiety have been reported in women with ASD, with maladaptive emotional regulation proposed as a contributing factor and prompting recommendations for sex-specific interventions [67]. While social and environmental influences likely contribute to anxiety outcomes in humans, the present findings suggest that heightened anxiety-like behaviour may also represent an intrinsic feature of the ASD-like phenotype, at least in zebrafish, independent of societal influences.

In terms of potential mechanisms, low abundance of several anxiety-related neurotransmitter proteins in the proteomic dataset limited the insights that could be drawn from the observed sex-specific behavioural outcome. Nevertheless, a significant reduction in tyrosine hydroxylase – the rate limiting enzyme for dopamine synthesis – was observed in ASD-like and female groups. This finding is suggestive of reduced dopaminergic tone and is consistent with the observed heightened anxiety-like outcome in our study. A potential mechanistic link in this context may involve anxiety arising from reduced dopaminergic signalling [4, 8, 66]. This differs from a hyperarousal-type anxiety phenotype (more commonly associated with serotonergic dysfunction) in that it is characterised by avolition (lack of motivation) and anhedonia (inability to feel pleasure) [9, 64]. Such a phenotype often manifests as increased irritability, social withdrawal and emotional blunting – features that overlap with core clinical characteristics of ASD [2, 18, 37].

Supporting the possibility of a hypodopaminergic contribution to the anxiety-like outcome in our model, the ventral tegmental area (VTA) – or posterior tuberal nucleus (PTN) in the case of zebrafish [59] – has been described as a key interface of stress and reward processing [10] and plays an important role in the dopaminergic regulation of GABAergic and glutamatergic signalling [5, 35]. Altered dopaminergic signalling in the VTA, a region primarily involved in the regulation of motivation and interactive behaviours, may contribute to E/I imbalance, and consequently ASD-associated anxiety [51]. As such, the potential involvement of the VTA/PTN in mediating sex-specific ASD phenotypes warrants further investigation. Of additional relevance, dopaminergic neurons within the VTA co-release glutamate via VGLUT2 [13]. In the current study, VGLUT2 abundance was lowest in ASD-like females, suggesting a potential mechanism that may contribute to the observed exaggerated anxiety-like phenotype in this group. However, given that current analyses were performed on whole brain tissue, future studies examining dopaminergic alternations within the VTA/PTN, as well as determining whether VGLUT2 abnormalities localise to this region would help establish the plausibility of this proposed mechanism.

In further support of a distinct female ASD phenotype, our proteome data suggest the existence of a female-specific dopaminergic-glutamatergic link. For example, the elevated GRM5 levels observed in the ASD-like female group, may reflect a compensatory response to reduced dopaminergic signalling (resulting from lower tyrosine hydroxylase) in this group. Indeed, dopaminergic signalling can regulate GRM5 signalling both directly and indirectly, primarily through receptor heterodimerisation [28]. Specifically, dopamine D1/2 receptors (D1/2R) form functional complexes with GRM5. The heteromeric complexes exhibit synergistic activation of phospholipase C signalling, enhancing intracellular calcium, producing more robust downstream signalling than either receptor would achieve alone [69]. Given the role of GRM5 in regulating dynamic regulation of synaptic function and neural plasticity [68], the observed increased in GRM5 abundance may be indicative of sex-specific alterations in synaptic signalling. However, targeted studies investigating dopaminergic-glutamatergic interactions in a sex-specific ASD context will be required to demonstrate the likelihood of the premise. Interestingly, modulation of the GRM5 receptor has been reported to ameliorate behavioural deficits in murine models of ASD [71, 77]. Similarly, modulation of GMR5-D2R-adneosine A2A receptor oligomers have been proposed as a strategy to reduce ASD-associated repetitive and asocial behaviours [17]. However, the extent to which these effects differ between sexes remains unknown. Consequently, studies aimed at elucidating the role of GRM5 modulation on behavioural outcome in a sex-specific ASD context would provide valuable insight into the therapeutic relevance of this pathway.

