Introduction

Blood coagulation is an important process in preventing blood loss from damaged vessels, but can also be responsible for thrombosis leading to ischaemic heart disease, stroke or venous thromboembolism.1 An informative measure of efficacy of the intrinsic coagulation pathway is activated partial thromboplastin time (aPTT), measured as time taken for a clot to form in plasma in the absence of tissue factor following introduction of an activator (eg, silica). An abnormally short aPTT can indicate a hypercoaguable state in acute coronary syndromes,2 and is associated with increased risk of venous thrombosis,3, 4, 5 whereas abnormally long aPTT may also indicate thrombotic risk in the case of the lupus anticoagulant.6 Addition of activated protein C (APC), which deactivates factors Va and VIIIa, and calculation of an APC resistance provides one measure of APC resistance,7 including effects of factor V Leiden mutation as its major determinant.8, 9 There is evidence to suggest aPTT is highly heritable,10 thus meriting investigation of its genetic basis, but to date the only high-density genetic association study of aPTT was a recently reported genome-wide association study (GWAS) of aPTT in 1477 subjects from the Lothian Birth Cohorts, which identified three novel loci associated with aPTT, namely: coagulation factor XII (F12), kininogen 1 (KNG1) and histidine-rich glycoprotein (HRG).11 The IL3581Thr variant in KNG1 has since been found to associate with risk of venous thrombosis as well as aPTT.12 No GWAS has been reported for APC resistance.

Materials and methods

Subjects were from the British Women’s Heart and Health Study, a prospective cohort study of heart disease in British women. Baseline recruitment was 1999–2001 (age 60–79 years), with blood samples for DNA, APC ratio and aPTT measurement taken from consenting individuals. Protocols and consents were approved by relevant research ethics committees.13 aPTT measurements were available on 2962 women (mean age 68.8 years, SD 5.5), and APC resistance on 2953 women (mean age 68.8 years, SD 5.5). Data are not available where either insufficient blood was available to assay, consent was not given or assays failed in the laboratory.

DNA was extracted from whole blood using a salting-out procedure.14 Genotyping was successfully performed on 3445 of 3838 available samples using the Illumina HumanCVD Beadchip.15 Principal components analysis was used to check self-reported ancestry, with 32 individuals excluded to avoid stratification issues, leaving 3413 samples for analysis. aPTT and APC resistance were assayed in an automated coagulometer (MDA-180, Organon Teknika, Cambridge, UK) using reagents and standards from the manufacturer as previously described.16 APC ratio was assayed without factor V-deficient plasma. Citrated plasma samples were stored at −80 °C for a maximum of 12 months before assay. Genotype and phenotype data were available on 2618 women for aPTT (mean age 68.9 years, SD 5.5) and 2610 women for APC resistance (mean age 68.9 years, SD 5.5).

Analysis of genetic association was performed using linear regression without covariables (adjustment for age had little effect; whereas clotting phenotypes are age dependent this cohort are all post-menopausal and within the relatively narrow age-range 60–79 years) using PLINK.17 Single-nucleotide polymorphisms (SNPs) out of Hardy–Weinberg equilibrium (P<0.0001) were excluded, as were any with a minor allele frequency below 0.1%, leaving 36 529 SNPs for analysis. Both traits were natural log transformed, outliers >2.5 SD from the mean were removed (on the basis that extreme values may represent either technical errors or biological abnormalities unrelated to common polymorphic variants, which are the focus of our analyses), and warfarin users excluded, leaving 2510 participants with non-missing data for aPTT (arithmetic mean 30.06 s, SD 1.103 s) and 2500 with non-missing data for APC resistance (arithmetic mean 2.924, SD 1.134). Exclusion of women on hormone replacement therapy (shown to associate with these measures16) was evaluated, but did not substantially change the results. A stringent (given non-independence of many SNPs) Bonferroni correction for 36 529 tests gives a threshold of P=1.37 × 10−6 as equivalent to a single-test P=0.05. Variable selection was performed in R using Akaike Information Criterion (AIC)18 with the stepAIC function of the ‘MASS’ library.

Results

Results of the HumanCVD BeadArray-wide association analysis with aPTT and APC resistance are presented in Table 1, with results of variable selection presented in Table 2. The SNPs most significantly associated with aPTT are in or near the F12 gene on chromosome 5. The top SNP rs2545801 is a non-coding SNP upstream of F12, P=1.39 × 10−59), with an (antilogged) per-allele effect of 1.05 s on aPTT (95% CI 1.04–1.06). Other gene regions showing association include the HRG/KNG1 region on chromosome 3 (top hit SNP rs710446, P=2.68 × 10−19), the ABO blood group (ABO) locus on chromosome 9 (rs657152, P=2.45 × 10−11) and the kallikrein B (KLKB1) region on chromosome 4 (rs4253304, P=1.67 × 10−7). Variable selection identified multiple statistically independent signals at each locus except KLKB1 (Table 2).

Table 1 Associations between SNPs and either aPTT or APC resistance
Full size table
Table 2 Variable selection results
Full size table

Table 1 also presents genetic association results for APC resistance. The most strongly associated SNP with APC resistance is the factor V Leiden mutation (rs6025, P=4.2 × 10−104) in the factor V (F5) gene on chromosome 1. The other region associated with APC resistance is the HRG region on chromosome 3 (top SNP rs16860992, P=2.29 × 10−15).

