Perinatal C-Reactive Protein Levels Detected in Deciduous Teeth Are Associated With Autism-Related Traits and Behavior Problems in School-Aged Children Authors: Vishal Midya, Brooke G McKenna, Amy L Salisbury, Mauro Martinez, Manish Arora, Patricia A Brennan Published: 2026 ABSTRACT AbstractBackgroundPrevious studies have linked prenatal maternal inflammatory markers, including C-reactive protein (CRP), to child neurodevelopmental outcomes; however, findings are mixed. Deciduous tooth assays allow for a direct measure of CRP in the fetal milieu, and initial findings suggest that children diagnosed with autism spectrum disorder (ASD), compared with nondiagnosed control participants, have higher levels of CRP during the third trimester of gestation.MethodsThe present study attempts to replicate this finding in a nonclinical sample and test whether prenatal CRP is associated with mother-reported measures of ASD-related traits and other child behavior problems. Mother-child dyads (N = 93) were followed prospectively from pregnancy through school age. We obtained deciduous (baby) teeth and assayed them for weekly CRP levels during the perinatal period and linked these exposures to maternal reports of behavior problems on the Child Behavior Checklist.ResultsWe showed that higher ASD-related traits, including social problems and thought problems, as well as attention and aggression problems, are associated with higher CRP levels in the third trimester. Behavior problems were also associated with elevated CRP levels in the first several months after birth and decreased CRP levels later in infancy.ConclusionsOur findings replicate and extend the findings of the case-control study of ASD. These findings suggest a possible role for perinatal inflammation in children’s neurodevelopmental outcomes and a need to further probe sensitive developmental windows and potential protective factors in this association.Keywords: Attention, Autism spectrum disorder, CRP, Neurodevelopment, Prenatal exposure Autism spectrum disorder (ASD), a neurodevelopmental condition characterized by deficits in social communication and restrictive and repetitive behaviors, is currently estimated to have a prevalence rate of over 3% in the United States (1). The etiology of ASD and its related traits remains obscure, with both genetic and perinatal exposures potentially playing a significant role. The maternal immune activation (MIA) hypothesis posits that prenatal maternal inflammation engenders a biological risk pathway through which prenatal events impact the fetal brain and increase the risk for ASD as well as other neurodevelopmental disorders. Both maternal chronic inflammation and infections during pregnancy have been implicated, and gestational timing may also influence the impact of inflammation on neurodevelopment (2). Human studies examining prenatal immune-related factors and later child neurodevelopment have had several key methodological limitations (3). One is that peripheral measures of maternal immune-related biomarkers may not accurately reflect the levels of inflammatory biomarkers that directly reach the developing fetus. Another weakness of the existing literature is that studies typically rely on measures taken at only 1 or 2 time points during pregnancy, which may miss sensitive periods of risk. Even longitudinal studies that collect inflammatory biomarkers during regular research visits may fail to capture periods of acute stress, illness, or infection when inflammation is most elevated, thereby underrepresenting meaningful variation at the higher end of the inflammatory distribution. Limitations in methodology might explain the conflicting findings in the previous literature, with some studies showing that ASD and related behaviors are related to higher maternal inflammatory markers during pregnancy (4), some showing they are related to lower inflammatory markers during pregnancy (5), and some showing no associations once maternal health confounds are controlled (6). A recent study addressed limitations in previous pregnancy and neonatal studies by applying immunohistochemical methods to deciduous teeth to measure fetal and postnatal inflammation with high temporal resolution (7). Deciduous (i.e., baby) teeth begin developing in early gestation, with enamel and dentine forming incrementally in layered growth patterns analogous to tree rings that permanently record the timing of prenatal and early-life environmental exposures, enabling the retrospective assessment of exposure windows (Figure 1). The study by Dumitriu et al. (7) focused on the retrospective assessment of daily C-reactive protein (CRP) levels, a protein released by the liver in response to inflammation, to test associations with ASD