2.1 Study design

Providence SJHC approved this retrospective study under SJHC/SJCI Joint Institutional Review Board (IRB): Universal Consent (Providence Health and Services Portland IRB: JWCI-18-0401) and WIRB Copernicus Group, Inc. (WCG) IRB: MORD-RTPCR-0995. All participants provided written informed consent. This study enrolled a total of 88 patients who underwent robot-assisted laparoscopic prostatectomy and ePLND. All patients were pathologically diagnosed with PCa by two experienced surgical pathologists specialised in urology (D.L.K. and W.M.A.). Patients were excluded from the study if they received previous androgen deprivation therapy (ADT), radiation therapy, or any other treatment for PCa. PCa patients with follow-up periods shorter than one year were excluded. The clinical and pathological information is described in Table 1.

2.2 PCa tissue sample analyses

Formalin-fixed paraffin-embedded (FFPE) tissue blocks were cut into 5 µm-thick sections for hematoxylin and eosin (H&E) staining and microdissection. The microdissected FFPE samples were analysed using the HTG EdgeSeq miRNA whole transcriptome assay (WTA) and HTG EdgeSeq mRNA human transcriptome panel (HTP) assay (HTG). The HTP and WTA assays are extraction-free and targeted arrays that use histopathologically defined single-sections of FFPE and Illumina next-generation sequencing (NGS). The HTP and WTA assays provide a transcriptome panel for 19 398 human mRNA transcripts and 2083 human miR targets, respectively.^33-35 From this point on, we will refer to the data obtained from the HTP or WTA assays as HTP or WTA datasets. For detailed information for sample processing, library generation, data parsing and raw counts, please refer to Supporting information.

2.3 Bioinformatics analysis

Data analyses were performed using R version 4.2.1 (R Core Team 2022, https://www.R-project.org/); ggplot2 (version 3.4.3) was used for data visualisation. The Mann–Whitney U and Fisher's exact tests were used to evaluate differences in pN0 and pN1 patients. Differentially expressed (DE) mRNA and miRs were detected using DESeq2 (version 1.38.3) with a cutoff of a false-discovery rate (FDR) < .05 and absolute Log₂ fold change (FC) > |1.5|. For DE analyses were performed with raw counts that were normalised using the package's default median-of-ratios method, without further transformation of the data. For DE analysis, only LN positivity was included as a factor (pN0 vs. pN1). Kaplan–Meier survival curves were generated in R using packages survival (3.5-7) and survminer (0.4.9). For survival analyses the PCa patients were dichotomised into two groups based on the median expression values or the best cut-off values based on the ROC curve. Correlation analysis of mRNA and miR expression data was conducted in a pairwise manner between a given gene of interest and each miR in the panel, and correlation analysis of mRNA expression data was conducted between each given gene of interest. In both analyses, the Spearman's Rho coefficients and p-values were calculated using the R native cor.test package, and FDR adjustment was conducted by using the p.adjust package. Immune cell fractions were calculated using quanTIseq algorithm.³² For the classification of pN status by gene marker expression, receiver operating characteristics (ROC) curve analysis and plots were generated using the R package pROC (1.18.5), while logistic regression model classifiers were fit using the native glm R function. Model p-values were calculated using the log likelihood ratio test in comparison to the null model, while leave-one-out cross-validation accuracy values were calculated using the R package boot (1.3-28.1). All figures were unified using Adobe Illustrator CC (Adobe) or CorelDraw graphics suite 8X (Corel).

3.1 Patient characteristics and study design

A total of 88 patients diagnosed with primary PCa tumours were included in this retrospective study (Table 1; Figure 1A). There were 44 PCa patients diagnosed with pN1 and 44 patients diagnosed with pN0 status. The median PSA levels at surgery were higher in pN1 than pN0 cases (9.10 ng/mL and 6.75 ng/mL, respectively; p < .05). As expected, pN1 patients showed significantly higher pathological T stage and GG, but no significant differences were observed in age at diagnosis (Table 1). The recurrence rate was higher in pN1 as opposed to pN0 cases (70% vs. 23%; p < .05). The pN1 group showed a significantly shorter RFS (HR = 4.00; 95% confidence interval, 2.10–7.77, p < .0001; Figure S1B).

