Associations between chronic placental inflammation, fetal brain development and later autism traits Authors: Iris Menu, Lanxin Ji, Bosi Chen, Tanya Bhatia, Mark Duffy, Sofia Trapaga, Suzanne M Jacques, Faisal Qureshi, Adam Eggebrecht, Muriah D Wheelock, Christopher J Trentacosta, Moriah E Thomason Published: 2026 ABSTRACT AbstractPrenatal inflammation is a recognized risk factor for autism spectrum disorder (ASD), but the neural pathways linking in utero immune exposure to later outcomes remain unclear. We examined whether chronic placental inflammation (CPI), a sustained immune response within the placenta, is associated with fetal brain functional connectivity differences and later ASD traits. A cohort of 105 pregnant individuals (85.7% African American/Black, 1.0% Asian American, 4.8% Bi-racial, 4.8% White, 1.0% Other, 2.9% Missing) underwent fetal resting-state fMRI between 24 and 38 weeks' gestation. After delivery, placental tissues were evaluated by expert pathologists for CPI lesions. Fetuses exposed to CPI (n = 44) showed significant differences in resting-state functional connectivity (RSFC) compared to non-exposed fetuses (n = 61) across ten network pairs, with the greatest effects in prefrontal, cerebellar, visual, and motor regions. These connectivity differences were associated with behavioral outcomes: cerebellar-medial visual RSFC was the strongest prenatal predictor of ASD traits at age 3. An exploratory structural equation modeling indicated that CPI was associated with reduced cerebellum–medial visual RSFC, which was associated with higher ASD trait scores, though the indirect pathway did not reach statistical significance. This study provides the first human evidence that chronic placental inflammation is associated with large-scale fetal brain network differences that are also associated with autism traits, suggesting that prenatal immune activity may be relevant to neurodevelopmental trajectories.Keywords: Fetal brain development, Chronic placental inflammation, Prenatal inflammation, Resting-state functional connectivity, Fetal programming Introduction The prenatal period is a critical window for brain development, during which complex structural and functional changes unfold. For over a decade, resting-state functional MRI (rs-fMRI) has enabled researchers to study the fetal brain by detecting blood oxygen level-dependent (BOLD) signal fluctuations that indicate brain activity. Temporal correlations of these signals across brain regions reflect resting-state functional connectivity (RSFC) (Biswal and Uddin, 2025), which emerges during fetal development with distinct regional patterns (Schöpf et al., 2012; Thomason et al., 2013) shaped by environmental factors (De Asis-Cruz et al., 2020; Thomason et al., 2019, 2021). Maternal immune activation (MIA), triggered by infection or chronic inflammatory conditions, is one important influence on brain development. In humans, MIA has been associated with increased risk for neurodevelopmental disorders in offspring, particularly autism spectrum disorder (ASD) (Jiang et al., 2016). Beyond diagnostic risk, elevated maternal cytokine levels during gestation have been linked to greater ASD symptom severity in offspring (Irwin et al., 2019; Jones et al., 2017). Mechanistically, maternal inflammatory events are thought to induce fetal neuroinflammation and disrupt critical brain maturational processes (Libbey et al., 2005; Meyer et al., 2011), leading to long-term effects on offspring health and development (Boulanger-Bertolus et al., 2018; Hantsoo et al., 2019; Nazzari and Frigerio, 2020; Rasmussen et al., 2019). Studies in non-human mammalian models establish that prenatal inflammation alters brain structure (Ramirez et al., 2020; van de Looij et al., 2012; Vasistha et al., 2020) and white matter connectivity (Galinsky et al., 2020; Saadani-Makki et al., 2009; Sharabi et al., 2019; van de Looij et al., 2012). Complementary human work has associated MIA with differences in neonatal brain structure (Graham et al., 2018; Rasmussen et al., 2019; Sanders et al., 2024) and function (Graham et al., 2018; Rudolph et al., 2018; Spann et al., 2018), and these newborn RSFC changes predict subsequent variations in infant executive functions (Graham et al., 2018; Rudolph et al., 2018), cognitive processes often impaired in neurodevelopmental disorders (Demetriou et al., 2019; Yeo et al., 2021). Nonetheless, prior studies rely on postnatal human data or non-human models, limiting direct insight into how prenatal inflammation shapes the developing human brain in utero. Examining the human fetal brain is crucial for gaining insights into the early origins and mechanisms of inflammation's impact on neurodevelopment. It captures the earliest effects during a critical period of rapid growth particularly sensitive to environmental influences, identifies early deviations that may not be apparent later, and provides a more direct measure unconfounded by postnatal factors (van den Heuvel and Thomason, 2016). When considering which regions may be most vulnerable, emerging evidence highlights the cerebellum, a structure undergoing rapid perinatal growth and central to functional network integration (Stoodley, 2016), as a particularly sensitive target of MIA (Newman et al., 2024). Cerebellar abnormalities are also among the most consistent neuroimaging findings in individuals with ASD (Vacher et al., 2021), suggesting a potential mechanistic link between early immune exposure and later autism-related outcomes. Assessing in utero immune exposures presents major methodological challenges. Rodent models, where inflammation is typically induced using agents such as lipopolysaccharide (Galinsky et al., 2020) or