A Platinum Resistance-Related lncRNA Signature for Risk Classification and Prognosis Prediction in Patients with Serous Ovarian Cancer

Accurate risk stratification for patients with serous ovarian cancer (SOC) is pivotal for treatment decisions. In this study, we identified a lncRNA-based signature for predicting platinum resistance and prognosis stratification for SOC patients. We analyzed the RNA-sequencing data and the relevant clinical information of 295 SOC samples obtained from The Cancer Genome Atlas (TCGA) database and 180 normal ovarian tissues from the Genotype-Tissue Expression (GTEx) database. A total of 284 differentially expressed lncRNAs were screened out between platinum-sensitive and platinum-resistant groups by univariate Cox regression analysis. Then, a signature consisting of eight prognostic lncRNAs was used to construct a lncRNA score model by least absolute shrinkage and selection operator (LASSO) regression and multivariate Cox regression analysis. The ROC analysis showed that this signature had a good predictive performance for chemotherapy response in the training set (AUC = 0.8524) and the testing and whole sets with 0.8142 and 0.8393 of AUC, respectively. Dichotomized by the risk score of lncRNAs (lncScore), the high-risk patients showed significantly shorter progression-free survival (PFS) and overall survival (OS). Based on the final Cox model, a nomogram comprising the 8-lncRNA signature and 3 clinicopathological risk factors was then established for clinical application to predict the 1, 2, and 3-year PFS of SOC patients. The gene set enrichment analysis (GSEA) revealed that genes in the high-risk group were active in ATP synthesis, coupled electron transport, and mitochondrial respiratory chain complex assembly. Overall, our findings demonstrated the potential clinical significance of the 8-lncRNA-based classifier as a novel biomarker for outcome prediction and therapy decisions in SOC patients with platinum treatment.


Introduction
Ovarian cancer is the most lethal gynecological malignancy with a 5-year survival rate of 46% in women around the world [1]. According to the global cancer statistics 2020, there were an estimated 313,959 new cases and 207,252 cancer deaths worldwide in 2020, representing the 8th most commonly occurring cancer and the leading cause of cancerrelated death among women worldwide. As the most prevalent histologic subtype, SOC accounts for over 70% of all ovarian cancers and is composed of high-grade and lowgrade SOC. As for the treatment of ovarian cancer, platinum-based chemotherapy has been introduced to clinical practice as the backbone drug for more than 4 decades. Platinum, that directly interacts with DNA of cancer cells after hydration, has been employed as a therapeutic standard-of-care for ovarian cancer since the late 1970s [2]. Although 70% of ovarian cancer patients initially respond favorably to the frst application of platinum-based chemotherapy, over 80% of them eventually relapse at some stage [3]. Chemoresistance remains the major obstacle to improving prognosis [4]. Te outcome of patients with platinum resistance is generally poor, with a low response to further chemotherapy and a median survival of less than 12 months. Shortage of ideal platinum resistance-related biomarkers for risk classifcation and prognosis prediction is one of the major challenges in clinics.
Long noncoding RNA (lncRNA) is a class of RNA longer than 200 nucleotides that do not encode proteins and have no typical open reading frame [5]. It is now clear that lncRNAs participate in pathological processes by acting as enhancer RNAs (eRNAs), recruiting chromatin-modifying complexes, regulating the activity of transcription factors, and afecting the spatial conformation of chromosomes [6]. Accumulating evidence has highlighted that dysregulation of lncRNAs was involved in the chemoresistance and progression of ovarian cancer [7][8][9], and lncRNAs could represent novel biomarkers for prognostic signature in ovarian cancer [10,11]. However, studies are limited regarding the association of lncRNAs signature and platinum chemoresistance in SOC. From this perspective, there should be some lncRNAs that could serve as risk stratifcation biomarkers for platinum resistance in ovarian cancer. In this study, we analyzed the sequencing datasets of SOC from TCGA and GTEx databases and identifed a prognostic risk model consisting of eight platinum sensitivity-related lncRNAs. We also explored the potential molecular mechanisms and critical pathways that were associated with drug resistance in SOC induced by dysregulated lncRNAs.

