Clinical performance of an analytically validated assay in comparison to microarray technology to assess PITX2 DNA-methylation in breast cancer

Significant evidence has accumulated that DNA-methylation of the paired-like homeodomain transcription factor 2 (PITX2) gene can serve as a prognostic and predictive biomarker in breast cancer. PITX2 DNA-methylation data have been obtained so far from microarray and polymerase chain reaction (PCR)-based research tests. The availability of an analytically validated in vitro methylation-specific real-time PCR assay format (therascreen PITX2 RGQ PCR assay) intended for the determination of the percent methylation ratio (PMR) in the (PITX2) promoter 2 prompted us to investigate whether the clinical performance of these different assay systems generate comparable clinical outcome data. Mathematically converted microarray data of a previous breast cancer study (n = 204) into PMR values leads to a PITX2 cut-off value at PMR 14.73. Recalculation of the data to experimentally equivalent PMRs with the PCR PITX2 assay leads to a cut-off value at PMR 12 with the highest statistical significance. This cut-off predicts outcome of high-risk breast cancer patients to adjuvant anthracycline-based chemotherapy (n = 204; Hazard Ratio 2.48; p < 0.001) comparable to microarray generated results (n = 204; Hazard ratio 2.32; p < 0.0001). The therascreen PITX2 RGQ PCR assay is an analytically validated test with high reliability and robustness and predicts outcome of high-risk breast cancer patients to anthracycline-based chemotherapy.

receptor status are usually assayed by immunohistochemistry (IHC), HER2 by IHC or by Fluorescence In-Situ Hybridization (FISH) or uPA/PAI by enzyme linked immunosorbent assay (ELISA). Tumors classified as high-risk cancer with poor prognosis include triple-negative and HER2 positive breast cancer, patients with lymph node involvement 4 and patients classified through multigene mRNA signatures such as Endopredict ® (Myriad Genetics Inc., Salt Lake City, Utah, USA), OncotypeDX ® (Genomic Health, Redwood City, CA, USA) or Mammaprint ® (Agendia Inc., Irvine, CA, USA) to be at high risk to develop metastasis 5,6 .
Standard of care treatment of those high-risk breast cancer patients is an anthracycline-based chemotherapy. In addition, an adjuvant endocrine therapy will be administered if tumor cells are ER and/or PR positive. However, not all patients respond to anthracycline-based chemotherapy, and part of the patients suffer from side-effects without any benefit. Until now biomarkers to predict outcome to anthracycline-based chemotherapy are of high unmet medical need.
DNA-methylation of the PITX2 (paired-like homeodomain transcription factor 2) gene as a predictive and prognostic biomarker for patient selection has received increasing attention not only in breast cancer [7][8][9] . PITX2 is a transcription factor, which is involved in pituitary-specific gene regulation and left-right patterning during embryonic and organogenic development 9 . It has been shown that DNA-methylation of the PITX2 promoter gene predicts risk of distant metastasis in node-negative, hormone receptor-positive breast cancer [10][11][12] . Significant evidence has accumulated that PITX2 methylation predicts outcome in lymph node-positive, ER positive, HER2 negative breast cancer patients to adjuvant anthracycline-based chemotherapy and, thus, will support the clinicians to the most effective therapy option 13,14 . In these studies, different technological platforms for DNA-methylation analysis were used and performed on different types of tissue specimens. Those studies include a proprietary microarray platform developed by the company Epigenomics AG (Berlin, Germany) for on DNA methylation based biomarker screening using bisulfite converted DNA (bisDNA) specific probes for the methylated and unmethylated status of CpG residues 11,13 or a 'research use only' (RUO) real-time PCR assay covering three of the most relevant CpG residues for clinical response prediction in tamoxifen treated breast cancer patients 11 .
Recently, we developed an analytically validated PITX2 DNA-methylation assay (therascreen PITX2 RGQ PCR 15 , which is by now CE-marked and commercially available 9 . Here we describe results of the technical evaluation study of the therascreen PITX2 RGQ PCR assay according to Clinical and Laboratory Standards Institute (Clinical and Laboratory Standards Institute -CLSI; https://clsi.org/) guidelines for lymph node positive, ER positive, HER2 negative breast cancers. The therascreen PITX2 assay is deployable for routine diagnostic setting and the determined PMR value will aid clinicians to identify patients who are more versus less likely to benefit from adjuvant anthracycline-based chemotherapy. The goal of the present study was to investigate whether the clinical relevance of PITX2 DNA-methylation determined by the novel therascreen PITX2 DNA-methylation assay delivers the same results compared to the previously applied microarray technology 13 and research PCR assay 11 .

