Genomeâwide analysis of genetic determinants of circulating factorÂVIIâactivating protease (FSAP) activity

Summary

Essentials.

• Knowledge of genetic regulators of plasma factor VII activating protease (FSAP) levels is limited.

• We performed a genomeâwide analysis of variants influencing FSAP activity in Scandinavian cohorts.

• We replicated an association for Marburgâ1 and identified an association for a HABP2 stop variant.

• We identified a novel locus near ADCY2 as a potential additional regulator of FSAP activity.

Summary

Background

FactorÂVIIâactivating protease (FSAP) has roles in both coagulation and fibrinolysis. Recent data indicate its involvement in several other processes, such as vascular remodeling and inflammation. Plasma FSAP activity is highly variable among healthy individuals and, apart from the lowâfrequency missense variant MarburgâI (rs7080536) in the FSAPâencoding gene HABP2, determinants of this variation are unclear.

Objectives

To identify novel genetic variants within and outside of the HABP2 locus that influence circulating FSAP activity.

Patients/Methods

We performed an exploratory genomeâwide association study (GWAS) on plasma FSAP activity amongst 3230 Swedish subjects. Directly genotyped rare variants were also analyzed with geneâbased tests. Using GWAS, we confirmed the strong association between the MarburgâI variant and FSAP activity. HABP2 was also significant in the geneâbased analysis, and remained significant after exclusion of MarburgâI carriers. This was attributable to a rare HABP2 stop variant (rs41292628). Carriers of this stop variant showed a similar reduction in FSAP activity as MarburgâI carriers, and this finding was replicated. A secondary genomeâwide significant locus was identified at a 5p15 locus (rs35510613), and this finding requires future replication. This common variant is located upstream of ADCY2, which encodes a protein catalyzing the formation of cAMP.

Results and Conclusions

This study verified the MarburgâI variant to be a strong regulator of FSAP activity, and identified an HABP2 stop variant with a similar impact on FSAP activity. A novel locus near ADCY2 was identified as a potential additional regulator of FSAP activity.

Introduction

FactorÂVIIâactivating protease (FSAP) is a plasma serine protease targeting various substrates. FSAP is mainly produced by the liver, and is present in the circulation as an inactive proenzyme. Activation is triggered by histones released from apoptotic or necrotic cells 1. The protein was initially separately shown to be involved in fibrinolysis, as an activator of singleâchain proâurokinase, and as an activator of FVII 2, 3, 4. More recently, FVII was recognized as a poor FSAP substrate 5, 6, and tissue factor pathway inhibitor (TFPI) was identified as a novel substrate. In line with this, impaired FSAP modulation of TFPI levels was suggested as an explanation for the defective thrombus formation observed in mice made deficient for FSAP (FSAP^â/â) 7. Other studies have shown that FSAP has a role not only in vascular compartments 8, 9, 10, but also in liver fibrosis 11, 12, inflammation 13, 14, 15, 16, and cancer 17, 18. InÂvivo experiments in wildâtype and FSAP^â/â mice support a role for FSAP in vascular remodeling, liver fibrosis, neointima formation, and arteriogenesis 11, 19, 20, 21.

Epidemiological studies have shown that circulating FSAP activity is increased in women as compared with men, and is further enhanced by pregnancy or the use of oral contraceptives 22, 23, 24. FSAP activity is also increased in subjects with deep vein thrombosis 25 orÂwith coronary heart disease 26 as compared with controls. We have found that traditional vascular risk factors explain very little of the variation in plasma FSAP activity, i.e. <Â10% in healthy individuals 27. We have also reported on increased FSAP activity in ischemic stroke cases as compared with controls 27. Furthermore, a region near the FSAPâencoding gene hyaluronanâbinding proteinÂ2 (HABP2) was recently identified as being associated with youngâonset stroke 28.

An early study of interindividual plasma levels of FSAP in healthy subjects discovered individuals with markedly reduced FSAP activity that was not related to antigen levels 23. Low levels of FSAP activity were found to be associated with the minor allele of the soâcalled MarburgâI (MI) singleânucleotide polymorphism (SNP) (rs7080536), which introduces an amino acid change, Gly534Glu (NP_004123.1), in the FSAP protein 29. The MIâSNP has been associated with several disease processes, such as carotid stenosis 30, stroke 31, and liver fibrosis 12. Associations between the MIâSNP and venous thrombosis have also been reported 32, 33, but inconsistent results do exist 25, 34, 35, 36.

To our knowledge, only three studies have searched for genetic variants associated with FSAP activity, and these studies were restricted to variants within HABP2 16, 27, 29. We hypothesized that there are additional genetic variants that contribute to the variation in circulating FSAP activity. Hence, we set out to test this hypothesis by conducting the first genomeâwide association study (GWAS) of this quantitative trait. We used a genotyping platform that is enriched with exome content, and searched for both common and rare genetic variants associated with FSAP activity.

