Introduction
Cancer remains a major global health challenge,1 and the identification of robust molecular biomarkers is essential for improving risk stratification, prognostic assessment, and individualized treatment. Although many molecular markers have been proposed, their applicability across tumor types is often limited by tumor heterogeneity, database-specific effects, and insufficient biological validation. Therefore, integrative analyses that combine transcriptomic, protein-level, epigenetic, genomic, immune, and pharmacological information may help prioritize candidate biomarkers for subsequent experimental and clinical validation.
The transmembrane emp24 domain-containing protein (TMED)/p24 family participates in vesicular transport between the Golgi apparatus and the endoplasmic reticulum (ER) and contributes to secretory pathway homeostasis.2,3 TMED3, a member of the p24 protein family, has been implicated in unconventional protein secretion and cancer-related signaling. Previous studies have suggested that TMED3 may function either as an oncogenic factor or as a tumor suppressor depending on tumor context.4–6 More recent mechanistic studies further indicate that TMED3 can promote malignant phenotypes in ovarian cancer, prostate cancer, and glioma through pathways involving SMAD2, FOXO1a/FOXO3a phosphorylation, and ZBTB7A signaling.7–9 These findings provide a biological rationale for evaluating whether TMED3 has broader prognostic relevance across cancer types.
Therefore, the aim of the present study was to evaluate the expression pattern, prognostic significance, promoter methylation, genetic alterations, copy number variations (CNVs), protein–protein interactions (PPI), CD8+ T-cell infiltration, and Comparative Toxicogenomics Database (CTD)-derived chemical-gene interactions of TMED3 across six cancer types, including bladder cancer (BLCA), head and neck squamous cell carcinoma (HNSC), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), liver hepatocellular carcinoma (LIHC), and lung adenocarcinoma (LUAD). Because this was an exploratory computational study, all associations were interpreted as candidates requiring independent validation.
Materials and methods
All databases were accessed between January 1, 2022 and June 1, 2026.
UALCAN database analysis
The UALCAN database (http://ualcan.path.uab.edu ) was used to compare TMED3 transcript expression levels across multiple human cancers and their corresponding normal tissues to evaluate differential expression.10,11
Kaplan-Meier Plotter database analysis
In this study, the Kaplan-Meier Plotter database (http://kmplot.com/analysis ) was employed to evaluate the association between TMED3 expression levels and overall survival (OS) as well as recurrence-free survival (RFS) in patients with six selected cancer types.12
The optimal cutoff value was determined using the “auto select best cutoff” function to maximize survival differences between groups, and hazard ratios (HRs), 95% confidence intervals (CIs), and log-rank P-values were recorded.
GENT2 database analysis
The GENT2 database (http://gent2.appex.kr/ ) integrates public gene expression datasets to systematically compare gene expression profiles between normal and tumor tissues.13
Human Protein Atlas (HPA) database analysis
HPA (https://www.proteinatlas.org/ ) is a public database integrating antibody-based immunohistochemistry, transcriptomic, and proteomic data to reveal protein expression patterns in normal and cancer tissues.14,15 The HPA database was used to evaluate TMED3 protein expression levels in normal tissues and different cancers. Based on immunohistochemistry staining intensity and the proportion of positive cells, TMED3 protein expression was semi-quantitatively classified into four levels: not detected, low, medium, and high.
MEXPRESS database analysis
The MEXPRESS database (https://www.mexpress.be ) was used to evaluate the correlation between TMED3 transcript expression and promoter methylation levels in 450 BLCA, 612 HNSC, 980 KIRC, 380 KIRP, 468 LIHC, and 861 LUAD tumor samples.16
cBioPortal database analysis
CNVs and other genetic alterations of TMED3 across the selected cancer subtypes were analyzed using the cBioPortal for Cancer Genomics database.17,18 Alteration types, including missense mutations, amplification, and deep deletion, were summarized descriptively.
