Comprehensive proteogenomic characterization of UC samples
We collected formalin-fixed paraffin-embedded (FFPE) tumor samples and paired morphologically normal human urothelium (MNU) samples from 116 urothelial carcinoma of bladder (UC) patients, comprising 45 non-muscle-invasive bladder cancers (NMIBCs) and 71 muscle-invasive bladder cancers (MIBCs). The patients were recruited from Zhongshan Hospital (Shanghai, China), and none had a history of presurgical treatment. The samples were re-reviewed by three expert genitourinary pathologists, who classified these as pure urothelial and histological variants. Clinical data, including the gender, age at diagnosis, tumor grade, survival, etc., were summarized in Additional file 13: Table S1. Additionally, comparing to previous published UC dataset, Beijing Cohort [19] and TCGA Cohort [12] revealed the similarity of patients’ basic features including age, gender, history of treatment among the three cohorts, yet some distinctive features were also observed (Table 1). To be more specifical, at demographic level, all the patients in our cohort and Beijing cohort were from Asian, whereas only 7% patients in TCGA cohort were from Asian. Histologically, more early-stage patients (TCGA cohort: Ta-T1, n = 1, 0.1%; Beijing cohort: Ta-T1, n = 47, 39%; our cohort: Ta-T1, n = 45, 38%, chi-square test, p < 2.2E−16) were included in our cohort and Beijing cohort (Table 1). A schematic of the experimental design is shown in Fig. 1A. The samples were characterized using clinical data and four molecular profiling platforms (Fig. 1B; Additional file 1: Fig. S1A).
Table 1 The baseline characteristics of patients among different cohorts Using whole-exome sequencing data with a mean depth of 159X in 113 tumors and 128X in 43 paired MNUs (Additional file 1: Fig. S1B), we totally detected 106,622 mutations including 97,950 single-nucleotide variants (SNVs) and 8672 small insertion–deletions (indels) (Additional file 13: Table S1) (Methods). Twenty-four genes showed statistically significant levels of recurrent somatic mutation by analysis using MutSig (q < 0.1; Methods) (Additional file 13: Table S1), which included six well-known bladder cancer-related genes: TP53 44%, KMT2D 25%, PIK3CA 19%, ARID1A 17%, RB1 5%, and ELF3 4%. These six SMGs and 20 additional TCGA-reported potential UC driver mutations are shown in Fig. 1C. Correlation analysis was performed using mutational frequencies from other cohorts including Beijing cohort [19] and TCGA cohort [12]. As a result, the mutation frequencies of the hotspot genes detected in our cohort were more closely correlated with Beijing cohort than TCGA cohort (Fig. 1E). Moreover, there was no marked difference between the TMBs in our cohort and the Beijing cohort (Fig. 1F) [19]. This might be due to that both Beijing cohort and our cohort include only Asian patients. Additionally, among the hotspot mutations, the mutational frequency of RB1 was significantly higher in TCGA cohort than that of in Beijing cohort and our cohort (Fig. 1D; Fisher’s exact test, p < 0.05). This finding was in consistent with previous papers that RB1 displayed higher mutational rates in MIBC [12], since 99% patients in TCGA patients were MIBC, whereas 38% patients in our cohort were NMIBC, respectively.
We identified four mutational signatures by Sigminer (Fig. 1G; Additional file 13: Table S1; Methods). Signatures 1–4 corresponded to the known COSMIC (Catalog of Somatic Mutations in Cancer) signatures: SBS5, SBS1, SBS30, and SBS13. We compared tumor mutation burden (TMBs) of the four mutational signatures; the results showed that SBS30, which represents DNA-based excision repair, had the highest level of TMB (Fig. 1H). In addition, SBS13, representing the activity of APOBEC, was associated with the worst prognosis (Fig. 1I). For comparison, the same enrichment analysis was performed in the TCGA cohort [12], in which four mutational profiles (SBS5, SBS2, SBS13, and SBS10b) were identified (Additional file 1: Fig. S1C). The mutational signatures best matching to those in TCGA cohort were (1) SBS5; unknown (clocklike signature), and (2) SBS13; APOBEC cytidine deaminase (Additional file 1: Fig. S1D). In addition, SBS5 and SBS13 were also identified in Beijing cohort [19].
Label-free quantification measurement of all patient samples (157 tumors and paired 76 MNUs) resulted in a total of 16,440 protein groups with a 1% false discovery rate (FDR) at the protein and peptide levels (Additional file 1: Fig. S1E–G; Additional file 13: Table S1; Methods) [20, 21], and an average of 6990 protein groups per tumor sample and 5945 protein groups per MNU sample (Additional file 1: Fig. S1H). The tumor proteome and the MNU proteome exhibited a unimodal distribution, and the correlations among 157 tumor samples ranged between 0.52 and 0.87 (Additional file 1: Fig. S1I, J). Phosphoproteomic analysis was conducted on 111 tumor samples and 46 MNU samples which revealed 5789 phosphoproteins and 33,233 phosphosites in tumors, as opposed to 3246 phosphoproteins and 11,668 phosphosites identified and quantified in MNUs (Additional file 1: Fig. S1K; Additional file 13: Table S1; Methods). The average Spearman’s correlation coefficient, calculated for all quality control runs of HEK293T cell samples, was 0.9, showing that the MS data were of high quality (Additional file 1: Fig. S1L).
Transcriptional sequencing was carried out on forty-three tumors and paired 22 MNU samples; we identified 17,091 genes per tumor sample and 14,738 genes per MNU sample, with fragments per kilobase of transcript per million mapped reads (FPKM) of more than 1 (Additional file 1: Fig. S1M). The number of genes identified as corresponding to the proteome (unique peptide ≥ 2), phosphoproteome, and transcriptome was 2221 for MNC samples and 4344 for tumor samples (Additional file 1: Fig. S1N). In addition, we calculated the correlation between 5001 mRNA–protein pairs for UC tumors and 3983 mRNA–protein pairs for MNU samples (Fig. 1J; Additional file 13: Table S1). The median correlation value of MNU was 0.15, whereas tumors had a higher median value of 0.26. This result is similar to that of previous studies investigating ccRCC and higher-grade serous ovarian cancers [22, 23]. In MNU samples, 74.6% of mRNA–protein pairs showed positive Spearman correlation coefficients associated with pathways, such as the cellular amide metabolic pathway and organonitrogen compound biosynthetic pathway, whereas genes showing negative correlations were enriched in RNA localization. In tumor samples, 86.7% of mRNA–protein pairs showed positive Spearman correlations with pathways including the cell adhesion molecules cams and valine, leucine and isoleucine degradation, whereas genes with negative correlations were enriched in the spliceosome. The discordance between transcriptomics and proteomics suggests that proteomics data possessed unique oncogenic features that cannot be obtained from genomic and transcriptomic data.
