Influence of 23 coronary artery disease variants on recurrent myocardial infarction or cardiac death: the GRACE Genetics Study

Abstract

Aims

A pooled analysis of 14 genome-wide association studies revealed 23 susceptibility loci for coronary artery disease (CAD), thereby providing the most comprehensive genetic blueprint of CAD susceptibility. Here, we evaluated whether these 23 loci also predispose to recurrent myocardial infarction (MI) or cardiac death following an acute coronary syndrome (ACS).

Methods and results

A total of 2099 ACS patients enrolled in the Global Registry of Acute Coronary Events (GRACE) UK–Belgian study were prospectively followed for a median of 5 years (1668 days). C-allele carriers of the rs579459 variant, which is located upstream of the ABO gene and correlates with blood group A, were independently associated with recurrent MI [multivariable-adjusted hazard ratio (HR) 2.25, CI = 1.37–3.71; P = 0.001] and with recurrent MI or cardiac death [multivariable-adjusted (HR) 1.80, CI = 1.09–2.95; P = 0.021] within 5 years after an index ACS. The association of rs579459 was replicated in 1250 Polish patients with 6 months follow-up after an index ACS [multivariable-adjusted (HR) 2.70, CI = 1.26–5.82; P = 0.011 for recurrent MI]. Addition of rs579459 to a prediction model of 17 clinical risk factors improved risk classification for recurrent MI or cardiac death at 6 months as calculated by the integrated discrimination improvement method (P = 0.037), but not by C-statistics (P = 0.096).

Conclusion

In this observational study, rs579459 was independently associated with adverse cardiac outcome after ACS. A weak improvement in clinical risk prediction was also observed, suggesting that rs579459 should be further tested as a potentially relevant contributor to risk prediction models for adverse outcome following ACS.

rs579459, ABO locus, Genetics, Acute coronary syndrome, Myocardial infarction, Plaque rupture

Introduction

Recurrence of a myocardial infarction (MI) after a first coronary event is frequent and increases morbidity and mortality of coronary artery disease (CAD). To predict the risk of recurrent ischaemic and possibly lethal events, clinical risk scores have been developed, among which the GRACE Risk Score prevailed.¹ However, based on clinical variables only, risk prediction is not sufficiently robust, in particular for recurrent MI. Given the large burden of recurrent ischaemic events, improved prediction modalities are urgently needed.

The genetic basis of CAD is actively being explored. Genome-wide association (GWA) studies successfully identified genetic loci associated with the risk of CAD or MI. Interestingly, a pooled analysis of 14 GWA studies of CAD comprising 22 233 individuals with CAD (cases) and 64 762 controls of European descent, followed by validation of the association signals in 56 682 additional individuals, confirmed 10 loci as established risk factors for CAD, and further identified 13 new loci associated with CAD.² In total, this study consisted of 40 different cohorts of CAD patients, thereby representing by far the largest study for CAD susceptibility and providing the most complete genetic blueprint of susceptibility factors contributing to CAD.

Although the association between these 23 loci and CAD has indisputably been established, it has not been explored whether any of these loci contribute to the occurrence of acute coronary events in existing CAD. Since acute coronary events (both first and recurrent events) are the clinical manifestation of plaque instability and rupture, whereas CAD defines the underlying subclinical process characterized by plaque prevalence, genetic variants contributing to acute coronary events may differ from those predisposing to CAD (e.g. because they predispose to plaque instability rather than plaque formation).³ However, similar to some of the clinical risk factors, such as hypertension, genetic factors predisposing to both phenotypes may also overlap. The identification of variants predisposing to adverse cardiac outcome in CAD patients that experienced a first acute coronary syndrome (ACS) is particularly relevant, as they provide novel insights into the pathogenesis of recurrent coronary events, can be used to refine prospective identification of ACS patients at increased risk for adverse outcome, and may assist in the clinical decision-making of secondary prevention measures.

To test the hypothesis that genetic risk variants for CAD may also increase the risk of recurrent ischaemic events in CAD patients, we used the Global Registry of Acute Coronary Events (GRACE) program, which is a multinational study of patients hospitalized with an ACS across the geographically distinct populations in the UK, Belgium, and Poland. We first assessed which of the 23 CAD risk variants predispose to recurrent MI or a combined endpoint consisting of recurrent MI or cardiac death in the combined UK–Belgian discovery cohort with a median of 5 years of follow-up. Next, we replicated significant findings in an independent Polish validation cohort with a median follow-up of 6 months. Finally, we also assessed whether inclusion of these variants improved clinical risk classification models for recurrent MI or cardiac death.

