PITX3 promoter methylation is a prognostic biomarker for biochemical recurrence-free survival in prostate cancer patients after radical prostatectomy

Background Molecular biomarkers that might help to distinguish between more aggressive and clinically insignificant prostate cancers (PCa) are still urgently needed. Aberrant DNA methylation as a common molecular alteration in PCa seems to be a promising source for such biomarkers. In this study, PITX3 DNA methylation (mPITX3) and its potential role as a prognostic biomarker were investigated. Furthermore, mPITX3 was analyzed in combination with the established PCa methylation biomarker PITX2 (mPITX2). Methods mPITX3 and mPITX2 were assessed by a quantitative real-time PCR and by means of the Infinium HumanMethylation450 BeadChip. BeadChip data were obtained from The Cancer Genome Atlas (TCGA) Research Network. DNA methylation differences between normal adjacent, benign hyperplastic, and carcinomatous prostate tissues were examined in the TCGA dataset as well as in prostatectomy specimens from the University Hospital Bonn. Retrospective analyses of biochemical recurrence (BCR) were conducted in a training cohort (n = 498) from the TCGA and an independent validation cohort (n = 300) from the University Hospital Bonn. All patients received radical prostatectomy. Results In PCa tissue, mPITX3 was increased significantly compared to normal and benign hyperplastic tissue. In univariate Cox proportional hazards analyses, mPITX3 showed a significant prognostic value for BCR (training cohort: hazard ratio (HR) = 1.83 (95 % CI 1.07–3.11), p = 0.027; validation cohort: HR = 2.56 (95 % CI 1.44–4.54), p = 0.001). A combined evaluation with PITX2 methylation further revealed that hypermethylation of a single PITX gene member (either PITX2 or PITX3) identifies an intermediate risk group. Conclusions PITX3 DNA methylation alone and in combination with PITX2 is a promising biomarker for the risk stratification of PCa patients and adds relevant prognostic information to common clinically implemented parameters. Further studies are required to determine whether the results are transferable to a biopsy-based patient cohort. Trial registration: Patients for this unregistered study were enrolled retrospectively.


Background
Prostate cancer (PCa) is the most common cancer in men in the western hemisphere. In 2015, 220,800 new cases and 27,540 tumor-related deaths were predicted for the USA [1]. In the last couple of decades, prostatespecific antigen (PSA) screening has increased the number of early detected PCa [2]. However, the natural course of these tumors is highly variable. A majority of cases progresses slowly, remains years to decades in a clinically dormant state, and may be safely kept under active surveillance. Others develop fast and lead to locally aggressive growth and metastasis after short courses of disease. In the long run, these patients might benefit from a more radical treatment when diagnosed at a very early stage. Clinicopathological parameters, i.e., PSA values, tumor size, number of positive biopsies, and Gleason grading groups, as suggested by the International Society of Urological Pathologists (ISUP), guide the decision-making process when determining whether a patient may benefit from radical prostatectomy or can instead be closely monitored. However, in many cases, this approach has not proven satisfactory in that patients either suffered from overtreatment or experienced very early PSA relapses after surgery [3,4]. Therefore, new prognostic tools are still urgently needed to distinguish between the aggressive and indolent subtypes of PCa.
As potential biomarkers, epigenetic modifications such as hyper-or hypomethylation of tumor-related genes have lately emerged as one of the key alterations in cancer development [5][6][7]. Aberrant patterns of methylation have aroused interest in the molecular subclassification of urologic tumors and might potentially serve as prognostic and predictive biomarkers in PCa [8,9]. Furthermore, DNA is a highly robust cellular element that can be extracted reliably from different materials, e.g., fresh tissue, formalin-fixed paraffin-embedded tissue (FFPET), and body fluids [10,11].
Methylation of the paired-like homeodomain transcription factor 2 (PITX2) has been successfully proven a powerful prognostic biomarker in several cancer entities such as lung cancer [12], hormone-receptornegative breast cancer [13][14][15][16], and PCa [17][18][19]. PITX2 is initiated by Wnt/β-catenin and is involved in the control of cell proliferation [20]. PITX2 regulates the expression of the androgen receptor (AR) and insulin-like growth factor (IGF) receptor genes, leading to the regulation of signaling pathways involving AR and IGF during PCa progression [21].
The paired-like homeodomain transcription factor 3 or pituitary homeobox 3 (PITX3) is a transcription factor belonging to the same protein family as PITX2 [22]. PITX3 has been shown to be transiently expressed in the eye lens and skeletal muscle during embryogenesis [23,24]. Very recently, it has been reported that exposure to environmental xenoestrogens may lead to neonatal DNA methylation re-programing effects in the prostate including dysregulation of PITX3 methylation [25]. This may potentially foster carcinogenesis. Moreover, PITX3 has previously been found to be aberrantly methylated in breast cancer patients [26].
These findings prompted us to investigate PITX3 promoter methylation in PCa in a publically available dataset from The Cancer Genome Atlas (TCGA) [27] (training cohort) and an independent primary PCa patient cohort from the University Hospital Bonn (validation cohort).

