Introduction

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental condition, with a prevalence of 1% [1]. ASD is characterized by impairments in social communication and interaction, as well as repetitive or stereotyped behaviors [2], associated with various co-occurring features. Despite the identification of over thousands of genes associated with ASD, the etiology of most cases remains idiopathic. Early behavioral interventions address social features, but heavily rely on intensive and costly training [3, 4]. Although atypical antipsychotics have gained approval for targeting some co-occurring features, their efficacy is often limited, and they are associated with side effects [5, 6]. To date, clinical trials for drugs specifically targeting core features have encountered setbacks due to their lack of efficacy [7, 8]. The failure of these trials can be attributed to the inherent heterogeneity observed across the spectrum. The success of future trials depends on stratifying this diversity into subgroups within the autism spectrum and identifying reliable markers. Common pathological mechanisms in ASD include global alterations in synaptic activity and plasticity, such as imbalances in GABA and glutamate or reduced oxytocin (OT) levels [9,10,11,12,13]. Among these alterations, meta-analyses of multi-omic studies identify molecular markers such as members of the immediate early genes (IEGs), neurotrophic factors, and components of the OT pathway [14,15,16,17,18,19].

OT and its paralog vasopressin (AVP), along with their receptors, regulate various social interactions throughout life, including maternal and socio-sexual behaviors [20,21,22,23]. These neuropeptides are predominantly synthesized in the somas of OT and AVP neurons located in the paraventricular (PVN) and supraoptic (SON) nuclei of the hypothalamus [24,25,26,27]. They project to and release these peptides in response to stimuli in brain structures of the emotional and social circuit, where their receptors are expressed, or in the periphery via the pituitary gland [28,29,30]. Among these structures, the prefrontal cortex (PFC), and the caudate putamen (CPU) and nucleus accumbens (NAC) of the striatum are interconnected and play a key role in both motor behaviors and social interaction and reward [31,32,33,34,35]. Disruptions in these regions are associated with core autism-like features in mice. There is emerging evidence suggesting a potential link between dysregulation of the OT system and the etiology of ASD (for reviews [36,37,38,39]), particularly the association with pathogenic variants in oxytocin (OXT), oxytocin receptor (OXTR), and vasopressin receptor 1A (AVPR1A) genes (SFARI Gene). Additionally, OXT mRNAs are particularly susceptible to deadenylation and degradation [19] and both OXT and its receptor, OXTR, show reduced levels in the brains of individuals with ASD or related conditions [40, 41] and mouse models displaying autism-like behaviors [39, 42]. Notably, deletion of Oxtr and Avpr1a genes in mice has been shown to induce autism-like behaviors (for review [8]). Nevertheless, the administration of OT has yielded inconsistent effects on social traits in individuals with ASD [8, 43, 44]. Intriguingly, a study has suggested that the concentration of OT in the blood of individuals diagnosed with autism could predict the outcome of oxytocin administration [45]. It remains unclear whether mRNAs within this family can serve as shared or specific molecular markers for individuals with ASD.

Multiple studies evidenced global alterations in neuronal activation and synaptic plasticity in ASD [17, 19, 46, 47]. IEGs are rapidly transcribed, and translated within 30 min to 2 h in response to various stimuli, including social interactions [48,49,50]. Following this initial response, IEGs orchestrate long-term plasticity, notably by initiating the second wave of late genes within 4–24 h for some IEGs that are transcription factors. IEG mRNAs represent key molecular markers of neuronal activation. Dysregulation of their expression has been observed in both plasma and brain tissues of people with ASD [51,52,53,54,55] and animal models mimicking ASD features [56,57,58]. Whether specific IEGs can serve as predictive markers for distinct autism subtypes remains an open avenue for investigation.

In this study, we used three distinct genetic mouse models—Shank3, Fmr1 and Oprm1 knockout (KO) mice—alongside mice subjected to early chronic social isolation, encompassing diverse etiologies to model the autism spectrum. Our investigation revealed that each model exhibited a distinctive behavioral signature, mimicking the complexity of the spectrum. We pinpointed Egr1, Foxp1, Homer1a, Oxt and Oxtr as robust molecular markers within the OT and IEG gene families. These markers exhibited specific responses to social interactions in wild-type (WT) mice and significant dysregulations in the aforementioned four mouse models across five structures of the social circuit. Using these five markers, we successfully demonstrate the first proof-of-concept for the stratification of mouse models, differentiating subtypes of autism-like behaviors. The identification of these specific markers offers a promising avenue for tailoring personalized medical help to individuals with autism.

