Deep-coverage whole genome sequences and blood lipids among 16,324 individuals

Pradeep Natarajan, Gina M. Peloso, Seyedeh M. Zekavat, May E. Montasser, Andrea Ganna, Mark Chaffin, Amit V. Khera, Wei Zhao, Jonathan M. Bloom, J Engreitz, Jason Ernst, Jeffrey R. O’Connell, Sanni Ruotsalainen, Maris Alver, Ani Manichaikul, W. Craig Johnson, James A. Perry, Timothy Poterba, Cotton Seed, Ida Surakka, Tōnu Esko, Samuli Ripatti, Veikko Salomaa, Adolfo Correa, Ramachandran S. Vasan, Manolis Kellis, Benjamin M. Neale, Eric S. Lander, Goncalo Abecassis, Braxton D. Mitchell, Stephen S. Rich, James G. Wilson, L. Adrienne Cupples, Jerome I. Rotter, Cristen J. Willer, Sekar KathiresanbioRxiv 2017 · DOI

Abstract

Deep-coverage whole genome sequencing at the population level is now feasible and offers potential advantages for locus discovery, particularly in the analysis rare mutations in non-coding regions. Here, we performed whole genome sequencing in 16,324 participants from four ancestries at mean depth >29X and analyzed correlations of genotypes with four quantitative traits – plasma levels of total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol, and triglycerides. We conducted a discovery analysis including common or rare variants in coding as well as non-coding regions and developed a framework to interpret genome sequence for dyslipidemia risk. Common variant association yielded loci previously described with the exception of a few variants not captured earlier by arrays or imputation. In coding sequence, rare variant association yielded known Mendelian dyslipidemia genes and, in non-coding sequence, we detected no rare variant association signals after application of four approaches to aggregate variants in non-coding regions. We developed a new, genome-wide polygenic score for LDL-C and observed that a high polygenic score conferred similar effect size to a monogenic mutation (~30 mg/dl higher LDL-C for each); however, among those with extremely high LDL-C, a high polygenic score was considerably more prevalent than a monogenic mutation (23% versus 2% of participants, respectively).

Introduction

The genetic architecture of blood lipid levels—a primary risk factor for cardiovascular disease—has traditionally been viewed through two lenses: rare, high-impact mutations in single genes (monogenic) and common variants with small individual effects (polygenic). While Genome-Wide Association Studies (GWAS) have identified hundreds of common loci, a significant portion of lipid heritability remains unexplained. This "missing heritability" was hypothesized to reside in the genome's "dark matter," specifically rare variants and non-coding regulatory regions that are often invisible to standard array-based studies.

By utilizing deep-coverage whole-genome sequencing (WGS) across a diverse population of 16,324 individuals, this study sought to bridge the gap between monogenic and polygenic risk, exploring whether comprehensive sequencing of non-coding regions could reveal new drivers of dyslipidemia.

Methods and Data Integration

The study sequenced the entire genomes of 16,324 participants at a mean depth of >29X, ensuring high accuracy for both common and rare variants. To ensure broad applicability, the researchers pooled data from several major cohorts, including the Framingham Heart Study (FHS), Old Order Amish (OOA), Jackson Heart Study (JHS), Multi-Ethnic Study of Atherosclerosis (MESA), FINRISK, and the Estonian Biobank. Validation was performed using the Nord-Trøndelag Health Study (HUNT) and the Atherosclerosis Risk in Communities (ARIC) study.

The analytical framework employed several sophisticated techniques:

  • Single Variant Association: To identify common variants, including those not captured by previous imputation panels.
  • Rare Variant Aggregation: Using the Sequence Kernel Association Test (SKAT) to group rare variants by gene or regulatory region to increase statistical power.
  • Non-Coding Analysis: Four distinct strategies were used to link non-coding variants to genes, including sliding windows and Activity by Contact (ABC) models, which utilize 3D genomic data to identify enhancers that interact with specific gene promoters.
  • Polygenic Score (GPS) Development: A Bayesian approach (LDpred) was used to construct a score based on 2 million variants to estimate cumulative genetic risk.

Results

Variant Discovery and Novel Loci

The sequencing effort identified a staggering 189 million unique variants. While most common variant signals aligned with previously known loci, WGS identified five novel lipid-associated loci. Notably, a 1-bp deletion at 9p24.1 was found to affect HDL-C; this variant had been previously overlooked by standard array-based methods, demonstrating the added value of deep sequencing.

