Variation in Glucose Homeostasis Traits Associated With P2RX7 Polymorphisms in Mice and Humans

Abstract

Context:

Extracellular nucleotide receptors are expressed in pancreatic B-cells. Purinergic signaling via these receptors may regulate pancreatic B-cell function.

Objective:

We hypothesized that purinergic signaling might influence glucose regulation and sought evidence in human studies of glycemic variation and a mouse model of purinergic signaling dysfunction.

Design:

In humans, we mined genome-wide meta-analysis data sets to examine purinergic signaling genes for association with glycemic traits and type 2 diabetes. We performed additional testing in two genomic regions (P2RX4/P2RX7 and P2RY1) in a cohort from the Prevalence, Prediction, and Prevention of Diabetes in Botnia (n = 3504), which includes more refined measures of glucose homeostasis. In mice, we generated a congenic model of purinergic signaling dysfunction by crossing the naturally hypomorphic C57BL6 P2rx7 allele onto the 129SvJ background.

Results:

Variants in five genes were associated with glycemic traits and in three genes with diabetes risk. In the Prevalence, Prediction, and Prevention of Diabetes in Botnia study, the minor allele in the missense functional variant rs1718119 (A348T) in P2RX7 was associated with increased insulin sensitivity and secretion, consistent with its known effect on increased pore function. Both male and female P2x7-C57 mice demonstrated impaired glucose tolerance compared with matched P2x7-129 mice. Insulin tolerance testing showed that P2x7-C57 mice were also less responsive to insulin than P2x7-129 mice.

Conclusions:

We show association of the purinergic signaling pathway in general and hypofunctioning P2X7 variants in particular with impaired glucose homeostasis in both mice and humans.

Diabetes mellitus is a major contributor to morbidity and mortality worldwide. Type 2 diabetes, the most prevalent subtype, results from insufficient insulin secretion in the setting of reduced insulin sensitivity; strong evidence exists for an inherited component of type 2 diabetes risk. In contrast to classic monogenic disorders, the genetic architecture of type 2 diabetes appears to be comprised of multiple variants, each with a modest impact on diabetes risk. Despite substantial progress in our knowledge of the genetic basis of type 2 diabetes with genome-wide association studies (GWAS), there remains a large portion of unexplained genetic heritability (1). In addition, although GWAS efforts identify genomic regions that contain significant association signals, it is rarely evident how these loci contribute to type 2 diabetes pathogenesis. Candidate gene studies, particularly those informed by animal models and in vitro mechanistic studies, have the advantage of beginning with a plausible hypothesis that is testable functionally.

ATP is the principal molecule of energy metabolism. Although almost all ATP and its metabolites are found inside cells, low levels of ATP can be measured in the extracellular space (2). Discovery of two families of receptors for extracellular nucleotides (P2X and P2Y) has confirmed a signaling role for extracellular ATP (3, 4). Extracellular nucleotide levels change dramatically in certain conditions such as inflammation, cellular injury, or active cellular release (5, 6) and are also subject to active regulation by membrane-bound ectonucleotidases (7). Mice null for several extracellular nucleotide (purinergic) receptors have abnormal glucose homeostasis traits (8, 9). Both P2X and P2Y receptors are expressed in pancreatic β-cells. Moreover, ATP is a component of secretory granules and is released by β-cells during glucose stimulation. ATP signaling helps organize insulin secretion pulsatility by coordinating the response to glucose by different islets (10).

We hypothesized that purinergic signaling influences glucose regulation. We screened publicly available databases of human subjects for evidence that purinergic signaling genes might influence either type 2 diabetes risk or glycemic traits. We performed additional association testing of missense variants in candidate genes in a cohort with more refined measurements of glycemic control. Taking advantage of a naturally occurring hypomorphic mutation in P2X7, a receptor strongly expressed on pancreatic islets (9–11), we generated a mouse model of purinergic signaling dysfunction and examined its effect on glucose homeostasis. Together our data show evidence for purinergic signaling in the regulation of blood glucose levels.

