Background
Ovarian cancer (OC) is the seventh most frequent malignant tumor type in women worldwide and is the leading cause of death from gynecological cancer, accounting for 4% of cancer-related deaths (GLOBOCAN) [
1]. The majority of patients are diagnosed at an advanced stage due to unspecific clinical manifestations. There are 4 main histologic subtypes that have been described: serous (approximately 70%), endometrioid, clear cell and mucinous. In recent years, evidence has shown that there are unique molecular features, treatment responses and prognoses for each of the subtypes. Genetic alterations involving DNA homologous recombination repair system (BRCA1/2, genes of the Fanconi anemia and DNA mismatch repair pathways) are the most investigated and have been identified in more than 30% of OCs. Other relevant alterations include defective Notch, PI3K, RAS-MEK and forkhead box protein M1 (
FOXM1) signaling pathways, as well as mutations in
TP53, MTOR or
MYC in high-grade serous or endometrioid OCs, mutations in
ARID1A,
PIK3CA and
PTEN in clear-cell carcinomas, and
KRAS,
BRAF or
CDKN2A mutations in mucinous carcinomas [
2].
Complete cytoreductive surgery that achieves the resection of all macroscopically visible disease is a major factor that determines the chances of success in the treatment of OC. Chemotherapy is always given after surgery since most of the patients will eventually relapse, except in cases of nonaggressive tumors and in very early stage tumors. Platinum agents constitute the most active group of chemotherapy drugs in ovarian cancer, and over the last decades, multiple studies have progressively optimized the efficacy and tolerability of the treatment. Combination schemes of cisplatin and taxanes demonstrated a higher survival benefit over monotherapy and other combinations, and the cisplatin analogue carboplatin confirmed similar efficacy and substantially better tolerance than cisplatin. Therefore, intravenous carboplatin in combination with paclitaxel every 3 weeks constitute the standard first-line treatment for OC [
3]. Pegylated liposomal doxorubicin [
4,
5] or docetaxel [
6] are alternatives for patients who are not candidates for paclitaxel, and these treatments showed similar efficacy with a different toxicity profile. More recently, targeted therapies directed against angiogenesis (bevacizumab) and PARP inhibitors have demonstrated benefit in ovarian cancer, expanding available therapeutic options [
2].
The clinical response rates to these drugs regularly exceed 60%, and the median time to the onset of recurrence usually exceeds 1 year even in the subset of women with suboptimal cytoreduction [
2,
7‐
9]. In spite of surgery and chemotherapy administration, approximately 80% of the patients will relapse. Recurrent disease is generally incurable, and it is classified as platinum-resistant (recurrence < 6 months after last platinum dose) or platinum-sensitive (> 6 months). Platinum-resistant o sensitive status is one of the most important prognostic factors in recurrent disease and it is also a predictive factor of response of retreatment with platinum-based schemes. Platinum-resistant tumors show dismal outcomes with median overall survival less than 12 months [
3]. Therefore, the search for new compounds that may be active in platinum-resistant tumors (primary or acquired after treatment) is a necessity for these patients. In addition, the identification of platinum response biomarkers would help to discriminate patients, avoiding the administration of high doses of cytotoxic compounds to patients who would not obtain a real benefit.
Many mutations have been found to be responsible for the resistance to platinum drugs (TCGA, [
7,
10‐
15]), although their complexity makes the analysis of ovarian cancer resistance difficult. We hypothesize that the many known mutations that confer platinum resistance are distributed among different pathways, which may activate a few common essential effector genes. Ultimately, these effector genes may be responsible for the “ovarian cancer resistance physiology”, which may be measurable, predictive and targetable.
In this study, we performed a bioinformatic analysis with public databases to analyze transcriptional alterations that were common in ovarian tumors, mainly linked to recurrence. It was hypothesized that these alterations were causally connected to resistance to platinum therapy. After individual validation, the data suggest that these genetic alterations are involved in the acquisition of stemness properties that are linked to the resistance to therapy in ovarian tumor cells. Finally, the inhibition of key regulators of the stemness phenotype can recover sensitivity to platinum in stem cell surrogate assays.
Methods
Study approval
Written informed consent was provided by all patients. This project was approved by the Research Ethics Committee of the Hospital Universitario Virgen del Rocio (CEI 0309-N-15). All tissue samples and patient information were treated in accordance with the Declaration of Helsinki.
