Background

Ovarian cancer (OC) is the main cause of gynecological cancer-related death worldwide [1]. Because of the lack of clinical manifestations in the early stage, 70% of patients are diagnosed in the middle or late stages [1]. It is also difficult to perform radical surgery, leading to a high OC mortality rate of OC [1]. OC is characterized by obvious tissue heterogeneity with genomic characteristics, which shows that it easily develops resistance to chemotherapy and leads to a high tumor recurrence rate [1]. Therefore, improving the prognosis of OC patients by exploring novel diagnoses and therapies is an urgent problem to address.

Studies have shown that genome-level epigenetic modifications, such as DNA methylation, histone modification and RNA editing, are crucial to tumorigenesis. N6-methyladenosine (m6A) modification was first detected in the poly (A) RNA component in 1974 and is considered to be the most abundant posttranscriptional modification of mRNA [2]. m6A modification is mainly associated with three types of proteases that serve as regulators, namely, methyltransferases (writers), demethylases (erasers), and signal transducers (readers) [2]. Numerous studies have shown that different levels of m6A regulators are associated with the self-renewal of tumor stem cells, the proliferation of cancer cells, and the sensitivity to chemotherapy [2]. A previous study showed that high-frequency genetic alterations of m6A RNA methylation regulators are crucial for the progression [3] and that m6A modification contributes to PARPi resistance in OC [4]. In OC, YTHDF1 (reader) facilitates the expression of EIF3C [5] and the stem cell-like phenotype of cisplatin resistance [6] in a m6A-dependent manner, thereby strengthening tumorigenesis and metastasis. ALKBH5 (eraser) [7] and YTHDF2 (reader) [8, 9] were shown to regulate the carcinogenesis of OC by modulating m6A levels. IGF2BP1 (reader) was demonstrated to augment the translation of serum response factor through m6A modification [10]. METTL3 (writer) can regulate m6A methylation and thus regulate the malignancy of OC [11,3a, Additional file 1: Table S1). Interestingly, p53 may regulate m6A factors and LI-m6As through trans interactions (p53-IGF2BP1-AC010745.4/CACNA1G-AS1/ERVH48-1/HOXA-AS3/PTENP1-AS). On the basis of ten lncRNA-m6A pairs, we further found four of them included in the lncRNA-miRNA-mRNA relationship pairs and the top one lncRNA-m6A pair with most miRNAs was UBA6-AS1-YTHDC1(Fig. 3b). These findings may provide insights into the regulatory mechanisms of lncRNAs correlated with m6A-regulatory genes.

Fig. 3
figure 3

Network based on trans interactions and ceRNA. a Trans interaction for lncRNA-TF-mRNA relationship pairs. b ceRNA for lncRNA-miRNA-mRNA relationship pairs. Blue arrowheads: TFs; red rhombus: LI-m6As; yellow ellipses: m6A regulators; blue rectangles: miRNAs

Full size image

Construction and validation the prognostic signature

A LI-m6A prognostic signature was constructed according to the LASSO Cox analysis of 10 significant prognostic LI-m6As. The λ selection diagram is shown in Fig. 4a, b. λ between λmin and λ1se were considered appropriate. The model constructed by λ1se was the simplest, that was, it used a small number of genes, while λmin had a higher accuracy rate and used a larger number of genes. The λmin was selected to build the model for accuracy in our study. Patients in the training and validation cohorts were divided into low- or high-risk subgroups based on the median of RSs. The K–M survival curves of risk groups revealed that OS in the high-risk group was markedly lower than that in the low-risk group in both cohorts (Fig. 5a, b). The ROC curves proved that the risk model signature based on four m6A-RLs was accurate in predicting the OS of OC patients (Fig. 5c, d).

Fig. 4
figure 4

a LASSO Cox analysis of 10 significant prognostic LI-m6As. b λ selection diagram. The two dotted lines indicated two particular values of λ. The left side was λmin and the right side was λ1se. The λmin was selected to build the model for accuracy in our study

Full size image
Fig. 5
figure 5

Construction and validation of the prognostic signature. a, b The K-M survival curves of two risk groups based on the risk model in the training cohort (a) and validation cohort (d). c, d The ROC curves for predicting 1-, 3-, and 5-year survival in in the training cohort (c) and validation cohort (d)

Full size image

Four LI-m6As with corresponding coefficients and HR values (Table 2), namely, AC010894.3, ACAP2-IT1, CACNA1G-AS1, and UBA6-AS1, were generated to calculate RSs.. A heatmap of the associations between the expression levels of the four LI-m6As and RSs showed that the expression of ACAP2-IT1 and CACNA1G-AS1 increased with increasing RS, whereas the expression of AC010894.3 and UBA6-AS1 decreased with increasing RS in both the training and validation cohorts (Fig. 6a, b). The K–M survival curves confirmed that lower expression of ACAP2-IT1 and CACNA1G-AS1 and higher expression of AC010894.3 and UBA6-AS1 were associated with better OS of OC patients (Fig. 6c-f). Hence, ACAP2-IT1 and CACNA1G-AS1 were risk factors, while AC010894.3 and UBA6-AS1 were protective factors.

