2.1 Patient summary and study design

The primary case-control study for mRNA analyses consisted of 165 GG2-4 patients treated by radical prostatectomy in Helsinki University Hospital in 1992-2015. The selection process of patients has been previously described.³⁴ Seventy-nine patients with metastatic progression or PCa-specific death, during the median follow-up of 11 years, were selected to present cases. Patients with similar clinical baseline characteristics, but no metastasis during the follow-up (n = 86), were assigned as controls, representing a clinically indolent phenotype of GG2-4 PCa. Additionally, 12 samples from distant PCa metastases, seven samples of castration-resistant PCa (CRPC) and six neuroendocrine PCa (NE) were included. Nineteen benign prostate samples from men without detected PCa, but with benign prostatic hyperplasia, were also studied. One of the metastases was from a para-aortic lymph node, from the level of renal vein and was extracted 10 years after RP. The other metastases were from bone (vertebra, humerus, femur, n = 5), bone marrow (iliac crest, n = 1), spinal cord (n = 1), liver (n = 1), appendix (n = 1), testis (n = 1) and epididymis (n = 1). All patients in the CRPC group, 10 in the metastasis group and three in the NE had received hormonal treatment before the samples were extracted. Patients in other groups were hormone-naïve at the time of sample extraction. During quality control, eight aggressive cases, four controls and three metastasis samples were excluded from the final analyzes. The histological slides of all the patients were rereviewed by an expert uropathologist (TM) and the index cancer lesions were annotated for RNA extraction and tissue microarray (TMA) construction using the formalin-fixed paraffin-embedded (FFPE) blocks. Detailed demographics of analyzed samples are available in Table 1.

2.2 RNA extraction and transcript analysis

RNA extraction was performed using previously published methodology.³⁴ In brief, one to two 1 mm diameter punches were sampled from FFPE blocks. The punches were deparaffinized, homogenized and proteinase K digested. Then RNA was extracted using QIASymphony RNA kit (QIAGEN, Venlo, Netherlands). RNA concentration was analyzed using RiboGreen kit (Invitrogen, Waltham, MA) and the RNA integrity (range = 1.0-7.6, mean = 2.24, SD = 0.82) was measured with Agilent Bioanalyzer kit (Agilent Technologies, Santa Clara, CA). Transcript counts were measured using the NanoString nCounter platform (NanoString Technologies, Seattle, WA). The transcripts were hybridized with probes of a custom 93-gene NanoString Reporter and Capture CodeSet. Among others, probes included those for KLK1-15 and PAR genes, androgen receptor gene (AR), and 12 housekeeping genes (Table S1). The housekeeping genes were ACTB, ALAS1, B2M, CLTC, G6PD, GAPDH, GUSB, HPRT1, POLR2A, SDHA, TBP and TUBB.

2.3 Tissue microarray composition and immunohistochemical staining

The samples punched into TMA for an immunohistochemistry (IHC) study overlapped largely with the samples punched for mRNA extraction. The IHC study additionally contained 11 samples of matched primary PCa for the metastasis samples. For case, control, CRPC and NE specimens, two punches of 1 mm in diameter of cancer tissue were sampled. Benign tissue adjacent to cancer tissue, was also sampled from case and control blocks. For metastasis samples, two punches were extracted from FFPE blocks from metastases, and two samples from the index tumor in the prostate. For patients without any detected PCa, two punches of benign tissue were extracted. Punches were sampled from a location as close as possible from those used for RNA extraction, so the TMA cores can be considered representative of the mRNA punches.

TMA sections of 3.5 μm were first hematoxylin and eosin stained in a clinical laboratory (Huslab, Helsinki, Finland). The TMA slides were scanned using Pannoramic 250 Flash III scanner (3DHISTECH, Budapest, Hungary) with 20X magnification (NA 0.8). Digitized slides were then consensus-assessed by an expert uropathologist (TM) and a nonpathologist researcher (TPKL) for percentage area of total epithelium and cancerous epithelium in relation to total epithelium. TMA cores in the case group had mean total epithelial area of 67% (SD = 18.7) of which 93% (SD = 16.1) was cancerous. In the control group, the mean total epithelial area was 67% (SD = 18.4) of which 89% (SD = 23.7) was cancerous.

