Association Testing of the Protein Tyrosine Phosphatase 1B Gene (PTPN1) With Type 2 Diabetes in 7,883 People
http://www.100md.com
糖尿病学杂志 2005年第6期
1 Department of Molecular Biology, Massachusetts General Hospital, Boston, Massachusetts
2 Department of Medicine (Diabetes Unit), Massachusetts General Hospital, Boston, Massachusetts
3 Program in Medical and Population Genetics, Broad Institute of Harvard and MIT, Cambridge, Massachusetts
4 Department of Medicine, Harvard Medical School, Boston, Massachusetts
5 Department of Clinical Sciences—Diabetes and Endocrinology, University Hospital Malm, Lund University, Malm, Sweden
6 Department of Clinical Sciences, University Hospital Malm, Lund University, Malm, Sweden
7 Department of Medicine, Helsinki University Central Hospital, Folkhalsan Genetic Institute, Folkhalsan Research Center, and Research Program for Molecular Medicine, University of Helsinki, Helsinki, Finland
8 University of Montreal Community Genomic Center, Chicoutimi Hospital, Quebec, Canada
9 McGill University and Genome Quebec Innovation Centre, Montreal, Canada
10 Genomics Collaborative Division, SeraCare LifeSciences, Cambridge, Massachusetts
11 Department of Genetics, Harvard Medical School, Boston, Massachusetts
12 Divisions of Genetics and Endocrinology, Children’s Hospital, Boston, Massachusetts
ABSTRACT
Protein tyrosine phosphatase (PTP)-1B, encoded by the PTPN1 gene, inactivates the insulin signal transduction cascade by dephosphorylating phosphotyrosine residues in insulin signaling molecules. Due to its chromosomal location under a chromosome 20 linkage peak and the metabolic effects of its absence in knockout mice, it is a candidate gene for type 2 diabetes. Recent studies have associated common sequence variants in PTPN1 with type 2 diabetes and diabetes-related phenotypes. We sought to replicate the association of common single nucleotide polymorphisms (SNPs) and haplotypes in PTPN1 with type 2 diabetes, fasting plasma glucose, and insulin sensitivity in a large collection of subjects. We assessed linkage disequilibrium, selected tag SNPs, and typed these markers in 3,347 cases of type 2 diabetes and 3,347 control subjects as well as 1,189 siblings discordant for type 2 diabetes. Despite power estimated at >95% to replicate the previously reported associations, no statistically significant evidence of association was observed between PTPN1 SNPs or common haplotypes with type 2 diabetes or with diabetic phenotypes.
The PTPN1 gene encodes protein tyrosine phosphatase (PTB)-1B (EC 3.3.3.48), an enzyme that binds the insulin receptor (1) and dephosphorylates phosphotyrosine residues of the activated receptor and the insulin receptor substrate 1, thus downregulating the insulin signaling cascade (2). Mice deficient for PTPN1 display enhanced insulin sensitivity and resistance to diet-induced obesity (3), as well as general leanness due to an increased basal metabolic rate (4). In vitro inhibition of PTP-1B by small molecules leads to greater insulin sensitivity (5,6). These functional features, coupled with its genomic location under the chromosome 20q13 type 2 diabetes linkage peak (7), support PTPN1 as a candidate gene that might harbor variants influencing susceptibility to insulin resistance and type 2 diabetes.
Several groups have examined genetic variants in the human PTPN1 gene for association with type 2 diabetes. A rare ProLeu change at position 387 was reportedly associated with type 2 diabetes in a Danish sample of 527 cases and 542 control subjects (nominal P = 0.036) (8) but not in a Chinese sample of 329 cases and 238 control subjects (9) or a Finnish sample of 257 cases and 285 control subjects (10). A common insertion of a guanosine at position 1484 of the 3' untranslated region (UTR) (1484insG) was associated with features of the metabolic syndrome in two Italian samples comprising an aggregate of 812 normoglycemic individuals (11) but was not associated with type 2 diabetes, BMI, fasting glucose, or fasting insulin in the Danish sample (8) or a large Swedish sample of 2,309 nondiabetic subjects (12).
Recently, a more extensive evaluation of noncoding genetic variants at this locus has been reported (13). In two nonoverlapping case/control samples totalling 610 and 475 individuals, Bento et al. (13) found that several single nucleotide polymorphisms (SNPs) were associated with type 2 diabetes, with odds ratios (ORs) in each sample near 1.3 and nominal P values <0.01. Examination of the patterns of linkage disequilibrium (LD) in this region revealed very limited haplotype diversity, with most SNPs falling within a single LD block (defined by the authors as the linear DNA segment where pairwise D' between SNPs is >0.8) that spans 100 kb from the promoter region to the 3' UTR. Of eight common haplotypes, two of them (GTCCTGTO and its reciprocal ACTTCAGO) capture 80% of haplotypic diversity in this locus; as predicted from the SNP results, either the former haplotype protected against type 2 diabetes and/or the latter haplotype conferred risk (P < 0.005) (13).
In a companion article (14), the same group evaluated these SNPs and haplotypes in PTPN1 for association with quantitative glycemic traits in an independent sample. Palmer et al. (14) found that the protective haplotype led to higher insulin sensitivity and lower fasting glucose under a recessive/multiplicative model in control individuals, whereas the risk haplotype had opposite effects under a dominant model (14). These results in a third nonoverlapping sample supported the conclusion that common variation in PTPN1 substantially contributed to type 2 diabetes risk and related traits.
As previously noted (15eC17), replication of genetic association studies with adequately powered samples is critical given the low prior probabilities for any given gene and to assess the robustness of findings in regard to specific aspects of study design and interpretation. In type 2 diabetes, consistently reproducible associations have been documented for the P12A polymorphism in the peroxisome proliferatoreCactivated receptor , the E23K polymorphism in the ATP-sensitive potassium channel Kir6.2, and, more recently, SNP-44 in CAPN10 (18). We set out to test the reproducibility of the association of PTPN1 with type 2 diabetes in a large collection of diabetes and control samples.
RESEARCH DESIGN AND METHODS
The diabetic sample subjects have been described elsewhere (19,20). Briefly, they comprise 1,189 Scandinavian siblings discordant for type 2 diabetes; a Scandinavian case-control sample totaling 942 subjects individually matched for age, BMI, and geographic region; a case-control sample from Sweden totaling 1,028 subjects who were individually matched for sex, age, and BMI; an individually matched case-control sample totaling 254 subjects from the Saguenay Lac-St. Jean region in Quebec; and two case-control samples individually matched for age, sex, and self-reported ethnicity from the U.S. (2,452 subjects) and Poland (2,018 subjects), both collected by Genomics Collaborative, Inc. (GCI).
Genotyping and phenotypic analysis.
Genotyping was performed by primer extension of multiplex products with detection by matrix-assisted laser desorption ionizationeCtime of flight mass spectroscopy (21) using a Sequenom platform. Our genotyping success rate was 97.3%, and our consensus rate (based on 11,799 duplicate genotypes) was 99.9%. The detailed list of SNPs and their genotyping assays are shown in online appendix Table 1 (available at http://diabetes.diabetesjournals.org). Genotype counts for the various samples tested in this study are shown in online appendix Table 2. This information is also posted on our website (http://genetics.mgh.harvard.edu/AltshulerWeb/publicationdata/Florez_PTPN1.html).
Plasma glucose (fasting and during an oral glucose tolerance test [OGTT]) was measured by a glucose oxidase method on a Beckman Glucose analyzer (Beckman Instruments, Fullerton, CA). Insulin was measured by radioimmunoassay. A 75-g OGTT was performed in a subset of the control Scandinavian subjects (n = 766). The whole-body insulin sensitivity index (ISI) was calculated as in Matsuda and DeFronzo (22), and because of nonnormality it was logarithmically transformed. Insulin resistance by homeostasis model assessment was calculated as in Matthews et al. (23).
Haplotype structure.
To evaluate the haplotype structure of the PTPN1 gene, we genotyped 42 SNPs in a multigenerational panel of Centre d’Etude du Polymorphisme Humain (CEPH) Caucasian pedigrees totaling 120 chromosomes selected by the HapMap project (24). SNPs were initially selected from genotypes already available from the HapMap and from variants previously studied by others; additional SNPs were added to refine areas of low SNP density or clarify the extent of LD. The list of SNPs included the eight SNPs that displayed the strongest nominal associations with type 2 diabetes in Bento et al. (13) (kindly provided by D.W. Bowden), 9 SNPs previously studied in this gene (8,10), and 16 SNPs previously chosen by the HapMap project; the rest were obtained from dbSNP.
