Altered kinship vocal dynamics in marmosets with valproic acid-induced model of autism.

Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice iScience . 2026 Mar 20;29(4):115419. doi: 10.1016/j.isci.2026.115419 Search in PMC Search in PubMed View in NLM Catalog Add to search Altered kinship vocal dynamics in marmosets with valproic acid-induced model of autism Koki Mimura Koki Mimura 1 Department of Ultrastructural Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo 187-8502, Japan 2 Department of Mental Disorder Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo 187-8502, Japan Find articles by Koki Mimura 1, 2, 4, ∗ , Keiko Nakagaki Keiko Nakagaki 1 Department of Ultrastructural Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo 187-8502, Japan Find articles by Keiko Nakagaki 1 , Hirofumi Morishita Hirofumi Morishita 2 Department of Mental Disorder Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo 187-8502, Japan 3 Department of Psychiatry, Neuroscience, and Ophthalmology, Mindich Child Health and Development Institute, Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA Find articles by Hirofumi Morishita 2, 3 , Noritaka Ichinohe Noritaka Ichinohe 1 Department of Ultrastructural Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo 187-8502, Japan Find articles by Noritaka Ichinohe 1, ∗∗ Author information Article notes Copyright and License information 1 Department of Ultrastructural Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo 187-8502, Japan 2 Department of Mental Disorder Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo 187-8502, Japan 3 Department of Psychiatry, Neuroscience, and Ophthalmology, Mindich Child Health and Development Institute, Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA ∗ Corresponding author [email protected] ∗∗ Corresponding author [email protected] 4 Lead contact Received 2025 Mar 11; Revised 2025 Nov 16; Accepted 2026 Mar 18; Collection date 2026 Apr 17. © 2026 The Authors This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). PMC Copyright notice PMCID: PMC13068532  PMID: 41971986 Summary Autism spectrum disorder (ASD) involves social communication impairments and repetitive behaviors. Language abnormalities in ASD, such as echolalia and idiosyncratic speech, heighten caregiver stress and affect communication dynamics within the kinship system. However, the influence of ASD-related traits on family-level interactions remains poorly understood in animal models. Here, we established an ASD model in common marmoset via prenatal valproic acid (VPA) exposure, and analyzed 28,418 kinship vocalizations from VPA-exposed and unexposed (UE) pups with their parents. VPA families exhibited increased isolation calls, decreased affiliative calls, disrupted repetition patterns, and reduced developmental maturations. These alternations intensified after weaning and correlated with parental weight loss, suggesting heightened caregiver stress. VPA pups also displayed premature locomotion independence, indicating broader social disruptions. Our findings highlight VPA marmosets as valuable models for investigating ASD-like traits at both individual and kinship levels, with kinship vocalizations serving as potential non-invasive biomarkers of ASD-related communication impairments and family stress. Subject areas: biological sciences, behavioral neuroscience, developmental biology Graphical abstract Open in a new tab Highlights • Prenatal VPA exposure alters kinship-level vocal dynamics in marmoset families • VPA families show increased isolation calls and reduced affiliative vocal patterns • Attenuated vocal maturation reflects ASD-relevant communication alterations Biological sciences; Behavioral neuroscience; Developmental biology Introduction Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by difficulties in social communication and restricted and repetitive behaviors. 1 These symptoms typically emerge in early childhood and intensify as social demands increase. 2 Impaired language development in patients with ASD shows characteristic alterations, including echolalia and atypical vocal patterns, reflecting broader impairments in communicative flexibility. 3 , 4 Such communication deficits not only impact the quality of life of individuals with ASD but also impose considerable stress on caregivers, exacerbating family dynamics. 5 , 6 , 7 Compared with families of children with other developmental disorders, families with ASD face heightened kinship stress, particularly during critical early developmental social milestones such as weaning and preschool entry. 8 Despite advances in neuroscience research using animal models of ASD, our understanding of kinship dynamics remains limited. Existing animal models, predominantly rodent models, have been instrumental in elucidating the neural mechanisms underlying ASD. 9 However, rodents lack the social complexity and cooperative kinship structures that characterize human families. To bridge this gap, the common marmoset ( Callithrix jacchus ), a small nonhuman primate, offers distinct advantages. Marmosets possess a kinship system that closely parallels that of humans and is characterized by paternal and elder involvement in childcare, altruistic behaviors, and rich vocal communication. During their early development, they acquire complex social skills such as third-party reciprocity and fairness within the family environment. 10 , 11 These unique traits make marmosets a valuable model for investigating the developmental dynamics of kinship systems. Building on these strengths, we developed a marmoset model of ASD using prenatal exposure to valproic acid (VPA), an antiepileptic drug clinically linked to ASD-like symptoms when exposure occurs during the fetal neural tube development stage. 12 In rodents, prenatal VPA exposure is a well-established model of ASD that results in altered neural development and social impairments. 9 In marmosets, VPA exposure induces ASD-like phenotypes, 13 including atypical neural development, 14 , 15 , 16 social impairments, 17 , 18 , 19 and heightened daily stress. 20 The VPA-exposed marmoset model incorporates typically developing parents and ASD-like pups, thus providing a unique framework for investigating how ASD traits affect kinship dynamics over time. ASD symptoms often emerge during key developmental milestones, such as weaning, which occurs between postnatal months 2 and 3 (PM 2–3) in marmosets. 21 This period offers a valuable opportunity to examine behavioral phenotypes and kinship interactions that may reflect ASD-like traits. We hypothesized that kinship behaviors during weaning could serve as sensitive, non-invasive indicators of stress and social communication impairment in marmoset families with VPA-exposed pups (VPA families). A unique feature of marmosets is their rich vocal repertoire, which earns them the nickname-song monkeys. They produce more than 10 distinct call types, 22 including the isolation-indicating phee call 23 and the socially engaging trill call. 24 These vocalizations provide a window into social and emotional dynamics, making them particularly suited for studying the interactions between the pup models of ASD and typically developing family members. Existing approaches to vocal communication analysis in animal models often require expensive methods to identify individual callers, limiting their scalability and ecological validity. 23 , 25 , 26 , 27 To address these challenges, drawing on insights from natural language analysis in ASD, 28 , 29 we focused on call frequency and sequence patterns, instead of individual identification, to capture the overarching social dynamics within the kinship system. In this study, we aimed to explore how VPA exposure affects kinship behavior and stress, particularly during the weaning period. By recording and analyzing locomotion, vocalizations, and body weights across marmoset families from PM 1 to 5.5, we sought to uncover atypical patterns associated with ASD traits ( Figure 1 A). Our approach not only advances the understanding of ASD-related family dynamics but also holds potential for clinical applications, offering insights into stress and communication impairments in human families affected by ASD. Figure 1. Open in a new tab Characteristics of VPA families in terms of behavior and parental body weight (A) Pups born to mothers that received intragastric (i.g.) administration of VPA during pregnancy are referred to as VPA pups. Kinship behaviors, including locomotion and vocalizations, are recorded by transferring the father, mother, and either a VPA or UE pup in their home cage to a soundproof room. (B) Behavior data included 16 pups (VPA = 9; UE = 7) from five families. The inverted Y-branch represents twin pairs. (C) Body weights of pups, overlaid with typical behavioral milestones (top, blue) and the recording period (PM 1–5.5, bottom, orange). (D) Scatterplot shows the carrying time of pups (time being transported by parents) during 30-min recording sessions. Two twin pairs, VPA8/VPA9 and UE7/UE6, are highlighted with distinct line types and labeled. (E) Boxplots of carrying time during the first recording session around PM1 (gray squares in D). Markers are differentiated by parental pairs. VPA families exhibit longer