TIMER2.0 (Tumor Immune Estimation Resource, version 2, http://timer.cistrome.org/) (Li et al., 2020) and GEPIA2.0 (Gene Expression Profiling Interactive Analysis, version 2, http://gepia2.cancer-pku.cn/#analysis) (Tang et al., 2017) were used to compare the expression levels of DLD between normal and neoplastic tissues at corresponding sites. The expression levels of DLD genes are expressed using a log2 (TPM+1) scale, where TPM represents per million transcripts. TIMER2.0 analysis was based on data from The Cancer Genome Atlas (TCGA). GEPIA2.0 analysis was based on data from the TCGA and a genotype-tissue expression (GTEx) dataset. For GEPIA2.0, the p-value cutoff was 0.05, and the |Log2FC| cutoff was 1. The relationship between tumor pathological stage and DLD expression was analyzed with this bioinformatic tool. Then, the protein expression and DNA methylation of DLD were analyzed in the Clinical Proteomic Tumor Analysis Consortium (CPTAC) dataset through the UALCAN (http://ualcan.path.uab.edu/analysis-prot.html) platform (Chen et al., 2019; Chandrashekar et al., 2022). Information on the number of GEPIA2.0 samples is shown in Supplementary Table S1.

IHC staining images of the DLD protein in seven tumor tissues and corresponding normal tissues were obtained from the HPA (Human Protein Atlas, http://www.proteinatlas.org/) to further confirm the differences in DLD expression at the protein level. In addition, we collected paraffin-embedded tissue sections from 12 patients with glioma from the Department of Pathology, Tianjin Huanhu Hospital. These sections were analyzed with reference to previous studies (Yang et al., 2018; Huang et al., 2019). The antibody used was Anti-Lipoamide Dehydogenase (1:150, Abcem, ab124926, Cambridge, United Kingdom). Subsequently, Fiji software was used to assess the mean optical density values and the percentage of positive area of the images (Schindelin et al., 2012). The clinical information of glioma patients are included in Supplementary Table S9.

GEPIA 2.0 was used to explore the prognostic value of DLD, including the value for predicting overall survival (OS) and disease-free survival (DFS), for all types of tumors. Cox regression analysis was performed to determine whether DLD expression was associated with OS or DFS. Survival analysis was conducted using the Kaplan‒Meier method, and the median value was applied as the cutoff to divide samples into high and low expression groups. Survival curve hypothesis testing was performed by the log-rank test. p-value < 0.05 was considered significant.

The cBioPortal (https://www.cbioportal.org/) tool was used to analyze the mutational status of DLD, including mutation frequency, mutation type, and mutation site information, in all TCGA tumor subtypes. Survival data, including OS, DFS, disease-specific survival (DSS), and progression-free survival (PFS) results, were then compared across all TCGA cancer types. Information regarding DLD mutations is summarized in Supplementary Table S2. Patient characteristics for the DLD mutation analyses are included in Supplementary Table S3.

The TIMER 2.0 tool was used to investigate the relationship between DLD expression and immune cell infiltration. The following algorithms were chosen for this analysis: TIMER, EPIC, QUANTISEQ, MCPCOUNTER and XCELL.

The CancerSEA (http://biocc.hrbmu.edu.cn/CancerSEA/home.jsp) tool was used to explore the correlations between DLD expression and different tumor biological functions at the single-cell level (Yuan et al., 2019). Based on the conditions of p-value < 0.01 and correlation>0.3, we obtained different biological functions with significant results in five tumors. The correlations and corresponding p values of cancer types and biological functions are shown in Supplementary Table S4. The distribution of DLD expression in different cancers is presented by using t-distribution random neighborhood embedding (t-SNE) plots.

The STRING tool (https://string-db.org/) was used for DLD-related protein‒protein interaction analysis (Franceschini et al., 2013). Then, the top 100 DLD-associated genes were obtained and subjected to Pearson correlation analysis via GEPIA2 (Supplementary Table S5). Finally, GO and KEGG enrichment analyses were performed for these 100 genes by using the R package ClusterProfiler (version 3.14.3), and the results obtained are displayed in Supplementary Table S6 and Supplementary Table S7. The screening conditions were FDR<0.1 and p-value < 0.05.

Analysis of the differential expression of DLD across cancers was performed by the Wilcoxon test or Student’s t-test. Correlation analysis between DLD expression and other variables was performed using the Spearman or Pearson test. p< 0.05 was considered significant.

