Prostate‐derived circulating microRNAs add prognostic value to prostate cancer risk calculators

Abstract

Prostate cancer is the second leading cause of malignancy‐related deaths among American men. Active surveillance is a safe option for many men with less aggressive disease, yet definitively determining low‐risk cancer is challenging with biopsy alone. Herein, we sought to identify prostate‐derived microRNAs in patient sera and serum extracellular vesicles, and determine if those microRNAs improve upon the current clinical risk calculators for prostate cancer prognosis before and after biopsy. Prostate‐derived intracellular and extracellular vesicle‐contained microRNAs were identified by small RNA sequencing of prostate cancer patient explants and primary cells. Abundant microRNAs were included in a custom microRNA PCR panel that was queried in whole serum and serum extracellular vesicles from a diverse cohort of men diagnosed with prostate cancer. The levels of these circulating microRNAs significantly differed between indolent and aggressive disease and improved the area under the curve for pretreatment nomograms of prostate cancer disease risk. The microRNAs within the extracellular vesicles were the most informative and improved the AUC to 0.739 compared to the existing nomogram alone, which has an AUC of 0.561. The microRNAs in the whole serum improved it to AUC 0.675. In summary, quantifying microRNAs circulating in extracellular vesicles is a clinically feasible assay that may provide additional information for assessing prostate cancer risk stratification.

INTRODUCTION

Prostate cancer (PCa) is the most diagnosed cancer and the second most common cause of death due to malignancy among men in the United States (US), with one in eight men in the US being diagnosed during their lifetime (Siegel et al., 2020). Because many men with PCa will die from other causes, a current challenge is determining which patients will develop aggressive, metastatic PCa, and which patients have indolent, localized PCa (Preventive Services Task Force et al., 2018). The 5-year relative survival rate drops from 100% in patients with localized PCa to 32.3% in patients with metastatic disease (Cancer Facts & Figures 2023, 1930), lending urgency to developing improved prognostic strategies.

PCa is diagnosed by a multiple-core prostate biopsy with aggressiveness determined by the Gleason pathological grading system which includes a primary and secondary grade, ranging from 3 to 5, with 5 representing anaplastic cells (Epstein et al., 2016). The grades are classified into a clinical Grade Group (GG) ranging from 1 to 5, with GG1 as the lowest grade PCa (Gleason 3+3) and GG5 as the highest grade PCa (Gleason 4+5, 5+4, or 5+5) (Epstein et al., 2016). The challenge with biopsies is that even with MRI guidance, they are a somewhat random sampling of the peripheral zone, which can lead to false-negative results. Thus, a low Gleason grade on biopsy tissue does not rule out the presence of a high-grade tumour that was not biopsied (Quintana et al., 2016).

For PCa prognosis after biopsy, urologists combine GG with other clinical parameters to stratify patients into risk groups (Mohler et al., 2019). These risk groups predict the risk of biochemical recurrence (BCR), a rise in serum prostate-specific antigen (PSA) after primary treatment, which is indicative of metastases. Risk group and life expectancy are used to guide treatment recommendations in a shared decision process with the patient. Treatment options for localized PCa include active surveillance, radical prostatectomy (RP), radiation therapy (RT), cryotherapy and high-intensity–focused ultrasound (Mohler et al., 2019). Compared with active surveillance, RP and RT decrease PCa metastasis and mortality rate (Aas et al., 2020; Hamdy et al., 2016; Lei et al., 2015), but are associated with adverse side effects such as urinary incontinence and erectile dysfunction (Sanda et al., 2008). Moreover, biopsy GG does not always align with GG of the RP as 30%–45% of patients have upgraded and 10% have downgraded GG at RP (Quintana et al., 2016), contributing to the difficulties faced by patients and clinicians when making PCa treatment choices.

Multiple nomograms and tissue-based tests have been developed to provide personalized risk predictions at various stages of diagnosis (Ankerst et al., 2018; Cooperberg et al., 2011; Kattan et al., 2003; Mohler et al., 2019). The Prostate Biopsy Collaborative Group (PBCG) risk calculator is assessed before a biopsy to predict the likelihood of a positive biopsy (Ankerst et al., 2018). Following a PCa-positive biopsy, the Cancer of the Prostate Risk Assessment (CAPRA) nomogram, is used to aid in treatment decisions by predicting adverse pathology (AP) of the prostate (Cooperberg et al., 2011, 2015). These nomograms take into account family history, serum PSA and other clinical parameters. Multiple biopsy tissue-based prognostic tests have been developed to further guide treatment decisions (Cucchiara et al., 2018; Quintana et al., 2016), yet Salami et al. found significant differences between patient-matched GG1 and GG4 prostate cores when comparing three tissue-based tests, indicating that transcription profiles from lower-grade biopsy cores do not capture the profiles from higher-grade cores (Salami et al., 2018).