Effects of copy number alterations
We profiled 113 tumors for somatic copy number alteration (CNA) using whole-genome sequencing and examined the regulatory effects of 25,961 CNAs on mRNA, protein, and phosphoprotein expression. CNAs affect mRNA, protein, and phosphoprotein abundance in either “cis” or “trans” modes, corresponding to the diagonal and vertical patterns (Fig. 2A; Additional file 2: Fig. S2A). Interestingly, cis-regulatory effects of CNA (Fig. 2A; diagonal lines) on mRNAs and proteins were more prominent than those of phosphoproteins (Additional file 2: Fig. S2A). A total of 5186, 2841, and 494 significant correlations (cis-effects) were observed for mRNA, proteins, and phosphoproteins, respectively, with only 139 significant cis-effects overlapping among all three omics levels (Fig. 2B). These 139 overlapping genes were significantly enriched in positive regulation of GTPase activity, regulation of cell cycle, focal adhesion, and the ErbB signaling pathway (Fig. 2C), suggesting that core pathways were affected by genomic aberrations. Apart from the overlapped cis genes, the cis-effects on mRNA, protein, and phosphoprotein were enriched in different pathways. Specifically, mRNA-specific cis-effects (n = 3682) were enriched in RNA processing, whereas protein-specific cis-effects (n = 1315) were enriched in metabolic pathways, and phosphoprotein-specific cis-effects (n = 161) were enriched in certain signaling pathways (Additional file 2: Fig. S2B). To further nominate functionally important genes within CNA regions, we focus on the 593 cancer-associated genes (CAGs), in which 555 were identified in our cohort (Fig. 2D; Additional file 14: Table S2). A total of 10 significant positive correlations (RBL1, TPR, MTOR, IRF6, TBX3, RB1, PRKCD, MTUS1, CDK12, and ERBB2) were observed on mRNA, proteins, and phosphoprotein levels. Moreover, besides cis-effects, these ten genes also impacted the expression of proteins enriched in RNA splicing, cell proliferation through trans-effects (Fig. 2E). We further investigated the cis- and trans-effects of these ten CAGs in TCGA cohort [12]. As a result, in consistent with our findings, besides elevating their cognate proteins, these ten CAGs also impacted the expression of proteins participated in cell proliferation, RNA splicing, and transcription pathways through trans-effects (Fig. S2C).
CNA analysis showed the most frequent gains in chromosomes 1p, 1q, 3q, 5p, 7p, 7q, 8q, 17q, 19p, 19q, 20p, and 20q and losses in chromosomes 4q, 5q, 8p, 9p, 9q, 10q, 11p 11q, 18p, and 22q (Fig. 2F; Additional file 14: Table S2), and this result was consistent with those of previous studies [24,25,26]. To be more specific, ten out of twelve most frequent gains detected in our cohort (chromosome 3p, 5p, 8q, and 17q amplifications, etc.) were also identified in TCGA [12] and Beijing cohort [19] (Additional file 2: Fig. S2D, E). Meanwhile, seven out of ten most frequent loss (chromosome 8p and 9p deletions, etc.) detected in our cohort were also identified in TCGA and Beijing cohort (Additional file 2: Fig. S2D, E). In addition, we identified amplifications in driver oncogenes, including MYCL (1p34.3, 9 cases), PPARG (3p35.3, 13 cases), ERBB2 (17q12, 10 cases), and CCNE1 (19q12, 11 cases) (Additional file 2: Fig. S2F; Additional file 14: Table S2), and deletions in key tumor suppressors, such as CDKN2A (9p21.3, 26 cases), RB1 (13q14.2, 13 cases), and NCOR1 (17p12, 14 cases) (Additional file 2: Fig. S2F; Additional file 14: Table S2). We found that 5p gain and 7q gain were associated with both poor overall survival (OS) and inferior progressive-free survival (PFS) (Fig. 2G, H; Additional file 3: Fig. S3A). On chromosome 5p, 27 and 30 significant positive cis-effects were observed on mRNA and protein levels, respectively (Additional file 3: Fig. S3B; Additional file 14: Table S2). RICTOR, a cis-effect gene assigned to 5p at the protein level, is a critical regulator of cell migration and invasion in bladder cancer cell lines [27]. In chromosome 7q, which is associated with chromosomal instability and many types of neoplasia [28], 19 and 58 significant positive cis-effects were observed at mRNA and protein levels, respectively (Additional file 3: Fig. S3C; Additional file 14: Table S2). Proteins overrepresented due to 7q gain were significantly enriched in DNA replication and G1/S transition of the cell cycle (Wilcoxon rank-sum test, p < 0.05, 7q gain/WT ratio > 2).
The levels of proteins encoded by the genes linked to significantly positive cis-effects on chromosome 5p (C9) and 7q (SND1 and ELN), associated with poor prognoses, but the mRNA expression levels were not (Fig. 2I). SND1, a transcriptional co-activator overexpressed in tumors (Additional file 3: Fig. S3D; Wilcoxon rank-sum test, FDR < 0.01), is also associated with poor patient prognoses in the TCGA BLCA cohort (Fig. 2J). Overexpression of SND1 has been detected in various cancer types in TCGA (Additional file 3: Fig. S3E). Notably, SND1 expression significantly increased with pathological stage in both TCGA BLCA and our cohort (Additional file 3: Fig. S3F, G). SND1 is reported to interact with transcription factors, such as STATs and E2F1, modulating the expression of genes that promote carcinogenesis [29,30,31]. We found that the abundance of SND1 protein in urothelial bladder tumors positively correlated with the cell cycle KEGG (Kyoto Encyclopedia of Genes and Genomes) gene set (Fig. 2K; Additional file 3: Fig. S3H; Spearman’s r = 0.27, p = 3.5 × 10E−3) and the protein abundance of MCM2, a cell proliferation marker (Additional file 3: Fig. S3I; Spearman’s r = 0.3, p = 1 × 10E−3). STAT3, a transcription factor that interacts with SND1 [32], showed the highest correlation with the abundance of SND1 protein in tumors (Fig. 2L, M; Spearman’s r = 0.32, p = 7.3 × 10E−4). A significantly positive correlation between SND1 and STAT3 was also confirmed in the TCGA BLCA cohort (Additional file 3: Fig. S3J; Spearman’s r = 0.16, p = 1.0 × 10E−3). Furthermore, the predicted STAT3 activity, inferred via mRNA expression of its target genes (Methods), positively correlated with the protein abundance of SND1 (Additional file 3: Fig. S3K; Spearman’s r = 0.29, p = 6.7 × 10E−3), and patients showing higher expression of STAT3 protein in tumors appeared to have worse prognostic outcomes (Additional file 3: Fig. S3L; log-rank test, p = 8.9 × 10−3). Expression of many genes participating in the cell cycle, which are STAT3 targets, were upregulated along with increasing STAT3 activity (Fig. 2N, P; Spearman’s r = 0.50, p = 9.4 × 10E−4). We also confirmed the significantly positive correlation between STAT3 and the cell cycle KEGG gene set in the TCGA BLCA cohort (Fig. 2O; Spearman’s r = 0.28, p = 8.9 × 10E−9). In addition, phosphorylation of STAT3 also significantly correlated with the protein abundance of SND1 (Additional file 3: Fig. S3M; Spearman’s r = 0.26, p = 8.2 × 10E−3). STAT3 is reported to be phosphorylated by CDK5 [33]. Surprisingly, we found that CDK5 also exerted a cis-effect on chromosome 7 at the protein level (Fig. 2R; Spearman’s r = 0.24, p = 1.1 × 10E−2). Further analysis indicated that phospho-STAT3 was positively correlated with the protein abundance of CDK5 in tumors (Fig. 2Q, R; Spearman’s r = 0.24, p = 1.8 × 10E−2). Supporting our findings, STAT3 was also reported to promote cell proliferation in bladder cancer cell lines, WH, UMUC-3, and 253 J [34]. Our data identified two cis-effects on chromosome 7q, SND1 and CDK5. Upregulation of SND1 expression modulated STAT3 activity, while CDK5 further phosphorylated STAT3, which was related to cell proliferation (Fig. 2S).
