henhaobio
Sign in

NatureGenetic & Genomic MedicinePeer-reviewed journal article

Exome sequencing identifies rare coding variants in 10 genes which confer substantial risk for schizophrenia

Tarjinder SinghiD, Timothy PoterbaiD, David CurtisiD, Huda AkiliD, Mariam M. AlEissaiD, Jack D. BarchasiD, Nicholas James BassiD, Tim B. BigdeliiD, Gerome D. BreeniD, Evelyn J. BrometiD, P.F. BuckleyiD, William E. BunneyiD, Jonas Bybjerg‐GrauholmiD, William ByerleyiD, Sinéad B. ChapmaniD, Wei J. CheniD, Claire L. ChurchhouseiD, Nick J. CraddockiD, Caroline M. CusickiD, Lynn E. DeLisiiD, Sheila Dodge, Michael EscamillaiD, Saana EskelineniD, Ayman H. FanousiD, Stephen V. FaraoneiD, , David St ClairiD

PDFDOIPreprintData
In-depth full-text digest available · Jump to digest ↓

Abstract

By meta-analyzing the whole-exomes of 24,248 cases and 97,322 controls, we implicate ultra-rare coding variants (URVs) in ten genes as conferring substantial risk for schizophrenia (odds ratios 3 - 50, P < 2.14 x 10^-6), and 32 genes at a FDR < 5%. These genes have the greatest expression in central nervous system neurons and have diverse molecular functions that include the formation, structure, and function of the synapse. The associations of NMDA receptor subunit GRIN2A and AMPA receptor subunit GRIA3 provide support for the dysfunction of the glutamatergic system as a mechanistic hypothesis in the pathogenesis of schizophrenia. We find significant evidence for an overlap of rare variant risk between schizophrenia, autism spectrum disorders (ASD), and severe neurodevelopmental disorders (DD/ID), supporting a neurodevelopmental etiology for schizophrenia. We show that protein-truncating variants in GRIN2A, TRIO, and CACNA1G confer risk for schizophrenia whereas specific missense mutations in these genes confer risk for DD/ID. Nevertheless, few of the strongly associated schizophrenia genes appear to confer risk for DD/ID. We demonstrate that genes prioritized from common variant analyses of schizophrenia are enriched in rare variant risk, suggesting that common and rare genetic risk factors at least partially converge on the same underlying pathogenic biological processes. Even after excluding significantly associated genes, schizophrenia cases still carry a substantial excess of URVs, implying that more schizophrenia risk genes await discovery using this approach.

In-depth summary

Claude Opus 5.5 · max effort · 8 Nov 2026 · from the full text of version 1 (18 Sept 2020) · CC BY 4.0

Exomes of 24,248 schizophrenia cases and 97,322 controls implicate ultra-rare damaging variants in 10 genes, including GRIN2A and GRIA3, with odds ratios of 3 to 50.

Schizophrenia GWAS loci are mostly non-coding and risk-associated copy number variants span many genes, which hides the causal genes and direction of effect; earlier sequencing studies lacked power and robustly implicated only SETD1A. Pooling harmonized exomes at far larger scale lets ultra-rare coding variants, especially interpretable protein-truncating ones, pinpoint individual risk genes and link them to neurodevelopmental disorders and common-variant loci.

Key results

  1. ShownUltra-rare protein-truncating variants (PTVs) and damaging missense variants in 10 genes (SETD1A, CUL1, XPO7, TRIO, CACNA1G, SP4, GRIA3, GRIN2A, HERC1, RB1CC1) reach exome-wide significance for schizophrenia, with 22 more genes at false discovery rate (FDR) < 5%.

    24,248 cases vs 97,322 controls (46,885 of them gnomAD summary-count controls) plus de novo mutations from 3,402 trios; P < 2.14 × 10^-6 (0.05/23,321 tests); odds ratios 3 to 50 (wide CIs) for PTVs and Class I variants, i.e. PTVs plus MPC > 3 missense; de novo Class I rate ratio 8.3 (95% CI 4.9 to 13; P = 2.1 × 10^-11) in the 52 genes with case-control P < 0.001. · Abstract; Analysis approach; Risk genes implicated by rare protein-coding…

  2. ShownLarge-effect ultra-rare variants hit the glutamate receptor subunits GRIN2A (NMDA) and GRIA3 (AMPA). Genome-wide burden concentrates in postsynaptic and chromatin-modification gene sets and in brain-specific genes, which supports glutamatergic synaptic dysfunction in schizophrenia.