in a case-control design. Across 2 samples, children who were later diagnosed with ASD had significantly increased levels of CRP during the third trimester compared with neurotypical control participants. C-reactive protein (CRP) levels in the rings of deciduous teeth can be assayed and used to quantify longitudinal CRP exposure across perinatal development. Daily CRP levels can then be examined in relation to child developmental outcomes to determine whether perinatal exposure to inflammation confers risk for neurodevelopmental or other symptom outcomes. Importantly, Dumitriu et al. (7) focused on children with clinically diagnosed ASD, and their samples excluded participants with other neurodevelopmental disorders, such as attention-deficit/hyperactivity disorder (ADHD). This sample selection strategy makes it unclear whether third-trimester fetal CRP is specifically associated with diagnosed ASD or whether it might also be relevant to subclinical ASD–related traits or other neurodevelopmental or child behavior problems (3). The current study fills this research gap by examining perinatal CRP levels assayed from deciduous teeth in association with parent-reported child behavior problems and ASD-related traits in a nonclinically referred sample. In line with the MIA hypothesis, we hypothesized that third-trimester CRP levels measured in child deciduous tooth assays would be positively associated with ASD-related traits, attention problems, and other child behavior problems. Methods and Materials Participants The present study recruited mother-child dyads from 2 existing samples that were followed from the prenatal period through early childhood at 2 sites: Women & Infants Hospital in Rhode Island and the Emory Women’s Mental Health Program in Atlanta, Georgia. Both samples originated from studies examining the differential effects of prenatal serotonin reuptake inhibitor (SRI) exposure and maternal depression on infant and child neurodevelopment. The original inclusion and exclusion criteria for the 2 samples were nearly identical. At Emory, women and infants were included if the mothers were ages 18 to 44 years. At Women & Infants Hospital, mothers were included if they were ages 18 to 40 years. At both sites, mothers were required to have no positive urine drug screens during pregnancy, no primary diagnoses of psychosis, and no life-threatening medical illnesses. Infants at both sites were excluded if they had serious cardiac, respiratory, or central nervous system malformations or if their gestational age was <32 weeks at birth. Sample characteristics and participant demographics are provided in Table 1. Table 1Sample CharacteristicsSample CharacteristicsOverall, N = 93Study Sitep ValueBrown, n = 15Emory, n = 78Maternal Age at Pregnancy, Years33.71 (4.73)30.13 (4.69)34.40 (4.45)<.01Length of Breastfeeding Duration, Weeksa50.58 (52.73)27.93 (29.95)55.05 (55.19).11Alcohol Use During Pregnancy Yes16 (17.20%)0 (0.00%)16 (20.51%).07 No77 (82.80%)15 (100.00%)62 (79.49%)Tobacco Use During Pregnancy Yes4 (4.30%)1 (6.67%)3 (3.85%).51 No89 (95.70%)14 (93.33%)75 (96.15%)Mode of Deliverya Cesarean36 (39.56%)3 (20.00%)33 (43.42%).15 Vaginal55 (60.44%)12 (80.00%)43 (56.58%)Maternal Education Completed graduate/professional school42 (45.16%)4 (26.67%)38 (48.72%).16 No graduate/professional school51 (54.84%)11 (73.33%)40 (51.28%)Child’s Age at the Time of the CBCL Administration, Years7.57 (1.97)11.33 (1.23)6.85 (1.03)<.01Maternal Prepregnancy BMI25.48 (5.61)25.05 (5.70)25.57 (5.62).62Maternal SRI Used in Pregnancy Yes45 (48.39%)6 (40.00%)39 (50.00%).58 No48 (51.61%)9 (60.00%)39 (50.00%)Maternal Depression During Pregnancy Yes37 (39.78%)5 (33.33%)32 (41.03%).77 No56 (60.22%)10 (66.67%)46 (58.97%)CBCL ASD-related traits score165.03 (20.11)160.33 (18.07)165.94 (20.46).05 Withdrawn T score52.17 (6.21)52.40 (5.14)52.13 (6.42).78 Social problems T score53.32 (7.66)50.93 (2.58)53.78 (8.23).08 Thought problems T score59.54 (9.39)57.00 (11.24)60.03 (8.99).07 Attention problems T score53.95 (6.64)52.20 (4.02)54.28 (7.00).18 Anxiety depression T score53.20 (5.70)53.53 (5.59)53.14 (5.76).92 Somatic T score54.75 (7.10)53.40 (5.53)55.01 (7.36).39 Aggression problems T score51.75 (4.34)50.47 (1.81)52.00 (4.64).13Values are presented as mean (SD) or n (%).ASD, autism spectrum disorder; BMI, body mass index; CBCL, Child Behavior Checklist; SRI, serotonin reuptake inhibitor.aMissing data on 2 samples. Sample Characteristics Values are presented as mean (SD) or n (%). ASD, autism spectrum disorder; BMI, body mass index; CBCL, Child Behavior Checklist; SRI, serotonin reuptake inhibitor. Missing data on 2 samples. Procedures Women were invited to participate in the Tooth Fairy Study at both sites, coinciding with