Introduction

Autism spectrum disorder (ASD) is a major neurodevelopmental condition, and its main symptoms include persistent deficits in social communication and interaction, restricted interest, and repetitive behavioral patterns1. The prevalence rate of ASD is estimated to be around 3% in East Asia and the USA2,3, but some researchers argue that it is now on the rise4,5. While there is no definitive cure for ASD, studies have shown that early intervention in at-risk children leads to better adaptation to society. Therefore, early screening of children at risk of ASD is now considered the core strategy for ASD treatment6,7.

ASD has long been considered a genetic disorder8. However, investigations of candidate genes have achieved limited success. Thus, it is now accepted that environmental and genetic factors and the interactions between them should be considered when clarifying the etiology of ASD9. Indeed, a seminal study by Hallmeyer et al.10 based on a large cohort of dizygotic and monozygotic twins revealed a higher contribution of shared environment to ASD than previously expected. Several environmental factors have been identified as risk factors for ASD, including pesticides11, dioxins12, and traffic-related air pollutants13,14.

PCBs are organic chemicals with biphenyl structures in which hydrogen atoms are replaced by chlorides15. Among the 209 PCB congeners, some congeners, termed dioxin-like PCBs, function by binding to aryl hydrocarbon receptors, mimicking the function of the highly toxic dioxin16. The remaining PCBs, known as non-dioxin-like PCBs, do not rely on the aryl hydrocarbon receptor pathway. However, they still have effects on neural system, such as interfering with Ca2+ signaling17,18 (Pessah et al.19).

PCBs are persistent and bioaccumulate, and they are still being detected in human tissues and breast milk decades after the worldwide ban20,21. PCBs pass through the blood–brain barrier and placenta22, possibly interfering with neural development. In addition, children are particularly vulnerable to PCB toxicity23, as the immature metabolic system of children cannot efficiently detoxify or secrete PCBs. Thus, there is concern about the interference by PCBs with the immature brain during early human development.

A potential link has been proposed between PCB exposure and ASD24, and PCBs are among the most studied chemicals involved in the etiology of ASD18 (Pessah et al.19). However, empirical studies have not provided unanimous support for claims regarding the association between PCB exposure and ASD25,26,27,28. Specifically, some epidemiological studies have reported an increased risk of ASD or autistic behavior patterns due to prenatal exposure to background levels of PCBs26,27,29,30, while others have found no association28.

The main objective of this study was to investigate whether prenatal PCB exposure is associated with the later emergence of ASD-like behavioral tendencies. To achieve this, we analyzed longitudinal data collected from an urban area in Japan. We measured the concentrations of PCB congeners in umbilical cord blood as markers of prenatal PCB exposure. Based on the dataset, we assessed the association between the concentrations of PCB congeners in the umbilical cord blood and behavioral tendencies at 18 months old and 5 years old.

In a previous study conducted by our team31, we successfully predicted ASD risk at 18 months old by analyzing body movement patterns observed in neonates, a potential biomarker of neurodevelopment (for a review Prechtel et al.32; Einspieler et al.33), which might lead to the development of early screening method in the future. The second objective of the present study was to examine whether it might be possible to predict the later emergence of autistic tendencies by combining data on prenatal PCB exposure and features of spontaneous bodily movement.

Methods

Data collection was conducted as part of a longitudinal cohort study in Chiba prefecture, Japan, called “The Chiba Study of Mother and Child Health (C-MACH)”34. The C-MACH is a birth-cohort project that aims to investigate the effects of environmental chemicals on the health and development of mothers and infants34. The participants of the current study were recruited in one hospital in Chiba prefecture, Japan, that collaborated with C-MACH. The total number of eligible mother-infant pairs were 182. The number of drop-outs by 5 years old was 57. The participants were recruited with no exclusion criteria. They were asked to answer questionnaire and to provide biological samples, such as cord blood and saliva, from early pregnancy to 5 years old according to the predetermined schedule.

Mother-infant pairs were included in the analysis if data on prenatal PCB exposure and MCAT scores were available. Among the all the participants in the cohort, a set of data including umbilical cord blood, MCHAT and background data was available for 115 mother-infant pairs. Among these mother-infant pairs, SDQ and bodily movement data were available for 103 and 54 mother-infant pair, respectively.