These SNPs span 112 kb, from 28 kb upstream of the transcription start site to 10 kb downstream of the end of the PTPN1 3' UTR. Three of the SNPs attempted were technical failures (<75% genotyping percentage), 1 failed Hardy-Weinberg equilibrium, and 7 were found to be monomorphic in this population; thus, the final set comprised 31 (74%) working SNPs that were used to evaluate patterns of LD. The average spacing between these 31 SNPs is 3.6 kb. Haplotype blocks were determined as in Gabriel et al. (25). In the disease samples, case and control chromosomes were phased together using the expectation-maximization algorithm of Excoffier and Slatkin (26) modified to process larger data files using the partition-ligation approach as previously described (27). Each individual chromosome was assigned the phased haplotype of highest maximum likelihood.
Tag SNP selection.
To correlate our findings with those of the literature, we chose to genotype all eight SNPs provided by D.W. Bowden (rs941798, rs3787345, rs754118, rs2282147, rs718049, rs718050, rs3787348, and 1484insG) and the two missense SNPs (P387L and G381S) in our disease panels (T420M was monomorphic in our reference panel). We selected additional tag SNPs by a newly developed algorithm named Tagger (Paul I.W. de Bakker et al., unpublished observations; http://www.broad.mit.edu/mpg/tagger/), which selects single markers and multimarker combinations of tags based on pairwise LD between markers (set at minimum r2 > 0.8). This method allows the user to "force in" markers of interest, such that the additional tags selected capture those markers not already tagged by the forced-in set. This procedure resulted in 18 tag SNPs (the above 10 plus 8 additional markers; Figs. 1 and Table 1); if no SNPs had been forced in a priori, Tagger would have yielded 12 tag SNPs. The 18 tag SNPs were validated by an independent method (28) and were found to be robust predictors of all single SNPs (minimum RS2 > 0.95) and haplotypes (minimum Rh2 > 0.85).
Statistical analysis.
Power calculations were performed with the program of Purcell et al. (29), available at http://statgen.iop.kcl.ac.uk/gpc. To examine the association of SNPs and haplotypes with type 2 diabetes, we used simple 2 analysis in the case-control samples and the discordant allele test (30) in the sibpairs. Haplotypes were compared individually against all others; only haplotypes with frequencies >5% in the reference panel were examined. Results from the various samples were combined by Mantel-Haenszel meta-analysis of the ORs (16). Homogeneity among study samples was tested using a Pearson 2 goodness-of-fit test as previously described (16).
Phenotype comparisons.
We concluded that under the most parsimonious model, the positive findings obtained by Bento et al. (13) and Palmer et al. (14) could be reduced to two statistically indistinguishable and complementary results: the association of one high-frequency PTPN1 haplotype with risk of type 2 diabetes (and related variables of fasting glucose and insulin sensitivity) and the converse result of protection from disease and related quantitative traits by another high-frequency reciprocal haplotype. We therefore formulated our study as a test of a single hypothesis. In secondary analyses, we tested this putative association under both a neutral, multiplicative genetic model and the genotypic model reported in Palmer et al. (14). Only nondiabetic subjects were included in our phenotypic analyses. Comparisons of phenotypic variables between genotypic groups were made by Student’s t test (two groups) or ANOVA (more than two groups).
RESULTS
Characterization of common sequence variation at PTPN1.
Consistent with the results of Bento et al. (13), genotyping of 31 SNPs defined a single segment of high LD ("haplotype block") spanning 83 kb (between SNPs rs6063528 and rs3787348; Fig. 1); due to their low minor allele frequency, the two missense SNPs G381S and P387L were outside the boundaries of this block as defined by Gabriel et al. (25). The block comprises five haplotypes of >5% frequency, which together capture 90% of haplotypic diversity at this locus. Of note, genotyping additional SNPs has permitted us to further refine the haplotype structure, such that the risk haplotype of 33eC34% frequency in Bento et al. (13) is subdivided into haplotypes B and D in our reference panel, with frequencies of 22 and 6%, respectively, in the CEPH reference sample. (In the GCI U.S. sample, these haplotypes are present at 26 and 9% frequencies in control subjects, in agreement with the control samples of Bento et al. [13].) The 31 working SNPs are well captured by a set of 18 tagging markers, which include both missense SNPs and those previously studied by Bento et al. (13), and which were genotyped in this study (see RESEARCH DESIGN AND METHODS).
Power calculations.
To estimate the sample size required to replicate the association of the high-frequency haplotype in Bento et al. (13) with protection from type 2 diabetes, we assumed a haplotype frequency of 45%, a type 2 diabetes disease prevalence of 8%, and a genotypic relative risk of 1.3 under a recessive model. Under these parameters, we estimated that our combined sample of 3,347 case/control pairs and 1,189 discordant sibs would provide >95% power to reject the null hypothesis of no association at P < 0.05. If we reduce the genotypic relative risk to 1.2 in order to allow for the likely overestimate of the initial report due to the "winner’s curse," we estimate power at >75% to reject the null hypothesis of no association at P < 0.05; this power is raised to >80% under a multiplicative model. Similarly, we estimated that the size of our phenotypic sample provides 90% power to detect a 2-mg/dl difference in fasting plasma glucose under the genotypic model proposed by Palmer et al. (14).
Association study.
Association results for each of the 18 SNPs and five common haplotypes are presented in Table 1. No association was observed to any of the SNPs or haplotypes spanning PTPN1 with type 2 diabetes in the overall sample. Results that were nominally significant in one of the three subsamples (rs6020546, rs2230605, and P387L) were not reproduced in the other two subsamples or displayed a trend in the opposite direction. No heterogeneity among subsamples was observed.
In an attempt to understand the differences between our results and those previously published, we examined the frequencies of the two major haplotypes described by Bento et al. (13) in the various samples studied in this and in previous reports. We note that the frequencies of haplotype A and the aggregate of haplotypes B and D are quite similar in all the control samples (from both this study and that of Bento et al. [13]) and in the diabetic samples studied in this report. The two samples showing a different haplotypic frequency are the case series in Bento et al. (13) (Table 2).
Genotype-phenotype correlations.
Since the single SNP results of Palmer et al. (14) could be explained by the effects of a single high-frequency haplotype (and/or the converse effect of an alternate, high-frequency haplotype), we confined analysis of associations to diabetic subphenotypes to this single hypothesis. We evaluated the effects of the presence of haplotype A (the putative protective haplotype) or haplotype B (the major contributor to the putative risk haplotype) on the two measures of glucose homeostasis found by Palmer et al. to be significantly affected. We measured fasting plasma glucose in control individuals from both the GCI U.S. and the Scandinavian samples, and the whole-body ISI in control individuals from the Scandinavian sample. We examined both parameters under a neutral genotypic model (comparing carriers of haplotype A [A/A or A/X] with carriers of haplotype B [B/B or B/X] and excluding A/B heterozygotes from the analysis) and under the genotypic model thought to be most consistent with the findings of Palmer et al. (comparing A/A homozygotes with carriers of haplotype B, including A/B heterozygotes [B/B, A/B, or B/X]).
Taking the entire sample as a whole, we did not find significant evidence for association of PTPN1 haplotypes with either measure of glucose metabolism. In agreement with Palmer et al. (14), in the GCI U.S. sample, fasting plasma glucose was elevated in haplotype B carriers under the neutral model (A carriers vs. B carriers: 90.2 ± 14.4 vs. 94.1 ± 14.1 mg/dl [mean ± SD], one-sided P = 0.015); this difference was not statistically significant under the recessive genetic model of Palmer et al. (A/A homozygotes vs. B carriers: 91.5 ± 16.8 vs. 92.7 ± 14.6 mg/dl, one-sided P = 0.26). In the Scandinavian sample, however, fasting plasma glucose was not significantly different according to haplotype status (A carriers vs. B carriers: 95.0 ± 11.5 vs. 93.7 ± 10.9 mg/dl; A/A homozygotes vs. B carriers: 95.1 ± 11.1 vs. 94.0 ± 11.1 mg/dl).
Because the GCI case/control samples were not but the Scandinavian samples were matched for BMI, we developed the post hoc hypothesis that differences in patient phenotypes might contribute to these divergent results. There was a nominally significant difference in the average BMI of control individuals between the Genomics Collaborative U.S. and Scandinavian samples (27.2 ± 4.9 vs. 26.7 ± 3.7 kg/m2, respectively, two-sided P < 0.02). Moreover, the observed difference in fasting plasma glucose between carriers of the A and B haplotypes in the Genomics Collaborative U.S. sample was most obvious for the highest quartile of BMI under the neutral model (P < 0.0001; Table 3); the difference was also present under the recessive model (P < 0.003; Table 3). In the Scandinavian sample, although there was no statistically significant difference in fasting plasma glucose according to genotypic group across each BMI quartile, fasting plasma glucose rose across BMI quartiles in carriers of the B haplotype (P < 0.02; Table 3).