Patient cohort
A cohort of paraffin-embedded tissue samples from 21 patients with ovarian cancer were obtained from the biobank of the Hospital Universitario Virgen del Rocío-Instituto de Biomedicina de Sevilla (Sevilla, Spain) for RNA expression studies and for a correlation analysis of the clinicopathological features. Samples were obtained from biopsies of patients subjected to platinum treatment who were evaluated for their response according to the RECIST criteria, and normal tissue, platinum-resistant and platinum-sensitive tumor samples were obtained. Tumor samples were sent to the pathology laboratory for diagnosis and were prepared for storage with formalin fixation and paraffin embedding. Samples were stained with hematoxylin/eosin, and RNA was extracted and obtained from tumor tissue.
Public databases of clinical samples
RT–qPCR
Total RNA from paraffin-embedded tissue samples was purified using a Recover All Total Nucleic acid isolation Kit (Invitrogen) according to the manufacturer’s instructions, but with slight modifications; specifically, digestion was performed for 3 h at 50 °C and 15 min at 80 °C. Total RNA from tumorspheres and total adherent cultured cells (total culture samples) was purified using a ReliaPrepTM RNA Tissue Miniprep System (Promega, Fitchburg, WI, USA) according to the manufacturer’s instructions. Reverse transcription was performed with 0.5 μg of mRNA using a High-Capacity cDNA Reverse Transcription C-KIT (Life Technologies) according to the manufacturer’s recommendations. The PCR mixture (10 μl) contained 2 μl of the reverse transcription reaction product diluted 1:6, 2.5 μl of water, 5 μl of GoTaqR Probe qPCR Master Mix (Promega) and 0.5 μl of the appropriate TaqMan Assay (20X) containing primers and a probe for the mRNA of interest (Applied Biosystems). We used the following probes (Applied Biosystems): ADRB3 (Hs_00609046_m1), ANG (Hs04195574_s1), BTG2 (Hs00198887_m1), ESD (Hs00382667_m1), FBXL7 (Hs00202348_m1), RAD51 (Hs00947967_m1), ST13 (Hs00832556_s1), ST7L (Hs00373316_m1), DUSP4 (Hs01027785_m1), AP1M2 (Hs01091817_m1), CKAP4 (Hs_00199135_m1), C-KIT (Hs00174029_m1), DUSP1 (Hs00610256_g1), PAX8 (Hs01015257_g1), NOTCH3 (Hs01128541_m1), CD133 (Hs01009257_m1), NANOG (Hs04260366_g1), CXCR4 (Hs00607978_s1), ABCG2 (Hs01053790_m1) and GAPDH (Hs03929097_g1). We analyzed the quality of RNA obtained from the tumor samples and normalized expression levels to the housekeeping gene GAPDH.
Cell culture
Cells were cultured according to the manufacturer’s recommended procedure. Briefly, SKOV3 and OVCAR8 were cultured in RPMI and incubated at 37 °C with 5% CO2 in a humidified atmosphere.
Tumorsphere assay
Cells were washed once with PBS and then harvested with 0.025% trypsin-EDTA. A total of 5 × 103 cells of each cell line were resuspended in 1 ml of complete MammoCult medium (contains the MammoCult Basal medium, MammoCult Proliferation Supplement, fresh hydrocortisone and heparin; STEMCELL Technologies) and seeded in ultralow attachment 24-well plates (Corning #3473). Cultures were incubated in a 5% CO2 humidified incubator at 37 °C for 4 days. Tumorspheres were then visualized by inverted microscopy (Olympus IX-71) and were counted. Experiments were independently repeated a minimum of three times in triplicate.
Cytotoxic MTT assay
A total of 5 × 103 SKOV3 or OVCAR8 cells were seeded to form tumorspheres and then treated 24 h later with platinum drugs or/and Notch or C-KIT inhibitors (DATP or imatinib, respectively). After 96 h, the cell viability was measured with MTT.
Quantification and statistical analysis
All statistical analyses were performed using GraphPad Prism 4. The distribution of quantitative variables among different study groups was assessed using parametric (Student’s t-test) or nonparametric (Kruskal–Wallis or Mann–Whitney) tests, as appropriate. Experiments were performed a minimum of three times and were always performed as independent triplicates. Survival data from the patient databases were analyzed with the log-rank Mantel-Cox statistical test.
Analyses of cancer patient databases
We performed meta-analyses of the PrognoScan public patient datasets (
http://dna00.bio.kyutech.ac.jp/PrognoScan/) to analyze the expression levels in tumor and non-tumor databases for ovary tissue samples. Statistical significance versus normal samples was considered to be
P < 0.05. Patient survival was analyzed using the R2 Genomics analysis and visualization platform (
http://hgserver1.amc.nl), developed by the Department of Oncogenomics of the Academic Medical Center (AMC) (Amsterdam, Netherlands). Kaplan-Meier plots showing patient survival were generated for databases with available survival data using the scan method, which search for the optimum survival cut-off based on statistical analyses (log-rank test), thus finding the most significant expression cut-off. To analyze the protein network, we use the web portal
https://string-db.org.
