The landscape of anoikis-related genetic features in OC
As depicted in Fig 1a, 44 AIRGs were differently expressed between OC tissue and normal ovarian tissue. We analyzed the correlation of these molecules in TCGA-OV cohort and found that most genes were significantly correlated in the positive direction (Fig. 1b), indicating that the interactions among AIRGs may play a crucial role in OC progression. We further explore the somatic mutations of AIRGs and LRP1 was identified as the molecules with the highest frequency of mutations (Fig. 1c). We also determine the CNV-related mutations of AIRGs. The investigation demonstrated that most of these 44 AIRGs display significant amplification, whereas AP1S2, BLMH, FBLN1 and CELSR1 exhibited evident “loss” (Fig. 1d). These findings supported that AIRGs correlated with genomic background of OC.
Derivation of an anoikis-related prognostic model for OC patients
Univariate Cox regression analysis of 43 AIRGs was performed and we observed that 11 genes including EML2, LRP1, CYBRD1, RPS6KA2, TFCP2L1, PI3, ISG20, STARD13, AP1S2, PRSS16 and CITED2 were most significantly correlated with prognosis of OC patients (p<0.01, Fig. 1e). We selected these 11 genes and employed the LASSO algorithm to identify them as prognostic gene sets for OC (Fig. 2a). The coefficients of 11 AIRGs were described in Table 1, and the risk scores of AIRGs (AIRS) were derived as follows: = (0.2471* EML2 expr.) + (-0.0624* LRP1 expr.) + (0.0761* CYBRD1 expr.) + (0.2028* RPS6KA2 expr.) + (0.0058* TFCP2L1 expr.) + (0.1385* PI3 expr.) + (-0.2746* ISG20 expr.) + (0.1270* STARD13 expr.) + (-0.1127* AP1S2 expr.) + (-0.2930* PRSS16 expr.) + (0.1849* CITED2 expr.). 365 OC patients were stratified into high-risk subgroups (n=200) and low-risk subgroups (n=165) based on the optimum cut-off value calculated by AIRS, and high-risk groups has more blue dots representing dead status (Fig. 2b). 11 selected AIRGs distribution were depicted in the heatmap above (Fig. 2c). Subsequently, KM analysis was used to compared the overall survival (OS) of the high- and low-risk set, which showed that the probability of death of lower AIRS subgroup was markedly lower (p<0.05) (Fig. 2d). Time-dependent ROC curve was plotted to assess the prognostic signature. The AUC values of the 1- , 2-, 3-, and 5-year OS in TCGA cohort were 0.636, 0.727, 0.711 and 0.771, respectively (Fig. 2e).
Table 1 The regression coefficient of 11 AIRGs
Gene
|
Coefficient
|
EML2
|
0.247116275
|
LRP1
|
-0.062364789
|
CYBRD1
|
0.076122022
|
RPS6KA2
|
0.202757962
|
TFCP2L1
|
0.005819503
|
PI3
|
0.13845466
|
ISG20
|
-0.274565273
|
STARD13
|
0.126978231
|
AP1S2
|
-0.112690043
|
PRSS16
|
-0.292971349
|
CITED2
|
0.184851831
|
Validation of prognostic and predictive role of risk model in OC
Next, the risk signature was confirmed in the ICGC-OC-AU cohort and GSE26712 cohort by using the same algorithm. 88 ICGC-OC patients and 185 GSE26712-OC patients were classified into high-risk subgroup and low-risk subgroup, respectively (Fig. 3a). The mortality incidence was lower in the low-risk categories versus high-risk categories, which was in good agreement with TCGA-OV signature. Also, KM analyses and prognostic AIRGs distribution were almost in accord with the outcome of TCGA cohort (Fig. 3b, c). ROC curves exhibited 1- , 2-, 3-, and 5-year AUC in ICGC cohort were 0.568, 0.609, 0.626 and 0.683, while 1- , 2-, 3-, and 5-year AUC in GSE26712 cohort were 0.586, 0.597, 0.636 and 0.682 (Fig. 3d). Based on TCGA-OC and ICGC-OC cohorts, the univariate and multivariate Cox regression formula were utilized to evaluate of the prognostic capability of AIRS. And the results validated AIRs as a potent prognostic indicator for OC patients (HR>1, p<0.05) (Fig. 4a, b). Thus, we concluded that AIRS was a powerful independent prognostic factor in OC. ROC curves showed that compared with other clinicopathological parameters, AIRS had a superior predictive ability for the OS of OC patients (Fig. 4c). The heatmap of 11 constituent AIRGs of AIRS, pathological characteristics of TCGA OC dataset were shown, and we found that AIRS were tightly correlated with in the survival status of OC patients (Fig. 4d). It is known that the mesenchymal subtype of OC suffered from the shortest survival time among ovarian tumor subtypes (Lv et al. 2020). Interestingly, the proportion of mesenchymal patients in AIRS-high groups was obviously higher than that in AIRS-low groups (Fig. 4e), indicating the AIRS were correlated with malignant mesenchymal subtype in the positive direction. Additionally, 8 genes were positively associated with the AIRS, whereas 3 genes exhibited the opposite trend (Fig. 4f). A nomograph featuring 5 clinicopathologic factors was established to further explore the clinical value of the AIRS (Fig. 5a). The ROC curves and calibration plot of 1-, 2-, 3-, and 5-year OS revealed the nomogram forecasted prognosis with high accuracy and reliability (Fig. 5b, c). 1- , 2-, 3-, and 5-year DCAs of the AIRS and other clinical variables unveiled that the prognostic model was capable of predicting OS of OC patients effectively (Fig. 5d).
