Impact of Genomic Mutations on the Transcriptional Pathways and Tumor Microenvironment Landscape of Localized Early Prostate Cancer Authors: William Lautert‐Dutra, Camila M Melo, Luiz P Chaves, Francisco C Souza, Vida Talebian, Cheryl Crozier, Mary Anne Quintayo, Linda Liao, Dan Dion, Fabiano P Saggioro, Felipe S Avante, Rodolfo B dos Reis, Jeremy A Squire, Jane Bayani Published: 2026 ABSTRACT ABSTRACTBackgroundThe management of intermediate‐risk early prostate cancer (PCa) is challenging due to the difficulty in distinguishing indolent from aggressive tumors. This study explores the association between genomic alterations and the tumor and its microenvironment (TME) and implications for disease progression.MethodsWe performed multi‐omic profiling in a cohort of 53 localized PCa using targeted sequencing, transcriptional, and proteomic spatial profiling.ResultsSomatic mutations and copy number alterations in RB1 (21%), PTEN (18%), and TP53 (9%) were identified. Kaplan–Meier analysis revealed that alterations in the RB and Cell Cycle pathways, particularly aberrations in PTEN, TP53, or RB1, were associated with shorter biochemical recurrence‐free survival (p < 0.001). Spatial proteomic analysis demonstrated a complex immune landscape in patients with mutations. The tumor compartment demonstrated higher expression of immune checkpoint markers, T‐cell activation proteins, and proliferation markers; and a TME that is enriched with CD8 + T cells and antigen‐presenting cells, but also with immunosuppressive M2 macrophages, suggesting adaptive immune resistance.ConclusionsOur analysis demonstrates that genomic alterations in PTEN, TP53, or RB1 are not only prognostic for poor outcomes but are also associated with a unique, immunologically complex TME in this Brazilian cohort.Keywords: biomarkers, digital spatial biology, immune, progression, proteomics Introduction Patients with localized prostate cancer (PCa) can experience excellent overall survival outcomes with those identified as very low‐ or low‐risk with a 5‐year survival at 98% [1]. Intermediate‐risk PCa is challenging, as some tumors behave indolently while others possess high‐grade phenotypes and outcomes [2]. Unfortunately, 20%–30% of localized cancers will progress [3, 4]. Moreover, while current clinical variables and pathological grading systems provide valuable guidance, they are often unable to distinguish between indolent and aggressive intermediate‐risk cancers. Identifying which intermediate‐risk tumors have high‐risk molecular features is essential for guiding treatment strategies and avoiding overtreatment. Thus, there is an ongoing need for advanced molecular biomarkers to enhance prognostic accuracy and guide therapeutic decisions in intermediate‐risk PCa. Genomic and transcriptomic studies have revealed significant PCa heterogeneity [5, 6, 7]. Improved risk stratification at biopsy and/or radical prostatectomy (RP) is therefore vital, not only to inform on treatment options, but also to alleviate patient anxiety. In advanced and metastatic settings, precision oncology biomarker research has resulted in several targeted therapeutics. For example, the identification alterations in DNA repair and response genes now predict benefit to Poly (ADP‐ribose) polymerase inhibitors (PARPi) [8] with biomarkers being evaluated for predictive benefit of PI3K/AKT/mTOR inhibitors, CDK4/6 inhibitors [9], and immunotherapies (reviewed by He et al.) [10]. Early PCa is characterized by genomic alterations, including changes in driver genes such as PTEN, TP53, RB1, SPOP, and ETS family gene fusions [6]. These changes not only directly impact the behavior of the transformed cell, but also with distinct patterns of interaction with other cells in their surrounding microenvironment [11]. Understanding the consequences of these genomic alterations, particularly the relationship between cancer cell phenotypes and the tumor microenvironment (TME), is essential for identifying new biomarkers and rational therapeutic strategies. The landscape of mutation and copy‐number changes in PCa has been extensively investigated [6], in addition to transcriptomic changes [12], both in the context of bulk analyses and single cell analysis [3, 5, 6, 7, 13, 14]. Similarly, tissue‐based proteomic profiling has revealed novel biomarkers [15]. Many of these studies rely on bulk profiling, which limits the ability to appreciate the impact of cellular context. Advances in microscopy and sequential or simultaneous staining with labeled antibodies now permit spatial proteomics [16, 17, 18]. This study aims to characterize the impact of genomic alterations on the molecular and spatial architecture of localized early PCa by combining targeted sequencing of hallmark pan‐cancer genes, transcriptomic analysis, and spatial proteomic profiling. Through this approach, we seek to map how specific driver events are associated with both tumor‐intrinsic phenotypes in this cohort and the immune/stromal landscape, providing new insights into the consequences of PCa genomics and the spatially resolved biology of the TME. This could provide potentially novel mechanisms of immunosuppression, escape, and progression, in addition to novel biomarkers. Methods Patient Cohort All 53 samples included in the cohort were primary PCa collected by radical prostatectomy (RP) following National Comprehensive Cancer Network (NCCN) [19] clinical practice guidelines in the Department of Surgery and Anatomy, Urology Division at Ribeirao Preto Medical School, Brazil, between 2007 and 2015 (Table 1). The samples and clinical classifications were processed according to the previously established method [20]. Patient outcome data were collected to the last follow‐up date. The patients were classified according to the