Moreover, distinct patterns of social avoidance were observed in the ASD-like groups during the SPT. In humans with ASD, the degree and presentation of social difficulties vary considerably between individuals and across sexes [45], with males generally presenting with more severe social withdrawal. The current study outcomes align with a male bias for asocial behaviour. However, the relative impact of the ASD-like phenotype on social behaviour appeared greater in females than in males, such that the movement patterns of the ASD-like females more closely resembled that of control males. In line with this notion, neural ‘masculinisation’ has previously been reported in females with ASD [44], a finding frequently reviewed as support for the extreme male brain theory of ASD [6]. Our findings also parallel observations typically described as social isolation in other zebrafish VPA models [7, 48, 65, 84]. However, when considered alongside the latency and transition data reported here, an alternative interpretation may be that these behaviours reflect impaired execution of social engagement rather than a complete absence of social motivation. While speculative at this point, this interpretation is consistent with the female autism phenotype theory, which proposes that females with ASD may retain relatively higher social motivation despite exhibiting atypical social behaviour [41]. Notably, the use of a preclinical model that precludes societal adaptions such as masking/camouflaging [16], which may limit direct translational comparisons with human ASD. Nevertheless, an important strength of the current study is the demonstration that superficially similar behavioural outcomes, albeit associated with nuanced sex differences, can arise from distinct neurobiological outcomes. Thus, behavioural convergence does not necessarily imply shared underlying mechanisms, providing further insight into sex-specific ASD presentation. Finally, reduced conspecific preference may not exclusively reflect diminished social interest. Alternative explanations include heightened anxiety-like behaviour, altered stress responsivity, differences in novelty processing or changes in attention toward social stimuli. Collectively, these findings may indicate a broader alteration in social responsiveness; however, more in-depth behaviour analyses will be required to distinguish between these possibilities.

Turning to male-specific observations, relative suppression of the GABAergic (major inhibitory) neurotransmitter system was observed in the current study. In line with this, reduced expression of glutamate decarboxylase has been reported in predominantly male brains of individuals with ASD [21, 80]. As the final enzymes in GABA synthesis, decreased glutamate decarboxylase activity is expected to reduce GABA availability, a finding that has also been reported in ASD cohorts, although study cohorts had significant male bias (59 males vs. 14 females) [36, 62]. Beyond potential alterations in neurotransmitter availability, our study also identified changes in post-synaptic receptor expression. Specifically, several GABA_A receptor subunits (β and γ) were significantly reduced exclusively in the ASD-like males relative to their control counterparts. Similar, reductions in GABA_A receptor expression have been reported in individuals with ASD, although these cohorts likewise were significantly predominantly male bias [22, 23]. Given the relative paucity of female-specific clinical data, our findings may provide valuable insight into potential mechanisms to evaluate that may contribute to the sex-specific ASD phenotypes.

Furthermore, aspartate aminotransferase - the principal enzyme involved in neuronal glutamate synthesis (major excitatory) - was significantly higher in the ASD-like male group (compared to both control and ASD-like females). This finding may suggest either capacity for increased glutamate synthesis, or alternatively a compensatory response to altered glutamatergic signalling. However, given reports of elevated brain glutamate levels in males with ASD [58], as well as in rodent preclinical ASD models [83], increased glutamatergic activity appears the more plausible interpretation. Such alterations may therefore represent a key feature of male ASD-associated neuropathology. Interestingly, Zieminska et al. [83] demonstrated hippocampal glutamate levels were only elevated in male or combination (male and female) VPA-exposed rodent groups, despite both sexes exhibiting comparable behavioural deficits in ultrasonic vocalisation assays. This aligns with our study in that similar behavioural outcomes seem to arise from distinct neurochemical alternations.

