The RNA sequencing data (level 3) and clinical information of the NCI's Genomic Data Commons The Cancer Genome Atlas (TCGA) colon cancer cohort (TCGA cohort; 286 primary tumor, 26 paired and 15 unpaired normal colon specimens) were downloaded from the University of California Santa Cruz Xena browser (https://xenabrowser.net/datapages/). The levels of mRNA were shown as log₂(x+1) values (x: transformed RNA-Seq by Expectation Maximization normalized counts). Patients were defined as CMTM6^High or CMTM6^Low based on CMTM6 mRNA levels (the top 30% and the bottom 30%, respectively). The DEGs between the two groups were analyzed by the R package edger,27 and significant differences were defined with fold change >1.5 and p value <0.05. Genes with upregulated expression in the CMTM6^High group were subjected to gene ontology (GO) enrichment analysis via the online tool DAVID V.6.8 (https://david.ncifcrf.gov/).28 False discovery rates <0.05 were considered significant. In addition, the infiltration of immune cells in the tumor microenvironment was analyzed using a single-sample GSEA (ssGSEA),29 in which immune cell types were identified by specific gene markers,30 and the enrichment score in the ssGSEA represented the relative abundance of each type of immune cell.

A total of 233 patients were involved in this study and signed informed consent forms. The patients were pathologically and clinically diagnosed with CRC in Sun Yat-sen University Cancer Center (SYSUCC) from May 2007 to December 2015. The median age at surgery was 60 years (ranging from 28 to 86 years). All patients underwent surgery immediately without any neoadjuvant therapy, and 156 received adjuvant chemotherapy, while the other 77 did not (their clinicopathological parameters are shown in online supplemental table S1). Formalin-fixed, paraffin-embedded sections of tumor tissues were obtained from the pathology department of SYSUCC and re-evaluated by two pathologists according to the tumor-node-metastasis staging system of the eighth edition of the American Joint Committee on Cancer. The follow-up data of the patients were collected from the follow-up department of SYSUCC, and patients received regular follow-up (every 3 months for the first 2 years after surgery, every 6 months in the following 2 years and every year thereafter). Progression-free survival (PFS) was defined as the time span from the date of surgery to the date of cancer progression or death, and overall survival (OS) was defined as the time span from the date of surgery to the date of death. The follow-up was censored on December 31, 2019, and patients who did not experience progression or death during the follow-up period were censored at the last follow-up date. At the end of follow-up, 20.6% (48/233) of patients had progression, and 18.9% (44/233) of patients died from CRC. The median PFS and OS were 47.1 and 49.53 months, respectively (both ranging from 1.67 to 88.6 months).

Immunohistochemistry (IHC) staining to detect CMTM6, PD-L1, cluster of differentiation (CD) 8, CD4 and FoxP3 was performed by a professional pathologist according to previous reports.3 18 31 Briefly, after deparaffinization, rehydration, antigen retrieval, endogenous peroxidase inactivation and non-specific binding blockade, 4 µM-thick sections were incubated with primary antibodies (anti-CMTM6: Sigma–Aldrich, HPA026980; anti-PD-L1: Cell Signaling Technology (CST), #13684; anti-CD8: CST, #85336; anti-CD4: Abcam, ab252199 and anti-FoxP3: Abcam, ab20034) at 4°C overnight. Then, the slides were incubated with a corresponding secondary antibody for 30 min at 37°C and visualized with a DAKO EnVision Detection System (Dako). Finally, the slides were counterstained with hematoxylin (CST, #14166), dehydrated and cover-slipped.