presence of biochemical recurrence (BCR), defined as PSA levels > 0.2 ng/mL within 6 months after RP. The risk of prognostic PCa recurrence was estimated after first‐line surgery using the Cancer of the Prostate Risk Assessment Score (CAPRA‐S) [21]. Patients were grouped by CAPRA‐S scores: low (0–2), intermediate (3–5), and high (6–12). This retrospective study was approved by the Ethics Committee in Research of Hospital of Ribeirao Preto, Sao Paulo, Brazil (HCRP) number CAAE 60032122.8.0000.5440 and the Ethics Board of the University of Toronto (Protocol: 00043323). Furthermore, this study was performed in accordance with the Declaration of Helsinki. Table 1Summary of histopathological and clinical features of the Brazilian PCa cohort.VariableCohort (n)BCRNo32Yes19CAPRA‐SLow‐CAPRA‐S20Intermediate‐CAPRA‐S21High‐CAPRA‐S10pGS5—687378492ISUP182243134254TNMpT1c—pT22pT2a2pT2b3pT2c30pT3a6pT3b7Average age at surgery66Abbreviations: BCR, biochemical recurrence; CAPRA‐S, Cancer of the Prostate Risk Assessment Score; ISUP, International Society of Urological Pathology grading system; pGS, prostatectomy Gleason score; TNM, tumor size, lymph node, and metastasis status. Summary of histopathological and clinical features of the Brazilian PCa cohort. Abbreviations: BCR, biochemical recurrence; CAPRA‐S, Cancer of the Prostate Risk Assessment Score; ISUP, International Society of Urological Pathology grading system; pGS, prostatectomy Gleason score; TNM, tumor size, lymph node, and metastasis status. Tissue Handling and Targeted Sequencing Formalin‐fixed paraffin‐embedded (FFPE) tissues were processed at the Ontario Institute for Cancer Research, Toronto, Canada (OICR) for dual DNA and RNA extraction, as previously described [22]. Briefly, representative sections were stained with haematoxylin and eosin (H&E) and histologically reviewed by a pathologist (by FPS) to identify the highest Gleason pattern. Areas containing 70%–95% tumor enrichment were identified for macrodissection. Nucleic acids were isolated from macrodissected tumor‐rich areas. Targeted Sequencing using the Oncomine Comprehensive Assay Plus: Libraries were manually prepared from 50 ng of genomic DNA using the Oncomine Comprehensive Assay Plus (OCAPlus) kit according to the manufacturer's instructions (Thermo Fisher Scientific, Mississauga, ON, CA). Barcoded libraries were loaded onto Ion 550 chips and templated using the Ion Chef and sequenced on the Ion S5 XL System (Thermo Fisher Scientific). Raw sequencing data were mapped to the human genome hg19, indicated as the reference genome in the Ion Reporter v5.18 (Thermo Fisher Scientific). The Oncomine Comprehensive Plus ‐ w2.3 ‐ DNA ‐ Single Sample workflow was used for analysis. The coverage, mean depth uniformity, and alignment were used as references for quality assessment. Analysis: Samples having > 500,000 total reads, > 100,000 reads for each of the two primer pools, and a mean read length > 60 bp were included in the analysis. Due to formalin fixation‐based nucleotide deamination leading to C > T and G > A base changes, low‐frequency artifacts are predicted to occur in all samples with an allelic frequency ranging from 5% to 10% [23]. To address the presence of false‐positive changes, annotated variant call format (VCF) files were filtered following these thresholds: (1) variant allele frequency (VAF) > 10% and < 80%, (2) p < 0.0001, and (3) coverage > 350. In contrast, variants isolated from highly deaminated samples were excluded if they did not reach VAF > 15% and < 80%, p < 0.0001, coverage > 350, and Phred score of 300. Furthermore, mutations detected in all samples [24] were excluded. Copy‐number somatic variants (CNVs) were selected for RB1, TP53, and PTEN, with num ≥ 8 and ≥ 9 for CNV amplifications and CNV deletions, respectively. To further evaluate the molecular impact of the somatic mutation on PCa patients, variants and copy‐number changes were categorized according to their respective pathways (Supporting Table 1). The pre‐defined “Prostate Cancer” gene lists available on the CBioportal was also used as a reference (https://www.cbioportal.org/). Gene Expression Profiling Data from the NanoString PanCancer, Immune Profiling, and Breast Cancer 360 panels (NanoString Technologies Inc., Seattle, WA, USA), previously published by our group, were used for this study [20, 25]. Each dataset was normalized individually using the housekeeping (HK) genes in their respective panels. Specifically, for each sample i, we computed the sum of the HK gene counts, housekeepingSumi. These were used to compute: 1. normalizationFactor i = mean(housekeepingSum i )/housekeepingSum i Then we normalized the count for sample i and probe j as follows: normalizationFactor i = mean(housekeepingSum i )/housekeepingSum i Then we normalized the count for sample i and probe j as follows: 2. count ij ′ = count ij × normalizationFactor i count ij ′ = count ij × normalizationFactor i To combine the NanoString datasets, we further normalized using the positive and negative control probes present in all datasets. Specifically, for each sample i across all datasets, we computed the sum of the control probe counts, controlSum i . These were used to compute: 3. normalizationFactori′ = Mean(controlSumi)/controlSumi normalizationFactori′ = Mean(controlSumi)/controlSumi Then we normalized the count for sample i and probe j as follows: 4. count ij ″ = log 2 (count ij ′ × normalizationFactor i ′) count ij ″ = log 2 (count ij ′ × normalizationFactor i ′) We assessed the effectiveness of this across‐dataset normalization by measuring the similarity between the datasets restricted to the samples and genes they have in common. Specifically, for each pair of datasets, we computed the concordance across common samples, for each