independent activity durations than UE families ( p < 0.05, Brunner-Munzel test). (F) Scatterplots with loess regression lines depicting scaled body weight changes for the fathers and mothers of recorded pups. Body weights are scaled relative to the average during PM -3 to −1. Paternal body weight significantly decreased during the caregiving period for VPA pups ( p < 0.05, analysis of covariance). (G) Scatterplots with loess regression lines of scaled body weight during the parenting period (VPA = 17; UE = 15) for the same parental pairs. To confirm robustness, the analysis includes data from other caregiving periods experienced by the same parents. Significant decreases in paternal body weight for VPA pups are observed at PM 2, 4, and 5 ( p < 0.05, Tukey’s honestly significant difference test). The blue squares in F and G indicate the weaning period of pups, consistent with those shown in C. VPA, valproic acid; UE, unexposed; PM, pups’ age in postnatal months. Results Early reduction of carrying time in VPA pups We investigated the effect of VPA on kinship-level behavioral expression in pups throughout their early developmental stages. Kinship behavior was recorded using a video camera mounted above the home cage in a soundproof room, allowing one pup and its parents to move freely for 30 min ( Figure 1 A). This study included nine VPA-exposed pups (seven males and two females) and seven unexposed (UE) pups (five males and two females) as controls ( Figure 1 B and Table S1 ). Body weight, monitored as a physical growth indicator, was not significantly different between the groups during PM 0 to 7 (f[1, 13] = 1.482, p = 0.247, repeated-measures analysis of covariance (ANCOVA); Figure 1 C). A total of 109 recording sessions (UE = 46; VPA = 63) were conducted biweekly from PM 1 to 5.5 spanning three key behavioral milestones: locomotion independence (∼PM 2), weaning (PM 2 and 3), and increased social behaviors such as grooming and head-to-head contact between pups and parents (PM 4 onwards; Figure 1 C, blue arrows). The first milestone—pup carrying by the parents—typically decreased PM 3 in most pups, with a few exceptions in both groups ( Figure 1 D). Interestingly, the effects of VPA exposure were evident as early as the initial recording session around PM 1. VPA pups exhibited significantly shorter carrying time compared to UE pups (UE = 6, VPA = 7, BM statistic = 2.62, df = 8.46, p = 0.029, Brunner-Munzel test; Figure 1 E). In this novel recording context, the reduced tendency of VPA pups to remain carried by their parents, or the lack of facilitation by parents, suggests potential alterations in their kinship system. Paternal body weight loss during the weaning period of VPA pups To evaluate whether VPA exposure affected not only pup behavior but also the broader kinship system, parental body weights were continuously monitored. Maternal body weight transitioned from decreasing to increasing around PM 3, corresponding to the weaning period, with no significant effects attributed to VPA exposure (PM 0–7, f[1, 48] = 0.64, p = 0.428, ANCOVA). By contrast, fathers showed a significant reduction in body weight during the caregiving period for VPA pups (PM 0–7, f[1,42] = 16.7, p = 1.96e-4, ANCOVA; Figure 1 F). To confirm the robustness of this trend, a systematic analysis was conducted on parental body weight during caregiving, including data from births not associated with the recorded sessions (VPA = 13, UE = 15; Table S2 ). This analysis replicated this pattern, with significant reductions in paternal body weight observed at PM 2 (estimate = −0.046, adjusted p = 0.048), 4 (estimate = −0.057, adjusted p = 0.022), and 5 (estimate = −0.062, adjusted p = 0.025, Tukey’s honestly significant difference (Tukey-HSD) test; Figure 1 G). Unlike maternal weight, which is influenced by factors such as childbirth, lactation, and subsequent pregnancies, paternal weight loss is less likely to be affected by these confounders. These findings suggest that caregiving for VPA pups may require more effort from fathers than caregiving for UE pups. This weight loss trend became apparent after the pups achieved independent locomotion behavior, suggesting the need to investigate additional metrics to better understand the changes in kinship dynamics during this period. The number of kinship calls in VPA families did not decrease with pup development A total of 28,418 calls were systematically collected from kinship behavior records, providing a dataset of vocalizations, recognized as a commonly used non-invasive indicator for assessing collective emotional states and enabling longitudinal comparisons. All vocalization analyses were conducted without identifying the subject that produced the calls. In UE families, the frequency of kinship calls decreased with pup development ( n = 46, f[1,33] = 5, p = 0.032, F-test), which is consistent with previous studies reporting a decline in pup call frequency under isolated conditions as development progressed. However, this decline was not observed in VPA families, where the call frequency remained stable over time ( n = 63, f[1,45] = 0.03, p = 0.954, F-test; Figure 2 A). These results from the simple regression models were supported by linear mixed model (LMM) analysis, which was conducted to examine the model while accounting for the random effects of individual and parental differences. Based on model selection using the Akaike information criterion (AIC), the best model for UE families showed a negative correlation with PM, including the random effects of individual differences (model #13; Tables S3 and S4 ), with a fixed effect of −81.3. Model 13, the best model, was validated using a likelihood ratio test (LRT) against model 4, which excluded the treatment effect, confirming a significant improvement in the data explanation (deviance residual (DR) = 12.7, df = 1, p = 3.61e-4, LRT). For VPA families, a negative correlation with PM was observed, including the random effects of parental differences (model #14, DR = 9.30, df = 1, p = 0.0229, LRT vs. model #5), but the fixed effect was significantly smaller at −0.186. The lack of a typical developmental decrease in the kinship vocal frequency and pup development in VPA families suggests potential alterations or immaturity in the rules of vocal usage. In addition, in this analysis and throughout all subsequent results of kinship call analyses, no significant effects of sex differences were detected. Figure 2. Open in a new tab Abnormal kinship vocal call frequencies in VPA families (A) Scatterplot showing the number of kinship vocal calls. Data points are connected by thin lines for each measured family pair. (B) Typical spectrogram of the nine major call types. These data are sampled from individual datasets separated by dotted lines and combined for visualization ( Table S2 ). (C) Call type proportions aggregated across all ages. (D) Examples of kinship vocalizations recorded during 30 min of free activity for VPA4 (top) and UE7 (bottom) pups, along with their parents (BE-DO in Figure 1 B). The x axis shows timestamps, and the y axis indicates call types. Columns represent recordings at PM 1, 3.75, and 5 or 5.25. (E) Scatterplots of call frequency. Significant atypical trends are detected in VPA families for three types of calls: ock , phee , and trill (∗, p < 0.05, analysis of covariance) ( Table S5 ). VPA, valproic acid; UE, unexposed; PM, pups’ age in postnatal months. Affiliative calls decreased while anxiety-related calls increased in VPA families To evaluate the social stress conditions in VPA families, we identified diverse call types within kinship vocalizations and focused on the frequency of calls linked to emotional states. Excluding 125 unclassifiable calls, the remaining calls were categorized into nine distinct types: ock , tsik , egg , phee , peep , trill , trillphee , chirp , and twitter ( Figure 2 B and Table S5 ). Of all calls, trill— known as a social affiliative call—was the most frequent at 50.3% ( n = 14,426), whereas phee— known as an anxiety-related call—accounted for only 3% ( n = 955). No significant differences were observed in the composition ratios of calls pooled across the entire measurement period between the UE and VPA families (statistic = 0.225, df = 9, p = 1.00, chi-square test; Figure 2 C). However, the visualization of representative examples revealed distinct distributions of call types, with an increase in phee observed in VPA families and an increase in ock in UE families ( Figure 2 D). Further statistical comparisons of call frequencies by type highlighted notable differences: In VPA families, the frequency of trill significantly decreased (f[1,102] = 5.56, p = 0.020, ANCOVA), whereas that of phee significantly increased (f[1,68] = 7.50, p = 0.008, ANCOVA). A significant negative correlation with PM was also observed for trill (f[1,102] = 8.64, p = 0.004, ANCOVA). Additionally, notable changes were observed in ambiguous calls such as ock and twitter . In VPA families, the frequency of ock significantly decreased (f[1,98] = 5.71, p = 0.019, ANCOVA) and most of twitter (38/39) calls were observed in VPA families ( Figure 2 E). These changes in call types were supported by the LMM analysis, where significant VPA exposure