Whole-genome analysis of plasma fibrinogen reveals population-differentiated genetic regulators with putative liver roles Free

Fibrinogen is a critical coagulation factor and acute phase reactive protein. Under normal conditions, fibrinogen is abundant in circulation, yet during the acute phase inflammatory response, interleukin--6 (IL-6) and IL-1 mediate transcriptional cascades that increase circulating fibrinogen levels up to 3-fold above baseline.^1,2 Fibrinogen measures are a clinical predictor of thrombotic diseases, including coronary heart disease, myocardial infarction, venous thromboembolism, and ischemic stroke.^3,4 Although animal models have shown a causative relationship between fibrinogen and thrombosis,^5,6 this has been difficult to confirm in human genetic studies.^7-9 

Circulating fibrinogen levels are estimated to be 21% to 67% heritable,^10,11 with most estimates falling in the 30% to 50% range,^12,13 and heterogeneous across diverse populations.^14-16 Individuals identifying as African American have higher reported baseline levels of fibrinogen,^14-16 with 1 study suggesting higher fibrinogen heritability in African Americans (44%) vs non-Hispanic Whites (28%).¹⁷ Although genome-wide and exome-wide sequencing studies have identified several loci associated with fibrinogen measures, these variants explain a maximum of 3.7% of variance in European populations.^7,18,19 Little is known regarding genetic regulation of fibrinogen across diverse populations.

Unlike genotyping arrays, which often have better coverage of variants common in European populations, whole-genome sequencing (WGS) allows nontargeted genomic interrogation across all populations.²⁰ Broader coverage provided by WGS increases confidence in minor allele calls, improving power to detect associations with rare and low-frequency variants, and to distinguish multiple signals in the same region.^21,22 Furthermore, large-scale WGS-based reference panels improve imputation quality, increasing power in meta-analysis.^21,22 Incorporating more sensitive genomic measurements, such as WGS, in genetic analyses of fibrinogen may reveal additional associations across a range of allele frequencies and effect sizes in diverse populations.

In this study, we performed genome-wide analyses integrating WGS data from National Heart, Lung, and Blood Institute’s Trans-Omics for Precision Medicine (TOPMed) program²³ and TOPMed-imputed genotyping data from the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium²⁴ to identify new genetic variants associated with circulating fibrinogen. To determine putative regulatory mechanisms and prioritize potentially causal genes, we performed in silico annotation, colocalization analyses, transcriptome-wide association study (TWAS), and TWAS fine mapping. Additionally, we tested the hypothesis that polygenic risk scores (PRSs) for fibrinogen are also associated with coagulation and inflammation-related phenotypes in European and African ancestry participants from the VA Million Veteran Program (MVP).²⁵ 

To investigate the genetic architecture of circulating fibrinogen, we performed a multipopulation genome-wide association study (GWAS), followed by TWAS and phenome-wide association study (PheWAS). An overview of the study design is in Figure 1 and supplemental Figure 1, available on the Blood website. These analyses were undertaken within National Heart, Lung, and Blood Institute’s TOPMed Program and the CHARGE Consortium’s Hemostasis Working Group. In total, 163 912 participants contributed to primary genomic analyses, including 11 283 African-ancestry (AFR), 741 Asian-ancestry, 149 619 European-ancestry (EUR), and 2061 Hispanic participants. Details of the 40 participating studies are in Table 1 and the supplemental Data. The contributing studies were all approved by their relevant institutional review boards.

Population group assignment was determined by each study internally. Studies often used some combination of self-reported race or ethnicity data (which in many cases was used for stratification at the genotyping stage, making future pooled analysis challenging) and comparison of ancestry principal components to commonly used reference panels such as 1000G (often with exclusion of any extreme outliers). We acknowledge that this is not concordant with the most up-to-date standards for defining ancestry clusters based on similarity to reference panels²⁶ but do not have access to individual-level data for most participants and are thus reliant on these study assignments.

Fibrinogen was measured in g/L. Most studies measured levels of functional fibrinogen using the Clauss method,²⁷ whereas the remaining 7 studies used a variety of approaches, including nephelometry and enzyme-linked immunosorbent assay, to measure fibrinogen antigen, a measure of overall fibrinogen quantity. Study-specific phenotyping methods can be found in Table 1 and the supplemental Data. Measures of plasma fibrinogen for TOPMed studies were harmonized to ensure common units (g/L) and absence of unexpected distributions or an excess of outliers (supplemental Figure 1). Data were then uploaded to Analysis Commons²⁸ for centralized genetic analysis.

