Phenotypic Consequences of the Peroxisome Proliferator-Activated Receptor-{gamma} Pro12Ala Polymorphism: The Weight of the Evidence in Genetic Association Studies

Jose C. Florez

Diabetes Unit and Departments of Medicine and Molecular Biology Massachusetts General Hospital Boston, Massachusetts 02114; Program in Medical and Population Genetics Broad Institute of Harvard and Massachusetts Institute of Technology Cambridge, Massachusetts 02141; Department of Medicine Harvard Medical School Boston, Massachusetts 02115

Address all correspondence and requests for reprints to: Jose C. Florez, M.D., Ph.D., Diabetes Unit and Department of Molecular Biology, Wellman 8, Massachusetts General Hospital, Boston, Massachusetts 02114. E-mail: jcflorez{at}partners.org.

The peroxisome proliferator-activated receptor-{gamma} (PPAR{gamma}) is a ligand-activated transcription factor involved in lipid and glucose metabolism, fatty acid transport, and adipocyte differentiation (1). In 1997, Yen et al. (2) identified a coding missense mutation, an alanine substitution for proline at codon 12, which was present in diverse human populations with a minor allele frequency around 12% in Caucasians. Given the effect of PPAR{gamma} on glucose homeostasis and its role as the target of the novel class of thiazolidinedione antidiabetic medications, a number of groups explored whether this polymorphism might be associated with type 2 diabetes (T2D). An initial positive report (3) was followed by several negative studies; however, when samples of sufficient size were examined, the association with T2D was consistently reproducible: every study that enrolled at least 1000 subjects demonstrated an odds ratio (OR) near 1.25 in favor of the more frequent proline allele (reviewed in Ref.4). Because the risk allele is so abundant, the population-attributable risk (the fraction of disease that would disappear if this variant did not exist) has been estimated to be as high as 25% (5). Because an increasing number of studies confirms this association, the P value for the combined metaanalysis has become highly significant (P < 2 x 10–8) (4). Thus, PPAR{gamma} P12A became the first genetic variant to be convincingly associated with common T2D, and it illustrated the need to perform association studies of sufficient sample size to demonstrate a consistent role of genetic variants that have a modest effect on disease risk.

A similar approach has been successfully applied to other single nucleotide polymorphisms (SNPs) in the diabetes field. In T2D, several large-scale association studies have shown that the E23K variant in the islet ATP-sensitive potassium channel Kir6.2 is also associated with T2D (6, 7, 8, 9). An updated metaanalysis shows that the less frequent K allele is associated with T2D with an OR of 1.15 (P < 4 x 10–8), and suggests that transmission of risk occurs under a recessive model. In addition, several metaanalyses suggest that SNP-44 in the gene CAPN10 may also be associated with T2D (10, 11). In type 1 diabetes, convincing associations have long been demonstrated for the human leukocyte antigen region and the insulin variable number tandem repeat polymorphism, with more recent confirmation of the cytotoxic T lymphocyte associated-4 gene (reviewed in Ref.4).

Several groups have explored whether PPAR{gamma} P12A may also impact the incidence or rate of progression of diabetes-related complications, such as nephropathy, obesity, and atherosclerosis. In these studies, it should be noted that due to the low frequency of the alanine allele, most models have examined this association under a dominant model, comparing Pro/Pro homozygotes to carriers of the Ala allele (Pro/Ala and Ala/Ala).

The Berlin Diabetes Mellitus study examined 445 patients with T2D and found that carriers of the Ala allele had lower urinary albumin excretion than noncarriers, a difference that was magnified by longer duration of diabetes (12). A follow-up study in a different population of 316 diabetic patients showed that the presence of the Ala allele was associated with protection from nephropathy, defined as either need for dialysis or proteinuria with a serum creatinine greater than 2 mg/dl (13). Importantly, both studies controlled for the duration of diabetes and levels of glycosylated hemoglobin. Although the sample sizes were relatively small and the P values modest, the concordant results of these two independent groups suggest that PPAR{gamma} P12A may indeed influence the incidence or progression of nephropathy in patients with T2D.

