Epigenetic inactivation of DNA repair genes as promising prognostic and predictive biomarkers in urothelial bladder carcinoma patients

We sought to examine epigenetic inactivation of DNA damage repair (DDR) genes as prognostic and predictive biomarkers for urothelial bladder cancer (UBC) as there are currently no reliable prognostic biomarkers that identify UBC patients who would benefit from chemotherapy. Genome-wide DNA methylome using the cancer genome atlas-bladder cancer (TCGA-BLCA) datasets (primary tumors = 374 and normal tissues = 37) was performed for 154 DDR genes. The most two significant differentially methylated genes, Retinoblastoma binding protein 8 (RBBP8) and MutS homologue 4 (MSH4), between primary tumors and normal tissues of TCGA–BLCA were validated by methylation-specific PCR (MSP) in UBC (n = 70) compared to normal tissues (n = 30). RBBP8 and MSH4 expression was measured using qRT-PCR. We developed a predictive model for therapeutic response based on the RBBP8- and MSH4-methylation along with patients’ clinical features. Then, we assessed the prognostic significance of RBBP8 and MSH4. RBBP8- and MSH4 methylation and corresponding gene downregulation significantly associated with muscle-invasive phenotype, prolonged progression-free survival (PFS) and increased susceptibility to cisplatin chemotherapy in UBC. Promoter methylation of RBBP8 and MSH4 was positively correlated with each other and with their corresponding gene repression. The best machine-learning classification model predicted UBC patients’ response to cisplatin-based chemotherapy with an accuracy of 90.05 ± 4.5%. Epigenetic inactivation of RBBP8 and MSH4 in UBC could sensitize patients to DNA-damaging agents. A predictive machine-learning modeling approach based on the clinical features along with RBBP8- and MSH4-methylation might be a promising tool for stratification of UBC responders from nonresponders to chemotherapy.


Introduction
Bladder cancer (BLCA) is the most prevalent malignancy of the urinary tract and has the highest recurrence rate ranging from 50 to 90% (Siegel et al. 2019). Urothelial bladder carcinoma (UBC) accounts for 94% of bladder cancer cases and can be categorized as either muscle-invasive urothelial bladder carcinoma (pT2, pT3, or pT4; MIBC) or nonmuscleinvasive urothelial bladder carcinoma (pTa or pT1; NMIBC). The majority of NMIBC are associated with high risk of recurrence and progression to MIBC (Halperin et al. 2019). Cancer-specific survival in patients with MIBC is unfavorable despite treatments with radical cystectomy with or without perioperative cisplatin chemotherapy (Alfred Witjes et al. 2017). Thus, there is a need for novel prognostic and predictive biomarkers that will aid in identifying high-risk UBC patients who may benefit from chemotherapy. UBC is a heterogenous disease that is associated with genetic and epigenetic instability that drive the progression and aggressiveness of cancer (Martinez et al. 2019).
Epigenetic changes chiefly differential DNA methylation (DNAm) pattern represents the major form of epigenetic modifications that control gene expression early in Communicated by Shuhua Xu. carcinogenesis and holds promise as prognostic biomarker for cancer due of its well-recognized association with various aspects of human cancer (Alvarez et al. 2011;Patil and Herceg 2019). It has been reported that aberrant promoter methylation of DNA damage repair (DDR) genes play a crucial role in cancer risk diagnosis, prognosis and stratification of patients with distinct risk of treatment response (Magzoub et al. 2019).
The identification of aberrantly methylated and differentially expressed genes might provide potential epigenetic biomarkers for UBC. In this study, we evaluated differential DNAm levels of 154 DDR genes using the cancer genome atlas (TCGA)-BLCA DNA-methylome data. Consequently, we aimed to investigate the prognostic value of the most significant differentially methylated genes RBBP8 and MSH4 in an institutional cohort of UBC patients and to develop classification model for prediction of pathological response to therapy in UBC patients based on the RBBP8 and MSH4 methylation.
Retinoblastoma binding protein 8 (RBBP8), also known as C-terminal binding protein (CtBP)-interacting protein (CtIP) encodes extensively expressed nuclear endonuclease. Accumulating studies have reported that RBBP8 is required for DNA double-stranded break (DSB) repair by homologous recombination (HR) in G2/M phases through interaction with BRCA1 and MRE11-RAD50-NBN (MRN) complex (Huertas and Jackson 2009;Sartori et al. 2007). RBBP8 interacts with tumor suppressor genes such as BRCA1 and the pRb family members through binding domains that are frequently mutated in human cancers (Chinnadurai 2006). Some studies suggested that disruption of BRCA1-RBBP8 interaction results in cell cycle arrest modulation (Li et al. 2000;Wu-Baer and Baer 2001). Mismatch repair (MMR) genes play a crucial role in DNA repair mechanism. Loss of function of MMR genes by mutation, loss of heterozygosity or promoter hypermethylation affect its role in repairing intranucleotide error (Spetsotaki et al. 2017). MutS homolog (MSH)4 plays a crucial role in maintaining genomic stability through nonhomologous end joining (NHEJ) pathway to DSB (Chu et al. 2013). It has been reported that a single nucleotide polymorphism (SNP)-SNP interaction between MSH4 Ala97Thr/MLH3 Leu844Pro increases breast cancer susceptibility (Conde et al. 2009). Promoter hypermethylation and downregulation of MSH4 have also been shown in head and neck squamous cell carcinoma. However, the role of aberrant MSH4 expression have not been previously reported in bladder tumors (Chaisaingmongkol et al. 2012). In this study, we demonstrate for the first time the prognostic and predictive role of RBBP8 and MSH4 hypermethylation for UBC disease.

