Common and Rare Genetic Variants Explain Distinct Diagnostic Variance in Pediatric Attention Deficit Hyperactivity Disorder

Abstract

Purpose.

Pediatric attention deficit hyperactivity disorder (ADHD, MIM: 143465) is highly heritable, yet the genetic architecture of the condition remains poorly understood. The current study tested the hypothesis that rare and common genetic variants reflect distinct genetic pathways to ADHD.

Methods.

Genome sequencing was completed for 150 pediatric ADHD cases and 370 controls. ADHD polygenic scores were derived and compared across five methods, including two published GWAS and two publicly available catalogs. Likely pathogenic rare variants were identified with a previously published customized annotation and classification pipeline followed by manual curation using established ACMGG variant interpretation guidelines.

Results.

ADHD cases had higher ADHD polygenic scores and lower IQ polygenic scores. Likely pathogenic variants for ADHD were identified in 13% of cases and 0.5% of controls. ADHD polygenic scores among cases without rare variants were higher than cases carrying rare variants. ADHD cases were predicted by ADHD and IQ PGS, ancestry, and rare variant status with 70% area under the curve.

Conclusions.

The genetic etiology of ADHD is likely multifactorial, with independent contributions from common and rare variants. Genome wide association studies of ADHD may have increased power to detect common genetic loci if individuals with rare variants are excluded.

Introduction

Pediatric attention deficit hyperactivity disorder (ADHD, MIM: 143465) is a prevalent and highly heritable disorder with approximately 75% broad based heritability¹. Despite this, the genetic architecture of ADHD remains largely unknown. Both rare^2,3 and common^4,5 genomic risk factors have been identified, but these account for only a minority of diagnostic variance. The single nucleotide polymorphism (SNP) liability-based heritability is only 28% and common genetic loci for ADHD overlap considerably with other psychiatric and cognitive traits⁶. Recently, our group reported that up to 50% of individuals with ADHD may have rare disruptive variants to genes implicated in neurodevelopment². In the current study, we examine whether the combination of rare and common genetic risk factors explains additional variance in pediatric ADHD.

Polygenic scores (PGS), which estimate the degree of common genetic liability for a disorder within a particular individual, have been proposed as a potential clinical tool to aid behavioral diagnoses and identify affected individuals prior to their enrollment in formal schooling⁷. PGS for ADHD have been associated with categorical ADHD diagnoses, continuously measured ADHD traits^8,9, and neurobiological correlates of ADHD⁸. However, effect sizes for published ADHD PGS are modest, typically explaining < 4% of the phenotypic variance in the disorder^8,10. Moreover, specificity of the ADHD PGS is weak, with strong associations between ADHD PGS and comorbid psychiatric symptoms, such as substance use disorders, bipolar disorder, and anxiety^8,9,11.

The known polygenic contributions to ADHD are highly heterogeneous. To date, only two genome wide association studies (GWAS) have identified statistically significant loci associated with ADHD^12,13. These investigations found a total of 33 statistically significant loci, only six of which overlapped across studies. Lack of polygenic overlap across GWAS points to methodology as another likely source of variability in PGS estimates for ADHD. PGS scores may be calculated from a single GWAS or from aggregate data published in publicly available catalogs^14,15. Different PGS calculations for the same disorder often do not correlate with one another, as was demonstrated by an investigation of polygenic risk for psychiatric and medical diagnoses in the UK Biobank¹⁶. Moreover, the significance thresholds used to identify variants of interest from existing ADHD GWAS range from p = 5 × 10⁻⁸ to p = 1.0 ⁸. A common approach in the extant literature is exploratory testing of multiple significance thresholds, with selection of a final cutoff based on optimized case-control differentiation⁸.

