3.1 TXNIP expression analysis data
To conduct a comprehensive analysis of human TXNIP (NM_006472.6 for mRNA, NP_006463.3 for protein, data relating with its genome location, conserved functional domain and the phylogenetic tree were obtained(Fig. S1A-S1C). As shown in Fig. S1D, the levels of mRNA expression of TXNIP shows a low tissue specificity with all normalized expression values in detected tissues > 1. Then, a low cell type specificity of TXNIP expression in different blood cells is also observed (Fig. S1E).
To figure out whether TXNIP expression correlates with cancer, we firstly evaluated TXNIP expression in different tumors and adjacent normal tissues using the online Oncomine database. The results showed that TXNIP expression was lower in several cancer groups than in normal tissues, including breast cancer, colorectal cancer (CRC), head and neck cancer, lung cancer, ovarian cancer, sarcoma (SARC), lymphoma, liver cancer or bladder cancer (Fig. 1B, all P < 0.05), and more detailed results were summarized in Table S1. Moreover, we further used TIMER2 and GEPIA2 web servers to evaluate the RNA sequencing data of TXNIP in TCGA and GTEx. Data confirmed that significantly more tumor tissues expressed lower levels of TXNIP mRNA than the corresponding control tissues (Fig. 1C, 1D), of which included BLCA, UCEC, breast invasive carcinoma (BRCA), LUAD, lung squamous cell carcinoma (LUSC), colon adenocarcinoma (COAD), rectum adenocarcinoma (READ), stomach adenocarcinoma (STAD), kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC), prostate adenocarcinoma (PRAD), thyroid carcinoma (THCA) (P < 0.001), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), esophageal carcinoma (ESCA) (P < 0.01), OV, DLBC, ACC, skin cutaneous melanoma (SKCM), uterine carcinosarcoma (UCS), and liver hepatocellular carcinoma (LIHC) (P < 0.05). Results of the TXNIP proteomic expression profile analysis showed absolutely lower expression of TXNIP total protein in the primary tissues of breast cancer, colon cancer, ovarian cancer, LUAD and UCEC (Fig. 1E, all P < 0.001) than in normal tissues. Taken together, the data confirmed that the TXNIP gene was down-regulated in multiple cancers compared to normal samples.
To detect the transcriptional level of TXNIP in CRC, we collected 50 pairs of fresh tumor specimens with matched normal tissues from CRC patients in our hospital. The qPCR analysis (n = 50) revealed that the expression level of TXNIP mRNA in CRC was significantly lower than that in adjacent tissues (Fig. 1F, P < 0.0001). Further, western blotting analysis of 18 pairs of randomly selected CRC samples showed that tumor tissues exhibited markedly lower level of TXNIP than the adjacent tissues (Fig. 1G, P = 0.0133). TXNIP IHC staining of 22 pairs of CRC samples drew consistent conclusion. TXNIP was lowly expressed in 12 out of 22 tumor tissues and highly expressed in 19 out of 22 normal tissues (Fig. 1H). Immunohistochemical score of TXNIP showed that the expression level of TXNIP in tumor tissues was significantly lower than that in paired normal tissues (Fig. 1I, P = 0.0003).
3.2 Survival analysis data
Next, we analyzed the prognostic value of TXNIP expression across cancers in GEPIA2. As shown in Fig. 2A, OS analysis data showed a correlation between low TXNIP expression and poor prognosis for cancers of KIRC (P = 2.0e-04) within the TCGA project. Meanwhile, lowly expressed TXNIP was linked to poor prognosis of DFS for the TCGA cases of CHOL (P = 0.010), KIRC (P = 0.002), SARC (P = 0.019), and uveal melanoma (UVM, P = 0.026) (Fig. 2B). Interestingly, high expression of the TXNIP gene was related to poor OS prognosis for OV and STAD (Fig. 2A, all P < 0.05), suggesting the prognostic value of TXNIP may has tissue or tumor specificity.