Materials and methods

Animals

All mouse breeding, care, and experimental procedures were in accordance with the European and French Directives and approved by the local ethical committee CEEA Val de Loire No. 19 and the French Ministry of Teaching, Research and Innovation (APAFIS #18035-2018121213436249). Mouse models were selected based on their distinct etiologies and behavioral phenotypes, which are well-documented in the literature and consistently observed across laboratories: the mu-opioid receptor deletion—Oprm1 KO (JAX stock #007559) [59], the synaptic anchoring protein deletion—Shank3 KO (JAX stock #017688) [32], the X-linked mRNA binding protein deletion—Fmr1 KO and females Fmr1+/− (provided by Rob Willensem [60]) and the IEG deletion—Arc KO (JAX stock #007662) [61], modeling ASD genetic cases. Additionally, “isolated” animals exposed to chronic social isolation after weaning for at least 4 weeks prior to behavioral tests were used to mimic idiopathic cases. Animals were maintained on a mixed 50% C57BL/6J—50% 129S2 background. Each mouse line was outcrossed with fresh mixed backgrounds every 5–10 generations, and between these outcrosses, with WT mice from other lines, to prevent inbreeding and ensure consistency among mice across independent batches and lines (Fig. S1). Mice were kept on a regular 12-h light/dark cycle, with food and water available ad libitum, at a controlled temperature of 21 °C and 50% humidity in conventional housing conditions. After breeding heterozygous F0 parental mice, 3–4 unrelated WT and KO couples were housed separately to avoid the effects of the social environment [62], and all offspring were raised in social groups of 2–4 individuals per sex for testing. Two months old naive male and female mice were randomly assigned to housing conditions by an experienced experimenter.

Behavior experiments

All behavioral tests were sequentially performed in a dim light quiet room, using standardized equipment, and are detailed in the supplementary methods. Briefly, social interactions were performed in the three-chambered and in reciprocal social interaction tests with a sex- and age-matched unknown mouse or a cage mate in an open field. Non-social interactions were tested with an object also in open field arenas. Perseveration and cognitive flexibility in a spatial task were assessed in the Y-maze test, while repetitive and stereotyped behaviors were examined in the motor stereotypy test, assays that have been previously characterized for mouse models of ASD [63]. While experimenters were not blinded to the conditions during the tests to prevent any mixing of mice, scoring was conducted manually for reciprocal social interaction and three-chambered tests by a trained experimenter who was blinded to the conditions, or automatically for the Y maze and open field tests. All animal research criteria have been reported in accordance with the ARRIVE guidelines [64]. The WT group consisted of distinct batches of WT mice raised in social groups, serving as the respective controls for each mouse line or housing condition and batch.

Quantitative PCR

Mice were dissected at basal conditions or 0.75 (45 min), 2 or 6 h following a social or an object interaction. Within 5 min, 1 mm-thick brain slices were generated using a coronal mouse brain matrix (Vivo-tech Aniphy, VI-68707). Tissue samples of the prefrontal cortex (PFC), nucleus accumbens (NAC), caudate putamen (CPU), paraventricular nucleus (PVN), and supraoptic nucleus (SON), the five brain regions of the social circuit selected for their interconnections and their roles in social interaction impairments and motor stereotypies (anatomical coordinates in Fig. S1E, F) were collected using a 2 mm diameter puncher (with two punches for lateralized regions) and were immediately frozen until further use. After tissue homogenization using a polytron (Grosseron, France, PT1200E), total RNAs were extracted according to the Direct-zol™ RNA Microprep and Miniprep kit (ZymoResearch, Ozyme, France, respectively R2063 and R2050) and quantified using ND2000 nanodrop (Thermofisher, France). The cDNAs were generated from 0.5 or 0.25 µg of total RNAs using the SensiFast reverse transcriptase kit (Ozyme-Bioline, BIO-65054) and quantitative PCR (qPCR) was performed in triplicates according to 2X ONE Green Fast qPCR premix (Ozyme, France, OZYA008-1000) with 1 µL of cDNA and 1 µM of each validated couple of primers (Table S1). The following qPCR protocol was applied for 40 cycles: 95 °C for 5 s, 60 °C for 15 s, and 60 °C for 30 s.

