Autism spectrum disorder (ASD) is a highly heritable neurodevelopmental condition^[1], yet a substantial proportion of its genetic architecture remains unresolved. Short-read whole-genome sequencing (SR-WGS) has successfully identified numerous de novo coding protein-truncating and large copy number variants, which account for only a fraction of the overall ASD risk^[2]. This missing heritability reflects a complex genomic architecture involving rare non-coding variants, cryptic structural rearrangements, polygenic background, and repetitive sequences. Among these potential contributors, tandem repeat (TR) variants represent a highly variable genomic component that remained poorly understood during the SR-WGS era^[3]. Advances in long-read whole-genome sequencing (LR-WGS) now allow us to interrogate these complex regions, offering a critical opportunity to investigate how TR variations^[4] and their associated parent-of-origin transmission biases elucidate the diverse downstream molecular mechanisms that shape the broader inherited architecture of ASD [Figure 1].

In the etiology of ASD, it is crucial to distinguish between highly penetrant Mendelian repeat-expansion disorders and the broader, genome-wide TR burden. While Mendelian expansions typically drive deterministic monogenic syndromes, the broader TR risk architecture encompasses diverse variants, ranging from rare large expansions to subtle non-coding variations, that collectively shape baseline developmental susceptibility. Driven by advances in sequencing technologies and bioinformatic tools, empirical evidence for this genome-wide impact has rapidly accumulated^[5,6]. For instance, rare TR expansions detected by SR-WGS are estimated to account for approximately 2.6% of ASD risk, a proportion that likely remains underestimated^[5]. This is because SR-WGS is inherently biased toward shorter, simpler repeats, making the accurate resolution of larger, structurally complex, or heavily expanded TR alleles a formidable challenge. Consequently, current clinical consensus guidelines emphasize that while SR-WGS serves as a powerful, cost-effective tool for high-throughput screening, bioinformatic estimates of repeat size should be confirmed by locus-specific molecular assays before being used as the basis for clinical diagnosis^[7]. To address these constraints, LR-WGS technologies have emerged, generating reads spanning kilobases to megabases^[8]. This technology can resolve complex structural variants, unambiguously determine parental origin, and capture previously inaccessible large repeat tracts^[9,10]. Nevertheless, the widespread integration of LR-WGS into routine clinical diagnostics remains hindered by high sequencing and computational costs, which limit cohort scalability^[11]. Furthermore, substantial technical hurdles persist, including the detection of low-frequency somatic mosaicism, the characterization of sequence interruptions within repeat tracts, and the difficulties that alignment algorithms face in precisely defining the boundaries of variable-number tandem repeat variants^[12-14]. Ongoing innovations in both sequencing and computational frameworks are poised to overcome these barriers, thereby transforming our understanding of TR variations and enabling a new era of precise genetic counseling and targeted intervention.

Parent-of-origin effects (POEs) refer to the phenomenon in which the phenotypic impact of a genetic variant is dictated by its parental mode of inheritance^[15]. Although classically associated with genomic imprinting, POEs are recognized as an important aspect of mutational behavior and clinical penetrance in TR disorders. Fragile X syndrome (FXS), one of the most common inherited monogenic causes of ASD, results from a full mutation (> 200 CGG repeats) in the FMR1 gene^[16]. Maternal carriage of a premutation allele substantially elevates the risk of full mutation expansion in the offspring, whereas paternal transmission of a similarly sized allele rarely drives such progression^[17]. Moreover, contraction of the premutation repeat displays a paternal bias. The full mutation expansion triggers aberrant methylation of the FMR1 promoter and subsequent transcriptional silencing, ultimately causing a deficiency in fragile X messenger ribonucleoprotein essential for synaptic plasticity^[18]. In contrast to FMR1, other disease-associated loci exhibit divergent dynamics, with paternal transmission driving greater instability. For example, expansion of the HTT CAG repeat occurs almost exclusively through the male germline, and a similar paternal bias has been observed within specific allelic ranges of the DMPK CTG repeat^[19]. Intriguingly, recent genome-wide findings demonstrate that abnormal ASD-risk TR variants across the entire genome also display significant, though often heterogeneous, parental biases^[20]. These biased yet divergent parental dynamics challenge the traditional assumption of parental equivalence in gene-disease associations. Therefore, deciphering POEs in ASD-associated TRs requires a conceptual shift: we should not merely evaluate the presence of these repeat variants but fundamentally characterize their parental lineage of transmission. For a highly heterogeneous condition such as ASD, this parental-stratification lens holds the potential to unveil critical signals that have historically been obscured by bulk, parent-agnostic genetic analyses.