Moreover, the TXNIP prognostic value was evaluated by the Kaplan-Meier plotter database using Affymetrix microarrays data (Fig. 2C–O, Fig. S2A–H). And results demonstrated that low levels of TXNIP was closely linked to poor OS, relapse-free survival (RFS), post-progression survival (PPS), and distant metastasis-free survival (DMFS) prognosis for breast cancer (Fig. 2C–F, all P < 0.001); poor OS (P = 0.000), RFS (P = 0.017), progression-free survival (PFS, P = 0.027), and disease specific survival (DSS, P = 0.001) prognosis for liver cancer (Fig. 2G–J); and poor OS (P = 2.7e-10), PPS (P = 0.000), and first progression (FP, P = 0.001) prognosis for lung cancer (Fig. 2K–M). However, we failed to detect a correlation between TXNIP expression and the OS, PFS and PPS prognosis of ovarian cancer (data not shown, all P > 0.05). Based on the datasets of the Kaplan-Meier plotter, we observed that high expression of TXNIP was associated with poor clinical outcomes of OS and PPS for gastric cancer, especially in subgroup analysis of “stage 2 to 4”, “lauren classification/intestinal/diffuse” and “treatment/surgery alone” (Table S2). We also analyzed the correlation of TXNIP expression and the selected clinicopathological factors in liver (Table S3), lung (Table S4), and breast cancer (Table S5), and finally observed distinct conclusions concerning these tumors. These findings suggest that low TXNIP expression implies reduced survival in breast, lung and liver cancer.
In addition to TXNIP microarray analysis in the Kaplan-Meier plotter database, the impact of TXNIP expression to survival rates of patients was evaluated using the PrognoScan(Table S6 and Fig. S2I–T). Notably, decreased expression of TXNIP indicated poorer survival prognosis in seven out of eight cancers, including breast, bladder, skin, lung, brain, eye and soft tissue cancers (Fig. S2I–S, all P < 0.05). Unexpectedly, analysis of two cohorts (GSE14333, GSE17536) showed that high TXNIP expression was linked to poorer DFS prognosis of CRC patients (Table S6 and Fig. S2T, all P < 0.05). Accordingly, it is conceivable that low TXNIP expression is an independent risk factor and leads to a poor prognosis in breast, bladder, and lung cancer patients.
3.3 Genetic alteration analysis data
The genetic alteration status of TXNIP in different tumor samples of the TCGA cohorts was examined by the cBioPortal website. As shown in Fig. 3A, bladder (n = 411), liver (n = 372) and lung (n = 566) cancer contribute to the top 3 tumors with the highest alteration frequencies of TXNIP, with an alteration frequency of 12.17%, 9.95%, and 9.89%, respectively. It is worth noting that the “amplification” type of copy number aberration (CNA) is the primary genetic alteration type in almost all tumors. The three-dimensional structure of TXNIP protein as well as the sites and types of the TXNIP genetic alteration are further presented in Fig. 3B and 3C. The missense mutation of TXNIP appears to be the main type of genetic alteration in PRAD (Fig. 3C). N389del/L386F alteration, detected in ACC, SKCM, UCEC and PRAD (Fig. 3C), is able to induce inframe and missense mutation of the TXNIP gene, leading to the subsequent deletion of N (asparagine) at the 389 site of TXNIP protein, the substitution of L (Leucine) with F (Phenylalanine) at the 386 site of TXNIP protein, respectively.
The tumor mutational burden (TMB) is a quantitative biomarker, which can reflect the number of mutations in tumor cells22. Microsatellite instability (MSI), a mode of genomic instability, is closely related with carcinogenesis, prognosis and response to immunotherapy23,24. To assess the role of TXNIP in tumorigenesis, we analyzed the correlation between TXNIP expression and TMB/MSI across all tumors of TCGA. As shown in Fig. 3D, we observed a negative correlation between TXNIP expression and TMB for BRCA (P = 1.6e-12), LIHC (P = 0.013), LUSC (P = 0.007), PRAD (P = 3.2e-05), BLCA (P = 0.036), DLBC (P = 0.021), KIRC (P = 0.011), LUAD (P = 2e-09), STAD (P = 0.000), and THCA (P = 4.5e-07). TXNIP expression is also negatively correlated with MSI of BRCA (P = 0.044), LIHC (P = 0.021), LUSC (P = 0.000), PRAD (P = 0.002), ESCA (P = 0.019), SKCM (P = 2.8e-06), STAD (P = 2e-05), head and neck aquamous cell carcinoma (HNSC) (P = 2.6e-05), brain lower grade glioma (LGG) (P = 0.039) but is positively correlated with that of COAD (P = 1.3e-08) and READ (P = 0.012) (Fig. 3E). This suggests that TXNIP may be associated with tumorigenesis and treatment.
