Integrative analysis of the cancer genome atlas and cancer cell lines encyclopedia large-scale genomic databases: MUC4/MUC16/MUC20 signature is associated with poor survival in human carcinomas

MUC4 z-score expressions were extracted from databases available at cBioPortal for Cancer Genomics [2, 3]. This portal stores expression data and clinical attributes. The z-score for MUC4 mRNA expression is determined for each sample by comparing mRNA expression to the distribution in a reference population harboring typical expression for the gene. The query “MUC4” was realized in CCLE (881 samples, Broad Institute, Novartis Institutes for Biomedical Research) [1] and in all TCGA datasets available (13,489 human samples, TCGA Research Network (http://​cancergenome.​nih.​gov/​)). The mRNA expression from selected data was plotted in relation to the clinical attribute (tumor type and histology) in each sample. MUC4 expression was analyzed in normal tissues by using the Genome Tissue Expression (GTEX) tool [19, 20]. Data were extracted from GTEX portal on 06/29/17 (dbGaP accession phs000424.v6.p1) using the 4585 Entrez gene ID.

We established a list of 187 genes that are correlated with MUC4 expression in CCLE dataset out of 16208 genes analyzed with cBioportal tool on co-expression tab. These genes harbor a correlation with both Pearson’s and Spearman’s higher than 0.3 or lower than − 0.3. Functional annotation and ontology clustering of the complete list of genes were performed using David Functional Annotation Tool (https://​david.​ncifcrf.​gov/​) and Homo sapiens background [21, 22]. Enrichment scores of ontology clusters are provided by the online tool.

Interaction of proteins correlated with MUC4 was determined using String 10 tool (https://​string-db.​org/​) [23]. Edges represent protein–protein associations such as known interactions (from curated databases or experimentally determined), predicted interactions (from gene neighborhood, gene fusion or co-occurrence), text-mining, co-expression or protein homology. The network was divided in 3 clusters based on k-means clustering.

Using (https://​portals.​broadinstitute.​org/​ccle), we extracted mRNA expression of MUC4, methylation score (Reduced Representation Bisulfite Sequencing: RRBS) and copy number variations of the genes of interest. The mRNA expression of MUC4 was plotted in relation to log2 copy number or RRBS score.

Survival analysis was performed using the SurvExpress online tool available in bioinformatica.mty.itesm.mx/SurvExpress (Aguire Gamboa PLos One 2013). We used the optimized algorithm that generates risk group by sorting prognostic index (higher value of MUC4 for higher risk) and split the two cohorts where the p-value is minimal. Hazard ratio [95% confidence interval (CI)] was also evaluated. The tool also provided a box plot of genes expression and the corresponding p value testing the differences.

GSE28735 and GSE16515 pancreatic cancer microarrays were analysed from the NCBI Gene Expression Omnibus (GEO) database (http://​www.​ncbi.​nml.​nih.​gov/​geo/​). GSE28735 is a dataset containing 45 normal pancreas (adjacent non tumoral, ANT) and 45 tumor (T) tissues from pancreatic ductal adenocarcinoma (PDAC) cases. GSE16515 contains 52 samples (16 had both tumor and normal expression data, and 20 only had tumor data. Data were analysed using GEO2R software. The dataset GSE28735 used Affymetrix GeneChip Human Gene 1.0 ST array. The dataset GSE16515 used the Affymetrix Human Genome U133 Plus 2.0 Array. GSE13507 contains 165 bladder cancer and 58 ANT samples. GSE30219 contains 14 normal lung, 85 adenocarcinomas and 61 squamous cancer samples. GSE40967 contains 566 colorectal cancers and 19 normal mucosae. GSE27342 contains 80 tumors and 80 paired ANT tissues. GSE4587 contains 2 normal, 2 melanomas and 2 metastatic melanomas. GSE14407 contains 12 ovarian adenocarcinomas and 12 normal ovary samples.