Integrated multi-omics analyses of tumors and MNUs
Multi-omic profiles of both tumors and MNUs were derived, presenting a unique opportunity to explore multi-omic remodeling upon tumorigenesis. We compared differences between tumors and MNUs at different omics levels. Principal component analysis (PCA) of RNA-seq (27,828 genes), proteome data (5546 proteins), and phosphoproteome data (1672 phosphoproteins) showed a clear separation of tumors and MNUs at all three omics levels (Additional file 4: Fig. S4A–C; Additional file 15: Table S3). Differential gene analysis between tumors and MNUs resulted in 1726 mRNA, 2676 proteins, and 784 phosphoproteins (Additional file 4: Fig. S4D–F; Wilcoxon rank-sum test, FDR < 0.01, T/MNU ratio > 2 or < 1/2; Additional file 15: Table S3). For further comparison, we include data from recent publish proteomic landscape of 16 major types of human cancer [35]. By performing comparative analysis, we observed that 225 of the 288 urothelial cancer-type-specific proteins identified by Zhou Y. et al. were also identified in our proteomic data (Additional file 4: Fig. S4G). Further pathway enrichment analysis using these differential genes indicated that upregulation of the cell cycle pathway and downregulation of the transmembrane transport pathway in tumors occurred at all three omics levels (Fig. 3A; Additional file 4: Fig. S4H–J), suggesting that the differences between tumors and MNUs were reflected at all omics levels. Key factors in the cell cycle and transmembrane transport pathways, such as CDK1 and SLC14A1, were correlated with clinical outcomes (Additional file 4: Fig. S4K). Notably, the urothelial bladder-specific proteins nominated by Zhou et al. [35], such as those of the UPK family (uroplakins; UPK3A and UPK3BL among others), were observed to be downregulated in tumors (Fig. 3A; Wilcoxon rank-sum test, FDR < 0.01, T/MNU ratio < 0.5), and patients with higher expression of those uroplakin proteins in tumors appeared to have better prognostic outcomes (Fig. 3B; log-rank test, p < 0.05). These proteins are specifically expressed in the urothelial epithelium, necessary for urothelial bladder permeability barrier [36]. The loss of these proteins in tumors implied the cellular atypia due to initiation of urothelial bladder cancer in the urothelial epithelium.
Interestingly, we found divergence in tumors and MNUs differences among the three omics levels. For example, the 174 proteins (Wilcoxon rank-sum test, FDR < 0.01, T/MNU ratio > 2) that showed greater changes in protein abundance than in corresponding mRNA abundance (Wilcoxon rank-sum test, FDR < 0.01, T/MNU ratio < 2) were mainly involved in pathways related to protein processing in endoplasmic reticulum and cell–cell adhesion (Fig. 3C). The correlation between the differences in tumor-MNU protein and mRNA abundance was intermediate compared to previous findings (Fig. 3C, Spearman’s r = 0.27; p = 2.2 × 10−16) [37, 38]. To assess the potential clinical relevance of genes showing significant mRNA–protein correlations, we examined whether these were associated with patient survival. Interestingly, genes associated with significant differences in survival, especially at both protein and mRNA levels, showed the strongest mRNA–protein correlations (Fig. 3D; Additional file 15: Table S3). Further pathway enrichment analysis showed that genes with significant survival differences at both protein and mRNA levels participated in focal adhesion, ECM-receptor interaction, and PI3K-Akt signaling pathways (Fig. 3E), suggesting that alterations in key signaling, especially at all omics levels, were associated with tumorigenesis and tumor progression.
To further investigate the dominant signal transduction pathway, we studied the phosphoproteome. The results revealed that 991 phosphosites mapped to 379 phosphoproteins showing greater changes than corresponding protein abundance (Fig. 3F; Wilcoxon rank-sum test, FDR < 0.01, T/MNU ratio > 2) were significantly enriched in pathways, including the cell cycle and cell–cell adhesion (MCM2-S13, TP53BP1-S1367, and MYH9-S1943 among others) (Fig. 3G). Kinase substrate enrichment analysis (KSEA) of the phosphoproteome of tumors and MNUs revealed the dominant kinases that were activated in tumors, including CSNK2A1, AURKA, VRK1, PRKDC, MAP2K7, and ERBB2 (Fig. 3H; Methods). Among these, many activated kinases, such as PRKDC, MAP2K7, and ERBB2, are targets of approved inhibitors. Further investigation of differentially changed phosphosites showed that elevated substrates involved in the cell cycle, ErbB signaling, and cell–cell adhesion were observed in tumors (Fig. 3I). In addition to kinase targets, plasma membrane proteins are also attractive therapeutic targets in cancer treatment. Therefore, we further performed supervised analysis to filter out plasma membrane proteins (Fig. 3J; Methods). Eighteen proteins (FLNA, PKN1, LAMA4, etc.) met the screening criteria and were annotated for the degree to which short hairpin RNA (shRNA)-or CRISPR-mediated depletion reduced survival and proliferation in urothelial cancer cell lines [39, 40] (Fig. 3K; Additional file 15: Table S3). We further investigated these 18 proteins by assessing immunohistochemistry (IHC) expression in the Human Protein Atlas (HPA). Five of these 18 exhibited overrepresented tumor-specific staining in urothelial bladder cancer samples (Additional file 4: Fig. S4L, M), including PKN1, which had approved drug and were also associated with prognosis (Additional file 4: Fig. S4N). However, the other 13 proteins did not yield staining information or show low staining in HPA, and merit further investigation. An overview of significantly enriched pathways between the tumors and MNUs in urothelial bladder cancer served as guide for future studies and therapeutic opportunities (Fig. 3L).