Identification of EML2 expression and functions in OC
Considering that univariate Cox regression algorithm uncovered EML2 was a risk factor for the survival time of OC patients. The survival curve was devoted to unveil that high EML2 expression is unfavorable in OC patients (Fig. 6a). Interestingly, EML2 functioned as a prognostic risk gene in multiple cancer types, including OC, uterine carcinosarcoma (UCS), uterine corpus endometrial carcinoma (UCEC) and kidney renal clear cell carcinoma (KIRC) confirmed by the outcome derived from TISCH database (Fig. 6b). As shown in Fig. 8C, immunohistochemistry staining validated that the expression level of EML2 protein were obviously elevated in epithelial ovarian neoplasms in comparison with normal ovarian tissues (Fig. 6c). 7 dataset including OV_EMTAB8107, OV_GSE118828, OV_GSE130000, OV_GSE147082, OV_GSE154600, and OV_GSE158722 from the TISCH were applied to evaluate the correlation between the expression of EML2 and TME (Fig. 6d). The results reflected that the higher expression levels of EML2 were detected in malignant cells compared to immune cells and stromal cells (Fig. 6e). Gene set enrichment analysis (GSEA) was concomitantly carried out to identified the EML2 functions. The cases with high expression of EML2 were distinctly linked to cancer pathways, containing acute myeloid leukemia (AML), non-small cell lung cancer (NSCLC) and pancreatic cancer, activated VEGF, MAPK, mTOR, insulin, GnRH, neurotrophin signaling axis and immune-related signaling such as Fc epsilon R1-mediated signaling. Furthermore, some cancer-promoting biological processes enriched in high-EML2 group including cell motility, endocytosis, cell-matrix adhesions and actin cytoskeleton modifications were described (Fig. 7).
Correlation of AIRS and tumor microenvironment characteristics
The “CIBERSORT” was applied to elucidate the association of immune infiltration composition and AIRS. Significantly, the proportion of different immunocyte varied between different subtypes (Fig. 8a). The infiltration of M1 macrophages, activated T follicular helper cells (Tfh), activated CD4 memory T cells, activated dendritic cells (DCs) were greater in low-risk subgroup. Nonetheless, M2 macrophages and activated Mast cells exhibited greater activity in high-risk subtypes (Fig. 8b). What is more, the selected genes in the risk model were closely related to immunocyte infiltrations, and ISG20 presented the highest correlation coefficient with M1 macrophages (Fig. 8c). With a decreasing AIRS, the abundance of M1 macrophages, activated Tfh cells, activated CD4 memory T cells and activated DCs were upregulated. The proportion of M2 macrophages and activated mast cells (MCs) were positively relevant to AIRS (Fig. 8d). Besides, the correlation between high- and low-risk set in stromal score, immune score, ESTIMATE score was obtained by ESTIMATE method, pointing to that AIRS was correlated with stromal score in the positive direction statistically (Fig. 8e). In addition, we discovered that seven m6A regulatory molecules were significantly different between various subgroups. Specially, YTHDF3, YTHDC2, ZC3H13, WTAP, YTHDF1 and FTO was overexpressed in high risk set, but HNRNPC showed the opposite performance (Fig. 8f).
Genetic mutation alterations in OC
Mutation analysis was performed to compare the somatic mutations in high- and low- AIRS sets of OC patients. The top 20 genes in two sets displayed various mutation frequencies. In high-risk populations, TP53 (96%), TTN (25%), CSMD3 (13%), FLG (9%) and RYR2 (9%) had the highest mutation rates (Fig. 9a). And the 5 most frequently mutated genes in subtype with low AIRS were TP53 (94%), TTN (28%), CSMD3 (12%), MUC16 (10%) and FLG2 (9%) (Fig. 9b). As shown in Fig. 9c and d, the level of TMB was upregulated in low-risk subtype, implicating that low-risk set has more mutation genes. Several researches had revealed that tumor patients who presented elevated value of TMB possibly profited from immunotherapy(L. Liu et al. 2019). Thus, low-risk subtype who underwent immunotherapy may displayed good therapy outcomes. Moreover, there were differentially mutation genes between the two groups including DSCAML1, GRIN2A, CSMD1, MYO18B, NCOA3, TRIO, NLRP3 and TACC2 examined in the Fig. 9e.
Correlation analysis of AIRS and therapeutic applications
The value of AIRS to assess the immunotherapy efficacy was identified by TIDE (Fig. 10a, b). The relationship between AIRS and TIDE score was illustrated in the Fig. 10c. High-risk set presented upregulated TIDE, which was linked to tumor immune escape (Chen et al. 2021). Thus, low-risk populations’ response to immune checkpoint inhibitors (ICI) may be better. Next, an immunotherapy dataset, IMvigor210, was selected to further assess the influence of AIRS on forecasting the immunotherapeutic response. The survival curve showed that the patients in the immunotherapy cohort with high-AIRS exhibited dismal clinical prognosis, suggesting that these populations were incapable of benefitting from anti-PD-L1 therapy (Fig. 10d). Apparently, the proportion of in the stable disease (SD)/ progressive disease (PD) subtype was much higher in the high-risk populations (Fig. 10e, f). There was strong association between AIRS and immunotherapeutic results significantly (Fig. 10g). Additionally, the high-AIRS patients with low tumor neoantigen burden (TNB) and low TMB presented the poorest prognosis, respectively (Fig. 10h, i). These supported that the populations with low-AIRS may have promising ICI response and prolonged OS versus high-AIRS group. Finally, we further studied the important role of AIRS on forecasting therapeutic response in OC cases. Cisplatin and Irinotecan were more sensitive to the patients with high-AIRS (Fig. 10j). Cediranib and Foretinib exhibited greater response in OC patients with low-AIRS (Fig. 10k).