Differential Transcriptome Analysis of Diabetes-Resistant and -Sensitive Mouse Islets Reveals Significant Overlap With Human Diabetes Susceptibility Genes

The susceptibility for type 2 diabetes (T2D) in humans is, to a substantial degree, inherited. Accordingly, genome-wide association studies (GWAS) of common single nucleotide polymorphisms in large cohorts have established associations between ∼100 genes and T2D (1–4). However, these associations explain only a small portion of the total heritability. Furthermore, little direct functional evidence links the genetic variants with the pathogenesis of the disease. In several mouse models, diabetes exhibits all features of the human pathogenesis with obesity-associated insulin resistance, failure of insulin secretion, and subsequent β-cell loss. Thus, it appears reasonable to assume that the cellular mechanism of the pathogenesis of diabetes in mice and humans is similar and that at least some of the genes involved are identical. Furthermore, identification of such an overlap would validate the GWAS results by providing biological plausibility. Thus, we tested this hypothesis by aligning human GWAS results with data from a genome-wide screen of islets from a mouse model of T2D. This screen was based on a comparison of a diabetes-sensitive with a diabetes-resistant obese mouse strain subjected to sequential dietary fat and carbohydrate challenges. The alignment indicated that 20 of the 106 genes identified in the GWAS were also detected in the screen of mouse islets.

Male NZO/HIBomDife mice from our own colony and male B6.V-Lep^ob/ob/JBomTac (B6-ob/ob) mice (Charles River Laboratories, Calco, Italy) were housed as described previously (5). All experiments were approved by the Ethics Committee of the State Ministry of Agriculture, Nutrition and Forestry (State of Brandenburg, Germany).

After weaning, all mice received a carbohydrate-free diet (−CH) containing 68% fat weight for weight (w/w) and 20% protein (w/w), with a total metabolizable energy of 29.3 kJ/g. At the age of 18 ± 1 weeks, subgroups of the animals received a carbohydrate-enriched diet (+CH) containing 28% fat (w/w), 20% protein (w/w), and 40% (w/w) metabolizable carbohydrates with an energy content of 21.9 kJ/g (6).

Immunohistochemistry of insulin was performed as described earlier (5).

Isolation of islets was performed by a modified protocol of Gotoh et al. (7) from 18 ± 1-week-old NZO and B6-ob/ob mice. Total islet RNA preparation was performed with the RNAqueous-Micro Kit (Life Technologies, Darmstadt, Germany). RNA integrity was assessed with the RNA 6000 Nano Kit (Agilent, Santa Clara, CA). RNA sequencing (RNAseq) and evaluation of data were performed by LGC Genomics (Berlin, Germany).

Differences in blood glucose were tested with a one-way ANOVA. All other comparisons were tested with the nonparametric Kruskal-Wallis test and Dunn corrected for multiple comparisons. All comparisons were considered significantly different at P < 0.05. RNAseq read counts were tested by two statistical evaluations [DESeq (8) and edgeR (9)] of the R software package in combination with a Benjamini-Hochberg error correction. Here, all comparisons were considered significantly different at P < 0.05 and a log₂ fold change (FC) >|0.6|. For testing the specificity of the overlap of differentially expressed islet genes with genes identified by GWAS, we used contingency tables in combination with the Fisher enrichment exact test and the χ² test (10).

The use of a specific feeding regimen of initial carbohydrate restriction followed by a carbohydrate challenge (Fig. 1A) induced glucolipotoxicity in NZO mice, resulting in a synchronized fast β-cell loss through apoptosis within 16 days (5). To test how a diabetes-resistant obese mouse strain reacts to these conditions, we put B6-ob/ob mice on the same dietary regimen and compared them with NZO mice. Before the switch from the carbohydrate-free to the diabetogenic carbohydrate-containing diet, both strains had developed severe obesity. In accordance with earlier findings (5,6), the carbohydrate deprivation protected NZO mice from hyperglycemia (Fig. 1B). Before the diet switch, NZO mice had elevated blood glucose concentrations (9.8 ± 0.6 mmol/L) compared with B6-ob/ob mice (6.6 ± 0.3 mmol/L), and their plasma insulin levels were similar to B6-ob/ob mice (Fig. 1C). Two days after the carbohydrate challenge, NZO mice developed hyperglycemia, with blood glucose concentrations increasing continuously until day 32 (>22 mmol/L) followed by an increased insulin secretion until day 8 and a striking decrease thereafter. In contrast, B6-ob/ob mice responded with a transient and moderate elevation of blood glucose levels until day 4 (peak value 12.2 ± 1.2 mmol/L) and returned to normoglycemia afterward (Fig. 1B), whereas insulin levels increased significantly during carbohydrate feeding (Fig. 1C).

