Researchers from Ontario's McMaster University have used a drug target-based approach to create gene expression signatures that can predict response to two classes of breast cancer chemotherapy drugs.
Earlier this month, the team published in BMC Medical Genomics a proof of principle that these expression indices were significantly associated with response to anthracycline and taxane drugs. In the paper, the group described its method of measuring the expression of a drug or drug class's main molecular target along with a swath of co-expressed gene transcripts, to create the predictive indices.
According to the study's lead author Robin Hallett, the researchers are now planning to apply the method to newer targeted cancer therapies.
"A lot of pharmaceutical companies are pushing these targeted agents," Hallett told PGx Reporter this week. "But in a lot of cases, patients can be resistant or the drugs don’t work as well as hoped," he said.
"What we hope we can do is build target indices for these next-generation cancer therapeutics … that hopefully can inform how to use the drugs with better efficacy."
In the meantime, the group is also working on refining the chemotherapy predictors reported in the study, Hallett said. With the right collaborators, the group might eventually move to validate the signatures clinically, he added, although there are no concrete plans for that yet.
Hallett said that in the proof-of-principle study, the researchers aimed to establish that gene expression measurements linked to a chemotherapy drug target could predict therapy response as measured by complete pathological response, or complete disappearance of a patient's tumor.
Hallett divides breast cancer patients into three groups: "patients who don’t need treatment, patients who need treatment and we have treatment that works, and patients who need treatment and we don’t have treatments that work for them."
In recent years, gene expression signatures have yielded accurate and useful tests — like Genomic Health's Oncotype DX and Agendia's MammaPrint — for parsing patients that fall into the first group versus those in the second or third.
But it has been harder for researchers to use gene expression to distinguish between the latter groups, patients who need treatment and have options and those who need it but don't respond to available drugs.
"If you have a tumor and these tests say you don’t need chemotherapy, that’s great, you’d probably opt not to get it," Hallett said. "But if you have tumor and [the test says you are at a higher risk,] for those patients … we still need to identify those that benefit from the chemotherapy we have, and those that don’t benefit so that we can get them onto something a little more novel a little sooner."
In their study, the researchers used several hundred samples to create expression indices associated with the molecular targets of two types of chemotherapy — anthracylcines and taxanes.
Anthracycline drugs target the protein TOP2A, while taxanes target tubulins. The researchers decided to focus on these two targets and the transcripts of genes co-expressed with them, hoping to show that this target-based approach could yield a better predictor than previous studies.
"If you have a drug and you have a cell you want to kill with the drug, you’d better hope the cell needs what you are targeting," Hallett said. "So, that’s really the approach we took."
Hallett said the researchers hypothesized that they could increase the accuracy of their predictors by measuring co-expressed genes along with the drugs' direct targets, taking a broader view of pathways potentially involved in the biology of the cancer cells' reliance on the target.
"We extended to build a network of genes we think represents the biology in cells associated with [the target], so it's kind of a much more comprehensive measurement of how relevant that signaling is to the cell," he said.
Testing the resulting target-based indices in a retrospective sample cohort of 278 breast cancer patients treated with a regimen of paclitaxel, 5-fluorouracil, adriamycin, and cyclophosphamide, the researchers found that the TOP2A index was significantly associated with complete pathological response to the combined treatment with an area under the receiver operating curve of 0.73.
Testing another signature involving B-tubulin and associated transcripts in a smaller 14-patient cohort that received only docetaxel, the group found it could predict complete pathological response to the taxane with an AUC of 0.89. Combining the two signatures in the original 278-patient cohort and in two additional combination therapy cohorts allowed the researchers to predict response to combined chemotherapy with between 0.75 and 0.78 AUC, they reported.
To investigate whether their target-based signatures were similar to existing predictors of chemotherapy response, the group then examined the overlap between their TOP2A and B-tubulin indices and a validated gene expression predictor called DLDA30, finding that no probe sets from the B-tubulin set, and only one from the TOP2A set, were present in the DLDA30 predictor.
However, according to Hallett, it's not clear yet whether the group's approach will be able to yield a clinically useful predictive test.
One challenge to establishing a signature associated with specific chemotherapies is that measures that seem to be predicting response may really just be mirroring a cancer's estrogen receptor status, which is itself strongly associated with proliferation and response to chemotherapeutics. ER-negative tumors tend to be more proliferative and less responsive to chemo.
"If you collect data on 500 tumors and you try to identify genes that match whether a tumor responded or not, you typically identify genes that are associated with the estrogen receptor, rather than with chemotherapy response itself," Hallett said. "A lot of people just rediscover that big division in those two types of breast cancer."
Because of this, Hallett said, one of the study's reviewers asked the team to show in the paper that their signatures weren't just measuring ER status or proliferation-associated genes.
To establish this, the team first tested whether the TOP2A index was truly specific to its anthracycline target by trying it as a predictor of response to the taxane-only treated patients. Indeed, the TOP2A index was not significantly associated with response to the docetaxel.
The group also created a "proliferation index" linked to the well-characterized proliferation-associated gene E2F1 and showed that the performance of both the TOP2A and B-tubulin indices were more accurate.
"I actually do think our signatures are measuring proliferation to a certain extent," Hallett said. "But they’re also measuring something else, and we think they’re measuring how much the cells in a tumor require the target of the chemotherapy drug the patient is being administered."
With AUCs close to 0.8, Hallett said, the indices for TOP2A and B-tubulin are definitely significant. However, he said, while the predictors had high sensitivity, their specificity was much lower. "A lot of the patients you identify who you think might respond actually don’t respond," he said.
Another limitation to the groups' findings, he said, is that pathological response is not the real clinical endpoint of interest. His team did not study whether the indices were able to predict long-term survival. "That’s the ultimate thing you're trying to measure with these tests," he said.
In the near term, Hallett and his team view the results from the proof-of-principle study as a stepping stone for follow-on investigations. The researchers now hope to apply the same approach to newer targeted cancer therapies, which may yield stronger predictive signatures.
Hallett said he has also been working to refine the signatures the group created for TOP2A and B-tubulin, by narrowing down the co-expressed gene transcripts to those that are most strongly associated with response.