Proteogenomic profiles distinguished NMIBC from MIBC
According to the T-category and histologic grading, UC is divided into NMIBC and MIBC (T-category) or high-grade and low-grade (histologic grading) [41]. Our cohort contained 116 UC patients, comprising 45 NMIBCs (Ta [n = 11], T1 [n = 34]) and 71 MIBCs (T2 [n = 46], T3 [n = 22], T4 [n = 3]), that were mainly invasive (including propria membrane infiltration and muscle infiltration). As for the histologic grading, MIBCs were all high-grade, while 7 of NMIBCs were low-grade and 38 were high-grade. We showed representative hematoxylin and eosin (H&E)-stained slides of low-grade NMIBC (NMIBC LG)/high-grade NMIBC (NMIBC HG) and MIBC samples (Additional file 5: Fig. S5A–C). MIBC has a poor prognosis compared with NMIBC due to early occult metastatic dissemination [42], which was also observed in our data (Fig. 4A). PCA analysis of RNA-seq (27,752 genes), proteomic (5683 proteins), and phosphoproteomic (2014 phosphoproteins) data separated MIBC samples from NMIBCs (Fig. 4B; Additional file 5: Fig. S5D, E), revealing the molecular differences between MIBCs and NMIBCs. Pathway enrichment analysis of differentially expressed molecules (Additional file 5: Fig. S5F, G; Wilcoxon rank-sum test, p < 0.01, MIBC/NMIBC ratio > 2 or < 1/2) showed that NMIBC-enriched molecules were involved in oxidative phosphorylation and lipid metabolism, including glycerophospholipid and arachidonic acid metabolism, whereas molecules enriched in MIBC mainly participated in regulation of actin cytoskeleton and complement and coagulation cascades (Fig. 4C; Additional file 5: Fig. S5H–J). To further investigate the differences among low-grade NMIBC, high-grade NMIBC, and MIBC, we surveyed the differential expressed proteins in the featured pathways in the low-grade NMIBC, high-grade NMIBC, and MIBC. Interestingly, we found that the expression level of proteins participated in MIBCs-enriched pathways gradually increased from low-grade NMIBC, high-grade NMIBC to MIBC, while the expression level of proteins participated in NMIBCs-enriched pathways gradually decreased (Additional file 5: Fig. S5K, L; Kruskal–Wallis test, p < 0.0001). For example, proteins that function in regulation of actin cytoskeleton (EHD2, ELN, LCP1, etc.) showed a gradual increasing trend from low-grade NMIBC, high-grade NMIBC to MIBC and presented relative high-risk scores for a mortality prognosis of UC (Additional file 5: Fig. S5M). In contrast, proteins in oxidative phosphorylation and lipid metabolism (NDUFA7, CYP2J2, GPX2, etc.) presented a gradual decreasing trend from low-grade NMIBC, high-grade NMIBC to MIBC and showed low-risk scores for a mortality prognosis of UC (Additional file 5: Fig. S5N). This finding reveals that the differences identified in NMIBC and MIBC were also found in low-grade NMIBC and high-grade NMIBC and presented a gradual tendency from low-grade NMIBC, high-grade MIBC to MIBC.
To determine the divergence of genomic drivers in MIBC and NMIBC, we compared the differences in genomic variations between them. FGFR3 mutations were observed more frequently in NMIBCs, whereas TP53 and RB1 displayed higher mutation rates in MIBCs (Fig. 4C), which was consistent with previous findings [42]. At the arm event level, chromosome gains, such as chromosome 5p, 7p, and 20q gains, were more predominant in MIBCs than in NMIBCs (Fig. 4C; Fisher’s exact test, p < 0.05). Among the significant differential arm-level CNA events, only 5p gain was associated with poor prognosis in the entire cohort (Fig. 4D; log-rank test, p < 0.05). Surprisingly, we found that the poor prognoses of 5p gains were observed only in NMIBC patients but not in MIBC patients (Fig. 4E; log-rank test, p < 1 × 10−3). We further investigated the proportion of 5p gains in NMIBC and MIBC patients and found that the percentage of patients with 5p gain in MIBCs (80%) was higher than that in NMIBCs (20%), implying the function of 5p gain in the progression from NMIBC to MIBC. Pathway enrichment analysis using proteomic data showed upregulation of proteins involved in antigen processing and presentation, actin filament-based movement, and regulation of GTPase activity in the 5p gain group, compared with the WT group (Fig. 4F; Additional file 16: Table S4), which was consistent with MIBC-enriched pathways.
To further investigate the potential mechanism, we focused on the cis-effects of chromosome 5p. A total of 27 and 30 significantly positive cis-effects were observed at the mRNA and protein levels, respectively, in which nine cis-effects overlapped between both (Fig. 4G). Among these nine cis-effects, only Trio Rho guanine nucleotide exchange factor (TRIO) was significantly upregulated in MIBC as compared to NMIBC at both the mRNA and protein levels (Fig. 4G, H; Wilcoxon rank-sum test, p < 0.05, MIBC/NMIBC ratio > 1.5; Additional file 16: Table S4). TRIO encodes a large protein that functions as a GDP to GTP exchange factor for Rho GTPases, which plays a role in cell invasion and growth by promoting actin remodeling [43, 44]. The abundance of TRIO at both mRNA and protein levels in urothelial bladder tumors was positively correlated with the reorganization of the actin cytoskeleton gene ontology (GO) biological processes (BP) gene set (Fig. 4I; Additional file 5: Fig. S5O). Expression of RHOG, a TRIO-activating Rho GTPase [45], positively correlated with TRIO at the mRNA level (Fig. 4J, K; Spearman’s r = 0.42, p = 5.6E−3). A significantly positive correlation between TRIO and RHOG was also observed in the TCGA BLCA cohort (Additional file 5: Fig. S5P; Spearman’s r = 0.25, p = 3.3E−7). In addition, Rho-associated protein kinases (ROCKs) are reported as the best-characterized downstream effectors of Rho GTPases [46]. The correlation between RHOG and ROCK1 was significantly positive at both mRNA and protein levels (Fig. 4L; Additional file 5: Fig. S5Q). We then performed kinase activity analysis based on the levels of substrate phosphorylation and compared specific activated kinases between MIBC and NMIBC (Methods). As a result, ROCK1 was found to specifically activate kinases in MIBC and was targeted by FDA-approved drugs (Fig. 4M). The expression of ROCK1 substrates (ACTG1 T318, MYLK S1776, CALD1 S202, etc.), which facilitate regulation of the actin cytoskeleton and cell motility, was also upregulated in MIBC (Fig. 4N). Therefore, TRIO activated RHOG and then RHOG activated downstream effectors ROCK1, thereby increasing the reorganization of the actin cytoskeleton (Fig. 4O, P). In sum, our data revealed the potential role of 5p gain in progression from NMIBC to MIBC, through mechanism of modulating actin cytoskeleton implicating in tumor cells invasion.
