Differential DNA methylation in human islets from T2D versus non-diabetic donors

We next examined if any individual sites exhibit differential DNA methylation in pancreatic islets from T2D compared with non-diabetic donors. After correcting for multiple testing using a false discovery rate (FDR) analysis we identified 3,116 CpG sites that were differentially methylated between T2D and non-diabetic islets with FDR less than 5% (q<0.05), which means that 156 false positives are expected by chance [18]. The distribution of the absolute differences in DNA methylation between T2D and non-diabetic islets is shown in Figure 1D. To gain biological relevance, we filtered our DNA methylation results requiring absolute differences in methylation ≥5% between T2D and non-diabetic islets. We found 1,649 CpG sites, of which 1,008 are located in or near 853 unique genes and 561 are intergenic, that had absolute differences in methylation ≥5% in T2D versus non-diabetic islets and these were used for all further analyses (Figure 1D and Table S1). When the differences in DNA methylation between diabetic and non-diabetic islets of these 1,649 CpG sites are expressed as fold-change instead of absolute differences, we observe differences in methylation ranging from 6 to 59%. While 1,596 of these 1,649 CpG sites (97%) showed decreased DNA methylation, only 53 sites (3%) showed increased DNA methylation in T2D compared with non-diabetic islets (Figure 1E). The majority of CpG sites showing decreased DNA methylation in T2D versus non-diabetic islets had an intermediate degree of methylation and they were overrepresented among CpG sites with 20–70% methylation (Figure 1F). On the other hand, CpG sites showing increased DNA methylation in T2D versus non-diabetic islets had a lower degree of methylation and they were overrepresented among CpG sites with 20–40% methylation (Figure 1G). Our data suggest that CpG sites with an intermediate degree of DNA methylation are more dynamic to change in human islets.

Chromosomal and genomic distribution of differentially methylated CpG sites in T2D islets

There is an accumulation of genetic variation on certain chromosomes associated with disease [19], [20]. However, it remains unknown if there is an over- or underrepresentation of differential DNA methylation on certain chromosomes linked to diabetes. We therefore determined the chromosomal distribution of the 1,649 sites that exhibit differential DNA methylation in human T2D islets (Figure 2A). Using Chi2 tests, we found that the number of differentially methylated sites were overrepresented on chromosomes 1 and 2, and underrepresented on chromosome 19, in comparison to the chromosomal distribution of all analyzed sites on the Infinium HumanMethylation450 BeadChip. These results are not explained by the distribution of analyzed sites on the array or gene density on the chromosomes.

Open in a separate window

Figure 2

Distribution of individual sites that exhibit differential DNA methylation in human islets from 15 T2D versus 34 non-diabetic donors.

(A) The chromosomal location of the 1,649 sites that exhibit differential DNA methylation in islets of T2D versus non-diabetic donors in comparison to all analyzed sites on the Infinium HumanMethylation450 BeadChip. (B) Distribution of significant CpG sites in relation to nearest gene regions. (C) Distribution of significant CpG sites in relation to CpG island regions. *The distribution of the significant sites compared with all analyzed sites on the Infinium HumanMethylation450 BeadChip is significantly different from what is expected by chance based on Chi2 tests. P-values have been corrected for multiple testing using Bonferroni correction. (D) The average degree of DNA methylation of analyzed non-CpG sites in human pancreatic islets. Genes with previous known function in (E) pancreatic islets, (F) the exocytotic process and (G) apoptosis that exhibit differential DNA methylation in T2D versus non-diabetic human islets with an absolute difference in methylation ≥5%. * q<0.05. Data are mean ± SEM.

Previous cancer studies suggest that differential DNA methylation mainly occurs in CpG island shores rather than in CpG islands and promoter regions [21], [22]. However it remains unknown if this is also the case in T2D patients. We therefore evaluated the distribution of differentially methylated sites in T2D versus non-diabetic islets, either based on their relation to the nearest gene and functional genome distribution (Figure 2B) or based on the CpG content and neighbourhood content (Figure 2C). We found that differentially methylated CpG sites were overrepresented in intergenic regions and underrepresented in the TSS200, 1^st exon and 3′UTR in comparison to the probe distribution on the Infinium HumanMethylation450 BeadChip (Figure 2B). The distribution of the probes on the Infinium HumanMethylation450 BeadChip as well as the differentially methylated CpG sites in T2D islets in relation to a CpG island is further shown in Figure 2C. We found that ~60% of the differentially methylated CpG sites in T2D islets are located in the open sea while ~25% are located in the CpG island shores and only ~7% are located in CpG islands (Figure 2C). Moreover, differentially methylated CpG sites were overrepresented in the open sea and underrepresented in CpG islands in comparison to the probe distribution on the Infinium HumanMethylation450 BeadChip (Figure 2C).

The Infinium HumanMethylation450 BeadChip covers 16,232 previously known differentially methylated regions (DMRs) that were selected based on the previously described criteria [21], [23]. We found 156 CpG sites with decreased and 4 CpG sites with increased DNA methylation in known DMRs in T2D compared with non-diabetic islets (Table S2), which is more than expected (P<0.001).

