Abstract
Background/Aim: DNA mismatch repair (MMR) deficiency has received increasing attention as a biomarker of anti-PD-1 treatments of solid tumors including gastric cancer (GC). However, efficient screening has not been established. Patients and Methods: A total of 513 patients were tested for the expression of MMR proteins by immunohistochemistry to identify MMR deficient GC. Development of a prediction model was attempted using the common clinicopathological features. Results: In total, 11% (57/513) of the patients showed loss of expression of either one or more MMR proteins (MMR protein deficiency; MMR-D). Multivariate analysis demonstrated that age (≥70 years), sex (female), tumor location (lower 1/3), depth invasion (low, T1/T2/T3), and absence of distant metastasis were significantly independent predictive factors of MMR-D GCs. The MMR-D GC probability estimated by the prediction model ranged from 0.4% to 62.2%, and the area under the curve of the receiver operating characteristics curve was 0.82 (95% confidence interval=0.75-0.87). Conclusion: Our prediction model can sufficiently and efficiently identify MMR-D GCs using clinical features.
- Defective mismatch repair
- gastric cancer
- anti-PD-1 blockade
- microsatellite instability
Gastric cancer (GC) carcinogenesis has been studied in detail in the past decades. GC has been classified into four subtypes based on the Epstein-Barr virus, microsatellite instability (MSI), genomic stability, and chromosomal instability status (1). Among these subtypes, MSI-high (MSI-H) has received the most attention clinically, given that anti-PD-1 treatment is expected to be effective in unresectable GCs with MSI-H (2-5). MSI-H is a phenotype resulting from a deficiency in the MMR system (dMMR). Most MSI-H tumors show MMR protein deficiency in the nuclei (MMR-D) by using immunohistochemistry (IHC). The prevalence of MMR-D/MSI-H GC in East Asia has been reported to be approximately 10%, and the frequency tends to be even lower in unresectable or recurrent cases (6-9). Therefore, efficient screening for dMMR GC has been attempted, since models predicting MMR-D/MSI-H GC have not yet been developed. Thus, we extracted clinical pathological features to narrow down MMR-D GC, and developed an efficient cost-effective algorithm that can be used as a prediction model for MMR-D GC.
Patients and Methods
The study was approved by the local ethics committee of Saitama Medical Center, Saitama Medical University, Japan (approval numbers: 924-VIII, 925, and 926-V). All patients who underwent primary tumor resection for GC at the Medical Center of Saitama Medical University from March 2005 to December 2016 were included in the study after obtaining informed consent. Clinicopathological information was collected either from medical records or directly from patients. The pathological stage was determined according to the 7th version of the TNM classification (10). All histologically confirmed cases of GC were described according to the Japanese Classification of Gastric Carcinoma, which is similar to the 7th version of the TNM classification (11). Tumor grade was divided into two groups according to the Nakamura and Sugano classification: differentiated type and undifferentiated type (12). The exclusion criteria included cases with intramucosal cancer or with a small tumor size that was difficult to be analyzed by IHC, residual cases, and those with a special type of histological classification (11). Overall, a total of 513 GC patients were subjected in this study.
IHC for MMR proteins. IHC was performed on formalin-fixed paraffin-embedded (FFPE) 4-μm-thick tissue sections. The sections were pretreated in citrate buffer (pH 6.0) at 121°C for 15 min for antigen retrieval and then, stained for the four MMR proteins (MLH1, MSH2, MSH6, and PMS2) according to the manufacturer’s protocol. The primary antibodies used for detecting MMR protein expression were the anti-hMLH1 antibody (clone ES05, 1:50; Dako, Glostrup, Denmark), anti-hMSH2 antibody (clone FE11, 1:50; Dako), anti-hMSH6 antibody (clone EP49, 1:50; Dako), and anti-hPMS2 antibody (clone EP51, 1:40; Dako). The normal staining patterns for MLH1, MSH2, MSH6, and PMS2 are nuclear. The presence of nuclear staining in the tumor cells and non-neoplastic cells (such as normal endothelial cells, lymphocytes, or stromal cells) was used as a positive control and considered to have MMR protein expression (MMR-P). In contrast, the total absence of nuclear staining in the tumor cells in the presence of nuclear staining of non-neoplastic cells (such as normal gastric epithelial cells, lymphocytes, or stromal cells) was considered to be MMR-D. The results were assessed and a consensus was established among the pathologists (T.A.) and the first author (O.S.).
Statistical analysis. All statistical analyses were performed with EZR (Saitama Medical Center, Jichi Medical University, Japan), and all reported p values are two-tailed. EZR is a graphical-user interface for R (The R Foundation for Statistical Computing, Vienna, Austria, version 3.3.0) (13). This interface is a modified version of R commander (version 2.2-4) that includes statistical functions that are frequently used in biostatistics. Statistical significance was defined as a p-value<0.05. Data are expressed as medians and ranges, as appropriate. Categorical data were dichotomized where appropriate, and comparisons between groups were carried out using Fisher’s exact probability test. Continuous variables between the groups were compared using the Mann– Whitney test.