TOPMed WGS methods were described previously.²³ Briefly, WGS was conducted at 6 sequencing centers (mean depth >30×, Illumina HiSeq X Ten instruments). Joint variant discovery and genotype calling were conducted by the TOPMed Informatics Research Center across all TOPMed studies using the GotCloud pipeline, resulting in a single genotype call set encompassing all TOPMed (TOPMed Freeze 6). Variant quality control was also performed centrally by the Informatics Research Center. Sample quality control was performed by the TOPMed Data Coordinating Center. Further details are in the supplemental Data.

Genotype array data for CHARGE studies and for TOPMed study participants without WGS were imputed using standard methods to the densest available imputation panel (35 using TOPMed Freeze 5b²³; 4 using Haplotype Reference Consortium²⁹).

TOPMed WGS genetic analyses were conducted using mixed models and inverse normalized and rescaled residuals, adjusting for age, sex, population group × study, TOPMed sequencing phase, study-specific parameters, and 10 ancestry-informative principal components. Single variant and aggregate gene-based tests were implemented using the SMMAT function of GENESIS on the Analysis Commons cloud computing platform.^28,30 Aggregate tests included only variants with minor allele frequency (MAF) <5% and minor allele count (MAC) ≥1 and used 3 strategies for variant selection: (1) loss of function (LOF), (2) LOF or deleterious missense (LDM), and (3) coding, promotor, or enhancer variants.

Studies without sequencing data undertook single-variant analyses within each population group using their software of preference, and the same model described for TOPMed WGS analyses then provided summary statistics for central meta-analysis. Quality control of the study-specific single-variant GWAS summary statistics was undertaken using EasyQC³¹ using the following variant exclusion criteria: estimated MAC (MAC × imputation quality) <6, absolute effect size (β) >5, standard error >10, sample size <30, or imputation quality <0.30. Meta-analysis was completed using GWAMA³² with genomic control applied to each study but not to the meta-analysis results. Meta-analysis was completed (i) within each population group for studies with imputed data, (ii) within TOPMed WGS studies, and (iii) combined in a multipopulation mega-analysis that included results from (i) and (ii). Statistical significance was set at P < 5.0E-09.³³ 

For all genome-wide significant regions from the multipopulation mega-analysis, conditional analyses were undertaken using cojo-slct³⁴ within GCTA³⁵ (supplemental Figure 3). The linkage disequilibrium (LD) reference panel used all TOPMed WGS samples that contributed to this GWAS. The 95% credible sets were calculated for the same regions using the Wellcome Trust Case Control Consortium method.³⁶ Percentage variance explained was estimated with summary-level data reported by GCTA³⁵ for the 69 conditionally independent variants from the mega-analysis, using the approximation derived by Shim et al.³⁷ See supplemental Methods for more details.

The Ensembl Variant Effect Predictor (VEP) (https://useast.ensembl.org/Tools/VEP)³⁸ was used to determine nearest gene and top predicted consequence for each of the 69 conditionally independent variants and their LD proxies (as determined by r² > 0.8 in TOP-LD European and/or African ancestry reference panels^23,39). VEP was also used to annotate Combined Annotation Dependent Depletion (CADD) Phred^40,41 and LoFtool⁴²-predicted impact scores and additionally SIFT⁴³ and PolyPhen^44,45 scores for coding variants. InterPro was used to determine amino acid substitution.⁴⁶ 

The same set of variants used for VEP annotation were queried, using rsID, against the National Human Genome Research Institute-European Bioinformatics Institute GWAS catalog v1.0.2, (downloaded 29 October 2022; https://www.ebi.ac.uk/gwas/downloads). Mapped traits corresponding to related quantitative measures were manually cataloged to broad categories (ie, liver enzymes, blood cell count and morphology traits, lipids, etc).

We performed colocalization to query gene-trait associations using fastENLOC^47,48 with precomputed Genotype-Tissue Expression (GTEx) multiple-tissue expression quantitative trait loci (eQTL) annotations and fibrinogen GWAS posterior inclusion probabilities (PIPs) as inputs for each of the most relevant tissues: liver and whole blood. We considered a single-nucleotide polymorphism (SNP) level colocalization probability >0.1 or a regional colocalization probability (RCP) >0.5 as strong evidence of colocalization.