Non-CpG methylation in human pancreatic islets

It has previously been established that DNA methylation in differentiated mammalian cells mainly occurs on cytosines in CG dinucleotides [24]. While the Infinium HumanMethylation450 BeadChip mainly analyzes DNA methylation in CpG sites, it also generated methylation data for 3,039 non-CpG sites in the human islets. However, only 1,189 of these non-CpG probes can be mapped with a perfect match to the correct genomic location annotated by Illumina [25] and methylation data of those probes were subsequently used for further analysis. These 1,189 sites were most predominant in the intergenic region, gene body and open sea (Figure. S2A–B). To test if non-CpG sites are methylated in human islets, we calculated the average degree of DNA methylation of the analyzed non-CpG sites. The average degree of methylation of non-CpG sites was 5.9–21.7% in human islets (Figure 2D and Table S3). Out of these non-CpG sites, only two sites were significant after correcting for multiple testing (q<0.05) and none of these sites had an absolute difference in methylation >5% in T2D versus non-diabetic islets.

Biological features of the genes that exhibit differential methylation in T2D islets

We next performed a KEGG pathway analysis to identify biological pathways with enrichment of genes that exhibit differential DNA methylation in T2D versus non-diabetic islets. A total of 853 genes, represented by CpG sites with differential DNA methylation ≥5% in T2D islets (Table S1), were analyzed using WebGestalt. Relevant enriched KEGG pathways in T2D islets include pathways in cancer, axon guidance, MAPK signaling pathway, focal adhesion, ECM-receptor interaction and regulation of actin cytoskeleton (Table 2). We further performed a separate KEGG pathway analysis only including genes that exhibit increased DNA methylation in T2D islets and we then found an enrichment of genes in the complement and coagulation cascades; C4A and C4B (observed number of genes=2, expected number of genes=0.13 and P _adjusted=0.0175). We also tested if any of the genes in Table 2 exhibit differential expression in human β- compared with α-cell fractions using published data by Dorrell et al [26]. However, among the genes included in Table 2, there were no significant differences in expression in β- versus α-cells.

Previous GWAS have identified SNPs associated with T2D and/or obesity [3]. These SNPs have been linked to candidate genes, representing genes closest to respective risk SNPs. However, the SNPs identified in GWAS only explain a small proportion of the estimated heritability for T2D, proposing that there are additional genetic factors left to be discovered. These may include genetic factors interacting with epigenetics [27]. We therefore tested if any of 40 T2D candidate genes and 53 obesity genes identified by GWAS were differentially methylated in the human T2D islets [3]. The Infinium HumanMethylation450 BeadChip covers 1,525 CpG sites representing 39 of the T2D candidate genes and 1,473 CpG sites representing all 53 obesity genes. However, one should keep in mind that for a number of these SNPs it still remains unknown if the closest gene is the gene involved in T2D or obesity and if the identified SNP is the functional SNP. Therefore, to cover most regions harboring a genetic variant associated with T2D or obesity, we also investigated the level of DNA methylation for all CpG sites in a region 10 kb up- and downstream of intergenic SNPs associated with T2D (n=28) and obesity (n=41) (www.genome.gov/gwastudies. Accessed: March 18, 2013). We identified 44 methylation sites, representing 17 T2D candidate genes and one intergenic SNP that were differentially methylated in T2D versus non-diabetic islets with a FDR less than 5% (q<0.05, Table 3). Twenty-one of these sites, representing ten genes, had absolute differences in methylation >5% in T2D versus non-diabetic islets, which correspond to a fold change ranging from 7 to 28%. Only three sites in three different obesity genes were differentially methylated in T2D islets and one of these sites had an absolute difference in methylation >5% (Table 3).

Finally, based on literature search, genes with known functions in pancreatic islets and/or β-cells [28]–[40] (Figure 2E), the exocytotic process [41]–[44] (Figure 2F) and apoptosis [45] (Figure 2G) were found among the genes that showed differential DNA methylation in T2D islets.

Differential mRNA expression and differential methylation in T2D islets

DNA methylation of certain genomic regions may silence gene transcription [8], [12]. We therefore used microarray mRNA expression data to examine if any of the 853 genes that exhibit differential DNA methylation in T2D islets also exhibit differential mRNA expression in islets from the same donors. We found that 102 of the 853 differentially methylated genes were also differentially expressed in T2D compared with non-diabetic islets (Table S4). While 77 (~75%) of the differentially expressed genes had an inverse relationship with DNA methylation, e.g. decreased DNA methylation was associated with increased gene expression in T2D islets, 26 (~25%) had a positive relationship with DNA methylation, e.g. decreased DNA methylation was associated with decreased expression (Figure 3A and Table S4). Figure 3B–C describes the genomic distribution of the differentially methylated CpG sites that are located in/near genes that also exhibit differential expression in T2D islets. Interestingly, there was an overrepresentation of CpG sites in the 5′UTR only when differential DNA methylation and gene expression show an inverse relationship (Figure 3B). In addition, CpG sites in the open sea and northern shore were overrepresented while sites in the CpG islands were underrepresented when DNA methylation and gene expression show an inverse relationship (Figure 3C). These data suggest that differential DNA methylation in certain genomic regions may contribute to an inverse regulation of gene expression. In addition, we found an overrepresentation of differentially methylated CpG sites in the gene body regardless of whether methylation and gene expression show a positive or inverse relationship (Figure 3B), this is known as the DNA methylation paradox which still remains unexplained [46].

Open in a separate window

Figure 3

Relation between DNA methylation and gene expression in human pancreatic islets.