Protocol

All experiments and analyses were performed in accordance with relevant guidelines and regulations. The protocol of this study was approved by the Biomedical Research Ethics Committee of the Graduate School of Medicine, Chiba University (ID 451: application date, 8 November 2013; ID 462: application date, 4 December 2013; ID 502: application date, 28 May 2014). Informed consent had been obtained from the participating mothers and the caregivers of all the infant/child participants. Prospective mothers were recruited from three obstetrics clinics and enrolled in the cohort before 13 gestational weeks. Mothers completed questionnaire booklets during the first and third trimesters. The background information of mother-infant pairs was obtained from their medical records or the questionnaire booklets. Umbilical cord serum samples were collected, and some infants participated in video recordings of spontaneous bodily movement 2–4 days after delivery31.

Autism behavioral tendencies were assessed at 18 months of age using the Modified Checklist for Toddlers with Autism (MCHAT35,36,37). When the infants reached 18 months old, mothers completed the Japanese version of the MCHAT35. MCHAT was designed as the primary screening tool of the likelihood of being autistic for toddlers38. There was in total of 23 items describing behavioral characteristics that are often observed in children with ASD. The caregivers answered whether their child shows behavioral characteristics described in each item by binary choice (yeas or no). Among the 23 items, ten items especially sensitive to the likelihood of being ASD are identified as “critical items”. Since the Japanese version of the MCHAT had been standardized and validated for toddlers above 18 months old, the present study analyzed only the MCHAT data collected at 18 months old.

Furthermore, we examined the association between PCB exposure and the Strength and Difficulty Questionnaire (SDQ39,40) at 5 years of age to assess the long-term effects of the prenatal chemical environment. Mothers completed the Japanese version of the SDQ40 when the child reached 5 years of age. SDQ is the behavioral screening questionnaire developed to assess multiple domains of behavioral problems in children from 2 to 17 years old39. There are in total of 25 items describing behavioral patterns of children. Caregivers were asked to rate how well each item applies to their children in 3 levels. Based on the raw score, the strength and weakness in the five domains of psychological attributes, i.e. emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship problems and prosociality, are evaluated. The total difficulty score, hereinafter referred to as SDQ total score, is calculated by summing up the scores of emotional symptoms, conduct problems, hyperactivity/inattention and peer relationship problems.

PCB measurement

Concentrations of 23 main PCB congeners, that are reported to be more frequently detected in cord blood than the other congeners41, were measured using gas chromatography42,43. Among the 23 congeners, either the measurement was unavailable or the congener was not detected in every participant for PCB60 and PCB87. Thus, further analyses were conducted for the remaining 21 congeners (PCB 28, 66, 74, 99, 105, 118, 126/178, 138, 146, 153, 156, 170, 177, 180, 183, 187, 194, 199, 201, 206, 209). To avoid inflating the false positive rate in the subsequent analysis, we used principal component analysis (PCA) to condense the PCB exposure data. The fat-based concentrations (ng/g) were used in the analysis since PCBs have lipophilic nature. N.D. (none detected) values in the dataset were substituted with zeroes, and all values were logarithmically transformed after adding values of one. The data were then standardized using means and standard deviations and submitted to PCA. It is common practice in PCA to retain PCs so that the cumulative explained variance amounts to at least 80% of the total variance44. To safeguard against loss of critical information, we retained eleven PCs, as these PCs accounted for over 85% of the variance. These PCs are referred to as PCB PCs hereinafter.

In order to examine whether dioxin-like and non-dioxin-like PCB congeners exerted differential influences on the later behavioral development, PCB PCs were synthesized for dioxin-like and non-dioxin-like PCBs separately. After raw concentrations were preprocessed in the way as described above, the PCB congeners were grouped into dioxin-like PCBs and non-dioxin-like PCBs. The dioxin-like PCBs included PCB118, PCB105, PCB126/178, and PCB15645, while remaining congeners were grouped into non-dioxin-PCBs. The standardized concentration data of the dioxin-like PCBs were entered into PCA. With the criteria of examining 85% variance of the data, three PCs were retained. These PCs are hereinafter referred to as dioxin-like PCB PCs to distinguish them from PCB PCs synthesized from the concentrations of all the congeners measured. Likewise, the concentrations of non-dioxin-like PCB congeners were submitted to PCA, and nine non-dioxin-like PCB PCs were retained.