effects were detected across trill , phee , and ock . Random effects were also observed for pups in terms of trill (model #17 in Table S3 selected by LMM analysis, DR = 17.9, df = 2, p = 1.30e-4, LRT vs. model #13) and for parents in terms of phee and ock (model #18; phee , DR = 21.5, df = 2, p = 2.16e-5; ock , DR = 7.72, df = 2, p = 2.11e-2, LRT vs. model #14; Tables S3 and S6 ). The frequency of representative calls, which is known to be associated with emotional states, showed a consistent trend of increased social stress and decreased affiliative mood throughout the measurement period. Call repetition frequency of kinship vocalization was altered in VPA families Given that impairments in language communication in ASD are characterized by echolalia or unnatural repetition of words, leading to reduced communicative flexibility, we examined the sequential organization of calls in kinship vocalizations. Specifically, we analyzed all pairs of calls with inter-call intervals (ICIs) ≤30 s ( Figure 3 A), categorizing them as either repeated or non-repeated and further classified into short ICIs (ICI ≤10 s) or long ICIs (10 < ICI ≤30 s). Thresholds of 10 s and 30 s corresponded approximately to the 85th and 96th percentiles of the ICI distribution, respectively ( Figure 3 B). In UE families, the number of short ICI repeated calls decreased with pup development ( n = 5539, f[1,43] = 15.8, p = 2.69e-4, F-test), whereas that of long ICI non-repeated calls increased ( n = 1289, f[1,43] = 9.75, p = 0.003, F-test) ( Figure 3 C). These developmental trends were supported by the LMM analysis, with model #10 being the best fit ( Table S3 ). By contrast, no such trends were observed in VPA families (short ICI repeated: n = 7017, f[1,61] = 0.02, p = 0.89, F-test; model 5, DR = −8.40, df = 1, p = 1.00, LRT vs. model #14; long ICI non-repeated: n = 2213, f[1,61] = 0.401, p = 0.53; model #1). These findings suggest a disruption of sequential rules in the kinship vocalizations of VPA families, diverging from the structured developmental patterns observed in UE families ( Tables S3 and S7 ). Figure 3. Open in a new tab Alteration of call repetition rules in VPA families (A) Example spectrogram of kinship vocalizations from UE2 and its parents at 1.5 p.m. The inter-call interval (ICI) is defined as the time between the end of a pre-call and the start of a post-call. (B) Density distribution of the ICI. Two calls with ICI ≤30 s are classified as related calls and further divided into short (ICI ≤10 s) and long (10 < ICI ≤30 s) intervals. The percentiles of the ICI for UE and VPA families are indicated above the density plot in black and red, respectively. (C) Longitudinal comparison of call repetition rules during development. Related calls are categorized into four types based on the ICI (short or long) and whether they are repetitions of the same call type (repeated) or different types (non-repeated) ( Tables S6 and S7 ). (D) Entropy of 2-call phrases categorized by repetition and ICI. (E) Proportional representation of each call type in repeated calls within a short ICI (≤10). Significant VPA exposure effects are observed for three repeated call types based on LMM analysis: trill-trill (UE > VPA), trillphee-trillphee (UE < VPA), and chirp-chirp (UE < VPA) ( Table S7 ). VPA, valproic acid; UE, unexposed; PM, pups’ age in postnatal months. Disruption of structured repetition patterns in short ICI calls of VPA families To characterize the compositional changes in repeated calls, we analyzed the distribution patterns of 2-call phrases, focusing specifically on short ICI repeats. Entropy analysis (average information, H ) of the four categories of 2-call phrases revealed significantly higher diversity in VPA families, reflecting a departure from the organized vocal structure observed in UE families (f[1,102] = 10.18, p = 0.002, ANCOVA; model #6 in Table S3 and Figure 3 D). A detailed analysis of the 10 call types (9 categories plus “other calls”) revealed specific differences in repetition patterns. In UE families, trill-trill accounted for approximately 75% of the short ICI repeats, whereas in VPA families, this proportion significantly decreased to approximately 50% (f[1,95] = 22.2, p = 8.43e-6, ANCOVA; model #9). Conversely, VPA families exhibited significant increases in trillphee-trillphee (f[1,16] = 5.41, p = 0.034, ANCOVA; model #15) and chirp-chirp (f[1,76] = 12.8, p = 6.13e-4, ANCOVA; model #6) repeats. Although the frequency of peep-peep was higher in UE families than in VPA families (f[1,78] = 12.8, p = 6.13e-4, ANCOVA), the random effects of parental differences absorbed this variation, resulting in no significant group differences in the LMM analysis (model #5) ( Figure 3 E and Table S8 ). These findings demonstrate that VPA families deviate from socially meaningful and consistent repetition patterns, such as trill-trill , commonly observed in UE families. Instead, VPA families exhibited an increase in non-affiliative repetitions. This breakdown of structured vocal patterns parallels ASD-like echolalia, which is characterized by atypical repetitive vocal usage in social communication. Collectively, these results highlight how VPA exposure disrupts the organization of socially relevant vocal sequences, leading to more variable and less structured vocal interactions. Reduced developmental changes in multi-call phrase frequency distributions in VPA families VPA families exhibited developmental delays or stagnation in kinship vocalizations, with significantly smaller changes in the frequency of single calls and 2-call phrases with specific repetition rules. Within-subject longitudinal comparisons were performed to validate this pattern. Data were divided into three developmental stages with balanced recordings from VPA families: stage 1 (PM 1–2.5, UE = 18, VPA = 21), stage 2 (PM 2.5–3.5, UE = 15, VPA = 21), and stage 3 (PM 3.5–5.5, UE = 13, VPA = 21). Jensen-Shannon divergence (JSD) was calculated to evaluate the changes in the frequency distributions of single calls and multi-call phrases (up to six consecutive calls, ICI ≤30). Within-subject comparisons showed a significantly lower JSD in VPA families than in UE families for 4- to 6-call phrases between stages 2 and 3 ( p < 0.05, Brunner-Munzel test). No significant differences were observed between stages 1 and 2 ( Figure 4 A). These findings indicate that focusing on multi-call phrase frequency distributions highlights developmental stagnation in kinship vocal usage in VPA families, which is evident in the form of reduced changes within individuals, particularly during later weaning stages. Figure 4. Open in a new tab Discrimination of VPA families based on kinship vocal patterns (A) Boxplots show the expected changes in entropy (JSD) with the development of the related call sequence frequency. The sequences were classified as 1- to 6-call phrases based on the number of consecutive calls (ICI ≤10). Comparisons are made between developmental stage 1 (PM 1–2.5) and stage 2 (PM 2.5–4; top) and between stages 2 and 3 (PM 4–5.5; bottom). Significant differences are observed between stages 2 and 3 for 4-call (BM statistic = −3.12, df = 8.94, p = 0.012), 5-call (BM statistic = −3.12, df = 8.94, p = 0.012), and 6-call phrases (BM statistic = −4.19, f = 8.93, p = 2.3e-3, Brunner-Munzel test). (B) Mahalanobis distance from UE families for the frequency distribution of 4-call phrases is calculated based on the top 1–5 PC scores across the three developmental stages. The number of data points included in each developmental stage is shown at the top. These stages correspond to typical behavioral milestones in marmosets (blue arrows; see Table S8 ). (C) Results of the discriminant analysis, showing whether data points deviate from the expected 95% distribution range of UE families (gray box in B). Outlier data points are filled in, whereas non-deviating points are displayed as a hollow. (D and E) Validation of discriminant models to determine UE or VPA. False negative rates (red) and false positive rates (black) are calculated from C (D) or from the frequency of single calls (E). (F) Scatterplots with regression lines of the best model for the correlation between the scaled parental body weight ( Figure 1 F) and log10 of Mahalanobis distance in B ( Table S11 ). VPA, valproic acid; UE, unexposed; ICI, inter-call interval; PC, principal component; PM, pups’ age in postnatal months; JSD, Jensen-Shannon divergence. Developmental deviations in 4-call phrasing in VPA families The finding that multi-call phrase frequencies prominently reflect the characteristics of VPA families opens up avenues for their use as biomarkers to distinguish families with ASD-like pups. To test this potential, we performed a discriminant analysis focusing on the frequency of 4-call phrases. When aggregating representative 4-call phrases, 4- trill phrases were the most frequent across all developmental stages in UE families. However, in VPA families, 4- chirp phrases and 4- peep phrases consistently showed high frequencies, whereas these phrases tended to be less frequent in UE families during stages 2 and 3 ( Table S9 ). From the dataset, 1713 unique 4-call phrases were identified, and their count data were reduced to five dimensions (explaining 98.45% of the variance; Table S10 ) using principal component analysis (PCA). We calculated the Mahalanobis distances from the UE mean vector to the individual data across the three developmental stages. This enabled the assessment of the deviation of each data point ( Figure 4 B). Deviations were defined as exceeding the 95th percentile of the UE distribution and were observed in 63.5% (40/63) of VPA families, with a tendency to increase with pup development ( Figure 4 C). The false negative rate, indicating the proportion of VPA families indistinguishable from UE families, was the highest during stage 1 at 81% (17/21). However, this value steadily declined as the pups developed, reaching 9.5% (2/21) in stage 3 ( Figure 4 D, indicated by red). The false positive rate, which indicates the proportion of UE families classified as not UE families, progressively declined and reached zero in stage 3 ( Figure 3 D, indicated by black). For comparison, a similar discriminant analysis was conducted using single-call frequency data. The 10-dimensional single-call data were reduced to four principal components to match the explained variance (98.4%; Table S11 ) in the 4-call analysis. The single-call analysis showed a low false positive rate of 0.07 in stage 3, aligning closely with the theoretically expected value of 0.05 for random deviations. The false negative rate remained at 0.59, which is close to chance (21/34 = 0.617), indicating poor discriminative performance ( Figure 4 E). These findings underscore the developmental divergence of 4-call phrase patterns in VPA families, highlighting their potential as non-invasive biomarkers for ASD-like pups and their families. Notably, this utility is achieved without the need to identify the individual producing each call, a step that typically incurs significant costs in conventional vocal analysis, making this approach both efficient and practical. Parental weight loss correlates with kinship vocalization deviation The relationship between parental caregiver stress, indicated by weight loss, and deviations in kinship vocal communication, measured by the frequency-based divergence of 4-call phrases, was examined using a correlation analysis. In VPA families, a significant negative correlation was observed for both fathers and mothers (model #5 in Table S12 ; father, fix effect = −3.841, DR = 5.80, df = 1, p = 0.015; mother, fix effect = −3.093, DR = 5.05, df = 1, p = 0.024, LRT vs. model #2), indicating that greater weight loss was associated with more pronounced vocalization deviations. No such correlation was found in UE families (model #1 in Table S12 and Figure 4 F). These findings suggest that stress-related body weight loss in parents is linked to atypical kinship vocal patterns in VPA families. Discussion This study revealed that prenatal VPA-exposure not only affects individual marmosets but also disrupts kinship dynamics, characterized by reduced carrying time, atypical vocalization patterns, and significant paternal weight loss. VPA pups exhibited shortened parental carrying even during the developmental phase, whereas control pups continued to rely on parental transport. The nature of kinship vocalizations in VPA families may reflect heightened stress among family members, marked by unnatural call repetitions and minimal changes across the pre-, during-, and post-weaning periods. The frequency of call phrases in VPA families significantly deviated from the convergent patterns observed in control families after weaning. This deviation enabled the successful identification of VPA families, suggesting the potential use of kinship vocal metrics as non-invasive biomarkers. Furthermore, the inverse correlation between call phrase deviation and parental body weight highlights how altered communication may contribute to caregiver stress within the family. These findings underscore the value of prenatally VPA-exposed marmosets as a model for studying both individual ASD phenotypes and the broader impact of ASD-like traits on family dynamics, with kinship vocalizations offering a sensitive tool for assessing stress and communication impairments. VPA pups exhibited reduced carrying time despite having weight trajectories, a key indicator of physical development, comparable to those of the control groups. As marmosets are arboreal primates, infants spend most of their time being carried by their parents immediately after birth, and normal carrying plays a critical role in development. This early reduction in parental carrying time suggests disruptions in parent-pup communication during the early developmental stages. We have previously reported that VPA pups exhibit reduced social gaze and social attention by PM 3, including impairments in attachment formation, 20 and the findings of the present are consistent with these observations. Such behavior may reflect ASD-like phenotypes, potentially mirroring hypersensitivity or hypo-responsiveness to physical contact observed in early childhood, as well as challenges in attachment formation. 30 To better understand the implications of early independence, it is essential to investigate whether this behavior arises naturally or is influenced by the novel recording environment and whether it reflects the avoidance of parental carrying by the pup or reduced facilitation by parents. Because pup carrying restricts parental locomotor ability and fathers are typically the primary carriers, 25 one might expect reduced carrying in VPA pups to mitigate paternal weight loss. However, fathers in VPA families still exhibited significant weight loss around the weaning period, suggesting that disruptions in parent-infant communication may contribute to increased caregiving stress. These findings emphasize the need for behavioral metrics, such as kinship vocalizations, to assess caregiver stress and family dynamics, particularly in later developmental stages. The limited developmental changes in kinship vocalizations in VPA families were evident in their failure to exhibit typical vocal maturation observed in control families. Although the initial vocalization patterns during the pre-weaning stage (PM 1–2.5) were comparable between the VPA and control families, significant differences emerged during later stages (PM 4–5.5). This parallels a key feature of ASD in humans, where language delays become apparent between 12 and 24 months of age as social demands increase. 31 In VPA families, critical traits such as declining call frequency, maturing call repetition patterns, and developmental changes in phrase distribution were significantly reduced, resulting in kinship vocalizations in an immature state. This developmental stagnation, combined with the paradox of premature locomotion independence, likely heightened caregiving stress and disrupted kinship dynamics. Such disruptions mirror the broader features of ASD, in which impaired communication and increased caregiving challenges significantly impact the kinship system. 32 These findings underscore the utility of VPA-exposed marmosets as a model not only for individual ASD-like traits but also for family-level dynamics, providing insights into the interplay between communication impairments and caregiving stress in the context of ASD. Further analysis revealed significant disruptions in the composition of calls across hierarchical levels: single calls, paired calls, and extended phrases. In the single-call analysis, trill calls accounted for half of all calls, suggesting that the recordings captured the affiliative social mood. This aligns with the findings that phee calls dominate in isolated conditions, 23 whereas the number of trill calls increases in affiliative contexts. 24 However, VPA families exhibited a pattern resembling isolated conditions, with an increased number of phee calls and a decreased number of trill calls, suggesting weakened social communication. The decline in the frequency of ock calls, often associated with contact behavior, 22 further reflects reduced communication within the kinship system. Additionally, the occasional presence of ambiguous twitter calls in VPA families, which were almost absent in control families, underscores the broader disruption in communication dynamics caused by VPA exposure. These findings highlight the social and emotional effects of VPA exposure and provide insights into how ASD-like traits disrupt natural kinship communication. Paired calls were analyzed based on repetition and the ICI length, revealing insights into kinship vocal dynamics. In UE families, trill-trill pairs, which are strongly associated with affiliative interactions, were consistently predominant, replicating the findings of previous studies. By contrast, VPA families displayed deviant and less affiliative repetitions such as trillphee-trillphee and chirp-chirp . These atypical patterns may reflect tendencies analogous to echolalia or self-repetition, which are the hallmarks of ASD language characteristics. 33 Furthermore, our analysis highlighted the critical role of extended sequences consisting of four or more calls in distinguishing between VPA and control families. This distinction holds regardless of whether the sequences represent individual vocalizations or social exchanges, underscoring the need to extend the vocal structures in kinship communication analyzes. 