Introduction
Height is heritable and provides insights into the genetic architecture of human traits. Thousands of common and low-frequency variants have been identified with height using GWAS. However, these common variants only explain a limited fraction of heritability. A recent study shows that rare variants (RVs) are a major source of the missing heritability of height. Large-scale whole-genome sequencing (WGS) studies, such as the multi-ethnic NHLBI Trans-Omics Precision Medicine (TOPMed) Program, enable the assessment of associations between height and rare variants across the genome, especially for the noncoding genome.

Hypothesis
Rare variant aggregations are associated with height.

Methods
We applied our newly developed STAARpipeline workflow to rare variant (MAF < 0.01) association analyses using 87,652 individuals from TOPMed Freeze 8 WGS data, including gene-centric analysis and non-gene-centric analysis using a variety of coding and noncoding masks. The gene-centric analysis provides five coding and eight noncoding functional categories. The non-gene-centric analysis includes sliding window analysis with fixed sizes and dynamic window analysis with data-adaptive sizes.

Results
In the gene-centric analysis of coding RVs, we identified 7 genome-wide significant associations at the Bonferroni-corrected level 5.00E-07 (=0.05/20,000/5). After conditioning on known height-associated variants, the association between missense RVs in ACAN remained significant at the same level 5.00E-07. In the gene-centric analysis of noncoding RVs, we identified 8 genome-wide significant associations at the Bonferroni-corrected level 3.12E-07 (=0.05/20,000/8). The association of RVs in the promoter of HMGA1 remained significant at the same level 3.12E-07 in conditional analysis by adjusting for known height-associated variants. In 2-kb sliding window analysis, we identified 25 genome-wide significant associations at the Bonferroni-corrected level 1.88E-08 (=0.05/2.66E06). After conditioning on known height-associated variants, the strengths of all associations reduced and 4 of these associations remained significant at level 1.00E-05. Two of them are in the coding region of ACAN, and the other two are in the upstream region of HMGA1. The results of dynamic window analysis are similar to sliding window analysis. We identified 11 genome-wide significant associations, and 4 of these associations remained significant at level 1.00E-05 in conditional analysis.

Summary
Two new RV associations, missense RVs in ACAN and RVs in the promoter of HMGA1 with height, were identified using the TOPMed WGS Freeze 8 data through STAARpipeline.