Abstract
Background/Aim: Due to the absence of screening protocols, high-grade serous ovarian cancer (HGSOC) patients are frequently diagnosed at an advanced stage, which significantly reduces the survival rate. Moreover, relapse occurs in approximately 70% of HGSOC patients after primary treatment. Predicting resistance to primary chemotherapy remains a challenge. In the research setting, transcriptomic analyses have emerged as powerful tools for predicting which HGSOC patients are likely to benefit from primary treatment. The aim of this review was to investigate the literature demonstrating the potential of transcriptomic signatures as biomarkers for assessing the risk of resistance to platinum-based chemotherapy. Materials and Methods: We conducted a three-step search process on PubMed to systematically review English-language articles published between 2020 and 2024. From the 123 articles retrieved, we included 11 articles that investigated transcriptomic signatures by RNA sequencing in tissues from chemo-sensitive and -resistant HGSOC patients. Results: We report the clinicopathological data of 727 patients in the experimental cohorts, transcriptomic signatures, and technical aspects. Finally, the review lists 15 publicly available datasets used in the included studies. Furthermore, we investigated the overlap of 167 differentially expressed genes retrieved across the various articles. Conclusion: We believe this review might offer valuable insights for further studies focusing on predicting platinum resistance and personalized treatments. In addition to discussing the latest findings and potential candidates, we highlight the challenges of validating biomarkers across studies and publicly available datasets. Transcriptomic signatures represent a potential tool for patient stratification, prognosis, and the potential adoption of long-term therapies, such as poly (ADP-ribose) polymerase inhibitors (PARPis).
Among gynecological malignancies, ovarian cancer (OC) is the leading cause of death and the fifth most common cancer type. GLOBOCAN 2022 registered 324.398 patients diagnosed with OC and 206.839 OC-related deaths (1). Approximately 550 new cases are diagnosed annually in Denmark (2), and 70-80% of patients are diagnosed in advanced stages, defined with International Federation of Gynecology and Obstetrics (FIGO) staging system as stage II-IV (3). The majority of serous carcinomas are diagnosed at stage III (51%) or stage IV (29%), with 5-year cause-specific survival rates of 42% and 26%, respectively (4). Ninety percent of OCs are of epithelial origin [epithelial ovarian cancer (EOC)] and are characterized by an extended heterogeneous profile (5).
According to their molecular, histochemical, and immunochemical profiles, five histological subtypes were identified in the 5th WHO edition (2020) (6). Seventy percent of patients are diagnosed with HGSOC, which is characterized by increased genome instability. A total of 96% of HGSOCs are mutated in TP53, 20% are mutated in BRCA1/2, and approximately 11% are homologous recombination deficient (HRD) (3, 7, 8). Pennington et al. (7) reported that the HRD-positive score in primary ovarian tumors was strongly correlated with platinum and poly-ADP ribose polymerase inhibitor (PARPi) sensitivity and thus improved survival rates. Kang et al. (9) showed the association between CCNE1 amplification and a low response rate to platinum chemotherapy and a lack of sensitivity to PARPis, which resulted in shorter survival in a cohort of 3029 HGSOC patients. Additionally, they corroborated previous evidence of the mutually exclusive occurrence of CCNE1 amplification with BRCA1/2 mutations (9). The remaining 30% of EOC subtypes, also called type I tumors, include the histological subtypes endometrial carcinoma (EC), ovarian clear cell carcinoma (OCCC), low-grade serous ovarian carcinoma (LGSOC), and mucinous ovarian carcinoma (MOC) (6). Type I carcinomas typically can arise from endometriomas, adenofibromas, or borderline tumors (10-12). First-line current treatment is primary debulking surgery (PDS) followed by six cycles of platin-based chemotherapeutic agents. Neoadjuvant chemotherapy (NACT) followed by interval debulking surgery (IDS) is recommended for OC patients when preoperative evaluation indicates that complete tumor debulking with PDS is not feasible. Patients who are platinum sensitive and have a BRCA1/2 mutation or HRD status are potential candidates for treatment with PARPis, such as monotherapy or combination therapy with bevacizumab (13). Before the clinical application of PARPis, tumor relapse occurred in approximately 70% of HGSOC patients within three years after primary treatment (14). The potential of PARPis to significantly improve disease prognosis in patients with HGSOC was demonstrated by several clinical trials. Notably, compared with a placebo, the phase III SOLO1/GOG 3004 (15) clinical trial showed a significant advantage in both progression free survival (PFS) and overall survival (OS) for patients treated with olaparib alone. Similar benefits were subsequently demonstrated in the PAOLA-1/ENGOT-ov25 (16) clinical trial, in which patients received a combination of olaparib and bevacizumab.
