Visfatin was recently reported as a novel adipokine encoded by the NAMPT (PBEF1) gene. This study was aimed at investigation of the possibility that single-nucleotide polymorphisms (SNPs) in the visfatin gene are associated with either obesity or type 2 diabetes (T2D). A set of eight “tag-SNPs” were selected and ABI SNPlex assays designed for genotyping purposes. A total of 1,709 severely obese subjects were typed (896 class III obese adults and 813 children) together with 2,367 T2D individuals and 2,850 controls. For quantitative trait analysis, an additional 2,362 subjects were typed for rs10487818 from a general population sample. One rare SNP, rs10487818, located in intron 4 of NAMPT was associated with severe obesity, with a minor allele frequency of 1.6% in controls, 0.4% in the class III obese adults and, remarkably, 0% in the severely obese children. A highly significant association was observed for the presence or absence of the rare allele, i.e., (A,A) vs. (A,T + T,T) genotypes, in children (P = 6 × 10−9) and in adults (P = 8 × 10−5). No other significant (P < 0.05) association was observed with obesity or T2D for this or any other SNP. No association with BMI or waist-to-hip ratio was observed in a general population sample (n = 5,212). This is one of the first rare SNPs shown to be protective against a common polygenic disease and provides further evidence that rare alleles of strong effect can contribute to complex diseases such as severe obesity.
Visfatin was first reported as a novel adipocytokine secreted preferentially by visceral fat tissue compared to subcutaneous fat in both humans and mice (1) and is encoded by a gene officially now known as nicotinamide phosphoribosyltransferase (NAMPT, Entrez ID: 10135), but previously known as pre-B cell colony–enhancing factor gene, PBEF1 (ref. 2). Visfatin was also identified as having insulin-mimetic properties, though subsequent reports indicate that visfatin is expressed in all adipose tissue samples examined (3,4,5) and have not confirmed visfatin's insulin-like properties (6). Currently, there is sufficient evidence to consider that visfatin is expressed by the macrophages infiltrating adipose tissue and is produced in response to inflammatory signals (4,7). However, the initial reports led to the hypothesis that visfatin might be involved in type 2 diabetes (T2D) and/or obesity, and subsequent studies have demonstrated some significant associations.
Plasma visfatin concentration is significantly increased in T2D (P = 0.002) (ref. 8) and was reported to be independently associated with T2D, after correction for a wide range of factors such as age, gender, BMI (P = 0.007). Plasma visfatin concentrations are also higher in obese nondiabetic children compared to lean control children (P < 0.001) (ref. 9). Plasma concentration of visfatin is also positively associated with visceral visfatin mRNA expression (P < 0.0001), BMI (P = 0.004), and percent body fat (P = 0.048) (ref. 10) in adults. Positive associations between visceral visfatin mRNA expression, BMI (P = 0.001), and percent body fat (P = 0.004) were also revealed by the same study. The association with BMI was later confirmed by another group (11), but they also reported that plasma visfatin concentration was negatively associated with BMI (P = 0.02), although the authors attributed this result to differences in visfatin expression in different fat depots.
Four studies of polymorphic markers in the visfatin gene have been reported to date. The first (12) did not report any association with T2D but did observe an association between the −948 G>T SNP and fasting insulin (P < 0.05). In the second (13), a significant association was demonstrated between rs9770242, rs1319501 and fasting plasma insulin levels (P = 0.011) and fasting glucose levels (P = 0.02 and P = 0.017, respectively). The third report (14) described an association between −948 G>T and T2D (P = 0.021), but no association with BMI, waist circumference, serum glucose levels, or fasting insulin levels. The fourth study reported no association with BMI, waist-to-hip ratio, measures of glucose, insulin, or lipid metabolism (15).
Given the evidence for a relationship between visfatin expression, BMI, and possibly T2D, the study reported here was designed to look for SNP variants in visfatin that were associated with obesity and T2D. To this end, 1,709 obese cases, 2,367 type 2 diabetic subjects, and a total of 2,850 controls were genotyped using eight haplotype-tagging SNPs and analyzed for statistical associations.