Recent work suggests that an additional source of heterogeneity in ADHD PGS analyses may be the contribution of rare genetic variants. Rare variants associated with well-characterized pediatric ADHD cases (absent autism or intellectual disabilities) include both de novo and inherited variants, particularly to genes involved in histone methylation². One study found that children with ADHD who had large, rare copy number variants (CNVs) had lower ADHD PGS than those without¹⁷, which would support distinct contributions of common and rare variants to the disorder. By contrast, another study found that among healthy controls, individuals with CNVs had lower ADHD PGS than those without, while there were no differences in PGS by CNV status among ADHD cases¹⁸. Lastly, a study of individuals with ADHD that found an increased burden of rare missense and disruptive variants to genes implicated by GWAS¹⁹ suggests that common and rare variants for ADHD are overlapping

Individual variability in the phenotypic presentations of ADHD, including the timing of diagnosis and degree of functional impairment, can be explained at least in part by other risk and protective factors. For example, high effortful control, a temperament-based measure of self-regulation and goal directedness, was shown to mitigate the effects of family risk for ADHD on ADHD outcomes in youth enrolled in the longitudinal ABCD study²⁰. Similarly, higher IQ was associated with improved pharmacological treatment response in a large sample of children with ADHD²¹. As IQ scores are likewise a highly heritable trait, it stands to reason that polygenic load for IQ may attenuate the association between polygenic risk for ADHD and child diagnostic outcomes.

In the current study, we investigated moderators of ADHD PGS in a sample of 150 children with ADHD and 370 comparison controls on whom we had completed genetic sequencing. First, we identified rare, likely disruptive variants in this cohort. Second, we compared ADHD PGS effect sizes using PGS derived from 4 distinct variant lists: GWAS published in 2019¹³ and 2023¹², the publicly available PGS catalog, and the NHGRI-EBI GWAS Catalog²². Third, we tested our hypothesis that PGS would be lower in ADHD cases with rare disruptive variants, indicating two distinct etiological pathways to ADHD (common versus rare genetic liability). Finally, we examined whether a genetic protective factor, IQ PGS (derived from a publicly available dataset) moderated the effect of ADHD genetic liability on diagnostic outcomes.

Methods

Participants

ADHD participants were 4–17 year old children enrolled in the PrecisionLink Biobank for Health Discovery at Boston Children’s Hospital²³. Participants at Boston Children’s Hospital were invited to enroll, which included providing research access to electronic health records and linking biospecimens, when available. Participants completed informed consent with a research coordinator prior to contributing samples. Participants with available genetic data as of 2022 whose medical record indicated an ADHD diagnosis (ICD-10 codes F90.0, F90.1, F90.0, or F90.0) and did not indicate autism spectrum disorder or intellectual disability (ICD-10 codes F70, F84) were included in the study. Three participants with known large copy number variants (CNVs) were excluded.

Control participants were unaffected family members of individuals enrolled in the Children’s Rare Disease Collaborative at Boston Children’s Hospital²⁴ who had complete GS as of November 2022. The control cohort included family members who had available clinical diagnostic records that did not indicate any of the ADHD diagnostic codes listed above. Cohorts in this collaborative include neurological and medical diseases, typically with dominant genetic causes²⁵; thus, unaffected family members are not expected to carry genetic burden.

All procedures were approved by the Boston Children’s Hospital Institutional Review Board. Sex and ancestry were inferred through genetic analyses (Table 1). Eight control participants did not have sufficient X chromosome data to infer sex; these individuals were excluded from sex comparison analyses.

Table 1.

Participant Demographics

	ADHD	Controls
N	150	370
Female	48 (32%)	196 (53%)
Ancestry: African	12 (8%)	5 (1%)
Ancestry: Admixed American	3 (2%)	9 (2%)
Ancestry: East Asian	2 (1%)	9 (2%)
Ancestry: European	133 (89%)	347 (94%)
Heterozygote Variant	19 (13%)	2 (0.5%)

Genome sequencing and variant calling

Blood derived DNA collected from participants underwent short read genome sequencing (GS) to an average coverage of 40X as part of the Children’s Rare Disease Collaborative at Boston Children’s Hospital²⁴. Using the Illumina DRAGEN Bio-IT Platform using on-premises servers and version 3.9.3 of the Germline DNA pipeline, raw sequencing data were processed under read mapping, QC, and variant calling. Read alignment, read trimming, sorting, duplicate marking, and small variant calling were done with default parameters and using reference genome build hg38. All individuals included in the analysis (150 cases and 370 controls) were jointly variant called and genotyped²⁵, to ensure consistency of genotype calling across all individuals.