The flow chart for the pancancer analysis of DLD expression is shown in Figure 1. As shown in Figure 2A, the expression level of DLD was downregulated in 8 tumor types including BLCA, BRCA, COAD, KIRC, PCPG, PRAD, READ, and THCA. In contrast, DLD expression was significantly upregulated in 5 tumors including CHOL, KICH, LIHC, LUSC, and STAD. For the results based on the data from the TCGA and GTEx evaluated with GEPIA2.0, we found that the expression level of DLD was upregulated in 4 cancer types including DLBC, GBM, PAAD, and THYM. Correspondingly, the DLD expression level was significantly downregulated in LAML (Figure 2B). Next, analysis of data from the CPTAC database showed that the DLD protein expression level differed in various cancers apart from lung cancer (Figure 2C). In addition, the violin plots in Figure 2D show the relationship between the DLD expression level and pathological stage. The results revealed that the DLD expression level was correlated with pathological stage in KIRC, KIRP, LUAD, READ, THCA, and UCS. No statistically significant results were found in other types of malignancies (Supplementary Figure S1). We also obtained IHC staining results from the HPA database and compared them with the results from the TIMER2.0 database. Similar results were found in both analyses. As displayed in Figures 3A–G, IHC analysis demonstrated moderate or high staining in normal bladder, breast, intestine, kidney, and thyroid tissues, corresponding to weak or moderate staining in the corresponding tumor tissues. In contrast, DLD staining was weak or moderate in normal lung and liver tissues but high in the corresponding tumor tissues.

According to the expression level of DLD, samples from the TCGA and GTEx datasets were divided into high and low expression groups. The results in Figures 4A, B show that elevated levels of DLD expression were associated with poorer OS in patients with BRCA (p = 0.032, HR = 1.4), KICH (p = 0.0095, HR = 9.6), LUAD (p = 0.024, HR = 1.4) or OV (p = 0.029, HR = 1.3). In the other three tumors, COAD (p = 0.019, HR = 0.56), KIRC (p = 2.1e-05, HR = 0.51) and KIRP (p = 0.038, HR = 0.51), elevated expression levels of DLD were associated with better OS. With regard to PFS, elevated levels of DLD expression were associated with longer DFS in both tumor types (KIRC and KIRP). Opposite results were obtained for the two other cancers, LGG and MESO.

The cBioPortal tool was used to explore the mutational status of DLD across cancers. We found that DLD had the highest mutation rate (6.04%) in ESCA. Among the evaluated mutation, the “Amplification” type had the highest rate at 5.49% (Figure 5A). In addition, “Mutation” was the main type of genetic alteration in DLD, mainly located in the “PDB1zy8” domain. A potentially clinically significant missense mutation, an alteration at the V212Sfs*32/Ffs*12 locus, was detected in 5 samples (Figure 5B). Subsequently, to further investigate the relationship between genetic alterations and prognosis, we performed survival analyses including OS, DFS, DSS, and PFS analyses. As shown in Figures 6A, B, patients suffering from KIRC with DLD mutations showed poorer DSS (p = 4.004e-5), OS (p = 9.005e-4), and PFS (p = 5.441e-3). In lung squamous cell carcinoma, patients with DLD genetic alterations showed poorer PFS as well (p = 0.0347).

Epigenetic alterations, such as DNA methylation, could be part of the causal chain in cancer progression. Therefore, we explored the altered methylation levels of DLD in tumor tissues compared to normal tissues in pancancer data using the UALCAN tool. Our findings suggested that DLD methylation levels were significantly different in 12 tumor types (Figures 7A–L). Among these tumor types, BLCA, COAD, LIHC, LUSC, PAAD, READ, and TGCT showed significantly decreased DLD promoter methylation, while BRCA showed a significant increase. In the remaining types of tumors, no significant differences were observed in the methylation level of DLD (Supplementary Figure S2).

The correlation between DLD expression and immune cell infiltration was analyzed by using the TIMER2.0 tool. Our results revealed that the expression of DLD was positively correlated with B-cell infiltration in PAAD, PCPG, and PRAD (Figure 8A); CD4⁺ T-cell infiltration in HNSC-HPV- (Figure 8B); CD8⁺ T-cell infiltration in UVM (Figure 8C); cancer-associated fibroblast infiltration in HNSC (Figure 8D); monocyte infiltration in KIRC (Figure 8E); mast cell infiltration in BRCA-LumA and KIRP (Figure 8F); and neutrophil infiltration in BRCA, KIRC, LIHC, LUAD, PRAD, and STAD (Figure 8H). Moreover, the expression of DLD was negatively correlated with myeloid dendritic cell infiltration in TGCT (Figure 8G) and cancer-associated fibroblast infiltration in KIRC, KIRP, and THCA (Figure 8D). However, as summarized in Supplementary Figures S3, S4, DLD expression was not significantly correlated with other tumor-infiltrating immune cells.

