Integrative proteomics in prostate cancer uncovers robustness against genomic and transcriptomic aberrations during disease progression

ARTICLE Integrative proteomics in prostate cancer uncovers robustness against genomic and transcriptomic aberrations during disease progression Leena Latonen1,2, Ebrahim Afyounian 1, Antti Jylhä3, Janika Nättinen3, Ulla Aapola3, Matti Annala 1, Kati K. Kivinummi1, Teuvo T.L. Tammela4, Roger W. Beuerman3,5,6,7,8, Hannu Uusitalo3,9, Matti Nykter1,10 & Tapio Visakorpi1,2 To understand functional consequences of genetic and transcriptional aberrations in prostate cancer, the proteomic changes during disease formation and progression need to be revealed. Here we report high-throughput mass spectrometry on clinical tissue samples of benign prostatic hyperplasia (BPH), untreated primary prostate cancer (PC) and castration resistant prostate cancer (CRPC). Each sample group shows a distinct protein profile. By integrative analysis we show that, especially in CRPC, gene copy number, DNA methylation, and RNA expression levels do not reliably predict proteomic changes. Instead, we uncover previously unrecognized molecular and pathway events, for example, several miRNA target correlations present at protein but not at mRNA level. Notably, we identify two metabolic shifts in the citric acid cycle (TCA cycle) during prostate cancer development and progression. Our proteogenomic analysis uncovers robustness against genomic and transcriptomic aberrations during prostate cancer progression, and significantly extends understanding of prostate cancer disease mechanisms. DOI: 10.1038/s41467-018-03573-6 OPEN 1Prostate Cancer Research Center, Faculty of Medicine and Life Sciences and BioMediTech Institute, University of Tampere, Tampere 33014, Finland. 2FimLab Laboratories, Tampere University Hospital, Tampere 33101, Finland. 3Department of Ophthalmology, Faculty of Medicine and Life Sciences, University of Tampere, Tampere 33014, Finland. 4Department of Urology, University of Tampere and Tampere University Hospital, Tampere 33521, Finland. 5Singapore Eye Research Institute, Singapore 169856, Singapore. 6Duke-NUS Neuroscience, Singapore 169857, Singapore. 7Duke-NUS Medical School Ophthalmology and Visual Sciences Academic Clinical Program, Singapore 169857, Singapore. 8Ophthalmology, Yong Loo Lin Medical School, National University of Singapore, Singapore 119228, Singapore. 9Tays Eye Centre, Tampere University Hospital, Tampere 33521, Finland. 10 Science Center, Tampere University Hospital, Tampere 33521, Finland. These authors contributed equally: Ebrahim Afyounian, Antti Jylhä, Janika Nättinen. Correspondence and requests for materials should be addressed to M.N. (email: matti.nykter@uta.fi) or to T.V. (email: tapio.visakorpi@uta.fi) NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications 1 1234567890():,; Prostate cancer is the most common male malignancy in Western countries, and the second most common cancer among men overall1. Currently, no curative treatment exists for castration resistant prostate cancer (CRPC)2. To understand the etiology of the disease and to find more specific drug targets, the driver mutations and expressional changes in prostate cancer have been examined through extensive genomic and transcriptomic characterization3–7. Although significant insight has been gained through these efforts, it is clear that not all molecular alterations influencing the tumor outcome can be captured through these approaches. Proteins are regulated at multiple levels, and their expression is not always reflecting the levels of mRNA8,9. Thus, a comprehensive understanding of the molecular events in cancer require thorough investigation of the proteome10. Recent developments in mass spectrometric methods11–13 have enabled high throughput analysis of clinical patient samples, and the first integrative studies involving large scale, mass spectrometry-based proteomics of human cancer have recently been published14–16. For prostate cancer, recent proteomic advancements have included high scale, mass spectrometry-based studies performed in diagnostic body fluids17,18, as well as primary tumors19 and the tumor microenvironment20. So far, the only integrative proteogenomic analysis of clinical prostate cancer involved genomic and transcriptomic data of CRPC combined with phosphoproteomic analysis21. Despite the merits of this study in interrogating the active signaling pathways in CRPC, the large-scale proteomic view of PC and CRPC, and reflections of them to the disease progression are still lacking. Here, we provide the first integrative view on human prostate cancer with the proteome of clinical patient samples of benign prostatic hyperplasia (BPH), untreated primary prostate cancer (PC) and locally recurrent CRPC. Our analysis adds a new level to the current knowledge of prostate cancer development and 229575 153 a c b d Relative protein expression 10 BPH_335 BPH_337 BPH_456 BPH_656 BPH_664 BPH_671 BPH_677 BPH_689 BPH_718 BPH_719 PC_5934 PC_8131 PC_12517 PC_15420 PC_15760 PC_18307 PC_9324 PC_14670 PC_17447 PC_470 PC_10286 PC_15194 PC_20873 PC_4786 PC_4906 PC_4980 PC_17163 CRPC_278 CRPC_305 CRPC_348 CRPC_435 CRPC_489 CRPC_530 CRPC_531 CRPC_539 CRPC_541 CRPC_543 CRPC_697 20 30 Expressed proteins (n=3394) Differentially expressed proteins PC vs BPH (n=728) Differentially expressed proteins CRPC vs PC (n=382) BPH PC BPH PC PC CRPC CRPC PC vs BPH CRPC vs PC Fig. 1 Proteomic analysis reveals distinct protein expression patterns in PC and CRPC. aHeat map of all protein expressions identified and quantified by mass spectrometry in the proteomic analysis of BPH and prostate cancer samples (PC and CRPC). Each column of heat map represents a patient