Skip to content

RatXcan

Summary

RatXcan extends the PrediXcan/TWAS framework to outbred heterogeneous stock (HS) rats, a well-characterized mammalian model organism population, enabling direct tests of whether gene-trait relationships discovered in human GWAS are conserved in another mammal. Because laboratory rat colonies have much higher relatedness than typical human GWAS cohorts, RatXcan adds a random effect encoding genetic relatedness (equivalent to the infinitesimal polygenic model), which simultaneously corrects for polygenicity-driven inflation — the same statistical issue identified in a companion paper from the same lab. Applied to body length and BMI, RatXcan-derived rat gene associations were significantly enriched among the orthologous human height and BMI genes reported by PhenomeXcan.[1]

Method

  1. Training stage: gene expression prediction models trained from genotype and transcriptome data across five rat brain regions (88 HS rats).
  2. Association stage: predicted transcriptome estimated in an independent target set of 5,401 densely genotyped HS rats (from prior body length and BMI GWAS), tested for association with each trait using a mixed-effects model with a random effect capturing both genetic relatedness and the infinitesimal polygenic background — jointly addressing relatedness-driven false positives and the polygenicity-driven inflation previously reported for standard TWAS.
  3. Cross-species comparison: rat gene associations (combined across the five brain tissues via the ACAT method, chosen for robustness to misspecified tissue correlations) compared against human gene associations for the orthologous trait, queried from PhenomeXcan (which aggregates via MultiXcan across GTEx tissues).

Findings

  • Genetic architecture parallels: cis-heritability of gene expression in rat brain tissues (mean 13.5–15.5%) closely resembled that of orthologous genes in human GTEx whole blood, and both species showed better cross-validated prediction with sparser (more lasso-like) elastic-net models than fully polygenic (ridge-like) ones — the underlying cis-regulatory architecture of gene expression appears broadly similar between rats and humans.
  • Trait associations: 11 Bonferroni-significant genes (3 loci: chromosomes 7, 10, 14) for rat body length; 10 significant genes (3 loci: chromosomes 1, 10, 18) for rat BMI. The strongest BMI hit, Prlhr (P=1.52×10⁻⁹), replicated a known obesity/energy-expenditure gene from independent rat GWAS cohorts; its human ortholog PRLHR was independently associated with human BMI (P=1.76×10⁻⁶) and body fat percentage in PhenomeXcan.
  • Cross-species enrichment: genes significantly associated with rat body length were significantly enriched among human height-associated genes (Fisher test, P=0.016), and rat BMI genes were similarly enriched among human BMI genes (Fisher test, P=0.013) — direct empirical evidence that gene-trait relationships are at least partly conserved between rats and humans for these traits, relevant to the broader debate over the validity of rodent models for human complex-trait genetics.

Availability

Code: github.com/hakyimlab/rat_genomics_paper_pipeline_2024 (archived on Zenodo; data on Zenodo). Rat brain-tissue expression prediction models: predictdb.org. Interactive results browser: imlab.shinyapps.io/RatXcan.

See Also

Citations

[1] Santhanam, N., Sanchez-Roige, S., Mi, S., Liang, Y., Chitre, A.S., Munro, D., et al., Palmer, A.A., & Im, H.K. (2025). RatXcan: A framework for cross-species integration of genome-wide association and gene expression data. PLOS Genetics, 21(3), e1011583.