TMA sections of 3.5 μm were immunostained using rabbit-anti-KLK12 antibody, previously validated for IHC of prostatic tissue.³⁵ The slides were incubated at +37°C overnight and at +56°C for 1 hour before deparaffinization in xylene and rehydration in a decreasing ethanol series. Antigen retrieval was performed with Dako PT Link Pretreatment module (Agilent Technologies) using Dako Envision TM Flex Target Retrieval solution, low pH (Agilent Technologies). The staining was carried out in an Autostainer 480 (LabVision Corporation, Fremont, CA) using Dako REAL Envision Detection system Peroxidase/DAB+, Rabbit/Mouse (Agilent Technologies). The KLK12 antibody, diluted at 1:500, was incubated for 1 hour at room temperature and the secondary antibody (of the Dako REAL Envision Detection system) for 20 minutes at room temperature. In control sections the primary antibody was replaced by Rabbit IgG control antibody (Vector Laboratories, Newark, CA). Finally, tissues were counterstained with hematoxylin (Histolab, Espoo, Finland), washed with water and mounted using Aquatex (Merck Millipore, Burlington, MA). The slides were again scanned with a Pannoramic 250 Flash III slide scanner, with 20X magnification (NA 0.8). Staining intensities of the tissue samples were evaluated by three observers (RK, HK and TM). Only the cancerous areas were evaluated in cancerous TMA spots, and benign areas in cancer-adjacent benign spots. Apart from this classification, staining evaluation was blinded regarding the clinical data. Staining intensity in the epithelial cells was assessed in an ordinal scale as negative (0), weak (1), moderate (2) or strong (3) and their combinations, for example, moderate-strong (2.5). Only the areas with the strongest staining for the spot were assessed, given that those represented >5% of the total cancerous epithelial area in the spot. One aggressive case, two metastasis samples, one sample of matched primary PCa and four samples of benign prostate tissue were excluded due to not being scorable.

2.4 Cell culture

LNCaP cells (ATCC, CRL-1740, RRID: CVCL_0395), originally isolated from lymph node metastasis of human prostate adenocarcinoma, were maintained in RPMI-1640 media (Lonza, Basel Switzerland) supplemented with 10% fetal bovine serum (Biowest, Nuaillé, France), 100 units/mL penicillin, 0.1 mg/mL streptomycin, 1 M glucose, 1 M HEPES, 100 mM sodium pyruvate, L-glutamine and nonessential amino acid cell culture supplement (Lonza) at +37°C, 5% CO₂. AR-negative bone metastasis-derived PC-3 cells (ATCC, CRL-1435, RRID: CVCL_0035) were cultured in Ham's F12 Nutrient mixture (F-12) (Gibco) supplemented with 10% fetal bovine serum, 100 units/mL penicillin, 0.1 mg/mL streptomycin, and 7.5% sodium bicarbonate (Gibco). Culture media was changed every 2 to 3 days. Upon reaching 80% confluence, cells were detached from flasks using trypsin-EDTA solution (Lonza). Trypsin was inactivated with complete growth media and cells were centrifuged. Cell pellet was resuspended in fresh media and cells were counted using Countess II automated cell counter (Invitrogen). All cell lines were authenticated using short tandem repeat (STR) profiling within the last 3 years. All experiments were performed with mycoplasma-free cells.