Data modeling and correlations

All statistical analyses were performed using R (version 4.2.2) [65]. Animal outliers (±3 standard deviations) were removed. Variability between cohorts (Fig. S1) was assessed using Principal Component Analysis (PCA) with the FactoMineR package [66]. For qPCR, batch corrections were applied using the ComBat sva package [67] (Fig. S1A–C), while the behavior across the distinct WT batches remained consistent and did not require correction (Fig. S1D). For behavioral data and OT peptide levels, Kruskal-Wallis tests with Dunn’s post hoc tests were conducted using the rstatix package [68]. P-values were adjusted with Benjamini–Hochberg correction [69]. A complete linear model was fitted for qPCR data and IEG protein levels, encompassing variables such as time, social interaction, or mouse line, and their interactions. Post-hoc tests, based on estimated marginal means with Sidak p-value correction, were conducted using the emmeans package [70] to compare social interaction at each time point, the different time points for each interaction, and mouse models to WT mice. In addition, the analyses were conducted separately as PCA revealed distinct clusters among the structures: one comprising PFC, NAC, and CPU, and the other involving PVN and SON (Fig. S1C). Sex differences in behavioral and qPCR data were analyzed only in WT mice, as sample sizes for the other mouse models were too small for comparison. In WT mice, clustering of genes and median expressions was performed using the pheatmap package. Furthermore, the integration of qPCR and behavioral data was performed using DIABLO (Multiblock (s)PLS-DA) implemented in the mixOmics package [71, 72]. The compromise parameter was set at 0.75 to maximize the correlation between qPCR and behavioral datasets. Linear discriminant analysis (LDA) was performed using Egr1, Foxp1, Homer1a, Oxt and Oxtr, on each mouse with less than 5 missing values across all structures, using the MASS package [73]. LDA predicted the class membership of Arc KO mice that were not considered during model fitting (i.e., the model with most similarities). Finally, on the data of Shank3 and Fmr1 KO mice and only on the OT markers, hierarchical clustering on the principal components of PCA (HCPC) was performed using the FactoMineR package [66].

Results

Distinct behavioral profiles across autism spectrum in four mouse models

We examined the behavioral profiles of Fmr1 and Shank3 KO mice, modeling Fragile X and Phelan–McDermid syndromes, respectively, along with Oprm1 KO and chronically isolated mice (Fig. 1, S2–S4, Table S2), all previously published as mouse models displaying autism-like and neurodevelopmental conditions [32, 74,75,76]. To capture a comprehensive view of their social responses, we exposed these models to distinct social interactions: one involving a family member (e.g. a cage mate, “SI mate”) and the other with an unknown conspecific (“SI unknown”), mimicking the daily life of individuals with autism. Additionally, we included a non-social interaction condition with an object (“NSI object”), as control.

Fig. 1: All four models exhibited distinct behavioral features along the autism spectrum.
figure 1

In the sociability (A), the social novelty (B) or mate preference (C) phases of the three-chambered test, the cumulative time in nose contacts of Oprm1, Fmr1, Shank3 KO and isolated mice are compared to WT animals. In the sociability phase, Fmr1 KO mice exhibit reduced time spent in nose contact with an unknown WT mouse (chamber A) compared to WT mice, but not with the object (chamber B). In the social novelty phase, Shank3 KO mice display no preference for the novel mice (unknown WT mouse, chamber B) over the familiar mouse (already explored for 10 min in the precedent phase, chamber A). Oprm1 KO mice show a lack of preference for the cage mate (chamber A) over the familiar mouse (already explored for 10 min in the precedent phase, chamber B). In the reciprocal social interaction tests, the cumulative time in nose contacts with a genotype-, sex- and age-matched unknown mouse (D) is reduced in Fmr1 KO mice (green), but not in Oprm1 (blue), Shank3 KO (yellow) and isolated (burgundy) mice, compared to WT animals (gray). In motor stereotypies, the time spent in self-grooming (E) is increased in Shank3 KO mice and reduced in Oprm1 and Fmr1 KO animals, while the number of head shakes (F) is increased in Shank3 KO mice and reduced in isolated mice, compared to WT mice. Data are presented as individual data, mean ± sd (Table S2). Kruskal–Wallis tests followed by Dunn post hoc tests were performed with stars as line effect and hash as chamber effect (p = adjusted p-value with Benjamini–Hochberg correction). * or # p < 0.05, ** or ## p < 0.01, *** or ### p < 0.001, **** or #### p < 0.0001.

Full size image

In the open field, none of the models exhibited altered behavior following SI mate or NSI object (Table S2). Fmr1 KO mice demonstrated robust social impairments, indicated by a decrease in both total time and mean duration engaged in nose contacts with an unknown mouse both in the reciprocal and three-chambered tests (Fig. 1A, D). This impairment was also observed in heterozygous females in the sociability phase of the three-chambered test (Fig. S2A–C). In contrast, Shank3 KO mice displayed impaired social novelty, evident in their lack of preference for a new mouse over a familiar one (Fig. 1B). Surprisingly, Oprm1 KO and isolated mice did not exhibit the expected social impairments as reported previously [74, 76]. Under standard conditions, Oprm1 KO mice only displayed a lack of mate preference in the three-chambered test (Fig. 1C). Elevating light intensity from 15 to 40 lx to strengthen anxious-like behavior revealed social interaction impairments with an unknown animal in Oprm1 KO mice, as evidenced by a reduction in the time spent in nose contact during both the sociability and social novelty phase of the three-chambered tests, as opposed to WT mice tested in dim light conditions (Fig. S2D–F). Additionally, sex effect analyses indicated that WT males exhibited increased time in nose contact during the social novelty phase of the three-chamber test (Fig. S3A–D). This finding suggests a potential manifestation of induced social impairments or social anxiety—a phenotype situated at the periphery of the autism spectrum.