common gene (Supporting Figure 1A). Moreover, we fitted the gene concordance as a function of the gene median normalized count (median for each dataset and then the mean of dataset medians) to understand when concordance values are lower (Supporting Figure 1B–D). Due to the non‐linear trend, we did a LOESS fit [26] after trimming the genes with the bottom and top 2.5% median counts, with the loess() function in R with default arguments. Finally, we merged the datasets by restricting to common samples, and for common genes, computing the mean normalized count across datasets. For genes that are not present in multiple datasets, the normalized counts are used as is. The above analyses were performed using R v4.3.3 with the xlsx v0.6.5, NanoStringNorm v1.2.1 [27] and BoutrosLab.plotting.general v7.1.0 [28] packages. A final dataset comprising 1703 gene targets was used for analysis. Differential expression analysis was conducted using the R package limma v3.60.6 [29] to generate empirical Bayes moderated t‐statistics and p‐values for each gene on the integrated panel's dataset. As model contrasts, we used mutated PCa hallmark pathways with at least four patients in the mutated group. We also assessed the differentially expressed genes in the BCR and CAPRA‐S groups for internal validation. Molecular subtyping was performed using the proprietary Prosigna algorithm (NanoString Technologies Inc.). GeoMx Digital Spatial Profiling (DSP) Tissue microarray (TMA) slides were stained and normalized according to the protocol described in GeoMx DSP Slide Preparation User Manual v2.1 (NanoString Technologies, Inc) and the GeoMx DSP Analysis Suite (v.2.4.2.2). Core antibodies for Immune Cell Profiling and five additional modules were used: (1) Immuno‐oncology Drug Targets; (2) Immune Activation Status; (3) Immune Cell Typing; (4) Pan Cancer; and (5) PI3K/AKT signaling (NanoString Technologies, Inc.) (Supporting Table 2). Morphological antibodies, anti‐PanCK, and anti‐CD45. The TMA sections were used with the nuclear stain, SYTO 13 (GeoMx Nuclear Stain Morphology Kit, NanoString). Each TMA was processed in the GeoMx DSP instrumentation. Each TMA core served as a region of interest (ROI) (~600 µm in diameter) where markers PanCK and CD45 were used to segment the ROIs into tumor and TME areas of interest (AOIs), respectively. The minimum nuclei and surface area counts were set at 20 and 1600, respectively. The protein counts were normalized by applying the ratio of the geometric mean to the surface area of each compartment to all the counts. Second, the counts were then normalized to HK proteins S6, GAPDH, and Histone H3. In addition, rabbit and mouse isotypes are also included as negative controls to correct for background noise. A hierarchical cluster heatmap was created using GeoMx DSP Analysis Suite, where the log‐transformed, z‐scores for the immune proteins were determined. Protein marker cell scores were calculated by averaging cell‐specific proteins if the cell type was represented by multiple proteins (Supporting Table 3). Statistical Analysis The data processing and downstream analysis were completed in RStudio software (R Foundation for Statistical Computing, R v4.4.1 “Race for Your Life”). All results were plotted using ggplot2 v2.3.5.1 or ComplexHeatmap v2.20. Genes were considered differentially expressed when log2 foldchange > |1|. Further, the dichotomized groups based on mutated PCa hallmark pathways were used to estimate the time‐to‐detection of BCR. The log‐rank test was used to estimate the time‐to‐detection of BCR in Kaplan–Meier curves (BCR free‐survival, BCRFS) using the survival R package v3.8‐3 and plotted using survminer v0.5. Two‐tailed Welch's T‐test was used for statistical comparison between cell marker and signature classes at p < 0.05. Results Aberrations in RB1, PTEN, or TP53 Defines a Biological High‐Risk Group Associated With Poor BCR‐Free Survival A summary of the study cohort is shown in Table 1. Targeted sequencing identified 33 patients with somatic mutations and copy number changes. A total of 65 somatic mutations across 45 of 501 genes in the targeted panel were detected. The most frequently altered genes included RB1 (21%), PTEN (18%), TP53 (9%), ASXL1 (9%), BCOR (9%), RAD52 (9%), KMT2C (9%), and KMT2A (9%) (Figure 1A). Notably, the alterations in the RB1 and PTEN genes were predominantly expressed as copy number losses. Additionally, changes in other genes such as CDK12 (3%), BRCA2 (3%), ATM (3%), MPK3K4 (3%), and PIK3R1 (3%) were also observed. When comparing the mutation frequency across the CAPRA‐S groups, the high‐risk group (n = 10) had 19 mutations, the intermediate‐risk group (n = 21) had 30 mutations, and the low‐risk group (n = 20) had 14 mutations. As some tumors contained multiple mutations, this corresponds to an average of 1.9, 1.4, and 0.7 mutations per patient in the CAPRA‐S high, intermediate, and low‐risk groups, respectively. Landscape of genomic changes using the OCAPlus targeted pan‐cancer panel and implications for prognosis. (A) Summary of mutational and copy‐number changes identified across the study cohort, highlighting frequent alterations in the key tumor suppressor genes RB1, PTEN, and TP53. (B) Summary of pathways impacted by mutational and copy‐number changes, highlighting the stratification of patients by BCR status. (C) Genomic copy‐number plot from Brazil‐011 and ‐019 showing chromosomal gains and losses in addition to focal amplifications and deletions. Kaplan–Meier plots showing BCRFS based on alterations in the RB pathway (D) and Cell Cycle Pathway (E). In both cases, patients with alterations in genes contributing to the pathway experienced poor BCRFS p = 0.001 and p = 0.002, respectively. [Color figure can be viewed at wileyonlinelibrary.com] To