Bodily movement analysis

Spontaneous bodily movements of infants positioned in a supine position were video-recorded 2–4 days after delivery. From these recordings, 26 bodily movement patterns were quantified using a markerless system of infant motion analysis46. Detailed descriptions of these bodily movement features can be found in Doi et al.31,47. For this study, we focused on the data of bodily movement features during the awake state, as we had recordings from a larger number of infants in this state compared to the sleeping state.

Analysis

Association between PCB exposure and MCHAT

Mother-infant pairs were initially grouped into high- and low-ASD-risk groups based on MCHAT scores. The high-risk group was defined as having any three out of 23 items or any one of the 10 critical items36. We incorporated maternal age (yrs old), paternal age (yrs old), gestational age (days), weight at delivery (g), sex of the infant, educational background of the mother and household income as the background information. The sex of the infant was coded as a binary categorical variable. There were several sparce cells in the table of educational background and household income. So, we decided to dichotomise these data and treat them as categorical variables. Specifically, for educational background, mothers who graduated universities or graduate school were coded as “high”, while the remaining as “low”. Similarly, mothers whose household income was equal to or above 6 million yen were scored as “high”, while the remaining “low”. The other background variables were treated as the continuous variables. Continuous variables were z-scored, while ordinary and categorical variables were min–max normalized.

At the first stage of the analysis, we analyzed the differences between the groups in potential confounding variables, i.e. maternal age, paternal age, gestational age, weight at delivery, sex of the infant, educational background of the mother and household income. Differences in continuous variables were tested using the Brunner-Munzel test, while differences in categorical variables were tested using the chi-squared test with Yate’s correction. We compared the factor scores of each PCB PCs between the high- and low-ASD-risk groups using the Brunner-Munzel test.

Logistic regression analyses were carried out to ascertain the association between each of PCB PCs and MCHAT when confounders were adjusted. The potential confounders were maternal age, paternal age, gestational age, weight at delivery, sex of the infant, mother’s final education, and household income.

Features to be included into the final model was selected by the procedures below. For the selection of main predictors, i.e. PCB PCs, feature selection was performed using logistic regression analysis with L1 norm regularization, with PCB PCs as predictors. The data were imbalanced; the low-risk infants outnumbered high-risk ones by 3:1. Thus, naïve random oversampling48 was done before estimating the logistic regression model. Specifically, the data of the minority class, i.e. high-risk group, was duplicated at random to equate the number of data points between classes. PCB PCs with non-zero coefficients were retained as the predictors of the final model.

The true confounding variables to be entered into the final model were selected based on the “change-in-estimate” (CIE) criteria with cut-off value of 10%49. At the first stage of this procedure, unadjusted coefficient in a logistic regression model was estimated for each PCB PC when all the PCB PCs were entered into the model as the predictors. Then, at the second stage, each confounder was entered into the model one-by-one in addition to the main predictors, i.e. PCB PCs. If the coefficient of any of the main predictors changed more than 10% when a potential confounder was entered into the model, the entered potential confounder was determined to be the true confounder to be adjusted in the final model. After all these procedures, a final model was fitted to the data by entering all the retained main predictors and confounders. The classification performance of the final model was evaluated by receiver-operator-characteristics (ROC) analysis. The association between MCHAT and dioxin-like PCB PCs and non-dioxin-like PCB PCs were modelled with the essentially same method using either dioxin-like PCB PCs or non-dioxin-like PCB PCs as the main predictors.

Association between PCB exposure and SDQ

The first analysis examined the association between SDQ scores and MCHAT scores by comparing the SDQ total score between high- and low-ASD-risk children identified based on MCHAT scores using the Brunner-Munzel test.

In the second stage of analysis, the association between PCB PCs and the standardized SDQ total score was assessed by multiple linear regression analysis. For the selection of the main predictors, i.e. PCB PCs, feature selection was performed using Lasso regression, with all the PCB PCs as predictors. PCB PCs with non-zero coefficients were retained as the predictors of the final model. The confounding variables to be included into the final model was selected by the CIE criteria in essentially the same manner as described in Sect. “Association between PCB exposure and MCHAT”. The multiple linear regression model was fitted to the data including the main predictors and the true confounders retained in the feature selection. We used the total SDQ as the target variable because all the SDQ subscales showed a strong positive correlation with the total SDQ score.

Prediction of developmental outcome based on PCB exposure and neonatal bodily movement

Twenty-six features of spontaneous bodily movement patterns were z-scored and then compressed into PCs through PCA. These resultant PCs are hereinafter referred to as “GM PCs.” Seven GM PCs were retained for further analysis, accounting for more than 85% of the variance.

