Integrated Approach for Biomarker Discovery and Mechanistic Insights into the Co-Pathogenesis of Type 2 Diabetes Mellitus and Non-Hodgkin Lymphoma

Introduction

T2DM is a chronic disorder characterized by insulin resistance and progressive pancreatic β-cell dysfunction, leading to metabolic disturbances and hyperglycemia.¹ It is the most prevalent form of diabetes, accounting for 90–95% of all diabetes cases.² Over the past few decades, the incidence of T2DM has risen at an alarming rate, posing a significant global public health challenge.³ Environmental factors including obesity, sedentary lifestyles, and unhealthy diets, as well as genetic predispositions, contribute to the various pathophysiological disruptions responsible for glucose homeostasis dysfunction in T2DM.⁴ This condition is associated with numerous complications, including cardiovascular disease, nephropathy, neuropathy, and retinopathy, which contribute significantly to global morbidity and mortality.⁵

Emerging evidence highlights a connection between T2DM and an increased risk of various cancers such as liver,⁶ gallbladder,⁷ prostate,⁸ gastric,⁹ lung,¹⁰ and oral cancers.¹¹ Moreover, among hematological malignancies, previous research suggests that T2DM is linked to a heightened risk of NHL, leukemia, and myeloma.^12–14 It is worth mentioning that a growing body of research has identified a potential association between T2DM and NHL, a heterogeneous group of malignant lymphoid tumors.^15–18 NHL is the most prevalent hematological malignancy worldwide, accounting for approximately 3% of all cancer diagnoses and deaths, and its incidence has steadily increased in recent decades.¹⁹ The development of NHL is influenced by complex factors, including genetic susceptibility, immune dysfunction, and viral infections, such as those caused by the Epstein-Barr virus and human immunodeficiency virus.^19–21 However, the precise mechanisms underlying the association between T2DM and NHL remain poorly understood. Further research is crucial to elucidate the underlying mechanisms and identify potential biomarkers that could enable targeted therapeutic strategies for coexisting conditions.

In recent years, advancements in high-throughput technologies, such as microarray analysis, have enabled researchers to profile gene expression patterns comprehensively, thereby offering valuable insights into the molecular mechanisms underlying complex diseases.^22,23 Coupled with the power of machine learning algorithms, these tools allow the identification of critical biomarkers and key genes associated with disease progression, as well as the discovery of potential therapeutic targets.^24,25 Machine learning facilitates the integration and analysis of large, multidimensional datasets and provides a robust framework for distinguishing meaningful biological signals from noise.^26–28 In this study, we aimed to identify the hub genes and elucidate the mechanisms underlying the co-pathogenesis of T2DM and NHL by integrating microarray analysis, machine learning, and experimental validation.

Materials and Methods

Data Collection

This study was conducted in accordance with the Declaration of Helsinki and received approval from the Ethics Committee of Shanghai Yangpu Hospital (LL‐012). Microarray datasets were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). The GSE25724 dataset, which included samples from T2DM patients and non-diabetic controls, was used for T2DM-related analysis. The GSE25638 dataset, which included samples from patients with various types of NHL and normal controls, was used for NHL analysis. Detailed information on these datasets is provided in Supplementary Table S1. To ensure consistency and reliability across datasets, raw data were normalized to address potential batch effects. Peripheral blood samples were obtained from three groups at Shanghai Yangpu Hospital: five patients with T2DM, five patients diagnosed with NHL, and five healthy donors. T2DM samples were selected based on the American Diabetes Association diagnostic criteria. Inclusion criteria required patients to meet at least one of the following: fasting plasma glucose ≥126 mg/dL (7.0 mmol/L), 2-hour plasma glucose ≥200 mg/dL (11.1 mmol/L) during an OGTT, or HbA1c ≥6.5%. NHL samples were included after diagnostic confirmation via morphological examination, immunohistochemistry, flow cytometry, and molecular genetic testing. Healthy donors were required to be free from significant illnesses (eg, chronic diseases, autoimmune disorders, infections, or malignancies). The general exclusion criteria included patients with significant comorbidities (eg, autoimmune diseases, active infections, or other malignancies), individuals with other types of diabetes (eg, type 1, gestational, or secondary diabetes), patients taking medications affecting glucose metabolism or immune function (eg, corticosteroids), individuals under 18 or over 75 years of age, pregnant or lactating women, and cases with poor-quality blood samples (eg, hemolysis or clotting).