Results
PITX3 and PITX2 promoter methylation in prostate tissues from TCGA training cohort For the analysis of PITX3 promoter methylation (mPITX3) in the training cohort, results obtained from two Illumina Infinium HumanMethylation450 BeadChip beads from the TCGA dataset (cg12324970 and cg23095743) were used. Both beads were located within the CpG island of PITX3 (Fig. 1a). Firstly, PCa (n = 498) and normal adjacent tissue (NAT, n = 50) samples from the training cohort were analyzed with respect to mPITX3. Patient samples showed a significantly lower level of mPITX3 in NAT compared to PCa samples (p < 0.001, Fig. 2a). A histogram of mPITX3 showed a bell curve with a minor depression at ≈68 % (Fig. 3a). mPITX3 levels as a continuous variable were related to prognostic clinicopathological variables and were found to be significantly correlated with the ISUP Gleason grading group (ρ = 0.112; p = 0.012), pathologic tumor (pT) category (ρ = 0.123; p = 0.006), presurgical PSA (ρ = 0.134; p = 0.003), and the AR activity score (ρ = 0.154; p = 0.005) as obtained from TCGA [27] in the training cohort. In order to analyze the suitability of mPITX3 for the stratification of patients at risk for biochemical recurrence (BCR), mPITX3 was dichotomized by an optimized cutoff (mPITX3 low < 68.2 % ≤ mPITX3 high ; Table 1) which was identified by an iterative approach. In the training cohort, mPITX3 high was significantly associated with BCR in the univariate Cox proportional hazards model (hazard ratio (HR) = 1.83 (95 % CI 1.07-3.11); p = 0.027; (Table 2)) and the Kaplan-Meier analysis (likelihood ratio (LR) = 5.05; p = 0.025, Fig. 3b).
PITX3 carries a homeodomain which is highly homologous with PITX2 [28]. In a second step, PITX2 promoter methylation (mPITX2) was therefore analyzed in an equivalent manner. A PITX2 quantitative methylationspecific real-time polymerase chain reaction (PCR) (qMSP) assay has been established and validated in a previous study using other patient material [29]. Three beads from the HumanMethylation450 BeadChip which are located in close proximity of the established qMSP assay were selected. mPITX2 showed a rather asymmetrical distribution (Fig. 3c). Associations of mPITX2 levels with clinicopathological variables in the training cohort are shown in Table 3. In brief, mPITX2 levels correlated with age, T category, ISUP Gleason grading group, surgical margin, and ETS-related gene (ERG) fusion status. Dichotomization by an optimized cutoff (mPITX2 low < 34.3 % ≤ mPITX2 high ) revealed a significant prognostic value. In the training cohort, mPITX2 high was significantly associated with BCR in the univariate Cox proportional hazards model (HR = 2.20 (95 % CI 1.25-3.87); p = 0.006) and the Kaplan-Meier analysis (LR = 7.95; p = 0.005, Fig. 3d).
Since both parameters showed excellent prognostic performance, the combination of mPITX2 and mPITX3 was tested in the TCGA collective. Here, mPITX2 low and mPITX3 low cases showed significantly longer BRC-free survival compared to patients with mPITX2 high and/or mPITX3 high (LR = 12.70, p = 0.002; Fig. 4a).
