henhaobio
Sign in

NatureGenomicsPeer-reviewed journal article

Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program

Daniel TaliuniD, Daniel HarrisiD, Michael D. KessleriD, Jedidiah CarlsoniD, Zachary A. SzpiechiD, Raul Torres, Sarah A. Gagliano TaliuniD, André CorveloiD, Stephanie M. GogarteniD, Hyun Min KangiD, Achilleas PitsillidesiD, Jonathon LeFaiveiD, Seung‐been LeeiD, Xiaowen Tian, Brian L. BrowningiD, Sayantan DasiD, Anne‐Katrin EmdeiD, Wayne E. ClarkeiD, Douglas P. LoeschiD, Amol Carl ShettyiD, Thomas W. BlackwelliD, Albert V. SmithiD, Quenna WongiD, Xiaoming LiuiD, Matthew P. ConomosiD, , Ruth J. F. LoosiD

PDFDOIPreprint
In-depth full-text digest available · Jump to digest ↓

Abstract

Summary paragraphThe Trans-Omics for Precision Medicine (TOPMed) program seeks to elucidate the genetic architecture and disease biology of heart, lung, blood, and sleep disorders, with the ultimate goal of improving diagnosis, treatment, and prevention. The initial phases of the program focus on whole genome sequencing of individuals with rich phenotypic data and diverse backgrounds. Here, we describe TOPMed goals and design as well as resources and early insights from the sequence data. The resources include a variant browser, a genotype imputation panel, and sharing of genomic and phenotypic data via dbGaP. In 53,581 TOPMed samples, >400 million single-nucleotide and insertion/deletion variants were detected by alignment with the reference genome. Additional novel variants are detectable through assembly of unmapped reads and customized analysis in highly variable loci. Among the >400 million variants detected, 97% have frequency <1% and 46% are singletons. These rare variants provide insights into mutational processes and recent human evolutionary history. The nearly complete catalog of genetic variation in TOPMed studies provides unique opportunities for exploring the contributions of rare and non-coding sequence variants to phenotypic variation. Furthermore, combining TOPMed haplotypes with modern imputation methods improves the power and extends the reach of nearly all genome-wide association studies to include variants down to ~0.01% in frequency.

In-depth summary

Claude Opus 5.5 · max effort · 8 Nov 2026 · from the full text of version 1 (6 Mar 2019) · CC BY 4.0

Over 400 million variants, 97% below 1% frequency, emerge from deep whole-genome sequencing of 53,831 diverse TOPMed participants; TOPMed haplotypes enable imputation down to ~0.01% frequency.

Rare variants dominate human genetic variation and are potentially more deleterious, yet earlier surveys relied on arrays, exomes, low-coverage genomes, or smaller, less ancestrally diverse samples. TOPMed contributes high-coverage genomes of phenotyped, diverse participants through dbGaP, a variant catalog in the BRAVO browser, and an imputation panel letting array-based studies reach far rarer variants than 1000 Genomes or Haplotype Reference Consortium panels.

Key results

  1. ShownThe catalog is dominated by novel, rare alleles; singleton fractions rise with predicted functional severity, and near-equal per-genome counts of non-synonymous and synonymous variants fit purifying selection purging many non-synonymous alleles before they become common.

    410,323,831 variants (381,343,078 single-nucleotide variants [SNVs], 28,980,753 indels) at mean depth 38.2X; 78.7% absent from dbSNP build 149; 46.0% singletons. Singleton share: frameshift 59.7%, non-synonymous 48.1%, synonymous 43.0%. Per genome: 3.78 million variants, 30,207 novel. Non-synonymous to synonymous 2.1:1 overall but 11,743 vs 11,768 per genome. · 410 million genetic variants in 53,831 samples

  2. ShownA TOPMed reference panel of 60,039 individuals (239,756,147 SNVs and indels) imputes rare variants far more accurately than 1000 Genomes or Haplotype Reference Consortium (HRC) panels, with gains across ancestries except South Asian.

