The human gene expression data of AD and T2DM were collected from the Gene Expression Omnibus (GEO) database (Barrett et al., 2013). For AD, GSE159699 includes 12 disease and 10 normal samples while GSE28146 data include 22 disease and 8 normal samples (Blalock et al., 2011; Nativio et al., 2020). For T2DM, GSE38642 consists of nine diseases and 54 normal islets samples while the GSE164416 contains 39 disease and 18 control islets samples, which are selected from 133 samples among four diabetes statuses (Taneera et al., 2012; Wigger et al., 2021). For each disease, RNA-seq and microarray platforms were separately used for the two datasets. Specifically, GSE159699 and GSE164416 were detected using RNA-seq platform while GSE28146 and GSE38642 were measured using microarray platform. The platform information is detailed in Table 1.

The Affymetrix GeneChip data are preprocessed by RMA (Robust Multi-array Analysis) (Bolstad et al., 2003). The annotation soft tables downloaded from the corresponding GPL platform were used for assigning Probe ID to Gene Symbol. Probes with ambiguous or multiple gene symbols were removed. Averaged the expression intensity when replicated probes mapping to the identical gene. The entire expression matrix was log2 transformed. Fold Change (FC) and Mann-Whitney test were used to identify the differentially expressed genes (DEGs). Genes with |FC|>1.5 and Mann-Whitney test p-value<0.05 were defined as DEGs. Considering that different array platforms have different gene coverage, we just studied genes presented in all the analyzed datasets. Reads per kilo base per million mapped reads (FPKM) was used to measure the expression intensity for the RNA-seq data and the data was preprocessed using the methods mentioned in the original papers. All the analysis was carried out using R-4.1.2.

The human protein-protein interaction (PPI) data were derived from the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING version 11.5) (Szklarczyk et al., 2019). Interaction with a score greater than 0.4 were used to build a high-confidence network with 38 edges and 166 nodes. Network degree was defined as the number of neighbors linking a protein. PPI network was generated and illustrated by Cytoscape (version 3.9.1) (Shannon et al., 2003).

Monte Carlo simulation, also known as multiple probability simulation, is a mathematical technique used to estimate the possible outcomes of an uncertain event. p-values are calculated to see whether observed values are unusually large or small for the null distribution. This calculation compares the observed value to the upper/lower tails of the null distribution to explore whether the observed value is significantly large/small for the distribution. p = ( m + 1 ) / ( n + 1 ) (1) where n is the total number of Monte Carlo simulations, m is the number of simulations for which the statistic was greater than or equal to the observed statistic. One (1) is added to the numerator and denominator because the observed statistic is included in the reference distribution.

In this study, to access the statistical significance of an observed SS score, we randomly selected two gene sets with the same sizes as the two original sets from the background genes detected by both diseases. We then calculated p-value using these two gene sets. After 10,000 Monte Carlo random experiments, the significance level (or p-value) for an observed SS was calculated as the proportion of random scores higher than the observed score.

The hypergeometric distribution model was used to evaluate the significance of enrichment analysis. The probability of observing at least k genes annotated in a specific term is calculated as follows: p = 1 − ∑ k = 0 q − 1 ( t k ) ( m − t n − k ) ( m n ) (2) where k is the number of genes of interest, n is the total number of detected genes, m is the term size or the number of genes in a term, and t is the number of the overlap genes annotated in the term. The resulting p-value was then multi-test adjusted by the BH correction (FDR< 0.05).

Functional enrichment analysis of Gene Ontology (GO) (Gene, 2021) and Kyoto encyclopedia of genes and genomes (KEGG) (Kanehisa et al., 2021) pathway was performed to determine significantly enriched gene functions using the R package ‘clusterpProfiler’ (R version 4.1.2) (Yu et al., 2012). Three pathways, Glucagon Signaling Pathway, Carbohydrate Digestion and Absorption and Central Carbon Metabolism in Cancer, are collected and highlighted from KEGG.

Five semantic similarity methods were used to evaluate the functional similarity between two gene sets, i.e., Wang, Rel, Jiang, Lin and Resnik (Yu et al., 2012). ‘GOSemSim’ was used for the calculation of these scores.

Datasets GSE159699 and GSE28146 were used to identify the differentially expressed genes (DEGs) for AD. Genes with fold change (FC) larger than 1.5 and p-value less than 0.05 were identified as DEGs for subsequent analysis. In total, 2323 and 800 DEGs were identified for GSE159699 and GSE28146 (Figures 1A,B), respectively, in which 156 were commonly detected as DEGs (Figure 1C) and hereafter we defined it as AD DEGs. For these genes, the expression difference between the AD and control samples are shown in Figure 1D and Figure 1E for GSE159699 and GSE28146, respectively, which illustrates that the two groups of samples can be clearly stratified.

These genes are mainly involved in biological processes of regulation of amine transport, regulation of trans-synapic signaling, learning of memory, etc. (Figure 1F) and locate in transport vesicle, synaptic vesicle, synaptic vesicle membrane, etc. (Figure 1G). Also, they are implemented in biological pathways of synaptic vesical cycle, collecting duct acid secretion, glucolysis, etc. (Figure 1H).