Discussion
The cytotoxic activity of platinum complexes produces DNA alterations and increases the oxidative levels in tumor cells. This cytotoxic activity causes intra- and interstrand crosslinks and the formation of DNA adducts, provoking conformational changes that impair DNA replication. On the other hand, the increase of ROS species may induce DNA and mitochondrial damage, leading to a decrease in ATP activity. In addition, platinum-derived compounds produce alterations in cellular transport. Therefore, genetic events that alter any of these mechanisms may limit the efficacy of platinum compounds. Many mutations or alterations of the methylation profiles and epigenetic signals are involved in platinum resistance (TCGA database, cBioportal) [
10‐
13,
32]. In addition, a recent analysis of a large number of patients with high-grade serous ovarian tumors showed a high degree of complexity, a high number of genomic aberrations and genetic alterations, and high levels of intra- and intertumoral heterogeneity [
33‐
35]. Hence, the use of these alterations, most of them occurring with a very low frequency, as biomarkers to predict sensitivity or resistance to platinum-derived compounds is not currently useful. Since resistance to platinum treatments is one of the main causes of poor survival among ovarian cancer patients, the identification of prognostic and predictive biomarkers, as well as the understanding of the mechanisms driving resistance, is urgently needed.
We identified transcriptional targets that are possibly the common endpoints of the genetic alterations that are linked to platinum resistance in ovarian tumors. We found 15 genes that were transcriptionally altered, 6 of which were overexpressed and 9 of which were downregulated, belonging to different families. Compared to that in normal tissue, CKAP4, DUSP1, PAX8, NOTCH3, C-KIT and AP1M2 were upregulated in tumor tissue, while ADRB3, ANG, BTG2, ESD, ST13, ST7L, RAD51, DUSP4 and FBXL7 were highly downregulated in tumor tissue. Individually, most of these genes showed prognostic value in terms of overall survival in ovarian cancer patients, where platinum-derived compounds are still the main therapy. The genes that did not show statistically significant differences still showed a clear trend in our analysis. On the other hand, the complete profile of all of the genes also showed a clear predictive capability for the prognosis of overall survival or relapse-free survival, independent of the tumor stage. This profile itself could be used to stratify patients due to its predictive value for platinum resistance. It would also be interesting to combine our profile with other clinical predictors, such as CA125, stage, histological type or the degree of differentiation, to provide an accurate clinical assessment with increased prognostic and predictive value that could help to stratify patients in clinical practice.
Numerous attempts have been made to identify signatures associated with ovarian cancer therapy resistance [
16,
33‐
41]. Most of the multiple signature limitations involve the number of cases, representability of the different tumor types, further analysis of treatments, etc. A few of these signatures underwent further gene regulatory network analysis, identifying a unique network of genes that may potentially support current clinical practice, similar to our study. For example, Chudasama et al. found two proteins, RAD51AP1 and FSTL1, that were significantly overexpressed in ovarian cancer samples [
36]. In our study, we observed the downregulation of
RAD51 in tumor samples, which may account for the RAD51AP1 increase. Liu et al. reported a list of 21 genes from a literature search that may be involved in ovarian cancer drug resistance [
33‐
35]. One of these genes was
NOTCH3, which was also identified in our analysis. Interestingly, 8 of the genes from that study are a part of our network (
FOS, JUN, BCL2, KRAS, MAPK1, MYC, NOTCH3 and
STAT3), and most of the other genes are directly related to genes with similar functions to those found in our network (Bad and Bax are related to BCL2; EGFR and ERBB2 are receptor Ser/thr kinases that are related to Ras and the MAPK pathway; and Src, PIK3CA and AKT are also related to the Ras/MAPK pathway). It is important to take into account that most of the markers that have been proposed were related to the molecular mode of action of platinum, either preceding binding, directly related to the formation of adducts or related to the activation of the signaling pathway that is induced by DNA damage [
11]. A small number of markers represented off-target effects that are not related to the mechanism of action of platinum at any level; only one of our 15 markers may be directly related to the DNA signaling pathways,
RAD51, while the rest represent off-target effects. Interestingly, the MAPK pathway is at the core of our network and has been repeatedly described in the literature [
10,
11,
33‐
35]; therefore, the MAPK pathway may represent one of the main targets that could be leveraged to overcome platinum resistance. Our data on the cytotoxicity of suboptimal doses of the treatments on the tumorspheres suggest that combinations of platinum drugs with C-KIT, MAPK, PI3K or DNA-damage inhibitors may be a suitable therapeutic strategy to increase activity, avoid recidiva or the metastasis of platinum-resistant ovarian tumors.