the developmental period when children naturally shed their deciduous teeth. At Emory, women were contacted via e-mail, regular mail, or phone and asked to participate by logging into Research Electronic Data Capture (REDCap) web-based application, signing an informed consent document, and completing a 2-minute survey about the availability of up to 4 of their children’s teeth for collection. Those mothers who consented and confirmed that they had a tooth from their child that had shed naturally were sent a tooth collection kit, which included small, barcoded plastic vials (one for each tooth) along with an informational card about where in the mouth the tooth originated and the approximate date of tooth loss. Teeth that were extracted, fell out due to an injury, or had cavities were excluded. Participants were compensated $20 per tooth for their time and effort. Similar recruitment efforts were made for participants from the Women & Infants Hospital sample. For those who expressed an interest in participating, identical data collection protocols and procedures for collecting teeth were implemented through Emory. Prior to enrollment in the current study, the Emory cohort mothers had completed an online questionnaire (REDCap) follow-up that was initiated when the children were in elementary school (ages 6 years and older). During this follow-up, they reported on their children’s current behavior problems. To parallel the behavioral data already collected through Emory, mothers in the Women & Infants Hospital sample who returned their children’s teeth were also asked to complete a REDCap questionnaire about their children’s behavior problems. The Emory University Institutional Review Board approved the Tooth Fairy Study, and the Emory University and Women & Infants Institutional Review Boards approved previous phases of data collection. Measures: Teeth Assay Decalcification and Sectioning The 93 teeth included for assay were selected based on structural integrity. The detailed methods described below are adapted from Dumutriu et al. (7). Teeth were first cut along a vertical labial-lingual/buccal-lingual plane using a Bühler Iso Met low speed saw, and the cut fragments were imaged using a Keyence stereomicroscope. To prepare thin sections for immunohistochemistry, the samples were decalcified with EDTA. The process took place at a physiological pH of 7.0 to 7.4 and at temperatures ranging from 22 to 37 °C, using gentle rotary motion. Teeth took, on average, 5 weeks to fully decalcify using a 4.13% EDTA solution, refreshed every 7 days. Larger tooth fragments required longer decalcification times, up to 8 weeks. Once decalcification was complete, the teeth were cryoprotected in 30% sucrose, mounted in Optimal Tissue Cutting media, and flash-frozen on dry ice. Starting at the midline of the tooth, 20-mm tissue sections were cut on a LEICA CM3050S cryostat at 20 °C. The cut sections were placed in a 4-day additional EDTA wash to continue decalcification and improve antibody penetration. Immunohistochemistry For each tooth, two 20-mm adjacent sections were selected for staining. Sections were washed 3 times for 5 minutes in phosphate-buffered saline (PBS), followed by 5 minutes in 0.6% Triton X-100 in PBS and 20 minutes in 10% Triton X-100 in PBS. The samples were then incubated in a blocking buffer for 1 hour, containing 5% normal donkey serum to block nontarget reactive sites, 0.2% cold-water fish skin gelatin to minimize background noise, and 2% bovine serum albumin to reduce nonspecific binding in PBS. Slices were incubated with the primary antibody in blocking buffer overnight at 4 °C. One section of each tooth was stained for CRP (1:400, ab32412) and one section for dentin matrix protein 1 (DMP1) (1:1000, MABD19). After 24 hours, sections were washed 5 times for 5 minutes in 0.5% Triton X-100 in PBS. Sections were then incubated for 1 hour with Invitrogen fluorescent secondary antibodies (1:400, donkey-anti-rabbit-555, donkey-anti-mouse-555) while being protected from light. Afterward, sections were washed 3 times for 5 minutes with 5% Triton X-100 in PBS, followed by 2 additional PBS washes. Tooth sections were mounted using Pro-Long Gold mounting medium (Thermo Fisher). Antibody specificity was validated using antigen controls and omission of the primary antibody. CRP and DMP staining were performed at different time points to prevent cross-contamination of samples. Confocal Imaging Sections were imaged with a Yokogawa CSU-W1 spinning disk confocal microscope mounted on a Nikon Ti2-E stand using a 10 × 0.45 NA Plan Apochromat objective (Nikon). Sections were excited with a 561-nm laser, and emission was collected in the 579- to 630-nm band. Volumes were acquired with an image