Rare Variant Identification

All VCFs were processed through our semi-automated custom analytical pipeline (previously described^2,26), including the following steps. First, VCFs were merged and filtered for high-quality calls (GQ >30, PASS filter, AAF > 0.15) using BCFtools. Candidate variants, defined as those with the greatest likelihood of being damaging met the following criteria: a max population allele frequency (maxAF) across all gnomAD v2.1.1 populations of 0.002%, AD>2, DP>7, AC gnomAD exomes <5, AC gnomAD genomes < 5, number of homozygous occurrences in gnomAD 0, damaging prediction (LOF1, LOF2, LOF3, NsynD3). Additional filtration was applied based on the variant type: LOF alleles required the gene to be constrained in the gnomAD 2.1.1 database with a pLI>0.8 or loss-of-function o/e < 0.3; NsynD3 alleles required the impacted genes to have missense constrain with either a misZ>=3.08 or o/e < 0.3. Candidate variants that could explain the ADHD diagnoses in the individuals were identified from these filtered variants through manual curation by applying the ACMGG criteria, along with previous reports and studies of the variants extracted from HGMD, ClinVar, OMIM, and SFARI gene databases, resulting in a set of variants of uncertain significance including both LOF and likely damaging missense variants in genes with established and putative roles in NDDs.

Gene functional enrichment analyses were performed using the 19 genes with rare variants possibly explaining the ADHD using the Enrichr²⁷ tool which allows for assessment of enriched processes and pathways from databases including the 2023 WikiPathway database and Gene Ontology molecular processes. The p-value provided by Enrichr is corrected for multiple hypotheses using the Benjamini-Hochberg procedure. The combined score integrates the p-value and the odd ratio by multiply the scores as a compromise between the methods and was reported to provide an improved ranking of enrichments.

ADHD Polygenic Scores

For comparison and completeness, five methods were used to compute ADHD PGS. In the first two approaches, effect sizes were obtained from the top results reported by Demontis et al., in either 2019¹³ or 2023¹². From the 2019 table, we were able to match 9 out of 12 variants to those called from our GS sequencing; we found high linkage disequilibrium (LD; r² ≥ 0.7) proxy variants for the other three. From the 2023 table, we matched 25 of the 27 variants and did not find any proxies for the remaining two. In the third approach, we downloaded SNP effect sizes for ADHD from the PGS Catalog score ID PGS002746 (source publication PGP000358²⁸) and found 458,934 matching variants of 513,659. Then PLINK²⁹ was used to generate polygenic risk scores for each sample in this study. For the fourth and fifth approaches, we downloaded summary statistics from the NHGRI-EBI GWAS Catalog for study GCST90275136³⁰ and found 4,092,115 matching variants in our data, out of 8,785,478 possible variants. For this set of GWAS results, we first used PRSice-2³¹ to calculate PGS for each sample in the study. PRSice-2 calculated the model fit using 10,002 p-value thresholds ranging from 5e-08 to 1 and selected 9.25e-03 as the threshold providing the best model fit (p-value = 4.08e-04). For another approach, we used PRS-CSx with a single European reference panel to calculate weights for each variant in the summary statistics. We acquired weights for 1,067,688 variants overlapping with the reference panel and used PLINK to calculate PGS for each sample in the study.

IQ Polygenic Scores

PGS for intelligence were computed using effect sizes of 25,424 matching variants (out of 26,145) from the PGS Catalog score ID PGS001919 (source publication PGP000263³²) and 4,577,641 matching variants (out of 9,295,118) from the GWAS Catalog study GCST006250³³.

