Initially, the phenotypic characteristics of 62 “6q-evidence” families (defined by an individual pedigree Zscore > 1.0 in the 2-LOD drop interval flanked by markers D6S434 and D6S1704) were compared with the remaining 35 families from our previously published genome scan for childhood obesity¹. The “6q-evidence” obese children have a trend towards higher area under the glycemia curve after glucose administration and a significantly lower insulinogenic index (Supplementary Table 1). Compared to none of the other families, 3.1% of the “6q-evidence” obese children are glucose intolerant or diabetic and 13.8% of 6q linked obese children parents have type 2 diabetes mellitus (T2D) compared to 3.2% of parents in other families (p=0.018). Thus, the obesity susceptibility gene(s) on chromosome 6q may be also involved in glucose homeostasis.

The “6q-evidence” 2-LOD drop interval¹ covers 41.4 Mb and includes 166 referenced genes. This was narrowed to a 2.4 Mb interval between markers D6S1656 and D6S270 using overlapping published linkage results on chromosome 6q16.1-q27 with either obesity², insulin secretion^3,4 or T2D^5-8. Within this interval, the best candidate was ectonucleotide pyrophosphatase phosphodiesterase ENPP1 (also known as the Plasma Cell glycoprotein-1 PC-1). ENPP1 is believed to directly inhibit insulin-induced conformational changes of the insulin receptor, thereby affecting its activation and downstream signaling^9,11.

The microsatellite marker D6S1656 in intron 1 of the ENPP1 gene, linked to childhood obesity in our initial genome scan, was analyzed in a second replication set of 68 families with childhood obesity and modest evidence of linkage (MLS=0.83, p=0.04) was observed. Allele 10 of D6S1656 was also significantly under-transmitted to affected children in both the initial and replication sets (44 transmitted vs 69 untransmitted, p uncorrected =0.02).

All coding regions of the ENPP1 gene were sequenced, plus 1.3 Kb upstream of the ATG start codon and downstream of the TGA stop codon in 48 obese children from “6q-evidence” families and in 24 non-obese adults. Eight single-nucleotide polymorphisms (SNPs) were identified in the upstream sequence, six in intron/exon junctions, four were missense, three were synonymous mutations and twenty were identified in the downstream sequence (Supplementary Table 2). Among these forty-one polymorphisms, twenty-four were present in the public databases and twenty-two of the variants had a minor allele frequency (MAF) higher than 5%. Pairwise LD among the twenty-four most common SNPs was used to select a set of ten “haplotype tagging” SNPs to be typed in the whole set of samples.

Twenty-five intronic or intergenic fragments showing a high degree of homology (>70%) across Fugu rubripes, rat and human genomes were also sequenced and a further eleven SNPs of MAF >5% were identified. One SNP, T>G +5954 TGA, that showed a trend for association with childhood obesity (p<0.1) in a test set of 421 obese children and 298 control individuals was added to the set of SNPs analyzed in the whole sample set.

Eleven SNPs were genotyped in 2,430 individuals, made up of 529 unrelated obese children, 696 unrelated morbidly obese adults, and 1,205 lean normoglycaemic adults. Lean adults were used as controls for both sets of cases because they demonstrated a long term resistance against obesity. Associations were observed between severe forms of obesity and six of the SNPs: IVS2 delG +8, K121Q, IVS8 T>G +27, IVS20 delT –11, A>G +1044 TGA and T>G +5954 TGA (0.00008<p<0.03; 1.21<OR<1.37) (see Supplementary Table 3, Table 1). The global p-value, assessed by 10⁵ permutations of the obesity status among individuals, was 0.001. Analysis of the pooled data identified the strongest association with severe forms of obesity of the K121Q SNP (OR=1.37, 95% confidence interval [1.17-1.61], p=0.00008). The odds ratio under a recessive model increased to 3.29 [1.83-5.93] (p=0.00003) with a significant departure from the additive model (p=0.02).