Acquisition of Gene Expression and Clinical Data.
Te lncRNA sequencing profles of 180 normal ovarian tissues from the GTEx database (https://www.gtexportal.org/home/ datasets) and the normalized lncRNA sequencing and mRNA sequencing of 295 SOC samples from the TCGA dataset (https://portal.gdc.cancer.gov/repository? facetTab=cases) were downloaded. Te normalized gene expression data of the TCGA-SOC database after log 2transformed were used for analysis. Samples from the TCGA database were divided randomly into a training set (n = 179) and an internal validation set (n = 116) at a ratio of 3 : 2 by the "caret" package Version 6.0-86 in R software (Version 4.0.2). Te corresponding clinical data such as gender, age, pathologic grade, tumor-node-metastasis (TNM) stage, treatment type, treatment or therapy, prior malignancy, and survival data were also obtained from the TCGA.
Our inclusion criteria for the study cases were as follows: the primary tumor located in the ovary; histopathologically diagnosed with SOC; data for primary outcome evaluation and complete follow-up were available; the histological grade and clinical stage were clear; and all patients were treated with cisplatin. According to version 2.2020, National Comprehensive Cancer Network (NCCN) guidelines in ovarian cancer [12], 295 SOC cases extracted from the TCGA were divided into platinum-sensitive groups (nonrelapse and platinum-sensitive) and platinum-resistant groups (platinum-resistant and platinum-refractory). Nonrelapse means that no recurrence occurred in patients from the primary treatment until the end of follow-up. Patients who relapsed 6 months or more after initial chemotherapy were termed platinum-sensitive [13]. Tose whose disease recurred in less than 6 months were termed platinum-resistant, whereas patients who were defned as platinum-refractory for cancer progression after two consecutive chemotherapy regimens without ever sustaining a clinical beneft were also classifed in the platinumresistant group [14]. Te clinical beneft rate (CBR) is defned as without recurrence of ovarian cancer within 6 months in patients with complete remission (CR), partial remission (PR), or stable disease (SD) after primary therapy.

Analysis of Diferentially Expressed lncRNAs.
We integrated the expression matrices from the TCGA and GETx portal and identifed the diferentially expressed lncRNAs (DE-lncRNA) between SOC and normal tissues (marked as "TNlncRNAs") using the R package "DESeq2" package (version 1.28.1) [15] and the DE-lncRNAs between the platinum-sensitive and platinum-resistant groups (marked as "RSlncRNAs") using the R package "edgeR" (version 3.30.3) [16]. Te R package "ggplot2" [17] (version 4.0.2) was then used to display the candidate DE-lncRNAs. Te threshold for DE-lncRNA selection was based on a P value < 0.05 and |log 2 (FC)| > 1. Ten, we extracted the platinum-related lncRNAs based on the intersection set of TNlncRNAs and RSlncRNAs for further analysis using the "Venn" package (version 1.9).

Establishment and Validation of the Prognostic Gene
Signature. To identify a multi-lncRNA signature with a good predictive performance for prognosis, DE-lncRNAs that were correlated with patients' PFS in the training set were screened using univariate Cox regression analysis. Te lncRNAs with a Wald P value < 0.01 or |HR − 1| ≥ 0.35 were retained by the "survival" R package (version 4.0.2). Te prognostic lncRNAs with nonzero coefcients were further screened by combining least absolute shrinkage and selection operator (LASSO) regression analysis with 10foldcross-validation in the training set using the "glmnet" R package (Version 4.0.2).

Multi-lncRNA Signature Model Establishment and
Validation. Te risk coefcients (coefs) for the candidate lncRNAs were subsequently calculated using a multivariate Cox proportional hazards regression analysis. We calculated the risk score of lncRNAs (lncScore) by the following formula: where Exp lncRNAi indicates the expression level of lncRNA in patients with SOC. Ten, the receiver operating characteristic (ROC) curve was drawn to assess the sensitivity and specifcity of the hub 2 Journal of Oncology lncRNAs for chemotherapy response in the training set by calculating the area under the curve (AUC) with the "heatmap" package [15]. Te cutof value of the risk score for classifying patients into the high-risk group and low-risk group was determined when the ROC curve reached the optimum sensitivity and specifcity among the training cohort. Ten, we validated its classifcation properties for chemotherapy response in the testing cohort and the whole cohort. Kaplan-Meier curve was performed to evaluate the survival time of the high-risk and low-risk groups, and then, the log-rank method was used to verify the results.

Independent Prognosis Analysis and Nomogram
Construction. To verify the efcacy of the candidate lncRNA signature in predicting the prognosis and clinicopathological features such as age at diagnosis, clinical stage, pathological grade, tumor residual disease, and venous invasion, univariate and multivariable Cox regression analyses were used to select independent prognostic factors. Te cox regression coefcients were introduced to generate nomograms with the "rms" package version 6.0-0 [18] in R software (version 4.0.2). Te survival probability estimated by Kaplan-Meier analysis was used to correct the nomogram [19].