Identification of DEGs

Differential expression analysis was conducted on the above-mentioned datasets to identify DEGs using the “LIMMA” (version 3.56.2) package. DEGs were selected based on the criteria of adjusted p-value < 0.05 and |log fold change| > 0.585. The “VennDiagram” (version 1.7.3) package was used to identify genes common to both T2DM and NHL.

Construction of a PPI Network

To analyze the regulatory interactions among the identified DEGs, a PPI network was constructed using the STRING database (http://string-db.org). The resulting network was then imported into Cytoscape software (version 3.8.2) for further analysis. Within Cytoscape, the MCODE algorithm was used to identify key functional modules. The selection criteria for the MCODE analysis were set as follows: degree cutoff = 2, node score cutoff = 0.2, k-core value = 2, and maximum depth = 100. A topological analysis was performed using the CytoHubba plugin to rank the most significant genes in the PPI network. Based on a previous study,²⁹ we identified ten genes with the highest degree values as key genes. To further investigate the relationships between these key genes, a co-expression network was constructed using GeneMANIA (http://www.genemania.org/), a tool that provides insights into the functional associations between gene sets.

GO and KEGG Analyses

GO and KEGG analyses were conducted to explore the underlying mechanisms. The enrichment analysis helps to identify the potential mechanisms through which these genes contribute to disease development, offering theoretical support for further investigation. A significance threshold of p < 0.05 was applied in both analyses. The results were visualized using the “clusterProfiler” (version 4.8.1) and “enrichplot” (version 1.20.0) packages.

Selection of Hub Genes Using Machine Learning Algorithms

Advanced machine learning techniques were used to identify hub genes from the key genes in the PPI network, ensuring a rigorous and reliable selection process. The LASSO regression, recognized for variable selection and regularization to prevent overfitting, uses one standard error criterion to balance model complexity and performance.³⁰ The SVM algorithm is an effective supervised learning method known for its ability to handle high-dimensional data by maximizing the separation between classes.³¹ To enhance predictive accuracy, we applied SVM-RFE, which iteratively refines the feature set by removing less important features. In addition, the RF algorithm, an ensemble learning method, effectively manages unbalanced data and estimates feature importance.³² The final set of hub genes was derived from an intersection analysis of the results of the LASSO logistic regression, SVM-RFE, and RF methods. The implementation and visualization of these machine learning algorithms were carried out using the “glmnet” (version 4.1.7), “e1071” (version 1.7.13), and “randomForest” (version 4.7.1.1) packages, respectively.

GSEA and Correlation Analysis

GSEA was performed to identify the pathways significantly associated with the hub genes. Moreover, ssGSEA was performed to quantify the enrichment levels of 28 immune cells in the T2DM and control groups, as well as in the NHL and control groups, offering a comprehensive overview of immune cell involvement. Spearman correlation analysis was used to investigate the relationship between the hub genes and infiltrating immune cells. The analysis and visualization were conducted using several packages, including “vioplot” (version 0.4.0), “reshape2” (version 1.4.4), “ggplot2” (version 3.4.2), and “org.Hs.eg.db” (version 3.17.0).

qRT-PCR

Total RNA was extracted from the samples using TRIzol reagent (Takara, Japan), according to the manufacturer’s instructions. The isolated RNA was reverse-transcribed into cDNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific, USA). Subsequently, qRT-PCR was conducted in 96-well plates using the SYBR Green PCR Master Mix (KAPA, Japan) and the Applied Biosystems 7900HT Fast Real-Time PCR System (Thermo Fisher Scientific, USA). Gene expression levels were calculated using the 2^−ΔΔCt algorithm, with GAPDH as the internal control. Primer sequences used for qRT-PCR are listed in Supplementary Table S2.

Statistical Analysis

Data analysis and visualization were performed using R software (version 4.3.0) and GraphPad Prism (version 8.0.1). For group comparisons involving normally distributed quantitative variables, the Student’s t-test was applied, while the Wilcoxon test was used for non-normally distributed data. Statistical significance was determined using the following thresholds: *p < 0.05, **p < 0.01, and ***p < 0.001.