However, resistance to therapies remains the major challenge in cancer treatment (17). Resistance can be ascribed to the restoration of the homologous recombination repair (HRR) pathway, the mitigation of replication stress, and the deregulation of influx or efflux pumps that suppress the tumor microenvironment (TME) (18).
In addition to genomic stratification, in the research setting, transcriptomic signature-based approaches have recently exhibited their capability to predict chemotherapy response in patients with HGSOC. In the forthcoming review, we provide a comprehensive overview of the latest advances in determining transcriptomic patterns that might predict chemotherapy resistance. Our aim is to highlight limits and future perspectives to pave the way toward personalized therapies according to HGSOC patients’ gene expression profiles by RNA sequencing (RNA-seq).
Materials and Methods
Sources and selection criteria. Articles demonstrating the potential of transcriptomic signatures as biomarkers of first-line chemotherapy resistance in OC patients were selected for this study. A three-step search process was conducted on PubMed, in which English-language research articles published between January 1, 2020, and March 5, 2024 were systematically reviewed, with a focus on titles and abstracts.
Published articles were identified by the following query on PubMed: (“high grade serous ovarian cancer”[Title/Abstract] OR “ovarian cancer”[Title/Abstract]) AND (“therapy”[Title/Abstract] OR “chemotherapy”[Title/Abstract] OR “platinum”[Title/Abstract]) AND (“response”[Title/Abstract] OR “resistance”[Title/Abstract]) AND (“transcriptomic”[Title/Abstract] OR “transcriptome” [Title/Abstract] OR “expression profile”[Title/Abstract]) AND 2020/01/01:2024/03/05[Date - Publication]. This PubMed search resulted in 123 articles. Based on the abstract content, 112 articles were excluded based on the following criteria: single-cell RNA sequencing (scRNA-seq), in vitro models, xenografts and animal models, immune-related signatures, immunohistochemistry, cancer stem cell and stromal cell only, non-HGSOC, and review articles. We excluded articles reporting scRNA-seq, as its implementation in clinical routine faces several challenges, despite its potential for providing detailed insights into cellular heterogeneity and disease mechanisms. The process of isolating single cells, preparing libraries, and sequencing might introduce technical noise and variability. Moreover, it generates vast amounts of data at a single-cell level, which is more complex and extensive compared to traditional bulk RNA-seq. We ultimately included 11 articles related to the following inclusion criteria: patients with HGSOC, a transcriptomic profile measured with RNA-seq, patient tissues, and English-language research articles. The selected articles were screened for clinicopathological features, sample storage methods, percentage of tumor cellularity, sequencing platforms, data availability, and transcriptomic signatures. The flowchart in Figure 1 illustrates in detail all steps taken for the selection of articles.
Analysis of differentially expressed genes. To investigate the overlap between DEGs candidates from various studies, the occurrence of genes associated with chemotherapy resistance was calculated with R Studio.
