CXCL12 promoter methylation and PD-L1 expression as prognostic biomarkers in prostate cancer patients

Background The CXCR4/CXCL12 axis plays a central role in systemic metastasis of prostate carcinoma (PCa), thereby representing a promising target for future therapies. Recent data suggest that the CXCR4/CXCL12 axis is functionally linked to the PD-1/PD-L1 immune checkpoint. We evaluated the prognostic value of aberrant CXCL12 DNA methylation with respect to PD-L1 expression in primary PCa. Results CXCL12 methylation showed a consistent significant correlation with Gleason grading groups in both cohorts (p < 0.001 for training and p = 0.034 for testing cohort). Short BCR-free survival was significantly associated with aberrant CXCL12 methylation in both cohorts and served as an independent prognostic factor in the testing cohort (hazard ratio = 1.92 [95%CI: 1.12–3.27], p = 0.049). Concomitant aberrant CXCL12 methylation and high PD-L1 expression was significantly associated with shorter BCR-free survival (p = 0.005). In comparative analysis, the CXCL12 methylation assay was able to provide approximately equivalent results in biopsy and prostatectomy specimens. Materials and Methods CXCL12 methylation was determined by means of a methylation specific quantitative PCR analysis in a radical prostatectomy patient cohort (n = 247, training cohort). Data published by The Cancer Genome Atlas served as a testing cohort (n = 498). CXCL12 methylation results were correlated to clinicopathological parameters including biochemical recurrence (BCR)-free survival. Conclusions CXCL12 methylation is a powerful prognostic biomarker for BCR in PCa patients after radical prostatectomy. Further studies need to ascertain if CXCL12 methylation may aid in planning active surveillance strategies.


IntroductIon
Prostate cancer (PCa) is a highly prevalent disease that remains clinically silent in the majority of cases and is mostly curable when localized. Advanced stages of PCa initially respond well to anti-androgen therapy but will usually progress to castration resistance with poor prognosis. Obviously, new prognostic tools are needed to allow for an early recognition of aggressive versus indolent forms of PCa. Furthermore, the development of novel targeted therapies for advanced PCa, e.g. immunotherapies, necessitates the development of future predictive biomarkers as companion diagnostics.
Originally, CXCL12 was found to control hematopoietic cell trafficking and guide the homing of stem cells within the bone marrow [10,11]. Thereby, cells with CXCR4 expression migrate towards compartments with high CXCL12 production. In the bone marrow, CXCL12 signals lead to cell chemotaxis until they are terminated through CXCR4 receptor internalization [12]. In the long-run, steady CXCL12 binding causes CXCR4 desensitization resulting in a resting, non-deploying state of hematopoietic stem cells and tumor cells, likewise [13]. This implies that a microenvironment with a constant CXCL12 production induces a down-regulation of CXCR4 and thereby antagonizes the process of metastasis [10]. Gene methylation adds to the major regulatory alterations interfering with CXCL12 homeostasis in tumor epithelium, leading to a silenced CXCL12 signal in various human malignant tumors [14][15][16][17].
Moreover, the CXCR4/CXCL12 axis seems to be tightly linked to the immune checkpoint regulator programmed death 1 (PD-1) and programmed deathligand 1 (PD-L1) that co-operate to suppress anti-cancer immunity [18]. In the clinical setting, anti-PD-1 and anti-PD-L1 antibodies have shown promising outcomes in cancer patients. However, a subset of patients fails to respond to checkpoint inhibition [19]. Feig et al. recently managed to overcome anti-PD-L1 drug resistance in a murine pancreatic adenocarcinoma model by blocking CXCR4. These findings motivated us to investigate the CXCR4/CXCL12 axis within the cross-relation to PD-L1 expression in PCa results

Analytical assay performance
Methylation levels of CXCL12 were determined by quantitative methylation-specific real-time PCR detecting methylated and unmethylated CXCL12 promoter sequences. The assay was designed within a CpG island upstream of the CXCL12 gene ( Figure 1A). The analytical performance of the assay was tested using mixtures of bisulfite converted artificially methylated and unmethylated DNA. Figure 1B shows that the assay allowed for the accurate quantification of CXCL12 methylation within the whole spectrum of 0 -100% methylation (r 2 = 0.98).