A TWAS was performed using S-PrediXcan⁴⁹ to identify associations between cis elements of gene expression and plasma levels of fibrinogen in mechanistically related tissues, namely, artery (aortic, coronary, or tibial), liver, and whole blood.⁵⁰ Prebuilt prediction models, based on GTEx v8 multivariate adaptive shrinkage in R, were used to estimate variants’ weight on gene expression levels in chosen tissues.^49-51 Because the reference models were created on the basis of the European-ancestry population, we limited the analysis to EUR-only GWAS results. S-PrediXcan results were evaluated using S-MultiXcan⁵² with significant TWAS signals determined using Bonferroni correction for the total number of genes across all models.

TWAS fine mapping using FOCUS (Fine-mapping Of CaUsal gene Sets) was used to identify the likely causal gene under GWAS signals. FOCUS avoids false TWAS signals caused by coregulation and pleiotropic effects of variants at GWAS risk loci⁵³ by modeling marginal TWAS z-scores of all genes in the same region considering variant LD correlations and tagged pleiotropic effects of variants on the trait. Given generated z-scores, the PIP for a gene to be causal is derived and then used to form a credible set of putative causal genes. GTEx v8 MASH-R models were used as the source of eQTL weights and the European-based PROCARDIS database as the reference for LD correlations. PIP ≥0.95 was used to identify a gene as putatively causal.

Genetic risk scores (GRSs) were derived using the independent variants identified by the GCTA analysis described above and as seen in supplemental Figure 3. Three scores were derived, (i) weighting by the variant β from the multipopulation mega-analysis, (ii) weighting by the variant β from the EUR-only meta-analysis, and (iii) weighting by the variant β from the AFR-only meta-analysis. GRSs were then standardized to standard deviation units. A PheWAS was performed for each of the 3 GRSs within EUR and AFR participants of the VA MVP,²⁵ for International Classification of Diseases–based PheCodes⁵⁴ with at least 500 cases. Logistic regression models were adjusted for age, sex, and the first 5 population-specific principal components, and statistical significance was determined on the basis of Bonferroni correction for the number of independent PheCodes (EUR P_Bonferroni = 5.18E-05; AFR P_Bonferroni = 7.25E-05) (see supplemental Data). Additional sensitivity analyses were completed by creating the GRS removing variants in (i) the FGG gene region or (ii) the fibrinogen gene cluster region.

WGS and fibrinogen measures were available for 32 572 individuals across 13 TOPMed studies. Imputed genetic data and fibrinogen measures were available for 131 340 individuals within 39 additional CHARGE or TOPMed studies. A total of 163 912 individuals were included in the multipopulation mega-analysis (Table 1). The mean fibrinogen value across studies was 3.32 g/L. Mean values were slightly higher in African-ancestry individuals compared with European (AFR = 3.59; EUR = 3.30), but the 95% confidence intervals overlapped in TOPMed participants (AFR = 1.67-5.56; EUR = 1.44-4.93). No difference in mean was observed between studies measuring functional fibrinogen using Clauss methods and those measuring total fibrinogen (Table 1). The heritability estimate from EUR TOPMed studies with WGS was 20%, which aligns with the lower end of estimates from family and twin studies.

Multipopulation, single-variant mega-analysis identified 54 genetic loci associated with circulating fibrinogen (Figure 2; supplemental Table 1). Among these, 18 loci have not been identified in prior studies (Table 2; Figure 3). The TM6SF2 locus has been identified in a multiphenotype analysis leveraging fibrinogen and coronary artery disease variants,⁵⁵ but not in a GWAS of fibrinogen alone. All 18 variants achieved nominal significance in a previously published analysis,¹⁸ and the direction of effect was consistent with our findings (Table 2). Associated regions contained an average of 21 SNPs in their 95% credible set (range, 1-93). Credible sets are summarized in supplemental Table 1, and full SNP lists are presented in supplemental Table 2. Approximate conditional analysis in GCTA revealed 6 loci harboring multiple conditionally distinct lead variants (Table 3; supplemental Table 3). At 4 loci with previously reported fibrinogen associations, FGG, PDLIM4, RPL22L1, and RAB37, we detected 7, 5, 3, and 2 distinct lead variants, respectively, exceeding the maximum from any prior report. supplemental Table 4 presents the LD between conditionally distinct lead variants and previously published top SNPs. Among newly associated loci, only SOCS3 and HPN contained multiple distinct lead variants.