Results
Overview of reviewed articles. We conducted a three-phase article review process, enabling us to include 11 studies, starting from 123 initial candidate articles, as shown in the flowchart in Figure 1. Selected research articles have investigated the potential of transcriptomic signatures to predict chemotherapy sensitivity in patients with OC. We summarized these findings in six tables, offering a comprehensive overview of the progress achieved from January 2020 to March 2024. Table I outlines the clinicopathological characteristics of each study, providing the total number of patients enrolled in both the discovery and validation cohorts along with the reference dataset, including information on tumor histology, BRCA1/2 status, median age, and the FIGO stage at diagnosis. Table I also shows the therapies employed. Due to the difficulties in grouping patients based on their definitions of resistance and sensitivity to chemotherapy, Table II lists the exact definitions extracted from each article. Table III provides prognostic features, such as residual disease (RD) after surgery, resistance and disease relapse or recurrence, the platinum-free interval (PFI), PFS, and OS. Table IV presents the sample preservation techniques, percentages of tumor dissection, and transcriptomic platforms exploited in both the discovery and validation cohorts. Table V summarizes the selection process for identifying candidate DEGs together with the validation method employed, as described in the original articles by their authors. Moreover, we report the list of the differentially expressed genes (DEGs) associated with platinum-based treatment for each reference study. Nonetheless, the DEGs presented are not the complete lists derived from each RNA-seq analysis; rather, candidate DEGs represent the result of subsequent data analysis and validation, as determined by the authors of the original articles. The methods engaged for candidate gene validation varied across the analyzed articles. Some authors opted for validating data through other transcriptomic platforms, such as RT-qPCR or array-based technologies. Others employed statistical methods integrated with survival or prognosis information for validation. Finally, Table VI lists the Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA) (8, 19) datasets used in the analyzed articles, including the number of patients, tumor histology, extraction method, and sequencing platform used.
Cohort characteristics. Of the 11 selected studies, eight performed analyses in new discovery cohorts for a total of 727 patients diagnosed with HGSOC. Publicly available datasets, such as TCGA (8) and GEO (19) (Table VI) datasets, were used in five studies as the validation cohort (VC) (20, 21), discovery cohort (DC) (22, 23), or both (24). BRCA status was considered for 345 patients [n=231 BRCA1/2 wild-type (BRCA1/2wt); n=65 BRCA1m or BRCA2m, n=49 unknown significance]. Among the experimental cohort, 416 patients had FIGO stage III/IV HGSOC, and the median age at diagnosis was 59 years (range=56-59.4 years). Primary surgery followed by first-line platinum-based chemotherapy was performed for 93.2% of the patients, while the remaining patients underwent NACT followed by IDS. PARPi combination therapy was evaluated in two studies involving a total of 61 patients (25, 26). Disease recurrence or relapse was described in four studies (20, 25-27), while sensitivity to first-line treatment was described in eight (20, 21, 23, 24, 27-30). Recurrence and sensitivity were differently defined according to the choice of the authors of the original publication, as shown in Table II. The histology of patients selected from the GEO and TCGA datasets was described as EOC (22), SOC (23, 24) and HGSOC (30). Moreover, patients diagnosed at an advanced FIGO stage were either resistant or sensitive to first-line platinum treatment. Among the reviewed articles, only six documented at least one of the following clinical parameters: PFS (n=1 for chemotherapy, n=1 for PARPi) (19, 24), PFI (n=2) (24, 28), OS (n=5) (19, 22-24, 28, 29) or CA125 concentration (n=2) (27, 29). The samples were stored as formalin-fixed paraffin-embedded (FFPE) (n=5) or frozen tissue (n=5) samples. As inclusion criteria, each study either performed an RNA-seq analysis or retrieved RNA-seq data from the GEO and TCGA datasets to discover or validate gene expression profiles performed in a local laboratory. As shown in Table IV, the discovery and validation cohorts were analyzed through RNA-seq (n=7), array-based technologies (n=2), and RT-qPCR (n=3). The data from the GEO repositories were mostly array-based (except for GSE102073, for which RNA-seq analysis was performed), while the data from the TCGA were obtained via array and RNA-seq technologies (Table VI).
Gene expression profile. By reviewing the articles, we identified approximately 167 candidate DEGs that were correlated with resistance to first-line treatment or with transcriptomic changes before and after chemotherapy. Table V summarizes the workflow for DEGs selection, along with the validation method and the list of candidate genes associated with chemotherapy resistance. Differentially methylated genes (22, 26, 27), genomic mutational profiles (20, 25, 28, 29), long non-coding RNAs (22) and miRNAs (28) were also considered to develop more effective signatures. The articles’ lists of genes (column four in Table V) were compared to find (if present) overlapping genes across studies. However, the comparison did not reveal shared candidate genes across the considered studies.