The biological role of DNA methylation has been extensively studied. As a common epigenetic modification, its regulation is reported to be closely related to tumorigenesis25. Consequently, we next investigated the correlation between TXNIP expression and that of four epigenetic regulators (DNMT2, DNMT1, DNMT3A, and DNMT3B) via the Sangerbox platform. The expression of TXNIP is negatively correlated with that of DNMTs in KIRC, KIRP, PRAD and PCPG (all P < 0.05). The deficiency in mismatch repair (MMR), a common DNA repair pathway, may increase mutations and result in MSI and carcinogenesis26. Furthermore, TXNIP expression and the levels of five MMR genes, especially MLH1, MSH2, MSH6 and PMS2, are positively correlated in KIRC, kidney renal papillary cell carcinoma (KIRP), THCA, HNSC, PRAD, pancreatic adenocarcinoma (PAAD) and pheochromocytoma and paraganglioma (PCPG) (Fig. 3G, all P < 0.01). These results conformably demonstrate the important role of TXNIP as a tumor suppressor in certain types of cancer, suggesting that its deficiency may play a crucial role in oncogenesis.
Ubiquitination and phosphorylation are the enzymatic post-translational protein modification that regulate cellular biological processes in various ways and have been implicated in a number of cancer types14-16. It is of note that dysregulation of specific protein ubiquitination may contribute substantially to cancer development and metastasis16. Here we outlined the possible ubiquitination and phosphorylation sites of TXNIP protein, as shown in Fig. 3H. This observation merits further molecular assays for further exploration, which may explain the mechanism of low expression of TXNIP protein.
3.4 Immune infiltration analysis data
Tumor-infiltrating immune cells, an indication of the host immune reaction to tumor antigens, play a critical role in tumor progression and antitumor activity27-29. As the major TIIC population, tumor infiltrating lymphocytes are independent predictors of sentinel lymph node status and cancer survival28-30. As shown in the heatmap of Figure 4A, TXNIP expression was significantly correlated with TIICs in various types of cancers. Specifically, in the tumors of BRCA-LumA (n = 568), CESC (n = 306), ESCA (n = 185), HNSC (n = 522), HNSC-HPV- (n = 422), HNSC-HPV+ (n = 98), KIRC (n = 533), LIHC (n = 371), SARC (n = 260), SKCM (n = 471), and STAD (n = 415), TXNIP expression is negatively correlated with non-polarized M0 macrophages, but positively correlated with polarized M1 and M2 macrophages, especially the M2. We thus hypothesized that TXNIP may activate the tumor-associated macrophages (TAMs) along with shifting macrophage phenotype to a more anti-inflammatory state. Among these studied cancers, TXNIP expression has significant correlations with infiltrating levels of CD4+ T cells in 22 types of cancer, CD8+ T cells in 19 types of cancer, macrophages in 22 types of cancer, neutrophils in 25 types of cancer, and dendritic cells in 23 types of cancer (Fig. 4B-4D and Fig. S3A-AC, all P<0.05).
We then selected the specific cancers in which TXNIP was correlated with oncologic outcomes and infiltrating immune cells. Using the prognostic results related to TXNIP from the GEPIA2, Kaplan-Meier-plotter and PrognoScan analyses, we eventually selected BRCA, LUAD and KIRC for further research on immune infiltration via TIMER2. The TXNIP expression had significant positive correlations with six types of infiltrating immune cells in BRCA (Fig. 4B, all P<0.01), LUAD (Fig. 4C, all P<0.001), and KIRC (Fig. 4D, all P<0.001). Tumor purity, defined as the proportion of tumor cells in the tumor sample, was reported to influence the genomic sequencing analysis of immune infiltration in clinical tumor samples31. After purity adjustment, we found that the correlation of TAMs genes (CCL2, CCL5 and CD68) obviously changed, especially the values in BRCA (n = 1100), which were the most significant changes (Table S7). These findings strongly demonstrate that TXNIP could recruit immune cells in the TME in BRCA, LUAD and KIRC.
