Methods and Procedures

Study subjects

The GENNID study (Genetics of NIDDM) is a multicenter study established by the American Diabetes Association in 1993 and comprises a comprehensive, well-characterized resource of T2D families from four ethnic groups (whites, Mexican Americans, African Americans, and Japanese Americans). The study protocol and ascertainment scheme has been described in detail (18,19). Data were collected by the GENNID study group using standardized methods and included a medical history questionnaire, a family history questionnaire, a physical exam, fasting blood samples, laboratory data, and a diabetes history form.

For this analysis, data on age, gender, and self-reported diabetes was obtained from the medical history questionnaire; height, weight, waist circumference, and systolic and diastolic blood pressure were obtained from the physical examination; and glucose, insulin, high-density lipoprotein (HDL), and triglycerides were obtained from the laboratory data. Clinical MetS was defined by the National Cholesterol Education Program Adult Treatment Panel III guidelines (20).

Genetic markers and genotyping

A 10-cM genome-wide linkage scan using standard sets of highly informative short-tandem-repeat polymorphisms (screening set 6 for Mexican Americans and screening set 8 for all other groups) were performed at the Marshfield Medical Research Foundation (21) as part of the GENNID study. Additional markers were genotyped by GlaxoWellcome (19), to fill in gaps in the genome for a total of 389–395 markers, averaging 8.2–9.0 cM apart and with an average marker heterozygozity of 0.75. Genotype and MetS phenotype data are available for n = 69 white, n = 53 Mexican American, n = 65 African American, and n = 15 Japanese-American extended kindreds.

Statistical analysis

Mendelian inconsistencies. All genetic markers were checked for Mendelian consistency using PEDCHECK (http://watson.hgen.pitt.edu/register/docs/pedcheck.html) (22). A total of 14 white, 22 African-American, 1 Mexican-American, and 3 Japanese-American individuals were identified as Mendelian inconsistent (>100 marker errors) and were deleted from their respective family pedigree. An additional 56 white, 52 African-American, 11 Japanese-American, and 26 Mexican-American family members had errors for <25 markers. For these latter individuals, genotype information for the inconsistent markers was excluded. We also excluded two white individuals with genetic data inconsistent with their reported gender and one individual from each of three monzygotic twin pairs.

Principal component factor analysis. Eight variables were defined as core MetS "features" and included in the PCFA: triglyceride (mg/dl), HDL cholesterol (mg/dl), fasting glucose levels (mg/dl), fasting insulin levels (mg/dl), systolic blood pressure (mm Hg), diastolic blood pressure (mm Hg), waist circumference (cm) and weight (kg). Values for waist circumference, triglyceride, weight, fasting insulin, and fasting glucose were skewed and therefore log-transformed prior to analysis. Subjects with values >4 s.d. from the mean trait values were excluded to improve normality (n = 4 for systolic blood pressure, n = 2 for HDL, n = 1 for diastolic blood pressure, and n = 1 for triglycerides). After these exclusions, all the individual variables were adjusted for age and sex using linear regression, and the residuals were included in PCFA for those individuals with complete MetS data, age, and gender. Briefly, PCFA was performed separately in each ethnic group using Intercooled STATA 8.2 for Windows (StataCorp, College Station, TX) (6). Principal component analysis was used to extract the initial components as linear combinations of the variables included in the analysis. These components were rotated using a varimax orthogonal rotation and interpreted by considering only those variables with a factor loading |0.40| (equivalent to sharing 15% of the variance with the factor). Finally, factor scores for each of the rotated components were calculated as weighted sums of the individual MetS features (23). The factor scores are used as the primary quantitative traits reflecting the MetS in the variance components linkage analysis and will be referred to as MetS factors in the following sections.

Variance components linkage analysis. Heritability, h², was estimated for each MetS factor under a polygenic model adjusted for age, sex, and self-reported diabetes as implemented in SOLAR (http://www.sfbr.org/solar/) (24). All marker allele frequencies were estimated with a maximum likelihood method assuming Mendelian inheritance using the software package SOLAR (25). Exact conditional probabilities were computed every 1 cM using the Lander–Green algorithm as implemented in the software program MERLIN (http://www.sph.umich.edu/csg/abecasis/merlin/index.html) (26). Variance components linkage analysis was implemented in SOLAR as described in Almasy et al. (27). An LOD score of 1.9 is taken as suggestive evidence for linkage and an LOD score of 3.3 as evidence for significant linkage (28). For those regions with suggestive evidence for linkage to the MetS factors, variance components linkage analysis was also performed for the individual features.