We confirmed that the transcriptional levels of the genes found in our meta-analysis were also altered in samples from patients with platinum-resistant tumors from our own patient cohort, reinforcing their value as prognostic markers. Furthermore, we also found increased levels of CSC markers in the samples from platinum-resistant patients, connecting the resistance to platinum treatment to the CSC phenotype. We think that the CSC phenotype also has a clear prognostic value. The analysis of the CSC-related genes that were identified in our work (
ALDH1,
CD133,
NANOG,
CXCR4 and
ABCG2) showed that outside of their relevance for defining the CSC phenotype, these genes have no relevant prognostic value, either alone or as a group (data not shown). This lack of prognostic value is probably due to the large amount of heterogeneity that was observed in the expression of these genes in human tumors. However, our profile of 15 genes, which are not mechanistically related to the mechanism of action of cisplatin and are related to the off-target resistance signaling pathways [
11], have a clear prognostic ability to identify patients with ovarian tumors who are likely to develop resistance to platinum therapy.
Network analysis of these genes showed that most of the genes are integrated into 4 main clusters, three of which are linked directly to stemness. These networks are the NOTCH network, the MEK/MAPK network and, especially, the Yamanaka core (NANOG, SOX2, OCT4 and KLF4). These networks are connected through C-KIT, STAT3, NOTCH1 and MYC. These 4 genes lie at the middle of the three stemness networks and may be essential nodes to explore for therapeutic interventions.
Multiple markers, such as the Hoechst side population, CD133+, CD117 (c-KIT)+, ALDH1+ or CD44+ cells, have been described and used to identify CSCs from ovarian tumors [
16‐
22]. However, the concept of the ovarian CSC is controversial and has not been properly demonstrated. A unique CSC population has not been described, and it has not been concluded whether CSCs are responsible for ovarian cancer resistance. Some markers, such as CD133+, CXCR4+ [
42] or ALDH1 + CD133+ [
21], also define different CSC populations in ovarian tumors. In some cases, as in CD133+/CXCR4+ cells, these populations are also related to an increased expression of the stemness transcriptional core,
SOX2,
OCT4,
KLF4 and
NANOG [
42]. It has also been suggested that the expression of some of these markers depends on the environmental conditions [
21]. These data agree with the apparent heterogeneity of the CSC markers that was present in different resistant tumors, which probably reflects the selection of one of the different subpopulations, but each of these markers are related to the physiology of CSCs [
43‐
49]. However, a complete demonstration has not been provided. Our work directly analyzed 15 genes that were identified with transcriptional screening, 3 of which are directly related to stemness (
NOTCH3,
C-KIT and
PAX8) and 5 of which were direct markers of CSCs (
ALDH1,
ABCG2,
CXCR4,
CD133 and
NANOG) in ovarian tumor samples from platinum-resistant patients, corroborating this relationship. All of these data suggest that these CSC markers are not mutually exclusive and that they may appear as a result of multiple genetic changes that occur during the process of tumorigenesis.
The heterogeneity of the different CSC markers found among the resistant tumors was remarkable (Fig.
4). The correlation among the different markers indicated that different CSC populations may have been present. Most of the CSC markers correlated with
ALDH1, and we also found some correlation between
CXCR4 and
CD133 (Additional file
1: Figure S8), suggesting a general CSC population was represented by
ALDH1, and that there was a CXCR4 + CD133 subpopulation, as has been previously reported [
21,
42]. However, these trends also presented some heterogeneity. This may indicate that resistance is provided by different redundant pathways or by different pools of CSCs, most likely with distinct characteristics. Whether these pathways or CSC pools have different features and how this translates into the heterogeneity of the tumors needs to be further explored.
As previously mentioned, the networks that we identified are connected through C-KIT/MAPK/AKT or NOTCH. These genes lie at the middle of the networks and may be essential nodes that could be used to explore novel therapeutic interventions. In fact, we tested whether their inhibition produced any effect on CSC resistance. Our data showed that the inhibition of any of these nodes at suboptimal doses in combination with cisplatin or carboplatin significantly reduced the growth of the tumorspheres, elements that are enriched for CSCs. On the other hand, we expect much better results upon the inhibition of more than one node of the pathway, and this should preferably be evaluated in vivo; however, this hypothesis remains to be tested.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.