Perinatal insults, genetic susceptibility, and the pathogenesis of neurodevelopmental disorders: from epigenetic remodeling to neural circuit dysfunction—a narrative review

Abstract

Background and Objective

Neurodevelopmental disorders (NDDs)—including autism spectrum disorder (ASD), attention-deficit/hyperactivity disorder (ADHD), intellectual disability (ID), epilepsy, and schizophrenia—affect approximately 10–15% of children worldwide. Emerging evidence suggests that NDDs arise from complex interactions between genetic susceptibility and environmental insults, particularly during perinatal periods. This narrative review aims to synthesize current evidence on the molecular and circuit-level mechanisms linking perinatal insults and genetic vulnerabilities to NDD pathogenesis, and to propose an integrative framework across epidemiology, genetics, epigenetics, and neurobiology.

Methods

We searched PubMed/MEDLINE, Web of Science, and Google Scholar for articles published from January 2000 to April 2026, combining terms for NDDs, genetic susceptibility, perinatal exposures, and molecular mechanisms. We prioritized original research, meta-analyses, systematic reviews, and consortium studies; human studies were given precedence, with animal and organoid studies included only when providing unique mechanistic insights.

Key Content and Findings

The review covers five themes: (I) genetic architecture from rare variants to polygenic risk; (II) gene-environment interplay, distinguishing true interaction from genetic confounding; (III) epigenetic mediation (DNA methylation, histone modifications, noncoding RNAs); (IV) circuit dysfunction from synaptic/ion channel deficits to brain-wide networks (cortex, cerebellum, thalamus, brainstem); and (V) emerging therapeutic strategies with a translational roadmap. Its distinctive innovation lies in integrating these layers across developmental time, proposing epigenetic remodeling as a key molecular transducer linking early insults to persistent circuit dysfunction.

Conclusions

Critical gaps persist—ancestry bias, lack of unified quantitative risk models, and limited therapy translation. Future priorities include diversifying genomic cohorts, establishing multi-omic birth cohorts, developing early-detection biomarkers for the perinatal window, and evaluating targeted interventions in genetically defined subgroups. This integrative framework aims to guide developmentally sensitive, precision-oriented approaches to NDD research and clinical care.

Introduction

Neurodevelopmental disorders (NDDs) encompass a group of early-onset conditions, including autism spectrum disorder (ASD), attention-deficit/hyperactivity disorder (ADHD), intellectual disability (ID), epilepsy (which is included by its shared genetic pathways such as synaptic/ion channel genes), and schizophrenia (which we include here under the neurodevelopmental model, while noting that its evidence base—largely indirect and derived from adult outcomes—differs from that of the pediatric NDDs), that affect approximately 10–15% of children worldwide and confer lifelong functional impairment (1-3). While once considered discrete diagnostic entities, accumulating evidence reveals substantial clinical overlap, frequent comorbidity, and shared genetic risk across NDDs, suggesting a continuous neurodevelopmental spectrum rather than categorical boundaries (3-5).

The etiology of NDDs is fundamentally multifactorial. Studies have demonstrated moderate heritability—37–50% for ASD—but have also revealed that shared environmental factors account for up to 55% of variance in liability, challenging purely genetic models and implicating gene-environment interplay as a core mechanism (6,7). The genetic architecture of NDDs spans the entire allelic frequency spectrum: rare de novo loss-of-function (LoF) and missense mutations in hundreds of genes converging on synaptic function, chromatin remodeling, and transcriptional regulation (8-11); recurrent copy number variants (CNVs) with incomplete penetrance at loci such as 16p11.2 and 22q11.2 (12-14); and common polygenic variation that collectively contributes to risk and modifies phenotypic expression in carriers of rare mutations (7,15,16).

Concurrently, epidemiological studies have identified diverse perinatal risk factors—maternal infection, inflammation, hypoxia, nutritional deficiency, toxin exposure, and stress—that elevate NDD risk (15,17-19). In this review, we refer to these collectively as “perinatal insults”, defined as adverse exposures occurring from conception through the first postnatal year, including the prenatal (all trimesters), peripartum (labor/delivery), neonatal (first 28 days), and early infancy (up to 12 months) periods. The perinatal period represents a critical window of brain vulnerability, during which genetic programs governing neuronal proliferation, migration, synaptogenesis, and myelination are exquisitely sensitive to disruption (19,20). However, most exposed individuals do not develop NDDs, and many affected individuals lack identifiable environmental adversity. This discordance points inexorably toward gene-environment interaction (G×E): genetic susceptibility may render certain individuals vulnerable to insults that are benign in resistant hosts (7,15,21). Indeed, maternal polygenic risk for ADHD confounds associations with pregnancy-related exposures (15), while a genetic inflammation score interacts with prenatal phthalate levels to predict ASD and ADHD outcomes (18).

The molecular mediators of this interplay remain incompletely understood, but emerging evidence highlights epigenetic mechanisms—DNA methylation, histone modifications, and noncoding RNAs—as dynamically responsive to environmental signals and frequently disrupted in NDDs (22,23). Rare epigenetic variations (epivariations) at dosage-sensitive genes are enriched in affected individuals (23), and mutations in chromatin remodelers such as CHD8, BRPF1, and EP400 are strongly associated with syndromic NDDs (22,24-26). At the systems level, genetic and environmental perturbations converge on neural circuit formation: NDD risk genes are enriched in mid-fetal cortical projection neurons and interneurons (24,27), and disorder-associated CNVs produce distinct brain endophenotypes detectable by neuroimaging (27-29).

Despite extraordinary progress, critical gaps persist. Most NDD heritability remains unexplained; non-European and consanguineous populations are severely underrepresented (30,31); and translating genetic discoveries into mechanism-based therapies has proven exceptionally difficult (32,33). This review synthesizes current evidence on the interplay between perinatal insults and genetic susceptibility in NDDs—from epigenetic modifications to neural circuit dysfunction—and proposes an integrative framework for future research bridging developmental timing, multi-omic profiling, and precision intervention (Figure 1). We present this article in accordance with the Narrative Review reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0615/rc).

Figure 1.

Integrative framework linking perinatal insults, genetic susceptibility, epigenetic remodeling, neural circuit dysfunction, and neurodevelopmental disorders. The figure illustrates the hierarchical cascade proposed in this review. Perinatal insults and genetic susceptibility converge via gene-environment interactions to drive epigenetic remodeling (DNA methylation, histone modifications, noncoding RNAs, chromatin remodeling). These alterations lead to neural circuit dysfunction (synaptic/ion channel deficits, cortical malformations, cerebellar/subcortical involvement) and ultimately to neurodevelopmental disorders (ASD, ADHD, ID, epilepsy, schizophrenia). Arrows indicate the proposed causal cascade. ADHD, attention‑deficit/hyperactivity disorder; ASD, autism spectrum disorder; HIE, hypoxic-ischemic encephalopathy; ID, intellectual disability; PRS, polygenic risk score.

Methods

This narrative review was conducted to provide an integrated overview of the interplay between perinatal insults, genetic susceptibility, and NDD pathogenesis. We searched PubMed/MEDLINE, Web of Science, and Google Scholar for articles published from January 2000 to April 2026. The search strategy combined terms across four domains (NDDs, genetics, perinatal exposures, and molecular mechanisms) and is detailed in Table 1. We prioritized original research, meta-analyses, systematic reviews, and consortium studies. Human studies were given precedence; animal/organoid studies were included only when providing unique mechanistic insights. Single case reports without molecular data, non-English articles, and conference abstracts were excluded. Candidate gene findings without replication are discussed with explicit caveats. Two authors (J.L. and Y.Z.) independently screened titles and abstracts; disagreements were resolved by consensus with a third author (J.H.) as needed. As a narrative review, no quantitative synthesis was performed.

Table 1. Search strategy summary.

Items	Specification
Date of search	April 15, 2026
Databases and other sources searched	PubMed/MEDLINE, Web of Science, Google Scholar. Additional studies via reference lists and major consortia
Search terms used	Group 1 (NDDs): “neurodevelopmental disorders”, “ASD”, “ADHD”, “intellectual disability”, “epilepsy”, “schizophrenia”. Group 2 (Genetics): “genetic susceptibility”, “rare variants”, “de novo mutations”, “CNV”, “PRS”, “GWAS”. Group 3 (Perinatal): “perinatal”, “prenatal”, “maternal infection”, “inflammation”, “hypoxia”, “preterm birth”, “HIE”, “phthalates”. Group 4 (Mechanisms): “epigenetics”, “DNA methylation”, “histone modification”, “noncoding RNA”, “synaptic dysfunction”, “neural circuits”. Filters: English; human prioritized; January 2000–April 2026
Timeframe	January 2000–April 2026
Inclusion criteria	Original research, meta-analyses, systematic reviews, consortium studies; human studies prioritized; animal/organoid studies with unique mechanistic insights
Exclusion criteria	Single case reports without molecular data; conference abstracts; non-English articles; non-replicated candidate gene findings (discussed with caveats)
Selection process	Two authors (J.L., Y.Z.) independently screened; consensus with third author (J.H.) for disagreements
Additional considerations	Narrative review; no quantitative synthesis or formal risk-of-bias assessment performed

Genetic architecture of NDD

Rare variants and de novo mutations

Trio-based exome and genome sequencing have fundamentally reshaped the landscape of rare variant discovery in NDDs. By enabling unbiased comparison between affected probands and their unaffected parents, these approaches have firmly established de novo mutations—particularly protein-truncating and deleterious missense variants—as a major causal mechanism, accounting for 30–50% of solved cases in well-phenotyped cohorts (11,34,35). Beyond diagnostic yield, the aggregation of tens of thousands of trios has revealed unifying principles governing rare NDD-associated variation.

First, de novo LoF variants are significantly enriched in NDD probands and concentrated in genes highly intolerant to functional variation in the general population (8,10,36). Constraint metrics such as probability of loss-of-function intolerance (pLI) and loss-of-function observed/expected upper bound fraction (LOEUF) have become indispensable for prioritization; genes in the top decile of constraint are up to six-fold more enriched for high-confidence NDD risk variants than those in the bottom decile (10). Second, recurrence—the observation of independent de novo mutations in the same gene across unrelated individuals—provides the strongest statistical evidence for causation. Meta-analyses of over 20,000 trios have now implicated more than 250 genes with genome-wide significant enrichment, a number that continues to grow (9,10,26).

Despite extraordinary heterogeneity—over 1,500 risk genes identified to date—the encoded proteins converge on a limited set of biological pathways (9). Synaptic function constitutes the largest category, encompassing postsynaptic scaffolds (SHANK3, DLG4), ion channels (SCN2A, KCNQ2, CACNA1I, GRIN2B), and vesicular trafficking regulators (SCAMP5, STXBP1) (11,37-41). Chromatin remodeling and transcriptional regulation form a second major hub, including histone modifiers (CHD8, KMT5B, SETD1A, EP400), chromatin readers (BRPF1, ZMYND11), and transcription factors (TCF7L2, FEZF2, FOXP1) (22,24,25,36,42-45). The convergence of rare risk genes on synaptic function, chromatin remodeling, and transcriptional regulation has been further solidified by large-scale exome sequencing efforts. For instance, Satterstrom et al. [2020] analyzed over 35,000 ASD samples and identified 102 high-confidence autism risk genes, the majority of which converge on neuronal development and synaptic transmission (46). Similarly, Singh et al. [2022] extended this paradigm to schizophrenia, demonstrating that rare coding variants in genes intolerant to LoF mutations contribute significantly to schizophrenia risk, with notable overlap with ASD and ID genes (47). Neuronal development and migration pathways—cytoskeletal dynamics (TUBG1, MAPK8IP3, MARK2), axon guidance (EPHA7, DCC), and cell adhesion (NRXN1, CNTN6)—represent a third convergent group (14,48-53). This convergence implies that core disease mechanisms are considerably fewer than the genes that disrupt them, offering tractable targets for therapeutic development (32,33,42) (Figure 2).

Figure 2.

Functional convergence of representative NDD genes onto neural circuit dysfunction. This concentric diagram shows that hundreds of NDD-associated genes converge on a limited set of mechanisms. Outer ring: gene categories (ion channels, synaptic proteins, chromatin remodelers, transcription factors, splicing factors/noncoding RNAs). Middle ring: representative genes (SCN2A, SHANK3, CHD8, FEZF2, RNU4‑2). Inner ring: core disrupted mechanisms (altered excitability, synaptic defects, epigenetic dysregulation, abnormal developmental timing). Center: neural circuit dysfunction, the common final pathway to NDDs. This convergence supports targeting shared downstream mechanisms rather than individual genes. NDD, neurodevelopmental disorder.

A central and still unresolved challenge is the marked variability in clinical outcomes among carriers of identical pathogenic variants. Penetrance is highly context-dependent. Truncating mutations in CHD8, for instance, exhibit near-complete penetrance for ASD or developmental delay, whereas missense alleles in the same gene are associated with milder or atypical presentations (42,54). The phenotypic spectrum of SCN2A mutations ranges from severe early-infantile epileptic encephalopathy to isolated ASD without epilepsy, depending on mutation subtype and affected protein domain (11,40). This variable expressivity is increasingly attributed to genetic background. Common polygenic risk scores (PRS) modify phenotypic severity in carriers of high-penetrance alleles, including 22q11.2 and NRXN1 deletions (7,14,55). Moreover, a second rare variant—another CNV or deleterious single-nucleotide variant—can exacerbate severity, a phenomenon termed dual diagnosis or oligogenic inheritance (7,14,55). Such observations challenge the traditional monogenic model and suggest that many NDDs reflect the summation of multiple genetic hits along a continuous liability distribution (7,16).