Because LOD scores may be biased when assumptions of multivariate normality are not met, we also present empirical P values. Empirical P values were calculated at the maximum LOD score for each MetS factor using the "lodadj" and "empp" commands in SOLAR. The empirical P value (empp) is defined as the probability of obtaining the observed LOD score or greater from a simulated distribution of LOD scores. For each of the MetS factors, 100,000 simulations were run under the null hypothesis of no linkage with a fully informative marker (24,29). Multiple testing was accounted for by applying a Bonferroni correction to adjust the empirical P values for the MetS factors (empp 3) and individual features (empp 8), referred to as the adjusted empirical P value (adjusted empp).

Results

The characteristics of the GENNID families included in the genome-wide scan are presented in Table 1. Average family sizes ranged from 3.9 (African Americans) to 8.2 (Japanese Americans) family members. Japanese-American family members tended to be older and Mexican-American family members younger than the other groups. Prevalence of self-reported T2D and clinical MetS also varied by ethnic group. The highest prevalence of self-reported T2D was among African Americans, and the highest prevalence of clinical MetS was observed among the Mexican-American sample. The prevalence of both conditions was lowest among the Japanese Americans.

Table 1 - Characteristics of GENNID individuals stratified by ethnic group.

Full table (48K)

Factor analysis

The results of the PCFA are provided in Table 2 for each ethnic group. Three factors were retained, and exhibited similar factor loading patterns across the four ethnic groups. In all groups, the first factor is interpreted as a weight/waist factor ("wtfactor") due to the strong positive loadings for weight and waist circumference. Fasting insulin was also shown to load on the wtfactor in all groups. In the Japanese-American sample, HDL also loaded on the wtfactor. The second factor is interpreted as a blood pressure factor ("bpfactor") due to the strong loadings for systolic and diastolic blood pressure, with very consistent factor loadings across the four ethnic groups. The third factor is interpreted as a lipid factor ("lipidfactor") due to the strong factor loadings for HDL and triglycerides across all ethnic groups. In addition, fasting glucose and fasting insulin also load on this factor in all groups. Together these factors account for 70% of the total variance in the data for each group (Table 2).

Table 2 - Factors and factor loadings^a from principal components factor analysis, by GENNID ethnic groups.

Full table (59K)

Heritability and standard error estimates for the MetS factors after adjustment for age, gender, and diabetes status are also presented in Table 2 by ethnic group. In general, the three MetS factors were highly heritable across all ethnic groups. The residual heritabilities were significant for each of the three MetS factors in each ethnic group, except the bpfactor in Japanese Americans (Table 2). Covariates accounted for a small proportion of the trait variance: 1–6% of the wtfactor, 2–15% of the lipidfactor, and 1–8% of the bpfactor across ethnic groups (data not shown).

Linkage analysis

The strongest linkage signal for each ethnic group is shown in Figure 1. Note that Figure 1 shows -log₁₀ (adjusted empp) rather than LOD scores on the y-axis. On this scale, a "suggestive" linkage signal is represented by a -log₁₀ (adjusted empp) value of 2.77, which is equivalent to a P value of 0.0017 or an LOD score = 1.9 (28). Supplementary Table S1 online provides detailed results from the genome-wide linkage scan by ethnic group, including LOD scores, adjusted empp and chromosomal location for regions with an LOD score >1.90, along with the LOD scores and adjusted empp for the individual features that comprise each feature. The adjusted empp allows for a more direct comparison of the strength of the linkage signal between the factors and the individual features.

Figure 1.

Selected linkage results by ethnic group. Panels (a–d) show linkage results of chromosomes with the maximum genome-wide LOD score within each ethnic group: white, African Americans, Mexican Americans, and Japanese Americans, respectively. On the x-axis the centiMorgan (cM) location is plotted for a given chromosome, and on the y-axis, the -log₁₀ (empirical P value adjusted for multiple traits) is shown; the maximum -log₁₀(adjusted empirical P value) corresponds to the maximum LOD score. Each trait is specified by a line type noted in each subplot legend box.

