Abstract
Researchers used transcriptome analysis to predict patients' responses to treatments with 80% accuracy on average. The procedure relies on identifying pairs of interacting genes that are lethal to cells when both are inactivated or that enable tumors to evade treatments. The scientists showed that their approach accurately predicts patient responses to targeted therapies and checkpoint inhibitors for a variety of cancers.
A new method of analyzing tumor transcriptome data has enabled scientists to predict patients' responses to targeted therapies and checkpoint inhibitors with 80% accuracy on average (Cell 2021;184:2487–502.E13). Researchers are now designing clinical trials to test this approach to selecting the best treatment for each patient.
A few clinical trials have evaluated strategies for incorporating transcriptome information into treatment decisions. One arm of the 2019 WINTHER trial, for instance, assigned therapies to patients based on tumor RNA expression (Nat Med 2019;25:751–8).
Eytan Ruppin, MD, PhD, of the NCI, and colleagues devised a different strategy that involves identifying synthetically lethal pairs of genes, which kill a cell only if both are inactivated. They reasoned that if one gene of a pair is downregulated in a tumor, the patient is likely to respond to a drug that targets its partner.
To home in on the pairs, the scientists gathered data from previous screens of cancer cell lines in which researchers had knocked out or knocked down specific genes with RNAi, CRISPR/Cas9, or chemical inhibitors. Experiments like these typically generate hundreds of thousands of possible synthetic lethal pairs, most of which are not clinically relevant, so the researchers turned to other types of data to winnow the candidates. For instance, they analyzed patient outcomes from The Cancer Genome Atlas (TCGA) to identify gene pairs whose inactivation was associated with better survival. They ended up with 25 potential gene pairs for each drug they considered.
The team then evaluated their procedure by analyzing three datasets from patients with melanoma who had received BRAF inhibitors. They calculated a score for each patient based on the number of BRAF synthetic lethal partners with reduced expression in a given tumor. Patients with high scores—indicating more downregulated gene partners—responded better to treatment. The team's analysis of one dataset, for instance, determined that progression-free survival was a median of 88 weeks longer for patients with high scores than for patients with low scores. Synthetic lethal scores predicted whether patients would respond to BRAF inhibitors in 71% of cases.
The scientists then applied the approach to 23 additional datasets from clinical trials of chemotherapy or targeted therapies by identifying the synthetic lethal partners of the targets of each drug and determining the synthetic lethal score from transcriptomic data for each patient sample. Although the method performed poorly for patients treated with chemotherapy, it predicted the response to targeted therapies with more than 70% accuracy for most of these datasets.
To predict responses to immune checkpoint inhibitors, the researchers altered their approach by searching for synthetic rescue interactions, where a tumor cell remains viable by reducing expression of one gene when its partner (in this case, PD-1, PD-L1, or CTLA4) is targeted. To test the power of this approach, the scientists calculated synthetic rescue scores from 21 datasets that contained information on patients who had received checkpoint inhibitors. The scores predicted treatment response, they found.
Combining their analyses, the researchers determined that synthetic lethal and synthetic rescue scores predicted overall or progression-free survival with 80% accuracy, on average, among patients who had received targeted therapies or checkpoint inhibitors.
The work “provides further support that it's not just about genomics—transcriptomics is also important and should be part of what we are doing in the clinic,” says co-author Razelle Kurzrock, MD, of the University of California, San Diego. The researchers are now collaborating with other scientists to organize clinical trials employing the strategy in breast, prostate, bladder, and other cancers.
“This is a step in the right direction,” says Mark Cowley, PhD, of the University of New South Wales in Sydney, Australia, who wasn't connected to the study. “The consistency of the findings across different drug classes and cancer types suggests the results are robust.”
Yves Lussier, MD, of the University of Utah in Salt Lake City, who wasn't involved in the research, describes the approach as “a clever idea.” But he cautions that it “doesn't reach the clinical level of accuracy” and will probably have to be combined with other methods to help make treatment decisions. –Mitch Leslie
[Abstract:] Researchers used transcriptome analysis to predict patients' responses to treatments with 80% accuracy on average. The procedure relies on identifying pairs of interacting genes that are lethal to cells when both are inactivated or that enable tumors to evade treatments. The scientists showed that their approach accurately predicts patient responses to targeted therapies and checkpoint inhibitors for a variety of cancers.