3.2 Transcriptomic patterns found in primary PCa of patients with confirmed LNM

Previous studies demonstrated the utility of HTG EdgeSeq mRNA and miR assays for the molecular profiling of FFPE tumour tissues.^{29, 30, 35-38} In this study, primary PCa tumours were obtained from patients pathologically confirmed as pN0 or pN1. Then, primary PCa tumours from both groups were profiled using HTP dataset and compared to identify transcriptomic changes associated with the presence of LNM (Figure 1A and Figure S1A). differentially expressed gene (DEG) analysis comparing pN1 versus pN0 PCa samples showed that six genes were significantly upregulated and six were significantly downregulated. The probe for CTAG_family targets multiple genes, and therefore, the probe was removed from further analysis to avoid confusion (Figure 1B; Table S1). Two different gene clusters were identified using correlation analysis: cluster 1 (PLAT [plasminogen activator tissue type], GP2 [Glycoprotein 2], GLO1 [Glyoxalase I] and JAG1 [Jagged canonical notch ligand 1]) and cluster 2 (NPR3 [Natriuretic Peptide Receptor 3], VGLL3 [Vestigial like family member 3], NAT1 [N-Acetyltransferase 1], SEMA3F [Semaphorin 3F], PAH [Phenylalanine hydroxylase] and CHRNA2 [Cholinergic Receptor nicotinic alpha 2 subunit], Figure 1C). To validate the DEGs found in pN1 cases using the HTP dataset, the patients from the TCGA-PRAD were divided into pN0 (n = 312) and pN1 (n = 70, Table 2), and both groups were compared to identify DEGs (Figure 1D). We observed that six (GP2, PLAT, CHRNA2, NPR3, VGLL3 and PAH) out of the 11 DEGs were consistently changing in the TCGA-PRAD dataset (Tables S1 and S2). Next, the mRNA levels of the 11 DEGs found in the HTP dataset were divided into high- and low-expression groups based on the median value of mRNA levels for each gene. Then, each group was evaluated for the number of patients with pN1 status. Of note, only low-NAT1, low-NPR3, low-CHRNA2, low-PAH, low-SEMA3F, or low-VGLL3 levels were associated with an increased number of pN1 cases (Figure 2A–K; Table 3). Then, each of the 11 DEGs found in the HTP dataset was correlated to RFS data obtained for the 88 PCa patients. Of note, pN1 PCa patients showed a shorter RFS than pN0 (Figure S1B). In Kaplan–Meier analyses, five genes were associated with significant differences in RFS (Figure 2L–V; Table 3). Afterwards, the associations between the expression levels of each gene and the number of patients with recurrent and non-recurrent PCa were evaluated. Only low-NAT1, low-NPR3, low-CHRNA2, low-PAH, or high-JAG1 levels were associated with an increased number of recurrent cases (Figure S2A–K; Table 3). Then, the results from the HTP dataset were validated using the RNA-Seq data from the TCGA-PRAD dataset. As expected, patients with pN1 status showed a non-significant trend to shorter progression-free survival (PFS, Figure S1D). Notably, low-CHRNA2, low-NPR3, low-SEMA3F, low-VGLL3, high-GLO1, or high-JAG1 levels were associated with an increased number of pN1 cases (Figure S3A–K). In the TCGA-PRAD dataset, low-CHRNA2 (p = .03), low-PAH (p = .029) and high-JAG1 (p = .049) levels in PCa tissue samples were consistently associated with shorter PFS (Figure S3L–V; Table 2).

The mRNA expression levels of the 11-gene signature were analysed in the 33 tumour types of the TCGA database to assess specificity of the PCa mRNA signature. Of note, primary PCa tumours showed high JAG1, GP2, GLO1, NPR3, VGLL3, CHRNA2 and SEMA3F mRNA levels in primary PCa tissue samples (Figure S4–5). In addition, CHRNA2 and GP2 showed a PCa tissue-specific expression (Figure S4–5). Overall, these results suggest that specific transcriptomic changes found in primary PCa tumours of pN1 cases may be associated with more aggressive primary PCa tumours with higher rates of recurrence, while the downregulation of PAH, NPR3 and CHRNA2 mRNA levels may be more implicated in the metastatic development of PCa LNM.