Development of the MpC model to predict MMR-D GC. Multiple logistic regression analysis was used to model the probability that the case was MMR-D. The model selection was based on backward step selection, and the odds ratio (OR) and 95% confidence interval (CI) for this model were calculated. A receiver-operating characteristic (ROC) curve was drawn, and the area under curve (AUC) was obtained. Based on the results of the logistic regression analysis, we developed a model to predict MMR-D probability called “MpC” (MMR protein loss by clinicopathology). This model was constructed using the method reported as an “MSI path” for the prediction model of MMR-D/MSI-H in CRC (14-17).
Results
Detection of MMR-D GCs. IHC analysis revealed that MMR-D existed in 57 (11.1%) of the 513 GCs. Loss of expression of both MLH1 and PMS2, loss of expression of MLH1, PMS2, and MSH6, and isolated loss of PMS2 was observed in 53 patients, three patients, and one patient, respectively. The differences in clinicopathological characteristics between patients with MMR-D GC and those with MMR-P GCs are shown in Table I.
Unlike patients with MMR-P GC, patients with MMR-D GC were statistically older (p<0.01), MMR-D GC was not observed in patients under the age of 56 years (Figure 1), and a higher proportion of cases were observed in women (p<0.01) (Table I). Moreover, MMR-D GCs were located in the lower region, whereas MMR-P GCs were located in the middle or upper region (p<0.01). Furthermore, advanced tumor stages were found at a lesser frequency in MMR-D GC patients compared to MMR-P GC patients; significant differences were observed in tumor stages estimated based on the T-factor (p<0.01), N-factor (p=0.024), and M-factor (p<0.01) for MMR-D GC and MMR-P GC patients.
Univariate and multivariate analysis for the prediction of MMR-D GC. Table II shows the univariate and multivariate analyses for the factors associated with MMR-D GC. Univariate analysis revealed that age (≥70 years, p<0.01), sex (female, p<0.01), tumor location (lower, p<0.01), depth of tumor invasion (T1/2/3, p<0.01), lymph node metastasis (N0, p=0.02), and distant metastasis (M0, p<0.01) were significantly associated with MMR-D GC. Multivariate analysis of these six factors by backward stepwise selection indicated that age (≥70 years, OR=2.80, 95%CI=1.46-5.38, p<0.01), sex (female, OR=5.63, 95%CI=2.97-10.7, p<0.01), tumor location (lower, OR=3.67, 95%CI=1.95-6.92, p<0.01), depth of tumor invasion (T1/2/3, OR=2.24, 95%CI=1.13-4.42, p=0.02), and distant metastasis (M0, OR=3.28, 95%CI=1.06-10.1, p=0.03) were independent variables of MMR-D GC. The AUC of the ROC was 0.82 (95%CI=0.75-0.87).
Development of the MpC model. We rounded the β coefficients to simplify the logistic model, which is shown in Table II, and the MpC scores were calculated using the following formula:
Where (A) Age (0=under 70, 1=70 and over 70), (B) Gender (0=male, 1=female), (C) Location (0=upper and middle, 1=lower), (D) Depth of invasion (0=T4, 1=T1 or T2 or T3) and (E) Distant metastasis (0=M1, 1=M0).
The prediction model for MMR-D GC is shown in Figure 2. The MpC score was calculated based on the log-odds scale and converted to probability using the formula: 1-exp (–5.53 + MpC score)/[1 + exp (–5.53 + MpC score)]
This allowed for the ease of using categories of scores using the MpC conversion scale provided at the bottom of Figure 2. The sensitivity and specificity for a given MpC score with ROC are shown in Figure 3. The sensitivity and specificity were 70% and 82%, respectively, for a cutoff value of 3.5 points. Assuming a cutoff value of 3.5 or higher, 26% (43/163) of the 163 cases with a cutoff value of 3.5 or higher were MMR-D, and data on 75.4% (43/57) of all MMR-D GCs were extracted. Conversely, 96% (336/350) of the cases with a cutoff of less than 3.5 were MMR-P GCs, and 73% of all MMR-P GCs could be excluded. The actual prevalence of MMR-D GC in each score range (range=0-0.9, 1.0-1.9, 2.0-2.9, 3.0-3.9, 4.0-4.9, 5.0-6.0) was 0%, 4%, 2%, 9%, 26% and 32%, respectively (Figure 4).