Material and Methods
Sample preparation, DNA extraction and bisulfite conversion. Sample preparation from Fresh-Frozen (FF) tumor tissue: after surgical removal, breast tumor tissues were placed immediately on ice until examination by the pathologist. Thereafter, tissue was snap-frozen and stored in liquid nitrogen until further use. FF tumor tissue was processed to tumor cell nuclei pelleted at 100,000 × g at Erasmus Medical Center, Rotterdam as described 13,16 . Samples were transferred to Therawis Diagnostics GmbH, Munich. Genomic DNA (gDNA) was extracted from 10-30 mg FF tumor cell nuclei pellets using the QIAamp® DNA Mini Kit (QIAGEN, Hilden, Germany; Catalog No.: 51304/51306).
Formalin-fixed paraffin-embedded (FFPE) tumor tissue samples were prepared by fixation for 12-24 hrs in 10% neutral buffered formalin and embedded in paraffin. Tissue blocks were stored at room temperature and archived at the Institute of Pathology, Klinikum rechts der Isar, Technische Universität München until further use. gDNA was purified from one to two 5 µm FFPE tissue sections with a total surface area ≥100 mm 2 using the QIAamp DNA DSP FFPE kit (Qiagen, Hilden, Germany; Catalog No.: 60404).
DNA quantification was performed with the QIAxpert spectrophotometer (Qiagen, Hilden, Germany) using the QIAamp DNA plugin with internal blank calibration for elution buffer ATE with the OD 260 readout of total nucleic acids. DNA bisulfite conversion of the gDNA was performed using the Epitect Fast DNA Bisulfite Kit (Qiagen, Hilden, Germany, Catalog No.: 59824/59826) using 200-1000 ng input into the bisulfite conversion according to the therascreen PITX2 RGQ PCR Kit workflow 15 . This equals a bisDNA input in the qPCR into quantitative PCR (qPCR) of 53-267 ng bisDNA/well. All samples are measured in duplicates.

Methylation-specific custom microarray.
A customized methylation-specific oligonucleotide array was developed by Epigenomics AG for 61 genes for initial screening in a population of 384 breast cancer samples 13 . CpG sites from regulatory regions of the candidate genes were amplified in multiplex PCR reactions labeled with the fluorochrome Cy5 from bisulfite-treated genomic DNA with bisulfite-converted DNA specific primer pairs. In total, 64 PCR amplificants representing 61 genes were pooled and hybridized to the microarray on which detection oligonucleotides for methylated (CG) and nonmethylated (TG) gene copies were spotted. This allowed for simultaneous quantitative measurement of unmethylated and methylated copies of the genes. Microarrays included 4 oligonucleotide pairs for each of the 64 PCR amplificates (total of 256 pairs). Each probe pair covered between one and three CpG dinucleotides in the regulatory regions of the respective candidate gene. The methylation score for each CpG site was calculated from the fluorescence intensity values of the methylated (=FIm) and unmethylated (=FIu) oligonucleotides. To stabilize the variance, the score was transformed using the generalized log transformation (gLOG): methylation score = gLOG(FIm/FIu) 17 . For statistical analysis, methylation scores for each amplificate were determined by averaging measurements from all probe pairs belonging to one amplificate using the median. Multiple amplificates from the same candidate gene entered data analysis independently.  11 . After bisulfite exposure to distinguish between methylated and unmethylated PITX2, the percent methylation ration (PMR) of three CpG motifs of the PITX2 gene promoter 2 is quantified by qMSP and calculated by the Rotor-Gene AssayManager® software with Gamma Plug-in plus a kit-specific PITX2 Assay Profile for automated analysis and quality control including all validity criteria. Validity criteria for run controls and sample testing include cycle threshold (Ct), amplification curve anomalies, PMR range for the controls (PITX2 RGQ PCR reference 50, PITX2 RGQ PCR reference low, PITX2 RGQ PCR negative control (NC), PITX2 RGQ PCR NTC (NTC)) and delta-PMR threshold for sample duplicates as described in detail in the handbook 18 . The PMR (percent methylation ratio) is calculated applying the following formula: 15 . The PITX2 qMSP assay was optimized for the usage as an in vitro diagnostic assay following the CLSI guidelines (https://clsi.org/; see also 18 ). PCR analysis is performed on Rotor-Gene Q MDx 5plex HRM instrument (Qiagen, Hilden, Germany).