Results

Identification of DEGs

The flowchart in Figure 1 illustrates the analyses performed in this study. In the GSE25724 dataset, we identified 1807 DEGs between patients with T2DM and controls, including 754 upregulated and 1053 downregulated genes (Figure 2A and B). Similarly, we identified 2487 DEGs between patients with NHL and controls in the GSE25638 dataset, consisting of 1494 upregulated and 993 downregulated genes (Figure 2C and D). Intersection analysis of DEGs from the T2DM and NHL datasets identified 81 common genes, of which 52 were upregulated and 29 were downregulated (Figure 2E and F). GO analysis revealed that these common DEGs were significantly involved in immune-related biological processes, including response to type II interferon, lymphocyte differentiation, complement activation, mononuclear cell differentiation, and immune response activation (Figure 2G). KEGG analysis further revealed significant enrichment in immune-related pathways, including primary immunodeficiency, complement and coagulation cascades, pertussis, hematopoietic cell lineage, and Staphylococcus aureus infection (Figure 2H).

Figure 1 Flowchart of this study.

Figure 2 Identification of differentially expressed genes (DEGs). (A and B) Heatmap (A) and volcano plot (B) of the DEGs of T2DM. (C and D) Heatmap (C) and volcano plot (D) of the DEGs of NHL. (E) Venn diagram of the common upregulated DEGs. (F) Venn diagram of the common downregulated DEGs. (G) The barplot of the Gene Ontology analysis. (H) The barplot of the Kyoto Encyclopedia of Genes and Genomes analysis.

Construction of the PPI Network

The PPI network was constructed to analyze the shared DEGs between T2DM and NHL (Figure 3A). Using the MCODE plug-in in Cytoscape, we identified two interconnected gene modules comprising 14 common DEGs (Figure 3B and C). GO analysis indicated that these genes were significantly enriched in immune-related biological processes, including complement activation, immune response activation, and type I interferon-mediated signaling pathway (Figure 3D). KEGG analysis further showed significant involvement of immune-related pathways, such as complement and coagulation cascades, Staphylococcus aureus infection, and primary immunodeficiency (Figure 3E).

Figure 3 Construction of the protein-protein interaction (PPI) network and module analysis. (A) The PPI network of the common differentially expressed genes between type 2 diabetes mellitus and non-Hodgkin lymphoma, with downregulated genes denoted by green dots and upregulated genes indicated by red dots. (B and C) Two interconnected gene modules. (D) The barplot of the Gene Ontology analysis. (E) The barplot of the Kyoto Encyclopedia of Genes and Genomes analysis.

Subsequently, we identified the top ten genes with the highest degree values as key genes using topological analysis: ZAP70, IL7R, GZMM, CD8A, GBP2, HSPG2, LTBP2, IFITM1, SERPING1, and SAMHD1 (Figure 4A). To gain deeper functional insights, we used the GeneMANIA database to construct a co-expression network and analyze the interactions among these genes (Figure 4B). Functional analysis showed that these genes were significantly involved in immune-related processes, including lymphocyte differentiation, T cell differentiation, positive regulation of leukocyte activation, positive regulation of cell activation, response to type I interferon, cellular response to type I interferon, and receptor signaling pathway via STAT.

Figure 4 The top 10 genes in the protein-protein interaction network. (A) The ten genes with the highest degree values. Genes are represented using a color gradient, with darker hues indicating higher degree values. Bright red corresponds to the highest degree, while yellow represents the lowest. (B) The co-expression network of key genes derived from the GeneMANIA database.

Selection of Hub Genes Using Machine Learning Algorithms

Multiple machine learning algorithms were used to identify hub genes from the set of ten key genes. Using LASSO regression, four genes were identified based on the variables (Figure 5A and B; Supplementary Table S3). The SVM-RFE method identified six genes with an accuracy of 0.98 and an error of 0.02 (Figure 5C and D; Supplementary Table S4). Moreover, the RF algorithm was used to analyze the relationship between the error rate and the number of classification trees, ultimately identifying seven genes of relative importance (Figure 5E and F; Supplementary Table S5). By cross-referencing the results of all three algorithms, we identified three overlapping hub genes: GZMM, HSPG2, and SERPING1 (Figure 5G).

Figure 5 Selection of hub genes using machine learning algorithms. (A and B) Regression coefficients (A) and cross-validation curve (B) derived from the least absolute shrinkage and selection operator regression algorithm. (C and D) Cross-validation accuracy (C) and error (D) in the support vector machine-recursive feature elimination algorithm. (E) Relationship between error rate and quantity of trees in the random forest algorithm. (F) Importance of feature genes. (G) Venn diagram of the hub genes from the intersection of results from the above-mentioned algorithms.