Full figure and legend (19K)

In the white sample, two suggestive linkage peaks for the wtfactor were identified. The strongest overall finding in our study was observed on chromosome 2 at 122 cM (1-LOD support interval of 117–127 cM) with an LOD score of 2.88 (adjusted empp = 0.00048) (Figure 1a). The second signal was identified on chromosome 1 at 157 cM with an LOD score of 2.43 (1-LOD support interval of 151–164 cM, adjusted empp = 0.00162). Interestingly, log-weight and log-waist also showed suggestive linkage to the same location on chromosome 2 (adjusted empp <0.0017), while only log-weight showed suggestive linkage to the wtfactor peak on chromosome 1 (See Supplementary Table S1 online).

Among the African-American families five linkage peaks with LOD >1.9 were observed, all showing suggestive evidence for linkage to the bpfactor. These peaks were observed on chromosomes 1, 11, 14, 16, and 18 (Supplementary Table S1 online). The strongest evidence for linkage (LOD = 2.43; adjusted empp = 0.0032) was observed on chromosome 18 (85 cM; 1-LOD support interval of 81–91 cM) (Figure 1b), followed by a signal on chromosome 16 (32 cM, 1-LOD support interval of 14–41 cM) with an LOD score of 2.25 (adjusted empp = 0.0054). The LOD scores for the composite bpfactor were greater than the maximum LOD scores under this linkage peak for either systolic or diastolic blood pressure for chromosomes 1, 11, 16, and 18. In contrast, the LOD score on chromosome 14 was larger for systolic blood pressure (LOD > 2.2) than for the composite bpfactor.

Results for the Mexican-American families indicate evidence for linkage on chromosomes 3, 14, and 15 (Supplementary Table S1 online). The strongest evidence for linkage (LOD = 2.67; adjusted empp = 0.0012) was observed on chromosome 3 (206 cM; 1-LOD support interval of 190–223 cM) for the bpfactor (Figure 1c), followed by a signal (LOD = 2.41; adjusted empp = 0.0026) on chromosome 15 (25 cM; 1-LOD support interval of 17–33 cM) for the bpfactor. A suggestive signal (LOD = 1.90, adjusted empp = 0.0016) for the lipidfactor was observed on chromosome 14 (65 cM; 1-LOD support interval of 45 –79 cM).

Finally, results for the Japanese-American families indicate evidence for linkage on chromosomes 5 and 16 with the wtfactor and the lipidfactor, respectively (Supplementary Table S1 online). The strongest evidence (LOD = 2.33; adjusted empp = 0.0033) for linkage was observed on chromosome 16 (75 cM; 1-LOD support interval of 68–81 cM) with the lipidfactor (Figure 1d). A more modest signal (LOD = 1.94, adjusted empp = 0.0032) on chromosome 5 (166 cM; 1-LOD support interval of 152–185 cM) was identified for the wtfactor. Suggestive signals for waist and HDL (LOD > 1.90) were observed for these locations on chromosomes 5 and 16, respectively. The generally modest adjusted empirical P values reflect the smaller sample size of this group.

Discussion

In this study, three quantitative MetS factors were identified: (i) a weight/waist factor, (ii) a blood pressure factor, and (iii) a lipid factor. Results from this analysis indicate that the factors and factor loadings identified by PCFA are consistent across the four ethnic groups when the same set of risk factors is included in the PCFA. Evidence for linkage to each of these factors was observed, although the linkage signals varied across the four ethnic groups. These results suggest heterogeneity in MetS susceptibility.

Overall, the strongest evidence for linkage in this study was observed on chromosome 2 for the wtfactor phenotype among the white sample and on chromosome 3 for the bpfactor in the Mexican-American sample. The region identified on chromosome 2 (2q12.1–2q13) includes a candidate gene (INSIG2) previously identified by the Framingham group as influencing adult and childhood obesity on chromosome 2q14 (30). Although, the INSIG2 obesity association has not been consistently replicated (31,32,33,34,35), the correspondence of this candidate gene for obesity and our linkage finding is intriguing. The region on chromosome 3 corresponds to a region (3q27) that has been identified as harboring a MetS susceptibility locus (36). There are several interesting candidate genes in this region, including adiponectin (ADIPOQ), which over the past 10 years has been established as a key player in the processes that affect the clustering of MetS phenotypes (37,38). Additional work is needed to fine-map the regions identified in this study and to further evaluate INSIG2, ADIPOQ, and other genes in these regions as potential MetS susceptibility genes.