Materials and Methods

Mining of publicly available GWAS databases

The Meta-Analysis of Glucose and Insulin-Related Traits Consortium (MAGIC) and the Diabetes Genetics Replication and Meta-analysis (DIAGRAM) consortiums were both large-scale collaborative meta-analyses of multiple GWAS studies. DIAGRAM combined case-control data of 12 171 type 2 diabetes cases and 56 862 controls (1). In MAGIC, data were combined from up to 46 186 subjects for fasting glucose, fasting insulin, and homeostasis model assessments of β-cell function and insulin resistance (HOMA-IR) (12); 15 234 subjects for glucose level 2 hours after the glucose load (2 h glucose) (13); and 10 701 subjects for fasting proinsulin (14) (further methodological details are available in the primary publications). The institutional review boards of the respective cohorts' institutions approved the study protocols, and informed consent was obtained from each subject prior to participation.

We identified all known single-nucleotide polymorphisms (SNPs) within each of our 18 candidate genes: the seven P2X genes (P2RX1, P2RX2, P2RX3, P2RX4, P2RX5, P2RX6/P2RXL, P2RX7), the eight P2Y genes (P2RY1, P2RY2, P2RY4, P2RY6, P2RY11, P2RY12, P2RY13, P2RY14) and three ectonucleotidase genes (ENTPD1, ENTPD2, ENTPD3). We obtained effect estimates and P values for all SNPs provided for public download from both the DIAGRAM and MAGIC databases and examined each for association with type 2 diabetes risk and glycemic quantitative traits, respectively. We calculated an a priori significance threshold for each gene using a Bonferroni correction for the number of independent tests taking linkage disequilibrium (LD) among SNPs into account (Supplemental Table 1), following the method proposed by Nyholt (15) and Li and Ji (16).

Testing for association with detailed measures of glucose homeostasis

We performed additional testing in two genomic regions (P2RX4/P2RX7 and P2RY1 loci) in a cohort from the Prevalence, Prediction and Prevention of Diabetes in Botnia (PPP-Botnia) (17), which includes the more refined measures of glucose homeostasis not present in MAGIC, namely the insulin sensitivity index (ISI), corrected insulin response (CIR), and two disposition indices (ISI × CIR and ISI × insulinogenic index). The local research ethics committees approved the studies, and all participants gave informed consent. Genotyping on 3252 individuals was performed by allelic discrimination with a TaqMan assay on the ABI 7900 platform (Applied Biosystems), with an average call rate of 95.6%. All SNPs except for one were in Hardy-Weinberg equilibrium (rs25644, Hardy-Weinberg equilibrium, P = .006).

Phenotype-genotype associations were analyzed using linear regression analyses assuming additive models, with the minor allele set as the effect allele. Where an additive association was found, this analysis was followed up by analyses with dominant and recessive models. All nonnormally distributed phenotypes were natural log transformed prior to analysis. All analyses were adjusted for age, sex, and, when applicable, body mass index. Statistical analyses were performed using SPSS version 20.0.0 (SPSS Inc) and R version 2.14.2 (http://www.R-project.org/).

Mouse model

Some mouse strains have a P2rx7 polymorphism with functional consequences that mirror the human polymorphism A348T. Substitution of a leucine for proline at amino acid 451 in the C-terminal tail of P2X7 results in decreased calcium flux and pore-forming capacity (18). C57BL6 mice are highly insulin resistant, whereas 129SvJ mice are insulin sensitive, a difference likely due to a large number of loci (19, 20). Because these strains differ at the P2rx7 locus, we hypothesized that this might contribute to the difference in insulin sensitivity and created congenic (ie, differing at only one locus) mice on a 129SvJ background containing either the 129SvJ or C57BL6 P2rx7 allele but otherwise essentially identical.

Mice were purchased from Taconic. P2X7 congenic mice were generated by crossing 129SvJ mice (with proline at amino acid 451 of P2X7) with C57BL6 (N substrain) mice that have a naturally hypomorphic P2X7 variant (leucine at amino acid 451) with minimal pore-forming capacity (18, 21). Mice heterozygous for the P2X7 P451L substitution were backcrossed 8–11 times onto a 129SvJ background, genotyping for the status of the gene P2rx7 at each backcross. We bred P2rx7 heterozygotes and performed all experiments with littermates either homozygous for the hypomorphic C57BL6 P2X7 or homozygous for the fully functional 129SvJ P2X7 variant at this locus. The two congenic strains are approximately 99.9% identical by descent, with most or all of the remaining difference at the P2rx7 locus. Any remaining genetic differences at other loci should be evenly distributed between groups because littermates were used.

