Common variation in the adiponectin gene has an effect on ... - Nature

2 downloads 0 Views 232KB Size Report
Jan 20, 2011 - Newcastle upon Tyne, UK and 4Institute of Human Genetics, Newcastle University, ... morphism has independent effects both on waist–hip ratio.
Journal of Human Hypertension (2011) 25, 719–724 & 2011 Macmillan Publishers Limited All rights reserved 0950-9240/11 www.nature.com/jhh

ORIGINAL ARTICLE

Common variation in the adiponectin gene has an effect on systolic blood pressure PJ Avery1, SK Patel2, IM Ibrahim3, M Walker3 and BD Keavney4 1

School of Mathematics and Statistics, Newcastle University, Newcastle upon Tyne, UK; 2Department of Medicine (Austin Health), University of Melbourne, Heidelberg Repatriation Hospital, Heidelberg West, Victoria, Australia; 3Diabetes Research Group, School of Clinical Medical Sciences, Newcastle University, Newcastle upon Tyne, UK and 4Institute of Human Genetics, Newcastle University, Newcastle upon Tyne, UK

The genotype at the C-11377G single-nucleotide polymorphism (SNP) (rs266729) in the adiponectin gene promoter has been shown to affect the prevalence of coronary atherosclerosis and incidence of vascular events in men, and to affect carotid intima media thickness. We have examined the relationship between this polymorphism and blood pressure in a cohort ascertained to express variability in blood pressure measurements. We studied a cohort of 255 families comprising 1425 individuals ascertained via a hypertensive proband. Blood pressure was measured by ambulatory monitoring. The C-11377G SNP was genotyped using a TaqMan assay. There was evidence of association between this SNP and log systolic

blood pressure (SBP), having adjusted for significant covariates including gender, age and drug treatment; P ¼ 0.009, 0.014 and 0.022, respectively, for daytime, nighttime and clinic measurements. Replacing C by G caused an increase of 1.63, 1.83 and 1.61%, respectively, per gene copy. There were smaller effects on diastolic blood pressure and waist–hip ratio, which were of borderline significance. Genotype at the C-11377G (rs266729) polymorphism has independent effects both on waist–hip ratio and SBP. This may help in understanding the complex role that the adiponectin gene has in atherosclerosis. Journal of Human Hypertension (2011) 25, 719–724; doi:10.1038/jhh.2010.122; published online 20 January 2011

Keywords: adiponectin; promoter; blood pressure

Introduction A susceptibility locus for type II diabetes and the metabolic syndrome has been mapped to chromosome 3q27.1 The ADIPOQ gene is located at position 3q27 and encodes the protein adiponectin. Decreased circulating adiponectin levels have been observed in patients with cardiovascular disease, hypertension, obesity and type II diabetes compared with healthy controls.2–4 Adiponectin has been shown to suppress the transformation of macrophages into foam cells5 and is detected in catheterinjured vessels but not in intact vessels.6 Low adiponectin levels are associated with endothelial dysfunction,7 increased carotid intima media thickness8 and increased ambulatory systolic blood pressure (SBP)9 whereas higher plasma levels have been associated with a lower risk of myocardial infarction.3 A number of single-nucleotide polymorphisms (SNPs) in the ADIPOQ gene have been described.10,11 Correspondence: Dr PJ Avery, School of Mathematics and Statistics, Newcastle University, Herschel Building, Newcastle upon Tyne, Newcastle NE1 7RU, UK. E-mail: [email protected] Received 16 April 2010; revised 26 November 2010; accepted 5 December 2010; published online 20 January 2011

A common variant of C-11377G (rs266729) in the proximal promoter region of the ADIPOQ gene has been associated with circulating adiponectin levels8,12 and type II diabetes.10,13 Another study has shown this SNP to be associated with coronary stenosis and vascular events in men.14 Thus there is a possible link between this SNP and cardiovascular disease. The G allele at rs266729 is associated with a reduction in adiponectin levels10,15 and an increase in the incidence of coronary artery disease.14 It has also been found that this allele leads to an increase in carotid intima media thickness, a measure of systemic atherosclerotic burden.16 We have genotyped this SNP in a cohort of families ascertained for variability in blood pressure to ascertain whether there is an effect on blood pressure.