A further 184 multiplex families were then genotyped for K121Q. This consisted of the 97 childhood obesity genome scan families plus 87 nuclear families with adult severe obesity, which showed linkage of serum leptin levels to chromosome 6q24¹². Using the Transmission Disequilibrium Test (TDT) the 121 Q-allele was significantly over-transmitted to obese offspring (transmitted: 76, non-transmitted: 48, p=0.01), supporting the case-control result.

To confirm ENPP1 specificity for these associations, a total of fifty-three SNPs in the chromosome 6q region, spanning 580 kb and including the ARG1, CRSP3, ENPP3, ENPP1 and CTGF genes, were typed in the initial set of 421 obese children and 298 control individuals used above (average density: 1 SNP/10.9 kb). Three distinct regions of linkage disequilibrium (LD) were noted: the first contained ARG1, CRSP3 and ENPP3, the second ENPP1 alone, and the third the CTGF gene and 176Kb of non coding region (Fig. 1). Eight SNPs were associated with both childhood obesity and “6q-evidence” childhood obesity (p<0.05). Seven of the eight SNPs (K121Q, Celera dbSNP hcV1207989, C>T +164 TGA, Celera dbSNP hcV1207974, A>G +1044 TGA, G>T +1101 TGA, C>T +1157 TGA) mapped within ENPP1 (Fig. 2), suggesting that the observed association with childhood obesity is due to ENPP1 SNPs.

Open in a separate window

Figure 1

Pairwise LD between fifty-three SNPs in a 580 kb region including the ENPP1 gene in 421 obese children and 298 control individuals. Regions of high and low LD (delta) are presented by red and blue shading, respectively. The graph is to the physical map scale.

Open in a separate window

Figure 2

Allelic association of fifty-three SNPs in the 580 kb region including the ARG1, CRSP3, ENPP3, ENPP1 and CTGF genes. Blue circles correspond to the −log₁₀ (p-value) for the comparison of allelic distribution between 421 obese children and 298 control individuals. Red circles correspond to the −log₁₀(p-value) for the comparison of allelic distribution between 62 “6q-evidence” families obese children and 298 control individuals. The figure is not to scale according to the physical map location; indeed the SNPs are equidistant to allow easy visualization of the association test results.

A two SNP analysis between the K121Q polymorphism and the 6 other obesity-associated SNPs was used to assess whether these SNPs had an independent effect on the risk of obesity. A likelihood ratio test showed that only IVS20delT-11 and G+1044TGA SNPs significantly modulate the effect of K121Q (p=0.03 and p=0.04 respectively). It is worth pointing out that for these two SNPs, the model with the best fit was also a recessive one. This may account for the small observed deviation from Hardy-Weinberg Equilibrium (HWE) for these SNPs as genotyping errors have been ruled out by resequencing (data not shown).

To estimate the potential effects of combinations of the three SNPs on the risk of obesity, haplotype analysis was performed using the K121Q, IVS20 delT-11 and A>G +1044 TGA SNP data from the whole set of 2,430 French Caucasian subjects. Eight haplotypes were predicted, five having a MAF >5%. The three allele wild-type haplotype (K121Q/IVS20 delT-11/A>G +1044 TGA, KTA), was less frequent in obese subjects than in controls (60.3% vs 64.0%, p=0.002) (Table 2). In contrast, the three allele risk haplotype (K121Q/IVS20 delT-11/A>G +1044 TGA, QdelTG), was strongly associated with severe forms of obesity (11.2% vs 7.5%, OR=1.58, p=0.00001, empirical p-value<0.0001 for 10⁵ simulations). Interestingly, the risk haplotype effect was of similar magnitude in both morbidly obese adults and childhood obesity (10.8% vs 7.9%, OR=1.50, p=0.006, and 11.7% vs 7.1%, OR=1.69, p=0.0006, respectively). The association was also supported by TDT analysis in the 184 families (Chi2=5.68, p=0.01, Table 2) whereas transmission distortion was excluded in a set of 458 French Caucasian trios with unaffected children (Chi2=0.53, p=0.46)¹³. Finally, the risk haplotype was associated with obesity (10.3% vs 7.9%, OR=1.37, p=0.02) in an additional set of 717 adult subjects with a less severe form of obesity (BMI between 30 and 40 kg/m²).