Analytic Approach

Statistical analyses were done in R v4.1.0. ADHD and IQ PGS demonstrated normal distributions with skew and kurtosis < |2.0|. Diagnostic, sex and ancestry group differences in PGS were computed with independent-samples t-tests and one-way analyses of variance (ANOVA). Effect sizes were estimated with Cohen’s d. The optimal ADHD PGS was selected as that with the largest effect size and was used in subsequent analyses. Hierarchical logistic regression was estimated to evaluate the additive and interactive contributions of IQ PGS to ADHD outcomes. Power analysis indicated 85% power to detect a small effect (d = 0.29) between cases and controls. However, we were underpowered to compare PGS between rare variant heterozygote cases and non-heterozygous cases and controls (85% power to detect a large effect, d = 0.71). Thus, we focused on effect sizes rather than p-values for these exploratory comparisons.

Next, we estimated receiver operating characteristic (ROC) analysis (R package ROCR) using the formula in the optimal step of the hierarchical logistic regression. The sample was randomly split into 1,000 test (70%) and train (30%) datasets. Mean area under the curve (AUC) and 95% confidence interval estimates were estimated from the results.

Results

Rare Variant Discovery

Rare SNV and short insertion and deletion variants possibly explaining the ADHD diagnoses were identified in 13% (19/150) of the cases (Supplemental Table 1), including 7% with likely pathogenic variants (11/150), 4% (6/150) with LOF variants in neurodevelopment-related genes with less established disease associations, and 1% (2/150) with possibly damaging missense variants in disease-associated genes. Interestingly, genes impacted by these variants were enriched for histone modification (p = 0.03) pathways in the 2023 WikiPathway database and the Histone H3 Methyltransferase Activity (GO:0140938, p=0.013) molecular process, consistent with prior findings in ADHD^2,34, see Figure 1. In the control group, two participants were identified with likely pathogenic splicing variants (Supplemental Table 2). One individual carried a variant in the PLCB1 gene (HGNC:15917, NM_015192.4:c.1250+1G>A: NC_000020.11:g.8708753G>A), which is associated with epileptic encephalopathy. The second individual harbored a previously classified likely pathogenic variant (ClinVar ID: 3019100) in the KCNQ5 gene (HGNC:6299, NM_001160132.2:c.1277+2T>C: NC_000006.12:g.73129851T>C), linked to encephalopathy and ID. Consequently, these two control individuals were excluded from subsequent analyses.

Figure 1.

Rare variant gene enrichment analyses. Enrichment of A) molecular function gene ontology terms and B) functional pathways from Wikipathway 2023 genes with rare variants in our ADHD cohort. Combined score is calculated as the log product of Fisher’s exact test p-value and rank deviation z-score.

ADHD PGS Source Identification

ADHD PGS was significantly higher in cases compared to controls using the Demontis et al. 2023 and GWAS Catalog weights, with both the PRSice-2 and PRScsx methods (Table 2). The largest effect was found with the GWAS Catalog computed with PRSice-2; thus, this calculation was used in the subsequent analyses.

Table 2.

Case Control ADHD-PGS Effect Size By Source

ADHD PGS Source	t(df)	p	Cohen’s d
Demontis 201914	−0.35 (208)	.72542	−0.04
Demontis 202313	2.13 (218)	.03418	0.23
PGS Catalog16	1.76 (223)	.08031	0.18
GWAS Catalog22, PRSice-232	3.24 (274)	.00132	0.32
GWAS Catalog22, PRS-CSx33	2.69 (270)	.00761	0.27

Note: Independent samples t-tests were computed with unequal variances assumed.

Demographic and Diagnostic Associations with ADHD and IQ PGS

ADHD PGS did not differ by sex (t[508] = 0.62, p = .5337, d = 0.06). As expected, the omnibus test revealed a significant effect of ancestry on ADHD PGS: F(3,514) = 24.70, p < .00001. Post-hoc pairwise comparisons using Tukey’s HSD revealed lower ADHD PGS among those with European or Admixed American descent, compared to those with East Asian or African descent (d range = −1.64 to −1.65, ps < .002). Considering the 2% of control cases with likely pathogenic variants, the Relative Risk = 26.