polycytidylic acid (Vasistha et al., 2020), are the predominant system for investigating MIA due to their practicality and genetic tractability (Careaga et al., 2017; Hall et al., 2023; Kwon et al., 2022), but don't reflect the complexity of human immune responses. In contrast, human studies often rely on peripheral maternal cytokine measurements, an indirect proxy confounded by timing, medication, and maternal factors (Goetzl et al., 2002; Shelton et al., 2015). A promising alternative to these approaches is human placental pathology, which enables quantification of acute and chronic responses while distinguishing infectious from sterile inflammation (Redline, 2006; Roberts et al., 2024). Chronic placental inflammation (CPI), characterized by persistent maternal anti-fetal immune activation, is particularly relevant for understanding the etiology of ASD and other neurodevelopmental disorders, as it provides a window into sustained intrauterine inflammatory exposure (Kim et al., 2010, 2015). CPI has been associated with adverse perinatal outcomes (Derricott et al., 2016; Kim et al., 2015; Redline, 2007) and neurodevelopmental consequences (Korzeniewski et al., 2014; Torrance et al., 2010), including impaired executive function (Menu et al., 2025) and elevated ASD risk (Mir et al., 2021). The placenta functions as a biological interface linking maternal immune processes to the fetal environment (Fig. 1). To investigate this pathway, we recruited 105 pregnant mothers during their second and third trimesters and characterized inflammation using placental histopathology. We hypothesized that CPI would be associated with RSFC variations in brain regions linked to behavioral impairment, such as the cerebellum, posterior cingulate, medial prefrontal cortex, and subcortical areas. To evaluate whether these prenatal connectivity differences are linked to later neurodevelopment, we assessed ASD traits at age 3 and tested whether altered fetal brain connectivity mediates the association between placental inflammation and subsequent ASD traits, providing insight into potential prenatal neural pathways contributing to autism. Conceptual model, based on prior literature, of prenatal inflammation pathways to neurodevelopmental risk. Maternal infection, environmental stressors, or other health challenges initiate maternal immune activation, which in turn drives a placental inflammatory response. Altered placental signaling then disrupts fetal brain development, most notably in cerebellar structure and function, leading to downstream cognitive and behavioral manifestations such as executive-function deficits and ASD-like features. These processes unfold across three temporal stages: the prenatal environment (shaped by timing of exposure), fetal programming, and child neurodevelopment, with genetic and epigenetic factors moderating vulnerability at each stage. Methods Procedure This study followed a prospective longitudinal design across three timepoints: (1) fetal resting-state fMRI acquired between 24 and 38 weeks' gestation, (2) placental pathology evaluation following delivery, and (3) behavioral assessment at age 3 years. Pregnant women were recruited from the Detroit Medical Center during routine obstetrical appointments and were eligible if they had a singleton pregnancy, no suspected fetal central nervous system abnormality, no growth restriction as determined by 20-week ultrasound, high-quality fetal MRI data, and an available placental pathology specimen. The Wayne State University Institutional Review Board approved all study procedures, and all participating mothers provided written informed consent. Participants A total of 105 mothers were included in this study. They were 18-41 years old (M = 25.69, SD = 5.18) at the time of enrollment and self-identified as predominantly Black (85.7%; 1.0% Asian, 4.8% Bi-racial, 4.8% White, 1.0% Other, 2.9% Missing). Male children comprised 61.9% (n = 65) of the sample, and most were Black (81.0%; 1.0% Asian; 12.4% bi-racial; 4.8% White; 1.0% other). Gestational age at birth ranged from 24.9 to 42.1 weeks (M = 38.2, SD = 2.9). Prenatal chronic inflammation assessment Placentas were placed in formalin immediately following delivery and transported twice daily to the Department of Pathology. All specimens were evaluated grossly and microscopically in a standardized manner, consistent over the study period, following the Amsterdam Placental Workshop Group Consensus Statement (Khong et al., 2016). Sampling included a minimum of 3 full-thickness sections from the central two-thirds of the disc, two umbilical cord sections, a membrane roll, and three basal plate slivers, with grossly identified lesions submitted separately. After delivery, two expert perinatal histologists (S.J. and F.Q.) reviewed placental specimens for presence or absence of chronic inflammatory lesions, specifically, villitis (VUE), chronic chorioamnionitis (CCA) and chronic deciduitis (CD; Kim et al., 2015) (Fig. 2). The presence of at least one of these lesions was used as the prenatal inflammation criterion for analyses. Representative examples of placental lesions isolated. (A) Chronic villitis of unknown etiology. Maternal T lymphocytes (CD8+ T cells) (long arrow) have infiltrated and destroyed the capillaries in a chorionic villus, a fetal structure in the placenta responsible for oxygen and nutrient exchange. An adjacent chorionic villus without villitis shows intact capillaries filled with red blood cells (short arrow).