Proteomic subtypes of UC and signature proteins
Consensus clustering identified three proteomic subtypes based on 5489 proteins present in more than 30% of 116 tumors (Fig. 5A; Additional file 6: Fig. S6A; Methods). They were designated U-I (n = 37), U-II (n = 23), and U-III (n = 56) with distinct molecular and clinical features (Fig. 5A–C; Additional file 17: Table S5). Patients in U-I had the best OS and PFS, whereas patients in U-III had the worst OS and PFS (Fig. 5B; log-rank test, p < 0.05). Combined with the clinical data, tumors with papillary and NMIBCs were mostly enriched in the U-I (Fig. 5A, Fisher’s exact test, p < 0.05), whereas patients in U-III had a higher degree of nerve invasion, metastasis, and vascular invasion (Fig. 5A, Fisher’s exact test, p < 0.05). Univariate Cox regression analysis of proteomic subtypes and clinical features is shown in Additional file 6: Fig. S6B. The results revealed that proteomic subtypes (Additional file 6: Fig. S6B; hazard ratio, 2.0; 95% confidence interval [Cl], 1.3–3.1; p < 0.001; Additional file 17: Table S5) were authenticated as an independent prognosticator, after controlling for stage, nerve invasion, and vascular invasion.
Subgroup-specific pathway enrichment analysis indicated different features among the three proteomic subgroups. Subgroup U-I was characterized by the highest level of metabolism-related pathways, such as propanoate metabolism, lipid metabolism process, and arachidonic acid metabolism (CYP2J2, ALOX5, AKR1C3, etc.) (Fig. 5C). Subgroup U-II was more related to tumor proliferation, including cell cycle (MCM2, ANAPC4, CHEK1, etc.), RNA splicing, and DNA repair. Subgroup U-III was characterized by pathways, such as those pertaining to extracellular matrix disassembly, complement and coagulation cascades, and PI3K-AKT signaling pathway; some of these were associated with tumor environment and immune response. Genes linked to representative pathways among different proteomic subgroups at different omics levels are shown in Fig. 5A. Notably, some genes were correlated well with the clinical outcomes (Additional file 6: Fig. S6C). Furthermore, multiple previously described UC subtype markers are shown in Additional file 6: Fig. S6D. FGFR3 signatures (IRS1, FGFR3, PTPN13, etc.) were highly expressed in U-I (Wilcoxon rank-sum test, p < 0.05), while differentiation signatures (PPARG, SPINK1, DHRS2, etc.) were expressed at higher levels in U-I and U-II. Interestingly, tRNA aminoacylation (GARS, RARS, and TARS) was highest expressed in U-II. EMT signatures (TGFBI, VIM, and CAV1), wild-type p53 signatures (DES, FLNC, CNN1, etc.), and CSC signatures (NES, CD47, and THY1) were overexpressed in U-III. These results emphasized the association between increased biosynthetic, translation, and turnover rates and rapid tumor proliferation.
To directly translate our findings into laboratory tests for tumor classification, we performed differential expression analysis and functional analysis and identified 24 proteins biomarkers that showed dominant expression in a specific proteomic subgroup and were functionally relevant to the main function of the distinctive subgroup (Additional file 6: Fig. S6E). We further performed survival analysis and found eight (CYP2J2, PRKCB, COL1A1, etc.) were correlated with poor prognosis (Additional file 17: Table S5). We then randomly selected three protein marker candidates (CYP2J2, MLH1, and PRKCB) to validate their expression in specific proteome subgroup (U-I, U-II, and U-III). As a result, in consistent with our proteomic data, PRKCB was confirmed to be overrepresented in U-III, MLH1 was overrepresented in U-II, and CYP2J2 was overrepresented in U-I (Additional file 6: Fig. S6F). These suggested that the panel of biomarker candidates could be used to distinguish different subtypes in clinic.
To explore the correlation between proteomic subtypes and mutational signatures, 12 main signatures were identified (Additional file 7: Fig. S7A; Additional file 17: Table S5; Methods). Signature 1 (aging-related) was a dominant identified in 82 patients (Additional file 7: Fig. S7A; Additional file 17: Table S5). The other major signatures were Signature 2 (APOBEC-a, n = 33), Signature 6 (defective mismatch repair, n = 66), Signature 13 (APOBEC-b, n = 23), and Signature 16 (strong transcriptional strand-bias for C>T, n = 8). The most dominant signature, Signature 1, was mostly identified in subgroup U-I (Additional file 7: Fig. S7A; Kruskal–Wallis test, p < 0.05). Signature 2 (APOBEC-a) and Signature 13 (APOBEC-b) were mainly identified in subgroups U-II and UI-III (Fisher’s exact test, p < 0.05). Four signatures were significantly associated with survival, nerve invasion, TNM stage, vascular invasion, and papillary (Additional file 7: Fig. S7B; Wilcoxon rank-sum test, p < 0.05). Signature 1 was mostly found in patients carried papillary carcinoma (Additional file 7: Fig. S7B; p < 0.05), Signature 6 was mostly observed in patients with lower TNM stage (Additional file 7: Fig. S7B; p < 0.05), and Signature 16 was mostly found in patients with vascular invasion (Additional file 7: Fig. S7B; p < 0.05).
Furthermore, we conducted clustering analyses on the tumor transcriptome (n = 43, consensus clustering) and phosphoproteome (n = 105, consensus clustering) (Methods; Additional file 6: Fig. S6A; Additional file 17: Table S5) and also identified three subtypes in each dataset. Moderate concordance among proteomic, transcriptomic, and phosphoproteomic subgroups was uncovered (59.0% between proteome and transcriptome and 39.5% between proteome and phosphoproteome). The phosphoproteomic subgroup with poor overall survival was consistent with the proteomic subgroup U-III. An analysis combining the mRNA, protein, and phosphoprotein helps to decipher the diverse biology and heterogeneity of the molecular processes within UC.
We compared the three clusters with the results obtained from earlier classifiers (Fig. 5D). Comparison between UNC classification [11] and ours indicated that U-I and U-II (as revealed by our classifier) matched well with the luminal subtype, while U-III matched well with the basal-like subtype. In detail, the luminal markers (KRT18, KRT7, GPX2, etc.) were overexpressed in U-I and U-II, whereas basal markers (KRT14, KRT5, STAT3, etc.) were overexpressed in subgroup U-III (Fig. 5E). Comparison between MDA classification and our results indicated that U-I matched well with the luminal subtype, U-II matched well with the TP53-like subtype, while U-III subgroup matched well with the basal subtype (Fig. 5D). Comparison between MDA classification [10] and ours indicated that urobasal A and genomically unstable subtypes were enriched in U-I, the infiltrated subtype was enriched in U-II, and SCC-like and urobasal B subtypes were enriched in U-III (Fig. 5D). Comparison between TCGA classification [8] and ours indicated that U-I agreed well with luminal-papillary and luminal (Fig. 5D), U-II agreed well with luminal-infiltrated, while U-III was consistent with basal–squamous types. These results revealed that our proteome subgrouping showed consistencies with transcriptome-based subgrouping. Since proteins are the major executors of biological functions, the proteomic subgrouping reinforces previous transcriptome data and facilitates the discovery of variant proteins, serving as the resource of the biomarker candidates and therapeutic targets.
