The Influence of Rare Genetic Variation in SLC30A8 on Diabetes Incidence and β-Cell Function

Abstract

Context/Objective:

The variant rs13266634 in SLC30A8, encoding a β-cell–specific zinc transporter, is associated with type 2 diabetes. We aimed to identify other variants in SLC30A8 that increase diabetes risk and impair β-cell function, and test whether zinc intake modifies this risk.

Design/Outcome:

We sequenced exons in SLC30A8 in 380 Diabetes Prevention Program (DPP) participants and identified 44 novel variants, which were genotyped in 3445 DPP participants and tested for association with diabetes incidence and measures of insulin secretion and processing. We examined individual common variants and used gene burden tests to test 39 rare variants in aggregate.

Results:

We detected a near-nominal association between a rare-variant genotype risk score and diabetes risk. Five common variants were associated with the oral disposition index. Various methods aggregating rare variants demonstrated associations with changes in oral disposition index and insulinogenic index during year 1 of follow-up. We did not find a clear interaction of zinc intake with genotype on diabetes incidence.

Conclusions:

Individual common and an aggregate of rare genetic variation in SLC30A8 are associated with measures of β-cell function in the DPP. Exploring rare variation may complement ongoing efforts to uncover the genetic influences that underlie complex diseases.

Individuals of European descent who carry the C vs T allele in the missense single-nucleotide polymorphism (SNP) rs13266634 at SLC30A8 (encoding zinc transporter 8 ZnT8]) have elevated type 2 diabetes (T2D) risk (1), impaired β-cell function, and higher proinsulin levels (adjusted for fasting insulin) (2, 3), and zinc intake appears to modify the glycemic effects of this locus (4). ZnT8 transports zinc molecules, essential for insulin storage and processing, into insulin granules (5). In vitro and mouse models have demonstrated that disruption of Slc30a8 zinc transport alters insulin crystallization and results in decreased insulin secretion (6–9). The risk variant at rs13266634 was not significantly associated with diabetes incidence in the Diabetes Prevention Program (DPP) (10). Therefore, we sequenced regions in SLC30A8 in 380 DPP participants and subsequently genotyped discovered variants in the full DPP population. By examining variants individually and in aggregate, we aimed to identify variants at SLC30A8, beyond the index variant rs13266634, associated with diabetes incidence and impaired β-cell function, and test whether zinc intake modified these effects.

Subjects and Methods

The DPP enrolled 3819 U.S. participants at high risk of developing type 2 diabetes (overweight with elevated fasting glucose and impaired glucose tolerance) (11), from which a subset, 3445 participants, consented to genetic testing. Of these participants, we examined 2997 who were randomized to placebo, metformin 850 mg twice daily, or lifestyle intervention with a goal weight loss of ≥7% and ≥150 min/wk of physical activity for association with all the outcomes. Participants in a fourth troglitazone treatment arm (n = 585) were included in genotyping but not included in the association testing due to early termination (12). Power calculation for alleles of various frequencies and effect sizes on diabetes incidence can be found in the Supplementary Table 1 in Moore et al (10). Ethical approval was obtained by local human research committees, and all participants gave informed consent.

Diabetes incidence was determined by a diagnostic fasting or 2-hour glucose after an oral glucose tolerance test, confirmed by a second test (11). We measured β-cell function including the insulinogenic index (InsIndex [units per milliliter]/[milligrams per deciliter] = Δinsulin_30−0min/Δglucose_30−0min) (13) and the oral disposition index (DI_o [milligrams per deciliter]⁻¹ = InsIndex × 1/fasting insulin) (14). We provided the association with fasting glucose and proinsulin (adjusted for fasting insulin) for all analyses (Supplemental Tables 1–3, published on The Endocrine Society's Journals Online website at http://jcem.endojournals.org) to corroborate previous findings (15, 16). Baseline zinc intake was determined using a modified Block Food Frequency Questionnaire (17).