(B) Chronic chorioamnionitis. Clusters of maternal T lymphocytes (CD8+ T cells) (long arrow) have infiltrated the chorioamniontic connective tissue of the fetal membranes beneath the amnionic epithelium (short arrow), which lines the amniotic sac. (C) Chronic deciduitis. A cluster of maternal plasma cells (long arrows) is seen in the decidualized endometrium of the basal plate. The decidualized endometrium shows prominent cytoplasm, resulting from elevated hormones in pregnancy (short arrow). (Hematoxylin and eosin, magnification 200×). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) ASD traits measures Mothers completed the Child Behavior Checklist for Ages 1 1/2 -5 (CBCL) at the 3-year follow-up visit. The CBCL assesses emotional and behavioral problems in children, yielding syndrome scales and DSM-oriented scales. The CBCL DSM-oriented Autism Spectrum Problems scale is a 13-item subscale mapping onto DSM ASD criteria. Items assess social reciprocity deficits, communication difficulties, and restricted/repetitive behaviors. Studies show this scale effectively identifies children with ASD features requiring further evaluation (Levy et al., 2019; Myers et al., 2014; Rescorla et al., 2019), making it a valuable tool for pediatric and school-based screening. Of the 105 participants included in prenatal analyses, CBCL DSM-ASD scores at age 3 were available for 71 (67.6%); the remaining 34 participants (32.4%) were missing this outcome measure, reflecting attrition at the age-3 follow-up visit. Neuroimaging data acquisition and analysis MRI scanning Resting-state fetal functional MRI (rs-fMRI) data was acquired using a 3 T S Verio 70 cm open-bore system with a 550 g abdominal 4-channel Siemens Flex coil from fetuses between 24 and 38 weeks of gestation (mean gestational age at scan = 30.70 weeks; see also Table 1). For each participant, two runs were collected using an echo planar image sequence sensitized to T2* weighted blood oxygenation level dependent (BOLD) signal changes. Rs-fMRI sequence parameters for scans were one of the following, varying across participants: dataset i) single-echo, TR/TE = 2000/30 ms; resolution = 3.4 × 3.4 × 4 mm3. ii) multi-echo, TR = 2000ms; TE = 18, 31.07, 44.14ms (3 echoes); flip angle: 80°; voxel size: 3.5 x 3.5 × 3.5 mm3; dataset iii) multi-echo, TR = 2000ms; TE = 18, 34.06, 50.12ms (3 echoes); flip-angle: 83°; voxel-size: 3.487 x 3.487 × 3.5 mm3. For each participant a total of between 360 and 720 frames were collected in 12 to 24 min, and when possible, this sequence was repeated. Table 1Summary of data and participant characteristics.Table 1No CPI (n = 61)CPI (n = 44)F/χ2pBirth outcomes, M (SD) Gestational age at birth (weeks)38.01 (3.45)38.43 (2.05)0.510.48 Birth weight (grams)2948.17 (826.79)3016.91 (564.00)0.230.64Preterm status at birth, n (%)1.990.16 Full-term48 (78.68)40 (90.91) Preterm13 (21.31)4 (9.09)Sex, n (%)0.001.00 Female23 (37.7)17 (38.6) Male38 (62.3)27 (61.4)Maternal Ethnicity, n (%)6.860.14 Asian0 (0.0)1 (2.3) Bi-racial1 (1.6)4 (9.1) Black54 (88.5)36 (81.8) White4 (6.6)1 (2.3) Other0 (0.0)1 (2.3) Undeclared2 (3.3)1 (2.3)Maternal Education, n (%)3.190.67 No GED/High-school diploma8 (13.1)10 (22.7) GED/High-school diploma23 (37.7)17 (38.6) Some college25 (41.0)15 (34.1) 2-yr college degree1 (1.6)1 (2.3) 4-yr college degree1 (1.6)0 (0.0) Doctorate1 (1.6)0 (0.0) Undeclared2 (3.3)1 (2.3)Gross Annual Income, n (%)6.520.37 < $10,00036 (59.0)18 (40.9) $10,000 - $20,0008 (13.1)11 (25.0) $20,000 - $30,0009 (14.8)4 (9.1) $30,000 - $40,0001 (1.6)2 (4.5) $40,000 - $50,0000 (0.0)1 (2.3) $50,000 - $60,0001 (1.6)1 (2.3) $100,000 - $120,0001 (1.6)1 (2.3) Undeclared5 (8.2)6 (13.6)Imaging Characteristics, M (SD) Maternal age at scan (years)24.88 (4.55)26.81 (5.82)3.660.06 Fetal age at scan (weeks)30.86 (3.71)30.48 (3.40)0.290.59 Translational mean movement (mm)0.14 (0.10)0.12 (0.07)1.520.22 Rotational mean movement (mm)0.01 (0.01)0.01 (0.00)0.900.35 Mean framewise displacement0.52 (0.28)0.55 (0.29)0.420.52 Retained frames per subject233.74 (105.92)244.18 (101.92)0.260.61 Summary of data and participant characteristics. Fetal fMRI data preprocessing Fetal RSFC data require extensive pre-processing prior to generation of a whole brain connectivity matrix, or connectome. Fetal brains were automatically isolated from the rest of the image via a convolutional neural network-based segmentation model. Images from the multi-echo dataset were combined using a dataset specific T2* based combination approach. Following brain extraction and combination, all data was motion corrected and motion parameters were estimated. Frames where motion-correction failed or with high motion (DVARS > 132.69 and FD > 1.5 mm, two standard deviations above the mean) were censored. On average, 238.11 frames were retained after censoring (see Table 1 for more details). Data was semi-automatically denoised using ICA through FSL's MELODIC. Finally, brains were reoriented into the standard imaging plane using SPM 12 (Statistical Parametric Mapping; http://www.fil.ion.ucl.ac.uk/spm), and normalized step-wise using intermediary age-matched templates to a 2 mm isotropic resolution fetal template using ANTs and SPM 12 (Karolis et al., 2023; Serag et al., 2012) to enable comparisons across subjects in the same coordinate space. Data were then spatially smoothed at 4 mm FWHM. Single-echo data underwent the same preprocessing pipeline as the multi-echo data, except that the multi-echo combination step was omitted. Whole-brain functional connectome construction We defined 197 brain regions of interest (ROIs) using spatially constrained spectral clustering, a data-driven technique creating approximately equal-sized, contiguous regions across cortex, subcortex, and cerebellum (Craddock et al., 2012). Using the CONN toolbox (v21.a; www.nitrc.org/projects/conn), we computed BOLD signal covariance between each region pair. Realignment parameters and their first-order temporal derivatives