Molecular Changes in Tissue Proteome during Prostate Cancer Development: Proof-of-Principle Investigation

(1) Background: Prostate cancer (PCa) is characterized by high heterogeneity. The aim of this study was to investigate molecular alterations underlying PCa development based on proteomics data. (2) Methods: Liquid chromatography coupled to tandem mass spectrometry was conducted for 22 fresh-frozen tissue specimens from patients with benign prostatic hyperplasia (BPH, n = 5) and PCa (n = 17). Mann Whitney test was used to define significant differences between the two groups. Association of protein abundance with PCa progression was evaluated using Spearman correlation, followed by verification through investigating the Prostate Cancer Transcriptome Atlas. Functional enrichment and interactome analysis were carried out using Metascape and String. (3) Results: Proteomics analysis identified 1433 proteins, including 145 proteins as differentially abundant between patients with PCa and BPH. In silico analysis revealed alterations in several pathways and hallmarks implicated in metabolism and signalling, represented by 67 proteins. Among the latter, 21 proteins were correlated with PCa progression at both the protein and mRNA levels. Interactome analysis of these 21 proteins predicted interactions between Myc proto-oncogene (MYC) targets, protein processing in the endoplasmic reticulum, and oxidative phosphorylation, with MYC targets having a central role. (4) Conclusions: Tissue proteomics allowed for characterization of proteins and pathways consistently affected during PCa development. Further validation of these findings is required.


Introduction
With more than 1.2 million new cases and almost 360,000 disease-related deaths among men in 2018, prostate cancer (PCa) has contributed to 13.5% of the diagnosed cancer cases and 6.7% of cancer-associated deaths in men [1]. The notable difference between incidence and mortality is related to the fact that significant proportion of prostate tumours are slow-growing, organ-confined tumours that pH 7.6. Subsequently, the samples were centrifuged at 16,000g for 10 min at room temperature and the supernatant was kept in clean tubes. Protein concentration was measured with Bradford assay. Protease inhibitors (Roche, Basel, Switzerland) were added at a final concentration of 3.6% v/v, and the samples were stored at −80 • C until further use. Ten micrograms of each sample were loaded onto sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE). Protein separation and tryptic digestion were performed as described in [12][13][14]. Tryptic digests were dried, resuspended in mobile phase A buffer (0.1% formic acid, pH 3) and processed with LC-MS/MS analysis, as previously described [13,14]. Briefly, the peptide mixture was first loaded into a Dionex Ultimate 3000 RSLS nano flow system (Dionex, Camberly, UK), including an integrated Dionex nano trap column (0.1 × 20 mm, 5 µm, C18). Two percent acetonitrile/0.1% formic acid was applied as a mobile phase, and the flow rate was set to 5 µL/min. Acclaim PepMap C18 nano column (75 µm × 50 cm, 2 µm 100 Å) was configured as an analytical column, with a flow rate of 300 nL/min. Peptides were eluted with a gradient of mobile phase A (0.1% formic acid in water) versus mobile phase B (0.1% formic acid in 80% acetonitrile) as follows: 1% B for 5 min rising to 5% B at 10 min and then to 25% B at 360 min and 65% B at 480 min. Electrospray ionisation was applied using a Proxeon nano spray source (positive ion mode) connected to an Orbitrap LTQ Velos (Thermo Finnigan, Bremen, Germany). The instrument was operated in MS/MS mode with the scanning range set at 380 to 2000 m/z. An ionization voltage of 2.6 kV and temperature of the capillary at 275 • C were applied, while the resolutions of ions were 60,000 and 7500 in MS1 and MS2 (for higher-energy collisional dissociation, HCD), respectively. The top 20 method was implemented for selection of the precursor ions from each scan, followed by fragmentation using HCD at 35% collision energy. Dynamic exclusion was applied with a repeat count of 1 and an exclusion duration time of 30 s.

