A set of molecular markers predicts chemosensitivity to Mitomycin-C following cytoreductive surgery and hyperthermic intraperitoneal chemotherapy for colorectal peritoneal metastasis

Cytoreductive surgery (CRS) and hyperthermic intraperitoneal chemotherapy (HIPEC) is associated with significant perioperative morbidity and mortality. We aim to generate and validate a biomarker set predicting sensitivity to Mitomycin-C to refine selection of patients with colorectal peritoneal metastasis (CPM) for this treatment. A signature predicting Mitomycin-C sensitivity was generated using data from Genomics of Drug Sensitivity in Cancer and The Cancer Genome Atlas. Validation was performed on CPM patients who underwent CRS-HIPEC (n = 62) using immunohistochemistry (IHC). We determined predictive significance of our set using overall survival as a surrogate endpoint via a logistic regression model. Three potential biomarkers were identified and optimized for IHC. Patients exhibiting lower expression of PAXIP1 and SSBP2 had poorer survival than those with higher expression (p = 0.045 and 0.140, respectively). No difference was observed in patients with differing DTYMK expression (p = 0.715). Combining PAXIP1 and SSBP2 in a set, patients with two dysregulated protein markers had significantly poorer survival than one or no dysregulated marker (p = 0.016). This set independently predicted survival in a Cox regression model (HR 5.097; 95% CI 1.731–15.007; p = 0.003). We generated and validated an IHC prognostic set which could potentially identify patients who are likely to benefit from HIPEC using Mitomycin-C.

Comparison of gene expression with survival (either OS or RFS) across five The Cancer Genome Atlas (TCGA) cohorts (BLCA, COADREAD, CESC, HNSC, OV) allowed us to narrow down the list of genes to four which were significantly associated with RFS and OS (DTMK, PAXIP1, SSBP2, HMGB1) in patients treated with cross-linking agents as first-line chemotherapy (Supplementary Table S2). All four of these genes demonstrated correlation between gene expression and Mitomycin-C half maximal inhibitory concentration (IC 50 ), and could potentially be used to predict for chemosensitivity to Mitomycin-C (Fig. 1). A combined 4-gene molecular marker set yielded an improved prediction model highly selective for resistant versus sensitive cell lines (p < 0.001) (Fig. 2).
Validation of predictive molecular marker set. To validate our predictive molecular marker set derived from the analysis of the GDSC database, we proceeded to examine the association between level of protein expression (as determined by IHC staining of patient tumour tissue) and OS and disease-free survival (DFS) in our cohort of patients with CPM (n = 62). 3 of the targets (PAXIP1, SSBP2 and DTYMK) were successfully optimized for IHC, while 1 target (HMGB1) was not successful. The baseline clinical characteristics of our cohort are presented in Table 1. Representative IHC staining of the three targets can be found in Supplementary Fig. S1.
Survival analysis for each of the markers was performed. Patients exhibiting lower expression of PAXIP1 had significantly poorer OS as compared to those with higher protein expression (median OS, 34.6 months vs. 43.3 months for low and high expressions respectively, HR 2.595, 95% CI 1.022-6.589, p = 0.045) (Fig. 3). Although patients with lower expression of SSBP2 potentially have poorer OS, this did not reach statistical significance (median OS, 29.8 months vs. 42.3 months for low and high expressions respectively, HR 1.886, 95% CI 0.812-4.378, p = 0.140) (Fig. 4) Fig. S2).
We next sought to determine the effect of a combined set of PAXIP1 and SSBP2 on the prediction for Mitomycin-C chemosensitivity using OS as a surrogate marker. Median OS for patients exhibiting zero, one and two dysregulated protein markers are 43.3, 42.3 and 18.4 months, respectively (p = 0.033). The addition of a dysregulated marker results in an increased HR by 1.866 (95% CI 1.052-3.310). Patients were then grouped according to either one or no dysregulated marker versus two dysregulated markers. We found a significant association between the latter and poorer OS, where median survival time was 42.3 months versus 18.4 months (HR 2.844, 95% CI 1.212-6.674, p = 0.016) (Fig. 6). A similar association was observed for DFS (median DFS 22 months www.nature.com/scientificreports www.nature.com/scientificreports/ vs. 12 months for one or no dysregulated marker and two dysregulated markers respectively, HR 2.775, 95% CI 1.191-6.464, p = 0.018) ( Supplementary Fig. S2).