Methods The collection strategy of this family study has been described previously.17 Families were ascertained in 1993–1996 in the Oxfordshire region of the United Kingdom through a proband diagnosed with essential hypertension. The following criteria defined eligibility as a proband: a mean SBP over 140 mm Hg and mean diastolic blood pressure over 90 mm Hg on

Adiponectin gene and systolic blood pressure PJ Avery et al 720

daytime ambulatory blood pressure monitoring; or greater than three office blood pressure readings greater than 160 mm Hg systolic and 95 mm Hg diastolic; or treatment with two or more antihypertensive drugs. These relatively stringent criteria were applied to provide maximum security that probands were indeed at the upper end of the population blood pressure distribution. Secondary hypertension was excluded using the standard screening protocol applied in the hypertension clinic. In order to be suitable for the study, families were required to consist of at least three siblings clinically assessable for blood pressure if at least one parent of the sibship was available to give blood for DNA analysis, and to consist of at least four assessable siblings if no parent was available for DNA analysis. Quantitatively assessed sibships were recruited either in the generation of the proband or his/her offspring. Where members of the sibship were found to be hypertensive, families were extended and the spouses and offspring of hypertensive sibs collected. The majority of the individuals in the family collection, therefore, have blood pressures within the conventionally accepted ‘normal range’, and the family collection includes some extended families, though most are nuclear families. The study received ethical clearance from the appropriate review committees, and corresponded with the principles of the Declaration of Helsinki. All participants gave informed consent to participate in the study. Blood pressure was measured using ambulatory monitoring for a period of 24 h in all subjects willing to undergo monitoring, using the A&D TM 2421 monitor according to a previously described protocol.17 A full clinical history was taken, which included the medical history of the subject and lifestyle factors including consumption of alcohol and tobacco, and habitual physical exercise. Anthropometric measurements including height, weight and waist and hip circumferences were performed. Anthropometric measurements of height, weight, waist and hip circumferences were carried out according to standard methods. DNA was extracted from blood samples using standard methods. Bilateral common carotid ultrasonography was performed to measure the mean far-wall carotid intima medial thickness in a subset of 953 participants using standard approaches as previously described.18 The 11377C4G SNP (rs266729) was genotyped using a TaqMan allelic discrimination assay (Applied Biosystems, Warrington, UK). Primers and probes were designed using the Assay by Design Service (Applied Biosciences, Warrington, UK). A 25 ml PCR was set up with 10 ng of genomic DNA, 1  TaqMan universal PCR MasterMix with AmpErase UNG and 1  primer and probes assay mix. Samples were amplified on a GeneAmp 9700 PCR machine (Applied Biosciences). PCR cycling conditions were as follows: 50 1C for 2 min followed by Journal of Human Hypertension

95 1C for 10 min, then 40 cycles of 92 1C for 0.15 s, 60 1C for 1 min. Following PCR, an allelic discrimination assay was performed on an ABI Prism 7000 sequence detection system (Applied Biosystems). Of the genotyped samples, 5% were duplicates and there was at least one negative control per 96-well DNA plate. The accuracy of the genotyping was determined by the genotype concordance between duplicate samples. We obtained a 100% concordance between the genotyped duplicate samples. The genotyping success rate was 498%. Mendelian inheritance of all the genotypes, and Hardy–Weinberg equilibrium for each marker, were checked using PEDSTATS. Additional checks based on unlikely recombination patterns within families were carried out using the error-checking option in MERLIN v.1.1.1. Errors were corrected when possible by reference to the raw genotyping data, and when this was not possible genotypes were excluded from analysis. We used MINITAB v15 to examine the phenotypes for normality. All the variables, except waist–hip ratio, required log transformation to adequately conform to a normal distribution. We used MINITAB to adjust the phenotypes for the significant covariates selected from age, age-squared, sex, smoking, alcohol consumption, habitual physical activity and antihypertensive drugs (beta-blockers and diuretics) using linear regression, as described previously.19,20 The effect of the drug treatment was obtained by analyzing the values of the variables adjusted for the other covariates in the hypertensive individuals. Virtually all of the individuals receiving betablockers or diuretics were hypertensive. For each drug, only about a third of hypertensive individuals were receiving the drug and so it was possible to obtain good estimates of the effect of drug treatment. A small number of extreme individuals taking three or four drugs to reduce blood pressure were excluded from this analysis. The estimated effect of the drug was then added to the adjusted blood pressure score for each individual taking the particular agent, adjusting the blood pressure score upward to reflect the average effect of the drug in the population studied. The log-transformed, covariateadjusted residuals were entered into the quantitative trait genetic association analyses, which were performed using a variance-components approach within the package MERLIN which takes account of shared polygenic effects in members of the same pedigree, as described previously.19,20