were removed via covariate regression. White matter and CSF signals were regressed out using anatomical component correction (aCompCor) (Behzadi et al., 2007). Band-pass filtering (0.008< f < 0.09 Hz) removed low frequency drifts and high frequency artifacts. Functional connectivity values were calculated as pairwise zero-lag bivariate correlations between all 19,306 ROI pairs, then Fisher-z transformed. This yielded a 197 × 197 symmetrical adjacency matrix (functional connectome) for each participant. These matrices were used to define functional brain networks using methods previously developed by our group and applied in infant (Eggebrecht et al., 2017) and fetal (Thomason et al., 2019, 2021; Wheelock et al., 2019) RSFC studies. Using the largest available fetal dataset (N = 168), including cases without placental samples, the Infomap community detection algorithm (Rosvall and Bergstrom, 2008) was applied to generate a functional network model comprising 15 subnetworks (Fig. 3). Quality control analyses indicated no association between motion metrics and Euclidean distance among ROI pairs (Fig. S2–S3), suggesting minimal motion-related bias (Ciric et al., 2017). Data driven whole brain networks across the fetal brain. Panel A shows an Infomap-sorted mean functional connectivity matrix, derived from the correlation structure between the 197 functionally defined ROIs, obtained from 168 fetuses. Panel B presents the spatial location of each ROI, displayed on the brain's surface, color-coded based on the assigned functional network. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Statistical analyses Network level analyses Enrichment analyses were conducted to identify clustering of strong RSFC-chronic inflammation associations within and between network pairs using the Network Level Analysis (NLA) framework (Li et al., 2024). Edge-level RSFC-inflammation associations were assessed using Welch's two-sample t-tests, with an uncorrected p < 0.05 threshold used for visualization, controlling for gestational age at scan. Network-level significance was determined via permutation testing (10,000 permutations) with FWER correction; only networks significantly enriched for RSFC–inflammation associations (p < 0.05) were considered significant (Backes et al., 2014; Eggebrecht et al., 2017). All analyses and visualizations were performed in MATLAB 2020b using the NLA Toolbox (v1.0.0). Elastic Net regression model To investigate the relationship between fetal brain connectivity and ASD traits, we employed an Elastic Net regression model. Edge-level functional connectivity for edges identified as significant from whole-brain enrichment analysis served as predictors, while CBCL DSM-ASD subscale scores were considered as outcomes. Multiple imputation (predictive mean matching, mice package; Buuren and Groothuis-Oudshoorn, 2011) addressed missing data. Data were split into training (80%) and testing (20%) subsets, and this train/test procedure was repeated across 1000 iterations to obtain a stable distribution of performance metrics. Within each iteration, the model was optimized via 10-fold cross-validation to identify optimal lambda. Performance was assessed using R2, MAE, and RMSE, reported as the mean, median, and 95% CI across iterations. Feature importance was determined separately via 1000 bootstrap iterations, extracting nonzero coefficients from each model to compute selection frequencies for the most predictive brain connectivity edges. Structural equation model Structural equation modeling (SEM) investigated whether RSFC identified by Elastic Net mediated the relationship between prenatal inflammation and CBCL ASD traits. Because this candidate mediator was identified through a prior data-driven procedure rather than pre-specified, this analysis was treated as exploratory. This analysis relies on two key assumptions: that there is no unmeasured confounding of the exposure-mediator, mediator-outcome, or exposure-outcome relationships beyond the covariates included, and that CPI, though ascertained from pathology at delivery, reflects an inflammatory process already present during the fetal scan window, consistent with the chronic nature of the lesions assessed. The model included direct and indirect (RSFC-mediated) effects, with covariates for gestational age at scan and child age at assessment. An interaction term between inflammation and RSFC was not included in the primary model, as this analysis was designed to test mediation rather than moderation. The 95% CI for the indirect effect was obtained via BCa bootstrap (1000 resamples) given the known non-normal sampling distribution of indirect effects. The N used in this analysis was 71 of 105 participants; the remaining 34 were excluded due to missing CBCL DSM-ASD scores at the age-3 follow-up. Inflammation was modeled as a binary variable using lavaan (version 0.6.19; Rosseel, 2012) with diagonally weighted least squares (DWLS) estimation. Model fit was assessed via chi-square, RMSEA (good: < 0.05), and SRMR (good: < 0.05) (Schermelleh-Engel et al., 2003). For RMSEA, we additionally drew on MacCallum and colleagues (1996), who classify RMSEA between 0.08 and 0.10 as indicating acceptable fit, particularly for models with few degrees of freedom. Standardized regression coefficients, indirect effects, and explained variance (R2) were reported. Post hoc analyses were performed on R version 4.4.1 (R Core Team, 2021). Results 41.9% of fetuses were exposed to chronic inflammation in our sample First, expert pathologists (S.M.J. and F.Q.) conducted blinded assessment of placental specimens. Among our 105 samples, 44 (41.9%) presented signs of at least one chronic inflammatory lesion. Specifically, 15.2% of