Methods

Study design

This study included 3230 participants from the Malmà Diet and Cancer (MDC) study and the Sahlgrenska Academy Study on Ischemic Stroke (SAHLSIS).

The MDC study is a populationâbased prospective study that has been described in detail elsewhere 37. In brief, all men and women living in the Malmà area in southern Sweden and born in 1923â1950 were invited to participate. The participation rate was 41% 38. The present study is based on the MDC Cardiovascular Cohort, which randomly selected participant from the MDC study 39. We included subjects without both prevalent (i.e. at baseline) and incident cardiovascular disease (CVD) with biobanked plasma available for FSAP activity measurements (nÂ=Â2030).

The SAHLSIS is a caseâcontrol study for which FSAP activity has been reported 27. In brief, 600 patients with ischemic stroke at ages 18â69Âyears were consecutively recruited at stroke units in western Sweden 40. Controls (nÂ=Â600) were selected from populationâbased health surveys or registers to match the cases with regard to age, sex, and geographical region 40. Only control subjects without a history of CVD or signs of ischemic heart disease on electrocardiogram were included 41.

Sample sizes and baseline characteristics for the two studies are summarized in TableÂ1. The studies were approved by the ethics committee at the respective universities. All participants or their next of kin gave informed consent.

Table 1.

Characteristics of participants

	MDC study	SAHLSIS	Total
	(nÂ=Â2030)	(nÂ=Â1200)	(nÂ=Â3230)
Ischemic stroke cases, n (%)	0 (0)	600 (50)	600 (18)
Age (years), median (IQR)	58 (53â63)	59 (52â65)	58 (52â63)
Male sex, n (%)	797 (39)	770 (64)	1567 (49)
Hypertension*, n (%)	1227 (60)	578 (48)	1805 (56)
Diabetes mellitusâ, n (%)	170 (8)	147 (12)	317 (10)
Current smoking, n (%)	429 (21)	342 (29)	771 (24)
Hyperlipidemiaâ, n (%)	1822 (90)	816 (68)	2638 (82)
BMI (kgÂmâ2), median (IQR)	25.3 (23.1â27.7)	26.0 (23.8â28.7)	25.5 (23.4â28.2)
hsCRPÂ (mgÂLâ1), median (IQR)	1.2 (0.6â2.7)	1.9 (1.0â4.1)	1.5 (0.7â3.2)
FSAP activity (mUÂmlâ1), median (IQR)	938 (778â1100)	1152 (981â1334)	1008 (822â1192)
Genotyping platform	HumanOmniExpress Exome BeadChip v1.0	HumanOmniExpress Exome BeadChip v1.0, and HumanOmni 5M Exome v1.0Â	Imputed to the UK10KÂ+Â1000 Genomes Phase 3

BMI, body mass index; FSAP, factorÂVIIâactivating protease; hsCRP, highâsensitivity Câreactive protein; IQR, interquartile range; MDC, Malmà Diet and Cancer; SAHLSIS, Sahlgrenska Academy Study on Ischemic Stroke. *Hypertension was defined as pharmacological treatment for hypertension and/or a systolic blood pressure of âÂ160ÂmmÂHg and/or a diastolic blood pressure of âÂ90ÂmmÂHg. âDiabetes mellitus was defined as dietary or pharmacological treatment for diabetes and/or a fasting glucose level of âÂ7.0ÂmmolÂL^â1 or a fasting blood glucose level of âÂ6.1ÂmmolÂL^â1. âHyperlipidemia was defined as pharmacological treatment for hyperlipidemia and/or a total fasting serum cholesterol level of >Â5.0ÂmmolÂL^â1 and/or an LDL level of >Â3.0ÂmmolÂL^â1. ÂhsCRP levels in the two studies were determined as described previously 61, 62. ÂIncluded in the NINDS Stroke Genetic Network study, nÂ=Â444 ischemic stroke cases.

FSAP activity measurement

Venous blood samples were collected in tubes containing 10% by volume of 0.13ÂmolÂL^â1 sodium citrate. Aliquots of plasma were stored at âÂ80ÂÂC. For the prospective MDC study, blood samples were drawn during the baseline examinations in 1991â1996. For the SAHLSIS, blood sampling was performed in 1998â2003; at enrollment for controls, and at 3âmonth followâup for cases 27, 40. Plasma levels of FSAP activity were measured with an immunocapture activity test, as previously described 23, 27. These measurements were performed in 2010 and 2014 for the SAHLSIS and the MDC study, respectively. The interassay and intraâassay coefficients of variation (CVs) for FSAP activity were 14.7% and 4.7%, respectively, in the SAHLSIS, as reported in 27, and the interassay CV was 8.2% in the MDC study. FSAP activity values followed a normal distribution, and were not transformed for analysis.

