MG7-Ag, hTERT, and TFF2 identified high-risk intestinal metaplasia and constituted a prediction model for gastric cancer.

Abstract

To the Editor: Gastric cancer (GC) ranks third in incidence and mortality both worldwide and in China.[1,2] Intestinal metaplasia (IM) significantly increases risk of GC so that identifying high-risk IM patients who will progress to GC is crucial. Currently, the effect of many risk-stratification methods for gastric precancerous lesions (GPLs) was minimal. Monoclonal gastric cancer 7 antigen (MG7-Ag) combined with cyclooxygenase-2 has been shown early-warning value for the progression of GPLs.[3] The expression of human telomerase reverse transcriptase (hTERT) was proven to be related to state of cell proliferation in IM tissue.[4] Loss expression of trefoil factor family 2 (TFF2) from spasmolytic polypeptide-expressing metaplasia to IM may lead to a hyperproliferation and deleterious mutations.[5] We tried to construct and verify a multimolecular prediction model included MG7-Ag and hTERT and TFF2, which could identify high-risk IM patients and have early-warning value for GC. Our study was approved by the Ethics Committee of Xijing Hospital, Air Force Medical University. In the first stage, we conducted a retrospective case–control study. Seventeen patients diagnosed with gastric IM progression to GC in the follow-up from July 2009 to September 2019 were viewed as the case group (intestinal metaplasia progressing to gastric cancer [IM-GC] group). The following three kinds of matched cases were control groups: IM-NoGC group (patients with IM not progressing to GC), CG-NoGC group (patients with chronic gastritis not progressing to GC), and GC group (GC samples from patients with IM progression to GC, a self-control group of IM-GC group). Each control group included 17 patients. The inclusion criteria for IM-GC group were as follows: (1) age at diagnosis between 18 years and 75 years and (2) the interval between diagnosis of IM and GC should be no ≥6 months. The two control groups were matched with the IM-GC group on age, sex, and the time interval of pathological biopsy. The modeling cohort consisted of the IM-GC group and the IM-NoGC group in the first stage (34 cases). Logistic regression analysis was performed on establishing models. In the second stage, we evaluated early-warning value of multimolecular prediction model of MG7-Ag, hTERT, and TFF2 for GC. The validation cohort consisted of 31 patients in Xijing Hospital from October 2019 to December 2020 including 8 patients in the IM-GC group and 23 patients in the IM-NoGC group. We imported the measured scoring data into different models for verification, evaluation, and comparison. All the baseline data of patients are displayed in Supplementary Tables 1 and 2, https://links.lww.com/CM9/B368. A semiquantitative analysis was applied to the immunohistochemical staining results. The final score was the product of the positive cell number score and the positive intensity score. Evaluation of positive cell number score: 0 to 5%, 6% to 25%, 26% to 50%, 51% to 75%, and 76% to 100% were 0, 1, 2, 3, and 4 points, respectively. Evaluation of dyeing strength: no-staining, light-yellow, brown-yellow, dark-brown were 0, 1, 2, and 3 points, respectively. The final score was the product of the positive cell number score and the positive intensity score, which recorded as negative (0), weakly positive (1–4), moderately positive (5–8), and strongly positive (9–12). In the retrospective case–control study, Friedman test was applied for difference comparison both among four groups and between two groups. Factors associated with IM progression to GC were analyzed by univariate logistic regression analysis and receiver operating characteristic (ROC) curve analysis. The Kappa coefficient was used to evaluate the consistency between single factor and histopathological method. Area under ROC curve (AUC) and Youden index was used to obtain sensitivity, specificity, and truncation values. In prospective analysis, Mann–Whitney U test was used for comparison between groups. Kruskal–Wallis test was applied for stratified comparison of degree of IM among different groups. In verification, distinction of model was evaluated by ROC curve. Calibration of model was evaluated by Hosmer–Lemeshow goodness-of-fit test and Calibration Plot. In the retrospective case–control study, MG7-Ag showed overexpression in the GC group and low or no expression in other groups, that was significantly different (P < 0.05). There was no significant difference in MG7-Ag expression between the IM-NoGC and CG-NoGC groups (P > 0.05, Supplementary Figure 1E, https://links.lww.com/CM9/B368). Overexpression of hTERT was found in the GC and IM-GC groups, while low expression was observed in other groups. There was a significant difference in hTERT expression between the IM-GC and IM-NoGC groups (P < 0.001). No significant difference in hTERT expression was found between the GC and IM-GC groups or between the CG-NoGC and IM-NoGC groups (P > 0.05, Supplementary Figure 1J, https://links.lww.com/CM9/B368). Low expression of TFF2 was observed