    Mean imputation quality at frequency 0.001: African 0.91 vs ~0.15 (1000 Genomes, HRC); European 0.92 vs 0.26 (1000 Genomes) and 0.59 (HRC). Frequency floor for r>0.3: ~0.004-0.006% vs ~0.13-0.21% (1000 Genomes) and 0.05% (HRC, Europeans only). Rare (<0.5%) variants per genome imputable: African 93% vs 47% (1000 Genomes); European 95% vs 49% (1000 Genomes) and 72% (HRC). · TOPMed imputation resource; Evaluation of imputation accuracy

  3. ShownImputing UK Biobank with the TOPMed panel recovers far more rare putative loss-of-function (pLoF) variants than HRC imputation and detects associations (CHEK2, PALB2, BRCA2 burden with breast cancer; USH2A and IFT140 burdens with retinal dystrophy and kidney cysts) previously requiring sequencing.

    409,694 White British participants, 1,403 PheCode traits; 49,892 rare pLoF imputed vs 17,727 pLoF from HRC imputation. Breast cancer (12,636 cases, 200,417 controls): CHEK2 frameshift OR 2.04 (minor allele count 2,111), PALB2 stop-gain OR 4.48 (count 316), BRCA2 burden of 23 pLoF (cumulative frequency 0.11% in cases vs 0.05% in controls); all genome-wide significant. · TOPMed imputation resource; Imputation of the UK Biobank to the TOPMed panel…

  4. ShownDeep, jointly called genomes capture more variants than low-coverage whole-genome sequencing (WGS) or exomes and cover coding bases more fully, raising per-person pLoF discovery; disease-associated gene sets carry fewer rare pLoF variants than gene size predicts.

    430 Framingham Heart Study samples: 23.8 million variants vs 20.5 million from low-coverage WGS (16% increase); ~17% more coding variants than low-coverage WGS or exomes. Coding bases at >10X: ~98.8% in TOPMed vs ~88.9% in ExAC (Exome Aggregation Consortium). 228,966 pLoF variants in 18,493 genes; rare pLoF depleted in COSMIC (31%), OMIM (4%) and ClinVar (4%) genes. · TOPMed WGS; Protein loss of function variants; Sequencing depth at protein-coding regions; Low coverage WGS and high coverage WES in…

  5. SuggestedClosely spaced singletons far exceed independent-mutation expectations; an exponential mixture resolves four spacing classes consistent across ancestries, whose spectra suggest translesion synthesis (closest class) and double-strand-break repair (subtelomeric class).

    32,058,136 singleton SNVs in 2,000 unrelated individuals (equal African and European ancestry): 2% lie <100bp apart vs <0.07% expected if independent. Class 1 (~2-8 bp, ~1.5%) enriched for A>T and C>A; class 2 (~500-5,000 bp, ~12-24%) enriched for C>G with subtelomeric hotspots on 8p, 9p, 16p, 16q; classes 3 and 4 (~43-49%, ~31-37%) match background spectra. · Insights into mutation processes; Mixture model parameter estimation; Identification of cluster class 2 hotspots

Methods, data and reproducibility

TOPMed genomes were aligned to GRCh38 (BWA-MEM), jointly called (GotCloud, support-vector-machine site filter) and mined for constraint, singleton clustering (exponential mixtures), unmapped-read assembly, CYP2D6 star alleles, rare-variant and identity-by-descent (IBD) sharing, demography and selection; phased haplotypes formed a Minimac4 imputation panel.