MicroRNAs (miRs) are highly stable in circulation (Mitchell et al., 2008) and are promising prognostic biomarkers for PCa in prostate tissues and multiple biological fluids as shown by our group and others (Cozar et al., 2019; Ghafouri-Fard et al., 2020; Mihelich et al., 2015). Our group previously examined 14 serum miRs in a cohort of 150 men with high-grade PCa, low-grade PCa or benign prostate hyperplasia and determined they were predictive of indolent PCa or a cancer-free status with a negative predictive value of 100% and positive predictive value of 58.8% (Mihelich et al., 2015). Circulating miRs are often enclosed within extracellular vesicles (EVs) (Cozar et al., 2019), a heterogeneous group of vesicles with lipid bilayer membranes which includes exosomes (van Niel et al., 2018). EVs are released from cells and have functions in normal physiology and in various disease processes (Aalberts et al., 2014; Linxweiler & Junker, 2020). Due to the stability and functionality of miRs in EVs, they have gained substantial attention as liquid biomarkers and mediators of various cancers, including PCa (Saber et al., 2020; Wang et al., 2020).

Herein, we sought to identify prostate-derived miRs in EVs from patient cells and tissues and then to determine the prognostic ability of serum and serum EV miRs in a diverse cohort of 203 treatment-naïve patients with biopsy-confirmed PCa. The premise was that miRs present in normal prostate may emerge in the circulation in PCa, like PSA, but have more specificity than PSA. The primary endpoint was assessing the ability of circulating miRs to add prognostic value to the current clinical standard, the CAPRA nomogram, for determining risk of adverse pathology status at RP. The secondary endpoint involved determining if the circulating miRs added to the prognostic value of the PBCG nomogram for discriminating low-grade from high-grade PCa at biopsy.

METHODS

Patient-derived prostate tissue and cell cultures

Patient prostate cells and explants were derived from patient radical prostatectomy specimens via informed consent prior to surgery (UIC Internal Review Board-approved protocol #2006-0679). The tissue was either sliced for explant culture or dissociated for cell propagation as previously described by our lab (Richards et al., 2019). Briefly, dissociated cells were grown in Prostate Cell Growth Media (PrEBM) (Lonza, Basel, Switzerland) or MCDB105, for prostate epithelial (PrE) and stromal (PrS) cells, respectively. Cell type authentication was by RT-qPCR; epithelial-specific markers were keratin 5 (KRT5) (F: CCATATCCAGAGGAAACACTGC, R: ATCGCCACTTACCGCAAGC) and keratin 18 (KRT18) (F: CACAGTCTGCTGAGGTTGGA, R: CAAGCTGGCCTTCAGATTTC), and the stromal-specific markers were vimentin (VIM) (F: CGAAAACACCCTGCAATCTT, R: TCCTGGATTTCCTCTTCGTG) and Androgen Receptor (AR) (F: CCAGGGACCATGTTTTGCC, R: CGAAGACGACAAGATGGACAA). For explant cultures, 300 µm tissue slices were generated using a Tissue Slicer (Model MD6000, Alabama R&D, Munford, AL) and cultured on titanium grids that were constantly rotated in and out of PrEBM containing 50 nM R1881.

EV isolation from cells

Prior to EV isolation, cells were cultured in EV-free medium (bovine supplement components omitted) for 48 h. PrE and PrS EVs were isolated from conditioned medium using differential ultracentrifugation; 300 × g for 10 min (min), supernatant centrifuged at 2000 × g for 20 min, supernatant was ultracentrifuged at 10,000 × g for 30 min, supernatant ultracentrifuged at 100,000 × g for 16 h to pellet EVs. The EV pellet was washed in 1X phosphate-buffered saline (PBS) (no magnesium or calcium) (Corning Inc., Corning, NY), ultracentrifuged for 70 min at 100,000 × g and resuspended in 100 µL 1X PBS. All centrifugation and ultracentrifugation steps were performed at 4°C (Théry et al., 2006).

EV isolation from tissues

EVs were isolated from 2.5 mL of 48 h conditioned media from tissue slice explants using ExoQuick-TC (System Biosciences, Palo Alto, CA) per the manufacturer's protocol. The EV pellet was resuspended in 100 µL 1X PBS (no magnesium, no calcium) or purified via a bipartite resin column according to the ExoQuick-TC ULTRA protocol.

EV isolation from serum

Serum was collected in BD gold-top (serum-separator) vacutainers, incubated for 30 min at room temperature (RT), centrifuged at 1315 × g for 10 min at 4°C. RNase inhibitors were added to the separated serum; 40 units/µL RNaseOUT (Life Technologies, Carlsbad, CA) and 20 units/µL SUPERase-IN (Life Technologies, Carlsbad, CA). EVs were isolated from 500 µL of fresh serum using the ExoQuick protocol (System Biosciences, Palo Alto, CA). Remaining serum aliquots were snap-frozen and stored at −80°C.

Nanoparticle tracking analysis (NTA)

EV suspensions were run on the NanoSight NS300 according to the instrument protocol (MAN0541-01-EN-00, 2017). For each sample, three 30-s videos were captured using a 488 nm laser and a syringe pump speed of 100 (Malvern Instruments, United Kingdom) and automatically analysed using the built-in software NTA 3.2 (Malvern Instruments, Westborough, MA) using a detection threshold of 5. For the particle concentration and size analyses, particle counts were binned in 10 nm increments. The results from the control sample (PBS) were subtracted from each sample run to determine the concentration of EVs within each 10 nm size bin.