2.5 RNA interference

For RNAi, cells were seeded 24 h prior to transfection with siRNAs (all from QIAGEN). Four different siRNAs targeting KLK12 (KLK12_1, SI00124201, TGGGAACTTCTTGGAACTTTA; KLK12_2, SI00124208, TCCCTGGAGTCTACACCTATA; KLK12_3, SI00124215, TCCCGGGAGAATCACGAGCAA; KLK12_8, SI05119912, ATAGTCTGGAATAAATATAAA) and KLK15 (KLK15_1, SI00134974, CCCGCCTGACATGGAACAGAA; KLK15_3, SI00134988, CACCTGTTTAATGCCAAGATA; KLK15_5, SI02642647, CCCGTGTGATCTTGAACAAGA; KLK15_6, SI03033485, AAGCGCGATGGCCCAGAGCAA) were used in two to three independent experiments for each siRNA. AllStars Negative Control and AllStars Hs Cell Death Control siRNAs were used as controls. siRNAs were transfected, at a final concentration of 10 nM, using Lipofectamine RNAiMAX transfection reagent (Invitrogen) in serum-free OptiMEM-1 medium (Invitrogen) according to the manufacturer's instructions.

The efficiency of siRNAs was determined using quantitative reverse transcription-polymerase chain reaction (RT-qPCR). One milliliter of cold Matrigel (Corning, Corning, NY) was pipetted into wells of 6-well plates on ice and allowed to solidify at +37°C for an hour. LNCaP cells (2 × 10⁴ cells/well) were seeded and, after 24 hours incubation, transfected as above. Cells were washed three times with ice-cold PBS on ice 72 hours after transfection, and 2 mL of Cell Recovery solution (Corning) was added into each well. The cells and Matrigel were detached from the bottom of the well using cell scraper and the plate was incubated on ice for 30 minutes. The contents of the wells were collected into 50 mL tube, and the wells rinsed with 2 mL of Cell Recovery solution. The tubes were incubated on ice for 30 minutes followed by the centrifugation at 1000g for 5 minutes at +4°C. LNCaP cells were also seeded on a 6-well plate (250 000 cells per well) and incubated at +37°C for 24 hours prior to transfection as above. Seventy-two hours after transfection, the cells were washed once with PBS and detached using trypsin-EDTA solution (Lonza). Trypsin was inactivated with complete growth media and cells were centrifuged. For both on Matrigel and on plastic grown cells, the supernatants were discarded, and cell pellets were resuspended in 1 mL of ice-cold PBS, transferred into the microcentrifuge tubes and centrifuged at 10 000g for 5 minutes at +4°C. The washing step was repeated, and the cell pellets were stored at −80°C until RNA extraction. Total RNA was isolated using RNeasy Mini kit (QIAGEN) and quantified using DS-11+ spectrophotometer (DeNovix, Inc, Wilmington, DE).

cDNA was obtained using SensiFAST cDNA Synthesis Kit (Meridian Bioscience, Cincinnati, OH) according to the supplier's protocol. RT-qPCR was conducted using TaqMan gene expression assays (Applied Biosystems, Waltham, MA) on Applied Biosystems 7500 Fast Real-time PCR system according to the manufacturer's instructions. TaqMan assays used for KLK12, KLK15 and GAPDH were Hs00377603_m1, Hs00930396_g1 and Hs02786624_g1, respectively. Relative mRNA levels were analyzed as the ratio of KLK12 or KLK15 to GAPDH using 2^-ΔΔCt method.

2.6 Overexpression of KLK12 and KLK15 in PC-3 cells

KLK12 (UniProt: Q9UKR0) and KLK15 (Q9H2R5) encoding cDNAs were cloned as synthetic genes in pcDNA3.1(+) plasmid using EcoRI/NotI sites (GenScript Biotech, Rijswik, Netherlands) (see Supplementary Material). The codons were optimized for mammalian expression (GenScript) and the propeptides, following the signal peptides of KLK12 and KLK15, were replaced by a furin-activable propeptide (APLRLRR), which has previously been successfully used for activation of KLK3 in PC-3 cells.³⁶ For controls, codons for Ser²⁰⁰ in KLK12 and Ser²⁰⁹ in KLK15 in active sites were replaced by a codon encoding Ala.

