Approximately one-third of newly-diagnosed serous ovarian cancer patients do not respond to their initial neoadjuvant platinum-based chemotherapeutic regimen. By cataloging the genomic and clinical characteristics of these non-responding patients, it is hoped a more thorough understanding of their disease will be gained, thus permitting the development of effective strategies to treat them.
To that end, a team led by Jesus Gonzalez Bosquet, MD, PhD, Department of Obstetrics and Gynecology, University of Iowa Hospitals and Clinics, University of Iowa, evaluated a group of 88 platinum-resistant and platinum-refractory patients to see if they could be classified in a systematic manner using both clinical and genomic factors.
“Much work has been done by researchers to classify the genomic factors for cancer patients or alternatively, to catalog their clinical characteristics; however, we sought to develop a model which effectively integrated both clinical and genomic patient data,” Gonzalez Bosquet explained. “The goal of our study was to classify these non-responders to their initial chemotherapy using both clinical and molecular biological features.”
In this study, thorough molecular, biological, and clinical information was gathered for 88 platinum-resistant and platinum-refractory serous ovarian cancer patients at the University of Iowa Hospital who were included in The Cancer Genome Atlas (TCGA).
“These patients had genomic analysis done on their tumors using next-generation DNA sequencing techniques,” Gonzalez Bosquet detailed. Additionally, all relevant clinical information was gathered from the patients for incorporation into the researchers' model.
“TCGA is a tremendous resource, having a great deal of biological and bioinformatics data available for use. However, it does not always incorporate very relevant clinical data. Much of the clinical data that we obtained was fairly basic, such as surgical treatment, BMI, or diabetic status; however, that information, although simple, still is very relevant to the patient's outcome,” he noted.
“One of the first things we had to do was determine which variables were relevant and which ones were irrelevant,” he continued. “The main issue with this was that it required complex weighting of the different variables used; we would make variations to the system and then see how the data separated the clusters. This was definitely an iterative process, involving several modifications.
“Initially, we identified a set of 422 different genes that could predict chemo-response in these patients; through optimization of the prediction process we were able to narrow this large set down to 34 relevant genes,” Gonzalez Bosquet explained. By using tools such as GeneGo and clusterProfiler, the researchers found the 34 genes chosen had roles in important cellular functions such as DNA repair and replication, protein metabolism, and absorption, as well as cell adhesion and cytoskeletal remodeling (a particularly important function in the case of metastatic disease).
“Once we were able to differentiate between responders and non-responders, the next step was to stratify these non-responding patients based on their clinical-molecular characteristics in the hope that it will give us an insight into possible alternative treatments,” Gonzalez Bosquet stated.
“In the initial stages of the study, genome-wide unsupervised ‘cluster of clusters’ analysis was performed, which integrated gene copy number variation, messenger RNA (mRNA) expression levels, gene somatic mutations, and DNA promoter methylation levels to determine the patterns present in these non-responding patients,” Gonzalez Bosquet explained. “To this framework, clinical variables, such as surgical treatment, microRNA expression levels, and TP53 mutation status were incorporated.”
In this fashion, the researchers were able to separate the chemotherapy non-responding patients into three distinct clusters. Once these clusters were obtained, pathway enrichment analysis was performed for each grouping.
“We utilized pathway enrichment analysis to examine groups of genes, to determine their expression levels and this allowed us to see if there was any overexpression in those sets,” Gonzalez Bosquet noted. “In doing so, it was hoped that processes would present themselves for which there would be appropriate means for therapeutic targeting.”
In this study, Kyoto Encyclopedia of Genes and Genomes pathway analysis was utilized to determine the gene expression levels of the gene clusters identified as well as the relevant biochemical pathways involved. “We wanted to use all the available tools to fill in the gaps for missing information,” Gonzalez Bosquet explained.
When these analyses were applied, the non-responding serous ovarian cancer patients were separated into three distinct clusters. The first group was characterized clinically by having a larger than expected number of stage IV malignancies with less than optimal cytoreduction.
“This group also displayed underexpression of miRNA as well as mRNA, hypomethylation of their DNA, ‘loss of function’ mutations for TP53, and overexpression of platelet-derived growth factor receptor derived processes,” he said.
The second cluster of non-responders typically had low expression levels of miRNA's, generalized DNA hypermethylation, mutations in MUC17, and a large degree of activation in the WNT/β-catenin pathway. “The third group typically consisted of stage III patients who had optimal de-bulking, miRNA overexpression, mixed methylation patterns, and TP53 ‘gain of function’ mutations,” Gonzalez Bosquet noted.
Regarding the TP-53 gain of function mutations, he said there was some debate around that topic. “Some researchers feel that the mutated TP-53 switches from a role of a tumor suppressor to that of a tumor promoter; however, there is considerable debate about this in the scientific community.”
From the findings obtained in the pathway analyses done for the three clusters of non-responding serous ovarian cancer patients, the researchers were able to postulate some potential therapies for them based on their unique biochemical processes.
“For the first cluster, we feel that anti-VEGF (vascular endothelial growth factor) and broadly-targeted tyrosine kinase inhibitor therapies might be a suitable therapy,” Gonzalez Bosquet said. “For the second group, an anti-PD-1 or anti-PD-L1 immunotherapy might be a reasonable treatment. The third cluster of non-responding patients could potentially benefit from a combined therapy of a proteasome inhibitor with a histone deacetylase inhibitor.”
The need to find new and effective therapies for patients who do not respond to the standard platinum-based agents (typically cisplatin or carboplatin) is especially critical, as there are no efficacious treatments for these patients once those compounds no longer work.
“What we set out to do was characterize these non-responding patients using all of the tools at our disposal to characterize their disease and hopefully develop new strategies so that they can be successfully treated,” Gonzalez Bosquet noted.
“We obtained relevant genomic and clinical information and incorporated these into a working clinical model which tended to outperform the very useful TCGA analyses. One of the most import features of our model is its modularity; as new modes of information become available we can readily incorporate them into our model. This is very important, because as time goes on, the increased computing power and genomic testing capabilities as well as decreasing testing costs have resulted in an unprecedented growth in the amount of genomic information available.
“The potential therapies we mentioned for the three separate clusters are merely theoretical, based on the pathway analyses we did,” Gonzalez Bosquet concluded. “Much validation needs to be done before we can translate these findings into effective therapies for non-responding ovarian cancer patients.”
Richard Simoneaux is a contributing writer.