    GRIN2A P = 7.37 × 10^-7, Class I OR 24.1 (95% CI 5.36 to 221); GRIA3 P = 5.98 × 10^-7, Class I OR 20.1 (95% CI 4.28 to 188); 33 of 1,732 gene sets significant (P < 2.9 × 10^-5), e.g. postsynaptic density P = 1.2 × 10^-12, chromatin modification P = 1.8 × 10^-12; GTEx frontal cortex P = 1.63 × 10^-8. · Risk genes implicated by rare protein-coding…; Shared genes and processes identified by common…

  3. ShownSchizophrenia rare-variant burden is enriched in autism (ASD) and severe neurodevelopmental disorder (DD/ID) genes, yet 9 of the 10 top genes show little DD/ID de novo PTV signal; TRIO, GRIN2A and CACNA1G carry PTV-driven schizophrenia risk but clustered de novo missense DD/ID risk.

    Burden in 299 DD/ID genes: Pmeta = 1.5 × 10^-14, Class I OR 1.44 (95% CI 1.3 to 1.6); in 102 ASD genes: Pmeta = 3.7 × 10^-7, OR 1.45 (1.23 to 1.72). Across 31,058 DD/ID trios, SETD1A had 8 de novo PTVs vs 0.41 expected (P = 1.3 × 10^-8) but the other nine top genes only 8 vs 4.98 (P = 0.13); six of 22 further FDR < 5% genes are DD/ID-associated. · Shared and distinct genetic risk with other…; Discussion

  4. ShownGenes prioritized by fine-mapping schizophrenia GWAS loci carry excess ultra-rare coding variants in cases, and exome signals help prioritize genes within multi-gene GWAS loci, indicating partial convergence of common and rare variant risk.

    69 fine-mapped protein-coding genes from 270 loci (GWAS of 69,369 cases, 236,642 controls): Pmeta = 1.1 × 10^-3, Class I OR 1.42 (95% CI 1.17 to 1.73). GRIN2A (GWAS OR = 1.057) and SP4 have clear GWAS signals; FAM120A and STAG1 (FDR < 5%) stand out among neighbors in multi-gene GWAS regions. · Risk genes implicated by rare protein-coding…; Shared genes and processes identified by common…

  5. ShownAbout 5.7% of cases are estimated to carry a risk-relevant ultra-rare PTV in loss-of-function-intolerant genes, mostly inherited rather than de novo. The burden persists after removing all 32 FDR < 5% genes, implying most rare-variant risk genes remain undiscovered.

    In 3,063 pLI > 0.9 genes, 0.057 (95% CI 0.049 to 0.065) extra singleton PTVs per case; in 2,304 trios, 0.0394 inherited (74%) vs 0.0121 de novo (26%) per proband, and 7 de novo plus 13 transmitted variants in the 10 genes (0.86% of patients carriers); constrained-gene OR 1.26 (P = 7.6 × 10^-35) remains 1.23 (P = 4.3 × 10^-27) after excluding the 32 genes. · Data description, generation, and quality…; Contribution of ultra-rare PTVs to schizophrenia…

Methods, data and reproducibility

Exomes from eleven collections were jointly called (BWA-Picard-GATK, alongside gnomAD) and supplemented with gnomAD control counts; per-gene Fisher's exact burden tests with within-stratum permutation of ultra-rare PTVs and MPC-defined damaging missense variants were meta-analyzed with de novo Poisson tests via weighted Z-scores.