Among the 69 conditionally distinct lead variants associated with fibrinogen, 5 variants (rs10936662, rs28577061, rs11077357, rs7507218, and rs1672981) have a minor allele at least 10 percentage points more frequent in African ancestral populations compared with European (supplemental Table 3). At the FGG locus, GCTA identified a conditionally distinct lead variant (rs28577061), which has not been associated with fibrinogen in prior studies and is common in AFR but extremely rare or absent in other TOPMed populations (MAF_AFR = 0.1801; MAF_EUR = 0.008). At 4 loci, RPL22L1, HPN, SOCS3, and SSPB4, the primary lead variants are common in all assessed populations, but at increased frequency in African ancestry populations. Together, the 69 independent variants discovered increased the phenotypic variance explained for circulating fibrinogen across populations from 3.7%¹⁸ to 4.8%, with the fibrinogen gene cluster alone explaining 2.2%. Effect sizes were highly correlated between studies that measured functional fibrinogen and those that measured total fibrinogen (supplemental Figure 4; supplemental Table 5).

Aggregate tests using low-frequency and rare TOPMed WGS variants yielded associations in the fibrinogen gene cluster region: FGG was significant when aggregating LOF and LDM variants, whereas FGA was only significant when aggregating LOF variants and FGB with LDM variants. No genes were significant when aggregating all low-frequency and rare variants in the coding, promotor, or enhancer regions (supplemental Table 6). A list of variants in this region included in the aggregation tests is presented in supplemental Table 7.

To characterize each genetic signal associated with fibrinogen, we queried all conditionally distinct variants and their LD proxies (r² > 0.8 in TOPMed-based EUR and/or AFR ancestry reference panels),^23,39 in multiple publicly available data sets. Most signals we identified contained previously reported lead variants for liver-enzyme measures, lipid measures, and/or blood-cell traits. Notably, 23 signals contained a previously reported lead variant for C-reactive protein (CRP).

VEP³⁸ annotation shows that 18 signals contained at least 1 missense variant. Additionally, 13 signals contained variants with CADD PHRED scores exceeding 20, indicating the variant is in the top 1% of predicted deleterious mutations. SERPINA1, ZFP36L2, and TLR10 stand out as newly associated loci in this category. We also observed that several signals overlap potential regulatory regions in liver. A total of 52 signals (45 loci) contain at least 1 variant mapping to a “consensus” region of open chromatin in liver tissue samples, 42 signals overlap a GeneHancer regulatory element reported active in HepG2 hepatocytes, 5 signals harbor previously reported liver chromatin accessibility quantitative trait loci (QTL) variants, 6 signals hold GTEx liver expression QTL, and the primary signal in HPN contains a GTEx liver splice QTL. A summary of these annotations for each signal can be found in supplemental Table 8 and full results in supplemental Table 9.

Variant-level colocalization analysis in fastENLOC identified 153 variant-tissue pairs with evidence (SNP-level colocalization probability >0.1) for a shared genetic basis with GTEx eQTLs for 93 distinct genes in aortic artery, tibial artery, coronary artery, liver, and/or whole blood (supplemental Table 10). Regional analysis in the same set of tissues found 46 region-tissue pairs, implicated in regulation of 41 distinct genes, with evidence for statistical colocalization with fibrinogen regions (RCP > 0.5) (supplemental Table 11). We note that 5 of the 6 fibrinogen signals harboring sentinel GTEx liver eQTL variants statistically colocalize with the eQTL signals in fastENLOC, whereas the PLEC eQTL signal was at the significance threshold (RCP = 0.5).

Gene-tissue pairs across 5 tissues (aortic, coronary, or tibial artery, whole blood, and liver) were included in MetaXcan TWAS analyses for fibrinogen (supplemental Figure 5). Genetically determined expressions of 64 gene-tissue pairs were significantly associated with fibrinogen levels after Bonferroni correction and filtering for models with at least 2 contributing SNPs (supplemental Table 12). A total of 15 gene-tissue pairs identified by TWAS were further prioritized via FOCUS fine-mapping (supplemental Table 12). TNKS was the only gene prioritized by both TWAS and fine-mapping in liver analyses. Notably, this gene was further implicated in fibrinogen regulation through variant and region-based liver expression colocalization analysis (supplemental Table 11). In whole blood analyses, TWAS and fine-mapping prioritized AFT1, C5orf56, FGB, MS4A4E, and SLC22A5. Of these, AFT1 was further supported by variant-level colocalization analysis (supplemental Table 10).