For MUC4 expression analysis, paired and unpaired t test statistical analyses were performed using the Graphpad Prism 6.0 software (Graphpad softwares Inc., La Jolla, CA, USA). p < 0.05 was considered as statistically significant. Receiving operator characteristic (ROC) curves and areas under ROC (AUROC) were evaluated by comparing tumor and ANT values. cBioportal provided Pearson and Spearman tests were performed to analyze correlation of other genes, RRBS score and log2 copy number with MUC4 expression. DAVID tool provided p value of each ontology enrichment score. SurvExpress tool provided statistical analysis of hazard ratio and overall survival. A Log rank testing evaluated the equality of survival curves between the high and low risk groups.

MUC4 expression was analyzed from databases available at cBioPortal for Cancer Genomics [2, 3]. We queried for MUC4 mRNA expression in the 881 samples from CCLE [1] (Fig. 1). The oncoprint showed that MUC4 was altered in 195 samples out of 881 (22%). 188 were amplification (n = 120) or mRNA upregulation (n = 88) (Additional file 1: Figure S1). Results were sorted depending on the tumor type. We mainly observed an important z-score expression of MUC4 in carcinoma samples (n = 538 samples, p = 0.001) (Fig. 2a). MUC4 Expression scores were subsequently sorted depending on the organ (Fig. 2b). As expected, pancreatic cancer cell lines harbor the highest MUC4 expression (n = 35, z-score = 2.166, p = 0.0006 against theoretical control median = 0). Other cell lines from different tissues (lung NSC, esophagus, bile duct, stomach, upper digestive, colorectal, ovary, and urinary tract) showed statistically significant alteration. We also performed a similar analysis on 13 489 human samples retrieved from TCGA by using the cBioportal platform. An important MUC4 expression z-score was observed in bladder urothelial carcinoma, cervical squamous cell carcinoma/endocervical adenocarcinoma, colorectal carcinoma, esophageal carcinoma, head and neck squamous cell carcinoma, lung adenocarcinoma, lung squamous cell carcinoma, ovarian serous cystadenocarcinoma, pancreatic adenocarcinoma, prostate adenocarcinoma, stomach adenocarcinoma and uterine corpus endometrial carcinoma (Fig. 3). Expression of MUC4 in normal tissues was analyzed using the GTEX project tool, MUC4 was expressed in lung, testis, small intestine, terminal, ileum, prostate, vagina, minor salivary gland and esophagus mucosa and transverse colon (Additional file 2: Figure S2). Altogether, this shows that MUC4 high expression is observed in carcinoma and notably in pancreatic cancer.

Using the co-expression tool on expression data extracted from the 881 samples of CCLE [1], we obtained a list of genes that are co-expressed with MUC4. Genes that harbor a correlation with both Pearson’s and Spearman’s higher than 0.3 or lower than − 0.3 were selected. 187 genes are positively (n = 178) or negatively (n = 9) correlated with MUC4 expression. The better correlated genes were Adhesion G Protein-Coupled Receptor F1 (ADGRF1, Pearson’s correlation = 0.56) and Lipocalin2 (LCN2, Pearson’s correlation = 0.54) (Table 1). We also observed that expression of other membrane-bound mucins MUC16 and MUC20 are positively correlated with MUC4. Correlation between MUC16 and MUC20 was also observed (not shown). Only few genes were negatively correlated such as ZEB1 transcription factor or ST3 Beta-Galactoside Alpha-2,3-Sialyltransferase 2 (ST3GAL2) (Table 2).

