Polygenic and developmental profiles of autism differ by age at diagnosis

Abstract

Although autism has historically been conceptualized as a condition that emerges in early childhood^1,2, many autistic people are diagnosed later in life^3-5. It is unknown whether earlier- and later-diagnosed autism have different developmental trajectories and genetic profiles. Using longitudinal data from four independent birth cohorts, we demonstrate that two different socioemotional and behavioural trajectories are associated with age at diagnosis. In independent cohorts of autistic individuals, common genetic variants account for approximately 11% of the variance in age at autism diagnosis, similar to the contribution of individual sociodemographic and clinical factors, which typically explain less than 15% of this variance. We further demonstrate that the polygenic architecture of autism can be broken down into two modestly genetically correlated (r_g = 0.38, s.e. = 0.07) autism polygenic factors. One of these factors is associated with earlier autism diagnosis and lower social and communication abilities in early childhood, but is only moderately genetically correlated with attention deficit-hyperactivity disorder (ADHD) and mental-health conditions. Conversely, the second factor is associated with later autism diagnosis and increased socioemotional and behavioural difficulties in adolescence, and has moderate to high positive genetic correlations with ADHD and mental-health conditions. These findings indicate that earlier- and later-diagnosed autism have different developmental trajectories and genetic profiles. Our findings have important implications for how we conceptualize autism and provide a model to explain some of the diversity found in autism.

Fig. 1. Trajectory analyses in three of the four birth cohorts.

a–c, Longitudinal growth mixture models of SDQ total scores in autistic individuals, demonstrating the presence of two groups in the MCS (a), LSAC-B (b) and LSAC-K (c) cohorts. Shaded areas indicate 95% confidence intervals of the line of best fit. d–f, Stacked bar charts show the proportion of individuals who had been diagnosed as autistic at specific ages, categorized by membership in the latent trajectories identified from the growth mixture models in MCS (d), LSAC-B (e), and LSAC-K (f) cohorts. Darker colours indicate male individuals and lighter colours indicate female individuals. P-values are from χ² tests (two-sided) comparing the distribution of age at autism diagnosis between the two latent trajectories (pooling the two sexes).

Fig. 2. Heritability of age at autism diagnosis.

a, SNP-based heritability (h²) for age at autism diagnosis in the SPARK cohorts, calculated using single-component genome-wide complex trait analysis with a genomic-relatedness-based restricted maximum-likelihood approach (GCTA-GREML) for the SPARK Discovery cohort (orange dashed line, n = 16,786), SPARK Replication cohort (purple dashed line, n = 8,558) and a meta-analysis of the two (light blue solid line, n = 25,344), and iPSYCH, calculated using linkage disequilibrium score regression coefficient (LDSC) (solid green line, n = 18,965). b, SNP-based heritability (GCTA-GREML) in the SPARK cohorts after accounting for various clinical and sociodemographic factors. A ‘+’ indicates the baseline model in addition to the specified covariates. The x axis has been truncated at 0 and 0.25. In a and b, central points represent SNP-based heritability estimates and error bars indicate 95% confidence intervals. Sample sizes for b are provided in Supplementary Table 10. PC, genetic principal component; RBS-R, Repetitive Behavior Scale-Revised; SCQ, Social Communication Questionnaire; SES, socioeconomic status.

Fig. 3. Median age at autism diagnosis and genetic correlations with age at autism diagnosis across different GWAS cohorts.

Left, median age at diagnosis (years) with error bars representing median absolute deviation. Circle size indicates the number of autistic individuals (cases) in the GWAS, and exact sample sizes are provided in Box 1. Beige circles represent GWAS unstratified by age at diagnosis; red circles represent GWAS stratified by age at diagnosis. Lighter (more transparent) circles indicate studies with no information about age at autism diagnosis, and the median ages have been inferred from other available information (PGC-2017 (ref. ) and Grove et al.). Right, genetic correlations with age at autism diagnosis for both SPARK (blue, meta-analysis, n = 28,165) and iPSYCH (green, n = 18,965) datasets, with error bars representing 95% confidence intervals.

Fig. 4. Two genetic latent factors in autism.

a, Left, genetic correlation heatmaps of all GWASs of autism as described in Box 1. Asterisks indicate significant genetic correlations after Benjamini–Yekutieli adjustment. Right, median age at autism diagnosis for the same GWAS (indicated by the number on top of the circle). Error bars indicate median absolute deviation; the size of the circles indicate the sample size. For both panels, GWASs have been ordered based on hierarchical clustering of the genetic correlations. b, Structural equation model illustrating the two-correlated genetic-factor models for autism, using six minimally overlapping autism GWAS datasets. F1, factor 1; F2, factor 2. One-headed arrows depict the regression relationship pointing from the independent variables to the dependent variables; the numbers on the arrows represent the regression coefficients of the factor loadings, with standard errors provided in parentheses. Covariance between variables is represented by two-headed arrows linking the variables. The numbers on the two-headed arrows can be interpreted as genetic-correlation estimates with the standard error provided in parentheses. Residual variances for each GWAS dataset are represented using a two-headed arrow connecting the residual variable (u) to itself. Standard errors are shown in parentheses.