identify the pathways impacted, mutational and copy number alterations were grouped into the following pathways: TP53/p53, DNA damage response, PI3K/AKT/mTOR, Cell Cycle Control, RB pathway, AR and steroid signaling, Notch signaling, TGF‐β, MYC signaling, Invasion and Metastasis, and WNT signaling (Figure 1B). The most frequently altered pathways included Cell Cycle Control, PI3K/AKT/mTOR, and Notch signaling pathways, each accounting for 18% of the mutations. The PI3K/AKT/mTOR pathway, represented primarily by mutations in PTEN and PIK3R1, showed significant alterations. The Notch pathway was represented by NOTCH3, EP300, SPEN, and FBXW7. The RB pathway followed closely in 16% with mutations. Not surprisingly, the RB and Cell Cycle Pathways were driven by RB1 deletion, reflecting the critical role of this gene in cell cycle regulation. The TP53/p53 signaling pathway, represented by TP53 and ATM mutations, exhibited 8.9% of the mutations. The DNA damage response pathway accounted for 6.7% of mutations, which were composed of alterations in BRCA2 and ATM. Mutations in AR Steroid Signaling and WNT signaling were less frequent, each representing 4.4% of the alterations. Rare mutations were observed in pathways associated with invasion and metastasis, MYC signaling, and TGF‐β, each constituting 2.2% of the total. The genomes of most cases displayed relatively low levels of aneuploidy, with only two cases (Brazil‐011 and Brazil‐019) showing significant aneuploidy (Figure 1C). These two cases were notable for possessing both copy number alterations and/or mutations in TP53 and PTEN. Additionally, Brazil‐011 exhibited an RB1 deletion alongside the alterations in PTEN and TP53. Of the patients that experienced a BCR, 60% had an alteration in any of RB1, PTEN, or TP53. Due to the relatively low frequency of individual alterations in a cohort of this size (e.g., TP53 at 9% and PTEN at 18%), individual statistical comparisons were underpowered. However, because RB1, PTEN, and TP53 alterations frequently co‐occur in aggressive PCa and converge on pathways that drive progression, cases with alterations in any one of these genes formed a tumor‐suppressor driven biologically high‐risk group [30]. To investigate the link between pathways affected by somatic mutations and BCR progression‐free survival analysis (BCRFS), an analysis of mutated pathways (Supporting Table 1) was performed. Kaplan–Meier (KM) analysis revealed a significant association of the RB (Log‐Rank test p = 0.01) and Cell Cycle (Log‐Rank test p = 2.2 × 10−4) signaling pathways with reduced BCRFS (Figure 1D,E). When patients with any of RB1, PTEN, or TP53 aberrations were compared to those without aberrations, those patients also experienced significantly worse BCRFS (Log‐Rank test p = 1.7 × 10−4). There was no statistical difference between patients with mutations impacting the PI3K/AKT/mTOR pathway (Log‐Rank test p = 0.33) or Notch pathways (Log‐Rank test p = 0.57). Spatial Proteomics Reveals a Complex TME To characterize the protein‐level architecture of the tumor and its TME, we applied GeoMx DSP to our patient cohort. Figure 2A shows representative images of tumor cores where two areas of interest (AOIs) were collected per core, defined by green‐fluorescing PanCK‐positive regions for the tumor epithelial regions, and pink‐fluorescing CD45 regions defining the TME. Data was collected from 43 patients (n = 48 regions of interest, ROI). For each ROI and associated AOI (n = 96 total), hierarchical clustering of the 61 proteins (Figure 2B) shows the tumor compartment AOIs grouped separately from the TME AOIs. Across all samples profiled, the tumor compartment was defined by relatively higher expression of CD127, EpCAM, and Phospho‐GSK3B (S9). In contrast, the TME showed relatively higher expression of proteins consistent with its stromal and immune components, including immune cell markers (CD8, CD4, CD30, and CD20) and stromal markers (Fibronectin and SMA). Differential proteomic analysis showed no significant differences within the tumor and TME compartments when stratified by BCR status (Figure 2C,D, Supporting File 1). When grouped according to their protein marker classes (Supporting Table 3), the cluster patterns recapitulated the findings when using all gene proteins in the proteomic panel (Supporting Figure 2), with no discrimination based on BCR or CAPRA‐S status. Although the immune profile across all ROIs demonstrated the generally immune cold landscape of PCa, across most of the tumor segments profiled, there was relatively higher expression of epithelial markers, Treg and naïve and memory cells, with most other classes of immune cells more highly expressed in the TME (Supporting Figure S5). Summary of DSP proteomic profiling. (A) Representative images of the low‐density TMA (one of five total) and patient cores from patients Brazil‐019 and ‐047. PanCK‐positive cells are identified by green fluorescence, CD45‐postive cells are identified by pink fluorescence and cell nuclei stained by blue fluorescing STYO13. Each core provides two AOIs defining the proteome of the PanCK‐masked region and the CD45‐masked region. (B) Hierarchical clustering results from DSP proteomic profiling from 43 patients showing the clustering of AOIs (n = 96) into two groups: those characteristic of the TME and those characteristic of the tumor compartment. Differential protein expression shows relatively low expression of immune‐related proteins in the tumor compartment compared to the TME. Stratification of patients according to BCR status and spatial expression between the TME and tumor compartments is shown in (C) for patients who experienced a BCR; and (D) for patients who did not experience a biochemical recurrence. Proteins labeled in red are significantly