Considering that protein kinases have been developed as viable drug targets of cancer therapy, we next inferred kinase activities based on differentially abundant phosphosites in each proteomic subtype, by performing kinase–substrate enrichment analysis (Methods). Significant differences between the inferred activated kinases were observed among the three proteomics subtypes, of which U-I was predominantly featured in the TK kinase group (ERBB3, SRC, YES1, etc.), U-II was characterized by CMGC (CDK1, CDK4, WEE1, etc.), and U-III was characterized by two major kinase groups, AGC (PRKCB, PRKCE, and STK10, etc.) and CMGC (GSK3A, GSK3B, CDK5, etc.) kinase groups (Fig. 5F; Additional file 7: Fig. S7C). Further investigation into the differentially altered phosphosites showed that elevated substrates involved in the ERBB pathway (EGFR-pY1110, SRC-pY149, MAPK1-pY187, etc.) were observed in U-I, DNA repair, and the cell cycle (LIG1-pS76, MCM4-pS3, MCM2-pS13, etc.) in U-II, and Interleukin-2 production and extracellular matrix organization (VIM-pS34, ACTG1-pS323, and MYH9-pS1714, etc.) in U-III (Fig. 5G). These findings suggest that different proteomics subtypes are featured with different kinase and could be treated with corresponding kinase inhibitors. For example, Patritumab, an ERBB3 inhibitor, and Saracatinib, a SRC inhibitor, have the potential to be utilized for patients in U-I, Dinaciclib, a CDK1 inhibitor, and Palbociclib, a CDK4 inhibitor, for patients in U-II, and Enzastaurin, a PRKCB inhibitor, for patients in U-III (Fig. 5G). Importantly, among the potential inhibitors nominated by us, Saracatinib was demonstrated potent antimigratory and anti-invasive effects in vitro and inhibited metastasis in a murine bladder cancer model [47]. CDK4/6 inhibitors (Palbociclib) have been tested in bladder cancer [48] and reported to be potential therapeutic agents for RB positive bladder cancer [49].
Genomic information showed that subgroup U-I and U-II carried a higher mutation rate of FGFR3 (Fig. 5A; Fisher’s exact test, p < 0.05). The mutational hotspots in FGFR3 in our cohort were similar to those in the TCGA BLCA cohort (Additional file 7: Fig. S7D, E). Notably, most tumors carrying FGFR3 mutations also harbor PIK3CA mutations, while TP53 and FGFR3 mutations are mutually exclusive in bladder cancer. To investigate how mutations in FGFR3 drive its clinical features, we examined the expression of FGFR3 at different omics levels in patients with or without FGFR3 mutations. The results showed that FGFR3 expression was higher in patients carrying FGFR3 mutations at mRNA, protein, and phosphoprotein levels (Fig. 5H; Wilcoxon rank-sum, p < 0.05; Fold change > 2). To further establish a connection between genetic alterations and corresponding downstream pathways, we explored the correlation between the protein abundance of FGFR3 and enriched pathways. It has been reported that FGFR3 regulates mTORC1/2-cSREBP1 through PI3K/AKT-dependent and PI3K/AKT-independent signaling [50]. Notably, we found that the protein abundance of FGFR3 was positively correlated with the mTOR pathway (Fig. 5I). Additionally, higher expression of FGFR3 was positively correlated with higher phosphorylation of putatively druggable kinases AKT, RICTOR, and RPS6KB1 (Fig. 5J). The summary of the FGFR3 mutation associations is shown in Fig. 5K.
Immune cell infiltration in UC tumors
The tumor microenvironment component in our cohort was studied using xCell based on proteomic data of 116 tumors, which had been used in proteomic consensus clustering (Methods; Additional file 18: Table S6). These molecularly based cell-type classifications were supported by ESTIMATE analysis [51] (Methods; Additional file 18: Table S6), yielding the Pearson correlation coefficients of 0.68 and 0.69 for protein and mRNA data of immune- and stromal-derived signatures, respectively (Additional file 8: Fig. S8A). Consensus clustering of the cell signatures identified two NMIBC subtypes and three MIBC subtypes as follows: Cold-mixed (n = 22); Cold-tumor (n = 27); Metabolism (n = 11); Cell cycle (n = 30); and Hot-tumor (n = 26) (Fig. 6A, C). Among these, Cold-mixed and Cold-tumor were obtained from NMIBC. Comparing this with proteomic subtypes, we observed that hot-tumor was enriched in U-III, while the Cold-tumor cluster was compatible with U-I (Fig. 6A, E). In addition, combined with clinical data, we observed the immune subgroups significantly differed in OS and PFS (Fig. 6D; Additional file 8: Fig. S8B; log-rank test, p < 0.05).
The Cold-tumor cluster tumors displayed low immune and stromal scores (Fig. 6A; t test, p < 0.05), higher frequencies of FGFR3 mutations, ART1 mutations, and 9q34.3 deletions (Fisher’s exact test, p < 0.05), increased mTOR signaling and ERBB signaling (Wilcoxon rank-sum, p < 0.05), as well as the enrichment of endothelial cells and keratinocytes (t test, p < 0.05). In particular, keratinocytes were upregulated in tumors carrying FGFR3 mutations compared with the wild-type (Fig. 6B).
The Hot-tumor cluster was characterized by the highest immune score and multiple types of immune cells, including neutrophils, eosinophils, and multipotent progenitors (MPP) (Fig. 6A; t test, p < 0.05). Proteomic analysis has showed that upregulation of immune-related pathways, including that of JAK STAT signaling (Fig. 6A, C; Kruskal–Wallis test, p < 0.05), is involved in tumor cell recognition and tumor-driven immune escape [52]. Expression levels of the immune evasion markers, CD40, WARS, SAMD9L, GBP1, and GBP5, were upregulated in this cluster (Fig. 6A; Wilcoxon rank-sum, p < 0.05; Fold change > 2), as were the mRNA signatures (CD274 and CTLA4) associated with T cell exhaustion, implicating for immune checkpoint therapy.
The Cold-mix cluster was distinguished from the Cold-tumor cluster by stronger signatures for Pro B-cells, CMP, and pDC (Fig. 6A; t test, p < 0.05), upregulation of O-glycan biosynthesis, and interaction between L1 and ankyrins (Fig. 6A; Kruskal–Wallis test, p < 0.05), and by containing a higher degree of male patients (Fisher’s exact test, p < 0.05). The Cell cycle cluster was characterized by CD4+ T cells, CD4+ memory T cells, and Th1 cells (Fig. 6A; t test, p < 0.05), and regulation of the cell cycle and mismatch repair pathways (Fig. 6A; Kruskal–Wallis test, p < 0.05). The Metabolism cluster showed a certain degree of similarity with the Hot-tumor cluster. This cluster was characterized by fibroblast cells and NKT (Fig. 6A; t test, p < 0.05), and regulation of metabolism pathways, such as histidine metabolism and beta-alanine metabolism (Fig. 6A; Kruskal–Wallis test, p < 0.05). We also found that proliferator-activated receptor (PPAR) signaling pathway (Fig. 6A; Kruskal–Wallis test, p < 0.05) was regulated in this cluster.