Methods and Procedures
Eight tagged SNPs in the region of NAMPT were genotyped in 4,459 unrelated French white individuals, including 2,850 lean normal glucose-tolerant subjects (BMI < 27 kg/m2 and normal glucose tolerance), 813 obese children (BMI 97th percentile for gender and age), and 896 adults with class III obesity (BMI ≥ 40 kg/m2). Obese children were collected through a multimedia campaign run by the CNRS UMR8090 (N = 645) as well as in the Toulouse Children's Hospital (N = 82) and the Paris Trousseau Hospital (N = 86). Class III obese adults were recruited through a multimedia campaign run by the CNRS UMR8090 and the Department of Nutrition of the Paris Hotel Dieu Hospital. Control subjects for the initial phase of this study were selected from participants in the DESIR (Data from the Epidemiological Study on the Insulin Resistance syndrome) prospective study (n = 2,850) (ref. 16). The DESIR study is a general population sample, which is a 9-year follow-up study that aims to clarify the development of the insulin resistance syndrome. Subjects were recruited from ten health centers in the western central part of France, and all participants signed informed consent. The study protocol was approved by all local ethics committees, and an informed consent was obtained from each subject before participating in the study. We genotyped a further 2,362 subjects for rs10487818 to include the whole of the DESIR cohort in the analysis of this SNP. In order to assess possible association of SNPs with T2D, 2,367 French European diabetic subjects were genotyped as well. Type 2 diabetic subjects were identified as currently undergoing treatment for diabetes or were diagnosed as diabetic using World Health Organization criteria. The majority of type 2 diabetic subjects were recruited while undergoing treatment at the Endocrinology-Diabetology Department at Corbeil-Essonnes Hospital (n = 2,012), whereas the Centre National de la Recherche Scientifique–Institut Pasteur Unit in Lille recruited 355 probands from families with T2D. The gender ratio was 1:1.45, male:female, and the average BMI was 30.0 ± 5.9 kg/m2 (see ref. 17 for details).
The SNPs were selected on the basis of the HapMap, using the Tagger (18) functionality within Haploview (19) to identify haplotype-tagging SNPs using an r2 > 0.8 and a minor allele frequency >0.1%. Using aggressive tagging (2- and 3-marker haplotypes), this resulted in the selection of five SNPs, capturing 20 SNPs with a mean r2 value of 0.945. In addition, two SNPs with higher minor allele frequency were added to decrease the average physical spacing of the markers in the gene and a low-frequency SNP rs10487818 was added, as this was the only other SNP reported in HapMap that was not in the tagged set at the time. The positions of these SNPs in the gene and their HapMap frequencies are detailed in Table 1.
Table 1. Haplotype tagging-SNP list for genotyping the visfatin gene
ABI SNPlex genotyping
SNPs were genotyped using the Applied Biosystems (Foster City, CA) SNPlex technology, which is based on the Oligonucleotide Ligation Assay combined with multiplex PCR amplification. Experimental methods were as per the manufacturer's instructions. Genotype assignment was performed by capillary electrophoresis analysis using an ABI 3730xl DNA Analyzer and ABI GeneMapper v3.7 software.
Genotyping was considered to be successful if the overall success rate for the SNP was >80% and the Hardy–Weinberg equilibrium did not depart significantly from the expected distribution (P > 0.05 for χ2-test between expected and actual values).
Analysis was carried out using SPSS v15 (SPSS, Chicago, IL). For the qualitative trait of obesity, association analysis was carried out using either a 3 × 2 χ2-test or a 2 × 2 Fisher's exact test, depending on the cell values. Analysis of BMI and waist-to-hip ratio was carried out using a linear regression model with age and sex included as covariates. Testing for association with T2D was carried out using binary logistic regression, adjusting for the effects of age, sex, and BMI. Results are given in the tables without correction for multiple testing, but the effects of applying Bonferroni correction are discussed.
All SNPs had >80% genotyping success rate and showed no deviation from Hardy–Weinberg equilibrium, except one. This SNP, rs6947766, had an overall (cases plus controls) P value of 0.04, indicating modest divergence from Hardy–Weinberg equilibrium. This SNP was nonetheless included in the analysis because only the controls showed any significant deviation from Hardy–Weinberg equilibrium (P = 0.02), with none of the obesity or T2D case groups deviating from equilibrium, suggesting that this was not due to genotyping error but a possible real difference between the groups. In this study, when either genotypes (Table 2) or presence or absence of alleles were analyzed (Table 3), chi-squared analysis of the obese adults and children demonstrated that only one SNP was significantly associated with obesity. Analysis of the T2D cases, after compensation for the effects of obesity, demonstrated no significant associations at all (P < 0.05) (data not shown).
Table 2. Association analysis of SNP genotypes with obesity using a 3 × 2 χ2-test
Table 3. Analysis of SNPs using minor allele presence/absence for association with obesity using Fisher's exact test
One SNP, rs1047818, exhibited significant genotypic association with the qualitative trait of severe obesity (as tested using the Fisher's exact test because of low minor allele frequency). There was a very strong protective association of this allele, which was markedly stronger in the severely obese children compared to the class III obese adults (P = 2 × 10−6 compared to P = 2 × 10−4, see Table 3). The significance of the effect was more pronounced when the analysis was carried out for the presence or absence of the rare allele (dominant model) in the obese children, again being more significant than from the obese adults (P = 6 × 10−9 vs. P = 9 × 10−5, see Table 3). No significant association with BMI or waist-to-hip ratio was identified after genotyping the DESIR cohort (n = 5,212) (data not shown). As the major association was to a low-frequency allele, no haplotype analysis was attempted for any of the traits. Linkage disequilibrium (LD) across the gene is illustrated in Figure 1 for (i) D' and (ii) R2, demonstrating negligible LD between rs1047818 and the other genotyped SNPs, due to its low frequency.
