Full Breakdown
Heritability and Genetic Architecture of Postpartum Psychosis from All of Us Whole-Genome Sequencing
6/19/2026, 12:42:33 AM
Core Study and Primary Findings
A family-based quasi-experimental design applied Falconer’s Liability Threshold Model and PCGC regression to estimate genetic contribution to postpartum psychosis (ICD-10 F53.1) using whole-genome sequencing from the All of Us Research Program. The combined version-7 and version-8 datasets included 461 cases and 4,610 ancestry-matched controls; heritability was estimated assuming a 0.15% prevalence.
Context and Rationale
Postpartum psychosis occurs within three months after delivery. Prior studies have been limited by small samples and lack of genome-wide data; the All of Us cohort enables the first large-scale, population-based assessment of genetic liability for this condition.
Sample Selection and Quality Control
From the All of Us database, 431 postpartum psychosis cases were identified; 310 had whole-genome data, and additional cases from Swedish registers raised the total to 461. Controls were drawn from 170,161 women without mental-disorder diagnoses and down-sampled to ~5,600 ancestry-matched individuals using PCs 1-5. Quality control removed outlier variant counts, low call-rate samples, close relatives, and sex mismatches, leaving 301 cases and 8,628 controls for version-7 and 160 cases with 1,600 controls for version-8. A 1:10 case-control match produced the final analytic set of 461 cases and 4,610 controls.
Heritability Estimation
PCGC regression, which corrects for ascertainment bias, was applied to the European-ancestry subset (198 cases, 2,013 controls) with the first 20 PCs as covariates. Assuming a disease prevalence of 0.15%, the analysis yielded a point estimate of SNP-based heritability for postpartum psychosis.
Rare Variant Burden in HMGCR and DNMT1
Rare deleterious variants in HMGCR and DNMT1 were tested for association with psychiatric disorders in 240,009 All of Us whole-genome samples (excluding postpartum psychosis cases). Cases were defined by ICD-10 codes beginning with “F”. Burden testing used SKAT-O, adjusting for sex and PCs 1-10, with a Bonferroni-corrected significance threshold of p < 0.00069 (72 ICD-10 codes). Significant findings were replicated in the Mount Sinai BioMe Biobank (58,990 exome samples) using a Bonferroni-adjusted p < 0.05.
Gene-Level Association via TADA
De novo-like events—singletons in All of Us absent from gnomAD—were quantified. An unseen-species model estimated the total number of risk genes. The TADA framework combined rare coding variants (PTVs, missense categories, synonymous) and LOEUF scores to compute Bayes factors. Genes with FDR < 0.05 were classified as high-confidence risk genes for postpartum psychosis.
Limitations and Data Gaps
The analysis lacked parent-offspring trios, preventing direct detection of true de novo mutations. Heritability estimates were limited to European ancestry, restricting generalizability. Psychiatric cases were defined by ICD-10 codes.
Implications and Future Directions
The study provides a quantitative heritability estimate for postpartum psychosis and identifies high-confidence risk genes, suggesting shared genetic pathways with other psychiatric and autoimmune disorders. Future work will broaden ancestry representation, incorporate longitudinal phenotyping, and validate risk genes in independent cohorts.