Genotyping, imputation, and quality control (QC)

DNA was extracted from whole blood, and genotyping was performed with either the HumanOmniExpress Exome BeadChip v1.0 at the Broad Institute or the HumanOmni 5M Exome v1.0 as part of the Stroke Genetics Network (SiGN) study (Illumina, San Diego, CA, USA), and is described in detail elsewhere 42, 43. These arrays have anÂoverlap of the Omni content (coverage of common genomeâwide variation) and of the exome content (240k probes). The QC criteria were filtering for sex mismatch, call rate of <Â0.95, identical by descent sharing (>Â0.375), population outliers, excess autosomal heterozygosity, and deviation from HardyâWeinberg equilibrium (PÂ<Â10^â3). After separate QC for each dataset had been performed, genotypes available on both platforms were merged and subjected to repeat QC according to the above criteria. Chromosomes were phased by the use of shapeit2 (v2.r790 44) and imputed against the UK10KÂ+Â1000 Genomes PhaseÂ3 reference panel at the Sanger Imputation Service mapped to hg19 45. Approximately 10ÂÃÂ10⁶ variants with a minor allele frequency (MAF) of >Â0.01 and an information score of >Â0.3 were brought forward for analysis. After QC, the total number of subjects included in the analyses was 3126.

Genomeâwide association analyses

Genotypeâphenotype association analyses were performed according to a prespecified analysis plan on variants with an MAF of >Â0.01. The association between variants and FSAP activity was assessed with a linear model, on the assumption of an additive effect of each risk allele, with plink versionÂ1.9. All analyses were adjusted for age, sex, study (i.e. the MDC study or the SAHLSIS to adjust for studyâassociated biases such as variations in year of sample collection, plasma storage time, and batch effects on FSAP activity measurements), and ischemic stroke caseâcontrol status. Selected MDC study participants were free of both prevalent and incident stroke, so all MDC study subjects were assigned control status. The thresholds of genomeâwide significance and suggestive association were set at the conventional levels of PÂ=Â5ÂÃÂ10^â8 and PÂ=Â1ÂÃÂ10^â5, respectively. On the basis of our sample size and on the assumption of an additive model, the minimal effect size (Îâcoefficient) per allele that is detectable with 80% power is 0.27 for a MAF of 10%, 0.37 for a MAF of 5%, and 0.78 for a MAF of 1%. Conditional analyses were performed on the lead SNPs for their respective chromosomal regions.

Geneâbased analyses

All directly genotyped functional variants (missense, nonsense and splice variants) with a MAF of <Â0.05 that passed QC were included in the geneâbased tests. Genes were required to contain at least two variants to be included in the analysis (71Â134 variants in 11Â425 genes in total). Geneâbased tests were performed with the optimal combination of the sequence kernel association test and the burden test (SKATâO) 46 in r package skat v.1.0.7, with adjustment for age, sex, caseâcontrol status, and study. A Bonferroniâcorrected Pâvalue threshold of 4.4ÂÃÂ10^â6 was used (0.05/11Â425 genes).

Owing to the association of multiple rare variants in HAPB2 with FSAP activity, we removed one rare variant at a time and repeated SKATâO to determine the impact of each variant on the geneâbased association.

Genotyping of variants associated with FSAP activity in additional cohorts

The MIâSNP, rs35510613 and rs41292628 were genotyped in 665 subjects from the Venous Thromboembolism in Pregnancy (VIP) study from Norway 47, 48 and in 276 healthy subjects from the Danish Risk Score (DanRisk) study 49, which have measured FSAP activity with the same assay as was used in the present study. In brief, the VIP study included 313 women with pregnancyârelated venous thromboembolism and 353 controls. The DanRisk study included 155 women and 121 men born in either 1949 or 1959. Genotyping was performed at the University of Oslo (Norway) and at LGC genomics (UK) with KASPar genotyping chemistry. The studies were approved by the respective Norwegian and Danish regional committees on medical health research ethics, and all participants gave their written informed consent to participate.

Annotation and functional prediction of variants

Genetic variants of interest were visualized in the UCSC Genome Browser, with regional association plots 50, in HaploReg v4.1 51, in the Genbank SNP database, and in the Exome Aggregation Consortium (ExAC) 52. Prediction of functional effects of SNPs (PolyPhen and SIFT) were retrieved from the ExAC. Genetic variants associated with gene expression levels were identified in the GenotypeâTissue Expression Project (GTEx), and expression levels were analyzed in the GTEx and in BioGPS 53. Genetic variants with a correlation with lead SNPs (r ²Â>Â0.6) were identified in HaploReg v4.1, and additional functional predictions were assessed with RegulomeDB 54.

Cell culture

Mouse primary hepatocytes were isolated from BALB/c mice by collagenase perfusion as described previously 55. Hepatocytes were cultivated in Dulbecco's modified Eagle's medium High Glucose/F12 (1Â:Â1) supplementedÂwith 5Âmm sodium pyruvate (Thermo Fisher Scientific, Fermentas, Rockford, IL, USA), 10Âmm HEPES (SigmaâAldrich, St. Louis, MO, USA), 1Âmm lâglutamine (Thermo Fisher Scientific), 0.05% (v/v) NaH₂CO₃ (SigmaâAldrich), 10Âmm glucose (SigmaâAldrich), 10% fetal bovine serum (Thermo Fisher Scientific), 10ÂunitsÂmL^â1 penicillin, and 10ÂÎgÂmL^â1 streptomycin (Invitrogen, Darmstadt, Germany). Cells were maintained in a humidified atmosphere of 5% CO₂ at 37ÂÂC, and treated with 8â(4âchlorophenylthio)adenosine 3â,5ââcyclic monophosphate sodium (8âCPT) or forskolin (both from SigmaâAldrich).