Transmission electron microscopy (TEM)

EVs suspended in 10–15 µL 1X PBS was deposited onto a 300-mesh Formvar/Carbon-coated copper EM grid and fixed with 2% Uranyl acetate solution (heavy metal stain). The grid was dried and examined immediately via TEM. EV samples were examined with the JEOL JEM-1400F transmission electron microscope operating at 80 kV. Digital micrographs were acquired using an AMT NanoSprint1200-S camera and AMT software (version 7.01).

CD63 ELISA

EV suspensions were diluted and 45 µL was examined with the ExoELISA-ULTRA CD63 Kit according to the manufacturer's instructions (System Biosciences, Palo Alto, CA). The sample absorbance minus the blank wells was used to determine concentrations based on the standard curve included in the kit.

Western blot

The EV pellet was resuspended in RIPA buffer with protease/phosphatase inhibitor (Cell Signaling Technology, Danvers, MA, USA). Protein (20 µg) was loaded in Laemmli buffer, run on a 4%–12% Bis-Tris electrophoresis gel (Life Technologies, Carlsbad, CA), and transferred to polyvinylidene difluoride (PVDF) membrane (Merck Millipore, Carrigtwohill, CO). Blot was blocked with 5% dry milk in Tris-buffered saline plus 0.05% Tween (TBST) and incubated overnight at 4°C with 1:1000 primary antibody (CD63 and CD81, rabbit anti-human, EXOAB-CD63A-1/EXOAB-CD81A-1, System Biosciences, Palo Alto, CA) in TBST. Membranes were washed, incubated in 1:20,000 secondary antibody (Goat Anti-Rabbit HRP) and visualized with SuperSignal West Atto Ultimate Sensitivity Chemiluminescent Substrate (Thermo Fisher Scientific, Waltham, MA) on the iBright 1500 (Thermo Fisher Scientific, Waltham, MA).

MicroRNA RNAseq library preparation and sequencing

PrE and PrS cells were cultured to 70% confluency, conditioned media was collected and EVs were isolated by differential ultracentrifugation, as described above. PrE and PrS cells were lysed, and RNA was collected using QIAzol (Qiagen, Germany). Tissue slices were homogenized in QIAzol using the BeadRuptor 4 (OMNI International, Bedford, NH). RNA was isolated from all intracellular samples with the miRNeasy Mini Kit (Qiagen, Germany) and from EVs using the Serum/Plasma miRNeasy Kit (Qiagen, Germany). MiR quantification from all samples was completed with microRNA Qubit (Thermo Fisher Scientific, Waltham, MA).

MiR libraries were generated from 100 to 150 ng of cell using the QIAseq miRNA Library Kit according to the protocol (Qiagen, Germany). Library quality control and concentrations were determined using a High Sensitivity D1000 ScreenTape with TapeStation Analysis Software A.02.02 (Agilent Technologies, Santa Clara, CA) Libraries (10 µM) were submitted to the Roy J. Carver Biotechnology Center at the University of Illinois at Urbana-Champaign, where they were titrated for equal loading using the MiSeq followed by sequencing on a HiSeq 4000 with 100 nucleotide forward single reads.

MicroRNA differential expression analysis

Raw FASTQ sequencing reads were inputted into the GeneGlobe Data Analysis Center (Qiagen, Germany). Reads were trimmed using cutadapt (Martin, 2011), which removed the 3′ adapter and low-quality bases. MiR sequences less than 16 base pairs (bp) and UMI sequences less than 10 bp were removed. All samples passed the initial quality control checkpoints in the GeneGlobe Data Analysis Center. Reads were then mapped to miRBase (version 22) using bowtie (Langmead et al., 2009), which allows up to two mismatches. Raw counts were generated by calculating the number of UMIs mapped to each miR sequence. Downstream analyses were performed using R version 3.4.1 (Team, R. C. 2013).

PrE, PrS, tissue slice (TSC), PrE EV, PrS EV and TSC EV samples were initially grouped together with Serum EV samples, and piwi-interacting RNA (piRNA) were removed. Next, miRs were filtered such that only miRs with at least five raw reads in at least 50% of all samples were retained, leaving 2114 of the original 2523 miRs. The data were then separated into the following four datasets: (1) PrE EV, PrS EV, TSC EV, Serum EV; (2) PrE EV, PrS EV, TSC EV; (3) Serum EV; and (4) PrE, PrS, TSC. For each individual dataset, a DGEList object was created in EdgeR (Robinson et al., 2010), normalization factors were calculated using the default trimmed mean of M-values (TMM) between each pair of samples, counts per million (CPM) function with or without log2 scaling was used to generate CPM expression levels of each miR, and files were saved. Log2 scaled CPM data were used to generate all heatmaps and principal component plots (PCA) using the heatmap3, ggplot2 and cowplot packages (Wickham, 2009; Zhao et al., 2014). Differential expression analyses were performed using EdgeR. For each dataset, a DGEList object was created, normalization factors were calculated using TMM, and genewise statistical tests were conducted using the estimateDisp, glmQLFit, glmQLFTest and topTags functions to identify genes differentially expressed between any of the groups within each dataset. For heatmaps, only differentially expressed genes with a statistically significant adjusted p-value ≤0.01 by Benjamini and Hochberg method were plotted (Benjamini & Hochberg, 1995; Sun et al., 2015).