The CancerSEA tool was used to analyze the expression level of DLD at the single-cell level and then further explore the relationships between DLD and different tumor biological functions across cancers. Figure 9A demonstrates the correlations of 14 tumor biological functions with DLD in multiple cancer types. In AML, DLD expression was positively correlated with 7 tumor biological behaviors including apoptosis, differentiation, EMT, inflammation, metastasis, proliferation, and quiescence (Figure 10A). With regard to UM, DLD expression was negatively correlated with 7 tumor biological behaviors including apoptosis, DNA damage, DNA repair, inflammation, invasion, metastasis, and quiescence (Figure 10B). In RCC, 4 biological behaviors including apoptosis, differentiation, hypoxia, and stemness were positively correlated with DLD expression (Figure 10C). For RB, biological functions such as angiogenesis, differentiation, and inflammation were positively correlated with DLD expression (Figure 10D). In LUAD, our findings suggested that DLD expression was positively correlated with 4 tumor biological behaviors including the cell cycle, DNA damage, DNA repair, and proliferation and negatively correlated with angiogenesis (Figure 10E). In addition, the distribution of DLD expression at the single-cell level in the above 5 tumor types is demonstrated in t-SNE plots shown in Figures 11A–E.

To explore the potential molecular mechanisms involving DLD in tumorigenesis and progression, we further obtained the DLD-interacting molecular network by using the STRING tool (Figure 12A). Then, the top 100 genes associated with DLD expression in pancancer data were acquired, and the top 6 were selected to draw the scatter plot in Figure 12B. Finally, GO and KEGG enrichment analyses were performed on these top 100 DLD-related genes, and the top 10 results are shown in bubble plots. As shown in Figure 12C, GO enrichment analysis indicated that DLD-related genes were mainly associated with mitochondria-related cellular components, aerobic respiration and the tricarboxylic acid cycle. The pathways obtained by KEGG enrichment analysis are also displayed in Figure 12D.

We explored the relationships between DLD expression and 60 two-class immune checkpoint pathway marker genes including inhibitory (24) and stimulatory (36) genes (Thorsson et al., 2018), as well as 150 five-class immune pathway marker genes including chemokine (41), receptor (18), MHC (21), immunoinhibitory (24), and immunostimulatory (46) genes (Hu et al., 2021). Our results showed that DLD expression was positively correlated with immune checkpoint- and immunomodulatory activity-related genes in most tumors (Figures 13A-B).

Finally, CellMiner data were used to explore the relationship between the DLD gene and sensitivity to common antitumor drugs. The correlation between DLD expression and the IC50 value of a drug was calculated by Pearson correlation analysis, and we obtained 33 significantly correlated drugs (p-value < 0.05). The results are summarized in Supplementary Table S8, and the top 16 drugs were selected for plotting in scatter plots (Figure 14B). Our results indicated that DLD expression was negatively correlated with dasatinib, erlotinib, BMS-690514, and neratinib sensitivity and positively correlated with Cpd-401, MONENSIN SODIUM, IDEBENONE, ZM-336372, raloxifene, econazole nitrate, tepotinib, BP-1-102, IDOXURIDINE, HYPOTHEMYCIN, tipifarnib, and tic10 sensitivity. Furthermore, PubChem (https://pubchem.ncbi.nlm.nih.gov/) was used to obtain the 3D structures of these 16 drugs (Figure 14A).

By previous differential expression analysis of DLD, we found that the expression of DLD was higher in glioma tissues than in normal tissues. Then, the analysis of IHC results showed that DLD expression was significantly correlated with the grade of glioma (Figure 15A). The mean optical density values and the percentage of positive area responded to the expression of DLD in glioma tissues (Figure 15B). For the percentage of positive area, the results were statistically significant (p-value < 0.05), except for the comparison between grade I and grade III, grade II and grade III (p-value > 0.05). Similarly, for the mean optical density values, the results were statistically significant except for the comparison between grade I and grade II.