Methods

Study populations and phenotype descriptions

The GRACE registry is a prospective observational study of the spectrum of ACS patients hospitalized between 1999 and 2010. The GRACE Genetics Study consists of 3585 patients of European ancestry whose DNA was collected by eight participating centres in the UK, Belgium, and Poland between 2001 and 2010. Collected blood samples were transferred to the core laboratory (The Queens Medical Research Institute, Edinburgh, UK), where DNA was extracted and stored at −80°C.

Full methodological details used in the GRACE registry were previously published.⁴ Briefly, patients were eligible if they were admitted with a clinical diagnosis of ACS and at least one of the following: electrocardiographic changes consistent with ACS, increases in serum biochemical markers of cardiac necrosis and/or documented CAD (see Supplementary material online for complete details). Demographic characteristics, medical history, presenting symptoms, biochemical and electrocardiographic findings, treatment, and outcomes were collected. All cases were assigned to one of the following categories using predefined criteria: (i) ST-elevation MI (STEMI; including left bundle-branch block), (ii) non-STEMI, (iii) unstable angina (UA), and (iv) other cardiac or (v) non-cardiac diagnoses.⁴

To determine the occurrence of cardiovascular complications, all patients were prospectively followed at 6 months (UK, Belgian, and Polish patients) and at a median of 5 years (UK and Belgian patients only).⁵ More details of the follow-up methods are given in the Supplementary material online. Outcome events were prospectively defined using standard definitions for the entire study, independent of local interpretations. The primary endpoint was recurrent MI within 5 years of follow-up in the combined UK–Belgian discovery cohort and recurrent MI within the first 6 months following the day of hospital admission in the Polish validation cohort. We excluded events that occurred prior to inclusion. A second endpoint consisted of recurrent MI or cardiac death. Non-cardiac death was deliberately excluded as an outcome event, since various other causes of death, which are non-cardiovascular in origin, might dilute the association with cardiac-specific events. Deaths were classified as cardiac unless a clear non-cardiac cause was diagnosed.⁴

Local hospital ethics committees approved the GRACE Genetics Study. All patients provided written informed consent for follow-up contact and DNA sampling at enrolment, in accordance with the declaration of Helsinki.

Genotyping of the 23 coronary artery disease susceptibility variants

We genotyped 23 genetic variants reported by Schunkert et al.² (Table 1). Genotyping was performed at the Vesalius Research Center (Leuven, Belgium) in a blinded manner, using iPLEX technology on a MassARRAY Compact Analyser (Sequenom, Inc., CA, USA; see Supplementary material online for complete genotyping details).⁶

Table 1

Overview of the 23 genotyped risk loci for coronary artery disease

Band	Variant	Gene(s) in region	Genotyping success rate (%)	Allele frequency (risk allelea)
1p13.3	rs599839	SORT1	97.7	0.79 (A)
1p32.2	rs17114036	PPAP2B	99.4	0.91 (A)
1q41	rs2133189	MIA3	97.6	0.74 (T)
2q33.1	rs6725887	WDR12	95.9	0.14 (C)
3q22.3	rs9818870	MRAS	97.1	0.17 (T)
6p21.31	rs17609940	ANKS1A	99.0	0.79 (G)
6p24.1	rs12526453	PHACTR1	96.2	0.66 (C)
6q23.2	rs12190287	TCF21	96.9	0.68 (C)
6q25.3	rs3798220	LPA	98.7	0.02 (C)
7q32.2	rs11556924	ZC3HC1	99.4	0.63 (C)
9p21.3	rs4977574	CDKN2A, CDKN2B	99.6	0.53 (G)
9q34.2	rs579459	ABO	99.7	0.21 (C)
10q11.21	rs1746048	CXCL12	97.9	0.88 (C)
10q24.32	rs12413409	CYP17A1, CNNM2, NT5C2	99.9	0.93 (G)
11q23.3	rs964184	ZNF259, APOA5-A4-C3-A1	65.9	0.11 (G)
13q34	rs4773144	COL4A1, COL4A2	99.5	0.45 (G)
14q32.2	rs2895811	HHIPL1	99.4	0.43 (C)
15q25.1	rs3825807	ADAMTS7	97.0	0.53 (T)
17p11.2	rs12936587	RASD1, SMCR3, PEMT	99.6	0.58 (G)
17p13.3	rs216172	SMG6, SRR	97.4	0.37 (C)
17q21.32	rs46522	UBE2Z, GIP, ATP5G1, SNF8	95.4	0.54 (T)
19p13.2	rs1122608	LDLR	92.1	0.75 (G)
21q21.11	rs9982601	MRPS6	98.1	0.14 (T)

^aFor each genotyped variant, the frequency of the risk allele for coronary artery disease, as identified by Schunkert et al., ² in the GRACE UK–Belgian discovery cohort is provided.