PPI network construction and pathway analysis
TMED3-interacting proteins were screened using STRING (Homo sapiens, confidence ≥ 0.400), with the PPI network visualized via Cytoscape (https://string-db.org/ ).19,20 GO and KEGG enrichment analyses were performed using DAVID (Homo sapiens, adjusted P < 0.05) (https://davidbioinformatics.nih.gov/ ).21
Analysis of TMED3 expression and CD8+ T-cell immune infiltration levels
Immune cell infiltration levels were analyzed using the TIMER2.0 database (http://timer.cistrome.org/ ). The abundance of immune cells was estimated using the CIBERSORT-ABS algorithm, and Spearman correlation analysis was performed to evaluate the relationship between TMED3 expression and CD8+ T-cell infiltration levels across multiple cancer types.22
Chemical-gene interaction network analysis
The CTD (http://ctdbase.org/ ) links toxicological information on chemicals, genes, phenotypes, diseases, and exposures.23 A TMED3-related chemical-gene interaction network was constructed using CTD and Cytoscape software. Because CTD-derived chemical-gene relationships are based on curated associations from existing studies, the analysis was considered exploratory and was not interpreted as direct evidence of drug efficacy in TMED3-high tumors.
Statistical methods
Differential expression analysis was performed using the UALCAN (TPM values) and GENT2 (normalized expression profiles) databases. Student’s t-test was used to compare TMED3 expression levels between tumor and normal tissues when provided by the source database. Correlation analysis was conducted using MEXPRESS to evaluate the association between TMED3 mRNA expression and promoter methylation and using TIMER to assess the association between TMED3 expression and CD8+ T-cell infiltration. Survival analysis was performed using the Kaplan-Meier Plotter database to assess OS and RFS, with intergroup comparisons conducted using the log-rank test. Genetic alteration analysis was performed using cBioPortal to describe the frequency and type of TMED3 mutations and CNVs. Functional enrichment analysis was conducted using DAVID. The analyses relied on database-reported nominal P-values. Raw result matrices required to calculate adjusted q-values across all tested comparisons were not consistently available from the public database outputs; therefore, adjusted q-values were not calculated or reported. Accordingly, P < 0.05 was treated as a nominal significance threshold, and findings involving multiple comparisons should be interpreted as exploratory rather than confirmatory.
Results
TMED3 transcriptional expression in human cancers and normal tissues
In this study, transcriptional expression levels of TMED3 across 24 major human cancer subtypes were analyzed using the UALCAN database based on The Cancer Genome Atlas dataset. The results revealed that TMED3 mRNA expression levels were significantly higher in BLCA, KIRP, LIHC, LUAD, and KIRC compared with their corresponding adjacent normal tissues. Conversely, TMED3 expression was significantly downregulated in HNSC relative to normal tissues (P < 0.05) (Fig. 1).
Prognostic potential of TMED3 in multiple cancers
To further evaluate the prognostic value of TMED3 expression, associations between TMED3 expression levels and OS as well as RFS were analyzed in six cancer types (Fig. 2). In BLCA, neither OS (HR = 1.22, 95% CI: 0.88–1.68, P > 0.05) nor RFS (HR = 1.67, 95% CI: 0.78–3.55, P > 0.05) showed a statistically significant association. In HNSC, OS was significantly associated with TMED3 expression (HR = 1.52, 95% CI: 1.16–1.99, P < 0.05), whereas RFS was not (HR = 1.99, 95% CI: 0.76–5.24, P > 0.05). In KIRC, OS showed a significant association (HR = 1.77, 95% CI: 1.32–2.39, P < 0.05), while RFS did not (HR = 5.22, 95% CI: 0.69–39.75, P > 0.05). In KIRP, OS was significantly correlated with TMED3 expression (HR = 2.46, 95% CI: 1.26–4.79, P < 0.05), but RFS was not (HR = 1.95, 95% CI: 0.91–4.17, P > 0.05). In LIHC, OS showed a significant association (HR = 1.49, 95% CI: 1.05–2.11, P < 0.05), whereas RFS did not (HR = 0.89, 95% CI: 0.62–1.30, P > 0.05). In LUAD, OS was significantly associated with TMED3 expression (HR = 1.56, 95% CI: 1.09–2.21, P < 0.05), while RFS did not reach statistical significance (HR = 1.50, 95% CI: 0.98–2.31, P > 0.05).
Overall, elevated TMED3 expression was associated with a trend toward shorter OS across the six cancer types analyzed. This association reached nominal statistical significance in HNSC, KIRC, KIRP, LIHC, and LUAD (P < 0.05), but not in BLCA. In contrast, RFS did not reach nominal statistical significance in any of the six cancer types; therefore, RFS-related findings were interpreted cautiously and were not described as statistically significant.