Glucose tolerance tests (GTTs) were performed at 8 weeks; both GTT and insulin tolerance testing (ITT) were performed at 16–20 weeks (females) and 24 weeks (males). For the GTT, we injected mice that had been fasted overnight with 1 g/kg of glucose and measured glucose levels with the Ultra One Touch system (LifeScan, Inc) at 0, 15, 30, 60, and 120 minutes; area under the curve (AUC) was calculated for each mouse. For the ITT, we injected mice with recombinant human insulin (females, 0.33 U/kg; males, 0.5 U/kg) after a 4-hour fast and measured glucose at 0, 15, 30, 45, 60, and 90 minutes. We normalized the curves based on starting (t = 0) glucose and measured the area above the curve for each mouse. All results were analyzed by t test (GraphPad Prism software).

Results

To examine the effect of purinergic signaling genes on glucose homeostasis and type 2 diabetes risk in humans, we searched the publicly available MAGIC and DIAGRAM databases. We examined a total of 488 variants within our 17 candidate genes for association with seven glycemic traits. For the glucose homeostasis traits, we found five variants that reached the a priori level of significance (Table 1): the minor allele (T) of rs7861020 (ENTPD2) was associated with lower 2-hour glucose level on a 2-hour oral glucose tolerance test, whereas the minor allele (C) of rs16864613 (P2RY1) was associated with the elevated 2-hour glucose level; the minor alleles (both T) of rs4938864 (P2RX3) and rs11655312 (P2RX5) were associated with a higher fasting glucose level, and the minor allele (A) of rs11065464 (P2RX7) was associated with an elevated fasting proinsulin level. All of these variants are located in intronic regions, with the exception of rs16864613, which is located in the 3′ untranslated region of P2RY1.

Table 1.

Nominally Significant SNPs Associated With Glucose Homeostasis Traits

SNP	Gene	Effect/Other Allele	Trait	β	P Value
rs7861020	ENTPD2	T/G	2-Hour glucose	−0.072	.0048
rs4938864	P2RX3	T/C	Fasting glucose	0.017	.0031
rs11655312	P2RX5	T/C	Fasting glucose	0.022	.0047
rs11065464	P2RX7	A/C	Proinsulin	0.028	.0013
rs16864613	P2RY1	C/G	2-Hour glucose	0.150	.0015

For type 2 diabetes risk, we found 14 variants in three genes that reached the a priori level of significance (Table 2), all of which were located in intronic regions. In P2RX3, we found a block of variants significantly associated with diabetes risk: rs4938863, rs1815769, and rs4938864; of note, the allele of rs4938864 associated with increased diabetes risk (T) was associated with higher fasting glucose levels in MAGIC. In P2RX4, there were two independent variants that reached significance: rs7298368, rs10774588, and two groups of significant variants in perfect LD: rs1629287, rs10849856, rs1169727, and rs1180051, and rs1151880, rs3815989, rs4980998, and rs10849859. In P2RX5 there was one significant variant, rs222774.

Table 2.

Nominally Significant SNPs Associated With Type 2 Diabetes Risk

SNP	Gene	Risk/Other Allele	Odds Ratio (95% CI)	P Value
rs4938863	P2RX3	G/A	1.08 (1.02–1.14)	.0047
rs1815769	P2RX3	C/G	1.08 (1.02–1.14)	.0043
rs4938864	P2RX3	T/C	1.09 (1.03–1.15)	.0034
rs7298368	P2RX4	T/C	1.09 (1.04–1.15)	3.2 × 10−04
rs1629287	P2RX4	A/G	1.07 (1.03–1.11)	6.0 × 10−04
rs10849856	P2RX4	T/A	1.07 (1.03–1.11)	7.3 × 10−04
rs1169727	P2RX4	G/A	1.07 (1.03–1.11)	7.5 × 10−04
rs1151880	P2RX4	C/T	1.10 (1.04–1.16)	7.9 × 10−04
rs3815989	P2RX4	T/C	1.09 (1.03–1.16)	.0015
rs4980998	P2RX4	T/C	1.09 (1.03–1.15)	.0019
rs1180051	P2RX4	G/A	1.05 (1.02–1.09)	.0027
rs10774588	P2RX4	A/G	1.08 (1.03–1.13)	.0028
rs10849859	P2RX4	T/C	1.08 (1.03–1.14)	.0031
rs222774	P2RX5	G/A	1.06 (1.02–1.10)	.0035

Abbreviation: CI, confidence interval.