Despite substantial progress, most individuals with NDDs remain undiagnosed after exome sequencing, particularly those from non-European and consanguineous populations (30,31,56). A substantial portion of this missing heritability resides outside protein-coding regions. The recent discovery that recurrent de novo mutations in the noncoding gene RNU4-2—a component of the major spliceosome—represent one of the most common causes of syndromic ID exemplifies the untapped diagnostic potential of the noncoding genome (30,57,58). Similarly, homozygous splice-site variants in LSM1 is associated with autosomal recessive NDD (59). These findings underscore the necessity of whole-genome sequencing (WGS) and systematic reanalysis of unsolved cases (58,60). Long-read sequencing technologies are further expanding the detectable mutational spectrum by resolving complex structural variants, mobile element insertions, and tandem repeat expansions invisible to short-read platforms (60,61).

The translational impact of identifying a pathogenic de novo variant is immediate and multifaceted: it ends the diagnostic odyssey, enables accurate recurrence risk counselling, and often informs surveillance for associated comorbidities (34,35,62). In a subset of cases, molecular diagnosis directly guides treatment—for example, ketogenic diet in GLUT1 deficiency or avoidance of sodium channel-blocking agents in SCN1A-related Dravet syndrome (11,63). However, such genotype-guided acute management is currently available for only a minority of NDDs; most mechanisms-based therapies [e.g., gene therapy, antisense oligonucleotides (ASOs), histone deacetylase (HDAC) inhibitors] remain experimental and are not yet standard clinical care. Beyond diagnostics, each newly discovered NDD gene provides an entry point for mechanistic dissection and therapeutic development. Patient-derived induced pluripotent stem cell (iPSC), CRISPR-engineered rodent models, and human brain organoids have illuminated how specific mutations disrupt neuronal excitability, synaptic transmission, transcriptional programs, and circuit assembly (37,64-66). These platforms are now being leveraged to test ASOs, small molecules, and gene therapy approaches in a genotype-specific manner (32,64,65).

In summary, rare de novo variants constitute a well-established and clinically actionable class of NDD risk alleles. Their study has illuminated fundamental biological pathways disrupted in these disorders and provided a foundation for genotype-driven stratification, mechanistic investigation, and precision medicine. The continued expansion of sequencing efforts to encompass diverse populations, noncoding regions, and complex structural variation will be essential to complete the genetic atlas of NDDs and translate these discoveries into improved outcomes for affected individuals and families. Table 2 summarizes representative high‑confidence risk genes for NDDs according to their primary biological function, including synaptic transmission, chromatin remodeling, ion channel activity, transcriptional regulation, and neuronal migration.

Table 2. High-confidence risk genes for NDDs by functional category.

Functional category	Representative genes	Associated NDDs	Mutation type
Synaptic function	SHANK3, DLG4, NRXN1, STXBP1, SCAMP5	ASD, ID, epilepsy	De novo LoF, CNV
Ion channel function	SCN2A, KCNQ2, CACNA1I, GRIN2B, KCNB1	Epilepsy, ASD, ID	Gain/LoF, missense
Chromatin remodeling	CHD8, KMT5B, EP400, BRPF1, SETD1A	ASD, ID, epilepsy	LoF, missense
Transcriptional regulation	TCF7L2, FEZF2, FOXP1, ZMYND11	ASD, ID, ADHD	De novo LoF
Neuronal migration & axon guidance	TUBG1, MAPK8IP3, MARK2, EPHA7, DCC	ID, epilepsy, microcephaly	Missense, CNV

This table summarizes representative high-confidence risk genes for NDDs according to their primary biological function. Genes were selected based on meta-analyses of large-scale exome and genome sequencing studies. Functional categories include synaptic transmission (postsynaptic scaffolds, vesicular trafficking), ion channel activity (sodium, potassium, calcium channels), chromatin remodeling and histone modification, transcriptional regulation, and neuronal migration/axon guidance. Mutation types are indicated as de novo LoF, missense, or CNVs. ADHD, attention-deficit/hyperactivity disorder; ASD, autism spectrum disorder; CNV, copy number variation; ID, intellectual disability; LoF, loss-of-function; NDD, neurodevelopmental disorder.

CNVs

CNVs—genomic deletions or duplications exceeding one kilobase—represent a mechanistically complex and clinically actionable class of NDD-associated variation. Unlike single-nucleotide variants, CNVs frequently encompass multiple genes and regulatory elements, producing pleiotropic effects with highly variable expressivity and incomplete penetrance (12,13). Chromosomal microarray analysis has long served as a first-tier diagnostic test, yielding pathogenic findings in 10–20% of affected individuals; genome sequencing now enables nucleotide-resolution detection and has begun to resolve the structural complexity of recurrent rearrangement hotspots (13,60,67,68).

A subset of CNVs arises recurrently at segmental-duplication-mediated hotspots, facilitating genotype-phenotype correlation across large cohorts. The 16p11.2 locus exemplifies dosage sensitivity: deletion confers risk for ASD, ID, macrocephaly, and obesity, whereas duplication predisposes to schizophrenia, ADHD, microcephaly, and underweight (12,13,28). Neuroimaging endophenotypes—distinct alterations in cortical surface area, white matter microstructure, and functional connectivity—replicably differentiate deletion from duplication carriers (27,28). The 22q11.2 deletions, the most common genetic cause of schizophrenia, exhibits a remarkably variable phenotype encompassing palatal, cardiac, and endocrine anomalies alongside a 25–30% lifetime psychosis risk, with many carriers remaining asymptomatic (12,13). Other recurrent CNVs (e.g., 15q13.3, 1q21.1, 16p13.11) display analogous patterns of pleiotropy and reduced penetrance (12,68,69).

Population reference datasets such as gnomAD have transformed CNV interpretation. Systematic comparison of CNV frequencies in NDD cohorts versus tens of thousands of controls has refined penetrance estimates for 83 recurrent loci and clarified the pathogenic status of several previously uncertain CNVs (12,14). For instance, 15q24 duplications, 17q11.2 duplications, and distal 22q11.2 duplications now show statistically significant associations with NDDs, whereas 15q11.2 and proximal 2q13 duplications appear benign (12). NRXN1 deletions provide a paradigmatic illustration of locus-specific heterogeneity: deletions involving exons 6–24 are highly penetrant (32.4%) and strongly associated with ID and schizophrenia, whereas exon 1–5 deletions confer substantially lower risk (12.6%) and are observed in controls; intronic deletions do not elevate NDD risk (14,55).

Identifying the specific driver genes within large CNV intervals remains a fundamental challenge. Several complementary strategies have converged to address it. First, individuals with small, intragenic CNVs disrupting single genes within recurrent loci have implicated YTHDF3, EPHA7, DLG3, and IL1RAPL1 in NDD etiology (48,53,70,71). Second, integrative genomic analyses demonstrate that genes within NDD-associated CNV regions are significantly enriched for brain expression, evolutionary constraint, and interaction with known NDD genes (13,72). A recent meta-analysis of 11,614 NDD cases and 4,031 controls identified 41 NDD-risk CNV loci—including 24 novel regions—and prioritized 162 dosage-sensitive driver genes, 22 of which had not previously been linked to NDDs (13). Notably, long non-coding RNAs are significantly overrepresented in ASD-associated CNVs, implicating non-coding transcripts in disease pathogenesis (72).

Although many pathogenic CNVs arise de novo, a substantial proportion are inherited from mildly affected or apparently asymptomatic parents, a finding with profound implications for genetic counselling (48,55,73). Clinical expression is often modulated by additional genetic factors: NRXN1 deletion carriers with a second clinically relevant CNV exhibit more severe phenotypes, and this “second hit” burden is enriched in lower-penetrance 5' deletions (14,55). Common PRS for schizophrenia are over-transmitted to 22q11.2 deletion carriers who develop psychosis compared with those who remain unaffected, indicating that common variants modify penetrance even in disorders traditionally considered monogenic (7).

Emerging evidence implicates mechanisms beyond simple gene dosage. High-throughput chromatin conformation capture has revealed that CNVs can disrupt topologically associating domains, rewiring long-range enhancer-promoter interactions and altering expression of genes outside the CNV interval itself (74). This phenomenon, demonstrated for 4q31.3 CNVs affecting the FBXW7 regulatory landscape, expands the definition of pathogenicity and underscores the need to integrate three-dimensional genomic architecture into variant interpretation (74,75).

Clinically, the interpretation of rare, non-recurrent CNVs and susceptibility loci with incomplete penetrance remains challenging. The French AchroPuce network has proposed a dedicated “PIEV” classification for NDD-associated CNVs with reduced penetrance and variable expressivity, accompanied by national guidelines to harmonise reporting and counselling (67). Prenatal identification of susceptibility CNVs is particularly complex; the severity of ultrasound anomalies—not the CNV itself—primarily determines pregnancy continuation, emphasising the need for multidisciplinary counselling and long-term follow-up (73). Emerging technologies, notably long-read WGS, are poised to overcome historical limitations in detecting mosaic events, complex rearrangements, and small intragenic deletions (60).

In summary, CNVs constitute a well-established and clinically actionable component of NDD genetic architecture. Large-scale reference cohorts and meta-analyses have refined penetrance estimates and resolved the pathogenic status of numerous CNVs (12). Integration with functional genomic and three-dimensional chromatin data is progressively moving the field from locus-level associations to gene-level mechanistic understanding (13,72,74). Continued application of long-read sequencing, epigenetic profiling, and population-scale genotyping across diverse ancestries will be essential to fully capture CNV contributions to NDD etiology and to translate these discoveries into improved diagnostic precision, prognostic counselling, and therapeutic development. Table 3 provides a summary of the most common recurrent CNVs associated with NDDs, including their genomic loci, type, associated phenotypes, penetrance estimates, and proposed driver genes.

Table 3. Recurrent CNVs associated with NDDs.

CNV locus	Type	Associated phenotypes	Penetrance estimate	Key driver gene(s)
16p11.2	Del/Dup	ASD, ID, schizophrenia, obesity, macro/microcephaly	~25–30%	MVP, MAPK3
22q11.2	Del	Schizophrenia, ID, cardiac defects, palatal anomalies	~25–30%	TBX1, COMT
15q13.3	Del	ASD, ID, epilepsy, schizophrenia	~10–15%	CHRNA7
1q21.1	Del/Dup	ID, macro/microcephaly, ASD, congenital anomalies	~15–20%	GJA5, PRKAB2
NRXN1	Del (exonic 6–24)	ASD, ID, schizophrenia, developmental delay	~32% (3' exons)	NRXN1

This table presents recurrent CNVs that are robustly associated with NDDs. Loci were selected from large-scale cohort studies and meta-analyses. For each CNV, the genomic locus, type (Del or Dup), associated clinical phenotypes, estimated penetrance, and proposed driver gene(s) are provided. Penetrance estimates vary depending on cohort ascertainment and mutation size; values shown represent ranges from published meta-analyses. For NRXN1 deletions, penetrance is highest for deletions involving exons 6–24 (3' region), whereas 5' deletions and intronic deletions confer lower or no risk. ASD, autism spectrum disorder; CNV, copy number variant; Del, deletion; Dup, duplication; ID, intellectual disability; NDD, neurodevelopmental disorder.

Common variants and polygenic risk

If rare de novo mutations and CNVs define the high-penetrance end of neurodevelopmental risk, a substantial proportion of heritability is attributable to common variation. Twin studies indicate that approximately 30–50% of liability for autism, ADHD, and schizophrenia is explained by genetic factors beyond rare, highly penetrant lesions (6,7). Genome-wide association study (GWAS) findings show that thousands of common alleles with individually small effects collectively shape susceptibility at the population level (7,16). Large-scale GWAS meta-analyses have now robustly identified common variant contributions to ASD (76), ADHD (77), and schizophrenia (78), with hundreds of independent loci implicating overlapping biological pathways. The principal framework for quantifying this diffuse signal is the PRS, derived from GWAS summary statistics. For example, the schizophrenia PRS derived from Trubetskoy et al. [2022] explains approximately 7–10% of liability on the observed scale (78), while Grove et al. [2019] reported a similar heritability capture for ASD common variants (76). Although current PRS has limited utility for individual-level prediction, it is informative for characterizing shared liability, developmental continuity, and genetic confounding in environmental epidemiology (15,16,79).

A central example is the schizophrenia-cognition paradox. Schizophrenia is clinically associated with cognitive impairment, yet schizophrenia PRS shows a modest positive genetic correlation with educational attainment. Pleiotropic meta-analytic work suggests that this reflects biological heterogeneity: variants associated with adult synaptic pruning may drive the positive correlation, whereas early neurodevelopmental variants show the expected inverse association with cognition and educational outcomes (80). This supports a non-unitary genetic architecture. Unlike the direct prospective pediatric evidence available for ASD, ADHD, and ID, the evidence for schizophrenia discussed here is largely indirect, based on adult outcomes, postmortem studies, and retrospective data. Consistently, genomic structural equation modeling indicates that dyslexia, ADHD, and autism are better represented by at least two correlated latent dimensions—a core neurodevelopmental factor (autism/ADHD) and an attention-learning factor (ADHD/dyslexia) (81). Cross-disorder analyses further identify pleiotropic loci, including 17q21.31 (rs2696609), where opposite effects on schizophrenia and autism risk have been reported (4,5,82).

PRS also strengthens causal inference by exposing inherited confounding. Maternal ADHD PRS predicts both smoking during pregnancy and offspring ADHD diagnosis, challenging purely environmental interpretations (15). The same maternal liability associates with multiple prenatal exposures, while paternal valproate use is enriched among fathers with higher polygenic liability to epilepsy, ADHD, and autism, complicating attribution of direct teratogenic effects (83). Mendelian randomization provides complementary evidence: ADHD liability appears to increase risk of coronary artery disease, heart failure, and stroke, whereas atrial fibrillation may increase ADHD risk; autism liability has also shown effects on atrial fibrillation and heart failure (84). In contrast, support for causal links between autoimmune disease and NDDs remains limited despite observational comorbidity (85). Reported bidirectional associations between ADHD/autism and fractures warrant further mechanistic evaluation (86).