Results The study consisted of 1425 patients (47.6% of which were male). Approximately 40% of those assessed were hypertensive. In all, 31.5% of the hypertensive individuals were taking beta-blockers and 33.7% were taking diuretics. Only 0.6% of ‘normal’ individuals were taking beta-blockers and a

Adiponectin gene and systolic blood pressure PJ Avery et al 721

similar percentage were taking diuretics. The characteristics of the study population, significant covariates for each phenotype, the amount of variability explained by the significant covariates and the heritability of the adjusted traits are summarized in Table 1. The frequency of the G allele at rs266729 was 0.27 with no significant deviation from Hardy–Weinberg equilibrium at Po0.01. This is similar to that found in the RISC cohort,16 wherein the G allele frequency was 0.24. The frequency is also similar to the frequency of G in the HapMap CEU population, which was 0.30 (http://www.hapmap.org). The effect of this SNP was assessed both using regression and using MERLIN, which allows for the relatedness of the individuals. In general, the MERLIN P-values are slightly higher. This SNP appears to have an effect on adjusted systolic, and to a lesser extent on diastolic, blood pressure measurements measured in the clinic, during the day or at night. Similarly, there appears to be an effect on waist–hip ratio. The effect on blood pressure is reduced slightly by adjusting for waist– hip ratio but is still significant (see Table 2). As the blood pressure variables were log-transformed, the estimated coefficients essentially give the percentage increase caused by replacing C by G at this SNP. The effect is thus a rise of 1–2% in each of the blood pressure variables per gene copy. The increase in waist–hip ratio is about 0.8% per gene copy. There were no significant effects on mean intima media thickness or on various other measurements made on this cohort. If you look at the distribution of the three genotypes for those individuals with adjusted log daytime SBP above or below the upper quartile, you find that there is a significant difference in the genotype frequencies (w2 ¼ 7.79 on 2 degrees of freedom, P ¼ 0.020) with an excess of GG homozygotes and a shortage of CC homozygotes in those individuals above the upper quartile. If you look at those above and below the upper quartile for waist– hip ratio, you find a similar but less strong effect (w2 ¼ 6.34 on 2 degrees of freedom, P ¼ 0.042).

Discussion We have shown that the C11377G SNP (rs266729) in the adiponectin promoter is associated with small effects on SBP and central adiposity. The G allele, which has been shown by others to be associated with lower plasma adiponectin levels, is associated with around a 1–2% increase in SBP and 0.8% increase in waist–hip ratio per allele. Once the effect on waist–hip ratio is adjusted for, the effect on blood pressure remains significant, indicating that the effects are independent. A previous report14 showed association between this same SNP and angiographic coronary artery disease in a prospective study of 402 men. The G

allele was significantly associated with disease severity and future vascular events independent of serum adiponectin levels and traditional risk factors for cardiovascular disease. Patel et al.16 demonstrated recently an association between the G allele of C11377G and higher carotid intima medial thickness, after adjustment for adiponectin levels and a variety of cardiovascular risk factors. However, previous studies have not elucidated the potential mechanism for the association. The pleiotropic effect we have observed on blood pressure and central adiposity suggest that the SNP acts via reduced tissue or circulating adiponectin levels to affect a number of different pathways related to cardiovascular risk. Previous studies have not identified the rs266729 SNP as a risk factor for higher blood pressure16 or waist–hip ratio.15 However, a recent French study has shown that it affected the change in waist–hip ratio in a longitudinal study.15 With respect to blood pressure, our study utilized ambulatory monitoring, which increases the precision with which blood pressure is measured and could have substantially increased the power of the study to detect effects. With respect to the effects we observed on both blood pressure and waist–hip ratio, these are small, and would have been unlikely to withstand the rigorous corrections for multiple testing of many thousands of SNPs that must be implemented in recent genome-wide association studies of either blood pressure or obesity. The small effect we have observed is, however, potentially important with respect to a more complete understanding of the role of adiponectin in disease. Previous studies have demonstrated that lower adiponectin levels are associated with a variety of cardiovascular risk factors including endothelial dysfunction,7 increased carotid intima media thickness8 and increased ambulatory blood pressure9 (interestingly, as we observed for adiponectin genotype, the association with adiponectin levels appeared to be restricted to systolic pressure). Treatment of patients with type II diabetes with the PPAR-g agonist rosiglitazone results in a twofold increase in plasma adiponectin levels,21 and specific adiponectin receptor agonists or adiponectin sensitizers have been suggested as potential treatments to reduce cardiometabolic risk. However, whether the association between adiponectin levels and risk reduction is causal or confounded remains in doubt. Our study provides evidence in favour of a causal role for adiponectin. Genotypes at C11377G, which has previously been shown to be associated with differences in plasma adiponectin levels, are associated with blood pressure and waist–hip ratio. The allele that results in lower adiponectin levels is associated with higher levels of these cardiovascular risk factors. As genotypes are determined at conception, and should not be affected either by confounding or reverse causality, this constitutes strong evidence for a causal effect (an argument sometimes termed ‘Mendelian randomisation’). Journal of Human Hypertension