The impact of the ENPP1 risk haplotype on the linkage with childhood obesity observed in the genome scan was identified in several ways. Initially, a higher frequency of the risk haplotype in individuals from “6q-evidence” families (16.2% vs 7.1%, OR=2.37, p=0.004) compared to other families (12.2% vs 7.1%, OR=1.65, p=0.16) was detected. After removing 15 affected sib-pairs sharing the risk haplotype (total sib-pairs = 135), the multipoint MLS LOD score dropped from 4.06 to 1.6 at marker D6S287, and a new maximal score of 2.63 appeared 16-Mb centromeric to the original linkage peak, at marker D6S301. The Genotype IBD Sharing Test (GIST)¹⁴ also suggested a trend for a possible effect of the haplotype under an additive model (p=0.07) which became significant for a recessive model (p=0.03).

At least three ENPP1 SNPs are involved in the association with obesity. The Q121 variant is believed to inhibit insulin signalling¹⁵ more effectively than the wild-type version, but the functional effects of the risk haplotype are unknown. The ENPP1 protein has a proteolytic cleavage site, is cleaved at the surface of cells, and is known to be present in the circulation¹⁶. Thus, the protein serum level represents a good estimation of its tissue expression¹⁶. Serum ENPP1 protein levels measured in 279 children encompassing a wide weight range showed a positive correlation with the Z score of BMI (Pearson correlation coefficient=0.1, p=0.05, Supplementary Figure 1). Eighty-nine lean children were then selected (mean BMI, 18.4 ± 2.5 kg/m²; mean age, 13.3 ± 2.6 y; n=50/39 girl/boy), to fix the confounder BMI, and analyzed for the effect of the three SNPs on ENPP1 levels. The presence of at least one copy of Q121, IVS20delT-11 and G+1044TGA alleles was associated with a highly significant increase of ENPP1 levels (28.6 ng/ml vs 24.1 ng/ml, p=0.008) (Fig. 3), suggesting that the obesity-associated haplotype not only impairs insulin binding but also enhances ENPP1 levels of expression.

Open in a separate window

Figure 3

ENPP1 serum level in 89 lean children according to the presence of the risk haplotype.

The contribution of the risk haplotype on the variation of obesity-related phenotypes was then assessed in 474 obese children where data were available. Obese children with the obesity-associated risk haplotype showed a 0.17 mmol/l increase in fasting glycemia (p=0.002) with a higher prevalence of glucose intolerance/T2D (OR=3.43, p=0.02). Parents carrying the risk haplotype had a 2.35 fold increased risk to develop T2D (p=0.005). This risk was higher in the subset of obese parents (OR=3.26, p=0.0005) with no increase in the risk of T2D observed in non obese parents carrying the risk haplotype (OR=0.86, p=0.9). An additional non-overlapping cohort of 752 unrelated T2D French Caucasian subjects with familial history of the disease was compared to the previously used 556 middle aged non-obese normoglycemic subjects (average age: 55±6 years). The T2D group had a significant excess of the risk haplotype (10.7% vs 7.1%, OR=1.44, p=0.005) further supporting a potential effect of ENPP1 SNPs on glucose homeostasis in French Caucasians. This finding was replicated using 1261 unrelated Austrian subjects consisting of 503 T2D subjects and 758 non obese normoglycemic subjects (9.8% vs 6.3%, OR=1.68, p=0.001). The Mantel-Haenszel adjusted odds ratio, under a fixed effects model, was used in the pooled cohorts of 2569 European subjects, and strengthened the association of the risk haplotype with T2D (10.4% vs 6.6%, combined OR=1.56, p=0.00002). In summary, these findings indicate that both obesity and T2D, especially in obese subjects, are associated with ENPP1 genetic variability and specifically one three-allele risk haplotype. This provides genetic evidence for the recently described link between obesity in childhood and the high risk for T2D in the teens or early adulthood¹⁷, providing the first common molecular mechanism for this deleterious association.

