Introduction

Low-grade inflammation, oxidation, and vascular remodeling are cardinal components in the pathophysiology of atherosclerosis.¹ Heme oxygenase-1 (HO-1) is the inducible, rate-limiting enzyme of heme degradation and exerts potent anti-inflammatory, antioxidative, and antiapoptotic effects in response to various stressors.^2,3 Compelling evidence for a protective effect of HO-1 on the vasculature derives from animal studies, demonstrating that HO-1 suppresses the development of atherosclerotic lesions^4–6 and thrombi.⁷ Moreover, prominent endothelial damage was observed in rare human HO-1 deficiency,⁸ as well as in HO-1 knockout mice.⁹

There is a (GT)_n dinucleotide repeat polymorphism (variable number tandem repeat, VNTR) in the HO-1 gene promoter region, and higher repeat numbers translate into lower enzyme expression.^10–13 A deficiency in HO-1–mediated vascular protection in subjects with greater repeat lengths was proposed to predispose to atherosclerosis and its clinical sequelae myocardial infarction (MI) and stroke.¹⁴ Studies examining the association between (GT)_n repeat length and cardiovascular disease (CVD) have to date been restricted to selected patient series, mainly subjects admitted for coronary angiography or patients with coronary artery disease (CAD) or peripheral arterial disease, and yielded inconsistent results. A summary of the literature is presented in Table 1. Apart from differences in study design, patient characteristics, and end point definitions, heterogeneous results may arise from the different cut-offs applied to categorize repeat number.

We present here the first prospective study on the potential relationship of the HO-1 (GT)_n polymorphism with CVD conducted in the general community.

Materials and Methods

Materials and Methods are available in the online-only Data Supplement.

Results

HO-1 genotyping resulted in unambiguous results for 812 of 816 subjects for which DNA samples were available (call rate, 99.5%). Duplicate measurement of 95 random DNA samples yielded 100% concordant findings. The distribution of (GT)_n repeat lengths ranged from 12 to 44 repeats and was trimodal, with peaks at 23, 30, and 37 repeats, constituting 20.5%, 40.9%, and 3.9% of alleles (Figure 1). The most common allele combinations were 30/30 and 23/30, observed in n=145 (17.9%) individuals each.

• Download figure
• Download PowerPoint

Categorization of study subjects by VNTR length (S, <23; M, 23–31; L, ≥32) resulted in only 2 subjects homozygous for short alleles (SS genotype), and we therefore merged SS and SM genotype groups to form SS/SM (n=35), MM (n=665), ML (n=101), and LL (n=11) groups. Distributions of baseline characteristics according to these 4 groups are shown in Table 2. Levels of standard risk factors emerged as independent of HO-1 genotype.

Crude incidence rates (95% confidence intervals [CIs]) for CVD were 6.5 (0.0, 15.3), 13.2 (10.8, 15.8), 13.0 (7.1, 19.8), and 65.1 (24.1, 130.4) events per 1000 person-years for SS/SM, MM, ML, and LL groups, respectively. Accordingly, 55% of subjects in the LL group developed hard CVD end points (stroke, MI, or vascular death) in the 15-year follow-up period. End point–specific event counts during the survey period in LL subjects and in other subjects were 4 and 61 for stroke, 2 and 51 for MI, and 0 and 20 for vascular death not caused by stroke or MI.

Under adjustment for age and sex, subjects homozygous for the longest repeat lengths (LL) faced a substantially elevated risk for CVD compared with MM subjects (hazard ratio (HR) [95% CI], 5.46 [2.39, 12.50]; P<0.0001; Table 3). A recessive model best fitted the data and revealed a HR [95% CI] of 5.45 [2.39, 12.42] (P<0.0001) in a comparison of LL to the rest of the study population. Effects remained virtually unchanged under further multivariable adjustment, were similar when excluding 50 subjects with prior CVD (HR [95% CI], 4.44 [1.63, 12.10]; P=0.0036), and were highly significant for the extended CVD end point as well (P<0.0001). Analyses of individual disease end points yielded a HR [95% CI] of 7.87 [2.84, 21.86] (P<0.0001) for stroke and 2.18 [0.52, 8.96] (P=0.282) for MI.

In sensitivity analyses, we used penalized cubic splines to examine the precise scale of relationship between VNTR length of each allele and CVD irrespective of predefined cut-offs. This gave significant results for the shorter allele (P=0.0073) and provided a post hoc confirmation of our a priorily fixed cut-off of 32 (Figure 2). When applying alternative and mostly lower cut-offs previously used in the literature (Table 1), findings were not significant, underscoring that high risk was confined to subjects homozygous for the longest HO-1 VNTRs.

• Download figure
• Download PowerPoint

Finally, subjects in the LL group tended to experience atherosclerosis progression (incidence of new plaques or growth of existing ones) more frequently (82% versus 46%, odds ratio [95% CI], 4.72 [0.91, 36.68]; P=0.089) and showed a significantly larger change in the atherosclerosis score over 5 years (median difference in atherosclerosis score [interquartile range], 2.1 [0.8, 5.6] versus 0.0 [0.0, 2.2] mm; P=0.001), suggesting that the enhanced burden of CVD is at least in part mediated by accelerated atherogenesis. Subjects in the LL group also showed a trend toward elevated baseline levels of oxidized phospholipids (OxPL) on apolipoprotein B (apoB)-100 (median OxPL/apoB levels [interquartile range], 11364 [4160, 18330] versus 4844 [3174, 12284] relative light units; P=0.055). Results were similar when the Δatherosclerosis score and OxPL/apoB were log-transformed (P=0.014 and P=0.073, respectively). Differences between subjects in the LL group and the rest of the sample with regards to incident CVD, Δatherosclerosis score, and OxPL/apoB are summarized in Figure 3.