RNA isolation and quantitative PCR analysis

Total RNA was extracted from hepatocytes with the total RNA Miniprep Kit (SigmaâAldrich). Reverse transcription was performed with the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Darmstadt, Germany). For realâtime PCR, the SensiMix SYBR Kit (Bioline, Luckenwalde, Germany) was used. Habp2 transcript levels were analyzed as described previously, and normalized against the reference gene Gusb 55. Primer sequences are shown in TableÂ2.

Table 2.

List of PCR primers used for realâtime PCR of Habp2

Target gene	Forward primer sequence	Reverse primer sequence
Gusb	AAAATGGAGTGCGTGTTGGGT	CCACAGTCCGTCCAGCGCCTT
Habp2	TTCCCGACACAGACGGAGA	GTCGTCCGGACCTATTTCA

Gusb, the gene encoding Îâglucuronidase.

Statistical analysis

We used linear regression to determine the amount of variance in FSAP activity explained by the covariates included in the GWAS model, i.e. age, sex, caseâcontrol status, and study. We then performed a model conditioned on these covariates to determine the remaining variance explained by a particular SNP or SNP combination, or by traditional vascular risk factors (hypertension, diabetes mellitus, smoking, and hyperlipidemia), body mass index, and highâsensitivity Câreactive protein (hsCRP, logâtransformed). Associations between FSAP activity and genotypes in the VIP and DanRisk studies were analyzed with linear regression adjusting for age, sex, and caseâcontrol status, or with Student's tâtest, as appropriate. Changes in Habp2 transcript levels in response to treatment as compared with control were analyzed with Student's tâtest.

Results

We conducted a GWAS and evaluated rare and lowâfrequency functional variants by using geneâbased tests in 3230 individuals of predominantly northern European ancestry to identify genetic loci influencing circulating FSAP activity. Characteristics of participants and genotyping arrays are summarized in TableÂ1. The majority of the participants (i.e. >Â80%) were populationâbased and free of CVD. Subjects homozygous for the minor allele of the MIâSNP had approximately fiveâfold lower FSAP activity than subjects homozygous for the major allele (TableÂ3), which is in line with previous reports 56. Above the variance in FSAP activity explained by the covariates adjusted for in the GWAS (i.e. age, sex ischemic stroke caseâcontrol status, and study), cardiovascular risk factors, anthropometrics and hsCRP explained only 3.6% of the variance in FSAP activity.

Table 3.

Plasma factorÂVIIâactivating protease (FSAP) activity for different MarburgâI genotypes

Genotype	Median FSAP activity (mUÂmLâ1)	FSAP activity IQR (mUÂmLâ1)	No.
MI:GG	1032	862â1208	2898
MI:AG	592	510â712	224
MI:AA	207	49â366	4

IQR, interquartile range; MI, MarburgâI (rs7080536).

GWAS

After adjustment for age, sex, study, and ischemic stroke caseâcontrol status, 164 variants were genomeâwide significant, and 239 additional variants were suggestively associated with FSAP activity (Fig.Â1; TableÂS1). The quantileâquantile plot for the GWAS revealed more variants with lower observed Pâvalues than expected (Fig.Â1B). The genomic control (lambda) was 1.016. Associated loci with PÂ<Â1ÂÃÂ10^â6 are shown in TableÂ4 and outlined below.

Figure 1.

Genomeâwide association analyses of factorÂVIIâactivating protease (FSAP) activity. (A) Manhattan plot of associations for FSAP activity. The dotted line shows genomeâwide significance (5ÂÃÂ10^â8). The plot is truncated at a Pâvalue of 10^â20. (B) Quantileâquantile plot for associations. (C, D) Regional association plots of the MarburgâI (MI)âsingleânucleotide polymorphism (SNP) (rs7080536) (C) and of rs1579587 (D), which showed suggestive association (PÂ=Â5.1ÂÃÂ10^â6) when the 10q25 region was adjusted for MIâSNP. Linkage disequilibrium (r ²) is indicated by the color scale.

Table 4.