Because of the difference in fasting plasma glucose between carriers of haplotypes A and B in the most obese subjects of the GCI U.S. sample, we wondered whether an association of either haplotype with type 2 diabetes might be present in this group. When the association analysis was restricted to the 325 control individuals and 816 diabetic subjects with BMI >29.5 kg/m2, no association was observed (haplotype A: OR 1.02 [95% CI 0.82eC1.18], two-sided P = 0.84; haplotype B: 0.96 [0.85eC1.28], P = 0.70). In addition, we detected no effect of genotype on BMI (data not shown).
We explored whether insulin resistance, as measured by the whole-body ISI, might be affected in the 766 control Scandinavian subjects for whom we had OGTT data. No statistically significant difference in insulin sensitivity was found among genotypic groups according to presence of either haplotype A or B (online appendix Table 3). In contrast to the results of Palmer et al. (14), a trend toward greater insulin sensitivity was found among carriers of haplotype B under the neutral model (two-sided P = 0.057; online appendix Table 3).
Finally, we examined whether the specific 1484insG variant, previously reported to induce insulin resistance (11), affected similar glycemic measures in our control subjects. We detected no statistically significant difference between nondiabetic Scandinavian carriers and noncarriers of this polymorphism in fasting glucose (94.4 ± 10.9 vs. 94.9 ± 11.2 mg/dl [mean ± SD]), fasting insulin (8.04 ± 4.9 vs. 8.03 ± 4.7 mU/l) or insulin resistance by homeostatis model assessment (1.94 ± 1.2 vs. 1.95 ± 1.2). Similarly, fasting glucose was not significantly different between 1484insG carriers and noncarriers in the GCI U.S. samples (91.9 ± 14.9 vs. 91.1 ± 14.4 mg/dl). Nondiabetic Scandinavian siblings discordant for genotype at this locus also failed to show any statistically significant differences in these three parameters (data not shown).
DISCUSSION
We set out to test the previous hypothesis of association of SNPs and haplotypes in PTPN1 with type 2 diabetes (13) and with fasting plasma glucose and insulin sensitivity (14). We confirmed the extensive LD among common variants in the PTPN1 gene (13) but failed to detect an association of any SNP or common haplotype with type 2 diabetes in either our individual samples or their combination by meta-analysis.
Lack of replication can have several possible explanations. First, it is possible that the initial association may have been a statistical fluctuation. This seems less likely since the authors replicated their findings in a second population, and two measures of glucose homeostasis varied according to PTPN1 haplotype in the predicted direction (14). To evaluate the significance of such a finding it will be useful to simulate the frequency, in the absence of a true association, of obtaining such P values across multiple studies after examining many SNPs and haplotypes in each of several phenotypes (type 2 diabetes and related subphenotypes).
A second possibility is that the current study constitutes a false-negative attempt at replication. In the setting of a true initial association, failure of replication can be due to 1) insufficient power (16), 2) heterogeneity among subsamples, and/or 3) heterogeneity within subsamples that renders them inadequate for case/control analysis. Although our study was well powered to replicate the model of Bento et al. (13), it is possible that the initial association may have been overestimated due to the "winner’s curse." Regardless, the best estimate of the OR is bounded by the 95% CIs of all available data. With respect to genetic heterogeneity among subsamples, formal tests for heterogeneity were negative in this and previous studies that have analyzed these samples (19,20,27). Finally, with respect to within-sample heterogeneity, we note that these samples have shown consistent results when genotyped for the two most widely reproduced associations with type 2 diabetes, P12A in peroxisome proliferatoreCactivated receptor (31 and N.B., J.C.F., J.N.H., D.A., unpublished observations) and E23K in Kir6.2 (19,27).
A third possibility is that the association signal reported by Bento et al. (13) was valid, but there is heterogeneity among the various populations examined 1) in the extent of LD in this region, 2) in the phenotypic characteristics of patients studied, or 3) in the frequencies of genetic or environmental modifiers. In regard to LD, the LD pattern in PTPN1 observed in our reference panel and disease samples is nearly identical to that observed by Bento et al. (13). With regard to phenotypic characterization, our case and control subjects were matched for age, sex, self-reported ethnicity (in all cases), and BMI (in the Scandinavian and Canadian subsamples). In contrast, the first case/control population of Bento et al. (13) consisted of cases with both type 2 diabetes and end-stage renal disease (mean BMI 28.5 kg/m2) who were matched to nondiabetic control subjects; the second case/control population of Bento et al. (13) consisted of relatively obese diabetic subjects from the Diabetes Heart Study (mean BMI 33.7 kg/m2), matched to a reference sample of nondiabetic control subjects obtained from a publicly available repository. The sample of Palmer et al. (14) consisted of nondiabetic Hispanic subjects (mean BMI 29.2 kg/m2). Since the definition of case status and the methods to match controls differs among studies, it is possible that one of these considerations and/or population stratification may account for the disparate results (32). In addition, we note that in one of our phenotypic analyses, the whole-body ISI derived from the OGTT is not a perfect correlate of the ISI obtained from the intravenous glucose tolerance test employed by Palmer et al. (14).
Heterogeneity in gene-gene and/or gene-environment interactions could also contribute to divergent findings. We observed an association of a subset of the risk haplotype of Palmer et al. (14) (haplotype B in our study) with fasting plasma glucose in the GCI U.S. subsample, albeit confined in a post hoc analysis to the quartile with the highest BMI. The same effect was not seen in the Scandinavian sample, although fasting plasma glucose rose with BMI in Scandinavian carriers of the B haplotype. Because the Scandinavian subjects in the highest quartile of BMI were leaner than their GCI U.S. counterparts (BMI 31.58 ± 2.54 vs. 33.74 ± 3.80 kg/m2, respectively, P < 10eC10; Table 3), one might speculate that in Scandinavian subjects of an even heavier BMI, fasting plasma glucose may also have been higher in carriers of the B haplotype than in carriers of the A haplotype. But since these observations were post hoc, they should be considered an exploratory, hypothesis-generating exercise.
Thus, our different findings could be explained by the presence of unaccounted modifier genes in the various populations that affect diabetes-related phenotypes. In this scenario, variants that are enriched in individuals of Hispanic ancestry or genes that influence renal function or obesity might interact with PTPN1 to induce insulin resistance and raise fasting plasma glucose. If the effect is large enough, an association with type 2 diabetes might be seen in a subset of samples but not in others. Given these multiple explanations, further refinement of patient phenotypes, matching for all possible confounders and most importantly replication in additional well-powered samples, will be necessary to evaluate the hypothesized effect of PTPN1 on type 2 diabetes and related traits.
ACKNOWLEDGMENTS
J.C.F. is supported by National Institutes of Health Research Career Award 1 K23 DK65978-01. J.N.H. is a recipient of a Burroughs Wellcome Career Award in Biomedical Sciences. D.A. is a Charles E. Culpeper Scholar of the Rockefeller Brothers Fund and a Burroughs Wellcome Fund Clinical Scholar in Translational Research, the latter of which supported this work. D.A., M.J.D., and J.N.H. are recipients of The Richard and Susan Smith Family Foundation/American Diabetes Association Pinnacle Program Project Award. L.G., T.T., and the Botnia Study are principally supported by the Sigrid Juselius Foundation, the Academy of Finland, the Finnish Diabetes Research Foundation, The Folkhalsan Research Foundation, EC (BM4-CT95-0662, GIFT), the Swedish Medical Research Council, the JDF Wallenberg Foundation, and the Novo Nordisk Foundation.
We thank D.W. Bowden for sharing experimental protocols and unpublished data, the Botnia research team for clinical contributions, and the members of the Altshuler, Hirschhorn, Daly, and Groop labs for helpful discussions.
FOOTNOTES
L.G. and D.A. jointly supervised this work.
L.G. has received consulting fees from and served on advisory panels for Aventis-Sanofi, Bristol-Myers Squibb, Kowa, and Roche. D.A. has received consulting fees from and served on an advisory panel for GCI.