Important limitations remain. PRS research is predominantly Eurocentric, and cross-ancestry portability is reduced by differences in linkage disequilibrium structure and allele frequencies, with clear technical and equity implications (30,31,56). Even within European cohorts, variance explained by common variants in severe NDDs is modest (e.g., ~7.7% liability), but robust (7).

A growing body of evidence indicates that the effects of polygenic risk on neurodevelopmental outcomes are significantly modulated by the nature and timing of perinatal environmental exposures, with distinct patterns across disorder subtypes. For ASD, ASD PRS interacts with prenatal inflammation, with elevated genetic risk and maternal inflammatory markers jointly associated with more severe social communication deficits, whereas repetitive behaviors appear less influenced. For ADHD, ADHD PRS shows stronger associations with offspring ADHD in the context of prenatal tobacco exposure and maternal obesity, with some evidence of male-specific susceptibility, though replication remains inconsistent. For schizophrenia, the picture is more complex and rests on less direct evidence (PRS derived from adult GWAS applied retrospectively); several studies have reported interactions with hypoxic-ischemic events and preterm birth, but these findings require cautious interpretation given the indirect evidence base. For ID, the additive effect of multiple perinatal risk factors—preterm birth, low birth weight, maternal nutritional deficiency—appears to overwhelm compensatory mechanisms in individuals with high cognitive-impairment PRS, suggesting a threshold model.

Cross-disorder comparisons further illuminate the specificity of these interactions. The inflammation-phthalate interaction predicts both ASD and ADHD traits, but the effect on ASD is markedly stronger at higher phthalate levels. Similarly, prenatal smoking interacts with ADHD PRS to predict ADHD outcomes, while the same exposure shows weaker or inconsistent interaction with ASD PRS. These findings argue against a simple “general environmental sensitivity” model and instead support a framework in which genetic vulnerability and environmental exposures interact in a diagnosis-specific manner, shaped by developmental timing, sex, and the biological pathways linking exposure to core pathophysiology.

Notably, common and rare risks appear additive: polygenic background modifies penetrance and severity in carriers of high-impact variants, including 22q11.2 and NRXN1 deletions (7,55,80). Overall, neurodevelopmental genetics is best conceptualized as an integrated architecture in which rare variants provide high-effect mechanistic anchors and common variants shape penetrance, expressivity, and phenotypic boundaries.

Perinatal insults and gene-environment interplay

Prenatal exposures

The prenatal environment influences neurodevelopment is well established. Epidemiological studies have consistently associated gestational exposures—maternal infection, hypoxia, nutritional deficiency, endocrine-disrupting chemicals, anticonvulsants, alcohol, tobacco, and psychosocial stress—with elevated rates of autism, ADHD, ID, and schizophrenia (1,2,19). Among these, maternal immune activation (MIA) has emerged as a particularly well-studied exposure; rodent and nonhuman primate models have demonstrated that MIA disrupts fetal brain development via inflammatory cytokines, and human studies support associations with ASD and schizophrenia (87). Effect sizes are modest, however, and the majority of exposed children develop normally, whereas many affected children lack identifiable prenatal adversity. This discordance has shifted explanatory frameworks toward individual, genetically anchored variation in susceptibility.

Early attempts to incorporate genetics focused on candidate G×E, yet findings proved inconsistent. A systematic review of studies in children under six identified fourteen reports testing G×E effects on early psychopathology; twelve reported at least one nominally significant interaction, but replication was rare, and the specific genes and exposures varied widely (21). Serotonin transporter, monoamine oxidase A, and dopamine receptor polymorphisms featured prominently, moderating the effects of maternal stress, maltreatment, or prenatal smoking. The heterogeneity suggested that G×E relationships are likely context-dependent, developmentally timed, and poorly captured by single-variant tests.

The availability of genome-wide data in large pregnancy cohorts revealed a more fundamental challenge: genetic confounding. Many exposures long assumed to exert direct neurotoxic effects are themselves associated with the same polygenic risk that predisposes to the outcomes they are thought to cause. In the Avon Longitudinal Study of Parents and Children (ALSPAC), maternal PRS for ADHD were significantly associated with smoking during pregnancy, acetaminophen use in late gestation, lower blood mercury, and higher blood cadmium (15). Women with higher genetic liability for ADHD are thus more likely to engage in behaviours or encounter conditions classified as environmental risk factors. The association between prenatal paracetamol exposure and offspring ADHD, previously interpreted as evidence of endocrine disruption, became considerably less straightforward. Genetic confounding of similar magnitude has been demonstrated for paternal valproate use: fathers taking valproate for epilepsy have higher polygenic risk for epilepsy, ADHD, and autism than those on other antiseizure medications, and their children’s elevated NDD risk may reflect shared genetic susceptibility rather than direct spermatotoxic effects (83). These findings do not negate environmental causation but underscore the need for triangulation across designs that account for genetic nurture and assortative mating.

Genetic variation does more than confound exposure-outcome relationships; it also modifies their neurodevelopmental consequences. A pathway-based polygenic score for inflammation, derived from transcriptional activity of inflammatory response genes, was developed in the Barwon Infant Study and tested for interaction with prenatal phthalate levels (18). Children with high genetic susceptibility to inflammation exposed to higher concentrations of di(2-ethylhexyl) phthalate (DEHP) in utero exhibited markedly elevated risks of parent-reported ASD traits and doctor-diagnosed ASD by age 11.5 years. The attributable proportion due to interaction approached 40%, indicating synergistic rather than additive effects. This methodological innovation—anchoring a polygenic score in a specific molecular pathway—enhanced both power and interpretability.

Sex differences add further complexity. A secondary analysis of a magnesium sulphate trial for preterm birth prevention examined 80 polymorphisms in inflammation, coagulation, vasoregulation, excitotoxicity, and oxidative stress pathways among infants born before 32 weeks (17). An interleukin-6 polymorphism (rs2069840) was associated with adverse neurodevelopmental outcome (cerebral palsy, mental delay, or psychomotor delay) in females (odds ratio 2.6) but not males, and the sex-by-genotype interaction survived rigorous correction for multiple testing. No other gene-sex associations reached significance. Although replication is awaited, the finding aligns with evidence that male and female fetuses differ in inflammatory responses to intrauterine stress and that the genetic architecture of NDDs is partially sex-differentiated.

Mendelian randomisation offers a framework for addressing confounding and reverse causation, but its application to prenatal exposures is constrained by the scarcity of genetic instruments that specifically proxy modifiable environmental factors. To date, studies have largely focused on downstream consequences of NDDs—cardiovascular disease, fractures, autoimmune disorders—rather than upstream prenatal determinants (84-86). Nevertheless, the logic of using genetic variants as unconfounded proxies for exposure is gaining traction, and the development of instruments for maternal behaviours, metabolic traits, and environmental toxin levels is an active area of investigation.

The direction of causality is not always self-evident. Does prolonged infant screen time cause ADHD, or do early attentional difficulties—themselves genetically influenced—lead children to seek more screen exposure? Longitudinal data from a Japanese cohort supported the latter: ASD genetic risk was associated with longer screen time at 24–40 months, and ADHD genetic risk with an increase in screen time over the same period (88). Exposures measured in epidemiological studies are not randomly assigned; they are, in part, expressions of the child’s emerging phenotype. This insight reframes, rather than invalidates, the search for environmental causes. The perinatal environment is not a static backdrop for genetic programmes; it is dynamically coupled with a developing organism whose own genetic inheritance shapes its responses.

Progress will require several concurrent advances. Larger pregnancy cohorts integrating genome-wide data with deep, serial exposure phenotyping are needed to dissect genetic confounding from genuine interaction. Biomarker-based exposure assessment will reduce measurement error and enhance detection of effect modification. Developmental timing must be systematically addressed: the same genetic variant may confer resilience to an insult in the first trimester but vulnerability in the third, a hypothesis now testable using high-resolution transcriptomic atlases of human fetal brain (20,27). Finally, inclusion of diverse, non-European populations—in which allele frequencies, exposure distributions, and social contexts differ—is essential to distinguish universal mechanisms from context-bound associations. For now, the interplay between prenatal exposures and genetic susceptibility remains incompletely charted, but its contours are emerging as far more intricate than a simple catalogue of risk factors and protective alleles (Figure 3).

Figure 3.

Temporal interplay of perinatal insults, epigenetic remodeling, and neurodevelopment. This timeline from early gestation to infancy illustrates that the impact of genetic and environmental factors depends on developmental timing. Upper panel: neurodevelopmental events (neurogenesis, migration, synaptogenesis, myelination). Middle panel: windows of vulnerability to specific insults (infection, hypoxia, toxins) and peak expression windows of representative genes (SCN2A, EP400). Lower panel: potential intervention nodes (nutritional supplements, anti‑inflammatory agents, antisense oligonucleotides, therapeutic hypothermia). Timely interventions can mitigate risk.

Perinatal complications

Perinatal complications—preterm birth, hypoxic-ischemic encephalopathy (HIE), low birth weight, and neonatal seizures—are robust epidemiological predictors of adverse neurodevelopmental outcomes, including cerebral palsy, ID, and later psychiatric morbidity (1,2,17). Yet the relationship between insult and outcome is far from deterministic: most exposed infants escape major sequelae, whereas many affected children lack any identifiable perinatal adversity. This paradox points inexorably toward genetic modification of perinatal risk.

Preterm birth exemplifies these complexities. Extremely preterm delivery (<32 weeks) elevates risk for cognitive impairment and ASD five- to ten-fold, plausibly through vulnerability to haemorrhage, white matter injury, and excitotoxicity (17,19). Outcomes among infants of equivalent gestational age, however, vary strikingly. A secondary analysis of the BEAM trial identified a common interleukin-6 polymorphism (rs2069840) associated with a nearly threefold increased risk of death or neurodevelopmental impairment—but exclusively in females (17). This sex-specific effect survived rigorous correction and aligns with established sex differences in inflammatory responses to intrauterine stress, suggesting that genetic modulation of perinatal brain injury is both biologically plausible and measurably present years later.

HIE presents parallel questions. Despite therapeutic hypothermia, a substantial fraction of cooled infants develops cerebral palsy or cognitive impairment. Familial aggregation of poor outcome after HIE has long been noted; more recently, whole-exome sequencing of unexplained cases has identified pathogenic variants in genes previously associated with early-onset epileptic encephalopathies (e.g., KCNQ2, SCN2A, CDKL5) (11,51,63). These findings suggest that some instances of “perinatal asphyxia” reflect an underlying genetic predisposition that lowers the threshold for injury—a conceptual shift from HIE as exclusively acquired insult, with implications for counselling and prognostication.

Low birth weight and fetal growth restriction occupy a more ambiguous position. The association with later ADHD, autism, and cognitive deficits is well replicated, even among term infants (2,19). Historically attributed to nutritional deprivation or placental insufficiency, this correlation is now recognized to be partially confounded by pleiotropy: common genetic variants influencing intrauterine growth also confer liability to ADHD and schizophrenia (7). Observational studies therefore overestimate the causal neurodevelopmental impact of low birth weight per se. For schizophrenia specifically, these associations are derived primarily from retrospective cohort and case‑control studies; direct prospective evidence from birth to psychosis onset is more limited.

Neonatal seizures are both consequence of perinatal injury and independent predictors of adverse outcome. Among infants with acute symptomatic seizures secondary to HIE or stroke, a subset carries pathogenic variants in developmental and epileptic encephalopathy genes (11,51). Whether these represent alternative diagnoses, susceptibility alleles, or incidental findings is not always clear, yet the distinction carries therapeutic relevance: KCNQ2-related seizures may respond to sodium channel blockers, whereas phenobarbital—standard for HIE-associated seizures—is often ineffective. Genotype-guided management in the neonatal intensive care unit remains uncommon, but its rationale is strengthening.

The placenta has emerged as an underappreciated mediator. As the interface between maternal and fetal circulations, it regulates nutrient and oxygen delivery, modulates endocrine and immune signals, and is itself transcriptionally active. Placental expression of neurodevelopmental genes varies with fetal genotype and maternal exposures, and early evidence correlates this variation with offspring cognitive outcomes. One might speculate that the placenta transduces genetic risk and environmental insult into shared molecular signals—inflammatory cytokines, oxidative stress markers—that converge on vulnerable fetal neural circuits.

Causality remains contested. Does preterm birth cause neurodevelopmental impairment, or do shared genetic liabilities increase risk for both? Family-based designs comparing differentially exposed siblings suggest a partially causal relationship, attenuated but not eliminated after controlling for shared genetic and environmental background. Such within-family analyses counsel against a purely genetic interpretation while underscoring the need for triangulation across designs.

What emerges is not a simple catalogue of risk alleles but a dynamic, interactive picture. Perinatal complications are best conceptualized as stressors that interact with pre-existing genetic vulnerabilities—rare and common—to determine neurodevelopmental trajectories. The same hypoxic-ischemic insult well tolerated by most infants may produce substantial injury in a carrier of a deleterious SCN2A variant; the same degree of prematurity may yield widely divergent cognitive outcomes depending on polygenic background and placental inflammatory mediators. These interactions are likely sex-specific, developmentally timed, and modified by postnatal care. Their elucidation will require larger, deeply phenotyped cohorts, integration of maternal and fetal genotypes, and analytical methods capable of detecting effect modification across the allele frequency spectrum. The clinical imperative is to move from population-level estimates to individualized prognostication and, ultimately, to targeted neuroprotective strategies for infants genetically predisposed to perinatal brain injury.

