Dynamic neuro-immune regulation of psychiatric risk loci in human neurons

Abstract

The immune environment influences neurodevelopment and subsequent clinical trajectories for psychiatric outcomes in childhood and adolescence. Yet it remains unclear if the impact of maternal and fetal immune activation varies with distinct polygenic risk profiles. Therefore, here we catalog genotype and environment (GxE) interactions, contrasting allele-specific regulatory activity between inflammatory cues. We report a cue-specific neuronal massively parallel reporter assay (MPRA) of 152 loci from genome-wide association studies (GWAS) of ten brain traits/disorders, empirically dissecting the impact of interleukin-6 (IL-6) and interferon-alpha (IFNα) on transcriptional activity. In human induced pluripotent stem cell (hiPSC)-derived glutamatergic neurons, 1,156 active candidate regulatory risk sequences (MPRA-active CRSs) are resolved, including 267 with variant-specific effects (MPRA-emVars) and 61 with variant-by-cytokine interaction effects (interaction MPRA-emVars). Broadly, neuronal immune-mediated regulatory activity is associated with differences in transcription factor binding and chromatin accessibility, the gene targets of which show pleiotropic enrichments for brain, metabolic, and immune disorders. Dynamic genetic regulation mediates neuroimmune effects, informing our understanding of the genomics of psychiatric and neurological traits, mechanisms governing pleiotropy across disorders, and how immune mechanisms mediate genetic risk.

Stress and inflammation influence the developing brain and are associated with mental health outcomes. Here the authors show that common genetic risk dynamically interacts with immune responses in human neurons.

Introduction

Genome-wide association studies (GWAS) link hundreds of significant loci with risk for psychiatric traits^1–6 and neurodegenerative disease^7,8; the overwhelming majority are non-coding variants, common in the population-at-large, and thought to regulate the expression of one or more target genes⁹. Yet, pinpointing specific causal variants is complex, with true signal obscured by linkage disequilibrium patterns, and polygenic risk scores still incapable of reliably predicting individual outcomes¹⁰. Interactions between risk variants with each other^11,12 and the environment¹³ may underlie the variable penetrance and expressivity of genetic risk for complex brain disorders.

Neurodevelopment represents a key window during which environmental exposures interact with genetic risk¹⁴. Maternal immunometabolic stressors (e.g., infection¹⁵, trauma¹⁶, immune dysfunction¹⁷, obesity¹⁸, and diabetes¹⁹) are associated with subsequent neuropsychiatric disorder risk in childhood and adolescence. High levels of maternal cytokines (e.g., interleukin-6 (IL-6)^20,21) cause immune dysregulation in rodent offspring^21–23, concomitant with changes in neurodevelopment, gene expression, circuit function, and behavior^24–27. IL-6 likewise impacts gene expression, neurogenesis, and neuronal activity in cultured human neural cells^28–30. Likewise, fetal cytokines (e.g., type 1 interferons, including interferon-alpha (IFNα)) mediate placental response to infectious agents and can influence neurological and neurodevelopmental trajectories in humans³¹; for example, alterations in brain development are often present in patients with genetic interferonopathies³². Yet, what remains unclear is the extent to which cytokine effects vary between individuals with distinct polygenic risk profiles.

Traditional genomic studies assume genetic risk to be static, and so map risk variants without consideration of how regulation of gene expression can change, but examples of cell-type-^33–35, sex-^36–38, and developmental stage-^39–42 specific, stress-^43,44, inflammation-^45,46, body mass index-⁴⁷, and drug-^48–50 dependent genetic regulation of gene expression abound. We hypothesize that the regulatory activity of non-coding risk variants in human neurons is influenced by inflammatory cues, and that resulting cue-specific effects may alter susceptibility for complex brain disorders and diverse neurotypes. Which genetic variants show dynamic immune-responsive regulatory activity in neurons is unknown.

By coupling massively parallel reporter assays (MPRAs)^51,52 with human induced pluripotent stem cell (hiPSC) models^53–55, transcriptional activity can be empirically evaluated at scale in live human neurons. Here, we test the hypothesis that immune signaling (specifically IL-6 and IFNα) interacts with non-coding regulatory elements by characterizing the dynamic transcriptional activity of 3668 candidate regulatory sequences (CRSs) prioritized from 152 GWAS loci linked to ten brain traits/disorders. Altogether, we modeled dynamic immune contributions to neurodevelopment that precede symptom onset and disorder etiology by decades.

Results

Dynamic neuronal changes in response to immune signaling

Given that hiPSC-derived human Neurogenin-2 (NGN2)-induced glutamatergic neurons (iGLUTs)^56,57 most resemble their fetal counterparts⁵⁸, the influence of inflammatory cues during neurodevelopment was assessed using iGLUTs from two neurotypical donors (one female, one male) that were acutely (24- and 48 h) treated with IL-6 (25 ng/ml, 60 ng/mL), IFNα−2b (100 IU/mL, 500 IU/mL), or vehicle (0.1% FBS in ultrapure H₂O) before harvesting at 24 days in vitro (DIV) (experimental schematic: supplementary information (SI) Fig. 1). Treatment dose was informed by previous studies in human neural progenitor cells (NPCs) and neurons: IL-6^28,30, IFNα⁵⁹.

Receptors for all stressors (IL-6: IL6R, IL6ST; IFNα: IFNΑR1, IFNAR2) were expressed in iGLUTs (SI Fig. 2). In classical IL-6 signaling, IL-6 binds to membrane-bound IL6R, which then associates with glycoprotein 130 (GP130, encoded by IL6ST); comparatively, trans-signaling via hyper-IL-6 delivers IL-6 covalently bound to soluble IL6R. Notably, no qualitative differences between classical and trans IL-6 signaling pathways have been reported⁶⁰. Whereas some reports indicate that hiPSC-derived NPCs express low IL6R, do not respond to IL-6, and require hyper-IL-6 treatment^28,61, others find that IL6-R mRNA and protein is expressed in hiPSC-derived neurons⁶², increase with maturation⁶², and that IL-6 treatment impacts expression of genes regulating extracellular matrix, actin cytoskeleton and TGF-beta signaling⁶³. Here, we likewise demonstrate that DIV23 iGLUTs express IL6R (SI Fig. 2A–C), albeit at relatively low levels (0.5-1 normalized TPM), and demonstrate neuronal transcriptomic, epigenomic, and cellular response to classical IL-6 signaling as follows.

In glia, different immune stressors induce distinct cellular states^64,65; our findings indicate that this may also occur in neurons (SI Figs. 3–6). There was limited overlap between either significant (BH-FDR-corrected p-value p_FDR < 0.05) or nominally significant (unadjusted p_nom < 0.05) differentially expressed genes (DEGs) by cue, with IL-6 treatment more moderately impacting the transcriptome (12 down-regulated and 1 up-regulated DEG, p_FDR < 0.05: 1434 DEGs, p_nom < 0.05) relative to IFNα treatment (39 down- and 97 up-regulated DEGs, p_FDR < 0.05; 2000 DEGs, p_nom < 0.05) (SI Fig. 3A, B and SI Data 1.1–1.4). Individual cytokine significant effects at 24 and 48 h were significantly correlated (IL-6: r = 0.35, p = 0.003; IFNα r = 0.72, p = 5.7 × 10⁻²³⁰) (SI Fig. 3A–C), albeit with differences in the magnitude of effects over time. Nominal IL-6 DEGs were enriched for processes largely related to cell adhesion, whereas IFNα response genes were enriched for antigen binding, protein ubiquitination, and, as expected, interferon (-log(p) = 9.02; z-score=2.287) and neuroinflammatory signaling (-log(p) = 8.24) (SI Fig. 3D, E and SI Data 1.5).

Although neuronal immune responses were distinct between exposures, convergent mechanisms and shared enrichments between IL-6 and IFNα responses were also resolved. Both treatments yielded DEGs (p_nom< 0.05) enriched for pathways related to stress and immune response (e.g., mTOR signaling [IL-6(-log(p) = 3.7; IFNα (-log(p) = 6.25] and EIF2 signaling [IL-6(-log(p) = 10.2; IFNα (-log(p) = 24.7)]) (SI Data 1.5) and well-recapitulated fetal mouse brain signatures associated with four models of immune activation (poly(I | C)⁶⁶ (SI Data 1.6). These similarities likely reflect a set of convergent genes (258 down-regulated and 352 up-regulated) with perturbations in the same direction across neuroinflammatory contexts (meta-analysis p_FDR ≤ 0.05) (SI Data 1.7).

Chromatin accessibility changes were greatest with IL-6 (675 differentially active regions (DARs); p_FDR ≤ 0.05) and more modest with IFNα (12 DARs; p_FDR ≤ 0.05) (SI Figs. 4, 5; SI Data 1.9-1.11), unlike transcriptomic changes, which were greatest with IFNα. Yet, of those DEGs that overlapped with chromatin DARs, cue-responsive gene expression (average expression) and chromatin accessibility (ATAC peak score) were significantly, albeit weakly, positively correlated (IL-6 exposure: Pearson’s Correlation Coefficient r = 0.17, p = 0.001; IFNα exposure: r = 0.13, p = 0.0004, SI Fig. 5A). Comparative enrichment analysis revealed robust IL-6-specific enrichments for calcium-dependent signaling and O-linked glycosylation, IFNα-specific enrichments for interferon signaling and synaptic transmission, and shared enrichments for actin/cadherin binding (SI Fig. 5B, C).

Phenotypically, neither cytokine altered cell number or synaptic puncta density and IFNα significantly increased neurite outgrowth (p_bon< 0.001) in immature neurons (SI Fig. 6). Altogether, multimodal evidence indicated that acute exposure to IL-6 and IFNα in mature iGLUTs resulted in distinct neuronal responses.

Dynamic transcriptional regulation of allele-specific activity at GWAS loci in neurons

To test the extent that neuronal immune response altered genetic regulation of expression by disease-relevant loci, we designed a cross-disorder Lenti-MPRA⁵² library integrating GWAS summary statistics from ten brain disorders, neurotypes, and traits (Alzheimer’s disease (AD)⁶⁷, attention deficit hyper-activity disorder (ADHD)⁶⁸, anorexia nervosa (AN)⁶⁹, autism spectrum disorder (ASD)³, bipolar disorder (BIP)⁷⁰, major depressive disorder (MDD)⁷¹, obsessive compulsive disorder (OCD)⁷², post-traumatic stress disorder (PTSD)⁷³, schizophrenia (SCZ)¹, and neuroticism (NEU-P)⁷⁴) with post-mortem brain eQTLs⁷⁵ (coloc2⁷⁶ and S-PrediXcan⁷⁷) (SI Fig. 7; SI Data 2.1–2.7). We prioritized 4430 GWAS single nucleotide polymorphisms (SNPs), with the number selected per trait dependent on the number of significant loci per GWAS (SI Fig. 7C). The following benchmark variants were included: 310 empirically active MPRA-candidate regulatory sequence (CRS) controls⁷⁸ (164 positive (active with variant-specific) and 146 negative (active without variant-specific effects)), and 88 coloc2-controls (lack of colocalization or brain QTL regulation (pph3 > 0.9)). In total, we included 8,960 sequences (8860 SNP-centered elements, 100 scramble controls to measure basal activity of the minimal promoter), representing 4430 biallelic pairs (SI Data 2.8, 2.9).

