Our study samples and data were sourced from the NHANES (https://wwwn.cdc.gov/nchs/nhanes/) from 2001 to 2018. NHANES is a nationally representative survey of the non-institutionalized civilian population in the US, and the survey involved interviews conducted at participants’ homes and standardized physical examinations, including laboratory tests, performed at mobile screening centers (MEC).

To identify pertinent datasets for our investigation, we conducted a comprehensive search of the NCBI GEO database (https://www.ncbi.nlm.nih.gov/geo/) using specific medical keywords such as “Type 2 diabetes mellitus”, “Atherosclerosis”, “Homo sapiens”, “Expression profiling by array”, and “expression profiling analysis”. The objective was to pinpoint datasets that met stringent criteria: they had to contain archived information on both case and control groups, offer raw data for further analysis, and enable expression analysis using array methods. Additionally, our search was limited to datasets exclusively featuring data from Homo sapiens ( Figure 1 ).

Microarray analysis was performed using two different platforms: the GSE40231 dataset employed the GPL570 (Affymetrix Human Genome U133 Plus2 microarrays), whereas the GSE9006 dataset utilized the GPL96 (Illumina HumanHT-12 V4.0 Expression Bead Chip). The T2DM dataset (GSE9006) comprised gene expression data from peripheral blood mononuclear cells (PBMC) collected from 24 healthy people and 12 newly diagnosed with T2DM. We utilized 40 samples of Atherosclerotic Arterial Wall (AAW) and 40 samples of Non-Atherosclerotic Arterial Wall (NAW) from GSE40231 to identify DEGs, including Differentially Expressed mRNAs (DEmRNAs) and Long Non-Coding RNAs (DElncRNAs). Additionally, we validated the diagnostic efficacy of essential genes using datasets from two different platforms: the T2DM (GSE71416) from the GPL570 platform, which included 14 morbidly obese diabetic patients (cases) and six morbidly obese non-diabetic patients, and the AS dataset (GSE43292) utilizing 32 AAW samples and 32 NAW samples.

In this cohort study, we selected adult participants from the NHANES spanning from 2001 to 2018, totaling 91,351 individuals. The inclusion criteria mandated participants to be at least 20 years old and not pregnant, narrowing down the cohort to 48,943 participants. Following the exclusion of individuals with missing data on fasting blood glucose (FBG), hemoglobin A1c (HbA1c), triglyceride (TG), total cholesterol, low-density lipoprotein (LDL), and high-density lipoprotein (HDL), the final participant was 9,357.

To identity individuals withT2DM, we adhered to the American Diabetes Association’s diagnostic criteria, which include:(1)a self-reported physician diagnosis of diabetes; (2) the use of oral hypoglycemic agents or insulin for treatment; (3) a fasting plasma glucose level of at least 126 mg/dL; (4) an HbA1c level of at least 6.5%. Following these criteria, 1,829 participants were classified into the diabetes group, whereas 7,528 were allocated to the control group. Our study rigorously followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines to ensure the highest level of clarity, transparency, and rigor in reporting the observational study findings.

We used R software version 4.2.2 and processed raw matrices downloaded from the datasets. Data normalization was performed using the RMA algorithm after preprocessing and converting probe IDs to gene symbols using annotated platform files. Empty probes were removed, and values for genes with multiple probes were averaged to enhance result reliability.

Separate analyses were conducted for AS and T2DM datasets. The limma package was employed with stringent criteria (|logFC| >1 and Padj< 0.05) to identify DEGs. This approach identified genes with significant expression changes between case and control groups. An overlap analysis of DEGs from T2DM and AS datasets was also performed. We identified common DEGs to uncover potential molecular links between T2DM and AS. Venn Analytics was used for this analysis, allowing for a comprehensive evaluation of shared genes (11). These overlapping DEGs form the basis for further exploration into the molecular mechanisms connecting T2DM and AS.