Additional information for this article can be found in an online appendix at http://diabetes.diabetesjournals.org.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
CEPH, Centre d’Etude du Polymorphisme Humain; GCI, Genomics Collaborative, Inc; ISI, insulin sensitivity index; LD, linkage disequilibrium; OGTT, oral glucose tolerance test; PTP, protein tyrosine phosphatase; SNP, single nucleotide polymorphism; UTR, untranslated region
REFERENCES
Seely BL, Staubs PA, Reichart DR, Berhanu P, Milarski KL, Saltiel AR, Kusari J, Olefsky JM: Protein tyrosine phosphatase 1B interacts with the activated insulin receptor. Diabetes 45: 1379eC1385, 1996
Goldstein BJ: Protein-tyrosine phosphatases: emerging targets for therapeutic intervention in type 2 diabetes and related states of insulin resistance. J Clin Endocrinol Metab 87: 2474eC2480, 2002
Elchebly M, Payette P, Michaliszyn E, Cromlish W, Collins S, Loy AL, Normandin D, Cheng A, Himms-Hagen J, Chan C-C, Ramachandran C, Gresser MJ, Tremblay ML, Kennedy BP: Increased insulin sensitivity and obesity resistance in mice lacking the protein tyrosine phosphatase-1B gene. Science 283: 1544eC1548, 1999
Klaman LD, Boss O, Peroni OD, Kim JK, Martino JL, Zabolotny JM, Moghal N, Lubkin M, Kim Y-B, Sharpe AH, Stricker-Krongrad A, Shulman GI, Neel BG, Kahn BB: Increased energy expenditure, decreased adiposity, and tissue-specific insulin sensitivity in protein-tyrosine phosphatase 1B-deficient mice. Mol Cell Biol 20: 5479eC5489, 2000
Xie L, Lee SY, Andersen JN, Waters S, Shen K, Guo XL, Moller NP, Olefsky JM, Lawrence DS, Zhang ZY: Cellular effects of small molecule PTP-1B inhibitors on insulin signaling. Biochemistry 42: 12792eC12804, 2003
Wiesmann C, Barr KJ, Kung J, Zhu J, Erlanson DA, Shen W, Fahr BJ, Zhong M, Taylor L, Randal M, McDowell RS, Hansen SK: Allosteric inhibition of protein tyrosine phosphatase 1B. Nat Struct Mol Biol 11: 730eC737, 2004
Florez JC, Hirschhorn JN, Altshuler D: The inherited basis of diabetes mellitus: implications for the genetic analysis of complex traits. Annu Rev Genomics Hum Genet 4: 257eC291, 2003
Echwald SM, Bach H, Vestergaard H, Richelsen B, Kristensen K, Drivsholm T, Borch-Johnsen K, Hansen T, Pedersen O: A P387L variant in protein tyrosine phosphatase-1B (PTP-1B) is associated with type 2 diabetes and impaired serine phosphorylation of PTP-1B in vitro. Diabetes 51: 1eC6, 2002
Weng J, Yan J, Huang Z, Sui Y, Xiu L: Missense mutation of Pro387Leu in protein tyrosine phosphatase-1B (PTP-1B) is not associated with type 2 diabetes in a Chinese Han population (Letter). Diabetes Care 26: 2957, 2003
Santaniemi M, Ukkola O, Kesniemi YA: Tyrosine phosphatase 1B and leptin receptor genes and their interaction in type 2 diabetes. J Intern Med 256: 48eC55, 2004
Di Paola R, Frittitta L, Miscio G, Bozzali M, Baratta R, Centra M, Spampinato D, Santagati MG, Ercolino T, Cisternino C, Soccio T, Mastroianno S, Tassi V, Almgren P, Pizzuti A, Vigneri R, Trischitta V: A variation in 3' UTR of hPTP1B increases specific gene expression and associates with insulin resistance. Am J Hum Genet 70: 806eC812, 2002
Dahlman I, Wahrenberg H, Persson L, Arner P: No association of reported functional protein tyrosine phosphatase 1B 3' UTR gene polymorphism with features of the metabolic syndrome in a Swedish population. J Intern Med 255: 694eC695, 2004
Bento JL, Palmer ND, Mychaleckyj JC, Lange LA, Langefeld CD, Rich SS, Freedman BI, Bowden DW: Association of protein tyrosine phosphatase 1B gene polymorphisms with type 2 diabetes. Diabetes 53: 3007eC3012, 2004
Palmer ND, Bento JL, Mychaleckyj JC, Langefeld CD, Campbell JK, Norris JM, Haffner SM, Bergman RN, Bowden DW: Association of protein tyrosine phosphatase 1B gene polymorphisms with measures of glucose homeostasis in Hispanic Americans: the Insulin Resistance Atherosclerosis Study (IRAS) Family Study. Diabetes 53: 3013eC3019, 2004
Hirschhorn JN, Lohmueller K, Byrne E, Hirschhorn K: A comprehensive review of genetic association studies. Genet Med 4: 45eC61, 2002
Lohmueller K, Pearce CL, Pike M, Lander ES, Hirschhorn JN: Meta-analysis of genetic association studies supports a contribution of common variants to susceptibility to common disease. Nat Genet 33: 177eC182, 2003
Florez JC: Phenotypic consequences of the peroxisome proliferator-activated receptor- Pro12Ala polymorphism: the weight of the evidence in genetic association studies. J Clin Endocrinol Metab 89: 4234eC4237, 2004
O’Rahilly S, Barroso I, Wareham NJ: Genetic factors in type 2 diabetes: the end of the beginning Science 307: 370eC373, 2005
Florez JC, Sjogren M, Burtt N, Orho-Melander M, Schayer S, Sun M, Almgren P, Lindblad U, Tuomi T, Gaudet D, Hudson TJ, Daly MJ, Ardlie KG, Hirschhorn JN, Altshuler D, Groop L: Association testing in 9,000 people fails to confirm the association of the insulin receptor substrate-1 G972R polymorphism with type 2 diabetes. Diabetes 53: 3313eC3318, 2004
Winckler W, Graham RR, de Bakker PIW, Sun M, Almgren P, Tuomi T, Gaudet D, Hudson TJ, Ardlie KG, Daly MJ, Hirschhorn JN, Groop L, Altshuler D: Association testing of variants in the hepatocyte nuclear factor 4 gene with risk of type 2 diabetes in 7,883 people. Diabetes 54: 886eC892, 2005
Tang K, Fu DJ, Julien D, Braun A, Cantor CR, Koster H: Chip-based genotyping by mass spectrometry. Proc Natl Acad Sci U S A 96: 10016eC10020, 1999
Matsuda M, DeFronzo R: Insulin sensitivity indices obtained from oral glucose tolerance testing: comparison with the euglycemic insulin clamp. Diabetes Care 22: 1462eC1470, 1999
Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC: Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 28: 412eC419, 1985
The International HapMap Consortium: The International HapMap Project. Nature 426: 789eC796, 2003
Gabriel SB, Schaffner SF, Nguyen H, Moore JM, Roy J, Blumenstiel B, Higgins J, DeFelice M, Lochner A, Faggart M, Liu-Cordero SN, Rotimi C, Adeyemo A, Cooper R, Ward R, Lander ES, Daly MJ, Altshuler D: The structure of haplotype blocks in the human genome. Science 296: 2225eC2229, 2002
Excoffier L, Slatkin M: Maximum-likelihood estimation of molecular haplotype frequencies in a diploid population. Mol Biol Evol 12: 921eC927, 1995
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: Haplotype structure and genotype-phenotype correlations of the sulfonylurea receptor and the islet ATP-sensitive potassium channel gene region. Diabetes 53: 1360eC1368, 2004
Stram DO, Haiman CA, Hirschhorn JN, Altshuler D, Kolonel LN, Henderson BE, Pike MC: Choosing haplotype-tagging SNPS based on unphased genotype data using a preliminary sample of unrelated subjects with an example from the Multiethnic Cohort Study. Hum Hered 55: 27eC36, 2003
Purcell S, Cherny SS, Sham PC: Genetic power calculator: design of linkage and association genetic mapping studies of complex traits. Bioinformatics 19: 149eC150, 2003
Boehnke M, Langefeld CD: Genetic association mapping based on discordant sib pairs: the discordant-alleles test. Am J Hum Genet 62: 950eC961, 1998
Ardlie KG, Lunetta KL, Seielstad M: Testing for population subdivision and association in four case-control studies. Am J Hum Genet 71: 1478eC1480, 2002
Freedman ML, Reich D, Penney KL, McDonald GJ, Mignault AA, Patterson N, Gabriel SB, Topol EJ, Smoller JW, Pato CN, Pato MT, Petryshen TL, Kolonel LN, Lander ES, Sklar P, Henderson B, Hirschhorn JN, Altshuler D: Assessing the impact of population stratification on genetic association studies. Nat Genet 36: 388eC393, 2004
Lewontin RC: On measures of gametic disequilibrium. Genetics 120: 849eC852, 1988(Jose C. Florez, Christina)
2 Department of Medicine (Diabetes Unit), Massachusetts General Hospital, Boston, Massachusetts
3 Program in Medical and Population Genetics, Broad Institute of Harvard and MIT, Cambridge, Massachusetts
4 Department of Medicine, Harvard Medical School, Boston, Massachusetts
5 Department of Clinical Sciences—Diabetes and Endocrinology, University Hospital Malm, Lund University, Malm, Sweden
6 Department of Clinical Sciences, University Hospital Malm, Lund University, Malm, Sweden
7 Department of Medicine, Helsinki University Central Hospital, Folkhalsan Genetic Institute, Folkhalsan Research Center, and Research Program for Molecular Medicine, University of Helsinki, Helsinki, Finland
8 University of Montreal Community Genomic Center, Chicoutimi Hospital, Quebec, Canada
9 McGill University and Genome Quebec Innovation Centre, Montreal, Canada
10 Genomics Collaborative Division, SeraCare LifeSciences, Cambridge, Massachusetts
11 Department of Genetics, Harvard Medical School, Boston, Massachusetts
12 Divisions of Genetics and Endocrinology, Children’s Hospital, Boston, Massachusetts
ABSTRACT
Protein tyrosine phosphatase (PTP)-1B, encoded by the PTPN1 gene, inactivates the insulin signal transduction cascade by dephosphorylating phosphotyrosine residues in insulin signaling molecules. Due to its chromosomal location under a chromosome 20 linkage peak and the metabolic effects of its absence in knockout mice, it is a candidate gene for type 2 diabetes. Recent studies have associated common sequence variants in PTPN1 with type 2 diabetes and diabetes-related phenotypes. We sought to replicate the association of common single nucleotide polymorphisms (SNPs) and haplotypes in PTPN1 with type 2 diabetes, fasting plasma glucose, and insulin sensitivity in a large collection of subjects. We assessed linkage disequilibrium, selected tag SNPs, and typed these markers in 3,347 cases of type 2 diabetes and 3,347 control subjects as well as 1,189 siblings discordant for type 2 diabetes. Despite power estimated at >95% to replicate the previously reported associations, no statistically significant evidence of association was observed between PTPN1 SNPs or common haplotypes with type 2 diabetes or with diabetic phenotypes.