sample and each row represents a specific protein (n=3394). bVenn diagram showing the numbers of differentially expressed proteins in PC vs BPH and CRPC vs PC comparisons. Only a minority of the differentially expressed proteins overlap between the comparisons. c,dHeat maps of the differentially expressed proteins in bshow clearly distinctive patterns of protein expression between disease groups. PC compared to BPH samples (n=728) is shown in c, and CRPC compared to PC samples (n=382) is shown in d. Color key of relative expression in aapplies also to cand d ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 2NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications progression by identifying several molecular and pathway events not previously described based on transcriptomic data. Results Mass spectrometric analysis of proteomic profiles. Samples of 10 BPH, 17 untreated PC (Supplementary Table 1), and 11 CRPC (Supplementary Table 2) were analyzed. The CRPC samples came from patients that had been treated either by castration and/or antiandrogens and experienced urethral obstruction (ie. local recurrence) during the treatment. With sequential window acquisition of all theoretical fragment ion spectra mass spectrometry (SWATH-MS), we identified a total of 213,979 peptides, corresponding to 1,753,161 identified spectra in an assembly of 4601 protein groups using false discovery rate of 1%. Protein and peptide quantification data can be found from Supplementary Data 1. From this library, 3394 proteins had distinct peptides sequences with matching spectras to SWATH-MS analysis and were quantified in all samples (Supplementary Data 2). The SWATH-MS data was reproducible with mean intraclass correlation (ICC) coefficient of 0.98 between technical replicate MS analyses. Permutation tests (Spearman correlation) showed that 98.6% of the technical replicate MS analyses had a p-value < 0.05, demonstrating excellent quality. The represented protein classes (PANTHER protein class) and gene ontology groups (GO; molecular functions, cellular components, and biological processes) are shown in Supplementary Fig. 1a. The distribution of the proteins into different protein classes was largely according to expected as compared to Homo sapiens reference list (Supplementary Fig. 1a,b). The major overrepresented groups included the highly abundant nucleic acid binding (mainly RNA binding) and ribosomal proteins, oxidoreductases, and hydrolases. The major underrepresented groups were transcription factors and receptors, including immunoglobulins, consistent with the cell type-dependent expression of especially the latter group. Expression profiles of the identified proteins in the prostate tissue samples are shown in Fig. 1a. We wanted to assess changes occurring at the protein level during prostate cancer development and progression. As a model for benign tissue, we used BPH samples, against which primary PC samples were compared to identify early cancerous events. To identify events related to cancer progression and castration resistance, CRPC samples were compared to PC samples. We identified 728 proteins in PC vs BPH and 382 proteins in CRPC vs PC to be differentially expressed (Wilcoxon rank sum test with Benjamini & Hochberg adjustment p-value < 0.05 and median ratio (fold change) >1.5) between the comparison groups (Fig. 1b). While the overall protein classes of the differentially expressed proteins and their distribution to groups of molecular function, cellular component, and biological process were similar between PC vs BPH and CRPC vs PC comparisons (assessed by Panther analysis; data not shown), only a subset (n=153) of the differentially expressed proteins were common between the comparison groups (Fig. 1b). The expression profiles of the differentially expressed proteins clearly distinguished between the patient sample groups, as shown in Fig. 1c (PC compared to BPH) and Fig. 1d (CRPC compared to PC). These results show that the proteomic profile of prostate cancer is significantly altered during the course of the disease. Correlations of copy number and methylation with proteomics. We have previously performed whole genome sequencing for copy number analysis, DNA methylation sequencing, and whole transcriptome sequencing to majority of the samples used in the proteomic analysis described here (Supplementary Table 3) 7,22. We compared the correlation between gene copy number, and mRNA or protein expression levels between the common samples. While at the transcriptome level, the mRNA expression and copy number have an increased overall correlation in the CRPC samples compared to PC samples (Fig. 2a, Supplementary Fig. 2a), a similar global correlation change with gene copy number is not present at the proteomic level. Next, we compared the correlation between DNA methylation at differentially methylated regions (DMRs), and mRNA or protein expression levels in the same samples. Similarly as with the copy number data, the increased negative correlation between DNA methylation and mRNA expression at a global level in the CRPC samples compared to PC samples is not detected at the level of the proteome (Fig. 2b, Supplementary Fig. 2b). These results suggest that, a Correlation estimate Count Global effect of gene dosage on expression mRNA Protein mRNA / PC mRNA / CRPC Protein / PC Protein / CRPC 16 14 12 10 8 6 4 2 0–0.2 –0.1 0.0 0.1 0.2 b Count Correlation estimate mRNA Protein mRNA / PC mRNA / CRPC Protein / PC Protein / CRPC 16 14 12 10 8 6 4 2 0–0.2 –0.1 0.0 0.1 0.2 Global effect of DNA methylation on expression Fig. 2 Global expression changes associated with gene copy number and DNA methylation are visible at the transcriptomic but not at proteomic level. a