Variable selection (Table 2) suggests that all association with APC resistance in the F5 and solute carrier family 19 member 2 (SLC19A2) region on chromosome 1 is attributable to the functional factor V Leiden mutation, with no evidence of statistically independent effects for other SNPs. In the HRG region on chromosome 3 there are three potentially independent SNPs.

Discussion

Although the HumanCVD array is a candidate gene array, the coagulation pathway is well represented, with SNPs in the genes for the majority of intrinsic and extrinsic pathway proteins (Table 3). We confirmed previous reports11 of effects in F12 (our ‘top hit’ rs2545801 is the best HumanCVD tag of rs2731672, HapMap r2=0.93519), KNG1 (‘top hit’ rs710446) and HRG (rs9898, significantly associated, but not our top hit at this locus). We also found positive associations with aPTT at the G protein-coupled receptor kinase 6 (GRK6) gene, genomically adjacent to the F12 gene, although low LD between the ‘top hit’ SNPs at each locus suggests that these are marking independent effects (even if both the effects are actually in the F12 gene). GRK6 deactivates G protein-coupled receptors, and thus may potentially also have a biological effect in the clotting mechanism. The results of variable selection suggest that there may be more than one causal site at each of the three main loci (excluding KLKB1).

Table 3 Representation of coagulation factor genes on the HumanCVD array
Full size table

We also found significant associations between aPTT and SNPs at the ABO and KLKB1 loci. Blood group O versus non-O becomes associated with lower levels of factor VIII and von Willebrand factor (vWF) during childhood20 and continues into adulthood.21 Assuming this relationship is causal, and given that aPTT is prolonged with both severe von Willebrand Disease (vWF deficiency in type 1 and 3) and Haemophilia A (factor VIII deficiency), we hypothesise that ABO genotype could associate with aPTT through alteration of levels of vWF or factor VIII. There is also a previous report describing association of ABO OO genotype with aPTT using a combined linkage and association approach.22 Our highest ABO locus association is with rs657152, which is in high linkage disequilibrium (LD, r2=0.98)23 with rs8176719 (the O/non-O variant), and thus rs657152 closely marks the association of O blood group with clotting. SNP rs657152 is also in high LD (r2=0.93) with the myocardial infarction risk SNP reported by Reilly et al,24 and hence likely to tag the functional mechanism of that risk. KLKB1 encodes plasma kallikrein B (Fletcher factor) 1, a glycoprotein, which is involved in the intrinsic coagulation pathway,25 and also neighbours the F11 locus, encoding the factor XI protein, an important factor in the intrinsic coagulation pathway.

We also present results for genetic associations with APC resistance. The HumanCVD array directly assays the factor V Leiden mutation (rs6025), which is known to influence the APC resistance.26 This mutation shows the strongest genetic association with APC resistance in our data set (P=4.2 × 10−104). Although our other association signals in the F5 gene are with SNPs in little LD with rs6025 (eg, r2=0.104), the magnitude of signal with rs6025 and results of variable selection (Table 2) suggest these SNPs are simply showing a ‘bystander’ effect. The other locus containing SNPs associating with APC resistance is HRG (same SNPs as with aPTT, and showing consistent direction of effect on both tests). HRG has a complex role in coagulation, with both anticoagulant and antifibrinolytic properties reported.27, 28 In our data, we observe concordant effects of HRG genotype on both aPTT (clotting speed) and APC resistance (response to inhibition). Although there is a SNP in SLC19A2 (the gene for solute carrier family 19 (thiamine transporter), member 2) this is physically close to the F5 gene on chromosome 1, so may simply tag functional variation in F5. Although variable selection excludes the SLC19A2 SNP (Table 2), the LD between all our top hits in F5 (including rs6025) and the SLC19A2 SNP is very low (r2<0.006), suggesting this may either mark an independent effect in the F5 gene, or a biological relevance of thiamine transport in coagulation.

With the exception of the factor V Leiden mutation (rs6025), already known to influence APC resistance,26 the majority of these results represent relatively small genetic effects on aPTT and APC resistance. They therefore have very limited predictive value (especially as individual variants), but instead offer additional insight into the functional pathways underlying blood coagulation.

Our study has three principal limitations: (i) the HumanCVD array is not ‘genome-wide’. Although even genome-wide arrays do not capture all genetic variation they offer a relatively unbiased representation of the genome. Table 3 illustrates the extent to which this candidate gene array represents coagulation system genes. (ii) The population we have analysed is female, of European ancestry, and represents a fairly narrow age-range (post-menopausal, 60–79 years). The results may therefore not be generalisable to other ancestries, males or younger people. Further studies are needed to examine the associations of newly identified genotypes with risk of venous and arterial thrombosis. (iii) These phenotypes (in particular APC resistance) are infrequently measured on a cohort scale, and we were unable to identify a suitable replication cohort with both these measures and appropriate genotyping data. Appropriate caution should therefore be applied in interpreting results close to our significance threshold. Our replication of published aPTT GWAS results11 and the very strong statistical evidence (most of our reported P-values several orders of magnitude below the nominal HumanCVD significance threshold of 1 × 10−6) support the reliability of these findings.

In conclusion, we have both confirmed previous reports that F12/GRK6, KNG1 and HRG are associated with aPTT11 and identified new SNPs at ABO and new genomic locus KLKB1 associated with aPTT. We also present the first high-density genetic association analysis of APC resistance, and identify signals in the F5 and HRG genomic regions. Our findings suggest that KLKB1 and HRG may have potentially important roles in blood coagulation.