3.3 Validation of the mRNA signature in PCa primary tumours

The observations in the HTP and TCGA-PRAD datasets were further validated using the RNA-Seq data from GSE220095 dataset,³⁹ which contains FFPE tissue biopsies from primary tumours obtained from 155 PCa patients with clinical annotations (Figure 1A). PCa patients were divided into pN0 (n = 138) and pN1 (n = 17) groups (Table 4). As expected, pN1 PCa patients showed a significantly shorter time to biochemical recurrence (BCR, Figure S1F). The pN1 group showed 401 DEGs compared to the pN0 group (Figure 3A; Table S3). Four overlapping genes (CHRNA2, NPR3, VGLL3 and PAH) were consistently and significantly downregulated in pN1 groups in the three comparisons (Figure 3B–F, Figure S6A–H, and Table S4). Then, the mRNA levels of the four genes (CHRNA2, NPR3, VGLL3 and PAH) in the three datasets were analysed across GG. Notably, the four genes were significantly downregulated in the GG4-5 group compared to the GG1-2 group (Figure 3G–J and Figure S6I–P). In a separate analysis the four genes found in pN1 were categorised based on median mRNA levels into low and high groups. Of note, PCa patients with low mRNA levels for CHRNA2, NPR3 and VGLL3 showed significantly shorter time to BCR in the GSE220095 dataset (Figure S7A–D). Also, PCa patients with low levels of CHRNA2, NPR3 and VGLL3 showed a significantly increased number of cases with pN1 status (Figure S7E–H). These results showed that CHRNA2, NPR3, VGLL3 and PAH are downregulated in tissue biopsies taken from primary PCa tumours of pN1 patients, suggesting that the four-mRNA signature may have potential in the identification of pN1 PCa patients eligible for ePLND.

3.4 A four-gene mRNA signature distinguishes pN1 from pN0 PCa patients

To demonstrate the clinical utility, we determined the performance of each of the four genes in the signature using ROC curves. Overall, lower AUC values were observed in the TCGA-PRAD dataset compared to the HTP or GSE220095 datasets (Figure 4A–O). The highest AUC values were observed for CHRNA2 in the three datasets analysed (Figure 4A–O; Table S5). The four-gene signature showed good performance in the HTP, the TCGA-PRAD, as well as in the GSE220095 datasets (Figure 4E,J,O; Table S5), suggesting a potential value to distinguish patients with primary PCa tumours who have LNM. When PCa patients were divided based on ROC cutoff values (Table S6), instead of the median cutoff values, the associations between miRs and RFS or pN status significantly improved for most of the comparisons (Figure S8).

3.5 Overlapping mRNA signatures in aggressive primary PCa tumours

In this section, we focused on the identification of specific transcriptomic signatures in primary tumours from PCa patients that were recurrent versus non-recurrent. As expected, patients with recurrent PCa showed worse outcomes in the HTP, TCGA-PRAD and GSE220095 datasets (Figure S1C,E,G). Several DEGs were observed in each comparison (Figure S9A–C; Table S7–9); however, only non-SMC condensin II complex subunit D3 (NCAPD3) was commonly found across the three datasets (Figure S10D). Although this is not the focus of this study, functional assays showed that NCAPD3 is activated by the AR signalling and NCAPD3 upregulation promotes PCa progression.^40-42 Of note, NPR3, VGLL3 and PLAT were significantly changing in the HTP and GSE220095 datasets, but not in the TCGA-PRAD dataset (Figure S10D). To summarise, elevated mRNA levels of NCAPD3 may be associated with aggressive primary PCa tumours.

3.6 The miR and mRNA levels were negatively correlated in primary PCa tumours of patients who develop LNM

In this study, the WTA (miR) and HTP (mRNA) datasets provided paired profiles for each PCa tissue sample. Initially, the comparison between the miR profiles of primary PCa tumours from pN1 versus pN0 patients showed 103 miRs DE (Figure 5A, Figure S1H, and Table S10), of which only ten were upregulated in pN1 primary PCa tumours. Then, the profiles of the DE miRs were correlated with the 11 DEGs found between pN1 versus pN0. Only the DE miRs in pN1 group that significantly correlated (> .25 or <−.25) with any of the 11 DEGs in pN1 were included (Table S11). Ten miRs (miR-3188, miR-4632-3p, miR-6729-3p, miR-4745-3p, miR-4791, miR-5582-5p, miR-323a-5p, miR-6799-3p, miR-4695-3p and miR-548h-5p) were significantly downregulated in pN1 and negatively correlated with the five upregulated genes found in pN1 primary PCa (Table S12). On contrary, eight miRs (miR-125a-3p, miR-612, miR-615-3p, miR-663b, miR-664a-5p, miR-1291, miR-3687 and miR-4417) were significantly upregulated in pN1 and negatively correlated with the six downregulated genes found in pN1 (Figure 5B–I; Table 3, and Table S12). The eight upregulated miRs found in pN1 primary PCa tumours were stratified based on the miR detection levels into high and low groups and related to RFS or pN status. Only miR-612 and miR-615-3p were significantly associated with a shorter RFS (Figure 6A–H; Table 5). The number of pN1 cases significantly increased in PCa patients with high levels of any of the miRs found, except for miR-3687 (Figure 6I–P; Table 5). Each of the eight miRs showed good performance to distinguish pN1 from pN0 PCa patients in the WTA dataset (Figure 6Q–X; Table S5). Using ROC cutoff values (Table S6) improved the associations observed between miRs and RFS or pN status (Figure S10).