pixel size of 0.61 mm and a z-step size of 4 mm. Imaging Analysis Before decalcification, an image of each tooth was taken using a Keyence VHX-6000 digital microscope, and a neonatal line was identified, serving as a reference point for the subsequent creation of a temporal map in the decalcified dentin. This annotated image, along with the confocal microscopy data of DMP1 and CRP labeling, was registered using a custom pipeline developed in Fiji (https://github.com/lahammond/Tooth_Registration). Once registered, the DEJ (dentino-enamel junction) was used as a guide to measure pixel intensity along a line 100 mm from the DEJ using Fiji. Temporal mapping (time since birth) of fluorescent intensity along this line was calculated by measuring the distance from the incisal edge to the intersection of the neonatal line and the DEJ. To account for variability in CRP labeling, including batch effects and variability in antibody penetration related to the demineralization of tooth sections, CRP intensities were normalized to the average DMP1 intensity per tooth. Minimization of False Positive and False Negative Signals and Quality Control Our assessment used immunohistochemistry, as described above and in Dumitriu et al. (7), in which a specific CRP antibody was used to detect CRP present in dentin tissue. As a principle of immunohistochemistry, these reactions are specific to the selected molecule or target. In this way, it is a binary reaction: there is a reaction if the molecule is present in the dentin substrate, and there is no reaction if it is absent. In addition, we ran positive and negative controls following the manufacturer’s recommendation to test the reagents and ensure that we do not have false positive or false negative identification in dentin tissue. Therefore, the use of this reagent-based analysis ensured that our method truly detected CRP. However, given that we only ran 1 tooth slide per sample, we did not have enough data points to calculate a dispersion or error in the measurement. This can be improved upon with access to more tooth samples. We measure metals and CRP in different teeth from the same individual, showing similar patterns, given that each tooth type has different growth rates and calcification starts at different time points. Once the date of birth is precisely estimated using the neonatal line, we follow the standardized growth rate in the negative (prenatal) and positive (postnatal) directions. For both metals and CRP, we use a spot size of 35 μm in this calculation, with an error of around ±1 day. Child Behavior Assessment Child behavior problems were assessed using the Child Behavior Checklist (CBCL) for ages 6 to 18 years. The CBCL contains 112 items, which the mother rates as “not true,” “somewhat or sometimes true,” or “very or often true” about her child within the past 6 months. Items are then summed onto subscales reflecting different categories of behavior problems, including anxiety/depression, somatic complaints, social problems, thought problems, attention problems, and aggressive behavior. The CBCL has established reliability (test-retest reliability ≥0.90) and validity (Cronbach’s α = 0.72–0.97), and sex-based norm-referenced T scores can be calculated for each behavior problem subscale (8). In this study, T scores on the withdrawn, social problems, and thought problems subscales were summed to create a CBCL ASD-related traits score for each child. Although the CBCL was not designed to measure ASD and does not measure some of the core domains of ASD, such as social reciprocity, nonverbal communication skills, and restricted interests, this score has been shown to discriminate between children with and without ASD, particularly among high-functioning youth (9). Perinatal Covariates Perinatal variables that may be associated with the study outcomes were collected prospectively during the perinatal phases of the 2 cohort studies. Previous research suggests that demographic and health confounds exist in the relationship between maternal inflammation and child ASD (10). Therefore, multiple perinatal covariates were selected a priori and included in the study analyses. Medical and psychiatric records provided data on the method of delivery (vaginal vs. cesarean) and the use of selective SRIs (SSRIs) during pregnancy. Mothers reported their age, education, prepregnancy body mass index (BMI), and alcohol and tobacco use during pregnancy. Maternal depression during pregnancy was assessed via the Structured Clinical Interview for DSM-IV Psychiatric Diagnoses (11) and/or clinical cutoff scores on standardized depression measures (12,13). Our main analysis did not include maternal prepregnancy BMI as a covariate; however, we conducted sensitivity