As previously shown (5) and as indicated in Fig. 1D, NZO islets exhibited a rapid decrease of insulin content and alterations of islet histology between days 16 and 32. In contrast, B6-ob/ob mice maintained normal islet structure and insulin stores throughout the 32-day carbohydrate challenge (Fig. 1D).

To identify genes that are activated or repressed in the resistant B6-ob/ob islets and thereby participate in the prevention of diabetes, we isolated islets of NZO and B6-ob/ob mice at day 2 after initiation of the carbohydrate challenge and performed RNAseq-based transcriptome analyses. We obtained 63,251,849, 45,623,816, and 68,092,230 sequence reads for NZO and 71,641,278, 63,210,263, and 69,422,841 for B6-ob/ob islets and filtered the reads for high-quality ones by trimming off the base pairs with a low-quality score assigned to them during downline processing of RNAseq. Furthermore, genes with a read count <25 were excluded. About 67% of the reads passed the quality filter and were mapped to the mouse genome for gene annotation. For the comparison of NZO and B6-ob/ob transcripts, we used data after processing by two statistical evaluations (DESeq and edgeR). Using a cutoff value of 5 reads per kilobase per million, we identified 31,340 transcripts in NZO islets corresponding to 14,192 genes and 31,394 transcripts in B6-ob/ob islets corresponding to 14,166 genes. An alignment of these transcripts from NZO and B6-ob/ob islets revealed 2,328 genes with significant differences (log₂ FC >|0.6|); among these, 1,057 exhibited an elevated expression in NZO islets and 1,271 in B6-ob/ob islets.

By combining recent meta-analyses (2–4), we currently count 108 genetic variants that predict diabetes-associated traits (Table 1). Many of these variants are located in introns of known genes or between known genes, suggesting that they affect their expression. Therefore, as a functional validation of these genes, we studied the overlap between human diabetes genes that have mouse orthologs (106) and the genes differentially expressed in islets of diabetes-prone and diabetes-resistant mice. This alignment revealed a statistically significant overlap of 20 genes (P = 0.0103 [Fisher], P = 0.009 [χ²]) (Table 1). To further test the specificity of this correlation, we aligned the present data with 73 genes that associate with inflammatory bowel disease (IBD) (11) or with 185 genes that associate with the trait height (12) and detected an overlap of 9 genes (P = 0.57 [Fisher], P = 0.77 [χ²]) and 23 genes (P = 0.401 [Fisher], P = 0.49 [χ²]), respectively. The random overlap as estimated by contingency tables was 11 genes for T2D, 8 for IBD, and 20 for height, suggesting that GWAS-derived markers for IBD and height, in contrast to the diabetes loci, do not specifically overlap with islet gene expression data.

To further validate the candidates in the overlap, we studied the effect of the carbohydrate challenge by examining their expression in islet samples of both strains before and 2 days after the diet switch by quantitative RT-PCR (Fig. 2B). Six genes (e.g., Adamts9, Adcy5, Igf2bp2) were differentially expressed in response to the glucose challenge, 10 (e.g., Cdkn2a, Grb10, Tcf7l2) differed between the two mouse strains, and 4 (e.g., Ankrd55, Ysk4) differed both between strains and in response to the carbohydrate challenge. In addition, measurement of the expression of the 20 candidates in lean 6-week-old NZO and B6-ob/ob mice indicated that levels of several transcripts were associated with the development of obesity (e.g., Ysk4) (Supplementary Fig. 1).