Statistical analysis

Data are summarized as frequencies and percentages for categorical variables. Continuous variables are presented as medians with 25th and 75th percentiles. All continuous variables were analysed in a continuous fashion. Genotypes were compared using analysis of variance and χ² tests. Hardy–Weinberg equilibrium of the genotypes was assessed using Pearson's χ² test.

Cox proportional hazard analysis was used to determine the association of 23 genetic variants with recurrent MI and recurrent MI or cardiac death, while considering clinical variables known to be associated with these outcomes as covariates.^7–9 In addition to age and gender, which we considered non-modifiable risk factors for adverse ACS outcome, we additionally adjusted for history of angina, prior MI, diabetes, current smoking status, systolic blood pressure, heart rate, total cholesterol, initial serum creatinine level, elevated initial cardiac enzymes, ST segment depression, Killip class, PCI, CABG, thrombolytics as well as participating country. For the majority of the covariates, less than 1% of the data was missing. Covariates with a considerable number of values missing were imputed using a multiple imputation technique based on the Markov chain Monte Carlo method (see Supplementary material online for details on % of missing values and imputation procedures).⁹ Consistency with non-imputed data was confirmed by repeating the analysis with non-imputed data. In a first step, Cox proportional hazard analysis was performed without assuming a specific model of inheritance (genotypic test). For the variants that significantly associated with both endpoints, we tested the association under a dominant, additive, and recessive risk model. Next, the Lambda statistic was calculated to confirm the most appropriate genetic model of inheritance.¹⁰ All above-mentioned statistical analyses were performed using SPSS software version 19.0 (SPSS Inc., IL, USA). The P-value threshold for significance was adjusted for 23 multiple comparisons using the Bonferroni correction method, resulting in a significance threshold of P < 0.002 (P < 0.05/23 variants tested; two-tailed). In a second step, to account for competing risk of death when predicting the probability of an event during the long-term follow-up, we performed regression analyses of cumulative incidence functions (CIFs) for the UK–Belgian discovery cohort (for methods described in detail, see Supplementary material online). Analysis of CIFs was performed using SAS software version 9.2 (SAS Institute Inc., NC, USA).

Comparison of the GRACE score for improvement in reclassification with and without genetic information was done by the integrated discrimination improvement (IDI) method, which compares the average difference in correctly predicting the risk for patients who have a recurrent MI with those who do not.¹¹ In addition, we also calculated the IDI index for a second risk model composed of the 17 clinical variables that were considered covariates in our endpoint analyses.^7,8 The method applied for the calculation of the IDI index at 6 months and at 5 years is described in detail in the Supplementary material online. Finally, the C statistic was calculated by use of the risksetROC function in the ‘risksetROC’ package in R.

Results

Overall, 3585 patients were enrolled in the GRACE Genetics Study. Patients with no diagnosis, a non-cardiac diagnosis, or with no ACS at the time of hospital discharge (16, 117, and 103, respectively) were excluded. Of the remaining 3349 participants, 734 and 1365 were prospectively recruited in Belgian and UK centres, respectively. For the resulting 2099 Belgian and UK patients with confirmed ACS, follow-up status 24 h after initial presentation and at a median of 5 years was available. Genotyping succeeded for 22 out of the 23 selected variants (Table 1). In particular, we omitted rs964184 from further analyses because of a low genotyping success rate. For the 22 variants, no deviation from Hardy–Weinberg equilibrium was seen (P > 0.05). The genotyping success rate per patient ranged between 99.5 and 100%.

The median duration of follow-up in the 2099 patients was 1518 days [interquartile range (IQR) 656–1870] following admission. After excluding patients who died, median follow-up was 1668 days [(IQR) 838–1973]. A recurrent MI occurred in 278 patients (13.2%) and a recurrent MI or cardiac death in 340 patients (16.2%). In patients suffering a second event, the median time to recurrence was 169 days [(IQR) 17–796] for recurrent MI and 206 days [(IQR) 21–864] for recurrent MI or cardiac death. There were 260 (93.5%) recurrent MIs in the period up to 5 years of follow-up. Consistent with previous studies, patients with recurrent MI were older, more likely to have prior history of vascular disease, exhibited a higher GRACE risk score for an endpoint predicting death or recurrent MI, and were more likely to have 3-vessel disease (Table 2).

