The common VTE-protective G haplotype of F5 increases factor V-short, TFPI function, and risk of bleeding

Qualitative and quantitative abnormalities in plasma coagulation factors can mediate the risks for both bleeding^1,2 and thrombosis.^3,4 One of the causes for these abnormalities is the inheritance of DNA variants⁵ in the genes that encode coagulation factors. The type and strength of the consequent phenotype is dependent on the variant and the coagulation factor.

Factor V (FV) circulates as a 2196 amino acid (aa) glycoprotein with 6 domains (A1-A2-B-A3-C1-C2).⁶ In its activated form, FV acts as a procoagulant cofactor that binds FXa and accelerates the conversion of prothrombin to thrombin.⁷ For this to occur, thrombin must first activate FV through the proteolytic removal of the B-domain, thereby transitioning FV into FVa.⁸ FVa procoagulant functions are regulated by activated protein C (APC)–mediated cleavage, rendering FVa inactive.⁶ However, FV also has anticoagulant properties and acts as a cofactor for APC in the inactivation of FVIIIa⁶ and for tissue factor pathway inhibitor α (TFPIα) in the inhibition of FXa activity.⁶ An intact/partially intact FV B-domain is a requisite for FV anticoagulant function within both pathways.^6,8,9 

The best-known single nucleotide variant (SNV) in the FV gene (F5) is FV Leiden (FVL). FVL abolishes an APC cleavage site, which leads to delayed FVa¹⁰ and FVIIIa¹¹ inactivation and causes a threefold increased risk for venous thromboembolism (VTE) in heterozygotes.¹² This APC resistance, coupled with an allele frequency (AF) of 1% to 4% in individuals of South Asian, European, and Middle Eastern ancestry,¹³ means that it is the most important common genetic risk factor for VTE. At least 11 other F5 variants have been identified that confer APC resistance, most being specific to an ancestral group or extremely rare.¹⁴ These include FV Nara and Besançon. Although both SNVs are associated with reduced FV levels, they still exert a prothrombotic effect.^15-18 This is because they are resistant to APC proteolysis and associated with reduced TFPIα plasma levels, thereby overpowering any reduction in procoagulant functionality.^9,15,17-19 

There is another group of extremely rare F5 variants that influence TFPIα levels but through a different mechanism. These variants cause enhanced alternative splicing of F5, thereby increasing the expression of a B-domain truncated, lower molecular weight isoform (220 kDa vs 330 kDa). The isoform was first discovered in individuals heterozygous for the c.2350A>G variant causal for East Texas bleeding disorder (ETBD)²⁰ and named FV-short.^6,21 Although it has normal procoagulant functionality, FV-short forms a high-affinity complex with TFPIα that is mediated through the truncated FV B-domain.²¹ This explains why the pathologically raised (predicted ∼10-fold) FV-short in ETBD leads to an ∼10-fold increase in circulating TFPIα, the physiological regulator of the initiation of coagulation,²² thereby exerting a bleeding phenotype.²¹ Similar observations were later made in individuals who carry the FV-Amsterdam²³ and FV-Atlanta variants.²⁴ FV-short normally circulates in the plasma of individuals in the general populace at low concentrations.²¹ The absolute concentrations have not been formally determined, but are estimated to be in the sub nanomolar range, that is, similar to the plasma concentrations of TFPIα.^21,25 FV-short-TFPIα also binds to protein S in plasma and, together with FV-short and protein S, amplify the efficiency of the TFPIα anticoagulant pathway.^6,26,27 

Aside from FVL, little mechanistic insight is available for other common F5 variants that affect the risk of VTE. This is relevant because a genome-wide association study (GWAS) showed that, in addition to FVL, 2 other variants in F5 were independently associated with VTE risk, namely rs966751 and rs6032.¹² The latter is one of a cluster of 4 co-inherited SNVs (others are rs6021, rs4524, and rs4525) known as the G haplotype because each SNV represents an A to G substitution.^28-30 The AF is 27%, and each allele confers a reduction in the risk of VTE of 14%.¹² Prompted by the well-established effect of FVL on APC resistance, this has been the focus of mechanistic studies of the G haplotype to date.²⁸ However, the results have been conflicting between studies.^29-32 Given this uncertainty and the emergent evidence of other anticoagulant properties of FV, the aim of this study was to establish a putative mechanism for the VTE protection afforded by the G haplotype and to examine the clinically relevant effects.