Frequency of CXCL12 gene methylation in human prostate tissue samples
In a test case control study, we evaluated a total of 66 human prostate tissue samples from 25 radical prostatectomy specimens regarding their CXCL12 methylation (mCXCL12) status in normal prostate parenchyma, hyperplastic adenomyomatous tissue, and carcinomatous tissue. These 25 PCa specimens were representative of a wide spectrum of the disease and comprised the following Gleason grading groups (GG): GG  I 28%, GG II-III 28%, GG IV 24, and GG V 20%. Among  non-carcinomatous samples, we analyzed 22 normal  tissue samples remote from the tumor front and 19 benign hyperplastic cases. Compared to PCa samples, CXCL12 methylation was altogether significantly lower in benign and normal specimens compared to PCa samples (p < 0.05 and p < 0.01, Figure 2). Mean mCXCL12 values of normal and PCa tissue were narrowly confined to 39.5% [31.1%-48.1%] and 56% [45.4%-68.5%], respectively. Dispersion around the mean of mCXCL12 in PCa samples, however, was significantly more prominent (Χ 2 = 14.1, p < 0.001, Bartlett's test for equal variances). In detail, one group of patients showed values offsetting over approximately 70% methylation and another group was observed at a distance below approximately 30% methylation ( Figure 2).
Differences in the spectrum of mCXCL12 prompted us to investigate the clinical performance of mCXCL12 in two independent cohorts of patients that had undergone radical prostatectomy.

Methylation and expression of CXCL12 in the training cohort
Full clinicopathological characteristics of the training cohort are given in Table 1. Among prognostic clinicopathological variables, a significant association was found between high mCXCL12 and high Gleason grading groups (p < 0.001) as well as nodal positive (pN1) patients (p = 0.038; Table 1).
In the training cohort, mCXCL12 revealed a symmetric distribution covering a broad spectrum of values (0.04% -98.40%) with a major peak seen at 55% CXCL12 methylation, comparable with the results shown for the test case control series. Aiming at reproducing the characteristics of pathological mCXCL12 distribution in PCa, methylation values were trichotomized to obtain qualitative results. Two optimized cut-offs were introduced at 33.06% and 70.68%; mCXCL12 low , mCXCL12 medium , and mCXCL12 high referring to the respective groups.

CXCL12 methylation and biochemical recurrence-free survival analyses: training cohort
Subsequently, we analyzed whether mCXCL12 allows for the stratification of patients at risk for    Table 2). Trichotomized mCXCL12 stratified patients according to BCR in the Kaplan-Meier analysis (χ 2 = 6.13, p = 0.047, Figure 3A).
For univariate Cox proportional hazard analysis, mCXCL12 low and mCXCL12 high were combined in one group. In the training cohort, 44.5% of patients fell within the mCXCL12 low/high group and 55.5% fell within the mCXCL12 medium group. Abnormal mCXCL12 low/high was associated with significantly shorter BCR-free survival (hazard ratio (HR) = 2.11 [95%CI: 1.15-3.87], p = 0.016, Table 2). clinical performance of mCXCL12 in the testing cohort Table 1 displays the full clinicopathological characteristics of the testing cohort. The results from the testing cohort are entirely based upon data generated by the TCGA Research Network: http://cancergenome. nih.gov/. mCXCL12 significantly correlated with age (p = 0.004), Gleason grading group (p = 0.034), and surgical margins/pT3b-pT4 categories (p = 0.005).
Analogous to the approach in the training cohort, mCXCL12 low and mCXCL12high groups were combined to mCXCL12 low/high for univariate Cox proportional hazard analysis. In the testing cohort, 39.6% of patients fell within the mCXCL12 low/high group and 60.4% fell within the mCXCL12 medium group.