RT-PCR was performed on cDNAs from a wide range of human tissues including brain, muscle, liver, adipocyte (subcutaneous and omental), and purified pancreatic islet beta-cells. Taking into account that five ENPP1 isoforms are known, (http://www.ncbi.nih.gov/IEB/Research/Acembly/av.cgi?db=human&l=ENPP1), primers were designed to amplify the region between exons 7 and 12, common to at least 3 transcripts, and ubiquitous expression was found (Supplementary Figure 2). Primers were then designed to specific amplify the long mRNA isoform, (larger 3′UTR with 1170 bases downstream of the TGA stop codon) that includes the obesity-associated SNP A>G +1044 TGA. The long form was found to be only expressed in pancreatic beta-cells, adipocytes and liver, three key tissues for glucose homeostasis (Supplementary Figure 2).

The contribution of the obesity-associated ENPP1 risk haplotype to our observed linkage was assessed by several methods but only a moderate excess of transmission of this haplotype to affected offspring was found. This suggests that the risk haplotype is contributory but not sufficient to explain the linkage with childhood obesity. Additional SNPs in the non-coding regions of the ENPP1 locus may account for part of the observed linkage, including SNP T>G +5954 TGA, located in a highly conserved region. This variation predicts loss of binding to the Insulin Promoter Factor IPF1. Alternatively, more than one gene may explain the linkage with obesity and T2D on 6q. Recent data supports a contribution of ENPP1 SNPs¹⁸ to linkage for insulin fasting levels in Mexican-Americans³.

Only the ENPP1 K121Q exon 4 missense mutation has been suggested to associate with insulin resistance or T2D in limited studies¹⁹. The 3′UTR (A>G +1044 TGA) SNP belongs to an isoform specifically expressed in three highly insulin-responsive human tissues (pancreatic islet beta-cell, adipocyte and liver). Mice given an adenovirus expression construct overexpressing this gene in hepatocytes show insulin resistance and glucose intolerance¹⁰. Although the exonic 121Q amino acid substitution directly inhibits insulin receptor by a non enzymatic mechanism¹⁵, the other non-coding SNPs may have their effect by modifying gene expression, protein production or splicing. This hypothesis is favoured by the increased serum protein levels in children carrying the ENPP1 obesity risk haplotype and by the rise of ENPP1 levels with adiposity.

Higher protein expression may mimic the effects of insulin receptor inactivation in the brain where insulin has potent anorectic actions, or in the skeletal muscle, both leading to an increased fat mass^20,21. Obese children carrying the ENPP1 risk haplotype often have glucose intolerance and a family history of T2D. This suggests that the exaggerated insulin resistance conferred by inherited increased ENPP1 expression in the context of a Westernized obesogenic environment, may contribute to excessive fat accumulation. The ENPP1 Q121 allele was recently associated with increased BMI in the UK general population²². Data presented here supports the view of a causative effect of primary insulin resistance on childhood obesity. According to this hypothesis, insulin resistance-induced fasting hyperinsulinemia was shown to be a strong predictor for the subsequent development of obesity in children of various ethnic groups²³.

In conclusion, this study strongly supports a genetic link between ENPP1 gene variants and chromosome 6q-linked childhood polygenic obesity and also with adult obesity and T2D. This provides an insight into the molecular basis for the physiologic association between insulin resistance and obesity, and presents a new perspective for prevention and treatment of these conditions.