Functional Annotation of the complete list of genes and ontology clustering were performed using David Functional Annotation Tool. The gene clustering analysis is presented in Table 3. The complete gene ontologies that are statistically significant are provided in Additional file 3: Table S1. We observed the highest enrichment scores in gene clusters involved in cell adhesion (7.08) and tight junction (5.44) (Table 3). Notably, we observed the correlation of expression of MUC4 with genes encoding integrins (ITGB4 and ITGB6) and cadherin-type proteins such as CDH1, CDH3, Desmocollin 2 (DSC2). A strong enrichment of 91 transmembrane proteins was observed including EPH Receptor A1 (EPHA1), Epithelial cell adhesion molecule (EPCAM), Carcinoembryonic Antigen Related Cell Adhesion Molecule-5 and -6 (CEACAM5 and CEACAM6), C-X-C motif chemokine ligand 16 (CXCL16) and ATPase Secretory Pathway Ca²⁺ Transporting 2 (ATP2C2). As MUC4 is a glycoprotein, it is interesting to also note the correlated expression of enzymes involved in different steps of glycosylation such as sialyltransferases (ST3GAL2, ST6GALNAC1), beta-1,3-N-acetylglucosaminyltransferases (B3GNT5, B3GNT3), fucosyltransferases (FUT3, FUT2), and UDP-GalNAc transferase (GALNT3). MUC4 was also associated with genes associated with cell signaling containing SH2 domain (Cbl proto-oncogene C (CBLC), signal transducing adaptor family member 2 (STAP2), dual adaptor of phosphotyrosine and 3-phosphoinositides 1 (DAPP1), SH2 domain containing 3A (SH2D3A), protein tyrosine kinase 6 (PTK6), growth factor receptor bound protein 7 (GRB7), fyn related Src family tyrosine kinase (FRK), tensin 4 (TNS4)) or SH3 domains (MET transcriptional regulator (MACC1), Rho GTPase activating protein 27 (ARHGAP27), tight junction protein 2 (TJP2), Rho guanine nucleotide exchange factor-5 and -16 (ARHGEF5, ARHGEF16), protein tyrosine kinase 6 (PTK6), EPS8 like 1 (EPS8L1), tight junction protein 3 (TJP3) and FRK). Finally, several genes encoding proteins with a SEA domain (ADGRF1, ST14, MUC16) were correlated with MUC4 expression. Additionally, we analyzed protein–protein interactions of differentially expressed proteins with MUC4 with the String 10 tool. We showed that MUC4 is directly related with CEACAM5, CEACAM6, MUC16, MUC20 and glycosylation enzymes (ST3GAL2, B3GNT3, B3GNT5 and GALNT3) (Additional file 4: Figure S3). Altogether, we have identified genes with expression correlated with MUC4 involved notably in cell adhesion, cell–cell junctions, glycosylation and cell signaling. In order to understand the association between the observed aberrant expression of MUC4 and other molecular events, we explored the correlation between MUC4 expression in CCLE and DNA methylation (RRBS) of the top genes correlated with MUC4. We observed that MUC4 expression is negatively correlated with the methylation score of 16 out of 20 of the top genes (LCN2, MUC20, STEAP4, WFDC2, GJB3, SH2D3A, RNF39, PRSS22, HS3ST1, GPR87, TACST2, FAM83A, LAMC2, B3GNT3, CLDN7) (Fig. 4) suggesting that the association of MUC4 and the correlated genes could be mediated by methylation regulation. Only ADGRF1 RBBS is not correlated with MUC4 mRNA level. MUC16, SCEL and C1ORF116 scores were not available. Additionally we also evaluated the copy number variation association of the top genes with MUC4 expression. We only observed a weak amplification of MUC20 copy number (Pearson’s correlation = 0.13) and a weak deletion of MUC16 copy number (Pearson’s correlation = − 0.14) suggesting that the relationship between MUC4 expression and copy number variation of top genes is unlikely (Additional file 5: Figure S4).

To establish a correlation between MUC4 expression and patient survival, we have compared survival analysis and hazard ratio in population designated as MUC4 high risk and low risk in every organ from TCGA datasets (Table 4). We have used SurvExpress optimized algorithm that generates risk group by sorting prognostic index (higher value of MUC4 for higher risk). The algorithm splits the populations where the p-value testing the difference of MUC4 expression is minimal [4]. Pancreatic cancer presented the most important hazard ratio for MUC4 (HR = 3.94 [CI 1.81–8.61] p = 0.0005756) (Fig. 5a). MUC4 high risk was also significantly associated with survival in bladder cancer (HR = 1.48), colon cancer (HR = 2.1), lung adenocarcinoma (HR = 1.7), lung squamous carcinoma (HR = 1.69), ovarian cancer (HR = 1.33), skin cancer (HR = 1.87) and stomach cancer (HR = 1.58) (Fig. 5a). Acute myeloid leukemia (HR = 1.59) and liver cancer (HR = 1.4) almost reach statistical significance. Other datasets did not show any statistically significant differences.