Journal of Human Hypertension

86.0

47.0

Clinic SBP, mm Hg

Clinic DBP, mm Hg

0.647

0.78

73.7

121.3

60.1

103.9

72.8

0.757

0.85

82.0

134.0

66.0

113.0

79.7

26.4% (former smoker) 21.6% (1, 2 times per week) 132.6

51% (non-smoker) 21.4% (o1 time per week) 122.1

50.9 3

Median

35.7 0

Lower quartile

0.905

0.913

92.0

153.0

73.5

126.0

90.0

13.9% (X3 times per week) 146.0

22.0% (smoker)

60.9 12

Upper quartile

Abbreviations: DBP, diastolic blood pressure; IMT, intima media thickness; SBP, systolic blood pressure. a % explained by covariates on logged data (excluding drug treatment).

0.423

40.9

Night-time DBP, mm Hg

Mean IMT

72.0

Night-time SBP, mm Hg

0.56

54.0

Daytime DBP, mm Hg

Waist–hip ratio

92.0

43.1% (none)

18.7 0

Minimum

2.17

1.22

135.7

226.0

110.6

181.0

119.9

214.0

90.7 80

Maximum

854

1358

1177

1178

930

931

1181

1182

1413

1423

1425 1420

N

a

38.9

48.7

19.7

27.2

13.6

11.4

17.9

20.4





— —

R2

20.6

27.8

11.5

12.3

33.0

23.7

9.6

13.7





— —

H2

Age, gender, age*gender, age2, exercise (some versus none), alcohol, beta-blockers, diuretics Age, gender, age2, exercise, alcohol, smoking. beta-blockers, diuretics Age, gender, age*gender, age2, beta-blockers, diuretics Age, gender, alcohol, smoking. beta-blockers, diuretics Age, gender, age*gender, 2 age , exercise, alcohol, beta-blockers, diuretics Age, gender, age2, exercise, alcohol, smoking. beta-blockers, diuretics Age, gender, age*gender, exercise (some versus none), smoking (non versus others) Age, gender, age2, alcohol (females), exercise (some versus none)





— —

Covariates used

722

Daytime SBP, mm Hg

Exercise taken

Age Alcohol consumption, units per week Smoking

Variable

Table 1 Descriptive information on the study cohort

Adiponectin gene and systolic blood pressure PJ Avery et al

Adiponectin gene and systolic blood pressure PJ Avery et al 723

Table 2 Effect of the C-11377G SNP (rs266729) on adjusted logged blood pressure measurements and selected other adjusted variables Variable (adjusted for covariates: see Table 1)

Log clinic SBP Log daytime SBP Log night-time SBP Log clinic DBP Log daytime DBP Log night-time DBP Waist–hip ratio Log IMT mean

Effect of replacing C by G (Standard error)

0.0161 0.0163 0.0183 0.0101 0.0140 0.0147 0.0068 0.0041

(0.007) (0.006) (0.007) (0.007) (0.007) (0.008) (0.0030) (0.012)

Regression P-value

Merlin P-value

Estimate and P-value, having allowed for adjusted WHR

0.022 0.009 0.014 0.176 0.048 0.074 0.022 0.737

0.047 0.019 0.018 0.20 0.079 0.35 0.111 0.77

0.0152, 0.031 0.0154, 0.013 0.0166, 0.028 0.0094, 0.213 0.0135, 0.060 0.0138, 0.100 — 0.0021, 0.863

Abbreviations: DBP, diastolic blood pressure; IMT, intima media thickness; SBP, systolic blood pressure.