Correlation distributions of mRNA and protein expression with gene copy number. Lines represent effects in all analyzed genes in all samples, and show that gene dosage has higher positive correlation with mRNA expression than protein expression in prostate cancer on a global scale. Symbols on the bottom of the graph represent individual samples, and show how most of the CRPC samples have a higher positive correlation compared to PC samples at the mRNA level, as at the protein level no such difference between the disease groups is observed. bCorrelation distributions of mRNA and protein expression with DNA methylation. Lines represent effects in all analyzed genes in all samples, and show that DNA methylation has higher negative correlation with mRNA expression than protein expression in prostate cancer on a global scale. Symbols on the bottom of the graph represent individual samples, and show how most of the CRPC samples have a decreased correlation compared to PC samples at the mRNA level, as at the protein level no such difference between the disease groups is observed NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 ARTICLE NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications 3 on a global level, the genomic and epigenomic events that influence mRNA levels are not directly translated to protein expression in prostate cancer. The effect of altered methylation in prostate cancer on selected genes is, on the other hand, evident also at the proteomics data. There were 140 genes, which were differentially expressed either at mRNA or protein level, with a DMR close by (<10 kb). Within this group, there were several examples of methylation correlating with, and thus likely affecting, mRNA and protein expression. For example, the previously described increased DMR methylation in prostate cancer on genes ALDH1A2,GSTP1,GPX3, and CYB5R2 correlate with decreased expression of their mRNA and protein according to our data (Supplementary Fig. 3). We further identified increased promoter DMR methylation in prostate cancer correlating with decreased expression of mRNA and protein expression also on FBXO2,TGFB1I1, and TNS1 (Supplementary Fig. 4). Increased gene body methylation in prostate cancer correlating with decreased expression of mRNA and protein expression was identified on GNAO1,LGALS1,TNS1, and PPAP2B (Supplementary Fig. 5). Decreased methylation significantly correlating with increased expression was identified for ENO1, SOAT1, RPS2, and TACSTD2 (Supplementary Fig. 6). Altered DMR methylation found in prostate cancer samples identifies also genes that are less likely to affect directly the outcome of the cancer cells. This is due to either their expression primarily in stromal cells (e.g., CSRP1, CA3) or the fact that, a Count 70 60 50 40 30 20 10 0 Correlation estimate 1.0 –0.5 0.0 0.5 1.0 Gene-wise correlation of mRNA and protein expression µ=0.15 positive correlation 73.32% b Density 15 5 20 10 0 Correlation estimate 0.25 0.30 0.35 0.40 0.45 0.50 Sample-wise correlation of mRNA and protein expression All samples µ=0.37 BPH µ=0.37 PC µ=0.39 CRPC µ=0.35 All samples BPH PC CRPC PC to BPH CRPC to PC 555 269 156 354163 c DE mRNA DE mRNADE protein DE protein 360352 122 dmiRNA analysis 40 eTargets of differentially expressed miRNAs miRNA - target in transcriptome miRNA - target in proteome miRNA - target in both miRNA target Y X 22 21 20 19 18 17 16 15 14 13 12 11 10 98 7 6 5 4 3 2 1 Targets in transcriptome Targets in proteome Fig. 3 Transcriptomic and proteomic data show distinct patterns of expression at the RNA and protein level in prostate cancer. aCorrelation between mRNA and protein expression of individual genes. The graph shows correlations of all genes identified in proteomic analysis in all samples used in this study. Most of the genes (>73%) show positive correlation between expression of their mRNA and protein. μis the mean of the correlations. bDisease group-wise correlation between mRNA and protein expression of the genes identified in the proteomic analysis shows that in CRPC there is a decreased correlation between mRNA–protein expression pairs compared to primary PC. Compared to all samples (black line) and BPH (green line), the PC samples (blue line) have a higher correlation between their mRNA-protein expression pairs, while CRPC samples (red line) have a lower correlation. µis the mean of the correlations. cVenn diagram showing the numbers differentially expressed (DE) genes in PC vs BPH and CRPC vs PC comparisons identified based on mRNA or protein expression. The numbers of overlapping genes show that only a minority of the differentially expressed genes show expression changes in both mRNA and protein levels. dVenn diagram showing the numbers of genes that are negatively correlating with a targeting miRNA based on their expression at the mRNA or protein level. Only a minority of the miRNA targets are identified both at the mRNA and protein level, indicating that correlations at the protein level help to identify mostly a different pool of miRNA targets than correlations at the mRNA level. eCircos plot depicting genomic locations of miRNAs and their targets that are both negatively correlating at expression, as well as differentially expressed during prostate cancer progression (CRPC vs PC samples). Outer ring indicates chromosomes and cytobands, with chromosome numbers in the gray circles. Each line in the center maps a prostate cancer-related miRNA-target pair indicated through transcriptomic (blue lines), proteomic (red lines), or both (black lines) analyses. The blue circles mark the genomic location of the miRNAs, and the solid blue dots mark the targets ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 4NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications despite mRNA expression being affected, the expression level of the protein is not being affected by the differential methylation of the gene (e.g., CLU, CNTN1) (Supplementary Fig. 7). Interestingly, we also identified genes whose differentially increased methylation significantly correlated with increased expression in mRNA and/or protein level (GMDS, MCCC2, MIA3, and PYCR1) (Supplementary Fig. 8). Impact of mutations on protein expression. We identified amino acid altering mutations in expressed genes from the RNAsequencing data of the samples used in this study, and validated these from the DNA using targeted sequencing (Supplementary Table 4). For all somatic and germline variants, we evaluated the impact of the variant to mRNA and protein expression as described earlier14. While somatic mutations had a statistically significant impact to mRNA levels in relation to germline variants (Fisher’s exact test, p-value =0.0055, Supplementary Fig. 9), we observed no impact on protein expression levels between somatic and germline mutations or in relation to null distribution estimated from unmutated genes (Supplementary Fig. 9). To screen for proteins with potential involvement in mutation accrual during prostate cancer development and progression, we assessed correlations of protein expression in relation to mutation ab 4049 50 0 5 10 15 20 25 30 35 40 Other Cell cycle Plasma membrane transport Protein degradation Stress response DNA metabolism and repair Inflammation/immunity Translation Vesicle transport Signaling Cytoskeleton, attachment, migration, invasion Metabolism PC vs BPH CRPC vs PC No. of pathways cd Significance of pathway regulation (log p-value) Signaling pathways Translation activating and growth promoting RXR-related Prostate cancer related Cytoskeleton, migration, and invasion related GTPase signaling EIF2 signaling Regulation of eIF4 and p70S6K signaling mTOR signaling p70S6K signaling FXR/RXR activation LXR/RXR activation TR/RXR activation Androgen signaling PI3K/AKT signaling Actin cytoskeleton signaling Integrin signaling ILK signaling FAK signaling Cdc42 signaling Signaling by Rho family GTPases RhoA signaling RhoGDI signaling Rac signaling RAN signaling Metabolic pathways PC vs BPH CRPC vs PC PC vs BPH CRPC vs PC Protein 0.0 7.0 14.0 mRNA e DNA repair Cell cycle Telomere extension by telomerase DNA DSB repair by NHEJ BER pathway CHK proteins in checkpoint control Cell cycle regulation by BTG family Control of chromosomal replication G2/M DNA damage checkpoint Cell cycle and DNA repair PC vs BPH CRPC vs PC PC vs BPH CRPC vs PC Protein mRNA TCA cycle Mitochondrial dysfunction Ketogenesis Acetyl-CoA biosynthesis Fatty acid oxidation Glycolysis Glycogen degradation Oxidative ethanol degradation PC vs BPH CRPC vs PC PC vs BPH CRPC vs PC Protein mRNA CRPC vs PC PC vs BPH NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 ARTICLE NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications 5 burden of the samples, including the somatic point mutations, copy number alterations, and genomic rearrangements. The two proteins, expression of which correlated best with point mutation burden, were mitochondrial antioxidant regulator PRDX3 (peroxiredoxin 3) and CAD (carbamoyl-phosphate synthetase 2) functioning in de novo synthesis of pyrimidine nucleotides (Supplementary Table 5). For copy number alterations and genomic rearrangements, the best correlating proteins had functions mostly in mitochondria and cytoskeleton. Notably, the strongest negative correlations with the number of rearrangements were with expression of two Talin proteins (TLN2 and TLN1)(Supplementary Table 5). Comparison of expression profiles at RNA and protein levels. The expression levels of most of the proteins identified in our dataset were positively correlated with the expression level of their mRNA, as expected (Fig. 3a). However, when comparing the sample groups, we found that in CRPC, the correlation between individual mRNA-protein pairs was lower in general than in BPH or PC samples (Fig. 3b). We next tested whether similar genes are identified as differentially expressed based on both mRNA and protein expression data. In both PC vs BPH and CRPC vs PC comparisons, only a fraction of the differentially expressed genes were common between the identifications based on transcriptomic and proteomic data, the difference being larger in CRPC vs PC comparison (Fig. 3c). Of the commonly identified genes, 97 and 95% of the differential expressions detected were oriented to the same direction (up or downregulated) in both PC vs BPH and CRPC vs PC comparisons based on mRNA and protein expression data, respectively. According to these results, proteomic and transcriptional data help identify largely different events during prostate cancer development and progression. Next, we integrated small RNA sequencing data for PC and CRPC samples common between the proteomics and mRNA expression data. MicroRNAs regulate gene expression by binding to mRNA molecules and preventing translation, which leads to decreased target protein expression. miRNA binding to the target can induce degradation of the mRNA, however, also stabilization of the target mRNA has been reported23. To study how much of the observed gene expression in prostate cancer is potentially connected to regulation by miRNAs, we studied the pool of differentially expressed genes and their correlating miRNAs. As one miRNA can have several target mRNAs, and one mRNA can be targeted by several miRNAs, we considered individual miRNA-target pairs based on both transcriptome and proteome data, and the predicted or verified miRNA target annotations. Negative correlations between miRNA and differentially