MS Data Processing and Statistical Analysis
RAW mass spectrometry data were processed and analysed as follows [14][15][16]: Protein identification was conducted using Proteome Discoverer v1.4 (Thermo Finnigan, San Jose, CA, US) using the human SwissProt database and the SEQUEST search engine. The FASTA file was downloaded from the Uniprot database (https://www.uniprot.org/) on 20 June 2019 and included 20,431 reviewed entries. Only canonical sequences were considered. The following search parameters were applied: (a) precursor mass tolerance: 10 ppm and fragment mass tolerance: 0.05 Da; (b) full tryptic digestion; (c) maximum missed cleavage sites: 2; (d) static modifications: carbamidomethylation of cysteine; and (e) dynamic modifications: oxidation of methionine and proline, deamidation of asparagine and glutamine, carbamylation of lysine and N-terminal carbamidomethylation. Subsequently, individual datasets were exported at the peptide level using the following filters: (a) peptide confidence: high, medium and low; (b) peptide rank up to 5; (c) peptide grouping enabled and protein grouping disabled; and (d) ∆M ± 5 ppm. Data were further processed using a clustering approach that was described previously, with some minor modifications [15,16]: peptides from different proteomics analyses were grouped ("clustered") based on a predefined mass window of ±5 ppm and retention time of ±5%. To increase the validity of the reference sequence assignment to the cluster, only sequences that were determined with high confidence during Proteome Discoverer analysis (false discovery rate, FDR < 1%) were considered. Among them, the sequence with the highest sum of Xcorr across all samples was accepted as representative for the cluster. This resulted in the generation of the common peptide list based on all proteomics runs (dataspace) that was subsequently used for protein identification. For protein identification, when the peptide corresponded to multiple proteins, it was assigned only to one protein based on Occam's Razor approach (i.e., protein that was identified with the highest number of peptides in our dataspace). All identified proteins (independent from the number of peptide-spectrum matches and assigned peptides) were considered for quantification and normalization. Briefly, quantification of proteomics data was based on the precursor ion peak area. Abundance for a given protein was calculated based on the sum of all belonging peptide peak areas [15,16], followed by part per million (ppm) normalization. Statistical analysis was conducted using Mann Whitney test, and Diagnostics 2020, 10, 655 4 of 20 differences in protein abundance with p < 0.05 were defined as significant. The Spearman's rank-order correlation was employed to identify proteins associated with PCa progression. For the latter, the protein abundance was correlated with Gleason score (GS) (GS < 7, GS = 7 and GS > 7) from patients with PCa.

Differences in Protein Abundance Between Patients with PCa and BPH
A total of 276 proteins was found to be significantly altered (p < 0.05, Mann Whitney test) between samples from patients with PCa and BPH (Figure 2A). To increase confidence in the detected differences, only proteins detected in >30% of the samples were considered. Following these criteria and upon removing proteins originating from blood/plasma, 145 proteins (defined for the purpose of this manuscript as "PCa signature") were considered for subsequent evaluation (Figure 2A). The latter included 72 and 73 proteins being up-and downregulated in cancer in comparison to benign tissue, respectively. The list of differentially abundant proteins is presented in Table S2.

Differences in Protein Abundance between Patients with PCa and BPH
A total of 276 proteins was found to be significantly altered (p < 0.05, Mann Whitney test) between samples from patients with PCa and BPH ( Figure 2A). To increase confidence in the detected differences, only proteins detected in >30% of the samples were considered. Following these criteria and upon removing proteins originating from blood/plasma, 145 proteins (defined for the purpose of this manuscript as "PCa signature") were considered for subsequent evaluation (Figure 2A). The latter included 72 and 73 proteins being up-and downregulated in cancer in comparison to benign tissue, respectively. The list of differentially abundant proteins is presented in Table S2.

Association of Protein Abundance with PCa Progression
Significantly enriched Kegg pathways (Table 1) and MSigDB hallmarks ( Table 2) were represented by a total of 67 proteins, with most proteins that act in metabolic pathways identified with higher expression in malignant compared to nonmalignant prostate tissue. For those 67 proteins, association with cancer progression (as represented by increased GS) was investigated by Spearman correlation. Significant correlation of protein abundance with progression was observed for 25 proteins (i.e., 23 positively correlated and 2 negatively correlated; Table 3). In most cases, the direction of the association was in line with the direction of the fold change when comparing protein abundance between patients with PCa and BPH, with the exception of aldo-keto reductase family 1 member B1 (AKR1B1), carbonic anhydrase 2 (CA2) and acetyl-CoA carboxylase 1 (ACACA). These three proteins showed significantly lower abundance in cancer in comparison to controls (p < 0.05), while their abundance was positively correlated with GS (p < 0.05). For the shortlisted 25 proteins, association with disease progression was further evaluated in an independent transcriptomics dataset available from the Prostate Cancer Transcriptome Atlas [21]. The significant association with disease progression at the mRNA level was found for 21 out of the 25 proteins (p < 0.05, ANOVA, Table 3), supporting validity of the proteomics findings.