Interestingly, we found that 9q34.3 amplification, which was significantly positively correlated with the immune score (Fig. 6F; Spearman’s r = 0.22, p = 0.016), was observed more frequently in the Hot-tumor cluster than in the Cold-tumor cluster (Fig. 6E; Fisher’s exact test, p < 0.05). To further investigate the potential mechanism by which 9q34.3 affects immune activity in UC, we focused on the effect of cis- on 9q34.3. Eleven significantly positive cis-effects were observed at protein level, and three of these (QSOX2, TRAF2, and UAP1L1) were associated with poor prognosis (Fig. 6G; hazard ratio > 2, p < 0.05). Tumor necrosis receptor-associated factor 2 (TRAF2) was overexpressed in the Hot-tumor cluster tumors, compared to the Cold-tumor cluster at both mRNA and protein levels (Fig. 6H; Fold change > 2; Wilcoxon rank-sum test, p ≤ 0.061). Next, we found that the protein abundance of TRAF2 was positively correlated with the TNFR2 non-canonical NF-κB pathway (Additional file 8: Fig. S8C; Spearman’s r = 0.32, p = 5e−4) [53], and the mRNA abundance of TRAF2 was positively correlated with the mRNA abundance of TNF (Additional file 8: Fig. S8D; Spearman’s r = 0.37, p = 0.016) and TNFR2 (Additional file 8: Fig. S8E; Spearman’s r = 0.54, p = 3.2e−4). These results indicate the activation of NF-κB1. We further found that the predicted NF-κB1 activities, inferred by the mRNA expression of its target genes (Methods), were positively correlated with the protein abundance of TRAF2 (Additional file 8: Fig. S8F; Spearman’s r = 0.32, p = 0.039). Many target genes of NF-κB1 were upregulated along with the increase in the NF-κB1 activity (Fig. 6I). Some studies had revealed that several signaling cross talk pathways, such as p53, STAT3, and NF-κB, regulate PD-L1 expression [54]. Surprisingly, the mRNA abundance of PD-L1, a target of NF-κB1, was positively correlated with the mRNA abundance of TRAF2 (Additional file 8: Fig. S8G; Spearman’s r = 0.39, p = 0.012). The significant correlation between TRAF2 and PD-L1 was also observed in the TCGA BLCA cohort (Additional file 8: Fig. S8H; Spearman’s r = 0.32, p = 2.2e−11). Furthermore, the mRNA abundance of TRAF2 was positively correlated with CD8 enrichment score (Additional file 8: Fig. S8I; Spearman’s r = 0.33, p = 0.033), indicating that TRAF2 plays a role in peripheral CD8+ T cell [55].
Based on the observation above, we performed further investigation on PD-L1 immunotherapy in our cohort. To be more specific, by surveying the clinical data of patients in our cohort, we found that 47 patients were detected for PD-L1 expression by IHC, in which 27 patients were defined as PD-L1 positive (TPS: > 1%), whereas 20 patients were defined as PD-L1 negative (TPS: < 1%). Further comparative analysis between PD-L1 positive and negative patients confirmed the positive association between the elevated expression of TRAF2 and increased PD-L1 expression (Fig. 6J; Additional file 8: Fig. S8J).
To further validate the positive association between TRAF2 and PD-L1, we collected FFPE samples from an independent validation cohort containing 14 UC patients treated with PD-L1 inhibitors (5 responders (PR, partial response), 9 non-responders (PD, progressive disease)). Clinical data are summarized in Additional file 18: Table S6. We examined the expression of TRAF2 and PD-L1 on the tissue level by immunohistochemistry of consecutive slides and observed significant elevated expression of both TRAF2 and PD-L1 in responders compared to non-responders (Fig. 6K, L; Additional file 8: Fig. S8K). This result further confirmed that the elevated expression TRAF2 is associated with the increased expression of PD-L1 and is related to patients’ responses. A summary of the TRAF2 amplification associations is shown in Fig. 6M.
Clinical features associated with proteomic and phosphoproteomic profiles
To explore the biological characteristics of our cohort in an unbiased proteome-wide manner, weighted correlation network analysis (WGCNA) was performed using 6692 proteins present in more than 10% of the 116 tumors (Methods). The clustering dendrogram of the samples is shown (Additional file 9: Fig. S9A–C; Additional file 13: Table S1). Co-expression analysis yielded 15 consensus modules (Fig. 7A), ranging in size from 157 proteins (MEmidnightblue module) to 2257 proteins (MEturquoise module). The modules were subsequently analyzed by pathway enrichment to characterize the associated biology (Fig. 7B; Additional file 19: Table S7).
Among these modules, MEred was significantly correlated with chromosome 5p gain (Fig. 7A; r = 0.18, p < 0.05). The genes in this module were enriched in regulation of cell proliferation (Fig. 7B; p = 2.21E−04) and positive regulation of GTPase activity (p = 2.21E−04). In addition, we found that serum albumin value and serum urea levels were correlated with poor prognosis (Additional file 9: Fig. S9D–E; log-rank test, p < 0.05). It has been reported that serum albumin shows potential as a reliable biomarker of inflammation. We further found that serum albumin was significantly correlated with the MEgreenyellow module, where pathway analysis of the genes in this module showed that abnormal serum albumin was associated with endocytosis (Fig. 7B; p = 6.89E−04), intracellular protein transport (p = 7.5E−03) and vesicle-mediated transport (p = 0.015). Furthermore, the MEblue module was significantly correlated with OS, in which 211 out of 607 genes were significantly correlated with clinical outcomes. The genes in this module were enriched in focal adhesion (Fig. 7B; p = 2.85E−34), complement and coagulation cascades (p = 2.29E−26), and the PI3K-AKT signaling pathway (p = 8.09E−11). We further performed supervised analysis to filter potential drug targets (https://www.proteinatlas.org/), and 23 genes (GARS, CFI, MYLK, etc.) met the criteria, in which six (GARS, CAV1, P4HA2, etc.) were reportedly correlated with poor prognosis and overrepresented in the staining of urothelial bladder cancer samples in the HPA database (Additional file 9: Fig. S9F–G; Additional file 19: Table S7).
Heterogeneity of histopathological characteristics adds complexity to the diversity of bladder cancer. Histologically, UC is divided into papillary (papilloma, low malignant potential, and papillary carcinoma) and non-papillary (urothelial carcinoma in situ and invasive) categories [56]. The non-papillary category is further classified into several differentiation forms, such as glandular differentiation, squamous differentiation, micropapillary differentiation, and invasive urothelial carcinoma not otherwise specified (NOS). Papillary carcinoma, NOS, glandular differentiation, squamous differentiation, and micropapillary differentiation accounted for 29.3%, 42.0%, 11.5%, 7.0%, and 3.8% of all tumor samples, respectively (Additional file 9: Fig. S9H). These results are consistent with those of previous studies [57, 58].