(A) The number of genes that exhibit both differential DNA methylation and gene expression in pancreatic islets from 15 T2D versus 34 non-diabetic donors. One gene (SLC44A4) was counted twice in the figure because it showed increased expression in T2D islets and was associated with probes that showed both increased and decreased DNA methylation. Distribution of differentially methylated CpG sites located in/near genes that also exhibit differential expression in T2D islets with an inverse relationship or a positive relationship in relation to their (B) functional genome distribution and in relation to (C) CpG island regions. Decreased DNA methylation and increased mRNA expression of (D) CDKN1A and (E) PDE7B in pancreatic islets of T2D versus non-diabetic donors. (F) A diagram of the two luciferase reporter plasmids used to test the effect of DNA methylation on CDKN1A and PDE7B promoter activity is shown. The two plasmids contain 1500 bp of either the CDKN1A or the PDE7B promoter regions inserted into a pCpGL-basic vector. Methylated (grey and black bars) or mock-methylated (white bars) promoter constructs were transfected into clonal β-cells for 48 hours prior to luciferase assay. The data were normalized with co-transfected renilla luciferase control vector and are the average of three separate experiments with five replicates each. In each experiment, cells were transfected with an empty pCpGL-basic vector as a background control. * P<0.05. Data are mean ± SEM.

Functional analysis of differentially methylated genes

We continued to functionally test if DNA methylation affects gene expression. We selected two genes with differential DNA methylation of multiple CpG sites and in which DNA methylation showed an inverse relationship with gene expression for luciferase experiments (Table S4 and Figure 3D–E). Reporter gene constructs were subsequently produced for CDKN1A and PDE7B. Promoter sequences of the selected genes were inserted into a luciferase expression plasmid that completely lacks CpG dinucleotides. The constructs could thereby be used to study the effect of DNA methylation on luciferase activity in transfection assays. Each construct was mock methylated or methylated with the methyltransferases SssI or HhaI that methylates all CpG sites or the internal cytosine residue in a GCGC sequence only, respectively. SssI methylation thereby results in highly methylated constructs and HhaI methylation gives point methylated constructs in which only a fraction of the CpG sites are methylated. The number of CpG sites that may be methylated by these enzymes in each respective construct is shown in Figure 3F. Clonal β-cells were then transfected with the mock-methylated or methylated constructs. The highest reporter gene expression was generated by the mock-methylated control constructs including the promoter regions (Figure 3F). Furthermore, methylation of the human CDKN1A and PDE7B promoter regions suppressed reporter expression significantly (P<0.05). While methylation of the promoter regions by SssI suppressed reporter gene expression to 28.0±11.8% for CDKN1A and to 22.1±2.9% for PDE7B, point methylation by HhaI suppressed reporter expression to 47.2±7.1% for CDKN1A and to 55.9±9.4 for PDE7B (Figure 3F).

Overexpression of Cdkn1a, Pde7b and Sept9 in clonal β-cells results in impaired insulin secretion

We identified 75 genes that exhibit decreased DNA methylation and increased gene expression in pancreatic islets of T2D compared with non-diabetic donors when performing the genome-wide DNA methylation analysis (Table S4). To model the situation in humans and elucidate the mechanisms whereby these genes may contribute to impaired β-cell function and the development of T2D, we overexpressed three of these genes; Cdkn1a, Pde7b and Sept9, in clonal β-cells (Figure 4A). These genes were selected based on their potential role in diabetes and islet function and because they showed both differential DNA methylation of multiple CpG sites and differential gene expression in T2D islets. Overexpression of Cdkn1a and Pde7b led to a significant decrease in glucose-stimulated insulin secretion in clonal β-cells, while Sept9 had no significant effect (Figure 4B and Figure S3A). Moreover, while the direct response to the membrane-depolarizing agent KCl was unaffected, the fold-change of insulin secretion at KCl-stimulation divided by insulin secretion at low glucose was decreased in clonal β-cells overexpressing Cdkn1a, Pde7b or Sept9 (Figure S3B–C). Cdkn1a (also known as p21) encodes a potent cyclin-dependent kinase inhibitor that regulates cell cycle progression [31] and we therefore tested if overexpression of this gene would affect cell proliferation in clonal β-cells. Indeed increased Cdkn1a levels resulted in decreased β-cell proliferation (Figure 4C). Pde7b encodes a cAMP-specific phosphodiesterase [47] and cAMP potentiates insulin secretion [48]. We next stimulated Pde7b overexpressing cells with glucose in combination with IBMX, a general phosphodiesterase inhibitor. Addition of IBMX normalized glucose-stimulated insulin secretion in Pde7b overexpressing β-cells (Figure 4D) suggesting that the cAMP hydrolyzing activity of Pde7b underlies the secretory defect.

Open in a separate window

Figure 4

Impact of Cdkn1a, Pde7b, Sept9 and Exoc3l on clonal β- and α-cells.