analyses in which we included this as a covariate in the model. Statistical Analyses To test our primary hypothesis, we used a reparametrized version of the classical distributed lag model (DLM) that incorporates a reversed temporality module (14,15). DLM is a well-established regression-based approach that without explicitly specifying the parametric structure, models the time-varying effect of an exposure on the outcome while simultaneously controlling for covariates. A reversed-DLM (rDLM) formulation interchanges the outcome and the time-varying exposure, eventually arriving at the more familiar classical mixed-model structure (16). This structure, therefore, does not necessarily allow for causal inference; instead, the associations can be interpreted as covariate-adjusted correlations. In our case, the inference was based on the difference in CRP intensity levels relative to the outcomes. Furthermore, we used the generalized additive mixed model (GAMM) with smooth functions chosen as the basis. In this analysis, we primarily used a penalized B-spline (17). To model the within-subject correlated longitudinal time points, we incorporated a random intercept using the gamm4 R package. The statistical significance of the time-varying association (in this case, CRP intensity levels) was estimated by constructing strict 99% across the function credible intervals (CrIs) (18). Given the nature of simultaneous hypothesis testing via time-varying curves, we chose a strict alpha level of 0.05% to ensure sufficient penalization and minimize false positives at a moderate sample size. A time region was denoted as a critical time window when its corresponding credible band did not cross the null line. We dichotomized the CBCL outcomes for the regression-type analysis to determine the time-varying effect of the tooth CRP intensity levels. For each outcome, we used the 80th percentile as the cutoff, dividing the values above and below it into high and low groups, respectively. All analyses were conducted in R (version 4.3.0). Results Sample characteristics are presented in Table 1. There were no site differences in birth characteristics, maternal prenatal depression or antidepressant use, or child behavior problem T scores. However, mothers in the Emory sample were more highly educated, older at the time of pregnancy, and breastfed their children longer than mothers in the Women & Infants Hospital sample. Children in the Emory sample were younger when their behavior problems were assessed. The study site was therefore controlled in all analyses. Additional covariates included maternal prenatal depression, SSRIs, alcohol and tobacco use, maternal age and education, method of delivery, length of breastfeeding, and child’s age at the time of CBCL administration. Between sites, there were no significant differences observed in CBCL outcomes. CRP data extend from the second trimester of pregnancy through the first year of infancy, with 82.5% of participants having CRP data that extend back into the second trimester, and 67% having CRP data that extend beyond 6 months of infancy. We analyzed the CRP data based on daily intensities, which allowed us to examine very specific windows of risk across our overall time range. CRP intensities were log-transformed and scaled; we observed moderate-to-high correlations among the child behavior problem measures, consistent with other studies in the literature (see Figure 2). (A) Log-transformed time-varying C-reactive protein (CRP) intensities during the prenatal and postpartum periods; CRP intensities were measured from around 3 to 4 months before birth and continued for over a year. (B) Spearman rank correlations between the Child Behavior Checklist (CBCL) behavior problems. ASD, autism spectrum disorder. Primary Analysis Covariate-adjusted GAMM-based reverse temporal models were repeated for each CBCL behavior score to test our hypothesis that third-trimester CRP levels would be positively associated with ASD-related traits, attention problems, and other child behavior problems. For ease of interpretation, the corresponding differences between the high- and low-behavior-problem groups are presented at each daily time point. Results are first presented for each group separately, followed by the differences in CRP intensity levels between groups. Patterns of CRP intensity over time differed significantly for the high and low ASD-related traits groups (Figure 2). For the ASD-related traits score, CRP was higher in the high ASD group around the third trimester and the first 3 months postpartum (from around 70 days before birth to almost 5 months after birth). After 5 months, the pattern shifted, with CRP becoming significantly lower in the high ASD-related traits