Pharmaceutical drugs targeting dyslipidemia and cardiovascular disease (CVD) may increase the risk of fatty liver disease and other metabolic disorders. To identify potential novel CVD drug targets without these adverse effects, we perform genome-wide analyses of participants in the HUNT Study in Norway (n = 69,479) to search for protein-altering variants with beneficial impact on quantitative blood traits related to cardiovascular disease, but without detrimental impact on liver function. We identify 76 (11 previously unreported) presumed causal protein-altering variants associated with one or more CVD- or liver-related blood traits. Nine of the variants are predicted to result in loss-of-function of the protein. This includes ZNF529:p.K405X, which is associated with decreased low-density-lipoprotein (LDL) cholesterol (P = 1.3 × 10⁻⁸) without being associated with liver enzymes or non-fasting blood glucose. Silencing of ZNF529 in human hepatoma cells results in upregulation of LDL receptor and increased LDL uptake in the cells. This suggests that inhibition of ZNF529 or its gene product should be prioritized as a novel candidate drug target for treating dyslipidemia and associated CVD.

Meta-analysis of genetic association studies increases sample size and the power for mapping complex traits. Existing methods are mostly developed for datasets without missing values, i.e. the summary association statistics are measured for all variants in contributing studies. In practice, genotype imputation is not always effective. This may be the case when targeted genotyping/sequencing assays are used or when the un-typed genetic variant is rare. Therefore, contributed summary statistics often contain missing values. Existing methods for imputing missing summary association statistics and using imputed values in meta-analysis, approximate conditional analysis, or simple strategies such as complete case analysis all have theoretical limitations. Applying these approaches can bias genetic effect estimates and lead to seriously inflated type-I or type-II errors in conditional analysis, which is a critical tool for identifying independently associated variants. To address this challenge and complement imputation methods, we developed a method to combine summary statistics across participating studies and consistently estimate joint effects, even when the contributed summary statistics contain large amounts of missing values. Based on this estimator, we proposed a score statistic called PCBS (partial correlation based score statistic) for conditional analysis of single-variant and gene-level associations. Through extensive analysis of simulated and real data, we showed that the new method produces well-calibrated type-I errors and is substantially more powerful than existing approaches. We applied the proposed approach to one of the largest meta-analyses to date for the cigarettes-per-day phenotype. Using the new method, we identified multiple novel independently associated variants at known loci for tobacco use, which were otherwise missed by alternative methods. Together, the phenotypic variance explained by these variants was 1.1%, improving that of previously reported associations by 71%. These findings illustrate the extent of locus allelic heterogeneity and can help pinpoint causal variants. It is of great interest to estimate the joint effects of multiple variants from large scale meta-analyses, in order to fine-map causal variants and understand the genetic architecture for complex traits. The summary association statistics from participating studies in a meta-analysis often contain missing values at some variant sites, as the imputation methods may not work well and the variants with low imputation quality will be filtered out. Missingness is especially likely when the underlying genetic variant is rare or the participating studies use targeted genotyping array that is not suitable for imputation. Existing methods for conditional meta-analysis do not properly handle missing data, and can incorrectly estimate correlations between score statistics. As a result, they can produce highly inflated type-I errors for conditional analysis, which will result in overestimated phenotypic variance explained and incorrect identification of causal variants. We systematically evaluated this bias and proposed a novel partial correlation based score statistic. The new statistic has valid type-I errors for conditional analysis and much higher power than the existing methods, even when the contributed summary statistics contain a large fraction of missing values. We expect this method to be highly useful in the sequencing age for complex trait genetics.

There is great interest in understanding the impact of rare variants in human diseases using large sequence datasets. In deep sequence datasets of >10,000 samples, ~10% of the variant sites are observed to be multi-allelic. Many of the multi-allelic variants have been shown to be functional and disease-relevant. Proper analysis of multi-allelic variants is critical to the success of a sequencing study, but existing methods do not properly handle multi-allelic variants and can produce highly misleading association results. We discuss practical issues and methods to encode multi-allelic sites, conduct single-variant and gene-level association analyses, and perform meta-analysis for multi-allelic variants. We evaluated these methods through extensive simulations and the study of a large meta-analysis of ~18,000 samples on the cigarettes-per-day phenotype. We showed that our joint modeling approach provided an unbiased estimate of genetic effects, greatly improved the power of single-variant association tests among methods that can properly estimate allele effects, and enhanced gene-level tests over existing approaches. Software packages implementing these methods are available online.