The PTPN1 gene encodes protein tyrosine phosphatase (PTB)-1B (EC 3.3.3.48), an enzyme that binds the insulin receptor (1) and dephosphorylates phosphotyrosine residues of the activated receptor and the insulin receptor substrate 1, thus downregulating the insulin signaling cascade (2). Mice deficient for PTPN1 display enhanced insulin sensitivity and resistance to diet-induced obesity (3), as well as general leanness due to an increased basal metabolic rate (4). In vitro inhibition of PTP-1B by small molecules leads to greater insulin sensitivity (5,6). These functional features, coupled with its genomic location under the chromosome 20q13 type 2 diabetes linkage peak (7), support PTPN1 as a candidate gene that might harbor variants influencing susceptibility to insulin resistance and type 2 diabetes.
Several groups have examined genetic variants in the human PTPN1 gene for association with type 2 diabetes. A rare ProLeu change at position 387 was reportedly associated with type 2 diabetes in a Danish sample of 527 cases and 542 control subjects (nominal P = 0.036) (8) but not in a Chinese sample of 329 cases and 238 control subjects (9) or a Finnish sample of 257 cases and 285 control subjects (10). A common insertion of a guanosine at position 1484 of the 3' untranslated region (UTR) (1484insG) was associated with features of the metabolic syndrome in two Italian samples comprising an aggregate of 812 normoglycemic individuals (11) but was not associated with type 2 diabetes, BMI, fasting glucose, or fasting insulin in the Danish sample (8) or a large Swedish sample of 2,309 nondiabetic subjects (12).
Recently, a more extensive evaluation of noncoding genetic variants at this locus has been reported (13). In two nonoverlapping case/control samples totalling 610 and 475 individuals, Bento et al. (13) found that several single nucleotide polymorphisms (SNPs) were associated with type 2 diabetes, with odds ratios (ORs) in each sample near 1.3 and nominal P values <0.01. Examination of the patterns of linkage disequilibrium (LD) in this region revealed very limited haplotype diversity, with most SNPs falling within a single LD block (defined by the authors as the linear DNA segment where pairwise D' between SNPs is >0.8) that spans 100 kb from the promoter region to the 3' UTR. Of eight common haplotypes, two of them (GTCCTGTO and its reciprocal ACTTCAGO) capture 80% of haplotypic diversity in this locus; as predicted from the SNP results, either the former haplotype protected against type 2 diabetes and/or the latter haplotype conferred risk (P < 0.005) (13).
In a companion article (14), the same group evaluated these SNPs and haplotypes in PTPN1 for association with quantitative glycemic traits in an independent sample. Palmer et al. (14) found that the protective haplotype led to higher insulin sensitivity and lower fasting glucose under a recessive/multiplicative model in control individuals, whereas the risk haplotype had opposite effects under a dominant model (14). These results in a third nonoverlapping sample supported the conclusion that common variation in PTPN1 substantially contributed to type 2 diabetes risk and related traits.
As previously noted (15eC17), replication of genetic association studies with adequately powered samples is critical given the low prior probabilities for any given gene and to assess the robustness of findings in regard to specific aspects of study design and interpretation. In type 2 diabetes, consistently reproducible associations have been documented for the P12A polymorphism in the peroxisome proliferatoreCactivated receptor , the E23K polymorphism in the ATP-sensitive potassium channel Kir6.2, and, more recently, SNP-44 in CAPN10 (18). We set out to test the reproducibility of the association of PTPN1 with type 2 diabetes in a large collection of diabetes and control samples.
RESEARCH DESIGN AND METHODS
The diabetic sample subjects have been described elsewhere (19,20). Briefly, they comprise 1,189 Scandinavian siblings discordant for type 2 diabetes; a Scandinavian case-control sample totaling 942 subjects individually matched for age, BMI, and geographic region; a case-control sample from Sweden totaling 1,028 subjects who were individually matched for sex, age, and BMI; an individually matched case-control sample totaling 254 subjects from the Saguenay Lac-St. Jean region in Quebec; and two case-control samples individually matched for age, sex, and self-reported ethnicity from the U.S. (2,452 subjects) and Poland (2,018 subjects), both collected by Genomics Collaborative, Inc. (GCI).
Genotyping and phenotypic analysis.
Genotyping was performed by primer extension of multiplex products with detection by matrix-assisted laser desorption ionizationeCtime of flight mass spectroscopy (21) using a Sequenom platform. Our genotyping success rate was 97.3%, and our consensus rate (based on 11,799 duplicate genotypes) was 99.9%. The detailed list of SNPs and their genotyping assays are shown in online appendix Table 1 (available at http://diabetes.diabetesjournals.org). Genotype counts for the various samples tested in this study are shown in online appendix Table 2. This information is also posted on our website (http://genetics.mgh.harvard.edu/AltshulerWeb/publicationdata/Florez_PTPN1.html).
Plasma glucose (fasting and during an oral glucose tolerance test [OGTT]) was measured by a glucose oxidase method on a Beckman Glucose analyzer (Beckman Instruments, Fullerton, CA). Insulin was measured by radioimmunoassay. A 75-g OGTT was performed in a subset of the control Scandinavian subjects (n = 766). The whole-body insulin sensitivity index (ISI) was calculated as in Matsuda and DeFronzo (22), and because of nonnormality it was logarithmically transformed. Insulin resistance by homeostasis model assessment was calculated as in Matthews et al. (23).
Haplotype structure.
To evaluate the haplotype structure of the PTPN1 gene, we genotyped 42 SNPs in a multigenerational panel of Centre d’Etude du Polymorphisme Humain (CEPH) Caucasian pedigrees totaling 120 chromosomes selected by the HapMap project (24). SNPs were initially selected from genotypes already available from the HapMap and from variants previously studied by others; additional SNPs were added to refine areas of low SNP density or clarify the extent of LD. The list of SNPs included the eight SNPs that displayed the strongest nominal associations with type 2 diabetes in Bento et al. (13) (kindly provided by D.W. Bowden), 9 SNPs previously studied in this gene (8,10), and 16 SNPs previously chosen by the HapMap project; the rest were obtained from dbSNP.