An overview of the methods is provided here. For more detailed information, refer to the supplemental Methods.

RNA sequencing (RNA-seq) of whole blood was undertaken in 4732 European ancestry blood donors who participated in the INTERVAL study, who were analyzed as previously described, and among whom a subset of 4651 individuals had data on the genotype of the G haplotype available.^33-36 The median read depth of the INTERVAL RNA-seq samples was 24.0 million unique reads (interquartile range [IQR], 21.5-26.9). Splice junctions were identified using RegTools³⁷ and filtered for the cryptic FV-short donor and acceptor splice sites (GRCh38 [Genome Reference Consortium Human Genome Build 38] coordinates Chr12:169542739 and Chr12:169540633, respectively).²⁴ The number of reads that supported the 2106 base pair (bp) excision event between the FV-short splice sites was determined for each participant.

Participants in the National Institute for Health and Care Research (NIHR) BioResource (BioResource) who were homozygous for the G haplotype (F5-G/G) or the reference sequence (F5-ref) were approached by the BioResource team to participate in this study (research ethics committee reference 10/H0304/65). A medical history was taken. Blood was drawn in citrate tubes for coagulation factor quantitation and citrate-corn trypsin inhibitor for calibrated automated thrombography (CAT). Plasma was isolated using a standard protocol (supplemental Methods; supplemental Table 1). Researchers were blinded to the G haplotype genotype at the study visit and for all experiments.

F5-G/G and F5-ref plasma pools were created from the citrated plasma obtained from the BioResource participants (n = 7 and n = 13 for F5-G/G and F5-ref, respectively) and subjected to immunoprecipitation. For this, antibodies (abs) against TFPI (PAHTFPI-S; Prolytix) were immobilized to tosylactivated magnetic beads (150 μg ab per 5 mg beads) as per manufacturer’s instructions. Plasma (1 mL) was incubated with 5 mg of beads for 2 hours at room temperature. The supernatant was removed, and the beads were washed with 0.1% bovine serum albumin in 0.01 M sodium phosphate with 0.150 M NaCl (pH 7.4). Proteins bound to the beads were eluted with 40 μL lithium dodecyl sulfate buffer (LDS buffer; Life Technologies) and analyzed by western blotting alongside known quantities of recombinant FV-short or TFPIα.^27,38 Eluted FV/FV-short and TFPIα were detected using abs against FV (AHV-5146; Prolytix) and TFPIα (mix of anti-TFPIα K1, K2 and C-terminus; Sanquin). Using recombinant FV-short as a standard, the FV-short coimmunoprecipitated with TFPIα was quantified using Bio-Rad Image Lab 6.1. Supernatants (900 μL) from the first co-immunoprecipitation (co-IP) were subjected to a new co-IP, repeating the same wash and elution steps as previously. This was done either on the neat supernatant or after incubating with recombinant TFPIα (1 nM) for 2 hours at room temperature.

The LDlinkR R package, version 1.2.2,³⁹ was used to identify SNVs in linkage disequilibrium (LD) with rs4524 in the 1000 Genomes Project–phased genotype data^40,41 of individuals of European ancestry.³⁹ The initial list that was generated was filtered to retain only SNVs in complete LD as defined by a minor allele frequency (MAF) of 0.2727 (ie, equal to that of rs4524), R squared (R²) of 1.00, and D prime (D') of 1.00.^40,41 

Plasma FV levels were quantified using an FV Ab set according to the manufacturer’s instructions (FV-EIA; Affinity Biologicals). The total and free protein S and the total TFPI and TFPIα³⁸ plasma levels were quantified using in-house enzyme-linked immunosorbent assays.

CAT was performed on corn trypsin inhibitor–citrated plasma from the BioResource participants in the presence of 4 μM phospholipid vesicles and 16.6 mM CaCl₂ as described.^42,43 Thrombin generation was initiated by 1 or 13 pM tissue factor (TF; Dade Innovin) in a final volume of 120 μL and monitored using 0.42 mM fluorogenic thrombin substrate Z-Gly-Arg-AMC-HC (Bachem) in the presence or absence of saturating inhibitory anti-TFPI Abs (78 μg/mL; PAHTFPI-S; supplemental Figure 3). The ratio of the variables, determined in the presence and absence of anti-TFPI Abs, were used to quantify TFPI function. The mean of 3 technical replicates was calculated for each CAT parameter of the plasma sample from each participant.