ADHD cases had significantly lower IQ PGS compared to controls, using the GWAS Catalog: t(516) = 4.92, p = .00001, Cohen’s d = −0.50. IQ PGS did not differ by sex (t[508] = 1.19, p = .236; d = 0.11). IQ PGS varied significantly across ancestry (F[3,514] = 69.32, p < .00001). IQ PGS was higher among participants with European or Admixed American descent compared to East Asian or African descent; and higher among individuals with Eastern Asian compared to African descent (d range = 2.27 to 3.37, ps < .001).

Rare Variant Sensitivity Analyses

ADHD PGS

ADHD PGS was compared across rare variant heterozygote cases, non-heterozygote cases, and controls (Figure 2). The omnibus effect was significant: F(2,515) = 6.14, p = .0023. Post-hoc pairwise comparisons revealed that non-heterozygote cases had higher ADHD PGS compared to controls (d = 0.36, p = .0015). Non-heterozygote cases had higher ADHD PGS than heterozygote cases, with a similar effect size, although as expected this did not reach statistical significance (d = 0.29, p = .4028). In contrast, ADHD PGS among heterozygote cases was comparable to that of controls (d = −0.036, p = .9845).

Figure 2.

Distributions of ADHD polygenic scores across Controls (left), cases without rare variants (ADHD−; middle), and cases carrying rare variants (ADHD+; right). ADHD PGS are reported as z-scores (mean=0, standard deviation = 1). Cohen’s d effect sizes are reported to indicate differences between two groups. **p = .001.

IQ PGS

There was a significant omnibus effect of the three groups on IQ PGS: F(2,515) = 12.3451, p < .00001 (Figure 3). Post-hoc pairwise comparisons revealed a significantly lower IQ PGS in non-heterozygote cases compared to controls (d = −0.50, p < .0001). Non-heterozygote cases had a lower IQ PGS compared to heterozygote cases, but the effect size was small (d = −0.17, p = .6737). IQ PGS was lower among heterozygote cases than controls, with a small effect size (d = −0.26, p = .4227).

Figure 3.

Distributions of IQ polygenic scores across Controls (left), cases without rare variants (ADHD−; middle), and cases carrying rare variants (ADHD+; right). IQ PGS are reported as z-scores (mean=0, standard deviation = 1). Cohen’s d effect sizes are reported to indicate differences between two groups. ***p < .0001.

Hierarchical Logistic Regressions

A series of hierarchical logistic regressions with ADHD case/control status as the dependent variable was computed to examine additive and interactive effects of ADHD and IQ PGS on diagnostic outcomes (Table 3). Ancestry and rare variant heterozygote status were included as covariates. In the first step, ADHD PGS was associated with case-control status over and above ancestry. In the second step, IQ PGS was added; IQ PGS was independently associated with case control status, and the effect of ADHD PGS now only approached significance. In the third step, the interaction between ADHD and IQ PGS was not statistically significant.

Table 3.

Hierarchical Logistic Regressions

	B	SE	p	z	R 2
Step 1					0.14
African Ancestry	1.76	0.97	.0691	1.82
Eastern Ancestry	−1.17	1.33	.3756	−0.89
European Ancestry	0.50	0.80	.5318	0.63
Heterozygote	17.74	515.48	.9725	0.03
ADHD PGS	0.34	0.11	.0027	3.00
Step 2					0.15
African Ancestry	0.88	1.02	.3890	0.86
Eastern Ancestry	−1.49	1.33	.2635	−1.12
European Ancestry	0.56	0.80	.4897	0.69
Heterozygote	17.78	513.86	.9724	0.04
ADHD PGS	0.22	0.12	.0739	1.79
IQ PGS	−0.43	0.14	.0015	−3.17
Step 3					0.15
African Ancestry	1.00	1.05	.3430	0.95
Eastern Ancestry	−1.44	1.34	.2827	−1.074
European Ancestry	0.54	0.80	.4985	0.68
Heterozygote	17.77	513.75	.9724	0.04
ADHD PGS	0.22	0.12	.0684	1.82
IQ PGS	−0.45	0.14	.0016	−3.15
ADHD PGS x IQ PGS	0.05	0.12	.6621	0.44