the total sample showed evidence of VUE, 26.7% presented CCA, and 20% presented CD lesions (Fig. S1). Two MRI comparison groups were defined: participants with chronic placental inflammatory lesions (CPI; N = 44) and CPI-naïve controls (no CPI; N = 61), with no significant differences in demographic or imaging characteristics between groups (Table 1). Differences in fetal neural network connectivity in fetuses exposed to chronic inflammation Functional connectivity enrichment analysis revealed distinct patterns of fetal brain connectivity associated with CPI across ten network pairs; see Fig. 4. Specifically, four network pairs showed reduced connectivity in the CPI group: left prefrontal-right operculum, cerebellum-cingulate, cerebellum-medial visual, and ventromedial prefrontal-right motor. Conversely, six network pairs showed increased connectivity: medial visual-ventromedial prefrontal, cerebellum-left prefrontal, medial visual-left prefrontal, left motor-medial visual, left motor-cerebellum, and left temporal-ventromedial prefrontal (Table S1). These results were unchanged when maternal age and sex were included as covariates. Prenatal chronic inflammation-exposed and prenatal chronic inflammation-naïve participant groups differences in neural connectivity between and across networks. Panel A illustrates the edge-level Welch test results. Each black dot represents a connection between two parcels that significantly differs between the inflammation and non-inflammation groups. At the network level, ten network pairs showed significant differences as depicted in the Panel B connectogram. Within the connectogram, significance of the between-network effects is reflected in line color, ranging from yellow (p < 0.001) to dark blue (p < 0.05). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Fetal cerebellar connectivity as key predictor of ASD traits Elastic Net regression identified a sparse set of fetal RSFC edges that predicted ASD traits at age 3. To address missing data and obtain a robust estimate of performance, we used multiple imputation (m = 20, predictive mean matching) and repeated the train/test procedure across 1000 total resamples (20 imputed datasets × 50 splits each), yielding a mean R2 of 0.15 (median = 0.12, 95% CI [0.00, 0.43]) and mean RMSE of 2.49, supporting a promising and consistently detectable, if modest, prenatal predictive signal for later ASD traits. Variable selection frequencies, similarly pooled across imputed datasets, identified the most consistently selected predictors as cerebellum-medial visual (95.6% selection), left prefrontal-right operculum (94.7%), cingulate-medial visual (80.6%), and cerebellum-cingulate connections (78.4%), highlighting prominent contributions of cerebellar and fronto-opercular connectivity (Fig. 5A). Additional motor- and prefrontal-visual connections were selected with moderate frequency but smaller effect sizes. Cerebellar connectivity further showed a posterior-to-anterior gradient in association strength, with stronger effects for posterior targets and weaker or positive associations for more anterior regions (Fig. 5C). Association between fetal brain networks and ASD traits markers. Panel A shows results from an ElasticNet regression predicting CBCL ASD scores from fetal RSFC, with strongest contributions from cerebellar–medial visual and prefrontal–operculum connectivity. Panel B depicts a posterior-to-anterior gradient in cerebellar RSFC, where the strength of association with ASD traits varied spatially, as reflected in the distribution of regression coefficients. Panel C presents a structural equation model testing the pathway from prenatal inflammation to ASD traits, showing that CPI was significantly associated with fetal RSFC and RSFC was in turn significantly associated with later behavioral outcomes, though the indirect effect linking CPI to behavioral outcomes through RSFC did not reach statistical significance (p = 0.09). Prenatal inflammation affects ASD traits through cerebellum-medial visual RSFC SEM tested whether fetal RSFC mediated the association between prenatal inflammation and ASD traits at age 3, focusing on cerebellum–medial visual connectivity identified by Elastic Net (Fig. 5B). The model showed acceptable fit (χ2[2] = 3.54, p = 0.17; 0.08 < RMSEA < 0.10; SRMR < 0.001). Prenatal inflammation was associated with reduced cerebellum–medial visual RSFC (standardized β = −0.25, p < 0.05), which in turn predicted higher ASD trait scores (standardized β = −0.32, p < 0.01). The indirect pathway (standardized β = 0.08, p = 0.09) and direct effect of inflammation on ASD traits did not reach statistical significance (p = 0.91). Covariates for fetal age at scan and age at assessment showed no significant effects (all ps > 0.15). Consistently, a BCa bootstrap 95% confidence interval for the indirect effect (1000 resamples) was [−0.01, 1.00]. Discussion Our findings provide intrauterine evidence in humans that chronic placental inflammation is associated with altered fetal brain network development, offering insight into how inflammation-driven processes may shape neural circuitry maturation. Given longstanding interest in inflammation as a contributor to ASD, we examined whether patterns of fetal brain connectivity predicted ASD symptoms in toddlerhood. Although inflammation has been linked to ASD in prior work, existing studies have been limited to animal models or postnatal human samples that can only infer, but not directly assess, fetal brain involvement. By directly testing fetal brain-ASD associations, this study examined specific prenatal neural pathways that may link inflammation to later neurodevelopmental outcomes. We found both reductions and increases