Analytical assay design and performance of the mPITX3 real-time PCR Following the analysis of the training cohort, a PITX3 quantitative methylation (QM) assay was designed within the CpG island upstream of the PITX3 gene in the same region as the beads selected from the HumanMethylation450 BeadChip analyzed by TCGA Research Network (Fig. 1a). In contrast to the established qMSP used to quantify PITX2 methylation as described earlier [29], QM assay refers to an assay which is based on two primers which do not cover any CpG sites and therefore amplify unmethylated as well as methylated DNA. This assay contains two detection   Frequency and prognostic value of mPITX3 and mPITX2 in the training (n = 498) and validation (n = 300) cohorts. PITX3 and PITX2 DNA methylation was analyzed in prostate carcinoma patients from two cohorts. Methylation frequencies (a, c, and e) and Kaplan-Meier analyses of BCR-free survival in patients stratified according to dichotomized mPITX3 and mPITX2 levels are shown (b, d, and f). a mPITX3 analysis in the training cohort revealed a symmetric, bell-shaped distribution covering a broad spectrum of values (22-92 %). An optimal cutoff was elaborated by an iterative approach (68.2 %) stratifying patients into mPITX3 hyper-(mPITX3 high ) and hypomethylated (mPITX3 low ) cases. b Patient survival in the training cohort according to mPITX3 low and mPITX3 high status. Patients with mPITX3 low tumors show a better prognosis. Approximate mean BCR-free survival: 93 months (mPITX3 low , 95 % CI 85-100 months, n = 301) and 76 months (mPITX3 high , 95 % CI 63-90 months, n = 117; LR = 5.05; p = 0.025), respectively. c mPITX2 analysis in the training cohort revealed an uneven distribution covering an altogether lower spectrum of values than mPITX3 (5-79 %). An optimal cutoff was elaborated by an iterative approach (34.3 %) stratifying patients into mPITX2 hyper-(mPITX2 high ) and hypomethylated (mPITX2 low ) cases. d Patient survival in the validation cohort according to mPITX2 low and mPITX2 high status. Patients with mPITX2 low tumors show a better prognosis. Approximate mean BCR-free survival: 96 months (mPITX2 low , 95 % CI 88-105 months, n = 220) and 78 months (mPITX2 high , 95 % CI 67-89 months, n = 198; LR = 7.95; p = 0.005), respectively. e mPITX3 analysis in the validation cohort revealed a flattened, bell-shaped distribution covering (5-100 %). An optimal cutoff was elaborated by an iterative approach (61.8 %) stratifying patients into mPITX3 hyper-(mPITX3 high ) and hypomethylated (mPITX3 low ) cases. f Patient survival in the validation cohort according to mPITX3 low and mPITX3 high status. Patients with mPITX2 low tumors show a better prognosis. Approximate mean BCR-free survival: 125 months (mPITX3 low , 95 % CI 118-132 months, n = 145) and 103 months (mPITX3 high , 95 % CI 91-115 months, n = 105; LR = 11.17; p = 0.001), respectively. Patient survival in the validation cohort according to mPITX2 low and mPITX2 high status is reported elsewhere [30]   probes: One detection probe specifically binds to unmethylated DNA while the other probe specifically and competitively binds to methylated DNA. The assay performance was validated using a dilution series of bisulfite-converted artificially methylated and unmethylated DNA. The assay allowed for an accurate quantification of mPITX3 within the whole spectrum from 0 to 100 % methylation (r 2 = 0.98, Fig. 1b).