Discussion
This proof-of-concept study aimed to characterize protein changes in tissue that occur during PCa development and to identify affected molecular pathways. A high number of proteins could be identified as significantly affected, many of these in agreement with reports on the transcriptome level. In silico analysis of proteins differentially abundant between PCa and BPH clearly indicated that the vast majority of pathways were related to metabolism. Metabolic reprograming is one of the main cancer hallmarks [22], and its role in the PCa initiation, progression and resistance to therapies has been extensively studied [23][24][25]. Further investigation of the individual metabolic pathways revealed that many of them are well-established hallmarks of PCa, including, among others, fatty acid metabolism, TCA cycle and oxidative phosphorylation. Fatty acid metabolism has been recognized as a dominant process responsible for energy production in PCa cells due to slow glycolysis [26]. Fatty acid synthesis is required for energy production, membrane synthesis and posttranslational modifications [27]. Along these lines, previous studies have demonstrated also acceleration of the TCA cycle in the context of PCa [25], representing one of the key features of prostate malignant transformation. In contrast, normal prostate epithelial cells are specialized to produce and secrete citrate, a component of prostatic fluid, due to inhibition of the TCA cycle. In concordance with the activation of TCA cycle, our analysis revealed also an activation of oxidative phosphorylation. In comparison to other malignancies, PCa depends more on oxidative phosphorylation instead of glycolysis [25,28]. In addition to the metabolic changes, other pathways and biological functions that have been previously linked to PCa were also predicted based on proteomics analysis. Among others, IL-17 signalling [29], protein processing in ER [30,31], androgen signalling [32], PI3K/AKT/mTOR signalling [33] or MYC targets [34] were significantly enriched based on the proteomic PCa signature. The role of the most promising pathways is further discussed below. The identification of pathways/hallmarks that have been previously reported in the context of PCa serves as a positive control for our study and supports its validity.
To date, numerous molecular subtypes have been defined for PCa [6][7][8]. Thus, classification of samples based on clinical characteristics might not be optimal, also reflected by the moderate success of treatment. Taking this into consideration, we investigated PCa progression (expressed as increase in GS) based on the hypothesis that cancer progresses as a continuum and that proteins that are truly associated with this process are gradually and consistently changed. To further enhance the validity of the findings, the association with disease progression was also assessed at the mRNA level. This complementary assessment through cross-correlation of different -omics traits is expected to better reflect the clinical reality and to increase the validity of individual observations. The approach of cross-correlating proteomics and transcriptomics data has previously shown increased validity [35]. Such cross-omics analysis followed by interactome analysis highlighted prominent alterations in MYC targets, oxidative phosphorylation and protein processing in the ER. Interestingly, these processes seem to be interconnected, with MYC targets linking the latter two pathways. These pathways/hallmarks appear activated during PCa development, and abundance of the associated proteins (also verified at the mRNA level) was found to increase along with disease progression. Multiple evidence to support the implication of these pathways/hallmarks in the context of PCa, along with their interconnections, was collected, as discussed below.
MYC is a proto-oncogene frequently overexpressed in PCa [36] and associated with PCa progression [37]. In principle, MYC increases the expression of genes involved in cell growth and proliferation, survival of cancer cells [37,38] and regulation of cellular metabolism [39]. It promotes transcription of androgen receptor (AR) and enhances the stability of full-length AR and its splice variants [40]. Our study revealed several MYC targets being consistently upregulated during PCa development, including RACK1, HSP 90, EPRS1, and PA2G4. RACK1 [41], HSP 90 [42] and PA2G4 [43] have been reported previously in the context of PCa. RACK1 is a scaffolding protein involved in recruitment, assembly as well as regulation of signalling molecules and has been observed to promote proliferation, invasion and metastasis of PCa both in vitro and in vivo [41]. The molecular chaperone HSP 90 is involved in protein folding and maintaining protein stability (including AR) and has been reported to be overexpressed in many cancers, including PCa [42]. Considering its interactions with AR, HSP 90 was proposed as a potential therapeutic target in PCa [42]. PA2G4 is involved in cellular proliferation and regulation of gene expression and is also a corepressor of the AR. Significant increase in the protein expression from normal to hormone refractory PCa was reported previously, also correlating with the nuclear expression of AR in normal adjacent and cancer tissue [43].