We sequenced 380 DPP participants (76 from each ethnic group). We oversampled participants who developed diabetes to enrich our analysis for diabetes-related variants. The sequenced individuals were included in the subsequent association analyses of 2997 participants.

We used Sanger sequencing on an ABI3730 DNA analyzer for 2× coverage of 8 exonic regions, 5′-untranslated region with 50 base pairs (bp) around each intron/exon junction, 1000 bp upstream and downstream of SLC30A8, and 1000 bp surrounding rs13266634. We genotyped 69 SNPs discovered in sequencing and 10 SNPs annotated to SLC30A8 but not identified during sequencing in 3445 participants. After quality control (excluding nonconcordance between genotyping and sequencing, failed assay design, call rate <95%, failed Hardy-Weinberg equilibrium with a P < .001), 61 SNPs (44 of which were novel) were further analyzed.

Twenty-two common SNPs (minor allele frequency [MAF] ≥0.01 in at least 1 ethnic group) were examined using Cox proportional hazard models for association with diabetes incidence and analysis of covariance for association with the quantitative traits. Results were stratified by treatment group for a genotype × treatment group interaction P < .05. We adjusted for sex, age at randomization, baseline body mass index, self-reported ethnicity, and, if applicable, treatment group and respective baseline trait. Follow-up analyses included the SNP as a class variable obtaining marginal means and comparison of differences between genotypic groups.

We used 5 methods to test the association between 39 rare genetic variants (MAF <0.01 in all ethnicities) and the outcome.

Three genetic risk scores (GRSs) were constructed by summing the number of minor alleles over the sample: 1) a GRS including all 60 SNPs, not including rs1326634; 2) a missense GRS including 4 novel missense variants (8_118228561, 8_118239185, 8_118252509, and 8_118254036) and 1 known missense variant (rs16889462); 3) and a rare GRS including the 39 rare SNPs. A combined multivariate and collapsing (CMC) method coded each participant as having a variant with an MAF <2% as present or no rare variants as absent (18). The Sequence Kernel Association Test (SKAT) allows variants to have different directions and magnitude of effects (19). The GRS and CMC method were used in the models described above to test the associations with the outcomes. SKAT was used for testing associations with the β-cell function traits only. All scores were tested for interaction by treatment group and stratified if P < .05 except SKAT, which does not allow for interaction terms and was stratified up-front by treatment group. In a follow-up analysis, we used a Wilcoxon rank sum test to examine the association between the individual rare variants and quantitative traits stratified by treatment group.

We tested whether zinc intake modified diabetes risk conferred by SLC30A8 variants by adding additional covariates: baseline zinc intake, total caloric intake, an interaction term for baseline total zinc intake × genotype or GRS covariate, and factors that affect intestinal zinc absorption (iron intake, log calcium intake, polyunsaturated-to-saturated fat intake, and log dietary fiber) (20). For interaction P < .05, we stratified by genotype and obtained hazard ratios per 1 mg/d difference in baseline zinc intake.

Given the high previous probability of association with T2D, we used a traditional α-level of .05 for statistical significance.

Results

Sixty-one SLC30A8 variants (44 novel) passed quality control (Supplemental Table 4) and were analyzed. Five novel missense variants were probably damaging on bioinformatic analysis (Supplemental Table 5), and a subset of SNPs had predicted regulatory consequences (Supplemental Table 6).

Common variants were not associated with diabetes incidence (Supplemental Table 7). The minor alleles of rs2464591, rs2466296, rs2466297, and rs2466299 (r² > 0.9; HapMap CEU and YRI) were associated with a positive ΔDI_o (P < .0005), whereas the minor allele of rs2466293 was associated with a negative ΔDI_o (Table 1). The rs16889462 was associated with improved InsIndex in the AG/AA vs the GG in the metformin group but not in the other treatment groups (Supplemental Table 2)_. The minor allele of rs3802177 was associated with a higher InsIndex, lower baseline PI(FI) (proinsulin adjusted for fasting insulin) levels, and a greater decrease in fasting glucose during the first year (Table 1 and Supplemental Tables 1 and 2).