Several other groups have performed genome-wide linkage analysis on MetS phenotypes defined by PCFA and have reported evidence for genetic influences on these phenotypes. Studies have been conducted in Mexican-American families (7,11), white families (8,10), African-American families (12), Native Americans (14), and multiethnic samples stratified by race/ethnicity (9,15,16). In general, there are very few overlaps in reported signals across these studies, and none of the previous suggestive or significant linkage results for composite MetS phenotypes overlaps with regions identified in our study. However, the results of these studies are difficult to directly compare. In particular, while all previous studies have included features related to obesity, glucose intolerance, lipids, and blood pressure, there has been variation in the individual features included in the PCFA.

While the potential impact of these differences on linkage scans has not been fully evaluated, an interesting example can be seen by contrasting two linkage studies performed by Kraja and colleagues (15,16). In both analyses, Kraja et al. used family data from HyperGEN and defined MetS phenotypes using PCFA. Ten individual MetS features were included in the PCFA in one paper, while the second paper included eleven MetS features. In both analyses, four factors were identified and used as composite phenotypes in the linkage analysis (obesity-insulin, lipids-insulin, central obesity, and blood pressure). In some cases, there was direct overlap in the linkage regions identified between the two studies, while in other cases linkage signals were not replicated even though it appears that the same population was used. The lack of replication may be due in part to differences in the composite factors identified by PCFA.

Another potentially important difference with regard to defining composite MetS phenotypes by PCFA relates to whether the retained factors are rotated or not. For example, six studies performed an orthogonal rotation, most often a varimax rotation, whereas three of the previous linkage studies did not used a rotation step (9,10,13). While there is debate about the most appropriate approach, work in this field will benefit from more standard practices in carrying out factor analysis (39).

We performed PCFA with varimax rotation in four ethnic groups, using the same set of MetS features, retaining the same number of factors, and applying the same statistical methods. This standardized approach resulted in very similar composite phenotypes for MetS in all ethnic groups. However, despite these efforts different patterns of linkage emerged across ethnic groups. First, the specific MetS factors showing suggestive evidence for linkage differed across groups. For example, among the African-American families only the bpfactor showed evidence for linkage, while among the whites, it was only the wtfactor. Second, evidence for linkage for specific MetS factors was observed for different regions across groups. Among the African-American families, five suggestive regions for the bpfactor were identified that did not overlap with the two bpfactor regions identified in the Mexican-American sample.

A strength of the current study was the use of a set of large, well-characterized families from four different ethnic groups, allowing comparisons of linkage signals across these groups. Based on the different linkage patterns observed across the four groups, it is tempting to speculate that specific genomic regions may play a more prominent role in MetS disease susceptibility in different ethnic groups. However, it is not clear whether these differences truly represent evidence for genetic heterogeneity. While it is possible that the slight differences in factor loadings may account for differences in linkage patterns, it is more likely that these signals reflect differences in power among the different groups (due to both sample size and differences in LD patterns between the groups), environmental effects, gene environment interactions, or combinations of these hypotheses. Evaluating the impact of each of these potential effects is a focus of our ongoing work.

In summary, several regions with suggestive evidence for linkage to composite MetS phenotypes were identified in each ethnic group. By identifying genetic factors influencing susceptibility to the MetS our understanding of the molecular mechanisms underlying the pathogenesis of MetS will be enhanced. This understanding is a necessary prerequisite for the rational development of preventive and improved therapeutic approaches for MetS. The challenge in years to come will be to understand how risk is influenced by the interaction of susceptibility genes with each other and with pertinent social and environmental factors and to establish how best to apply this understanding to provide individuals with clinically useful preventive, diagnostic, prognostic, and therapeutic information.

Disclosure

The authors declared no conflict of interest.