To confirm that our predictive molecular set was independent of clinicopathologic factors, we performed multivariate Cox regression analysis ( Table 2). After adjusting for age, gender, race, ECOG status, PCI score, and pathological T and N stages (8 th Edition), the presence of two dysregulated protein markers was found to be independently associated with a poorer prognosis within our cohort (HR 5.097, 95% CI 1.731-15.007, p = 0.003).  www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
With CRS-HIPEC gaining widespread acceptance as a potential standard of care for patients with CPM, the next frontier to accomplish is to improve long-term outcomes while minimizing perioperative morbidity, a goal in part achievable by tailoring treatment to the unique clinical characteristics and tumour biology of each patient. In the  www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ current paradigm of integrative genomic analysis, genomic data can serve as a powerful tool in the identification of patient subgroups for tailoring therapy and prognosis 20 . In addition, the PRODIGE 7 study suggests that the use of a standardized chemotherapeutic regime may not be beneficial for all patients. Instead, a personalized approach may be more suitable 21 .
The key tenets to an effective CRS-HIPEC procedure involve thorough and adequate cytoreduction to achieve maximal surgical debulking, followed by the instillation of intraperitoneal chemotherapy which achieves a peak peritoneal chemotherapeutic agent concentration up to over 1,000 times that achievable by systemic chemotherapy 22 . Mitomycin-C, a primary chemotherapeutic agent of choice for HIPEC in CPM, is subjected to the perennial issue of chemoresistance, best exemplified by the extensive experience in the use of Mitomycin-C in the treatment of bladder cancer 23,24 . Likewise, the ability to predict chemosensitivity to Mitomycin-C through integrative genomic analysis will serve as a potential patient selection and predictive tool in the use of CRS-HIPEC for patients with CPM. There is currently no practical means of routinely assessing chemosensitivity to Mitomycin-C amongst patients with CPM undergoing CRS-HIPEC in a clinical setting. As such, we adopted OS as the primary endpoint in our study, which serves as a surrogate measure of chemosensitivity to Mitomycin-C used in each patient's CRS-HIPEC procedure.
To this end, we designed our study to identify a set of molecular markers predictive of chemosensitivity to Mitomycin-C by examining the GDSC database, the largest public resource for information on drug sensitivity in cancer cells and molecular markers of drug response 25 . Through correlating Mitomycin-C IC 50 with the level of gene expression across a wide spectrum of molecular markers for each colorectal cancer cell line, we identified four potential molecular markers with a significant predictive value for chemosensitivity to Mitomycin-C.  www.nature.com/scientificreports www.nature.com/scientificreports/ To determine the reproducibility of our findings, we then sought to validate the utility of these prognostic markers against an internal cohort of patients with CPM, with a baseline demographic likely dissimilar to that making up the cell lines within GDSC based on a Western population. Of the initial four targets identified, three were successfully optimized on IHC. Subsequently, we found that two of the three markers were prognostic in our patient cohort, and the combined 2-marker set was significantly associated with OS following multivariate Cox regression analysis, outperforming conventional clinicopathologic prognostic factors such as PCI and CC score [10][11][12][13][14] .
Conceding that OS is influenced by many confounders such as other tumour biology factors and the administration and completion of adjuvant systemic chemotherapy, we assessed the effect of these potential predictive molecular markers on DFS. Unsurprisingly, the survival analysis showed similar results to that of OS, suggesting that the predictive markers identified were indeed predictors of chemosensitivity to Mitomycin-C in patients with CPM.
