Prostate Cancer Secretome and Membrane Proteome from Pten Conditional Knockout Mice Identify Potential Biomarkers for Disease Progression

Prostate cancer (PCa) is the second most common cause of mortality among men. Tumor secretome is a promising strategy for understanding the biology of tumor cells and providing markers for disease progression and patient outcomes. Here, transcriptomic-based secretome analysis was performed on the PCa tumor transcriptome of Genetically Engineered Mouse Model (GEMM) Pb-Cre4/Ptenf/f mice to identify potentially secreted and membrane proteins—PSPs and PMPs. We combined a selection of transcripts from the GSE 94574 dataset and a list of protein-coding genes of the secretome and membrane proteome datasets using the Human Protein Atlas Secretome. Notably, nine deregulated PMPs and PSPs were identified in PCa (DMPK, PLN, KCNQ5, KCNQ4, MYOC, WIF1, BMP7, F3, and MUC1). We verified the gene expression patterns of Differentially Expressed Genes (DEGs) in normal and tumoral human samples using the GEPIA tool. DMPK, KCNQ4, and WIF1 targets were downregulated in PCa samples and in the GSE dataset. A significant association between shorter survival and KCNQ4, PLN, WIF1, and F3 expression was detected in the MSKCC dataset. We further identified six validated miRNAs (mmu-miR-6962-3p, mmu-miR- 6989-3p, mmu-miR-6998-3p, mmu-miR-5627-5p, mmu-miR-15a-3p, and mmu-miR-6922-3p) interactions that target MYOC, KCNQ5, MUC1, and F3. We have characterized the PCa secretome and membrane proteome and have spotted new dysregulated target candidates in PCa.


