PD-L1 promoter methylation is a prognostic biomarker for biochemical recurrence-free survival in prostate cancer patients following radical prostatectomy

Background The rapid development of programmed death 1 (PD-1)/programmed death ligand 1 (PD-L1) inhibitors has generated an urgent need for biomarkers assisting the selection of patients eligible for therapy. The use of PD-L1 immunohistochemistry, which has been suggested as a predictive biomarker, however, is confounded by multiple unresolved issues. The aim of this study therefore was to quantify PD-L1 DNA methylation (mPD-L1) in prostate tissue samples and to evaluate its potential as a biomarker in prostate cancer (PCa). Results In the training cohort, normal tissue showed significantly lower levels of mPD-L1 compared to tumor tissue. High mPD-L1 in PCa was associated with biochemical recurrence (BCR) in univariate Cox proportional hazards (hazard ratio (HR)=2.60 [95%CI: 1.50-4.51], p=0.001) and Kaplan-Meier analyses (p<0.001). These results were corroborated in an independent validation cohort in univariate Cox (HR=1.24 [95%CI: 1.08-1.43], p=0.002) and Kaplan-Meier analyses (p=0.029). Although mPD-L1 and PD-L1 protein expression did not correlate in the validation cohort, both parameters added significant prognostic information in bivariate Cox analysis (HR=1.22 [95%CI: 1.05-1.42], p=0.008 for mPD-L1 and HR=2.58 [95%CI: 1.43-4.63], p=0.002 for PD-L1 protein expression). Methods mPD-L1 was analyzed in a training cohort from The Cancer Genome Atlas (n=498) and was subsequently measured in an independent validation cohort (n=299) by quantitative methylation-specific real-time PCR. All patients had undergone radical prostatectomy. Conclusions mPD-L1 is a promising biomarker for the risk stratification of PCa patients and might offer additional relevant prognostic information to the implemented clinical parameters, particularly in the setting of immune checkpoint inhibition.


INTRODUCTION
Recently, the blockade of the programmed death 1 (PD-1)/programmed death ligand 1 (PD-L1) signaling pathway, either with antibodies against the receptor PD-1 (Nivolumab, Pembrolizumab, Pidilizumab) or its ligand PD-L1 (BMS-936559, MPDL3280A, MEDI4736, MSB001718C), has shown promising results in several advanced cancers, e.g. malignant melanoma, non-small cell lung cancer, bladder cancer, head and neck cancer, and renal cell cancer [1][2]. While the activation of the PD-1/PD-L1 pathway allows for the tumor to successfully elude the host's immune system via T-cell exhaustion, a therapy strategically targeting the PD-1/PD-L1 pathway seems to promote the immune response against the tumor [3][4]. Immunohistochemically detected PD-L1 expression has further been reported to be a predictive biomarker for the treatment with anti-PD-1/PD-L1 treatment [4].
We have previously shown that PD-L1 is highly expressed in aggressive primary prostate cancer (PCa) and is an independent predictor of biochemical disease progression [5]. Although clinical trials for the treatment of PCa so far have yielded conflicting results [2,6], our findings indicate that PD-1/PD-L1 targeted therapy might be a novel treatment option for hormone-naive tumors. Due to interlaboratory and interobserver variation, however, the general reproducibility of immunohistochemical methods still remains challenging [7][8]. For the successful implementation of reliable biomarkers into clinical practice, the robust and reproducible quantification of DNA methylation instead of immunohistochemistry might be highly beneficial.
The epigenetic mechanism of DNA methylation plays a key role in several fundamental biological processes, e.g. development, cell differentiation, and gene silencing [9][10]. Furthermore, DNA methylation is often deregulated in human malignancies [11][12][13][14] suggesting that aberrantly methylated loci might be a valuable source for biomarkers [15][16]. As DNA methylation can also be robustly quantified in specimens with limited DNA abundance or formalin-fixed paraffin-embedded tissue (FFPET), in which DNA is highly degraded, methylationbased biomarkers prove to be promising diagnostic tools for clinical routine [17]. So far, various DNA methylation biomarkers have been successfully applied in a clinical setting [18]. The short stature homeobox 2 (SHOX2) and septin 9 (SEPT9), for instance, are employed as diagnostic and screening tools for the detection of lung cancer and colorectal cancer [19][20]. In addition, promoter methylation of the O-6-methylguanine-DNA methyltransferase (MGMT) has been shown to be predictive of response to alkylating agents in glioblastoma [21][22].
We have very recently shown that promoter methylation of the immune checkpoint receptor PD-1 is an independent prognostic biomarker for biochemical recurrence (BCR)-free survival in PCa patients following radical prostatectomy [23]. The aim of the present study was to quantify PD-L1 DNA methylation in prostate tissue samples and to evaluate its potential role as a prognostic biomarker in PCa.
mPD-L1 as continuous variable was further shown to be of prognostic significance in univariate Cox proportional hazard analysis for bead cg19724470 (HR = 1.02 [95%CI: 1.00-1.04], p = 0.024; Table 1). In order to analyze the suitability of mPD-L1 for the stratification of patients at risk for BCR, mPD-L1 was dichotomized using an optimized cut-off which was identified by an iterative approach. According to two beads (cg19724470 and cg15837913), mPD-L1 was significantly associated with BCR in Kaplan-Meier analysis (cg19724470: p < 0.001, cg15837913: p = 0.008, Figure 3A) Figure 3B).
A detailed analysis of the association/correlation of PD-L1 promoter methylation with clinicopathological parameters was further conducted by means of the best performing bead cg19724470. Hereby, PD-L1 promoter methylation was shown to correlate/associate with the prognostic variables age (p = 0.028), pathological tumor category (pT) (p < 0.001), the International Society of Urological Pathology (ISUP) grading group (p = 0.001), surgical margin (p = 0.013), and ETS-related gene (ERG) expression (p = 0.022, Table 2). www.impactjournals.com/oncotarget