Lead associated genetic loci (PÂ<Â1ÂÃÂ10^â6) for factorÂVIIâactivating protease activity

Locus	Lead variant	Chromosome: position of lead variant in hg19	Geneârelated position	Alleles (A1/A2)	Frequency	Î	Pâvalue
10q25.3	rs7080536	10: 115Â348Â046	Missense in HABP2	A/G	0.037	â429	7.0ÂÃÂ10â142
5p15.31	rs35510613	5: 7Â377Â210	Intergenic, upstream of ADCY2	â/G	0.77	â45	1.3ÂÃÂ10â8
12q21.31	rs75809015	12: 82Â823Â136	Intron in METTL25	G/A	0.012	194	2.1ÂÃÂ10â7
17q25.3	rs62073440	17: 81Â005Â762	Intron in B3GNTL1	C/T	0.11	59	4.6ÂÃÂ10â7
7p11.2	rs373067567	7: 57Â749Â605	Intergenic	CAT/C	0.61	â37	7.0ÂÃÂ10â7
16q24.1	rs61613787	16: 84Â728Â954	Intergenic, upstream of USP10	C/T	0.014	169	8.6ÂÃÂ10â7

ADCY2, adenylate cyclaseÂ2; B3GNTL1, UDPâGlcNAc:beta Gal Îâ1,3âNâacetylglucosaminyltransferaseâlikeÂ1; HABP2, hyaluronanâbinding proteinÂ2; METTL25, methyltransferaseâlikeÂ25; USP10, ubiquitinâspecific peptidaseÂ10.

The majority of the genomeâwide associations (nÂ=Â163) were located at the 10q25 locus near HABP2. The strongest association was found for the MIâSNP rs7080536 (PÂ=Â7.0ÂÃÂ10^â142; Fig.Â1C).

At the 5p15 locus, rs35510613 was genomeâwide significant (PÂ=Â1.3ÂÃÂ10^â8; Fig.ÂS1). This variant is located 19Âkbp upstream of the adenylate cyclaseÂ2 (ADCY2) gene. Two SNPs, rs12652415 and rs1609428, located in ADCY2 introns were in linkage disequilibrium (LD) with rs35510613 (r², 0.74 and 0.64; and Dâ, 0.89 and 0.96, respectively) and showed suggestive associations with FSAP activity (PÂ=Â1.1ÂÃÂ10^â6 and PÂ=Â4.3ÂÃÂ10^â6, respectively).

SNPs with suggestive associations with PÂ<Â1ÂÃÂ10^â6 were also located at 12q21 (rs75809015), 17q25 (rs62073440), 7p11 (rs373067567) and 16q24 (rs61613787) loci (TableÂ4).

Above the variance in FSAP activity explained by the covariates adjusted for in the GWAS (i.e. age, sex, ischemic stroke caseâcontrol status, and study), rs7080536 explained 18.6% of the remaining variance in FSAP activity, whereas rs35510613 explained 1%. Associations for these variants were also evaluated in a stratified analysis of study participants without stroke or CVD (i.e. all MDC study subjects and SAHLSIS controls). The effect sizes for both rs7080536 and rs35510613 were similar to those in the whole sample, and the same was true when MDC subjects only were analyzed (TableÂS2).

Conditional analyses on lead SNPs

To identify independent signals within each locus, conditional analyses were performed on the lead variants with PÂ<Â1ÂÃÂ10^â6. At the 10q25 locus, rs1539587 remained suggestively associated after adjustment for rs7080536 (PÂ=Â5.1ÂÃÂ10^â6). This variant is located in the neighboring nebulinârelated anchoring (NRAP) gene (Fig.Â1D). No other suggestive associations remained.

Geneâbased analyses

Next, we performed geneâbased association tests by using SKATâO. Two genes were significantly associated with FSAP activity after Bonferroni correction. HABP2 showed the strongest association (PÂ=Â8.2ÂÃÂ10^â120), and was represented by 10 variants. DCLRE1A (PÂ=Â3.0ÂÃÂ10^â16), which is also located on chromosomeÂ10q25, was represented by nine variants. The geneâbased analysis was repeated for chromosomeÂ10 without the carriers of the MIâSNP, and, in this analysis, only HABP2 remained significant (PÂ=Â8.9ÂÃÂ10^â9). NRAP (PÂ=Â1.5ÂÃÂ10^â5, represented by 22 variants) was ranked as the second most associated gene.

On testing of the contribution of each rare variant inÂHABP2 to the SKATâO Pâvalue, rs41292628 (MAF ofÂ0.002), which encodes a stop codon at amino acid positionÂ203, was identified as a determinant of FSAP activity (Fig.Â2). The FSAP activity in 13 carriers of thisÂstop variant was reduced as compared with that inÂnonâcarriers (IQR_{rs41292628:C/T}Â=Â404â592ÂmUÂmL^â1, IQR_{rs41292628:C/C}Â=Â825â1190ÂmUÂmL^â1, P _tâtestÂ<Â0.001). The association for rs41292628 was also independent of the NRAP rs1539587 variant (PÂ=Â2.5ÂÃÂ10^â16) after adjustment for both the MIâSNP and rs1539587 in a linear model. In subjects homozygous for the MI major allele (MI:GG), after adjustment for GWAS covariates, 3.5% of the remaining variance in FSAP activity was explained by rs41292628, rs35510613, and rs1539587.