Introduction
Prostate cancer (PCa) is the most frequent cancer and has the second-highest morbidity and mortality rate among men, with 1,276,106 (7.1%) new cases globally and 358,989 (3.8%) deaths by cancer [1,2]. In the United States, the estimated number of new cases of PCa diagnosed in 2021 was 248,530, with 34,130 deaths. PCa in the United States accounts for 26% of all new cancer cases [3]. Current statistics show that one in seven men will be diagnosed with prostate cancer during their life and that one in 39 men will die of the disease [4].
The introduction of novel androgen receptor (AR) antagonists for clinical treatment has improved outcomes; however, most metastatic castration-resistant prostate cancer (mCRPC) patients ultimately develop resistance to these therapies. Patients with localized and advanced prostate tumors are sensitive to androgen deprivation therapy (ADT) and are highly curable; patients with metastatic prostate cancer acquire resistance to ADT and succumb to this disease [5]. While a large number of prostate cancer cases are diagnosed at a localized stage and are curable, metastatic prostate cancer remains fatal. In the last decade, large-scale omics analysis has revealed well-established and new master regulators and pathways involved in the metastatic and lethal behavior of PCa [6]. mCRPC is incurable, with a median survival rate f two years from diagnosis, and available treatments extend life for few months [7]. mCRPC commonly exhibits genetic alterations involving the AR, cell cycle and cell survival pathways such as the phosphatidylinositol-3-kinase (PI3K) and protein kinase B (PKB/AKT) [8,9]. One of the most frequently deleted genes in PCa, which negatively regulates PI3K-AKT signaling, is the tumor suppressor phosphatase and tensin homolog (PTEN) that is consistently associated with more aggressive forms and worse prognosis of PCa [10,11]. Loss of PTEN function has been well documented in PCa and PTEN mutations have been found in 40% metastatic PCa tumors [12,13]. Genetically Engineered Mouse Model (GEMM) Pb-Cre4/Pten f/f mice have been used since 2003, exhibiting pathological features similar with human prostate cancers, which includes the progression from intraepithelial neoplasia to invasive well-and poor-differentiated adenocarcinoma [14]. Moreover, this model has been used to produce new GEMM by combining mutations and to explore diet manipulation effects on prostate cancer progression [15,16].
Gene expression analysis is an important tool for understanding the behavior of tumors. Gene expression signatures have been successfully applied to define subclasses of different types of cancers with different biological behaviors and responses to therapies [17][18][19][20][21]. Several studies have revealed gene expression signatures of PCa tumors that correlate with poor prognosis in retrospective analyses [22][23][24]. Some of these molecular signatures help stratify patients with a Gleason score of 7, improve prognostic prediction, and provide appropriate management plans for patients after radical prostatectomy [23][24][25].
Comprehensive studies of histological, genomic, and transcriptome analyses and their relationship with PCa are necessary. Abida et. al. (2019) [26] presented an integrative analysis of genomic alterations with expression and histological evaluation of tumors from patients with mCRPC, representing the clinical spectrum of advanced disease, and with tissues collected before and after treatment with androgen signaling inhibitors [26]. However, most molecular signatures do not require validation before clinical use. In addition, some signatures include too many genes, which are expensive and hard to use in the clinic [23]. In addition, the list of genes generated in these signatures generally does not overlap between studies, and no gene sets have been validated for clinical use [27][28][29]. The search for molecular gene signatures is based on the assumption that a clear distinction between tumors that will relapse and those that will not is possible using gene expression profiles [29]. Therefore, more studies are needed to identify and validate prognostic markers.
Therapeutic and diagnostic options for PCa are limited, and progress in drug development is delayed because most cancers are highly complex at different levels, including cellular, genomic, and metabolic. The current challenge in PCa diagnosis is the lack of alternative screening to replace the existing PCa biomarker, prostate-specific antigen (PSA). Although PSA is widely used, it cannot distinguish between indolent and aggressive PCa [30][31][32]. Therefore, exploring new types of biomarkers beyond the conventional AR and PI3K pathways and/or altered genes, such as PTEN, P53, and RB1, are highly important in prostate cancer research.
The tumor microenvironment plays an important role in the initiation and progression of tumors [33][34][35]. Transcriptome analysis revealed that stromal regions adjacent to the tumor express genes that allow for re-stratification of the tumor microenvironment [36]. Secreted and membrane proteins play an important role in cancer metastasis by stimulating cancer cell migration and invasion, consequently increasing cancer metastasis [35][36][37][38]. Therefore, investigating potential targets for diagnosis and prognosis that are available in PCa tumor stroma provides an opportunity to reframe and help treat this disease.
In this study, we used available data to perform an integrative analysis of the PCa secretome and tumor membrane proteome. Our design consisted of identifying potential biomarker targets at different stages of PCa progression, demonstrated here by the early stages of mouse Prostatic Intraepithelial Neoplasia (mPIN), Middle-stage tumor (MT), and Advanced-stage tumor (AT), focusing on the tumor microenvironment of PCa. From the list of targets (genes and proteins) available in the Human Protein Atlas (HPA) secretome, we investigated a commonly deregulated gene network in the transcriptome of Pb-Cre4/Pten f/f mice. Enrichment analysis, protein-protein interaction (PPI) network, and in silico tools allowed us to identify nine membranes and secreted proteins that were either downregulated or upregulated in PCa. We also compared the transcriptomic profiles of prostate adenocarcinoma (PRAD) and normal tissue samples using The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) data, which revealed that four genes that encode secreted proteins were downregulated in PRAD. Finally, the gene expression patterns and prognosis of patients with PCa were analyzed by comparing four published datasets with disease outcomes (decreased relapse-free survival, overall survival, and probability of freedom from biochemical recurrence), with subsequent validation of HPA protein expression in human PCa and normal prostate samples.