The library was transduced into mature iGLUTs (21 DIV); 24 hours later, neurons were acutely exposed (48 hours) to IL-6 (60 ng/mL), IFNα (500 IU/mL) or vehicle (vehicle: 0.1% FBS in ultrapure H₂O) before harvest (24 DIV) (two control donors, two biological replicates each, experimental schematic: SI Fig. 1). Technical replicates were highly correlated based on DNA and RNA counts (Pearson’s correlation coefficient, R = 0.98-1.00), log2 (RNA/DNA) ratios across replicates (R = 0.61-0.93, mean R = 0.83), and mean log2 (RNA/DNA) between donors (R = 0.87-0.95) (Fig. 1A–C, SI Fig. 8 and SI Data 2.11). Following filtering, 3440 (baseline), 3322 (IL-6), and 3366 (IFNα) CRSs and 44-48 scramble sequences were resolved (3071 shared sequences, 843 bi-allelic variants, 45 shared scrambles, mean n barcode/sequence = 45) (SI Fig. 8D and SI Data 2.11).

Fig. 1. Dynamic immune-responsive neuronal MPRA-emVars, related to SI Figs. 8–13, SI Data 2.1–2.16.

Mature iGLUTs transduced with cross-disorder/trait lenti-MPRA at baseline (A) (vehicle) or when treated with (B) 60 ng/mL interleukin 6 (IL-6) or (C) 500 IU/mL interferon-alpha (IFNα-2B), for 48hrs. (i) MPRA activity was strongly correlated (Pearson’s Correlation Coefficient) across biological (donor, d1-d2) and technical replicates (sequencing batch, rep1-rep 2) within each condition. (ii) Mean transcriptional activity across technical replicates was strongly correlated between donors. D Active MPRA-candidate regulatory sequences (CRSs) were identified as those with a transcriptional rate significantly (mpralm CRS quantification: p_FDR < 0.1) greater than the average activity (representing the baseline rate of the minimal promoter). 26–31% of MPRA sequences were transcriptionally active (baseline, n = 1,057 out of 3440, 31%; IL-6, n = 1092 of 3,322, 33%; IFNα, n = 764 of 3366, 22.7%). (i-iii) Number of active CRSs by condition. Proportion of active/inactive MPRA-CRS (dark orange/tan) is greater than for scramble (teal/pale yellow) and coloc2-controls (pph3 > 0.9 salmon/light green). E The mean proportion of repressed (i) and active (ii) sequences significantly differed by prioritization method. (repressed two-sided ANOVA p = 0.004, active two-sided ANOVA p = 0.000681). Proportion of active scramble sequences was significantly less than experimental benchmark CRS⁷⁸; post-hoc paired two-sided Student’s T tests: scramble-v-positive p = 0.013) or variants prioritized by colocalization (PPH4 > = 0.5) (n = 3; scramble-v-coloc2 p = 0.0008) and S-PrediXcan (n = 3; scramble-v-prediXcan p = 0.006). The proportion of active MPRA-CRSs was significantly greater for eQTL-colocalized GWAS variants (PPH4 > = 0.5) than non-colocalized controls (pph3 > 0.9) (n = 3; post-hoc paired Student’s T test: p = 0.02). F The proportion of experimental positive benchmark variants⁷⁸ identified as MPRA-expression modulating variants (emVars) was significantly greater than for negative benchmark variants⁷⁸ (paired Student’s T test, p-value = 0.00088). For boxplots, hinges represent the 25th and 75th percentile, whisker extends to the maximum or minimum value no further than 1.5 * IQR (inter-quartile range), central line indicates the median; independent MPRA experiments n = 3. Two-way ANOVA followed by paired two-sided T tests, testing assumptions of normalcy (Shapiro-Wilk’s Test) and equal variance (Levene’s Test). G Across contexts, 15–20.5% of tested variants were MPRA-emVars, with the greatest number of active MPRA-emVars at baseline (mpralm variant-dependent differential analysis: teal p_FDR < =0.1 threshold; terracotta; p_FDR < =0.05 threshold), SI Fig. 13C). Source data are provided as a Source Data file and on synapse (syn75962204).

Across all experiments, scramble sequences were significantly less active than experimental positive benchmark CRSs⁷⁸ (Student’s t test; scramble-v-positive: baseline p = 0.00095, IL-6 p = 0.0028, IFNα p = 0.00044) or prioritized CRSs (Student’s t test; scramble-v-prioritized CRSs: baseline p = 0.0012, IL-6 p = 0.0026, IFNα p = 0.00092) (SI Fig. 9B). Analysis methods were compared across three standard pipelines: MPRAnalyze⁷⁹, mpralm⁸⁰, and DEseq2^81,82: Transcriptional activity, variant-specific differential activity, and cue-by-variant differentially activity was highly correlated between mpralm and DEseq2, but DEseq2 results were affected by p-value inflation (SI Fig. 10, 11).

MPRA-CRSs were defined as MPRA-validated cis-regulatory sequences and measured as CRS with either significantly (mpralm p_FDR ≤ 0.1) higher transcriptional activity (active) or lower transcriptional activity (repressed) compared to the mean transcriptional activity. 26–41% of MPRA sequences were transcriptionally active (baseline, n = 1057 out of 3440, 31%; IL-6, n = 1092 of 3322, 33%; IFNα, n = 764 of 3366, 22.7%, Fig. 1D); in total, 1156 CRS were active in at least one context, and 660 were active across all contexts (SI Fig. 12).

The mean proportion of repressed (Fig. 1E, i) and active (Fig. 1E, ii) sequences was significantly different between scramble sequences, positive reference sequences⁷⁸, and prioritized CRSs (active ANOVA p = 0.000681, post-hoc pairwise Student’s T test, n = 3: scramble-v-positive p = 0.013, scramble-v-coloc2 controls p = 0.02, scramble-v-coloc2 p = 0.0008, scramble-v-PrediXcan p = 0.006) (SI Fig. 12B). Notably, relative to coloc2-controls (pph3 > 0.9), top eQTL-colocalized GWAS sequences (PPH4 > = 0.5) had a significantly greater proportion of active MPRA-CRSs (n = 3; post-hoc paired Student’s T test: coloc2-controls (pph3 > 0.9)-v-eQTL-colocalized GWAS p = 0.02) (Fig. 1E, ii). Moreover, CRSs prioritized by colocalization (coloc2 PPH4 ≥ 0.5) were more likely to be active compared to CRSs prioritized by transcriptomic imputation (coloc2-v-PrediXcan p = 0.078) (Fig. 1E, ii).

MPRA-emVars were defined as “MPRA-active CRSs with single-nucleotide (variant) changes that significantly altered transcriptional activity” and are calculated as variants with a significant difference in transcriptional activity between the reference allele and the alternative allele (mpralm variant-specific, p_FDR ≤ 0.1), restricted to those with positive CRS activity (mpralm quantification p_FDR ≤ 0.1, logFC > 0). The proportion of positive benchmark variants (selected based on empirical differential variant activity⁷⁸) that were MPRA-emVars was greater than the proportion of negative benchmark variants⁷⁸ (n = 3, paired Student’s T test, p = 0.00088, Fig. 1F). Of the 859 (base), 860 (IL-6), and 719 (IFNα) transcriptionally active biallelic MPRA sequences (SI Fig. 13A, B), 15–20% (20.5% baseline, n = 176; 18.5% IL-6, n = 159; 15% IFNα, n = 108) showed variant specific effects (BH-FDR-corrected p-value p_FDR < 0.1) (Fig. 1G and SI Data 2.12–14). Allelic shifts were significantly correlated across cues (no significance threshold: r = 0.85-0.88, p < 2.2 × 10¹⁶, p_FDR < 0.1: r = 0.96-0.98, p < 2.2 × 10¹⁶; SI Fig. 13C) and replicate those reported in human neural progenitor cells⁵⁴ (MPRA-emVar p_FDR < 0.1 threshold at baseline: r = 0.9, p = 0.0055, SI Data 2.17), although there are too few overlapping CRSs (n = 76) for this to represent robust validation (SI Data 2.18).

Concordance of iGLUT MPRA-emVars with eQTL effects in the adult and fetal brain

Effect sizes of concordant MPRA-emVars were significantly correlated with eQTL betas (fetal cortex Pearson’s R = 0.66-0.69; excitatory neurons R = 0.56-0.83; adult pre-frontal cortex R = 0.58-0.6, SI Fig. 14A). Of course, MPRAs are synthetic measures of transcriptional activity outside the endogenous context, whereas eQTLs measure variant regulatory activity in the native genome, where one SNP can regulate multiple genes (eGenes) with different magnitudes and directions of effect. In particular, the human fetal brain has more SNPs that map to multiple eGenes with opposing directions of regulation (SI Fig. 14B–D). To represent this nuance, we considered overlap of MPRA-emVars based on the absolute number of regulatory SNPs (Fig. 2A) and based on the eQTL effects (SNP-eGene associations) (Fig. 2B). At baseline, 22, 29, and 22% percent of MPRA-emVars overlapped with significant (Bonferroni or FDR < = 0.05) eQTLs in the adult cortex⁷⁵, fetal brain⁸³, or adult cortical excitatory neuronal⁸⁴, the majority of which were concordant in direction with at least one SNP-eGene pair in adult cortex (69%), fetal brain⁸³ (71%), and excitatory neurons⁸⁴ (55%, Fig. 2A). Concordance when considering all SNP-eGene pairs was highest in excitatory neurons⁸⁴ (68.8%, Fig. 2B), regardless of significance threshold (n = 3, paired Student’s t test; all variants p = 0.062, MPRA-emVars p = 0.084). Whereas strong concordance with adult cortical eQTLs was only observed when sub-setting for significant MPRA-emVars (n = 3, paired Student’s t test p = 0.035, Fig. 2B, i-iii). Overall, MPRA effects were concordant with eQTL data, except between response to IFNa and fetal PFC eQTLs (Fig. 2B, iv-vi). MPRA-emVars that overlapped with eQTLs (SNP-eGene pairs) identified by S-PrediXcan and coloc2 were highlighted (Fig. 2C).

Fig. 2. Neuronal MPRA-emVars largely validate brain eQTLs across inflammatory exposures, related to SI Fig. 15.

