henhao
Sign in
GeneticsbioRxiv
Preprint

Genetics of 38 blood and urine biomarkers in the UK Biobank

Nasa Sinnott-ArmstrongiD, Yosuke TanigawaiD, David AmariD, Nina MarsiD, Matthew AguirreiD, Guhan Ram VenkataramaniD, Michael WainbergiD, Hanna Maria OllilaiD, James P. PirruccelloiD, Junyang Qian, Anna ShcherbinaiD, FinnGeniD, Fátima RodrígueziD, Themistocles L. AssimesiD, Vineeta Agarwala, Robert John TibshiraniiD, Trevor J. HastieiD, Samuli RipattiiD, Jonathan K. PritchardiD, Mark J. DalyiD, Manuel A. RivasiD

Abstract

Clinical laboratory tests are a critical component of the continuum of care and provide a means for rapid diagnosis and monitoring of chronic disease. In this study, we systematically evaluated the genetic basis of 38 blood and urine laboratory tests measured in 358,072 participants in the UK Biobank and identified 1,857 independent loci associated with at least one laboratory test, including 488 large-effect protein truncating, missense, and copy-number variants. We tested these loci for enrichment in specific single cell types in kidney, liver, and pancreas relevant to disease aetiology. We then causally linked the biomarkers to medically relevant phenotypes through genetic correlation and Mendelian randomization. Finally, we developed polygenic risk scores (PRS) for each biomarker and built multi-PRS models using all 38 PRSs simultaneously. We found substantially improved prediction of incidence in FinnGen (n=135,500) with the multi-PRS relative to single-disease PRSs for renal failure, myocardial infarction, liver fat percentage, and alcoholic cirrhosis. Together, our results show the genetic basis of these biomarkers, which tissues contribute to the biomarker function, the causal influences of the biomarkers, and how we can use this to predict disease.

PDFDOICodeData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
102
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (5 Jun 2019) · CC BY 4.0

UK Biobank genetics of 38 blood and urine lab tests yields 1,857 independent loci; combining biomarker and disease polygenic scores improves FinnGen prediction of incident renal failure and alcoholic cirrhosis.

Apart from lipids, glycaemic traits and kidney-function markers, most routine clinical lab biomarkers had little measured genetic contribution. This work characterizes 38 UK Biobank serum and urine tests across coding, non-coding, HLA and copy-number variants and connects them to disease through genetic correlation, Mendelian randomization and combined polygenic scores tested in FinnGen.

Key results

  1. Shown1,857 independent loci are associated with at least one of the 38 biomarkers, including 488 large-effect protein-truncating, missense or copy-number variants, with population structure apparently well controlled.

    358,072 participants; association tests in 318,984 unrelated White British individuals over 27,816 protein-truncating, 87,407 protein-altering and 9,444,561 non-coding variants gave over 7,000 significant associations; LD Score intercepts 0.929 to 1.33; effect-size correlation > 0.5 with prior studies in 33 comparisons (42 comparison studies, 25 biomarkers). · Abstract; Genetics of biomarkers

  2. ShownAdding the 38 biomarker polygenic scores (PRS) to a trait's own PRS improves prediction. In FinnGen the gain is clearest for renal failure and alcoholic cirrhosis, where the trait's own PRS alone shows no association with incidence.

    FinnGen hazard ratio per SD, snpnet PRS vs multi-PRS: renal failure 0.99 (p = 0.46) vs 1.14 (p = 9.3 × 10^−13); alcoholic cirrhosis 1.00 (p = 0.98) vs 1.14 (p = 1.8 × 10^−4); myocardial infarction 1.18 vs 1.26. UK Biobank: liver fat adjusted r-squared 0.024, 0.050, 0.087 (covariates, + own PRS, + 38 biomarker PRSs); highest-risk renal failure prevalence nearly 30% (multi-PRS) vs 5% (own PRS). · Abstract; Multiple regression with PRSs for biomarkers…; Evaluation of multi-PRS prediction in an…

  3. ShownRare large-effect coding variants and copy-number variants (CNVs) tie biomarker shifts to disease risk, for example a protein-truncating variant (PTV) in COL4A4, an inframe deletion in GOT1 and rare CNVs overlapping HNF1B.

    123 PTV (48 rare) and 2,737 protein-altering associations outside the major histocompatibility complex at p < 1 × 10^−7. GOT1 deletion (MAF 0.05%): 2.4 sd lower aspartate aminotransferase (AST), pancreatitis OR 4.55. COL4A4 PTV: kidney problems OR 6.9 (p = 6.7 × 10^−13). HNF1B CNV burden: urea, eGFR, creatinine, cystatin C (p < 8.8 × 10^−13); renal failure OR 4.94 (p = 1 × 10^−7). · Biomarker associated variants prioritize…; CNVs influencing biomarkers

  4. ShownBiomarker heritability concentrates in kidney, liver and pancreas annotations. Single-cell marker genes narrow this to podocytes, tubule cells and hepatocytes, plus an enrichment near mesangial cell markers across many traits.