All immunostained sections were evaluated independently and blindly by two professional pathologists from SYSUCC, yielding reasonably consistent results. Five fields (more than 500 cells) in each specimen were selected randomly for analysis. The CMTM6 expression levels were scored as an IHC Score, which was calculated as the proportion score based on stained cell percentage (0, 0%; 1, 1%–25%; 2, 26%–50%; 3, 51%–75% and 4, 76%–100%) multiplied by the staining intensity score (0, negative; 1, weak; 2, moderate and 3, intense) (online supplemental figure S1A), and the median IHC Score was chosen as the cut-off value for defining high and low expression of CMTM6. The expression of PD-L1 was evaluated separately in cancer cells (CC), tumor stroma (TS) cells or whole tumor tissue (whole) on stained sections as previously described,24 and was defined as ‘positive’ if PD-L1 staining was present on ≥1% of cells (online supplemental figure S1B,C).18 32 The infiltration of CD4⁺, CD8⁺, and FoxP3⁺ T cells was measured as the percentage of cells staining positive in the invasive margin (IM) or TS (online supplemental figure S1B,D).18

Co-IP and mass spectrometry were performed as previously described.33 Briefly, 293 T cells (American Type Culture Collection (ATCC), Manassas, Virginia, USA) were transfected with pEnter-CMTM6-Flag, a plasmid expressing CMTM6 with a Flag-tag at the C-terminus (Vigene Biosciences, Jinan, China), using Lipofectamine 2000 (Invitrogen, Carlsbad, California, USA). Forty-eight hours later, cells were collected and lysed with lysis buffer (CST, Danvers, Massachusetts, USA) containing protease inhibitor cocktail (Roche, Basel, Switzerland) and phosphatase inhibitors (KeyGen Biotech, Nanjing, China). The cell lysates were centrifuged to obtain supernatants, which were incubated with M2 anti-Flag agarose beads (A2220, Sigma-Aldrich). After washing, the IP products were examined by mass spectrometry. Proteins detected in the IP product of CMTM6-Flag-expressing cells, but not in that of control cells, were subjected to GO enrichment analysis via DAVID V.6.8.28

Data were analyzed using SPSS V.23.0 (IBM Corporation) or GraphPad Prism 8.0 software (GraphPad Software, San Diego, California, USA).The correlation analysis was performed using the χ² test, while the survival analysis was performed by the Kaplan-Meier method to calculate the survival probability in terms of PFS and OS, and the log-rank test was used to examine intergroup differences. Univariate and multivariate analyses were executed via a Cox proportional hazard model. A p value <0.05 was considered to indicate statistical significance.

We first analyzed CMTM6 expression in 286 CRC and 41 normal colon specimens from the TCGA cohort and found that CMTM6 mRNA levels were dramatically higher in CRC tissues than in normal colorectal tissues (p<0.001; figure 1A). Next, we compared CMTM6 expression in CRC specimens with different consensus molecular subtypes (CMSs) or clinical stages. The results showed that CMTM6 expression in CMS1 tissues, characterized by immune infiltration and immune cell activation, was significantly higher than that in other tissues (p=0.002), especially that in CMS2 and CMS4 tissues (p<0.001 and p=0.024, respectively; figure 1B). In addition, CMTM6 mRNA levels were higher in early-stage CRC (stage I/II) than in advanced-stage CRC (stage III/IV; p=0.024; figure 1C).

Next, we examined CMTM6 expression in CRC tissues of 233 patients from SYSUCC by IHC assay. The results showed that CMTM6 protein levels tended to be higher in early-stage CRC (stage I/II) than in advanced-stage CRC (stage III/IV; p=0.052; figure 1D), similar to the findings in the TCGA cohort above. We classified these patients into two groups using the median CMTM6 IHC score as the cut-off value (<4, CMTM6^Low; ≥4, CMTM6^High) and found that CMTM6 expression levels were significantly correlated with tumor anatomic site (p=0.002) and pT classification (p=0.031; table 1). Kaplan-Meier analysis revealed that patients with CRC with CMTM6^High status had a significantly longer PFS (p=0.004) and OS (p=0.002) than those with CMTM6^Low status (figure 2E).