Table 2

Baseline clinical variables predicting a recurrent myocardial infarction in the GRACE UK–Belgian discovery cohort and in the GRACE Polish validation cohort

	UK–Belgian cohort					Polish cohort
Variables	Total	No recurrent MI	Recurrent MI	HR (95% CI)	P-value	Total
No. of patients (%)	2099	1839 (87.6)	260 (12.4)			1250
Demographics
Men, n (%)	1510 (72.2)	1321 (72.1)	189 (73.0)	0.99 (0.76–1.29)	0.97	732 (59.3)
Age, median (25th–75th percentiles), years	66 (57–74)	65 (56–74)	70 (59–79)	1.03 (1.02–1.04)	<0.001	65 (55–73)
BMI, median (25th–75th percentiles), kg/m2	27 (24–30)	27 (24–30)	27 (24–29)	1.00 (0.99–1.00)	0.85	27 (24–30)
Medical history, n (%)
Angina	913 (43.5)	758 (41.3)	155 (59.6)	1.80 (1.42–2.29)	<0.001	844 (67.7)
Diabetes	330 (15.8)	272 (14.8)	58 (22.3)	1.57 (1.18–2.10)	0.002	215 (17.2)
Prior MI	543 (26.0)	437 (23.9)	106 (40.8)	1.97 (1.55–2.51)	<0.001	352 (28.2)
Congestive heart failure	151 (7.2)	119 (6.5)	32 (12.4)	2.05 (1.26–3.31)	0.004	102 (8.3)
Stroke or transient ischaemic attack	152 (7.2)	130 (7.1)	22 (8.5)	1.18 (0.77–1.80)	0.45	51 (4.1)
Prior PCI	364 (17.4)	313 (17.0)	51 (19.6)	1.14 (0.84–1.53)	0.39	94 (7.6)
Prior CABG	230 (11.0)	188 (10.2)	42 (16.1)	1.49 (1.07–2.06)	0.02	37 (3.0)
Peripheral arterial disease	169 (8.0)	127 (6.9)	42 (16.1)	2.57 (1.66–3.98)	<0.001	79 (6.5)
Hypertension	983 (46.9)	842 (45.9)	141 (54.2)	1.31 (1.04–1.66)	0.024	759 (62.1)
Hyperlipidaemia	854 (40.6)	747 (40.6)	107 (41.8)	1.05 (0.76–1.43)	0.77	315 (25.8)
Combined history of atherosclerosis	1346 (64.1)	1146 (62.3)	200 (77.1)	2.03 (1.41–2.93)	<0.001	790 (63.2)
Family history of coronary artery disease	542 (25.8)	484 (26.3)	58 (22.4)	0.81 (0.50–1.31)	0.39	326 (26.1)
Smoking Status, n (%)					0.045
Current	676 (32.8)	600 (33.3)	76 (29.5)	1.02 (0.75–1.40)		354 (28.9)
Former	675 (32.8)	572 (31.7)	103 (39.9)	1.37 (1.03–1.82)		218 (17.8)
Never	710 (34.4)	631 (35.0)	79 (30.6)	Reference		653 (53.3)
Presenting characteristics
Positive initial enzymes, n (%)	1191 (57.2)	1031 (56.5)	160 (61.8)	1.38 (1.09–1.77)	0.009	442 (38.4)
ST elevation or LBBB, n (%)	864 (41.2)	764 (41.5)	100 (38.5)	0.92 (0.72–1.17)	0.49	270 (21.6)
ST depression or T wave inversion, n (%)	750 (35.7)	631 (34.3)	119 (45.8)	1.62 (1.28–2.06)	<0.001	376 (30.1)
Systolic blood pressure, median (25th–75th percentiles), mmHg	135 (120–155)	136 (120–155)	135 (118–154)	1.00 (0.99–1.00)	0.50	140 (120–160)
Diastolic blood pressure, median (25th–75th percentiles), mmHg	80 (70–90)	80 (70–90)	80 (70–90)	1.00 (0.99–1.01)	0.66	80 (70–90)
Pulse (25th–75th percentiles), bpm	72 (62–85)	72 (61–84)	75 (65–95)	1.01 (1.00–1.02)	<0.001	80 (70–90)
Killip class, n (%)					<0.001
I	1587 (77.3)	1431 (79.6)	156 (60.7)	Reference		990 (80.8)
II	418 (20.4)	338 (18.8)	80 (31.1)	2.15 (1.66–2.79)		196 (16.0)
III	43 (2.1)	23 (1.3)	20 (7.8)	6.31 (3.96–10.0)		33 (2.7)
IV	6 (0.3)	5 (0.3)	1 (0.4)	1.81 (0.25–2.93)		6 (0.5)
Creatinine, median (25th–75th percentiles), mg/dL	1.08 (0.93–1.25)	1.07 (0.92–1.23)	1.16 (0.98–1.40)	1.23 (1.09–1.37)	<0.001	0.98 (0.80–1.16)
Serum cholesterol, median (25th–75th percentiles), mg/dL	191 (161–225)	192 (162–226)	180 (150–211)	0.99 (0.99–1.00)	<0.001	201 (170–232)
GRACE risk score
In-hospital death or MI, median (25th–75th percentiles)	167.4 (125.8–208.8)	170.0 (167.4–172.6)	197.5 (189.5–205.4)	1.01 (1.00–1.01)	<0.001	160.3 (129.4–203.5)
Diagnosis
STEMI, n (%)	847 (40.4)	743 (40.4)	104 (40.0)	Reference		244 (21.1)
Non-STEMI, n (%)	716 (34.1)	613 (33.3)	103 (39.6)	1.15 (0.89–1.50)		548 (47.4)
UA, n (%)	536 (25.5)	483 (26.3)	53 (20.4)	0.66 (0.48–0.91)		363 (31.4)
Treatment
Cardiac catheterization, n (%)	1380 (65.7)	1212 (65.9)	168 (64.5)	0.94 (0.60–1.30)	0.71	813 (65.0)
PCI, n (%)	1211 (57.8)	1108 (60.3)	104 (40.0)	0.51 (0.40–0.65)	<0.001	445 (36.4)
Thrombolytics, n (%)	282 (13.4)	236 (12.8)	46 (17.7)	1.39 (1.02–1.89)	0.037	117 (9.4)
CABG, n (%)	129 (6.2)	111 (6.1)	18 (6.9)	1.14 (0.72–1.82)	0.58	6 (0.5)
Cardiac catheterization
3-vessel disease (more than 50% stenosis), n (%)	511 (24.3)	404 (22.0)	107 (41.3)	2.35 (1.58–3.51)	0.001	304 (24.3)
Left main (more than 50% stenosis), n (%)	155 (7.3)	129 (7.0)	26 (10.1)	0.70 (0.35–1.28)	0.22	91 (7.3)