Glucose-sensitive clonal INS-1 832/13 β-cells were used to study the impact of Cdkn1a, Pde7b, Sept9 and Exoc3l on insulin secretion and β-cell function. (A) Overexpression of Cdkn1a, Pde7b and Sept9 with pcDNA3.1 expression vectors in clonal β-cells resulted in elevated mRNA levels (black bars) compared with β-cells transfected with an empty pcDNA3.1 vector (white bars). * P<0.05. Data are mean ± SEM. Overexpression was also evident at the protein level as determined by immunoblot detection of the HA-tag situated on the c-terminal of the cloned cDNAs (rightmost panel) (B) Insulin secretion in response to 2.8 mM (white bars) and 16.7 mM (black bars) glucose in clonal β-cells overexpressing either Cdkn1a, Pde7b or Sept9 compared with control cells transfected with an empty pcDNA3.1 vector (n=5). * P<0.05. Data are mean ± SEM (C) Decreased cell proliferation in clonal β-cells overexpressing Cdkn1a (black bar) compared with cells transfected with an empty pcDNA3.1 vector (white bar) (n=4). * P<0.05. Data are mean ± SEM (D) Insulin secretion in response to 2.8 mM and 16.7 mM glucose (Gluc) or 16.7 mM glucose (Gluc) in combination with 100 µM IBMX in clonal β-cells overexpressing Pde7b (black bars) compared with cells transfected with an empty pcDNA3.1 vector (white bars) (n=3). * P<0.05. NS=not significant. (E) Increased DNA methylation and decreased mRNA expression of EXOC3L2 in pancreatic islets of 15 T2D versus 34 non-diabetic donors. (F) Transfection of clonal β-cells with siRNA targeting Exoc3l resulted in decreased Exoc3l mRNA expression (siExo3l, black bar) when compared with clonal β-cells transfected with negative control siRNA (siNC, white bar) (n=3). * P<0.05. Silencing of Exoc3l resulted in decreased Ca²⁺-dependent exocytosis; (G) Depolarizationevoked a decrease in membrane capacitance (ΔC_m) in single INS1-832/13 β-cells where Exoc3l was silenced (black trace) compared with control β-cells (grey trace) treated with negative control siRNA. (H) Histogram of the summed increase in membrane capacitance evoked by the two first depolarizations (Σ_1–2), the latter eight depolarizations (Σ_3–10) or all depolarizations in the train (Σ_all). Data are mean ± SEM of n=11 β-cells treated with control siRNA (white bars) and n=4 β-cells treated with siRNA against Exoc3l (black bars). * P<0.05. (I) Depolarization-evoked increase in current (I) in single INS1-832/13 β-cells treated with control siRNA (grey trace) or siRNA targeting Exoc3l (black trace). Notice that the rapid Na⁺-current is markedly diminished in siExoc3l treated clonal β-cells. (J) Reduced expression of Exoc3l has no effect on the Ca²⁺ influx. Charge (Q)-voltage (V_m) relationship in β-cells treated with control (white squares) siRNA and siRNA against Exoc3l (black circles). The charge is representative of the Ca²⁺-influx into the cell through the voltage-dependent Ca²⁺ channels. (K) As in J, but the peak-current (Ipeak)-voltage (Vm)-relationship was estimated as a representation of the voltage-dependent Na⁺ current. The Na⁺ current is significantly reduced in siExoc3l cells (P<0.05) versus control siRNA cells except for the depolarization to −40 mV. Data in J and K are mean ± SEM of n=11 β-cells treated with control siRNA and n=7 siExoc3l treated β-cells. (L) αTC1-6 cells were used to determine the impact of Cdkn1a, Pde7b and Sept9 on glucagon secretion. Overexpression of Cdkn1a and Pde7b resulted in significantly increased glucagon secretion at 1 mM glucose (white bars) (n=4, * P<0.05), while overexpression of Pde7b and Sept9 resulted in increased glucagon secretion at 16.7 mM glucose (black bars) (n=4, ¤ P<0.05) compared with control cells transfected with an empty pcDNA3.1 vector.

Silencing of Exoc3l in clonal β-cells results in decreased exocytosis

EXOC3L2 is one of the genes that exhibit increased DNA methylation and decreased gene expression in pancreatic islets of T2D compared with non-diabetic donors (Figure 4E and Table S4). The protein encoded by EXOC3L2 is part of the exocyst complex [42] and may consequently affect exocytosis of insulin from pancreatic β-cells. To model the situation in human T2D islets, and to examine if decreased levels of EXOC3L2 affect β-cell exocytosis, Exoc3l was silenced in clonal β-cells using siRNA. This resulted in a 60.7% reduction of Exoc3l levels (Figure 4F). Next, exocytosis was measured as changes in membrane capacitance using the patch-clamp technique. Silencing of Exoc3l resulted in decreased β-cell exocytosis (Figure 4G–H). In particular, the two first depolarizations, representing the rapid first-phase insulin secretion, were decreased in Exoc3l deficient β-cells (Figure 4H). Moreover, the Ca²⁺ current was unaffected in Exoc3l deficient β-cells (Figure 4I–J), demonstrating a direct effect of Exoc3l on the exocytosis machinery rather than on the Ca²⁺ current. We also observed decreased voltage dependent Na⁺ current in Exoc3l deficient β-cells (Figure 4I, K).

Overexpression of Cdkn1a, Pde7b and Sept9 in clonal α-cells results in increased glucagon secretion

As elevated glucagon levels in the fasted state contribute to hyperglycemia in patients with T2D [49], we next tested if selected candidate genes that exhibit decreased DNA methylation and increased expression in pancreatic islets of T2D compared with non-diabetic donors contribute to increased glucagon secretion in pancreatic α-cells (αTC1-6 cells). While overexpression of Cdkn1a and Pde7b resulted in increased glucagon release at 1 mM glucose compared to control transfected α-cells, Sept9 overexpression led to a borderline significant increase in glucagon secretion (Figure S3D and Figure 4L). Additionally, when stimulated with 16.7 mM glucose, Pde7b and Sept9 overexpressing α-cells secreted more glucagon than control transfected cells (Figure 4L).