G×E

Genes and environments do not operate in isolation has long been axiomatic in developmental psychopathology. Yet translating this axiom into empirically robust, mechanistically informative research has proven unexpectedly refractory. G×E—whereby genetic susceptibility moderates the neurodevelopmental consequences of environmental exposure, or environmental conditions modify the penetrance of genetic risk—has been simultaneously celebrated as the future of psychiatric genetics and dismissed as a graveyard of unreplicated candidate gene studies. The past decade has witnessed both a sober reassessment of early methodological shortcomings and the emergence of more rigorous, genome-wide approaches.

Conceptual foundations are frequently misunderstood. G×E is not merely the coexistence of genetic and environmental risk factors, nor is it equivalent to genetic confounding, in which genotype influences both exposure and outcome. True statistical interaction—departure from additivity—implies that the effect of environment differs across genotypes or vice versa. Biologically, this may reflect diverse mechanisms: toxin activation by a genetically variable enzyme, stressor dysregulation of a vulnerable neural circuit, or nutritional deficiency compounding impaired synaptic plasticity. The challenge is to specify these mechanisms with sufficient precision for testable translation into prevention or intervention.

The early wave of G×E research in NDDs was dominated by underpowered candidate gene studies, typically focused on monoamine system polymorphisms (5-HTTLPR, MAOA, DRD4, COMT) and their interaction with prenatal stress, maternal smoking, or obstetric complications (21). A systematic review of studies in children under six identified fourteen reports, twelve with nominally significant G×E effects, yet replication was rare, effect sizes modest, and publication bias likely. The lesson: hypothesis-driven G×E research requires either exceptionally strong biological priors or sample sizes an order of magnitude larger than those typically assembled.

A more fundamental limitation was the implicit assumption that genetic moderation operates through a small number of common variants of modest effect. The alternative—that G×E reflects the aggregate contribution of thousands of variants, each infinitesimally influencing environmental sensitivity—was not testable until recently. PRS have shifted this terrain. Rather than testing single polymorphisms, investigators can now ask whether individuals with high polygenic liability exhibit differential susceptibility to environmental adversity. This formulation is both more statistically powerful and biologically plausible, aligning with the continuous polygenic architecture of NDDs.

The inflammation-phthalate interaction exemplifies this approach (18). A pathway-based polygenic score for inflammation, derived from transcriptional responses enriched in brain tissue, was tested for interaction with prenatal phthalate exposure in the Barwon Infant Study. Children in the highest tertile of both genetic susceptibility and DEHP exposure had markedly elevated odds of ASD and ADHD outcomes, with a significant proportion of risk attributable to synergistic rather than additive effects. The score was anchored in biological pathway knowledge, not prior association signals, and tested prospectively in a population-based cohort—a template for future studies.

Developmental timing adds further complexity. The same genetic variant may confer vulnerability in the first trimester but resilience in the third; the same exposure may have radically different consequences during neuronal proliferation, migration, synaptogenesis, or myelination. Transcriptomic atlases reveal that NDD risk genes are not uniformly expressed across gestation but cluster in temporally and spatially restricted patterns corresponding to critical periods of circuit formation (20,24,27). G×E effects are likely similarly time-limited, necessitating prospective cohorts with repeated, high-resolution exposure measurement across gestation and infancy.

Sex differences constitute another recurring theme. The IL-6 polymorphism associated with adverse neurodevelopmental outcome in preterm females but not males—though awaiting replication—aligns with broader evidence of sex-differential genetic effects in autism and ADHD (17). Pooling male and female samples without testing for sex heterogeneity risks obscuring genuine G×E signals operating in only one sex.

Mendelian randomization offers a complementary approach to causal inference, using genetic variants as instrumental variables for exposures of interest. Applied to neurodevelopmental phenotypes, magnetic resonance (MR) has tested causal relationships between ADHD and cardiovascular disease, autism and fractures, and schizophrenia and cognitive ability (84-86). Its application to prenatal exposures is constrained by the scarcity of genetic instruments that specifically proxy modifiable environmental factors, yet development of maternal-specific instruments is an active area.

A persistent confound is the distinction between G×E and genetic nurture—the effect of parental genotype on offspring phenotype mediated through the rearing environment. Maternal genotype influences both exposure (smoking, medication use) and offspring genetic liability (via transmitted alleles), creating a mixture of direct genetic, indirect genetic, and environmental effects (15,79,89). Family-based designs—differentially exposed siblings, children of related mothers, or assisted reproductive technologies—are essential to disentangle these components. Maternal ADHD polygenic score predicts offspring ADHD risk even after accounting for transmitted alleles, indicating that genetic nurture contributes non-trivially to intergenerational transmission and that studies failing to control for it will overestimate both environmental causation and G×E (79,89).

Where the field now stands is a shift from underpowered candidate studies toward methodologically rigorous, genome-wide approaches. Polygenic scores, pathway-based scores, family designs, and Mendelian randomization are gradually replacing the search for single interacting polymorphisms. Sample size requirements for detecting interaction are now recognized to substantially exceed those for main effects, prompting large-scale consortia efforts. Developmental timing and sex differences are increasingly, if not yet systematically, incorporated.

The path forward requires integration across levels of analysis. Epidemiological cohorts with deep, repeated exposure phenotyping must couple with genome-wide genotyping and multi-omic profiling of placenta, cord blood, and neonatal tissue. Functional studies in human iPSC-derived neurons, cerebral organoids, and CRISPR-engineered rodents can test whether candidate G×E effects translate into altered gene expression, synaptic function, or circuit development under controlled conditions (64-66). Developmental neuroimaging can assess whether G×E effects on cognition and behavior are mediated by measurable alterations in brain structure and functional connectivity (27-29).

The goal is not simply to catalogue interactions but to understand the biological pathways through which genetic and environmental forces jointly shape the developing brain. That understanding may ultimately support stratified prevention strategies—identifying children at genetically conferred high risk for environmentally mediated neurodevelopmental impairment and targeting them for enhanced surveillance or early intervention. Whether this promise is realized depends on the field’s capacity to sustain the collaborative, interdisciplinary effort that the complexity of gene-environment interplay demands. Table 4 compiles representative G×E studies in perinatal NDDs, detailing the environmental exposure, genetic factor, outcome, interaction type, and study design, while distinguishing true statistical interaction from genetic confounding.

Table 4. Key gene-environment interactions (G×E) in perinatal NDDs.

Exposure	Genetic factor	Outcome	Interaction type	Study design
Prenatal phthalates (DEHP)	Inflammation pathway-based polygenic score	ASD, ADHD traits (age 11.5)	Synergistic (attributable proportion ~40%)	Prospective cohort (Barwon Infant Study)
Preterm birth (<32 weeks)	IL6 rs2069840 (female-specific)	Cerebral palsy, mental/psychomotor delay	Sex-specific interaction	RCT secondary analysis (BEAM trial)
Maternal smoking	Maternal ADHD polygenic risk score	ADHD in offspring	Genetic confounding (not true G×E)	Birth cohort (ALSPAC)
Paternal valproate use	Paternal epilepsy/ADHD/autism polygenic scores	NDD in offspring	Confounded by shared genetic liability	Norwegian Mother, Father, Child cohort
Prenatal stress/maltreatment	5-HTTLPR (SLC6A4), MAOA, DRD4	Early childhood psychopathology	Candidate G×E (rarely replicated)	Systematic review of 14 studies
Prenatal acetaminophen	Maternal ADHD polygenic risk	ADHD in offspring	Genetic confounding	Birth cohort (ALSPAC)
Prenatal infection/inflammation	ASD polygenic risk score	ASD social vs. repetitive behavior domains	Domain-specific interaction (social > repetitive)	Prospective cohort studies
Prenatal tobacco exposure	ADHD polygenic risk score	ADHD diagnosis	Gene-environment interaction (stronger in males)	Birth cohort (ALSPAC)
Obstetric complications/hypoxia	Schizophrenia polygenic risk score	Schizophrenia risk	Evidence suggestive but indirect; replication limited	Retrospective cohort studies
Multiple perinatal risk factors	Cognitive performance PRS	Intellectual disability	Additive/threshold model	Emerging evidence, awaiting replication

This table compiles representative studies investigating gene-environment interactions (G×E) or genetic confounding in the context of perinatal risk factors for NDDs. For each entry, the environmental exposure, genetic factor, observed outcome, interaction type (or confounding), and study design are provided. True statistical interaction (departure from additivity) is distinguished from genetic confounding, where the same genetic liability influences both exposure and outcome. The inflammation-phthalate interaction (first row) demonstrates a synergistic effect using a pathway-based polygenic score. The IL6 polymorphism showed sex-specific effects only in females. Candidate gene studies (5-HTTLPR, MAOA, etc.) have yielded inconsistent replication. Maternal and paternal polygenic scores have revealed substantial confounding in observational studies of smoking, acetaminophen, and valproate. ADHD, attention-deficit/hyperactivity disorder; ALSPAC, Avon Longitudinal Study of Parents and Children; ASD, autism spectrum disorder; DEHP, di(2-ethylhexyl) phthalate; NDD, neurodevelopmental disorder; RCT, randomized controlled trial.

The placenta as an active mediator of G×E interplay

The placenta, long viewed as a passive conduit for nutrient and oxygen transfer, is now recognized as an active endocrine, immune, metabolic, vascular, and epigenetic organ that integrates maternal and fetal signals and shapes neurodevelopmental trajectories (90,91). Situated at the maternal-fetal interface, the placenta responds to maternal infection, inflammation, hypoxia, nutritional status, and stress, while simultaneously expressing genes influenced by fetal genotype—making it a critical node for gene-environment interplay.

Placental dysfunction has been mechanistically linked to neuropsychiatric risk. Ursini et al. [2018] demonstrated that genetic risk for schizophrenia interacts with placental complications to increase disease susceptibility, with the placenta mediating the effects of hypoxia on fetal brain development (90). Vacher et al. [2021] showed that placental endocrine dysfunction—particularly involving neuroactive steroid synthesis—alters fetal brain gene expression and cortical maturation in a sex-dependent manner (91). Complementing these findings, Nugent et al. [2018] reported sex-specific epigenetic resilience in the placenta, with female placentas exhibiting distinct DNA methylation patterns that may contribute to the male predominance of certain NDDs (92).

Specific pathways through which the placenta mediates perinatal insults and transduces them into fetal neurodevelopmental risk are now being delineated. First, mTOR signaling integrates nutrient and growth factor signals to regulate fetoplacental growth and brain development; maternal undernutrition or overnutrition can dysregulate this pathway, altering fetal nutrient supply and neural progenitor proliferation. Second, hypothalamic-pituitary-adrenal axis dysregulation occurs when maternal stress elevates placental corticotropin-releasing hormone, altering fetal glucocorticoid exposure and programming neurodevelopmental trajectories through receptor-mediated effects on hippocampal and prefrontal development. Third, inflammatory cytokine cascades involving IL-6, TNF-α, and IL-1β can either cross the placenta directly or act locally within the placental bed to disrupt trophoblast function, impair fetal brain angiogenesis, and activate fetal neuroinflammatory responses. Fourth, oxidative stress pathways, triggered by hypoxia or metabolic dysfunction, impair placental barrier integrity, reduce nutrient and oxygen delivery, and generate reactive oxygen species that can directly damage fetal neural tissue. Fifth, epigenetic modifications at imprinted loci such as IGF2/H19 and MEST respond to environmental insults (maternal diet, toxins, stress) and correlate with offspring cognitive outcomes; these methylation changes can persist and program long-term gene expression patterns in the fetal brain. These pathways do not operate in isolation; rather, they converge to position the placenta as a critical transducer of G×E effects, converting maternal exposures and fetal genotypes into molecular signals that reach the developing brain.

Longitudinal placental omics studies are now beginning to link placental molecular profiles directly to childhood neurodevelopmental outcomes. A recent systematic review and meta-analysis by Guglielmo et al. [2025] found that placental inflammatory and metabolic gene expression signatures predict internalizing and externalizing behaviors in offspring, with effect sizes moderated by fetal sex and maternal prenatal stress (93). This emerging evidence reinforces the concept that placental dysfunction, whether driven by environmental insult, fetal genotype, or their interaction, disrupts fetal brain development through altered nutrient delivery, endocrine signaling, immune modulation, and epigenetic programming—establishing the placenta as a central integrator within the “perinatal insult → placenta → fetal brain” axis.

Despite these advances, key challenges remain: placental tissue is rarely collected in large birth cohorts; most studies are cross-sectional; and causal inference is limited by confounding. Prospective collection of placenta, cord blood, and neonatal dried blood spots, coupled with multi-omic profiling, will be essential to fully integrate placental biology into the G×E framework of NDDs.

Epigenetic mechanisms as molecular mediators

DNA methylation and histone modifications

If genetic variants establish the blueprint of neurodevelopment, epigenetic mechanisms execute its dynamic interpretation. DNA methylation, histone modifications, and chromatin conformation collectively govern spatiotemporal gene expression programmes that are both heritable and environmentally responsive—positioning epigenetics as a compelling molecular mediator of gene-environment interplay in NDDs. Technological advances now enable genome-wide epigenetic profiling in primary human tissues, including developing brain, catalyzing intense investigation over the past decade (22,23,94) (Figure 4).

Figure 4.

Epigenetic mechanisms as molecular mediators of gene-environment interplay. Perinatal insults (infection, phthalates, hypoxia) and genetic susceptibility (mQTLs, epivariations, imprinting) converge on the fetal epigenome. Center: three epigenetic layers—DNA methylation, histone modifications, noncoding RNAs—which regulate gene expression. Downstream: dysregulated gene expression leads to synaptic dysfunction and circuit miswiring, culminating in neurodevelopmental disorders. The placenta (icon) acts as a key interface. Solid arrows, direct effects; dashed arrows, modulatory influences. CpG, Cytosine‑phosphate‑Guanine; mQTL, methylation quantitative trait loci.