34 Relatedly, naturalistic vocal analyses in human ASD have also revealed alterations in social feedback dynamics. 35 Studies using wearable systems such as the language environment analysis (LENA) system 36 , 37 automatically extracted acoustic features from day-long recordings to estimate conversational turn-taking between children and caregivers, often using adult speech as an anchor to define response windows. Importantly, these approaches rely solely on acoustic features and cannot access the semantic or intentional content of speech, making their interpretation necessarily inferential. While such work has provided early evidence that naturalistic vocal metrics may hold promise for early ASD screening, their clinical and mechanistic applicability remains limited due to environmental and linguistic heterogeneity. However, human communication is profoundly shaped by linguistic complexity, environmental heterogeneity, and individual variability, which constrain mechanistic generalization from acoustic proxies alone. In contrast, the VPA marmoset model offers a developmentally homogeneous background and a simpler yet evolutionarily conserved vocal repertoire, allowing the isolation of the core temporal and social coordination principles of communication. The identification of extended four-call sequences as a major discriminator between VPA and control families may provide an abstract analytic framework that can inform future human studies as AI-based semantic and intentional modeling continues to advance. 38 In future work, linking these behavioral metrics with clinical data will be essential to validate their translational relevance. Caregivers of individuals with ASD are known to experience higher stress than those of children with other developmental disorders, with language alterations identified as one of the major contributing factors. 5 The ripple effects of ASD on family stress often become apparent during major social transitions such as school entry and have been shown to increase vulnerability to external stressors such as the COVID-19 pandemic, making this a major concern for clinical and welfare support. 8 Demonstrating that similar kinship vocalization patterns and stress-coupling dynamics exist in human ASD families—where parental stress and emotional resilience are major modulators of social reciprocity—would further justify the use of the VPA marmoset model. Such cross-species correspondence would not only strengthen mechanistic interpretation but also enable clinical application: These family-level vocal and stress measures could be used for early risk screening, longitudinal monitoring of intervention efficacy, and even subtyping, given the heterogeneity of human ASD. For example, such metrics could be implemented in naturalistic, home-based audio monitoring of at-risk infants to detect atypical vocal coordination, or used to quantify family-level changes in reciprocity during therapeutic interventions. This study has several limitations. A sex difference has been established in ASD prevalence, and sex differences have also been reported in the co-occurrence of intellectual disability. 39 The sex ratio of our marmoset cohort was roughly consistent with the reported male-to-female ratio of ASD. However, no significant effects of sex were detected in the kinship vocalization analyses. The limited statistical power was further compounded by missing data due to health conditions, and experimental conflicts led to an incomplete repeated-measures design. To address this, we applied LMMs and LRTs to account for individual and parental effects. Constraints in the soundproof room and experimental cage limited behavioral observations to basic metrics, such as whether the pups were carried by their parent. Parent-infant interactions—such as clinging or pups riding on the parent’s back—remain particularly challenging to capture and require multi-camera systems. 40 , 41 Our decision to avoid linking vocalizations to specific individuals preserved natural social interactions within a simple recording setup, but it also precluded detailed analyses of interactive versus solitary communication. Advances in machine learning enabling individual identification from vocal data offer potential solutions. 42 Nevertheless, label-free vocal analysis remains advantageous for clinical applications because it reduces labor and respects privacy. Furthermore, while this study suggests a possible association between prenatal VPA exposure, kinship vocal disruption, and increased caregiving stress within the family, the underlying neurophysiological mechanisms were not directly examined. VPA’s effects on neurodevelopment—including epigenetic modifications, altered neurogenesis, and disrupted synaptogenesis 16 , 43 , 44 —have been characterized across multiple systems. In our marmoset model, we have traced these molecular disruptions to specific circuit-level and behavioral consequences: hypogenesis of the anterior commissure, 14 which is essential for interhemispheric exchange of emotional information in early postnatal life, along with reduced synaptic density in the medial prefrontal cortex, 15 a region implicated in social behavior. These early abnormalities may prime the later-emerging alterations in social interaction observed after 2–3 months of age. During this developmental period, transient cortical hyper-synaptogenesis and excessive synaptic plasticity also occur, which could further disrupt the maturation of coordinated social behavior. In adult VPA marmosets, electrocorticography (ECoG) recordings have demonstrated abnormal predictive coding responses to auditory stimuli, 45 suggesting altered auditory perception and impaired generation of appropriate calls that may draw others into atypical interactive patterns. In addition, elevated cortisol levels previously observed in VPA individuals indicate heightened stress, which could propagate within the family through emotional contagion. Collectively, these findings suggest a multilevel cascade—from structural disconnection to functional desynchronization and social stress transmission—that warrants further investigation to clarify its integrative mechanisms. Future studies combining simultaneous family-level recordings of vocal exchanges, neural activity, autonomic signals, and stress markers will be crucial to capture the temporal coupling underlying social communication. Integrating these correlative approaches with causal manipulations, such as chemogenetic modulation 46 , 47 of medial prefrontal or commissural pathways, may reveal how disrupted circuit synchronization and stress propagation interact across family members. Ultimately, computational modeling grounded in predictive coding frameworks could provide a unifying account linking neural dynamics, emotional contagion, and social communication in VPA-exposed marmosets. In summary, this study demonstrates that prenatal VPA exposure disrupts both individual development and kinship dynamics in marmosets, characterized by premature locomotion independence, immature vocal communication patterns, and heightened caregiving stress. The observed stagnation in kinship vocalization development, coupled with atypical pup locomotion, highlights the broader impact of ASD-like traits on family interactions. By leveraging non-invasive vocal metrics without individual identification, this study offers a scalable approach for assessing social dynamics, with potential applications as a biomarker for ASD traits. These findings provide valuable insights into the interplay between individual phenotypes and family systems and lay the groundwork for future research to refine ASD models and their translation into human clinical contexts. Limitation of this study • Although we observed significant alterations in kinship vocalization patterns and paternal body weight, causal relationships between altered communication and caregiving stress cannot be definitively established. • Individual-level vocal identification was not performed, which precluded detailed analyses of dyadic interactions. • The underlying neural mechanisms linking prenatal VPA exposure to altered kinship dynamics were not directly examined in this study. Future studies combining simultaneous neural recordings and family-level behavioral analyses will be required to clarify these mechanisms. