OTTERS: a powerful TWAS framework leveraging summary-level reference data

Abstract

Most existing TWAS tools require individual-level eQTL reference data and thus are not applicable to summary-level reference eQTL datasets. The development of TWAS methods that can harness summary-level reference data is valuable to enable TWAS in broader settings and enhance power due to increased reference sample size. Thus, we develop a TWAS framework called OTTERS (Omnibus Transcriptome Test using Expression Reference Summary data) that adapts multiple polygenic risk score (PRS) methods to estimate eQTL weights from summary-level eQTL reference data and conducts an omnibus TWAS. We show that OTTERS is a practical and powerful TWAS tool by both simulations and application studies.

Fig. 1. OTTERS framework.

OTTERS estimates cis-eQTL weights from eQTL summary data and reference LD panel using four imputation models (Stage I), and conducts ACAT-O test to combine gene-based association test p values from individual methods with individual/summary-level test GWAS data (Stage II).

Fig. 2. Test R² (A) and TWAS power (B) comparison in simulation studies.

Various scenarios with proportions of true causal cis-eQTL $p_{causal}=\left(0.001,0.01\right)$ and gene expression heritability $h_e^2=\left(0.01,0.05,0.1\right)$ were considered in the simulation studies. Distribution of test R² in 5000 simulations per method per scenario was presented using box-plot (A). The median was shown as a black bar. The lower and upper hinges corresponded to the 25th and 75th percentiles. Whiskers extended from the hinge to the value no further than 1.5 of the interquartile range. Data beyond the end of the whiskers were plotted individually. The GWAS sample size in the x-axis of panel B was chosen with respect to $h_e^2$ values. The proportion of phenotype variance explained by gene expression ($h_p^2$) was set to be 0.025. TWAS was conducted using simulated GWAS Z-scores.

Fig. 3. Test R² by PRS-CS versus P+T (0.001), P+T (0.05), lassosum, SDPR and FUSION.

Test R² by PRS-CS versus P+T (0.001) (A), P+T (0.05) (B), lassosum (C), SDPR (D), and FUSION (E) with 315 GTEx V8 test samples, with different colors denoting whether test R² > 0.01 only by PRS-CS (red), only by the y-axis method (green), or both methods (blue). Genes with test R² > 0.01 by at least one method were included in the plot.

Fig. 4. Manhattan plot of TWAS results by OTTERS.

Manhattan plot of TWAS results by OTTERS using GWAS summary-level statistics of cardiovascular disease and imputation models fitted based on eQTLGen summary statistics. The x-axis represented the genomic position, and the y-axis represented –log₁₀(p values). p values were the genomic-control corrected p values from the Z-score test from TWAS (two-sided). Independently significant TWAS risk genes were labeled.