    Stratified LD Score regression: 20 of 38 biomarkers at least 20-fold enriched (p < 1 × 10^−5) in kidney, pancreas or liver. Marker genes (extended 100 kb) of scRNA-seq clusters from liver (adult human), kidney (adult mouse, fetal and adult human) and pancreas (adult human): enrichment in podocytes, proximal and distal tubule, hepatocytes, mesangial cells; gamma-delta T cells for HbA1c. · Cell type decomposition of genetic effects; Cell-type enrichment analysis; Single-cell enrichment analysis

  5. SuggestedMendelian randomization (MR-Egger) and the latent causal variable (LCV) model nominate causal links from biomarkers to disease. They recover expected links such as LDL cholesterol on coronary artery disease and propose others such as AST on hernia.

    Genetic correlation: 1,127 significant pairs between the 38 biomarkers and 146 traits (p < 1 × 10^−3). MR-Egger: 86 causal relationships at 10% FDR; LCV recovers 49 and adds 344. LDL on coronary artery disease 0.34 log odds per SD (LCV causal percent 0.75); urate on gout 1.2 LOR/SD; calcium on kidney stones 0.625 LOR/SD; AST on hernia −0.319 LOR/SD. · Correlation of genetic effects between…; Causal inference; Causal inference

Methods, data and reproducibility

In UK Biobank, 38 biomarkers adjusted for 127 covariates and statins were tested against coding, imputed non-coding, HLA and copy-number variants, then followed up with heritability, cell-type enrichment, phenome-wide association, genetic correlation and causal (MR-Egger, LCV) analyses. snpnet lasso polygenic scores were combined in multi-PRS regressions and tested in FinnGen with Cox models.

  • UK Biobank participants with 38 serum and urine biomarker phenotypes (32 measured, 2 derived continuous, 4 binary) · human · n = 358,072 (318,984 unrelated White British in the association analysis)
  • UK Biobank non-White-British groups used to test polygenic score transfer · human · n = 24,131 non-British White; 6,951 South Asian; 1,950 East Asian; 6,056 African
  • UK Biobank MRI-derived liver fat percentage · human · n = 4,617
  • FinnGen Data Freeze 3, registry-based incident disease (renal failure, myocardial infarction, cirrhosis) · human (Finnish participants) · n = 135,500 (abstract, results); 135,300 (Methods)
  • Published scRNA-seq cluster marker genes for liver, kidney and pancreas · human (liver, kidney, pancreas); mouse (adult kidney)
  • External summary statistics for diseases and medically relevant phenotypes used in genetic correlation and causal analyses · n = 146 (56 diseases, 90 phenotypes)
  • 1000 Genomes Phase 3 European LD reference · human (1000 Genomes reference) · n = 489
  • Prior biomarker association studies used for effect-size comparison · n = 42 comparison studies covering 25 biomarkers
Methods & tools
PLINK v2.00aLM / PLINK 1.9, VEP LOFTEE, PennCNV v1.0.4, HLA:IMP*2, bma (R package), LD Score regression (ldsc), HESS, TwoSampleMR (MR-Egger), LCV, snpnet (BASIL) with glmnet, Beagle 4.1 (SISu v3 panel), Global Biobank Engine
Code
https://github.com/rivas-lab/public-resources/tree/master/uk_biobank/biomarkers
Data & accessions
https://biobankengine.stanford.edu/
Software & environment
Versions given: PLINK v2.00aLM (2 April 2019, 17 Jul 2017, 31 Mar 2018 builds), PLINK 1.9, PennCNV v1.0.4; ldsc and plink commands printed; no container stated. FinnGen imputation followed protocol dx.doi.org/10.17504/protocols.io.nmndc5e (SISu v3 panel, Beagle 4.1).
Reproducibility gaps
The text says only that results are 'displayed' in the Global Biobank Engine and gives no download or access route for UK Biobank or FinnGen data. The GCTA-GREML heritability cross-check and the simulations behind the MR log-odds scaling are reported as 'data not shown'.
Limitations (6)
  • Biomarker polygenic scores lose accuracy outside the White British training group: median performance reductions of 6.4%, 30.0%, 43.2% and 72.0% in non-British White, South Asian, East Asian and African UK Biobank participants. · stated by the authors
  • Mendelian randomization can be confounded by shared etiology and pleiotropic instruments; LCV was added to separate genetic correlation from causation. · stated by the authors
  • Whether the combined risk models transfer still needs testing in more independent biobanks and prospective cohorts; larger datasets from under-represented populations are needed for generalizability. · stated by the authors
  • For lipoprotein(a) and total bilirubin, snpnet did not reach the optimal penalty within the default 100 iterations, so the 100th-iteration model was used. · stated by the authors
  • Multi-PRS outcomes were chosen after inspecting disease enrichment at the tails of UK Biobank score distributions, and the FinnGen myocardial infarction comparison (1.18 vs 1.26) is reported without p-values. · noted in this summary
  • The FinnGen sample size differs between the abstract and results (135,500) and the Methods (135,300), and two supporting checks (GCTA-GREML, MR scaling simulations) are not shown. · noted in this summary

Who should read:Read this if you need variant-level genetic associations for routine blood and urine lab tests, or want to see whether combining biomarker polygenic scores improves disease risk prediction beyond a single-trait score.

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
THISZanetti '19Widén '21Tanigawa '22Thompson '24Clifton '22Kulm '20
65% semantic matchGenetics
Daniela Zanetti, Abhiram S. Rao, Stefan Gustafsson+3 more·Kidney International·Mar 12, 2019· 51 citations