To examine whether MCHAT scores could be predicted based on the data available at the neonatal stage, a logistic regression analysis was carried out. The main predictors were selected in the essentially the same manner as described in Sect. “Association between PCB exposure and MCHAT”. That is, a logistic regression model with L1 norm regularization using 11 PCB PCs and 7 GM PCs as predictors. To address the imbalance of the classes, naïve random oversampling48 was done before running the logistic regression model. The variables with non-zero coefficient was retained as the main predictors in the final model. Confounders were selected by the CIE criteria as described above.

In the performance evaluation of the final model, the classification performance of the logistic regression model was cross-validated by the leave-one-out procedure. Finally, the classification performance was evaluated based on the area under the curve (AUC).

Results

Association between PCB exposure and MCHAT score

Comparisons of PCs between high and low ASD risk group

Participants with available PCB measurements and MCHAT scores were included in the first analysis. The results of the group comparisons of the background variables are summarized in Table 1.

Table 1 Background of Infant and Mothers in each ASD risk group.
Full size table

The factor loadings of each PC are summarized in Supplementary Material. The factor score of each PC was compared between the low- and high-risk infants using the Brunner-Munzel test. The results of statistical analyses are summarized in Table 2. As can be seen in this table, the factor scores of PCB PC3, 4 and 5 were significantly higher in the high-ASD-risk group than in the low-ASD-risk group.

Table 2 Group difference in the mean PCB PC factor scores between high and low ASD risk group.
Full size table

Logistic regression analysis of the association between PCB PCs and MCHAT

Logistic regression with L1 norm regularization retained 10 PCB PCs excluding PCB PC6. After the application of CIE criteria, maternal age, paternal age, delivery weight, sex, educational background, and household income were retained as the confounders. The results of the logistic regression analysis using these retained variables as predictors are summarized in Table 3. As can be seen in this table, PCB PC3 was positively associated with the classification of the ASD risk (β = 0.607, p = 0.023). The AUC was 0.79, and the ROC curve is shown in Fig. 1a.

Table 3 Estimation result of logistic regression model predicting ASD risk based on PCB PCs and the confounders.
Full size table
Fig. 1
Fig. 1
Full size image

(a) ROC curve of the ASD risk classification by logistic regression model. The target variable was the binary classification of ASD risk, and the predictors were the factor scores of PCB PCs and the confounders. (b) Scatterplot of the standardized SDQ total score and the SDQ total predicted by multiple regression model based on the confounders and PCB PCs.

Logistic regression analysis of the association between dioxin-like PCB PCs and MCHAT

Logistic regression with L1 norm regularization yielded non-zero coefficient for all the three dioxin-like PCB PCs. Thus, these dioxin-like PCB PCs were retained in the final model. After the application of CIE criteria, maternal age, delivery weight, sex and educational background were retained as the confounders. The results of the logistic regression analysis using these retained variables as predictors are summarized in Table 4. As can be seen in this table, the dioxin-like PCB PC2 was significantly associated with the ASD risk (β = 0.519, p = 0.026).

Table 4 Estimation result of logistic regression model predicting ASD risk based on the dioxin-like PCB PCs and the confouders .
Full size table

Logistic regression analysis of the association between non-dioxin-like PCB PCs and MCHAT

Logistic regression with L1 norm regularization yielded non-zero coefficient for all the nine non-dioxin-like PCB PCs. Thus, these non-Dioxin PCB PCs were retained in the final model. After the application of CIE criteria, maternal age, delivery weight, sex, and educational background were retained as the confounders. The results of the logistic regression analysis using these retained variables as predictors are summarized in Table 5. As can be seen in this table, non-dioxin-like PCB PC3 and 4 were significantly associated with the ASD risk (β = 0.744, p = 0.004 for PC3; β = 0.501, p = 0.029 for PC4).

Table 5 Estimation result of logistic regression model predicting ASD risk based on non-dioxin-like PCB PCs and the confouders .
Full size table

Association between PCB exposure and SDQ

The Brunner-Munzel test revealed no significant difference in SDQ total score (p value = 0.335) between high and low risk group identified by MCHAT, thus providing no support for the linkage between behavioral tendencies measured by MCHAT and SDQ total score.