A 22, 29, and 22% percent of MPRA-expression modulating variants (emVars) at baseline overlapped with significant (i) adult brain⁷⁵, (ii) fetal brain (12-19 post-conception week or PCW⁸³), and (iii) adult cortical excitatory neuronal⁸⁴ regulatory SNPs, respectively (Bonferroni or FDR ≤ 0.05). Of these, 69, 71, and 55% had concordant directions of effect. B The proportion of MPRA-emVars across conditions with concordant directions of effect with (i) adult dorsolateral prefrontal cortex (DLPFC), (ii) fetal brain, and (iii) excitatory eQTLs (SNP-eGene pairs). For boxplots, the lower and upper hinges represent the 25th and 75th percentiles, while the lower and upper whiskers that extend from the hinge to the maximum or minimum value no further than 1.5 * IQR (inter-quartile range), and the central line indicates the median. Individual samples (unique MPRA experiments n = 3) are represented by individual points. Average percent concordance is annotated within the boxes, differences in the proportions was evaluated between pairs using two-sided Student’s T Tests after testing for assumptions of normalcy (Shapiro-Wilk’s Test) and equal variance (Levene’s Test). (Biv-vi) The absolute number of MPRA-emVars that mapped to SNPs that regulated multiple eGenes with concordant (same direction between MPRA logFC and eQTL beta), discordant (opposing direction between MPRA logFC and eQTL beta), or both effects (up and down regulation dependent upon the eGene) in (iv) adult DLPFC, (v) fetal brain, and (vi) excitatory neurons. C Concordance of MPRA-emVars with their eQTL effects after filtering for S-PrediXcan and coloc2 prioritized eGenes at (i) baseline or (ii) after interleukin-6 (IL-6) or (iii) interferon-alpha (IFNα) exposure. Bars show the frequency of MPRA-emVars mapping to an eGene across tissues (purple = adult DLPFC, teal = adult Excitatory neuron, orange = Fetal PFC). The x-axis represents the total number of MPRA-emVars with the same direction of effect (positive values) or opposite direction (negative values). Created in BioRender. Retallick-Townsley, K. (2026) https://BioRender.com/lv1j0bi. Source data are provided as a Source Data file and on synapse (syn75962204).

Neuronal immune activation results in significant cue-by-variant interaction effects on transcriptional activity enriched for psychiatric traits and common co-morbidities

152 brain trait GWAS loci were represented by 843 biallelic variants measured in baseline, IL-6, and IFNα conditions; 52 GWAS loci had variant effects across all three conditions, and 22 GWAS loci were specific to one immune condition (14 IL-6, 6 IFNα, Fig. 3A, B). MPRA-emVars overlapped with significant variants in multi-trait or pleiotropic GWAS and highlight differential regulation of genetic risk by neuronal immune activation (e.g., IL-6 MPRA-emVars were annotated for SCZ-BIP pleiotropy and IFNα MPRA-emVars were annotated for Parkinson’s disease, Fig. 3C, SI Fig. 15 and SI Data 2.19).

Fig. 3. Neuronal inflammation alters the effect of psychiatric risk variants on transcriptional activity, related to SI Figs. 14–16, SI Data 2.12–2.16,2.19.

A 98 of 152 GWAS loci had ≥1 MPRA-expression modulating variant (emVar) at baseline; 14 more loci had emVar with interleukin-6 (IL-6) and 6 with interferon-alpha (IFNα). B 267 active MPRA-candidate regulatory sequences (CRSs) had variant-specific effect QTLs in ≥ 1 condition. Manhattan plot of MPRA-emVar summary statistics: colored by condition and polarized (direction of effect of the ref-alt allele) -log10(p_FDR). C Size of the points represents the proportion of MPRA-emVar to non-MPRA-emVars annotated for each GWAS trait. D Of active MPRA CRSs, 64 (IL-6, n = 61; IFNα, n = 29) had a significant cue-variant interaction. E (i) Representative stable MPRA-emVar: rs246002. (ii) Representative dynamic MPRA-emVar: rs3024744, transcriptionally repressed following IFNα exposure but not IL-6. (iii) Representative specific MPRA-emVar: rs12325245, transcriptionally repressed at baseline only. For boxplots, lower and upper hinges represent the 25th and 75th percentiles, lower and upper whiskers represent the maximum or minimum value no further than 1.5 * IQR (inter-quartile range), central line indicates the median. Individual points represent unique barcodes, shape corresponds to donor (d1-2) and replicate (rep1-2). F (i) rs12325245 (risk allele T; SLC38A7) is a GWAS variant associated with SCZ-metabolic syndrome pleiotropy; the non-risk allele had greater regulatory activity at baseline. (ii) MPRA-emVars included GWAS variants annotated for psych-psych pleiotropy or psych-cardiometabolic pleiotropy. G–I Activity-by-contact (ABC) enhancer-gene interaction scoring: (G) Track annotation of stable emVar, rs246002 at the PCDHA locus. H Heatmap of MAGMA enrichments across psychiatric, substance use (SUD), neurological, metabolic/autoimmune disorders, and cardiometabolic GWAS, colored by –log10(p_FDR). I ABC gene targets of interaction MPRA-emVars revealed cue-responsive regulation by IL-6 and IFNα, affected genes involved in neurodevelopment, neurological disorder, and cardio-metabolic traits. Each point represents a gene list of either (i) IFNα interaction MPRA-emVar gene targets or (ii) IL-6 interaction MPRA-emVar gene targets. Color of the point indicates the MPRA-QTL values used to calculate the ABC score; red dashed line represents a P_FDR < 0.1 significance cut-off (FDR multiple testing correction). Source data are provided as a Source Data file and on synapse (syn75962204).

Interaction MPRA-emVars were defined as single-nucleotide changes that significantly altered transcriptional activity in a differential manner between vehicle and either IL-6 or IFNα (mpralm cue*variant interaction test; p_FDR ≤ 0.1). 18% and 10% of MPRA active CRS resolved significant cue-by-variant interaction effects (IL-6-v-base: 61 of 337, IFNα-v-base = 29 of 287; q_storey < 0.1, Fig. 3D). Across cues, 267 unique variants showed significant regulatory effects in at least one condition (Fig. 3B), with 64 showing significant cue-by-variant interaction effects (Fig. 3D).

Whereas some top loci were stable across all conditions (e.g., rs246002, Fig. 3E, i: baseline p = 0.04, IL-6 p = 0.027, IFNα p = 7.4 × 10⁻⁶, Wilcoxon Rank Sum Test ref-vs-alt allele), others were dynamically regulated by immune activation (Fig. 3D–G, e.g., rs3024744, Fig. 3E, ii: consistent allelic effects, but variable magnitudes of effect, and rs12325245, Fig. 3E, iii: significant variant effects at baseline only; Kruskal-Wallis: p = 3.6 × 10⁻⁷, Post-hoc Dunn’s test: base-ref vs. IL-6-ref p_nom = 0.05, base-ref vs. IFNα-ref p_Holm = 0.025, base-alt vs. IFNα-alt p_Holm = 0.006). 44–52% of the 64 interaction MPRA-emVars were immune-specific: active CRSs with a significant variant interaction effect only at baseline or only after with IFNα and/or IL-6 exposure (Fig. 3D). For example, the SCZ-metabolic pleiotropy risk variant rs12325245-T (SLC38A7, Fig. 3E, ii, Fig. 3F) had significantly lower activity than the non-risk allele at baseline (p_FDR= 0.017) but showed no allelic effect following IL-6 exposure despite being a highly active MPRA-CRS (mpralm interaction p_FDR = 0.02) and was transcriptionally repressed following IFNα exposure. Conversely, 38–48% of interaction MPRA-emVars were immune-responsive: significant emVars across cues with significant IFNα and/or IL-6 interaction effects, and the direction of effect retained across cues. Notably, no interaction MPRA-emVars were significant QTLs at both baseline and after cue exposure, but with reversed direction of effect across cues (SI Data 2.15, 2.16).

Target genes regulated by cue-specific MPRA-emVars were predicted using an adapted activity-by-contact⁸⁵ model (STARE⁸⁶) that incorporated enhancer activity (MPRA activity and donor- and cue-matched neuronal chromatin accessibility), contact (Hi-C contact frequency), and predicted transcription factor (TF) binding affinities. Enhancer-to-gene mapping generally replicated the original eGene distance-based associations while also identifying more distal targets. For example, while rs246002 stably regulates transcription across conditions, incorporation of endogenous information at the PCDHA locus shows differential chromatin accessibility peaks across conditions and iGLUT-specific chromatin loops, integration of which allowed for more specific mapping of emVars to gene targets (Fig. 3G). MPRA-emVar ABC gene targets were highly enriched for psychiatric disorder risk genes, including SCZ, BIP, and cross-neuropsychiatric disorder pleiotropy (CxD) (MAGMA⁸⁷), as well as common comorbid metabolic and immune syndromes (Fig. 3H). Notably, when specifically considering interaction MPRA-emVar ABC gene targets (SI Data 2.21, 2.22), enrichments were stronger for neurodevelopmental and neurological disorders, and non-brain disorders such as type-II diabetes and hypertension (Fig. 3I).

Variant-specific disruptions in transcription factor binding affinities are associated with dynamic inflammatory-responsive regulation of psychiatric risk loci

CRSs can regulate gene expression through binding of sequence-specific transcription factors (TFs) to cognate motifs. In fact, TFs are a major mechanism of dynamic regulatory activity^35,43,44,88, with activity influenced by epigenetic state of the genomic locus^89,90 and TF affinity for the binding site^91,92, among other factors.

We scanned MPRA-emVars (p_FDR ≤ 0.1) for variant-specific disruptions in predicted TF binding affinity (q_Storey< 0.05) across 505 unique TFs expressed in mature iGLUTs (1573 motifs) and compared correlations between TF binding disruptions and MPRA variant-specific activity. We annotated significantly correlated TFs based on downstream gene expression changes, cue-specific changes in chromatin accessibility, and differential peak motif-enrichments (Fig. 4). Top TF motif enrichments, chromatin accessibility, and TF gene expression varied by context (Fig. 4A and SI Fig. 16A) and included expected immune-responsive TFs: IL-6 with interleukin-associated TFs (e.g., CREB1, SMAD3); IFNα with interferon targets (e.g., IRF9, STAT1) as well as immune-responsive TFs involved in neural development and differentiation (SI Fig. 16B, C and SI Data 2.20). For example, binding affinity z-scores of TFAP2C_5, involved in neural development, most highly correlated with activity following exposure to IFNα (r = 0.81, p = 2 × 10⁻⁴, n = 18, Fig. 4B).

Fig. 4. Dynamic regulation associated with variant-specific binding affinities and cue-specific changes in chromatin accessibility, related to SI Fig. 18 and SI Data 2.20.