The Non-Coding Paradox

Despite the application of advanced aggregation frameworks, the study found no significant rare variant association signals in non-coding regions (enhancers or promoters). This suggests that while non-coding regions are critical for gene regulation, the "target size" for high-impact rare mutations in these regions may be smaller than anticipated, or their effects may be too subtle to detect at the current sample size.

Monogenic vs. Polygenic Impact

A central finding of the study is the clinical equivalence of monogenic and polygenic risk. A "high" polygenic score (defined as the top 5% of the distribution) was associated with an increase of approximately 30 mg/dl in LDL cholesterol—an effect size nearly identical to carrying a classic Mendelian mutation in genes such as LDLR.

However, the prevalence of these two risk factors differed dramatically. Among participants with extremely high LDL-C, only 2% carried a monogenic mutation, whereas 23% had a high polygenic score. In European Americans, the Odds Ratio (OR) for extremely high LDL was 10.92 for monogenic mutations and 7.65 for a high polygenic score. This indicates that for the vast majority of individuals with severe dyslipidemia, the condition is driven by the cumulative effect of many common variants rather than a single "broken" gene.

Discussion and Limitations

The study highlights a significant "ancestry gap" in genomic medicine. The polygenic score was substantially more predictive in individuals of European ancestry than in those of African ancestry. This disparity underscores the urgent need for larger, more diverse sequencing efforts to ensure that polygenic risk assessment is equitable across populations.

Furthermore, the lack of findings in non-coding regions suggests that even 16,000 genomes may lack the statistical power to uncover rare non-coding burdens. Future research will likely require sample sizes in the hundreds of thousands to fully map the regulatory landscape of blood lipids.

In conclusion, while monogenic mutations remain a critical clinical focus, the high prevalence and significant impact of polygenic risk suggest that genome-wide scores should play a more prominent role in identifying and managing individuals at high risk for cardiovascular disease.

Summary updated

Abstract

Deep-coverage whole genome sequencing at the population level is now feasible and offers potential advantages for locus discovery, particularly in the analysis rare mutations in non-coding regions. Here, we performed whole genome sequencing in 16,324 participants from four ancestries at mean depth >29X and analyzed correlations of genotypes with four quantitative traits – plasma levels of total cholesterol, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol, and triglycerides. We conducted a discovery analysis including common or rare variants in coding as well as non-coding regions and developed a framework to interpret genome sequence for dyslipidemia risk. Common variant association yielded loci previously described with the exception of a few variants not captured earlier by arrays or imputation. In coding sequence, rare variant association yielded known Mendelian dyslipidemia genes and, in non-coding sequence, we detected no rare variant association signals after application of four approaches to aggregate variants in non-coding regions. We developed a new, genome-wide polygenic score for LDL-C and observed that a high polygenic score conferred similar effect size to a monogenic mutation (~30 mg/dl higher LDL-C for each); however, among those with extremely high LDL-C, a high polygenic score was considerably more prevalent than a monogenic mutation (23% versus 2% of participants, respectively).
BibTeX
@misc{natarajan2017deep,
  author = {Pradeep Natarajan and Gina M. Peloso and Seyedeh M. Zekavat and May E. Montasser and Andrea Ganna and Mark Chaffin and Amit V. Khera and Wei Zhao and Jonathan M. Bloom and J Engreitz and Jason Ernst and Jeffrey R. O’Connell and Sanni Ruotsalainen and Maris Alver and Ani Manichaikul and W. Craig Johnson and James A. Perry and Timothy Poterba and Cotton Seed and Ida Surakka and Tōnu Esko and Samuli Ripatti and Veikko Salomaa and Adolfo Correa and Ramachandran S. Vasan and Manolis Kellis and Benjamin M. Neale and Eric S. Lander and Goncalo Abecassis and Braxton D. Mitchell and Stephen S. Rich and James G. Wilson and L. Adrienne Cupples and Jerome I. Rotter and Cristen J. Willer and Sekar Kathiresan and Sekar Kathiresan},
  title = {{Deep-coverage whole genome sequences and blood lipids among 16,324 individuals}},
  year = {2017},
  doi = {10.1101/224378},
  howpublished = {bioRxiv},
}
Figure