  • TOPMed Freeze 5 deep WGS from 33 NHLBI-funded projects, excluding duplicates, monozygotic co-twins, low-depth or identity-questionable samples and incompatible consents · human · n = 53,831 · dbGaP controlled access (e.g., phs000956.v3.p1)
  • TOPMed imputation reference panel of phased haplotypes with singleton sites removed, spanning the autosomes and chromosome X · human · n = 60,039 individuals; 239,756,147 SNVs and indels
  • UK Biobank White British participants imputed with the TOPMed panel and tested against ICD-10-derived PheCodes · human (UK Biobank participants) · n = 409,694 (1,403 traits tested)
  • Framingham Heart Study participants with TOPMed WGS plus CHARGE low-coverage WGS and whole-exome data for platform comparison · human · n = 430 (263 unrelated)
  • Jackson Heart Study African American participants typed for CYP2D6 star alleles with Stargazer · human · n = 3,418 (1,923 unrelated for metabolizer estimates)
  • TOPMed Freeze 3 (GRCh37) genomes used for de novo assembly of unmapped and mismapped read pairs · human · n = ~18,000 (merged set described as ~17,000)
  • Unrelated subsets for singleton clustering (Freeze 5; African/European), demographic inference (Freeze 3; European) and integrated haplotype score (iHS) scans (Freeze 3) · human · n = 2,000; 2,416; 8,377, respectively
  • 1000 Genomes Phase 3 and Haplotype Reference Consortium panels as imputation baselines; Human Genome Diversity Project (HGDP) samples defining ancestry principal-component space · human · n = 2,504 (1000 Genomes); 32,470 (HRC); 938 (HGDP)
Methods & tools
BWA-MEM (alignment to hs38DH, GRCh38 with decoys), GotCloud joint calling with support vector machine site filter, Eagle 2.4 (haplotype phasing), Minimac4 (genotype imputation), LOFTEE v0.3-beta and VEP v94 (pLoF annotation, GENCODE v29), mixtools R package (expectation-maximization fit of exponential mixtures), Stargazer with GATK HaplotypeCaller/DepthOfCoverage and Beagle (CYP2D6), RefinedIBD and IBDNe (IBD segments, recent effective population size), moments (site frequency spectrum demographic inference), selscan (iHS selection scans), SAIGE (logistic mixed model association tests), PC-AiR, verifyBamID2, RFMIX (relatedness, ancestry); msprime (simulation)
Reagents & environment
Reads aligned to hs38DH (1000 Genomes GRCh38 build with decoys); GotCloud; Eagle 2.4/2.4.1; Minimac4 (m3vcf); LOFTEE v0.3-beta; VEP v94; GENCODE v29; Ensembl 94; dbSNP build 149; mixtools (R); SAIGE; selscan; moments; RefinedIBD/IBDNe
Experimental conditions
Illumina HiSeq X Ten at six centers (PCR-free KAPA libraries, target mean of at least 30X, 150-bp paired-end reads); 1,606 earlier genomes on HiSeq 2000/2500 (2 x 100 or 2 x 125 bp, sometimes PCR-amplified); no computing hardware or runtimes reported
Reproducibility gaps
Individual-level data are dbGaP controlled-access; no code repository or URLs given (center- and freeze-specific methods deferred to the TOPMed website); imputation panel only announced as forthcoming on the Michigan Imputation Server.
Limitations (6)
  • Genotype discordance estimates from 378 samples sequenced in duplicate (median 0.040% for passing SNVs) may be optimistic, since both members of each pair were processed and called jointly. · stated by the authors
  • Imputation gains did not extend to South Asians, who were sparsely represented in the panel (139 of 55,365 ancestry-assigned individuals); the authors expect a future panel with more South Asian samples to help. · stated by the authors
  • East Asian studies show the most singletons despite low heterozygosity; the authors propose a prolonged bottleneck then rapid growth, but note that modest East Asian sampling in TOPMed could also explain this. · stated by the authors
  • A lowered IBD LOD threshold (1.0 rather than the default 3.0), adopted because founder groups such as the Amish deflate LOD scores, inflates apparent IBD in low-variant-density regions, which had to be filtered out. · stated by the authors
  • Imputation accuracy was assessed only on chromosome 20, with 100 sequenced individuals per super-population and one array's marker set, and the UK Biobank rare pLoF associations were not replicated in an independent cohort. · noted in this summary
  • Analyses mix call sets (Freeze 3 on GRCh37; Freezes 5 and 6 on GRCh38) and subsets, and some counts disagree across sections (53,581 vs 53,831 samples; 40,273 vs 40,772 unrelated), complicating exact reproduction. · noted in this summary

Read this if you need ancestrally diverse whole-genome variant frequencies or a rare-variant imputation panel, or study functional constraint, mutation clustering, non-reference sequence or demography in human cohorts.

Citations
2,212
FWCI
190.17
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISKowalski '19Dai '25Higasa '25Jun '23Liu '25Huerta‐Chag '23
66% semantic matchGenomics
Madeline H. Kowalski, Huijun Qian, Ziyi Hou+76 more·PLoS Genetics·Dec 23, 2019· 277 citations