The two predictive markers identified and validated through this study play key biological roles in cellular proliferation. Neither of these genes has been described as potential predictors of Mitomycin-C response in CPM. SSBP2 and PAXIP1 are known to be involved in the DNA damage response and aid in maintaining genome stability [26][27][28][29][30] . Specifically, Shaw et al. 31 identified oligodendroglial tumours that responded to chemotherapy had lower expression of SSBP2. Correspondingly, PAXIP1 is involved in the sensitization of lung cancer cells to the WEE1 inhibitor, AZD 1775, in combination with cisplatin 29 . The mechanism of action of cisplatin lies in its association with DNA leading to the formation of DNA adducts, mainly intra-strand DNA adducts, in turn activating multiple signalling pathways and the apoptosis of cells 32 . This is akin to Mitomycin-C, which exerts its anti-tumour activity as a potent DNA cross-linker, resulting in the inhibition of DNA synthesis 33 . Taken together, the roles of SSBP2 and PAXIP1 in DNA damage repair support the hypothesis that dysregulation of these molecular markers could confer differential chemosensitivity profiles.   www.nature.com/scientificreports www.nature.com/scientificreports/ Additionally, literature has shown that SSBP2 contributes to pathogenesis of some cancers, such as colorectal cancer 34 , prostate cancer 27 , oesophageal squamous cell carcinoma 26 , acute myelogenous leukaemia 28 and glioblastoma 35 . Likewise, dysregulation of PAXIP1 expression has been reported to result in poorer patient outcomes. Similar to our findings, lower expression of PAXIP1 in breast cancer was associated with poorer prognosis 36 . Additionally, upregulation of PAXIP1 in hepatic cell carcinoma correlated with poorer clinicopathological characteristics, where it was an independent risk factor for poorer OS and RFS 37 .
The major strength of our study is that these predictive molecular markers can be easily assessed for in preoperative tumour biopsy samples through IHC during histological examination in the routine diagnostic and staging evaluation of patients with CPM. The presence of a pre-operative set allows clinicians to predict how a patient would respond to Mitomycin-C even before entering the operating theatre. This would yield additional predictive information to aid in the clinical decision making process to improve patient selection for a morbid procedure such as CRS-HIPEC, enabling better personalized treatment decisions. Moreover, IHC is a simple and cheap tool to achieve reproducible results even in the clinical setting, potentially resulting in a smooth transition from bench to bedside. With the evolution of automated IHC stainers and standardized staining protocols, the testing of this molecular marker set can be easily carried out worldwide. Furthermore, this study utilizes formalin-fixed paraffin-embedded (FFPE) samples to optimize the marker set. These samples are readily available from pathology repositories, providing us with an easy platform for validation. Moving forward, we aim to validate our molecular marker set in a larger patient cohort.
A limitation of our study is the small sample size. Additionally, IHC scoring has a subjective nature. Nonetheless, the presence of two or more blinded scorers interpreting the results limits the subjectiveness involved. There are also an increasing number of computational programs that serve to analyse IHC staining results, although these are associated with their own set of limitations.

Conclusion
In conclusion, our study identified and validated a molecular marker set that serves as a prognostic factor in CRS-HIPEC patients with CPM, which may be used as a surrogate marker of chemosensitivity to Mitomycin-C and OS. This set can be used in conjunction with other clinicopathologic prognostic factors to guide the optimal management of patients with CPM. However, due to the small sample size, a subsequent larger confirmatory study is required.

Methods
All experiments were performed in accordance with relevant guidelines and regulations, as approved by the SingHealth Centralized Institutional Review Board (CIRB 2015/2479/F). Informed consent was obtained from participants in this study.
Identification of potential predictive molecular markers. To identify clinically relevant predictive molecular markers which may predict chemosensitivity to Mitomycin-C, we interrogated the GDSC, an open access database of >1,000 genetically characterized human cancer cell lines screened with a wide range of anti-cancer therapeutics 25 . Each cell line is characterized by varying extents of dysregulated genes, with each specific anti-cancer therapeutic sensitivity data expressed in half maximal inhibitory concentration (IC 50 ). An initial list of genes whose expression was predictive of Mitomycin-C sensitivity was identified. Cell lines were grouped by Mitomycin-C sensitivity into resistant (IC 50 > 200 nM), intermediate (IC 50 > 50 nM) and sensitive (IC 50 < 50 nM), and those with intermediate sensitivity were excluded. Genes with expression that was predictive of chemosensitivity were identified for further analysis.