expressed in the tumor compartment, while proteins labeled in green are significantly expressed in the TME. Dotted lines delineate −log10 p‐value and log2 FC cutoffs. (E) Box‐plots show significantly expressed cell classes between patients with an aberration in any of RB1, PTEN, or TP53 vs. those without aberrations stratified by compartment: tumor and TME. [Color figure can be viewed at wileyonlinelibrary.com] To investigate whether there were protein expression differences, the data were similarly stratified between the biologically high‐risk group (defined by the presence of mutations in RB1, PTEN, or TP53) and those without alterations in those tumor suppressors. Classification by protein marker classes, showed there were significant differences for checkpoint markers (p = 0.009), dendritic cells (DCs) (p = 0.038), hematopoietic cells (p = 0.037), memory cells (p = 0.042), MHC2 cells (p = 0.033), proliferation (p = 0.011), TME markers (p = 0.044), T cell activation (p = 0.022), and T cells (p = 0.023), all elevated in the tumor compartment of mutated cases (Table 2 and Figure 2E). Similarly, differences in the stromal compartment were seen with higher scores for the mutant cases in antigen presentation (p = 0.031), CD8 T cells (p = 0.002), dendritic cells (p = 0.017), M2 macrophages (p = 0.044), memory cells (p = 0.033), monocytes (p = 0.036), and T cells (p = 0.030) (Figure 2E). Table 2Comparison of GeoMx digital spatial protein marker class scores between RB1/PTEN/TP53‐mutated cases and non‐mutated cases.Any of RB1/PTEN/TP53 vs. noneProtein marker classesTumor compartmentTME compartment p‐value < 0.05 p‐value < 0.05Antigen0.1710.561Antigen presentation0.052 0.031 Apoptosis0.9170.239B cells0.6170.417CD8 T cells0.053 0.002 Checkpoint 0.009 0.347CT antigen0.9100.227Cytotoxicity0.1980.215DC 0.038 0.017 Epithelial0.2250.565Fibroblasts0.0810.329Hematopoietic 0.037 0.675Hormone receptor0.3240.216Interferon0.6230.323M2 macrophage0.074 0.044 Macrophage0.0550.064Melanoma0.3960.913Memory 0.042 0.033 MHC2 0.033 0.088Monocyte0.080 0.036 Myeloid0.0990.157Myeloid activation0.1450.419Myeloid suppression0.2190.276Naive and memory0.2060.224Neutrophil0.3690.580NK cells0.8750.191Oncogene0.5320.738PI3K‐AKT0.6650.853Proliferation 0.011 0.185Stroma 0.044 0.306T cell activation 0.022 0.085T cells 0.023 0.030 Th cells0.0890.156Total immune0.1670.097Tregs0.3320.116Tumor0.1380.325Tumor suppressor0.2480.990 Note: Bolded values identify protein marker classes that were statistically significant. Comparison of GeoMx digital spatial protein marker class scores between RB1/PTEN/TP53‐mutated cases and non‐mutated cases. Note: Bolded values identify protein marker classes that were statistically significant. Bulk RNA Expression Profiling Stratified by RB1, PTEN, or TP53 Status Confirms Increased Proliferation Markers and Immune Dysregulation To validate and expand on the observed proteomic changes, particularly amongst the different immune cell classes, we utilized previous bulk gene expression data to provide a deeper view of the cellular programs altered by these driver genes and their contributions to poor clinical outcomes [20, 25]. Robust RNA expression results from all three expression panels were pooled (Supporting Figure 1) and used to explore differential expression of genes associated with key pathways affected by somatic mutations specifically Cell Cycle Control and RB pathways (Figure 3A). Differences in gene expression between patients possessing aberrations in the RB pathway included the relative overexpression of NDC80, HIST3H2BB, RAG1, SPC25, TMEFF2, EXO1, POLQ, TPTE, ANLN, and SSX1 which represent genes involved in cell division, chromosomal segregation, and DNA repair (Supporting File 2). Similarly, differential gene expression between those with and without aberrations involving Cell Cycle Control genes involved the relative overexpression of CD1C, MME, CDC80, RAG1, and TMEFF, which impacted both cell division and, interestingly, immune activation. When stratified according of the high‐risk PTEN, TP53, or RB1 tumor suppressor group (Figure 3A) (Supporting File 2), there was significant over expression of genes related to cell cycle and replication and downregulation of a handful of genes, including FOLS1, LIF, CCL2, IL1B, SOCS3 KRT17, and CEACAM1, broadly related to inflammation and endothelial cell differentiation, suggesting a greater impact on immune response collectively amongst cancers with mutations. These observations support the findings of our DSP cohort analysis, in which alteration in drivers genes may be associated with the suppression of key anti‐tumor inflammatory pathways, consistent with a trend toward an immunosuppressive environment observed via the M2 macrophage enrichment. Summary of gene expression profiling. (A) Volcano plots show differential gene expression based on bulk RNA analysis showing significantly overexpressed genes (labeled in red) amongst patients with aberrations in genes of the RB and Cell Cycle pathways, in addition to stratification according to RB1, PTEN, or TP53 status. Genes that show relatively lower expression in the of RB1, PTEN, or TP53 altered cases are labeled in blue. (B) Box‐plots of gene signatures from the BC360 panel showing significant differences according to of RB1, PTEN, or TP53 status. (C) Box‐plots demonstrating the range and mean subtype correlation for each of the intrinsic subtypes of the PAM50 breast signature. The entire PCa cohort was identified as Luminal A due to the strongest correlation values to the Luminal A subtype. Outlier cases are shown with those patients possessing alterations in RB1, PTEN, or TP53 shown in red. The associated table shows the mean correlation for each subtype and the raw correlation scores for the outlier