A Top predicted transcription factors (TFs) regulating MPRA-expression modulating variants (emVars) overlapped with cue-responsive ATAC peaks motif enrichments and temporally dynamic cue-responsive gene expression (24hrs-vs-48hrs). Dot plot and heatmap of predicted TFs, organized by magnitude of correlation with MPRA-emVars, that either overlapped with differentially accessible ATAC peaks (cue-vs-vehicle, green; color = peak concentration), ATAC motif enrichments (purple; color = -log10(p_Bon)) or differentially expressed genes (size = -log10(p_nom), shape = direction of logFC, fill = two-sided Pearson’s correlation coefficient (R)). B TFAP2C_5 is a top correlated motif following IFNα exposure, but not at baseline; individual SNPs represented as points, R=two-sided Pearson’s Correlation Coefficient. C TF average gene expression was significantly positively correlated with MPRA-motif correlations at 24 hrs and 48 hrs. For (B, C, E) lines were fit with a linear model (y ~ x) with bands displaying standard error at 0.95 confidence. D Comparative over-representation analysis (REACTOME pathways) between baseline, IL−6, and IFNα conditions (TFs selected from A). Size represents the total number of unique term enrichments affecting a signaling cascade, ordered and colored by broader category. Signaling cascades related to example TFs (B, E) are in bold. E Known downstream TFs of interferon and interleukin signaling regulate MPRA-emVars. (i) STAT1_2 (interaction emVars n = 2) and (ii) MYC_disc4 (interaction emVars n = 4) were highly correlated with MPRA variant-specific effects, and (iii) STAT1 and MYC were differentially expressed following IFNa and IL-6 exposure, respectively. F The FTCDNL1 locus is dynamically regulated by immune-responsive MPRA-emVars (rs769948, rs281789) that have variant-specific effects on STAT1 binding affinity. They overlap with annotated brain regulatory elements (ENCODE) and iGLUT-specific chromatin loops and are brain-eQTLs (GTEx v8 Frontal Cortex) of FTCDNL1 and TYW4. Integration of iGLUT-specific chromatin accessibility peaks and Hi-C loops (ABC) further map these dynamic emVars to regulation of distal targets > 500 KB away, including SATB2 (rs769948) and AOX1/AOX3P (rs281789). Source data are provided as a Source Data file and on synapse (syn75962204).

MPRA variant effects were positively correlated with cue-responsive TFs expressed in iGLUTs (strongest at 24hrs post exposure for IL-6; R = 0.71, p = 0.00066, 48hrs post exposure for IFNα; R = 0.58, p = 0.00044, Fig. 4C). Comparative gene set enrichment analysis of the TFs implicated in regulating responses at baseline and in response to IL-6 and IFNα resolved interleukin, STAT, and immune signaling as expected, but also highlighted enrichments for cellular stress response, metabolism, and neural differentiation (Fig. 4D). For example, predicted disruptions in motif binding affinities of STAT1 (r = 0.54, p = 0.0394, n = 15) and MYC (r = 0.73, p = 2.4 × 10⁻⁵, n = 27), both involved in NOTCH signaling cascades impacting neural differentiation, were highly correlated with context-responsive MPRA-emVar effects (Fig. 4E, i-ii), differentially enriched in context-specific ATAC accessible regions (STAT1, IFNα: q_Storey = 0.08, MYC, IL-6: q_Storey = 0.02), and corresponded to dysregulated TF expression (STAT1, logFC = 4.5, p_FDR = 4.46 × 10⁻³³; MYC: logFC = 0.58, p_nom = 0.01; Fig. 4E, iii).

Dynamic transcriptional regulation (MPRA-emVars), chromatin accessibility (ATAC peaks), chromatin contact (iGLUT Hi-C loops), and variant-specific predicted TF binding affinities influenced dynamic genotype-by-immune interactions at psychiatric GWAS loci (Fig. 4F). For example, the FTCDNL1 locus, associated with SCZ, metabolic syndrome, and cognition, was dynamically regulated by MPRA-validated brain-eQTLs with proximal and distal gene targets: the reference alleles at rs769948 (C) and rs281789 (G) were predicted to increase binding affinity for STAT1 (q_Storey= 0.0008, q_Storey = 0.037, Fig. 4F). rs769948-C significantly increased transcriptional activity at baseline (p_FDR= 8.9 × 10⁻³) and with IL-6 (p_FDR= 9.7 × 10⁻⁴), but was repressed following exposure to IFNα, whereas rs281789-G increased activity only at baseline (base: p_FDR = 0.09, interaction IFNα-v-base p_FDR = 0.057, IL6-v-base p_FDR = 0.03). While brain eQTLs mapped these SNPs to regulation of FTCDNL1/TYW4, enhancer-to-gene mapping with iGLUT-specific chromatin loops also predicted multi-gene regulation at STAB2 ( ~ 500 KB upstream; schizoaffective disorder and serum level immunoglobulin glycosylation GWAS target gene) and AOX1/AOX3P ( ~ 750 kb downstream; associated with response to antidepressants and drug metabolism, Fig. 4F). Overall, dynamic regulation of GWAS risk loci in response to neuronal inflammation is associated with variant-specific binding affinities and cue-specific changes in TF gene expression and chromatin accessibility.

Discussion

By testing the regulatory activity of 152 GWAS loci linked to brain diseases and traits across dynamic immune-regulated cues in live human neurons, we modeled neuronal immune response during neurodevelopment, demonstrating gene x environment interactions of broad relevance to subsequent neuropsychiatric outcomes. We characterized effects at 152 GWAS loci, testing 3668 CRSs in human neurons, of which 1156 were active in at least one condition. In total, we resolved 267 variant-specific MPRA-emVars across 859 (baseline), 860 (IL-6), and 719 (IFNα) active MPRA-CRS; 31% had significant cue-by-variant interaction effects (interaction MPRA-emVars). Overall, dynamic effects tended to affect the magnitude but not the direction of regulatory activity. Our neuron lenti-MPRA replicated MPRA-CRSs previously reported in NPCs, was concordant with brain eQTLs, and highlighted known risk-associated eGenes, but expanded upon these existing resources by incorporating dynamic regulatory activity.

Fetal brain development is exquisitely sensitive to inflammatory insults throughout pregnancy⁹³, with both maternal¹⁷ and fetal⁹³ immune dysfunction, as well as genetic defects resulting in type I IFN overproduction^94,95 linked to altered neurodevelopment and increased risk of psychiatric and neurological outcomes in offspring. We selected two cytokines, IL-6 and IFNα, based on evidence from animal models^20,21,96,97, cell culture^28–30,59, inflammatory biomarkers^98–100, brain imaging^101,102, and post-mortem brain analyses^26,103. Both cytokines are believed to act directly on neurons: maternal IL-6 accumulates in the placenta¹⁰⁴ and crosses the fetal blood-brain barrier¹⁰⁵, whereas IFNα is generated by fetal brain cells in response to infection³¹.

Despite being a simple acute treatment paradigm, we observed time-dependent effects worth considering in subsequent studies of dynamic genetic regulation. First, IL6R and IL6ST expression increased within 24 h of IL-6 treatment but declined nearly to baseline by 48 h; likewise, IFNΑR1 and IFNΑR2 expression decreased following exposure to IFNα, but returned to baseline by 48 h. These trends paralleled genome-wide effects; although gene expression signatures were strongly correlated across time points, there were fewer DEGs and decreased magnitude of effects over time, consistent with a rapid initial response and subsequent attempt to return to homeostasis (of note, mature iGLUTs were treated one time, without re-exposure). Transcription at 24 h post-treatment correlated more strongly with MPRA effects measured at 48 h. These dynamic patterns indicated that gene expression and chromatin accessibility changes should be monitored over time, with careful consideration of pseudotime effects in MPRA studies when dissecting mechanistic pathways impacted by cue-dependent genetic regulation.

Our library design prioritized eQTLs from the adult brain, but MPRA was conducted in iGLUTs that more resemble fetal-like neurons; the resulting MPRA-emVars showed significant concordance with both adult and fetal eQTLs. Patterns of gene expression change across neurodevelopment^106,107 and aging¹⁰⁸, which may underly critical periods of heighted susceptibility to environmental stressors. Such changes presumably arise from differences in genetic regulation, reflecting shifting patterns of TF expression that mediate enhancer activity^109,110. Age-specific eQTLs modify the functional impact of genetic risk throughout lifespan^37,111,112 and are particularly relevant given the variable age of onset of symptom presentation across brain disorders. A variety of methods now exist to maintain^113,114 or accelerate^115–118 aging in vitro, which could be incorporated into future MPRAs, towards dissecting the dynamic functional impact of genetic risk across neurodevelopment, brain maturation, and aging to inform mapping of brain-related GWAS and facilitate precision medicine.

Notable technical limitations in MPRA design and methodology reduce the broad generalizability of our GxE analyses. First, reflecting the present state of GWAS, the variants tested were identified in exclusively European-ancestry data and excluded the MHC locus. Moreover, many recent publicly available GWAS remain underpowered, overall biasing the MPRA library towards specific disorders (e.g., eating disorders, PTSD, and OCD had very few variants included). Second, technical limitations restricted CRSs to short DNA fragments flanking prioritized variants, potentially omitting crucial portions of larger regulatory regions. Recent technological innovations in MPRA (e.g., tiling MPRA design¹¹⁹) should facilitate examination of more complex regulatory sequences moving forward. Third, although we designed the library to test 8960 sequences, not all are represented in the final analysis; 7267 were represented with adequate barcode coverage in our physical lenti-MPRA library and of these, we acquired reads for ~ 6700, filtered to 3668 with adequate barcode coverage. MPRA design should be mindful of physical library complexity and neuronal culture sizes to ensure a comprehensive analysis of the full MPRA library. A notable limitation of lenti-MPRA design is the risk of CRS-barcode swapping, mitigated here by our use of 5’UTR barcoding. Fourth, although MPRA activity was highly correlated between donors (r = 0.87-0.95), polygenic donor genotype may influence functional genomics, especially when exploring different contexts, and warrants further exploration^120,121. Fifth, while MPRA-emVar effects were correlated with previous findings, the number of overlapping sequences was small ( < 10%) and do not provide robust validation; four previously published MPRAs^{54,78,122,123} based on psychiatric GWAS likewise had low to moderate overlap in variants (SI Data 2.17). Lastly, as CRSs were tested outside their endogenous context, results may not accurately inform regulatory activity of distal gene targets. Although new methods to integrate matched cue-specific epigenetic datasets refine predictions of gene targets of enhancer activity^85,124, further experimental validation through crisprQTL¹²⁵ or prime-editing¹²⁶ is warranted. Future studies of cue-specific GxE effects across additional variants, doses, longitudinal and recovery time-points, cues, cell-types (particularly brain-specific immune cells), and ultimately within more physiologically relevant brain organoids via emerging single-cell MPRA methods¹²⁷, and across an expanded number of donors via village-in-a-dish approaches¹²⁸ will be crucial for further dissection of how immune activation during fetal development contributes to brain disorder risk.

The influence of pro-inflammatory environmental factors on brain-related regulatory elements may explain biological mechanisms through which immune signaling contributes to increased susceptibility for complex brain disorders. We identified hundreds of GWAS variants that confer greater susceptibility to complex brain disorders following developmental exposure to neuroinflammation and mapped cue-specific risk-associated genes, informing the influence of GxE interactions across complex brain traits. Dynamic neuronal immune-response genetic regulation frequently reflected combinatorial effects of multiple variants within a GWAS locus, with distinct variants top-ranked across neuroimmune cues; moreover, loci with immune-specific genetically regulated transcription were annotated for pleiotropic effects across multiple traits. Likewise, downstream target genes of interaction MPRA-emVars were uniquely enriched for complex brain disorders and common comorbid metabolic and immune syndromes. The clinical impact of resolving GxE interactions includes preventative measures (e.g., improved prenatal care, social supports, and early life interventions for high-risk individuals) and care of current patients (e.g., drug repurposing, patient stratification by immune status, personalized prescription). Overall, when attempting to understand the genetic mechanisms of variable penetrance and pleiotropy, with broad relevance across complex traits and disorders, it is critical to consider the impact of dynamic regulation of gene expression.

