Will ADVANCE population genomic determinants improve upon biomarkers in predicting vascular complications of diabetes?

P. Hamet and J. Tremblay,Canada

Pavel HAMET, MD, PhD
Johanne TREMBLAY, PhDGene Medicine Service
Centre de recherche du Centre hospitalier de l’Université de Montréal (CRCHUM) and
Prognomix Inc, Montreal
Quebec, CANADA

Recent progress in genomics that allows the identification of millions of variations in an individual’s genome has given rise to new hope over the implementation of this knowledge in personalized medicine. The genetic architecture of complex diseases, including type 2 diabetes (T2D), is being uncovered by whole genome association studies. Despite tremendous progress in the last 3 years, their major limitation lies in their relatively low capacity to predict individual susceptibility in a general population. Fine phenotyping and careful consideration of various factors, including age, sex, as well as genetic and environmental backgrounds of the individual, are required to resolve this diagnostic challenge. Here, we summarize our initial strategies for dissecting the genetic determinants of renal and cardiovascular complications in T2D, using rich data from the ADVANCE (Action in Diabetes and Vascular disease: PreterAx and DiamicroN MR Controlled Evaluation) trial. We discuss the relative importance of traditional clinical biomarkers, such as cholesterol levels, hypertension, and body mass index, in the context of novel “genomic biomarkers,” as well as the need for their eventual integration into a more inclusive paradigm. Additional information contained in our DNA that is susceptible to modulation by environmental factors (such as disease state, medication, and lifestyle) includes epigenetic DNA methylation and telomeric shortening, a scar of biological aging. These can be added to DNA sequence variations at the single nucleotide polymorphism level, and this will accelerate the path toward personalized and predictive medicine where presymptomatic intervention becomes part of prevention.

Medicographia. 2009;31:307-313 (see French abstract on page 313)

Type 2 diabetes (T2D), a complex disease that is increasing in prevalence worldwide, represents a major global health burden1-4 in countries with high as well as low incomes. Diabetes is a major risk factor for the development of serious health problems, such as kidney failure, heart attack, and stroke.5,6 Currently, there are no tests available to promptly, accurately, and specifically determine which patients are more likely to encounter these complications (although several risk scores have been proposed for predicting events based on population studies, such as Framingham, or on data derived from clinical trials, such as the United Kingdom Prospective Diabetes Study [UKPDS]).7 A recent New Zealand investigation that compared 3 different methods of assessing cardiovascular disease risk and improved upon the UKPDS risk engine still only presents relatively low specificity.8 In fact, some recommendations are predictive in nature. For example, in the United States, the National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults proposes lowering the therapeutic threshold of the high-density lipoprotein (HDL)/low-density lipoprotein (LDL) ratio to _2.6 mmol/L in T2D, yet this is still only seen as secondary prevention.9

There is growing evidence that combining multiple genetic and clinical markers is the best way to develop a molecular test with clinically useful predictive power. However, no such test has emerged so far for the vascular complications of diabetes. Due to the multifactorial nature of diabetes and its complications, large sample collections and high-quality data sets combined with sophisticated study designs and robust statistical models are required to decipher the genetic determinants of susceptibility to diabetes complications. Toward this end, we are fortunate to have access to clinical and epidemiological data and to biological samples from ADVANCE, the largest clinical trial of T2D (involving over 11 000 patients) to date.10,11 To exploit these data and samples, we developed bioinformatic tools and, as a first step, we performed dense genotyping of genomic DNA in registered patients who participated in ADVANCE. To reduce genetic heterogeneity, we started with patients of Caucasian origin. The fundamental aims of our approach were to: (i) determine the relationships between genetic and clinical markers and diabetes complications; (ii) develop a predictive model for diabetic complications, based on genetic and clinical markers; and (iii) implement the test as a usable, clinically relevant tool.