cases. [Color figure can be viewed at wileyonlinelibrary.com] Moreover, when stratified by mutational status using bespoke gene signatures from the NanoString BC360 panel (Supporting Table 5) [20, 25]. Statistically significant differences were identified with the following signature scores: BCa TP53 (p = 0.002), BCa proliferation (p = 0.029), CDK4 expression (p = 0.027), and inflammatory chemokines (p = 0.0034), which possessed higher scores in the mutated cases, suggesting enrichment of aberrant TP53‐driven pathways, increased proliferation, higher CDK4, and inflammatory chemokine expression, respectively. Scores for mast cells (p = 0.025) and PDL1 (p = 0.024) were higher in the non‐mutated cases (Figure 3B and Supporting Table 5), showing altered immune response and markers. This complex transcriptional profile supports our proteomic findings of an immunologically active but suppressed cold TME. Finally, we re‐examined the PAM50 cancer subtype results from our previous work [20]. While the entire cohort was classified as Luminal A, a deeper analysis of the raw correlation data revealed four outlier cases with transcriptional profiled suggesting contributions from Luminal B and HER2‐Enriched (HER2‐E) subtypes (Figure 3C). Crucially, three of these cases (Brazil‐011, ‐019, and ‐020) fall within the RB1, PTEN, or TP53 high‐risk mutational group. Brazil‐011 possessed aberrations in all three genes, including a frameshift deletion of TP53, PTEN LOH, and loss of RB1; Brazil‐0019 had a gain in both PTEN and TP53; and Brazil‐020 experienced a loss of RB1. Moreover, both Brazil‐011 and ‐019 were the only two cases with marked aneuploid genomes, (Figure 1C) while the rest were diploid and genomically stable. This observation demonstrates that even within a favorable subtype classification, these key mutations are associated with an underlying aggressive molecular program that drives the tumor toward a worse prognosis. Discussion This study provides a novel approach to understanding the biology of localized early PCa, elucidating the relationship between genomic changes not only on tumor cells but also on the surrounding TME. While it is well‐established that aberrations in key driver genes profoundly impact the entire tumor ecosystem (reviewed in Melo et al., [11]), our work provides deeper hypothesis‐generating data from a spatially informed perspective, providing a potentially novel mechanisms of immunosuppression, escape, and progression, in addition to novel biomarkers. Targeted sequencing with a panel of pan‐cancer genes identified a similar distribution of mutations and copy‐number changes of cancer genes that are characteristic of primary PCa [6]. Known drivers, RB1, PTEN, and TP53 [5, 6, 7] affected 21%, 18%, and 9% of patients, respectively. Other variants in ASXL1, BCOR, RAD52, KMT2C, and KMT2A (9% each), with lower frequencies of CDK12, BRCA2, ATM, MPK3K4, and PIK3R1 (3% each) were also identified impacting DNA damage response and chromatin modification pathways (Figure 1A). Although limited for in‐depth copy‐number analysis, this targeted panel demonstrated sufficient evidence that most cases had largely stable and quiet diploid genomes with only two cases showing evidence of aneuploidy (Figure 1C). A pattern of chromosomal changes associated with both early, and more characteristically, advanced cancers [7], including net losses of 5q, 6q, 8p, 13q, and 16q; and net gains of 6p, 8q, 11q, 12q, and 17 were observed. We then investigated how these genomic alterations translate to clinical risk. A study by Hamid et al., [30], demonstrated that alterations were frequent in the key tumor suppressors RB1, PTEN, and TP53 and increased the risk of relapse in localized castrate‐sensitive and ‐resistant disease. Additionally, they reported that the presence of more than one mutation amongst the three was similarly associated with an incremental increase in the risk of relapse. Given the progressive nature of PCa biology and limited by cohort size, we used a pathway‐focused approach we identified actionable pathways associated with these genomic changes that offer the potential for therapeutic interventions typically reserved for advanced cancer (reviewed by He et al., [10]). Indeed, patients with aberrations in genes characterizing Cell Cycle control (p = 2.2 × 10−4) and RB (p = 0.01) pathways were associated with a significantly increased risk of BCR (Figure 1D,E). These pathways are critical regulators of tumor progression and treatment resistance in PCa [31], with the RB pathway playing a pivotal role in regulating cell cycle progression, with inactivation of RB1 by mutations being linked to aggressive phenotypes and poor outcomes. Similarly, aberrations in genes of the Cell Cycle pathway, including alterations in CDK2/4/6, and their inhibitors, is associated with DNA replication and unchecked cell cycle proliferation. These findings are similar to those previously reported [6] and offer the possibility that these at‐risk patients might benefit from therapies targeting those pathways [10]. The impact of these genomic alterations was supported by bulk expression data demonstrating relative increase of the CDK4 expression signature (p = 0.027) and proliferation signature (p = 0.029) in the high‐risk tumor suppressor driven cases with RB1, PTEN, or TP53 aberrations compared to non‐mutated cases (Figure 3B) (Supporting Table 4). Many of the differentially expressed genes showed an enrichment for those associated with proliferation, chromosomal segregation, and the cell cycle (Figure 3A). When we examined the intrinsic subtyping status using the PAM50 algorithm [32, 33] as evaluated in other PCa studies [34, 35, 36], three of the cases which had outlier raw correlation values across the four subtypes and were