Because visfatin was identified and reported as having insulin-mimetic properties (1), research has been aimed at determining whether sequence variation in the gene is associated with T2D and/or obesity. There have been a number of reports in the literature of associations between plasma visfatin levels and BMI or plasma glucose levels (8,9,10). Three SNPs, 948 G→A, rs9770242, and rs1319501, have recently been reported to be associated with fasting plasma insulin and glucose levels and to some extent with T2D (12,13,14), although a recent study failed to demonstrate any association between common SNPs in the visfatin gene and obesity- and diabetes-related quantitative traits (15). There has also been no reported association with this gene from any of the four T2D genome-wide association studies or the two very recent Genome Wide Association scans for obesity, defined by BMI (20,21,22,23,24,25).
Our results demonstrate strong association between the rare rs10487818 SNP and protection from obesity, and this is robust to Bonferroni correction for multiple testing. Reconciling the current literature with our reported results is relatively straightforward. For obesity, the novel strong result is simply explained by the fact that in reported studies so far, researchers have excluded SNPs with a minor allele frequency of <5%. This is a standard strategy for SNP screening, based on the “common disease, common variant” hypothesis (see ref. 26 for review), but as our result demonstrates, this assumption can easily lead to failure to detect rare alleles of strong effect. No association was observed with the quantitative trait of BMI in our general population sample. The failure to detect association with BMI may reflect the fact that BMI is only one measure of the complex phenotype of obesity, but it may also be a consequence of the low frequency of the allele resulting in low statistical power to detect an effect.
For T2D, there were no significant associations detected after adjustment for BMI, in spite of the case and control groups each being over 2,000 subjects. This confirms previous findings of very few reported associations between SNPs in the visfatin gene and T2D. No genome-wide association study of T2D has so far identified the visfatin gene as positively associated. However, with very low genome-wide thresholds for reporting significant associations (e.g., P < 10−5 or less), it is possible that such data exist but have not been reported.
If rs10487818 were a marker for disease susceptibility, then it would be easy to conclude that this was an example of a relatively common monogenic disorder, such as the one that has been previously observed for certain MC4R variants and obesity (27). However, as has been recently demonstrated, rare alleles of strong effect can contribute to complex diseases such as, e.g., infectious disease, without any evidence of Mendelian inheritance patterns (28). The rare visfatin allele reported here appears to be protective, something that we have also recently reported in common obesity for the I251L and V103I variants of the MC4R gene (29). Whereas, by definition, few subjects carry any particular rare allele, if that allele is of strong effect, it may be of clinical significance as part of a large set of rare alleles that can be used to screen subjects for an overall estimate of genetic risk.
Other research carried out by us has revealed no evidence that copy number variation could account for the association we have observed (30). However, there is also no evidence that rs10487818 is the causative SNP. It is in intron 4—a region that does not appear to show any strong evolutionary conservation. There is also no evidence that it affects existing splice recognition sites or creates a cryptic splice site. At this stage, it seems likely that it is in LD with another SNP that is causative.
Interestingly, the strongest association observed for rs10487818 is when the presence or absence of the allele is tested, i.e., 1/1 vs. 1/2 + 2/2. This may just be due to the small number of alleles available for analysis but the allelic association is notably less significant (Table 3). This may reflect a nonadditive effect of the two alleles of a causative SNP that rs10487818 is in LD with, and the presence of one allele is enough to have a dominant protective effect. As it has been reported that visfatin protein forms a functional homodimer (31), it is easy to see that affecting only half the copies of protein produced may be enough to affect the function of most, if not all, of the homodimers.
Our study demonstrates a clear association to the binary trait of obesity, in apparent contrast to previous studies, where association to BMI and T2D have been reported. The low minor allele frequency of our associated SNP means that even strong LD with previously reported SNPs might not be sufficient to reveal any indirect association with T2D or BMI. In order to identify the SNP or SNPs in LD with rs10487818, a significant resequencing of the visfatin gene needs to be carried out in control samples to determine the presence of other low-frequency variants that could be responsible for the observed association. It would also be very useful to type further sets of cases and controls, both children and adults, in order to replicate this strong association.
We wish to acknowledge the invaluable contribution to this work of the patients, their families, doctors, nurses, and other staff involved in the recruitment of the subjects. We thank the FONDACOEUR charity and “le Conseil Regional Nord Pas de Calais/FEDER” for their financial support. The DESIR study has been supported by INSERM, CNAMTS, Lilly, Novartis Pharma and Sanofi-Aventis, by INSERM (Réseaux en Santé Publique, Interactions entre les determinants de la santé), by the Association Diabète Risque Vasculaire, the Fédération Française de Cardiologie, La Fondation de France, ALFEDIAM, ONIVINS, Ardix Medical, Bayer Diagnostics, Becton Dickinson, Cardionics, Merck Santé, Novo Nordisk, Pierre Fabre, Roche, Topcon. This work was funded by the UK Medical Research Council.