group (from 152 days after birth to around 390 days after birth). A similar pattern emerged for social problems and thought problems. CRP was higher in the high ASD-related traits group in early life, but after approximately 5 months, the pattern shifted, with CRP being significantly lower. Withdrawal problems, on the other hand, demonstrated an opposing pattern (see Figure 3). (A–D) Reverse distributed lag models showing daily C-reactive protein (CRP) intensity levels in groups with high (noted in red, above 80th percentile) and low (noted in green, below 80th percentile) scores on the autism spectrum disorder (ASD)–related traits composite and its components (withdrawn, social problems, and thought problems scales). The solid line represents the mean CRP intensity in each group over time. The gray bands represent 99% credible intervals (CrIs). Models were adjusted for maternal age, maternal education, method of delivery, as well as maternal depression, selective serotonin reuptake inhibitor use, and alcohol and tobacco use in pregnancy. Timing differences between high- and low-symptom groups for the ASD-related traits score and related behaviors are shown in Figure 4. For the ASD-related traits score, a significant between-group difference was observed from 47 days before birth to 109 days after birth, with the largest difference observed around a month after birth (marginal mean [99% CrI] = 0.17 [0.10–0.23]). After 5 months, the directionality of the associations reversed, with CRP significantly lower in the high ASD-related traits group (from 152 to 389 days after birth). As described above, a similar pattern was observed for outcomes on social problems and thought problems, and the opposite pattern was observed for the withdrawn score. The statistically significant estimates of sensitive periods (as shown in Figure 4) for the components of the ASD-related traits score—withdrawal, social, and thought problems scores—are presented in Tables S1–S4. (A–D) Reverse distributed lag models (rDLMs) showing the difference in C-reactive protein (CRP) intensity between high (above 80th percentile) and low (below 80th percentile) scores on the autism spectrum disorder (ASD)–related traits composite and its components (withdrawn, social problems, and thought problems scales). The black line represents mean differences in CRP intensity between groups over time. Values above zero indicate higher CRP in the high behavior problem group, and values below zero indicate lower CRP in the low behavior problem group. The gray bands represent 99% credible intervals (CrIs) (to highlight an underlying Bayesian viewpoint). A sensitive period, where the difference between groups is statistically significant, is identified as an area of the rDLM where the gray band does not include zero. Models were adjusted for maternal age, maternal education, method of delivery, as well as maternal depression, selective serotonin reuptake inhibitor use, and alcohol and tobacco use in pregnancy. We also examined CRP intensity scores across the perinatal period in association with high and low attention, anxiety-depression, somatic, and aggression problem behavior groups. Attention and aggression showed associations with CRP intensity over time that were similar to the pattern noted for the ASD-related traits, social, and thought scores. Specifically, both the high attention and aggression problem groups had significantly higher CRP levels around and shortly after birth (Figures 5 and 6; Tables S5 and S6). In contrast, somatic and anxiety-depression scores showed a pattern of postnatal, rather than perinatal, group differences (see Figures 5 and 6). Particularly for somatic problems, a significant difference between the high and low groups was observed only after roughly 10 months from birth. (A–D) Reverse distributed lag models showing daily C-reactive protein (CRP) intensity levels in groups with high (above 80th percentile, noted in red) and low (below 80th percentile, noted in green) scores on the Child Behavior Checklist behavior problems subscales of attention, anxiety-depression, somatic, and aggression. The solid line represents the mean CRP intensity in each group over time. The gray bands represent 99% credible intervals (CrIs) (to highlight an underlying Bayesian viewpoint). Models were adjusted for maternal age, maternal education, method of delivery, as well as maternal depression, selective serotonin reuptake inhibitor use, and alcohol and tobacco use in pregnancy. (A–D) Reverse distributed lag models (rDLMs) showing the difference in C-reactive protein (CRP) intensity between high (above 80th percentile) and low (below 80th percentile) scores on the Child Behavior Checklist attention, anxiety-depression, somatic, and