Lasso regression yielded non-zero coefficients for PCB PC2, 3, 9, and 11. Thus, these four PCB PCs were retained in the final model as the main predictors. After the application of CIE criteria, the maternal age, paternal age, gestational age, delivery weight, sex, educational background, and household income were identified as the confounders. The results of the multiple regression analysis were summarized in Table 6. As can be seen in this table, no PCB PC was significantly associated with SDQ total socre. Nevertheless, there was a significant positive correlation between the predicted and the actual score of SDQ total score (r = 0.416, p < 0.001). A scatter plot of the predicted and actual scores is shown in Fig. 1b).

Table 6 Estimation result of multiple regression model predicting SDQ total based on PCB PCs and the confouders.
Full size table

Prediction of ASD risk based on the data available at the neonatal stage

Logistic regression with the L1 norm regularization retained all the PCB PCs except for PCB PC10 and all the seven GM PCs except for GM PC7 as the predictors. After the application of the CIE criteria, the maternal age, paternal age, gestational age, delivery weight, sex, educational background, and the household income were identified as the confounders. The classification performance of a classifier with all these main predictors and confounders was estimated by a leave-one-out cross-validation procedure. The AUC was 0.69.

Discussion

The present study examined the relationship between prenatal exposure to PCBs and later development using a longitudinal birth cohort study. The results showed a connection between PCB levels and the risk of developing ASD, as assessed by the MCHAT score at 18 months of age. PCBs were identified as predictors of the MCHAT score after taking into account the confounders, suggesting that prenatal exposure to PCBs influences ASD risk independently. By considering the pattern of spontaneous bodily movement, an early behavioral marker of ASD risk identified by Doi et al.31, it became possible to predict ASD risk based on neonatal stage data.

Research has reported that even low levels of background PCB exposure can have negative effects on socio-cognitive development in later years50,51, which may persist into preschool age52. Some researchers have investigated the potential link between ASD and prenatal exposure to PCBs18 (Pessah et al. 2020). The present results, supported by logistic regression analyses and comparisons of PCB PC factor scores between groups, provide evidence that prenatal exposure to PCBs is associated with the emergence of behavioral tendencies often seen in ASD. The factor scores of PCB PC3,4 and 5 were higher in high ASD risk group than low ASD risk group. Among these PCB PCs, the factor score of PCB PC3 was significantly associated with ASD risk after adjusting the effect of the confounders. Having established a connection between prenatal PCB exposure and MCHAT, we sought to determine if developmental outcomes at 18 months could be predicted based on neonatal stage information. Our analysis demonstrated that infants’ ASD risk could be classified to some extent using background variables, PCB PCs, and spatiotemporal characteristics of spontaneous bodily movement31.

These findings seemingly indicate that higher level of PCB is associated with elevated risk of ASD, but the interpretation is not so straight forward. The absolute value of the factor loading on PCB PC3 was above 0.3 for PCB74, PCB126/178, PCB199 and PCB209, indicating the potentials of these PCB congeners to influence the likelihood of the emergence of ASD-like behavioral tendency. PCB126 is classified into dioxin-like PCB45. Thus, though somewhat speculative, one possible interpretation is that the prenatal exposure to PCB126 increases the risk of ASD through interference to the process of sexual differentiation mediated by AhR pathway.

The overall pattern points to the possibility that non-dioxin-like PCBs as well as dioxin-like PCBs, exert modulatory effects on the likelihood of ASD-like behavioral pattern in toddlers. PCB74, PCB199 and PCB209, are non-dioxin-like PCBs. In addition, two of the non-dioxin-like PCB PCs showed significant association with ASD risk. Recent studies show non-dioxin-like PCBs exert varying effects on such processes as Ca2+-signaling and dendric growth/abortion (for a review, Klocke and Lein17; Pessah et al.18; Pessah et al. 2020). For example, Weyman et al.53 reported that non-dioxin-like PCB95 could influence the development of the neural connectivity by promoting dendric growth. This observation is reminiscent of the atypical pattern of neural connectivity in the autistic brain54,55, though there is, as far as we know, no direct evidence linking the environmental factors to the pattern of functional or morphological connectivity in ASD. In addition, some researchers argue for the potential involvement of the dysregulated Ca2+ signaling in the pathophysiology of neurodevelopmental conditions56,57. Considering these, further study is required to elucidate the exact effects of the non-dioxin-like PCB congeners on the development of neural morphology and function in order to clarify the mechanism through which the exposure to PCBs leads to the emergence of ASD-like behavioral pattern.