Specific miRs (such as miR-125a-3p) showed negative correlations with multiple genes NPR3, CHRNA2, NAT1, VGLL3, and SEMA3F, suggesting a common regulatory mechanism (Table S12). Conversely, four genes (NPR3, NAT1, VGLL3, and SEMA3F) were negatively correlated with more than one miR (miR-125a-3p, miR-663b, miR-4417 and miR-3687, Table S12). These results suggest that upregulated or downregulated miRs in pN1 cases may promote LNM by modifying the gene expression output of primary PCa tumours. Therefore, we assessed the TargetScan database to determine predictive miRs targeting any of the 11 mRNA previously identified. We observed that NPR3, which is downregulated in pN1, was a predictive target for miR-125a-3p, miR-663b and miR-664a-5p (Table S13). Of notice, miR-125a-3p, miR-663b and miR-664a-5p were upregulated in pN1 PCa tumours and showed a negative correlation with NPR3 (Figure 6I–P; Table S13). These results suggest that miRs may promote the downregulation of CHRNA2, NPR3, VGLL3 and PAH genes, which are associated with primary PCa tumours that developed LNM.

3.7 Combination of miR/mRNA signatures and PSA or GG levels distinguished primary PCa tumours that developed LNM

Then, the miR and mRNA signatures obtained from the HTP and WTA datasets were combined with the serum PSA levels or GG. Adding the serum PSA levels or GG to the four-gene signature improved the performance of the four-gene signature alone, the serum PSA levels alone, or GG alone in the three datasets (Figure 7A–L; Table S5). Then, we also compared the performance of the eight-miRs as a signature or the eight-miRs combined with serum PSA levels or GG. The eight-miRs signature demonstrated good performance, and the combination with the serum PSA levels or GG slightly improved the performance in the HTP dataset (Figure 7M–O; Table S5). Finally, we weighed the miR and mRNA using log likelihood p-value. The combination of miR-663b, CHRNA2, PAH and serum PSA levels or GG showed a good performance in the HTP dataset (Figure 7P–Q; Table S5). To summarise, this study identified unique miR and mRNA, as well as miR/mRNA signatures that combined with serum PSA levels or GG significantly improved the detection of primary PCa tumours that developed LNM.

3.8 Immune cells of primary PCa tumour microenvironment are associated with pN1 status or RFS

Previous studies have shown that PCa tumours are immune ‘cold’, and therefore, immunotherapies showed no advantage for the treatment of patients with primary PCa.⁴³ Cancer-associated fibroblast, macrophages, natural killer (NK) cells, mast cells and T cells are commonly found infiltrated in primary PCa tumours of patients and are associated with survival and metastasis.⁴⁴ Thus, we focused on understanding the types and fractions of the immune cells in primary PCa and their relation to clinical outcomes. Initially, we evaluated whether IFN-γ score⁴⁵ was a factor associated with recurrence, pN status, or RFS in PCa patients. However, the IFN-γ score showed no association with any of the variables analysed (Figure S11A–C). CD8⁺ and CD3⁺ T cell populations are associated with survival in different solid tumours.³² Of note, neither CD8⁺ nor CD3⁺ T cell populations were associated with pN status, recurrence, or RFS in the HTP dataset (Figure S12A–E). Then, we applied the quanTIseq algorithm³² to the HTP dataset to obtain immune cell fraction estimates (Figure S13A–B). Of the ten immune cell types analysed, only macrophages M1 were significantly upregulated in pN1 cases (p = .01, Figure S14A–K). Interestingly, low-macrophage M2 macrophages or low-NK cell fractions were significantly correlated with shorter time for RFS (Figure S13C–F). These results suggest that immune cell types such as M2-macrophages or NK cells are associated with shorter RFS. Only M1-macrophages showed increased levels in primary PCa tumours that developed LNM.