These SNPs span 112 kb, from 28 kb upstream of the transcription start site to 10 kb downstream of the end of the PTPN1 3' UTR. Three of the SNPs attempted were technical failures (<75% genotyping percentage), 1 failed Hardy-Weinberg equilibrium, and 7 were found to be monomorphic in this population; thus, the final set comprised 31 (74%) working SNPs that were used to evaluate patterns of LD. The average spacing between these 31 SNPs is 3.6 kb. Haplotype blocks were determined as in Gabriel et al. (25). In the disease samples, case and control chromosomes were phased together using the expectation-maximization algorithm of Excoffier and Slatkin (26) modified to process larger data files using the partition-ligation approach as previously described (27). Each individual chromosome was assigned the phased haplotype of highest maximum likelihood.
Tag SNP selection.
To correlate our findings with those of the literature, we chose to genotype all eight SNPs provided by D.W. Bowden (rs941798, rs3787345, rs754118, rs2282147, rs718049, rs718050, rs3787348, and 1484insG) and the two missense SNPs (P387L and G381S) in our disease panels (T420M was monomorphic in our reference panel). We selected additional tag SNPs by a newly developed algorithm named Tagger (Paul I.W. de Bakker et al., unpublished observations; http://www.broad.mit.edu/mpg/tagger/), which selects single markers and multimarker combinations of tags based on pairwise LD between markers (set at minimum r2 > 0.8). This method allows the user to "force in" markers of interest, such that the additional tags selected capture those markers not already tagged by the forced-in set. This procedure resulted in 18 tag SNPs (the above 10 plus 8 additional markers; Figs. 1 and Table 1); if no SNPs had been forced in a priori, Tagger would have yielded 12 tag SNPs. The 18 tag SNPs were validated by an independent method (28) and were found to be robust predictors of all single SNPs (minimum RS2 > 0.95) and haplotypes (minimum Rh2 > 0.85).
Statistical analysis.
Power calculations were performed with the program of Purcell et al. (29), available at http://statgen.iop.kcl.ac.uk/gpc. To examine the association of SNPs and haplotypes with type 2 diabetes, we used simple 2 analysis in the case-control samples and the discordant allele test (30) in the sibpairs. Haplotypes were compared individually against all others; only haplotypes with frequencies >5% in the reference panel were examined. Results from the various samples were combined by Mantel-Haenszel meta-analysis of the ORs (16). Homogeneity among study samples was tested using a Pearson 2 goodness-of-fit test as previously described (16).
Phenotype comparisons.
We concluded that under the most parsimonious model, the positive findings obtained by Bento et al. (13) and Palmer et al. (14) could be reduced to two statistically indistinguishable and complementary results: the association of one high-frequency PTPN1 haplotype with risk of type 2 diabetes (and related variables of fasting glucose and insulin sensitivity) and the converse result of protection from disease and related quantitative traits by another high-frequency reciprocal haplotype. We therefore formulated our study as a test of a single hypothesis. In secondary analyses, we tested this putative association under both a neutral, multiplicative genetic model and the genotypic model reported in Palmer et al. (14). Only nondiabetic subjects were included in our phenotypic analyses. Comparisons of phenotypic variables between genotypic groups were made by Student’s t test (two groups) or ANOVA (more than two groups).
RESULTS
Characterization of common sequence variation at PTPN1.
Consistent with the results of Bento et al. (13), genotyping of 31 SNPs defined a single segment of high LD ("haplotype block") spanning 83 kb (between SNPs rs6063528 and rs3787348; Fig. 1); due to their low minor allele frequency, the two missense SNPs G381S and P387L were outside the boundaries of this block as defined by Gabriel et al. (25). The block comprises five haplotypes of >5% frequency, which together capture 90% of haplotypic diversity at this locus. Of note, genotyping additional SNPs has permitted us to further refine the haplotype structure, such that the risk haplotype of 33eC34% frequency in Bento et al. (13) is subdivided into haplotypes B and D in our reference panel, with frequencies of 22 and 6%, respectively, in the CEPH reference sample. (In the GCI U.S. sample, these haplotypes are present at 26 and 9% frequencies in control subjects, in agreement with the control samples of Bento et al. [13].) The 31 working SNPs are well captured by a set of 18 tagging markers, which include both missense SNPs and those previously studied by Bento et al. (13), and which were genotyped in this study (see RESEARCH DESIGN AND METHODS).
Power calculations.
To estimate the sample size required to replicate the association of the high-frequency haplotype in Bento et al. (13) with protection from type 2 diabetes, we assumed a haplotype frequency of 45%, a type 2 diabetes disease prevalence of 8%, and a genotypic relative risk of 1.3 under a recessive model. Under these parameters, we estimated that our combined sample of 3,347 case/control pairs and 1,189 discordant sibs would provide >95% power to reject the null hypothesis of no association at P < 0.05. If we reduce the genotypic relative risk to 1.2 in order to allow for the likely overestimate of the initial report due to the "winner’s curse," we estimate power at >75% to reject the null hypothesis of no association at P < 0.05; this power is raised to >80% under a multiplicative model. Similarly, we estimated that the size of our phenotypic sample provides 90% power to detect a 2-mg/dl difference in fasting plasma glucose under the genotypic model proposed by Palmer et al. (14).
Association study.
Association results for each of the 18 SNPs and five common haplotypes are presented in Table 1. No association was observed to any of the SNPs or haplotypes spanning PTPN1 with type 2 diabetes in the overall sample. Results that were nominally significant in one of the three subsamples (rs6020546, rs2230605, and P387L) were not reproduced in the other two subsamples or displayed a trend in the opposite direction. No heterogeneity among subsamples was observed.
In an attempt to understand the differences between our results and those previously published, we examined the frequencies of the two major haplotypes described by Bento et al. (13) in the various samples studied in this and in previous reports. We note that the frequencies of haplotype A and the aggregate of haplotypes B and D are quite similar in all the control samples (from both this study and that of Bento et al. [13]) and in the diabetic samples studied in this report. The two samples showing a different haplotypic frequency are the case series in Bento et al. (13) (Table 2).
Genotype-phenotype correlations.
Since the single SNP results of Palmer et al. (14) could be explained by the effects of a single high-frequency haplotype (and/or the converse effect of an alternate, high-frequency haplotype), we confined analysis of associations to diabetic subphenotypes to this single hypothesis. We evaluated the effects of the presence of haplotype A (the putative protective haplotype) or haplotype B (the major contributor to the putative risk haplotype) on the two measures of glucose homeostasis found by Palmer et al. to be significantly affected. We measured fasting plasma glucose in control individuals from both the GCI U.S. and the Scandinavian samples, and the whole-body ISI in control individuals from the Scandinavian sample. We examined both parameters under a neutral genotypic model (comparing carriers of haplotype A [A/A or A/X] with carriers of haplotype B [B/B or B/X] and excluding A/B heterozygotes from the analysis) and under the genotypic model thought to be most consistent with the findings of Palmer et al. (comparing A/A homozygotes with carriers of haplotype B, including A/B heterozygotes [B/B, A/B, or B/X]).
Taking the entire sample as a whole, we did not find significant evidence for association of PTPN1 haplotypes with either measure of glucose metabolism. In agreement with Palmer et al. (14), in the GCI U.S. sample, fasting plasma glucose was elevated in haplotype B carriers under the neutral model (A carriers vs. B carriers: 90.2 ± 14.4 vs. 94.1 ± 14.1 mg/dl [mean ± SD], one-sided P = 0.015); this difference was not statistically significant under the recessive genetic model of Palmer et al. (A/A homozygotes vs. B carriers: 91.5 ± 16.8 vs. 92.7 ± 14.6 mg/dl, one-sided P = 0.26). In the Scandinavian sample, however, fasting plasma glucose was not significantly different according to haplotype status (A carriers vs. B carriers: 95.0 ± 11.5 vs. 93.7 ± 10.9 mg/dl; A/A homozygotes vs. B carriers: 95.1 ± 11.1 vs. 94.0 ± 11.1 mg/dl).
Because the GCI case/control samples were not but the Scandinavian samples were matched for BMI, we developed the post hoc hypothesis that differences in patient phenotypes might contribute to these divergent results. There was a nominally significant difference in the average BMI of control individuals between the Genomics Collaborative U.S. and Scandinavian samples (27.2 ± 4.9 vs. 26.7 ± 3.7 kg/m2, respectively, two-sided P < 0.02). Moreover, the observed difference in fasting plasma glucose between carriers of the A and B haplotypes in the Genomics Collaborative U.S. sample was most obvious for the highest quartile of BMI under the neutral model (P < 0.0001; Table 3); the difference was also present under the recessive model (P < 0.003; Table 3). In the Scandinavian sample, although there was no statistically significant difference in fasting plasma glucose according to genotypic group across each BMI quartile, fasting plasma glucose rose across BMI quartiles in carriers of the B haplotype (P < 0.02; Table 3).