The combined GRACE UK–Belgian discovery cohort is stratified for the occurrence of a recurrent MI in the period up to 5 years of follow-up post-index ACS. Data were analysed by a Cox proportional hazard model. Combined history of atherosclerosis comprehends medical history of angina, prior MI, stroke or transient ischaemic attack, prior PCI, CABG, or peripheral arterial disease. A 50% stenosis cut-off was used to determine the presence or absence of CAD, instead of a 70% stenosis cut-off, which is used to denominate clinically significant stenosis.

HR, hazard ratio; CI, confidence interval; BMI, body mass index; MI, myocardial infarction; PCI, percutaneous coronary intervention; CABG, coronary artery bypass graft; LBBB, left bundle branch block; STEMI, ST-segment elevated myocardial infarction; UA, unstable angina. The GRACE risk score for in-hospital death or MI is based on age, pulse rate, systolic blood pressure, creatinine, Killip class, cardiac arrest on admission, positive initial cardiac enzymes, and ST-deviation.

Association of the ABO locus with recurrent myocardial infarction or cardiac death

We then assessed whether any of the 22 variants were associated with recurrent events (see Supplementary material online, Table S1). Survival analysis in the combined UK–Belgian discovery cohort (n = 2099) revealed that rs579459 near the ABO gene was significantly associated with recurrent MI (n = 278) during a median follow-up of 5 years after the primary event (P = 0.004, Figure 1 for survival curves). The risk effect followed a recessive genetic model, as indicated by a lambda coefficient of 0. Presence of two at-risk C alleles predicted recurrent MI with an age- and gender-adjusted hazard ratio (HR) of 1.64 (CI = 1.00–2.68; P = 0.048) and a multivariable-adjusted HR of 2.25 (CI = 1.37–3.71; P = 0.001; Table 3). Repeating the analysis without imputation revealed comparable results [multivariable-adjusted (HR) 2.19, CI = 1.26–3.80; P = 0.005]. Regression analysis of the CIF values to account for competing risk of death revealed a multivariable-adjusted HR of 2.26 (CI = 1.29–3.97; P = 0.005) for recurrent MI in CC carriers. None of the other 21 loci was significantly associated with recurrent MI after Bonferroni correction. One SNP, i.e. rs12526453, initially exhibited a significant association but failed to survive correction for Bonferroni (P = 0.041).

