Integrative analysis of the transcriptome profiles observed in type 1, type 2 and gestational diabetes mellitus reveals the role of inflammation

Abstract

Background: Type 1 diabetes (T1D) is an autoimmune disease, while type 2 (T2D) and gestational diabetes (GDM) are considered metabolic disturbances. In a previous study evaluating the transcript profiling of peripheral mononuclear blood cells obtained from T1D, T2D and GDM patients we showed that the gene profile of T1D patients was closer to GDM than to T2D. To understand the influence of demographical, clinical, laboratory, pathogenetic and treatment features on the diabetes transcript profiling, we performed an analysis integrating these features with the gene expression profiles of the annotated genes included in databases containing information regarding GWAS and immune cell expression signatures.

Methods: Samples from 56 (19 T1D, 20 T2D, and 17 GDM) patients were hybridized to whole genome one-color Agilent 4x44k microarrays. Non-informative genes were filtered by partitioning, and differentially expressed genes were obtained by rank product analysis. Functional analyses were carried out using the DAVID database, and module maps were constructed using the Genomica tool.

Results: The functional analyses were able to discriminate between T1D and GDM patients based on genes involved in inflammation. Module maps of differentially expressed genes revealed that modulated genes: i) exhibited transcription profiles typical of macrophage and dendritic cells; ii) had been previously associated with diabetic complications by association and by meta-analysis studies, and iii) were influenced by disease duration, obesity, number of gestations, glucose serum levels and the use of medications, such as metformin.

Conclusion: This is the first module map study to show the influence of epidemiological, clinical, laboratory, immunopathogenic and treatment features on the transcription profiles of T1D, T2D and GDM patients.

Figure 1

Heatmap representative of type of diabetes, demographic, clinical, laboratory and treatment features of the patients. Qualitative variables were assigned by the absence or presence of the characteristic, and quantitative variables were assigned by values below or above the mean values. This information was used as array (experimental) set for the module map construction.

Figure 2

Principal component analysis (PCA) of the major types of diabetic patients, using the 8,469 informative genes obtained by the DBF-MCL algorithm. The separation of samples of each type of diabetes after filtering non-informative genes showed similarities among them, indicating that sample transcription profiles were not influenced by the batch effect.

Figure 3

Heatmap of the significant functional categories of the genes of the clusters obtained by non-informative filters, in the basis of a FDR ≤ 0.1. The Kegg categories were obtained from DAVID knowledge base with an enrichment P value ≤ 0.05 after Benjamini correction. The grey scale represents the logarithm of the enriched P value.

Figure 4

Venn diagrams show the differentially expressed genes after paired analysis of the three types of DM. The genes were identified by Rank Product analysis with P value ≤ 0.001 and a percentage of false prediction (pfp) ≤ 0.05. The analysis referring to upregulated genes is shown in panel A and that of downregulated genes in panel B.

Figure 5

Heatmap of the significative functional categories of the differentially expressed genes obtained by paired Rank Products analysis with P value ≤ 0.001 and percentage of false prediction (pfp) ≤ 0.05 (T1D vs. GDM; T2D vs. GDM and T1D vs. T2D). The Kegg categories were obtained from DAVID knowledge base with an enrichment pvalue ≤ 0.05 after Benjamini correction. The grey scale represents the logarithm of the enriched P value.

Figure 6

Heatmaps of the modules identified by Genomica tool, which compares gene lists of immune cells and diabetic association genes with demographic, clinical, laboratory and treatment features of patients (P value ≤ 0.05, corrected by the false discovery rate - FDR ≤ 0.05. The four module maps presented list of genes (induced or repressed), identified by: A) Non-informative filters (DBF-MCL algorithm) in the basis of FDR ≤ 10%; B) Rank products analysis of T1D vs. GDM; C) Rank products analysis of T2D vs. GDM; D) Rank Products analysis of T1D vs. T2D. Abbreviations: MF - macrophages ; B1a and B1b - subsets of B lymphocytes; BFo - follicular B lymphocytes; BMz - marginal zone B lymphocytes; Treg - regulatory T lymphocytes; DC - dendritic cells ; CD4 and CD8 - subsets of T lymphocytes.

Figure 7

Confirmation of microarray findings by qRT-PCR of (A) IL1B, (B) RGS1, (C) EGR2, (D) FOXO3A, (E) SOD2 and (F) HIF1A genes. Expression levels were normalized to HPRT1. The differences were evaluated by Mann–Whitney U test. * p < 0.05; ** p < 0.01 and *** p < 0.001 were considered significant.