Circulating adiponectin is present in at least three different complexes: high-molecular weight (12 to 36-mer), low-molecular weight (hexamer) and trimeric. These complexes have different function, and the ratio of the high molecular weight to other forms is decreased in obese patients. The effect of the C11377G SNP on the relative amounts of each complex present in plasma is unknown; it therefore remains possible that the SNP exerts its influence principally through affecting the propensity of adiponectin monomers to form up into the highmolecular weight complex. It is as yet unclear whether the C11377G SNP is itself causal of reduced adiponectin levels or whether the effect arises from a neighbouring SNP in linkage disequilibrium with it. Further studies will be needed to investigate this possibility. Lifestyle modifications and cardiovascular therapies such as angiotensin-converting enzyme inhibitors and angiotensin receptor blockers are known to increase plasma adiponectin level, and it is thought that this property may explain, in part, their propensity to improve endothelial function. Future pharmacogenetic studies to determine the relationships between the adiponectin response to these agents, C11377G genotype and improvements in cardiovascular risk factors would be of interest. Despite careful phenotyping, our finding will require replication in additional cohorts to be considered secure. As very large case-control genetic studies not involving ambulatory monitoring have thus far not detected the association we describe, it would be of particular interest to attempt replication in a quantitatively characterized cohort who have undergone ambulatory monitoring. Another issue is that we were unable to standardize the salt intake in our population of largely healthy community dwelling individuals. In conclusion, our finding that the C11377G SNP in the adiponectin gene promoter is associated with blood pressure and waist–hip ratio lends further support to the notion that adiponectin is involved causally in cardiometabolic risk and is an attractive target for direct therapeutic intervention.

What is known about topic Genotype at C-11377G SNP (rs266729) in the adiponectin gene promoter affects: K The prevalence of coronary atherosclerosis. K Incidence of vascular events in men. K Carotid intima media thickness. What this study adds Each G copy at C-11377G SNP (rs266729) in the adiponectin gene promoter causes an increase of: K About 2% in systolic blood pressure. K About 1% in diastolic blood pressure. K And about 1% in waist–hip ratio.

Conflict of interest The authors declare no conflict of interest.

Acknowledgements The authors acknowledge the support of the British Heart Foundation. We would also like to thank the referees of the first draft of this paper for their helpful comments.

References 1 Vionnet N, Hani ElH, Dupont S, Gallina S, Francke S, Dotte S et al. Genomewide search for type 2 diabetessusceptibility genes in French whites: evidence for a novel susceptibility locus for early-onset diabetes on chromosome 3q27-qter and independent replication of a type 2-diabetes locus on chromosome 1q21–q24. Am J Hum Genet 2000; 67: 1470–1480. 2 Hotta K, Funahashi T, Arita Y, Takahashi M, Matsuda M, Okamoto Y et al. Plasma concentrations of a novel, adipose-specific protein, adiponectin, in type 2 diabetic patients. Arterioscler Thromb Vasc Biol 2000; 20: 1595–1599. 3 Pischon T, Girman CJ, Hotamisligil GS, Rifai N, Hu FB, Rimm EB. Plasma adiponectin levels and risk of myocardial infarction in men. JAMA 2004; 291: 1730–1737. 4 Iwashima Y, Katsuya T, Ishikawa K, Ouchi N, Ohishi M, Sugimoto K et al. Hypoadiponectinemia is an Journal of Human Hypertension