The UK Biobank (UKB) is a longitudinal study of 500 000 individuals in the United Kingdom.⁴⁴ Participant-level data include genotyping and Hospital Episode Statistics (HES). The latter includes the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) code assigned to each public hospital admission an individual has received since 1997. We performed genotype-phenotype analyses on the first 200 000 UKB participants who underwent whole exome sequencing⁴⁵ and focused on 117 699 participants of genetically determined European ancestry who were enrolled at UKB assessment sites in England. Whole exome sequencing data were used to determine the G haplotype genotype. HES data from 1 April 1997 to 30 September 2020, were reviewed, and the ICD-10 codes that were applied to each hospital episode were used to determine whether the participant had VTE, bleeding, or cardiovascular disease.⁴⁶ A binomial general linear model was used to determine the odds of each phenotype in participants with F5-G/G in comparison with the rest of the population with age, sex, and ABO blood group included as additional covariates.

The recall-by-genotype study was approved by the Cambridge East Research Ethics Committee (REC 10/H0304/65).

The 4 common G haplotype SNVs are clustered at the 5′ end of exon 13 in close genetic proximity to the rare variants for ETBD, FV-Atlanta, and FV-Amsterdam (Figure 1A), which lead to the production of an excess of truncated forms of FV that are missing 623 aa (FV-Amsterdam²³) or 702 aa (ETBD²¹ and FV-Atlanta²⁴). The FV isoform with 702 missing amino acids is referred to as FV-short²¹ and leads to alternative splicing at cryptic donor and acceptor sites in exon 13. This leads to the excision of 2106 bps, in frame, from the F5 transcript, which correspond to the missing 702 aa in FV-short. Given that the G haplotype SNVs are in the genetic vicinity of the positions of the ETBD and FV-Atlanta variants and the FV-short donor site (the closest, G haplotype SNV, rs6021, is 50 bp 5′; Figure 1A), they may influence FV-short splicing efficiency.

In the original description of FV-short,²¹ its transcripts were detected in leukocyte RNA of both individuals with and without ETBD. Here, we analyzed the whole-blood RNA-seq data of 4651 genotyped blood donors from the INTERVAL study^33,34 and focused on the detection of the 2106 bp excision event between the FV-short donor and acceptor splice sites. We detected excision events in 119 (2.6%) participants (Table 1). This was at extremely low levels (88 with 1 read; 17 with 2 reads; 12 with 3 reads; and 2 with 4 reads supporting excision event), consistent with previous assertions that FV-short is an exceptionally low-abundance transcript.²⁴ We compared the proportions of participants for whom excision events could be detected among those homozygous for the G haplotype (F5-G/G), heterozygotes (F5-het), and those with the reference sequence (F5-ref). Excision events were detected in an approximately twofold higher proportion of participants with F5-G/G than in those with F5-ref with F5-hets occupying an intermediate position (Figure 1B; Table 1). Increasing the dosage of the G haplotype was significantly associated with a detectable excision event (irrespective of the number of reads) at the FV-short splice sites (logistic regression P = .003), which is consistent with those with F5-G/G having enhanced alternative splicing of F5 to form FV-short.

To ascertain whether enhanced excision at the FV-short splice sites affects FV-short levels in plasma, we invited 20 participants in the BioResource to donate blood; 7 were F5-G/G and 13 were F5-ref. All were male, aged 45 to 60 years, of European ancestry, did not smoke, did not have a VTE or bleeding history, and were not taking anticoagulants (supplemental Table 1). One (F5-ref) participant was FVL heterozygous with no FVL homozygotes.

The similarity of FV-short to FV and its low plasma concentration (predicted to be ∼1% of plasma FV) posed challenges for quantitation.⁶ Therefore, as a proof-of-concept analysis to explore FV-short plasma levels, we used co-IP of FV-short using anti-TFPI Abs, taking advantage of the high-affinity FV-short–TFPIα interaction. This same approach was used to identify FV-short in plasma from healthy individuals previously.²¹ Because of the large amount of plasma required, it was not feasible to use this approach on individual plasma samples. Instead, a plasma pool was generated for each group (F5-G/G and F5-ref). As in previous studies, FV-short was co-eluted with TFPIα. Smaller amounts of FV were also co-eluted, reflecting the preferential binding of TFPIα to FV-short over FV (Figure 1C-D).^21,48 The median amount of FV-short eluted from 1 mL of plasma pool was higher in participants with F5-G/G than in participants with F5-ref at 88.5 ng (IQR, 61.7-106) vs 40.6 ng (IQR, 29.8-48.5) (P = .029) (Figure 1E). This suggests that the enhanced excision associated with the G haplotype at the FV-short cryptic splice sites corresponds to higher levels of FV-short protein expression in those with F5-G/G.