To explore the biological roles of CMTM6 in CRC, DEGs were analyzed between the CMTM6^High and CMTM6^Low groups from the TCGA cohort, and the top 100 DEGs (50 upregulated and 50 downregulated genes) in the CMTM6^High group compared with the CMTM6^Low group were used to generate a heatmap (online supplemental figure S2A). GO enrichment analysis revealed that the top 50 upregulated genes in the CMTM6^High group were mainly enriched in immune or immune-related pathways, such as the adaptive immune response, immune response and T cell receptor signaling pathways (online supplemental figure S2B).

We knocked down the expression of the CMTM6 gene in RKO colon cancer cells (ATCC) and performed RNA sequencing analysis but failed to identify DEGs clustered in immune-associated signaling pathways (data not shown). These findings were different from those in CRC tissues from the TCGA cohort. We presumed that the difference may be due to the different cell composition of two subjects: RKO cells contained tumor cells only, while CRC tissues contained tumor cells, fibroblasts, vessel cells and various immune cells; thus, the enrichment of immune-related pathways in DEGs of CRC tissues was likely contributed by those tumor infiltrating immune cells. Considering that CMTM6 has been reported to maintain PD-L1 stability via protein-protein interactions, we performed a co-IP-mass spectrometry assay and found that potential proteins interacting with CMTM6, which were pulled down by Flag (CMTM6) in CMTM6-Flag-expressing 293 T cells but not in blank vector-transfected 293 T cells, were related to immune-related pathways (online supplemental figure S2C). Based on these findings, how CMTM6 plays an immunomodulatory role, in addition to its interaction with PD-L1, should be investigated and clarified in future research.

To better understand the roles of CMTM6 in the immune response, ssGSEA was conducted to evaluate the immune cell composition of CRC samples from the TCGA cohort. As shown in figure 2A, there were many more tumor-infiltrating T cells, such as activated CD4⁺ and CD8⁺ T cells, effector memory CD4⁺ and CD8⁺ T cells, central memory CD4⁺ T cells, gamma delta T cells, and Th1, Th2 and Th17 cells, but not regulatory T cells, in tumor tissues of the CMTM6^High group than in those of the CMTM6^Low group, indicating that CRC tissues with high levels of CMTM6 had a tumor microenvironment with an activated adaptive immune phenotype. However, there were no significant differences in natural killer (NK) cells, including CD56^bright and CD56^dim NK cells, eosinophils, macrophages, mast cells, neutrophils and plasmacytoid dendritic cells (DCs), between the CMTM6^High and CMTM6^Low groups; as an exception, there were differences in activated and immature DCs between the two groups. These data reflect that CMTM6 may play an important role in regulating the adaptive antitumor immune response in CRC.

Furthermore, we investigated the infiltration of CD4⁺, CD8⁺ and regulatory (FoxP3⁺) T cells in CRC tissues from the SYSUCC cohort by IHC assay. The results showed that there were significantly more CD4⁺ and CD8⁺ T cells in both the TS and IM of CMTM6^High tissues than in those of CMTM6^Low tissues (figure 2B,C), but there were no differences in FoxP3⁺ regulatory T (Treg) cells between the two groups (figure 2D), similar to the findings in the TCGA cohort.

Because CMTM6 has been reported as a PD-L1 regulator, we next investigated the relationship between PD-L1 expression and CMTM6 levels in CRC. CMTM6 expression was much higher (59.7% of CRC tissues with IHC score >4) than PD-L1 expression in CRC. There was a low frequency of PD-L1 expression in whole CRC tissues (PD-L1(whole)), and only 30% of tissues showed PD-L1 staining on ≥1% of cells (PD-L1⁺) (online supplemental table S2). CMTM6 expression levels (IHC scores) were significantly higher in PD-L1(whole)⁺ than in PD-L1(whole)⁻ CRC tissues (Student’s t-test, p<0.05; online supplemental figure S3); the χ² test showed that CMTM6 expression status (high or low) was not associated with the status of PD-L1(whole) (positive or negative) (p=0.146; table 1). Furthermore, the positive rates of PD-L1 in CC (PD-L1(CC)) and in TS (PD-L1(TS)) were 23.6% (55/233) and 39.5% (92/233), respectively (online supplemental table S2). CMTM6 expression was significantly higher in PD-L1(TS)⁺ than in PD-L1(TS)⁻ CRC tissues (Student’s t-test, p<0.05), but there was no difference between PD-L1(CC)⁺ and PD-L1(CC)⁻ tissues (online supplemental figure S3). Similar to PD-L1(whole) status, the PD-L1(TS) and PD-L1(CC) statuses were not associated with CMTM6 expression in CRC tissues (χ² test, p=0.103 and 0.532, respectively; table 1). These findings suggested a weak association between PD-L1 and CMTM6 protein levels in CRC. In addition, CMTM6 was identified to play a role in immune regulation beyond acting as a PD-L1 regulator, with PD-L1 expression being regulated by multiple factors other than CMTM6.