References

REFERENCES

1. Ford ES, Giles WH, Dietz WH. Prevalence of the metabolic syndrome among US adults: findings from the third National Health and Nutrition Examination Survey. JAMA 2002;287:356–359. | Article | PubMed | ISI |
2. Reaven GM. Banting lecture 1988. Role of insulin resistance in human disease. Diabetes 1988;37:1595–1607. | Article | PubMed | ISI | ChemPort |
3. Leyva F, Godsland IF, Worthington M, Walton C, Stevenson JC. Factors of the metabolic syndrome: baseline interrelationships in the first follow-up cohort of the HDDRISC Study (HDDRISC-1). Heart Disease and Diabetes Risk Indicators in a Screened Cohort. Arterioscler Thromb Vasc Biol 1998;18:208–214. | PubMed | ChemPort |
4. Pyorala M, Miettinen H, Halonen P, Laakso M, Pyorala K. Insulin resistance syndrome predicts the risk of coronary heart disease and stroke in healthy middle-aged men: the 22-year follow-up results of the Helsinki Policemen Study. Arterioscler Thromb Vasc Biol 2000;20:538–544. | PubMed | ISI | ChemPort |
5. Lindblad U, Langer RD, Wingard DL, Thomas RG, Barrett-Connor EL. Metabolic syndrome and ischemic heart disease in elderly men and women. Am J Epidemiol 2001;153:481–489. | Article | PubMed | ChemPort |
6. Edwards KL, Austin MA, Newman B, Mayer E, Krauss RM, Selby JV. Multivariate analysis of the insulin resistance syndrome in women. Arterioscler Thromb 1994;14:1940–1945. | PubMed | ChemPort |
7. Arya R, Duggirala R, Almasy L et al. Linkage of high-density lipoprotein-cholesterol concentrations to a locus on chromosome 9p in Mexican Americans. Nat Genet 2002;30:102–105. | Article | PubMed | ChemPort |
8. Liu M, Ramos-Castaneda J, Arvan P. Role of the connecting peptide in insulin biosynthesis. J Biol Chem 2003;278:14798–14805. | Article | PubMed | ChemPort |
9. Loos RJ, Katzmarzyk PT, Rao DC et al. Genome-wide linkage scan for the metabolic syndrome in the HERITAGE Family Study. J Clin Endocrinol Metab 2003;88:5935–5943. | Article | PubMed | ChemPort |
10. Tang W, Miller MB, Rich SS et al. Linkage analysis of a composite factor for the multiple metabolic syndrome: the National Heart, Lung, and Blood Institute Family Heart Study. Diabetes 2003;52:2840–2847. | Article | PubMed | ChemPort |
11. Cai G, Cole SA, Freeland-Graves JH, MacCluer JW, Blangero J, Comuzzie AG. Principal component for metabolic syndrome risk maps to chromosome 4p in Mexican Americans: the San Antonio Family Heart Study. Hum Biol 2004;76:651–665. | Article | PubMed |
12. Adeyemo AA, Johnson T, Acheampong J et al. A genome wide quantitative trait linkage analysis for serum lipids in type 2 diabetes in an African population. Atherosclerosis 2005;181:389–397. | Article | PubMed | ChemPort |
13. Bosse Y, Despres JP, Chagnon YC et al. Quantitative trait locus on 15q for a metabolic syndrome variable derived from factor analysis. Obesity (Silver Spring) 2007;15:544–550. | PubMed | ChemPort |
14. North KE, Miller MB, Coon H et al. Evidence for a gene influencing fasting LDL cholesterol and triglyceride levels on chromosome 21q. Atherosclerosis 2005;179:119–125. | Article | PubMed | ChemPort |
15. Kraja AT, Hunt SC, Pankow JS et al. Quantitative trait loci for metabolic syndrome in the Hypertension Genetic Epidemiology Network study. Obes Res 2005;13:1885–1890. | Article | PubMed |
16. Kraja AT, Rao DC, Weder AB et al. Two major QTLs and several others relate to factors of metabolic syndrome in the family blood pressure program. Hypertension 2005;46:751–757. | Article | PubMed | ChemPort |
17. Colhoun HM, McKeigue PM, Davey Smith G. Problems of reporting genetic associations with complex outcomes. Lancet 2003;361:865–872. | Article | PubMed | ISI |
18. Raffel LJ, Robbins DC, Norris JM et al. The GENNID Study. A resource for mapping the genes that cause NIDDM. Diabetes Care 1996;19:864–872. | Article | PubMed | ChemPort |
19. Ehm MG, Karnoub MC, Sakul H et al. Genomewide search for type 2 diabetes susceptibility genes in four American populations. Am J Hum Genet 2000;66:1871–1881. | Article | PubMed | ISI | ChemPort |