Prostate cancer patient cohort description

Serum was collected from two UIC Urology PCa patient cohorts; one cohort included patients on active surveillance as part of the Engaging Newly Diagnosed Men About Cancer Treatment Options (ENACT) trial (IRB protocol # 2015-1294), and the second cohort included patients who had chosen RP treatment for PCa (Prostate Cancer Blood Biorepository Resource [PCBBR], IRB protocol # 2017-0807). ENACT trial patients were 76 years or younger with a recent diagnosis of very low-, low-, or low-intermediate risk PCa. The only inclusion criterion for the PCBBR cohort was the decision to undergo RP. Due to the differences in inclusion criteria, the ENACT cohort consisted of lower-risk PCa patients (biopsy Gleason grade groups 1–2), and the PCBBR cohort consisted of relatively higher-risk PCa patients (biopsy Gleason grade groups 1–5). Age, PSA level, self-declared race, biopsy Gleason grade group, and pathological Gleason grade group data are included for each cohort. For the ENACT patient cohort, serum was collected at least 8 weeks post-biopsy and prior to RP treatment (if the patient terminated active surveillance treatment). Serum was collected at least 1 month after biopsy but prior to RP treatment for the PCBBR cohort. Serum samples were immediately processed upon collection, separated into 500 µL aliquots, snap-frozen and stored at −80°C. EVs were isolated immediately from serum and stored in liquid nitrogen.

Custom reverse-transcription quantitative polymerase chain reaction (RT-qPCR) miR panel

Five µL RNA was reverse transcribed to cDNA using the miRCURY LNA RT Kit (Qiagen, Germany). MiRs (61 plus 3 controls) were quantified on a custom PCR plate using the miRCURY LNA miR SYBR Green PCR Handbook (Qiagen, Germany). PCR (40 cycles) was run according to the manufacturer's instructions.

Normalization and filtering of qPCR values

NormFinder normalization values were generated by Gene Globe. The normalization panel for the serum samples, as determined by NormFinder, included hsa-miR-103a-3p, hsa-miR-107, hsa-miR-24-3p, hsa-miR-30c-5p, hsa-miR-93-5p, hsa-miR-19b-3p, hsa-miR-222-3p, hsa-miR-27a-3p, hsa-miR-23a-3p and hsa-let-7b-5p. The normalization panel for the serum EV samples, as determined by NormFinder, included hsa-miR-107, hsa-miR-24-3p, hsa-miR-30c-5p, hsa-miR-93-5p, hsa-let-7i-5p, hsa-miR-222-3p, hsa-miR-27a-3p, hsa-miR-23a-3p, hsa-miR-21-5p and hsa-miR-27b-3p. Because each normalization factor was based on the average of 10 miRs, these miRs were included in the analysis for significance based on outcomes.

The call rate per sample was computed as the fraction of miRs with Ct lower than 33, and samples with a call rate <50% were removed from the dataset. Similarly, miRs with a call rate <60% were removed. Normalized delta-Ct values were computed and miRs with Ct ≥33 were set to the minimum normalized expression over all other samples for the same miR. MiRs from serum and extracellular vesicles (EVs) were processed independently in the identical manner.

Statistics—Random forest model

Statistical analyses were completed at the UIC Research Resource Center Bioinformatics Core. The following binary outcomes were considered in preparing the models: (1) Biopsy Gleason grade group 1 (low-grade) and Gleason grade group ≥2 (high-grade) and (2) adverse pathology or non-adverse pathology at RP. For each binary outcome, samples with missing values were removed, and t tests were run for each miR using data from serum and EVs separately. MiRs with a p-value <0.05 were retained as features in the random forest model. We also included clinical features for different models: PBCG high-grade, low-grade and negative biopsy percentages were included independently or in combination with miRs for the prediction of Gleason grade group outcomes. CAPRA score was included independently or in combination with miRs for the prediction of adverse pathology.

For each outcome and feature dataset, we trained and tested the performance of the random forest model using the randomForest() function in the randomForest package in R (Liaw & Wiener, 2002) with default parameters. Leave-one-out cross-validation was performed, predicting the outcome of each sample based on a model trained on all other samples. ROC and AUC statistics were computed based on cross validation predictions using the roc() function in the pROC package in R (46). Confidence intervals of AUC statistics were computed via bootstrapping with 500 bootstrap replicates using the ci.auc() function. p-values comparing the pairwise performance of different models were computed using the Wilcoxon signed rank test between the predicted numerical values from each model.

Study approval

Patient specimens for this study were collected by three UIC IRB-approved protocols, one for prostate tissue/cells and two for sera, #2006-0679, #2015-1294 and #2017-0807.

Data availability

RNA sequencing and RT-qPCR files are available at GEO Accession numbers: GSE228062, GSE228371, GPL33285. R-code is available at .