GSEA and Correlation Analysis

To investigate the underlying mechanisms, GSEA was conducted to identify pathways significantly associated with hub genes. GZMM was predominantly enriched in complement and coagulation cascades, natural killer cell-mediated cytotoxicity, and T-cell receptor signaling pathway (Figure 6A). HSPG2 was enriched in complement and coagulation cascades, ECM-receptor interaction, and cancer pathways (Figure 6B). SERPING was primarily enriched in complement and coagulation cascades, cytokine-cytokine receptor interaction, and ECM-receptor interaction (Figure 6C). Collectively, these findings indicated that all hub genes were enriched in immune-related pathways. Further analysis using ssGSEA revealed significant differences in the enrichment levels of 19 of the 28 immune cells between T2DM patients and controls (Figure 6D). In patients with NHL, all 28 immune cells showed significant differences compared to the controls (Figure 6E). These findings suggested a strong association between immune cell infiltration and both diseases. Furthermore, correlation analysis demonstrated significant associations between the hub genes and immune cells (Figure 6F).

Figure 6 Gene set enrichment analysis (GSEA) and correlation analysis. (A-C) GSEA of the significantly enriched pathways of GZMM (A), HSPG2 (B), and SERPING1 (C). (D) Violin plot of immune cells between T2DM and control samples. (E) Violin plot of immune cells between NHL and control samples. (F) Heatmap of correlations between hub genes and immune cells. Statistical significance was determined using the following thresholds: *p < 0.05, **p < 0.01, and ***p < 0.001.

qRT‐PCR

To validate these findings, qRT-PCR was performed on clinical samples from patients with T2DM, patients with NHL, and controls. As anticipated, T2DM patients showed significantly higher expression levels of GZMM, HSPG2, and SERPING1 than the controls (Figure 7A–C). Similarly, NHL patients exhibited a consistent pattern, with all three hub genes showing significant upregulation compared to controls (Figure 7D–F).

Figure 7 qRT-PCR. (A-C) Expression levels of GZMM (A), HSPG2 (B), and SERPING1 (C) between T2DM and control samples. (D-F) Expression of GZMM (D), HSPG2 (E), and SERPING1 (F) between NHL and control samples. Statistical significance was determined using the following threshold: *p < 0.05.

Discussion

Previous studies have indicated that T2DM may increase the risk of developing NHL, although the mechanisms underlying this association remain unclear. In this study, functional analyses suggested an important role of immune-related pathways in the co-pathogenesis of T2DM and NHL. Identifying biomarkers is crucial to elucidate the molecular connections between these diseases. By integrating microarray analysis and machine learning, we identified three hub genes involved in the co-pathogenesis of T2DM and NHL. These genes demonstrated significant correlations with immune cells, further supporting the potential role of immune dysregulation in the shared pathogenesis. Experimental validation using clinical samples was performed to ensure the reliability of the hub genes in clinical practice. These findings provide valuable insights into the shared pathogenic mechanisms of T2DM and NHL, emphasizing the genetic changes that inform molecular pathways and could guide future research and therapeutic strategies.

Functional analyses highlighted the role of immune-related pathways in the co-pathogenesis of T2DM and NHL. Immune cell analysis further demonstrated a significant association between both diseases and immune cell infiltration, highlighting immune dysfunction as a common feature. Patients with T2DM demonstrate increased susceptibility to infections, resulting in higher morbidity and mortality rates than non-diabetic individuals.³³ This increased susceptibility is largely due to abnormalities in both the innate and adaptive immune responses that interact with each other during the progression of T2DM.^34–36 Specific changes, such as altered T cell and macrophage proliferation as well as impaired NK cell and B cell function, reflect an overall dysfunction in immune regulation in patients with T2DM.³⁷ These immune alterations suggest that patients with T2DM may be immunodeficient, increasing their vulnerability to various diseases, including cancer. On the other hand, NHL is associated with dysregulation of the immune system. Individuals with compromised immune systems, such as those with HIV/AIDS or those who have undergone organ transplantation, and patients with autoimmune disorders involving chronic immune activation and inflammation, are known to have an elevated risk of NHL.^38–40 Disruptions in immune function promote the neoplastic transformation of blood cells into lymphoid malignancies through mechanisms, such as chronic immune stimulation, persistent inflammation, defective immune surveillance, and impaired anticancer immunity.^40,41 Therefore, it is plausible that chronic immune dysfunction in T2DM may create an environment of sustained inflammation and impaired immune surveillance, facilitating the development of lymphoid malignancies such as NHL.