Study population
The study protocol was approved by institutional Review Board (IRB) of National Cancer Institute (NCI), Cairo, Egypt-as guided by the 2013 Helsinki Declaration (IRB NO.IRB00001568). All subjects provided signed informed consent for collection and analysis of their specimens. Patients with history of other malignancy and carcinoma in situ were excluded from the study. A total of 70 formalin fixed paraffin embedded (FFPE) bladder tissues of patients undergoing radical cystectomy were recruited from NCI, Egypt during the period from January 2016 to October 2018. 30 adjacently normal urothelium were included as normal controls (NC). The lack of significant inflammation or atypia confirm the diagnosis of normal tissues. Follow-up data were acquired prospectively from clinic visits and electronic patient records. All patients received adjuvant and/ or neoadjuvant cisplatin-based chemotherapy. Response to treatment was assessed based on the response evaluation criteria in solid tumors (RECIST) (Schwartz et al. 2016). For data analysis complete and partial response were grouped into responders while, stable and progressive disease were grouped into nonresponders.

In silico analyses
The DNAm, gene expression (RNA-seq) and the corresponding clinical data of TCGA−BLCA (primary tumors = 397 and normal tissues = 37) (https:// portal. gdc. cancer. gov/) were downloaded and assessed by TCGAassembler 2 (Wei et al. 2018). Briefly, after data download, we performed advanced processing to retrieve average DNA methylation values (B value) of CpG sites in a specific gene location (e.g. the promoter region) and mRNA expression in transcript per million (TPM). We determined the average methylation level for each of 154 DDR genes within the gene promoter region generated by KEGG database searches for DNA damage repair, MMR, NHEJ, HR, DDR checkpoint, base excision repair (BER), and nucleotide excision repair (NER) (Supplementary Table 1). The criteria for screening of significant differentially methylated genes were Beta value > 0.2 and corrected p value < 0.05 (independent t test plus Benjamini-Hochberg method). The differential expression of RBBP8 and MSH4 genes in bladder cancer samples compared to normal control were confirmed by GEO13507 dataset using GEOquery and limma package of R studio. Ensembl biomart provided all the necessary genomic DNA information required to identify RBBP8 and MSH4 gene core promoter region (ENSG00000101773 and ENSG00000057468, respectively) (https://m. ensem bl. org/ bioma rt/) that can be used to design the required methylation-specific PCR (MSP) primers. Consequently, refTSS (http:// reftss. clst. riken. jp/r) provided CpG island locations (22,933,933,894)

Total DNA and RNA extraction
Sections of FFPE tissue blocks were deparaffinized by xylene and rehydrated prior to nucleic acid extraction. Genomic DNA and RNA were extracted from FFPE tissues using Genedirex DNA extraction kit for tissue (GENEDI-REX, INC, Taiwan, China) and Genedirex total RNA extraction kit (GENEDIREX, INC, Taiwan, China), respectively, according to manufacturer's instructions.
Nucleic acid extraction for UBC tissues was performed in histologically confirmed areas containing a minimum of 70% tumor cells. Nucleic acid quality and concentration were assessed using NanoDrop 2000 (Thermoscientific, USA) with A260/A230 for DNA and A260/A280 ratio for RNA between 1.8 and 2.2.