Table 3

Association of rs579459 with recurrent MI and with recurrent MI or cardiac death during a median of 5 years post-index ACS in the GRACE UK–Belgian discovery cohort

rs579459	CC	CT	TT	Recessive modela		Additive modela		Dominant modela
Total, n (%)	89 (4.2)	712 (33.9)	1291 (61.5)	HR (95% CI)	P-value	HR (95% CI)	P-value	HR (95% CI)	P-value
Recurrent MI, n (%)	17 (6.1)	95 (34.3)	165 (59.6)	2.25 (1.37–3.71)	0.001	1.32 (1.06–1.63)	0.011	1.25 (0.97–1.60)	0.086
No recurrent MI, n (%)	72 (4.0)	617 (34.0)	1126 (62.0)
Recurrent MI or cardiac death, n (%)	17 (5.0)	114 (33.7)	207 (61.2)	1.80 (1.09–2.95)	0.021	1.19 (0.98–1.45)	0.081	1.14 (0.91–1.44)	0.25
No recurrent MI or cardiac death, n (%)	72 (4.1)	598 (34.1)	1084 (61.8)

Genotyping for rs579459 succeeded in 2092 subjects (99.7%). Data were analysed by a Cox proportional hazard model.

HR, hazard ratio; CI, confidence interval. Percentages are row percentages.

^aAdjustments for: age, gender, history of angina, prior myocardial infarction, diabetes, current smoking status, systolic blood pressure, heart rate, total cholesterol, initial serum creatinine level, elevated initial cardiac enzymes, ST segment depression, Killip class, percutaneous coronary intervention (PCI), coronary artery bypass graft (CABG), thrombolytics, and participating country.

Figure 1

Cumulative hazard curves for recurrent MI by rs579459 genotype. Cumulative hazard curves, with multivariable adjustment and by rs579459 genotype, for recurrent MI in the GRACE UK–Belgian discovery cohort. MI, myocardial infarction.

No differences in traditional CAD risk factors or in the GRACE risk score for death or MI were identified for rs579459 genotypes (see Supplementary material online, Table S2). There was no significant interaction between rs579459 and any of the covariates used to adjust our primary endpoint analysis (see Supplementary material online, Table S3). In particular, stratification of risk effects into UA and STEMI/non-STEMI subgroups revealed a HR = 2.73 (CI = 0.94–7.94; P = 0.065) and HR = 2.17 (CI = 1.21–3.87; P = 0.009), respectively. Overall, HRs in these subgroups were similar, suggesting that recurrent events in the full spectrum of ACS, rather than a particular subtype, were associated with rs579459.

Since cardiac death is often precipitated by ischaemic heart disease, we used a second combined endpoint consisting of recurrent MI or cardiac death. We did not investigate cardiac death as a single endpoint, since it only occurred in 62 patients during follow-up in the GRACE UK–Belgian cohort. In agreement with the primary endpoint analysis, rs579459 was significantly associated with recurrent MI or cardiac death (n = 340) within 5 years of the primary event. Cox regression analysis revealed a multivariable-adjusted HR of 1.80 (CI = 1.09–2.95; P = 0.021; Table 3) for recurrent MI or cardiac death in CC carriers (recessive model). Consistency with non-imputed data was confirmed (multivariable-adjusted: HR 1.79, CI = 1.04–3.09; P = 0.037). None of the other 21 SNPs was significant for the second combined endpoint, with the effect of the rs12526453 variant no longer reaching significance, also not at the uncorrected level (P = 0.086).

Replication of the association of the ABO locus with recurrent myocardial infarction

In an effort to replicate the association of the ABO locus with recurrent MI, we used 1250 independently recruited Polish patients with confirmed diagnosis of ACS. The median follow-up for these patients was limited to 6 months [191 days, (IQR) 185–208], with a recurrent MI occurring in 65 patients (5.2%) and a recurrent MI or cardiac death in 79 patients (6.3%). In patients suffering a second event, the median time to recurrence was 25 days (IQR 4–79) for recurrent MI and 26 days (IQR 4–97) for recurrent MI or cardiac death. In Table 2, demographic characteristics, medical history, and presenting clinical features are shown for the Polish validation cohort. Remarkably, Cox regression analysis, assuming a recessive risk model and correcting for exactly the same covariates as in the UK–Belgian discovery cohort, established that rs579459 was significantly associated with recurrent MI. As in the 5-year follow-up cohort, homozygous carriers of the rs579459 variant in the ABO gene were significantly associated with recurrent MI after 6 months (HR 2.70, CI = 1.26–5.82; P = 0.011; Supplementary material online, Table S4). Similar effects were seen for the second combined endpoint consisting of recurrent MI or cardiac death (HR 2.51, CI = 1.17–5.37; P = 0.018).