It has been reported that PD-L1 influences the tumor microenvironment by inhibiting the function of T lymphocytes in multiple cancers18 34; thus, we investigated the association of PD-L1 expression with the infiltration of T cells and the prognosis of patients with CRC. The results showed that the patients with CRC with PD-L1(TS)⁺ had longer PFS and OS than those with PD-L1(TS)⁻ (p=0.013 and 0.036, respectively), but there were no significant differences in survival (PFS and OS) between patients with PD-L1(whole)⁺ and those with PD-L1(whole)⁻ (p=0.140 and 0.243, respectively) or between patients with PD-L1(CC)⁺ and those with PD-L1(CC)⁻ (p=0.933 and 0.678, respectively; figure 3A–C). Further analysis of T cell infiltration revealed that there were no differences in the infiltration of CD4⁺, CD8⁺ or FoxP3⁺ T cells in either TS or IM of CRC tissues between two groups classified based on PD-L1 expression status (figure 3D–F, online supplemental figure S4), except that there were more CD4⁺ T cells in the TS of PD-L1(TS)⁺ CRC tissues than in that of PD-L1(TS)⁻ CRC tissues (figure 3D).

Univariate Cox regression analyses of patients with CRC in the SYSUCC cohort revealed that in addition to higher T, N and M stages, clinical stage and CA19-9 level (≥35 kU/L), lower CMTM6 (CMTM6^Low) (PFS, HR=2.267, 95% CI 1.276 to 4.029, p=0.005; OS, HR=2.483, 95% CI 1.353 to 4.559, p=0.003) and PD-L1(TS) (PFS, HR=2.302, 95% CI 1.174 to 4.514, p=0.015; OS, HR=2.046, 95% CI 1.034 to 4.049, p=0.004) were associated with a higher risk of disease progression and death (table 2). Multivariate Cox regression analysis showed that CMTM6 but not PD-L1(TS) was an independent predictor for PFS and OS in CRC (PFS: HR=1.833, 95% CI 1.005 to 3.343, p=0.048; OS: HR=1.953, 95% CI 1.040 to 3.669, p=0.037) (table 2).

To better analyze the relationship between the expression of CMTM6 and/or PD-L1 and the prognosis and immune microenvironment of patients with CRC, we considered PD-L1 expression in combination with CMTM6 expression to generate a potential tool for predicting survival and tested it in survival analysis. We found that the expression status of PD-L1(TS), but not that of PD-L1(whole) or PD-L1(CC), could effectively differentiate the survival of patients with CMTM6^High (PFS, p=0.033; OS, p=0.091), and the survival curves of patients with CMTM6^High/PD-L1(TS)⁻ almost coincided with those of patients with CMTM6^Low/PD-L1(TS)⁺ (PFS, p=0.757; OS, p=0.559; figure 4A, online supplemental figure S5). Thus, we classified these patients with CRC into three groups: group 1 (CMTM6^High/PD-L1(TS)⁺; low risk), group 2 (CMTM6^High/PD-L1(TS)⁻ or CMTM6^Low/PD-L1(TS)⁺; moderate risk) and group 3 (CMTM6^Low/PD-L1(TS)⁻; high risk) according to survival curves of PFS (p=0.001) and OS (p=0.002; figure 4B). The proportions of patients in groups 1, 2 and 3 were 26.2% (61/233), 46.8% (109/233) and 27.0% (63/233), respectively.