Regarding stereotyped behaviors, Shank3 KO mice spent more time in self-grooming, accompanied by an increased number of head shakes and a decrease in the time spent digging and in the number of rearing events compared to WT mice (Fig. 1E, F, S4A, B). Conversely, Oprm1 and Fmr1 KO mice demonstrated a reduction in self-grooming time, while isolated mice exhibited a decreased number of head shakes (Fig. 1E, F). WT males displayed a higher number of head shakes compared to females (Fig. S3A–D). Concerning co-occurring features, none of the models exhibited impaired cognitive flexibility in the Y maze, nor did they show locomotion impairments in the open field (Fig. S4C, D). Notably, Shank3 KO mice displayed anxious-like behaviors, spending more time in the periphery of the open field arena compared to WT mice (Fig. S4E).

In summary, the four models presented distinct behavioral features modeling the autism spectrum, with Fmr1 and Shank3 KO mice displaying the most severe phenotypes.

Identification of specific molecular markers of social interactions in WT mice

Given that molecular markers for social interactions remain unknown, our primary objective was to identify them, starting with WT mice. We focused on two mRNA families, recognized as potential molecular markers of autism [8, 17, 39, 77]: the OT family, encompassing Oxt and Avp, and their receptors (Oxtr, Avpr1a, Avpr1b), and IEGs and neurotrophic factors, including Arc, Egr1, Fos, Fosb, Jun, Homer1a, Foxp1, Gdnf and Bdnf. We investigated their kinetic profiles up to 6 h after SI mate, SI unknown, and NSI object, as well as 6 h following acute social isolation, a condition with negative social valence. We studied their profiles across five key structures within the social circuit: the paraventricular (PVN) and supraoptic nuclei (SON) of the hypothalamus, the caudate putamen (CPU) and nucleus accumbens (NAC) of the striatum, and the medial prefrontal cortex (PFC). To ensure the specificity of our findings, we ruled out circadian or sex biases within the two mRNA families (Fig. S5A, Table S3). Notably, none of these mRNAs displayed distinct patterns within a 6-h period or between males and females, except for Foxp1, which exhibited sexually dimorphic expression in the SON. Interestingly, Arc in the SON was the only mRNAs down-regulated by acute social isolation from cage mates (Fig. S5B).

Social criteria for selecting mRNAs included a significant regulation following SI unknown or SI mate, and distinct differences from NSI object in at least two separate brain structures (Figs. S6–S8). Consequently, we identified Oxt and Oxtr mRNAs within the OT family, along with Foxp1 and Homer1a within the IEG family, which were induced by the NSI object in no more than one (and distinct) structure. Among these, Oxtr mRNAs emerged as the most reliable marker, showing no induction in response to the NSI object and meeting the social criteria across three different brain structures. Notably, SI unknown at 45 min emerged as the most distinct stimulus with a rapid and transient increase in mRNAs at 45 min, such as Oxtr expression in the NAC and CPU and Homer1a in the PVN, or conversely, no IEG induction, compared to SI mate and NSI object (Fig. 2A, S6–S8, Table S3). When distinct from NSI object, SI mate induced sustained regulations, like Foxp1 mRNAs in the PFC. Although the NAC and SON exhibited the highest number of regulations, no single structure or mRNA distinctly stood out for a particular social stimulus, except for Oxtr, which was induced following SI unknown and not SI mate (Fig. 2A and S6–S8). Interestingly, SI unknown regulated most IEGs in the PVN and SON (Fig. S8), suggesting potential active processes of social plasticity induction. Remarkably, our datasets in WT and models revealed robust correlations between mRNAs, such as a 0.92 positive correlation between Oxt and Avp (Fig. S9A, B). Noteworthy, Foxp1, Homer1a and Egr1 displayed a negative correlation with Oxt and Avp (Fig. S9C), suggesting a potential interplay between them.

Fig. 2: Egr1, Foxp1, Homer1a, Oxt and Oxtr are the most discriminant mRNAs in WT mice and models with ASD-like features.
figure 2