Technical validation and replication of DNA methylation data in T2D islets

As a technical replication of the Infinium HumanMethylation450 BeadChip data, one islet sample was bisulfite converted and analyzed on the Infinium chips at two different occasions. The correlation for DNA methylation of 100 000 randomly chosen CpG sites on the two chips was then calculated and the methylation data showed a strong positive correlation (R²=0.99, P=2.2×10⁻¹⁶; Figure S4A).

In addition, seven of the CpG sites that exhibit differential DNA methylation in T2D islets were selected for technical validation of the Infinium HumanMethylation450 BeadChip data with pyrosequencing. The CpG sites selected for technical validation include cg21091547 (CDKN1A), cg27306443 (PDE7B), cg19654743 (SEPT9), cg04751089 (IRS1), cg20995304 (HDAC7), cg01649611 (THADA) and cg15572489 (PTPRN2) (Table 4). In agreement with the Infinium HumanMethylation450 BeadChip data, all seven CpG sites showed differential DNA methylation in T2D versus non-diabetic islets when analyzed with pyrosequencing with differences in methylation of a similar magnitude between the two groups (Table 4). Furthermore, the DNA methylation data generated with Infinium HumanMethylation450 BeadChip and pyrosequencing for these seven CpG sites correlated strongly (rho=0.84–0.94, P≤1.2×10⁻¹³; Table 4 and Figure S4B).

Confirmation of DNA methylation data in human T2D islets

To further validate our results, we tested if any of the CpG sites that exhibit differential DNA methylation in T2D islets of a recent study by Volkmar et al [14], also exhibit differential methylation in T2D islets in our study. While our study analyzed DNA methylation of 479,927 CpG sites distributed across the entire genome and 99% of RefSeq genes, the study by Volkmar et al. only analyzed 27,578 CpG sites distributed mainly in CpG islands in a subset of RefSeq genes and it was therefore only possible to compare some of our studied CpG sites. Nevertheless, our array covers 264 of the 276 CpG sites that exhibit differential DNA methylation in the study by Volkmar et al and 71 of these sites (~27%) were differentially methylated in our study with P<0.05, which is more than expected by chance (Chi2=66.7 and P<0.0001; Table S5). The data by Volkmar et al have not been corrected for multiple testing and it may therefore include false positive results, which may explain the difficulty to replicate some of their results. Yet, the DNA methylation data of the 264 CpG sites analyzed in both studies correlated significantly for both non-diabetic (rho=0.66, P<0.0001; Figure S5A) and diabetic (rho=0.68, P<0.0001; Figure S5B) islets.

As an additional control, we compared the results from the present study with our previous studies where we found differential DNA methylation of PPARGC1A, INS and PDX1 in human pancreatic islets of T2D versus non-diabetic donors by using a candidate gene approach [11]–[13]. For PPARGC1A we were unable to compare the methylation data between the two studies as our previous study was based on the average degree of methylation of multiple CpG sites in an analyzed genomic region [11]. We have previously found increased DNA methylation in four CpG sites of the INS gene in human pancreatic islets from T2D compared with non-diabetic donors [13]. Two of these four CpG sites (+63 and −180) were covered by probes on the Infinium HumanMethylation450 BeadChip and in agreement with the results in our previous study these two CpG sites show increased DNA methylation in T2D versus non-diabetic islets also in the present analysis (cg00613255, P=0.008 and cg25336198, P=0.003, Figure S5C). In the present study there were 13 additional CpG sites annotated to the INS gene with increased DNA methylation in T2D islets and P<0.05 (Figure S5C). On the other hand, none of the individual CpG sites showing differential methylation in PDX1 in T2D islets in our previous study [12] were covered by probes on the Infinium HumanMethylation450 BeadChip. However, using EpiTYPER, we could validate our previous PDX1 data [12], in the islet samples included in the present study (Figure S5D).

As a final control, we examined the degree of methylation in four known imprinted genes; SNRPN, MEST, KCNQ1 and IGF2. As expected these genes showed approximately 50% methylation in islets from non-diabetic donors (Figure S5E).

Impact of HbA1c, age and BMI on DNA methylation in human pancreatic islets

We then tested if known risk factors for T2D, including hyperglycemia, aging and obesity affect DNA methylation of the 1,649 CpG sites differentially methylated in T2D islets, already in non-diabetic subjects. The impact of HbA1c levels, age and BMI on DNA methylation of these 1,649 CpG sites was examined in pancreatic islets of 87 non-diabetic donors with HbA1c levels, age and BMI spanning between 4.3–6.4%, 26–74 years and 17.6–40.1 kg/m² respectively (Table S6). HbA1c levels were associated with differential DNA methylation of 142 CpG sites (Table S7). Moreover, age and BMI were associated with differential DNA methylation of 28 and 16 CpG sites, respectively (Tables S8, S9). Interestingly, increased age was associated with decreased methylation of CDKN1A and increased methylation of EXOC3L2 (Tables S8), which is in agreement with the results seen in T2D islets (Figure 3d and Figure 4e). Moreover, ~92% of the CpG sites that exhibit differential DNA methylation due to increased HbA1c, age or BMI in non-diabetic islets changed in the same direction as methylation in T2D islets. These data suggest that increased HbA1c levels, aging and/or obesity may affect DNA methylation of CpG sites which are differentially methylated in T2D islets already in islets of non-diabetic subjects.