Note: PGS=polygenic score. ADHD = attention deficit hyperactivity disorder. Heterozygote = individual with a rare variant. Polygenic scores were standardized within the sample prior to analyses to improve interpretation.

The second step of the hierarchical regression was identified as the optimal model, because the R² was larger than that of step 1 and comparable to that of step 3, and the interaction effect in step 3 was not significant. The mean AUC was 0.695 (95% CI: 0.693 – 0.698).

Sensitivity Analyses

The majority of participants in the sample were of European ancestry, and the ADHD GWAS derived from European datasets. Thus, we conducted sensitivity analyses using the subsample of participants with European ancestry (N = 118 non-heterozygote cases; N = 15 heterozygote cases; N = 345 controls). The results were consistent with the full sample. ADHD PGS was higher in non- heterozygote cases compared to controls (d = 0.29, p = .0212). Non-heterozygote cases had a higher ADHD PGS than heterozygote cases, with a medium effect that did not reach statistical significance (d = 0.44, p = .2097). Controls and heterozygote cases did not differ on ADHD PGS (d = −0.16, p = 0.7769). In the logistic regression models, IQ PGS was significantly negatively associated with case status, over and above heterozygote status and ADHD PGS; the interaction term between ADHD and IQ PGS was not significant. See Supplementary Materials for full results.

Discussion

As expected, a significant minority (13%) of the ADHD sample had identifiable rare variants that were likely associated with their ADHD diagnoses. Analysis of the rare variants revealed enrichment for genes involved in histone methylation, consistent with prior work^2,34. Results of the PGS analyses indicated that the NHGRI-EBI GWAS Catalog²² provided the most sensitive ADHD PGS for distinguishing cases and controls. Contrary to our expectations, the interaction between IQ PGS and ADHD PGS was not significantly predictive of case-control status. Rather, IQ and ADHD PGS each explained unique variance in case-control status. However, the variance explained by ADHD PGS was significantly reduced when IQ PGS was included in the model, indicating some overlap between these two genetic measures. Our results were consistent when the analyses were limited to participants with European ancestry.

Our exploratory analyses suggested that cases with a rare heterozygous variant had lower ADHD PGS than cases without a known variant. These results, although preliminary, point to an equifinal etiology for ADHD diagnosis, with distinct polygenic and rare variant genetic pathways to the disorder. This contrasts with a diathesis-stress model that would suggest that the combination of high PGS and rare variants lead to ADHD. Nonetheless, we were not able to rule out the possibility that PGS contributes to variable expression of ADHD symptom severity and psychiatric comorbidity profile among affected children with a rare variant, as has been suggested for other genetic disorders³⁵. Nor can we exclude the possibility that low ADHD PGS protects against the effect of more mild rare variants, as we only included variants meeting a classification of at least likely pathogenic by the established ACMGG criteria in our cases.

We found reduced ADHD PGS burden among rare variant heterozygote cases; if replicated in a larger sample, this result has implications for future GWAS. Specifically, individuals with candidate rare variants may dilute or alter the results of common variant associations. More refined, sensitive PGS may be obtained by excluding individuals with rare variants from GWAS samples. Conversely, our results suggest that individuals with ADHD with high PGS scores are less likely to carry pathogenic rare variants, consistent with previous reports for other traits³⁶.