The ABO locus and clinical prediction models for recurrent myocardial infarction or cardiac death

We also assessed whether rs579459, when added to the GRACE risk score or an established set of clinical risk factors, predicts the risk of a recurrent MI or cardiovascular death using the IDI index.¹¹ Inclusion of rs579459 into the GRACE risk score failed to significantly improve classification of recurrent MI, or recurrent MI or cardiac death, at 5 years in the combined UK–Belgian discovery cohort (P = 0.15 and P = 0.54, respectively).⁷ Next, since the majority of recurrent events in the discovery cohort occurred in the first 6 months, we pooled the UK–Belgian cohort with the Polish cohort and calculated the C statistic and the IDI index at 6 months in the combined study population. After incorporation of rs579459 in the GRACE risk model, the C statistic improved from 0.637 (CI = 0.597–0.677) to 0.644 (CI = 0.604–0.683; P = 0.32) for the prediction of recurrent MI and from 0.645 (CI = 0.607–0.683) to 0.654 (CI = 0.616–0.691; P = 0.20) for the prediction of recurrent MI or cardiac death. The relative IDI improvement was 10.7% (IDI 0.0015, P = 0.15) and 10.5% (IDI 0.0021, P = 0.097), respectively, for recurrent MI and for recurrent MI or cardiac death. After incorporation of rs579459 in our multivariable prediction model (see Methods), the C statistic improved from 0.745 (CI = 0.712–0.778) to 0.751 (CI = 0.718–0.784; P = 0.15) for the prediction of recurrent MI and from 0.750 (CI = 0.719–0.780) to 0.756 (CI = 0.726–0.787; P = 0.096) for recurrent MI or cardiac death. The relative IDI improvement was 5.8% (IDI 0.0038, P = 0.059) and 6.0% (IDI 0.0048, P = 0.037), respectively, for recurrent MI and for recurrent MI or cardiac death.

Discussion

The GRACE Genetics Study was used to investigate whether 23 established susceptibility loci for CAD predispose to recurrent MI or cardiac death following an ACS. We could establish rs579459 in the ABO locus on chromosome 9q34.2 as a novel risk factor for adverse cardiac outcome after an index ACS. In ACS patients receiving optimal secondary prevention treatments, we found that homozygous carriers of the at-risk C allele exhibited an increased risk of developing a recurrent MI or cardiac death within 5 years after the primary event. We validated the risk effect of rs579459 in an independent cohort, indicating that the association between rs579459 and recurrent ischaemic events does not represent a spurious finding. The fact that we did not identify such association for the other variants further suggests that CAD and recurrent MI or cardiac death after an index ACS constitute different phenotypes that may be influenced by distinct genetic variants. This hypothesis is also supported by the recent GWA study from Reilly et al.,¹² in which variants in the ABO locus were identified as risk factors for MI in patients with angiographic CAD, but not for angiographic CAD itself. Remarkably, although most published loci for MI had significant signals for angiographic CAD compared with controls, none was associated with MI in patients with angiographic CAD.¹²

The association between rs579459 and recurrent events was independent of conventional CAD risk factors. In particular, we did not observe an association between rs579459 and cholesterol, a link that has previously been established in a population-based cohort of healthy individuals.^2,13 Since statin use lowers the risk of death or non-fatal MI during follow-up of ACS patients with hypercholesterolaemia and could therefore act as a confounding factor for the association between rs579459 and recurrent MI, we additionally adjusted for statin use.¹⁴ The association between rs579459 and recurrent MI was, however, not influenced by statins. The fact that the association of rs579459 with ACS outcome was independent of all other risk factors suggests that it may be used in clinical risk prediction models for recurrent events. Inclusion of rs579459 in the GRACE risk score did not show a significant improvement, but inclusion of rs579459 in an extended multivariable prediction model did significantly improve risk classification for recurrent MI or cardiac death at 6 months (P = 0.037). It should be acknowledged, however, that the observed improvement was rather limited, presumably because the subgroup of homozygous carriers of the at-risk C allele is rather small, and became only significant when applying the IDI method for the recurrent MI or cardiac death endpoint. Nevertheless, these findings suggest that rs579459, together with other risk variants yet to be identified, should be considered potentially relevant contributors in risk prediction models for adverse outcome following ACS.