Table 2 The coefficient and HR value of the four m6A-related lncRNAs
Full size table
Fig. 6
figure 6

Prognostic analysis of the four LI-m6As. a, b Heatmap of the expression levels of the four m6A-RLs with RSs in the training (a) and validation cohort (b). c-f The K-M survival curves of LI-m6A AC010894.3 (c), ACAP2-IT1 (d), CACNA1G-AS1 (e), and UBA6-AS1 (f)

Full size image

Nomogram model construction and visualization

Univariate and multivariate Cox regression analyses of clinicopathology and the LI-m6A risk model showed that the signature was an independent risk factor for OC patients in the training and validation cohorts (Table 3). A nomogram was further constructed based on risk score, age, grade, and Figo state to predict the prognosis of OC (Fig. 7a, b). A calibration diagram was also used to verify the prediction ability of risk score for 1-, 3- and 5-year of survival (Fig. 7c–h). In the calibration plots, the blue solid line was the prediction for survival, and the diagonal dotted line was the actual survival. The closer the solid line was to the dotted line, the better the prediction ability was.

Table 3 Univaraite and multivariate Cox analysis of risk score and clinicopathology
Full size table
Fig. 7
figure 7

Nomogram model construction and visualization. a, b The Nomogram model based on risk score and clinical features for the training (a) and validation (b) cohort. c-h The calibration plots of the nomogram for predicting the probability of OS at 1 (c, f), 3 (d, g), and 5 (e, h) years in the training cohort (c–e) and validation cohort (f–h)

Full size image

Clinicopathology analysis

We carried out clinicopathological analysis to evaluate the prognostic capacity of the LI-m6As prognostic signature. Several clinicopathological characteristics, such as WHO grade I, grade IV and FIGO stage I, were excluded due to the small sample size (< 10 patients). The results showed that the OS of the clinicopathology subgroups, such as Grade II, Grade III, age <  = 55 or > 55 years old, Stage III, Stage IV, platinum sensitivity and platinum resistance, were significantly different in the low- or high-risk group (Fig. 8a-h). Potentially, the p value of the Stage II subgroup was indistinctive because of insufficient sample size (Additional file 2: Fig. S1).

Fig. 8
figure 8

The K-M survival curves of several clinicopathological characteristics based on the risk model. a Grade II. b Grade III. c Age <  = 55. d Age > 55. e Stage III. f Stage IV. g Platinum sensitivity. h Platinum resistance

Full size image

Biofunctional analysis of the risk groups

A total of 1581 DEGs between the risk subgroups were identified. The GO biological process analysis found that DEGs were mainly enriched in migration and proliferation, such as epithelial cell migration and vascular smooth muscle cell proliferation (Fig. 9a). Moreover, the performed KEGG pathway analysis revealed that the DEGs were mainly enriched in microRNAs in cancer and taste transduction (Fig. 9b). GSEA of GO indicated that gene sets enriched in ‘negative regulation of gene expression’, ‘negative regulation of metabolic process’, and ‘posttranscriptional regulation of gene expression’ were highly expressed in the high-risk group (Fig. 9c). GSEA of KEGG showed that genes enriched in ‘MicroRNAs in cancer’ were highly expressed the in high-risk group, while genes enriched in ‘Metabolic pathways’ were expressed at low levels in the high-risk group (Fig. 9d). The enriched GO terms and KEGG pathways were further annotated from “Metascape” website (Fig. 9e). The key module of the PPI network discerned by MCODE were shown in Fig. 9f. Seven hub genes were included in the interaction network.