Discussion
In the present review, 11 studies investigated the expression levels of genes associated with platinum sensitivity, patient survival, and prognosis. It is worth noting that data validation could be demanding because of the considerable heterogeneity across articles in terms of the definition of resistance and sensitivity to treatment. As shown in Table II, resistance can be defined as a PFI <6 months (21, 29); evidence of recurrence, relapse or progression (20, 27, 30) within six months; no evidence of response after three/four cycles of chemotherapy (28); or no response (23). Moreover, sensitivity to treatment is described as complete response (23, 28); absence of relapse, recurrence, or progression (20, 27, 30) within six months; or PFI>12 months (21, 29). Further definitions are provided as partial (pt)-sensitive, with signs of recurrence between six and 12 months (30) or partial response to first-line treatment (23). Performing a comprehensive comparative analysis between resistant and sensitive patients might be even more challenging because of the lack of definitions in four of the selected articles. To address this issue, it would be beneficial for future studies to agree on universal definitions of chemotherapy resistance and sensitivity applicable across cohorts, although this may not always be possible.
Candidate transcriptomic signatures predictive of chemotherapy resistance were identified in all the analyzed studies, even though they were obtained through different approaches. Although Javellana et al. (20) performed an RNA-seq analysis to identify DEGs associated with NACT resistance, they further evaluated only a list of seven candidate DEGs as the result of reshaping chemotherapy treatment. Conversely, instead of associating gene expression profiles with chemotherapy sensitivity, Gull et al. (26) aimed to identify genes that were both differentially expressed and differentially methylated between primary and recurrent tumors as well as between BRCA-carriers and BRCA-non carriers. Yang et al. (22) divided patients in two risk groups (high-risk and low-risk), describing survival and chemotherapy efficacy according to the differential expression of 17 long non-coding RNAs.
The association between platinum sensitivity and OS was studied by only two studies (23, 29), who calculated the median OS for each sensitive and resistant cohort. The initial platinum-resistant (IPR) signature (24), SIII signature (characterized by the transcriptomic profile of SDF2L1, PPP1R12A and PRKG1 genes) (30) and The Scarface Score (an integrative algorithm developed upon genomic and transcriptomic data) (25) were used to predict both patient prognosis and platinum sensitivity. Javellana et al. (20) associated the overexpression of genes identified after NACT (c-Fos and c-Jun) with worse OS and PFS by generating Kaplan–Meier plots with data from the GEO and TCGA. However, the study lacked data on the independent prognostic impact of c-Fos and c-Jun because no multivariate analysis was performed. The integration of survival data with chemotherapy sensitivity data could improve the prognostic accuracy of biomarkers through the development of more robust biomarkers for patient stratification.
Therapeutic approaches are described in almost every article [(22, 23) excluded]. Table I provides the treatment combinations as described in the reference article. Differences in regimen descriptions might induce biases when analyzing cohorts from public datasets. Across all the studies, the described regimens were all platinum based and administered as monotherapy or in combination with taxanes or other drugs. PARPis were administered in two cohorts in combination with adjuvant chemotherapy (25, 26), while bevacizumab in combination with platinum chemotherapy was evaluated only by Buttarelli et al. (21). Although few studies have analyzed immune infiltration levels in combination with tumor mutational burden (TMB) or HRD (24, 29) in response to chemotherapy, none of them included cohorts of patients treated with immune checkpoint inhibitors (ICIs). Nonetheless, integrating cohorts of patients receiving combinations of platinum and immune inhibitor agents might be carried out to improve patient prognosis and to design biomarkers that are predictive of ICI sensitivity. Currently, the NCT04034927 phase II trial has the exploratory objective of assessing the correlation between BRCA or HRD status and response to olaparib or tremelimumab (anti-CTLA-4) (31).