Because of the difference in fasting plasma glucose between carriers of haplotypes A and B in the most obese subjects of the GCI U.S. sample, we wondered whether an association of either haplotype with type 2 diabetes might be present in this group. When the association analysis was restricted to the 325 control individuals and 816 diabetic subjects with BMI >29.5 kg/m2, no association was observed (haplotype A: OR 1.02 [95% CI 0.82eC1.18], two-sided P = 0.84; haplotype B: 0.96 [0.85eC1.28], P = 0.70). In addition, we detected no effect of genotype on BMI (data not shown).
We explored whether insulin resistance, as measured by the whole-body ISI, might be affected in the 766 control Scandinavian subjects for whom we had OGTT data. No statistically significant difference in insulin sensitivity was found among genotypic groups according to presence of either haplotype A or B (online appendix Table 3). In contrast to the results of Palmer et al. (14), a trend toward greater insulin sensitivity was found among carriers of haplotype B under the neutral model (two-sided P = 0.057; online appendix Table 3).
Finally, we examined whether the specific 1484insG variant, previously reported to induce insulin resistance (11), affected similar glycemic measures in our control subjects. We detected no statistically significant difference between nondiabetic Scandinavian carriers and noncarriers of this polymorphism in fasting glucose (94.4 ± 10.9 vs. 94.9 ± 11.2 mg/dl [mean ± SD]), fasting insulin (8.04 ± 4.9 vs. 8.03 ± 4.7 mU/l) or insulin resistance by homeostatis model assessment (1.94 ± 1.2 vs. 1.95 ± 1.2). Similarly, fasting glucose was not significantly different between 1484insG carriers and noncarriers in the GCI U.S. samples (91.9 ± 14.9 vs. 91.1 ± 14.4 mg/dl). Nondiabetic Scandinavian siblings discordant for genotype at this locus also failed to show any statistically significant differences in these three parameters (data not shown).
DISCUSSION
We set out to test the previous hypothesis of association of SNPs and haplotypes in PTPN1 with type 2 diabetes (13) and with fasting plasma glucose and insulin sensitivity (14). We confirmed the extensive LD among common variants in the PTPN1 gene (13) but failed to detect an association of any SNP or common haplotype with type 2 diabetes in either our individual samples or their combination by meta-analysis.
Lack of replication can have several possible explanations. First, it is possible that the initial association may have been a statistical fluctuation. This seems less likely since the authors replicated their findings in a second population, and two measures of glucose homeostasis varied according to PTPN1 haplotype in the predicted direction (14). To evaluate the significance of such a finding it will be useful to simulate the frequency, in the absence of a true association, of obtaining such P values across multiple studies after examining many SNPs and haplotypes in each of several phenotypes (type 2 diabetes and related subphenotypes).
A second possibility is that the current study constitutes a false-negative attempt at replication. In the setting of a true initial association, failure of replication can be due to 1) insufficient power (16), 2) heterogeneity among subsamples, and/or 3) heterogeneity within subsamples that renders them inadequate for case/control analysis. Although our study was well powered to replicate the model of Bento et al. (13), it is possible that the initial association may have been overestimated due to the "winner’s curse." Regardless, the best estimate of the OR is bounded by the 95% CIs of all available data. With respect to genetic heterogeneity among subsamples, formal tests for heterogeneity were negative in this and previous studies that have analyzed these samples (19,20,27). Finally, with respect to within-sample heterogeneity, we note that these samples have shown consistent results when genotyped for the two most widely reproduced associations with type 2 diabetes, P12A in peroxisome proliferatoreCactivated receptor (31 and N.B., J.C.F., J.N.H., D.A., unpublished observations) and E23K in Kir6.2 (19,27).
A third possibility is that the association signal reported by Bento et al. (13) was valid, but there is heterogeneity among the various populations examined 1) in the extent of LD in this region, 2) in the phenotypic characteristics of patients studied, or 3) in the frequencies of genetic or environmental modifiers. In regard to LD, the LD pattern in PTPN1 observed in our reference panel and disease samples is nearly identical to that observed by Bento et al. (13). With regard to phenotypic characterization, our case and control subjects were matched for age, sex, self-reported ethnicity (in all cases), and BMI (in the Scandinavian and Canadian subsamples). In contrast, the first case/control population of Bento et al. (13) consisted of cases with both type 2 diabetes and end-stage renal disease (mean BMI 28.5 kg/m2) who were matched to nondiabetic control subjects; the second case/control population of Bento et al. (13) consisted of relatively obese diabetic subjects from the Diabetes Heart Study (mean BMI 33.7 kg/m2), matched to a reference sample of nondiabetic control subjects obtained from a publicly available repository. The sample of Palmer et al. (14) consisted of nondiabetic Hispanic subjects (mean BMI 29.2 kg/m2). Since the definition of case status and the methods to match controls differs among studies, it is possible that one of these considerations and/or population stratification may account for the disparate results (32). In addition, we note that in one of our phenotypic analyses, the whole-body ISI derived from the OGTT is not a perfect correlate of the ISI obtained from the intravenous glucose tolerance test employed by Palmer et al. (14).
Heterogeneity in gene-gene and/or gene-environment interactions could also contribute to divergent findings. We observed an association of a subset of the risk haplotype of Palmer et al. (14) (haplotype B in our study) with fasting plasma glucose in the GCI U.S. subsample, albeit confined in a post hoc analysis to the quartile with the highest BMI. The same effect was not seen in the Scandinavian sample, although fasting plasma glucose rose with BMI in Scandinavian carriers of the B haplotype. Because the Scandinavian subjects in the highest quartile of BMI were leaner than their GCI U.S. counterparts (BMI 31.58 ± 2.54 vs. 33.74 ± 3.80 kg/m2, respectively, P < 10eC10; Table 3), one might speculate that in Scandinavian subjects of an even heavier BMI, fasting plasma glucose may also have been higher in carriers of the B haplotype than in carriers of the A haplotype. But since these observations were post hoc, they should be considered an exploratory, hypothesis-generating exercise.
Thus, our different findings could be explained by the presence of unaccounted modifier genes in the various populations that affect diabetes-related phenotypes. In this scenario, variants that are enriched in individuals of Hispanic ancestry or genes that influence renal function or obesity might interact with PTPN1 to induce insulin resistance and raise fasting plasma glucose. If the effect is large enough, an association with type 2 diabetes might be seen in a subset of samples but not in others. Given these multiple explanations, further refinement of patient phenotypes, matching for all possible confounders and most importantly replication in additional well-powered samples, will be necessary to evaluate the hypothesized effect of PTPN1 on type 2 diabetes and related traits.
ACKNOWLEDGMENTS
J.C.F. is supported by National Institutes of Health Research Career Award 1 K23 DK65978-01. J.N.H. is a recipient of a Burroughs Wellcome Career Award in Biomedical Sciences. D.A. is a Charles E. Culpeper Scholar of the Rockefeller Brothers Fund and a Burroughs Wellcome Fund Clinical Scholar in Translational Research, the latter of which supported this work. D.A., M.J.D., and J.N.H. are recipients of The Richard and Susan Smith Family Foundation/American Diabetes Association Pinnacle Program Project Award. L.G., T.T., and the Botnia Study are principally supported by the Sigrid Juselius Foundation, the Academy of Finland, the Finnish Diabetes Research Foundation, The Folkhalsan Research Foundation, EC (BM4-CT95-0662, GIFT), the Swedish Medical Research Council, the JDF Wallenberg Foundation, and the Novo Nordisk Foundation.
We thank D.W. Bowden for sharing experimental protocols and unpublished data, the Botnia research team for clinical contributions, and the members of the Altshuler, Hirschhorn, Daly, and Groop labs for helpful discussions.
FOOTNOTES
L.G. and D.A. jointly supervised this work.
L.G. has received consulting fees from and served on advisory panels for Aventis-Sanofi, Bristol-Myers Squibb, Kowa, and Roche. D.A. has received consulting fees from and served on an advisory panel for GCI.