We conducted an integrated analysis using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) to explore biological functions and pathways associated with the identified genes (12–14). Visual representations were generated using the ggplot2 package in R4.2.2, facilitating a clear understanding of enriched pathways and their significance. Pathways with a P-value < 0.05 were considered statistically significant, indicating robust associations between T2DM and AS.

We applied weighted gene co-expression network analysis (WGCNA) to assess gene expression patterns in extensive T2DM and AS datasets (15). Genes with significant Padj values (P <0.05) and absolute logarithmic changes greater than 1 were chosen. Combining the soft thresholding-derived neighbor-joining matrix with a gene-gene correlation matrix,we explored gene connectivity, describing the network’s interconnectedness. Co-expression modules were identified through transformation of the neighbor-joining matrix into a topological overlap matrix, followed by gene hierarchical clustering dendrogram analysis, grouping genes with similar expression patterns and implying potential functional links. To pinpoint clinically relevant modules, we calculated module eigengene sand examined their correlation with clinical features, focusing on modules with positive correlations in both T2DM and AS datasets. Positive correlations between modules and diseases indicated strong associations between module genes and the respective disease.

LASSO (Least Absolute Shrinkage and Selection Operator) is a regression-based methodology that accommodates many covariates in the model. Notably, LASSO possesses a distinct feature of penalizing the absolute value of regression coefficients. Our study employed the ‘glmnet’ package in the R software to conduct LASSO analysis on the candidate hub genes and DEGs. This analysis aided in identifying the final hub genes that exhibited strong associations with the studied conditions.

To assess the potential diagnostic utility of the hub genes, we evaluated performance using the T2DM dataset (GSE9006 and GSE71416) and the AS dataset (GSE40231 and GSE43292). The ROC curves provide a graphical representation of the sensitivity and specificity of the hub genes as diagnostic markers. By analyzing area under curve (AUC), we can measure the accuracy with which central genes classify disease and control groups. The closer the AUC value is to 1, the higher the diagnostic accuracy.

Additionally, to investigate significant differences in gene expression levels between the groups, we employed t-tests. These tests allowed us to compare the expression levels of the hub genes in individuals with T2DM, AS, and controls.

The CIBERSORT algorithm was employed to calculate the proportions of various immune cells in the peripheral blood of patients with T2DM and non-T2DM participants and the arterial wall of patients with AS and non-AS participants. Using the R package “CIBERSORT” and the expression matrices, we determined the proportions of 22 immune cell types in the T2DM and AS disease groups and their respective control groups. To visually represent the proportions of the 22 immune cells in the disease and control groups for T2DM and AS, we generated heatmaps using the “corrplot” package. These heat maps provided a comprehensive view of the quantitative correlations between each disease condition’s different immune cell types. Additionally, we employed the “ggplot2” Rpackage to explore potential associations between immune cell proportions and the expression levels of specific diagnostic markers in the context of T2DM and AS.

The Connectivity Map (cMAP) database provided by the Broad Institute (https://clue.io)consist of drug-like compounds tested for gene expression (16). We uploaded all common DEGs in the GSE9006 and GSE40231 datasets to the cMAP database to screen small molecule candidates. We screened with a score greater than 90, suggesting they potentially have therapeutic effects on T2DM and AS.

All statistical analyses were done using R software (R version 4.2.2). Means and confidence intervals for quartile HbA1c and quartile FBG versus lipid indices were calculated by linear regression. Wilcoxon test was used for statistical analysis between diabetic and non-diabetic groups. We used Spearman’s correlation analysis to investigate the relationship between glycemia and lipid indices by calculating the means and confidence intervals of quartile HbA1c and quartile FBG versus lipid indices in the T2DM using multivariate logistic regression modeling. Statistical significance was determined based on P-values less than 0.05, 0.01, or 0.001. These thresholds helped identify genes with statistically significant differences in expression levels between the disease and control groups. By combining ROC curve analysis and t-tests, we can assess the diagnostic performance of the pivotal genes and determine which genes may be promising biomarkers for differentiating between T2DM, AS, and controls. P-values less than 0.05 (P < 0.05) indicate statistical significance. Significance levels are expressed as follows: *P < 0.05, **P < 0.01, ***P < 0.001. ***P < 0.001.