We next examined the associations of CMTM6/PD-L1(TS) coexpression with T lymphocyte infiltration. The data showed that the level of infiltrating CD4⁺ T lymphocytes in both the IM and TS of CRC tissues was highest in group 1 compared with groups 2 and 3 (IM, p<0.05 and<0.01; TS, p<0.01 and<0.001; figure 4C (upper)). In terms of CD8⁺ T lymphocytes, there were no significant differences between group 1 and group 2, but they were significantly higher in group 1 than in group 3 (IM: p<0.05; TS: p<0.001; figure 4C (middle)). However, there were no differences in the infiltration of FoxP3⁺ Treg cells among the three groups (figure 4C (lower)).

Considering that traditional chemotherapy (such as oxaliplatin) has been reported to induce immunogenic cell death, thereby improving the tumor immune microenvironment,35 we performed subgroup survival analysis stratified by whether patients received adjuvant chemotherapy or not. Kaplan-Meier analyses revealed that patients with CRC with CMTM6^High had a significantly longer PFS (p<0.001) and OS (p<0.001) than those with CMTM6^Low in the subgroup received adjuvant chemotherapy (figure 5A), but there were no significant differences in PFS (p=0.246) or OS (p=0.270) between patients with CMTM6^High and CMTM6^Low in the subgroup not received adjuvant chemotherapy (figure 5B). Similar phenomena were observed for PD-L1(TS) expression status in differentiating the survival of patients with CRC: p=0.019 for PFS and p=0.054 for OS in the adjuvant chemotherapy subgroup; p=0.442 for PFS and p=0.455 for OS in the non-adjuvant chemotherapy subgroup (figure 5C,D). Notably, PD-L1(whole), a factor without obvious prognostic value in the whole CRC cohort, displayed a significant predictive value for the survival of patients with CRC who received adjuvant chemotherapy (PFS, p=0.032; OS, p=0.070; online supplemental figure S6A), although PD-L1(CC) still showed no prognostic value for patients in the adjuvant chemotherapy subgroup (online supplemental figure S6B).

Because PD-L1(TS) showed more significant prognostic value than PD-L1(whole), we considered the expression of PD-L1(TS) and CMTM6 jointly and analyzed its association with the survival of patients with CRC who received adjuvant chemotherapy. The results were similar to those from the analysis of the whole cohort (figure 4A): in the adjuvant chemotherapy subgroup, patients with CMTM6^High/PD-L1(TS)⁺ had the best prognosis, those with CMTM6^Low/PD-L1(TS)⁻ had the worst prognosis, and those with CMTM6^High/PD-L1(TS)⁻ and those with CMTM6^Low/PD-L1(TS)⁺ showed similar survival curves, especially for PFS (online supplemental figure S7). Thus, patients in adjuvant chemotherapy subgroup were divided into three groups, as was done for the whole cohort, for subgroup survival analysis. The results showed that this new classification (the coexpression status of CMTM6 and PD-L1(TS)) could divide patients with CRC who received adjuvant chemotherapy into three groups with low, moderate or high risk (PFS, p<0.001; OS, p<0.001; figure 5E) but was unable to predict prognosis for patients who did not receive adjuvant chemotherapy (PFS, p=0.419; OS, p<0.442; figure 5F). These findings suggest that CMTM6/PD-L1 expression may have better prognostic value for patients with CRC receiving adjuvant chemotherapy than for those not receiving adjuvant chemotherapy. Similar to the analysis of T lymphocyte infiltration in the whole cohort (shown in figure 4C), there were higher infiltration of CD4⁺ or CD8⁺ T cells in the tumor tissues of group 1 than in those of group 2 and/or group 3 in adjuvant chemotherapy subgroup (online supplemental figure S8).