Gene Set Enrichment Analysis (GSEA).
Te 295 SOC samples were divided into the high-risk and low-risk groups according to the lncScore threshold. Te "edgeR" package was used to analyze the DE-mRNA expression profle between the two groups with FDR <0.05 and |log 2 FC| > 1. Te "clusterProfler" package (version 3.16.1) in R software (version 4.0.2) was employed to perform GSEA [20] for the signifcantly altered gene ontology (GO) function. P < 0.01 was considered as the cutof criterion. Finally, the "enrichplot" package (version 1.8.1) in R software (version 4.0.2) was used to demonstrate the enrichment functions that were statistically signifcant.

Te Study Flowchart and Patient Clinicopathological
Features. A detailed data processing fowchart is shown in Figure 1. Te 304 SOC samples with primary therapy outcome were retained. Tese samples consisted of 212 cases in the platinum-sensitive group, which included 52 cases of nonrecurrence and 160 cases of platinum-sensitive, with a median age of 58 years old (IQR, interquartile range: 51-67) and 83 cases in the platinum-resistant group which included 35 cases of platinum-resistant and 48 cases of refractory, with a median age of 59 years old (IQR: 50-67). Te clinicopathological characteristics of the patients in the training, testing, and whole cohorts are summarized in Table 1.  Table S1). By using a univariate Cox regression analysis, eight lncRNAs were found to be signifcantly associated with PFS (P value < 0.01 and |HR-1| > 0.35) and were defned as the prognostic lncRNAs in the training set (Table 2). Te prognostic signature was composed of the following lncRNAs:

Construction of the Eight lncRNAs-Based Prognostic
Model. We fnally identifed eight markers with LASSO coefcient >0.1 and developed a risk score formula by a linear combination of the expression values of the eight lncRNAs weighted by multivariate Cox regression coefcients (Figures 2(d) and 2(e); Table 2). Te risk score for each patient was calculated based on the following formula: (2) ROC curves were plotted to evaluate the predictive power based on specifcity and sensitivity. Te AUC of ROC for the prognostic signature is 0.8534, which shows excellent performance for the efcacy prediction of platinum-based chemotherapy. Te cutof value of the risk score for classifying patients into the high-risk group and low-risk group was determined when the ROC curve reached the optimum sensitivity (75.8%) and specifcity (82.4%) in the training cohort ( Figure 3). Te distribution of the risk score was plotted along with the corresponding survival outcome (Figure 3(b1)). Te expression of the eight lncRNAs in the training set is shown in Figure 3 showed that patients in the high-risk group had worse PFS than those in the low-risk group in the training cohort (HR = 1.81; 95% CI: 1.28-2.54; P = 0.0006; Figure 3(d1)).

Validation of the Eight lncRNAs-Based Prognostic Model.
To verify the predictive performance of this 8-lncRNA model, we divided patients from the testing and whole cohorts into high-risk and low-risk groups according to the cutof value determined in the training cohort. Te AUC of patients in the testing and the whole sets was 0.8023 and 0.8344, respectively (Figures 3(a2) and 3(a3)), implying that the lncScore of this model has a good performance in predicting the efcacy of platinum-based chemotherapy.
Te distribution of risk scores and survival status of patients in the testing and whole cohorts are shown in       (a1), testing cohort (a2), and the whole cohort (a3). (b) PFS status scatter plots and risk score distribution for patients in the training set (b1), testing cohort (b2), and the whole cohort (b3). Red and blue indicate high-risk and low-risk scores, respectively. (c) A heatmap for the 8-lncRNA signature in the training set (c1), testing cohort (c2), and the whole cohort (c3), and red and blue indicate high and low expression, respectively. Kaplan-Meier curves of PFS (d) and OS (e) for the high-and low-risk patients grouped by the 8-lncRNAs signature in the training set (d1, e1), testing cohort (d2, e2), and the whole cohort (d3, e3). Te time-dependent AUC analysis of the individual lncRNA and lncScore for predicting PFS (f ) and OS (g) in the whole set.
To develop a clinically applicable tool that can provide individualized estimates of 1, 2 or 3-year PFS and 3 or 5-year OS, a nomogram was subsequently established based on the fnal Cox model for PFS and OS. As shown in Figure 5 (for PFS) and Supplementary Figure S1 (for OS), the lncScore takes the biggest weight for prediction, followed by age at diagnosis, clinical stage, and the tumor residual disease. Most important, the calibration plots of the nomogram predicting PFS and OS performed well with the ideal model ( Figures 5(b)-5(d), Figures S1(b) and S1(c)).
Likewise, the lncScore showed the strongest predictive power in distinguishing platinum-sensitive and platinumresistance patients with SOC (AUC � 0.8295), whereas the AUC-ROC for postoperative tumor residual disease, clinical stage, age at diagnosis, and pathological grade ranged from 0.5031 to 0.6555 ( Figure 5(d)).