Fig. 9
figure 9

Functional analysis. a, b Significantly enriched GO terms (a) and KEGG pathways (b) of differentially expressed genes (DEGs). The color scale represented p value and the circle size indicated count. c, d Gene set enrichment analysis (GSEA) of GO (c) and KEGG (d). The crest on the left represents the high expression of the enriched genes in the high-risk group, while the crest on the right is the opposite. e Cluster ID. Nodes with the same cluster ID are typically close to each other and the same color indicates the same cluster ID. f The key module of PPI network. Seven hub genes were included in the interaction network. Circles indicate the genes in the PPI network, and the connection indicates the potential interaction between different mRNAs

Full size image

Immune microenvironment and checkpoint

According to five algorithms (Estimate, ssGSEA, Cibersort, MCPcounter and xCell), Stromal Score, Immune Score, ESTIMATE Score, and relative infiltration abundance of immune cells and stromal cells of each sample were estimated respectively. The results revealed that a total of 59 microenvironment cells and Stromal score had significant differences between the two groups (Fig. 10a, Additional file 3: Table. S2). Among them, Neutrophils cells, Fibroblasts cells, and Endothelial cells were the top 3 immune cells significantly enriched in the high-risk group. Given the importance of immune checkpoints in cancer treatment, the expressions of seven checkpoint genes were compared. We found that HAVCR2 and SIRPA had higher expressions in the low-risk group (Fig. 10b). In Fig. 10c, OC patients in low-risk group exhibited higher TIDE scores than those in high-risk group, indicating that OC patients with lower RSs were more sensitive to ICB therapy.

Fig. 10
figure 10

Immune microenvironment and checkpoint. a Heatmap of immune microenvironment revealed that a total of 59 immune cells and stromal score had significant differences between the two risk groups. b Expression of seven immune checkpoint genes between high and low-risk group. HAVCR2 and SIRPA had higher expressions in the low-risk group. c TIDE scores in the low-risk group were higher than those in the high-risk group. Data are shown as means ± S.D. ns not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Full size image

Sensitivity of chemotherapy drug

The IC50 levels of 135 drugs in OC patients were quantified and significant differences in 87 chemotherapy drugs were found between the risk groups (Additional file 4: Table. S3). Figure 11 exhibited the results of nine commonly used chemotherapeutic agents for OC. Our data showed that the IC50 levels of Cisplatin (Fig. 11a) were significantly higher in high-risk group than that in low-risk group. Inversely, the IC50 levels of Bleomycin (Fig. 11b), Docetaxel (Fig. 11c), Gemcitabine (Fig. 11d), Vinblastine (Fig. 11e) and Vinorelbine (Fig. 11f) were significantly lower in high-risk group than that in low-risk group, indicating that the OC patients in the high-risk group were more sensitive to these drugs. However, the sensitivity of the two risk groups to Paclitaxel (Fig. 11g), Rucaparib (Fig. 11h) and Veliparib (Fig. 11i) did not show a significant difference.

Fig. 11
figure 11

Sensitivity of chemotherapy drug. a-i Difference in the estimated IC50 levels of Cisplatin (a), Bleomycin (b), Docetaxel (c), Gemcitabine (d), Vinblastine (e), Vinorelbine (f), Paclitaxel (g), Rucaparib (h) and Veliparib (i). Data are shown as means ± S.D. ns not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Full size image

In vitro assays

lncRNA CACNA1G-AS1 was filtered as the candidate molecule to perform cell function assays. Real-time qPCR analysis indicated that CACNA1G-AS1 was significantly upregulated in 30 OC tissues and three OC cell lines (Fig. 12a, b). The transfection efficiencies of three siRNAs targeting CACNA1G-AS1 were detected, which revealed that siRNA-2 presented the highest transfection efficiency in SKOV-3 cells (Fig. 12c) and that siRNA-3 presented the highest transfection efficiency in A2780 cells (Fig. 12e). The CCK-8 assay showed that the viability of SKOV-3 (Fig. 12d) and A2780 cells (Fig. 12f) was suppressed after transfection of the corresponding siRNAs.

Fig. 12
figure 12

Cell function assays. a, b Real-time qPCR analysis detecting relative expression of lncRNA CACNA1G-AS1 in 30 OC tissues (a) and three OC cell lines (b). c The transfection efficacities of three siRNAs targeting CACNA1G-AS1 in SKOV-3 cells. d CCK-8 assays were performed after CACNA1G-AS1 was inhibited in SKOV-3. e The transfection efficacities of three siRNAs targeting CACNA1G-AS1 in A2780. f CCK-8 assays were performed after CACNA1G-AS1 was inhibited in A2780. Data are shown as means ± S.D. ns not significant, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Full size image