RESULTS

Distinct cell-type specific miR expression profiles for prostate-derived cells, tissues and EVs

The first goal was to identify prostate-derived miRs. The prostate is composed of multiple cell types, with a fibromuscular stroma surrounding the glandular epithelium from which cancer arises. To account for these multiple cell types, the miR profiles of patient prostate tissues, patient-derived primary prostate epithelial cells (PrE), patient-derived primary prostate stromal cells (PrS), and their EVs were characterized by small RNA next-generation sequencing (small RNAseq) (Figure 1a). EVs were included as they are a mechanism for the prostate to shed miRs into the circulation. Ex vivo prostate tissue slice cultures (TSC) were used as the source of whole tissue miRs and TSC EVs. EVs were isolated from conditioned media from patient-derived, TSCs, PrE, and PrS samples. We used four different patient sources for each of the sample types (12 unique donors in total), to minimize the potential for interpatient heterogeneity to bias miRs identified as coming from one cell type or another. Cell type was verified by gene expression of known markers (Figure S1).

[IMAGE OMITTED. SEE PDF]

TSC, PrE and PrS EVs were characterized by Nanosight Tracking Analysis (NTA), CD63 ELISA and transmission electron microscopy (TEM) (Figure 1b–d). Nanosight showed enrichment of EVs around 100 nm, consistent with exosomes, as well as other peaks. The EV peaks varied between the samples, with PrE EVs showing one large peak, and PrS and TSC EVs showing more heterogeneity with multiple size peaks (Figure 1b). EVs were CD63-positive in all samples (Figure 1c), and TEM of TSC, PrE, and PrS EVs showed sizes of 50–200 nm and a classic cup-shaped morphology (Figure 1d).

Small RNA libraries were created from the intracellular RNAs and the EVs, sequenced and analysed using EdgeR (38). Principal component analysis (PCA) of the counts showed pronounced clustering of miRs by sample type. This was true whether they were isolated from the cells/tissues or from the EVs (Figure 1e,f). PrS4 and PrS EV4, was the exception, as this sample did not cluster tightly with the other three PrS samples and clustered more closely with the PrE and PrE EV samples, respectively.

Unbiased hierarchical clustering heatmaps of miRs showed grouping by sample type (q ≤ 0.01, (Figure 1e,f, Tables S1–S2), which was similar to the PCA. Differential expression analysis in EdgeR identified common miRs and those with distinct expression in TSC, PrE and PrS samples and EV samples. MiRs were considered ‘detectable’ with at least 500 normalized counts per million (cpm). For intracellular miRs, TSC expressed the most miRs and the majority of cell-type specific miRs were also detected in the TSCs (Figure 1g and Table S3). Similarly, the TSC EVs had the most detectable miRs (Figure 1g and Table S4). RT-qPCR was used to validate a few of the cell type-specific miR expression profiles observed by small RNA-seq. Hsa-let-7a-5p was robustly expressed in all sample types (PrE, PrS, TSC and EVs), supporting the small RNAseq (Figure 1h). Hsa-miR-203a-3p was confirmed to be specific to PrE/PrE EVs and TSC/TSC EV (Figure 1h) whereas hsa-miR-199a-5p was specific to PrS/PrS EV and TSC/TSC EV (Figure 1h). MiR profiling showed that the stroma and epithelium have distinct EV miR populations and that almost all stromal and epithelial EV miRs were present in the EVs from ex vivo prostate tissue slices, which contain both epithelial and stromal components.

Prostatic miRs within serum extracellular vesicles as biomarkers in prostate cancer patients

EVs were isolated freshly from patient sera. NTA and TEM showed the serum EV size range was 50–300 nm (Figure 2a,b). Serum EVs expressed CD63 and CD81 but not GRP78 (intracellular protein) by western blot (Figure 2c). Serum miRs are known to be exceptionally stable (22), but the stability of serum miRs packaged in EVs has not been well described. To mimic potential delays and storage conditions that could occur in the clinic, EVs were immediately isolated or stored at 4°C, −20°C, or −80°C for 7 days. CD63 levels remained unchanged after a 7-day delay in EV isolation (Figure 2d) and the expression levels of known serum miRs remained consistent in all conditions (Figure S2).

[IMAGE OMITTED. SEE PDF]

To identify potential biomarkers for PCa, serum EV miR expression profiles from PCa patients (N = 20) (NGS in Table 1) were examined by small RNAseq. These patients did not overlap with the donors for the patient-derived models shown in Figure 1. PCA showed no clustering by the pathological Grade Group, but only six patients were higher than GG2 (Figure 2e, Table 1). MiRs from the serum EV miRs were compared to the profiles from the TSC EVs, PrE EVs and PrS EVs (Figure 2f) to identify potential prostate-derived miRs in the serum. There were 18 detectable miRs in common between serum EV, TSC EV, PrE EV and PrS EV (Figure 2f). By RT-qPCR, hsa-let-7a-5p and hsa-miR-199a-5p were present in serum EVs and hsa-miR-203a-3p was not detected, validating the small RNAseq findings (Figure 2g).

TABLE 1 Patient cohort characteristics.