Regarding cohort design, nearly every cohort included patients classified as having HGSOC. The histology subclass was not precisely defined in those articles where cohorts were built using publicly available data (22-24), but they just reported EOC or SOC, therefore the cohort may also include patients with other histological subtypes than HGSOC. The median age at diagnosis was approximately 59 years; however, no data about age were reported for patients recruited by Lee et al. (28) and TCGA patients selected in Yang et al. (22). The BRCA status was differentially considered across transcriptomic studies. Buttarelli et al. (21) defined a 10-gene expression signature that was predictive for chemotherapy resistance in a cohort of BRCAwt patients only. In addition to identifying DEGs between primary and recurrent tumors related to tumor development, Gull and colleagues (26) identified 25 hypermethylated and downregulated, as well as 41 hypomethylated and upregulated genes between BRCA carriers and BRCA non-carrier patients. Three articles did not use the BRCA profile as a patient selection criterion but evaluated it through genomic analysis assessing a broader genomic panel (20, 25, 29).
Genomic and methylation analyses were also conducted in several studies in addition to transcriptomic experiments to increase the efficiency of predicting sensitivity and survival (20, 25, 27-29). Among the evaluated public datasets, the BRCA mutational profile was described only in the GSE63885 and GSE19829 reference articles (32, 33). Since patient stratification based on the BRCA mutational profile or HRD has already proven powerful for the clinical use of PARPis, as reflected in OS, identifying and validating additional biomarkers to guide treatment decisions is important. These biomarkers, whether signatures or algorithms, might be derived from integrated genomic and transcriptomic profiles to predict chemotherapy resistance and be applicable in clinical settings. A preliminary attempt in this direction was made with the development of The Scarface Score, an algorithm designed by Fernández-Serra et al. (25) to determine candidate patients who are likely to be sensitive to further PARPi therapies. In fact, The Scarface Score is developed based on statistical models investigating single nucleotide polymorphisms, copy number variations and the gene expression profile associated with chemotherapy response.
All but one (22) study reported the FIGO stage, with a greater number of enrolled patients diagnosed with advanced-stage disease (FIGO III/IV). Given the correlation between advanced FIGO stages and increased tumor aggressiveness and mortality, it is important to stratify patients according to which gene expression signature is mainly associated with advanced-stage disease rather than with early-stage disease. However, none of the recorded studies correlated transcriptomic signatures stratified by early or late stages.
In the end, we highlighted a total of 167 candidate DEGs associated with platinum resistance. Of the RNA-seq analyses performed, we reported only genes associated with chemotherapy, which were selected through validation or statistical analyses within their respective studies. Surprisingly, none of the candidate genes were identified in more than one study. However, it is noteworthy to outline that we did not analyze the entire list of DEGs discovered with RNA-seq; rather our focus was on those candidate genes that were validated for each study using different transcriptomic and statistical analyses. Indeed, a limitation of our study is the absence of access to raw RNA-seq data for many of the studies analyzed. Only Buttarelli et al. (21), Javellana et al. (20), and Benvenuto et al. (30), provided raw or normalized sequencing data that could have been newly analyzed by us. Therefore, for consistency, we decided to discuss only genes that were further validated through other experimental strategies (RT-qPCR or array-based technologies) or were chosen in additional data analyses performed by the authors of the original publications, as summarized in Table V. Furthermore, the absence of common biomarkers could be also a consequence of different transcriptomic platforms utilized across studies. RNA-seq analysis was performed in each engaged cohort or public dataset to obtain transcriptomic data as part of the inclusion criteria. Additionally, array-based technologies were used to discover new gene expression profiles or to validate data, while RT-qPCR was utilized only for validation purposes. Different platforms induce differences in results due to batch effects within the same technology or across platforms (34) or due to differences in the sensitivities and specificities of the selected technologies (35). Moreover, differences can arise from the extraction methods used or library preparation protocols used for RNA-seq (36) as well as variations in pipelines for data analysis. The method of sample preservation can further impact tissue integrity and thus data variability. FFPE blocks are widely utilized to preserve tissues. However, they are characterized by poor quality because of formalin-induced cross-linking and fragmentation of nucleic acids (37). In contrast, frozen tissue guarantees higher RNA integrity (38). In addition to the tissue handling procedures, the size and heterogeneity of the analyzed tumor samples are also important to discuss. A low percentage of tumor cellularity [at least 20% (20)] might have resulted in a limited number of malignant transcripts for the analysis and difficulties in reproducing results across studies. To address this challenge, even though 100% tumor cellularity is optimal, dissection of at least 70% or 80% of the tumor mass is usually mandatory to derive transcriptomic signatures predictive of first-line therapy response and survival.