mCXCL12 in relation to preoperative serum PsA
It has been shown that CXCL12 can help to distinguish between benign lesions and PCa among patients presenting with low serum PSA [20], which prompted us to evaluate mCXCL12 with respect to the initial serum PSA in the full cohort of patients. Patients were stratified according to initial PSA levels. Mean mCXCL12 did not differ between low-level PSA secreting (PSA < 4 ng/ml), medium-level PSA secreting (PSA ≥ 4 ng/ml, ≤ 10 ng/ml), and high-level PSA secreting (PSA > 10 ng/ml) tumors in both cohorts under investigation (Table 1). mCXCL12 low/high was significantly associated with BCR in medium-level PSA secreting PCa in the univariate Cox proportional hazard model (Table 3).

Association with Pd-l1 expression in the training cohort
Very recently, Gevensleben et al. showed that PD-L1 expression served as an independent prognostic marker in PCa in the same cohort of patients [21]. We correlated mCXCL12 with the previously published PD-L1 expression data. Mean mCXCL12 did not differ between PD-L1 high and PD-L1 low PCa. However, Spearman's rank correlation revealed a trend towards an association of PD-L1 with mCXCL12 (ρ = 0.132, p = 0.084). Therefore, subgroups were defined according to PD-L1 expression in PCa: PD-L1 high PCa were flagged and allocated to mCXCL12 low , mCXCL12 medium , and mCXCL12 high groups, respectively. A total of 85 patients presented with PD-L1 low , while 87 PD-L1 high patients were split into 15 mCXCL12 low , 45 mCXCL12 medium , and 27 mCXCL12 high . In Kaplan-Meier analysis, PD-L1 low PCa showed the longest BCR-free survival (mean estimated 112 months), while mCXCL12 medium showed best BRC-free survival rates among PD-L1 high PCa (mean estimated 107 months), and PD-L1 high /mCXCL12 low , and PD-L1 high /mCXCL12 high showed short BCR-free survival times (mean estimated 52 months and 83 months, respectively; n = 151, χ 2 = 12.99; p = 0.005; Figure 3C).

concordance, accuracy, and robustness of mCXCL12 detection in biopsies and ectomies
It has previously been shown that methylation assays precisely report methylation status even in formalin-fixed paraffin-embedded (FFPE) tissues [22,23]. In order to test the utility of our assay for the analysis of biopsy specimens, we evaluated the robustness, accuracy, and concordance of mCXCL12 testing in small biopsy specimens and the matched ectomy in a biospy test case series including 10 patients. Therefore, we correlated mCXCL12 obtained from pooled biopsy cores of an individual patient with mCXCL12 detected in his total tumor volume after radical prostatectomy ( Figure 5A). mCXCL12 levels detected in biopsies significantly correlated with those obtained from the total tumor volume (r = 0.76; p = 0.019) indicating that mCXCL12 in biopsies represents the methylation level found in the tumor. In a second step, we tested whether mCXCL12 can robustly and accurately be tested in small samples and over a wide range of DNA yield. Patients of the biopsy test case series were divided into two groups: five patients with mCXCL12 ranging from 15.5% -41.2% were assigned to group A and 5 patients with mCXCL12 ranging from 42.7%-65.2% were allocated to group B. Tumor DNA samples of each group were pooled and measured tenfold with two different input amounts (30 ng and 6 ng, respectively) as displayed in Figure 5B