A significant reduction in patient’s survival was observed in bladder cancer (p = 0.01135), colon cancer (p = 0.00891), lung adenocarcinoma (p = 0.008187), lung squamous carcinoma (p = 0.03586), ovarian cancer (p = 0.0186), pancreatic cancer (p = 0.000219), skin cancer (p = 0.02384) and stomach cancer (p = 0.04751) as illustrated in Kaplan–Meier curves (Fig. 5b). Strikingly, pancreatic median survival was 593 days in MUC4^high cohort (n = 149) whereas the 50% survival was not reached in MUC4^low cohort (n = 27). In lung squamous carcinoma, the median survival of MUC4^high cohort (n = 116) was 1067 days whereas MUC4^low cohort (n = 59) presented a 2170 days median survival. It is interesting to note that the algorithm splits the population in two parts that were characterized as the most different regarding MUC4 expression. Therefore, there are a modest number of MUC4^low PDAC or lung adenocarcinoma patients and a low number of MUC4^high colon or stomach cancer patients. A similar survival analysis was performed on pancreatic cancer by dividing the patient population in two equal parts (88 vs 88), MUC4^high harbored a decreased survival that was close to statistical significance (p = 0.06784) (not shown). Therefore, MUC4 expression is associated with a poorer overall survival in different cancers including pancreatic cancer.

We also compared the survival and hazard ratio, in the same cancers whose survival is associated with MUC4 (bladder cancer, colon cancer, lung adenocarcinoma, lung squamous carcinoma, ovarian cancer, pancreatic cancer, skin cancer and stomach cancer), according to gene signatures corresponding to the five first gene ontology term from Additional file 3: Table S1 (GO 0031424: keratinization, GO 0007155: cell adhesion, GO 0019897: extrinsic component of plasma membrane, GO 0016323: basolateral plasma membrane and GO 0016324: apical plasma membrane) (Fig. 6a, Additional file 6: Table S2). These gene signatures were all significantly associated with survival in the TCGA dataset tested. The “keratinization” (GO 0031424) and “cell adhesion” (GO 0007155) signature are associated with HR comprised between 1.65 and 3.76 and between 2.15 and 3.23, respectively. The GO 0019897 signature is associated with weaker HR (1.55–2.30). “basolateral” (GO 0016323) and “apical plasma membrane” (GO 0016324) signatures harbor more increased HR (2.21–4.5 and 1.77–4.42, respectively) in these datasets.

We performed a similar analysis according to the top genes (ADGRF1, LCN2, MUC20, C1ORF116, SCEL, STEAP4) that harbored Pearson’s correlation with MUC4 superior to 0.5 (Fig. 6b, Additional file 7: Table S3). This signature is associated with survival in all TCGA dataset tested (HR comprised between 1.91 and 8.77). Notably, pancreatic cancer harbored the strongest association with survival according to this signature (HR = 8.77 [CI 2.15–35.83]). Overall, these bigger signatures harbored higher hazard ratio compared to MUC4 alone.