Table 1.

Significantly Associated SLC30A8 Genetic Variants Tested for Association With Glucose- and Insulin-Related Quantitative Traits^a

SNP (Common SNPs, MAF ≥1%)	Trait	Genotype	Treatment-Adjusted Means (95% CI)	P
8_118252680	ΔInsIndex	GG	0.024 (−0.037 to 0.084)	.04
		GA/AA	−0.130 ( −0.283 to 0.023)
	ΔDIo	AA	0.009 (0.005 to 0.013)	.0001
		AG	0.005 (0.001 to 0.009)
		GG	−0.003 (−0.008 to 0.003)
rs6469675	ΔInsIndex	AA	−0.018 (−0.091 to 0.055)	.05
		AG	0.026 (−0.048 to 0.100)
		GG	0.100 (−0.100 to 0.209)
rs2464591	ΔDIo	GG	0.002 (−0.001 to 0.006)	.0003
		GA	0.007 (0.003 to 0.011)
		AA	0.015 (0.009 to 0.022)
rs2466296	ΔDIo	GG	0.002 (−0.002 to 0.006)	.0002
		GA	0.006 (0.002 to 0.011)
		AA	0.015 (0.009 to 0.022)
rs2466297	ΔDIo	CC	0.002 (−0.001 to 0.006)	.0003
		CT	0.006 (0.002 to 0.011)
		TT	0.016 (0.009 to 0.022)
rs2466299	ΔDIo	GG	0.002 (−0.002 to 0.006)	.0002
		GA	0.007 (0.003 to 0.011)
		AA	0.015 (0.009 to 0.022)
rs3802177	InsIndex	GG	1.21 (1.15 to 1.27)	.04
		GA	1.28 (1.22 to 1.35)
		AA	1.27 (1.15 to 1.40)
rs13266634	InsIndex	CC	1.20 (1.14 to 1.26)	.02
		CT	1.27 (1.20 to 1.34)
		TT	1.30 (1.17 to 1.43)

^aCommon SNPs rs2466293 and rs16889462 had a significant genotype × treatment interaction for ΔFG (fasting glucose) and ΔInsIndex, respectively, are found in Supplemental Table 2. Minor alleles are underlined. SNPs above did not have a significant genotype × treatment interaction. Analysis was adjusted by age, sex, body mass index, and ethnicity and additionally adjusted for treatment group and corresponding baseline trait for the year-1 change in traits; Δ represents the change in trait between year 1 and baseline. P values reported are from the 1 degree of freedom additive model. Sample size by genotype for baseline traits analysis is detailed in Supplemental Table 1.

The rare variant GRS showed a tentative direct relationship with diabetes incidence (hazard ratio = 1.27 [1.00–1.61] per rare variant allele; P = .05) (Table 2). One hundred twenty-five participants carried only 1, 33 carried 2, and 1 carried 3 rare variants and was grouped with the 2-variant carriers.

Table 2.

P Values for Association Tests Using Various Methods to Test Aggregates of Rare Variants for Association With Diabetes Incidence and β-Cell Function^a

	GRS	Missense GRS	Rare GRS	CMC	SKATb
Diabetes incidence	.66	.24	.05	.31
Baseline InsIndex	.96	.41	.46	.44	.25
Baseline DIo	.88	.07	.26	.84	.49
ΔInsIndex	.82	.60	.14	.03	.98
ΔDIo	.0003	.80	.53	.30	.99

^aΔ represents the change in trait between year 1 and baseline.

^bSKAT analysis stratified by treatment group is in Supplemental Table 2.