DNA methylation—covalent addition of a methyl group to cytosine within cytosine-phosphate-guanine (CpG) dinucleotides—remains the most extensively characterized epigenetic mark in NDDs. Its relative stability and detectability in peripheral tissues have enabled numerous case-control studies, although interpretational challenges abound: peripheral methylation may not mirror brain, confounding by cell composition and medication is pervasive, and statistical power is often limited (22,23). Despite these caveats, coherent principles have emerged.

Rare epigenetic variations (epivariations)—locus-specific methylation changes of magnitude comparable to imprinting disorders—occur at unexpectedly high frequency. A systematic screen identified 84 and 268 rare epivariations in autism and schizophrenia cohorts, respectively, absent from nearly 5,000 controls (23). Several features supported pathogenic relevance: enrichment at disorder-implicated genes, elevated brain expression of epivariation-affected genes in autism probands, and overrepresentation in affected versus unaffected siblings within autism families. These epivariations disrupted dosage-sensitive genes overlapping those targeted by rare CNVs and de novo mutations, suggesting convergent vulnerability. That epivariations arise without identifiable DNA sequence changes implicates stochastic errors in methylation establishment during gametogenesis or early embryogenesis.

Yet genetic variation and DNA methylation are far from independent. Methylation quantitative trait loci (mQTLs)—genetic variants influencing CpG methylation—are genome-wide abundant, and many NDD-associated risk variants exert effects partly through methylation alteration (36,80). The schizophrenia risk variant in SETD1A, encoding a histone methyltransferase, presumably alters H3K4 methylation landscapes, although downstream targets remain incompletely mapped (36,64). The 17q21.31 locus, implicated in opposing autism and schizophrenia risk, contains genes whose expression is regulated by both genetic variation and DNA methylation, with developmental stage- and brain region-specific directionality (27,82). Dissecting these causal cascades—from sequence variant to methylation change to gene expression to circuit dysfunction—remains formidable but increasingly tractable through multi-omic integration and Mendelian randomization.

Histone modifications constitute an even more diverse regulatory layer. Acetylation, methylation, phosphorylation, ubiquitination, and propionylation of histone tails modulate chromatin accessibility and recruit effector proteins. Unlike DNA methylation, histone marks are distributed across promoters, enhancers, gene bodies, and intergenic regions, with exquisite cell type specificity—a property that historically limited human studies reliant on bulk tissue (22,24). Single-cell epigenomic technologies are now beginning to surmount this barrier.

The BRPF1-KAT6 complex provides a paradigmatic illustration. BRPF1 recruits KAT6A/KAT6B acetyltransferases to catalyze histone H3 lysine 23 (H3K23) acetylation and, unexpectedly, propionylation (22). Patients with syndromic ID carrying BRPF1 mutations exhibit marked impairment of H3K23 propionylation, a defect rescuable in vitro by propionate, butyrate, or HDAC inhibitors valproate and vorinostat. This finding identifies a previously unrecognized modification—histone propionylation—as neurodevelopmentally relevant, demonstrates that a specific pathogenic variant operates through a discrete epigenetic mechanism, and suggests mutation-directed therapeutic strategies, however clinically uncertain (22).

Chromatin remodeling complexes extend the epigenetic repertoire beyond histone mark writing and erasure. Mutations in EP400, encoding an ATPase subunit of a SWR1-like remodeling complex, is associated with epilepsy and NDDs through compound heterozygous and de novo variants (25). EP400 exhibits striking developmental switching: predominant expression in inhibitory neurons during early corticogenesis shifts to excitatory neurons in mature brain. Drosophila orthologue knockdown increased seizure susceptibility; ep400-deficient zebrafish displayed brain abnormalities, epileptic discharges, and transcriptional dysregulation of 84 epilepsy/NDD-associated genes, including 11 highly dosage-sensitive loci. EP400 thus operates as a master regulator of transcriptional networks critical for both early neurogenesis and later circuit function, its dose-dependent disruption—by coding mutations or CNVs affecting its regulatory landscape—perturbing these networks (25,74).

A recurring theme is the environmental sensitivity of epigenetic machinery. Histone acetylation fluctuates with metabolic intermediates (acetyl-CoA, butyrate, propionate) influenced by nutrition and gut microbiota; DNA methylation at CpG islands is alterable by folate status; chromatin conformation responds dynamically to neuronal activity. The perinatal period—characterized by rapid brain growth, active epigenetic reprogramming, and extrauterine signal cascade—is thus a window of particular vulnerability. Genetic variants impairing epigenetic fidelity may render the developing brain less resilient to environmental perturbations that challenge epigenetic homeostasis; the phthalate-inflammation polygenic interaction described earlier may plausibly operate through such a mechanism (18).

Several challenges persist. Tissue specificity remains formidable: blood methylation correlates weakly with brain, and even within brain, patterns diverge markedly across regions and cell types (24,27). Single-cell epigenomic technologies are addressing this limitation but are not yet scalable to large cohorts. Temporal dynamics are understudied; most investigations capture single time points, yet epigenetic marks evolve across development and in response to experience. Prospective birth cohorts with serial biospecimen collection—dried blood spots, buccal swabs, cord blood, placenta—offer a path forward, though such resources remain scarce. Finally, distinguishing cause from consequence is rarely straightforward: epigenetic differences in postmortem NDD brain may reflect the disorder itself, its treatment, or agonal processes rather than developmental origins.

What emerges is a view of epigenetic mechanisms not as alternatives to genetic or environmental explanations but as their molecular integration point. Genetic variants alter the sequence or expression of epigenetic writers, erasers, and readers; environmental exposures modify substrate availability on which these enzymes depend; stochastic epigenetic errors produce phenotypes mimicking sequence mutations. Convergence of these diverse inputs onto shared epigenetic pathways—histone methylation, DNA methylation, chromatin remodeling—helps explain how heterogeneous aetiologies yield overlapping clinical syndromes. It also suggests that therapeutic strategies targeting epigenetic regulators may have efficacy across varied genetic and environmental contexts, a proposition now entering preclinical and early-phase clinical testing (22,64). Whether such interventions can achieve sufficient specificity and safety for the developing human brain remains an open question—one whose answer awaits deeper mechanistic understanding. Table 5 summarizes the major epigenetic mechanisms implicated in NDDs—DNA methylation, histone modifications, noncoding RNAs, and genomic imprinting—along with their molecular effects, associated phenotypes, and environmental sensitivity.

Table 5. Epigenetic mechanisms implicated in NDDs.

Mechanism	Example	Effect on gene expression	Associated NDD	Environmental sensitivity
DNA methylation	Rare epivariations at dosage-sensitive genes	Silencing or activation	ASD, Schizophrenia	Folate, toxins, maternal stress
Histone propionylation	BRPF1-KAT6 complex dysfunction	Reduced H3K23 propionylation	Syndromic ID	Propionate, HDAC inhibitors
Histone acetylation	EP400 mutations (chromatin remodeler)	Dysregulation of epilepsy/NDD genes	Epilepsy, ID	Metabolic intermediates, neuronal activity
Noncoding RNA (snRNA)	RNU4-2 mutations (minor spliceosome)	Global splicing defect of U12-type introns	Syndromic ID, hypotonia, seizures	None reported
Long noncoding RNA	Brain-enriched lncRNAs in ASD CNVs	Altered transcriptional regulation	ASD	Unknown
Genomic imprinting	15q11.2 maternal vs. paternal inheritance	Parental-origin-specific monoallelic expression	Angelman, Prader-Willi	Maternal intrauterine environment

This table summarizes major epigenetic mechanisms involved in the pathogenesis of NDDs. For each mechanism, a representative example is provided along with its molecular effect on gene expression, the associated NDD phenotype, and known or suspected environmental sensitivity. DNA methylation epivariations are rare locus-specific changes that can mimic sequence mutations. Histone modifications include propionylation (BRPF1-KAT6) and acetylation (EP400). Noncoding RNA mechanisms encompass the minor spliceosomal RNA RNU4-2 (mutations cause syndromic intellectual disability) and long noncoding RNAs enriched in ASD-associated CNVs. Genomic imprinting refers to parent-of-origin dependent gene expression, classically illustrated by 15q11.2 in Angelman and Prader-Willi syndromes. ASD, autism spectrum disorder; CNV, copy number variant; HDAC, histone deacetylase; ID, intellectual disability; NDD, neurodevelopmental disorder.

Noncoding RNAs and splicing

The noncoding genome and spliceosomal machinery govern the post-transcriptional fate of gene expression, yet their contribution to NDDs was, until recently, more speculated than demonstrated. Technical obstacles—poor capture by exome sequencing, lower evolutionary conservation, and intractable functional validation—long relegated noncoding variants to the periphery of genetic discovery. A series of discoveries over the past five years has decisively overturned this view, establishing noncoding RNAs and splicing factors as not merely contributors to, but in some cases dominant causes of, syndromic NDDs.

RNU4-2 is the paradigmatic example. Encoding U4atac, a small nuclear RNA component of the minor spliceosome, RNU4-2 was long ignored by exome-centric pipelines. In 2024, two studies identified recurrent de novo mutations—most commonly n.64_65insT—as among the most frequent causes of syndromic ID, accounting for ~0.5% of previously undiagnosed cases (30,57,58). Affected individuals share a recognizable phenotype: severe developmental delay, hypotonia, microcephaly, seizures, and distinctive facial dysmorphism. The mutation disrupts splicing of U12-type introns, present in ~700 genes enriched for DNA damage response, cell cycle regulation, and neurodevelopment. Rather than ablating a single transcript, RNU4-2 mutations impose a global but subtle perturbation of RNA processing—a mechanism fundamentally distinct from the complete LoF characterizing most monogenic disorders.

LSM1 provides convergent evidence. A recurrent homozygous splice-site variant (c.231+4A>C) may contribute to syndromic developmental delay with urological, cardiac, and skeletal anomalies in multiple affected individuals from Muslim Arab and Ashkenazi Jewish populations (59). Functional studies confirmed exon 3 skipping and near-complete loss of wild-type LSM1 expression. The mutation arises from a recurrent hotspot rather than a common founder, suggesting a vulnerable genomic context predisposing to mutation. The clinical overlap with RNU4-2-related disorder—global delay, hypotonia, dysmorphism—alongside distinctive features indicates that different perturbations of RNA metabolism produce syndrome-specific signatures.

Long noncoding RNAs (lncRNAs) represent a more heterogeneous class. Highly enriched in brain with exquisite developmental timing, most lncRNAs remain functionally unannotated (72). A systematic analysis of ASD-associated CNVs using SNATCNV revealed that brain-enriched lncRNAs are significantly overrepresented in recurrent ASD CNV loci; one locus contained a single lncRNA gene with no protein-coding neighbors, providing unusually direct evidence that noncoding transcripts alone can mediate CNV-associated neurodevelopmental risk (72).

RNA helicases and splicing factors bridge noncoding and coding disease genes. De novo missense mutations in DDX23—clustered in a conserved RecA-like domain—cause syndromic NDD with tone abnormalities, global delay, ASD, seizures, and variable multisystem involvement (95). Mutations in HNRNPU, encoding a heterogeneous nuclear ribonucleoprotein, produce a recognizable syndrome of moderate to severe ID, early-onset febrile seizures, and distinctive dysmorphism (96). That trans-acting splicing factors yield pleiotropic phenotypes overlapping those caused by mutations in the noncoding RNAs they bind suggests convergence on shared RNA targets and pathogenic pathways.

Distinguishing mechanism matters for therapeutic strategy. RNU4-2 mutations disrupt RNA-RNA interactions rather than simply reducing RNA abundance (30,57); LSM1-related disorder requires near-complete loss of function below a critical threshold, with asymptomatic heterozygotes (59); DDX23 and HNRNPU mutations favor dominant-negative or gain-of-function effects (95,96). LoF disorders may be amenable to gene replacement or allele reactivation; gain-of-function disorders may require allele-specific silencing.

Genomic imprinting and maternal effects

Genomic imprinting—an epigenetic mechanism restricting gene expression to a single parental allele—represents a striking violation of Mendelian expectation. Long confined to a handful of rare syndromes (Angelman, Prader-Willi, Beckwith-Wiedemann), imprinting has historically occupied a specialist niche in neurodevelopmental research. This position is no longer tenable. Convergent evidence from population genetics, molecular epigenetics, and family-based designs now places parent-of-origin effects and, more broadly, maternal genetic influence at the center of gene-environment interplay in NDDs.

Imprinted genes are overrepresented in brain and enriched for neurodevelopmental pathways—neuronal differentiation, synaptic plasticity, energy metabolism. Their dysregulation produces characteristic phenotypes, and the clinical boundaries of imprinting disorders have expanded considerably with genome-wide methylation arrays (9,97,98). Beyond these classic syndromes, systematic analysis of CNVs in NDD cohorts reveals significant enrichment at imprinted loci, with phenotypic consequences critically dependent on parental origin (12,13). The 15q11.2 deletion exemplifies this: maternal inheritance causes Angelman syndrome, paternal inheritance Prader-Willi syndrome. Subtler parent-of-origin effects are documented at 15q13.3, 22q11.2, and 16p11.2, yet parental provenance—essential for accurate CNV interpretation—is frequently unavailable in routine testing.

A distinct but conceptually related body of work concerns maternal genetic nurture: the influence of maternal genotype on offspring phenotype through mechanisms other than transmitted alleles. These may be environmental (intrauterine milieu, perinatal care, rearing environment) or involve oocyte-deposited factors. A landmark Swedish register study partitioned ADHD liability among nearly one million individuals, estimating that maternal genetic nurture accounts for approximately 14% of phenotypic variance—one-fifth the magnitude of direct genetic effects (89). The design exploited differential relatedness of maternal versus paternal half-siblings; higher concordance in maternal half-siblings provided evidence for a nurture effect independent of transmission.