expressed targeted mRNAs in CRPC vs PC samples revealed 30 miRNAs and 205 individual miRNA-target pairs (Supplementary Table 6). Of these, 9 miRNAs were also differentially expressed (Supplementary Table 6). For 34 of the miRNA-target pairs, negative correlation was also found between miRNA and protein expression of the target, indicating a functional impact of miRNA regulation for these particular targets (Supplementary Table 6). To look for the miRNA targets for which the miRNA does not induce mRNA degradation, but effect primarily through inhibition of translation, we searched for negative correlations between miRNA and differentially expressed targeted proteins in the proteome of CRPC vs PC samples. This analysis identified additional 49 miRNAs and 268 individual miRNA-target pairs (Supplementary Table 7). Of these, 8 miRNAs were also differentially expressed (Supplementary Table 7). This pool of miRNA-target pairs represents a resource of novel associations in prostate cancer that have not been visible through previous transcriptome analyses. To understand the capacity that miRNAs have in regulating prostate cancer progression, we assessed the number of differentially expressed miRNAs and the fraction of the proteome they are collectively able to regulate. There were 95 miRNAs that were differentially expressed between CRPC and PC samples. Assuming negative correlation between a miRNA and its database-predicted or verified target either at the mRNA or protein expression level, the differentially expressed miRNAs in our dataset had the potential to target 16% of the genes in the study. There were 474 and 482 genes according to mRNA and protein expression, respectively, targeted by and negatively correlating with at least one regulating miRNA (Fig. 3d, Supplementary Data 3-4). Of these, only 122 genes were commonly identified (Supplementary Table 8). To look for the miRNA targets which most likely affect prostate cancer progression, we assessed the fraction of the miRNA-regulated genes that were differentially expressed. Of the above miRNA-regulated targets identified based on mRNA expression, 24% (n=115) were differentially expressed between CRPC and PC samples at the mRNA level (Supplementary Fig. 10a, Supplementary Table 9). Similarly, of the regulatory targets identified based on the proteomics data, 45% (n=218) were differentially expressed at the protein level (Supplementary Fig. 10b, Supplementary Table 10). There were 24 genes common between these groups (21% or 11% of the genes identified based on mRNA and protein expression, respectively). A genomic map of the differentially expressed miRNAs and their differentially expressed targets in CRPC vs PC samples based on transcriptomics and proteomics is shown in Fig. 3e. Collectively, these data indicate that by studying the miRNA-target correlations at the protein expression level we were able to identify a significant number of potential regulatory events, which were not identified based on mRNA expression data of clinical prostate cancer samples. Fig. 4 Proteomic analysis identifies novel pathways as regulated in PC and CRPC. aVenn diagram showing numbers of differentially regulated pathways according to Ingenuity Pathway Analysis in PC vs BPH and CRPC vs PC comparisons. Despite partial overlap, the different disease states have a significant number of pathways specifically regulated. bDifferentially regulated pathways in aaccording to pathway types. Metabolism is the largest group in both comparisons, with roughly a similar number of pathways differentially regulated. Numbers of most of the other pathway types that are differentially regulated between the disease states vary. c–eExamples of signaling pathways found to be differentially regulated according to proteomics (protein) or transcriptomics (mRNA) data in PC vs BPH and CRPC vs PC comparisons. cExamples of signaling pathways groups identified as regulated according to proteomic data. Especially translation activating, growth promoting pathways are identified as regulated solely based on proteomic data. RXR-related pathways are identified better by proteomics than transcriptomics to be regulated in PC. Pathways related to cytoskeleton, migration, and invasion, as well as GTPase signaling pathways are identified to be regulated in PC solely by proteomics, although in CRPC they are better identified as regulated by transcriptomics. dMetabolic pathways differentially identified as regulated based on proteomic and transcriptomic data include pathways identified as regulated in both PC and CRPC solely based on proteomics (TCA cycle, mitochondrial dysfunction, ketogenesis, acetyl-CoA biosynthesis), and pathways that are equally identified by proteomics and transcriptomics, but are specific for PC (fatty acid oxidation, glycolysis) or CRPC (glycogen degradation, oxidative ethanol degradation). eWhile DNA repair pathways regulated in PC and CRPC were identified based on proteomics only, the regulated cell cycle pathways were altered in CRPC and identified based on either proteomic or transcriptomic data. The color key below panel capplies to panels c,d, and e ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 6NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications To validate our analysis for miRNA targets detectable both at the mRNA or the protein level, we transfected PC-3 prostate cancer cells with pre-miRNA constructs and assessed the mRNA and protein levels of predicted targets. We selected two representative miRNAs that were differentially expressed to opposite directions during prostate cancer progression, namely miR-22 as downregulated and miR-493 as upregulated in CRPC compared to PC, and verified their successful transfection