Three fibrinogen GRSs were tested for association with 1426 PheCodes in EUR and 1061 PheCodes in AFR participants of the VA MVP (supplemental Table 13). Each score reached Bonferroni significance with multiple thrombotic and inflammatory disease phenotypes (n_EUR = 13 PheCodes, n_AFR = 1 PheCode; Table 4; supplemental Table 14). Top associations included venous thromboembolism and gout.

To further assess which variants within the GRS may drive individual associations, we queried each variant against the significant PheCode (supplemental Table 15). As anticipated, for “Circulatory System” and “Hematopoietic” PheCodes, most of the signal came from variants in the fibrinogen gene cluster, whereas significant variants across the GRS were associated with gout and dermatologic traits. Sensitivity analyses with PRSs removing variants (i) in the FGG gene region (associated with γ prime fibrinogen) or (ii) in the full fibrinogen gene cluster did not substantially change the odds ratio, but did impact the significance (supplemental Table 16). PheCodes relating to gout (phe_274, phe_274_1) and superficial cellulitis and abscess (phe_1089) became more significant, whereas PheCodes relating to coagulation defects and hypercoagulable state (phe_286, phe_286_8, and phe_286_81) became less significant.

Magnetic resonance (MR) analysis of fibrinogen and gout failed to provide evidence for a causal effect of fibrinogen levels on gout, and multivariable MR analysis incorporating CRP levels to account for possible confounding was inconclusive (supplemental Table 17).

Herein, we used WGS and genotype data from diverse participants to identify 69 conditionally distinct genetic variants across 54 loci associated with circulating fibrinogen. Our results corroborate previous reports that fibrinogen is highly polygenic and advance the field by identifying new variant associations, including rare variants and variants most prevalent in underrepresented populations. On the basis of results of in silico characterization, we suggest some of the new genetic regulators we identified may act directly to regulate coagulation factors (such as factor VII and factor XI), whereas others may impact fibrinogen broadly through pathways altering liver metabolism, inflammation, and immune function, as reflected in the broad overlap between fibrinogen-associated signals and prior analyses for liver-related traits and blood and immune-cell counts.

We identified several novel genetic associations driven by variants with higher allele frequencies in non-European populations, validating that our approach improved detection of putative genetic regulators most common in underrepresented populations. To our knowledge, we identified the first common variant-fibrinogen association driven almost entirely by African ancestry participants at an intergenic region near FGG. Previous African ancestry driven variants were rare in African and not present in European populations.¹⁹ Additionally, expanded representation of both African and European populations allowed us to detect 4 new signals of small effect, led by common variants substantially more frequent in those with African ancestry. Detection of population-differentiated associations indicates that for variants with higher predicted effects, such as those within the fibrinogen gene cluster, expanded sample sizes are now reaching power to detect variant associations driven by underrepresented populations. We further identified 4 common European-driven signals that are rare or uncommon in African ancestry populations, suggesting that genomic signals for circulating fibrinogen are not saturated in any ancestral population.^56,57 This is also reflected by the relatively modest variance explained (4.8%) vs estimates of total fibrinogen heritability (30%-50%).^12,13,17 Increased diversity and larger sample sizes are still needed to help close this gap.

Three novel regions (SERPINA1, ZFP36L2, and TLR10) harbor deleterious missense variants in genes with plausible connections to coagulation and/or inflammation. In the SERPINA1 gene, we identified a rare missense variant (rs17580) predicted to be highly deleterious (CADD Phred score = 32). SERPINA1 encodes α-1-antitrypsin, which protects other proteins and tissues from serine protease degradation via direct binding inhibition.⁵⁸ Like fibrinogen, α-1-antitrypsin is synthesized in hepatocytes in response to IL-6 and IL-1 inflammatory pathways.⁵⁹ Severe α-1-antitrypsin deficiency is associated with liver dysfunction and emphysema related to a lack of inhibition of neutrophil elastase.⁶⁰ The association of SERPINA1 with fibrinogen levels may reflect α-1-antitrypsin involvement in inflammatory pathways, or a more direct relationship between these proteins. Although in vitro assays have suggested α-1-antitrypsin can bind to plasma fibrinogen and be incorporated into fibrin networks,⁵⁸ little is known about the functional implications of these interactions and whether it also influences fibrin(ogen) degradation.