PITX3 promoter methylation in prostate tissues in the test study
In order to avoid artifacts which might result from a genome-wide methylation testing approach as used by the TCGA, the aforementioned findings from the TCGA cohort were confirmed in a small test study comprised of 71 samples from 25 prostatectomy specimens. mPITX3 levels were significantly lower in NAT and samples with benign prostatic hyperplasia (BPH) compared to PCa samples (p < 0.001, Fig. 2b). No difference of mPITX3 was detected in BPH compared to NAT samples.
Since combined mPITX2 and mPITX3 revealed significant additive prognostic information in the training cohort, the combination of mPITX2 and mPITX3 was also tested in the validation cohort. According to the results obtained from the training cohort, mPITX2 low and mPITX3 low cases showed significantly longer BRC-free survival compared to patients with mPITX2 high and/or mPITX3 high (LR = 12.14, p = 0.002; Fig. 4b).

Discussion
In this study, PITX3 was shown to be aberrantly methylated in prostate carcinomas. PITX3 was hypermethylated in PCa compared to normal adjacent prostate tissue in the training cohort and compared to both normal and benign prostatic hyperplasia in the test study. These findings are in line with previous reports on PITX3 methylation in breast carcinoma [26].
Furthermore, carcinomatous PITX3 hypermethylation was significantly associated with established clinicopathologic parameters characteristic of PCa. In detail, high ISUP Gleason grading group, advanced tumor stages, and high preoperative PSA values were related to high PITX3 methylation in both cohorts. In addition, PITX3 methylation correlated with a molecular AR activity score as obtained from TCGA Research Network [27],   [27] which was only available for the training cohort. An association with the ERG fusion protein or ERG protein expression could not be determined. Recently, dysregulation of PITX3 methylation has been linked to the environmental burden of xenoestrogens [25]. In this respect, PITX3 methylation may have an exceptional position among prognostic biomarkers. Of note, PITX3 methylation served as a prognostic biomarker for BCR in both the training and validation cohort of radical prostatectomy patients. In Kaplan-Meier analysis, high PITX3 methylation defined by an optimized cutoff for both patient groups was associated with a shorter BCR-free survival in the training and validation cohort. As a limiting condition, however, the follow-up period was shorter in the training cohort compared to the validation cohort, and the training cohort comprised significantly more high-grade carcinomas with an advanced stage and associated with earlier BCR. In the validation cohort, PITX3 methylation succeeded as a prognostic factor dichotomized by an optimized cutoff and as a continuous variable in the univariate Cox proportional hazards analysis.
In consideration of the fact that several recent studies have reported on a striking prognostic power of gene methylation of PITX2 [17,19], a close relative of PITX3, a combined analysis of PITX2 and PITX3 promoter methylation, was performed. Thereby, we intended to investigate possible interactions to compensate for gene methylation in either PITX member. Combined analysis of PITX2 and PITX3 promoter methylation revealed that low methylation in both genes was associated with favorable courses of disease in each cohort. Vice versa, patients with hypermethylated PITX2 and PITX3 promoters presented with the shortest BCR-free survival intervals after radical prostatectomy. Intermediate BRCfree survival intervals were observed in patients with low gene methylation in one PITX member and high methylation in the other PITX members. In respect thereof, we conclude that the analysis of PITX3 gene methylation adds to the prognostic information obtained from PITX2 analysis, suggesting that, in contrast to their overlapping functions in human development, they play a distinct role in the genesis and progression of PCa. This issue further needs to be confirmed in larger studies in which patient numbers allow for multivariate analysis. Furthermore, the prognostic value should be analyzed with regard to more clinically relevant endpoints, i.e., prostate cancer-specific survival, which unfortunately was not available for the present analyses. The present study indicates that PITX3 promoter methylation may be of great value for the tailoring of individual therapies and risk stratification. Even though PSA screening has led to a reduction of cases with advanced disease and disease-specific mortality, low-risk PCa rarely causes symptoms or affects survival if left untreated. Nevertheless, most men diagnosed with low-risk PCa in the USA receive up-front treatment, including prostatectomy or radiotherapy [31]. Hence, the early detection of low-risk PCa may lead to overdiagnosis resulting in overtreatment of patients with potential unnecessary side effects such as urinary dysfunction or impotence [3,4]. The present study combines the analysis of PITX3 promoter methylation in two independent . Patients with high methylation in PITX2 and PITX3 genes present with the earliest BCR events (n = 32). Patients with low methylation in one PITX gene member and high methylation in the other PITX gene member (n = 82) show the highest number of BCR events, however, more protracted than patients with high methylation in both PITX genes cohorts and by two different molecular assays; however, further studies are warranted to scrutinize the potential of PITX3 methylation as a biomarker prior to radical prostatectomy. Therefore, the assay's prognostic power needs to be evaluated in biopsies from PCa patients included in an active surveillance protocol.