The association between rs579459 and recurrent MI or cardiac death is notable since this variant is located about 3.5 kb upstream of the ABO gene. Previous studies also established a strong correlation between the rs579459 CC genotype and blood group A (e.g. >90% of homozygous carriers of the at-risk C-allele are blood group A carriers¹⁵) and showed that non-O blood group carriers, mainly those with blood group A, have a higher risk of ischaemic heart disease than subjects with blood group O.¹⁶ The link between the rs579459 at-risk C allele and blood group A is also indicated by a pairwise r² of 0.95 between rs579459 and the intronic ABO variant rs507666 that tags the A1 allele.^12,17 On the other hand, linkage disequilibrium with variants tagging the O and B alleles is low (pairwise r² = 0.37 and r² = 0.025, respectively, for rs612169 and rs8176672).^12,18 Future research is now warranted to assess whether rs579459 associates with recurrent MI through its link with the ABO blood group. The blood group itself has been proposed to influence the risk of coronary events at least in part by increasing circulating von Willebrand factor (VWF) levels, which play an important role in thrombosis.¹⁹ Higher VWF plasma levels have indeed repeatedly been found in non-O carriers in comparison with blood group O carriers.²⁰ On the other hand, it is also possible that rs579459 affects the risk of recurrent ischaemic events via other biological processes, especially since the ABO locus appears to have pleiotropic effects and might modulate various CAD-related pathways. For instance, variants in this locus have been associated with type-2 diabetes, venous thrombosis, coagulation factors (vWF and factor VIII), inflammatory risk biomarkers (E-selectin, P-selectin, and ICAM-1), LDL-cholesterol, and angiotensin-converting enzyme (as summarized in refs^2,12). Consequently, additional research is needed to understand the mechanisms underlying the association between rs579459 and recurrent MI and to assess the potential confounding effect of the ABO blood group.

The strength of the GRACE Genetics Study is its specific design to capture an unselected and representative population of ACS patients with carefully characterized baseline and prospective follow-up data. The prolonged and close follow-up is important since it allowed us to assess the influence of CAD risk variants on long-term outcome of ACS. In addition, the inclusion of three independently recruited and geographically distinct populations suggests that our findings are not specific to a particular population or to a specific collection of risk factors in one country. However, further external replication is needed to confirm our findings in the general ACS population and to assess interactions with covariates. In particular, larger patient cohorts will be needed to subset primary ACS events according to documented MI and confirm that findings are consistent with UA (where index and recurrent events have different pathophysiology) or non-STEMI/STEMI (where pathophysiology is similar) as the index event. An additional limitation of the study is that some covariates exhibited missing values. We applied, however, a widely accepted multiple imputation technique to adjust for this.⁹ In addition, results with and without imputation were of virtually identical significance. Next, since we do not have genome-wide SNP data for GRACE, we could not assess population substructure by multi-dimensional scaling or principal component analysis. Finally, we observed a recessive risk effect for recurrent MI or cardiac death, which is different from the additive effects observed in CAD patients.² Possibly, this difference is due to the fact that recurrent MI is compared with a primary ACS population in GRACE, which according to Schunkert et al.² is already enriched for the at-risk C-allele. It is therefore possible that rs579459 can only exert an additional risk effect in a recurrent MI population when two at-risk C-alleles are present.

In conclusion, we identified rs579459 in the ABO locus as a novel and independent risk factor for recurrent MI or cardiac death in patients with a history of ACS. Further studies are warranted to validate our findings, to determine the mechanisms underlying the association with recurrent MI, and to assess the potential contribution of rs579459, in combination with other variants, to clinical risk prediction models.

Supplementary material

Supplementary material is available at European Heart Journal online.

Funding

GRACE is supported by an unrestricted educational grant from Sanofi-Aventis to the Center for Outcomes Research, University of Massachusetts Medical School. Phenotypic characterization and DNA extraction were funded by an award from the British Heart Foundation (CH/92010) to K.A.A.F. Genotyping was funded by a ‘Krediet aan Navorsers’ grant (1516009N) awarded to D.L. by the Fund for Scientific Research Flanders (FWO-F). D.R.D. was funded by a British Heart Foundation Centre of Research Excellence Award. E.W. is supported by FWO-F.

Conflict of interest: none declared.

Acknowledgements

We thank the patients, physicians, and nurses participating in the GRACE study.

Author notes

†

These authors contributed equally to this work.