Patients and tumor samples. For the analytical and technical evaluation of the therascreen PITX2
RGQ PCR assay, Formalin-Fixed Paraffin-Embedded (FFPE) tumor tissue samples from invasive breast cancers (n = 131; thereof 61 ER+ and 68 ER−) were collected at the Institute of Pathology, Technische Universität München. From these cases incomplete clinical and follow-up data were available. Therefore, the FFPE samples were only used to optimize the workflow, perform the analytical and technical evaluation including determination of DNA extraction yields, pass rates (run pass rate and sample pass rate), and tumor heterogeneity.
For the establishment of the transformation curve with the therascreen PITX2 RGQ PCR assay, Fresh frozen (FF) tumor tissue samples (n = 121) processed to tumor cell nuclei pellets were provided by the Erasmus Medical Center (EMC), Rotterdam, The Netherlands.
From a study published by Hartmann et al. 2009, 384 patients were evaluated according to the following eligibility criteria: invasive breast cancer, tumor stage pT1 to pT3, histologically confirmed lymph node involvement (>pN1), availability of clinical follow-up data of at least 5 years, availability of PITX2 methylation-specific microarray data, and, finally, standard adjuvant anthracycline-based chemotherapy (no dose-dense therapy, no other primary systemic chemotherapy, except hormonal therapy, and no additional taxanes) (Fig. 1). From 121 of the 384 patients ( Fig. 1), both fresh-frozen tumor tissue specimens and microarray data of PITX2 DNA methylation were available, but not all samples met the inclusion criteria of the clinical study population Also, the available clinical data of this group of 121 patients were partly incomplete and, therefore, could not be used for outcome analysis. The 121 samples, which were collected at the Erasmus Medical Center in Rotterdam were used to analyze the correlation between mathematically converted microarray gLOG (FIm/FIu) data to PMR values [PMR calc = 2exp(gLOG (FIm/FIu))] and their respective PMR values experimentally (PMR exp ) determined with the therascreen PITX2 RGQ PCR assay according to the formula: PMR exp = 100/[1 + 2exp(Ct FAM(methylated) − Ct HEX(unmethylated) )] from fresh frozen tumor tissue samples.
204 patients (ER positive, HER2 negative) of the 384 patients were selected for the clinical validation study group (Fig. 1). This clinical validation study group was considered for determination of PITX2 cut-off points and to perform Kaplan-Maier analyses by using available microarray gLOG (Flm/Flu) data from fresh-frozen tissue samples transformed into PMR values by a) mathematically calculation (PMR calc ) and b) applying the established transformation curve to calculate therascreen PITX2 RGQ PCR assay equivalent PMR data (PMR equiv ) and using the clinical data of this group ( Fig. 1 -bold box). The histopathological and clinical criteria of this group are summarized in Table 1.
Experimental studies involving human tissue material. Any experiments on human tissue material cited which were conducted in cooperation with the Department of Obstetrics and Gynecology (Frauenklinik) and the Institute of Pathology of the Technische Universität München, Munich, Germany, and with Erasmus Medical Center, Rotterdam, The Netherlands, were done in accordance with the Declaration of Helsinki (1964) and in accordance to the Code of Conduct of the Federation of Medical Scientific Societies in the Netherlands (www.federa.org/codes-conduct). The laboratory experiments were performed with the human subjects' understanding, who provided written informed consent for using the respective tissue specimens. The present study was approved by the local Ethical Committee of the Technische Universität München -Faculty of Medicine, Munich, Germany.
