NEW YORK – A 64-protein signature appears to accurately predict which patients with high-grade serous ovarian cancer will not respond to first-line treatment with platinum-based chemotherapy, a recently published study suggests.
In a paper published Thursday in Cell, Fred Hutchinson Cancer Center physician-scientist Amanda Paulovich outlined how she and colleagues at the University of Arkansas and the Mount Sinai School of Medicine developed the 64-protein predictor. Using mass spectrometry, they characterized proteomic markers in 242 high-grade serous ovarian cancers that either responded or did not respond to treatment. Paulovich and colleagues hope to develop the predictor into a test that hospitals can perform on mass spectrometry equipment already set up in most pathology laboratories.
The standard treatment for newly diagnosed high-grade serous ovarian cancer is platinum-based chemotherapy before or after surgery. Around 85 percent of patients respond initially to this treatment, but between 10 percent and 20 percent have treatment-refractory disease and poor prognosis. Among non-responders, the median overall survival is around a year.
There has been no meaningful change in these outcomes for 40 years. Additionally, there is currently no way to predict which patients will have chemo-sensitive disease and which will have chemo-refractory disease. The latter are subjected to the toxicity of chemotherapy without benefit and because they tend to progress rapidly, their poor performance status makes them ineligible for clinical trials, which means chemo-refractory disease is not being studied.
"I got into all of this because I was very frustrated at the inadequacy of our diagnostics," Paulovich said. "Patients were highly variable. One patient's tumor would melt away with therapy, and the next patient on the same therapy for what looked like the same kind of tumor might show no response, and we had no way of predicting that."
Paulovich joined Fred Hutch in 2003 and set up a translational proteomic laboratory to develop proteogenomic methods that could be used in clinical settings to facilitate precision cancer treatment.
In her lab, Paulovich is focused primarily on developing a test that can identify at diagnosis which patients are likely to be chemo-refractory, so they can have a chance to participate in clinical trials. The lab is also working on finding experimental treatments for such patients as an alternative to first-line platinum-based chemotherapy.
In a 2021 publication in Oncogene, Paulovich and collaborators curated a database of 900 genes linked to platinum-based chemotherapy resistance discovered during the past 30 years. "All of those genes mapped to many different cellular processes, indicating that there's no single mechanism that cancer cells use to become resistant to platinum," Paulovich said, noting that there are a heterogeneous set of interlinked mechanisms that lead to platinum resistance. "That's probably why, despite 30 years of fantastic basic science on this problem, not a single biomarker to predict platinum refractoriness has been translated into clinical use."
Further complicating the problem, she added, is that current technologies are not designed to validate biomarkers for such complex phenotypes. Most biomarkers are validated using immunohistochemistry assays, which are difficult to use to assess a multi-marker signature. "You look at one protein at a time, and as a result, despite the fact that hundreds of candidate biomarkers get discovered, validation studies tend to focus on single markers because that's what the technology allows," Paulovich said.
Instead of using conventional methods like IHC, in the current study, which was funded by the National Cancer Institute's Office of Cancer Clinical Proteomics Research, Paulovich's team turned to tandem mass spectrometry which can detect and quantify all proteins in a sample. In addition, they used liquid chromatography, whole-genome sequencing, and RNA-seq to analyze pretreatment biopsies from three cohorts of high-grade serous ovarian cancer patients — a discovery cohort of 158 formalin-fixed paraffin samples, and two validation cohorts comprising 20 FFPE samples and 64 frozen samples.
All cohorts included a mix of tumor samples known to be chemo-sensitive or chemo-refractory. Although only about 20 percent of patients have refractory cancers, the researchers were able to enrich for patients with chemo-refractory tumors in the study because they had access to retrospectively collected samples from several academic centers.
When analyzing the data for single biomarkers associated with refractory disease, researchers identified one mutation in BRCA1, 22 RNAs, four proteins, and one phosphosite, which were consistent with previously published findings. "Despite these encouraging results, better prediction models are needed for clinical practice," Paulovich and her collaborators wrote in the Cell paper.
Next, using proteomic data from the FFPE discovery cohort, the researchers identified a set of 64 proteins from the metabolic, hypoxia, and NF-κB pathways that together were associated with treatment response. Next, they trained an ensemble prediction model on these proteins using machine learning algorithms. They then validated the model on the FFPE and frozen validation samples and showed that it could detect a subset of refractory high-grade serous ovarian cancers with 98 percent specificity.
They also identified five "highly stable" clusters in the data associated with certain protein pathways, and Paulovich said her team was able to replicate these associations in two independent patient cohorts as well as in patient-derived xenograft models of ovarian cancer. "We believe these clusters represent different mechanisms of refractoriness and may be implicating therapeutic vulnerabilities of different molecular subtypes of ovarian cancer," she said.
For example, researchers flagged cluster 3 as being centered on a metabolic pathway, in which there is high expression of proteins involved in fatty acid oxidation and oxidative phosphorylation. "You might predict that tumors that fall into that cluster might be sensitized to platinum, if you treat with platinum in combination with a drug that inhibits fatty acid oxidation, which is very highly expressed in those tumors," Paulovich noted. "We demonstrated that in mouse models."
Paulovich and her colleagues tested that hypothesis in a patient-derived xenograft model mapped to cluster 3, which also showed upregulation of the metabolic pathway. Those tumors could be re-sensitized to platinum-based therapy through pharmacological inhibition or through CRISPR knockout of an enzyme, CPT1A, which catalyzes a critical step in fatty acid oxidation.
As a clinical application, Paulovich envisions that there would be two tests — one to determine whether patients have chemo-refractory disease and another to determine the cluster those chemo-refractory patients' tumors fall into. Ideally, that second test could be used to enroll patients in a clinical trial studying treatments tailored to that cluster. The tests would be carried out using a type of targeted mass spectrometry, called multiple reaction monitoring, which is widely deployed in clinical laboratories for newborn screening and other purposes.
Paulovich said her group is now further validating the assays so they can be performed in her CLIA-certified lab, though the researchers don't have a partner with whom to develop a commercial test. "Diagnostics are woefully undervalued and under-reimbursed," she said. "It's hard to get a new assay, especially a new, multiplexed mass spectrometry assay, [CPT] coded for adequate reimbursement," she said, pointing out that platinum-based chemotherapy is also off-patent, which "leaves little incentive for commercial partners to invest in the development of such an assay. "