Bisulfite conversion and methylation-specific PCR (MSP)
DNA methylation of RBBP8 and MSH4 was determined by bisulfite conversion of unmethylated cytosines to uracil using methylation-specific PCR (MSP) (Huang et al. 2013). In brief, 100 to 400 ng of the extracted DNA was subjected to bisulfite conversion using the EpiJET Bisulfite Conversion Kit (ThermoFisher Scientific, USA) according to the manufacturer's protocol followed by amplification of 150-300 ng of the bisulfite-treated DNA a set of MSP primers (Supplementary Table 2). All PCR reactions were performed by the Veriti Thermo Cycler (Life Technologies, Carlsbad, CA, USA). The MSP products were separated on 2% agarose gels, stained with ethidium bromide, and visualized using UV transilluminator.

Quantitative real time PCR (qRT-PCR)
An one-step qRT-PCR was accomplished using SYBR® Green RT-qPCR Master Mix (Willowfort.co/UK) including 5 μM of oligonucleotide primers (Supplementary Table 3) and 150 ng of extracted RNA. All reactions were done in triplicates using 7500 Fast-Real time PCR system software (Applied Biosystem, USA) and postamplification curves was assessed for product specificity. The fold change (FC) expression was calculated relative to glyceraldehyde 3-phosphate dehydrogenase (GAPDH) housekeeping gene by 2 −ΔΔCt method ).

Prediction model
We used Python sklearn library to develop predictive model for response to therapy based on the most relevant nonredundant patients' characteristics along with RBBP8 and MSH4 methylation data as shown in Supplementary Fig. 1. The most relevant nonredundant clinical characteristics were selected using the rank ordering method of the SelectKBest class of python scikit-learn library. We tested different classification models (logistic regression model (LR), kernel support vector machine (SVM), K-nearest neighbor (KNN), Decision tree (DT) and Random Forrest decision tree (RF)) for the best prediction of treatment response. sklearn voting ensemble was used to find the best model's combination. Initially dataset was split into 80% training-set and 20% testset. Grid-search method was used to optimize each classifier respective hyperparameters and a10-fold cross validation on the training-set was used to evaluate the model on an independent validation sets to avoid model overfitting. Performance of each classifier was measured by its accuracy and area under the Receiver operating characteristic (ROC) curve (AUC).

Statistical analysis
All statistical analysis was performed using R studio Statistical Software (version 3.7, Vienna, Austria). The pwr package was used to adjust the power of the test. Comparison of differential expression between study groups and with patients' clinicopathological features was done using Wilcoxon rank test. Chi-square was used to investigate methylation in association with clinicopathological features and logistic regression was used to estimate association of different parameters with response. The multiple comparisons were adjusted for false discovery rate (FDR) (Benjamini and Hochberg 1995). Pearson correlation was used to measure the correlation coefficient. Progression-free survival (PFS) was calculated from the date of primary therapy to either recurrent or progressive disease, patients free of progressive disease were censored at the time of the last follow up. Kaplan-Meier survival analysis-log-rank test was used to compare survival time. Cox proportional hazard regression analysis were applied to evaluate the hazard of RBBP8 and MSH4 along with clinicopathological data on survival probability. Hierarchical clustering heatmap was used to show the methylation values of DDR genes based on Infinium 1 3 HumanMethylation450 BeadChip. All tests were two sides and significance was set at p < 0.05.

Clinicopathological features
The clinicopathological variable of UBC and NC are shown in Table 1. The mean age of UBC patients and NC was 62.2 ± 8.5 years and 61.9 ± 8.9, respectively; p = 1.0).

Differential expression of RBBP8 and MSH4 mRNA in TGCA and GEO datasets
As shown in Fig. 4, the TCGA−BLCA dataset showed that the expression of RBBP8 and MSH4 was significantly lower in bladder cancer cases as compared to normal samples (p < 0.001). Furthermore, GEO13507 dataset showed that RBBP8 and MSH4 were differentially expressed in bladder cancer cases compared normal tissues by 1.82 and 1.88 fold, respectively (p < 0.001). Validation of RBBP8 and MSH4 differential mRNA expression between UBC and NC As shown in Fig. 5, the median RBBP8 mRNA FC was significantly lower in UBC patients by 76.4% (p < 0.001) as compared to NC. For MSH4, median FC was also significantly lower in UBC by 67.4% (p < 0.001) as compared to NC (Fig. 6).

Association of RBBP8 and MSH4 mRNA expression with patients' characteristics
As shown in Fig. 5

Odd ratios of chemotherapy response
Using forward features selection, we selected the most relevant nonredundant clinical features.

Predictive machine-learning models for patient's stratification according to response
We used the rank ordering method of the SelectKBest class of python scikit-learn library to select the most relevant nonredundant features including tumor size, tumor grade, stage, lymph node metastasis.