in connectivity among inflammation-exposed fetuses, most prominently in prefrontal, cerebellar, visual, and motor regions. Prenatal inflammation is known to impair postnatal white matter integrity (Rasmussen et al., 2019; Sanders et al., 2024; Spann et al., 2024) and synaptic development (Giovanoli et al., 2016; Mirabella et al., 2021), which could underlie these altered connectivity profiles. Decreases may reflect disruptions to developing pathways, while increases could represent compensatory hyperconnectivity as the brain reroutes circuits in response to inflammatory insult. The regions showing altered connectivity play distinct but interconnected roles in development. The prefrontal cortex is closely linked with executive functions (Menon and D’Esposito, 2022), which are central to mental health (Demetriou et al., 2019; Diamond, 2013; Yeo et al., 2021) and have been previously linked to altered prefrontal activation and reduced frontoparietal network recruitment in ASD (May and Kana, 2020). Motor and visual networks, which undergo significant perinatal development (Fenn-Moltu et al., 2023; Ji et al., 2024) and are implicated in early childhood ASD (Chen et al., 2021), also appear affected by chronic inflammation. Finally, cerebellar dysfunction is evident in several developmental disorders, including ASD and ADHD, with early damage potentially causing long-term effects on movement, cognition, and affective regulation (Stoodley, 2016). Recent findings have highlighted the role of chorioamnionitis in disrupted cerebellar maturation in preterm rhesus macaques, suggesting that prenatal inflammation may contribute to cerebellar changes similar to those in ASD (Newman et al., 2024). We found that fetal RSFC between the cerebellum and medial visual networks was the strongest prenatal predictor of ASD traits at age 3, aligning with previous findings showing ASD-related alterations in cerebellum-visual-sensorimotor connectivity (Oldehinkel et al., 2019). This finding was consistent across complementary analytical approaches (canonical correlation analysis with multiple ASD measures; see Supplement). An exploratory SEM further showed that prenatal inflammation was associated with reduced cerebellum-medial visual connectivity, and that this connectivity measure was associated with higher ASD trait scores; however, the formal indirect (mediation) effect did not reach statistical significance (p = 0.09). This pattern is consistent with, but does not establish, cerebellar-medial visual connectivity as a mechanistic link between inflammation and later neurodevelopmental risk, and should be interpreted as hypothesis-generating. Importantly, the co-occurrence of these findings does not establish that CPI is responsible for the connectivity-ASD relationship: other unmeasured factors, such as genetic risk or chronic environmental stress, could independently affect both connectivity and ASD traits, with CPI as an incidental correlate. We report this mediation pathway as exploratory and hypothesis-generating rather than confirmatory for this reason. Future work using moderation analyses in larger samples will be needed to test whether this association is specific to CPI-exposed children. Moreover, child outcomes were measured using the CBCL DSM-ASD subscale, a parent-report broadband screener rather than a diagnostic instrument; no observational or diagnostic measure was administered, and no ASD diagnoses are reported. Parent-report measures may also perform differently in high-stress samples. The absence of a direct pathway between inflammation and ASD traits underscores the potential relevance of altered connectivity to neurodevelopmental outcomes, consistent with MIA models emphasizing circuit-level perturbations as key drivers of long-term outcomes (Estes and McAllister, 2016), while highlighting the need for replication before a mechanistic pathway can be confirmed. More broadly, while our sample of 105 mother–infant dyads is substantial for fetal fMRI research given the technical demands of this modality, it remains modest for the purposes of broader predictive modeling, and estimates of predictive performance should be interpreted with appropriate caution until replicated in larger cohorts. Overall, cerebellar connectivity exhibited a posterior-to-anterior gradient in its association with ASD traits: connectivity with posterior cortical networks showed strong negative associations, while connectivity with more anterior regions transitioned toward weaker associations. This pattern mirrors the ontogenetic sequence of brain maturation (Desrosiers et al., 2024), suggesting that disruptions in early-maturing circuits may cascade to influence higher-order systems later in development. Recent work shows that functional gradients in the cerebellum reflect hierarchical organization tied to both sensorimotor and cognitive capacities (Tikoo et al., 2025), supporting the idea that prenatal inflammation-induced shifts along this gradient could underlie ASD-relevant neurodevelopmental trajectories. Taken together, these findings support the cerebellum as a particularly vulnerable substrate for prenatal inflammation and provide evidence of these effects in humans. This aligns with growing evidence implicating the cerebellum in ASD risk and its sensitivity to maternal inflammation. Volumetric MRI studies have shown that reduced cerebellar size is associated with both sensorimotor abnormalities and poorer social functioning in children with ASD (Elandaloussi et al., 2023; McKinney et al., 2022). Post-mortem analyses reveal marked reduction in Purkinje cell number and size in individuals with ASD, while transcriptomic studies demonstrate high co-expression of ASD-linked