To find the divergence of tumors with different histological variants at the molecular level, proteomic data were analyzed using gene set enrichment analysis (GSEA) (Fig. 7C–D; Additional file 19: Table S7). The results revealed that metabolism-related pathways, such as oxidative phosphorylation and glycerophospholipid metabolism (PCYT2, MBOAT7, DGKA, etc.), were enriched in papillary carcinoma. Vascular endothelial growth factor receptor signaling (NRP1, ELMO1, NCK1, etc.), as well as cell adhesion mediated by integrin, was enriched in NOS. Furthermore, different differentiation variants were correlated with different pathways. For example, glandular differentiation was characterized by the citrate cycle (TCA) (PDHB, SDHA, PCK2, etc.) and N-glycan biosynthesis and alanine, as well as by aspartate and glutamate metabolism (IL4I1, GFPT1, ASNS, etc.). Squamous differentiation was characterized by keratinocyte differentiation (ANXA1, KRT16, SPRR1A, etc.) and hemidesmosome assembly (KRT14, LAMA3, KRT5, etc.). Micropapillary differentiation was distinguished by lysosomes (CTSC, SCARB2, LAMP1, etc.) and tight junctions (OCLN, CGN, MYH10, etc.).
To systematically identify druggable targets specific to histological variants, we performed functional enrichment analysis using phosphoproteomic data. The results showed that phosphoproteins showing high expression in papillary carcinoma were enriched in ERBB signaling and MAPK signaling. Phosphoproteins upregulated in NOS were enriched in focal adhesion and muscle contraction (Additional file 9: Fig. S9I). In addition, we pooled the tumors showing different differentiations, such as glandular differentiation, squamous differentiation, and micropapillary differentiation, into one group and named this the differentiation variation. We found that keratinocyte differentiation and cell division were enriched in the differentiation variation group (Additional file 9: Fig. S9I). Various kinase activities in tumors with different histological variations were assessed. ERBB3/ERBB4/RAF1 kinases were activated in papillary carcinoma compared to other variants, PAK3/PAK6/CDK1 kinases were activated in NOS, and PRKACA/PRKACB/PRKACG kinases were activated in differentiation variants (Fig. 7E). These kinases and corresponding substrates are shown in Fig. 7F–H. Among these kinases, CDK1 is reportedly the only essential member of the CDK subfamily, which plays an important role in cell cycle progression [59]. In summary, different histological variants were characterized by different pathways and activated kinases, providing evidence for the need for personalized treatment.
GARS promotes bladder cancer cell proliferation through non-canonical function
We found that the expression levels of GARS, which is known to be significantly increased in tumor tissues compared to MNUs, were also increased during UC progression (Fig. 8A). Since GARS has not been reported as being associated with the onset of bladder cancer, we explored the role of GARS in bladder cancer progression. Using western blotting, we confirmed that GARS protein levels were profoundly upregulated in tumor tissues (Fig. 8B). Other kinds of aminoacyl-tRNA synthetases, including AARS, TARS, and SARS, which were used as controls, were not significantly altered (Fig. 8B). Overexpressing GARS in the human urinary bladder carcinoma cell lines, T24 and 5637, promoted DNA synthesis (Additional file 10: Fig. S10A–B), cell cycle progression (Additional file 10: Fig. S10C–D), and cell proliferation (Additional file 10: Fig. S10E–F), while knocking down GARS inhibited these three processes. Moreover, although induction of genotoxic stress by cisplatin led to cell cycle progression block and decreased cell proliferation in both T24 and 5637 cells, increase in GARS in cisplatin-treated cells rescued the blocked cell cycle and exhibited a stronger pro-proliferative effect, compared to GARS in normal cell cultures (Additional file 10: Fig. S10A–F). The levels of other types of aminoacyl-tRNA synthetases, as well as the levels of 4EBP and S6K, were not altered between tumors and MNUs in UC, suggesting that the oncogenic role of GARS was not due to protein translation (Fig. 8B). Measuring the metabolites profile using LC–MS revealed that the pentose phosphate pathway was activated, while glycolysis was downregulated in GARS-overexpressing cells (Fig. 8C). These results suggest that upregulated GARS enhances DNA synthesis and promotes cell proliferation by activating pentose phosphate pathway flux. However, the mechanism by which GARS regulates glucose metabolism remains unknown.
GARS enhances metabolic flux in pentose phosphate pathway by inhibiting PGK1 and PKM2
In addition to mediating protein translation, GARS catalyzes the glycine modification of protein lysine residues (K-Gly) and transfers the glycine signal by altering the function of the modified protein K-Gly [60]. Accordingly, we found that K-Gly levels were significantly increased in tumor tissues, compared to MNUs, in UC patients (Fig. 8B). K-Gly levels were determined by the concentration levels of GARS and glycine. Nuclear magnetic resonance indicated that glycine levels between the tumor and MNUs of UC were not altered (Additional file 11: Fig. S11A), suggesting that increased K-Gly levels seen in UC were caused by upregulated GARS protein abundance. Furthermore, we validated that, in both cultured T24 and 5637 cells, overexpression of GARS led to increased global K-Gly levels in cells (Fig. 8D). In a previous study of our investigating protein lysine aminoacylation [60], we searched for lysine aminoacylation in a tryptic peptide library of human liver cancer and identified large numbers of proteins, as well as potential K-Gly modified proteins. The search results suggested that the enzymes in glycolysis were enriched, including PGK1 and PKM2 (Additional file 20: Table S8). To verify this, we first validated that the interaction between GARS and PGK1, as well as the interaction between GARS and PKM2, using co-immunoprecipitation assays with proteins exogenously expressed proteins in both 5637 and T24 cell lines (Fig. 8E). Moreover, in the tumor tissues, we used a co-immunoprecipitation assay to validate interactions between GARS and PGK1 as well as between GARS and PKM2 (Fig. 8F). Accordingly, we found that overexpression of GARS led to increased K-Gly levels of PGK1 and PKM2, in both 5637 and T24 cells (Fig. 8G). Although the abundances of PGK1 and PKM2 were not altered by GARS (Additional file 11: Fig. S11B), enzymatic activities of PGK1 and PKM2 were decreased in GARS-overexpressing T24 cells (Fig. 8H), suggesting that increased GARS levels enhanced the pentose phosphate pathway flux through attenuating the glycolysis flux by reducing the activities of PGK1 and PKM2. To inhibit the GARS-induced oncogenic effect on cells, we tested whether the structural analog of glycine was able to inhibit the K-Gly formation in cells, including alanine, beta-alanine, and sarcosine (Additional file 11: Fig. S11C). We found that only beta-alanine inhibited K-Gly formation in cultured cells (Fig. 8I). Furthermore, we found that beta-alanine inhibited the DNA synthesis (Additional file 11: Fig. S11D) and cell cycle progression (Additional file 11: Fig. S11E) and thus promoted the cell apoptosis (Additional file 11: Fig. S11F) and attenuated cell proliferation in both 5637 and T24 cells (Additional file 11: Fig. S11G). In addition, we noted that increased GARS abundance promoted the xenograft growth of T24 cells, whereas inhibition of K-Gly by beta-alanine delayed the xenograft growth in tumor cells, thereby validating the oncogenic role of K-Gly in the development of UC (Fig. 8J). Taken together, we found that upregulated expression levels of GARS promoted the UC progression through enhancing the pentose phosphate pathway by inhibiting activities of PGK1 and PKM2.