Survival analysis
The mean follow-up duration was 48.5 months (range, 14.49-56.67 months). First, the mRNA data of patients were classified into low vs high according to median FC. Kaplan-Meier survival analyses of UBC patients have shown reduced PFS in association with RBBP8 and MSH4 methylation (p = 0.0027 and p = 0.02, respectively, log rank) and reduced expression of corresponding genes (p < 0.001, for all, log rank). In MIBC patients, PFS was significantly related to RBBP8 methylation (p = 0.018, log rank) as well as reduced expression of RBBP8 (p = 0.018, log rank) and MSH4 (p = 0.003, log rank) (Fig. 8). In univariate survival analysis, PFS of UBC patients was significantly associated with tumor grade (p = 0.006), stage (p = 0.007) along with RBBP8-M (p = 0.029), MSH4-M (p = 0.023) and their corresponding gene expression (p < 0.001 and p < 0.001). In MIBC, prolonged PFS was significantly associated with RBBP8 methylation (p = 0.0257) and reduced expression of RBBP8 and MSH4 (p = 0.007). In multivariate survival-analysis, MSH4 expression was the only independent prognostic factor for PFS in MIBC patients (p = 0.012) on cisplatinbased chemotherapy (Table 6).

Discussion
In postgenomic era, it has been proposed global epigenetic aberrations maintained in carcinogenesis may play an important role in tumor heterogeneity (Sandoval and Esteller 2012). Promoter hypermethylation and mutational In the present study, we aimed to identify differential methylation of DDR genes in UBC as compared to NC and in MIBC as compared to NMIBC. Hierarchical clustering of genome-wide methylome of TCGA−BLCA samples identified 12 out of the 154 DDR genes whose promoter region close to TSS to be hypermethylated. Then, we found that RBBP8 and MSH4 were the most significant aberrantly methylated genes in TCGA−BLCA primary tumors compared to normal tissues. In silico analysis was validated by our MSP and qRT-PCR that detected a significant increase in RBBP8 and MSH4 hypermethylation and downregulation in UBC compared to NC. Interestingly, we found for the first time that RBBP8 and MSH4 methylation and their corresponding gene downregulation were significantly associated with progressive UBC which is in parallel with high tumor stage and muscle-invasive disease. This could not be assessed in the TCGA−BLCA datasets because most of TCGA cases were of the muscle-invasive subtype. In the present study, we demonstrated a significant positive correlation between RBBP8 and MSH4 methylation and reduced expression of their corresponding genes which suggests that RBBP8 and MSH4 hypermethylation account for their epigenetic inactivation in UBC. Moreover, correlation analysis have demonstrated a moderate positive between RBBP8-and MSH4-expression in UBC. These results indicate that MMR and HR methylation and gene inactivation are strongly correlated in UBC patients. This was not previously reported in UBC however, a previous study has shown a strong Fig. 4 Differential RBBP8 and MSH4 expression between UC and normal bladder tissues. a Boxplot showing differential RBBP8 expression between urothelial carcinoma and normal controls in TCGA−BLCA dataset. b Boxplot showing differential MSH4 expression between urothelial carcinoma and normal controls in TCGA− BLCA dataset. c Volcano plot for differential expression of the significant differentially methylated genes between UC and normal control. The x axis shows the log-fold change, and the y axis is some measure of the B score statistical significance correlation between mutational status of HR and MMR genes that subsequently leads to genomic instability in gastric carcinoma patients (Liu et al. 2019). Hypermethylation of RBBP8 has been previously detected as biomarker for bladder cancer patients (Mijnes et al. 2018). Microsatellite instability in urine has also been implicated as a noninvasive characteristics. NMIBC nonmuscle invasive urothelial bladder cancer, MIBC muscle-invasive urothelial bladder cancer. *Significant at p < 0.05. Association between RBBP8 and clinicopathological features was tested using Wilcoxon sum ran test tool for diagnosis of BLCA . Moreover, combined mutations in MSH4 and MLH3 were associated with increased risk of breast cancer (Conde et al. 2009). Collectively, this highlights the possible role of RBBP8 and MSH4 in BLCA susceptibility.