The plasmids were transfected in PC-3 cells using FugeneHD transfection reagent (Promega, Madison, WI). For selection of KLK12/KLK15 expressing cells, the cells were grown, at least for 28 days, in Ham's F12 media (Gibco) with supplements as above, containing also 200 μg/mL of geneticin (Gibco). For the detection of the expression of KLK12, KLK15 and their mutant forms, RNA was isolated and reverse-transcribed to cDNA as described above, and subjected to RT-qPCR and SybrGreen detection (Meridian bioscience, Cincinnati, OH). For primers for codon optimized KLK12 and KLK15 see Supplementary Material. The expression was assessed using ΔΔCt method (i.e., ΔCt of KLK transfected cells − ΔCt of wild-type cells, where ΔCt = threshold cycle [Ct] of the gene of interest − Ct of GAPDH). If the Ct was not detectable it was set to 40 (as seen sometimes for KLK12 and KLK15 in wild-type PC-3 cells). Secreted proteins were also detected in conditioned cell culture supernatants by immunoassays.¹¹ For this, 40 000 cells were grown on wells of 12-well plate for 7 days in 1.2 mL medium. KLK12 was also detected by immunocytochemical staining (ICC). For ICC the cells grown on glass surface were fixed for 5 min with acetone. After washing twice with PBS, essentially the same protocol, including the antibody, as used for IHC staining of KLK12 in tissue sections was followed, but without antigen retrieval and steps before that.

2.7 Cell growth assays

To assess the effect of siRNAs on cell viability and growth on plastic surface, 6000 LNCaP cells/well were seeded on a 96-well plate and left to grow for 24 hours prior to siRNA transfection as above. To study the effect of overexpression of KLK12 and KLK15 on cell proliferation, PC-3 cells (3000 cells/well) were seeded on a 96-well plate. After 3-4 days, 10 μL of Cell Counting Kit-8 (Dojindo Laboratories, Kumamoto, Japan) reagent was added into each well and the plate was incubated at +37°C for 1-2 hours. The absorbance at 450 nm was measured using Multiskan EX plate reader (Labsystems, Vantaa, Finland). Three independent experiments were conducted in triplicate.

For growth of cells on top of Matrigel, 150 μL of cold Matrigel was pipetted into wells of a 48-well plate on ice and allowed to solidify at +37°C for an hour. LNCaP cells were seeded and transfected as above for RNA isolation. Half of the media was replaced with fresh media every 3 days. The effect of KLK12 and KLK15 knockdown on the number and area of LNCaP cell colonies grown on Matrigel, was quantitated from images acquired using EVOS FL cell imaging system (Advanced Microscopy Group, Bothell, WA) with 2X objective. Images were analyzed using Fiji ImageJ.³⁷ The background subtraction was performed using the Gaussian blur method and image thresholding using Otsu method. The Analyze particles-command was used to analyze the cell clusters in the segmented images. The minimum particle size was set to 200 pixels to remove the background debris. Cell clusters touching the border were excluded from the analysis. Median size of cell clusters from each siRNA treated sample was multiplied by the number of the cell clusters to obtain a colony formation score.

2.8 Invasion and migration

The invasion assays were performed using Nunc Polycarbonate cell culture inserts (Thermo Scientific) and migration assays using Millicell cell culture inserts (Millipore, Burlington, MA). For invasion assays, the 24-well inserts were coated with 50 μL of Matrigel (0.5 mg/mL) and incubated at room temperature for 2.5 hours followed by the incubation with serum-free culture media for an hour. Transfected and wild-type PC-3 cells (300 000) were seeded in serum-free culture media on top of Matrigel in the upper chamber of the Nunc inserts. For migration assays, 300 000 cells were seeded in serum-free culture media in the upper chamber of the Millicell inserts. For both assays, F-12 culture media with 10% FBS was used as a chemoattractant in the lower compartments. After 2-4 days incubation, the cells from the upper chamber were removed with a cotton swab soaked in ultrapure water. Migrated cells were stained with 0.1% crystal violet for 10 minutes and rinsed with ultrapure water. Finally, the inserts were incubated in 200 μL of extraction solution (Cell Biolabs, Inc, San Diego, CA) for 10 minutes at room temperature. One hundred microliters of the extracts were then transferred to a 96-well plate and the absorbance was measured at 560 nm using Multiskan EX plate reader.