Hierarchical clustering of the OT (blue) and IEG (orange) mRNAs under basal conditions and 45 min, 2 or 6 h following interactions with an unknown mouse (SI unknown), a cage mate (SI mate) or an object (NSI object) in WT animals A identifies three clusters among the different interactions and time points: one cluster with SI unknown at 45 min, one with SI mate at 45 min, 2 h and NSI object at 45 min, and one cluster with the other conditions. mRNAs and structures were separated in 7 clusters (1 to 7 from left to right). Clusters 1 and 2 mainly discriminate mRNAs induced by SI unknown at 45 min in the OT family in the NAC, CPU and PFC and IEGs in the PVN and SON, respectively. Clusters 4 and 7 highlight mRNAs that are not induced by SI unknown at 45 min compared to NSI object and SI mate, while cluster 6 shows the mRNAs that are induced by both social stimuli. Clusters 3 and 5 represent mRNAs induced by SI mate and NSI object or NSI object, respectively. Hierarchical clustering under basal conditions and following SI unknown at 45 min in Oprm1, Fmr1, Shank3 KO, isolated and WT animals B reveals that Oprm1 KO mice clustered with WT under basal conditions, where IEGs were not induced in the PVN and SON (cluster 1) but were upregulated in other brain regions (clusters 3–4). In contrast, Fmr1 KO mice at basal conditions clustered with isolated mice, showing upregulation of the OT family and Arc (cluster 2) and lower IEG expression (cluster 3) in the NAC, CPU, and PFC. Following SI unknown, Fmr1 KO mice clustered with Shank3 KO mice and WT exposed to SI unknown at 45 min, marked by higher levels of IEGs in the PVN and SON (cluster 1). Blue and red indicate low and high expression levels, respectively. All data, mean and statistical values are represented in Figs. S6–S8, S10, S11, S14, S15 and Tables S3, S4. CPU, caudate putamen; NAC nucleus accumbens, PFC prefrontal cortex, PVN paraventricular nucleus, SON supraoptic nucleus.

Full size image

In summary, we identified Foxp1, Homer1a, Oxt and Oxtr as specific molecular markers of social interactions in WT mice. These markers were selected to assess their levels in the four mouse models, both under basal conditions and following SI unknown at 45 min—the most discriminant interaction and time point.

Distinct dysregulations within the oxytocin family among mouse models

Previous studies have proposed OT, AVP, or their receptors as potential common biomarkers of autism [8, 39, 77]. To explore shared dysregulations within the OT family, using Oprm1, Fmr1, Shank3 KO and isolated mice, we assessed alterations in Oxt, Avp, Oxtr and Avpr1a expression following SI unknown at 45 min and under basal conditions (Figs. 2B, 3A, S10–S12, Table S4). These mRNAs are largely clustered together, separating Oprm1 KO mice and WT under basal conditions from Fmr1 KO mice under basal conditions and isolated mice (Fig. 2B). Additionally, we identified shared dysregulations across at least two models—Avp in the CPU, Avpr1a in the SON and Oxt in the NAC (Figs. S10, S11). However, the majority of dysregulations were rather specific for each mouse model. Fmr1 KO mice displayed a global decrease in the expression of all four mRNAs in the PVN, along with Avp, Oxt, Oxtr in the NAC, and Oxt and Avpr1a in the SON (Figs. 3A, S10, S11). In contrast, isolated mice exhibited an overall increase in Avp, Oxt and Oxtr expression in the PFC (Fig. 3A). Shank3 KO mice did not show additional dysregulations, while in the CPU, Oxt and Avpr1a were also down-regulated in Oprm1 KO mice (Fig. S10).

Fig. 3: Dysregulations are model-specific for the OT family and widespread across models for IEGs.
figure 3

Dysregulations in the OT family are rather specific to a model when comparing induced 45 min after social interaction with an unknown mouse (SI unknown, dashed lines) over basal conditions (solid lines) in Oprm1 (blue), Shank3 KO (yellow), Fmr1 KO (green) and isolated (burgundy) mice, compared to WT mice (gray). In the PVN (A), Fmr1 KO mice exhibit reduced levels of Oxt, Avp, Oxtr and Avpr1a mRNAs under basal conditions, compared to WT mice, while in the PFC, Avp, Oxt and Oxtr mRNAs are up-regulated in isolated mice (burgundy), compared to WT mice. In the SON, among the IEGs (B), Fos, Foxp1 and Homer1a levels were not induced 45 min after SI unknown over basal conditions across the four mouse models for different reasons in these models. While they remained at basal levels following SI unknown in Oprm1 KO mice, they were already induced under basal conditions in the other three models. Data are presented as individual data, mean ± sd (Table S4). Linear models followed by post hoc tests based on estimated marginal means were performed with stars as line effect and hash as basal vs SI unknown 45 min effect (p = adjusted p-value with Tukey correction). * or #p < 0.05, ** or ##p < 0.01, *** or ###p < 0.001. PFC prefrontal cortex, PVN paraventricular nucleus, SON supraoptic nucleus.