Its impact on obesity is less clear. More than 30 studies have examined whether PPAR{gamma} P12A is associated with body mass index (BMI) or other measures of obesity, and the results are often conflicting. A recent metaanalysis comprising a total of 19,136 subjects has analyzed this issue in detail (14). When 40 datasets were pooled, it appeared that the effect size for BMI was 0.07 d+ units higher in carriers of the Ala allele (Pro/Ala and Ala/Ala) compared with Pro/Pro homozygotes (P = 0.02); however, there appeared to be significant heterogeneity among these studies. When separate analyses were conducted for individuals with BMI at least 27 kg/m2 and BMI below 27 kg/m2, Ala12 carriers had significantly higher BMI compared with noncarriers in the obese subgroup (0.11 d+ units higher, P = 0.0006), an effect that was not seen in the lean subgroup. The authors concluded that although the effect of PPAR{gamma} P12A on BMI was modest (accounting for about 1% of the total variance), the sum of the evidence suggests that PPAR{gamma} P12A may indeed be a genetic modifier of obesity. It is noteworthy that even P values as modest as P < 0.02 could only be attained when samples of several thousand individuals were analyzed together, in contrast to the P < 10–8 for the role of this same variant in T2D.

Cardiovascular disease is a common life-threatening complication of diabetes mellitus. Using the hard endpoint of myocardial infarction, Ridker et al. (15) examined a sample of 523 cases of myocardial infarction matched to 2092 controls, drawn from the Physicians’ Health Study. The presence of the Ala allele was associated with a protective effect against myocardial infarction, with an OR = 0.76 (P = 0.04) when adjusted for traditional cardiac risk factors, including diabetes. The OR and 95% confidence intervals were consistent with the overall finding when the analysis was restricted to the subgroup of patients free of diabetes at study entry, although the P value was no longer less than 0.05. It should be noted that this analysis again assumed a dominant model of inheritance (i.e. the benefit was conferred on carriers of the Ala allele, whether homozygotes or heterozygotes).

Given the very modest statistical significance (P = 0.04), without additional replication this potentially important finding remains an intriguing but unconfirmed hypothesis. Nevertheless, one might ask: how could PPAR{gamma} influence the risk of myocardial infarction, over and above its impact on T2D? Mounting pharmacological evidence suggests that PPAR{gamma} agonists can affect cholesterol metabolism and vascular smooth muscle physiology (16). Relevant to this question, in this issue of JCEM, Temelkova-Kurktschiev et al. (17) ask whether the P12A polymorphism in PPAR{gamma} protects against atherosclerosis in humans.

To address this hypothesis, the authors studied 622 participants in the RIAD study (Risk Factors in IGT for Atherosclerosis and Diabetes); importantly, subjects with diabetes were excluded. As a marker for atherosclerosis, they chose the well-validated phenotype of carotid intima-media thickness (IMT) and performed adequate technical controls. They then genotyped all subjects for PPAR{gamma} P12A and compared IMT measurements across the three possible genotypes. They found that carotid IMT was about 0.1 mm lower in Ala/Ala homozygotes when compared with individuals with the Pro/Pro and Pro/Ala genotypes (P < 0.05).

Does this single result withstand the effect of possible confounders? First, the above difference remained nominally significant (at P < 0.05) when adjusted for age, gender, and use of antihypertensive and lipid-lowering medications. Second, the authors controlled for an appropriately high number of additional relevant variables: interestingly, Ala/Ala homozygotes had lower BMI, lower insulin levels 2 h after an oral glucose tolerance test, lower levels of serum free fatty acids, and lower leukocyte counts than individuals with the Pro/Pro and Pro/Ala genotypes. When a multivariate linear regression analysis was performed, age, total cholesterol, diabetes, male gender, and leukocyte count positively correlated with carotid IMT, whereas high-density lipoprotein cholesterol and Ala/Ala genotype negatively correlated with carotid IMT in an independent manner. Importantly, BMI, 2-h insulin levels, and free fatty acids were not independent determinants of carotid IMT, indicating that the effect of PPAR{gamma} P12A on atherosclerosis may be distinct from its role in insulin resistance. Finally, the authors demonstrate the presence of PPAR{gamma} mRNA in human atherosclerotic lesions and macrophage-derived foam cells.

