Identification of correlations between mRNAsi and clinical characteristics
The mRNAsi score of 302 OC cases were obtained by using the OCLR algorithm, and sample data was divided into two groups based on the median mRNAsi score (Figure 1A). As shown in Figure 1B and 1F, we found no significant correlation between mRNAsi and the age or stage of OC cases, respectively. OC patients with venous or lymphatic invasion had higher mRNAsi scores compared to patients with no invasion (Figure 1C, D). Further, grade 3/4 cases had higher mRNAsi score compared to grade 1/2 cases (p 0.024) (Figure 1E). The Kaplan-Meier survival curve (Figure 1G) indicated that mRNAsi had no significant prognostic value for OC.
Filtrating of differentially expressed miRNAs, lncRNAs, and mRNAs
By using the “limma” package, 313 lncRNAs (239 downregulated and 74 upregulated), 38 miRNAs (16 downregulated and 22 upregulated), and 2 downregulated mRNAs were obtained and then visualized by a volcano plot (Figure 2A, B, C).
Construction of the lncRNA-miRNA-mRNA competing endogenous RNA networks
The ceRNA network was built based on differentially expressed lncRNA, miRNA, and mRNA using Cytoscape. We first built the interaction network based on gene pair correlations (Figure 2D). We then selected 5 lncRNA-miRNA-mRNA gene pairs to construct the ceRNA network (Figure 2E, Supplementary Table 1). The network contained 5 lncRNAs (ENSG00000230333, ENSG00000227220, ENSG00000234323, ENSG00000233821, ENSG00000266411), 4 miRNAs (hsa-miR-141-5p, hsa-miR-19b-1-5p, hsa-miR-30e-5p, hsa-miR-s29a-5p), and 1 mRNA (CHN2).
Analysis of the relationship between CALB2 and CHN2 copy number variation and mRNAsi score
Figure 3E depicts the locations of CALB2 and CHN2 on the chromosome. Initially, the relationship between the gene expression level and CNV of CALB2 and CHN2 were analyzed. There was a positive correlation between the gene expression level and CNV of CALB2 and CHN2, which was consistent with our assumption (Figure 3A, B). Transcriptomic data indicated that the expression of CALB2 and CHN2 was lower in the high-mRNAsi group than in the low-mRNAsi group, and, based on this, we predicted that the mRNAsi score may have a negative correlation with the CNV of CALB2 and CHN2. Thus, we investigated the relationship between mRNAsi score and the CNV of CALB2 and CHN2. The result indicated that the CNV of CHN2 was negatively correlated with the mRNAsi score (Figure 3C). However, there was no statistically significant correlation between the mRNAsi score and the CNV of CALB2 (Figure 3D).
Establishment of the prognostic risk score model by Lasso regression
Univariate Cox PHR was used to identify survival-related lncRNAs, miRNAs, and mRNAs from differentially expressed genes based on mRNAsi score grouping. In general, 60 lncRNAs and 3 miRNAs (p<0.05) were obtained, as shown in Supplementary Table 2. To further narrow down candidate survival-related genes, 11 candidate genes, including 9 lncRNAs and 2 miRNAs, were obtained by using the Lasso binary logistic regression model (Figure 4A, B). Regression analysis of multivariate Cox proportional hazards was then used to further narrow the scope of candidate genes. 9 candidate genes, including 7 lncRNAs and 2 miRNAs, were obtained and used to establish the risk score model. The risk score model was then built with the addition of expression level results and the regression coefficient of each gene in the Lasso regression model: risk score = 0.00014*ENSG00000230601+0.000213*ENSG00000275850+0.000174*ENSG00000225807-0.00052*ENSG00000234754+5.66E-05*ENSG00000259645+4.27E-05*ENSG00000227068 +4.85E-05*ENSG00000249706 -0.02221*hsa-miR-133a-3p+0.099624*hsa-miR-6788-3 (Supplementary Table 3). OC patients with lower risk scores had better OC outcome, as shown in Figure 4C. The ROC curve had an area of under curve of 0.658, which indicated that the prognostic risk score model possessed good predictive power for prognosis (Figure 4D). The risk score of each OC patient was calculated based on expression of the 9 selected genes. The survival status, risk score distribution, and expression pattern of the 9 genes are depicted in Figure 4E-G. I have changed this so that the way you write out Lasso is consistent throughout the text.
Construction and assessment of a nomogram for overall survival prediction for OC
A nomogram model based on the 9 candidate genes was established for the prediction of overall survival of OC at 1, 3, and 5 years (Figure 5A). The model’s calibration curve for the probability of overall survival at 1, 3, and 5 years had accurate predictive capacity (Figure 5B-D). A DCA curve was constructed to assess the nomogram model’s clinical benefits. We found that the nomogram model could be of benefit for predicting the 1-, 3- and 5-year overall survival based on the red oblique line of the DCA curve (Figure 5 E-G). Finally, a clinical impact curve was plotted to visualize the nomogram model’s clinical potential. The predicted number of high-risk patients was more than the high-risk number with an adverse event at 1, 3, and 5 years, indicative of the predictive power of the nomogram model (Figure 5 H-J).
Assessment of the relationship between mRNAsi and tumor-infiltrating immune cells
We calculated the relative ratio of 22 tumor infiltrating immune cell types by using the CIBERSORT algorithm. We obtained 21 tumor-infiltrating immune cell types, as there was no significant fraction of naïve CD4 T cells. CIBERSORT results are presented in Figure 6A. We then investigated differential immune cell infiltration between the high- and low-mRNAsi groups. As shown in Figure 6B, fractions of follicular T helper cells, M1 macrophages, and activated dendritic cells were much higher in the high-mRNAsi group compared to the low-mRNAsi group. However, the fraction of resting dendritic cells followed the opposite trend. The other 17 infiltrating immune cell types showed no significant difference. As revealed by PCA analysis results presented in Figure 6C, tumor-infiltrating immune cell fractions could effectively discriminate OC patients belonging to the high-mRNAsi group and low-mRNAsi group. We explored the relationship between mRNAsi score and immune infiltration by Pearson correlation analysis. The result indicated a positive correlation between mRNAsi score and plasma cells, activated dendritic cells, and follicular T helper cells (Figure 7A, E, F). Moreover, mRNAsi decreased in parallel with the increase of resting dendritic cells, as well as M2 macrophages and resting mast cells (Figure 7B, C, D).