Design and validation of a PD-L1 qMSP assay
To further evaluate the previous results in an independent dataset with an additional technology, a quantitative methylation-specific real-time PCR (qMSP) assay was designed within the promoter region of the PD-L1 gene in the location of the best performing cg19724470 bead ( Figure 1A). In brief, the assay represents a duplex real-time PCR for sensitive and quantitative detection of PD-L1 DNA methylation and a reference PCR for quantification of the total DNA using the ACTB locus. The assay performance was validated using a dilution  series of bisulfite-converted artificially methylated and unmethylated DNA. Samples were measured in triplicate. As indicated in Figure 1B, the PD-L1/ACTB qMSP assay allowed for a highly accurate quantification of the PD-L1 DNA methylation over a broad range of relative methylation of the template DNA, even at low methylation levels. The assay showed an accurate quantification of mPD-L1 within the whole spectrum from 0% to 100% methylation (r 2 = 0.956, Figure 1B).

Association of PD-L1 promoter methylation and protein expression (validation cohort)
High PD-L1 protein expression determined semiquantitatively via immunohistochemistry (IHC) has earlier been shown to be a strong predictor of BCR in the cohort under investigation (p = 0.004, HR = 2.37 [95%CI: 1.32-4.25]) [5]. The fact that high mPD-L1 and high PD-L1 protein expression have an adverse prognostic value in the validation cohort seems to be conflicting, since DNA methylation is regarded as a long-lasting negative regulator of mRNA expression and should thus be associated with silencing of the gene product. We therefore analyzed the correlation between DNA methylation and protein expression. Matched IHC and DNA methylation results were available for n = 209 patients. Our results suggested that PD-L1 protein expression is unrelated to mPD-L1 in the PCa specimens available for comparative examination (ρ = 0.05, p = 0.47), indicating that post-transcriptional and potentially post-translational regulatory mechanisms are involved in the control of PD-L1 protein expression. In order to test whether mPD-L1 and PD-L1 protein expression provide independent prognostic information, we performed a multivariate Cox proportional hazards analysis including both variables. Notably, PD-L1 methylation (HR = 1.22 [95%CI: 1.05-1.42] p = 0.008) and PD-L1 protein expression (HR = 2.58 [95%CI: 1.43-4.63], p = 0.002) analyzed as continuous variables both added significant prognostic information. We further analyzed the prognosis of PCa patients stratified according to their mPD-L1 status (high vs. low) in combination with the dichotomized PD-L1 protein expression status. Figure  5 shows that patients with high PD-L1 protein expression and high mPD-L1 present with shorter BCR-free intervals compared to patients with simultaneous low PD-L1 protein expression and low mPD-L1. Patients with either high PD-L1 protein expression and low mPD-L1 or low PD-L1 protein expression and high mPD-L1 showed an intermediate BCR-free interval. The finding that PD-L1 protein expression and methylation seem to be unrelated in our cohort prompted us to investigate additional mechanisms of post-transcriptional regulation of PD-L1.