Methods

Lenti-MPRA library design across neuropsychiatric trait GWAS

A Lenti-MPRA¹²⁹ library was designed by statistical fine-mapping of ten GWAS: (Alzheimer’s disease (AD)⁶⁷, attention deficit hyper-activity disorder (ADHD)⁶⁸, anorexia nervosa (AN)⁶⁹, autism spectrum disorder (ASD)³, bipolar disorder (BIP)⁷⁰, major depressive disorder (MDD)⁷¹, obsessive compulsive disorder (OCD)⁷², post-traumatic stress disorder (PTSD)⁷³, schizophrenia (SCZ)¹, and neuroticism (NEU-P)⁷⁴) using two complimentary methods incorporating dorsolateral prefrontal cortex (DLPFC)⁷⁵ expression quantitative trait loci (eQTLs): Bayesian co-localization (coloc2⁷⁶) and transcriptomic imputation (S-PrediXcan⁷⁷, SI Fig. 7A; SI Data 2.1–2.3). For the former, significant GWAS loci and DLPFC⁷⁵ eQTLs from the Common Mind Consortium (CMC) were tested for co-localization using a lenient significance threshold for GWAS loci (p < 1 × 10⁻⁶). The most probable causal eQTLs from moderately to highly colocalized loci (PPH4 ≥ 0.5) were selected, along with all SNPs in high LD (r²≥ 0.9, SI Data 2.2, 2.3); controls for coloc2 prioritization were selected from significant BIP GWAS loci that (i) did not colocalize (PPH3 > 0.9) and (ii) were not significant CMC DLPFC eQTLs. For S-PrediXcan, all SNPs within the predictor models of Bonferroni-corrected significant (p < 4.64 × 10⁻⁶;(0.05/10786)) trait-associated S-PrediXcan genes (eGenes) and all SNPs in high LD (r²≥ 0.8) with them were selected (SI Data 2.4, 2.5).

Several strategies were applied to incorporate appropriate controls. First, scramble sequences (n = 100) were added as negative controls. Only the scramble controls were used for normalization and as a measure of basal activity of the minimal promoter. For a point of comparison, two additional sets of benchmark variants were included. As empirical controls, 310 SCZ and AD GWAS SNPs producing the greatest (n = 164) and least (n = 146) transcriptional shifts in K562 chronic myelogenous leukemia lymphoblasts and SK-SY5Y human neuroblastoma cells⁷⁸ that overlapped with CMC DLPFC eQTLs, were included to compare results of previously tested sequences. Third, to query the regulatory activity of significant GWAS loci lacking eQTL associations, we included 88 significant BIP GWAS SNPs that (i) did not colocalize (PPH4 < 0.1/PPH3 > 0.9) and (ii) were not significant CMC DLPFC eQTLs.

After accounting for SNPs identified through multiple strategies and removal of SNPs with sequences containing restriction digest sites (AgeI/SfbI), we synthesized a library of 4430 SNPs (8860 variants), many of which overlapped with chromatin accessible peaks in the DLPFC and hiPSC-derived neurons, and 100 scramble controls. Oligonucleotides were synthesized by Agilent and cloned into the lenti-MPRA vector⁵².

We performed a power analysis using designmpra¹³⁰, which estimates the power of a t test to distinguish differential activity in a variant at a Bonferroni corrected alpha = .05 level. At 4430 bi-allelic pairs, 100 barcodes per SNP, activity standard deviation of 1 (typical range = 0.3-2), and 4 replicates, the power of a t test to detect differential variant shifts of (logFC) of 0.5 or greater at a Bonferroni corrected alpha = 0.05 level is 100%.

Lenti-MPRA library preparation and viral titration

The MPRA library was generated according to published lenti-MPRA protocols with slight modifications¹²⁹. Briefly, 200 base pair oligonucleotides flanking each prioritized SNP were synthesized by Agilent to create an MPRA library of 9244 neuropsychiatric associated variants. The Agilent oligo pool was PCR amplified, and a minimal promoter and spacer sequence added downstream of the CRS. This protocol uses a 5′ UTR barcoding method uses a shorter distance (102 bp) between the CRS and barcode than 3′ UTR barcoding methods (801 bp), reducing the risk of CRS–barcode swapping. Amplified fragments were purified and amplified again for 15 cycles to add a random 15 bp sequence to serve as a unique barcode. Barcoded fragments were inserted in the SbfI/AgeI site of the pLS-SceI vector (AddGene #13772) and then transformed into 10-beta competent cells (NEB, C3020) via electroporation. Bacterial colonies were grown overnight on Ampicillin-positive plates and midi-prepped for plasmid collection. The quality of the purified plasmid was evaluated by Sanger Sequencing of 16 colonies at random. CRS-barcode associations were identified by sequencing of the purified plasmid (MiSeq; paired-end; 15milion reads). Our final MPRA library consisted of 7267 CRS (77% of the designed) that were represented at a minimum of 10 barcodes. 2^nd-generation lentiviral packaging of the purified plasmid was performed by the viral core at Boston’s Children Hospital. To determine multiplicity of infection (MOI) and approximation of appropriate viral volume we infected day 14 iGLUTs (0, 1, 2, 4, 8, 10, 16, 32, 64 µL) with control lentivirus (pLS-SV40-mP-EGFP; AddGene #137724) and harvested for 48 hrs later. Following DNA isolation, we performed qPCR to calculate the MOI based the relative ratios of genomic DNA to inserted viral DNA (after subtracting background noise caused by residual backbone DNA).

Cell-type deconvolution of postmortem DLPFC

We checked for the relative abundance of cell-type in the postmortem CMC DLPFC by performing cell-type deconvolution with the R package dtangle¹³¹ and a single-cell expression reference panel from the cerebral cortex. In the CMC DLPFC, glutamatergic neurons make up the highest proportion of cells based on cell-type deconvolution.

NGN2-glutamatergic neuron induction from clonalized hiPSC lines for molecular experiments^56,57

Clonal hiPSCs from two neurotypical donors of European ancestry with average schizophrenia PRS and no history of psychiatric diagnoses (#3182 (XX) and #2607 (XY)) were generated by lentiviral transduction with pLV-TetO-hNGN2-eGFP-Neo (Addgene #99378), and lentiviral FUW-M2rtTA (Addgene #20342), followed by antibiotic selection and clonal expansion. Stably selected clones were validated to ensure robust cell survival, expression of fluorescent tags, and transgene expression. hiPSCs were maintained in StemFlex™ Medium (ThermoFisher #A3349401) and passaged with EDTA (Life Technologies #15575-020).

On day 1, medium was switched to non-viral induction medium (DMEM/F12 (Thermofisher, #10565018), 1% N-2 (Thermofisher, #17502048), 2% B-27-RA (Thermofisher, #12587010)) and doxycycline (dox) was added to each well at a final concentration of 1 µg/mL. At day 2, transduced hiPSCs were treated with 500 µg/mL G418 (Thermofisher, #10131035). At day 4, medium was replaced, including 1 µg/mL dox and 4 µM cytosine arabinoside (Ara-C) to reduce the proliferation of non-neuronal cells. On day 5, young neurons were dissociated with Accutase Cell Detachment Solution (Innovative Cell Technologies, # AT-104), counted and seeded at a density of 1 × 10⁶ per well of a Matrigel-coated 12-well plate. Medium was switched to Brainphys neuron medium (Brainphys (STEMCELL, # 05790), 1% N-2, 2% B27-RA, 1 μg/mL Natural Mouse Laminin (Thermofisher, # 23017015), 10 ng/mL BDNF (R&D, #248), 10 ng/mL GDNF (R&D, #212), 500 μg/mL Dibutyryl cyclic-AMP (Sigma, #D0627), 200 nM L-ascorbic acid (Sigma, # A4403)). For seeding, 10 mM Thiazovivin (Millipore, #S1459), 500 μg/mL G418 and 4 μM Ara-C and 1 μg/mL dox were added. At day 6, medium was replaced with Brainphys neuron medium with 4 μM Ara-C and 1 μg/mL dox. Subsequently, 50% of the medium was replaced with fresh neuronal medium (lacking dox and Ara-C) once every other day until the neurons were harvested at D24.

Cue-Specific RNAseq (SI Figs. 1, 2-5, 8 and SI Data 1)

Briefly, mature iGLUTs (22 DIV), were acutely exposed (24 or 48 h) to IL-6 (25 ng/mL (24 hr only) and 60 ng/mL), IFNα (100 UI/mL (24 hr only) and 500 IU/mL), or vehicle (0.1% FBS in ultrapure H₂O before harvest at 24 DIV (two control donors, three to four replicates per donor per cue/vehicle treatment; experimental schematic: SI Fig. 1).

Compound	Solvent	Supplier	Product #	Conc.
Human-recombinant IL-6	UltraPure H2O	Sigma/Aldrich	GF338	60 ng/mL
Human-recombinant INFa2-b	UltraPure H2O	Mount Sinai Pharmacy	NDC 0085-4350-01	500 IU/mL

RNA Sequencing libraries were prepared using the Kapa Total RNA library prep kit. Paired-end sequencing reads (100 bp) were generated on a NovaSeq platform. Raw reads were aligned to hg38 using STAR aligner¹³² (v2.5.2a), and gene-level expression were quantified by featureCounts¹³³ (v1.6.3) based on the Ensemble GRCh38 annotation model. Genes with over 10 counts per million (CPM) in at least four samples were retained. Surrogate variable analysis (SVA) was performed to assess the correlation of known covariates and identify surrogate variables contributing to variance using the R packages sva¹³⁴ and variancePartition¹³⁵ (e.g., batch, treatment, donor, replicate, sex, and RIN). Following identification of SVs and covariates to correct for in the model, raw read counts were normalized with voom¹⁰² and, due to the repeated measures study design, where individuals are represented by multiple independent technical replicates, differential expression analysis with repeated measures was performed by the Dream method from variancePartition¹³⁵. Bayes shrinkage (limma::eBayes) estimated modified t- and p- values. Gene-level significance values were adjusted for multiple testing using the Benjamini-Hochberg method to control the false discovery rate (FDR). Genes with FDR <  5% were considered significantly differentially expressed (limma::TopTable)¹³⁶ (SI Data 1.1–1.5). In these analyses, the t test statistics from the differential expression contrast were used to rank genes in the GSEA using the R package ClusterProfiler¹⁰⁶ (SI Data 1.6). Gene-set enrichment analysis using WebGestalt¹³⁷ was performed between p_nom < 0.05 DEGs and gene-sets curated from four rodent models of maternal immune activate: poly(I | C) exposure (conceptus, whole brain, amygdala, and frontal cortex)^24,66,138, IL-6 exposure (whole brain)²⁴, H1N1 Flu virus infection (whole brain)²⁴, and chronic unpredictable maternal stress (whole brain)¹³⁹.