The present data indicate that islets of diabetes-prone and diabetes-resistant mouse strains show opposite reactions in response to the carbohydrate challenge: B6-ob/ob mice stay normoglycemic presumably by adapting β-cell mass and raising the plasma insulin concentration, whereas NZO mice fail to compensate and become hyperglycemic within a few days of carbohydrate intervention. Similar to the two mouse models used in this study, only a subset of humans with risk factors (obesity, insulin resistance) develop T2D, whereas others do not, depending on a genetic predisposition (13). Thus, it appears reasonable to assume that diabetes-resistant individuals are protected by their capacity to increase β-cell mass. Indeed, by comparing data of human GWAS with mouse transcriptomics, we found 20 orthologs of 106 human diabetes genes to be differentially expressed in diabetes-susceptible and diabetes-resistant mice, including the validated candidate genes Tcf7l2, Grb10, and Igfbp2. Recently, Prokopenko et al. (14) provided functional evidence for a complex regulation of GRB10 in human islets by showing that suppression of GRB10 reduced insulin and glucagon secretion. The authors suggested that a tissue-specific methylation and imprinting of GRB10 influences glucose metabolism and contributes to T2D pathogenesis. An additional analysis of the expression of the 20 genes (Fig. 2B) showed complex patterns of differential expression, depending on response to glucose (6), the genetic background (11), or both (4). For elucidation of the diabetes resistance of the B6-ob/ob mouse, genes exhibiting differences both between strains and in response to glucose might be most promising (Ankrd55, Igf2bp2, Prc1, Ysk4). Other promising candidates are Cdkn2a, Cdkn2b, and Prc1. These genes are known to be involved in cell cycle regulation and proliferation (15,16) and exhibited a differential expression, depending on the mouse strain. Other candidates such as Dgkb, Adcy5, Ppp1r3b, and Kl have been shown to regulate signal transduction. Dgkb (diacylglycerol kinase B) is a regulator of the intracellular concentration of the second messenger diacylglycerol and thus plays a key role in cellular processes (17). Adcy5 (adenylyl cyclase 5) mediates G protein–coupled receptor signaling through the synthesis of the second messenger cAMP (18). Ppp1r3b (protein phosphatase 1, regulatory subunit 3B) encodes the catalytic subunit of the serine/threonine-protein phosphatase 1, is expressed in liver and skeletal muscle tissue, and is involved in regulating glycogen synthesis in these tissues (19). Its function in pancreatic islets is still unknown. Kl (β-klotho), a cofactor of fibroblast growth factor receptor that is activated by FGF21, is repressed in islets of the diabetic db/db mouse in response to high glucose concentrations (20). Finally, we found genes with unknown function that also appear in GWAS of other diseases, such as ANKRD55, which is also a multiple sclerosis risk gene (21). Recently, Pasquali et al. (22) examined the function of human islet cis-regulatory networks and demonstrated by chromatin immunoprecipitation sequencing that five important transcription factors (FOXA2, MAFB, NKX2.2, NKX6J, and PDX1) bind to distinct accessible chromatin states located in clusters of enhancers. Of note, single nucleotide polymorphisms associating with traits of T2D were found to be enriched in these clusters.

The overlap between the human and mouse diabetes-associated genes was statistically significant and observed in a comparison of two biologically highly plausible screening approaches. Furthermore, additional biological plausibility of the overlap is provided by transcripts involved in the regulation of cell cycle and proliferation, such as Cdkn2a, Cdkn2b, Grb10, Prc1, and Tcf7l2 (Fig. 2A). However, we also observed some overlap with GWAS results that are, if at all, only remotely associated with diabetes (IBD, height). Thus, we cannot exclude that the alignment of 106 GWAS hits with 2,328 mouse genes produced false positive in addition to true candidates. In conclusion, we believe that the present ex vivo comparison of islets from diabetes-susceptible and diabetes-resistant mouse strains is a useful tool to investigate the pathogenesis of diabetes and to validate diabetes genes identified in human studies.

Acknowledgments. The authors thank Anett Helms and Andrea Teichmann, both from the Department of Experimental Diabetology, German Institute of Human Nutrition, for skillful technical assistance.

Funding. The study was supported by grants from the German Research Foundation (DFG, GK1208), the German Ministry of Education and Research (BMBF: DZD, 01GI0922), and the State of Brandenburg.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. O.K., D.M., G.S., and R.W.S. contributed to the data analysis. H.-G.J. contributed to the study concept and reviewed and edited the manuscript. A.S. contributed to the study concept, data interpretation, and writing of the manuscript. A.S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.