Family or SNPs: what counts for hereditary risk of coronary artery disease?

This editorial refers to ‘Risk prediction by genetic risk scores for coronary heart disease is independent of self-reported family history’†, by H. Tada et al., on page 561. Twin studies document that the individual genetic constitution contributes to the risk of almost all human traits.1 With respect to coronary artery disease (CAD), i.e. globally the predominant lethal disease, such twin studies revealed that about half of the overall risk to suffer from a myocardial infarction is inherited.2 The foremost challenges arising from these facts are two-fold: first, is it possible to improve individual risk estimates by adding genetic measures to prediction tools and, secondly, is it possible to neutralize some of this genetic risk? The initial detection of genetic factors contributing to coronary risk dates back to the finding that a positive family history is an independent predictor for a first- and even a second-degree relative of an individual who had suffered from myocardial infarction at a young age.3–5 This conclusion is further strengthened by the study of Tada et al. reported in this issue of the journal.6 The focus of the present study was to compare the clinical values of family history and common risk alleles, which had been identified in recent genome-wide association studies (GWAS) for CAD,7–10 in predicting coronary events.6 In addition to family history, the authors studied two genetic risk scores based on previous GWAS findings for CAD with respect to prediction of coronary events in the Malmö Diet and Cancer Study, a large cohort prospectively followed for 15 years.6 The data support three relevant conclusions: (i) genetic risk scores strongly associate with coronary risk; (ii) genetic risk scores discriminate coronary risk particularly well in middle-aged subjects; and (iii) risk prediction offered by a genetic risk score is independent from the information conferred by a positive family history. The first and second conclusions drawn from the study by Tada et al. are in line with previous studies,11–14 which accompanied the increasing number of single nucleotide polymorphisms (SNPs) found to affect CAD risk with genome-wide significance (i.e. achieved a P <5 × 10–8).7–10 It is worth mentioning that genetic risk scores based on GWAS findings may be associated not only with coronary events but also with sudden cardiac death, which makes any predictive value even more relevant.14 The most recent analysis of GWAS increased the number of genome-wide significant SNPs for CAD to ∼58;10 27 and 50 of these were included in genetic risk scores evaluated by Tada et al.6 Somewhat disappointingly—but in line with previous studies11–15—risk prediction was only marginally improved by adding the genetic risk scores to a model that included age, gender, and other traditional risk factors including family history. Indeed, the c-statistics did not change much and the percentage of individuals (17%) correctly re-classified based on the genetic risk score was only modest. While this improvement was in addition to 20% of individuals reclassified based on the family history alone and thus affected overall a large number of individuals, it was sobering to realize that categorical risk classification above and below 7.5% of 10-year risk, i.e. a cut-off for implementation of therapeutic measures, was not substantially improved by the genetic risk score.15 Future analysis need to evaluate whether the restriction to genome-wide significant SNPs—as done by Tada et al.—is meaningful or whether inclusion of 1000, 5000, 10 000, or even 100 000 top SNPs in a genetic risk score more profoundly enhances prediction. It is interesting to read that the genetic risk score offered better risk prediction in younger individuals. In fact, comparing the highest with the lowest quintile of the genetic risk score, Tada et al. observed a 2.4-fold difference in coronary events during 15 years of follow-up in subjects younger than 57 years of age (P = 7.5 × 10–11).6 Such an enormous difference goes beyond the discriminatory effects of most traditional risk factors such as hypertension or diabetes. Similar data have been reported for premature cases of incident and prevalent CAD by Hughes et al. and the CARDIoGRAM Consortium.8,12 Given that genetic risk scores have to deliver better discrimination of risk before such tools can be considered for routine clinical decision-making, this aspect should be evaluated in more depth by future studies. It can be anticipated that studying larger panels of SNPs in middle-aged subjects at intermediate risk may finally sharpen decision-making in a clinically relevant fashion. In contrast, it may be futile to employ genetic risk scores in elderly subjects, in whom age—or rather the integral effect of all risk factors over time—is largely responsible for the coronary risk. The most striking finding from the study by Tada et al. is the fact that a positive family history and a high genetic risk score confer largely independent information regarding prediction of future coronary events (Figure 1). While similar observations have been reported for the prominent risk locus on chromosome 9p21,16 no previous analysis examined the topic in such a comprehensive fashion. The obvious question arising from these data is: how can it be explained that the predictive values of family history and common risk alleles, albeit that both were significantly associated with each other,6 are largely non-redundant? Hereditary risk of coronary artery disease. The figure contrasts the overlapping and