amongst the any of PTEN, TP53, or RB1 aberrant group despite their Luminal A classification (Figure 3C). This illustrates a heterogeneous continuum across subtypes where more aggressive Luminal B and HER2‐E‐like molecular contributions may exist. While the HER2‐E subtype is not a defined group in the context of PCa, the underlying biology driven by EGFR‐transcription family signaling is associated with more aggressive phenotypes and higher rates of proliferation [37], consistent with our observations. Interestingly higher rates of HER2 positivity appear to exist amongst PCa patients of African ancestry [38] and given the diverse ethnicity of Brazil [39], HER2 status or molecular phenotypes linked to HER2 signaling may be relevant to prognosis and treatment selection. The mutational spectrum and associated poor outcomes amongst patients in the tumor suppressor–driven cases are strongly associated with aggressive PCa proliferation [30], neuroendocrine differentiation, and resistance to standard therapies [3, 6, 40]. This phenotype is increasingly recognized as being driven by molecular mechanisms of lineage plasticity, in which the combined loss of RB1 and TP53 in PTEN‐deficient PCa, facilitates a transition from an androgen‐receptor dependent state to an aggressive, poorly differentiated phenotype [41, 42]. While PCas have a comparatively low overall mutational burden [43], we hypothesized that certain recurrent mutations could mediate cellular programs that reshape the local TME [11]. We explored whether the genomic landscape, by grouping the biologically high‐risk tumor suppressor altered cases against those that did not possess such changes could contribute to changes in the TME; namely, in immune cell infiltration, cytokine expression, and mechanisms of immune evasion. Gene expression analysis revealed changes in genes associated with immune response and modifications to the TME. Interestingly, while overexpression of the NDC80 kineticore complex (including NUF2, SPC24, and SPC25) is associated with poor outcomes in many cancers, including PCa, it has also been reported to be associated with Th2 cell infiltration, thereby supporting an immunosuppressive environment [44]. Our observed overexpression of CXCL14 in the high‐risk RB1, PTEN, or TP53 mutant group could be reflective of the recruitment of monocytes in the tumor stroma and polarization to the immunosuppressive and M2 macrophage phenotype or the association with cancer‐associated fibroblasts (CAFs) and endothelial cells which have been shown to express CXCL14 in PCa (reviewed by Tang et al., [45]). Moreover, it has been suggested that CXCL14 can recruit dendritic cells (DCs), which play a critical role in T cell activation. Similarly, increased expression of ARG2 has been associated with an androgen‐regulated immunosuppressive pathway in PCa [46]. Additionally, decreased expression of the tumor suppressor SOCS3 promotes tumor growth by activating the JAK/STAT3 pathway and enhances tumor cell survival and suppression of the anti‐tumor immune response [47]. Collectively, these bulk expression changes in PCa suggest that driver mutations actively shape a complex TME in that is predominantly immunosuppressive. Our DSP results support findings of our bulk transcriptional data that these PCas are predominantly immune‐cold (Figure 2D) [48], which has also been reported in other DSP proteomic profiling studies [49, 50, 51, 52] and reviewed by de Bono et al. [48]. We observed no significant differences in protein expression within either the tumor or the TME compartments based on BCR status (Figure 2E). However, when stratifying the samples according to the biologically high‐risk RB1, PTEN, or TP53 mutant group, we observed a different and more active immune landscape. In the tumor compartment of mutated cases, we observed elevated scores for checkpoint markers, DCs, hematopoietic cells, memory cells, MHC2 cells, proliferation, stromal markers, T cell activation, and T cells (Table 2 and Figure 3E). Similarly, differences in the stromal compartment were seen with higher scores for the mutant cases in antigen presentation, CD8 T cells, DCs, M2 macrophages, memory cells, monocytes, and T cells (Figure 2E). These findings suggest the potential impact that differentially expressed genes such as NDC80, CXCL14, and ARG may have on the immune landscape, as discussed above, while adding the advantage of spatial context through spatial profiling. This spatially defined immune profile suggests that a dynamic and complex TME may exist in tumors with individual or compounded PTEN, TP53, or RB1 mutations. The elevated expression of immunoregulatory and activation‐associated proteins that define these cell classes (Supporting Table 3), within the tumor compartment of the RB1, PTEN, or TP53 mutated group is in keeping with observations from other studies of the immune TME of local and advanced PCa [49, 50, 51, 52]. This dysregulated immune profile shows similarity to those found in neuroendocrine and the aggressive variant PCas for which alterations in one or more of these genes, particularly RB1, are frequent [49]. The canonical immune checkpoint and immune‐modulatory receptors, including CTLA4, PD‐1, LAG3, Tim‐3, and GITR suggest a role for chronic immune activation accompanied by adaptive immune resistance, consistent with infiltrating T cells experiencing repeated stimulation [48]. Elevated CD25, ICOS, CD27, and OX40L may reflect increased T cell activation and potential expansion of regulatory T cells, and CD44 suggests both activated T cells and cancer stem‐like properties in tumor cells [48]. In the TME of cancers with mutations of PTEN, TP53, or RB1, our data further identified the enrichment of M2 Macrophages, CD8 T cells, and T cells (Figure 2E). The