Identification of Gene Expression Profile in Prostate Cancer of Pb-Cre4/Pten f/f Mice
We performed an integrative analysis of the prostate cancer secretome and membrane proteome data to identify clinically relevant diagnostic and prognostic biomarkers. According to the criteria used and from the list of genes included, our analysis identified upregulated and downregulated genes in the mPIN, MT, and AT distributed in all Anterior Prostate (AP), Dorsal Prostate (DP), Lateral Prostate (LP), and Ventral Prostate (VP) lobes.
After identifying the list of upregulated and downregulated genes in each prostate lobe, we checked for shared genes present in the AP, DP, LP, and VP lobes. We identified a list of genes common to the four prostate lobes. The number of upregulated common genes for membrane proteins was as follows: in mPIN = 98 genes, MT = 124 genes, and AT = 136 genes (Figure 2A  A list of secreted protein genes common to all four prostate lobes was also performed. The number of genes commonly upregulated for secreted proteins was as follows: in mPIN = 33 genes, MT = 52 genes, and AT = 63 genes ( Figure 3A-C). The number of downregulated common genes for secreted proteins present in the four lobes was as follows: in mPIN = 25 genes, MT = 31 genes, and AT = 26 genes ( Figure 3D-F). Our list of downregulated and upregulated PCa genes in mPIN, MT, and AT was analyzed using the EnrichR platform to identify enriched ontological terms. The most significant categories were tyrosine kinase activity, an integral component of the plasma membrane for upregulated genes (Tables 1 and 2). The most enriched terms for downregulated genes were inorganic cation transmembrane transport, potassium channel activity, and ion transmembrane transport (Tables 3 and 4). This analysis also showed their involvement in biological processes, such as glycolysis, carbohydrate biosynthetic processes, glycosaminoglycan metabolic processes, and extracellular organization.

Protein-Protein Interaction (PPI) Network of Membrane and Secreted Proteins Enriched in Prostate Cancer
Venn diagrams showing the list of common genes identified in the four prostatic lobes of predicted membrane proteins are represented in Figure 2; upregulated and downregulated genes are shown in Figure 2A-C, and Figure 2D-F, respectively. The common genes identified in the predicted secreted proteins are represented in Figure 3 for upregulated ( Figure    GO analysis of upregulated membrane proteins revealed significant protein enrichment in the integral components of the membrane, cell surface, immune system processes, and cell surface receptor signaling pathways (Supplementary Materials Data; Figure S1). The GO terms of downregulated membrane proteins revealed significant protein enrichment in the sarcoplasmic reticulum membrane, transmembrane transporter activity, transmembrane transporter activity, and ion transport (Supplementary Materials; Figure S1).
GO analysis of upregulated secreted proteins revealed significant protein enrichment in the categories of immune system process, glycosaminoglycan metabolic process, and cell migration (Supplementary Materials; Figure S2). The GO terms of downregulated secreted proteins revealed significant protein enrichment in the extracellular region, glycosaminoglycan binding, and extracellular space (Supplementary Materials; Figure S2

Differential Gene Expression of Transcripts Translated into The Membrane and Secreted Proteins in Prostate Cancer
After identifying all proteins present in the prostatic PPI network in mPIN, MT, and AT and among lobes, we selected the common upregulated and downregulated protein clusters in the three stages of PCa progression. The results identified 4 downregulated membrane proteins: Myotonin-protein kinase or myotonic dystrophy protein kinase (DMPK); Phospholamban (PLN); Potassium voltage-gated channel subfamily KQT member 5 (KCNQ5) and Potassium voltage-gated channel subfamily KQT member 4 (KCNQ4). We have also found 3 downregulated secreted proteins: Myocilin (MYOC); Wnt inhibitory factor 1 (WIF1); and Bone morphogenetic protein 7 (BMP7). The results identified Tissue Factor (F3) and Mucin-1 (MUC1) common upregulated proteins were present in secreted and membrane proteins.
The gene expression levels of the targets identified in our final list were analyzed using the online Gene Expression Profiling Interactive Analysis (GEPIA) tool (Tang et al., 2017). This tool allows the comparison of transcriptome profiles from TCGA and GTEx using uniformly processed and unified RNA sequencing data from the Toil Pipeline. The expression profiles of genes encoding secreted and membrane proteins were analyzed using the GEPIA tool to identify prognostic biomarkers of PRAD. The analysis showed that three genes (DMPK, KCNQ4, and WIF1) were significantly downregulated in PRAD (Log 2 fold change cutoff = 1 and q-value cutoff = 0.01) when compared to normal tissues ( Figure 6).