The P2RX4 and P2RX7 genes are in an area of strong LD on chromosome 12 and of all genes examined contain an unusual proportion of missense coding variation (which may be more amenable to functional experimental validation). Because both showed evidence of association with glycemic traits and diabetes risk, we tested potentially functional SNPs in this region for association with glucose homeostasis traits in 3504 subjects from the PPP-Botnia study (Table 3). These tests included more sensitive measures of glucose homeostasis such as the ISI, CIR, and disposition index (DI_{CIR × ISI} and DI_{CIR × IG30}, see Materials and Methods) as well as proinsulin measurements. A missense mutation in P2RX4, rs25644 (S258G), was associated with a lower 2-hour glucose and a 2-hour proinsulin under the additive model, a lower 2-hour glucose, proinsulin, and 2-hour insulin and higher CIR and DI_{CIR × ISI} under the recessive model and lower IG₃₀ under the dominant model; this SNP is in perfect LD with SNPs associated with diabetes risk in Table 2, potentially representing the same signal. The P2RX7 missense mutation rs17525809 (V76A) was associated with lower fasting glucose under the additive and recessive models and lower HOMA-IR under the dominant model. rs208294 (Y155H) was associated with higher 2-hour insulin levels under both the additive and recessive models. rs7958311 (R270H) was associated with higher 2-hour glucose and 2-hour proinsulin under the additive model, higher 2-hour glucose under the recessive model, and higher fasting insulin and HOMA-IR and lower ISI under the dominant model. rs1718119, a P2X7 receptor missense mutation (A348T) with previously demonstrated altered pore function (22, 23) was associated with lower fasting insulin, HOMA-IR, and 30-minute glucose and higher ISI and DI_{CIR × ISI} under the additive model and with lower fasting insulin, HOMA-IR, and 30-minute glucose and higher ISI lower fasting insulin, HOMA-IR, and 30-minute glucose and higher ISI, DI_{CIR × ISI} and DI_{CIR × IG30} under the recessive model. rs2230912, another P2X7 receptor missense mutation (Q460R), was associated with higher 2-hour glucose, insulin, and proinsulin under both the additive and recessive models and a higher DI_{CIR × IG30} under the dominant model. There was no association of rs2230911 (T357S) or rs3751143 (E496A) with glycemic traits in the PPP-Botnia panel.

Table 3.

Missense SNPs in P2RX4 and P2RX7 Nominally Associated (P < .05) With Glucose Homeostasis Traits in the PPP Panel of 3504 Subjects

SNP	Gene/AA Change	MAF (Minor Allele)	Trait	Model	β	P Value
rs25644	P2RX4	0.13 (G)	2-Hour glucose	A	−0.156	.005
	S258G		2-Hour proinsulin	A	−0.049	.006
			2-Hour glucose	R	−0.216	.001
			2-Hour insulin	R	−0.075	.026
			2-Hour proinsulin	R	−0.070	.001
			CIR	R	0.066	.040
			DICIR × ISI	R	0.083	.015
			IG30	D	−0.202	.039
rs17525809	P2RX7	0.05 (C)	Fasting glucose	A	−0.087	.005
	V76A		Fasting glucose	R	−0.086	.008
			HOMA-IR	D	−0.372	.036
rs208294	P2RX7	0.39 (A)	2-Hour insulin	A	0.048	.019
	Y155H		2-Hour insulin	R	0.077	.008
rs7958311	P2RX7	0.24 (A)	2-Hour glucose	A	0.094	.028
	R270H		2-Hour proinsulin	A	0.031	.022
			2-Hour glucose	R	0.107	.044
			Fasting insulin	D	0.082	.036
			HOMA-IR	D	0.093	.024
			ISI	D	−0.090	.019
rs1718119	P2RX7	0.41 (A)	30-Minute glucose	A	−0.098	.012
	A348T		Fasting insulin	A	−0.036	.009
			HOMA-IR	A	−0.039	.006
			ISI	A	0.029	.031
			DICIR × ISI	A	0.050	.016
			30-Minute glucose	R	−0.137	.017
			Fasting insulin	R	−0.053	.008
			HOMA-IR	R	−0.059	.005
			ISI	R	0.040	.041
			DICIR × ISI	R	0.079	.009
			DIISI × IG30	R	0.062	.037
rs2230911	P2RX7	0.13 (G)	None
	T357S
rs2230912	P2RX7	0.14 (G)	2-Hour glucose	A	0.127	.019
	Q460R		2-Hour insulin	A	0.086	.003
			2-Hour proinsulin	A	0.040	.022
			2-Hour glucose	R	0.165	.007
			2-Hour insulin	R	0.103	.002
			2-Hour proinsulin	R	0.046	.019
			DI (IG30 × ISI)	D	0.199	.044
rs3751143	P2RX7	0.11 (G)	None
	E496A
rs16864613	P2RY1	0.04 (G)	2-Hour glucose	A	−0.258	.008
	3'-UTR		2-Hour glucose	R	−0.261	.010
			DICIR × ISI	D	0.823	.036
			DIISI × IG30	D	0.960	.012