Meta-analysis of gene expression across cues

We performed a meta-analysis and Cochran’s heterogeneity Q-test (METAL¹⁴⁰) using the p-values and direction of effects (t-statistic), weighted according to sample size across all sets of perturbations (Target vs. Scramble DEGs). Genes were defined as convergent if they (1) had the same direction of effect across cue exposure (2) were Bonferroni significant in our meta-analysis (Bonferroni adjusted p-value; p_Bon ≤ 0.05), and (3) had a non-significant Cochran’s Heterogeneity Test (SI Data 1.7).

Cue-Specific ATAC-Seq (SI Figs. 1, 4-5 and SI Data 1)

Briefly, mature iGLUTs (22 DIV), were acutely exposed (48 hours) to IL-6 (60 ng/mL), IFNα (500 IU/mL), or vehicle (0.1% FBS in ultrapure H₂O) before harvest at 24 DIV (two control donors, two replicates per donor per cue/vehicle treatment; experimental schematic: SI Fig. 1). Mature neurons were washed with 500uL of PBS (-Ca/-Mg)-0.5 mM EDTA per well of a 12-well plate. Then, 300uL dissociation solution (0.042 U/µL papain suspension (Worthington-Biochem LS003126) in HBSS (Thermofisher #14025076)-10mM HEPES (Thermofisher #J61275AE)-0.5 mM EDTA (Life Technologies #15575-020), pre-activated at 37 C for 5 min) supplemented with 0.017 µ/µL DNase (Thermofisher #EN0521) and 1x Chroman I was added to each well before incubating the plate at 37 °C for 10 min, shaking at 125 rpm. 600 μL deactivating solution (DMEM-FBS-Chroman I) was then added to each well, and cells were dissociated into single cells by pipetting gently. For each condition, cells from 4 wells of a 12-well-plate were combined into a single 15 mL conical tube for higher yield. After spinning at 600 × g for 5 min at room temperature, cells were resuspended in 310 µL of DMEM (Thermofisher #10566-016)-10% FBS. Then, the cell suspension was filtered through a 37 µm reversible strainer and frozen in DMEM-10% FBS-10% DMSO.

ATAC sequencing library prep and sequencing were performed by the Yale Sequencing Core. The adapter sequence for pair-end sequencing was removed using trim_galore¹⁴¹, and sequencing quality measure by FastQC¹⁴² and MultiQC¹⁴³. Each experiment contained two technical replicates and two biological replicates (four samples total). All R1 and R2 fastqs passed basic QC with FastQC; total sequences per sample ranged from 47.2 million to 79.7 million after removal of mitochondrial reads (average 61 M), with percent deduplicated ranging from 65–80% (average 74.38%). Average GC content was normally distributed with a mean of 44% and average sequence length 85–119. Data was aligned with Bowtie2¹⁴⁴ against the hg38 reference genome, including rare SNVs. Mitochondrial reads were removed, sam files sorted and indexed, and converted to compressed BAM files with samtools¹⁴⁵. Peak calling and reproducibility analysis were performed using MACS2 (2.1.0)¹⁴⁶ and IDR (version, soft-idr-threshold 0.05)¹⁴⁷, respectively, according to ENCODE ATAC-seq pipeline (v1 2019)¹⁴⁸ specifications. Fraction of reads in peaks (FRiP) scores were calculated for each sample from the alignment file (bam) and MACS2 narrowPeak files: FRiP scores ranged from 0.23-0.35, passing ENCODE ATACseq data standards (SI Data 1.8). All replicates passed the two ENCODE standards for IDR, with a ratio of pooled pseudoreplicate results to true replicate results (Np/Nt) as well as self-replicate peaks (N1/N2) within a factor of two. Self-consistency ratio: max(N1,N2)/ min(N1,N2) = 1.56-2.86; and Rescue Ratio: max(Np,Nt)/ min(NP,Nt) = 1.32-1.64. Consensus peaks were defined as ATAC peaks present in at least 4 out of 6 pairwise true-replicate IDR comparisons for each condition (4 samples total), with at least 50% overlap (bedtools intersect -wa -u -f 0.50) resulting in a range of 19,373-27,930 peaks per condition. A single merged file was generated for each condition for visualization purposes by merging bam files with samtools, then intersecting that file with the list of consensus peaks via bedtools intersect. Bed files for each merged peak were opened with Integrative Genomics Viewer (IGV_2.16.0)¹⁴⁹ for figure generation. Separate narrowpeak files for each ATAC replicate were also intersected with the list of consensus peaks per condition to be used for differential accessibility analysis with DiffBind¹⁵⁰. Counts for all replicates across conditions were normalized to full library size, donor was used as a blocking factor, treating within donor replicates as repeated measures, condition-vehicle comparisons were defined as individual contrasts, and differential sites for each contrast calculated using edgeR at a threshold of Benjamin-Hochberg method of multiple testing correction; p_FDR < 0.1, or a nominal p-value < 0.05 for subsequent analysis. Transcription factor binding site motif enrichment was performed using Homer (findMotifsGenome.pl -size given -mask) on differential accessible regions identified at both significance thresholds¹⁵¹. Peak annotation was performed using ChIPseeker¹⁵² (v 1.8.6), and gene set over representation analysis of genes mapped to open chromatin regions was performed with ClusterProfiler¹⁵³ using Gene Ontology, WikiPathway, KEGG, and REACTOME gene sets.

Cue-specific MPRA (SI Fig. 1, 7–18 and SI Data 2)

Briefly, the MPRA library was transduced into mature iGLUTs (21 DIV), 24 h later neurons were acutely exposed (48 h) to IL-6 (60 ng/mL), IFNα (500 IU/mL) or vehicle (0.1% FBS in ultrapure H₂O) before harvest at 24 DIV (two control donors, two biological replicates each, experimental schematic: SI Fig. 1). Day 21 iGLUTs were spinfected (1krcf for 1 hr @37 C, slow accel, slow deceleration) with lenti-MPRA library, (based on titrations of the control virus from the Gordon et al. 2020 lenti-MPRA Nature Protocol Supplement). 24 h after spinfections, full media was replaced to remove un-integrated virus. At 48 h post-infection, iGLUTs were treated with stress and inflammatory cues or basal media. The number of cells required pre-replicate was calculated according to the Gordon et al. 2020 protocol, with on average, 2 million hiPSCs seeded per well of a six-well plate and later batched together as one technical replicate for an average of 6 million cells per replicate. 72 h post-lentiviral-infection, and 48 h post-exposure to stress or inflammatory compounds, neuronal cells were washed three times and harvested using AllPrep DNA/RNA mini kit (Qiagen) and the libraries prepped as previously described. The libraries were sequenced as paired end reads on a NovaSeq 2 × 50 on S2 flow cell (3.3-4.1 B reads/cell) by the New York Genome Center.

Lenti-MPRA CRS-barcode association from MiSeq

Sequencing of purified plasmid DNA was sequenced on an Illumina MiSeq v.2 (15 million paired-end reads and generated fastq files with bcl2fastq (parameters: --minimum-trimmed-read-length 0 --mask-short-adapter-reads 0). Barcode-CRS association was performed as previously described using the association utility of MPRAflow v2.3.5¹²⁹ (run as: nextflow run association.nf –fastq-insert –fastq-bc –fastq-insertPE –mapq 3 –baseq 15).

Lenti-MPRA RNA/DNA counts

We demultiplexed the indexed DNA and RNA libraries and generated paired-end fastq files with bcl2fastq v2.20 and used the count utility of MPRAflow 2.3.5 both with and without the --mpranalyze flag included (run as: nextflow run count.nf -w –experiment-file –dir –outdir –labels –design –bc-length 14 –umi-length 16 --thresh 10) to compute the activity score for each element and produce count files formatted for analysis with MPRAnalyze (SI Data 2.10)⁷⁹. Roughly 6700 inserts were sequenced (92% of the library). We filtered out barcode-CRS pairs with low DNA counts ( < 15) and those with RNA counts, but no DNA counts, leaving 3668 CRSs with a minimum requirement of 10 barcodes each.

Lenti-MPRA comparison of regulatory activity across replicates, donors, and cue exposures

Transcriptional activity, measured as the normalized log2 DNA and RNA counts per CRS, across conditions was strongly correlated between replicates (Pearson’s Correlation Coefficient r = 0.98-1.00). Normalized log2(RNA/DNA) ratios were also strongly correlated (mean r = 0.83, max r = 0.93), with only 1 comparison (donor 2607 following IFNα exposure below a correlation of 0.83; at 0.61; SI Fig. 8). When technical replicates were averaged together and compared across donors, correlations were high (r = 0.87-0.95), and we retained all replicates in downstream analyses. (Fig. 1A and SI Data 2.11). Across sequences, the number of barcodes per unique CRS were highly correlated between CRS shared across all conditions (r = 0.997–0.999). There was no significant difference in the mean number of barcodes (n ~ 45) per insert or the proportion of CRSs by prioritization method or disorder association across the conditions.

After filtering, we re-performed a power analysis (designmpra¹³⁰) with the actual values post filtering. At 1000 bi-allelic pairs, with an average of 45 barcodes per SNP, activity standard deviation of 1 (typical range = 0.3-2), and 4 replicates, we have 80% power to detect differential variant shifts of (logFC) of ~ 0.5 at a Bonferroni corrected alpha = 0.05 level using a t test.

Comparison of three approaches for evaluating MPRA activity and variant specific shifts

We performed a comparison of quantification, variant-specific differential analysis, and cue-by-variant interaction testing across three methods: MPRAnalyze⁷⁹, mpralm⁸⁰, and DEseq2^81,82:

MPRAnalyze⁷⁹

quantification

Mean transcriptional activity across replicates was calculated using MPRAnalyze. First, library size correction factors were estimated across replicates and donors using estimateDepthFactors(obj, lib.factor = c(“replicate”,”donor”,) which.lib = ”both”, depth.estimator = ”upper quantile”). Activity was quantified using analyzeQuantification (obj = obj, dnaDesign = ~ donor + replicate, rnaDesign = ~ donor + replicate). Alpha values representing the transcriptional rate of each sequence were extracted from the fitted model. CRS were labeled active if the MAD score p_FDR ≤ 0.05.

variant-specific differential analysis

First, RNA/DNA counts were separated by reference and alternative alleles and merged. Library size correction factors were estimated across replicates and donors using estimateDepthFactors(obj, lib.factor=c(“replicate”,”donor”) which.lib = ”both”, depth.estimator = ”upper quantile”). Differential activity was calculated between alternate (alt) and reference (ref) alleles using the classic mode version of analyzeComparative(mpraobject, dnaDesign = ~ replicate + donor + variant + barcode, rnaDesign = ~ variant, reducedDesign = ~ 1, mode = ”classic”) and testLrt(obj).

cue-by-variant interaction analysis

For active MPRA CRS and significant MPRA-emVars at either baseline or in IL-6 or IFNα, cue-by-variant effects were tested using an interaction analysis. First, conditions were merged based on shared sequences, and library size correction factors were estimated across replicates, donors, and conditions using estimateDepthFactors(obj, lib.factor = c(“replicate”,”donor”,”cue”) which.lib = ”both”, depth.estimator = ”upper quantile”). Possible interaction effects between IL-6 or IFNα exposure and baseline (vehicle) was performed using a cue-by-variant interaction term and significance tested using analyzeComparative(mpraobject, dnaDesign = ~ replicate + donor + cue_variant, rnaDesign = ~ cue_variant, reducedDesign = ~ 1, mode = ”classic”) and MPRAnalyze::testLrt(obj)”.

mpralm⁸⁰

quantification

We assessed the activity of CRS compared to the average activity of all CRS with mpralm() (mpra v.1.16.0) followed by eBayes() abd topTable() (limma v3.50.3) after correcting for donor and replicate ( ~ 1 + donor + replicate). Sequences that had positive logFC ( > 0) and met a statistical significance threshold at FDR < 0.1 were labeled as an active CRS, while those with negative logFC and FDR < 0.1 were labeled as a repressed CRS. Only active CRSs were used in the downstream analysis of differential variant effects.

variant-specific differential analysis

We applied mpralm() (mpra v.1.16.0) followed by topTable() (limma v3.50.3) to detect allelic effects in MPRA active CRS (MPRA sequence pairs with one of the sequences active in any condition). We set a statistical significance threshold at FDR < 0.1 to define emVars.

cue-by-variant interaction analysis

Cue-by-variant interaction effects were assessed with ~ cue:variant + donor + rep. Only MPRA CRS with significant allelic differences (MPRA-emVars) in at least one condition were used to test for interaction effects. We set a statistical significance threshold at FDR < 0.1 to define interaction emVars.