2.9 Statistical analyses

Statistical analyses were performed using nSolver Analysis Software (NanoString Technologies), v. 4.0.70 and R, v.4.0.2 (R Development Core Team, Vienna, Austria). Detailed statistical methodology can be found in the Supplementary statistical methods.

Transcript data quality control was performed in nSolver software in compliance with NanoString's guidelines. Raw counts were exported into R. Limit of detection (LOD) was determined as mean + SD of negative control probes. The eight best performing housekeeping genes, including ACTB, B2M, CLTC, GAPDH, GUSB, POLR2A, SDHA and TUBB were used to analyze variation caused by technical sources. Next, differential gene expression analysis was performed. Differentially expressed genes were determined based on multiple hypothesis testing adjusted P < .05.

Transcript counts were variance-stabilizing transformed for further analyses. Metastasis-free and PCa-specific survival of patients in the case and control study groups was analyzed using Kaplan-Meier estimator and Mantel-Haenszel log-rank test. Random forests were used to classify metastatic disease and PCa-specific death. The data was randomly split to training and validation sets with stratification, which consisted of 67% and 33% of case and control samples respectively. Class imbalance of training set was overcome by synthetic minority oversampling technique. The model performance was assessed using receiver operating characteristic areas under the curves (AUCs). AUCs were compared using DeLong test. Goodness of models was also assessed by stratifying the validation set based on the random forest model classifications in Kaplan-Meier survival analysis.

When multiple spots representing same patients and tissue type were available, the ones showing strongest staining were used for analyses. The IHC scores of cancer spots were compared to adjacent benign tissue using Mann-Whitney U-test. The IHC scores of cancer adjacent benign spots of case and control samples were also compared to benign tissue samples from patients with no detected PCa. Spearman's rank order correlation of KLK12 IHC scores and transcript counts was calculated. Statistical analysis of cell growth assays was conducted using two-tailed one-sample t-test after assessing normality with Shapiro-Wilk test.

3.1 Low KLK15 expression is associated with poor prognosis

We set out to study the expression of all human KLKs and PARs in PCa on mRNA level. KLK3, -2, -4, -11, -15, -10 and -12, in decreasing order of mean mRNA expression, were expressed in PCa and benign prostate samples over LOD (Figure 1A). Expression of these KLKs over LOD was observed also in metastases and CRPC. The mean expression levels of other KLKs were close to background. Very high expression of KLK2, -3 and -4 was observed in PCa and benign prostate, however in NE samples the expression of KLK2-4 was low. On the contrary, KLK12 expression was highest in NE. Among the four human PARs, PAR1 and PAR2 were highly expressed in benign tissues and all subgroups of PCa. Low expression of PAR3 and PAR4 was also observed in PCa groups. AR was highly expressed in all samples aside from NE.

Among the significantly expressed KLKs, the expression of KLK4, -12 and -15 was increased, and that of KLK10 and -11 decreased, in cancer (cases and controls combined), compared to benign prostate (P < .05 for all) (Figure 1A and Table 2). However, when aggressive cases were compared to controls (Figure 1A and Table 2), the expression of KLK2, -3, -4 and -15 was found to be decreased, and that of KLK12 increased, in cases (P < .05 for all). Higher expression of all PARs was observed in cancer (cases and controls combined) compared to benign prostate (P < .05). The expression of PAR1 and PAR4 was increased in cases compared to controls, while PAR2 was decreased (P < .05). AR expression was similar in case and control groups (P = .73). AR appeared to be up-regulated in metastatic and CRPC samples compared to localized cancer and down-regulated in neuroendocrine cancers as shown in Figure 1A, although no statistical tests were performed due to low number of samples in these groups.