Zinc Finger Protein (ZFP)36L2 and toll-like receptor 10 (TLR10) are broadly linked to inflammatory pathways. Although ZFP36L2 is largely uncharacterized, the ZFP36L2 missense variant we observed has previously been associated with blood cell count and morphology traits,^61-64 and studies in mice show ZFP36L2 regulates blood cell development and adipogenesis.⁶⁵ Proteins of the ZFP36 family regulate transcript abundance through binding AU-rich elements to target transcripts for decay.⁶⁶ Notably, ZFP36 (tristetraprolin) family proteins are known to modulate transcript abundance of the inflammatory cytokine tumor necrosis factor-α,⁶⁷ which induces both the coagulation cascade and complement system.^68,69TLR10 encodes a TLR protein. Although TLR activation generally contributes to inflammation, immune responses, and activation of coagulation cascades, little is known about TLR10 specifically.⁷⁰ Although TLR10 may exert a protective effect against inflammation,^71,72 more work is needed to elucidate mechanistic links between this protein and coagulation.

Although the missense variants highlight intriguing new gene targets for future functional studies, most signals we identified are driven by common noncoding variants with small effect sizes. Many of these noncoding signals overlap HepG2-active regulatory regions, and a subset of signals overlap liver chromatin accessibility, expression, and/or splice QTL signals, suggesting these variants may exert subtle regulatory effects on circulating fibrinogen levels. Notably, at 2 newly associated loci, HPN and SOCS3, we identified 2 conditionally distinct variants, each with distinct functional annotations.

At the HPN locus, the primary signal, which maps to an intronic region of HPN, contains a liver splice QTL for HPN. HPN encodes hepsin, a type II transmembrane serine protease with “enhanced” expression in the liver (according to the Human Protein Atlas), believed to function in macromolecular metabolism.⁷³ Interestingly, studies in zebrafish and mammalian cell lines suggest hepsin activates coagulation factor VII.^74,75 Although mouse studies have not conclusively demonstrated hepsin’s involvement in coagulation, these studies have established hepsin’s involvement in regulating liver metabolism via hepatocyte growth factor and Met signaling pathways.^73,76-78 The secondary signal at this locus maps to an intergenic region and contains a liver eQTL for a nearby gene, TMEM147. TMEM147, which was also prioritized in liver, whole blood, and aortic artery TWAS analyses, encodes a widely expressed endoplasmic reticulum transmembrane protein implicated in various metabolic processes, including calcium transport.⁷⁹ Although it is unclear whether the 2 signals in this region influence fibrinogen through altered splicing and expression patterns of HPN and TMEM147 in the liver, these signals provide a compelling starting point for future functional studies.

Similarly, among the 2 independent, common-variant signals in the SOCS3, PGS1 locus on chromosome 17, the secondary signal houses a previously identified sentinel variant for altered chromatin accessibility in liver tissue. This variant is of particular interest given its proximity to SOCS3, a “suppressor of cytokine signaling” known to act upstream of IL-6 in the acute-phase response pathway that induces fibrinogen.² It is possible that the secondary signal at this locus is capturing genetic variation that modulates the accessibility of the SOCS3 gene, and potentially other nearby genes, in liver or other tissues, leading to downstream impacts on fibrinogen levels. SOCS3 methylation was recently associated with circulating fibrinogen levels in an epigenome-wide association study in the CHARGE consortium⁸⁰ with the associated probe within 11 kb of our best SOCS3 GWAS SNP, reinforcing the idea that chromatin accessibility may be at play.

Results of the PheWAS in MVP yielded expected associations with venous thromboembolism and several hematopoietic traits. However, the direction of the association within bleeding and thrombosis phenotypes was often opposite to the direction we expected given total fibrinogen’s procoagulant function (ie, typically increased genetically predicted fibrinogen is associated with decreased risk of bleeding and increased risk of thrombosis). Although seemingly paradoxical, this finding is in line with what has been observed in patients with congenital fibrinogen deficiency. This rare, life-threatening bleeding disorder results from a complete or partial loss of fibrinogen, and/or the presence of dysfunctional fibrinogen.^2,81 In patients with complete loss of fibrinogen, not only has increased bleeding (from loss of fibrin formation) been reported, but also increased risk of thrombosis (hypothesized to result from loss of the anticoagulant properties of the γ' isoform), demonstrating complex and sometimes competing effects of fibrinogen isoforms on coagulation biology.⁸² A previous MR study observed an inverse association between total fibrinogen levels with venous thromboembolism risk.⁹ We tested the hypothesis that this inverse association in our PheWAS was driven by the alternatively spliced γ prime fibrinogen isoform, which has established anticoagulant properties,⁸³ but removing the FGG locus SNPs from the tested PRS did not alter the direction of effect we observed. Our hypothesis is thus that genetically driven increases in total fibrinogen are accompanied by increases in γ prime fibrinogen, and that the antithrombotic effect of γ prime fibrinogen compensates for the prothrombotic effect of other isoforms or of total fibrinogen.