This study has several strengths. First, the subject cohort is extremely well phenotyped. Second, all relevant confounding variables are included in the analysis, and the finding stands both after adjustment in univariate analysis and in multivariate linear regression. And third, it tests a very specific hypothesis (does genotype at PPAR{gamma} P12A affect carotid IMT in humans?), such that no correction for multiplehypothesis testing is required. When a study evaluates many different polymorphisms for association with various phenotypes simultaneously, all nominal P values must be corrected for the number of hypotheses tested (either of phenotypic models or of SNPs examined); Temelkova-Kurktschiev et al. (17) focus their investigation to a single polymorphism/single phenotype, making such correction unnecessary.

The key question then remains: how should we interpret this finding? To what extent does a P value slightly less than 0.05 provide a substantial likelihood that the proposed model is correct, rather than a statistical fluctuation?

This question is important because, whereas many associations of genotype and phenotype have been published in the last decade, only a handful of them have been consistently replicated (18). This lack of reproducibility has led to a certain skepticism in the field and a widespread mistrust of any reported findings. Although even true associations can be "irreproducible" because of a false negative result in replication (see below), a much more common occurrence is that the initial positive result was a statistical fluctuation, rather than a true association: 1 in 20 studies will reach a P value of 0.05 by chance alone, and the one that does stands to benefit from publication bias. The study by Temelkova-Kurktschiev et al. (17) employed a modest sample size (622 subjects, of which only 11 were Ala/Ala homozygotes), and thus in the absence of a large genetic effect their marginally significant P value is not surprising. Is this result yet another false positive?

Evaluating the significance of a P value (such as 0.05) depends on the prior probability that a given variant is truly influencing the phenotype of interest. Within this Bayesian framework, the higher the prior probability that a given polymorphism is causing disease, the more convincing any given P value becomes. For instance, if the prior probability is 10% and we desire 90% power to detect a significant result, about two thirds of positive results will be real at P = 0.05; however, if the prior probability is lowered to 1%, only 15% of positive results will be real at that same conventional P value. This statistical phenomenon, well appreciated by clinicians when they order diagnostic tests, carries enormous weight in the field of human genetics.

There exist approximately 10 million SNPs in the human genome; a recent review has estimated the number of coding variants that are common (minor allele frequency higher than 5%) to be approximately 20,000 (19). Even if the class of causal mutations is restricted to these common coding variants (which, of course, it is not), in the absence of any prior informed suspicion that any one of them is involved on a given phenotype, less than 1 in 1000 positive results will be real at P = 0.05 when one of these variants is selected randomly; one would need a P value of less than 10–5 to have more than 80% of positive results being real. Therefore, if we are to increase general confidence in the results generated by genetic association studies, it is incumbent on us to raise the prior probability or to require much more stringent P values.

The prior probability rises when existing evidence is considered: is the gene in question involved in a relevant biological pathway? Does the variant affect protein function or expression? Are there previous associations to the phenotype under study? Do they follow the same genetic model? Do experimental models (animal or cellular) exist to support the hypothesis? Considering the many variants that exist in the human genome and the extremely low likelihood that any one of them causes a particular phenotype, we can speculate on the weight we ascribe to this prior probability: in the plausible scenario of a 1% prior, a P value of 0.001 would ensure that 90% of positive results are real.

In this context, the results of Temelkova-Kurktschiev et al. (17), while raising an interesting hypothesis, seem at best only modestly statistically significant. Although the authors examine a SNP of known consequence to T2D, they evaluate a new phenotype (carotid IMT) for which no association with PPAR{gamma} P12A has ever been documented; they consider a recessive genetic model, whereas most previous associations to related phenotypes assessed a dominant model of risk transmission; and they provide no functional data for the role of this particular polymorphism on the phenotype of interest. Given these concerns and the previous history of other reported associations at this level of statistical significance (18, 20), a P value just below 0.05 seems insufficient.

The authors appropriately point out that replication in a similar cohort is of paramount importance. Given their genetic model (that only Ala/Ala homozygotes carry the protective genotype), the low frequency of the Ala allele (on the order of 10–15% or less, making homozygotes present at frequencies of 1–2%) and the likely overestimate of this initial positive result, a much larger sample size will be needed. Indeed, another reason for the irreproducibility of genetic association studies lies in a high number of false-negative follow-up reports, which fail to replicate the original result because 1) the initial report may be an overestimate of the true effect due to publication bias and the "winner’s curse," and 2) subsequent studies are underpowered to detect a smaller, but real effect (20). In addition, it will be important to control for key confounders at study design rather than at the analysis stage: in particular, excluding subjects who take 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitors would be helpful, considering that a not-statistically significant larger proportion of Ala/Ala individuals in this study were on lipid-lowering medications (for an unknown length of time). And finally, providing molecular evidence of the effects of the Pro12Ala variant on PPAR{gamma} function will be extremely informative, not only for this phenotype but also for others in which PPAR{gamma} has been implicated.