Functional Analysis of the Prognostic lncRNA Model.
GSEA was carried out to explore the biological efects of this lncRNA model, and our results suggested that the high score of lncRNAs in the model showed signifcant enrichment in a crowd of biological process terms, mainly involved in ATP synthesis coupled electron transport, mitochondrial respiratory chain complex assembly, nuclear-transcribed mRNA catabolic process (nonsense-mediated), protein targeting to ER, and translational initiation ( Figure 6).

Discussion
Te existing FIGO (International Federation of Gynecology and Obstetrics) and AJCC (American Joint Committee on Cancer) cancer staging systems, as well as pathological grade, are the major factors afecting the therapeutic regimen decision, but they could not give enough information in predicting platinum resistance for patients with SOC. In recent years, dysregulated lncRNAs have been documented to be involved in tumor progression and have a great potential in the diagnosis and prognosis of ovarian cancer as novel independent molecular biomarkers [21]. Tere is growing evidence that lncRNAs could serve as potential biomarkers for the prediction of platinum-based chemoresistance [22][23][24]. Several studies have shown the potential of a lncRNA expression signature for predicting the chemotherapeutic sensitivity of ovarian cancer while narrowly restraining its cohort in advanced stages or high-grade SOC [25,26]. To date, the lncRNA expression profles-based prognostic signature for the prediction of platinum sensitivity in SOC patients have not been developed.
In this study, a comprehensive analysis of lncRNA expression profles was conducted to discriminate platinum-based chemotherapeutic outcomes in patients with SOC. Eight lncRNAs with predictive signatures of platinum-based chemotherapeutic sensitivity were identifed in a large number of SOC patients from the TCGA. Among these lncRNAs, an increased expression level of AL162713.1 was associated with a poor PFS, and the other 7 lncRNAs were positively correlated with PFS time. We validated the predictive power of this 8-lncRNA signature on an internal validation set of 116 TCGA patients as well as on the entire TCGA cohort. Te AUC value of the prognostic signature was 0.8295, which was similar to the previous lncRNA panel for platinum-resistance prediction in patients with highgrade SOC developed by Song et al. [26], whereas higher than those developed are presented in two published studies in advanced stage and high-grade SOC patients [21,25]. In stratifcation of the subjects according to the cutof value of lncScore, the survival curves showed a high level of consistency among the three datasets, which indicates that SOC patients with higher scores have a shorter duration of PFS and OS than patients in the lower score group. Moreover, the 8 core lncRNAs in this prognostic panel were found to be more reliable and accurate than a single lncRNA alone. Tis fnding is consistent with the common perception that a multiple noncoding RNA panel is generally more precise and robust in predicting the outcomes of patients with cancer [27].
With further validation, the subsequent multivariate Cox regression analysis showed that the prognostic signature value of the lncScore functioned as an independent risk factor for poor prognosis in SOC patients besides age at diagnosis and tumor residual disease >20 mm. Based on the fnal Cox model, a nomogram was established for clinical use by integrating the 8-lncRNA signature and three clinicopathological risk factors for the clinical assessment of the 3and 5-year PFS and OS probabilities for SOC patients. And from the nomogram, we can see that the patient with a higher lncScore, or with residual tissue of 1-10 mm or >20 mm, or higher clinical staging, would have a higher total of points and get a higher probability of disease progression in 1, 2, or 3 years postsurgery. Notably, the AUC-ROC value of the lncScore in the nomogram showed excellent prognostic value in predicting the efcacy of platinum-based chemotherapy compared with that of other existing risk factors, including postoperative tumor residual disease, clinical stage, and age at diagnosis, implying that this 8-lncRNA signature could be a clinically applicable tool for individualized estimation and guidance of clinical decisionmaking regarding the usage of platinum-based chemotherapy.
Among the eight lncRNAs with this signature, only MIR137HG has been illustrated in some literature. MIR137HG (gene ID: 400765) is the host gene for MIR137. It was demonstrated that recruiting DNA methyltransferase DNMT3a to the promoter of MIR137 in colorectal cancer cells could lead to suppression of MIR137 expression [28]. Earlier studies focused on its mutation involved in the development of liver cancer [29]. Recently, MIR137HG was introduced to be one of the lncRNA signatures for predicting the survival of patients with hepatocellular carcinoma [30].