Complementary evidence from the ALSPAC demonstrated that maternal ADHD polygenic score predicts offspring ADHD risk even after conditioning on the offspring’s own transmitted score—a finding inexplicable by direct genetic transmission alone (15,79). This association was partially attenuated, not eliminated, by adjustment for smoking, breastfeeding, and postnatal depression, suggesting multiple pathways for genetic nurture.

Paternal effects have received less attention. A recent Norwegian Mother, Father, and Child Cohort Study analysis found that fathers taking valproate for epilepsy have significantly higher polygenic risk for epilepsy, ADHD, and autism than fathers on other antiseizure medications (83). The association between paternal valproate use and offspring NDD risk—previously attributed to spermatotoxicity—was substantially confounded by this genetic overlap, illustrating that observational studies cannot, on their own, resolve questions of environmental causation.

The placenta represents a critical but understudied mediator. A fetal organ bathed in maternal blood, its transcriptome is influenced by both fetal and maternal genotypes and expresses abundant imprinted genes regulating nutrient transfer, endocrine signaling, and immune modulation. Placental dysfunction is a well-established risk factor for preterm birth, fetal growth restriction, and HIE—each associated with elevated NDD risk (17,19). Maternal genotypes that subtly impair placental function may thus increase NDD risk indirectly, through perinatal complications, rather than through direct effects on fetal brain development.

Sex differences further complicate the landscape. Several imprinted genes show sexually dimorphic brain expression, and phenotypic consequences of their disruption often differ between males and females. The IL-6 polymorphism associated with adverse neurodevelopmental outcome in preterm females but not males may exemplify a sex-specific maternal effect, though the polymorphism was genotyped in offspring, not mothers (17). A recent microcephaly analysis found greater severity in females, partially attributable to parental origin of X chromosome alleles (99).

Methodological challenges abound. Family-based designs capable of separating direct genetic effects from genetic nurture require large genotyped parent-offspring trios—scarce resources, particularly in non-European populations. Detection of parent-of-origin effects demands knowledge of allelic phase and parental provenance, not routinely available from standard genotyping arrays. Assortative mating—individuals with high NDD polygenic risk partnering together—correlates non-transmitted maternal alleles with transmitted paternal alleles, potentially producing spurious evidence of genetic nurture unless appropriately modelled (89).

The maternal genome thus influences offspring neurodevelopment through at least three distinct but interacting channels: direct transmission of risk alleles, epigenetic regulation of imprinted genes by parental origin, and construction of the intrauterine and rearing environment—itself partially genetically determined. These channels are not mutually exclusive; they are likely correlated and potentially synergistic. Disentangling them is not merely academic: interventions targeting the intrauterine environment may reduce NDD risk even if exposure-outcome associations are partially confounded, provided the intervention genuinely modifies environmental exposure. Optimal targeting and realistic impact estimation require quantitative understanding of the relative contributions of direct genetic, epigenetic, and environmental pathways.

Genomic imprinting, maternal genetic nurture, and perinatal environmental epidemiology converge on a fundamental insight: the maternal genome is not merely a vehicle for allelic transmission but an active constructor of the environment in which the next generation develops. The fetal genome, reciprocally, influences maternal physiology through placental signaling. This dynamic, bidirectional, developmentally regulated system defies reduction to simple additive risk models. Its elucidation demands sustained interdisciplinary collaboration and a willingness to abandon orthodox divisions between genetic and environmental causation, between inherited and acquired, between nature and nurture.

Biomarkers and early warning: toward early detection in the perinatal window

The perinatal period offers a unique window for early NDD detection and intervention, as the brain retains considerable plasticity before irreversible circuit dysfunction becomes entrenched. While no single biomarker has yet achieved clinical implementation, converging evidence from epigenetic, imaging, and electrophysiological modalities has yielded promising candidates.

Epigenetic biomarkers—DNA methylation patterns in cord blood, placenta, and neonatal dried blood spots—have been most extensively investigated due to their stability in accessible tissues and responsiveness to environmental exposures. Methylation at MECP2, BDNF, and SLC6A4 promoters has been linked to altered cognitive, social, or behavioral development in longitudinal cohorts, though replication remains inconsistent. Histone modification patterns and circulating microRNAs have shown preliminary associations with ASD risk but require validation in larger, diverse cohorts before clinical application.

Imaging biomarkers offer a distinct advantage over epigenetic approaches by directly visualizing developing brain structure and function. Neonatal magnetic resonance imaging (MRI) can quantify cortical thickness, white matter microstructure via diffusion tensor imaging, and cerebellar/subcortical volumes—all of which show group-level differences between infants who later develop NDDs and controls. Resting-state functional connectivity has been associated with later cognitive outcomes in preterm infants. However, high cost, the need for sedation, and limited accessibility currently restrict neonatal MRI to research settings and selected high-risk populations.

Electrophysiological biomarkers provide a more accessible and cost-effective bedside alternative. Electroencephalography (EEG) measures—spectral power, interhemispheric coherence, sleep spindle characteristics, and abnormal background activity—have shown predictive value for cognitive impairment and cerebral palsy in preterm and term infants. Automated EEG analysis pipelines are under development to standardize interpretation and enable real-time risk stratification. Yet despite their practical advantages, sensitivity and specificity remain suboptimal for individual-level prediction.

The limitations of each modality when used in isolation highlight the promise of multi-modal integration. A hypothetical algorithm combining neonatal DNA methylation signatures, resting-state EEG measures, and diffusion tensor imaging parameters could generate a composite risk score with improved predictive accuracy. Proof-of-concept studies are emerging, but substantial challenges remain: protocol standardization, validation across diverse populations, establishment of normative reference data, and demonstration that biomarker-guided early intervention improves outcomes in randomized trials.

Based on this evidence, a translational pathway from biomarker discovery to clinical implementation can be envisioned: (I) validate candidate biomarkers in prospective birth cohorts; (II) develop cost-effective, scalable assays—such as dried blood spot epigenetic screening and automated EEG analysis—for population-level application; (III) design screening algorithms to identify the top 5–10% of newborns at highest risk; and (IV) rigorously test biomarker-guided early intervention in randomized controlled trials (RCTs) with long-term follow-up. Realization of this pathway requires sustained investment, interdisciplinary collaboration, and—most critically—inclusion of diverse populations to ensure generalizability and avoid perpetuating existing health disparities.

From genes to neural circuits

Synaptic and ion channel dysfunction

Synaptic and ion channel dysfunction constitutes one of the most mechanistically transparent classes of NDDs. The past two decades have witnessed the convergence of human genetics, electrophysiology, and structural biology onto a discrete set of synaptic proteins and ion channels whose perturbation—by rare de novo mutations, inherited variants, or copy number alterations—recurrently underpins ASD, ID, epilepsy, and schizophrenia (11,33,42). The emerging picture is not one of indiscriminate disruption but of precise, often allele-specific, alterations in channel biophysics and synaptic signaling that produce distinct and sometimes opposing physiological consequences.

Ion channelopathies exemplify this precision. Voltage-gated sodium, potassium, and calcium channels govern action potential initiation, propagation, and neurotransmitter release; their dysfunction predictably alters neuronal excitability. SCN2A illustrates the complexity arising when the same gene harbors mutations with opposing functional effects: LoF variants reduce sodium current and are strongly associated with autism and ID without epilepsy, whereas gain-of-function missense mutations impair channel inactivation and produce early-infantile epileptic encephalopathy (11,40). SCN1A, preferentially expressed in inhibitory interneurons, presents complementary logic: LoF preferentially impairs interneuron excitability, disinhibiting excitatory circuits and causing Dravet syndrome, whereas gain-of-function mutations are associated with familial hemiplegic migraine (11,63).

Calcium channelopathies are dominated by gain-of-function mechanisms. Rare missense variants in CACNA1I, encoding Cav3.3 T-type channels, slow activation and inactivation kinetics, hyperpolarise voltage dependence, and increase window current—enhancing calcium influx and lowering burst threshold in thalamic reticular nucleus neurons (37). The magnitude of gain-of-function predicts clinical severity: the severe p.(Ile860Asn) variant produces pronounced biophysical alterations and developmental encephalopathy with seizures; the milder p.(Ile860Met) variant segregates with borderline intellectual functioning (37).

AMPA-type glutamate receptors, encoded by GRIA1–4, mediate fast excitatory neurotransmission. Recurrent gain-of-function mutations at paralogous residues—GRIA2 p. Gly792 and GRIA3 p. Gly803—slow channel deactivation, prolong excitatory postsynaptic currents, and enhance calcium permeability, shifting excitation-inhibition balance toward pathological hyperexcitability (39,40). The recurrent GRIA1 p.(Ala636Thr) variant produces remarkably broad phenotypic spectra—even among carriers of identical amino acid substitutions—strongly implicating modifier genes, environmental factors, or stochastic processes in disease expression (40).

Potassium channels regulate repolarization and firing frequency. KCNB1, encoding Kv2.1, was initially implicated through de novo missense mutations clustered in voltage sensor and pore domains; expansion to include inherited truncating variants revealed that C-terminal truncations confer milder epileptic phenotypes than pore-directed missense mutations, establishing a gradient of physiological severity paralleling clinical expression (38).

Beyond ion pores, synaptic scaffolding, adhesion, and vesicular cycling machinery are recurrently perturbed. SHANK3, encoding a master postsynaptic scaffold, is among the most robustly validated autism risk genes; its disruption produces Phelan-McDermid syndrome (54,100). NRXN1 deletions illustrate incomplete penetrance: 3' exonic deletions confer substantial risk, whereas 5' exonic and intronic deletions require additional genetic or environmental hits for phenotypic expression (14,55). STXBP1, regulating syntaxin-mediated vesicle fusion, is among the most frequently mutated genes in developmental and epileptic encephalopathies (11,63). SCAMP5, identified through unrelated probands sharing an identical de novo missense mutation, causes autism, ID, and seizures through dominant-negative effects (41).

Binary classifications—LoF versus gain-of-function, haploinsufficiency versus dominant-negative—inadequately capture physiological reality. The EEF1A2 D252H mutation exhibits biochemical LoF yet produces more severe phenotypes in homozygous mice than null homozygotes, indicating toxic gain-of-function (65). Conversely, some truncating mutations expected to cause simple haploinsufficiency exert dominant-negative effects through nonsense-mediated decay escape and production of interfering C-terminally truncated proteins.

Synaptic and ion channel dysfunction is best conceptualized not as discrete pathological categories but as a continuum of physiological impairment whose severity, anatomical distribution, and developmental timing determine clinical syndrome. Mutations profoundly reducing sodium current in inhibitory interneurons produce catastrophic epilepsy; mutations modestly enhancing T-type calcium channel activity in thalamocortical circuits produce absence seizures; mutations slowing AMPA receptor deactivation in forebrain pyramidal neurons produce ID with variable epilepsy. The same gene, mutated differently, can produce autism without seizures, epilepsy without autism, or both.

This genetic and phenotypic heterogeneity frustrates diagnosis but illuminates biology: the brain’s computational algorithms are continuously tunable, and even small molecular perturbations can, over developmental time, cascade into profound circuit dysfunction. The therapeutic corollary is that restoring physiological function—not merely suppressing seizures or modifying behavior—requires exquisitely precise interventions tailored to specific molecular lesions. Whether such precision is achievable at scale remains uncertain, but mechanistic clarity achieved for several channelopathies and synaptopathies provides reason for cautious optimism. Nevertheless, it is important to emphasize that most of these precision interventions are still in preclinical or early-phase clinical testing; only a handful (e.g., ketogenic diet for GLUT1 deficiency, sodium channel blocker avoidance in Dravet syndrome) are currently clinically actionable. Table 6 lists representative genes encoding ion channels and synaptic proteins that are recurrently mutated in NDDs, highlighting how different functional changes (loss‑of‑function vs. gain‑of‑function) in the same gene can produce distinct clinical phenotypes.

Table 6. Representative genes disrupting synaptic and ion channel function in neurodevelopmental disorders.

Gene	Protein	Functional change	Clinical phenotype
SCN2A	Sodium channel Nav1.2	LoF → ASD + ID; GoF → epilepsy	Seizure-free ASD vs. severe infantile epilepsy
CACNA1I	T-type calcium channel Cav3.3	GoF (increased window current)	Developmental encephalopathy, epilepsy, ID
GRIA2/GRIA3	AMPA receptor	GoF (slowed deactivation)	ASD, ID, epilepsy
SHANK3	Postsynaptic scaffold	Haploinsufficiency	Phelan-McDermid syndrome (ASD + ID)
NRXN1	Presynaptic adhesion molecule	Exon 6–24 deletion (high penetrance)	Schizophrenia, ID, ASD

For each gene, the encoded protein, the predominant functional change LoF or GoF, and the associated clinical phenotype are listed. The table illustrates that different mutations in the same gene (e.g., SCN2A) can produce strikingly different syndromes, ranging from autism without epilepsy to severe infantile epileptic encephalopathy. ASD, autism spectrum disorder; GoF, gain-of-function; ID, intellectual disability; LoF, loss-of-function.

Cortical development and circuit formation

Cortical development constitutes the architectural foundation upon which neural circuits are built—and its disruption attacks that architecture directly. The human cerebral cortex, a laminated mosaic of billions of neurons diversified into hundreds of subtypes arrayed across cytoarchitectonic areas, emerges through precisely orchestrated sequences of progenitor proliferation, neuronal migration, laminar specification, axon guidance, dendritic arborisation, synaptogenesis, and activity-dependent refinement. Each step is vulnerable to genetic perturbation, and convergent evidence now implicates discrete stages of cortical neurogenesis as nodal points of vulnerability across diverse NDD aetiologies.