Association between DNA methylation and expression in islets of non-diabetic donors

We also tested if there are significant associations between DNA methylation and gene expression of the 102 genes that exhibit both differential DNA methylation and gene expression in T2D versus non-diabetic islets, already in non-diabetic subjects. A linear regression model was used including batch, age, gender, BMI, HbA1c, islet purity and days of culture as covariates. In islets of the 87 non-diabetic donors, we found significant associations between DNA methylation and gene expression for 55 CpG sites in/near 43 genes (Table S10) out of the 149 CpG sites in/near 102 genes that exhibit both differential DNA methylation and gene expression in T2D versus non-diabetic islets (Table S4). The association between DNA methylation and gene expression for these 43 genes was in the same direction in both the 87 non-diabetic donors as in the T2D versus non-diabetic donors (Table S10 and Table S4). We further tested for associations between DNA methylation and gene expression in non-diabetic islets including all analyzed CpG sites on the Infinium HumanMethylation450 BeadChip in/near the 102 genes that show both differential DNA methylation and gene expression in T2D versus non-diabetic islets. We then found significant associations between DNA methylation and gene expression for 663 CpG sites in/near 57 genes out of the 102 studied genes.

Expression of enzymes that regulate DNA methylation and demethylation

To identify factors that may contribute to differential DNA methylation in human pancreatic islets, we further tested if risk factors for T2D affect islet expression of a number of enzymes which are known to regulate DNA methylation and demethylation in mammals [50]. We found that the islet expression of DNMT3b, which is involved in de novo DNA methylation, correlated negatively with age (rho=−0.25, P=0.02). In addition, exposure to lipids (1 mM palmitate) for 48 hours in vitro decreased the expression of two methyltransferases, DNMT3a (control islets 111.3±2.3 vs. lipid treated islets 107.7±2.0, P=0.039, n=13) and DNMT1 (control islets 182.4±5.0 vs. lipid treated islets 154.6±6.3, P=0.00005, n=13), in human pancreatic islets. Lipid exposure also increased islet expression of GADD45A (control islets 489.4±23.7 vs. lipid treated islets 612.8±56.3, P=0.010, n=13), which encodes a protein involved in demethylation [50].

Human pancreatic islet donors

Human pancreatic islets from 15 donors with T2D and 34 donors not diagnosed with diabetes were included in the genome-wide analysis of DNA methylation and mRNA expression. Donors were considered to have T2D if they had been diagnosed with the disease prior to their death. Selection criteria for non-diabetic donors were to have an HbA1c below 6.0%. Clinical characteristics of these donors are given in Table 1. Moreover, the impact of HbA1c levels, age and BMI on DNA methylation was studied in pancreatic islets from 87 non-diabetic donors. Their characteristics are given in Table S6. Human pancreatic islets were provided by the Nordic Network for Islet Transplantation, Uppsala University, Sweden.

Sample processing of human islets

Human islets were prepared by collagenase digestion and density gradient purification. Prior to nucleic acid purification, islets were cultured for 2.7±0.15 days as previously described [63]. DNA and RNA were extracted with the All Prep DNA/RNA kit (Qiagen, Hilden, Germany) and purity and concentration were determined by using a nanodrop (NanoDrop Technologies, Wilmington, USA). The purity of the islet preparations was determined by expression of endocrine (somatostatin and glucagon) and exocrine (pancreatic lipase, amylase α2A, chymotrypsin 2) markers and dithizone staining [63]. β-cell content in human islets of donors with available embedded islets (6 T2D and 13 non-diabetic donors) was analyzed using transmission electron microscopy. Hand-picked islets where fixed in 2.5% glutaraldehyde in freshly prepared Millonig and post-fixed in 1% osmium tetroxide before being dehydrated and embedded in AGAR 100 (Oxford Instruments Nordiska, Lidingö, Sweden) and cut into ultrathin sections as described [5]. The sections were put on Cu-grids and contrasted using uranyl acetate and lead citrate. The islet containing sections were examined in a JEM 1230 electron microscope (JEOL-USA. Inc., MA). Micrographs were analyzed for β-cell content with ImageJ and in-house software programmed in Matlab using methods previously described [14], [79]. Islet cell types were distinguished by means of granular appearance: β-cell granules have a dense core surrounded by a white halo and α-cells have small dense granules. The ratio of β-cells in each islet was calculated by division of the total number of β-cells by the sum of the β-cell and α-cell numbers. Furthermore, glucose-stimulated insulin secretion from human pancreatic islets was measured as described [80].