Full size image

To explore the potential for shared dysregulations beyond the initial OT family, we extended our assessment to include Cd38, involved in OT secretion, and Avpr1b, along with generalist enzymes involved in peptide biosynthesis (Pcsk1, Pcsk2, Pcsk5 and Cpe) and degradation (Lnpep, Ctsa and Rnpep; Fig. S12, Table S4). Strikingly, only three shared dysregulations—Cpe, Ctsa, Rnpep in the NAC—were observed among the two models (Fig. S12). Once again, the majority of dysregulations demonstrated model specificity. In addition to the shared dysregulations, Fmr1 KO mice exhibited six unique down-regulations, particularly in the NAC and PVN. Isolated mice displayed mostly up-regulations in the PVN and PFC, along with up- and down-regulations in other structures. Shank3 KO mice showed a down-regulation of Cd38 in the NAC and PFC. Notably, none of these mRNAs were affected in Oprm1 KO mice. To identify the most discriminant mRNAs for model stratification, we applied stringent criteria, requiring a minimum of 6 total and 4 unique dysregulations across three structures and three models. Within the OT family, only Oxt, Avp and Oxtr met these criteria, spanning four structures and four models (Figs. S10–S12). We assessed the impact of these dysregulations on OT peptide levels in Fmr1 KO, Shank3 KO and isolated mice that shared mRNA dysregulations. Although OT concentrations were higher in the PVN as expected and variable between cohorts, no difference was observed in the models compared to their respective WT (Fig. S13A). Plasma OT concentration and urine levels were not correlated with brain concentrations, including PVN and SON (Fig. S13B).

In conclusion, Oxt, Avp and Oxtr emerge as the most discriminant markers within the OT family among the four models. These findings highlight their potential utility in stratifying mouse models based on their unique molecular signatures, rather than revealing shared mechanisms.

Widespread dysregulations in IEG expression across mouse models

Akin to the OT family, we investigated dysregulations in Arc, Egr1, Fos, Foxp1 and Homer1a across the four models following SI unknown at 45 min and under basal conditions. Our results revealed extensive dysregulations in IEGs across all five structures and mouse models (Figs. 2B, 3B, S14–S15, Table S4). However, the underlying causes of these dysregulations varied among the mouse models (Fig. 2B). In the PVN and SON, IEGs remained at basal levels following SI unknown in Oprm1 KO mice clustering with WT under basal conditions, while in the other models, IEGs were already induced under basal conditions compared to WT mice (Figs. 2B, 3B). Conversely, in the CPU and PFC, the opposite trend was observed (Figs. S14, S15). Clustering separated Fmr1 KO mice under basal conditions and isolated mice (lower IEG expression in the NAC, CPU, and PFC) from WT and Fmr1 KO mice following SI unknown at 45 min and Shank3 KO mice (upregulation of IEGs in the PVN and SON; Fig. 2B). Surprisingly, Arc was the only IEG that was frequently clustered with the OT family (Fig. 2B). We identified 6 shared dysregulations in all four models (SON: Fos, Foxp1 and Homer1a; PVN, Egr1 and Homer1a; CPU: Foxp1), as well as four additional dysregulations in three models (SON: Egr1; PFC: Homer1a; NAC: Fos; CPU: Homer1a; Figs. 3B, S14, S15). Applying the same criteria as the OT family, we identified Egr1, Fos, Foxp1 and Homer1a, with Homer1a standing out with 16 total and 7 unique dysregulations across all five structures and all four models.

We evaluated the protein levels of Arc, c-Fos, FoxP1 and Egr1 in basal conditions and two hours after SI unknown in Fmr1 KO and Shank3 KO mice, the two models with the most dysregulations (Fig. S16). Although IEG proteins are also dysregulated, particularly in the PVN of Shank3 KO mice and the NAC of Fmr1 KO mice, mRNA and protein levels did not match each other (Figs. S14–S16). However, c-Fos protein and Fos mRNAs were upregulated following SI or in basal conditions, respectively, in the PVN of Shank3 KO mice. Interestingly, FoxP1 mRNAs were reduced while protein was higher in basal conditions in the CPU of Fmr1 KO mice (Fig. S16), suggesting potential active translation processes and mRNA degradations.

In conclusion, our study unveils widespread dysregulations in IEGs levels across the models. Egr1, Fos, Foxp1 and Homer1a emerge as the most dysregulated IEGs in the four mouse models, providing valuable markers for model stratification.

Molecular stratification of mouse models using Egr1, Foxp1, Homer1a, Oxt and Oxtr

The integration of SI unknown and qPCR data across the four mouse models aimed to link mRNA expression 45 min after SI unknown with specific models and behavioral parameters (Fig. S17). In component 1 (isolated vs. Oprm1 KO mice), the analysis unveiled a positive association between Foxp1 in the SON of isolated mice and their increased nose contacts, which negatively correlated with Oprm1 KO mice (PFC: Fos and Homer1a) and their increased rearing events (Figs. S17A, B and S18). Additionally, Component 2 identified a positive association between grooming behavior and Shank3 KO mice, which negatively correlated with Oxt, Avp and Oxtr in the PFC (Figs. S17C and S18). These findings suggest a connection between Avp, Fos, Foxp1, Homer1a, Oxt and Oxtr and core autism-like features in specific models and confirmed the opposition between spatial exploration and social interaction.