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the read contact, Koki Mimura ( [email protected] ). Materials availability This study did not generate new unique reagents. Data and code availability • The data generated in this study have been deposited in the Open Science Framework Database ( http://doi.org/10.5281/zenodo.18839062 ). • Custom code generated in this study is available at GitHub repository ( http://doi.org/10.5281/zenodo.18839062 ). • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments We thank Y. Okawa, R. Saito, S. Okamura, A. Tsuchiya, I. Yamamoto, and Y. Nishimura for technical assistance. We also thank Dr. J. Noguchi (NCNP), Dr. K. Yoshitake (NCNP), Dr. K. Shimatani (Institute of Statistical Mathematics), Dr. S. Nakamura (Tokyo University of Agriculture and Technology), and Dr. T. Minamimoto (National Institutes for Quantum Science and Technology) for their comments on an earlier version of the manuscript. This study was supported by JSPS Research Fellowships for Young Scientists 14J10961 (to KM), MEXT/JSPS KAKENHI under the grant no. J22K07338 (to KM), AMED under grant nos. JP24wm0625124 (to KM), JP24wm0625206 (to HM), JP19dm0207066 (to NI), and the Intramural Research Grant for Neurological and Psychiatric Disorders from NCNP under grant nos. 2-7 and 5-8 (to NI). Author contributions Conceptualization, K.M. and N.I.; formal analysis, K.M.; investigation, K.M. and N.K.; resources, K.M., N.K., and N.I.; writing (original draft), K.M.; visualization, K.M.; supervision, I.N.; project administration, K.M.; funding acquisition, K.M., H.M., and N.I.; writing – review and editing, all authors. Declaration of interests The authors declare no competing interests. Correspondence and requests for materials should be addressed to Koki Mimura or Noritaka Ichinohe. STAR★Methods Key resources table REAGENT or RESOURCE SOURCE IDENTIFIER Deposited data Marmoset kinship vocalization data This paper DOI: http://doi.org/10.5281/zenodo.18839062 Marmoset body weight data This paper DOI: http://doi.org/10.5281/zenodo.18839062 Software and algorithms R (version 4.3.3) The R Foundation for Statistical Computing https://www.r-project.org ; RRID: SCR_001905 rstatix package (version 0.7.2) Kassambara https://cran.r-project.org/package=rstatix ; RRID: N/A lme4 package (version 1.1–35.5) Bates et al. https://cran.r-project.org/package=lme4 ; RRID: SCR_015654 seewave package (version 2.2.3) Sueur et al. https://cran.r-project.org/package=seewave ; RRID: SCR_003799 Syrinx (version 2.4i) Syrinx-PC (John Burt) https://syrinxpc.com/ ; RRID: N/A Open in a new tab Experimental model and study participant details Animals In this study, we used 16 marmosets born from nine litters along with their parents (six males and five females; Figure 1 B). The parental experience with birth and caregiving ranged from 1 to 7 times (mean ± standard deviation (SD): mother, 3.7 ± 1.9; fathers, 4.2 ± 2.0). The age of the mothers at the time of birth was 5.7 ± 1.8 years, and the age of the fathers was 7.8 ± 3.9 years ( Table S1 ). The sex of the pups is summarized in Table S1 . The experimental period lasted from the birth of the pups (PM 0) to PM 5.5, during which time the parents and pups were housed together in the same cage (500 × 600 × 800 mm [width × depth × height]). Marmosets and their dams were housed in family cages and provided with food (CMS-1, CREA-Japan Inc., Tokyo, Japan) and water ad libitum . The subjects were kept at room temperature of 29 ± 2.0 °C and maintained on a 12 h:12 h light–dark cycle. The lights in the breeding room were turned on at 7:00 a.m. and off daily at 7:00 p.m. The marmosets in the facility were familiar with human contact and approached the experimenter to obtain food rewards without hesitation. Ethical approval All experimental and animal care procedures were approved by the Animal Research Committee of the National Center of Neurology and Psychiatry (NCNP), Tokyo (#2011-013, #2014-006, and #2017-004), and were performed in accordance with the United States National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publication Nos. 80–23) and the Guide for Care and Use of Laboratory Primates published by the National Institute of Neuroscience, NCNP. Method details Valproic acid (VPA) treatment Of the nine litters, five were obtained by administering VPA to the mothers during pregnancy, resulting in VPA-exposed offspring. These parent–offspring groups were defined as the VPA family group. The remaining four litters were from the mothers that were not exposed to VPA and served as the unexposed (UE) control family group. VPA marmosets were obtained using the same procedure as previously described. 13 , 18 Briefly, the dams were mated in their paired cages, and their blood progesterone levels were periodically monitored to determine the timing of pregnancy. Blood samples (<0.3 mL) were collected twice weekly from the femoral vein of un-anaesthetized animals placed in a restrainer (CL-4532, CLEA, Japan, Inc.). The dams received seven intragastric administrations of VPA sodium salt (200 mg/kg/day; Sigma-Aldrich, St. Louis, MO, USA) from days 60–66 after conception ( Figure 1 A). VPA was dissolved in 10% glucose solution immediately prior to administration. Body weight analysis The body weights of the pups and parents were measured regularly during routine health checks. Pup weight comparisons, which had minimal missing data, were analyzed using repeated-measures ANCOVA. Postnatal month (PM), VPA treatment, and their interaction were included as fixed effects, and subject was specified as a repeated factor to account for within-subject dependence ( Figure 1 C). The parental body weight was calculated as a monthly average and scaled for comparison, with the period from PM -3 to −1 normalized to 1. Data from specific parent pairs were excluded from the analysis: father KU, who died suddenly during the caregiving period for pups UE2 and UE3, and the parent pairs CH and NI, who lacked experience with UE pups. Based on these criteria, body weight data were collected from the birth experiences of all the other parents ( Table S2 ). Parental body weight group comparisons were conducted using repeated-measures ANCOVA, as was performed for pups ( Figure 1 E). When data sufficiency was allowed, further analysis was performed using Tukey’s honest significant difference (HSD) test for pairwise comparisons ( Figure 1 G). Vocal recording and call classification Kinship vocalizations were recorded using a video recorder (HDR-CX630V, SONY) with the parents and one pup inside their home cage, which was moved to a soundproof room. The recordings were taken between 10:00 a.m. and 1:00 p.m., which corresponds to an active period for marmosets following the feeding provided between 9:30 and 10:30 a.m. The recording began immediately after the door was closed and continued for 30 min, during which time the animals were free to move within the cage. Recordings were conducted every 1–2 weeks from the pups’ PM 1 to 5.5. However, owing to the parents’ health conditions and conflicts with other experiments, there were missing data; thus, the habituation of the families to the recording environment was not standardized. By using Syrinx version 2.4i ( https://syrinxpc.com/ ) and R package {seewave} version 2.2.3, 48 all recorded calls were visually inspected on a spectrogram and manually classified by experts (K.M. and a technical staff Y.O.) and recorded their starting and ending timestamps, referencing characteristic features. The evaluator K.M. was not blinded to the experimental conditions (UE or VPA), whereas Y.O. performed the classifications under blinded conditions. For a subset of 250 randomly extracted calls, intra-rater and inter-rater reliability were assessed, yielding intraclass correlation coefficient (ICC) values of 0.75 and 0.868, respectively. Locomotion analysis During the recording sessions, whether pups were being carried by a parent was determined through video observation. Instances where the pup’s limbs were completely off the floor or walls, relying entirely on either the father or mother for transport, were categorized as “carried”. The total carrying time during each 30-min recording session was then quantified. The classification relied solely on top-view video footage, and ambiguous situations—such as unclear carried states or when the pup moved into blind areas—were included in the carried category. Carrying time around PM 1 was compared between the groups using the Brunner–Munzel test to assess the median differences nonparametrically. Linear mixed model (LMM) analysis Owing to the irregular occurrence of missing data, applying a simple repeated-measures analysis of variance (ANOVA) was not feasible. Therefore, hypothesis testing was conducted using the ANCOVA, and model selection was performed using an LMM. In the LMM analysis, pups’ age and VPA treatment conditions were regressed against various vocalization data, with pup sex, pup ID, and parent ID included as random effects. We ensured the robustness of the LMM analysis by conducting likelihood ratio tests (LRTs) based on the chi-square distribution of likelihood ratios between nested models. All nested models are presented in Tables S3 and S12 . The best model was selected based on the Akaike information criterion (AIC) to assess correlations between pups’ age and significance of the VPA treatment effect. Phrasing distribution analysis For two consecutive calls, the interval between the end of the pre-call and the start of the post-call was defined as the ICI. Calls were considered related when the ICI was ≤30 s. Among related calls, those with an ICI ≤10 s were classified as “short ICI,” whereas those with an ICI >10 s were classified as “long ICI” ( Figure 2 A). Additionally, when aggregating call phrases consisting of three or more calls, only those with all ICIs ≤30 s were included in the analysis ( Figure 3 ). To quantify the change in phrase diversity between two vocal recording sessions, we adopted the Jensen-Shannon divergence (JSD). JSD represents the