aggression behavior problem scales. The black line represents mean differences in CRP intensity between groups over time. Values above zero indicate higher CRP in the high behavior problem group, and values below zero indicate lower CRP in the low behavior problem group. The gray bands represent 99% credible intervals (CrIs) (to highlight an underlying Bayesian viewpoint). A critical window, where the difference between groups is statistically significant, is identified as an area of the rDLM where the gray band does not include zero. Models were adjusted for maternal age, maternal education, method of delivery, as well as maternal depression, selective serotonin reuptake inhibitor use, and alcohol and tobacco use in pregnancy. Sensitivity Analyses We conducted 3 sensitivity analyses to substantiate our results. First, we tested associations using the simple cubic spline (gamm::cr) instead of the chosen B-spline to ensure that our results were not dependent on the chosen spline-based model. Testing associations with this alternative approach resulted in no meaningful changes to the sensitive periods identified by our primary analyses. Second, to test the potential influence of outliers, we converted CRP levels into deciles and repeated all GAMM-based DLM analyses. This adjustment did not significantly alter the results (Figures S1 and S2), except for a slight change in time-varying patterns for thought problems (Figure S1). Third, the main analysis was conducted on the complete case, as there was missingness only in a few covariates in <5% of individuals. However, we repeated an additional sensitivity analysis on the whole imputed dataset (imputation conducted using the mice R package on a few covariates with about 2% missing values). We observed that the results remain unchanged. Fourth, we repeated our main analysis, including maternal prepregnancy BMI as a covariate in the model. The results are presented in Figures S3 and S4. Aside from some changes in the confidence bands, we did not observe any significant changes in the main associations. Fifth, we repeated the analysis only on the Emory samples to ensure that including the smaller site (Brown) does not bias or confound our results. The results are presented in Figures S5 and S6. We did not observe any significant confounding because of imbalance in cohort sizes. Lastly, we applied 3 different thresholds: an 80th-, a 75th-, and a 70th-percentile cutoff to CBCL outcomes to ensure our results were not explicitly dependent on the chosen cutoff. The results of the 70th-percentile cutoff are presented in Figures S7 and S8. Other than a few alterations to the confidence bands, the main results remain unaltered. In summary, our additional sensitivity analyses and adjustments show that the results did not substantially alter the identified sensitive periods for the ASD-related traits score and its subscales. Discussion Our results replicate and extend previous findings from a case-control study of ASD, showing a significant association between fetal CRP levels in the third trimester and maternal reports of ASD-related traits (7). The present study establishes a prospective association between CRP and ASD-related symptoms and provides more direct evidence that a fetus may be directly impacted by inflammation by measuring CRP in the fetal milieu. Overall, our findings support prior evidence of sensitive periods in development when CRP exposure confers risk for later ASD-related problems. We found that CRP exposure in both the late prenatal and early postnatal periods is associated with an elevated risk. High peripheral CRP levels have been consistently associated with ASD in children (19), and some research has linked maternal serum CRP in pregnancy to child ASD outcomes (4), lending support to the MIA hypothesis and suggesting that inflammation may play a role in the etiology of ASD. Our study extends the previous literature by identifying significant associations between tooth-assayed CRP levels before and after birth and child behavior problems beyond ASD-related traits, including attention problems and aggression. Our results are consistent with studies demonstrating an association between maternal prenatal CRP and children’s ADHD symptoms (20), and they add to the literature with a novel finding that fetal and infant CRP levels are also associated with children’s aggressive behaviors. Because behavior problem subscales on the CBCL are correlated with one another, our findings regarding ADHD and aggression may reflect comorbid conditions in children, including comorbid ASD symptoms. Previous studies have suggested that there are specific developmental periods when children may be particularly sensitive to CRP exposure. The