Ten key genes were identified in the PPI network of the DEGs shared between T2DM and NHL: ZAP70, IL7R, GZMM, CD8A, GBP2, HSPG2, LTBP2, IFITM1, SERPING1, and SAMHD1. ZAP70 is a cytoplasmic tyrosine kinase that plays an essential role in T-cell receptor signaling.^42,43 Previous research has identified ZAP70 as a promising biomarker for follicular lymphoma and implicated it in the development and progression of this disease through immune-related pathways.⁴⁴ IL7R, the receptor for interleukin-7, is critical for B cell development and T cell maturation.^45,46 GZMM, a serine protease expressed in cytotoxic lymphocytes, contributes to tumor cell destruction, inhibits cytomegalovirus replication, and plays a role in inflammation.⁴⁷ CD8A encodes the alpha chain of the CD8 glycoprotein, which is predominantly expressed in cytotoxic T cells and is vital for antigen recognition and immune responses.^48,49 GBP2, part of the GTPase family, is essential in innate immunity against bacterial, viral, and protozoan pathogens and is significantly upregulated by interferon-γ.^50,51 HSPG2 encodes perlecan, a protein that plays a key role in ECM and immune signaling.^52,53 LTBP2, a member of the ECM glycoprotein family, regulates the activity and function of transforming growth factor-beta, which has immune-modulating properties.^54,55 IFITM1 is an interferon-induced transmembrane protein that is involved in antiviral immunity, immune response regulation, and cellular membrane function.^56,57 SERPING1, a plasma protein, regulates immune responses and blood clotting by inhibiting key enzymes in the complement, coagulation, and fibrinolytic systems.⁵⁸ SAMHD1, a triphosphohydrolase, restricts viral replication by degrading intracellular deoxynucleoside triphosphate in nondividing cells.^59–61 All these genes share a connection with immune system regulation, highlighting the immune dysregulation underlying the co-pathogenesis of T2DM and NHL.

To further explore the biomarker genes involved in the co-pathogenesis of T2DM and NHL, we applied machine learning algorithms that were known for their ability to identify hidden patterns and enhance predictive model accuracy.^26,62,63 This approach led to the identification of three hub genes as key contributors to shared pathogenesis: GZMM, HSPG2, and SERPING1. Validation with clinical samples confirmed the reliability of these genes in clinical practice. GSEA revealed that all three hub genes were enriched in the immune-related pathways. Correlation analysis also revealed significant associations between these hub genes and immune cells, further supporting the hypothesis that immune dysregulation played a central role in the co-pathogenesis of T2DM and NHL. By identifying biomarker genes and exploring the involved mechanisms connecting these two diseases, this study provides a foundation for future research and potential therapeutic development. However, further research is required to elucidate the specific mechanisms by which these genes mediate the complex interplay between T2DM and NHL.

In previous studies, hub genes associated with T2DM and NHL were explored separately.^64–68 In contrast, our study focused on identifying shared hub genes and the mechanisms underlying these two diseases. Functional analyses highlighted the critical role of immune-related pathways in the co-pathogenesis of T2DM and NHL. Additionally, using a combination of microarray analysis and machine learning, we identified three hub genes that showed strong correlations with immune cells, further emphasizing the role of immune dysregulation in shared pathogenesis. Furthermore, hub genes were successfully validated in clinical samples. Our study introduced a novel integrative approach combining microarray analysis, machine learning, and experimental validation to provide a comprehensive understanding of the mechanisms underlying the association between T2DM and NHL. However, it is important to acknowledge the limitations of this study. First, the analyzed datasets from the GEO database primarily focused on B-cell NHL, because data on T-cell NHL were less comprehensive. Although the immunological similarities between the two subtypes suggest potential parallels, future studies should include T-cell NHL to validate the generalizability of our findings. Second, the relatively small sample size of the clinical specimens used for experimental validation may limit the robustness of our conclusions. Larger sample sizes are necessary in future studies to strengthen our results. Finally, although potential hub genes were identified, the exact mechanisms in mediating the interaction between T2DM and NHL remained unclear. Further experimental studies are required to fully elucidate these mechanisms. Addressing these limitations will be the focus of future studies to achieve a more comprehensive understanding of the shared pathophysiology between T2DM and NHL.

Conclusion

Our study suggested that immune-related pathways played a significant role in the shared pathogenesis of T2DM and NHL. By integrating microarray analysis, machine learning, and experimental validation, we identified and validated three key hub genes that were critical contributors to co-pathogenesis. These findings enhance our understanding of the complex interplay between T2DM and NHL and may pave the way for future therapeutic advancements.