Table 1 shows the demographic characteristics of the 9,357 participants included in the study, segregated into two groups: 1,829 individuals diagnosed with T2DM and 7,528 controls. The prevalence of T2DM was significantly higher in males (P=0.026), and the group of former smokers, hypertensive, alcohol drinkers, and less educated had a higher risk of developing the disease (P < 0.001). Patients with T2DM had a lower BMI than the healthy population (P < 0.001).

Significant differences in HDL, LDL, TG, total cholesterol, FPG, and HbA1c levels between the two groups of patients were analyzed by comprehensive generalized linear regression ( Figures 2A-F ). FBG, HbA1c, and TG levels were significantly higher in diabetic patients than in non-diabetic patients (P < 0.001) ( Figures 2A, B, D ). In addition, total cholesterol, LDL, and HDL levels were significantly lower in diabetic patients than non-diabetic patients( Figures 2C, E, F ). In Figure 3 , our analysis showed a significant positive correlation (P < 0.01) between FBG, HbA1c, and TG ( Figures 3B, F ). In contrast, there was a significant negative correlation (P < 0.01) between FBG, HbA1c, and HDL levels ( Figures 3D, H ). There was also a low positive correlation (P < 0.01) between HbA1c and total cholesterol, LDL ( Figures 3E, G ). Meanwhile, FBG also showed a lower positive correlation with total cholesterol and LDL as well ( Figures 3A, C ). These findings highlight the unique lipid profile of diabetic patients and emphasize the intricate relationship between glycemic control and lipid metabolism.

Table 2 stratifies FBG into quartiles, revealing a progressive increase in TG with higher FBG levels, indicating a heightened risk (OR 1.4, 95% CI [1.3-1.5]). Concurrently, HDL levels inversely correlate with FBG (P < 0.001). Similarly, Table 3 categorizes HbA1c into intervals, demonstrating significant TG elevation with increased HbA1c (OR 1.3, 95% CI [1.2-1.4]), alongside a decline in HDL with rising HbA1c levels (P < 0.001). Elevated FBG and HbA1c are associated with higher total cholesterol and LDL levels.

Among the 76 common DEGs, 21 were upregulated, an increased expression level, while 21 were down-regulated, indicating a decreased expression level ( Supplementary Figures 1A-C ). GO and KEGG enrichment analyses showed that lipid metabolism-related pathways significantly enriched the differential genes in both diseases. The results of GO analysis showed that the differential genes in both diseases were increased dramatically in response to fatty acid, long-chain fatty acid transport, lipid storage, triglyceride metabolic process, positive regulation of LDL receptor activity, fatty acid transmembrane transport, regulation of LDL particle clearance, LDL particle clearance, negative regulation of LDL particle receptor catabolic process and positive regulation of receptor-mediated endocytosis involved in cholesterol transport. KEGG analysis of differential genes was mainly enriched in the Adipocytokine signaling pathway, Glucagon signaling pathway, Insulin resistance, Insulin signaling pathway, Phospholipase D signaling pathway, Fatty acid degradation, Carbohydrate digestion and absorption, Fatty acid metabolism, regulation of lipolysis in adipocytes and VEGF signaling pathway ( Figure 4D ).

To construct a scale-free topological model, we chose a soft threshold β of 14 for the GSE40231 dataset and a soft threshold β of 9 for the GSE9006 dataset ( Supplementary Figures 2A-D ). These thresholds were instrumental in identifying gene modules that displayed positive associations with AS and T2DM. By applying hierarchical clustering and Spearman correlation analysis, we successfully identified three gene modules exhibiting positive associations with T2DM, encompassing 439 T2DM-related genes ( Figure 4A ). Likewise, we identified three gene modules that demonstrated positive associations with AS, encompassing 3084 AS-related genes ( Figure 4B ). Importantly, we observed 32 overlapping genes within the modules detected by the GSE40231 and GSE9006 datasets( Supplementary Figure 1D ). These shared genes are fascinating as they may be crucial in developing AS and T2DM.