Characteristics	NGS	ENACT	PCBBR
Age (Median with range)	63 (56–73)	62 (47–76)	62 (45–77)
PSA ng/mL (Median with Range)	6.15 (0.2–39.8)	5.58 (0.47–14.17)	7.05 (3–735)
Race, N (%)	African American	11 (55)	95 (67.4)	33 (53.2)
	Caucasian	5 (25)	25 (17.7)	15 (24.2)
	Hispanic	3 (15)	19 (13.5)	10 (16.1)
	Asian	0	2 (1.4)	0 (0)
	Multiple Races/Unknown	1 (5)	0 (0)	4 (6.5)
Biopsy Gleason grade Group, N (%)	1	2 (10)	119 (84.4)	7 (11.3)
	2	12 (60)	22 (15.6)	34 (54.8)
	3	3 (15)	0 (0)	11 (17.7)
	4	2 (10)	0 (0)	7 (11.3)
	5	1 (5)	0 (0)	3 (4.8)
Biopsy to RP Gleason grade comparison*, N (%)	Upgraded	NA	10 (40.0)	17 (29.3)
	Downgraded	NA	0 (0)	13 (22.4)
	Consistent	NA	15 (60.0)	28 (48.3)
Adverse pathology, N (%)	Yes	NA	8 (32.0)	36 (58.1)
	No	NA	17 (68.0)	20 (32.2)
	Unknown	NA	0 (0)	6 (9.7)

Whole serum and serum EV microRNAs add prognostic value to pre-treatment clinical nomograms

A prospective study was designed to examine the prognostic ability of post-biopsy serum miRs in UIC PCa patients (N = 203) (Figure 3a and Table 1). Serum was prospectively collected from two PCa patient cohorts with same-day EV isolation. As the adverse pathology classification of RP resections is a major predictor of disease recurrence after RP, the primary goal of this study was to identify miRs in presurgical serum that are prognostic for RP tumour pathology. Hence, the custom miR panel was run on the serum (N = 81) and serum EVs (N = 62) from patients who later underwent RP. Adverse pathology is defined as Gleason Grade Group 4 or 5 and/or extraprostatic disease at RP and is a hallmark of aggressive disease that is likely to progress (Kozminski et al., 2016).

[IMAGE OMITTED. SEE PDF]

Based on the RNA-seq data and current literature, we created a custom qPCR panel to quantify 61 miRs in whole serum and serum EV samples. The miRs included in the panel were the 14 miRs from our previously reported Serum miR Score (Mihelich et al., 2015), 20 new miRs from the small RNAseq analysis, and 27 miRs that were consistently reported in the literature as prognostic serum biomarkers for PCa (Bhagirath et al., 2018; Brase et al., 2011; Cheng et al., 2013; Cochetti et al., 2016; Fang et al., 2019; Haldrup et al., 2014; Kolluru et al., 2018; Kozminski et al., 2016; Li et al., 2016; Lieb et al., 2018; Lodes et al., 2009; Lyu et al., 2019; Mitchell et al., 2008; Selth et al., 2012, 2013; Singh et al., 2014; Tinay et al., 2018; Wa et al., 2019; Wach et al., 2019; Wang et al., 2014; Zhang et al., 2013; Zidan et al., 2018) (Figure 3b). Potential miR biomarkers were selected from the small RNAseq by expression level and detection in multiple samples. First, miRs with robust expression (>500 counts) in prostate-derived EVs (PrE, PrS, TSC) and serum EVs, were included in the panel. This resulted in nine new miRs and one of the miRs from the prior Serum miR Score. Next, we included 12 miRs that were robustly expressed in two of the prostate samples and serum EVs, eight were new and four were in the Serum miR Score. Lastly, we included four miRs that were very robustly detected (>5000 counts in one of the prostate samples and detectable (>100 counts) in the serum EVs.

MiRs were quantified and analysed for significant differences between patients with no adverse pathology and patients with adverse pathology in serum and serum EV samples. These significantly different miRs were determined and included in the random forest model analyses. CAPRA score, a clinical post-biopsy nomogram, was used for comparison (Cooperberg et al., 2009). The outputs for CAPRA are low risk (score 0–2), intermediate risk (score 3–5) and high risk (score 6–10) (Cooperberg et al., 2005). Random forest models showed that serum miRs and serum EV miRs added to the prognostic power of CAPRA for adverse pathology. Seven serum miRs significantly improved upon the CAPRA score AUC for adverse pathology, 0.675 (95% CI: 0.547–0.799), compared to the CAPRA score alone, with an AUC of 0.516 (95% CI: 0.379–0.643, p = 0.010) (Figure 4a,c and Table S5). 19 serum EV miRs were significantly different in patients with adverse pathology and improved upon the CAPRA model (Figure 4b,c and Table S5). Adding serum EV miRs to the CAPRA model increased the AUC to 0.739 (95% CI: 0.580–0.900) (p = 0.010). Five of the seven serum miRs were also significant in the EVs, with miRs 30a-5p and 122–5p, being unique to the serum.

[IMAGE OMITTED. SEE PDF]

The miRs from the prior ‘Serum miR Score’ from Mihelich et al. (2015), were analysed separately to determine the reproducibility of the miRs when combined with CAPRA to predict AP (Figure S3). The AUCs for AP were 0.713 and 0.662, for the EV miRs and serum miRs, respectively. Although both of these AUC results were better than CAPRA alone, they were not as high as the AUCs for the miRs selected from the overall analyses described above.