Our analysis showed a connection between MYC targets and protein processing in ER. In fact, activation of ER stress/the unfolded protein response has been linked to cancer, including also PCa [31]. The unfolded protein response represents protective mechanisms of cancer cells under unfavourable conditions (e.g., hypoxia, oxidative stress and eventually ER stress). During the unfolded protein response, inositol requiring-enzyme 1 alpha (IRE1α)/X-box-binding protein 1 (XBP1) signalling is activated [31]. It has been shown that androgens regulate expression of genes associated with ER stress, including the IRE1α-XBP1 arm [31]. Among the proteins contributing to the interactome network described in our study and associated with protein processing in ER, we found SEC61A1 (SEC61 translocon subunit), a component of the protein translocation machinery (together with SEC62 and SEC63) mediating transport across ER [44]. SEC61A1 is a downstream XBP1 target gene, downregulated upon inhibition of XBP1 splicing with MKC-3946 treatment in multiple myeloma cells [45]. Nevertheless, the role of SEC61A1 has not been investigated in the context of PCa. Another component of the ER translocation machinery (although not found in our proteomics analysis), SEC62, has been investigated in PCa. SEC62 knockdown reduced the migration and invasion of PCa cells [46], and the overproduction of SEC62 protein in PCa tissue was correlated with GS [46]. In addition, our analysis identified RPN1, one of the subunits of the oligosaccharyl transferase complex, also being associated with the SEC61 complex. RPN1 is involved in N-glycosylation and was found (among other genes) to be downregulated upon treatment of androgen-independent PCa xenografts with 17β oestradiol that led to inhibition of tumour growth [47]. Protein disulphide-isomerase (P4HB) was another protein identified in our network as a member of protein processing in the ER pathway. P4HB is responsible for the formation, opening and reorganization of disulphide bonds. When present in high concentration, it acts as a chaperone that protects from the generation of misfolded protein aggregates. The P4HB gene was found to be significantly increased in PCa in comparison to the normal prostate gland [48], in agreement with our data.
Based on the available literature and in line with our results, protein processing in ER appears to be interconnected with MYC signalling [49]. Treatment with an IRE1α inhibitor inhibited PCa growth in vitro and in vivo, indicating that the IRE1α/XBP1 pathway promotes PCa through activation of c-MYC signalling [50]. Based on these findings, inhibition of IRE1α/XBP1 was proposed as a possible strategy for PCa treatment. In another study, interaction between IRE1α and RACK1 was crucial for the activation of IRE1α/XBP1 signalling upon unfolded protein response induced by sorafenib in hepatocellular carcinoma cells [51], further supporting the interconnection between MYC signalling and ER processes.
The data presented here also highlighted oxidative phosphorylation as a hallmark in PCa and indicated its possible connection with MYC targets. Changes in mitochondrial metabolism covering oxidative phosphorylation are one of the cancer hallmarks [52]. A recent study demonstrated alterations in oxidative phosphorylation in paired benign and malignant human prostate tissue samples [28], supporting the validity of our findings. Among the proteins mapped to oxidative phosphorylation in our network, ALDH6A1, an enzyme catalysing the oxidative decarboxylation of malonate and methylmalonate semialdehydes to acetyl-and propionyl-CoA, was found to be overexpressed in PCa tissues in comparison to normal prostatic tissue and was also significantly associated with lymphatic invasion in PCa [53]. In addition, SLC25A6, involved in exchanging ADP from cytoplasm with ATP from mitochondria through the mitochondrial membrane, was also found in our analysis among the proteins involved in oxidative phosphorylation. Furthermore, evidence supporting the connection of MYC with oxidative phosphorylation exists. Among others, MYC is known to be involved in the stimulation of mitochondrial biogenesis [54]. Previous transcriptomics analysis revealed several nuclear genes encoding mitochondrial proteins being downstream targets of MYC, including among others proteins involved in oxidative phosphorylation [54], which supports our findings.
Our study presents with some limitations. Since this was a proof-of-concept study, the sample size was not adequate to perform multiple testing correction. Therefore, identification of differentially expressed proteins was based on the unadjusted p-values. Even though, our findings were in line with the existing literature and, for several proteins, a consistent association with disease progression was observed at both the protein and mRNA levels. Further validation of the findings in the context of a statistically well-powered study is required.

Conclusions
Tissue proteome analysis allowed characterization of molecular changes associated with PCa development. Among the most promising pathways and biological functions consistently affected in disease onset and progression were protein processing in ER, oxidative phosphorylation and MYC targets. These pathways also appear to be linked with each other, which is supported by existing literature. However, their interconnections through the proteins identified in our study are mostly novel and require further validation. These molecular linkers may also serve as potential candidates for drug targeting.
Supplementary Materials: The following are available online at http://www.mdpi.com/2075-4418/10/9/655/s1. Table S1. Clinical data: Baseline clinical data for patients with PCa and BPH are provided. Table S2. List of differentially abundant proteins: number of peptides belonging to protein, fold change, p-value (Mann Whitney test) and normalised protein abundance in individual samples are given. Fold change was calculated as a ratio between average protein abundance in patients with PCa and average protein abundance in patients with BPH.