Despite current advances in surgical procedures and neoadjuvant chemotherapy, a large proportion of patients with characteristics. NMIBC nonmuscle invasive urothelial bladder cancer, MIBC muscle-invasive urothelial bladder cancer. *Significant at p < 0.05. Association between MSH4 and clinicopathological features was tested using Wilcoxon sum ran test 1 3 NMIBC disease are at a high risk of progression to MIBC (Van Rhijn et al. 2009). In MIBC, patients are associated with increase in metastatic spread that eventually leads to less favorable outcome (Hautmann et al. 2006;Shariat et al. 2006). In recent decades, there has been a little progress in systematic chemotherapy for UBC (Sonpavde et al. 2016) except for a few including immunotherapeutic approaches (Inman et al. 2017). The first-line therapeutic approach of MIBC includes neoadjuvant-platinum-based combination chemotherapy or radiotherapy with or without concomitant systematic chemotherapy (Konety and Joslyn 2003;Poletajew et al. 2016). The delay of radical cystectomy in MIBC patients who do not respond to cisplatin is one of the drawbacks of neoadjuvant chemotherapy (Gore et al. 2009). Moreover, the lack of proper biomarkers that could predict muscle invasion and identify patients who will respond to cisplatin decreases the ability to select an appropriate treatment for BLCA (Bertz et al. 2014;Otto et al. 2011). Recently, epigenetic modifications have been implicated in modulation of treatment response in cancer (Lu et al. 2020).
In the current study, we aimed to assess epigenetic modification of RBBP8 and MSH4 as possible prognostic and predictive biomarkers for UBC patients (Lu et al. 2020). Our principal finding was the significant association of RBBP8 and MSH4-methylation corresponding to gene inactivation with increase in response by 57.1% and 65.0%, respectively. In line of our results, loss of function mutations in DDR genes including ATM, FANCC and ERCC2 have been associated with increased sensitivity to cisplatin-based chemotherapy and immunotherapies in MIBC (Abbosh and Plimack 2018). However, MMR status and chemosensitivity status has not been previously addressed in UBC.
BRCA1 promoter methylation has been associated with favorable response to platinum-based treatment in breast and ovarian cancer (Stefansson et al. 2012) which is the standard curative approach in UBC management. DNA-damaging agent-like platinum-based chemotherapy elicit its effect through induction of inter-strand crosslink that leads to DNA DSB that are regularly repaired by HR and MMR (Rycenga and Long 2018). CtIP/RBBP8 protein is known to modulate the functions of BRCA1 in transcriptional regulation and DNA repair. Thus, loss of function mutation of DDR will abolish cell's ability to repair DSB and consequently resulting in o treatment susceptibility (Hoa et al. 2015;Makharashvili et al. 2014).
With current interest in precision medicine, we aimed to develop a predictive model for response to cisplatin-based therapy in UBC based on the RBBP8 and MSH4 methylation along with patients' characteristics which might satisfy patients' needs to more personalized medicine. We found that the best predictive model for stratification of patients according to response to therapy was a combination of KNN, RF and DT presenting an accuracy of 90.0%.
Survival analyses have also shown that that RBBP8 and MSH4 methylation and their corresponding gene downregulation significantly correlated with longer PFS in UBC patients. Moreover, RBBP8 methylation along with RBBP8 and MSH4 mRNA expression were associated with longer PFS in MIBC patients. Based on the multivariate survival analysis, we found that MSH4 downregulation was the sole independent predictor of PFS in MIBC patients.
Thus, it can be inferred that unrepaired double-stranded crosslinks because of loss of RBBP8 and MSH4 function may also increase cellular sensitivity to cisplatin-based chemotherapy and improve patients' outcome especially in MIBC. In a recent study, deficiency of RBBP8 expression has been implicated to increase sensitivity to cisplatin-based therapy in BLCA (Mijnes et al. 2018). Moreover, RBBP8 deficiency has been associated with increase the susceptibility of breast and ovarian cancer to poly ADP ribose polymerase (PARP) inhibitors in a manner similar to BRCA1 mutations (Lin et al. 2014;Wang et al. 2016) which suggests To our knowledge, this is the first study to address a relationship between epigenetic inactivation of MSH4 and response to cisplatin-based chemotherapy in UBC patients. However, inactivation of MMR genes like MSH3 and MSH5 by methylation and single nucleotide polymorphisms (SNPs) has been associated with increased sensitivity to cisplatinbased chemotherapy in small cell lung carcinoma (J.-Y. Liu et al. 2017).
In conclusion, our data showed that epigenetic inactivation of RBBP8 and MSH4 by hypermethylation was significantly frequent in MIBC subtype and significantly associated with favorable outcome in terms of PFS and increase sensitivity to cisplatin-based chemotherapy. Our machinelearning model revealed that RBBP8 and MSH4 methylation could provide a tool for prediction of UBC patients who might respond to platinum-based chemotherapy taking in consideration patients' clinical data. This study should be expanded to multiple centers for further verification of the potential role of RBBP8 and MSH4 in UBC a step towards personalized medicine.