With logistic regressions, we tested the hypothesis that polygenic load for a protective trait, such as high IQ, could mitigate the association between ADHD PGS and diagnostic outcome. We found partial support for this hypothesis with confirmation of the additive model, wherein the effect of ADHD PGS on diagnostic status was reduced when IQ PGS was included. However, there was no interaction between the two polygenic scores, indicating that PGS for high IQ does not protect against increased polygenic vulnerability for ADHD. While a prior report found an association between ADHD PGS and performance on cognitive tests¹², our analysis confirmed the complementary association, i.e., between IQ PGS and ADHD diagnosis. Altogether, our results suggest that the combination of moderately elevated ADHD PGS and moderately decreased IQ PGS may increase risk for ADHD diagnosis. Thus, this line of inquiry presents a third possible genetic pathway to ADHD, namely polygenic vulnerability for ADHD in combination with polygenic vulnerability for lower cognition.

Previous studies have established a link between DNA methylation patterns and ADHD, with significant methylation changes being identified between individuals with ADHD and healthy controls^37–39. Our results further support this connection, demonstrating an enrichment of rare damaging variants in histone methyltransferase genes in ADHD. Importantly, many histone methyltransferase genes are implicated in other neurodevelopmental disorders and exhibit variable expressivity and incomplete penetrance^40–42. Even more, other histone methyltransferase genes are known to display unique DNA methylation signatures, suggesting that underlaying rare variation within these genes can impact the transcriptional regulation of many other genes, possibly explaining the phenotypic variability amongst individuals⁴³. Therefore, rare methylation-related variants could lead to altered expression of genes identified by GWAS, without the need for an elevated PGS, though studies with larger family-based genomic sequencing cohorts will be required to fully assess the possible link between PGS and rare variation in methylation genes.

We compared five common approaches to deriving ADHD PGS. There was significant variability in effect sizes across these methods, all of which have been used in prior publications. We hypothesize that the GWAS published in 2019 did not confer a large effect size in our sample due to the small number of alleles that were reported⁵. However, we were surprised by the lack of significant difference found for ADHD PGS derived from the PRS Catalog, although the effect was in the expected direction. The largest effect sizes were found with the GWAS catalog. The effect was likely larger when using PRSice-2 because this method allowed us to include more variants than with PRS-CSx. PRS-CSx requires rsIDs which are not always available, particularly for more recently identified variants. Altogether, the variability in effect sizes across methods likely reflects the heterogeneity among and within ADHD GWAS samples, and underscores that currently, PGS measurement lacks sufficient reliability and validity for use as a standalone instrument in ADHD clinical practice⁷.

Our results should be interpreted in the context of the study limitations. First, our sample size was modest, limiting power to conduct replication analyses or detect small effects. This is in part a result of our careful phenotyping and genotyping of the samples, which did not include any children with ASD or ID diagnoses. Additionally, detection of rare variants in the ADHD and control samples was carried out in a comprehensive manner. As expected, the number of rare variant cases in the sample was relatively small, which limited our interpretation of comparisons between cases with and without heterozygous variants. Yet, we did identify a large relative risk for rare variants in the ADHD sample (RR = 26), underscoring the importance of accounting for this etiology in genomic studies of the disorder. In future work, we will seek to replicate our results in larger, publicly available GS datasets.

In conclusion, pediatric ADHD is a phenotypically and genetically heterogeneous disorder. Polygenic risk and rare variants may represent distinct etiological pathways to ADHD, and PGS for IQ may increase predictive value for diagnostic likelihood. Rare variant discovery analysis supported our previous report that methylation genes play an important role in ADHD.

Data Availability

The data are available internally to all BCH researchers and clinicians through the genomics analysis platform. To facilitate external access, the CRDC data has also been made available to the Genomic Information Commons project (https://www.genomicinformationcommons.org/). Code available upon request. Clinical and genomic data are available to BCH researchers and their collaborators via request to the Biobank Sample and Data Access Committee. Through the Genomic Information Commons (GIC) (https://www.genomicinformationcommons.org/), PrecisionLink Biobank genomic data can be queried and requested by researchers at any of the established member sites.