Another unexpected finding from the PheWAS was a positive association between genetically predicted total fibrinogen and gout. Interestingly, this finding is in line with emerging literature, which has shown that patients with active gout have increased thrombin generation markers.^84,85 Our MR analysis did not support a causal association of fibrinogen on gout, and multivariable Mendelian randomization analysis to investigate if inflammation plays a profound role in the association between fibrinogen and gout (using CRP as a proxy) was inconclusive. Given the extensive overlap between our signals and previously reported GWAS signals for phenotypes reflecting liver health and inflammation, such as liver enzyme and lipid measures, we suggest that this association may be capturing variant impacts on broader liver metabolic pathways.

In conclusion, we identified 18 novel loci, collectively harboring 20 distinct variants, associated with fibrinogen measure. We report the first common African-driven fibrinogen association, and several additional associated variants with population-differentiated genetic architecture. Furthermore, we demonstrate overlap between these signals and liver regulatory elements, as well as GWAS phenotypes reflecting altered liver metabolism and inflammation. Future studies investigating coregulation and epistatic effects will likely provide new insight on the shared genetic architecture, and biological interplay, of hemostasis and inflammation.

The authors gratefully acknowledge the studies and participants who provided biological samples and data for TOPMed.

Molecular data for the Trans-Omics in Precision Medicine (TOPMed) program was supported by the National Institutes of Health (NIH), National Heart, Lung, and Blood Institute (NHLBI). See the TOPMed Omics Support Table in the supplemental Data for study-specific omics support information. Core support, including centralized genomic read mapping and genotype calling, along with variant quality metrics and filtering, was provided by the TOPMed Informatics Research Center (NIH, NHLBI grant 3R01HL-117626-02S1; contract HHSN268201800002I). Core support, including phenotype harmonization, data management, sample-identity quality control, and general program coordination, was provided by the TOPMed Data Coordinating Center (NIH, NHLBI grants R01HL-120393; U01HL-120393; contract HHSN268201800001I). CHARGE is supported by NIH, NHLBI grant R01HL-105756, and the CHARGE and TOPMed Hemostasis Working Groups by NIH, NHLBI grants R01HL-134894, R01HL-139553, and R01HL-141291. This research is based on data from the Million Veteran Program, Office of Research and Development, Veterans Health Administration, and was supported by award I01-BX004821. The study was supported by the Novo Nordisk Foundation (grant NNF18CC0034900). D.-A.T. was supported by the EPIDEMIOM-VT Senior Chair of Excellence from the University of Bordeaux initiative of Excellence. M.S.-L. is supported by a Miguel Servet contract from the ISCIII Spanish Health Institute (CPII22/00007) and cofinanced by the European Social Fund. Additional study-specific funding and acknowledgments can be found in the supplemental Data. The Genotype-Tissue Expression Project was supported by the Common Fund of the NIH Office of the Director, and by the NIH, National Cancer Institute, NIH, National Human Genome Research Institute, NIH, NHLBI, NIH, National Institute on Drug Abuse, NIH, National Institute of Mental Health, and NIH, National Institute of Neurological Disorders and Stroke.

This publication does not represent the views of the Department of Veteran Affairs, the US National Institutes of Health, the National Heart, Lung, and Blood Institute, the US Department of Health and Human Services, or the US government.