DEseq2

CRS quantification and allelic analysis with DEseq2 as described⁸².

quantification

We assessed the activity of CRS compared to the average activity of all CRS with DESeq2 using a nested model: ~ material + material:donor + material:replicate. Here, ‘material’ describes the aggregate allele-independent effect observed in RNA compared to DNA (expression effects). ‘material:donor’ and ‘material:replicate’ describes the sample-specific effects nested within RNA/DNA. Active CRS were defined as those with logFC > 0 and FDR < 0.1.

variant-specific differential analysis

For sequence pairs where at least one pair was an active CRS, we tested for variant specific effects with DEseq2 using a nested model: ~ material + material:donor + material:replicate + material:allele. Here, ‘material:allele’ describes the allele-specific effects nested with RNA/DNA. We set a 10% FDR threshold to define MPRA-emVars.

cue-by-variant interaction analysis

For emVars that were significant in at least one condition, we tested for cue-by-variant interaction effects with DEseq2: ~ material + material:donor + material:replicate + material:allele + material:allele:cue. Here, ‘material:allele:cue’ describes the cue-by-variant interaction effects nested with RNA/DNA. We set a 10% FDR threshold to define interaction emVars.

We compared quantification results across methods and observed that 43% of active CRS were identified by all three methods, with the greatest overlap (~ 93%) between mpralm and DEseq2. MPRanalyze identified 463 CRS that were not indicated as being active by either mpralm or DEseq2. We also compared the logFC and -log10(p-values) for each analysis across methods (SI Figure 9-10). DEseq2 and mpralm methods had the most agreement, with Pearson correlations between 0.9-0.97 for log fold change estimates and correlations between 0.8-0.91. for the -log10(p-value). MPRAnalyze showed the least agreement, moderately correlated with both mpralm and DEseq2 (logFC R = 0.59–0.67, -log10(p-value) R = 0.48–0.68). We calculated expected p-values for each method and compared them with the observed p-values to evaluate the degree of inflation. For both the variant-specific differential and interaction analyses, DEseq2 results had notably inflated p-values (SI Fig. 10E). Given these findings, all downstream analyses were performed on results from mpralm.

Analysis of concordance and comparison to published MPRA studies

We first reviewed the overall number of tested CRS and the number of donors, replicates, scramble controls, and barcodes across 8 previously published MPRAs (SI Data 2.18). Our number of total replicates (4/condition; 2 donor, 2 technical replicates pooled across 6 wells) exceeds that of Kreimer et al. 2022 (1 donor, 3 reps)¹¹⁰, Rummel et al. 2023 (3 reps)⁵³ and are comparable to the replicates in Deng et al. 2024 (5 replicates)¹²³. Based on a minimum threshold of 10 barcodes per variant across all replicates, our mean number of barcodes (mean n = 45 barcode) is comparable to the mean in Inoue et al. 2019 (n = 70)⁵², Rummel et al. 2023 (n = 45)⁵³, and Deng et al. 2024 (n = 64)¹²³ and exceeds those in Tewhey et al. 2016 (n = 20)⁵¹, Myint et al. 2019 (n = 10)⁷⁸, and Mulvey et al. 2021 (n = 10)¹²². We specifically examined correlations between MPRA variant shifts in our study and in others testing psychiatric GWAS SNPs^{54,78,122,123} across different significance thresholds (SI Data 2.17). We compared activity and absolute overlap, reporting comparisons where there was enough overlap for Pearson’s correlation analysis (SI Data 2.17).

Of note, overlapping variants were not significantly correlated, or there were too few to provide robust cross-study validation. Many factors including the GWAS used, prioritization methods, size of the MPRA library, cell-type or environmental context, and the quantification methods (mpralm, MPRAnalyze, DESeq2, etc.) may explain this lack of overlap and correlation. However, this highlights a need to intentionally include a greater number of previously tested SNPs in MPRA library design and to replicate findings in the same cell types and treatment paradigms independently.

Analysis of allelic shifts in MPRA activity and comparison to eQTL datasets

We first assessed the absolute overlap of all MPRA sequences post filtering and bulk adult PFC⁷⁵, fetal brain⁸³, and the CMC adult DLPFC eQTLs. For overlapping MPRA-emVars and eQTLs, we defined the percent concordant as the number of significant overlap SNPs with the same direction of effect in the MPRA-emVar and eQTL over the total number of overlapping SNPs. We further pared down this list by focusing on eQTL-eGene pairs that matched our prioritization by S-PrediXcan and coloc2. We calculated the Pearson’s Correlation Coefficient (r) between MPRA-emVar logFC and eQTL betas from post-mortem single-cell⁸⁴ and bulk adult PFC⁷⁵, and fetal brain⁸³.

Variant-specific predicted binding affinity scoring with atSNP and MotifBreakR

To identify TF that may influence GxE interaction related to stress, performed we motif affinity testing for SNPs with significant allelic shifts as measured by lenti-MPRA (MPRA-emVars) using atSNP¹⁵⁴. ENCODE derived TF motif binding PWM matrices were loaded and filtered for TF with expression in D24 iGLUTs, leaving 1573 motifs across 505 TFs. Genomic context (30 bp half window size) of significant MPRA-emVars (p_FDR ≤ 0.1) for each condition were pulled with atSNP::LoadSNPData. SNP allele affinity scores were calculated for each motif with atSNP::ComputeMotifScore and p-values computed using atSNP::ComputePValues. Rank p-values were used to assess the significance of the SNP effect on the affinity change. Bonferroni multiple testing correction was performed along with calculation of Storey’s q-values and local FDR. Motif with significant SNP effects were tested for TF enrichments using hypergeometric tests with multiple testing correction across the total number of TFs tested. To identify TFs with binding patterns that predict variant-specific activity, SNP-motif pairs were filtered based on q_Storey < 0.05. Pearson’s Correlation Coefficients (r) were then calculated between MPRA-emVar activity (ref-vs-alt) and the logOdds that the reference allele enhanced affinity binding was assessed for each motif. To visualize allelic effects on predicted TF binding affinities for specific MPRA-emVars, we used the motifbreakR package¹⁵⁵, filtered for strong allelic effects of TFs expressed in DIV24 iGLUTs.

Activity-by-Contact prediction of cue-specific regulatory element target genes

To predict target genes of condition-specific enhancer activity, we scored enhancer-gene interactions using STARE⁸⁶. STARE combines an adapted Activity-By-Contact (ABC)⁸⁵ interaction modeling with predicted TF binding affinities in regions to summarize these affinities at the gene level. For each condition, we filtered significant MPRA-emVar (p_FDR< 0.1) or significant interaction MPRA-emVar (p_FDR< 0.1) for overlap with matched cue-specific chromatin accessibility peaks (p_nom< 0.05) in iGLUTs. Enhancer activity for these variants was then calculated as the abs(logFC) compared to the alternative allele or baseline, respectively, multiplied by the scaled peak AUC, thus incorporating crucial information about proximal cue-specific chromatin activity. Distal chromatin contact was measured from Hi-C data in iGLUTs¹⁵⁶, which included one overlapping donor (2607). Genes were initially filtered by ABC score > 0.02 (default), but more stringent filtering was performed based on ABC score median thresholds prior to downstream enrichment analysis.

Over-representation analysis and biological theme comparison of correlated TFs and ABC genes

To identify pathway enrichments unique to cue-specific regulatory activity, we performed biological theme comparison using ClusterProfiler¹⁵³ and gene set enrichment for GWAS catalog risk genes using the GENE2FUNC query tool of FUMA GWAS¹⁵⁷.

Enrichment analysis of convergence for risk loci using MAGMA

We tested ABC genes for enrichment with genetic risk of psychiatric, neurological, cardiometabolic, and immune disorders/traits [Psychiatric: attention-deficit/hyperactivity disorder (ADHD)⁶⁸, anorexia nervosa (AN¹⁵⁸, including binge-purge AN-BP and restrictive subtypes AN-R¹⁵⁹), autism spectrum disorder (ASD)^3,160, alcohol dependence (AUD)¹⁶¹, bipolar disorder (BIP, BIP-I, BIP-II)⁷⁰ cannabis use disorder (CUD)¹⁶², major depressive disorder (MDD)⁷¹, obsessive-compulsive disorder (OCD)⁷², post-traumatic stress disorder (PTSD)¹⁶³, and schizophrenia (SCZ)¹, Cross Disorder (CxD)¹⁶⁴, Tourette’s¹⁶⁵, and neurotic personality traits⁷⁴; neurologic: Alzheimer disease (AD)⁶⁷, Parkinson disease (PD)¹⁶⁶, amyotrophic lateral sclerosis (ALS)¹⁶⁷, Multiple Sclerosis (MS)¹⁶⁸, Epilepsy (Epi)¹⁶⁹, migraine¹⁷⁰, chronic pain¹⁷¹; Inflammatory-Gastrointestinal: Diverticular Disease (DivD)¹⁷², Gastro-esophageal reflux disease (GORD)¹⁷², Peptic Ulcer Disease (PUD)¹⁷², Inflammatory Bowel Disease (IBD)¹⁷², Irritable Bowel Syndrome (IBS)¹⁷²; Cardiometabolic: Type-I Diabetes (T1D)¹⁷³, Type-2 Diabetes (T2D)¹⁷⁴, Metabolic Syndrome (MetS)¹⁷⁵, Atrial Fibrillation (Afib)¹⁷⁶, hypertension (HyperTen)¹⁷⁷. Anthropometric: Body Mass Index (BMI)¹⁷⁸, left-handedness (left hand)¹⁷⁹, GWAS summary statistics using multi-marker analysis of genomic annotation (MAGMA)⁸⁷. SNPs were mapped to genes based on the corresponding build files for each GWAS summary dataset using the default method, snp-wise = mean (a test of the mean SNP association). A competitive gene set analysis was then used to test enrichment in genetic risk for a disorder across gene sets with an p_FDR < 0.05.