by TaqMan RT-qPCR (Supplementary Fig. 11a). As positive controls for miRNA targeting at the mRNA level, we performed RT-pPCR on two predicted targets of miR-493 that were identified as negatively correlated based on our analysis at the target transcript level (Supplementary Data 3). Supplementary Fig. 11b shows that, as expected, the mRNA levels of ENDOD1 and GOLM1 are significantly decreased by miR-493 expression. Further, the negatively correlating miRNA-target pairs identified only in the proteomic analysis show decreased protein expression in MS/MS quantification, but no decrease in mRNA levels in RT-qPCR assay, as shown for miRNA-target pairs miR-22—KHRSP1 and miR-493—DNML1 (Supplementary Fig. 11c and d, respectively). These results confirm that our miRNA-target analyses based on the proteomics data have identified miRNA targets that are not identified at the mRNA level. Proteomic analysis reveals novel regulated pathways. To test whether proteomics reveal pathway alterations in prostate cancer that have not previously been found by interrogation of mRNA expression changes, we next performed pathway analysis comparison between mRNA and protein expression data from the same samples. Supplementary Fig. 12a shows that roughly similar numbers of pathways were found significantly regulated based on proteomics and RNA expression data when comparing PC to BPH, and slightly more based on proteomics in CRPC to PC comparison. However, only a minority (16–26%) of the pathways found in each comparison category were common between RNA and proteomics data. These results show that proteomic data is able to reveal pathway regulations not visible at the RNA expression level, especially when comparing CRPC to PC. We further analyzed which signaling pathways were deregulated during prostate cancer development and progression at the proteomic level. Comparing PC samples to BPH, 99 pathways were found regulated according to Ingenuity Pathway Analysis, while 90 pathways were regulated in CRPC vs PC (Fig. 4a, Supplementary Table 11). Fifty pathways were common between these comparisons. The pathway categories were similar in both comparisons, with metabolic pathways being the most prominent (Fig. 4b). In PC vs BPH, cytoskeleton, attachment, and motilityrelated pathways were the second largest group, while in CRPC vs PC it was the signaling pathways. Exclusively in PC vs BPH, there were protein degradation pathways found significantly regulated, while in CRPC vs PC, certain cell cycle pathways were significantly regulated. The pathways common between the PC vs BPH and CRPC vs PC comparisons (Supplementary Table 11) included mostly metabolic pathways, as well as cytoskeleton, attachment and motility-related pathways (62% of the common pathways). It is noteworthy that all significantly regulated DNA metabolism and repair pathways, and most of the vesicle transport pathways, were common between the comparison groups. In contrast, only a few of the regulated signaling pathways were common between the comparison groups, all of which represented Rho GTPase signaling pathways (Supplementary Table 11). In PC vs BPH, the top significantly regulated pathways included EIF2, eIF4 and p70S6K signaling, as well as FXR/RXR and LXR/RXR activation (Supplementary Table 11, Fig. 4c). While the former pathways promote growthand survival through alterations in levels of several translation initiation factors and ribosomal proteins, the latter signal to metabolic pathways regulated by farnesoid X receptor (FXR), liver X receptor (LXR), and retinoid X receptor (RXR). When comparing CRPC to PC samples, the most significantly altered pathways during progression of prostate cancer include ILK signaling and glucose metabolism-related pathways (Fig. 4c, d, Supplementary Table 11). Next, we wanted to further understand the differences in pathway regulation at mRNA and protein levels. Despite being largely different pathways, the biological functions of the pathways most often found by either RNA expression or proteomics were similar, with metabolic, signaling, and cytoskeleton and cell movement-related pathways being the most common (Fig. 4c, d, Supplementary Fig. 12b, Supplementary Fig. 13a,b). Examples of differentially identified signaling pathways are shown in Fig. 4c. Most prominently, the translationactivating and growth-promoting EIF, p70S6, and mTOR signaling, and cytoskeleton-related signaling in PC vs BPH were found solely based on proteomics data. RXR-related signaling in PC vs BPH, as well as several cytoskeleton-related signaling pathways in CRPC vs PC, were found by both transcriptomics and proteomics similarly. Interestingly, GTPase signaling was significantly regulated in PC vs BPH by proteomics and in CRPC vs PC by transcriptomics. The group of metabolic pathways that was regulated in both PC vs BPH and CRPC vs PC comparisons was extensive (Supplementary Fig. 12b). Despite the relatively low overlap in individual pathways between the comparisons (between disease groups, and between proteomic and transcriptomic data), all the analyses identified pathways from the major groups of energy, amino acid, and lipid metabolism (Supplementary Table 11; examplesshowninFig.4d, Supplementary Fig. 13b). It is noteworthy that the mitochondria-related metabolic pathways and ketogenesis were identified as differentially regulated only by proteomics, while e.g., glycolytic and glycogen degradationrelated pathways were identified by both transcriptomics and proteomics (Fig. 4d). One of the most prominent group of metabolic pathways in prostate cancer were amino acid metabolic pathways (Supplementary Fig. 13b). Interestingly, while different cell cycle regulatory pathways were found regulatedbytranscriptomicandproteomicdata,DNArepair