Association of PD-L1 mRNA and methylation with microRNA expression (training cohort)
Known epigenetic post-translational modification mechanisms of PD-L1 include the differential expression of microRNA (miR). In particular miR-197, miR-200, miR-570, miR-34a, and miR-513 have been shown to play  a key role in the regulation of PD-L1 expression (reviewed by [24]). MiRs are non-coding single-stranded RNAs 22-24 nucleotides in length, which can bind and degrade or stabilize mRNA. Data on miR expression were available only for the training cohort. MiR-197 and miR-200a-c positively correlated with PD-L1 mRNA and inversely correlated with mPD-L1, suggesting that mPD-L1 was associated not only with decreased mRNA expression but also with lower levels of mRNA destabilizing miR-197 and three of the miR-200 family members miR-200a-c (Table 4). MiR-570, on the other hand, was only correlated with mPD-L1. For miR-34a, an inverse correlation with mPD-L1 and mRNA expression was shown. MiR-513 was not differentially expressed with regard to methylation and mRNA expression of PD-L1.

DISCUSSION
Targeting the immune checkpoint pathway PD-1/ PD-1 has emerged as a promising therapy providing significant clinical activity and durable benefits with minimal toxicities in several solid tumors [1][2]. The rapid development and approval of PD-1/PD-L1 inhibitors has generated an urgent need for predictive biomarkers assisting the selection of patients most likely to respond to therapy. Several clinical trials have suggested immunohistochemically detected PD-L1 expression as a predictive biomarker for the treatment with anti-PD-1/PD-L1 inhibitors [4,25]. The use of PD-L1 immunohistochemistry as a predictive biomarker, however, is confounded by multiple unresolved issues e.g. differing antibodies and scoring cut-offs, tissue preparation, as well as interlaboratory and interobserver variability.
Aberrant DNA methylation of certain loci has been reported to play a major role in carcinogenesis [26]. Such epigenetic modifications have further been shown to be a robust and reliable diagnostic tool [19,[21][22]27], therefore representing a promising source for cancer biomarkers. We have very recently shown that promoter methylation of the immune checkpoint receptor PD-1 is an independent prognostic biomarker for biochemical recurrence (BCR)-free survival in PCa patients following radical prostatectomy [23]. So far, no data are available on the role of PD-L1 promoter methylation in PCa. The aim of this study therefore was to quantify PD-L1 DNA methylation in prostate tissue samples and to evaluate its potential as a biomarker in PCa.
DNA methylation of promoter regions is often associated with epigenetic silencing [12]. In normal tissue around 80% of CpGs are methylated and mainly CpG islands in promoter regions of active genes are hypomethylated [28]. In cancer there is a shift in the pattern of DNA methylation towards a global hypomethylation, whereas certain CpGs, often in promoter regions of tumor suppressor genes, are thought to become hypermethylated [28][29]. Nevertheless, recent publications show that the aberrant DNA methylation, especially in cancer, is much more complex (reviewed by [12]).
In the present study, five Illumina Infinium HumanMethylation450 BeadChip beads taken from the TCGA dataset and located in the promoter region of the PD-L1 gene were analyzed. In this training cohort, mPD-L1 levels were increased significantly in PCa Table 2: Patients' characteristics and correlation/association of PD-L1 DNA promoter methylation (mPD-L1) with clinicopathological parameters in the training (n = 498, Probe cg19724470) and validation cohort (n = 299). mPD-L1 was dichotomized according to the respective optimized cut-off (mPD-L1 low vs. mPD-L1 high ) compared to normal tissue, indicating that PD-L1 might be differentially methylated in prostatic malignancies. Further investigations revealed that two of the five examined beads (cg19724470 and cg15837913) showed a significant correlation of mPD-L1 with BCR in univariate Cox proportional hazards analysis and Kaplan-Meier analysis. Interestingly, differential methylation was found in the border regions and not the centre of the CpGdense promoter area. One of the two beads (cg19724470) further showed an inverse correlation with PD-L1 mRNA expression and an association with prognostic clinicopathological variables including the ISUP grading group, pT, pre-surgical prostate-specific antigen (PSA), resection margin, and ERG status. In order to validate these findings with an additional technology in an independent cohort, a quantitative methylation qPCR assay was designed within the region of the best performing bead (cg19724470). In this validation cohort, high PD-L1 DNA methylation, analyzed as a continuous variable and as a dichtomized variable using an optimized cut-off, was significantly associated with an increased risk for BCR in patients in univariate Cox  [34]); validation cohort: nuclear ERG protein expression (as previously published in [35]) § Training cohort: AR activity score (as adopted from [34]); validation cohort: AR protein expression (as previously published in [35])   proportional hazards analysis and its prognostic value was further confirmed by Kaplan-Meier analysis. Surprisingly, our study revealed no inverse correlation of PD-L1 promoter methylation and protein expression in PCa. This may point at a complex post-transcriptional and post-translational regulation of PD-L1. According to our findings, mPD-L1 is inversely correlated with mRNA transcription and associated with miR as cellular component, which modify the downstream processing of PD-L1 mRNA. Differential expression of miR may therefore potentially interfere with the linear translation of PD-L1 mRNA into PD-L1 protein. Thus, further studies are warranted to unravel the complex interplay between DNA methylation and protein expression in PCa. In addition, post-translational modifications which affect the stability of the protein (e.g. glycolization) and in turn alter the affinity of the anti-PD-L1 antibody to membranous PD-L1 might further be causative for ostensibly conflicting results [30]. When combining PD-L1 protein expression and methylation, however, both parameters added independent prognostic information, suggesting that PD-L1 methylation might be a valuable companion biomarker in addition to immunohistochemistry. We would therefore strongly recommend the integration of its analysis in running clinical trials.
A limitation of DNA methylation analysis techniques that needs to be mentioned is the inability to differentiate heterogeneous methylation patterns in different cell types present within samples. Methylation analysis is performed using tissue lysates and not specific cell subtypes, which can differ in methylation. We have previously shown that methylation levels in breast cancer differ among ductal epithelia of adjacent normal tissue, stromal cells, tumor infiltrating lymphocytes, and even within the invasive tumor itself [31]. Further studies employing microdissected specimens are therefore warranted to analyze tissue-and cell-specific epigenetic regulation in detail.
Taken together, our findings might be of great value for the tailoring of individual therapies and risk stratification, especially within the framework of therapies targeting the PD-1/PD-L1 signaling pathway. Although the experience with PD-1/PD-L1 inhibition in PCa is still limited [4], PD-1 DNA methylation may have a predictive value for future treatment with checkpoint inhibitors in PCa. However, further studies are warranted to determine the potential of PD-L1 methylation as a biomarker, particularly in the setting of immune checkpoint inhibition.