Finally, we crossed all our previous conclusions from WT mice and the four mouse models, identifying Egr1, Foxp1, Homer1a, Oxt and Oxtr, as the most consistent mRNAs. Avp was excluded due to its high correlation with Oxt that could bias the model towards their shared dysregulations. Employing these five robust molecular markers, we conducted a proof-of-concept for potential stratification among the four mouse models (Fig. 4). The linear discriminant analysis (LDA) integrated molecular data from these markers across the five brain structures in the four models under basal conditions and following SI unknown (Figs. 4A, B, S18). The analysis unveiled distinct classifications, identifying Oprm1 KO (LD1) and isolated mice (LD2) as separate models, while the model positioned WT mice in the middle of the four models, as expected. Although Shank3 and Fmr1 KO mice clustered together, they exhibited individual characteristics (LD3). Among the markers, LDA coefficients revealed that levels of Oxt (LD1–3) and Homer1a in the CPU and SON (LD1), as well as Homer1a in the NAC (LD2–3) exerted the most significant influence on the stratification, followed by Oxtr across the structures.

Fig. 4: Stratification of the four models using the five molecular markers Egr1, Foxp1, Homer1a, Oxt and Oxtr.
figure 4

Linear discriminant analysis (LDA; A) revealed that Oprm1 (blue, LD1 component) and isolated (burgundy; LD2 component) mice clustered apart from the other two models. LD3 component discriminates between Shank3 KO (yellow) and Fmr1 KO (green) mice. While challenged with WT (gray) and Arc KO mice (black), the stratification predicted that both mice were similar to Fmr1 KO mice, followed by Oprm1 KO mice for WT mice and Shank3 KO mice for Arc KO mice. The heatmap with LDA coefficients (B) indicates that the five molecular markers, Egr1, Foxp1, Homer1a, Oxt and Oxtr, contributed to the stratification of the models. Hierarchical Clustering on Principal Components (HCPC; C) revealed three subgroups among Shank3 KO and Fmr1 KO mice (left panel). Cluster 1, which contains Shank3 KO and Fmr1 KO mice, displays lower levels of Oxt and Oxtr in the CPU, NAC and PFC, as well as increased levels of Oxtr in the PVN. Blue, red and white indicate low, high levels or no significant contribution, respectively. PFC prefrontal cortex, NAC nucleus accumbens, CPU caudate putamen, PVN paraventricular nucleus, SON supraoptic nucleus.

Full size image

To challenge our stratification, we tested WT and Arc KO mice, previously documented to manifest social interaction impairments [61, 78]. Based on the five markers, LDA predicted that WT and Arc KO mice would be the closest to Fmr1 KO mice (with class membership probability of 57 and 60%, respectively), followed by Oprm1 KO mice (18%), isolated mice (14%) and Shank3 KO mice (10%) for WT mice and Shank3 KO mice (20%) and Oprm1 KO and isolated mice (10%) for Arc KO mice (Fig. 4A). Indeed, Arc KO mice displayed an intermediate phenotype between Fmr1 and Shank3 KO mice, showing social interaction impairments coupled with increased self-grooming (Fig. S19), further confirming their relevance as a mouse model with ASD-like features. Additionally, we employed our stratification to predict the potential responsiveness of subgroups (e.g., Fmr1 and Shank3 KO mice) to treatment administration (Fig. 4C). Solely considering the Oxt and Oxtr markers, we pinpointed cluster 1, consisting of Fmr1 KO and Shank3 KO mice with low levels of Oxt and Oxtr, suggesting a potential positive response to oxytocin treatment in this subgroup.

In conclusion, this study showed the first successful proof-of-concept for the stratification of four mouse models using five molecular markers.

Discussion

Our findings elucidate distinct dynamics in the expression patterns of the two mRNA families across five brain structures in response to two types of social interactions in WT mice. Notably, SI mate elicited a sustained pattern of expression, unlike NSI object, which displayed a rapid and transient increase, indicating a response to the novelty of the environment, or no response. SI unknown exhibited a unique pattern within 45 min, setting it apart from the other two stimuli. Furthermore, our results suggest a distinct molecular mechanism of activation for each social interaction in these five structures. SI unknown predominantly impacts the OT family in the CPU, NAC and PVN, as well as the IEG family in the PVN, SON and PFC. Conversely, SI mate primarily influences the OT family in the PVN, along with the IEG family in the PFC. While previous studies have highlighted the role of the CPU in facilitating mate interactions as a habitual behavior, and the NAC in orchestrating responses to SI unknown [33], our results reveal a more intricate and nuanced molecular mechanism across these structures.

Surprisingly, our findings unveiled a remarkably high positive correlation between Oxt and Avp mRNA levels in both WT mice and mouse models. Oxt and Avp, being paralog genes, are located only 11 kbp apart from each other (4 kbp in humans). Despite the identification of specific elements in this intergenic region in vitro that potentially promote neuron specificity [79], our results suggest that both mRNAs undergo similar regulation, and common regulatory elements may have been evolutionarily conserved. The robust correlation observed between Oxt and Avp mRNAs emphasizes the importance of considering both peptides in research studies.