different qualities of information represented by a positive family history and by a high score for common risk alleles. While a self-reported family history represents multiple measurable and non-measurable, genetic and non-genetic risk factors to which a family is exposed, it is limited by the lack of precision in reporting definitive coronary artery disease in family members as well as by dilution of effects from one generation to the next. On the other hand, common risk alleles can be determined accurately in an individual—but reflect at present only a small spectrum of inherited susceptibility. Positive family history (FH) and common risk alleles add together in predicting coronary events, particularly in young individuals.6 In the Malmö Diet and Cancer Study, the cumulative incidence of events after 15 years of follow-up in individuals with a positive self-reported FH was 1.34-fold greater than in those without (P = 6.70 × 10–16), and 1.78-fold greater among those in the highest vs. lowest quintile of GRS50 (P = 1.60 × 10–21), and those positive for both risk factors vs. those negative for both had a 2.51-fold greater incidence of events (P = 6.60 × 10–28).6 The Manhattan plot (upper right) represents P-values (y-axis) for millions of single nucleotide polymorphisms (SNPs) across all chromosomes (x-axis).8 On a critical note, one may argue that the analysis of the Malmö Diet and Cancer Study is limited by the fact that family history was self-reported (rather than validated by medical records) and without age restriction (rather than limited to an age at the time of myocardial infarction of <55 years in male and <65 years in female relatives17).6 The predictive value of such self-reported family history may be less precise, and, indeed, its hazard ratio for CAD in the Malmö Diet and Cancer Study was only 1.43 rather than 1.7–2.7, numbers that have been reported, for example, by the Framingham Heart Study or the Newcastle Family History Study.6,17,18 Without validation, some seemingly affected (or non-affected) parents or siblings will be falsely categorized, thereby impairing the accuracy of reported family history for predicting future events.18,19 Perhaps more importantly, the Malmö Diet and Cancer Study did not restrict—as a defining criterion for a ‘positive family history’—manifestations of coronary disease to those affected at a relatively young age.3,17,18,20 Indeed, the younger the age in an affected family member the stronger is condensation of genetic variants and therefore the discrimination of coronary events in first-degree relatives.18 Nevertheless, it is clear from the work of Tada et al. that a positive family history and common risk alleles reflect partially different aspects of genetic information.6 Indeed, the two measures of genetic risk differ by multiple features and thus confer different information for predicting disease burden (Figure 1). Two obvious examples may illustrate this fact: the genetic risk scores analysed by Tada et al. do not map monogenic causes for CAD such as mutations causing familial hypercholesterolaemia or dysfunctional nitric oxide (NO) signalling.20–22 a clear strength for taking family history. Vice versa, irrespective of whether or not a family member inherited a causal variant, all are labelled equally by a positive family history. In this regard genotyping of variants by an array may have clear advantages. In addition there are multiple other subtler differences between the two (Figure 1). Aspects that are much more strongly reflected by familial background, i.e. family history, include shared environmental factors (e.g. pollution with fine particles, noise, etc.), dietary habits, or socio-economic background, but also complex genetic features such runs of homozygosity.23 Moreover, sharing of the mircobiome and epigenome, i.e. potential risk factors for CAD, or specific coronary manifestations of coronary disease may be evident in siblings and thus covered by taking the family history.24 Finally, family members of CAD patients are easily identified in order to put preventive strategies in place.25 On the other hand, counting of common risk alleles in a genetic risk score very precisely characterizes a given person. Indeed, the study by Tada et al. suggests that this information, albeit being arbitrarily restricted to 50 SNPs that had obtained genome-wide significance in large-scale association studies at the time the analysis was carried out,7–10 is even better in predicting future coronary events than a positive family history. Interestingly, the finding that common risk alleles and family history reflect different genetic aspects of the trait CAD6 is somewhat in contrast to data regarding LDL-cholesterol SNPs. In this related field, a phenocopy of ‘familial hypercholesterolaemia’, a monogenic disease with autosomal inheritance that is even defined by its positive family history, is often found as a consequence of aggregation of multiple SNPs with small effect sizes.26 What are the clinical implications of the fact that a positive family history and a high genetic risk score are additive in predicting coronary risk (Figure 1)? In the long run, with addition of more common and all known rare variants on specific arrays, direct testing of our genetic variability may be more informative than taking family history. However, if we want to make use of all hidden familial and molecular heritage, we need to cover both for predicting coronary risk. I wish to thank the authors of the manuscript under discussion for providing the data and the Kaplan–Meier curves for Figure 1. Conflict of interest: none declared.