Adiponectin gene and systolic blood pressure PJ Avery et al 724

5

6

7

8

9

10

11 12

13

independent risk factor for hypertension. Hypertension 2004; 43: 1318–1323. Ouchi N, Kihara S, Arita Y, Nishida M, Matsuyama A, Okamoto Y et al. Adipocyte-derived plasma protein, adiponectin, suppresses lipid accumulation and class A scavenger receptor expression in human monocytederived macrophages. Circulation 2001; 103: 1057–1063. Okamoto Y, Arita Y, Nishida M, Muraguchi M, Ouchi N, Takahashi M et al. An adipocyte-derived plasma protein, adiponectin, adheres to injured vascular walls. Horm Metab Res 2000; 32: 47–50. Shimabukuro M, Higa N, Asahi T, Oshiro Y, Takasu N, Tagawa T et al. Hypoadiponectinemia is closely linked to endothelial dysfunction in man. J Clin Endocrinol Metab 2003; 88: 3236–3240. Lo J, Dolan SE, Kanter JR, Hemphill LC, Connelly JM, Lees RS et al. Effects of obesity, body composition, and adiponectin on carotid intima-media thickness in healthy women. J Clin Endocrinol Metab 2006; 91: 1677–1682. Vasunta RL, Kesa¨niemi YA, Ukkola O. Plasma adiponectin concentration is associated with ambulatory daytime systolic blood pressure but not with the dipping status. J Hum Hypertens 2010 (e-pub ahead of print0. Vasseur F, Helbecque N, Dina C, Lobbens S, Delannoy V, Gaget S et al. Single-nucleotide polymorphism haplotypes in the both proximal promoter and exon 3 of the APM1 gene modulate adipocyte-secreted adiponectin hormone levels and contribute to the genetic risk from type 2 diabetes in French Caucasians. Hum Mol Genet 2002; 11: 2607–2614. Gibson F, Froguel P. Genetics of the APM1 locus and its contribution to type 2 diabetes susceptibility in French Caucasians. Diabetes 2004; 53: 2977–2983. Laumen H, Saningong AD, Heid IM, Hess J, Herder C, Claussnitzer M et al. Functional characterization of promoter variants of the adiponectin gene complemented by epidemiological data. Diabetes 2009; 58: 984–991. Wang X, Zhang S, Chen Y, Liu H, Lan C, Chen X et al. APM1 gene variants 11377C/G and 4545G/C are

Journal of Human Hypertension

14

15

16

17

18

19

20

21

associated respectively with obesity and with nonobesity in Chinese type 2 diabetes. Diabetes Res Clin Pract 2009; 84: 205–210. Hoefle G, Muendlein A, Saely CH, Risch L, Rein P, Koch L et al. The –11377 C4G promoter variant of the adiponectin gene, prevalence of coronary atherosclerosis, and incidence of vascular events in men. Thromb Haemost 2007; 97: 451–457. Dolley G, Bertrais S, Frochot V, Bebel J-F, Guerre-Millo M, Tores F et al. Promoter adiponectin polymorphisms and waist/hip ratio variation in a prospective French adults study. Int J Obes 2008; 32: 669–675. Patel S, Flyvbjerg A, Koza`kova` M, Frystyk J, Ibrahim IM, Petrie JR et al. Variation in the ADIPOQ gene promoter is associated with carotid intima media thickness independent of plasma adiponectin levels in healthy subjects. Eur Heart Jour 2008; 29: 386–393. Palomino-Doza J, Rahman TJ, Avery PJ, Mayosi BM, Farrall M, Watkins H et al. Ambulatory blood pressure is associated with polymorphic variation in P2X receptor genes. Hypertension 2008; 52: 980–985. Mayosi BM, Avery PJ, Baker M, Gaukrodger N, Imrie H, Green FR et al. Genotype at the -174G/C polymorphism of the Interleukin-6 gene is associated with common carotid artery intimal medial thickness; family study and meta-analysis. Stroke 2005; 36: 2215–2219. Baker M, Gaukrodger N, Mayosi BM, Imrie H, Farrall M, Watkins H et al. Association between common polymorphisms of the proopiomelanocortin gene and body fat distribution: a family study. Diabetes 2005; 54: 2492–2496. Gaukrodger N, Mayosi BM, Imrie H, Avery P, Baker M, Connell JM et al. A rare variant of the leptin gene has large effects on blood pressure and carotid intimamedial thickness: a study of 1428 individuals in 248 families. J Med Genet 2005; 42: 474–478. Yang WS, Jeng CY, Wu TJ, Tanaka S, Funahashi T, Matsuzawa Y et al. Synthetic peroxisome proliferatoractivated receptor-gamma agonist, rosiglitazone, increases plasma levels of adiponectin in type 2 diabetic patients. Diabetes Care 2002; 25: 376–380.