Whole serum and serum EV microRNAs significantly differ by biopsy Gleason grade group

Prebiopsy Gleason grade group differentiation was a secondary analysis, as serum samples were collected from patients at least 1 month post-biopsy. All patients were treatment-naïve; therefore, this analysis was completed under the assumption that the serum and serum EV miR landscapes were the same pre- and post-biopsy. PBCG, a clinically utilized prebiopsy risk calculator, was used for comparison to classify high-risk PCa or low-risk PCa by biopsy (15). Random forest models showed that serum miRs and serum EV miRs added to the prognostic power of PBCG in distinguishing between low- and high-grade PCa prebiopsy. 15 serum miRs (Figure 4d) and 13 serum EV miRs (Figure 4e) differed between patients with low-grade and high-grade PCa. PBCG combined with the significant serum miRs had an AUC of 0.705 (95% CI: 0.631–0.783) compared to PBCG risk alone (AUC of 0.600, 95% CI: 0.521–0.686) in patients with low-grade and high-grade PCa (p = 0.164) (Figure 4f and Table S6). PBCG combined with the 13 significant serum EV miRs had an AUC of 0.695 (95% CI: 0.589–0.792) compared to PBCG risk alone, which had an AUC of 0.600 (95% CI: 0.521–0.686) in patients with low-grade and high-grade PCa (p = 0.080) (Figure 4f and Table S6). Although there were serum and serum EV miRs that were significantly different in patients with low-grade and high-grade PCa at biopsy, the miRs combined with PBCG did not reach a statistically significant difference in AUCs compared with PBCG alone.

DISCUSSION

In this study, we report miR profiling of patient-derived prostate cells and tissues and the extracellular vesicles released from these sample types for the first time. These prostate-derived miRs were present in the serum and serum EVs of PCa patients, leading us to develop a miR panel with prognostic value for aggressive PCa using serum collected after prostate biopsy. We determined that miRs within whole serum and serum EVs added prognostic value to the CAPRA nomogram for adverse pathology and that serum EV miRs were more prognostic than whole serum miRs.

Initial characterization of prostate-derived miRs by small RNA-seq revealed new cell-type specific expression of several miRs. Given the regulatory role of miRs in cell fate (Park et al., 2010), it was unsurprising that the miR profiles of patient-derived prostate samples showed clustering based on cell type (stromal vs. epithelial) and sample type (cell vs. whole tissue slices). Kumar et al. previously identified prostate stromal- and epithelial-specific miRs using microdissection of prostate tissue (Kumar et al., 2018). Both Kumar and our study identified miR-141 as epithelial-specific and miR-143-3p as highly expressed in stromal samples. We newly identified the miR-199 family as stromal-specific, which has not been previously reported in the prostate. The miR-199 family has been shown to have low expression in prostate tumours (Zhong et al., 2017). With the new knowledge that miR-199 is predominantly expressed in the prostate stroma, these reported changes in miR-199a expression in PCa compared to benign prostate tissue may be due to low stromal cells in areas of tumour compared to benign (Zhong et al., 2017).

One unexpected finding was that the distribution of the EV sizes varied between the prostate-derived and serum samples. EVs from PrE (epithelial cells) and sera had a large main peak centred around 100 nm, which is consistent with enrichment for exosomes. In contrast, PrS (stromal cells) and TSCs had multiple peaks in addition to the main one at 100 nm. This may be because PrE cells are a pure population, whereas PrS and TSC have a mixture of cell types that may produce differently sized EVs. We cannot draw conclusions given the limited samples run on the Nanosight, but future work may look into the significance of EV size diversity.

The cargo sorting of miRs into EVs was initially thought to be a random process that allowed cells to dispel molecular waste (Yuana et al., 2013). However, multiple studies have determined that the sorting process is highly selective, with the secreted EV miRs remaining functional after being transferred to another cell (Melo et al., 2014; Janas et al., 2015). In our study, robust intracellular cell type-specific miRs did not always result in high EV miRs. Only 24 of the 107 robustly expressed epithelial intracellular miRs were also highly expressed in epithelial EVs. For the stromal compartment, 50 of the 69 robustly expressed stromal miRs overlapped with stromal EV miRs. These findings support the selective packaging of miRs into EVs, particularly in the epithelial cells, as not all intracellular miRs were found in EVs and not all EV miRs were highly expressed intracellularly (Liu et al., 2021).

Pretreatment serum miRs and serum EV miRs added to the prognostic value of the CAPRA nomogram for the adverse pathology outcome, which was the primary clinical outcome of this study. These new findings improve upon our prior study in which the predictive miRs in the ‘Serum miR Score’ had only provided additional predictive information to patients with low-risk PCa (Mihelich et al., 2015). In this current study, serum and serum EV miRs were significantly different in patients with AP, indicating their potential to provide clinical information to patients with both low- and high-risk PCa before making treatment decisions. Additionally, the relative expression levels of the significantly different miRs in this new panel were both higher and lower in patients with AP, compared to all 14 miRs having higher expression in low-risk PCa in our prior study. A strength of this new study is that we included a racially diverse patient cohort and a full range of Gleason grade groups as opposed to the previous study cohort of only white men and no intermediate-risk patients.