All the abovementioned aspects may contribute to the absence of common transcriptomic markers across studies; therefore, a comprehensive integrative analysis of the mentioned articles could increase the statistical inference and eventually provide a robust prognostic biomarker predictive of therapy sensitivity. Additionally, given that gene enrichment analyses, such as Kyoto Encyclopedia of Genes and Genomes (KEGG) or Gene Ontology (GO) enrichment analysis, were performed for nearly all the analyzed articles [(23, 25) excluded], it would be advisable to carry out an integrated pathway enrichment analysis of the 167 identified genes. This approach would determine whether the candidate DEGs are involved in the same pathway responsible for platinum resistance. The next step could involve broadening the literature search to explore the genes identified in this scoping review and assess whether they have been examined and validated in other studies. Weberpals et al. (29), identified CCNA2 as a DEG, while other researchers have linked its protein product, cyclin A2, to more aggressive EOC tumor behavior and suggested it may serve as a predictor of patient response to first-line platinum-based chemotherapy (39).
Another potential obstacle encountered during data validation is the analysis of public datasets. Although the master TCGA dataset includes information from 587 SOC patients, several studies have only exploited data from cohorts of varying sizes without revealing selection criteria (40). The lack of detailed information on cohorts obtained from the TCGA renders accessing clinicopathological information from selected patients quite challenging. Similarly, retrieving data from GEO datasets is difficult when patient numbers differ between datasets and articles (22, 23, 30).
For our analysis, we did not perform any quality examination of the articles. However, as indicated in the above tables, several studies lacked clinical information on the enrolled patients. The absence of data could impede future external data validation. Moreover, in addition to performing integrated data analysis, future studies might focus on comparative pathway enrichment analysis to evaluate whether the identified deregulated genes are involved in chemotherapy resistance-related pathways. Another limitation of our study is the restricted time frame we applied for performing the PubMed query, which might have led to the exclusion of interesting articles due to time constraints. The recently published article by Lai et al., (41) compared scRNA-seq data from two GEO datasets with bulk RNA-seq data from TCGA-OV, eventually identifying a candidate common biomarker, FBXO2, that was strongly associated with chemoresistance, poor OS and poor PFS in patients with HGSOC. This finding was then further validated in cell lines. However, the limitation of this study is that scRNA-seq and bulk RNA-seq were performed in different populations, with a considerable difference in sample sizes (188 samples for bulk RNA-seq vs. seven for scRNA-seq). Even though the article reports promising findings, future research might consider analyzing the same patient cohort using both technologies to ensure consistency and validate the results.
Conclusion
Identifying biomarkers that predict resistance to first-line chemotherapy remains a challenge for patients with HGSOC. Transcriptomic signatures represent a potential tool for patient stratification and for the design of personalized treatments. Predicting platinum sensitivity may be crucial for patient prognosis and the potential adoption of long-term therapies, such as PARPis. Through our review, we identified 167 potential candidate genes, summarized the latest findings, and highlighted the challenges of validating biomarkers across studies and publicly available datasets.
Acknowledgements
S.C. received traineeship abroad in the Erasmus+ Programme.
Footnotes
Authors’ Contributions
S.C. contributed to the study planning, data collection, data analysis, results discussion; and writing of the final manuscript, J.L.J. contributed to the study planning, data collection, data analysis, results discussion; and final manuscript review, E.H. contributed to the study planning, data collection, data analysis, results discussion; and final manuscript review.
Availability of Data and Materials
All the data are from public repositories and published articles. The datasets analyzed in the present study are available in the GEO repository (https://www.ncbi.nlm.nih.gov/geo/) and TCGA repositories such as the UCSC Xena Browser (https://xenabrowser.net/).
Conflicts of Interest
The Authors declare that they have no competing interests.
- Received July 29, 2024.
- Revision received September 18, 2024.
- Accepted September 23, 2024.
- Copyright © 2024 International Institute of Anticancer Research (Dr. George J. Delinasios), All rights reserved.
This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY-NC-ND) 4.0 international license (https://creativecommons.org/licenses/by-nc-nd/4.0).