Additionally, MIR137HG was identifed as a core lncRNA included in the tumorigenesis-related ceRNA network and as a potential prognostic biomarker for laryngeal squamous cell carcinoma and muscle-invasive bladder cancer [31,32]. It is with regret that existing research rarely reported the biological functions of MIR137HG in cancer. Lyu et al. performed GO and KEGG enrichment analyses and revealed that MIR137HG may function as an oncogene in muscleinvasive bladder cancer by regulating epithelial cell diferentiation, cytokine production, the PPAR signaling pathway, and TNF signaling pathway [31]. In the study by Zhong et al., MIR137HG was identifed as an immune-related lncRNA and used in a prognostic lncRNA signature in neuroblastoma, which was mainly enriched in cancerrelated pathways and immune-related pathways [33]. Tese results are in line with what was found in this study, which suggests that MIR137HG might be an lncRNA that afects the prognosis in pan-cancers.
Afterward, an in-depth study by GSEA enrichment analysis for the characteristics and biological behavior of the constructed lncRNA model was performed to understand the potential mechanisms involved in platinum resistance in SOC. Earlier evidence indicated that intracellular ATPdependent processes contributed to cisplatin resistance of tumor cells, and the elevated ATP level appeared to be a consequence of enhanced mitochondrial ATP production [34]. Meanwhile, ATP-binding cassette transporters were implicated in the infux or efux of platinum-based chemotherapeutic agents [35]. Consistent with the literature, the current research found that the high-risk group of chemoresistance was more active in ATP synthesis coupled electron transport, further supporting this signature's involvement in platinum resistance.
Mitochondria are known to play a central role in regulating cellular metabolism and producing adenosine triphosphate (ATP). Interestingly, mitochondrial respiratory chain complex assembly was also found to be active in the high-risk group. It was documented that mitochondrial defects and the dysfunctions of energy production contributed to the resistance to platinum drugs in cancers [36], ovarian cancer included [37]. Tis fnding is consistent with that of Dong et al., who indicated that mitochondrial dysfunction was associated with platinum resistance [38].
In addition, GSEA results showed that some diverse functions, like NMD, were clustered in the high-risk group. NMD is a conserved mRNA surveillance mechanism that triggers the degradation of aberrant mRNAs harboring premature termination codons. In some cases, tumors exploited NMD by positively selecting for nonsense mutations to downregulate tumor suppressor genes, as has been shown for BRCA1 and BRCA2, which were associated with inherited susceptibility and platinum resistance in ovarian cancer [39][40][41][42][43]. However, the fndings of the current study do not support these two following studies. Although BRCA1-Δ11q alternative splice isoform has been shown to contribute to cisplatin-resistance in ovarian cancer, the alternative splicing rate or BRCA1-Δ11q level was not altered by frameshift mutations in exon 11-induced NMD or its containing-transcripts.
Another recent research demonstrated that inhibition of NMD induced by stress could lead to upregulation of the cystine/glutamate exchanger SLC7A11 [44], which has been associated with platinum resistance in ovarian cancer [45]. Accordingly, this result needs to be interpreted with caution. Te present results indicated that the lncScore model based on the eight lncRNAs can efectively stratify the patients with SOC into platinum-sensitive and -resistant groups, which updated the current prognostic risk prediction model for platinum resistance. Almost inevitably, there were some shortcomings in the current study. First, the cases were collected only from the TCGA database without other independent datasets, which might be skewed by selection bias. Furthermore, this study employed a retrospective design, so it requires prospective cohorts in clinical trials to further address this issue. And third, the biological function of the lncRNAs included in this 8-lncRNA signature has not been fully characterized before, and further functional studies, both in vitro and in vivo, are required.
In summary, by analyzing the available transcriptome sequencing data from the public data repositories, our study presents a panel of platinum sensitivity-related 8-lncRNA signatures with superior accuracy for prognosis and good predictive performance in SOC patients with platinum treatment. Furthermore, a nomogram combining the panel of these 8 lncRNAs with clinicopathologic characteristics showed excellent calibration consistency, improving the model's practicality for clinical applications. In view of the contribution to drug resistance-related biological processes, 8 lncRNAs could be potential therapeutic targets.

Data Availability
Publicly available datasets were analyzed in this study. Te data used to support the fndings of this study are available at https://portal.gdc.cancer.gov/.

Conflicts of Interest
Te authors declare that they have no conficts of interest. Figure S1: Construction and validation of the lncScoreclinicopathologic nomogram for survival prediction in SOC patients. Table S1: List of the 220 platinum-related lncRNAs based on the intersection set of TNlncRNAs and RSlncRNAs. (Supplementary Materials)