Datasets GSE164416 and GSE38642 were used to identify DEGs for T2DM. 1790 and 65 DEGs were identified for GSE159699 and GSE28146 (Figures 2A,B), respectively. The 28 DEGs commonly identified from them were T2DM DEGs (Figure 2C). For these genes, the expression difference between the T2DM and control samples is clear based on these T2DM DEGS for GSE164416 and GSE38642 (Figures 2D,E).

These genes are enriched in biological processes of positive regulation of T cell differentiation in thymus, regulation of chemokine production regulation of lymphocyte apoptotic process, etc. (Figure 1F) and resident in azurophil granule membrane, coated vesicle, endocytic vesicle, lysosomal membrane, etc. (Figure 1G). Also, they are implemented in biological pathways of fatty acid metabolism, insulin secretion, arginine biosynthesis, etc. (Figure 1H).

Dozens of DEGs were commonly detected by different datasets for each disease, whereas no common DEGs were identified between AD and T2DM (Figure 3A), given that the two diseases are strongly associated. However, these DEGs are closely related to each other from the function perspective. Specifically, the semantic similarity (SS) scores among the four datasets are generally over 0.8, especially for the three datasets GSE28146, GSE159699 and GSE164416 (SS > 0.9, Figure 3B). Figure C shows the SS scores between the 156 AD DEGs and 28 T2DM DEGs. Using Monte Carlo simulation, we randomly selected the same number of DEGs 1,000 times and calculate the SS scores to build a simulated SS distribution. The detected SS score between the 156 AD DEGs and 28 T2DM DEGs is 0.649, which is significantly higher than most of the simulated ones (p < 0.01, Figure 3D), indicating that AD and T2DM are close to each other from the point of biological function.

Using the other four SS methods, also, we calculated the SS scores and simulated SS distributions (Figures 3E–H), resulting in the SS scores of 0.748, 0.732, 0.76 and 0.421 for Rel, Jiang, Lin and Resnik, respectively. As expected, the SS scores of the simulated data are consistently lower than the real SS scores (p < 0.01), demonstrating that the DEGs of the two diseases are involved in some common pathways or functional modules.

To investigate the associations between the two diseases from the point of protein interaction, we constructed a protein-protein interaction (PPI) network using the AD DEGs and T2DM DEGs. A network consisting of 184 genes, 156 AD DEGs and 28 T2DM DEGs, was illustrated in Figure 4A and the genes with high degree were highlighted (AD in orange and T2DM in green). 38 edges were observed connecting the two gene set. To test whether the two gene sets are closely connected with each other, we randomly selected the same number of genes for each set 1,000 times and calculated the connectivity between them. Distribution of the random data was built and only a few simulations were larger than 38 (Figure 3B).

The hub genes or genes with the highest degree for AD and T2DM were shown in Figures 4C,D, respectively. The AD hub genes consist of G6PC2, PCSK1, ACADSB, and BDNF, while SLC2A2, PFKFB2, GLP1R, SLC1A1, and CHL1 are the hub genes of T2DM. It is apparent that the hub DEGs are connected much denser than the other DEGs, suggesting they are involved in biological pathways linking AD and T2DM, such as carbohydrate digestion and absorption, glycolysis, glucagon signaling pathway, etc. (Figure 4E). Among these hub genes, SLC2A2, a T2DM hub gene, was up-regulated and all the others were down-regulated. Also, it has the largest connectivity of 6 and connects most of the AD hub genes, suggesting that it is a pivot mediating or triggering the development from T2DM to AD. Additionally, we observed that most of the hub DEGs were down-regulated, seven out of eight in AD and four out of five in T2DM (Figure 5).

The glucagon signaling pathway is a process of a series of elevated blood glucose enzymatic reactions triggered by the binding of glucagon that produced by pancreatic islets alpha cells to the glucagon receptor on the surface of liver cells. Glucagon signaling pathway mainly assists glucagon to exert its role of raising blood glucose to sustain blood glucose homeostasis in the body and synergizes with insulin. In this pathway, GLUT2, encoded by SLC2A2, is working as a transmembrane carrier protein that enables protein facilitate glucose movement across cell membranes. Two AD hub genes/proteins, G6PC2 and LDHA that are interacted with GLUT2, are implemented in the pathway (Figure 4F).

In the carbohydrate digestion and absorption pathway, GLUT2 is critical and ubiquitous in carbohydrate transport (Figure 6). Glucose and galactose are initially transported into the enterocyte by SGLT1 located in the apical brush border membrane (BBM) and then exit through the basolateral membrane by GLUT2 or release out through exocytosis by HK1 and G6PC2 located in Endoplasmic Reticulum (ER). In intestinal glucose absorption, transport by SGLT1 induces rapid insertion and activation of GLUT2 in the BBM by a PKCβII-dependent mechanism. Moreover, trafficking of apical GLUT2 is rapidly promoted by glucose, which acts through T1R2 + T1R3/alpha-gustducin to activate PLCβ2 and PKCβII.