• Download figure
• Download PowerPoint

To corroborate our main result, we gathered data from 3 additional prospective cohorts: The Cooperative Health Research in the Region of Augsburg (KORA) F3 and F4 studies, and the Salzburg Atherosclerosis Prevention program in subjects at High Individual Risk (SAPHIR) study. As is visible in Table 4, these cohorts differed in most baseline characteristics. In particular, the additional 3 cohorts had substantially lower prevalences of the LL genotype and also substantially lower CVD incidence rates (P=0.011 for heterogeneity after adjustment for age and sex). As a consequence, we were unable to perform a strict independent replication of our key result. However, when pooling data from all 4 studies, the subjects in the LL group versus other subjects remained at strongly and significantly elevated risk for CVD (HR [95% CI], 3.26 [1.50, 7.33]; P=0.004; 326 events in 7848 subjects). Moreover, when pooling data from the Bruneck and the SAPHIR study, for which data on an extended end point additionally including revascularization procedures and peripheral vascular disease were available, the LL group was also strongly associated with this end point (HR [95% CI], 3.98 [1.76, 9.03]; P<0.001; 275 events in 2524 subjects). Both of these associations remained similar and significant under extended multivariable adjustment.

Discussion

In a prospective cohort study, we observed a substantially increased risk of CVD (hazard ratio [95% confidence interval], 5.45 (2.39, 12.42); P<0.0001) in subjects homozygous for long HO-1 VNTRs, indicating a recessive gene effect. This recessive nature of association is in line with experimental data, suggesting the shorter allele to be decisive for HO-1 upregulation in human umbilical vein endothelial cells.¹⁰ Excess risk in our study was restricted to a small segment of the population (LL genotype, 1.4%).

This is the first prospective study on the relationship of the HO-1 VNTR with CVD conducted in the general population. To the best of our knowledge, the previous studies were conducted in high-risk populations, such as patients with preexisting CVD, coronary stenting, or hemodialysis (Table 1). One Chinese study was population-based but cross-sectional in design.¹³ Many of the previous reports on this matter used lower VNTR cut-offs, most commonly 25 to 27. Of these, 3 large studies,^25–26 including 1800 to 3000 patients, found no relationship between HO-1 VNTR repeat length and their primary end points restenosis^25,26 or CAD,²⁸ but a large number of smaller studies did. Putting these data in perspective with our study, it should be considered that HO-1 induction occurs in response to stress conditions,^10,11,33 and a more severe deficit in HO-1 might be necessary in the general (low-risk) population to evoke deleterious effects, whereas a less severe deficit could suffice in higher-risk patients. This interpretation is consistent with several reports that found an association between HO-1 VNTR length and vascular end points only in high-risk sub groups, such as diabetic subjects or smokers.^11,13,16,29

The dependency of HO-1 protein expression on HO-1 VNTR length has to date been investigated primarily in cell lines. It was found that baseline as well as oxidative stress-induced HO-1 protein levels decreased approximately monotonically parallel to increasing length of the shorter HO-1 allele.¹⁰ This extends earlier findings of reduced HO-1 transcriptional activity with increasing VNTR length.^11,12 One study found lower increase of HO-1 protein in response to oxidative stress but higher HO-1 baseline expression in cells with long alleles,³⁴ whereas another found higher HO-1 protein expression associated with short alleles only under conditions of oxidative stress.¹³ There is to date no direct study of this dependency in humans. However, it has been reported that diabetic subjects homozygous for long alleles had increased CAD risk, reduced bilirubin levels, and increased serum ferritin levels and that the association with CAD risk disappeared with multivariable adjustment for bilirubin and ferritin.²⁷ These findings are consistent with reduced HO-1 activity in subjects with long alleles and also with reduced HO-1 activity potentially mediating the effect on CAD risk.

Several lines of evidence suggest that the key finding of our study is valid: (1) the association between HO-1 VNTR and CVD was of particular strength (HR, 5.45; lower confidence bound, 2.39) and highly significant (P=5.51×10^–5). It would even retain significance in an exploratory setting, testing for all previously used VNTR cut-off values and adjusting for these multiple comparisons (Bonferroni corrected P=4.95×10^–4). (2) The elevated CVD risk observed in the LL HO-1 group was robust in several sensitivity analyses (Table 3). (3) The LL group was at elevated CVD risk also in a pooled analysis of 7848 subjects. (4) Vascular protection conferred by HO-1^2,3 is impressively demonstrated by the prominent vascular damage observed in human HO-1 deficiency.⁸ (5) The deficit in HO-1 upregulation in response to cell stress with higher HO-1VNTR number rests on solid experimental evidence.^10–13(6) Subjects with the LL HO-1 genotype in our study had higher levels of OxPL/apoB (P=0.055), which is consistent with decreased HO-1 activity. (7) Finally, we observed a high risk of atherosclerosis progression in the LL HO-1 group, providing a pathophysiological explanation for the elevated CVD risk.

Strengths of our study include its prospective design with long-term high-quality follow-up and representativeness for the general population. Among its weaknesses is the limited number of subjects in extreme repeat length groups, a weakness that extends to the additional population-based cohorts that we used, which precluded subgroup analyses.

In conclusion, subjects with ≥32 tandem repeats on both HO-1 alleles represent a hitherto neglected vascular high-risk group featured by a substantial burden of CVD, amplified progression of atherosclerosis, and impaired antioxidant defense.

Footnotes