Understanding how TR variants drive the complex neurodevelopmental phenotypes characteristic of ASD represents a major challenge in the field. Emerging evidence suggests that many TR expansions confer ASD risk through a loss-of-function mechanism driven by localized epigenetic collapse^[21]. In individuals with FXS, expansions of the CGG repeat in the FMR1 promoter trigger hypermethylation and subsequent chromatin compaction, leading to complete transcriptional silencing of the locus^[22]. Similar epigenetic silencing has been documented in other loci associated with neurodevelopmental delay and ASD, including AFF2^[23], XYLT1^[24], FRA10AC1^[25], CBL^[26], and DIP2B^[27]. Interestingly, a recent study integrating 18,236 SR-WGS and 950 LR-WGS samples revealed that massive CGG expansions and promoter hypermethylation at DIP2B are frequently observed in phenotypically normal individuals, showing no robust clinical association with neurological deficits^[28]. These findings challenge the deterministic assumption that repeat expansion invariably leads to pathology, highlighting the critical influence of genetic modifiers or epigenetic thresholds on clinical penetrance. While these epigenetic collapse models primarily result in a loss of normal gene expression, other pathogenic TR variants operate via toxic gain-of-function routes^[22]. For instance, the intronic GCC expansion in AFF3 appears to drive pathogenesis through mechanisms such as RNA toxicity or repeat-associated non-AUG (RAN) translation^[29]. The concept of dominant-negative interference is further exemplified by myotonic dystrophy type 1 (DM1), a condition frequently comorbid with ASD. In DM1, transcribed CUG repeats form stable secondary structures that sequester conserved RNA-binding proteins essential for alternative splicing in the brain, precipitating widespread mis-splicing across the transcriptome^[30]. Although primarily a neuromuscular disorder, DM1 serves as a paradigmatic model for how repeat-induced RNA toxicity could similarly disrupt neurodevelopmental pathways in ASD. More broadly, these observations suggest that the molecular etiology of pathogenic TR variants extends well beyond simple promoter silencing, RNA toxicity, or RAN translation, encompassing a considerably broader spectrum of regulatory disruption. These diverse mechanisms represent frontier areas that warrant rigorous functional validation in future ASD cohort studies to elucidate the precise genotype-phenotype correlations.

Translating the expanding landscape of TR variants into clinically actionable insights presents a formidable bottleneck in ASD genetics. The path from technological discovery to routine clinical implementation is hindered by several interconnected biological, technical, and ethical challenges. First, determining whether a specific TR expansion is benign or pathogenic remains exceptionally difficult. In contrast to single-nucleotide variants, TR variants lack robust, standardized pathogenicity thresholds. This uncertainty is compounded by incomplete or variable penetrance, as exemplified by the massive DIP2B expansions observed in phenotypically normal individuals^[28]. Consequently, defining the boundary between normal polymorphic variation and disease-causing expansion remains an elusive endeavor. Beyond biological ambiguity, our current ability to interpret TR variants is severely restricted by incomplete reference databases. Traditional linear reference genomes fail to capture the complex, multi-allelic nature of repetitive loci and suffer from a profound ancestry bias. Transitioning to pangenome references, which utilize graph-based structures to represent diverse human genomic architectures, is critical to overcoming these gaps. By incorporating multi-ancestry data, pangenomics will enable the comprehensive mapping of population-specific TR alleles, thereby mitigating ancestry bias and establishing accurate baselines for diverse populations. Finally, if POEs and risk-associated TRs are integrated into preconception screening, genetic counselors will face unprecedented ethical challenges. Conveying risks characterized by highly variable phenotypes, unpredictable parental expansion dynamics, and poorly understood penetrance thresholds to prospective parents will require nuanced communication strategies. Such efforts are essential to prevent unnecessary anxiety and avoid guiding families toward ill-informed reproductive choices.

The inherited architecture proposed in this perspective constitutes a foundational framework rather than a theoretical abstraction, as TR variants, phasing, and POEs represent fundamental syntactic elements of the human genome. These components critically dictate how genetic information is transmitted, regulated, and expressed. While these layers of complexity remained largely illegible during the era of SR-WGS, LR-WGS has begun to decipher this nuanced framework within the context of ASD. Realizing the full clinical potential of this model hinges on prioritizing four actionable milestones: expanding large, family-based LR-WGS cohorts; establishing standardized benchmarks for TR calling; deploying high-throughput functional validation assays; and implementing ancestry-aware reference resources. Although deciphering this multidimensional landscape remains a formidable challenge, these advances lay the groundwork for integrating these genomic insights into more precise diagnostic frameworks and into informed clinical management of individuals with ASD.