Construction of Cox model based on IRGP
Data for a total of 1526 HGSOC samples that received platinum-based treatment were obtained from the TCGA-OV, GPL96-combined, GPL570-combined, and GPL6480-combined cohorts. TCGA-OV was used as the training set and GPL96-combined, GPL570-combined, and GPL6480-combined cohorts were used as the validation set. We obtained 238 common immune genes from the intersection of the four cohort immune genes (Fig. 2a). An IRGP analysis was performed by pairwise construction of 238 genes in each cohort. The IRGP that meets the requirements of the four cohorts was then intersected to obtain 2672 common IRGPs (Fig. 2b). The relationship between the 2672 IRGPs and overall survival rates was evaluated using the TCGA-OV cohort, and the results indicated that 141 IRGPs were associated with the prognosis of HGSOC (p < 0.01). The lambda and coef diagrams of IRGPs (Fig. 2c) were constructed using the lasso algorithm. As the lambda value increases, the coefficients of some IRGPs become zero, which means that the scores of these IRGPs do not affect the model. We then used 10-fold CV to calculate the partial likelihood deviance of the model (Fig. 2d). The deviance was smallest when 35 genes were used. The coef of each IRGP was obtained according to the corresponding lambda value (Fig. 2e). The 35 IRGPs and the corresponding coef values constitute the IRGPI prognostic model (Additional fle 1: Table S4). The IRGPI values were calculated for each sample in the four cohorts. These IRGPs include 52 genes. Through pathway and process enrichment analysis, we found that these genes are mainly concentrated in "cell chemotaxis," "cytokine-mediated signaling pathway," and "positive regulation of cell migration" (Additional fle 1: Table S5).
Verification of the prognostic ability of IRGPI of overall survival rate in the training set
IRGPI significantly stratified HGSOC patients in TCGA-OV, with poor prognosis and high risk for high IRGPI (p < 0.001, Fig. 2f). The AUC values were 0.637, 0.711, 0.766, 0.826, and 0.897 for 1, 3, 5, 7, and 9 years, respectively (Fig. 2g). Univariate Cox analysis showed that age, stage, and IRGPI had predictive ability (p < 0.05). Multivariate Cox analysis showed that age and IRGPI could be used as independent prognostic factors (p < 0.001). IRGPI also effectively stratified the risk of HGSOC patients under different clinical types (Age < = 60; age > 60; grade 2; grade 3; stage 3; stage 4; p < 0.001, Fig. 2j). Overall, IRGPI can be used to independently evaluate the overall survival rate of patients with HGSOC.
Prognostic ability of the overall survival rate of IRGPI in the validation sets
IRGPI can be used as a prognostic factor for the GPL96-combined, GPL570-combined, and GPL6480-combined cohorts. In the GPL96-combined cohort, the results of Kaplan-Meier analysis showed worse prognosis of patients with the high-IRGPI group compared to the low-IRGPI group (p = 0.001, Fig. 3a); The IRGPI distribution and patient survival status plots are shown in Fig. 3b; The AUC values for 3, 5, and 7 years were 0.651, 0.637, and 0.569, for the GPL96-combined, GPL570-combined, and GPL6480-combined cohorts, respectively (Fig. 3c). Similarly, in the GPL570-combined cohort, Kaplan-Meier analysis showed worse prognosis of patients for the high-IRGPI group compared to that of the low-IRGPI group (p = 0.002, Fig. 3d); Fig. 3e shows that IRGPI distribution and patient survival status plots; and the AUC values for 3, 5, and 7 years were 0.574, 0.560, and 0.641, respectively (Fig. 3f). Finally, in the GPL6480-combined cohort, Kaplan-Meier analysis showed worse prognosis of patients for the high-IRGPI group compared to the low-IRGPI group (p < 0.001, Fig. 3g). Figure 3h shows the IRGPI distribution and patient survival status plots, and the AUC values for the three, five, and seven years were 0.609, 0.620, and 0.630, respectively (Fig. 3i). Overall, IRGPI successfully stratified the risk of HGSOC patients for the three validation sets.
IRGPI can also predict progression-free survival (PFS) probability in patients with HGSOC
Including only samples with PFS information, 1011 cases of HGSOC were further analyzed, including samples from the TCGA-OV (N = 363), GPL570-combined (N = 277), and GPL6480-combined cohorts (N = 371). Kaplan-Meier analysis showed that IRGPI allowed risk stratification of HGSOC in TCGA-OV (p < 0.001, Fig. 4a), GPL570-combined (p = 0.007, Fig. 4b), and GPL6480-combined cohorts (p < 0.001, Fig. 4c). Low PFS correlated with high IRGPI.
We analyzed the differences in the infiltration levels of immune cells between the high-IRGPI and low-IRGPI groups. The radar map presented in Fig. 5a shows a significant difference in the level of infiltration in the TCGA-OV cohort for M1 macrophages, gamma delta T cells, and follicular helper T cells (p < 0.001). In the GPL96-combined cohort, significant differences were seen for M1 macrophages, CD4 memory resting T cells, monocytes, and resting dendritic cells (p < 0.001, Fig. 5b); in the GPL570-combined cohort, differences were seen for M1 macrophages and CD4 memory resting T cells (p < 0.001, Fig. 5c); in the GPL6480-combined cohort, differences were seen for M1 macrophages, activated mast cells, plasma cells, monocytes, CD4 memory activated T cells, CD4 memory resting T cells, and follicular helper T cells (p < 0.001, Fig. 5d). Among these immune cells, only M1 macrophages showed significant differences in high-IRGPI and low-IRGPI groups for all four cohorts. M1 macrophages exhibited higher infiltration level in the low-IRGPI group than that in the high-IRGPI group (p < 0.001, Fig. 5e).
IRGPI was applied to divide the samples into high-IRGPI and low-IRGPI groups to analyze differences in the enriched GSEA KEGG pathways between the two groups. The results showed few enrichment pathways in the high-IRGPI group, and no pathway is enriched in four cohorts at the same time (Fig. 6a). In contrast, in the low-IRGPI group, many pathways were enriched, and ten pathways were enriched in all four cohorts (Fig. 6b). The bubble charts presented in Fig. 6c-f show many of these enriched pathways are immune-related signal pathways. Among the four cohorts, the “Antigen processing and presentation” was the most highly enriched (Fig. 6g-j). This result suggests that the better prognosis of the low-IRGPI patients is related to the activity of this pathway.