Various rare variant methods showed an association with β-cell function (Table 2). SNP × treatment interactions were nonsignificant (P > .05) for all methods. For each additional GRS minor allele, there was a 0.001 (SE 0.0002) change in the DI_o (P = .0003). This association was no longer significant after removing rs2466293, rs2464591, rs2466296, rs2466297, and rs2466299 from the GRS (P = .2). With the CMC method, carriers of at least 1 minor allele of the 39 rare variants had a mean ΔInsIndex of 0.034 (95% CI, −0.027 to 0.096), whereas carriers of no rare variants had −0.079 (95% CI, −0.187 to −0.028) (P = .03). No statistically significant associations were seen between the individual rare variants and these traits (Supplemental Table 3).

Three SNPs (8_118252314, 8_118252435, and rs16889462), 2 of which were novel, modified the effect of total zinc intake on diabetes risk but had no clear additive trend with each additional minor allele (Supplemental Table 8).

Discussion

We examined 44 novel variants through targeted sequencing of the T2D candidate gene SLC30A8 in the DPP. In aggregate, rare variants appear to influence diabetes risk and related traits, illustrating that both rare and common genetic variation may influence diabetes risk. Zinc intake did not appear to modify the genetic predisposition to diabetes at this locus, suggesting a limited role for dietary manipulations in modifying genetic risk.

We examined 39 rare variants, unique to certain ethnicities (Supplemental Table 4). Despite the probably damaging prediction by bioinformatics analysis, the individual missense variants and the missense variant GRS were not associated with diabetes incidence or quantitative traits. Further functional studies where point mutations are introduced into SLC30A8 constructs for transfection into β-cells may elucidate whether these variants attenuate or enhance protein function. Similarly, phenotyping individuals with definite loss-of-function mutations should be informative with regard to the direction of SLC30A8 variation on glycemic regulation in humans. We did not identify any loss-of-function variants in SLC30A8 in this multiethnic cohort, which underscores the drive for conservation and therefore relevance of this gene for metabolism.

Although none of the common SNPs was associated with diabetes incidence, we found associations between SLC30A8 variants and insulin secretion traits. The minor alleles of rs2464591, rs2466296, rs2466297, and rs2466299 were associated with an improvement in β-cell function, illustrated by an increase in DI_o. Conversely, the rs2466293 minor allele was associated with a decrease in β-cell function. Given the nonsignificant treatment × genotype interaction, it appears that these variants influence glycemia similarly among all the intervention groups. These findings exemplify that SLC30A8 variation comparably influences improvements or deteriorations of β-cell function over a year of follow-up, independent of insulin-sensitizing interventions that reduce diabetes risk.

We employed 5 methods to examine the contribution of rare genetic variation on diabetes risk and B-cell function. The GRS was associated with ΔDI_o, and the CMC method revealed an association between carriers of rare variants and ΔInsIndex. These results suggest that rare SLC30A8 variation may have functional significance beyond the index SNP, rs13266634. The GRS and CMC methods are limited in that they presume that the rare allele is deleterious, which may not always be true despite our ascertainment having been largely conducted in participants who went on to develop diabetes. Therefore, this assumption may dilute the true impact of the rare GRS. This limitation is addressed with the SKAT method, which allows variants to have different directions and magnitude of effects (19); here we did not see an association with β-cell function. Although limited by power, none of the rare SNPs appear to have very large effects, but the aggregate burden of rare SLC30A8 variation influences β-cell function. These findings provide the basis for future studies with the exome chip implemented in larger populations where the rare variants can be adequately tested individually.

Limitations

We were able to enhance statistical power by constructing aggregate variant scores, but this method does not model the behavior of individual variants. As these methods continue to evolve, functional experiments will be needed to further elucidate the mechanism by which rare variants influence phenotypes. Additionally, this study lacks a validation cohort and nominal associations found in this study warrant follow-up elsewhere. Furthermore, our sequencing efforts started before the introduction of next-generation sequencing techniques and sequenced only targeted regions of SLC30A8; thus, novel variants may have been overlooked.

Our study showed that an aggregate of rare variants in SLC30A8 may increase diabetes risk and influence measures of β-cell function. This study supports the pursuit of rare variation to better understand the genetics of complex traits.