pathways were found solely by proteomics analysis (Fig. 4e). Other interesting groups of differentially identified pathways based on RNA and protein expression were vesicular trafficrelated and protein degradation pathways (Supplementary Fig. 13c,d). Table 1 TCA cycle proteins with altered expression levels in prostate cancer Symbol Entrez gene name PC vs BPH CRPC vs PC ACO2 aconitase 2 3.141 0.472 CS citrate synthase 1.705 n.s. FH fumarate hydratase 1.598 n.s. IDH3A isocitrate dehydrogenase 3 (NAD (+)) alpha n.s. 0.653 MDH2 malate dehydrogenase 2 2.167 1.912 OGDH oxoglutarate dehydrogenase 1.653 0.608 SUCLA2 succinate-CoA ligase ADP-forming beta subunit 1.909 n.s. SUCLG1 succinate-CoA ligase alpha subunit 2.091 0.469 Fold changes in protein expression are shown. n.s., not significantly altered NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 ARTICLE NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications 7 Changes in TCA during prostate cancer evolution. Based on our analysis, metabolic changes are prominent during both development and progression of prostate cancer. One of the most interesting pathways identified by our proteomic data was the tricarboxylic acid cycle (TCA; also referred to as the citric acid cycle, or the Krebs cycle), which was altered in both PC vs BPH and CRPC vs PC comparisons. This pathway was not found regulated by RNA expression data, suggesting changes taking place primarily at the protein level. Furthermore, although alterations in certain enzyme activities in TCA have previously been shown to occur during prostate cancer development24, our proteomics results indicated a previously undescribed, two-step modulation of the TCA cycle. The TCA pathway proteins that were considered regulated by the pathway analysis were mostly Coenzyme A NAD+ Succinyl-CoA CO2 NADH Ubiquinone Ubiquinol Fumarate Succinate dehydrogenase NAD+ NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH CO2 2-oxoglutarate Isocitrate dehydrogenase NAD+ H+ NADH Malate dehydrogenase Oxaloacetate Acetyl-CoA H2O H+ Citrate Coenzyme A Citrate synthase Cis-aconitate H2O Aconitate hydratase H2O D-threoisocitrate Aconitate hydratase (S)-malate H2O Fumarate hydratase Succinate Coenzyme A ATP ADP Phosphate Succinate-CoA ligase 2-ketoglutarate dehydrogenase complex aPC vs BPH bACO2 8 6 4 2 0 BPH PC CRPC *** *** Expression in mass spectrometric analysis/relative units c 0 1 2 3 4 5 LNCaP BPH PC CRPC ACO2 panactin WB: WB: ACO2 expression in mass spectrometric analysis/relative units 719 470 697 17163 689 4786 530 718 d PC n=197 CRPC n=84 ACO2 0 1 2 3 100 80 60 40 20 0 Relative % of samples in the staining category p<0.0001 Coenzyme A NAD+ Succinyl-CoA CO2 NADH Ubiquinone Ubiquinol Fumarate Succinate dehydrogenase NAD+ NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH NADH CO2 2-oxoglutarate Isocitrate dehydrogenase NAD+ H+ NADH Malate dehydrogenase Oxaloacetate Acetyl-CoA H2O H+ Citrate Coenzyme A Citrate synthase Cis-aconitate H2O Aconitate hydratase H2O D-threoisocitrate Aconitate hydratase (S)-malate H2O Fumarate hydratase Succinate Coenzyme A ATP ADP Phosphate Succinate-CoA ligase 2-ketoglutarate dehydrogenase complex CRPC vs PC MDH2 8 6 4 2 0 BPH PC CRPC *** *** Expression in mass spectrometric analysis/relative units MDH2 WB: WB: panactin 0 2 4 6 8 LNCaP BPH PC CRPC MDH2 expression in mass spectrometric analysis/relative units 719 470 697 17163 689 4786 530 718 PC n=153 CRPC n=75 MDH2 100 80 60 40 20 0 Relative % of samples in the staining category p<0.0001 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 8NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications altered to opposite directions in PC vs BPH and CRPC vs PC comparisons: upregulated in PC vs BPH, and downregulated in CRPC vs PC (Table 1, Fig. 5a). An exception was malate dehydrogenase 2 (MDH2), levels of which continued to increase in CRPC (Table 1, Fig. 5a). Comparison of protein and RNA expression of TCA genes8,23 in all three groups of samples revealed that the TCA proteins are divided into three classes: (1) proteins whose mRNA and protein expression go hand in hand indicating primary regulation by gene expression (CS, FH, IDH3A, IDH2, and SUCLG2), (2) proteins, whose protein levels are not changed (IDH3B, IDH3G), and (3) proteins, that exhibit regulation at the protein level not correlating with mRNA (ACO2, MDH2, OGDH, SUCLA2, and SUCLG1) (Supplementary Fig. 14). From the latter group of proteins, ACO2, OGDH, SUCLA2, and SUCLG1 were all upregulated at the protein level, but not at the mRNA level, in PC vs BPH, while being downregulated in CRPC vs PC either at the mRNA or protein level. Increase in MDH2 protein expression in PC vs BPH did correlate with an increase in mRNA levels, but the increase in CRPC vs PC did not, suggesting posttranslational regulation (Supplementary Fig. 14). To study more closely the events identified in the TCA cycle, and to validate the results of the proteomics data, we selected two TCA proteins showing significant but different alterations at their protein expression between the prostate cancer sample groups to study further. As a representative of the most common alteration pattern we chose aconitase 2 (ACO2) which showed statistically highly significant (p< 0.001, Mann–Whitney test) upregulation of the protein in PC vs BPH, as well as statistically highly significant (p< 0.001, Mann–Whitney test) downregulation in CRPC vs PC (Fig. 5b). As a second protein we chose MDH2 exhibiting the deviant behavior amongst the TCA proteins, as it was upregulated statistically significantly both in PC vs BPH (p< 0.001; Mann–Whitney test) and further upregulated in CRPC vs PC (p< 0.05; Mann–Whitney test) (Fig. 5b). We performed western