Correlation of TMED3 expression with clinicopathological features in six cancer types
This study analyzed TMED3 transcriptional expression in BLCA, HNSC, KIRC, KIRP, LIHC, and LUAD and examined associations with clinicopathological features, including tumor stage, race, sex, and age subgroup. In BLCA, TMED3 was significantly upregulated in stages II–IV, in Caucasian and Asian patients, in males and females, and in the 41–60 and 61–80 year age groups compared with normal controls (P < 0.01), whereas stage I, African American, and 21–40 year subgroups did not reach statistical significance, likely in part due to small sample sizes. In HNSC, TMED3 was significantly downregulated across most clinicopathological subgroups (P < 0.05). In KIRC, subgroup-level results differed from the overall UALCAN tumor–normal comparison, suggesting that KIRC findings should be interpreted cautiously and verified in independent cohorts. In KIRP, LIHC, and LUAD, TMED3 was generally upregulated in most evaluated subgroups (P < 0.05), although several subgroups with limited sample sizes did not reach statistical significance.
In summary, TMED3 dysregulation was observed across multiple clinicopathological subgroups; however, the direction and robustness of this dysregulation varied by cancer type and subgroup. These findings support further investigation of TMED3 while emphasizing the need for independent validation.
Validation of TMED3 transcriptional levels using independent cohorts
TMED3 transcriptional expression in BLCA, HNSC, KIRC, KIRP, LIHC, and LUAD was further evaluated using the GENT2 database. Compared with normal tissues, TMED3 was significantly upregulated in BLCA, KIRC, KIRP, LIHC, and LUAD (P < 0.05), whereas it was downregulated in HNSC (P < 0.05) (Fig. 3). Although KIRC showed increased TMED3 expression in both the UALCAN and GENT2 analyses, subgroup-level results were not fully consistent; therefore, these findings should be interpreted cautiously and validated in independent cohorts.
Analysis of TMED3 protein expression levels
TMED3 protein expression in cancer and normal tissues of BLCA, HNSC, KIRC, KIRP, LIHC, and LUAD was analyzed using the HPA database. The results showed that, compared with corresponding normal tissues, TMED3 protein expression appeared higher in BLCA, HNSC, and LUAD tumor tissues (Fig. 4). In contrast, clear protein-level upregulation of TMED3 was not observed in KIRC, LIHC, or KIRP tumor tissues.
Analysis of TMED3 promoter methylation
Promoter methylation plays a critical role in regulating gene expression and may contribute to cancer-related transcriptional dysregulation. MEXPRESS analysis was used to explore whether TMED3 expression was associated with promoter methylation levels. The analysis suggested an inverse relationship between TMED3 promoter methylation and TMED3 expression in BLCA, HNSC, KIRC, KIRP, LIHC, and LUAD based on database-reported nominal P-values. Because methylation–expression analyses involve multiple CpG sites and multiple cancer types, these findings should be regarded as exploratory and should be validated using independent methylation datasets.
Analysis of TMED3 genetic alterations using cBioPortal
Analysis based on the cBioPortal database revealed that, among the six cancer cohorts examined, LUAD exhibited the highest frequency of TMED3 alterations (1%), characterized primarily by gene amplification and missense mutations (Fig. 5). HNSC followed with an alteration frequency of 0.8%, displaying a heterogeneous pattern comprising missense mutations, amplifications, and deep deletions. In contrast, BLCA and KIRP showed lower alteration frequencies of 0.5% and 0.4%, respectively, while no genetic alterations in TMED3 were detected in KIRC or LIHC (0%). Collectively, these distribution patterns indicate that the overall frequency of TMED3 genetic alterations across multiple tumor types is relatively low, suggesting that CNVs are unlikely to be the primary drivers of TMED3 expression changes in these malignancies.
PPI network construction and pathway enrichment analysis
Enriched genes interacting with TMED3 were identified using STRING and Cytoscape resources. Functional interaction network analysis revealed that TMED3 was closely associated with 24 distinct genes (Fig. 6a). Pathways involving TMED3-enriched genes were analyzed using the DAVID tool (Fig. 6b).