Progress in genomics

Until recently, genetics was a science of monogenetic disorders following Mendelian law. Online Mendelian Inheritance in Man engine statistics listed, as of January 3, 2009, a total of 19 184 entries, with only 1677 remaining Mendelian phenotypes of unknown molecular basis. However, much more has to be accomplished in the area of complex diseases with their polygenic and environment-modifiable characters. One encouraging step, a major technological microarray breakthrough, has allowed us to determine hundreds of thousands of single nucleotide polymorphisms (SNPs) in several thousand subjects and in genomic association studies for a wide variety of pathologies, including heart disease, rheumatoid arthritis, colorectal cancer, and autoimmune disorders. In fact, Science declared human genetic variation to be the breakthrough of the year in 2007.12

Nevertheless, paradoxically, many investigations into complex disorders, including diabetes, hypertension, and dyslipidemia, are still hunting for a major gene, a monogenic component within the complexity. A good example here is the search for the genomic characteristics of C-reactive protein (CRP) as a cardiovascular risk factor. When a genomic determinant of CRP levels in cardiovascular disease–affected subjects is observed on a chromosomal location different from the CRP gene itself, some will suspect that this is a de facto finding, which excludes any role of the CRP gene and follows the monogenetic reasoning of Mendelian randomization.13 Thus, for instance, our collaborators uncovered a major genomic determinant of CRP levels in German subjects with myocardial infarction, an observation we validated in our French Canadian cohort at the same locus on chromosome 10, which has no CRP gene in this region.14 The appropriate explanation is a model of a network of intermediate phenotypes impacted by a separate set of genes, as illustrated in Figures 1A and 1B.12 In spheres such as hypertension and diabetes, this brings us back to Irving Page’s kaleidoscope mosaic where the “nodes” are genes and phenotypes.15

Figure 1 (A, B)
Figure 1. Network of intermediate phenotypes impacted by a
separate set of genes. (A) Model of causal association between intermediate phenotypes as governed by genes and disease. (B) Realistic modeling of the intermediate phenotype affected by several genes, with their direct and indirect impact on a complex disease.

An additional and sizable limitation to genomics gaining clinical usefulness is the disproportionate sophistication between today’s genomic armamentarium compared with the paucity of its standardized and fine phenotyping counterparts. In our search for the genomic determinants of hypertension, we have attempted to alleviate this situation by collecting over 200 cardiovascular and metabolic phenotypes16 in a panel of extended families of French Canadian origin, which has helped us to discover 46 significant quantitative trait loci for blood pressure and cardiometabolic traits that, according to Allen W. Cowley Jr, represent “the highest number of loci contributing to cardiovascular-related and metabolic traits that has been reported to date within a single population”.17 This work, completed in 2005, was performed with only 450 polymorphic markers. The number of genotyping tools has since grown exponentially: thus, over the last 2 years, we have enriched the genotyping of our French Canadian families with more than 50 000 SNPs, and our current efforts to genotype ADVANCE subjects are being performed with the Affymetrix Genome- Wide Human Array 6.0 that includes a total of 1.8 million genetic markers, more than 906 600 SNPs and 940 000 probes for the detection of copy number variations. In our study of the French Canadian founder population with 50 000 SNPs, we were able to identify sex- and/or age-specific loci at the level of single nucleotides.18 The lesson we learned from this systematic, genome-wide, sex-specific linkage study was that the traditional methods of sex adjustment would miss many of the significant genomic contributions in one or other of the sexes, making sex-separate analysis a must in the genomic investigations of a wide range of traits, such as triglyceride levels, which we demonstrated in experimental animal models19 initially, and of cardiovascular death risk prediction in diabetic men vs women.20

Table I
Table I. Overview of genome-wide scan studies for type 2 diabetes.
Abbreviations: AAO, age at onset; BMI, body mass index; JSNP, Japanese single nucleotide polymorphism; NIDDM, noninsulin-dependent diabetes mellitus; SNP, single nucleotide polymorphism; T2D, type 2 diabetes. Reproduced from reference 22: Prokopenko I, McCarthy MI, Lindgren CM. Trends Genet. 2008;24:613-621. Copyright © 2008, Elsevier Ltd.