Additional information for this article can be found in an online appendix at http://diabetes.diabetesjournals.org.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
CEPH, Centre d’Etude du Polymorphisme Humain; GCI, Genomics Collaborative, Inc; ISI, insulin sensitivity index; LD, linkage disequilibrium; OGTT, oral glucose tolerance test; PTP, protein tyrosine phosphatase; SNP, single nucleotide polymorphism; UTR, untranslated region
REFERENCES
Seely BL, Staubs PA, Reichart DR, Berhanu P, Milarski KL, Saltiel AR, Kusari J, Olefsky JM: Protein tyrosine phosphatase 1B interacts with the activated insulin receptor. Diabetes 45: 1379eC1385, 1996
Goldstein BJ: Protein-tyrosine phosphatases: emerging targets for therapeutic intervention in type 2 diabetes and related states of insulin resistance. J Clin Endocrinol Metab 87: 2474eC2480, 2002
Elchebly M, Payette P, Michaliszyn E, Cromlish W, Collins S, Loy AL, Normandin D, Cheng A, Himms-Hagen J, Chan C-C, Ramachandran C, Gresser MJ, Tremblay ML, Kennedy BP: Increased insulin sensitivity and obesity resistance in mice lacking the protein tyrosine phosphatase-1B gene. Science 283: 1544eC1548, 1999
Klaman LD, Boss O, Peroni OD, Kim JK, Martino JL, Zabolotny JM, Moghal N, Lubkin M, Kim Y-B, Sharpe AH, Stricker-Krongrad A, Shulman GI, Neel BG, Kahn BB: Increased energy expenditure, decreased adiposity, and tissue-specific insulin sensitivity in protein-tyrosine phosphatase 1B-deficient mice. Mol Cell Biol 20: 5479eC5489, 2000
Xie L, Lee SY, Andersen JN, Waters S, Shen K, Guo XL, Moller NP, Olefsky JM, Lawrence DS, Zhang ZY: Cellular effects of small molecule PTP-1B inhibitors on insulin signaling. Biochemistry 42: 12792eC12804, 2003
Wiesmann C, Barr KJ, Kung J, Zhu J, Erlanson DA, Shen W, Fahr BJ, Zhong M, Taylor L, Randal M, McDowell RS, Hansen SK: Allosteric inhibition of protein tyrosine phosphatase 1B. Nat Struct Mol Biol 11: 730eC737, 2004
Florez JC, Hirschhorn JN, Altshuler D: The inherited basis of diabetes mellitus: implications for the genetic analysis of complex traits. Annu Rev Genomics Hum Genet 4: 257eC291, 2003
Echwald SM, Bach H, Vestergaard H, Richelsen B, Kristensen K, Drivsholm T, Borch-Johnsen K, Hansen T, Pedersen O: A P387L variant in protein tyrosine phosphatase-1B (PTP-1B) is associated with type 2 diabetes and impaired serine phosphorylation of PTP-1B in vitro. Diabetes 51: 1eC6, 2002
Weng J, Yan J, Huang Z, Sui Y, Xiu L: Missense mutation of Pro387Leu in protein tyrosine phosphatase-1B (PTP-1B) is not associated with type 2 diabetes in a Chinese Han population (Letter). Diabetes Care 26: 2957, 2003
Santaniemi M, Ukkola O, Kesniemi YA: Tyrosine phosphatase 1B and leptin receptor genes and their interaction in type 2 diabetes. J Intern Med 256: 48eC55, 2004
Di Paola R, Frittitta L, Miscio G, Bozzali M, Baratta R, Centra M, Spampinato D, Santagati MG, Ercolino T, Cisternino C, Soccio T, Mastroianno S, Tassi V, Almgren P, Pizzuti A, Vigneri R, Trischitta V: A variation in 3' UTR of hPTP1B increases specific gene expression and associates with insulin resistance. Am J Hum Genet 70: 806eC812, 2002
Dahlman I, Wahrenberg H, Persson L, Arner P: No association of reported functional protein tyrosine phosphatase 1B 3' UTR gene polymorphism with features of the metabolic syndrome in a Swedish population. J Intern Med 255: 694eC695, 2004
Bento JL, Palmer ND, Mychaleckyj JC, Lange LA, Langefeld CD, Rich SS, Freedman BI, Bowden DW: Association of protein tyrosine phosphatase 1B gene polymorphisms with type 2 diabetes. Diabetes 53: 3007eC3012, 2004
Palmer ND, Bento JL, Mychaleckyj JC, Langefeld CD, Campbell JK, Norris JM, Haffner SM, Bergman RN, Bowden DW: Association of protein tyrosine phosphatase 1B gene polymorphisms with measures of glucose homeostasis in Hispanic Americans: the Insulin Resistance Atherosclerosis Study (IRAS) Family Study. Diabetes 53: 3013eC3019, 2004
Hirschhorn JN, Lohmueller K, Byrne E, Hirschhorn K: A comprehensive review of genetic association studies. Genet Med 4: 45eC61, 2002
Lohmueller K, Pearce CL, Pike M, Lander ES, Hirschhorn JN: Meta-analysis of genetic association studies supports a contribution of common variants to susceptibility to common disease. Nat Genet 33: 177eC182, 2003
Florez JC: Phenotypic consequences of the peroxisome proliferator-activated receptor- Pro12Ala polymorphism: the weight of the evidence in genetic association studies. J Clin Endocrinol Metab 89: 4234eC4237, 2004
O’Rahilly S, Barroso I, Wareham NJ: Genetic factors in type 2 diabetes: the end of the beginning Science 307: 370eC373, 2005
Florez JC, Sjogren M, Burtt N, Orho-Melander M, Schayer S, Sun M, Almgren P, Lindblad U, Tuomi T, Gaudet D, Hudson TJ, Daly MJ, Ardlie KG, Hirschhorn JN, Altshuler D, Groop L: Association testing in 9,000 people fails to confirm the association of the insulin receptor substrate-1 G972R polymorphism with type 2 diabetes. Diabetes 53: 3313eC3318, 2004
Winckler W, Graham RR, de Bakker PIW, Sun M, Almgren P, Tuomi T, Gaudet D, Hudson TJ, Ardlie KG, Daly MJ, Hirschhorn JN, Groop L, Altshuler D: Association testing of variants in the hepatocyte nuclear factor 4 gene with risk of type 2 diabetes in 7,883 people. Diabetes 54: 886eC892, 2005
Tang K, Fu DJ, Julien D, Braun A, Cantor CR, Koster H: Chip-based genotyping by mass spectrometry. Proc Natl Acad Sci U S A 96: 10016eC10020, 1999
Matsuda M, DeFronzo R: Insulin sensitivity indices obtained from oral glucose tolerance testing: comparison with the euglycemic insulin clamp. Diabetes Care 22: 1462eC1470, 1999
Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC: Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 28: 412eC419, 1985
The International HapMap Consortium: The International HapMap Project. Nature 426: 789eC796, 2003
Gabriel SB, Schaffner SF, Nguyen H, Moore JM, Roy J, Blumenstiel B, Higgins J, DeFelice M, Lochner A, Faggart M, Liu-Cordero SN, Rotimi C, Adeyemo A, Cooper R, Ward R, Lander ES, Daly MJ, Altshuler D: The structure of haplotype blocks in the human genome. Science 296: 2225eC2229, 2002
Excoffier L, Slatkin M: Maximum-likelihood estimation of molecular haplotype frequencies in a diploid population. Mol Biol Evol 12: 921eC927, 1995
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: Haplotype structure and genotype-phenotype correlations of the sulfonylurea receptor and the islet ATP-sensitive potassium channel gene region. Diabetes 53: 1360eC1368, 2004
Stram DO, Haiman CA, Hirschhorn JN, Altshuler D, Kolonel LN, Henderson BE, Pike MC: Choosing haplotype-tagging SNPS based on unphased genotype data using a preliminary sample of unrelated subjects with an example from the Multiethnic Cohort Study. Hum Hered 55: 27eC36, 2003
Purcell S, Cherny SS, Sham PC: Genetic power calculator: design of linkage and association genetic mapping studies of complex traits. Bioinformatics 19: 149eC150, 2003
Boehnke M, Langefeld CD: Genetic association mapping based on discordant sib pairs: the discordant-alleles test. Am J Hum Genet 62: 950eC961, 1998
Ardlie KG, Lunetta KL, Seielstad M: Testing for population subdivision and association in four case-control studies. Am J Hum Genet 71: 1478eC1480, 2002
Freedman ML, Reich D, Penney KL, McDonald GJ, Mignault AA, Patterson N, Gabriel SB, Topol EJ, Smoller JW, Pato CN, Pato MT, Petryshen TL, Kolonel LN, Lander ES, Sklar P, Henderson B, Hirschhorn JN, Altshuler D: Assessing the impact of population stratification on genetic association studies. Nat Genet 36: 388eC393, 2004
Lewontin RC: On measures of gametic disequilibrium. Genetics 120: 849eC852, 1988(Jose C. Florez, Christina)