Discussion
Here, we used IHC to evaluate MMR deficiency. In contrast to MSI testing, IHC has lower cost and is widely available in routine diagnostic laboratories. Moreover, direct germline mutation testing can also be carried out using IHC. Therefore, IHC is more widely used than MSI testing (18-19). With the development of high-quality antibodies for IHC and the unification of MSI markers, excellent concordance rate between IHC and MSI testing for MMR deficiency has been reported, including 96-98% and 97% concordance rates reported for CRC (20-23) and GC (24), respectively. Therefore, we deduced that IHC can be sufficiently used to assess the MMR status of GCs. The prevalence of MMR deficiency in GCs was reportedly 15% by IHC, 7% to 25% by MSI test, and 9% to 21% by next-generation sequencing (1, 6, 24-27). Recently, systematic reviews reported that the frequency of dMMR in GCs varies by population, race, region, and stage. A lower deficiency has been reported in Asia than in the West (10.8% and 17.6%, respectively) (289). Here, the frequency of MMR deficiency in GC was 11%, which is in agreement with previous reports on Asian races (6-9, 256, 2930).
Furthermore, in this study, we developed a prediction model to identify MMR-D GCs using the following clinicopathological factors: age, sex, tumor location, tumor infiltration depth, and distant metastases. Although these clinicopathological factors have been proposed to be associated with MMR-D/MSI-H GCs (7, 256, 301-356), prediction models using a combination of these factors for MMR-D/MSI-H GC prediction are non-existent. The five clinicopathologic factors used in our model are easily obtainable at any facility during the regular medical care for GC. The model can predict the probability of MMR-D GCs with high accuracy (AUC 0.82). Moreover, the use of weighted factors increased the accuracy of the prediction formula presented here. Thus, this prediction model can assess the need for evaluation of the MMR status in individual cases without incurring additional costs. GC patients with MMR-D could not be perfectly isolated by the MpC model. However, according to this model, 96% of the GC patients (336/350) with an MpC score of less than 3.5 had MMR-P GC, indicating a negligible need for IHC/MSI testing. Meanwhile, 30% (41/137) of GCs with a MpC score of 3.5 or higher were MMR-D GCs, indicating the need for further IHC/MSI testing. Thus, by using a combination of our prediction model with IHC of MMR, as a complement to the regular GC care routine, we showed that the development of an inexpensive and efficient MMR status evaluation system can be established at any facility.
In GCs, MMR deficiency is attracting much attention as an accurate prediction biomarker for more appropriately selecting patients with GC most responsive to anti-PD-1 therapy. PD-1 blockade using pembrolizumab, has been shown to be effective against MMR-D/MSI-H solid tumors (4, 367-389), and is currently approved for previously treated advanced GC patients in the United States. In a recent prospective phase 2 clinical trial (3940), dramatic responses to pembrolizumab were observed in MMR-D/MSI-H GCs than in MMR-P/MSS GCs [overall response rate (ORR) 85% vs. 18%, p<0.01, respectively]. Nivolumab is another monoclonal antibody against PD-1 that can be used for GC, but unlike pembrolizumab, the MMR status is not required, and is widely used. Interestingly, recent study demonstrated that nivolumab has a higher therapeutic effect on MMR-D/MSI-H GCs (401). According to this study, the ORR of GC patients who were given nivolumab after two or more chemotherapy regimens was significantly higher in MMR-D/MSI-H GCs than in those with MMR-P/MSS GCs (75% vs. 13%, p<0.01). Therefore, MMR status may also be an important predictor of therapeutic efficacy for nivolumab.
This study has the following limitations: 1) selection bias could be caused by its retrospective design, 2) the study sample size of 57 MMR-D/MSI-H GCs was small, and 3) it was a single-institutional study. 4) In the present study, the model was developed based only on the analysis group, and verification using a validation cohort was not possible; hence, validation of the model is necessary. Nevertheless, given the lack of similar reports on prediction models for GCs, this model would be useful to clinicians.
In conclusion, the prediction model that we developed can sufficiently and efficiently identify MMR-D/MSI-H GCs using only factors obtained in daily clinical practice. We believe that this prediction model can be applied in all clinical situations as an efficient and cost-effective tool for screening GC patients who are likely to benefit from PD-1 blockade.
Acknowledgements
The Authors gratefully acknowledge the work of past and present members of their laboratory.
Footnotes
Authors’ Contributions
O.S., M.F., and H.I. conceptualized and designed the study.
O.S., M.F., E.M., T.Y., T.A., K.A., and H.I. involved in acquisition of data. O.S., T.Y., T.A., K.A., and H.I. involved in analysis and interpretation of data. All the Authors were involved in manuscript for all aspects of the work.
Conflicts of Interest
K.A. reports research funding received from MSD, Ono and Falco Biosystems; honoraria received from Chugai, Taiho, MSD. H.I. received a research grant from Taiho. The other Authors declare no conflicts of interest in relation to this study.
- Received December 10, 2020.
- Revision received December 22, 2020.
- Accepted January 11, 2021.
- Copyright © 2021 International Institute of Anticancer Research (Dr. George J. Delinasios), All rights reserved.