Figure 2.

Analysis for factorÂVIIâactivating protease activity associations of rare variants (minor allele frequency of <Â5%) in HABP2. The âÂlog₁₀(Pâvalue) is shown from optimal combination of the sequence kernel association test and the burden test (SKATâO) of HABP2. For each rare variant presented in the graph, the variant was removed from the test and the Pâvalue for the geneâbased SKATâO analysis was determined. The analyses were performed in individuals homozygous for the major allele of the MarburgâI singleânucleotide polymorphism (MIâSNP:GG). The dashed line shows the Bonferroniâcorrected threshold Pâvalue of 0.00011 (0.05/447 genes) for geneâbased tests on chromosome 10 among MIâSNP:GGâcarrying subjects.

Genotypes for top associated variants in the VIP and DanRisk studies

The genotyping success rate was >Â95% for all variants. The known association between the MIâSNP and FSAP activity was detected in both studies (PÂ<Â0.001). One carrier of the stop variant was identified in the VIP study (MAF_rs41292628Â=Â0.0009) and six in the DanRisk study (MAF_rs41292628Â=Â0.011). Similarly to the discovery cohort, carriers of the stop variant had reduced FSAP activity (VIP, FSAP_{rs41292628:C/T}Â=Â770ÂmUÂmL^â1 versus interquartile range [IQR]_{rs41292628:C/C}Â=Â910â1210ÂmUÂmL^â1; and DanRisk, IQR_{rs41292628:C/T}Â=Â14â62% of reference plasma versus IQR_{rs41292628:C/C}Â=Â60â85% of reference plasma, P _tâtestÂ<Â0.001). In contrast, the association for rs35510613 on the 5p15 locus was not confirmed in either the Norwegian or the Danish study (PÂ>Â0.2 for both).

Annotation and functional prediction of top associated variants

For the top associated variants (TableÂ4), we also evaluated expression quantitative trait locus (eQTL) presence, and the putative regulatory function of variants in LD (TablesÂS3 and S4). There were, in total, 12 SNPs with r ²Â>Â0.6 with top associated variants (PÂ<Â1ÂÃÂ10^â6) in the CEU population dataset in the HaploReg database (TableÂS3). Of these, 11 were included in the GWAS. None of the top associated variants had a RegulomeDB score of <Â5. For rs62073440 and for 5p15 variants in LD with the lead SNP in this region (rs1265241 and rs1609428), eQTLs were identified towards B3GNTL1 and nonâproteinâcoding CTDâ2296D1.5 (TableÂS4), respectively.

The missense MIâSNP is known to influence FSAP activity, and was thus not evaluated further here, apart from the inÂsilico annotation provided in TableÂS3. The NRAP rs1539587 is also a missense mutation. The estimated effect of the minor allele on NRAP protein function is deleterious and benign according to PolyPhen and SIFT, respectively. This variant is also predicted to alter a putative regulatory motif sequence (TableÂS3). In BioGPS, the NRAP transcript was solely expressed in heart tissue, whereas GTEx indicated high gene expression also in skeletal muscle (Fig.ÂS2). However, except for the genomic proximity, we could not identify any clear biological or functional link between FSAP and NRAP. In contrast, the 5p15 locus, containing ADCY2, may be of functional relevance. Adenylate cyclase catalyzes the conversion of ATP to the secondary messenger molecule cAMP. Therefore, we investigated whether cAMP modifiers influence Habp2 expression inÂvitro, as described below. Given that circulating FSAP is mainly produced in the liver 2, hepatocytes were investigated.

Habp2 expression in hepatocytes in response to cAMP modifiers

In order to determine whether Habp2 mRNA levels were affected by cAMP modifiers, primary mouse hepatocytes were incubated with 8âCPT, an activator of the downstream cAMP activatorâdependent kinase (protein kinaseÂA [PKA]), or with forskolin, which stimulates cAMP production. Increases in Habp2 transcript levels were found after 48Âh of incubation with 8âCPT (PÂ<Â0.05 for both 50ÂÎm and 200ÂÎm) and after 24Âh of forskolin treatment (PÂ<Â0.05 for both 50ÂÎm and 200ÂÎm) relative to control (Fig.Â3).

Figure 3.

Habp2 expression in primary mouse hepatocytes in response to cAMP modifiers. (A) Cells were stimulated with 50ÂÎm or 200ÂÎm 8â(4âchlorophenylthio)adenosine 3â,5ââcyclic monophosphate sodium (8âCPT) in nÂ=Â4 biological replicates (nÂ=Â2 technical replicates). (B) Cells were stimulated with 20ÂÎm or 50ÂÎm forskolin in nÂ=Â5 biological replicates (nÂ=Â2 technical replicates). Data are presented as meanÂÂÂstandard error of the mean. Significance is compared with control (CTRL) at each time point *PÂ<Â0.05; **PÂ<Â0.005. FORSK, forskolin.