CAFs in the stroma of the TME were reported to participate in modulating the function of various TIICs31. As shown in Fig. 4E, we observed a statistical positive correlation of TXNIP expression and the estimated infiltration value of CAFs for the TCGA tumors of BRCA, BRCA-Basal, COAD, LIHC, LUSC, READ, STAD, and testicular germ cell tumors (TGCT) (all P < 0.05). The scatterplot data of the above tumors are presented in Fig. 4F-4M.
To further assess the stromal content (StromalScore), overall immune infiltration (Est_ImmuneScore), and the combined score (ESTIMATEScore) of the tumor tissues, the ESTIMATE algorithm was employed32,33. As shown in Fig. 5A-5C, the TXNIP expression was significantly positively correlated with the immune and matrix scores in the tumors of LUAD, LUSC, and STAD (all P<0.001). Mounting preclinical and clinical evidence discloses that immune checkpoint blockade therapy is emerging as one of the most promising strategy in oncology18. Next, in order to identify the potential therapeutic targets and further interpret the immune microenvironment, we analyzed the expression of a set of immune checkpoints in 33 types of studied tumors. Many types of tumors, especially CHOL, LUSC, and SKCM (Fig. 5D), highly express multiple immune checkpoint molecules, indicating that combinatorial treatment with different checkpoint inhibitors may be required for optimal tumor control. Increasing evidence suggests that tumor neoantigen plays a pivotal role in antitumor immune response and cancer immunotherapies19. Our data showed that the numbers of neoantigen in tumors like CESC (Fig. 5E), KIRP (Fig. 5G), and glioblastoma multiforme (GBM, Fig. 5F) was significantly positively associated with the expression of TXNIP (all P < 0.05). However, tumors such as STAD (Fig. 5H), LUAD (Fig. 5I), BRCA (Fig. 5J), and PRAD (Fig. 5K) displayed a negative association between the numbers of neoantigen and TXNIP expression (all P < 0.05).
3.5 Enrichment analysis of TXNIP-related partners
To further conduct the pathway enrichment analyses concerning the TXNIP-correlated proteins or genes, a group of public tools were applied, of which includes the STRING tool and the GEPIA2 tool. We first collected 100 binding proteins targeting TXNIP using the STRING tool, which were obtained in the setting of “active interaction sources: Experiments”. And the protein-protein interaction (PPI) network of these molecule is shown in Fig. 6A. Next, the top 100 TXNIP expression-related genes was extracted from the TCGA data using the GEPIA2 tool. As displayed in Fig. 6B, the expression level of TXNIP was significantly correlated with that of CALCOCO1 (calcium binding and coiled-coil domain 1, R = 0.63), KLF9 (Kruppel like factor 9, R = 0.62), TGFBR3 (transforming growth factor beta receptor 3, R = 0.64), TNS2 (tensin 2, R = 0.59), TSC22D3 (TSC22 domain family member 3, R = 0.59) and ZBTB16 (zinc finger and BTB domain-containing protein 16, R = 0.56) genes (all P = 0.000). The positive correlations between TXNIP and one or more of the above six genes was also observed in most cancer types according to the heatmap data (Fig. 6C). A specific molecule, ZBTB16, was obtained from the intersection analysis of the above two groups (Fig. 6D).
The KEGG and GO enrichment analyses were successively performed based on the two datasets. Results of the KEGG enrichment analysis suggested that “ubiquitin mediated proteolysis” ranked first among those enriched pathways, suggesting that it may be related to the role of TXNIP in tumor pathogenesis (Fig. 6E). Most of the aforementioned genes were closely linked to the cellular biological process of PTMs, such as protein dephosphorylation or phosphorylation, polyubiquitination, phosphoric ester hydrolase activity, protein phosphorylated amino acid binding, receptor tyrosine kinase binding, according to our GO enrichment analysis data (Fig. 6F). Meanwhile, pathways including cell-cell signaling by Wnt, stress-activated protein kinase signaling cascade, insulin receptor signaling pathway, and vascular endothelial growth factor receptor signaling pathway were related to those genes (Fig. S4).