The secondary clinical outcome of this study was to determine the ability of circulating miRs to add prognostic value to a current prebiopsy nomogram, PBCG, which could move the miR-based clinical test earlier in the diagnostic pipeline, potentially reducing biopsies in low-risk patients. One limitation is that the serum samples were collected at least one month post-biopsy, so this analysis was completed assuming that the composition of circulating miRs was the same pre- and post-biopsy. In our study cohort, both serum and serum EV miRs significantly differed in patients with low- versus high-grade PCa; however, they only trended towards adding to the prognostic ability of PBCG (p = 0.08). As previously discussed, there is inherent noise in the Gleason grade of prostate biopsies given that thorough sampling of the tumours is not possible. Thus, larger cohorts with prebiopsy serum and post-RP pathology are needed to test biomarkers in this space.

Considerable overlap occurred in the significant miRs between serum and serum EV sample types. Some overlap in miRs was expected because serum EVs are a component of whole serum and will also be present in the whole serum. Additionally, there was an overlap of prognostic miRs in this study compared to the miRs included in the prognostic ‘Serum miR Score’ from our previous study (Mihelich et al., 2015). Two miRs from the prior score were prognostic for AP in both serum and serum EVs (hsa-let-7a-5p and hsa-miR-451a). In the serum EV model, five additional miRs that were significant for AP overlapped with those included in the Serum miR Score (Mihelich et al., 2015), demonstrating reproducibility of serum miRs over multiple platforms and cohorts. Furthermore, when we queried only the miRs within the previously published Serum miR Score (Mihelich et al., 2015), they also significantly improved the AUCs, demonstrating reproducibility of the miRs in multiple cohorts and platforms. Therefore, this study provides evidence that circulating serum EVs have both overlapping and unique miRs compared to circulating miRs in whole serum and may add additional prognostic information for PCa.

Currently, there are no standardized normalization methods for miRs across biological sample types, with even fewer reports regarding EV miRs. For our serum analysis, there were 10 miRs in the normalization panels, as determined by both geNorm and NormFinder, with seven of the miRs being consistent in both panels. This degree of overlap between normalization methods was consistent with other studies (Kok et al., 2015).

This study was limited by the number of patients who chose to undergo RP treatment and, therefore, the number of samples available for prognostic analysis and model configuration. Of 203 patients, only 81 underwent RP treatment. As our patient cohort was smaller than anticipated, we did not use training and testing subgroups for prognostic model generation. However, we used random forest modelling, which allows for the inclusion of all samples in the model creation due to the bagging feature. Given this limitation, these findings would need to be validated in a prospective cohort of patients with PCa.

Going forward, the serum EV panel of 19 miRs is the strongest biomarker panel to study further as it improves upon CAPRA the greatest in predicting AP. Nineteen miRs is within the standard range of other gene expression panels currently used in the clinic, such as Decipher, which has 22 genes (Klein et al., 2016). Considering the high rates of Gleason grade group discordance between biopsy and RP, the addition of serum EV miR quantitation could be used to guide patient decisions to undergo or forgo radical prostatectomy. With more accurate pretreatment risk stratification, patients with low-risk disease may have more confidence in active surveillance protocols and avoid overtreatment, whereas high-risk patients could be appropriately treated with more aggressive measures. Future studies should assess the clinical significance of these prognostic tests and how they affect treatment recommendations and patient decisions (Murphy et al., 2021).

AUTHOR CONTRIBUTIONS

Morgan L. Zenner and Larisa Nonn conceptualized the study, secured funding, analysed data and wrote the manuscript. Morgan L. Zenner, Brenna Kirkpatrick, Michael J. Schlicht, Alejandra Cavazos Saldana and Candice Loitz conducted experiments. Trevor R. Leonardo and Mark Maienschein-Cline did informatic and statistical analyses. Klara Valyi-Nagy and Michael Abern secured patient samples and consent forms. Peter H. Gann edited manuscript and provided patient sera.

ACKNOWLEDGEMENTS

We thank the patient participants who donated their specimens for this research. Blood samples were acquired with the assistance of Dr. Ruben Sauer and Ms. Patrice King-Lee. BioRender was used to prepare Figure 1a and graphical abstract. UIC Research Resource Center Cores, Transmission Electron Microscopy Core, and Genomics Core assisted with data collection. This research was funded by the Department of Defense Prostate Cancer Research Program W81XWH-16-1-0382 (L.N.), Ruth L. Kirschstein NRSA for Individual Predoctoral MD/PhD Degree Fellowship (NIH/NCI F30 CA243197) (M.L.Z.), and the UIC Center for Clinical and Translational Studies Pre-doctoral Education for Clinical and Translational Scientists (PECTS) Fellowship (M.L.Z.).

CONFLICT OF INTEREST STATEMENT

The authors have declared that no conflict of interest exists.

References

Aalberts, M., Stout, T. A., & Stoorvogel, W. (2014). Prostasomes: Extracellular vesicles from the prostate. Reproduction (Cambridge, England), 147, R1–14.

Aas, K., Dorothea Fosså, S., Åge Myklebust, T., Møller, B., Kvåle, R., Vlatkovic, L., & Berge, V. (2020). Increased curative treatment is associated with decreased prostate cancer‐specific and overall mortality in senior adults with high‐risk prostate cancer; results from a national registry‐based cohort study. Cancer Medicine, 9(18), 6646–6657. [DOI: https://dx.doi.org/10.1002/cam4.3297]