Survival Analysis and Risk Assessment
The gene (KCNQ4, PNL, F3, and WIF1) expression patterns and prognosis of patients with PCa were analyzed by comparing four published datasets (MSKCC, Cambridge, Stockholm, and MCTP). In these analyses, overexpression of the KCNQ4 gene expression with cut-off >7.83 (red line) and ≤7.83 (blue line) was associated with a reduced time of biochemical recurrence (p = 0.037) in the MSKCC dataset ( Figure 7A). In the MSKCC and Stockholm datasets PLN (p = 0.019) expression with cut-off >5.28 (red line) and ≤5.28   [42,43] from an integrative study. (A)-Kaplan-Meier curve with the probability of freedom from biochemical recurrence of PCa with (red) or without (blue) KCNQ4 overexpression with cut-off = 7.83 from the MSKCC study [8]; the difference was statistically significant, p = 0.037. (B)-Kaplan-Meier curve with the probability of freedom from biochemical recurrence of PCa with (red) or without (blue) PLN overexpression with cut-off = 5.28 from the MSKCC study [8]; the difference was statistically significant, p = 0.0019. (C)-Kaplan-Meier curve with the probability of freedom from biochemical recurrence of PCa with WIF1 expression with cut-off <6.02 (black line), WIF1 gene expression with cut-off <6.41 (red line), and patients with WIF1 gene expression with cut-off >6.41 (blue line) (p = 0.0046) from the Stockholm study [44]; the difference was statistically significant, p = 0.0046. (D)-Kaplan-Meier curve with the probability of overall survival of PCa patients with (red) or without (blue) F3 alteration from the metastatic prostate adenocarcinoma (MCTP, Nature 2012) study [45]; the difference was statistically significant, p = 0.0005807.

In Silico Validation of Protein Expression in Membrane and Secreted Proteins in Human Prostate Cancer
We analyzed the protein expression in human PCa and normal prostate samples from two upregulated (F3 and MUC1) and downregulated (MYOC and KCNQ5) targets found in the membrane and secreted protein list. The expression of F3 and MUC1 proteins using the HPA database showed increased (high or medium) immunostaining intensity in PCa tumor tissues; however, the expression remained low or undetectable in normal prostate tissues (Figure 8). Other dysregulated proteins identified in the HPA analyses were KCNQ4, DMPK, and PLN. At the time of this study, there was no immunohistochemistry tissue data available in the HPA database on PCa samples for BMP7 and WIF1 proteins.