genes in cerebellar circuits (Fatemi et al., 2002; Peter et al., 2016). In animal models, placental inflammation induced by intrauterine lipopolysaccharide administration resulted in increased apoptosis and blood-brain barrier disruption in fetal cerebellum in sheep (Hutton et al., 2007), while similar exposure in rabbits led to intensive microglial activation, increased pro-inflammatory cytokines, and less developed Purkinje cell layers with associated functional impairment (Zhang et al., 2019). Resting-state fMRI human studies have identified altered cerebro-cerebellar connectivity in individuals with ASD, particularly involving visual and prefrontal cortices (Frosch et al., 2022; Hanaie et al., 2018), and longitudinal studies in rhesus monkeys demonstrate that prenatal immune challenge leads to persistent alterations in these circuits, including reduced frontal volumes, cognitive deficits, and increased repetitive behaviors, paralleling features observed in ASD and schizophrenia (Bauman et al., 2014; Vlasova et al., 2021). While these converging lines of evidence underscore the cerebellum's utility in early detection and mechanistic modeling, its role in ASD may be more reflective than causal. Ziats and Rennert (2013) argue that the cerebellum's distinctive anatomy, prolonged development, and dense cellular architecture may make it especially susceptible to widespread neurodevelopmental disruptions. In this view, cerebellar abnormalities function as an “anatomical beacon” of more subtle pathologies distributed across the brain, making them informative biomarkers for detecting atypical brain development. A significant distinction of our study is the use of placental pathology to assess prenatal inflammation, rather than reliance on maternal biomarkers alone. Placental pathology allows characterization of site, type, and severity of inflammatory lesions at the maternal-fetal interface. VUE, CCA, and CD are biologically distinct processes, characterized by different cellular responses (e.g., maternal CD8+ T-cell infiltration in VUE/CCA vs. plasma cell infiltration in CD), and were combined into a single CPI category in this study. Unlike discrete immune challenges used in rodent models, these human placental lesions represent cumulative, progressive inflammatory processes over gestation. Because these lesions frequently co-occurred and each subtype (alone or in combination) comprised a small number of cases, we were not adequately powered to test whether connectivity findings differ across lesion subtypes. This composite approach may obscure subtype-specific effects, and future studies in larger cohorts should examine whether specific lesion types, or specific combinations thereof, differentially relate to fetal connectivity alterations. By differentiating these mechanisms, research can begin to disentangle how maternal, placental, and fetal immune processes interact to shape neurodevelopment and long-term outcomes, and thereby refine both our mechanistic models and potential interventions. On another note, the 15-subnetwork parcellation was derived from a sample of 168 fetal scans, which fully includes the 105 participants analyzed here; the remaining 63 scans were added to improve the stability of the parcellation. Because the parcellation is therefore not fully independent of the connectivity findings reported, we cannot rule out some degree of circularity in defining network boundaries. Datasets of sufficient size to support fully independent parcellation and analysis samples remain uncommon in fetal resting-state fMRI; as larger datasets become available, replication using independently derived parcellations will be important for confirming the robustness and specificity of these findings. Additionally, because gestational age at scan was modeled as a linear covariate, we may not have fully captured how fetal brain connectivity changes across this developmental window, which spans 24 to 38 weeks; future studies could explore whether allowing for non-linear developmental trajectories reveals additional patterns in this relationship. This study focused on mother-fetus dyads with primarily African-American/Black mothers, who often experience higher stress levels (Giurgescu et al., 2013), greater stress-induced responses (Christian et al., 2013), and increased rates of prenatal inflammation (Borders et al., 2015). Chronic stressors such as racial discrimination, unemployment, and poverty impact functional immunity (Segerstrom and Miller, 2004), and among African-American women, those reporting more racial discrimination had higher circulating cytokines during pregnancy (Giurgescu et al., 2016). Chronic placental inflammation has been associated with maternal household income (Keenan-Devlin et al., 2017) and African-American ethnicity among very preterm infants (Matoba et al., 2019). Additionally, family socioeconomic status has been found to influence the relationship between prenatal inflammation and white matter development (Sanders et al., 2024). In the present study, socioeconomic status, approximated by maternal education and income, was balanced between CPI-exposed and CPI-naïve groups and was therefore not included as a covariate in the neuroimaging analyses. This study also did not have comprehensive data on several other maternal and pregnancy variables that may independently influence both CPI and fetal neurodevelopment, including maternal infection during pregnancy, antibiotic use, or autoimmune conditions; residual confounding from these factors cannot be excluded. It is important to note that social disadvantage in this cohort is unlikely to function solely as a confound in maternal reporting. Chronic socioeconomic stress is itself an