To allow for the selection of molecular predictive markers for chemosensitivity to Mitomycin-C, we next selected genes of prognostic significance on TCGA, a publicly available database of matched transcriptomic and clinical data, to allow cross relevance and to identify predictive markers that is associated with a known clinical endpoint, i.e. OS. As there were a limited number of patients treated with the cross-linking agent Mitomycin-C across the TCGA cohorts, we identified a panel of genes associated with survival in patients treated with platinum agents (as DNA cross-linkers). The number of genes was narrowed down by assessing for genes with survival association in patients treated with platinum agents across the following 5 TCGA cohorts, bladder urothelial carcinoma (BLCA), colorectal adenocarcinoma (COADREAD), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), head and neck squamous cell carcinoma (HNSC), and ovarian serous cystadenocarcinoma (OV). study population. The selected molecular predictive markers were validated on a cohort of 62 patients with CPM who underwent CRS-HIPEC at the National Cancer Centre Singapore between April 2001 and February 2017. All treatment decisions were made through a multidisciplinary tumour board comprising surgical, medical and radiation oncologists, radiologists and pathologists.
The extent of peritoneal tumour dissemination was scored at the time of surgery using the peritoneal cancer index (PCI) 38 . CRS was performed to remove all macroscopically visible tumour deposits, which entailed peritonectomy and en bloc organ resections where indicated 7 . The residual tumour burden following CRS was evaluated using the completeness of cytoreduction (CC) score: CC-0, no tumour; CC-1, tumour <2.5 mm; CC-2, tumour 2.5-25 mm; CC-3, tumour >25 mm 38 . HIPEC with Mitomycin-C at 10 mg/m 2 (maximum dose 70 mg) was administered using a closed technique for 60 minutes while maintaining intraperitoneal temperature at 41-43 °C in accordance with the consensus guidelines from The American Society of Peritoneal Surface Malignancies 39 . All HIPEC procedures performed at our institution are done during the same setting as the preceding CRS. Post-operative morbidity and mortality were graded according to the Clavien-Dindo classification 40 .
www.nature.com/scientificreports www.nature.com/scientificreports/ Immunohistochemistry. FFPE specimens were cut into 4μm sections and incubated for 1 hour at 60 °C before they were used for IHC staining. Three out of four predictive molecular targets were successfully optimized for IHC with staining performed on the FFPE sections. All IHC was performed using the Bond System (Leica Microsystems, Ltd, Milton Keynes, UK) according to the manufacturer's recommendations. Antibody sources and optimum IHC staining conditions are described in Supplementary Table S3.
The staining results were determined semi-quantitatively, based on the staining intensity and percentage of tumour cells stained, on a scale of 0 to 9 by at least two independent researchers (JWST, WW and YC) blinded to the outcome. In cases of discrepancies between the score assigned by the researchers, a third researcher (OCAJ) scored the sections independently to determine the final assigned degree of target expression. Staining intensity was stratified into 4 groups; negative (score 0), weak (score 1), moderate (score 2) and strong (score 3) (Supplementary Fig. S1). Similarly, percentage of tumour cells stained was determined as <10% (score 0), 10-50% (score 1), 50-80% (score 2) and >80% (score 3). Scores for staining intensity and percentage of stained tumour cells were multiplied to obtain the final immunoreactivity score. The immunoreactivity scores for each protein target were binarized to low (0-3) and high (4-9) expression.
Clinical endpoints and statistical analyses. Survival outcomes were examined via Kaplan-Meier analysis. The Cox proportional hazard model was used to examine our target molecular markers against other clinicopathologic variables as predictive factors for OS. All statistical analyses were performed using SPSS (Version 18.0, SPSS Inc., Chicago, USA). A p-value of <0.05 was considered statistically significant.

Data Availability
All data generated or analysed during this study are included in this published article (and its Supplementary  Information Files).