Further evidence of male-specific glutamatergic dysregulation was provided by the observation that GRM4 protein expression was lowest in ASD-like males. Interestingly, in a gastrointestinal context, reduced GRM4 has recently been reported in colonic biopsy samples from male, but not female, children with ASD [73]. Although the relative association of intestinal and brain GRM4 levels remains unclear, the authors suggest that the gut represents a critical modulator of neurodevelopmental processes in ASD, and that gut-brain axis dysregulation may contribute to ASD pathophysiology. Consequently, GRM4 abnormalities may represent an additional mechanism associated with a more male-typical ASD phenotype. Together, these findings highlight convergent alterations in neurotransmitter systems associated with E/I signalling. Although functional downstream signalling was not assessed in the current study, the observed pattern of reduced inhibition and enhanced excitation is consistent with potential E/I imbalance in ASD-like males.

Additional behavioural insight into the interplay between anxiety and aggression was provided by the MBT results. We suggest that the ASD-like behavioural profile observed in this test reflects irritability underpinned by anxiety-like caution/uncertainty, rather than overt aggression. This interpretation is consistent with evidence from non-ASD studies suggesting that heightened anxiety can suppress immediate aggressive responses while maintain vigilance or approach-related behaviours [1, 40]. Similarly, other zebrafish VPA models have either reported no significant changes in aggression [84] or decreased aggression [48]. However, the latter study employed a VPA dose more than 13-times higher than that used in the current study and thus may reflect different underlying mechanisms. Although aggression is a common behavioural trait amongst individuals with ASD [24, 34], it is perhaps of value to consider that some manifestations of perceived aggression may instead reflect escape- or avoidance-driven behaviours arising from social anxiety. This notion is in line with a study reporting that children with ASD frequently engaged in aggressive-like behaviours to gain access to preferred ritualistic or repetitive behaviours [63]. In this context, identifying the underlying motivation for aggressive behaviours may be critical for appropriate therapeutic intervention, as treatments targeting aggression without addressing the precipitating factors may inadvertently worsen behaviour symptoms. Interestingly, Reese et al., [63] also reported sex-differential motivations for disruptive behaviour, further supporting the notion that superficially similar behavioural outcomes may arise from distinct underlying mechanisms. This concept aligns closely with the findings of the present study, where behavioural similarities were accompanied by divergent neurochemical profiles.

In terms of neurotransmitter involvement, low serotonin and GABA, together with high dopamine and norepinephrine signalling, have been implicated as key regulators of aggression, with serotoninergic dysregulation being regarded as most important [27]. Indeed, partial/complete knockdown of the rate limiting enzyme of serotonin synthesis in the brain – tryptophan hydroxylase 2 – results in more aggressive behaviour in rodents [31, 38, 53, 72]. In the current study, although no pairwise group differences were detected, the ASD-phenotype was associated with a significant main ANOVA effect on tryptophan hydroxylase 2 levels (Fig. SM1), suggesting general reduction in serotonergic signalling. This finding aligns with reports of low brain serotonin levels during preschool years in children with ASD [15]. Nevertheless, the extent to which reduced serotonergic signalling contributed to irritability, social anxiety or related behavioural manifestations in adults with ASD remains unclear and warrants further investigation.

A limitation of the current study is the incomplete detection of low-abundance proteins within specific neurotransmitter systems (i.e. dopaminergic and serotonergic). As such, the full extent of dysregulation of these pathways could not be inferred, and additional sex-specific changes may have gone undetected. Be that as it may, to our knowledge, this data represents the first proteomic characterisation on the zebrafish VPA model of ASD and highlights several promising avenues for more targeted mechanistic investigations. Moreover, while whole brain analysis may have reduced sensitivity to region-specific effects, it enabled the integration of behavioural and neurotransmitter findings at the level of individual fish without the need for tissue pooling. As such, the existence of sex-specific neurological phenotypes in ASD seems plausible and may have important implications for the development of sex-specific informed therapeutic strategies.