in the GC and IM-GC groups, while overexpression was found in other groups. There was a significant difference in TFF2 expression between the IM-GC and IM-NoGC groups (P = 0.001). TFF2 expression was not significantly different between the GC and IM-GC groups or between the CG-NoGC and IM-NoGC groups (P > 0.05, Supplementary Figure 1O, https://links.lww.com/CM9/B368). After logistic regression analysis (inclusion criterion: P < 0.1), hTERT and TFF2 staining scores were found to be correlated with the progression from IM to GC [Supplementary Table 3, https://links.lww.com/CM9/B368]. An ROC analysis showed that hTERT and TFF2 staining scores could distinguish high-risk IM patients from low-risk IM patients (P < 0.01, Supplementary Table 4, https://links.lww.com/CM9/B368). The AUC values of hTERT and TFF2 were calculated as 0.815 and 0.933, respectively. We included hTERT and TFF2 in the model as independent variables. Because of the predictive ability of MG7-Ag showed by past studies, we finally included hTERT, TFF2, and MG7-Ag to construct two prediction models [Supplement Table 5, https://links.lww.com/CM9/B368]. Model 1 (MG7-Ag + hTERT + TFF2): ln[P/(1 – P)] = 5.403 + 0.202 × MG7-Ag + 0.576 × hTERT – 1.212 × TFF2. Model 2 (hTERT + TFF2): ln[P/(1 – P)] = 5.915 + 0.584 × hTERT – 1.212 × TFF2. In verification, there was a significant difference in hTERT expression between the IM-GC and IM-NoGC groups in the validation cohort (P < 0.05). Overexpression was found in the IM-GC group. Neither MG7-Ag nor TFF2 expression showed a significant difference between the IM-GC and IM-NoGC groups (P > 0.05). However, the proportion of moderately to strongly positive MG7-Ag expression in the IM-GC group was higher than that in the IM-NoGC group (37.5% > 4.35%). Similarly, the proportion of strongly positive TFF2 expression in the IM-NoGC group was higher than that in the IM-GC group (26.09% > 12.5%). Models 1 and 2 were verified in the modeling and validation cohort [Figure 1A, B]. In modeling cohort, AUC value of Models 1 and 2 were both 0.971. These results indicated that in internal validation, the probability of prediction from IM to GC by Models 1 and 2 both reached 97.1%. By contrast, Model 1 had a slightly higher prediction sensitivity (88.2% > 82.4%). The specificity of Models 1 and 2 were both 100%. In the validation cohort, AUC value of Models 1 and 2 were 0.87 and 0.84. The sensitivity and specificity of Model 1 were 75.0% and 87.0%, respectively. The sensitivity and specificity of Model 2 were 100% and 65.2%, respectively. The probability of prediction from IM to GC by Models 1 and 2 reached 87.0% and 84.0%, respectively. By contrast, the results showed that Model 1 had higher accuracy.Figure 1: ROC curve and calibration degree diagram of multimolecular prediction model of multimolecular prediction models. (A, B) ROC curve of Models 1 Model 2 in internal and external verification. (C, D) Calibration degree diagram of Model 1 in internal and external verification, respectively. (E, F) Calibration degree diagram of Model 2 in internal and external verification, respectively. hTERT: Human telomerase reverse transcriptase; ROC: Receiver operating characteristic; TFF2: Trefoil factor family 2.In the internal and external verification of the two prediction models, both AUC values of the models were >0.75, which indicated good discrimination ability. The chi-squared values of Model 1 in the internal and external verification were 4.592 and 3.576, respectively (both P > 0.05). The chi-squared values of Model 2 in the internal and external validation were 1.318 and 3.297, respectively (P > 0.05). These results suggested that these two prediction models showed good calibration ability in both internal and external verification. The same conclusion was obtained in the scatter diagrams of the visual standard degree [Figure 1C–F]. In our study, through a strict study design, the control groups (IM-NoGC and CG-NoGC) were included by matching the case group with age, sex, and diagnosis time. The heterogeneity was reduced; the comparability between the case and control groups and the reliability of the conclusion were increased. The three-molecule prediction model showed good early-warning value for GC among IM patients. Its accuracy for prediction was higher than that of the two-molecule prediction model and better than previous reports. The annual progression rate of GC in our study reached approximately 8%, and those of past studies were <3%.[6] We suggested that a combination of factors could complement and optimize the prediction model. There were still some deficiencies in our study. All of the cases came from a single center. The sample size was small. To conclude, in this study, MG7-Ag, hTERT, and TFF2 showed different expression levels in IM patients with different outcomes. This study may have an important role on predicting GC risk among IM patients. Funding This work was supported by the Shaanxi Foundation for Innovation Team of Science and Technology (No. 2018TD-003) and the Project from State Key Laboratory of Cancer Biology (No. CBSKL2019ZZ07). Conflicts of interest None.