Abbreviations: A, additive model; D, dominant model; MAF, minor allele frequency; PPP, Prevalence, Prediction and Prevention; R, recessive model.

We were also able to test the P2RY1 SNP in the 3′ untranslated region, rs16864613, that previously associated in MAGIC (2 h glucose) and found an association in PPP-Botnia with lower 2-hour glucose under both the additive and recessive models and higher DI_{CIR × ISI} and DI_{CIR × IG30} under the dominant model.

The above human findings provided experiment-wide but not genome-wide statistical significance; supportive evidence was sought through functional studies in a congenic mouse model. There is a natural polymorphism in mice of several strains with a severely hypomorphic P2rx7 gene product (18, 21). The hypomorphic mouse P2X7 channel has diminished pore function, allowing decreased flux through the cell membrane of molecules up to 900 Da in size, similar to the properties of known human P2RX7 common coding variants (22). We created congenic mice by introgressing the hypomorphic C57BL6 P2rx7 allele with impaired pore function onto the 129SvJ background and tested these congenic mice for abnormalities in glucose homeostasis.

There were no significant differences in weights, fasting glucose, or fasting insulin levels for mice with the two different P2rx7 alleles at baseline, although mice with the P451L variant tended toward higher insulin and glucose levels (Supplemental Table 2). The GTT uncovered significantly impaired glucose tolerance in the P451L variant mice (Figure 1). At 8 weeks of age, the glucose AUC was 17% larger for female P451L-variant mice than for wild-type (129SvJ) mice (17675 ± 951 minmg/dL vs 15 103 ± 715 minmg/dL; P < .05). For older mice at 16–20 weeks of age, the glucose AUC was 13% larger for P451L variant mice than wild-type mice (18333 ± 618 minmg/dL vs 16 167 ± 484 minmg/dL; P < .05). Similar results were seen for male mice (Supplemental Figure 1). Impaired glucose tolerance suggested increased insulin resistance or decreased insulin secretion in the P451L variant mice.

Figure 1.

Glucose tolerance tests in female P2rx7 congenic mice. A, A GTT in 8-week-old female P451L mice (filled circle, solid line) and wild-type mice (open circle, dashed line) (n = 11 per group). B, Area under the curve for P451L mice (black bar) and wild-type mice (white bar) for GTTs in panel A. C, GTT in 16- to 20-week-old female P451L mice (filled circle, solid line) and wild-type mice (open circle, dashed line). D, AUC for P451L mice (black bar) and wild-type mice (white bar) for GTTs in panel C (n = 10–11 per group). Wild-type mice demonstrate a smaller rise in blood glucose after glucose bolus than P451L mice, indicating greater glucose tolerance.

We tested insulin tolerance and found that the P451L variant mice were less responsive to insulin (Figure 2). The area above the curve for the ITT was 3109 ± 194 minmg/dL for wild-type vs 2260 ± 185 minmg/dL for P451L variant (P < .005), a difference of 27%, in female mice. Again, similar results were seen for male mice (Supplemental Figure 2). We also tested whether there was any difference in insulin secretion after a glucose bolus but found no significant differences between the two congenic strains at several different glucose doses and measurement time points (Supplemental Table 2).