Functional enrichment analysis of TMED3-related genes suggested that TMED3 is connected with proteins involved in ER–Golgi trafficking, p24 family transport functions, cytoskeletal organization, and ARF-family-related vesicle transport. In particular, TMED2 and TMED4 were enriched in p24 family protein-related categories, whereas ARF-family members were linked to vesicular transport and Golgi-related regulatory pathways. These results are consistent with the known role of TMED family proteins in secretory pathway regulation and provide a functional context for TMED3-associated tumor biology (Fig. 6).
Association between TMED3 expression and CD8+ T-cell infiltration
To investigate whether TMED3 may be associated with the tumor immune microenvironment, the relationship between TMED3 expression and CD8+ T-cell infiltration was analyzed in BLCA, HNSC, KIRC, KIRP, LIHC, and LUAD using the TIMER database. High TMED3 expression was significantly negatively correlated with CD8+ T-cell infiltration in BLCA, HNSC, and LUAD. In contrast, correlations in KIRC and KIRP were not statistically significant, whereas LIHC showed a significant positive correlation. These results suggest that the relationship between TMED3 expression and CD8+ T-cell infiltration is cancer-type specific and cannot be interpreted as uniformly negative across cancers (Fig. 7). Because the analysis is correlational, it cannot establish that TMED3 directly regulates CD8+ T-cell recruitment or function.
Chemical-gene interaction network analysis of TMED3
To explore potential chemical-gene associations involving TMED3, a CTD-based chemical-gene interaction network was constructed and visualized using Cytoscape. The results indicated that azacitidine and doxorubicin were associated with increased TMED3 expression, whereas MK-2206 was associated with decreased TMED3 expression (Fig. 8). These relationships should be regarded as database-derived hypotheses and should not be interpreted as evidence that these agents are effective treatments for TMED3-driven tumors.
Discussion
The functions of the TMED gene family and their mechanisms in regulating tumor development and progression remain incompletely understood. Beyond the TMED3-related mechanistic studies summarized in the Introduction, TMED3 has also been implicated in breast cancer,24 endometrial cancer,25 osteosarcoma,26 and chordoma.27 Together with the established ER–Golgi transport role of TMED/p24-family proteins, these findings support the need for cancer-type-specific evaluation of TMED3 rather than assuming a uniform cross-cancer function. In this study, TMED3 expression was systematically assessed using public cross-cancer datasets, and the findings suggest that TMED3 dysregulation may have prognostic relevance in several cancers. Importantly, the observed expression patterns were not fully uniform across analyses, particularly in KIRC subgroup-level results, indicating that tumor heterogeneity, dataset composition, and tumor-context-specific effects should be considered.28
TMED3 protein expression was also evaluated using HPA immunohistochemistry data. Higher TMED3 protein expression was observed in BLCA, HNSC, and LUAD tumor tissues compared with corresponding normal tissues, whereas clear protein-level upregulation was not observed in KIRC, LIHC, or KIRP. This partial concordance between transcript and protein results suggests that post-transcriptional regulation and tumor-level RNA–protein discordance may influence TMED3 assessment, in addition to antibody-based detection differences, sample heterogeneity, and database-specific limitations.29,30 In addition, promoter methylation and CNVs were examined because epigenetic regulation and genomic alterations can affect gene expression in cancer.31–33 The results suggest that promoter hypomethylation may be associated with increased TMED3 expression in some cancer types, whereas CNV frequencies were generally low in the six cancers analyzed (Fig. 5). Therefore, methylation may be a more plausible regulatory mechanism than CNVs for TMED3 expression in these datasets, although this conclusion requires independent methylation and expression validation.
PPI and pathway analyses indicated that TMED3-related genes are mainly involved in secretory pathway and vesicle transport processes, which is consistent with the canonical biological role of the transmembrane emp24 domain-containing/p24 family described above. CD8+ T-cells are key mediators of antitumor immunity, but their infiltration and function can be altered during tumor progression and immune exhaustion.34 In this study, TMED3 expression showed significant negative correlations with CD8+ T-cell infiltration in BLCA, HNSC, and LUAD, no significant correlations in KIRC and KIRP, and a positive correlation in LIHC, suggesting a cancer-type-specific immune association. These findings suggest that the immune association of TMED3 may differ by cancer type and should not be interpreted as uniformly reflecting reduced antitumor immune infiltration.35 However, the directionality and mechanism of this association cannot be determined from TIMER-based correlation analysis alone.