temporal window of high CRP that was associated with ASD-related traits in the current study included not only the third trimester of gestation but also the first several months after birth. This pattern is an extension of that observed in the Swedish cohort by Dumitriu et al. (7). Given the window of exposure before and after birth, it is unclear whether the relevant impacts occur prenatally or postnatally or result from more chronic inflammation across this developmental period. Studies that use person-specific analyses rather than group data are needed to test these distinctions and their relationship to ASD. Our study also yielded the unexpected finding that ASD-related traits and other child behavior problems were associated with significantly lower CRP in later infancy (after 6 months of age). There are several potential explanations for this finding. One possibility is that lower CRP during this developmental window resulted from recovery from the inflammatory event that preceded it in the neonatal period. Alternatively, if there was a major stressor or infection shortly after birth, parents may have been particularly protective of their child and limited the child’s exposure later in infancy, thereby lowering their child’s rates of stress and inflammation during this time. It is also possible that this significant decrease reflects disrupted immune regulation, an explanation proposed for earlier findings that ASD was associated with significantly lower maternal CRP in the second trimester of pregnancy (5) and significantly lower cytokine levels at birth (21). Studies leveraging person-specific designs may also be helpful for parsing these effects. Moreover, CRP is not a specific indicator of a particular cytokine, infection, or illness; future studies are needed to test which specific prenatal inflammatory factors or biomarkers are associated with ASD and other child neurodevelopmental problems. Our results should be interpreted considering several methodological limitations. First, participants in our sample were initially recruited for a longitudinal study on prenatal depression and medication use, which means the children were at higher risk for psychopathology than children in community samples. We adjusted for maternal prenatal medication and depression status in our analyses, but it is possible that perinatal CRP might be associated with child behavior problems in our sample due to associated genetic or familial risk factors that our confounding variables did not capture. Our sample size was relatively small and did not allow us to explore child sex as a moderator. Previous research has suggested sex-specific associations between MIA and child behavior outcomes, which may have been masked in the total sample (22). Because we focused on a community sample of <100 children, we were unable to examine ASD or other diagnostic outcomes, which limits the clinical utility of our findings. Because deciduous teeth do not develop until the second trimester, we were unable to examine associations between CRP exposures in the first trimester and ASD-related outcomes. Finally, although we were able to replicate previous findings linking third-trimester CRP levels and ASD-related outcomes (7), our unexpected findings regarding behavioral risks associated with lower CRP levels later in infancy require replication. Conclusions Despite the study limitations, our findings, using direct measures of fetal and infant CRP exposure, provide strong support for the MIA hypothesis regarding children’s neurodevelopmental outcomes and suggest that the timing of inflammation during development is an important area for future investigation. Acknowledgments and Disclosures This study was funded by the Brain and Behavior Research Foundation (to PAB) and the National Institute of Mental Health and Health Disparities (Grant No. RC1MH088609 [to PAB]). We would like to express our gratitude to the mothers and children who made this research possible. We are also extremely appreciative of Julie Carroll and the members of the BUILD lab for assisting with data collection. MA is an officer of a Mount Sinai academic start-up company, Linus Biotechnology Inc. VM is a consultant for Linus Biotechnology. MA and VM’s conflicts of interest are managed by the Mount Sinai Health System in accordance with institutional policy. All other authors report no biomedical financial interests or potential conflicts of interest. Footnotes Supplementary material cited in this article is available online at https://doi.org/10.1016/j.bpsgos.2026.100805. Supplementary Material Figures S1–S8 and Tables S1–S6 Key Resources Table References Associated Data Supplementary Materials Figures S1–S8 and Tables S1–S6 Key Resources Table