Figure 2.

Insulin tolerance tests in female P2rx7 congenic mice. A, ITT in 16- to 20-week-old female P451L mice (filled circle, solid line) and wild-type mice (open circle, dashed line) (n = 12–14 per group). B, Quantification of decrease in blood glucose after insulin injection for P451L mice (black bar) and wild-type mice (white bar) for ITTs in panel A. Wild-type mice are more sensitive to insulin than P451L mice.

After presentation of the above findings at a scientific meeting (24), it was brought to our attention that P2RX7 had been among the top loci (P < 10⁻⁵) identified in the initial stage of a recently completed GWAS of type 2 diabetes in Chinese subjects, but no significant association was seen in either of the two replication cohorts (25). Additional genotyping of rs1718119 in the Chinese population did not yield a significant association either with risk of developing type 2 diabetes or with glucose homeostasis traits in control subjects; however, the minor allele of rs1718119 is in rather low frequency (0.09) in the Chinese population, compared with PPP-Botnia (0.41), limiting the power to see statistically significant associations in this cohort.

Discussion

Genetic studies have identified several gene variants associated with glucose homeostasis, but the vast majority of the genetic component of these traits remains unexplained (26). Here we hypothesized that a pathway involved in sensing the extracellular levels of ATP, the principal molecule of energy metabolism, might play a role in glucose homeostasis. We tested SNPs in purinergic signaling genes for association with glucose regulation in two large-scale consortia of human participants, and we found a suggestive association of several genes with both glycemic traits and type 2 diabetes risk. In a smaller human cohort that included more specialized measures of glucose homeostasis, we identified several missense variants in P2RX7 that nominally associate with altered glycemic control. In a congenic mouse model with either a normal or hypomorphic allele of the extracellular ATP receptor P2X7, we found significant differences in glucose regulation under stress (glucose or insulin tolerance tests) but not under basal conditions.

Although we have corrected for the number of independent tests performed under a frequentist framework, our human genetic associations do not reach the commonly accepted threshold of genome-wide statistical significance (P = 5 × 10⁻⁸). This threshold is derived from the empirical observation of approximately 10⁶ independent tests among common SNPs in the European genome (27) and is derived from the Bayesian perspective of an equal prior probability of association for all common variants. Under the same Bayesian perspective, animal experiments that implicate a specific gene or pathway in the same phenotype raise the prior probability, such that the genome-wide threshold becomes overly stringent. The extent to which the use of prior probabilities may be incorporated into statistical tests of genetic association remains a matter of active investigation (27); our results should be interpreted in that light.

P2X7 is a ligand-gated ion channel that is activated by extracellular ATP (28). When bound by ATP, P2X7 fluxes cations from the extracellular space, leading most notably to a rise in intracellular calcium (29). P2X7 has a second functional property not shared by the other P2X family receptors: prolonged ATP binding leads to development of large pores in the cell membrane of a size that will permit passage of molecules up to 900 Da (28). These P2X7 initiated pores, probably formed by pannexins (30), can lead to a wide array of cellular responses such as inflammasome activation, release of inflammatory or antiinflammatory mediators, cell death, and other processes (31, 32). There is also evidence linking P2X7 to β-cell function and insulin release (9, 10), although it has not been determined whether this effect is due to calcium flux or pore-forming capacity.

Our genetic association studies appear to substantiate the relevance of P2RX7 in glucose regulation in humans. rs1718119 codes for an alanine to threonine substitution at amino acid 348 of P2X7, located in a transmembrane domain near the long intracellular C-terminal tail of P2X7. In vitro, others have shown the threonine variant to have increased pore function of approximately 2.5-fold compared with the alanine variant (22, 23). Carriers of the threonine variant are more insulin sensitive and demonstrate improved disposition indices in our studies. The DI relates the insulin secretion to insulin resistance, particularly the ability of islets to compensate for insulin resistance (33). Because DI measurements require insulin measurements at multiple GTT time points, few studies measure this key trait, limiting our ability to replicate our findings with sufficient statistical power. In a cohort of greater than 3000 well-characterized subjects, the association of rs1718119 with DI reached nominal significance. Several other missense SNPs in P2RX7 (rs17525809, rs208294, rs7958311, and rs2230912) showed an association with glycemic traits in this cohort, providing further support that P2X7 regulates β-cell function. An earlier GWAS of type 2 diabetes in Chinese showed a nominal association with the P2RX7 locus (25); follow-up genotyping showed no significant association of rs1718119 with either diabetes risk or glycemic traits in this cohort. However, rs1718119 is present in relatively low frequency in this population and is in low LD with the SNPs initially associated with diabetes risk; it is possible that the signal seen in the initial GWAS tags another P2RX7 locus with either altered function or expression.