dIscussIon
Recent studies have attracted notice to the CXCR4/ CXCL12 axis in metastasized PCa [24,25]. In breast and colon cancer, lower intrinsic levels of the chemoattractant chemokine CXCL12 due to hypermethylation of its promoter disrupt cellular feedback mechanisms to internalize membranous CXCR4 [14,15]. This, in consequence, results in higher tumor cell motility and an augmented capability for metastasis [14,15]. Our data support the postulate, that CXCL12 methylation downregulates tumor intrinsic CXCL12 protein expression in PCa as well, thereby fostering metastasis.
Due to its negative effect on CXCR4 expression, tumor intrinsic CXCL12 may be a valuable biomarker in clinical situations that require decision-making. Immunohistochemical CXCL12 staining, however, did not discriminate between PCa with favorable and adverse outcome in the present study. Abnormal methylation of the CXCL12 gene, in contrast, was associated with a shorter BCR-free survival of primary PCa. In two independent cohorts of patients after radical prostatectomy, CXCL12 hypo-and hypermethylation served as a prognostic biomarker identifying patients with early BCR and qualified as an independent prognostic factor in the testing cohort.
The fact that the cohort is based on radical prostatectomy specimens, however, is a major limitation of this study. Gene methylation can be robustly determined even in little formalin-fixed pathological material [22,23]. If the prognostic value of CXCL12 methylation can be transferred to a biopsy-based cohort, its analysis may be a promising tool in order to identify patients with innocuous tumors that can safely be followed by active surveillance protocols. Conversely, aberrant methylation of CXCL12 may indicate a high metastatic potential. In consequence, affected patients possibly will benefit from radical treatment, including radical prostatectomy. In a paired analysis of CXCL12 methylation based on biopsy and matched ectomy specimens, we were able to show that the results of mCXCL12 assessments are concordant. In addition, our assay seems to be able to reliably determine CXCL12 methylation in biopsy material. The variation coefficients of our results were slightly higher the lower the amount of applied DNA was, however, in homoscedastic testing, no difference could be observed. In practice, growing variances in small samples may easily be counteracted by increasing the number of replicates in the analysis. If the assay will be established for future routine use, a broad spectrum of biopsy material will be needed to precisely set the measuring range and achieve tight tolerances. Immunotherapy has been emerging as a promising treatment strategy in cancer patients [26][27][28]. Among major immune checkpoints, e.g. PD-1 and CTLA4, the pharmacologic blockage of CXCR4 and its endogenous ligand CXCL12 has appeared as a prime target for blocking strategies in the treatment of cancer patients [10]. Recent studies have focused on pharmacological CXCR4 inhibition within the cross relation of anti-PD-1/anti-PD-L1 drug resistance [29,30]. In vitro and murine in vivo studies have provided data that ADM3100 induced blockage of CXCR4 restores sensitivity to PD-1 and CTLA-4 checkpoint inhibitors in anti-PD1 and anti-CTLA-4 drug-resistant tumors and thus might help to overcome resistance to immune therapy [10,18,30]. Consequently, a combined anti-CXCR4/CXCL12 and anti-PD1/PD-L1 (Ulocuplumab, BMS-936564/ Nivolumab, BMS-936558) therapy is under clinical investigation in metastatic solid tumors (CXCessoR4 trial, ClinicalTrials.gov Identifier: NCT02472977). In our study, patients with high PD-L1 expression and aberrant CXCL12 methylation presented with significantly shorter BRC-free survival intervals than patients with either low PD-L1 expression or high PD-L1 expression plus normal CXCL12 methylation. Accordingly, a combined anti-PD-L1/PD and antagonistic CXCR4/CXCL12 treatment seems to be a promising approach in PCa as well. For the identification of eligible patients, methylation of CXCL12 together with immunohistochemical staining of PD-L1 might be a powerful tool. Whether carcinomatous CXCL12 methylation predicts responsiveness to therapeutic intervention targeting the CXCR4/CXCL12 axis remains the object of future research.   Patients were stratified according to CXCL12 DNA methylation and BCR-free survival was analyzed in three subgroups of patients: PSA < 4 ng/ml, PSA 4-10 ng/ml, and PSA > 10 ng/ml. ND: not determined due to low number of events.