Statistical methods. Fractional polynomial approach was used to analyze the correlation between PMR data calculated from gLOG values of methylation-specific microarray data (PMR calc ) and PMR data measured with the therascreen PITX2 RGQ assay (PMR exp ) values to determine the best-fitting function 19 . gLog(2) fluorescence intensity values (gLOG Flm/Flu)) of PITX2 on a methylation specific oligonucleotide microarray 20 were transformed into PMR values (PMR calc ) by the formula: PMR calc = 2exp(gLOG) to allow for comparability with PMR data (PMR exp ) from the therascreen PITX2 RGQ PCR assay according to the respective probe  Fig. 1). Disease-free survival (DFS) was the primary endpoint and defined as the time from primary surgery to the first documented event, which includes any of the following: local recurrence of disease (n = 19) or distant metastasis (n = 70) or contralateral breast cancer (n = 5). Analysis was performed for DFS follow-up time of 120 months. The PITX2 cut-off value for disease-free survival (DFS) was established with the "maximum-selected log-rank statistics" using the maxstat.test function as implemented by the program library "maxstat" of the program "R" (R Development Core Team 2012) 21 . The PITX2 cut-off value is the Percentage Methylation Ratio (PMR) of PITX2 which separates good survivors and poor survivors (responder vs. non-responder) by using the log-rank statistics algorithm (R Software version: R version 3.4.1, 2017; the R Foundation for Statistical Computing).

DNA yield from FFPE tumor sections.
To verify the successful application of the therascreen workflow from FFPE material to PCR-based results, we used 131 FFPE breast cancer samples (63 estrogen receptor positive and 68 estrogen receptor negative), which were collected at the Institute of Pathology, Technische Universität München, as an independent cohort. From those samples, the genomic DNA (gDNA) was extracted from one to two 5 µm FFPE-tumor tissue sections with a total surface area ≥100 mm 2 . DNA quantification was performed with the QIAxpert spectrophotometer (Qiagen, Hilden, Germany) using the QIAamp DNA plugin with internal blank calibration for elution buffer ATE with the OD 260 readout of total nucleic acids. For 110 tumor samples more than 400 ng of gDNA were obtained per sample, indicating an overall gDNA extraction pass rate of 84% for subsequent bisulfite conversion at the recommended gDNA input according to the assay format of 400 ng input ( Table 2). The gDNA yield for the minimal gDNA input of 200 ng was achieved in 125 samples (95.4%). Samples with a tissue area below 100 mm 2 (n = 5) had yields between 85 and 440 ng. In those cases, several pooled 5 µm sections for gDNA extraction were required.

Pass rates for qPCR runs and samples (DNA input).
Six runs were performed to analyze 104 FFPE tissue tumor samples. The assays were performed with the recommended DNA input of 400 ng for bisulfite conversion (n = 104) resulting in a final input of 106 ng bisulfite-converted DNA per well into qPCR. The six assay   Tumor heterogeneity. To investigate the heterogeneity of PITX2 DNA-methylation within the same tumor sample, 5 consecutive tissue sections of the same tumor were analyzed. As denoted in Table 4, six analyzed tumor specimens gave coefficients of variations (CV) from 7.59% to 29.70%. However, the intra-and inter-assay variability range of the therascreen assay format is between 3.90-28.72% 18 . The variability range observed in consecutive tumor sections with 7.59% to 29.70% is in the same range. Therefore, the contribution of tumor heterogeneity to PITX2 PMR value variability seems to be rather low.

Comparison of cut-off and clinical data derived from microarray analysis and therascreen PITX2
RGQ PCR assay format. Hartmann et al. 2009 analyzed the performance of the PITX2 marker in 241 ER-positive, HER2-negative, and lymph-node positive tumor tissue samples showing that PITX2 promoter 2 (PITX2P2) hypermethylation was associated with a high risk of distant recurrence (time to distant metastasis) in the patient cohort (HR = 1.66; p = 0.002) and poor DFS (HR = 1.47; p = 0.0084) with the amplificate designed for the promoter of transcripts A and B of PITX2 (PITX2P2). The hazard ratio (HR) in this study was calculated relative to an increment from the lower quartile to the upper quartile 13 .