expected difference between two frequency distributions, resolving the asymmetry present in the simple expectation of the Kullback–Leibler divergence (KLD). The JSD between the two frequency distributions P and Q is defined as D J S ( P ‖ Q ) = 1 2 ( D K L ( P ‖ P 2 + Q 2 ) + D K L ( Q ‖ P 2 + Q 2 ) ) (Equation 1) where D KL represents the KLD. The KLD from state P to Q is calculated for each phrase a i as follows: D K L ( P ‖ Q ) = ∑ i p P ( a i ) log p P ( a i ) p Q ( a i ) (Equation 2) where p P (a i ) and p Q (a i ) denote the probability of phrase a i in distributions P and Q , respectively. The KLD quantifies the expected gain of self-information I when moving from distribution P to Q . The self-information I of state A = { a 1 , a 2 , …} is defined as follows: I A = − log p A ( A ) (Equation 3) Finally, the entropy H of state A is the expected value of I A given by the following equation: H A = E [ I A ] = − ∑ i p A ( a i ) log p A ( a i ) (Equation 4) The JSD values were computed using the jsd() function in the {philentropy} package of R. 49 Discriminant analysis In the discriminant analysis shown in Figures 4 B–4F, the Mahalanobis distance was calculated using the statsmahalanobis() function, PCA was performed using the statsprcomp() function, and the JSD was computed using the philentropyjsd() function. The data matrix, consisting of the number of occurrences of 4-call phrases with ICIs ≤30 across all PMs, was reduced to a score matrix of the first to fifth principal components using PCA based on the covariance matrix, which accounted for 98.4% of the data variance ( Table S8 ). PCA was also performed on the frequency of single calls for comparison; in this case, the first to fourth principal components were used to match the data variance (98.4%, Table S10 ). These data were then divided into three stages—stage 1 (PM 1–2.5), stage 2 (PM 2.5–4), and stage 3 (PM 4–5.5)—and Mahalanobis distances were calculated for each stage with the UE group as the reference. The threshold for identifying outliers was set to 11.0705 for 4-call phrases and 9.4877 for single calls, corresponding to the 95th percentile of the chi-square distribution with five and 4° of freedom, respectively. The Mahalanobis distance of data point A from group B is defined as follows: D M ( A ) = ( A − μ B ) T Σ B − 1 ( A − μ B ) (Equation 5) where μ B is the mean vector of group B , and Σ B − 1 is the inverse of the covariance matrix of group B . The false negative and false positive ratios were defined as the proportions of VPA and UE family data points, respectively, within a specified period where the Mahalanobis distance did not exceed (false negative) or exceeded (false positive) the threshold, which was determined as the 95% distribution range expected from UE family data points, indicating the absence or presence of observed deviations. Quantification and statistical analysis All data processing and statistical analyzes were performed using R version 4.3.3. For data processing, we employed the {tidyverse} (version 2.0.0) 50 and the {data.table} (version 1.15.0) packages. Data visualization was performed using the {ggplot2} package (version 3.5.1). For statistical analyses, type II analysis of variance (ANOVA), analysis of covariance (ANCOVA), and simple linear regression analyses were conducted using the {rstatix} package (version 0.7.2). Non-parametric hypothesis testing was performed using the {lawstat} package (version 3.6). Linear mixed-effects models (LMMs) were fitted using the {lme4} package (version 1.1–35.5) to account for repeated measures and hierarchical data structures. Details of the statistical models applied to each analysis are provided in the corresponding figure legends. Published: March 20, 2026 Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.115419 . Contributor Information Koki Mimura, Email: [email protected]. Noritaka Ichinohe, Email: [email protected]. Supplemental information Document S1. Tables S1–S12 mmc1.pdf (110.7KB, pdf) References 1. American Psychiatric Association . Diagnostic and Statistical Manual of Mental Disorders. 5th edition. American Psychiatric Publishing; 2013. [ DOI ] [ Google Scholar ] 2. Özerk G., Özerk K., Silveira-Zaldivara T. Developing social skills and social competence in children with autism. Int. Electron. J. Elem. Educ. 2021;13:341–363. doi: 10.26822/iejee.2021.195. [ DOI ] [ Google Scholar ] 3. Luyster R.J., Zane E., Wisman Weil L. Conventions for unconventional language: Revisiting a framework for spoken language features in autism. Autism Dev. Lang. Impair. 2022;7 doi: 10.1177/23969415221105472. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Schaeffer J., Abd El-Raziq M., Castroviejo E., Durrleman S., Ferré S., Grama I., Hendriks P., Kissine M., Manenti M., Marinis T., et al. Language in autism: domains, profiles and co-occurring conditions. J. Neural Transm. 2023;130:433–457. doi: 10.1007/s00702-023-02592-y. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Valicenti-McDermott M., Lawson K., Hottinger K., Seijo R., Schechtman M., Shulman L., Shinnar S. Parental stress in families of children with autism and other developmental disabilities. J. Child Neurol. 2015;30:1728–1735. doi: 10.1177/0883073815579705. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Lecavalier L., Leone S., Wiltz J. The impact of behaviour problems on caregiver stress in young people with autism spectrum disorders. J. Intellect. Disabil. Res. 2006;50:172–183. doi: 10.1111/j.1365-2788.2005.00732.x. [ DOI ] [ PubMed ] [ Google Scholar ] 7. Voliovitch Y., Leventhal J.M., Fenick A.M., Gupta A.R., Feinberg E., Hickey E.J., Shabanova V., Weitzman C. Parenting stress and its associated components prior to an autism spectrum disorder (ASD) diagnostic evaluation. J. Autism Dev. Disord. 2021;51:3432–3442. doi: 10.1007/s10803-020-04804-w. [ DOI ] [ PubMed ] [ Google Scholar ] 8. White S.W., Stoppelbein L., Scott H., Spain D. It took a pandemic: Perspectives on impact, stress, and telehealth from caregivers of people with autism. Res. Dev. Disabil. 2021;113 doi: 10.1016/j.ridd.2021.103938. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Chaliha D., Albrecht M., Vaccarezza M., Takechi R., Lam V., Al-Salami H., Mamo J. A systematic review of the valproic-acid-induced rodent model of autism. Dev. Neurosci. 2020;42:12–48. doi: 10.1159/000509109. [ DOI ] [ PubMed ] [ Google Scholar ] 10. Locke-Haydon J., Chalmers N.R. The development of infant-caregiver relationships in captive common marmosets (Callithrix jacchus) Int. J. Primatol. 1983;4:63–81. doi: 10.1007/BF02739360. [ DOI ] [ Google Scholar ] 11. Caldwell C.A., Whiten A. Scrounging facilitates social learning in common marmosets, Callithrix jacchus. Anim. Behav. 2003;65:1085–1092. doi: 10.1006/anbe.2003.2145. [ DOI ] [ Google Scholar ] 12. Christensen J., Grønborg T.K., Sørensen M.J., Schendel D., Parner E.T., Pedersen L.H., Vestergaard M. Prenatal Valproate Exposure and Risk of Autism Spectrum Disorders and Childhood Autism. JAMA. 2013;309:1696–1703. doi: 10.1001/jama.2013.2270. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Yasue M., Nakagami A., Banno T., Nakagaki K., Ichinohe N., Kawai N. Indifference of marmosets with prenatal valproate exposure to third-party non-reciprocal interactions with otherwise avoided non-reciprocal individuals. Behav. Brain Res. 2015;292:323–326. doi: 10.1016/j.bbr.2015.06.006. [ DOI ] [ PubMed ] [ Google Scholar ] 14. Mimura K., Oga T., Sasaki T., Nakagaki K., Sato C., Sumida K., Hoshino K., Saito K., Miyawaki I., Suhara T., et al. Abnormal axon guidance signals and reduced interhemispheric connection via anterior commissure in neonates of marmoset ASD model. Neuroimage. 2019;195:243–251. doi: 10.1016/j.neuroimage.2019.04.006. [ DOI ] [ PubMed ] [ Google Scholar ] 15. Watanabe S., Kurotani T., Oga T., Noguchi J., Isoda R., Nakagami A., Sakai K., Nakagaki K., Sumida K., Hoshino K., et al. Functional and molecular characterization of a non-human primate model of autism spectrum disorder shows similarity with the human disease. Nat. Commun. 2021;12:5388. doi: 10.1038/s41467-021-25487-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Noguchi J., Watanabe S., Oga T., Isoda R., Nakagaki K., Sakai K., Sumida K., Hoshino K., Saito K., Miyawaki I., et al. Altered projection-specific synaptic remodeling and its modification by oxytocin in an idiopathic autism marmoset model. Commun. Biol. 2024;7:642. doi: 10.1038/s42003-024-06345-9. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Nakagami A., Yasue M., Nakagaki K., Nakamura M., Kawai N., Ichinohe N. Reduced childhood social attention in autism model marmosets predicts impaired social skills and inflexible behavior in adulthood. Front. Psychiatry. 2022;13 doi: 10.3389/fpsyt.2022.885433. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Yasue M., Nakagami A., Nakagaki K., Ichinohe N., Kawai N. Inequity aversion is observed in common marmosets but not in marmoset models of autism induced by prenatal exposure to valproic acid. Behav. Brain Res. 2018;343:36–40. doi: 10.1016/j.bbr.2018.01.013. [ DOI ] [ PubMed ] [ Google Scholar ] 19. Kawai N., Yasue M., Banno T., Ichinohe N. Marmoset monkeys evaluate third-party reciprocity. Biol. Lett. 2014;10 doi: 10.1098/rsbl.2014.0058. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 20. Nakamura M., Nakagami A., Nakagaki K., Yasue M., Kawai N., Ichinohe N. Prenatal valproic acid-induced autism marmoset model exhibits higher salivary cortisol levels. Front. Behav. Neurosci. 2022;16 doi: 10.3389/fnbeh.2022.943759. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Schultz-Darken N., Braun K.M., Emborg M.E. Neurobehavioral development of common marmoset monkeys. Dev. Psychobiol. 2016;58:141–158. doi: 10.1002/dev.21360. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Agamaite J.A., Chang C.-J., Osmanski M.S., Wang X. A quantitative acoustic analysis of the vocal repertoire of the common marmoset (Callithrix jacchus) J. Acoust. Soc. Am. 2015;138:2906–2928. doi: 10.1121/1.4934268. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Oren G., Shapira A., Lifshitz R., Vinepinsky E., Cohen R., Fried T., Hadad G.P., Omer D. Vocal labeling of others by nonhuman primates. Science. 2024;385:996–1003. doi: 10.1126/science.adp3757. [ DOI ] [ PubMed ] [ Google Scholar ] 24. Landman R., Sharma J., Hyman J.B., Fanucci-Kiss A., Meisner O., Parmar S., Feng G., Desimone R. Close-range vocal interaction in the common marmoset (Callithrix jacchus) PLoS One. 2020;15 doi: 10.1371/journal.pone.0227392. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Yano-Nashimoto S., Truzzi A., Shinozuka K., Murayama A.Y., Kurachi T., Moriya-Ito K., Tokuno H., Miyazawa E., Esposito G., Okano H., et al. Anxious about rejection, avoidant of neglect: Infant marmosets tune their attachment based on individual caregiver’s parenting style. Commun. Biol. 2024;7:212. doi: 10.1038/s42003-024-05875-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. Gultekin Y.B., Hildebrand D.G.C., Hammerschmidt K., Hage S.R. High plasticity in marmoset monkey vocal development from infancy to adulthood. Sci. Adv. 2021;7 doi: 10.1126/sciadv.abf2938. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 27. Gultekin Y.B., Hage S.R. Limiting parental feedback disrupts vocal development in marmoset monkeys. Nat. Commun. 2017;8 doi: 10.1038/ncomms14046. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 28. Mo C., Yin J., Fung I.C.-H., Tse Z.T.H. Aggregating twitter text through generalized linear regression models for tweet popularity prediction and automatic topic classification. Eur. J. Investig. Health Psychol. Educ. 2021;11:1537–1554. doi: 10.3390/ejihpe11040109. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. Chojnicka I., Wawer A. Social language in autism spectrum disorder: A computational analysis of sentiment and linguistic abstraction. PLoS One. 2020;15 doi: 10.1371/journal.pone.0229985. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 30. Cascio C.J., Moore D., McGlone F. Social touch and human development. Dev. Cogn. Neurosci. 2019;35:5–11. doi: 10.1016/j.dcn.2018.04.009. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 31. Estes A., Zwaigenbaum L., Gu H., St. John T., Paterson S., Elison J.T., Hazlett H., Botteron K., Dager S.R., Schultz R.T., et al. Behavioral, cognitive, and adaptive development in infants with autism spectrum disorder in the first 2 years of life. J. Neurodev. Disord. 2015;7:24. doi: 10.1186/s11689-015-9117-6. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Yesilkaya M., Magallón-Neri E. Parental stress related to caring for a child with autism spectrum disorder and the benefit of mindfulness-based interventions for parental stress: a systematic review. Sage Open. 2024;14 doi: 10.1177/21582440241235033. [ DOI ] [ Google Scholar ] 33. Xie F., Pascual E., Oakley T. Functional echolalia in autism speech: Verbal formulae and repeated prior utterances as communicative and cognitive strategies. Front. Psychol. 2023;14 doi: 10.3389/fpsyg.2023.1010615. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Bosshard A.B., Burkart J.M., Merlo P., Cathcart C., Townsend S.W., Bickel B. Beyond bigrams: call sequencing in the common marmoset (Callithrix jacchus) vocal system. R. Soc. Open Sci. 2024;11 doi: 10.1098/rsos.240218. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Putnam O.C., Markfeld J.E., Wright S.T., Feldman J.I., Goldblum J., Karpinsky M., Neal A.J., Swanson M.R., Harrop C. The use of Language ENvironment Analysis in autism research: A systematic review. Autism. 2025;29:1095–1117. doi: 10.1177/13623613241290072. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Oller D.K., Niyogi P., Gray S., Richards J.A., Gilkerson J., Xu D., Yapanel U., Warren S.F. Automated vocal analysis of naturalistic recordings from children with autism, language delay, and typical development. Proc. Natl. Acad. Sci. USA. 2010;107:13354–13359. doi: 10.1073/pnas.1003882107. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 37. Warlaumont A.S., Richards J.A., Gilkerson J., Oller D.K. A Social Feedback Loop for Speech Development and Its Reduction in Autism. Psychol. Sci. 2014;25:1314–1324. doi: 10.1177/0956797614531023. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 38. Nadwodny N., Yoder P.J., Ingersoll B.R., Wainer A.L., Stone W.L., Eisenhower A., Carter A.S., RISE Research Network The Language ENvironment Analysis ( LENA ) System in Toddlers With Early Indicators of Autism: Test–Retest Reliability and Convergent Validity With Clinical Language Assessments. Autism Res. 2025;18:1568–1579. doi: 10.1002/aur.70062. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 39. Werling D.M., Geschwind D.H. Sex differences in autism spectrum disorders. Curr. Opin. Neurol. 2013;26:146–153. doi: 10.1097/WCO.0b013e32835ee548. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Kaneko T., Matsumoto J., Lu W., Zhao X., Ueno-Nigh L.R., Oishi T., Kimura K., Otsuka Y., Zheng A., Ikenaka K., et al. Deciphering social traits and pathophysiological conditions from natural behaviors in common marmosets. Curr. Biol. 2024;34:2854–2867.e5. doi: 10.1016/j.cub.2024.05.033. [ DOI ] [ PubMed ] [ Google Scholar ] 41. Mimura K., Matsumoto J., Mochihashi D., Nakamura T., Nishijo H., Higuchi M., Hirabayashi T., Minamimoto T. Unsupervised decomposition of natural monkey behavior into a sequence of motion motifs. Commun. Biol. 2024;7:1080. doi: 10.1038/s42003-024-06786-2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Phaniraj N., Wierucka K., Zürcher Y., Burkart J.M. Who is calling? Optimizing source identification from marmoset vocalizations with hierarchical machine learning classifiers. J. R. Soc. Interface. 2023;20 doi: 10.1098/rsif.2023.0399. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Phiel C.J., Zhang F., Huang E.Y., Guenther M.G., Lazar M.A., Klein P.S. Histone Deacetylase Is a Direct Target of Valproic Acid, a Potent Anticonvulsant, Mood Stabilizer, and Teratogen. J. Biol. Chem. 2001;276:36734–36741. doi: 10.1074/jbc.M101287200. [ DOI ] [ PubMed ] [ Google Scholar ] 44. Uchino S., Hirasawa T., Tabata H., Gonda Y., Waga C., Ondo Y., Nakajima K., Kohsaka S. Inhibition of N-methyl-d-aspartate receptor activity resulted in aberrant neuronal migration caused by delayed morphological development in the mouse neocortex. Neuroscience. 2010;169:609–618. doi: 10.1016/j.neuroscience.2010.05.024. [ DOI ] [ PubMed ] [ Google Scholar ] 45. Chao Z.C., Komatsu M., Matsumoto M., Iijima K., Nakagaki K., Ichinohe N. Erroneous predictive coding across brain hierarchies in a non-human primate model of autism spectrum disorder. Commun. Biol. 2024;7:851. doi: 10.1038/s42003-024-06545-3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Mimura K., Nagai Y., Inoue K.I., Matsumoto J., Hori Y., Sato C., Kimura K., Okauchi T., Hirabayashi T., Nishijo H., et al. Chemogenetic activation of nigrostriatal dopamine neurons in freely moving common marmosets. iScience. 2021;24 doi: 10.1016/j.isci.2021.103066. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 47. Wood C.M., Tizkar R.B., Fort M., Zhang X., Mulvihill K.G., Liao N., Cockcroft G.J., McIver L.B., Sawiak S.J., Roberts A.C. Dysfunction in primate dorsolateral prefrontal area 46 affects motivation and anxiety. Science. 2025;389:840–844. doi: 10.1126/science.adx4142. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Sueur J., Aubin T., Simonis C. Seewave, A free modular tool for sound analysis and synthesis. Bioacoustics. 2008;18:213–226. doi: 10.1080/09524622.2008.9753600. [ DOI ] [ Google Scholar ] 49. Drost H.-G. Philentropy: information theory and distance quantification with R. J. Open Source Softw. 2018;3:765. doi: 10.21105/joss.00765. [ DOI ] [ Google Scholar ] 50. Wickham H., Averick M., Bryan J., Chang W., McGowan L., François R., Grolemund G., Hayes A., Henry L., Hester J., et al. Welcome to the Tidyverse. J. Open Source Softw. 2019;4:1686. doi: 10.21105/joss.01686. [ DOI ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Document S1. Tables S1–S12 mmc1.pdf (110.7KB, pdf) Data Availability Statement • The data generated in this study have been deposited in the Open Science Framework Database ( http://doi.org/10.5281/zenodo.18839062 ). • Custom code generated in this study is available at GitHub repository ( http://doi.org/10.5281/zenodo.18839062 ). • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Articles from iScience are provided here courtesy of Elsevier ACTIONS View on publisher site PDF (7.0 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top