Mucins have been proposed as potential biomarkers for carcinoma. Notably, previous work suggested that combination of mucins expression may be useful for early detection and evaluation of malignancy of pancreatobiliary neoplasms [24]. Moreover, MUC16/CA125 antigen is an already routinely used serum marker for the diagnosis of ovarian cancer [16]. Therefore, we decided to intentionally focus on the two other membrane bound mucins MUC16 and MUC20 that were correlated with expression of MUC4. We analyzed the survival curves of the high risk group (MUC4/MUC16/MUC20^high, n = 159) and low risk group (MUC4/MUC16/MUC20^low, n = 17) from the pancreas TCGA dataset. The MUC4/MUC16/MUC20^high risk group was associated with an increased hazard ratio (HR = 6.5 [2.04–20.78], p = 0.001582) and a shorter overall survival (p = 0.0003088) (Fig. 7a). Median survival was similar as in MUC4^high cohort (593 days). The MUC4/MUC16/MUC20^high group harbored a statistically significant increase of MUC4, MUC16 and MUC20 expression (Fig. 7b). We also analyzed overall survival in every other PDAC database available in Surexpress. We show that MUC4^high group was associated with a statistically significant reduced overall survival and increased hazard ratio in both ICGC and Stratford (GSE21501) cohorts (Fig. 7c). In Zhang cohort (GSE28735), MUC4^high group was associated with a reduced overall survival that was close to statistical significance (p = 0.08971). In other organs, the MUC4/MUC16/MUC20^high group was associated with an increased hazard ratio and reduced overall survival in bladder cancer, colon cancer, lung adenocarcinoma, lung squamous adenocarcinoma, skin cancer, stomach cancer (Additional file 8: Figure S5A). Notably, the MUC4/MUC16/MUC20^high group in colon cancer (HR = 2.26 [1.51–3.4]) showed a median survival of 1741 days whereas the low risk group did not reach the 50% survival. Similarly, the MUC4/MUC16/MUC20^high group in stomach cancer showed a median survival of 762 days whereas the low risk had a median survival of 1811 days. No significant difference was observed for ovarian cancer (p = 0.2081). Moreover, a reduced overall survival was observed in liver cancer (p = 0.04789) and acute myeloid leukemia (AML) (p = 0.02577) (Additional file 8: Figure S5B) in which we did not show any statistical difference when sorting the patients for MUC4 alone. Overall, we observed that MUC4/MUC16/MUC20 signature harbored an increased hazard ratio compared with MUC4 alone for pancreatic cancer and to a lower extent in bladder cancer, colon cancer, lung squamous cancer and stomach cancer.

We analyzed MUC4, MUC16 and MUC20 expression in pancreatic tumor (T) and paired adjacent non tumoral tissues (ANT) from GSE28735 (Fig. 6) and GSE16515 (not shown) datasets [25, 26]. We confirmed MUC4 overexpression in tumor tissues (p < 0.0001). MUC16 and MUC20 mRNA level were also increased (p < 0.0001 and p = 0.0062) in tumor samples (Fig. 8a). As previously observed in CCLE dataset, MUC4 expression was correlated with MUC16 (p = 0.0006) and MUC20 (p = 0.0621) in GSE28735 (Additional file 9: Figure S6). We also analyzed MUC4, MUC16 and MUC20 expression in datasets of other cancers (Additional file 10: Figure S7). MUC4 expression is increased in bladder cancer vs ANT (GSE13507, p < 0.01). MUC20 is increased in lung adenocarcinoma vs normal samples (GSE30219, p < 0.05). MUC4 and MUC20 expression is increased in colorectal cancer vs normal mucosae (GSE40967, p < 0.01). MUC16 and MUC20 relative expression is increased in ovarian adenocarcinoma (GSE14407, p < 0.01 and p < 0.05 respectively). ROC curves of MUC4, MUC16, MUC20 and MUC4 + MUC16 + MUC20 combination were established using GSE28735 dataset. The combination of MUC4 + MUC16 + MUC20 produced a high specificity of 97.78% (88.23–99.94) and a mild sensitivity of 55.56% (40–70.36) (likelihood ratio = 25) (Fig. 8b). Similar results were obtained for GSE16515 with 93.75% specificity and 69.44% sensitivity (LR ± 11.11) (not shown). MUC16 AUROC was similar to that of MUC4 + MUC16 + MUC20 in GSE28735 dataset but harbored a lower specificity/sensitivity in GSE16515.

Altogether, this suggests that MUC4/MUC16/MUC20^high signature would be useful in stratification of patients with worst prognosis in several carcinoma and notably pancreatic, stomach and colon cancers.