In order to examine if the clinical relevance (hazard ratio) are similar with data generated by microarray analysis or compared to data generated with the therascreen PITX2 assay for DFS the following strategy was pursued: 204 of 241 patient data were eligible (Fig. 1) and used for our analyses. The microarray data (gLOG data; Hartmann et al. 2009) from fresh frozen (FF) tissue of high-risk breast cancer patients ( Fig. 1; n = 204) were converted by mathematical calculation with 2exp(gLOG) function into PITX2 PMR calc data and used to determine the best PITX2 cut-off value (PMR calc ) for DFS as shown in Fig. 2a. A pronounced single peak is shown reaching the highest statistical difference (z = 4.15) of two groups of patients with long and short DFS at a cut-off value of PMR calc 14.73. Kaplan-Meier analysis applying the cut-off value of PMR calc 14.73 identifies patients who benefit from anthracycline-based chemotherapy (n = 204; ±endocrine treatment) with a significant longer DFS (HR 2.32, p < 0.0001) and those with less benefit from anthracycline-based chemotherapy (Fig. 2b).
PITX2 DNA-methylation also identifies patients in the group of breast cancer patients who received in addition to anthracycline-based chemotherapy endocrine therapy (n = 96) who have a significant longer DFS and those patients who have a significantly shorter DFS (HR 2.285, p = 0.024) (Fig. 2c). In order to confirm these results with a PCR-based analysis method we determined the PITX2 DNA-methylation applying the analytically validated therascreen PITX2 assay in 121 tumor cell nuclei pellets from fresh frozen (FF) tissue samples available from the same study (Fig. 1). However, as the clinical data for these 121 samples were partly incomplete, we established a transformation curve with the PMR values derived from microarray (PMR calc ) and qPCR (PMR exp ; assessed with the therascreen PITX2 RGQ PCR Kit) from these 121 samples. This resulted in a moderate correlation coefficient factor of r = 0.722 (r-squared = 0.5219; n = 121) as shown in Fig. 3. The relative high number of zero values is due to the limit of detection of therascreen PITX2 RGQ PCR test which results in zero values for all PMRs between 0 and 4. The best fitting regression analysis function was determined by using the fractional polynomial approach (see statistical methods) and used to calculate PMR calc values into therascreen PITX2 RGQ assay equivalent PMR values (PMR equiv) sample by sample from the Clinical Study Population (n = 204; Fig. 1).
The observed hazard ratios applying the cut-off derived from converted microarray data with PMR calc = 14.73 and from therascreen PITX2 RGQ PCR assay equivalent PMR data with PMR equiv = 12 are quite comparable and 9 of 204 patients (4.4%) of the clinical validation group switch from the low-risk group into the high-risk group with applying PMR equiv = 12. In current standard-of-care, this would lead to treatment of these patients with anthracycline-based chemotherapy.

Discussion
The clinical relevance of the PITX2 DNA-methylation status in breast cancer has been described in several studies 8,[10][11][12][13][14]24 . The precise role of PITX2 DNA-methylation in breast carcinogenesis remains not fully understood. One publication described a prognostic role of PITX2 DNA-methylation for the clinical course of breast cancer 10 . In the studies performed by Maier et al. 11 and Harbeck et al. 11,12 , PITX2 DNA-methylation showed the strongest correlation with metastasis-free survival in node negative tamoxifen-treated breast cancer patients. The study  11 . These studies underline PITX2 DNA-methylation as a potential biomarker for predicting outcome in patients with node negative breast cancer. The role of DNA-methylation in clinical cancer research gained significant attention. DNA-methylation analysis comprises single gene analysis on simple endpoint PCR based approaches, quantitative methylation specific PCR and oligonucleotide microarrays [25][26][27] . Evidence has been accumulated that DNA-methylation based markers -depending of the method applied -can be predictive or prognostic for clinical outcome and might improve diagnosis and treatment 28-31 . Hartmann et al., 2009 showed for the first time the clinical relevance of PITX2 methylation status in ER positive, HER2 negative, lymph node positive breast cancer patients treated with anthracycline-based chemotherapy 13 . PITX2 hypermethylation was associated with high risk of recurrence in that patient cohort [HR 1.66, p = 0.002] and was also associated with poor DFS (HR 1.47, p = 0.0084) 13 . However, the transformation of a research-based assay into a fully analytically validated test system rarely occurred and, therefore, only a few examples transitioned into clinical routine setting.