Genetics of type 2 diabetes and its complications

One of the areas that has witnessed the fastest growth in knowledge regarding genomic determinants stemming from whole genomic association studies is that of T2D. This was initiated by Sladek et al21 and subsequently followed by several others (as summarized in Table I).22 The main conclusion to draw is that common variants with high penetrance do not contribute substantially to disease variance, but rather many modest contributions with relatively low odds ratios have to be considered, as illustrated in Figure 2, with defects in pancreatic â-cell function predominating in the overall picture. Here, we will focus on the renal complications of diabetes, bearing in mind their importance, as well as the relatively rich evidence from genetic contributions reported in the literature and summarized in Figure 3. To date, several genomic regions or individual genetic variants have been found to be linked or associated with the phenotypes closely related to diabetic complications.

Figure 2
Figure 2. Odds ratios of confirmed genes reported to be associated with type 2 diabetes.

Linkage studies

Family Investigation of Nephropathy and Diabetes (FIND), a recent genome-wide scan of glomerular filtration rate estimation, was performed in multiethnic diabetic populations.23 For all ethnicities combined, strong evidence of linkage was observed on chromosomes 1q43, 7q36.1, 8q13.3, and 18q23.3. Mexican American families, who comprised the major ethnic subpopulation in FIND, contributed to linkage on chromosomes 1q43, 2p13.3, 7q36.1, 8q13.3, and 18q23.3, whereas African American and American Indian families displayed linkage peaks on chromosomes 11p15.1 and 15q22.3, respectively. In FIND,24 the strongest evidence for linkage to diabetic nephropathy (DN) was detected on chromosomes 7q21.3, 10p15.3, 14q23.1, and 18q22.3. For albumin creatinine ratio (883 diabetic sibling pairs), the strongest linkage signals were located on chromosomes 2q14.1, 7q21.1, and 15q26.3. These results confirmed the linkage regions for DN on chromosomes 7q, 10p, and 18q from prior reports. In Mexican Americans, Puppala et al25 reported a linkage signal for glomerular filtration rate at a region on chromosome 2q near the marker D2S427 (corrected LOD score 3.3), which was shown to be influenced by genotype with diabetes interaction effects. A summary of linkage studies for DN appears in Figure 3.

Association (candidate gene) studies

A number of genes and genetic polymorphisms were tested for their association with DN, either because of their reported relevance in metabolic and signaling pathways connected to the pathophysiology of diabetic complications (functional candidates) or a combination of the former with their genomic position under a peak of ascertained linkage (positional candidates).

Can genomics contribute to the predictive power of biomarkers?

In the current literature, most genomewide association studies only report the significance of association with SNPs, while disregarding its clinical utility, ie, sensitivity and specificity, which can be combined to increase predictive power and which will one day be considered when genetics is implemented in personalized medicine.7,26 Predictive power is determined by “area under the curve” (AUC). This measure is widely used to quantify the predictive power of different classifiers in predictive studies; a value of 0.5 refers to a random prediction, and a value of 1.0 indicates an optimal prediction. The measure is visualized by receiver operating characteristic (ROC) curves, which are graphs that represent the true positive rate versus the false positive rate of different cutoff values. If the curve tends to the top left of the graph (high true positive rate, low false positive rate), the classifier is considered efficient. Prediction models rely on training (fitting) and test sets. The training set is used to fit models, and the test set serves to assess the classification efficiency of the model. To investigate the predictive ability of the best-associated SNPs on our phenotypes, we carried out 10 iteration experiments by dividing our data set randomly into training and testing sets. We observed that the predictive power of SNPs increases with the number of best, significantly-associated SNPs. The example in Figure 4 illustrates the ROC curves obtained with the support vector machine as a classifier and with 55 best-associated SNPs with diabetes complications (renal, cardiac, and cerebrovascular). At term, genomic and epigenomic data, such as DNA methylation and telomeric length, will be integrated with clinical data to give a personalized predictive risk score of diabetic outcomes. While our finding in an independent population remains to be validated, we can envisage a future where there is a change in current standard-of-care paradigms and in which we will have to wait for increases in “biomarkers,” such as microalbuminuria or creatinine, to be present before treatment is initiated as a mode of “secondary” prevention, as illustrated in Figure 5A. We have reason to be optimistic and propose that, in the future, an integrated strategy will allow the combination of clinical biomarkers with genomic ones and other types that will move us toward a scenario of “primary” prevention of complications, as described in Figure 5B.