20. Executive Summary of The Third Report of The National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, And Treatment of High Blood Cholesterol In Adults (Adult Treatment Panel III). JAMA 2001;285:2486–2497. | Article | PubMed | ISI |
21. Yuan B, Vaske D, Weber JL, Beck J, Sheffield VC. Improved set of short-tandem-repeat polymorphisms for screening the human genome. Am J Hum Genet 1997;60:459–460. | PubMed | ChemPort |
22. O'Connell JR, Weeks DE. PedCheck: a program for identification of genotype incompatibilities in linkage analysis. Am J Hum Genet 1998;63:259–266. | Article | PubMed | ISI | ChemPort |
23. Thomson GH. The Factorial Analysis of Human Ability. Houghton Mifflin: Boston, 1951.
24. Blangero J, Williams JT, Almasy L. Variance component methods for detecting complex trait loci. Adv Genet 2001;42:151–181. | Article | PubMed | ChemPort |
25. Blangero J, Almasy L. Multipoint oligogenic linkage analysis of quantitative traits. Genet Epidemiol 1997;14:959–964. | Article | PubMed | ISI | ChemPort |
26. Abecasis GR, Cherny SS, Cookson WO, Cardon LR. Merlin—rapid analysis of dense genetic maps using sparse gene flow trees. Nat Genet 2002;30:97–101. | Article | PubMed | ISI | ChemPort |
27. Almasy L, Blangero J. Multipoint quantitative-trait linkage analysis in general pedigrees. Am J Hum Genet 1998;62:1198–1211. | Article | PubMed | ISI | ChemPort |
28. Lander E, Kruglyak L. Genetic dissection of complex traits: guidelines for interpreting and reporting linkage results. Nat Genet 1995;11:241–247. | Article | PubMed | ISI | ChemPort |
29. Blangero J, Williams JT, Almasy L. Robust LOD scores for variance component-based linkage analysis. Genet Epidemiol 2000;19(Suppl 1):S8–S14. | Article |
30. Herbert A, Gerry NP, McQueen MB et al. A common genetic variant is associated with adult and childhood obesity. Science 2006;312:279–283. | Article | PubMed | ISI | ChemPort |
31. Dina C, Meyre D, Samson C et al. Comment on "A common genetic variant is associated with adult and childhood obesity". Science 2007;315:187; author reply 187. | Article | PubMed | ChemPort |
32. Hall DH, Rahman T, Avery PJ, Keavney B. INSIG-2 promoter polymorphism and obesity related phenotypes: association study in 1428 members of 248 families. BMC Med Genet 2006;7:83. | Article | PubMed | ChemPort |
33. Loos RJ, Barroso I, O'Rahilly S, Wareham NJ. Comment on "A common genetic variant is associated with adult and childhood obesity". Science 2007;315:187; author reply 187. | Article | PubMed | ChemPort |
34. Rosskopf D, Bornhorst A, Rimmbach C et al. Comment on "A common genetic variant is associated with adult and childhood obesity". Science 2007;315:187; author reply 187. | Article | PubMed | ChemPort |
35. Lyon HN, Emilsson V, Hinney A et al. The association of a SNP upstream of INSIG2 with body mass index is reproduced in several but not all cohorts. PLoS Genet 2007;3:e61. | Article | PubMed | ChemPort |
36. Kissebah AH, Sonnenberg GE, Myklebust J et al. Quantitative trait loci on chromosomes 3 and 17 influence phenotypes of the metabolic syndrome. Proc Natl Acad Sci USA 2000;97:14478–14483. | Article | PubMed | ChemPort |
37. Cole SA, Comuzzie AG. The end of the beginning. Am J Clin Nutr 2007;86:274–275. | PubMed | ChemPort |
38. Vasseur F, Meyre D, Froguel P. Adiponectin, type 2 diabetes and the metabolic syndrome: lessons from human genetic studies. Expert Rev Mol Med 2006;8:1–12. | PubMed |
39. Novak S, Stapleton LM, Litaker JR, Lawson KA. Re: "(Mis)use of factor analysis in the study of insulin resistance syndrome". Am J Epidemiol 2005;161:1184–1185. | Article | PubMed |

Acknowledgments

This work was funded by a grant (7-05-RA 34) from the American Diabetes Association (K.L.E., principal investigator). We also thank Brian Fish, Mame Diop, and Erin Ramos for their invaluable assistance in data management activities. We also thank Wendy Langeberg, Margaret Ragland, and Kristin Beima for their assistance with the project.