Low KLK2, -3 and -15, and high PAR1 and PAR4 expression was associated with poor metastasis-free survival (Table 2). Further, low KLK15 and high PAR1 and PAR3 were associated with poor PCa-specific survival (Table 2). Survival plots depicting association of KLK15 with metastasis-free and PCa-specific survival are shown in Figure 1B,C. High GG was associated with adverse metastasis-free survival in the binary comparison of GG2 vs GG3-4 (P = .0074), but not with PCa-specific survival (P = .071). However, only PAR2 showed significant differences in mRNA expression across GG2 and 3, and none of the analyzed transcripts were differentially expressed across pathological stages. Differential expression analysis results across GGs and pathological stages are shown in Table S2.

3.2 KLKs and PARs improve classification of progressing PCa over clinical parameters

When all KLKs and PARs, that showed expression levels over LOD, were combined into a random forest model, higher classification performance for metastasis and PCa-specific death was observed compared to that of clinical parameters GG, stage and PSA combined (AUC = 0.67 vs 0.53 and AUC = 0.67 vs 0.58, respectively) (Figure 2A,B). Combining the clinical parameters with KLKs and PARs improved classification of metastatic disease, but not PCa-specific death over the combination of KLKs and PARs alone (AUC = 0.70 and AUC = 0.64, respectively). The observed improvements were not statistically significant in DeLong test due to lack of statistical power in the validation set. In Kaplan-Meier analysis, those with panel-predicted metastasis in the combined model had shorter PCa-specific survival (P = .031), however, no statistically significant differences were observed in metastasis-free survival (P = .087) (Figure 2C,D).

The expression of KLK2 and -3 correlated with AR both in case and control groups, while for KLK4 and -15 such correlation with AR was observed only in the control group (ρ > 0.4, P < .001 for all). Correlations of AR with KLKs and PARs showing mean expression over LOD can be found in Table S3.

3.3 High KLK12 protein expression is associated with poor survival

We also studied KLK12 protein expression using IHC (Figure 3). The intensity of KLK12 expression was stronger in cancer-adjacent benign spots (mean score = 2.46) compared to cancerous spots (mean = 0.96, P < .0001), and similar to that in benign spots from patients without detectable cancer (mean = 2.73, P = .21). In case and control samples, high KLK12 scores were associated with poor metastasis-free and PCa-specific survival (Figure 3G,H). KLK12 IHC scores and transcript levels showed low or no correlation for cases (ρ = 0.25, P = .04), controls (ρ = −0.16, P = .15), benigns (ρ = 0.29, P = .29), CRPCs (ρ = 0.76, P = .049), distant metastases (ρ = 0.27, P = .51) and NEs (ρ = 0.34, P = .51).

3.4 Knockdown of KLK15 reduces the formation of LNCaP colonies on Matrigel

To investigate the functional effects of KLK12 and KLK15 expression on prostate adenocarcinoma, KLK12 and KLK15 were down-regulated in LNCaP cells using four different siRNAs for each gene (Figure 4). When the cells were grown on plastic, all siRNAs targeting either KLK12 or KLK15 reduced KLK12 and KLK15 expression, respectively, below 20% of that in control transfected cells, as determined by RT-qPCR. In the cells grown on Matrigel basement membrane preparation the knock-down efficiency of siRNAs was, in most cases, less pronounced, but still clearly distinguishable from the control (Figure 4). Two siRNAs targeting KLK12 and two siRNAs targeting KLK15 had a statistically significant effect on proliferation of the cells grown on plastic surface (for all P < .05, n = 3), while one of the siRNAs targeting KLK12 and three targeting KLK15 reduced the colony formation score of cells grown on Matrigel (for all P < .05, n = 4).

3.5 Overexpression of KLK12 and KLK15 does not affect growth, migration or invasion of PC-3 cells

Since LNCaP cells are poorly invasive we used for invasion studies PC-3 cells that were transfected with KLK12 and KLK15 expression constructs and, as controls, constructs directing expression of active site mutant forms of these KLKs. Unlike the wild-type PC-3 cells, the transfected cells exhibited significant mRNA and protein expression of KLK12 and KLK15 (Table S4 and Figure S1). The overexpression of KLK12 and KLK15 did not significantly change the cell proliferation on plastic surface (Table S4). Also chemotactic migration and invasion were unaffected by overexpression.