Neuronal migration is among the best-characterized phases. TUBG1, encoding γ-tubulin essential for microtubule organization, causes microcephaly, pachygyria, and epilepsy when mutated; patient-derived models reveal impaired migration and mitotic defects (51,101). The phenotypic spectrum, however, extends to microcephaly and ID without overt malformations, indicating underrecognised milder presentations (51). ARF3, encoding an ADP-ribosylation factor, provides an even rarer but highly instructive example: the recurrent p.(Asp67Val) variant produces distinctive Z-shaped brainstem morphology, cerebellar fusion, and white matter reduction—a constellation aligning with Golgipathies and underscoring vesicular trafficking in neuronal migration (102).

Axon guidance and connectivity constitute a second major node. EPHA7, an ephrin receptor expressed in developing cortex, was implicated through twelve patients with deletions—four involving only EPHA7—who exhibited developmental delay, ID, and variable dysmorphism (48). Inheritance from mildly affected parents in nine cases and incomplete penetrance suggest EPHA7 haploinsufficiency confers risk rather than deterministically causing disease. DCC, encoding the netrin-1 receptor, shows haplotype-based association with ASD susceptibility in a Chinese Han cohort; biological plausibility is strong, yet genetic evidence remains circumstantial (50).

Transcriptional regulation of cortical development has emerged as a convergent pathway of exceptional importance. FEZF2, required for corticospinal motor neuron specification, was implicated through seven heterozygous deleterious variants in unrelated individuals with developmental delay, ID, autism, and ADHD (43). Five arose de novo; one was paternally inherited from an affected father. ZNF865, a poorly characterized poly-zinc finger protein, was identified through de novo truncating variants in eighteen unrelated patients sharing global developmental delay, hypotonia, and dysmorphic features—four variants recurred across families, and all but one clustered C-terminally (103). The rapid assembly of this cohort attests to international data-sharing power.

Chromatin remodeling intersects cortical development at multiple levels. SUPT16H, encoding a FACT complex subunit, was implicated through de novo missense variants and a deletion in five patients with neurodevelopmental deficits and corpus callosum anomalies; all three with available imaging showed callosal thinning or hypoplasia, suggesting specific requirement for callosal axon development (101). HNRNPU haploinsufficiency causes a recognizable syndrome of moderate to severe ID, early-onset febrile seizures, and distinctive dysmorphism, though its critical RNA targets remain incompletely characterized (96).

Developmental timing is increasingly resolved at single-cell resolution. Single-nucleus RNA sequencing of postmortem ASD brain reveals cell type-specific transcriptomic changes in interhemispheric and callosal-projecting neurons, superficial lamina interneurons, and distinct glial reactive states (24). Autism-associated gene regulatory network drivers and their targets are enriched in both rare and common genetic risk variants, connecting genetic susceptibility to cellular and circuit alterations. Human-evolved regulatory elements—sequence-based and epigenetic—converge on cell types and laminae involved in cerebral cortical expansion and interact with NDD risk genes, suggesting that vulnerability reflects inherent fragility of recently evolved, highly specialized neuronal populations (104).

Translating cortical malformations to circuit-level dysfunction remains challenging. Corpus callosum agenesis is consistently associated with ID and autism, yet many individuals with isolated callosal agenesis exhibit normal intelligence. The distinction likely resides in whether the axon guidance defect is confined to callosal projections or reflects broader commissural and associative connectivity disruption. Diffusion tensor imaging in 16p11.2 deletion and duplication carriers reveals dose-dependent white matter microstructural alterations correlating with cognitive performance, providing tractable intermediate phenotypes linking genotype to circuit function (28,29). Integrating such imaging endophenotypes into genetic association studies—imaging genomics—offers a pathway from static lesion to dynamic network. Table 7 provides examples of genetic mutations linked to specific neuroimaging findings, circuit‑level dysfunction, and behavioral outcomes, bridging the gap between genotype and neural circuit pathophysiology.

Table 7. Neural circuit-level endophenotypes in genetic NDDs.

Genetic mutation	Neuroimaging finding	Circuit dysfunction	Behavioral/phenotypic outcome
16p11.2 Del/Dup	Altered cortical thickness, white matter microstructure (diffusion tensor imaging changes)	Excitation/inhibition imbalance, altered functional connectivity	ASD, ID, macro/microcephaly
SCN2A loss-of-function	Reduced functional connectivity (default mode network)	Decreased sodium current → reduced neuronal excitability	ASD, ID without epilepsy
SCN2A gain-of-function	Regional hyperexcitability (fMRI/EEG)	Impaired inactivation → persistent sodium current → hypersynchrony	Early-infantile epileptic encephalopathy
CACNA1I gain-of-function	Thalamic hyperconnectivity, altered sleep spindles	Increased T-type calcium current → lower burst threshold in thalamic reticular nucleus	Absence seizures, ASD, sleep disturbances
22q11.2 Del	Cortical thinning (prefrontal, temporal), reduced gyrification	Dopamine dysregulation, altered prefrontal-thalamic connectivity	Schizophrenia, ID, psychosis (25–30% lifetime risk)
NRXN1 deletion (3' exons)	Reduced gray matter volume, altered white matter integrity	Synaptic adhesion defect → impaired synaptic transmission	ASD, ID, schizophrenia (variable penetrance)
MAPK8IP3 (missense)	Cerebellar and thalamic structural alterations	JNK signaling disruption → impaired neuronal migration and synaptic plasticity	Global developmental delay, ID, ASD, movement abnormalities
ADGRB3 (BAI3) biallelic duplication	Cerebellar atrophy, reduced cerebello-frontal connectivity	Synapse formation defect → disrupted cerebellar-frontal circuit development	ID, ataxia, behavioral disorder, cognitive impairment
ARF3, WDR45B, PPP1R21-related brainstem pathology	Brainstem structural abnormalities (Z-shaped morphology, hypoplasia), ventriculomegaly	Brainstem arousal system dysfunction → altered cortical maturation via experience-dependent plasticity	Global developmental delay, seizures, autonomic dysfunction, sleep disturbances

This table links specific genetic mutations associated with NDDs to intermediate phenotypes at the neural circuit level, as assessed by neuroimaging (structural/fMRI, diffusion tensor imaging, EEG) and electrophysiology. For each genetic lesion, the table provides: the observed neuroimaging or electrophysiological endophenotype, the proposed circuit-level dysfunction (e.g., excitation/inhibition imbalance, altered connectivity, channel biophysics), and the resulting behavioral or clinical outcome. These endophenotypes serve as bridges between genotype and clinical manifestation, and they are increasingly used for patient stratification and as surrogate endpoints in clinical trials. ASD, autism spectrum disorder; EEG, electroencephalography; fMRI, functional magnetic resonance imaging; ID, intellectual disability; NDD, neurodevelopmental disorder.

Cerebellar and subcortical involvement

The cerebellum has, for much of modern neuroscience, labored under the presumption of exclusive motor function. That view has collapsed. Convergent evidence from neuroimaging, genetics, and clinical observation now recognizes the cerebellum as a structure of extraordinary functional heterogeneity, reciprocally connected with prefrontal, parietal, and limbic cortices; its disruption contributes to cognitive, affective, and social impairments as surely as to ataxia and dysmetria. Subcortical structures—basal ganglia, thalamus, brainstem—have undergone parallel reappraisal, though their integration into neurodevelopmental models remains less complete.

Cerebellar developmental disorders exhibit characteristic but variable phenotypes. ATP2B2, encoding a Purkinje-cell-enriched calcium ATPase, was recently implicated through heterozygous de novo missense variants and a dominantly inherited frameshift allele (105). Affected individuals share developmental delay, cognitive impairment, epilepsy, autistic traits, and motor disorders; one patient exhibited striking cerebellar atrophy. A recurrent p.(Glu457Lys) substitution, previously documented in a neurologically impaired mouse mutant, confirmed cross-species conservation. ADGRB3 (BAI3), encoding an adhesion G protein-coupled receptor (GPCR) involved in synapse formation and cerebellar development, was implicated through a biallelic intragenic duplication segregating in a consanguineous family with two sons affected by ID, ataxia, and behavioral disorder—extending the phenotypic spectrum of this previously psychiatric-associated locus to include cognitive and motor dysfunction (106).

The thalamus remains understudied, yet its position as principal relay between subcortex and cortex marks it as a nodal vulnerability. CACNA1I, with its gain-of-function effects on T-type calcium channels, is strongly expressed in thalamic reticular nucleus neurons; computational modelling predicts lowered burst threshold and increased action potential duration, providing a mechanistic link to absence seizures and the sleep disturbances commonly comorbid with NDDs (37). TUBG1 mutations, beyond their cortical effects, produce thalamic hypoplasia in some patients, though functional consequences await systematic investigation (51,101).

The brainstem, even more neglected, has begun to attract attention through rare but highly informative phenotypes. The Z-shaped brainstem morphology associated with ARF3 mutations is sufficiently distinctive to warrant inclusion in the differential diagnosis of pontocerebellar hypoplasias (102). WDR45B, encoding an autophagy-implicated WD40-repeat protein, was identified through homozygous truncating variants in six individuals from three unrelated families, all presenting with profound developmental delay, early-onset refractory epilepsy, progressive spastic quadriplegia, and contractures; neuroimaging revealed ventriculomegaly, reduced cerebral white matter, and cortical thinning (107). PPP1R21, involved in endosomal maturation, was associated through biallelic pathogenic variants in eleven additional individuals with syndromic NDD characterized by global developmental delay, hypotonia, coarse facies, and consistent patterns of delayed myelination and lateral ventricular abnormalities (108).

Motor system involvement is the rule, not the exception. SCYL2, encoding a pseudokinase that suppresses excitotoxicity, was initially associated with arthrogryposis multiplex congenita-4 through biallelic LoF variants (109). The subsequent identification of individuals homozygous for missense variants in the protein kinase domain who presented with developmental delay alone extended the phenotypic spectrum along a severity continuum. MAPK8IP3, encoding a JNK-signalling scaffold, was studied in thirty-two individuals with heterozygous pathogenic variants; missense mutations were associated with more severe symptoms than LoF alleles, suggesting dominant-negative or gain-of-function mechanisms (49).

What unites these disparate observations is the recognition that NDDs are not, for the most part, disorders of cortex alone. They are disorders of the entire brain—cortex, cerebellum, thalamus, brainstem, basal ganglia—and their manifestations reflect the particular vulnerabilities of each structure at specific developmental stages. The cerebellum is especially vulnerable during its protracted postnatal development, a temporal window overlapping critical periods for cognitive and social development, plausibly explaining its contribution to autism and schizophrenia (3,19,20). The thalamic reticular nucleus, with its dense T-type calcium channel expression and strategic position controlling thalamocortical oscillations, is a natural locus for epilepsy genes. The brainstem, site of autonomic regulation and arousal, may contribute to sleep disturbances, sensory processing abnormalities, and sudden unexpected death in epilepsy that disproportionately affect NDD populations.

The clinical implications are straightforward yet incompletely implemented. Neuroimaging in NDDs should include dedicated evaluation of cerebellar and brainstem morphology, not merely exclusion of cortical malformations. Movement disorders—dystonia, ataxia, chorea, stereotypies—should be systematically assessed and documented, not dismissed as behavioral. Hearing and vision should be formally evaluated, given the high prevalence of sensorineural impairment in disorders affecting cerebellar and brainstem development. The convergence of cerebellar, thalamic, and brainstem pathology in genetically defined neurodevelopmental syndromes is not merely academic; it is a guide to clinical care.

The recognition that NDDs affect distributed neural systems rather than isolated brain regions has profound implications for mechanistic understanding and clinical practice. The cerebellum, long viewed as exclusively motor, is now understood to participate in multiple large-scale circuits subserving cognitive, social, and affective functions. Its dense reciprocal connections with prefrontal, parietal, and temporal association cortices—routed through the thalamus—form cerebello-thalamo-cortical loops critical for cognitive flexibility, social cognition, and language processing. Disruption of these loops, whether through primary cerebellar pathology (ATP2B2, ADGRB3) or secondary effects of thalamic or cortical dysfunction, produces distributed network abnormalities manifesting as cognitive and social impairments characteristic of ASD and schizophrenia. Conversely, emerging evidence implicates cerebellar GABAergic interneuron dysfunction as an upstream driver of frontal lobe developmental abnormalities via cerebello-prefrontal projections.

The basal ganglia, often considered primarily motor, are similarly integrated into broader cognitive and motivational circuits. Rather than operating in parallel, cerebellar and basal ganglia circuits interact at multiple levels—through thalamic convergence, cortico-striatal connections, and reciprocal cerebello-basal ganglia pathways—to coordinate motor and non-motor learning, reward processing, and action selection. Disruption of this integration, as seen in disorders involving MAPK8IP3-mediated JNK signaling or SCYL2-related excitotoxicity, produces both motor features (dystonia, ataxia, stereotypies) and cognitive-behavioral manifestations (impaired executive function, reduced cognitive flexibility, aberrant reward processing) that frequently co-occur in NDDs.

The brainstem, with its ascending arousal systems—noradrenergic locus coeruleus, serotonergic raphe nuclei, and cholinergic pedunculopontine nucleus—exerts global modulatory influences on forebrain development. These systems regulate attention, sleep-wake cycles, sensory gating, and autonomic homeostasis; their disruption in ARF3, WDR45B, and PPP1R21-related disorders may drive sleep disturbances, sensory processing abnormalities, and autonomic dysregulation—among the most common and disabling NDD comorbidities. Critically, brainstem dysfunction during developmental windows can alter trophic and activity-dependent signals guiding cortical maturation, suggesting that subcortical pathology propagates to widespread cortical dysfunction through experience-dependent plasticity mechanisms.