independent postnatal exposure that can directly shape child neurodevelopmental and behavioral trajectories (Rakesh et al., 2023; Tooley et al., 2021). As such, postnatal social adversity may interact with the prenatal biological factors studied here, jointly associated with both CBCL/BITSEA scores and children's actual developmental outcomes. Disentangling these prenatal and postnatal contributions will require analyses of larger cohorts with extensive postnatal environmental measures. Caution is warranted when generalizing these findings to less stress-exposed populations; however, focusing research on populations with elevated stress and inflammatory burden is essential, as these individuals may benefit most from improved mechanistic understanding and targeted intervention. Multi-site replication across cohorts that vary in psychosocial stress exposure will be needed to determine whether the cerebellar–medial visual connectivity alteration identified here represents a general biomarker of CPI or a mechanism specific to high-adversity prenatal environments. Relatedly, future studies could benefit from including direct measures like perceived discrimination, stress questionnaires, or maternal cortisol levels. More broadly, identifying and integrating the predictors of prenatal inflammation would enable us to pinpoint patients at highest risk for developing neurodevelopmental or psychiatric disorders and implement early interventions effectively. Conclusion This study provides evidence that chronic placental inflammation is associated with differences in fetal functional connectivity, which is in turn associated with autism traits at age 3. By demonstrating specific differences in neural networks implicated in psychiatric and neurodevelopmental disorders, these findings help illuminate how in utero environmental exposures can influence early brain development. Notably, inflammation-related disruptions in fetal cerebellar-visual connectivity were associated with later ASD traits, pointing to a potential pathway through which early immune processes may shape emerging behavior. While further research is necessary to fully elucidate these relationships, our results contribute to understanding the developmental origins of ASD and the role of early inflammation in shaping brain-behavior trajectories before birth. CRediT authorship contribution statement Iris Menu: Conceptualization, Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review & editing. Lanxin Ji: Data curation, Formal analysis, Investigation, Validation, Writing – review & editing. Bosi Chen: Validation, Writing – review & editing. Tanya Bhatia: Data curation, Formal analysis, Investigation, Writing – review & editing. Mark Duffy: Data curation, Writing – review & editing. Sofia Trapaga: Formal analysis, Investigation, Writing – review & editing. Suzanne M. Jacques: Formal analysis, Investigation, Writing – review & editing. Faisal Qureshi: Formal analysis, Investigation, Writing – review & editing. Adam Eggebrecht: Software, Validation, Writing – review & editing. Muriah D. Wheelock: Software, Validation, Writing – review & editing. Christopher J. Trentacosta: Data curation, Funding acquisition, Investigation, Project administration, Validation, Writing – review & editing. Moriah E. Thomason: Conceptualization, Data curation, Funding acquisition, Project administration, Supervision, Validation, Writing – original draft, Writing – review & editing. Ethics declaration Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed. This study included organ or tissue donors. This study includes human biological material and consent was obtained by donors, or their next of kin or legal representatives, for use in this study and for publication of the article. The samples used in this research were not sourced from executed prisoners or prisoners of conscience. This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place. This study was approved by the Wayne State University Institutional Review Board. (Approval No. IRB-21-08-3875) Data availability statement Data from the fetal neuroimaging cohort are available on OpenNeuro (https://openneuro.org/datasets/ds003090/versions/1.0.0) and on the NIMH NDAR Data Archive (https://nda.nih.gov/edit_collection.html?id=2434). Code availability statement Scripts for fetal fMRI data preprocessing are available on Zenodo (https://zenodo.org/records/14051966). The NLA Toolbox is available on GitHub (https://github.com/WheelockLab/NetworkLevelAnalysis/releases/tag/v1.0.0. Any additional queries can be directed to the corresponding author. Funding statement This project was supported by awards from the National Institutes of Health, MH122447, MH141129, DA055338, ES032294, R00EB029343, and K99HD113873. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements The authors thank Alexis Taylor, Ellyn Reed and the Detroit Youth Across Development lab for contributions to data collection and data management, as well as Jiaxin Tu and Ari Segel for sharing their expertise on Infomap and the NLA toolbox, respectively. Importantly, the authors also thank participant families who generously shared their time and expressed interest in helping future babies to achieve their best possible health outcomes. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.bbih.2026.101351. Supplementary data The following is the Supplementary data to this article. Multimedia component 1 References Associated Data Supplementary Materials Multimedia component 1 Data Availability Statement Data from the fetal neuroimaging cohort are available on OpenNeuro (https://openneuro.org/datasets/ds003090/versions/1.0.0) and on the NIMH NDAR Data Archive (https://nda.nih.gov/edit_collection.html?id=2434).