Future studies employing more detailed behavioural paradigms may also improve on the detection of subtle phenotypic changes and treatment-related effects. However, before specific biomarker or therapeutic targets can be identified, it will be necessary to incorporate region-specific analyses of relevant brain structures, as well as to assess these changes dynamically over time to delineate primary pathology from compensatory mechanisms. Finally, the current study only looked at single level of regulation i.e. in terms of the nervous system. Future work should integrate these findings with other regulatory systems such as the endocrine system and gut microbiome, as well as established comorbid pathologies such as mitochondrial dysregulation. A more holistic understanding of the interactions among these systems may be necessary to fully elucidate the mechanisms underlying ASD symptomatology and inform the development of more effective therapeutic interventions.

Conclusion

Overall, our data suggest distinct neurotransmitter pathways may be disrupted in males and females in the zebrafish VPA model of ASD. Specifically, the female ASD-like phenotype appears to be primarily associated with dopaminergic-glutamatergic dysregulation, whereas the male ASD-like phenotype appears more closely associated with E/I imbalance, although these findings require further validation. Notably, against a background of inherent sex differences, the ASD-like phenotype in females appeared comparatively subtle, which may potentially contribute to delayed recognition or underdiagnosis. In this regard, our findings align most closely with the female autism phenotype theory. Finally, our behavioural data may provide a new framework for interpretating ASD-like behaviours, particularly with respect to their psychological drivers, and highlights that behavioural convergence does not necessarily imply neurophysiological similarities in the context of ASD.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

3MT

3-Methoxytyramine

ACN

Acetonitrile

AMPA

-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor

ASD

Autism spectrum disorder

CI

Confidence interval

COMT

Catechol-O-methyltransferase

DA

Dopamine

DAT

Dopamine transporter

DDC

Dopamine decarboxylase

DOPAC

3,4-Dihydroxyphenylacetic acid

dpf

Days post fertilisation

DR

Dopamine receptor

E3

Embryo medium

EAAT

Excitatory amino acid transporters

FA

Formic acid

FC

Fold change

FDR

False discovery rate

GABA

-Aminobutyric acid

GAD

Glutamate decarboxylase

GAT

GABA transporter

Gln

Glutamine

Glu

Glutamate

GRM

Glutamate metabotropic receptor

HVA

Homovanillic acid

IQR

Interquartile range

KA

Kainate receptor

KCC2

K-Cl cotransporter 2

L-DOPA

Levodopa

LC-MS

Liquid chromatography-mass spectrometry

LDTT

Light-dark transition test

MAO

Monoamine oxidase

MAR

Missing-at-random

MBT

Mirror biting test

MNAR

Missing-not-at-random

mpf

Months post fertilisation

NET

Norepinephrine transporter

NKCC1

Na-K-Cl cotransporter 1

NMDA

N-methyl-D-aspartate receptor

NTT

Novel tank test

PC

Principal component

PCA

Principal component analysis

PTN

Posterior tuberal nucleus

SAVC

South African Veterinary Council

SD

Standard deviation

SPT

Social preference test

TDM

Total distance moved

TEAB

Triethylammonium bicarbionate

TTS

Total time spent

VAChT

Vesicular acetylcholine transporter

VGAT

Vesicular GABA transporter

VGLUT

Vesicular glutamate transporter

VMAT

Vesicular monoamine transporter

VPA

Valproic acid

VTA

Ventral tegmental area

Author contributions

Conceptualisation – CS, LP; Study design – CS, LP, TM, Curation of funding – CS, LP; Supervision – CS, LP; Data generation and statistical analysis – TM, LP; Interpretation – CS, LP; First draft – TM, LP, CS; All authors edited and approved the final manuscript.

Funding

The South African National Research Foundation (Grant SRUG2203301243 - CS), L’Oréal-UNESCO Fondation (For Women in Science Young Talents Program – LP) and Harry Crossley Foundation (Grant – LP) are acknowledged for financial support. LP received a postdoctoral fellowship from the South African National Research Foundation. Professors Jean Millar and Robert Fenn are acknowledged for continued financial support of the Smith group.

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

All experimental protocols received ethical approval from the Research Ethics Committee for Animal Care and Use at Stellenbosch University (Ref: ACU-2025-29984).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.