Chemical-gene interaction network analysis suggested that azacitidine and doxorubicin may be associated with increased TMED3 expression, whereas MK-2206 may be associated with decreased TMED3 expression. These CTD-derived findings are useful for generating hypotheses about potential regulatory relationships but should not be considered evidence that these drugs are clinically useful for TMED3-high tumors. Experimental studies are required to determine whether modulation of TMED3 alters drug response and whether TMED3 has any actionable therapeutic relevance.
Limitations
Several limitations should be acknowledged. First, this study is based entirely on public database mining and bioinformatics analysis; no in vitro, in vivo, or clinical sample-based experiments were performed to validate the biological function of TMED3. This is important because omics-based biomarker discovery generally requires rigorous independent validation before clinical translation.36 Second, external validation was limited: GENT2 was used for transcript-level validation, but protein expression, methylation, immune infiltration, and survival findings were not comprehensively validated in independent cohorts or platforms. Third, several analyses involved multiple comparisons across cancer types, molecular features, and clinical outcomes. Because only database-reported nominal P-values were consistently available and raw result matrices required for unified multiple-testing correction were not available from all platforms, adjusted q-values could not be calculated or reported. Therefore, the possibility of false-positive associations cannot be excluded, and the statistical findings should be interpreted as hypothesis-generating. Fourth, CTD-derived chemical-gene interactions are correlational or literature-curated associations and cannot establish therapeutic efficacy. Finally, discrepancies between databases, such as the KIRC expression pattern, highlight the need for standardized datasets, sensitivity analyses, and experimental confirmation before TMED3 can be regarded as a clinically validated biomarker. In addition, post hoc sensitivity analyses could not be performed because the original raw datasets were not retrievable from all source platforms, further increasing uncertainty in the reported associations.
Conclusions
TMED3 is a cancer-type-specific prognostic candidate associated with shorter OS in HNSC, KIRC, KIRP, LIHC, and LUAD. Protein-level evidence shows higher expression in BLCA, HNSC, and LUAD, while promoter methylation and CD8+ T-cell correlations provide testable regulatory and tumor microenvironment-related hypotheses. Because the analyses relied on database-reported nominal P-values without unified multiple-testing correction and lacked experimental validation, these findings establish TMED3 as a priority candidate for independent mechanistic and clinical validation rather than a clinically validated biomarker or therapeutic target.
Declarations
Acknowledgement
The authors thank UALCAN, Kaplan-Meier Plotter, GENT2, the Human Protein Atlas, MEXPRESS, cBioPortal, STRING, TIMER2.0, DAVID, and the Comparative Toxicogenomics Database for providing free access to their platforms.
Ethical statement
This study was based exclusively on publicly available, de-identified data obtained from UALCAN, Kaplan-Meier Plotter, GENT2, the Human Protein Atlas, MEXPRESS, cBioPortal, STRING, TIMER2.0, DAVID, and CTD. No new human specimens, animal experiments, clinical interventions, or identifiable private information were collected or analyzed by the authors. Therefore, institutional ethics approval and informed consent were not required.
Data sharing statement
The data supporting the findings of this study are publicly available and de-identified. They were obtained from the UALCAN database (http://ualcan.path.uab.edu), Kaplan-Meier Plotter (http://kmplot.com/analysis), GENT2 (http://gent2.appex.kr/), Human Protein Atlas (https://www.proteinatlas.org/), MEXPRESS (https://www.mexpress.be), cBioPortal (http://www.cbioportal.org), STRING (https://string-db.org/), TIMER2.0 (http://timer.cistrome.org/), DAVID (https://davidbioinformatics.nih.gov/), and the Comparative Toxicogenomics Database (CTD; http://ctdbase.org/).
Funding
This project was supported by the Shanghai Shenkang Hospital Development Center Medical-Enterprise Collaborative Innovation Project (SHDC2022CRT009).
Conflict of interest
The authors declare that they have no competing interests.
Authors’ contributions
Study conception and design (ZK, LC), literature collection, manuscript organization (PJ, YY), data extraction, bioinformatics analysis, organization of results (JL), validation of the analytical workflow, interpretation of findings (XG, YL), drafting of the manuscript (ZK), study supervision, and critical revision of the manuscript for important intellectual content (LC). All authors reviewed and approved the final manuscript for submission.