This brain-wide circuit perspective carries immediate clinical implications. First, neuroimaging should systematically evaluate cerebellar morphology, thalamic volume, brainstem integrity, and basal ganglia signal, not merely cortical structure. Second, clinical assessment should include motor function (tone, coordination, praxis), sensory processing, and autonomic function—domains frequently overlooked when attention focuses on cognitive and behavioral features. Third, therapeutic strategies targeting subcortical nodes—deep brain stimulation of thalamus or cerebellum, or pharmacological modulation of brainstem arousal systems—may offer circuit-level intervention avenues, though these remain experimental. Fourth, the recognition that subcortical vulnerability at specific developmental stages can propagate to cortical dysfunction suggests that early intervention targeting subcortical pathology—before cortical symptoms emerge—may prevent or mitigate secondary cortical impairments, reinforcing the urgent need for early detection and developmentally timed intervention.

Emerging therapeutic strategies and translational challenges

Currently clinically actionable genotype-guided management

For a small but growing subset of NDDs, molecular diagnosis directly informs clinical management. The ketogenic diet is standard of care for GLUT1 deficiency (SLC2A1 mutations). In Dravet syndrome (SCN1A LoF), sodium channel blockers (lamotrigine, carbamazepine) are contraindicated, whereas stiripentol, clobazam, and valproate are recommended. Conversely, in KCNQ2-related neonatal epilepsy, sodium channel blockers including phenytoin and carbamazepine are often effective—illustrating how the same drug class can be either contraindicated or beneficial depending on the genetic lesion. In tuberous sclerosis, mTOR inhibitors (e.g., everolimus) reduce seizure frequency and subependymal giant cell astrocytomas. These examples, though confined to a minority of NDDs, demonstrate that genotype-guided management can reduce diagnostic odysseys, guide recurrence risk counselling, and improve seizure control and developmental outcomes. However, for the vast majority of affected individuals, care remains symptomatic and non-specific.

Emerging mechanism-based approaches in preclinical and early-phase testing

A wave of mechanism-targeted strategies is advancing through preclinical and early-phase development. ASOs can modulate splicing, knock down toxic transcripts, or restore protein expression. For Dravet syndrome, ASOs upregulating SCN1A splicing or restoring Nav1.1 have shown promise in mouse models, with clinical trials underway. In Angelman syndrome, ASOs aimed at unsilencing the paternal UBE3A allele have entered early human studies. Gene therapy using adeno-associated virus vectors is being explored for CDKL5 deficiency (AAV9-mediated replacement) and MECP2 duplication (gene silencing or X-linked reactivation). Small molecules under investigation include translational read-through compounds for nonsense mutations (e.g., ataluren) and HDAC inhibitors (vorinostat, valproate), though broad epigenetic effects raise safety concerns for the developing brain. Anti-inflammatory and neuroprotective strategies inspired by MIA models are being evaluated in high-risk preterm infants, though clinical translation remains distant. Most emerging modalities remain several years from clinical availability.

Unique challenges in NDD therapeutics

Several formidable obstacles distinguish NDD therapeutics from other fields. First, developmental timing: many NDDs arise from disruptions in neurogenesis, migration, or synaptogenesis that occur months or years before clinical diagnosis, by which time the critical window for reversing structural abnormalities may have closed. Second, blood-brain barrier delivery: ASOs, gene vectors, and biologics require invasive intrathecal or intracerebroventricular routes, limiting scalability and increasing risk. Third, model system limitations: iPSC-derived neurons and organoids lack mature circuit context, while animal models fail to capture human-specific developmental timing and complexity. Fourth, allelic pleiotropy: exemplified by SCN2A, where loss- and gain-of-function mutations produce opposite phenotypes, demanding allele-specific rather than gene-wide strategies. Fifth, clinical trial heterogeneity: small patient populations for individual genetic subtypes make traditional randomized trials logistically difficult and statistically underpowered, while validated surrogate endpoints sensitive to developmental change are lacking.

A translational roadmap

To accelerate clinical translation, we propose five converging priorities: (I) integrating WGS into expanded newborn screening to enable presymptomatic diagnosis; (II) developing and validating multimodal biomarkers (epigenetic, imaging, electrophysiological) for disease tracking, patient stratification, and surrogate endpoints; (III) establishing international patient registries and natural history studies for genetically defined subtypes; (IV) embracing adaptive trial designs accommodating small samples and biomarker-based enrichment, including platform trials testing multiple candidates across subtypes; and (V) prioritizing therapies targeting convergent downstream pathways (synaptic function, excitability, epigenetic regulation) over individual genes to maximize impact across genetically heterogeneous populations. The first generation of mechanism-based therapies is now entering clinical trials. Even if full reversal of established disease proves elusive, early intervention targeting core pathophysiology may mitigate secondary complications and improve quality of life. This progress, consolidating and accelerating, justifies cautious optimism while demanding sustained interdisciplinary collaboration and global equity in research participation.

Discussion

The past two decades have transformed our understanding of NDDs—from clinical description to biological definition, from phenomenology to mechanism. The genetic architecture now stands revealed in considerable detail: hundreds of high-confidence risk genes, recurrent CNVs with refined penetrance estimates, and a polygenic background that continuously modulates liability across the entire population. The perinatal environment, long recognized as an important determinant, has been recontextualised not as a simple catalogue of toxic exposures but as a developmentally timed interface with genetic susceptibility, wherein maternal genotype shapes intrauterine and postnatal environments through genetic nurture and imprinting. The molecular mediators—DNA methylation, histone modifications, noncoding RNAs, spliceosomal components—are increasingly well characterized, bridging epidemiology to cell biology. And the neurobiological consequences, from ion channel biophysics to circuit organization, are traceable through patient-derived cellular models, animal constructs, and human neuroimaging with ever-increasing resolution.

Yet this very success exposes the limitations of the frameworks that enabled it. Binary classifications—monogenic versus polygenic, rare versus common, LoF versus gain-of-function—obscure the continuous, interactive nature of genetic risk. A child carrying a pathogenic 16p11.2 deletions, high polygenic risk for schizophrenia (derived from adult GWAS and representing largely indirect evidence), and maternal polygenic liability for ADHD with prenatal smoking does not fit any single explanatory category. The field lacks a unified quantitative framework integrating diverse risk dimensions, developmental timing, and environmental context. Developing such a framework—a ‘polygenic plus’ architecture encompassing rare variants, CNVs, common variants, and their interactions—remains the most pressing theoretical challenge.

Beyond these conceptual limitations, several methodological concerns warrant critical appraisal. Throughout the evidence reviewed here, replication status varies widely; many candidate gene and early G×E findings have not been independently replicated, and sample sizes in many studies remain insufficient for detecting interaction effects. Publication bias may skew the literature toward positive associations, and multiple testing issues are often inadequately addressed. Moreover, the evidentiary weight differs substantially across study designs: human observational associations—even from large cohorts—do not equate to causal evidence, which requires Mendelian randomization, family-based designs, or quasi-experimental approaches. Animal models and organoid studies provide valuable mechanistic insight but cannot substitute for human prospective data. Ancestry bias persists across most genetic and epigenetic studies, with non-European populations severely underrepresented. Tissue specificity of epigenetic findings (e.g., blood versus brain) further limits causal inference. Where these caveats apply, we have integrated them into the relevant sections above; we highlight them here collectively to encourage cautious interpretation and to guide future research priorities.

Developmental timing has moved to center stage. Transcriptomic atlases reveal that NDD risk genes cluster in temporally and spatially restricted patterns corresponding to critical periods of neurogenesis, migration, synaptogenesis, and refinement. Environmental exposures exhibit analogous windows of susceptibility. Genetically defined disorders have their own temporal logic: Rett syndrome manifests regression after apparent normal development; channelopathies present seizures neonatally; 22q11.2 deletion syndrome confers 25% psychosis risk in early adulthood. Therapeutic strategies must therefore be developmentally specified. Prophylaxis may require presymptomatic diagnosis—which depends on newborn screening or early genetic testing, neither widely available for most NDDs.

The ancestral and geographical bias of NDD genetics is no longer defensible. The vast majority of GWAS and exome cohorts are European; reference datasets—gnomAD, UK Biobank, ALSPAC—overwhelmingly so. PRS perform poorly in non-European populations; allele frequencies, linkage disequilibrium, and even risk gene identities may differ. Consanguineous populations, disproportionately responsible for recessive NDD gene discovery, remain grossly understudied at scale. This inequity is not merely methodological; it is an ethical failure perpetuating disparities in diagnostic yield, prognostic accuracy, and therapeutic access.

Several priorities for the next decade are clear. First, integration of WGS into first-tier clinical diagnostics, with systematic reanalysis of unsolved cases and open data sharing. Second, expansion of prospective, deeply phenotyped birth cohorts with serial biospecimen collection—placenta, cord blood, neonatal dried blood spots, buccal swabs—to enable truly developmental analysis of gene-environment interplay and to facilitate the validation and standardization of candidate biomarkers for early detection. Third, development of quantitative frameworks integrating rare and common variation, genetic and environmental risk, and developmental timing into testable models of individual liability. Fourth, and most fundamentally, a deliberate, well-funded, globally collaborative effort to diversify ancestral and geographical representation in NDD genetics.

Whether this research programme will ultimately reduce the burden of NDDs remains uncertain. Some disorders, once established, may not be reversible; developmental trajectories deflected from their normative course may resist correction; the brain’s compensatory mechanisms, though powerful, have inherent limits. Yet it is equally possible that current therapeutic pessimism reflects the field’s immaturity rather than intrinsic tractability (Figure 5, Table 8).

Figure 5.

Translational roadmap for precision medicine in NDDs. A five‑step framework from discovery to clinical application. Step 1—genomic diagnosis: WGS, CMA, long‑read sequencing identify rare and noncoding variants. Step 2—mechanistic studies: iPSC, organoids, animal models validate pathogenicity. Step 3—biomarker development: neuroimaging, EEG, epigenetic signatures enable stratification. Step 4—targeted interventions: ASOs, gene therapy, small molecules. Step 5—long‑term follow‑up: cognitive and quality‑of‑life assessment, counselling. A feedback loop (dashed arrow) refines models. Clinical readiness increases downward. For most NDDs, steps 3–4 remain under development. ASO, antisense oligonucleotide; CMA, chromosomal microarray; EEG, electroencephalography; iPSC, induced pluripotent stem cell; NDD, neurodevelopmental disorder; WGS, whole-genome sequencing.

Table 8. Clinical translation and research gaps in neurodevelopmental disorders.

Domain	Current progress	Current clinical availability	Major challenges	Future directions
Genomic Diagnosis	WGS identifies noncoding variants (e.g., RNU4-2); CMA first-tier; long-read sequencing resolves complex SVs	Yes (diagnostic)	Underrepresentation of non-European populations; interpretation of VUS; cost	Expand WGS to diverse ancestries; implement systematic reanalysis; develop functional validation pipelines
Genotype-guided acute management	Ketogenic diet for GLUT1; sodium channel blocker avoidance for SCN1A; stiripentol for Dravet	Yes (limited)	Only applicable to small subset; lack of robust biomarkers for titration	Integrate pharmacogenomics; develop rapid genetic testing in NICU
Mechanism-based therapies	ASOs for Angelman/Dravet (early trials); gene therapy for CDKL5 (preclinical); HDAC inhibitors for BRPF1 (preclinical)	No (research only)	BBB delivery; developmental windows; off-target effects; animal model limitations	Prioritize delivery methods; identify surrogate endpoints; trial readiness networks
Prevention/early intervention	Inflammation-phthalate interaction; maternal diet/supplement trials; therapeutic hypothermia for HIE	No (research only)	Lack of validated biomarkers for high-risk stratification; confounded by genetic nurture	Establish birth cohorts with multi-omics; test anti-inflammatory/neuroprotective agents in high-risk infants
Biomarker development	DNA methylation (cord blood, placenta); imaging (DTI, resting-state fMRI); EEG spectral power	No (research only)	Tissue specificity; lack of normative reference data across populations; cost	Standardize protocols; validate in diverse cohorts; integrate multi-modal biomarkers
Genetic counselling	CNV penetrance databases (gnomAD); PIEV classification for reduced-penetrance CNVs	Yes (counselling)	Incomplete penetrance; parental origin unknown; emotional impact of susceptibility loci	Standardised reporting; include non-genetic modifiers; long-term follow-up

This table synthesizes the current state of diagnostic, therapeutic, preventive, and counselling approaches for neurodevelopmental disorders, alongside major challenges and future directions. It serves as a concise summary of the translational roadblocks discussed in the main text, including ancestry bias, poorly defined developmental windows, and incomplete penetrance. ASO, antisense oligonucleotide; BBB, blood-brain barrier; CMA, chromosomal microarray; CNV, copy number variation; DTI, diffusion tensor imaging; EEG, electroencephalography; fMRI, functional magnetic resonance imaging; HDAC, histone deacetylase; HIE, hypoxic-ischemic encephalopathy; NICU, neonatal intensive care unit; PIEV, classification for CNVs with reduced penetrance and variable expressivity; SV, structural variant; VUS, variant of uncertain significance; WGS, whole-genome sequencing.

Conclusions

This integrative framework reveals that NDDs arise from complex interactions among genetic susceptibility, perinatal environmental insults, and epigenetic mediation, converging on distributed brain-wide circuit dysfunction. Despite substantial progress, critical gaps remain—including ancestral underrepresentation, limited biomarkers for early detection, and modest translation to therapies. Prioritizing diverse genomic cohorts, prospective multi-omic birth cohorts, and rigorous evaluation of targeted interventions in genetically defined subgroups will be essential to translate mechanistic insights into improved clinical outcomes for affected individuals and families.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.