Purification of human β- and α-cell fractions

β- and α-cells were purified from pancreatic islets of three human donors (aged 54, 55 and 74 years old, with BMI 21.5–23.1 kg/m²), different from the donors described in Table 1 and Table S1, by using a method previously described by Parnaud et al [73]. Dissociation of islet cells was achieved by incubation with constant agitation for 3 minutes at 37°C in 0.05% trypsin-EDTA (Invitrogen) supplemented with 3 mg/ml DNAse I (Roche, Mannheim, Germany) followed by vigorous pipetting. Labelling and FACS sorting of the β- and α-cell fractions was performed as previously described [73]. Sorted β- and α-cells were applied to microscope slides and co-immunostained for insulin and glucagon in order to detect the amount of α-cells in the β-cell fraction, and vice versa. Using this method, a cell purity of 89±9% (mean ± SD) was achieved [81]. After DNA isolation using Qiagen DNEasy blood and tissue kit (Qiagen) and ethanol precipitation, DNA for genome-wide methylation profiling was available for β- and α-cell fractions from 3 and 2 donors, respectively.

Genome-wide DNA methylation profiling

Genome-wide DNA methylation profiling of human pancreatic islets and purified human β- and α-cell fractions was performed at the SCIBLU genomics center at Lund University with the Infinium HumanMethylation450 BeadChip (Illumina, Inc., San Diego, CA, USA) which interrogates 482,421 CpG sites, 3091 non-CpG sites and 65 random SNPs and covers 21,231 RefSeq genes [68]. 500 ng DNA from human pancreatic islets was bisulfite converted using the EZ DNA Methylation Kit D5001 (Zymo Research, Orange, CA, USA) according to the manufacturer's instructions. Bisulfite converted DNA was amplified, fragmented and hybridized to the BeadChips following the standard Infinium protocol. T2D islet samples were randomized across the chips and all samples were analyzed on the same machine by the same technician to reduce batch effects. After single base extension and staining, the BeadChips were imaged with the Illumina iScan. Raw fluorescence intensities of the scanned images were extracted with the GenomeStudio (V2011.1) Methylation module (1.9.0) (Illumina). The fluorescence intensity ratio was used to calculate a β-value which corresponds to the methylation score for each analyzed site according to the following equation: β-value=intensity of the Methylated allele (M)/(intensity of the Unmethylated allele (U)+intensity of the Methylated allele (M)+100). DNA methylation β-values range from 0 (completely unmethylated) to 1 (completely methylated). All samples had high bisulfite conversion efficiency (intensity signal >4000) and they were included for further analysis based on GenomeStudio quality control steps where control probes for staining, hybridization, extension and specificity were examined. The intensity of both sample dependent and sample independent built in controls was checked for the red and green channels using GenomeStudio.

We next exported the DNA methylation data from GenomeStudio and used Bioconductor [82] and the lumi package [83] for further analyses. Individual probes were filtered based on their mean detection P-value and those with a P-value>0.01 were excluded from further analysis. As a result, DNA methylation data for 483,031 (99.5%) probes, including 479,927 CpG sites and 3,039 non-CpG sites were used for further analysis. Because M-values are more statistically valid [84], β-values were converted to M-values using the following equation: M=log₂ β-value/(1−β-value). M-values were then used for further statistical analysis [84]. In order to correct for background fluorescence the median M-value of the built in negative controls was subtracted from M-values. Next a quantile normalization was performed as described [85]. The Universal Methylated Human DNA standard D5011 (Zymo Research), which is human DNA that has been enzymatically methylated in the CpG sites by M.SssI methyltransferase, was used as a positive control in every batch and it showed high levels of methylation (87.7±0.6%). The low intensity of Y chromosome loci in female samples was used as an additional control. Moreover, one human pancreatic islet sample was run in two different batches on two different days and used as a technical replicate. As the β-value is easier to interpret biologically, M-values were reconverted to β-values when describing the results and creating the figures.

Pathway analysis

The enrichment of KEGG pathways among genes that exhibit differential DNA methylation in T2D compared with non-diabetic islets was tested using WebGestalt (http://bioinfo.vanderbilt.edu/webgestalt, March 2012).

Microarray mRNA expression analysis

mRNA expression of the human pancreatic islets was analyzed using the GeneChip Human Gene 1.0 ST array from Affymetrix (Santa Clara, CA, USA) as previously described [63].

Luciferase assays

1500 bp of the human CDKN1A or PDE7B promoters (sequences are given in Table S12) were inserted into a CpG-free firefly luciferase reporter vector (pCpGL-basic) kindly provided by Dr Klug and Dr Rehli [86]. Amplification of CDKN1A and PDE7B DNA sequences and insertion into the pCpGL-basic vector was done by GenScript (Piscataway, NJ, USA). The constructs were either mock-methylated or methylated using two different DNA methyltransferases; SssI and HhaI (New England Biolabs, Frankfurt am Main, Germany). While SssI methylates all cytosine residues within the double stranded dinucleotide recognition sequence CG, HhaI only methylates the internal cytosine residue in GCGC sequence. INS-1 832/13 β-cells were co-transfected with 100 ng pCpGL-vector either with or without respective insert together with 2 ng of pRL renilla luciferase control reporter vector (pRL-CMV vector, Promega, Madison, USA) as a control for transfection efficiency and luciferase activity was measured as previously described [12].