Because it has not been demonstrated formally that all FV-short is bound to TFPIα, the supernatant was subjected to further co-IP, either in the presence or absence of exogenous recombinant TFPIα (1 nM). In both cases, a small amount of FV-short was coimmunoprecipitated (<10% of initial elution), suggesting that the majority of FV-short was bound to TFPIα in plasma and that most of the FV-short–TFPIα complexes had already been precipitated successfully (supplemental Figure 2). To assess whether the increased amounts of FV-short found in the F5-G/G plasma as opposed to the F5-ref plasma pool was because of altered FV expression, we measured the total FV levels in plasma from all 20 individuals; no difference was detected between the F5-G/G and F5-ref groups (Table 2), suggesting that increased excision at the FV-short splice sites in those with F5-G/G only altered FV-short expression.

Of the 4 G haplotype SNVs, rs4524 is most cited as being associated with a reduced VTE risk (supplemental Table 2). However, it is not necessarily the lead SNV identified in epidemiological studies that drives the biological effect.⁴⁹ Therefore, we searched for SNVs in the 1000 Genomes Project⁴¹ in complete LD and with an identical AF as rs4524 (supplemental Table 2). We determined that the G haplotype also includes the SNV rs10800453 (Figure 2A). In our previous GWAS of the plasma proteome in ∼3000 European ancestry blood donors in the INTERVAL study, the alternate allele of rs10800453, A (ie, the G haplotype), was associated with an increased total plasma TFPI level, and this was replicated in an additional cohort of ∼5000 INTERVAL participants.⁵⁰ Given the high-affinity interaction between FV-short and TFPIα, we hypothesized that the increased total plasma TFPI level seen with the G haplotype in the INTERVAL study was mediated by higher FV-short expression and increased TFPIα. To test this hypothesis, we measured the total TFPI and TFPIα in the 20 BioResource participants (Table 2). Both the total TFPI and TFPIα were increased in those with F5-G/G by 6.6% and 22.9%, respectively, however, this did not meet statistical significance, which we suspect is because of a smaller sample size as when compared with that of the INTERVAL study.

Any influence on the TFPIα pathway by FV-short is more likely to be detected in functional assays because these are sensitive to both TFPIα levels and functional enhancement by FV-short. For this, thrombin generation assays, measured by CAT and initiated by 1 pM TF, were performed in the presence and absence of inhibitory anti-TFPI Abs. Earlier and higher levels of thrombin generation was observed when the TFPI pathway was inhibited (Figure 2B). The lag time (LT) in the F5-G/G group was significantly longer than that in the F5-ref group (Figure 2B-C; Table 3) with F5-refs having a similar median LT as a control plasma pool collected from 15 nongenotyped healthy individuals separate from the BioResource (supplemental Methods).

The increase in LTs in those with F5-G/G were abolished upon TFPI inhibition (Figure 2D; Table 3), suggesting that TFPI function was responsible for the increase in the LT in participants with F5-G/G. This observation was further evident when the LT ratios (LT in the absence of anti-TFPI divided by the LT with anti-TFPI) were compared between the groups (Figure 2E; Table 3). Apart from a significant increase in the time to peak ratio, none of the other CAT variables (ie, peak and endogenous thrombin potential) were significantly different between the groups, suggesting a weaker association between the TFPIα–FV-short levels and these variables (Table 3). One individual in the F5-ref group had an abnormal LT (29.1 minutes) despite having normal TFPIα levels (0.39 nM), suggesting that the extended LT was not associated with the TFPI pathway. To confirm the association between TFPI and the prolonged LT in the F5-G/G group, we repeated the assays at 13 pM TF at which concentration the TFPIα (and FV-short) anticoagulant function is less effective, whereas the FV procoagulant functions remain (supplemental Figure 4).^43,51 The increased procoagulant trigger abolished the absolute difference in LT without anti-TFPI. However, there remained a significant difference in LT ratios, highlighting the strength of the association between F5-G/G and TFPI function (Table 4). Given that FV-short/TFPIα also circulates in complex with protein S,⁵² differences in these levels between those with F5-G/G and those with F5-ref could have explained the differences in LT, but there were no significant differences in the total or free protein S levels between the groups (Table 2). This is supported by the absence of an association between the G haplotype and protein S in our previous GWAS.⁵⁰ 