Discussion
In this study, we first analyzed the PCa transcriptome from a Pten knockout mouse model for genes encoding membrane proteins and secreted proteins in PCa. We identified important DEGs in the PCa extracellular matrix (ECM) pertaining to hitherto unexplored pathways. From an integrative analysis of the secretome and membrane proteome, we identified 9 altered targets in PCa progression stages. The gene expression profile of these markers was altered in human PCa patients with worse overall survival and a worse probability of biochemical recurrence. Our strategy was to identify potential prognostic biomarker targets at different stages of PCa progression, focusing on the tumor microenvironment of PCa.
Tumors in patients with PCa present large histological, genetic, and molecular heterogeneity. A patient may harbor more than one genomic and phenotypically distinct prostate cancer; that is, these tumors appear independently and follow separate evolutionary trajectories. These clonally independent tumors exhibit biological differences and contribute differently to disease progression and clinical outcomes [46][47][48]. Currently, data on PCa proteomics and transcriptomes, using different GEMM and human patient samples, have been explored and integrated to identify potential targets against this disease.
The tumor microenvironment is a dynamic network of cells and structures, including tumor cells. The surrounding stroma is comprised of cancer-associated fibroblasts (CAFs), immune cells, mesenchymal stem cells (MSCs), ECM, cytokines, chemokines, and growth factors secreted by these cells [33,49]. It is already known that the tumor microenvironment plays an important role in the formation and progression of metastasis. CAFs deposit and degrade ECM components and thus remodel it during cancer progression, promoting immune cell infiltration and cancer cell proliferation, migration, and invasion [33,34]. CAFs can significantly promote proliferation and migration of prostate cancer cell lines [50,51]. Studies seeking to understand and identify precise biomarker signatures are necessary to identify effective targeted therapeutics to reduce the clinical and lethal diseases related to the role of inflammation in PCa progression [52,53]. It is important to note that studies that identified biomarkers for PCa, derived from markers of stromal infiltration or stromal transcriptomic and proteomic profiles, have not pointed to any of the markers that we found in our study [54][55][56][57]. Our targets are not directly related to the inflammatory profile but rather to another class of proteins related to the tumor microenvironment and ECM.
Of note, the findings of these targets were from the PCa transcriptome of knockout animals for Pten (Pb-Cre4/Pten f/f GEMM), in which they present important histopathological characteristics [58,59]. In addition to prostatic intraepithelial neoplastic (PIN) lesions, larger heterogeneous areas of fully invasive, both well-and poorly differentiated adenocarcinomas associated with reactive stroma are present in Pb-Cre4/Pten f/f GEMM. This model also presents a loss of the basal membrane structure and disorganization of the smooth muscle cells but shows rare metastasis. Additionally, infiltrated inflammatory cells are commonly identified in these tumors [58][59][60].
We believe that we found a set of proteins downregulated in PCa that are biologically important in (sub)types of human cancers. Myotonin-protein kinase or myotonic dystrophy protein kinase (DMPK) is a serine/threonine-protein kinase necessary for the maintenance of muscle structure and function [61]. DMPK is mainly expressed in smooth, skeletal, and cardiac muscles, and overexpression of DMPK mediated by p53 promotes contraction of the actomyosin cortex, which leads to the activation of caspases and concomitant cell death by apoptosis [61,62]. DMPK also phosphorylated phospholamban (PLN), another downregulated protein determined in our analyses. PLN is a small, and reversibly phosphorylated transmembrane protein found in the sarcoplasmic reticulum. Depending on its phosphorylation state, PLN binds to and regulates the activity of Ca 2+ pumps [63]. These two proteins are downregulated during PCa progression. We believe that this was due to the loss of smooth muscle cells [58]. Our enrichment analysis also showed changes in the sarcoplasmic reticulum membrane and ion transport, which may be related to autophagy processes, calcium homeostasis, and endoplasmic reticulum stress, as previously reported [64,65].
We also identified potassium voltage-gated channel subfamily KQT member 5 (KCNQ5) and member 4 (KCNQ4), both of which are important in regulating neuronal excitability. Voltage-gated potassium channels are responsible for the repolarization phase of the membrane action potential and play crucial roles in the excitability of neurons and other cells (Li et al., 2021). Several studies have proposed the use of KCNQ5 gene for the early clinical detection of colorectal precancerous lesions and cancer [66][67][68]. Downregulated expression of KCNQ5 has also been observed in other diseases [69,70]. These proteins play an important role in potassium homeostasis and are related to the enriched terms of potassium channel activity and potassium ion transport at different levels of PCa progression presented in our results. In our analysis, patients with altered KCNQ4 and PLN genes showed the shortest time for biochemical recurrence. The downregulation of these genes may be related to PCa progression.