blotting on these proteins with representative samples of BPH, PC, and CRPC used in the proteomic analysis, and found similar changes than by mass spectrometry (Fig. 5c, Supplementary Fig. 15 and 16), validating the mass spectrometry detection and analysis results. We performed further validation on the differential regulation of these proteins during prostate cancer progression by immunohistochemical stainings on larger sample sets of clinical PC and CRPC. Grading of the immunohistochemical staining intensity (example staining intensities of grades 0–3 displayed in Supplementary Fig. 17) showed that relative percentage of samples with no or low staining intensities (0–1) of ACO2 increased in CRPC vs PC (Fig. 5d), indicating that the relative levels of ACO2 decreased in CRPC. On the other hand, the relative percentage of samples with higher staining intensities (2–3) of MDH2 increased in CRPC vs PC (Fig. 5d), indicating that the relative levels of MDH2 increased in CRPC. These results confirm the mass spectrometry results and show that the TCA cycle proteins ACO2 and MDH2 are differentially regulated at the protein level during prostate cancer progression. We further assessed potential mechanisms that could explain the distinct regulation of MDH2. We found that two miRNAs predicted to target MDH2, namely miR-22 and miR-205, were identified as differentially expressed in our analysis and were negatively correlating with MDH2 protein (Supplementary Data 4) but not mRNA (Supplementary Data 3) levels in the large scale datasets. We transfected PC-3 prostate cancer cells with these miRNAs, and verified the transfection efficiency with TaqMan RT-qPCR analysis (Supplementary Fig. 18a). We detected no significant alterations at MDH2 mRNA levels in RT-qPCR analysis upon elevated expression of the miRNAs (Supplementary Fig. 18b). In contrast, luciferase assay showed statistically significant decrease in reporter production from a MDH2 3′-UTR construct by both miR-22 and miR-205 overexpression (Supplementary Fig. 18c), indicating that these miRNAs are able to directly target MDH2 mRNA. Furthermore, MS/MS quantification showed a substantial decrease in MDH2 protein levels by both miR-22 and miR-205 expression (Supplementary Fig. 18c). These results validated the predictions of miR22 and miR-205 to directly target MDH2, and identified these miRNAs as prostate cancer-relevant, differentially expressed regulators of the TCA. Discussion We have provided the first extensive proteomic view of prostate cancer development and progression. With over 3000 individual proteins quantified in each of the BPH, PC and CRPC samples analyzed, we described the protein level alterations occurring in clinical prostate cancer, and found several previously undescribed biological events with important implications and potential for future studies. In addition, we provided novel views on the relationship of proteomic, genomic, and transcriptomic changes occurring during castration resistance. The comprehensive view obtained by our integrative analysis underlines the importance of protein level dissection of the molecular mechanisms supporting cancer growth and progression. Our results showed that neither the altered gene dosages, nor the global methylation changes were translated to the level of the proteome to the same extent as they influence the global RNA expression in CRPC. This suggests that, in the progressed stage, a large proportion of changes in gene copy number and differential DMR methylation are side products of the catastrophic state of cancer cell regulatory systems which are untranslated and thus, subsequently, left without a functional effect at the protein level. Yet, our data confirmed several previously identified regulatory DNA methylation events with associated expression changes occurring in prostate cancer. We also identified several previously Fig. 5 TCA cycle is differentially regulated during prostate cancer progression. aA schematic view of the TCA cycle protein expression changes in PC vs BPH and CRPC vs PC comparisons according to the Ingenuity Pathway Analysis. Differential expression of TCA enzymes (diamonds) are highlighted in green (downregulation) and red (upregulation). As mostly the same enzymes are involved in both PC and CRPC, the primary mode of expression change is upregulation in PC and downregulation in CRPC. bExamples of a typical (ACO2) and a unique (MDH2) TCA protein expression patterns as identified by mass spectrometry proteomics. ACO2 is upregulated in PC compared to BPH, and gets downregulated in CRPC compared to PC. MDH2 protein expression levels increase in PC compared to BPH, and continue to increase in CRPC. Boxplots show interquartiles with mean values, whiskers represent minimum and maximum values. ***p-value < 0.001 (Mann–Whitney test). cACO2 and MDH2 protein expression patterns verified in a subset of BPH, PC, and CRPC samples by western blotting. ACO2 and MDH2 protein expression according to the proteomic mass spectrometry analysis (upper panel bar graph) and in corresponding samples according to western blotting (WB; lower panels). Pan-actin is used as a loading control. dChange in ACO2 and MDH2 protein expression patterns during progression of prostate cancer verified by immunohistochemistry. Immunohistochemical analysis in clinical tumor samples of PC and CRPC show statistically significantly decreased ACO2 and increased MDH2 staining intensity in CRPC compared to PC and (Chi squared test; 0 = no staining, 1 =weak staining, 2 =intermediate staining, 3 =strong staining) NATURE COMMUNICATIONS | DOI: 10.1038/s41467-018-03573-6 ARTICLE NATURE COMMUNICATIONS | (2018) 9:1176 |DOI: 10.1038/s41467-018-03573-6 |www.nature.com/naturecommunications 9