To explore the clinical implications of increased FV-short expression and TFPI functionality conferred by the G haplotype, we compared 8699 UKB participants with this genotype with 109 000 individuals who were either F5-het or F5-ref. HES between 1997 and 2020 were analyzed, and the incidence of VTE and cardiovascular disease episodes during this period was compared between the groups. Bleeding episodes were also analyzed because of the proposed link between increased TFPIα and bleeding tendency.⁵³ A bleeding episode was defined by a hospital episode that was assigned at least 1 of 97 ICD-10 codes for a bleeding symptom (listed in Stefanucci et al).⁴⁶ There was a significant reduction in VTE in UKB participants who were F5-G/G (odds ratio [OR], 0.75; P = .0003). confirming its protective effect¹² (Figure 3). However, this was at the expense of a moderately increased bleeding risk (OR, 1.07; P = .011), highlighting the physiological importance of FV’s anticoagulant functions.

The first description of 4 linked SNVs comprising the F5 G haplotype was made at the turn of the 21st century.³⁰ Genotype arrays arrived shortly thereafter,⁵⁴ enabling the first analysis of associations between genotype and VTE phenotype at scale. Using this technology, a case-control analysis showed that for each copy of the F5 G haplotype carried, the OR for thrombosis was reduced by 0.25.⁵⁵ The VTE-protective effect of the G haplotype has now been replicated in at least 18 subsequent studies (supplemental Table 2). However, the mechanism that underpins this effect has not been resolved.

The G haplotype has been suggested to cause an increased efficiency of the APC pathway.³¹ However, the published findings have been inconsistent.^30,32,56 Therefore, we questioned the mechanistic relevance of APC to the G haplotype and sought an alternative explanation.^29,30,32 Since these publications, cofactor functions of FV within the TFPI pathway have been discovered.⁶ We reasoned that this function of FV could be of relevance for the G haplotype because of the genetic proximity of its constituent SNVs to the location of the rare ETBD, FV-Amsterdam, and FV-Atlanta variants. All these confer bleeding diathesis through pathologically elevated levels of B-domain–truncated FV, including FV-short, mediated through enhanced alternative splicing at cryptic splice sites in exon 13.^21,23,24 FV-short is also present in the plasma of individuals from the general populace.^21,52 However, the extremely low abundance of both the transcript²⁴ and plasma protein²¹ have hampered efforts to ascertain its relevance. Access to high read depth, whole blood, RNA-seq data from 4651 genotyped blood donors in the INTERVAL study meant that, for the first time, to our knowledge, we could demonstrate the existence of excision events at the FV-short splice sites outside of the context of rare bleeding disorders. We detected the excision events in 2.6% of all donors with an ∼2.3-fold increase in the proportion of individuals in the F5-G/G group in whom this could be detected when compared with the F5-ref group, consistent with a positive increase in the alternative splicing of FV to FV-short caused by the common G haplotype.

The similarity between FV-short and FV, combined with its low plasma concentration, means that absolute quantitation of circulating FV-short is not currently possible, because no FV-short–specific Abs are available. Instead, we used FV-short–TFPIα co-IP to quantify FV-short indirectly (Figure 1C-E) and found that the increased excision events in F5 led to increased circulating FV-short in F5-G/Gs. It is particularly interesting that the precipitated amounts of FV-short corresponded to concentrations of 0.44 nM and 0.2 nM in the original F5-G/G and F5-ref plasma samples that were subjected to precipitation, respectively, which is within the concentration range of plasma TFPIα (Table 2).^25,57 This supports the role of FV-short as an essential carrier of TFPIα and is the closest any study has come to quantifying plasma FV-short in the general populace to date.