By LASSO regression analysis, 15 genes were selected as candidate genes for each of the two diseases ( Figures 5A-D ). Among them, there were six overlapping genes in both diseases. To further explore the association between these two diseases, we first tested the diagnostic effect of critical genes and whether they were differentially expressed. After removing the mismatched genes, the results showed that two genes (ABCC5 and WDR7) were significantly upregulated in T2DM and AS samples compared with standard samples ( Figures 6A, B ), and the AUC values of these two genes were more outstanding than 0.6, which provided better diagnostic effects ( Figures 6C, D ). In addition, we validated the diagnostic efficiency and expression levels of these two critical genes in validation datasets (GSE71416 and GSE43292) ( Figures 6E-H ). The results suggest that upregulation of these essential genes may lead to T2DM and induce AS.

We performed an in-depth analysis of the proportions of 22 immune cell types using the CIBERSORT algorithm. We included 12 patients with T2DM and 24 control samples, which showed a high percentage of infiltration of CD4 native T cells, resting NK cells, CD8 T cells, and monocytes. Notably, patients with T2DM demonstrated increased proportions of CD4 native T cells, gamma delta T cells, and neutrophils compared to the control group. Conversely, the proportions of CD8 T cells, resting NK cells, monocytes were decreased ( Figures 7A, C ).

Subsequently, we extended our analysis to 40 patients with AS and 40 control samples. The results demonstrated higher infiltration percentages of naive T cells,gamma delta T cells, plasma cells, and M2 macrophages among the 22 immune cell types in patients with AS. Compared to the control group, patients with AS exhibited elevated proportions of naive B cells, plasma cells, follicular helper T cells, Tregs, activated NK cells, M2 macrophages, and resting mast cells and neutrophils. In contrast, the proportions of memory B cells, resting CD4 memory T cells, and gamma delta T cells were decreased ( Figures 7B, D ). In addition, we found that two hub genes, ABCC5 and WDR7, were positively correlated with Neutrophil native cells CD4 naive, and negatively correlated with Macrophages M1 in AS and T2DM ( Figures 7E, F ).

Based on the results obtained from the cMap database, we have identified potential small molecular agents with therapeutic implications based on the upregulated genes. Among them, the top small molecules with the highest absolute enrichment values are presented in Table 4 , including RITA, ON-01910, doxercalciferol, topiramate. These findings provide valuable insights into the potential therapeutic options for AS and T2DM.

Firstly, the sample size and coverage of our study, although substantial, might not adequately represent the broader population affected by T2DM and ASCVD. Therefore, our sample may not capture the full spectrum of demographic and clinical variability, including age, gender, ethnicity, and comorbid conditions, which could significantly influence the disease mechanisms and outcomes. Therefore, future studies should prioritize expanding the sample size and ensuring a more diverse and representative population to enhance the external validity of the findings. the observational nature of our study inherently limits our ability to establish causal relationships between the observed variables. While we have identified correlations that suggest potential mechanisms linking T2DM and ASCVD, these associations do not imply causality. The reliance on observational data, without the ability to control for all potential confounding variables, underscores the need for cautious interpretation of the results. Experimental studies, particularly randomized controlled trials, are essential to confirm the causal links between T2DM and ASCVD and to understand the underlying biological processes.

Lastly, our research did not encompass functional experimental validation of the specific genes implicated in our findings. This limitation highlights a gap in our study, as experimental validation is crucial for verifying the biological relevance and mechanistic role of these genes in the context of T2DM and ASCVD.

Future studies should leverage more comprehensive datasets, employ methodologies that enhance data quality and representation, and incorporate experimental validations. Such endeavors will undoubtedly enrich our understanding and contribute to developing more effective strategies for the prevention, management, and treatment of T2DM and ASCVD.