Many issues raised by this study are relevant to genetic association studies in general. In this regard, a consistent message seems to be emerging from the work of many investigators with respect to standards that will be required before declaring that a claimed result represents a true association. These include: 1) more stringent P values, reflecting that the prior probability that any given polymorphism is causing disease is much lower than 100%; 2) much larger sample sizes needed to attain such P values, usually achieved through multicenter collaborations; 3) correction for multiple hypotheses by permutation testing or evaluation of false discovery rates; 4) replication of previously reported associations by testing exactly the same genetic model and phenotype; and 5) providing a forum for the publication of negative results, so that metaanalyses represent the closest approximation of the true genetic effect without the skew of publication bias. As these guidelines are implemented, it is hoped that convincingly robust associations that withstand the scientific method will contribute to unraveling the genetic architecture of complex phenotypes.

Acknowledgments

I am grateful to David Altshuler, Leif Groop, Joel N. Hirschhorn, and Mark J. Daly for their excellent mentorship; to Christopher Newton-Cheh, Jonathan Rosand, and Christopher J. O’Donnell for helpful advice; and to David Altshuler for valuable comments on this manuscript.

Footnotes

This work was supported by National Institutes of Health Research Career Award 1 K23 DK65978-01.

Abbreviations: BMI, Body mass index; IMT, intima-media thickness; OR, odds ratio; PPAR{gamma}, peroxisome proliferator-activated receptor-{gamma}; SNP, single nucleotide polymorphism; T2D, type 2 diabetes.

Received July 16, 2004.

Accepted July 16, 2004.