Fig. 5. Genetic correlation between the two autism polygenic factors and a range of mental-health, neurodevelopmental and cognitive traits.

Central points indicate the estimate (genetic correlation), error bars indicate 95% confidence intervals and asterisks indicate significant P-values (two-sided) with Benjamini–Yekutieli adjustment. Sample sizes are shown in Supplementary Table 18.

Extended Data Fig. 1. Variance in age at diagnosis explained by various clinical and sociodemographic factors.

Variance explained (R² or η²) in age at autism diagnosis by clinical and sociodemographic factors, identified from the review of literature (1998–2023). Variables are grouped into sociodemographic (MAD, SES), socioemotional-behavioural (SDQ scores), sex, clinical (e.g., IQ, regression, language ability), and autism severity (e.g, SCQ, ADOS, ADI-R, RBS-R) categories. Circle size represents sample size, with larger circles indicating larger cohorts. Colored points denote variables analysed in the current study. Inset shows factors that explain greater than 10% of the variance in age at autism diagnosis. Note: None of these studies account for additional family, service access, and contextual factors known to influence diagnostic timing. Abbreviations: MAD, Maternal Age at Delivery; IQ, Intelligence Quotient; SES, Socio-economic Status; ADOS, Autism Diagnostic Observation Schedule; ADI-R, Autism Diagnostic Interview-Revised; RBS-R, Repetitive Behavior Scale-Revised; SCQ, Social Communication Questionnaire; SDQ, Strengths and Difficulties Questionnaire; SDQ Total, SDQ Total Difficulties.

Extended Data Fig. 2. Two aetiological polygenic models of autism.

In Model 1 (Unitary Model), we assume a single liability threshold polygenic model. In this model, autism emerges from a unitary polygenic aetiology. Autistic individuals diagnosed later have lower polygenic predisposition than individuals diagnosed earlier. In Model 2 (Developmental Model), we model two correlated age-dependent polygenic liabilities.

Extended Data Fig. 3. Schematic of the study aims.

The study consists of four linked aims to understand whether the developmental trajectories and polygenic etiology of autism differs by age at diagnosis. In Aim 1, we modelled socioemotional and behavioural trajectories among autistic individuals in birth cohorts and investigated their association with age at autism diagnosis. In Aim 2, estimated the SNP-based heritability of age at autism diagnosis and whether it attenuates when accounting for various clinical and demographic factors. In Aim 3, we investigated whether the varying patterns of genetic correlations observed among different GWAS of autism can be explained by different polygenic factors associated with age at diagnosis. In Aim 4, we investigated the genetic relationship between the two autism polygenic factors and mental health and developmental phenotypes.

Extended Data Fig. 4. Schematic diagram of the birth cohorts included in the study.

Schematic diagram of the cohorts included in the study and the ages when data was collected for SDQ scores (dots) and autism diagnosis (in boxes). Reports of autism diagnosis were available at ages: MCS - 5,7,11,14; GUI - 9,13,17; LSAC-B - 7,9,11,13,15; and LSAC-K: 11,13,15,17. MCS = Millennium Cohort Study; GUI = Growing up in Ireland (cohort ’98); LSAC-B = Longitudinal Study of Australian Children (Birth cohort); LSAC-K = Longitudinal Study of Australian Children (Kindergarten cohort). Sample sizes and the year of initial SDQ data collection for each cohort are shown on the ordinate axis. The age cutoff used in the Latent Growth Curve Models for each cohort is indicated by a red line. GUI was used only for Latent Growth Curve Models and excluded from Growth Mixture Models.

Extended Data Fig. 5. Schematic diagram of age at autism diagnosis GWAS and age stratified autism GWAS.

Schematic diagram illustrating the main GWAS conducted in the study using the SPARK and iPSYCH cohorts. We conducted two age at autism diagnosis GWAS. In addition, we conducted six case-control GWAS, where autistic individuals were stratified based on their age at autism diagnosis.

Extended Data Fig. 6. Distribution of age at autism diagnosis in iPSYCH and SPARK.

Frequency histograms of age at autism diagnosis in iPSYCH and SPARK. Median and median absolute deviation (MAD) for age at diagnosis, and sample sizes (N) have been provided.

Extended Data Fig. 7. Within-cohort genetic correlation between age at diagnosis stratified autism GWAS and mental health and cognition related phenotypes.

Genetic correlation between age at autism stratified GWAS in SPARK (meta-analysed from Discovery and Replication cohorts) and iPSYCH and other mental health and cognition related phenotypes. Points represent genetic correlation estimates and whiskers indicate 95% confidence intervals. Green represents the earlier diagnosed autism GWAS (iPSYCH before 9 and SPARK before 6), and purple represents later diagnosed autism GWAS (iPSYCH and SPARK after 10). Asterisk (*) indicates significantly different genetic correlation between the earlier and later diagnosed GWAS (P < 0.05, two-tailed Z test).