Discussion

We searched for genetic factors contributing to the variation in circulating FSAP activity by using GWAS and geneâbased analysis. We identified several signals at the 10q25 locus: (i) the lead SNP in the GWAS was the wellâknown missense variant in the FSAPâencoding HABP2 gene, i.e. MI (PÂ=Â7.0ÂÃÂ10^â142); (ii) suggestive significance remained for an SNP at the 10q25 locus in NRAP after conditioning for this MI variant; and (iii) in geneâbased analysis, HABP2 was genomeâwide significant, an HABP2 stop variant was identified as an additional determinant of FSAP activity, and NRAP showed suggestive association. In the GWAS, we also identified a novel genomeâwide significant locus on chromosomeÂ5 with the lead SNP located upstream of ADCY2. Taken together, variants within HABP2, NRAP and ADCY2 were found to explain 21.4% of the variance in FSAP activity, above the variance explained by the covariates in the GWAS (i.e. age, sex, ischemic stroke caseâcontrol status, and study). In comparison, cardiovascular risk factors, including hsCRP, explained only 3.6% of the variance in FSAP activity above the variance explained by the covariates included in the GWAS.

In the GWAS, we identified a high number of SNPs with genomeâwide significance at the 10q25 locus. The lead MIâSNP was reported to be associated with FSAP activity in 2002 29, and we confirmed here that homozygous carriers of the MI variant have approximately fiveâfold lower plasma FSAP activity than nonâcarriers 56. We also identified additional signals at the 10q25 locus that were independent of the MIâSNP, indicating a complex genomic architecture at this locus. After conditioning on the MIâSNP, rs1539587 in the neighboring NRAP gene remained suggestively associated with FSAP activity. After adjustment for GWAS covariates, the amounts of FSAP activity variance explained by the MIâSNP alone and together with rs1539587 were 18.6% and 19.2%, respectively.

HABP2 was also significantly associated with FSAP activity in the geneâbased analysis. When we omitted all MI carriers from this analysis, the association with HABP2 remained. This association was mainly attributable to a newly identified rare variant (rs41292628), which is located in the coding region and introduces a premature stop codon at amino acid positionÂ203 (NP_004123.1). This position is before the serine protease domain of FSAP, and thus total loss of function can be expected to result from the introduction of this stop codon. Carriers of the stop variant had FSAP activity in the same range as MI variant carriers. In line with our results, a study from the Netherlands reported on an anonymous serum donor with deficient FSAP protein who was homozygous for the stop variant 16. As a replication effort, we genotyped 941 subjects from Norway and Denmark. We identified seven heterozygous carriers of the stop variant for whom the FSAP activities were in the same range as for MI variant carriers. Our results also indicated a northâtoâsouth gradient of increasing MAF for the stop variant. As FSAP activity has been implicated in several pathophysiological processes (recently reviewed in 57), and the mechanisms remain elusive, future genetic studies that take into account both the HAPB2 MI and stop variant would be of interest. However, given the low frequency of these variants, this will require very large datasets genotyped with rare variant information (e.g. exome chips or exome sequencing).

Using the GWAS approach, we also identified a novel genomeâwide significance locus at 5p15. The lead SNP (rs35510613) is located upstream of ADCY2 and is in high LD with two SNPs in ADCY2 introns. The protein encoded by ADCY2 is one of 10 adenylate cyclase isoforms that catalyzes the formation of cAMP. We have shown that there is a significant correlation between plasma levels of FSAP antigen and FSAP activity 27, suggesting that, apart from variants affecting the FSAP protein itself, such as the MI and stop variants, gene variants influencing FSAP expression may influence FSAP activity. There are also data indicating that Habp2 mRNA and FSAP protein levels change in parallel 22, 58, 59, and experimental data from studies on hepatic rodent cells indicating that cAMP may be involved in regulating Habp2 expression 55.

In order to obtain more direct information on the potential involvement of the cAMPâdependent pathway in regulating Habp2 expression, we therefore performed inÂvitro studies on primary mouse hepatocytes. Using either forskolin or 8âCPT, we found increased mRNA levels of Habp2 in these cells. Although the results from both inÂvitro manipulations agree with a model in which increased cAMP levels may affect Habp2 transcription through PKA activation, direct functional characterization is needed to support a functional role for ADCY2 in regulating FSAP.

It is also of note that we were not able to replicate the association between rs35510613 and FSAP activity in the two other, smaller, Scandinavian studies. However, the replication effort was mainly targeted at the HABP2 stop variant, and we did not have sufficient statistical power to replicate the effect size for rs35510613. Thus, we believe that future independent replications in larger samples are warranted before further studies are performed to functionally characterize this variant.