Array-based methodologies often are better suitable for marker discovery studies, whereas, single marker based quantitative DNA-methylation based assays are more reliable, cost-effective, sensitive and robust for high-throughput analysis in diagnostic or clinical chemistry units.
Therefore, a real-time PCR-based PITX2 marker assay as RUO test was developed for verification studies of the PITX2 marker 11,12 . Harbeck et al., 2008 showed that this RUO qPCR-based test can be reliably used in high-throughput studies (n = 399). This PCR-based assay demonstrated high reproducibility in replicate measurements (r ≥ 0.95, n = 150). Furthermore, Harbeck et al. (2008) showed, that PITX2 PMR values correlated between matched fresh frozen and FFPE tissue samples (n = 89, r = 0.81). Therefore, an expansion of the PITX2 test for assessment of more easily available FFPE material was indicated. The authors also concluded that tumor heterogeneity does not impact the analysis of PMR (percent methylation ratio) values in FFPE tissue blocks 12 .
The present technical evaluation study was based on the development of the aforementioned Research Use Only (RUO) assay into a CE certified in vitro diagnostics (IVD) kit including optimization of the workflow and analytical and technical verification and validation of the assay for its use in routine diagnostics. The assay was also optimized for use of the more easily available FFPE material. For the therascreen PITX2 RGQ assay single-site and multi-site precision studies were performed according to CLSI guideline 22 for technical certification of the assay, proving high robustness (average coefficient of variation (CV) at about 16%). Because the average   coefficient of variation was determined to 16%, there is a grey area in absolute PMR values around the cut-off of PMR 12 between PMR 10 to 14. For patients with tumor PMR results of PITX2 DNA-methylation in this grey area, the physician would very likely take additional clinical parameters for a therapy recommendation into account. The overall sample pass rate of DNA extraction for qPCR analysis was 95%, e.g. samples yielded sufficient DNA amounts for input into bisulfite conversion and overall 95% of all samples gave valid results according to the certified assay profile validity criteria 18 . These data emphasize that the therascreen PITX2 RGQ assay can reliably be used for analysis of FFPE-derived material. The amplifyable copy numbers assessed based on CT values of each sample did not correlate with the overall DNA yield, showing that DNA integrity in the original FFPE sample is critical on test outcome. Nevertheless, one 5 µm section with more than 100 mm 2 tissue area is sufficient to achieve an overall success rate of more than 95% with the therascreen PITX2 qPCR workflow. The present study confirms applying the analytically validated therascreen PITX2 RGQ PCR assay that PMR values derived from fresh-frozen tissue samples correlate (r = 0.722; r-squared = 0.5219; n = 121) with the microarray results from the Hartmann collective 13 . We could also demonstrate a statistically highly significant correlation between PITX2 DNA methylation (therascreen PITX2 RGQ PCR assay equivalent PMR data with PMR equiv = 12) and clinical outcome (DFS) for patients treated with anthracycline-based chemotherapy.
Predictive tests to avoid undertreatment or overtreatment of high-risk breast cancer patients are urgently needed. Therefore, a test which can predict outcome in patients receiving anthracycline-based chemotherapy is of high unmet medical need because it would improve the compliance of patients, who will most likely benefit, to stay on therapy despite potential toxic side effects. Vice versa patients who will most likely not benefit might be considered for an alternative treatment option.
The therascreen PITX2 RGQ PCR kit is now commercially available, and it is intended to be used by qualified users trained in molecular biology techniques and in in vitro diagnostic procedures. The therascreen PITX2 RGQ PCR kit with the determined cut-off value of PMR 12 will allow identification of high-risk, lymph node-positive, estrogen-receptor positive, HER2-negative breast cancer patients treated with anthracycline-based chemotherapy with good versus poor outcome. Future studies using independent cohort of ER+ breast cancer FFPE tissue samples and also prospective clinical trials are required to further substantiate the clinical cut-off of PMR 12. In conclusion, PITX2 DNA-methylation determined with the analytically validated therascreen PITX2 RGQ PCR assay may improve personalized breast cancer management and treatment decision in high-risk (lymph node-positive, estrogen-receptor positive, HER2-negative) breast cancer.