  • SCHEMA (Schizophrenia Exome Sequencing Meta-Analysis) exomes of cases and controls without known psychiatric diagnosis, eleven global collections; 16,269 cases newly reported · human · n = 24,248 cases; 50,437 controls
  • Summary-level variant counts from gnomAD controls in non-psychiatric, non-neurological collections, jointly called with SCHEMA samples · human · n = 46,885 controls
  • Re-annotated de novo mutations from published schizophrenia parent-proband trios, plus 2,304 trios used to split PTV burden into inherited and de novo parts · human · n = 3,402 trios; 2,304 trios
  • Published de novo mutations from severe neurodevelopmental disorder (DD/ID) trios, implicating 299 genes · human · n = 31,058 trios
  • Published autism spectrum disorder (ASD) exome study implicating 102 genes at FDR < 10% · human · n = 11,986 ASD cases
  • Companion Psychiatric Genomics Consortium (PGC) schizophrenia GWAS: 270 loci, 69 with a fine-mapped protein-coding gene · human · n = 69,369 cases; 236,642 controls
  • Expression references: GTEx tissues, BrainSpan developmental human brain series, and mouse nervous system cell types from single-cell data · human (GTEx, BrainSpan); mouse (single-cell data)
  • 1000 Genomes samples with reported ancestry, plotted as reference in principal components ancestry analysis · human (1000 Genomes)
Methods & tools
BWA-Picard-GATK joint variant-calling pipeline, gnomAD v2 callset and pLI loss-of-function intolerance metric, MPC missense pathogenicity score, Fisher's exact gene burden test with within-stratum case-control permutation, Mantel-Haenszel test, Poisson rate test for de novo mutation enrichment, Weighted Z-score meta-analysis; Fisher's combined P value, Mann-Whitney U test, Principal components analysis for ancestry inference, Gene Ontology, REACTOME and KEGG gene sets, SynGO synaptic ontology, Statistical fine-mapping of GWAS loci (gene prioritization)
Reagents & environment
Standardized BWA-Picard-GATK pipeline for joint re-processing with gnomAD (permitted SCHEMA samples are in the gnomAD v2 release); analysis limited to coding exons well covered across major exome capture kits; tool versions not given in the main text.
Reproducibility gaps
No analysis code repository is cited. Results are shared as tables and through the SCHEMA browser, access to individual-level genotypes is not described, and the 46,885 external gnomAD controls were used only as summary-level counts.
Limitations (6)
  • Gene-level effect sizes are imprecise; the authors note wide confidence intervals (e.g., GRIN2A Class I OR 24.1, 95% CI 5.36 to 221; GRIA3 OR 20.1, 95% CI 4.28 to 188). · stated by the authors
  • The 46,885 external gnomAD controls offered only summary-level counts, so batch artifacts had to be handled indirectly via joint calling, coverage-filtered exons, within-stratum permutation and synonymous-variant null checks. · stated by the authors
  • De novo data were sparse despite 3,402 trios (325 genes with a de novo PTV; 449 with a Class I or II mutation), and de novo P values were combined with case-control results only for the 244 genes with case-control P < 0.01. · stated by the authors
  • Gene discovery is incomplete: a significant excess of ultra-rare PTVs remains outside the 32 FDR < 5% genes, and the authors state that most risk genes, effect directions and mechanisms are still unknown. · stated by the authors
  • Gene-level odds ratios (3 to 50) come from the same samples used for discovery at P < 2.14 × 10^-6, so they may be inflated by winner's curse; no independent case-control replication is reported. · noted in this summary
  • The inherited versus de novo split of PTV burden (74% vs 26%) rests on 2,304 trios with wide intervals (de novo excess 0.0121, 95% CI 0.0022 to 0.02), so the relative contributions are imprecise. · noted in this summary

Read this if you work on psychiatric or neurodevelopmental genetics and want gene-level rare-variant evidence for schizophrenia, its overlap with autism, DD/ID and GWAS loci, or exome meta-analysis design.

Citations
979
FWCI
125.66
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISSealock '26Singh '16Trubetskoy '22Chick '24Akingbuwa '22Lescai '19
84% semantic matchGenetics
Julia Sealock, Connor Dowd, Calwing Liao+96 more·bioRxiv·Sep 21, 2026· 1 citations