presence of CD8 + T cells and increased B2M, a component of MHC class I molecules indicates enhanced antigen presentation and cytotoxic T lymphocyte infiltration, suggesting attempted immune recognition of tumor cells. However, the increased expression of ARG1, CD163, CD68, and CD14 marking M2 Macrophages expansion and activation of myeloid‐derived suppressor cells (MDSCs) and tumor‐associated macrophages (TAMs) [48, 53], contribute to an immunosuppressive environment that could inhibit effective T cell–mediated cancer cell killing. This complex immunological landscape underscores why early PCa with these genetic alterations may eventually lead to recurrence and therapy resistance. Conclusion In summary, we have demonstrated in this small data set that genomic profiling with a targeted pan‐cancer panel and analysis driven in a pathway‐directed fashion together with transcriptional profiling with bespoke signatures could provide improved risk assessment, trigger additional molecular testing, and impact treatment decision making. We acknowledge that the limitations of this study arise from the relatively small sample size of 53 patients, as well as the constraints to the spatial proteomics study, including the number of proteins profiled. However, we demonstrated that the prognostic significance of established genomic markers of poor outcome is supported by both transcriptomic and proteomic analyses, with data from these approaches providing evidence of the complex immune environment and proliferation of intermediate risk PCa. This study provided data that is hypothesis generating and requires future studies in a larger cohort for validation. Author Contributions Study inception and experimental design: Jeremy A. Squire and Jane Bayani. Clinical and pathological review: William Lautert‐Dutra, Jeremy A. Squire, Luiz P. Chaves, Francisco C. Souza, Fabiano P. Saggioro, Felipe S. Avante, and Rodolfo B. dos Reis. Data generation: William Lautert‐Dutra, Cheryl Crozier, Mary Anne Quintayo, Linda Liao, Dan Dion, and Vida Talebian. Data interpretation and analysis: William Lautert‐Dutra, Camila M. Melo, Cheryl Crozier, Mary Anne Quintayo, Linda Liao, Dan Dion, Vida Talebian, Jeremy A. Squire, and Jane Bayani. Manuscript preparation: William Lautert‐Dutra, Camila M. Melo, Cheryl Crozier, Jeremy A. Squire, and Jane Bayani. Manuscript review: all authors. Conflicts of Interest Jane Bayani discloses previous research grant support from Thermo Fisher Scientific and NanoString Technologies. Jane Bayani has the patent application “A Molecular Classifier for Personalized Risk Stratification for Patients with Prostate Cancer” under consideration. Status: PCT, Filing date: June 18, 2021, International Application No.: PCT/CA2021/050837, PCT Application Title: Molecular Classifiers for Prostate Cancer. Previous US Provisional Status: Filing Date: June 18, 2020, US Provisional Patent No. 63/040.692, US Provisional Application Title: Use of Molecular Classifiers to Diagnose, Treat, and Prognose Prostate Cancer. Data Availability Statement The datasets were deposited in the database of Genomic and Phenotypes (dbGaP) under accession numbers phs000178.v11.p8.c1 and phs000915.v2.p2.c1. The original clinical information was deposited in the CbioPortal and GDC Portal. All other data supporting the conclusions of this article are included within the article and its additional files. Additional files related to our retrospective cohort are available at GSE244631 and in GSE317455. Supporting information Supporting File 1 Supporting File 2 Supporting File 3 Supporting File 4 Supporting File 5 Supporting File 6 Supporting File 7 Supporting File 8 Acknowledgments The authors wish to acknowledge Anna Y. Lee, Elizabeth Woroszchuk, and Adam Sunby of Diagnostic Development at the Ontario Institute for Cancer Research for their contributions. This study was supported by FAPESP grant number 2021/15011‐4 (RR and JAS) at the University of Sao Paulo. The Ontario Institute for Cancer Research is supported by funding provided by the Ministry of Colleges, Universities, Research Excellence and Security of the Province of Ontario. Finally, the authors wish to thank the patients for consenting to this research. The Ontario Institute for Cancer Research is supported by funding provided by the Ministry of Colleges, Universities, Research Excellence and Security of the Province of Ontario. JB discloses previous research grant support from Thermo Fisher Scientific and NanoString Technologies. JB has the patent application “A Molecular Classifier for Personalized Risk Stratification for Patients with Prostate Cancer” under consideration. Status: PCT, Filing date: June 18, 2021, International Application No.: PCT/CA2021/050837, PCT Application Title: Molecular Classifiers for Prostate Cancer. Previous US Provisional Status: Filing Date: June 18, 2020, US Provisional Patent No. 63/040.692, US Provisional Application Title: Use of Molecular Classifiers to Diagnose, Treat, and Prognose Prostate Cancer. This study was supported by FAPESP grant number 2021/15011‐4 (RR and JAS) at the University of Sao Paulo. References Associated Data Supplementary Materials Supporting File 1 Supporting File 2 Supporting File 3 Supporting File 4 Supporting File 5 Supporting File 6 Supporting File 7 Supporting File 8 Data Availability Statement The datasets were deposited in the database of Genomic and Phenotypes (dbGaP) under accession numbers phs000178.v11.p8.c1 and phs000915.v2.p2.c1. The original clinical information was deposited in the CbioPortal and GDC Portal. All other data supporting the conclusions of this article are included within the article and its additional files. Additional files related to our retrospective cohort are available at GSE244631 and in GSE317455.