TFPIα is an important regulator of the development of thrombosis with an increasingly appreciated association with FV, including positive correlations found between FV, TFPIα, and protein S plasma levels.^6,25,43,57 Furthermore, FV-short is currently believed to regulate TFPIα levels through their high-affinity interaction.⁶ Recently, the INTERVAL study revealed a link between 1 of the G haplotype SNVs, rs10800453, and TFPI plasma levels.⁵⁰ However, although a higher level of FV-short was found in the plasma of our F5-G/G group than in that of the F5-ref group, the trend toward increased levels of plasma TFPI and TFPIα did not reach statistical significance. The most likely explanation is the limited sample size, which was 20 as opposed to 3000 in the INTERVAL study. It is also possible that not all TFPIα in circulation is bound to FV-short. This is supported by a previous study that showed that ∼26% of TFPIα remains in plasma after complete depletion of FV.⁵⁷ 

The potential regulation of TFPIα plasma levels is only 1 role that FV-short plays in the TFPI pathway. TFPIα anticoagulant functions are also synergistically enhanced by FV-short, together with protein S.^{26,27,38,52,58} Thanks to recombinant proteins and in vitro assays, we know that FV-short enhances TFPIα anticoagulant function by forming a complex, together with protein S.^27,52 Therefore, in contrast with TFPIα quantification, testing TFPI functionality reflects increased TFPIα levels and functional enhancement by FV-short. We found that the F5-G/G group had extended LTs for coagulation when compared with the F5-ref group (Figure 2; Tables 3 and 4). Because thrombomodulin was not used in these assays, this effect was APC-independent. Instead, we showed that it is TFPI-dependent, as illustrated by the abolition of the LT difference between the groups when evaluated in the presence of anti-TFPI Abs.

The clinical implications of the increased FV-short levels and TFPI functionality in individuals with F5-G/G were apparent when investigated using the UKB samples.⁴⁴ The G haplotype was, as expected, associated with a significant reduction in VTE. However, although it reduced the risk for thrombosis, this came hand-in-hand with an increased risk for bleeding, suggesting that there is an important and delicate balance between FV-short–TFPIα levels and the function in thrombotic and bleeding regulation.

In future studies it is important to determine which of the G haplotype SNVs influence alternative splicing and how. It has been proposed that exon 13 harbors regulatory elements that control splicing at the alternative FV-short splice sites and that their removal by the rare FV-Atlanta deletion increases FV-short at least 200-fold.²⁴ Given that 3 of the G haplotype SNVs overlie the region of exon 13 that is deleted in FV-Atlanta, we speculate that 1 or more may be abrogating the function of these regulatory elements, albeit to a much lesser extent than FV-Atlanta. Future examination of this hypothesis may provide further insight into the molecular pathogenesis of the G haplotype and the physiological regulation of FV-short splicing.

In summary, we provide evidence that the VTE protection afforded by the F5 G haplotype is mediated by an enhancement of alternative splicing at the FV-short splice sites with a consequent increase in plasma FV-short and a TFPIα-dependent delay in thrombin generation. This is the first time, to our knowledge, that FV-short has been implicated as a determinant of thrombotic risk and paves the way for further studies that could consider modulation of plasma FV-short as a novel anticoagulant strategy.

The authors thank the National Institute for Health and Care Research (NIHR) BioResource volunteers for their participation and acknowledge the NIHR BioResource centers, National Health Service (NHS) Trusts, and staff for their contribution. The authors thank the NIHR, NHS Blood and Transplant (NHSBT), and Health Data Research UK as part of the Digital Innovation Hub Programme. The authors thank the NIHR BioResource staff members who organized the recall of the NIHR BioResource volunteers who participated in the “genetic analysis of platelets in healthy individuals” study (supplemental Methods). The authors thank NHSBT for sponsoring and hosting this study. Lindsay Walker (University of Cambridge) and Verity Kew (NHSBT) kindly assisted with appointment scheduling and logistics on the 3 study visit days.