Contribution: L.M.R., J.A.B., L.A., D.I.C., M.-H.C., J.P.L., O. Polasek, J.I.R., E.d.G., J.-F.D., L.E., N.F., M.G., A.L., S.S.R., F.R.R., I.R., J.C.S., L.C.B., A.B., H.C., K.C., J.E.C., M.D., P.E., M.N., V.O., B.M.P., J.M.S., D.J.S., A.v.H., H.W., W.F.W., Y.B.-S., J.B., D.B., S.R.C., J.G.E., E.F., C.H., M.A.I., J.W.J., S.L.R.K., L.A.L., R.A.M., B.D.M., D.O.M.-K., P.-E.M., P.P.P., A.R., P.M.R., T.D.S., N.J.W., J.F.W., D.-A.T., A.D.J., P.S.d.V., A.C.M., and N.L.S. participated in the design and/or funding of the contributing studies; J.E.H., L.M.R., L.R.Y., L.A., L.F.B., R.P.B., G.D.C., J.L., J.P.L., R.L.-G., M.M., O. Polasek, M.S., P.S., M.H.C., P.C., E.d.G., N.F., M.G., A.G., H.G., T.H., T.K., J.L., A.L., S.N., F.R.R., I.R., K.A.R., J.C.S., F.J.A.v.R., H.W., W.Z.A.B., B.E.C., K.C., J.D.C., M.P.M.d.M., M.D., P.E., C.F., N.G., S.H.E., I.K., M.N., J.R.O., V.O., B.M.P., K.R., J.A.S., D.J.S., A.v.H., G.W., Y.B.-S., J.B., D.B., S.R.C., J.G.E., E.F., T.H., S.L.R.K., L.A.L., B.D.M., P.-E.M., O. Pedersen, S.R., P.M.R., E.K.S., N.J.W., J.F.W., P.S.d.V., M.S.-L., A.C.M., and N.L.S. participated in phenotype data acquisition and/or quality control; L.R.Y., L.A., G.D.C., D.I.C., M.-H.C., D.B.E., J.H., J.-J.H., M.E.K., N.-Q.L., J.L., R.L.-G., M.M., A.M.-P., A.R., B.A.T.R., M.S., P.S., S.T., S.W., M.Z., P.A., M.H.C., G.D., E.d.G., J.-F.D., H.G., T.O.K., J.L., A.L., S.N., F.P.-V., F.R.R., K.A.R., F.J.A.v.R., H.W., W.Z., B.E.C., J.E.C., C.F., N.G., S.H.E., C.M., J.R.O., V.O., B.M.P., K.R., J.A.S., G.W., J.B., D.B., A.D., J.G.E., E.F., M.F., T.H., C.H., J.W.J., S.L.R.K., R.A.M., B.D.M., D.O.M.-K., P.-E.M., O. Pedersen, S.R., P.M.R., E.K.S., U.V., J.F.W., D.-A.T., A.D.J., M.S.-L., and A.C.M. participated in genotype data acquisition and/or quality control; J.E.H., J.N., J.H., A.S.H., L.R.Y., J.A.B., F.T., L.F.B., G.D.C., M.-H.C., D.B.E., M.G., J.-J.H., M.E.K., N.-Q.L., J.P.L., R.L.-G., J.L., A.M., M.M., R.E.M., A.M.-P., A.R., B.A.T.R., M.S., P.S., S.T., S.W., M.Z., P.A., G.D., G.E.D., M.G., P.K.J., S.N., R.N., F.P.-V., K.A.R., M.R.B., J.S.F., C.F., X.G., C.K., J.R.O., D.B., A.D., C.H., R.A.M., B.D.M., A.R., U.V., J.Y., D.-A.T., A.D.J., P.S.d.V., M.S.-L., A.C.M., and N.L.S. participated in data analysis and/or interpretation; and J.E.H., J.N., J.H., P.S.d.V., M.S.-L., A.C.M., and N.L.S. participated in drafting the initial manuscript; and all authors participated in critical review of the manuscript.

Conflict-of-interest disclosure: L.R. is a consultant for the TOPMed Administrative Coordinating Center (through WeStat). R.L.-G. is a part-time consultant for Metabolon, Inc. M.C. has received grant support from Bayer. B.P. serves on the Steering Committee of the Yale Open Data Access Project, funded by Johnson & Johnson. In the past 3 years, E.K.S. has received grant support from Bayer. The remaining authors declare no competing financial interests.

A complete list of the members of the VA Million Veteran Program and the National Heart, Lung, and Blood Institute Trans-Omics for Precision Medicine (TOPMed) Consortium appears in the supplemental Data.

Correspondence: Jennifer E. Huffman, 2 Ave de Lafayette, ATTN: VA MAVERIC, Boston, MA 02110; email: jennifer.huffman2@va.gov.