The mice with the P451L variant were glucose intolerant and insulin resistant relative to their wild-type littermates. Previous work by Glas et al (9) showed differences in glucose homeostasis traits in P2rx7-null mice that are complementary to our data, with P2rx7-null mice demonstrating impaired glucose tolerance compared with the wild type. Because Glas' experiments were performed on a C57BL6 background, this study compared a null allele with a severely hypomorphic P2rx7 allele, and differences in glycemic control between null and wild-type were seen only on a high-fat, high-sucrose diet. In our study, the difference in P2X7 function is more marked (severely hypomorphic vs fully functional P2X7), allowing us to observe differences on a normal chow diet.

Also of note in our study is the potential statistical evidence for P2RY1 polymorphisms influencing glucose homeostasis. The same minor allele in the 3′ untranslated region (UTR), P2RY1 variant rs16864613, was associated with increased 2-hour glucose in the MAGIC cohort and as well as in the PPP-Botnia study. This finding is also of great potential interest because others have previously shown that P2ry1-null mice demonstrate abnormal glucose homeostasis associated with enhanced insulin secretion (8). Future directions would include examining the impact of the 3′ UTR P2RY1 variant with altered function or expression levels.

We also noted a statistical association of P2RX3, P2RX4, and P2RX5 polymorphisms with diabetes risk. The same SNP in P2RX3 was associated with an increased fasting glucose in MAGIC and an increased diabetes risk in DIAGRAM. Similarly, the signal from the cluster of P2RX4 SNPs for diabetes in DIAGRAM appears to be the same as the signal from the coding P2RX4 (S258G; r² = 1) SNP in PPP-Botnia associated with altered glycemic control. A recent study of human β-cells demonstrated by in situ hybridization that P2RX3, P2RX5, and P2RX7 were strongly expressed and that ATP activates P2X3 receptors in the β-cell membrane, resulting in enhanced insulin secretion (11). Further work is needed to test whether the SNPs identified in our study (or those in LD) affect receptor function and to elucidate how they may affect glucose homeostasis and diabetes risk.

Our data show that low pore-forming capacity P2X7 variants are associated with impaired glucose homeostasis in both humans and mice. It is thought provoking to note that the extracellular ATP-hydrolyzing enzyme ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1) has also been associated with glucose homeostasis traits in human (34–37) and mouse studies (36, 38). Mice transgenic for Enpp1 are insulin resistant, whereas mice with Enpp1 knocked down demonstrate increased insulin sensitivity (38), consistent with the hypothesis that increased P2X signaling (via increased extracellular ATP) improves insulin resistance (39). A similar phenotype has been demonstrated in mice null for the extracellular ATP-hydrolyzing enzyme Entpd1 (ectonucleoside triphosphate diphosphohydrolase 1) and for humans with a SNP associated with low ENTPD1 expression (40). The observation that an extracellular ATP receptor polymorphism causes similar phenotypes to an extracellular ATP-metabolizing enzyme suggests that extracellular nucleotides in general are important for glucose homeostasis.

We posit that extracellular ATP and its metabolites might serve as a way for cells or organ systems to communicate with one another about their state of energy supply and demand. Situations leading to increased ATP release from cells, such as inflammation or injury, could influence glucose homeostasis through the purinergic signaling system. In addition, corelease of ATP with insulin may be a mechanism of enhancing the pancreatic β-cell response to increases in plasma glucose concentration. A cell surface receptor such as P2X7 may be an ideal drug target candidate for altering glucose homeostasis in humans with abnormalities of glucose metabolism.