In addition to membrane proteins, Myocilin (MYOC) was identified in our study. MYOC is a secreted glycoprotein that regulates the activation of different signaling pathways in adjacent cells to control different processes, including cell adhesion, cell-matrix adhesion, cytoskeleton organization, and cell migration [71]. Mutations in the MYOC gene are an important cause of glaucoma with dominant inheritance (Liuska [72,73]. How-ever, other types of cancer, such as thymoma, exhibit MYOC downregulation, thereby corroborating our results [74]. Wnt inhibitory factor 1 (WIF1) is a secreted protein that binds to WNT proteins and inhibits their activities. WNT signaling mainly controls cell proliferation, differentiation, and maintenance of stem cells (β-catenin-dependent pathway), cell polarity, and migration (β-catenin-independent signaling). The WNT/Ca 2+ signalling pathway is also associated with the release of Ca 2+ from intracellular stores [75,76]. A large body of evidence has shown that activation of the WNT signaling pathway contributes to the proliferation and transformation of malignant cells with metastatic activity [77,78]. The WNT protein is regulated by a variety of secreted extracellular proteins that interfere with the formation of the WNT-receptor complexes. Extracellular inhibition of the WNT signaling pathway, WIF1, plays an important role in controlling cell proliferation and acts as a tumor suppressor [79]. Owing to its biological function, interest in using WIF1 as a biomarker for the early detection, diagnosis, and prognosis of cancer has increased in recent years [80][81][82][83]. As shown here, WIF1 was associated with favorable overall survival in PCa, corroborating other studies [84].
Bone morphogenetic protein 7 (BMP7) (https://www.uniprot.org/uniprot/P18075 (accessed on 21 December 2021)) is a growth factor of the TGF-β superfamily that plays important role in various biological processes, including proliferation, differentiation, and apoptosis in many different cell types [85,86]. Bone morphogenetic proteins can act as either tumor suppressors or oncogenes, depending on the cellular context and tumor type [87,88]. Studies have suggested that BPM7 inhibition may represent a target for overcoming resistance to cancer immunotherapies [85], and the use of BPM7 overexpression is a strong predictor of the risk of tumor recurrence in gastric cancer [87].
The upregulated gene in the common membrane and secreted proteins, tissue factor (F3) (https://www.uniprot.org/uniprot/P13726 (accessed on 21 December 2021)), is a transmembrane glycoprotein and primary initiator of the extrinsic blood coagulation cascade and ensures rapid hemostasis in case of organ damage [89]. F3 has been associated with strong tumor growth enhancement and poor prognosis in cancer [90]. In our analyses, we found that PCa patients with altered F3 gene expression had reduced survival rate. F3 expression is increased in tumors and is associated with tumor progression, particularly in pancreatic [91,92], cervical [93], breast [94], and prostate cancers [95].
The transmembrane glycoprotein Mucin-1 (MUC1) is highly glycosylated and is normally expressed in glandular and luminal epithelial cells. MUC1 provides protection and creates a physical barrier to negatively charged sugars, limiting accessibility, and preventing pathogenic colonization [96,97]. MUC1 is overexpressed and has been identified as a potential target for diagnosis, prognosis, and therapy in most human cancers and plays an important role in tumor progression [96][97][98][99][100]. Recently, we reported a family of deregulated mucins, including MUC1, in PCa progression, where it was shown that mucin cells (mucinous metaplasia) are in AR-negative areas of proliferation, and that mucin-associated genes have a worse prognosis in PCa and have significant prognostic value for PCa patients [58]. miRNA controls gene expression by targeting mRNA based on sequence complementarity and can serve as oncomiR or tumor suppressor miRs by targeting mRNA that encode oncoproteins or tumor suppressor proteins [101,102]. Using miRNA-mRNA tumor expression data, we identified deregulated miRNA that was validated in the regulatory networks of the four target genes. Some miRNAs found in our analysis that regulate MYOC, KCNQ5, MUC1, and F3 mRNAs has been also described in other cancers, such as colorectal cancer cells [103] and identified as biomarkers in lung cancer [104], osteosarcoma [105], ovarian cancer [106] and penile cancer [107].
The miR-15a-3p miRNA has been associated with the three DEGs (KCNQ5, MUC1, and F3) in PCa in our analysis. The miR-15a-3p has been shown to suppress proliferation and migration inhibiting the expression of BCL2 and MCL1 in epithelial cells [108] and restrains the growth and metastasis of ovarian cancer cells by regulating Twist1 [106].
Evidence has shown that miR-15a-3p overexpression also suppressed cell proliferation via down-regulating Wnt/β-catenin signaling in PCa cells [109]. Moreover, the mir-671-5p was previously described to function as a tumor suppressor, inhibiting tumor proliferation by blocking cell cycle in osteosarcoma and negatively regulates SMAD3 to inhibit migration and invasion of osteosarcoma cells [110,111]. The mir-671-5p interacts with MYOC and KCNQ5 and may be down-regulating the expression of these genes in PCa.
Here, our strategy consisted of selecting upregulated and downregulated membrane and secreted proteins identified in PCa transcriptome analysis, which stratifies a group of unexplored proteins with high prognostic value and a potential target for therapy in a subgroup of patients. Although we tried to avoid bias in our study, certain limitations still need to be considered. Experimental in vivo and in vitro analysis should be performed to confirm our findings. Investigate the role and function of these miRNAs in PCa and their regulation of these genes are required. The experimental validation of the membrane and secreted proteins identified in this study could help correlate the results obtained herein with another group of patients' prognoses, diagnosis, and/or overall survival. Despite the above limitations, we have demonstrated a well-characterization of secretome and membrane proteome and have spotted new dysregulated target candidates in PCa.

