- Association of P4HB and GRP78 protein expression with clinicopathological characteristics in GC
We detected the expression of P4HB and GRP78 separately by immunohistochemistry (IHC) in 150 GC tissues. The representative images of P4HB and GRP78 staining are shown in Figure 1. A total of 74 patients (49.3%) were P4HB low expression and 76 patients (50.7%) were P4HB high expression, as for GRP78 expression, there were 71 low cases (47.3%) and 79 high cases (52.7%) respectively. Furthermore, the expression of P4HB was correlated with age (P = 0.021), Bormann type (P = 0.029), depth of tumor invasion (P = 0.001), lymph node metastasis (P = 0.018), postoperative adjuvant chemotherapy (P = 0.008), while GRP78 expression was only correlated with depth of tumor invasion (P = 0.041) and lymph node metastasis (P = 0.043) (Table 1). And P4HB expression was strongly correlated with GRP78 (P < 0.001). Subsequent Spearman rank correlation analysis of P4HB and GRP78 H-score proved there existed a significant positive correlation (R = 0.587, P < 0.001) (sFigure 1A). By investigating the correlation coefficients of different clinicopathological characteristics, the correlation coefficients were significantly elevated in tumor size ≥ 5 cm (P = 0.037), TNM Ⅲ/Ⅳ stage (P = 0.037) and positive vessel invasion (P = 0.049) subgroups, as shown in sTable 2.
- Bioinformatics analyses of P4HB and GRP78 expression in GC
We explored TCGA-STAD (stomach adenocarcinoma) data by GEPIA and analyzed of extracted GC cell lines data (amount to 37 lines) from CCLE successively. It turned out P4HB mRNA was also positively correlated with GRP78 in both GC tissue and cell databases (R = 0.61, P < 0.001, and R = 0.356, P = 0.031, respectively) (sFigure 1B and 1C), in accordance with the correlation of proteins. Then we performed GO and KEGG analysis of these two genes. The top 5 items of each GO term was shown in sFigure 1D. Only items involve with both two genes were taken into account. Finally, protein folding and response to ER stress were discovered in biological process (BP) term, ER chaperone complex, melanosome, pigment granule, ER-Golgi intermediate compartment and ER lumen were discovered in cellular component (CC) term, together with only cell adhesion molecule binding was discovered in molecular function (MF) term. KEGG pathway analysis showed both two genes were enriched in the pathway “protein processing in ER” (sFigure 1E). As the occurrence and development of GC was most likely related to ER stress and UPR, we made further efforts to search correlation of P4HB or HSPA5 with genes within UPR, originated from the “protein processing in ER” KEGG pathway map (hsa04141), by explore TCGA–STAD cohort in GEPIA (sFigure 2-3). Except for P4HB and HSPA5 both positively correlated with typical genes within UPR from STAD (Table 2), correlation coefficients of most upstream genes were higher than downstream genes, as shown in sFigure 4.
- Association of P4HB and GRP78 expression with OS in GC
We applied Kaplan-Meier survival analysis and log-rank test to compare OS according to P4HB and GRP78 expression. The results revealed that high expression of P4HB (Figure 2A) and GRP78 (Figure 2B) alone was related to an obvious shorter OS (P < 0.001 and P = 0.012, respectively). Further stratification analysis showed that the prognosis was the best when P4HB and GRP78 were both low expressed, the worst prognosis was correlated with high expression of both P4HB and GRP78, while the prognosis was in between when one of P4HB and GRP78 was high expressed (Figure 2C, P = 0.002). When analyzing of co-expression of P4HB and GRP78, OS of both or one high expression was sharp lower than that of both low expression (Figure 2D, P = 0.002).
Since P4HB and GRP78 were both reported to be associated with chemotherapy, we also explored the prognosis of fluorouracil-based postoperative adjuvant chemotherapy together with co-expression of P4HB and GRP78, which contained both variables might more precise. We observed the prognosis of group with postoperative adjuvant chemotherapy was better than the other group (Figure 2E, P = 0.011). High level of P4HB and/or GRP78 exhibited significant worse in the former group (Figure 2F, P = 0.007), but no prognostic significance in the latter (Figure 2G, P = 0.231). Once the group with postoperative adjuvant chemotherapy was further divided according to TNM stage, it exhibited no prognostic significance in TNM Ⅰ/Ⅱ stage (Figure 2H, P = 0.859), on the contrary, a significant worse OS in TNM Ⅲ/Ⅳ stage (Figure 2I, P = 0.004). This indicates that GC patients in TNM Ⅲ/Ⅳ stage may not benefit from postoperative adjuvant chemotherapy if P4HB and/or GRP78 high expressed.
- Construction of a nomogram based on stepwise Cox regression analysis
Univariate analyses were performed for clinicopathological characteristics to estimate postoperative survival in our study group. According to the Table 3, we observed that OS was significantly correlated with Bormann type (P = 0.004), tumor size (P = 0.028), differentiation (P = 0.004), depth of tumor invasion (P < 0.001), lymph node metastasis (P < 0.001), TNM stage (P < 0.001), postoperative adjuvant chemotherapy (P = 0.012), P4HB expression (P = 0.001), GRP78 expression (P = 0.013), P4HB and GRP78 co-expression (P = 0.002). These statistically significant variables were subsequent recruited to forward stepwise COX analysis. The result showed that differentiation (P = 0.020, HR = 1.846, 95% CI: 1.099-3.101), TNM stage (P < 0.001, HR = 3.124, 95% CI: 1.870-5.217), postoperative adjuvant chemotherapy (P = 0.005, HR = 0.507, 95% CI: 0.317-0.810) and P4HB and GRP78 co-expression (P = 0.002, HR = 2.304, 95% CI: 1.355-3.915) were independent prognostic factors in GC patients (Table 3). Then four independent prognostic factors were included to construct a nomogram for predicting OS probability of GC patients (Figure 3A). Harrell’s C-index for nomogram was 0.733 (95% CI: 0.680-0.786). And calibration plots for nomogram were close to the 45-degree line (the perfect calibration model) at 3-year and 5-year (Figure 3B and 3C), indicated the model was well calibrated.
- Evaluate the nomogram and its clinical value with TNM stage
To evaluate the discrimination ability and clinical uesfulness of the nomogram, we compared it with widely clinical used TNM stage. From Figure 4A and 4B, we found areas under ROC (receiver operating characteristic) curve (AUROCs) of the 3-year and 5-year OS probability for nomogram model were 0.790 and 0.760, respectively, much better than that for TNM stage (0.684 and 0.671, respectively). Since DCA has superiorities over AUROC for estimating alternative prognostic strategies, we performed the 3-year and 5-year DCA curves individually for the nomogram and TNM stage. As shown in Figure 4C and 4D, compared to TNM stage, even at a very small threshold probability, the nomogram offered a high net benefit in 3-year and 5-year, respectively.