iGLUT neuron induction from non-clonal hiPSC-derived NPCs for phenotypic assays^56,57

hiPSCs-derived NPCs were dissociated with Accutase Cell Detachment Solution (Innovative Cell Technologies, #AT-104), counted and transduced with rtTA (Addgene 20342) and NGN2 (Addgene 99378) lentiviruses in StemFlex media containing 10 mM Thiazovivin (Millipore, #S1459). They were subsequently seeded at 1 × 10⁶ cells/well in the prepared 6-well plate. On day 1, medium was switched to non-viral induction medium (DMEM/F12 (Thermofisher, #10565018), 1% N-2 (Thermofisher, #17502048), 2% B-27-RA (Thermofisher, #12587010)) and doxycycline (dox) was added to each well at a final concentration of 1 μg/mL. At day 2, transduced hiPSCs were treated with 500 μg/mL G418 (Thermofisher, #10131035). At day 4, medium was replaced, including 1 μg/mL dox and 4 μM cytosine arabinoside (Ara-C) to reduce the proliferation of non-neuronal cells. On day 5, immature NPC-iGLUTs were dissociated with Accutase Cell Detachment Solution (Innovative Cell Technologies, #AT-104), counted and seeded at a density of 1 × 10⁶ per well of a Matrigel-coated 12-well plate. Medium was switched to Brainphys neuron medium (Brainphys (STEMCELL, # 05790), 1% N-2, 2% B27-RA, 1 μg/mL Natural Mouse Laminin (Thermofisher, # 23017015), 10 ng/mL BDNF (R&D, #248), 10 ng/mL GDNF (R&D, #212), 500 μg/mL Dibutyryl cyclic-AMP (Sigma, #D0627), 200 nM L-ascorbic acid (Sigma, # A4403)). For seeding, 10 mM Thiazovivin (Millipore, #S1459), 500 μg/mL G418 and 4 μM Ara-C and 1 μg/mL dox were added. At day 6, medium was replaced with Brainphys neuron medium with 4 μM Ara-C and 1 μg/mL dox. Subsequently, 50% of the medium was replaced with fresh neuronal medium (lacking dox and Ara-C) once every other day until the NPC-iGLUTs were harvested at d21.

Neurite analysis

Day 7 NPC-iGLUTs were seeded as 1.5 × 10⁴ cells/well in a 96-well plate coated with 4x Matrigel at day 3, followed by half medium changes until the neurons were fixed at day 7. At day 5, NPC-iGLUTs were treated for 48 h with either IL-6 (25 ng/mL and 60 ng/mL), INFa-2b (100 IU/mL and 500 IU/mL), or matched vehicles. Following cue exposure, NPC-iGLUTs were fixed using 4% formaldehyde/sucrose in PBS with Ca²⁺ and Mg²⁺ for 10 min at room temperature (RT). Fixed cultures were washed twice in PBS and permeabilized and blocked using 0.1% Triton/2% Normal Donkey Serum (NDS) in PBS for two hours. Cultures were then incubated with primary antibody solution (1:1000 MAP2 anti chicken (Abcam, ab5392) in PBS with 2% NDS) overnight at 4 °C. Cultures were then washed 3x with PBS and incubated with secondary antibody solution (1:500 donkey anti chicken Alexa 647 (Life technologies, A10042) in PBS with 2% NDS) for 1 hour at RT. Cultures were washed a further 3x with PBS, with the second wash containing 1 μg/ml DAPI. Fixed cultures were then imaged on a CellInsight CX7 HCS Platform with a 20x objective (0.4 NA), and neurite tracing analysis performed using the neurite tracing module in the Thermo Scientific HCS Studio 4.0 Cell Analysis Software. 12 wells were imaged per condition across two control donor hiPSCs, with 9 images acquired per well for neurite tracing analysis. A one-way ANOVA with a post hoc Bonferroni multiple comparisons test was performed on data averaged for each well for neurite length per neuron using Graphpad Prism.

Synapse analyses

Commercially available primary human astrocytes (pHAs, Sciencell, #1800; isolated from fetal female brain) were seeded on D3 NPC-iGLUTs at 1.7 × 10⁴ cells per well on a 4x Matrigel-coated 96 W plate in neuronal media supplemented with 2% fetal bovine serum (FBS). NPC-iGLUTs were seeded over the astrocyte monolayer as 1.5 × 10⁵ cells/well at day 5 post-induction. Half changes of neuronal media were performed twice a week until fixation. At day 13, NPC-iGLUTs + astrocyte co-cultures were treated with 200 nM Ara-C to reduce the proliferation of non-neuronal cells in the culture. At day 18, Ara-C was completely withdrawn by full medium change followed by half medium changes until the neurons were fixed at day 21. At day 21, NPC-iGLUTs + astrocytes co-cultures were treated for 48hrs with IL-6 (25 ng/mL and 60 ng/mL), IFNα-2b (100 IU/mL and 500 IU/mL), or matched vehicles. Following exposure, NPC-iGLUTs were fixed and immune-stained as described previously, with an additional antibody stain for Synapsin1 (primary antibody: 1:500 Synapsin1 anti mouse (Synaptic Systems, 106 011); secondary antibody: donkey anti mouse Alexa 568 (Life technologies A10037)). Stained cultures were imaged and analyzed as above using the synaptogenesis module in the Thermo Scientific HCS Studio 4.0 Cell Analysis Software to determine SYN1 + puncta number, area, and intensity per neurite length in each image. 20 wells were imaged per condition across two control donor hiPSCs, with 9 images acquired per well for synaptic puncta analysis. A one-way ANOVA with a post hoc Bonferroni multiple comparisons test was performed on data averaged for each well for puncta number per neurite length using Graphpad Prism.

Antibody	Species		Supplier	Product #	Dilution
MAP2		Ck	Abcam	ab5392		1:500
SYNAPSIN1 Alexa 568 anti-Mouse		Ms Ms	Synaptic Systems Life technologies	106 011 A10037		1:500 1:500
Alexa 647 anti-Chicken		Ck	Life technologies	A10042		1:500

Multiple Electrode array (MEA)

Commercially available primary human astrocytes (pHAs, Sciencell, #1800; isolated from fetal female brain) were seeded on D3 NPC-iGLUTs at 1.7 × 10⁴ cells per well on a 4x Matrigel-coated 48 W MEA plate (catalog no. M768-tMEA-48W; Axion Biosystems) in neuronal media supplemented with 2% fetal bovine serum (FBS). At D5, iGLUTs were detached, spun down and seeded on the pHA cultures at 1.5 × 10⁵ cells per well. Half changes of neuronal media supplemented with 2% FBS were performed twice a week until day 42. At day 13, NPC-iGLUT + astrocyte co-cultures were treated with 200 nM Ara-C to reduce the proliferation of non-neuronal cells in the culture. At Day 18, Ara-C was completely withdrawn by full medium change. At day 26, NPC-iGLUTs + astrocytes co-cultures were treated for 48 h with IL-6 (25 ng/mL and 60 ng/mL), IFNα-2b (100 IU/mL and 500 IU/mL), or matched vehicles. Following exposure, the electrical activity of iGLUTs was recorded at 37 °C using the Axion Maestro MEA reader (Axion Biosystems). Recording was performed via AxiS 2.4. The batch mode/statistic compiler tool was run following the final recording. Quantitative analysis of the recording was exported as a Microsoft Excel sheet. 6-12 wells were analyzed per condition across two control donor hiPSCs, with 16 electrodes per well for MEA. Data is log-transformed and normalized within each plate, prior to one-way ANOVA with a post hoc Bonferroni multiple comparisons test.

Statement of ethics

Ethical approval was not required because the hiPSC lines, lacking association with any identifying information and widely accessible from a public repository, are thus not considered to be human subject research. Post-mortem brain data are similarly lacking identifiable information and are not considered human subject research.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

The paper was written by K.G.R., L.H., and K.J.B., with input from all authors. All high-throughput MPRA and RNA-sequencing data and downstream analyses were performed by K.G.R. L.D. provided code and guidance in conducting Bayesian colocalization analyses. H.Y. provided guidance in eQTL analyses. S.L and M.J. performed cue-specific ATAC experiments. K.G.R. and S.E.W. performed ATAC-seq analysis. P.M.D. performed dose-dependent morphological assays and analysis. M.F.G and K.G.R. performed cell-line clonalization. K.G.R. performed MPRA library preparation and MPRA experiments with the assistance of M.F.G., S.C. [Samuel Cartwright], S.C. [Sophie Cohen], and A.S.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by F31MH130122 (K.G.R), R01MH109897 (K.J.B.), R56MH101454 (K.J.B., L.H.), R01MH123155 (K.J.B.) and R01ES033630 (L.H., K.J.B.), R01MH124839 (L.M.H), R01MH106056 (K.J.B) U01DA047880 (K.J.B), R01DA048279 (K.J.B), DOD TP220451 (K.J.B. and L.H.), and by the State of Connecticut, Department of Mental Health and Addiction Services. This publication does not express the views of the Department of Mental Health and Addiction Services or the State of Connecticut. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript

Data availability

All source donor hiPSCs have been deposited at the Rutgers University Cell and DNA Repository (study 160; http://www.nimhstemcells.org/). The raw and processed high-throughput sequencing data generated in this study have been deposited on the Gene Expression Omnibus (GEO) database under accession code GSE341428 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE341428). The secondary data and summary statistics generated in this study are provided in the Supplementary Information/Source Data file and are available through Synapse (syn75962204; https://www.synapse.org/Synapse:syn75962204/wiki/). Hi-C data used for annotation in this study are previously published and available through www.synapse.org/#!Synapse:syn12979101 (registration required; Data Download—Study “iPSC-HiC” and through the PsychENCODE Knowledge Portal (https://psychencode.synapse.org/). The PsychENCODE Knowledge Portal is a platform for accessing data, analyses, and tools generated through grants funded by the National Institute of Mental Health (NIMH) PsychENCODE program. Data are available for general research use according to the following requirements for data access and data attribution: (https://psychencode.synapse.org/DataAccess). Track annotations are publicly available from BrainScope (https://brainscope.gersteinlab.org/) and IGV (https://igv.org/app/). GWAS summary statistics are publicly available from the Psychiatric Genomics Consortium (https://pgc.unc.edu/for-researchers/download-results/). Common Mind Consortium eQTL summary statistics can be accessed through a data cces request on the NIMH Data Archive (NDA; https://nda.nih.gov/). GWAS annotations were downloaded from the GWAS catalog (https://www.ebi.ac.uk/gwas/docs/file-downloads). Source data are provided in this paper.

Code availability

The full analysis pipeline (including code and processed data objects) used for analysis of RNA-seq, ATAC-seq, and MPRA data evaluation is publicly available through Synapse (syn75962204; https://www.synapse.org/Synapse:syn75962204).

Competing interests

K.J.B. is a scientific advisor to Rumi Scientific Inc. and Neuro Pharmaka Inc. M.F.G. now works for Alkermes. L.D. now works for Ionis Pharmaceuticals. All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77114-x.