M.C.S is currently supported by a British Society for Haematology Early-Stage Research Grant (38047) and was previously funded by a Medical Research Council (MRC) Clinical Research Training Fellowship (MR/R002363/1). L. Stefanucci was a PhD student supported by the British Heart Foundation (BHF) Cambridge Centre for Research Excellence (RE/18/1/34212). J.H.C. was supported by an MRC Clinical Research Training Fellowship (MR/P02002X/1). Research in the Ouwehand laboratory received funding from the International Society on Thrombosis and Haemostasis, Medical Research Council, NHSBT, and NIHR. M.G., M.F., and J.A. are supported by BHF grants (FS/18/53/33863 and PG/20/13/34994). M.F. is supported by the National Institute for Health and Care Research Exeter Biomedical Research Centre. This work was performed by using resources provided by the Cambridge Service for Data Driven Discovery operated by the University of Cambridge Research Computing Service, provided by Dell EMC and Intel using Tier-2 funding from the Engineering and Physical Sciences Research Council (EP/P020259/1), and Distributed Research using Advanced Computing (DiRAC) funding from the Science and Technology Facilities Council. The academic coordinating center for INTERVAL was supported by core funding from the NIHR Blood and Transplant Research Unit (BTRU) in Donor Health and Genomics (NIHR BTRU-2014-10024), NIHR BTRU in Donor Health and Behaviour (NIHR203337), UK Medical Research Council (MR/L003120/1), British Heart Foundation (SP/09/002; RG/13/13/30194; RG/18/13/33946), and NIHR Cambridge BRC (BRC-1215-20014; NIHR203312). This work was supported by Health Data Research UK, which is funded by the UK Medical Research Council, Engineering and Physical Sciences Research Council, Economic and Social Research Council, Department of Health and Social Care (England), Chief Scientist Office of the Scottish Government Health and Social Care Directorates, Health and Social Care Research and Development Division (Welsh Government), Public Health Agency (Northern Ireland), British Heart Foundation, and Wellcome. RNA sequencing in the INTERVAL study was funded as part of an alliance between the University of Cambridge and the AstraZeneca Centre for Genomics Research, and by the NIHR Cambridge Biomedical Research Centre (BRC-1215-20014).

A. Tokolyi was supported by the Wellcome Trust (PhD studentship 222548/Z/21/Z).

The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health and Social Care in England.

Contribution: M.C.S. and J.A. wrote the manuscript; M.G., W.H.O., and J.T.B.C. revised the manuscript; M.C.S. and NIHR BioResource recruited participants into the study; M.C.S. and J.C.S. processed participant samples; M.G. and J.C.S. performed experiments; M.C.S., M.G., A. Tokolyi, L. Stefanucci, L. Sun, J.H.C., E.P., N.S.G., and J.A. analyzed data; M.C.S., M.G., A. Tolios, W.H.O, J.T.B.C., and J.A. designed the study; M.C.S., A.S.B., K.D., W.T., W.H.O, J.T.B.C., and J.A. conceived the project; E.E.D., E.D.A., M.I., D.S.P., W.H.O., J.T.B.C., A.S.B., M.F., and J.A. supervised the project.

Conflict-of-interest disclosure: M.C.S. reports receiving an honorarium from Sobi. The current affiliation for L. Sun is Regeneron Genetics Center LLC, Tarrytown, NY. D.S.P. reports being a full-time employee and stockholder at AstraZeneca. A.S.B. reports receiving grants outside of this work from AstraZeneca, Bayer, Biogen, BioMarin, Merck, Novartis, Pfizer, and Sanofi. W.T. reports receiving honoraria from Pfizer, Bayer, Takeda, Sobi, Novo Nordisk, Alexion, Portola, CSL Behring, AstraZeneca, Sanofi, and Ablynx and serving on the scientific advisory board of Sanofi, Daiichi Sankyo, Ablynx, Grifols, LFB Biopharmaceuticals, AstraZeneca, and Takeda. W.H.O. and N.S.G. report serving as consultants for Thermo Fisher Scientific. J.A. reports obtaining research funding from AstraZeneca that is unrelated to this study. The remaining authors declare no competing financial interests.

A complete list of the members of the NIHR BioResource appears in “Appendix.”

Correspondence: Matthew C. Sims, Department of Haematology, University of Cambridge, National Health Service Blood and Transplant Building, Long Rd, Cambridge, CB2 0PT, United Kingdom; email: matthewsims@nhs.net; and Josefin Ahnström, Centre for Haematology, Department of Immunology and Inflammation, Imperial College London, Hammersmith Hospital Campus, 5th Floor Commonwealth Building, Du Cane Rd, London, W12 0NN, United Kingdom; email: j.ahnstrom@imperial.ac.uk.

NIHR BioResource study group members: Sofie Ashford, Jyoti Khadake, Rachel Linger, Paul Roberts, Hannah Stark, and Hannah Williams.