Conclusions
In summary, PITX3 DNA methylation is a promising biomarker for the risk stratification of PCa patients and adds relevant prognostic information to the common clinically implemented parameters. The prognostic power of PITX3 DNA methylation was validated in two independent radical prostatectomy cohorts. Adjunct to the analysis of PITX2 promoter methylation, hypermethylation of PITX3 provided supplemental information on the course of disease, indicating adverse patient outcome. This implies a distinct function of the PITX3 gene in the development of PCa. However, the establishment of PITX3 as a clinical prognostic marker needs to be established in further studies reappraising its transferability to biopsy-based patient cohorts.

Patients and clinical endpoint Test study
A set of 71 FFPE prostate tissue samples from 25 PCa patients who underwent therapy at the University Hospital of Bonn in 2011 were included. The samples included 25 PCa, 24 NAT, and 22 BPH specimens.

Patient training cohort
A patient cohort comprised of 498 patients from the TCGA Research Network. Two Illumina Infinium HumanMethy-lation450 BeadChip beads (cg12324970 and cg23095743) were used to calculate relative methylation levels of PITX3 by the formula 100 %*bead_M/(bead_M + bead_U). The average value of the ratios of the beads cg12324970 and cg23095743 was calculated. BCR-free survival was considered as the primary endpoint of the study. For PITX2, three Illumina Infinium HumanMethylation450 BeadChip beads (cg10391633, cg01616926, and cg19134945) were analyzed, accordingly.

Patient validation cohort
A patient cohort comprised of 300 patients with histologically confirmed PCa who underwent radical prostatectomy at the University Hospital Bonn between 1998 and 2008. BCR-free survival was considered the primary endpoint of the study and was determined as elevation PSA levels above 0.2 ng/ml.

Sample preparation
For the analysis of PITX3 methylation, the FFPET samples were processed according to the InnuCONVERT Bisulfite All-In-One Kit (Analytik Jena, Germany) as previously published [10]. To validate the assay performance, a serial dilution of bisulfite-converted artificially methylated DNA (CpGenome™ Universal Methylated DNA; Merck Millipore, Darmstadt, Germany) and unmethylated DNA from human sperm (NW Andrology & Cryobank Inc., Spokane, WA, USA) was used. As a calibrator sample DNA, a 1:1 mixture of bisulfiteconverted unmethylated and artificially methylated DNA was used.
mPITX2 and mPITX3 quantitative real-time PCR The DNA methylation of PITX2 and PITX3 was determined by means of qMSP and QM PCR assays, respectively. The PITX2 qMSP assay has been described earlier in detail [29]. Table 4 lists the primers and probes used for the QM PITX3 assay. Each sample was measured in triplicate with an input of 25 ng bisulfite-converted DNA per reaction. The PITX3 QM assay was performed using an AB 7500 Fast Real-Time PCR System (Life Technologies Corporation, Carlsbad, CA, USA), and the following temperature profile was used: 15 min at 95°C (first denaturation), followed by 45 cycles of 95°C for 15 s, 60°C for 2 s, and 55°C for 60 s. The thresholds and baselines for analysis were set as follows: 0.02 (threshold) and 3-22 (baseline) for the methylated and unmethylated probe. mPITX3 was calculated using the ΔΔCT method: Immunohistochemistry Immunohistochemical staining of ERG and AR was conducted at the Institute of Pathology in Bonn. Staining of the sections was performed using the LabVision Autostainer 480S system (Thermo Scientific, Waltham, MA, USA) along with the Thermo Scientific Reagents and the N-Histofine® DAB-3S detection kit. The AR staining was  performed as previously described [32]. For the ERG staining, the following antibody and dilution was used: clone EPR3864 (Biologo, Kronshagen, Germany; 1:100).

Statistical analyses
The statistical analyses were performed using SPSS, version 22 (SPSS Inc., Chicago, IL). The relationship between input DNA methylation and measured DNA methylation was assessed by linear regression. Statements regarding potential correlations of specific histology findings were made using the Spearman's rank correlation coefficient (ρ). BCR-free survival analyses were conducted by Kaplan-Meier and univariate Cox proportional hazards regression analyses. Kaplan-Meier analysis was conducted using the log-rank test and likelihood ratios (LR). p values lower than 0.05 were considered significant. For the comparison of independent groups, Wilcoxon-Mann-Whitney test (for two groups) and the Kruskal-Wallis test (for more than two groups) were applied.