sample preparation
For methylation analysis, ectomy samples were processed according to the InnuConvert Bisulfite All-In-One Kit (Analytik Jena, Germany) as previously published [31]. Bisulfite DNA from biopsies were prepared as described earlier [23]. Length of tumor infiltrate in one tissue biopsy ranged from 0.1 mm to 12 mm (median 4 mm). Number of tumorous cores in one set of biopsies ranged from 1-8 (median 4). Tumorous tissue was marked, micro-dissected, and further processed as described above. For comparison with methylation results obtained from matched biopsies, total tumorous tissue from radical prostatectomy specimens was micro-dissected and bisulfite converted.
For the analytical performance verification of the assay, a dilution series of bisulfite-converted artificially methylated DNA (CpGenome TM Universal Methylated DNA; Merck Millipore, Darmstadt, Germany) and unmethylated DNA (NW Andrology & Cryobank Inc., Spokane, WA, USA) was used.

CXCL12 quantitative methylation-specific real-time Pcr
Gene methylation of CXCL12 was quantified by a quantitative-methylation real-time PCR assay, with the primers and probes shown in Table 4. The assay was performed using an AB 7500 Fast Real-Time PCR System (Life Technologies Corporation, Carlsbad, CA, USA). If not stated otherwise in the result section, DNA input was 25 ng of bisulfite-converted DNA from FFPE ectomy tissue for each PCR reaction. 5 µl DNA from biopsy samples was applied without prior quantification. A calibrator sample (50:50 mixture of bisulfite-converted artificially methylated and unmethylated DNA) was used. Each sample was measured in triplicate. Sample measurements were considered valid when the following quality criterion was met: CT CXCL12-P-U or CT CXC12-P-M < 35 Other samples were excluded from the analysis due to insufficient DNA yield. The CXCL12 methylation was calculated using the ΔΔCT method: ΔCT = Δ CTCXCL12-P-U -ΔCT CXCL12-P-M , ΔΔCT = ΔCT sample -ΔCT calibrator , mCXCL12 = 100%/(1 + 2 ΔΔCT ).

Immunohistochemical analyses of Pd-l1 expression and CXCL12 expression
PD-L1 expression data were collected from our previous study [21]. In brief, the anti-PD-L1 antibody clone EPR1161(2) (Abcam, Cambridge, UK) was used. Specific membranous and cytoplasmic staining of epithelial tumor cells was considered positive [21]. The intensity of PD-L1-staining was scored semiquantitatively as negative (0), weak (1), moderate (2), or strong (3) by two independent and blinded pathologists. For each patient a PD-L1 expression value was computed. Up to five cores per patient were scored semiquantitatively and mean averaged. Patients' stratification for Kaplan-Meier analysis was conducted using the median PD-L1 expression of the total cohort.
Ninety-one cases could be evaluated in immunohistochemical CXCL12 staining. CXCL12 staining was observed in the nuclei as well as in the cytoplasma of basal cells of PCa glands. Nuclear and cytoplasmic staining was assessed as negative (0), weak (1), or strong (2). Scores were added and averaged over 2-3 tissue scores. For correlation analysis and survival analysis, scores were dichotomized by the median of the total cohort.

statistical analyses
Statistical analysis was performed using SPSS, version 22 (SPSS Inc., Chicago, IL). The relationship between input DNA methylation and measured DNA methylation was assessed by linear regression. In the training and testing cohort, comparative studies of dichotomized CXCL12 methylation values were tested using the Mann-Whitney U test for paired analysis and using the Kruskal-Wallis test in cases when more than two groups were compared. Statements regarding potential correlations were made using the Spearman's correlation coefficient ranked between two assessed variables. CXCL12-R 5′-aaactaacttatttacttttcatta -3′ CXCL12-P-M 5′-Atto647N-cgccatcgaaaaaccccgtccc-BHQ-2 -3′ CXCL12-P-U 5′-HEX-caccatcaaaaaaccccatccc-BHQ-1 -3′