Analysis of RNA-Seq Data of the Genetically Engineered Mouse Model (GEMM) for PCa: The Pten Conditional Knockout
We used RNA-seq data derived from the analysis of samples from the four prostatic lobes obtained from the GEMM Pten f/f , control, and Pb-Cre4/Pten f/f mice. We accessed RNA sequencing data derived from all prostate lobes using the NCBI Gene Expression Omnibus platform (GEO, https://www.ncbi.nlm.nih.gov/geo/ (accessed on 21 December 2021)), reference number GSE94574. Briefly, 72 samples were submitted for RNA-seq analysis, including 20 prostate samples from wild-type (WT), 16 mouse prostatic intraepithelial neoplasia (mPIN), 20 well-differentiated tumors (middle-stage tumor, MT), and 16 poorly differentiated tumors (advanced-stage tumor, AT). A minimum of four samples for each prostatic lobe and pathological condition for RNA-seq analysis were used. A detailed description of the histopathological aspects of each prostatic lobe and tumor stage of the mouse model has been previously described [59,112]. First, we explored genes that were differentially expressed in each lobe and at different stages of tumor progression (mPIN, MT and AT); we used Log 2 FC ≥ |+1| ≤ |−1| and adjusted the p-value < 0.05. The transcriptome used in this study was generated from animals provided by Dr. David

