In a long-term study of patients with invasive breast cancer, researchers have identified several genomic signatures for chemotherapy resistance and treatment sensitivity that successfully predict patient response and outcome.
The group, led by Christos Hatzis of diagnostic developer Nuvera Biosciences and Fraser Symmans of the University of Texas MD Anderson Cancer Center, identified gene expression signatures for chemoresistance, chemosensitivity, and endocrine sensitivity and then examined the signatures' ability to predict patient responses to adjuvant chemotherapy. A test based on the findings, described May 11 in the Journal of the American Medical Association, could help identify which patients will do well with current treatments, and which might benefit from entering a clinical trial, they reported.
"What we wanted to know was, who are we already curing, and who are we not?" Symmans told PGx Reporter. "This would be a gatekeeper to inform and enable patients who … are at remaining residual risk — and in some breast cancer [cases] quite considerable risk — but don’t know it. In this setting they would have the potential to consider a clinical trial that would add something to the current best treatment and possibly further increase their probability of survival," he said.
Distinguishing these two groups would not only benefit patients, Symmans explained, but also clinical trials. "In the adjuvant setting, very few patients go on to clinical trials because we don’t have the metrics to identify who really is at risk after chemo plus or minus endocrine treatment," he said.
While acknowledging that the work is still in its earliest stages, he noted that if a test could, for example, determine that "half the patients out there could really benefit from participating in a clinical trial… in that setting, you could see very rapid clinical trial accrual. You could see clinical trials completing early, the opportunity to test a number of promising treatments, and bring the best treatment to the standard of care much, much earlier."
In the study, conducted between June 2000 and March 2010, Symmans, Hatzis, and colleagues used Affymetrix U133A GeneChip microarrays to identify six predictive signatures in 310 samples of newly diagnosed breast cancer.
Symmans said that in addition to predicting response to chemotherapy, the group also wanted to measure how patients would do on subsequent endocrine therapy, and predict resistance to both therapies. "As important as it is to define response, it becomes equally important to define tumors that are just totally resistant to chemotherapy or endocrine therapy," he said.
The group's predictor strategy thus involved components for each of the three groups: response to chemotherapy, response to endocrine therapy, and resistance to either therapy (characterized by two subcomponents: residual disease and earlier relapse events).
"We identified genes under consolidation that best predicted response, or best predicted resistance, or were most strongly associated with early relapse events," Symmans said. "We repeated that process in the two different subsets, ER-positive and ER-negative, so six different gene sets were identified."
Then the group combined the signatures into a single testing algorithm for predicted sensitivity to adjuvant treatment of ERBB2-negative breast cancer with taxane and anthracycline chemotherapy, which they tested on an independent cohort of 198 samples to confirm performance.
With a primary prediction endpoint of distant relapse free survival (DRFS) at three years, patients in the validation cohort predicted to be treatment sensitive had a 56 percent probability of excellent pathologic response and a DRFS of 92 percent — an absolute risk reduction of 18 percent.
The study also compared four other predictors to the group's test algorithm to see how they stacked up, including PAM50, a 52-gene signature assay, as well as the 96-gene genomic grade index, and a 30-gene signature called DLDA30. The authors noted in the paper that they could not compare their signatures to Agendia's MammaPrint or Genomic Health's Oncotype Dx "because each uses a different technology lacking direct correlation with our Affymetrix microarray platform."
None of the comparison tests had a positive predictive value "significantly greater than the baseline response rate of 30 percent," the authors wrote. And, Symmans added, the comparison signatures also predicted good response for patients that actually "had the worst outcome, which was paradoxical."
Part of the explanation for this, he said, is that no single predictor is perfect. "It's not enough to use response — pathological response, chemotherapy response — as the only point of reference," Hatzis said. "You have to really consider these other aspects — resistance and endocrine sensitivity — to be able to get something that is useful."
Testing only for treatment sensitivity might miss tumors that have a separate measure of resistance, he said. Similarly, ignoring the difference between estrogen sensitive and non-estrogen sensitive samples can confound prediction by marking patients as having a poor response to chemotherapy who would actually have a good overall response with a combination of chemotherapy and hormone therapy.
"There is a sequential synergy in some patients between partial response to the chemo and the rest coming from the endocrine ... Its okay to have half a glass of red wine from chemo, and another half glass of white wine from endocrine, you still get your full glass," he said.
Adding in these additional factors, he said, gave the group's test better resistance to this type of paradoxical identification. "First we identify who's going to be resistant, then who's going to be highly responsive. That way when we get it wrong, we won't get it horribly wrong," Symmans added.
"That's the power of dissecting the problem down into the key elements that consider the effects of treatment on response and survival, then piecing them back together into a meaningful algorithm that becomes the test," he said. "This is what makes this an interesting study in this field."
The authors reported that in its current format the test "would be performed on fresh primary tumor sample obtained from clinical core needle biopsy, fine needle aspiration… or surgical resection." Additional studies are needed to gauge if the results can be generalized to other chemotherapy regimens.
Hatzis reiterated that one role of a test based on the group's results would be identifying the patients who might not have a good response to standard treatments, which might encourage them to enter trials for new experimental drugs. At the same time, the test could also help solidify the decision to undergo chemotherapy for those predicted to have a good response.
"This is also good for clinical trials," Symmans said. "The bottom line is if you are on a trial that takes current best standard chemo, versus that plus something new, concurrently, the patients that are highly responsive and will have an excellent survival from the standard treatment will be represented in both of the arms of that study… So they're just diluting out the benefit. You've lost your signal."
The drug may actually work for ten percent of the population, but "you're not going to see that drug working, because you can't add to an already outstanding result," he said. "To quote a colleague, you can't cure someone twice."
"[A test] along these lines, assuming everything worked out down the future, could change that landscape," he added.
The two authors said that though they are not yet at the point of creating a commercial assay, results from ongoing validation studies, which they plan to present at this June's American Society of Clinical Oncology meeting, have been "promising."
Currently, the group is testing technical reproducibility and the algorithm's sensitivity to tissue heterogeneity. They said that work is being done "behind the test to really get it ready to the market," and that they will soon start a prospective study.
"Within a year's time, we'll have a much more complete picture as to how clinically robust this test really is," Hatzis said.
”We hope as more studies come out, the more evidence will accumulate of the validity of the test and the robustness of the predictions," Symmans added.
Have topics you'd like to see covered in Pharmacogenomics Reporter? Contact the editor at mashford [at] genomeweb [.] com.