Figure 3
Figure 3. Linkage studies for diabetic nephropathy. Diabetic nephropathy linkage (vertical bars) and candidate genes (abbreviated) mapped from a recent review of the literature summarized until 2008 (Seda et al. Unpublished data, 2009).

Figure 4
Figure 4. ROC curves representing fitting (learning) versus testing as a function of true positive and false positive rates.
Abbreviations: AUC, area under the curve; SNP, single nucleotide polymorphism; ROC, receiver operating characteristic; SVM, support vector machine.

Figure 5
Figure 5. Diabetes complications: care and prevention. (A) Current standard of care, where patients are first evaluated with biomarkers, yet therapeutic intervention is reserved until the appearance of initial complications as secondary prevention. (B) Integrated strategy for diabetes complications prevention that we are proposing, which will combine clinical and genomic biomarker profiles via a bioinformatic risk engine. This will be able to classify subjects as being susceptible to type 2 diabetes complications and allow the administration of medication as primary prevention against these complications.
Abbreviations: ACE inhibitor, angiotensin-converting enzyme inhibitor.

Table II
Table II. World Health Organization criteria for screening. Adapted from reference 27: Wilson JMG, Jungner G. Geneva: WHO. 1968;86: 281. Copyright © 1968, World Health Organization.

On this “journey,” we will have to ascertain the need for such screening in populations, as specified by the World Health Organization (WHO) (Table II).27 We believe that while such screening strategies may today be a long way off in general populations28—mainly due to difficulties in predicting low incidences of diseases—a distinct situation exists for diabetes, where subjects and health professionals are acutely conscious of the relatively high incidence of potentially avoidable complications. Naturally, several components of this proposition will have to be tested prospectively.

Acknowledgements. The authors would like to acknowledge the support from members of the Genetic Substudy Committee of ADVANCE: Drs J Chalmers, S Harrap, S MacMahon, and M Woodward. The essential contribution of the Prognomix staff, Drs Ondrej Seda, Ghislain Rocheleau, Mounsif Haloui, and Maxime Caron, Johanna Sandoval, Carole Long, Evelyne Morin, Pierre Chrétien, and Roberto Bellini, is greatly appreciated as is the collaboration of the Biogenix bioinformatics team. Our thanks go to Carole Daneau and Andrée Lévesque for administrative help and to Ovid Da Silva for editing this manuscript. The work is financially supported by Prognomix Inc, the Canadian Institutes of Health Research, the Canadian Foundation for Innovation, and the IRAP program of the National Research Council of Canada. Pavel Hamet is a Canada Research Chair in predictive genomics.