References

  1. Spiegelman BM 1998 PPAR-{gamma}: adipogenic regulator and thiazolidinedione receptor. Diabetes 47:507–514[Abstract]
  2. Yen CJ, Beamer BA, Negri C, Silver K, Brown KA, Yarnall DP, Burns DK, Roth J, Shuldiner AR 1997 Molecular scanning of the human peroxisome proliferator activated receptor {gamma} (hPPAR {gamma}) gene in diabetic Caucasians: identification of a Pro12Ala PPAR {gamma} 2 missense mutation. Biochem Biophys Res Commun 241:270–274[CrossRef][Medline]
  3. Deeb SS, Fajas L, Nemoto M, Pihlajamaki J, Mykkanen L, Kuusisto J, Laakso M, Fujimoto W, Auwerx J 1998 A Pro12Ala substitution in PPAR{gamma}2 associated with decreased receptor activity, lower body mass index and improved insulin sensitivity. Nat Genet 20:284–287[CrossRef][Medline]
  4. Florez JC, Hirschhorn JN, Altshuler D 2003 The inherited basis of diabetes mellitus: implications for the genetic analysis of complex traits. Annu Rev Hum Genomics Hum Genet 4:257–291[CrossRef]
  5. Altshuler D, Hirschhorn JN, Klannemark M, Lindgren CM, Vohl MC, Nemesh J, Lane CR, Schaffner SF, Bolk S, Brewer C, Tuomi T, Gaudet D, Hudson TJ, Daly M, Groop L, Lander ES 2000 The common PPAR{gamma} Pro12Ala polymorphism is associated with decreased risk of type 2 diabetes. Nat Genet 26:76–80[CrossRef][Medline]
  6. Gloyn AL, Weedon MN, Owen KR, Turner MJ, Knight BA, Hitman G, Walker M, Levy JC, Sampson M, Halford S, McCarthy MI, Hattersley AT, Frayling TM 2003 Large-scale association studies of variants in genes encoding the pancreatic ß-cell KATP channel subunits Kir6.2 (KCNJ11) and SUR1 (ABCC8) confirm that the KCNJ11 E23K variant is associated with type 2 diabetes. Diabetes 52:568–572[Abstract/Free Full Text]
  7. Nielsen E-MD, Hansen L, Carstensen B, Echwald SM, Drivsholm T, Glumer C, Thorsteinsson B, Borch-Johnsen K, Hansen T, Pedersen O 2003 The E23K variant of Kir6.2 associates with impaired post-OGTT serum insulin response and increased risk of type 2 diabetes. Diabetes 52:573–577[Abstract/Free Full Text]
  8. Barroso I, Luan J, Middelberg RPS, Harding AH, Franks PW, Jakes RW, Clayton D, Schafer AJ, O’Rahilly S, Wareham NJ 2003 Candidate gene association study in type 2 diabetes indicates a role for genes involved in ß-cell function as well as insulin action. PLoS Biology 1:41–55[CrossRef]
  9. Florez JC, Burtt N, de Bakker PIW, Almgren P, Tuomi T, Holmkvist J, Gaudet D, Hudson TJ, Schaffner SF, Daly MJ, Hirschhorn JN, Groop L, Altshuler D 2004 Haplotype structure and genotype-phenotype correlations of the sulfonylurea receptor and the islet ATP-sensitive potassium channel gene region. Diabetes 53:1360–1368[Abstract/Free Full Text]
  10. Weedon MN, Schwarz PEH, Horikawa Y, Iwasaki N, Illig T, Holle R, Rathmann W, Selisko T, Schulze J, Owen KR, Evans J, Del Bosque-Plata L, Hitman G, Walker M, Levy JC, Sampson M, Bell GI, McCarthy MI, Hattersley AT, Frayling TM 2003 Meta-analysis and a large association study confirm a role for calpain-10 variation in type 2 diabetes susceptibility. Am J Hum Genet 73:1208–1212[CrossRef][Medline]
  11. Song Y, Niu T, Manson JE, Kwiatkowski DJ, Liu S 2004 Are variants in the CAPN10 gene related to risk of type 2 diabetes? A quantitative assessment of population and family-based association studies. Am J Hum Genet 74:208–222[CrossRef][Medline]
  12. Herrmann S-M, Ringel J, Wang J-G, Staessen JA, Brand E 2002 Peroxisome proliferator-activated receptor-{gamma}2 polymorphism Pro12Ala is associated with nephropathy in type 2 diabetes: The Berlin Diabetes Mellitus (BeDiaM) study. Diabetes 51:2653–2657[Abstract/Free Full Text]
  13. Caramori ML, Canani LH, Costa LA, Gross JL 2003 The human peroxisome proliferator-activated receptor {gamma}2 (PPAR{gamma}2) Pro12Ala polymorphism is associated with decreased risk of diabetic nephropathy in patients with type 2 diabetes. Diabetes 52:3010–3013[Abstract/Free Full Text]
  14. Masud S, Ye S 2003 Effect of the peroxisome proliferator activated receptor-{gamma} gene Pro12Ala variant on body mass index: a meta-analysis. J Med Genet 40:773–780[Free Full Text]
  15. Ridker PM, Cook NR, Cheng S, Erlich HA, Lindpaintner K, Plutzky J, Zee RYL 2003 Alanine for proline substitution in the peroxisome proliferatoractivated receptor {gamma}-2 (PPARG2) gene and the risk of incident myocardial infarction. Arterioscler Thromb Vasc Biol 23:859–863[Abstract/Free Full Text]
  16. Hsueh WA, Bruemmer D 2004 Peroxisome proliferator-activated receptor {gamma}: implications for cardiovascular disease. Hypertension 43:297–305[Abstract/Free Full Text]
  17. Temelkova-Kurktschiev T, Hanefeld M, Chinetti G, Zawadzki C, Haulon S, Kubaszek A, Koehler C, Leonhardt W, Staels B, Laakso M 2004 Ala12Ala genotype of the peroxisome proliferator-activated receptor {gamma}2 protects against atherosclerosis. J Clin Endocrinol Metab 89:4238–4242[Abstract/Free Full Text]
  18. Hirschhorn JN, Lohmueller K, Byrne E, Hirschhorn K 2002 A comprehensive review of genetic association studies. Genet Med 4:45–61[Medline]
  19. Carlson CS, Eberle MA, Kruglyak L, Nickerson DA 2004 Mapping complex disease loci in whole-genome association studies. Nature 429:446–452[CrossRef][Medline]
  20. Lohmueller K, Pearce CL, Pike M, Lander ES, Hirschhorn JN 2003 Meta-analysis of genetic association studies supports a contribution of common variants to susceptibility to common disease. Nat Genet 33:177–182[CrossRef][Medline]