Training cohort
The patient cohort based on data collected by The Cancer Genome Atlas (TCGA) comprised of 498 patients with histologically confirmed PCa obtained from several international centers involved in the TCGA project [32]. BCR-free survival was considered as the primary endpoint of the study. Informed consent was acquired from all patients included in the cohort in accordance with the Helsinki Declaration of 1975. PD-L1 mRNA expression data were available for 497 patients samples. PD-L1 promoter methylation was assessable for 498 specimens. Clinical follow-up was available for 410 individuals (mean follow-up period 21.6 months, range 1-133 months). Data on miR expression were obtainable for 494 patients.

Validation cohort
A total of 299 patients with histologically confirmed PCa who had undergone radical prostatectomy at the University Hospital Bonn between 1998 and 2008 were included in the study. BCR-free survival defined as postoperative PSA levels rising above 0.2 ng/ml were determined as the primary endpoint of the study. Clinical follow-up was available for 259 individuals (mean followup period 66.4 months, range 1-145 months).

Ethical approval
The section of the study including patient material from the University Hospital Bonn has been approved by the local Institutional Review Board at the University Hospital Bonn (Board number 071/14), which waived the need for written informed consent from the participants. All experiments were performed in accordance with the relevant guidelines and regulations.

Sample preparation and bisulfite conversion
For methylation analysis, FFPET samples were processed according to the InnuCONVERT Bisulfite All-In-One Kit (Analytik Jena) as previously published [33]. For assay validation, a dilution series of bisulfiteconverted, unmethylated sperm DNA (NW Andrology & Cryobank Inc., Spokane, WA, USA) and artificially methylated DNA (CpGenome™ Universal Methylated DNA; Merck Millipore, Darmstadt, Germany) and was used. DNA concentration was quantified by UV spectrophotometry using a Nanodrop ND-1000 spectral photometer (Nanodrop Technologies, Wilmington, DE, USA).

PD-L1 quantitative methylation real-time PCR
The DNA methylation of PD-L1 was determined using a qMSP assay. Primers and probes are shown in Table 5. Relative DNA methylation of the PD-L1 locus compared to total DNA (determined via ACTB reference in a duplex PCR reaction) was quantified using the AB 7500 Fast Real-Time PCR System (Life Technologies Corporation, Carlsbad, CA, USA). Each patient sample was measured in triplicate and an input of 50ng bisulfiteconverted DNA per reaction was measured. Thresholds and baselines were set as follows: 0.07 (threshold PD-L1), 3-22 (baseline PD-L1), 0.02 (threshold ACTB), and 3-21 (baseline ACTB). Percentage DNA methylation was calculated using the following formula: ΔCT = CT PD-L1 -CT ACTB , ΔΔCT = ΔCT Sample -ΔCT Calibrator , Methylation[%]=100%*2 ΔΔCT