1. Hossain MP, Goyder EC, Rigby JE, et al. CKD and poverty: a growing global challenge. Am J Kidney Dis. 2009;53:166-174.
2. King H, Aubert RE, Herman WH. Global burden of diabetes, 1995-2025: prevalence, numerical estimates, and projections. Diabetes Care. 1998;21:1414- 1431.
3. Molenaar EA, Massaro JM, Jacques PF, et al. Association of lifestyle factors with abdominal subcutaneous and visceral adiposity: the Framingham Heart Study. Diabetes Care. 2009;32:505-510.
4. Romon I, Jougla E, Balkau B, et al. The burden of diabetes-related mortality in France in 2002: an analysis using both underlying and multiple causes of death. Eur J Epidemiol. 2008;23:327-334.
5. ADVANCE Collaborative Group. Action in Diabetes and Vascular Disease: patient recruitment and characteristics of the study population at baseline. Diabet Med. 2005;22:882-888.
6. Nathan DM. Long-term complications of diabetes mellitus. N Engl J Med. 1993; 328:1676-1685.
7. Guzder RN, Gatling W, Mullee MA, Mehta RL, Byrne CD. Prognostic value of the Framingham cardiovascular risk equation and the UKPDS risk engine for coronary heart disease in newly diagnosed type 2 diabetes: results from a United Kingdom study. Diabet Med. 2005;22;554-562.
8. Metcalf PA, Wells S, Scragg RK, et al. Comparison of three different methods of assessing cardiovascular disease risk in New Zealanders with type 2 diabetes mellitus. N Z Med J. 2008;121:49-57.
9. Haffner SM. Dyslipidemia management in adults with diabetes. Diabetes Care. 2004;27(suppl 1):S68-S71.
10. Patel A, MacMahon S, Chalmers J, et al. Effects of a fixed combination of perindopril and indapamide on macrovascular and microvascular outcomes in patients with type 2 diabetes mellitus (the ADVANCE trial): a randomised controlled trial. Lancet. 2007;370:829-840.
11. Patel A, MacMahon S, Chalmers J, et al. Intensive blood glucose control and vascular outcomes in patients with type 2 diabetes. N Engl J Med. 2008;358: 2560-2572.
12. Pennisi E. Breakthrough of the year. Human genetic variation. Science. 2007; 318:1842-1843.
13. Tobin MD, Minelli C, Burton PR, et al. Commentary: development of Mendelian randomization: from hypothesis test to ‘Mendelian deconfounding’. Int J Epidemiol. 2004;33:26-29.
14. Broeckel U, Hengstenberg C, Mayer B, et al. A locus on chromosome 10 influences C-reactive protein levels in two independent populations. Hum Genet. 2007;122:95-102.
15. Page IH. A Sense of the History of Discovery. Science. 1974;186:1161.
16. Hamet P, Merlo E, Seda O, et al. Quantitative founder-effect analysis of French Canadian families identifies specific loci contributing to metabolic phenotypes of hypertension. Am J Hum Genet. 2005;76:815-832.
17. Cowley AW Jr. The genetic dissection of essential hypertension. Nat Rev Genet. 2006;7:829-840.
18. Seda O, Tremblay J, Gaudet D, et al. Systematic, genome-wide, sex-specific linkage of cardiovascular traits in French Canadians. Hypertension. 2008;51: 1156-1162.
19. Ueno T, Tremblay J, Kunes J, et al. Gender-specific genetic determinants of blood pressure and organ weight: pharmacogenetic approach. Physiol Res. 2003;52:689-700.
20. Berry JD, Dyer A, Carnethon M, et al. Association of traditional risk factors with cardiovascular death across 0 to 10, 10 to 20, and >20 years follow-up in men and women. Am J Cardiol. 2008;101:89-94.
21. Sladek R, Rocheleau G, Rung J, et al. A genome-wide association study identifies novel risk loci for type 2 diabetes. Nature. 2007;445:881-885.
22. Prokopenko I, McCarthy MI, Lindgren CM. Type 2 diabetes: new genes, new understanding. Trends Genet. 2008;24:613-621.
23. Schelling JR, Abboud HE, Nicholas SB, et al. Genome-wide scan for estimated glomerular filtration rate in multi-ethnic diabetic populations: the Family Investigation of Nephropathy and Diabetes (FIND). Diabetes. 2008;57:235-243.
24. Iyengar SK, Abboud HE, Goddard KA, et al. Genome-wide scans for diabetic nephropathy and albuminuria in multiethnic populations: the family investigation of nephropathy and diabetes (FIND). Diabetes. 2007;56:1577-1585.
25. Puppala S, Arya R, Thameem F, et al. Genotype by diabetes interaction effects on the detection of linkage of glomerular filtration rate to a region on chromosome 2q in Mexican Americans. Diabetes. 2007;56:2818-2828.
26. Chung WK. Implementation of Genetics to Personalize Medicine. Gender Med. 2007;4:248-265.
27. Wilson JMG, Jungner G. Principles and practice of screening for disease. Geneva: WHO. 1968;86:281.
28. Greenland P, Lloyd-Jones D. Defining a rational approach to screening for cardiovascular risk in asymptomatic patients. J Am Coll Cardiol. 2008;52:330-332.

Keywords: biomarkers; genetics; predictive value; type 2 diabetes; vascular complications