Discussion

As the most common internal modification of mRNAs in all higher eukaryotes, m6A has been proven to be abnormally expressed in a variety of tumors, and plays an important role in the regulation of a series of malignant biological behaviors such as cell proliferation, invasion and metastasis [2]. Previous studies have reported that several lncRNAs regulate the occurrence and development of cancer via m6A modification. For example, lncRNA RP11-138 J23.1 regulated by m6A positively induced the epithelial mesenchymal transition (EMT) of colorectal cancer cells [38]. Oncogenic lncRNA XIST is inhibited by METTL14 (writer) in colorectal cancer [39]. The lncRNA DANCR and IGF2BP2 (reader) have been proven to synergistically promote the pathogenesis of pancreatic cancer [40]. It has been unveiled that LINC00942 serves as an oncogene by promoting METTL14-mediated m6A methylation in breast cancer [41]. lncRNA GAS5 negatively regulated by YTHDF3 (reader) was demonstrated to be a tumor-suppressor in colorectal cancer [42]. LINRIS was proven to promote carcinogenesis of colorectal cancer by serving as a stabilizer of IGF2BP2 (reader) [43]. LNCAROD is aberrantly expressed and has been identified as an oncogenic lncRNA due to dysregulation of m6A modification [44]. The abovementioned evidence shows that m6A is modified in the process of lncRNA biological function. In addition, lncRNA can also regulate m6A modification. These are all part of the complex regulatory network of tumors. As a new discovery in scientific research, we believe that in-depth study of the biological mechanism of mutual regulation between lncRNAs and m6A will become a hot spot for discovering prognostic markers or therapeutic targets for malignant tumors. Nonetheless, lncRNAs involved in m6A regulation in OC are still unknown.

Four LI-m6As, namely, AC010894.3, ACAP2-IT1, CACNA1G-AS1, and UBA6-AS1, were included in our identified prognostic signature. They were correlated with METTL5 (writer), RBM15 (writer), IGF2BP1 (reader), and YTHDC1 (reader), respectively. Among the four m6A regulators, METTL5 was confirmed to serve as the enzyme for the m6A modification of 18S rRNA, thereby promoting breast cancer cell growth [45]. RBM15 was identified as a regulator that binds to METTL3 and WTAP and directs these two proteins to specific RNA sites for m6A modification [2]. Studies have shown that lncRNAs participate in mediating the occurrence and progression of cancer by targeting IGFBP1, thus becoming a new therapeutic target for cancer [46]. The m6A level of lncRNA pncRNA-D was recognized by YTHDC1 [47]. For the LI-m6As we identified, a study has shown that UBA6-AS1 serves as a malignant gene in glioblastoma by competitively binding to miR-7648. Previous studies have demonstrated that lncRNA CACNA1G-AS1 enhances the multiplication capacity via the miR-2392/C1orf61 pathway in hepatocellular cancer [49]. CACNA1G-AS1 promoted the proliferation and invasiveness of colorectal cancer by inhibiting the expression of p53 [50]. CACNA1G-AS1 exerts its malignant function by HNRNPA2B1 in non-small-cell lung cancer [51]. In our research, CACNA1G-AS1 showed high expression and may regulate proliferation, which would fill the gap in the OC literature. Our findings are probably worthy of further study by subsequent researchers who concentrate on the mechanism of m6A-related lncRNAs in OC.

We further explored the differences in immune microenvironment, immune checkpoint, chemotherapy drug sensitivity between the two risk groups. Our data showed that OC patients in the low-risk group may be more sensitive to Cisplatin and ICB therapy compared to the high-risk group, while patients in the high-risk group were more sensitive to Bleomycin, Docetaxel, Gemcitabine, Vinblastine and Vinorelbine. Findings of our study uncovered potential biomarker and therapeutic target for the risk model based on the LI-m6As.

Our study still has some limitations. First, there were only 30 OC patients without OS in our cohort, hence, more time and more samples are needed for follow-up. Second, cell function assays of CACNA1G-AS1 were preliminary and need further investigation to provide a better understanding. Definitively, the number of OC samples in TCGA is very limited; hence, more independent data sets are needed to validate our identified LI-m6As.

Conclusions

We found four lncRNAs (AC010894.3, ACAP2-IT1, CACNA1G-AS1, and UBA6-AS1) involved in m6A regulation and that have significant prognostic value in OC. Furthermore, the predictive signature based on the four lncRNAs can independently predict the therapeutic value of OC patients by combining molecular signatures and clinical characteristics. Furthermore, CACNA1G-AS1 was preliminarily identified as a malignant lncRNA through in vitro experiments.