Overexpression of Cdkn1a, Pde7b and Sept9 in clonal β- and α-cells

INS-1 832/13 β-cells were cultured as previously described [87] and αTC1-6 cells were cultured according to ATCC's instructions (ATCC, Manassas, VA). pcDNA3.1 expression vectors with rat cDNA for either Cdkn1a, Pde7b or Sept9, or the empty vector, were transfected into β- or α-cells with Lipofectamine LTX and Plus Reagent (Life Technologies, Paisley, UK), according to the manufacturer's instructions (sequences for Cdkn1a, Pde7b or Sept9 are given in Table S13). Overexpression was verified with real-time PCR using an ABI 7900 system (Applied Biosystems, Foster City, CA, USA) and a SYBR Green assay for Cdkn1a (fwd-primer: ATGTCCGACCTGTTCCACAC, rev-primer: CAGACGTAGTTGCCCTCCAG) or TaqMan assays (Life Technologies) for Pde7b (Rn00590117_m1) and Sept9 (Rn00582942_m1). Cyclophilin B (Rn03302274_m1 and Mm00478295_m1) was used as an endogenous control. Expression levels were calculated with the ΔΔCt method. Overexpression was also verified by Western Blot analysis and cells transfected with HA-tagged cDNAs for Cdkn1a, Pde7b and Sept9 were lysed in RIPA buffer (50 mM Tris pH 7.6, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton×100 and 1× protease inhibitor cocktail (P8340, Sigma-Aldrich, USA) and boiled with 6× sample buffer (60 mM Tris pH 6.8, 10% glycerol, 2% SDS, 10% β-mercaptoethanol and bromophenol blue). Samples were separated on gradient Mini-PROTEAN® TGX gels (Bio-Rad, Hercules, CA, USA) and transferred onto Hybond-LFP PVDF membranes (GE Healthcare, Piscataway, NJ, USA). Protein expression was detected with primary antibodies against HA tag (ab9110, Abcam Cambridge, UK) and β-actin (A5441, Sigma-Aldrich) and secondary DyLight 680/800 conjugated anti-mouse and anti-rabbit antibodies (35518 and 35571, Thermo Scientific, Rockford, USA) and blots were scanned in an ODYSSEY (Licor, Lincoln, NE, USA).

Insulin secretion and content

48 hours post transfection of INS-1 832/13 β-cells, insulin secretion with indicated secretagogues was determined during 1 hour static incubations as previously described [87]. Insulin content of cells was determined after acid ethanol extraction of the hormone. Insulin secretion was normalized to total insulin content.

Glucagon secretion

αTC1-6 cells were transfected as described above. 48 hours post transfection clonal α-cells were pre-incubated in HEPES balanced salt solution (HBSS, 114 mM NaCl; 4.7 mM KCl; 1.2 mM KH₂PO₄; 1.16 mM MgSO₄; 20 mM HEPES; 2.5 mM CaCl₂; 25.5 mM NaHCO₃; 0.2% BSA, pH 7.2) supplemented with 5.5 mM glucose. Secretion was then stimulated in 1 hour static incubation with HBSS supplemented with 1 or 16.7 mM glucose. Secreted glucagon was measured with a glucagon ELISA (Mercodia, Uppsala, Sweden) and normalized to total protein as determined by a BCA assay (Thermo Scientific).

Proliferation assay

INS-1 832/13 β-cells were transfected as described above. 72 hours post transfection the β-cells were washed with PBS and stained with 0.1% crystal violet in 0.15 M NaCl. Cells were then washed with water and allowed to dry. Methanol was added to wells and absorbance measured at 600 nm in an Infinite M200 plate reader (Tecan, Männerdorf, Switzerland).

RNA interference

INS-1 832/13 β-cells were transfected with Lipofectamine RNAiMAX (LifeTechnologies) according to the manufacturer's instructions with siRNA targeting Exoc3l (LifeTechnologies, ID: s146127) or negative control siRNA (5′-GAGACCCUAUCCGUGAUUAUU-3′). Following 24 hours incubation, cells were transferred onto Petri dishes and cultured another 24 hours. Exoc3l knock-down was verified with real-time PCR using an ABI 7900 system and assays for Exoc3l (Rn01432027_m1) and endogenous controls (Cyclophilin B, Rn03302274_m1 and Hprt, Rn01527840_m1) (Life Technologies).

Electrophysiological measurements of exocytosis using the patch-clamp technique

Electrophysiological measurements of exocytosis were performed on INS-1 832/13 β-cells as described [88].

Analysis of DNA methylation of selected genomic regions

Pyrosequencing was used to technically validate the Infinium HumanMethylation450 BeadChip DNA methylation data. EpiTect Bisulfite Kit (Qiagen) was used for bisulfite conversion of human islet DNA. Primers were designed using the PyroMark Assay design Software 2.0 (Qiagen). Sequences are included in Table S14. Bisulfite converted DNA was amplified with the PyroMark PCR kit. Pyrosequencing was performed with PyroMark ID 96 and PyroMark Gold Q96 reagents (Qiagen) according to the manufacturer's instructions. Data were analyzed with the PyroMark Q96 2.5.7 software program.

Sequenom's MassARRAY EpiTYPER protocol (Sequenom, San Diego, CA, USA) was used to measure DNA methylation of PDX1 in its distal promoter and enhancer regions according to our previous study [12].

We thank the Nordic Network for Clinical Islet Transplantation (JDRF award 31-2008-413), the tissue isolation teams and Human Tissue Laboratory within EXODIAB/Lund University Diabetes Centre for providing human pancreatic islets. We also thank SCIBLU (Swegene Center for Integrative Biology at Lund University) Genomics Facility for analysis of DNA methylation and gene expression, Dr. Marloes Dekker Nitert and Jelena Stamenkovic for technical support.