The major strength of this study is that we analyzed ~Â10Âmillion markers spread throughout the genome, covering both common variants (assessed via GWAS) and rare variants (assessed via geneâbased testing) in 3230 wellâcharacterized individuals who were ethnically relatively homogeneous 43, and the majority were ascertained by populationâbased recruitment methods. This is thus the most comprehensive genetic study of circulating FSAP activity to date. However, in the context of genomeâwide association analysis, the sample size is still modest, and we thus have limited power to detect relatively small effect sizes, even for common variants. Another limitation is that we included participants with ischemic stroke, who, as a group, have increased FSAP activity as compared with controls. However, in a sensitivity analysis excluding stroke cases, we found similar effect sizes for the lead variants in HABP2 and ADCY2. Another strength of the study is the use of a standardized protocol for blood sampling in both cohorts, and the fact that all samples were analyzed with the same FSAP activity assay. A limitation is that blood sampling and FSAP activity measurements were performed on different occasions. However, analysis of the largest cohort (MDC study) alone provided similar effect sizes for the top findings. We used a wellâcharacterized FSAP activity assay, but a limitation of this assay is that FSAP hydrolyzes singleâchain proâurokinase, generating urokinase plasminogen activator, the activity of which is used as a proxy for FSAP activity. It is not known whether this covers all aspects of FSAP activity. It is thus plausible that an assay with another substrate may generate somewhat different findings. However, we have shown that the MI FSAP protein has reduced activity as compared with wildâtype FSAP towards all identified substrates that we have tested so far 21, 56. With regard to the rare variant analysis, we used an exome array, so rare variants that impact on FSAP activity may remain to be discovered with exome sequence analysis. Additionally, it should be emphasized that the results were generated in a population of predominantly European ancestry, and our findings may not be generalizable to populations of different ethnicity.

We confirmed that the HABP2 MI variant is a strong regulator of FSAP activity, and we identified and replicated a HABP2 stop variant with a similar impact on FSAP activity. We also identified a novel locus near ADCY2 as an additional potential regulator of circulating FSAP activity that requires future independent replication. Interestingly, the identified genetic variants explained ≈ 20% of the variation in FSAP activity, whereas cardiovascular risk factors explained <Â4%. Additional larger genetic studies, potentially employing future novel assays for the measurement of FSAP activity 60, may provide further insights into the genetic regulation of this plasma protease, and thus pave the way for future studies on its role in different pathophysiological pathways.

Addendum

C. Jern, S. M. Kanse, O. Melander, M. Olsson, and T. M. Stanne conceived the research design of the present study. C. Jern and A. Pedersen were responsible for the SAHLSIS (sample and phenotype contribution). G. EngstrÃm and O. Melander were responsible for the MDC study (sample and phenotype contribution). P.M. Sandset and A. F. Jacobsen were responsible for the VIP study (sample and phenotype contribution). J. J. Sidelmann and N. P. RÃnnow Sand were responsible for the DanRisk study (sample and phenotype contribution). A. MartinezâPalacian and E. Kara performed inÂvitro experiments and measurements of plasma FSAP activity, respectively. S. M. Kanse supervised inÂvitro experiments and measurements of plasma FSAP activity. E. Lorentzen, M. Olsson, and A. MartinezâPalacian performed statistical analyses. M. Olsson, T. M. Stanne, C. Jern and S. M. Kanse interpreted the data. M. Olsson and C. Jern: drafted the manuscript. S. M. Kanse, T. M. Stanne, O. Melander, G. EngstrÃm, P. M. Sandset, and J. J. Sidelmann intellectually reviewed the manuscript. All authors contributed to the last revision process and approved the version to be published.

Disclosure of Conflict of Interests

The authors state that they have no conflict of interest.

Supporting information

Fig.ÂS1. Regional association plot of rs35510613, which showed significant association for FSAP activity.

Fig.ÂS2. Tissue distributions of NRAP gene expression.

TableÂS1. Genomeâwide (PÂ<Â5ÂÃÂ10^â8) and suggestive significant loci (PÂ<Â1ÂÃÂ10^â5) for FSAP activity.

TableÂS2. Association of rs7080536, and rs35510613 with FSAP activity in study participants without stroke or CVD.

TableÂS3. Functional annotations from HaploReg 4v1, level of association with FSAP activity in present GWAS, and RegulomeDB score for top associated lead genetic variants (PÂ<Â1ÂÃÂ10^â6), of rs1539587, and variants in linkage disequilibrium (LD; r ²ÂâÂ0.6 in EUR population in HaploReg 4v1) with these variants.

TableÂS4. Expression quantitative trait loci (eQTLs) from the GTEx Portal identified for top associated lead variants (PÂ<Â1ÂÃÂ10^â6) for FSAP activity, and variants highlighted in this study.

Data S1. STROBE Statement â checklist of items that should be included in reports of observational studies

Olsson M, Stanne TM, Pedersen A, Lorentzen E, Kara E, MartinezâPalacian A, RÃnnow Sand NP, Jacobsen AF, Sandset PM, Sidelmann JJ, EngstrÃm G, Melander O, Kanse SM, Jern C. Genomeâwide analysis of genetic determinants of circulating factorÂVIIâactivating protease (FSAP) activity. J Thromb Haemost 2018; 16: 2024â34.