Integration of Secretome and Membrane Proteome Analyses to Identify Prostate Cancer Biomarkers
The DEGs from RNA-seq were used to predict membrane and secretome proteins using a known list of 5520 genes of Predicted Membrane Proteins and 1708 genes of Predicted Secreted Proteins available in The Human Protein Atlas Secretome (https://www. proteinatlas.org/humanproteome/tissue/secretome) [113,114] (accessed on 21 December 2021). The Predicted Membrane Proteins are a selection of seven prediction algorithms used to create a majority decision-based method (MDM) using the combined results from the chosen tools to estimate the human membrane proteome [115]. The human secretome was predicted by a whole-proteome scan using three methods for signal peptide prediction: SignalP4.0, Phobius, and SPOCTOPUS, which have all been shown to produce reliable prediction results in comparative analysis and selected the genes that were altered in at least three prostatic lobes in mPIN, MT, and/or AT.

Protein-Protein Interaction Network and Functional Enrichment Analysis
We used EnrichR software from Ma'ayan Lab (https://maayanlab.cloud/Enrichr/) (accessed on 21 December 2021) to determine the enrichment of ontological terms and molecular pathways related to DEGs [116,117]. The cutoff criteria used for both analyses were adjusted to a p-value of ≤ 0.05. Gene ontology (GO) enrichment analysis of secreted and membrane proteins was grouped into a single list. The ontological terms downregulated and upregulated in the biological process, molecular function, and cellular component categories with the lowest adjusted p-values were selected.
We used the STRING database (https://string-db.org/) [39] to identify the proteinprotein interaction (PPI) network by individually analyzing the upregulated and downregulated genes. The minimum interaction score required was 0.700 (high confidence), and the nodes disconnected from the network were hidden to simplify the display. The PPI enrichment p-value indicated the statistical significance provided by STRING (accessed on 21 December 2021).
The ShinyGO application (version 0.741) (http://bioinformatics.sdstate.edu/go/) [118] (accessed on 21 December 2021) was used to explore the enrichment of ontological terms in GO (http://geneontology.org/) (accessed on 21 December 2021) categories for the biological process of proteins from PPI. The cut-off criterion used for both analyses was a false discovery rate (FDR) p-value < 0.05.

Gene Expression Profile in Prostate Cancer
Differential expression levels were calculated using a web-based gene expression profiling analysis (GEPIA) tool [40]. GEPIA analysis revealed that genes encoding secreted proteins are regulated in PRAD (http://gepia.cancer-pku.cn/detail.php) (accessed on 21 December 2021). DEGs between tumor and normal samples were determined by oneway analysis of variance (ANOVA), applying the log 2 fold-change > 1 and q-value < 0.01. Genes were considered positively or negatively regulated and indicated in red and green, respectively, in PRAD (n = 489-492) relative to normal tissue (n = 150-152).

Survival Analysis and Risk Assessment
After identifying the secretome and membrane proteome targets of knockout mice, we performed analyses using data from publicly available databases. We investigated gene expression using the Cambridge Carcinoma of the Prostate App (CamcAPP) database and developed the CamcAPP (https://bioinformatics.cruk.cam.ac.uk/apps/camcAPP/) [41] (accessed on 21 December 2021) and the cBioPortal for Cancer Genomics database (https:// www.cbioportal.org/) [42,43] (accessed on 21 December 2021), to determine the association of gene alterations with patient clinical data, such as tumor risk development, prognosis, and survival rates. Survival curves were constructed using the Kaplan-Meier method. The expression of genes was associated with disease outcomes (decreased relapse-free survival and an increased expression level of genes in advanced prostate cancer) in several published PCa datasets, namely Memorial Sloan-Kettering Cancer Center (MSKCC) [8] and Cambridge and Stockholm integrative studies [44] performed using CamcAPP [41]. The published PCa datasets used was Metastatic Prostate Adenocarcinoma (MCTP, Nature 2012) [45] performed using the cBioPortal for Cancer Genomics database [42,43]. The grouping of samples found by recursive partitioning (RP) was used to construct a Kaplan-Meier plot by CamcAPP.

In Silico Validation of Differentially Expressed Genes (DEGs)
After gene expression analysis, the deregulated genes identified in our analysis of GEMM Pb-Cre4/Pten f/f PCa were assessed using the HPA (https://www.proteinatlas.org/) database (accessed on 21 December 2021) [113,114] to identify the distribution and localization of proteins in normal and tumor prostate samples via immunohistochemistry.

Prediction of Commonly Dysregulated miRNAs-mRNA Targets
We used the miRWalk 3.0 tool (http://mirwalk.umm.uni-heidelberg.de/) [119] to perform the regulatory interaction between miRNA and mRNA (MYOC, KCNQ5, MUC1, and F3); the algorithm for target validation was used in other available databases of Homo sapiens. The miRNA was considered significant when involved with at least three of the four genes selected. An alluvial plot diagram was generated using the online tool SankeyMATIC (http://sankeymatic.com/) to demonstrate the interaction networks between the miRNA and mRNA.