Genome wide association analysis in a mouse advanced intercross line

Abstract

The LG/J x SM/J advanced intercross line of mice (LG x SM AIL) is a multigenerational outbred population. High minor allele frequencies, a simple genetic background, and the fully sequenced LG and SM genomes make it a powerful population for genome-wide association studies. Here we use 1,063 AIL mice to identify 126 significant associations for 50 traits relevant to human health and disease. We also identify thousands of cis- and trans-eQTLs in the hippocampus, striatum, and prefrontal cortex of ~200 mice. We replicate an association between locomotor activity and Csmd1, which we identified in an earlier generation of this AIL, and show that Csmd1 mutant mice recapitulate the locomotor phenotype. Our results demonstrate the utility of the LG x SM AIL as a mapping population, identify numerous novel associations, and shed light on the genetic architecture of mammalian behavior.

Fig. 1

Variants, MAFs and LD decay in the LG × SM AIL. Imputation provided 4.3 million SNPs. Filtering for LD (r² ≥ 0.95), MAF < 0.1, and HWE (p ≤ 7.62 × 10⁻⁶, Chi-squared test) resulted in 523,028 SNPs for GWAS. a SNP distribution and density of GWAS SNPs are plotted in 500 kb windows for each chromosome. As shown in Supplementary Figure 1, regions with low SNP density correspond to regions predicted to be nearly identical by descent in LG and SM. b MAF distributions are shown for 4.3 million imputed SNPs (gold; unfiltered) and for the 523,028 SNPs used for GWAS (orange; filtered). Mean MAF is the same in both SNP sets. c Comparison of LD decay in G50–56 (dark purple) and G34 (light purple) of the LG × SM AIL. Each curve was plotted using the 95th percentile of r² values for SNPs spaced up to 5 Mb apart

Fig. 2

GWAS for albinism verifies that the LOCO-LMM effectively controls type I and type II error. We conducted a GWAS for albinism, a Mendelian trait caused by the Tyr^c locus on mouse chromosome 7, using three models: a linear model, an LMM, and a LOCO-LMM. We also repeated each scan after excluding SNPs on chromosome 7. A Manhattan plot of results from the LOCO-LMM is shown in (a). Quantile-quantile plots of expected vs. observed p-values are shown for (b) a simple linear model that does not account for relatedness; c a standard LMM that includes all GWAS SNPs in the GRM (i.e., the random effect); and d a LOCO-LMM whose GRM excludes SNPs located on the chromosome being tested. Plots e–g show results after excluding chromosome 7 from the GWAS

Fig. 3

Manhattan plot and heritability for 120 traits measured in the LG × SM AIL. We identified 126 loci for behavioral and physiological traits using 1063 mice from G50–56 of the LG × SM AIL. A Manhattan plot of GWAS results is shown in (a). Associations for related traits are grouped by color. For clarity, related traits that mapped to the same locus (Supplementary Data 1) are highlighted only once. The dashed line indicates a permutation-derived significance threshold of −log₁₀(p) = 5.09 (p = 8.06 × 10⁻⁶; α = 0.05; likelihood ratio test). b For a representative subset of traits, SNP heritability estimates (percent trait variance explained by 523,028 GWAS SNPs) are shown. Precise estimates of heritability with standard error are provided for all traits in Supplementary Data 2

Fig. 4

Replication of an association between Csmd1 and locomotor activity. a Regional plot drawn from the full set of 4.3 million unpruned SNPs showing the association between rs33436747 and D5 activity levels (12,058 SNPs are plotted for this region). The location of Csmd1, 1.5-LOD interval (gold bar), areas of elevated recombination (green plus symbols), regions predicted to be nearly IBD between LG and SM (gray bars), and SNP MAFs (gray heatmap) are indicated. Points are colored by LD (r²) with rs33436747 The dashed line indicates a significance threshold of −log₁₀(p) = 5.09 (α = 0.05; likelihood ratio test). b PheWAS plot of associations between rs33436747 and other behavioral traits measured in G50–56 mice. Violin plots in (c–e) display the interquartile range with the median plotted as a horizontal line and the mean plotted as a point. P-values from ANOVA and two-sided t-tests are shown for each comparison. c Violin plot of quantile-normalized residuals of locomotor activity at the Csmd1 locus are plotted for G50–56 mice (ANOVA p = 7.85e−06; F = 11.89; df = 2, 1033). For each comparison, LG vs. SM p = 5.52e−04, t = −3.58, df = 94.34; LG vs. HET p = 1.11e−04, t = −3.88, df = 890.44; SM vs. HET p = 0.108, t = 1.62, df = 99.56. d Violin plot of quantile-normalized residuals of locomotor activity at the Csmd1 locus for G34 mice (ANOVA p = 3.51e-−09; F = −20.03; df = 2, 685). For each comparison, LG vs. SM p = 2.11e−04, t = −4.17, df = 32.45; LG vs. HET p = 6.72e−07, t = −5.03, df = 521.27; SM vs. HET p = 0.033, t = 2.22, df = 34.73. rs33436747 was not genotyped in G34; therefore, we plotted activity by genotype at the nearest SNP (rs33014260); 6764 bp upstream of rs33436747. e Violin plot of locomotor activity data (distance traveled in 0–30 min) for Csmd1 mutant (MUT) mice (ANOVA p = 0.034; F = 3.46; df = 2, 135). For each comparison, WT vs. MUT p = 0.026, t = −2.27, df = 71.03; WT vs. HET p = 0.519, t = −0.649, df = 68.20; MUT vs. HET p = 0.058, t = −1.93, df = 67.62. Data plotted in (c–e) is provided as a Source Data file