Previous studies investigating the effect of prenatal PCB exposure on ASD have yielded mixed results25,26,27,28. One potential cause for the discrepancy among the existing studies is that these studies vary in terms of the types of congeners considered and the surrogate measures of PCB toxicity. In light of these findings, it is necessary to examine the effect of each PCB congener separately in relation to ASD risk. One previous study27 classified PCB congeners based on known mechanisms of toxicity and examined their association with infant cognitive development. This type of analysis holds promise for future studies aiming to clarify the association between environmental toxicants and neurodevelopmental conditions.

A multiple regression model that considered background variables and some PCB PCs revealed a linear association between the SDQ total score and the predicted value of the SDQ total score. At first sight, this finding seemingly aligns with previous studies (Vreugdenhil et al.51, but see Winneke et al.58) and supports the notion of persistent effects of prenatal PCB exposure on socio-cognitive development. However, we did not find a significant association between PCB PCs and the SDQ total score. This contrasts sharply with the significant association between ASD risk and PCB PCs. It is likely that this discrepancy arises from the fact that the MCHAT and SDQ measure different aspects of a child’s behavior. Many items on the MCHAT focus on social behavior and well-known ASD symptoms37 such as lack of pretend play and emergence of joint attention. In contrast, the SDQ total score assesses behavioral problems across a wider range of domains39. Notably, there was no direct association between MCHAT scores and SDQ total scores in the present dataset. Among the 182 eligible mother-infant pairs (182 pairs), only 56.6% (103 pairs) provided all the data necessary for them to be included in the analysis of the association between PCB and SDQ. This low rate of inclusion is due to the drop out from the cohort and the failure to answer all the questionnaire items. Thus, the dataset of SDQ analysis seems overly represented by the mother-infant pairs who had high commitment to the longitudinal cohort. Taking this into consideration, we cannot exclude the possibility that this selection bias somehow weakened any links between prenatal exposure to PCB and SDQ.

Early screening of infants at high risk for ASD is now a crucial part of intervention strategies for children with ASD, as early intervention has proven effective in improving their social adaptation6,7. Our study provides promising evidence that incorporating environmental factors into the classification model can further enhance the accuracy of early screening for high-ASD-risk infants. We did not utilize clinical diagnosis data in this study for two main reasons. First, we recruited a relatively small number of participants, which led us to expect a limited number of children with ASD diagnoses. Secondly, ASD is often misdiagnosed (for example, Blumberg et al.59), making diagnostic labels less suitable for our study’s purpose. Instead, we assessed ASD risk using the MCHAT score provided by caregivers, rationalizing this approach by recognizing that autistic tendencies span a spectrum from subclinical to clinical levels60. However, the existence of a categorical boundary between individuals with and without an ASD diagnosis is debated in some studies (for example, Abu-Akel et al.61). Additionally, we did not conduct recommended follow-up interviews with caregivers36,37. Therefore, future research using longitudinal data from a larger sample of infants is needed to empirically evaluate if our approach effectively predicts ASD diagnosis.

Yet another weakness of the present study concerns generalizability of the findings. All the participants were recruited in a restricted rural region in Japan, and the number of participants who provided the data necessary for the current analysis was limited. Thus, it remains to be seen whether any effects observed in the current analysis could be replicated in the other cohorts.

In conclusion, our study reveals the long-lasting effects of prenatal PCB exposure on socio-cognitive development. Specifically, we found a link between levels of prenatal PCB exposure and ASD risk as assessed by MCHAT at 18 months. These findings offer potential for early screening of children at risk for neurodevelopmental conditions. It is important to note that the effects of PCBs on later behavioral tendencies are not uniform, challenging the simplistic notion that prenatal PCB exposure only has detrimental effects on neural development due to neurotoxicity. Instead, it appears that each PCB congener contributes differently to socio-cognitive development through various pathways, such as interference with sexual differentiation or the formation of neuronal morphology and connectivity18 (Pessah et al. 2020). Importantly, non-dioxin-like PCB as well as dioxin-like PCB seemingly makes significant contribution to the modulation of the neural development that may lead to the emergence of ASD-like behavioral pattern in the toddlerhood. However, due to the limited data available, we can only provide a phenomenological understanding at this point regarding the precise mechanisms through which each PCB congener influences neural development. To achieve a comprehensive understanding of how environmental chemicals affect human neural development, synthesis of findings from neurotoxicity research on model animals and epidemiological studies on early human development is necessary.