Within the OT system, our results revealed distinct molecular dysregulations in each mouse model. Particularly, levels of Oxt and Avp mRNAs, as well as Oxtr across the structures delineated the four mouse models. Noteworthy, Shank3 KO mice exhibited Cd38 downregulation in the PFC and NAC. Given its association with social memory and recognition impairments [80], and the social novelty phenotype displayed by Shank3 KO mice, our results position Cd38 as a potential marker of social memory-related conditions. In contrast to the OT family, our results underscore widespread IEG dysregulations across brain structures in all four models, confirming synaptic plasticity impairment as a major hallmark of ASD. Notably, Egr1, Foxp1 and Homer1a emerge as the most robust markers of social interactions in WT animals and the most dysregulated IEGs in the four models. These findings align with their previous association with ASD in both individuals with autism and mouse models mimicking the condition [56, 81,82,83,84,85].

IEG or OT at the protein/peptide level did not substantiate these findings, except c-Fos/Fos in the PVN of Shank3 KO mice. Indeed, protein and mRNA involve distinct cellular processes, translation and transcription, and dynamics [75, 86]. This is particularly evident for IEGs that are highly regulated and responsive to the environment and stimuli. Indeed, as observed for FoxP1/Foxp1 in the CPU of Fmr1 KO mice, mRNAs are rapidly translated and degraded upon translation while protein levels are increased. For the OT family, we observed that Oxt transcription is relatively stable over several hours in the PVN of WT mice. In contrast, OT peptide levels are released in the plasma and brain within min upon activation of OT neurons in response to social, and non-social contexts, including stressful conditions (here, mouse handling and euthanasia) [31, 87,88,89,90,91], pointing to distinct mechanisms and dynamics governing peptide release and mRNA regulations. Notably, brain Oxt mRNAs, similar to OT peptide concentration [92], did not correlate with peripheral OT levels. However, levels of IEGs in accessible fluids, such as blood or cerebrospinal fluid, could be linked to autism, as shown by elevated Arc proteins in the blood of children with ASD [54]. Unlike OT peptide, which can be degraded and still detected through OT dosage [93, 94], our findings suggest Oxt or Avp mRNAs might be reliable markers for long-term synaptic plasticity in OT neurons.

Our results revealed robust impairments in social interaction with an unknown conspecific in Fmr1 and Shank3 KO mice, accompanied by pronounced motor stereotypies in Shank3 KO mice, aligning with previous reports [95, 96]. Conversely, Oprm1 KO and chronically isolated mice did not exhibit the expected social impairments [74, 76]. Their phenotype may have been influenced by the experimenter’s sex [97], predominantly females in this study, as well as by experimental conditions designed to reduce animal stress. Indeed, a slight increase in light induced social impairments in Oprm1 KO mice. Recently, it has been proposed that housing conditions and the duration of social isolation may interfere with social impairments previously observed in chronically isolated animals [62, 98]. Interestingly, none of the mouse models exhibited impaired social interaction with a cage mate, aligning with reports suggesting that interactions within the family environment were more manageable for children with ASD [99]. Notably, the stratification of the models using the five molecular markers—Egr1, Foxp1, Homer1a, Oxt, and Oxtr—consistently aligned with behavioral data. This analysis accurately predicted the molecular underpinnings of the observed behavioral differences between Oprm1 KO and isolated mice and the other two models, as well as Arc KO mice. Despite their distinct profiles, it unveiled shared dysregulations in both Fmr1 and Shank3 KO mice, providing valuable insights into potential links with their respective social phenotypes and responsiveness to treatments.

In conclusion, our study offers valuable insights into the dynamics of OT and IEG mRNAs in WT animals and their dysregulations in mouse models displaying autism-like behaviors across five structures within the social circuit. Our findings not only enable the stratification of the four mouse models, but also allow the identification of subgroups within these models. Transcriptomic studies of postmortem brains from individuals with ASD have also identified dysregulations in EGR1, CFOS, JUN, FOXP1, OXT, and OXTR mRNAs [16,17,18,19], underscoring the relevance of this study to understanding the human condition. The stratification of other mouse models, such as Cntnap2, Magel2, or Oxtr KO mice [100,101,102], and different species, including Shank3 KO rats [103], would enhance the predictive value of the five molecular markers in individuals affected by autism. Exploring additional structures within this circuit, such as the amygdala and lateral septum, could further contribute to the stratification of the models. Nevertheless, future research employing omic approaches beyond the oxytocin and synaptic plasticity families is essential to identify more molecular markers and potentially uncover novel therapeutic targets. This study represents the first proof-of-concept for molecular stratification using brain tissues of mouse models, with the aim of improving the success of clinical trials and personalized treatment for ASD. Additionally, our work may offer the potential for more precise and faster ASD diagnostics, complementing behavioral screenings [2].