Skip to main content
Premium Trial:

Request an Annual Quote

Purdue University Developing Classifier to Better Predict Cancer Patients' Responses to Chemo


NEW YORK (360Dx) – Purdue University researchers are developing machine learning technology that analyzes tissue biopsies to enable the launch of a diagnostic classifier that could eventually clarify whether a cancer patient will respond to chemotherapy.

The researchers said that they've seen success in using this technology as part of a biodynamic digital holography system used to obtain phenotypic profiles of non-Hodgkin B-cell lymphoma biopsies taken from dogs treated with standard-of-care chemotherapy. Dogs are considered a good surrogate for humans for evaluating the efficacy of cancer drugs, and the Purdue researchers' method for testing patient response to therapy is also undergoing clinical studies of ovarian, breast, and esophageal cancer in humans. So far, those studies have shown similar levels of accuracy to the canine study, the researchers said.

John Turek, a professor of basic medical sciences at Purdue University and a developer of the technology, said in an interview that enrollment for all the human trials will continue until the analysis of patient responses achieve the desired level of statistical significance.

In the group's first pre-clinical study, published recently in the journal Biomedical Optics Express, the researchers reported an 84 percent success rate predicting response to therapy. The researchers tested 19 dogs previously diagnosed with B-cell lymphoma, which is molecularly and clinically similar to lymphoma in humans.

Existing chemosensitivity tests that rely on two-dimensional cell culture derived from cancer patient biopsies lose the important three-dimensional tumor microenvironment that controls many aspects of drug distribution and efficacy, the researchers said.

"There is a clear need to develop technologies that are able to accurately predict the response of tumors to chemotherapy," Daniela Matei, a professor of cancer research at Northwestern University's Feinberg School of Medicine, said in an interview. Matei's group has been providing tumor tissue initially grown in mice for the classifier's development, and more recently it has been providing biopsies taken from patients undergoing surgeries, she said.

The classifier under development at Purdue "uses a unique type of technology applicable to cells and tissues that can be tested ex vivo," Matei said, adding, "The beauty of this technology is that it can be done on very small fragments of tissue obtained through biopsy, and it contains not only the tumor material but also information about the tumor microenvironment to replicate what goes on inside the tumor."

Matei said that her group has been working with the Purdue technology for more than three years, and within six months it hopes to complete its work to obtain clinical responses and begin analyzing the data it has collected so that it can provide research conclusions. At present the results are "too preliminary" to share them, she said.

Turek noted that Indianapolis, Indiana-based Animated Dynamics, which he cofounded with David Nolte, a colleague at the university, has licensed the technology from Purdue. The firm expects to make the machine learning classifier available later this year through a CLIA-certified lab for use as a laboratory-developed test that tests for chemoresistance in humans and dogs.

The Purdue classifier captures movement inside living tissue and uses Doppler light scattering and fluctuation spectroscopy to detect how living tissue is affected by cancer drugs.

In practice, the researchers place pieces of biopsy tissue in a multiwall plate and apply various chemotherapy drugs. After shining infrared light into the middle of the tissue, the system captures the scattering of light between cells. It compares the signatures derived from light scattering to a library of signatures obtained over decades that reflect whether a patient — dog or human — is likely to respond to a given chemotherapy.

The researchers noted in their paper that they capture signatures reflective of "molecular mechanisms of action of therapeutic drugs that modify a range of internal cellular motions."

"Organelles, vesicles, the nucleus, and the cell membrane all move at different speeds and the scattered near-infrared light picks up a Doppler shift based upon speed," Turek said. "We detect the changes of speed over time in living 3D tissue when a drug is applied." The outcome, he noted, is a spectral fingerprint, or change in frequency over time, that reflects a drug's mode of action and the resistance or sensitivity of the biopsy to the drug.

A reference beam that is adjusted for depth within the tissue produces a three-dimensional characterization of the tissue.

Over years of research, the group has "used various tool compounds with known mechanisms of action to understand the spectral signatures" of cell activities, such as apoptosis, necrosis, and mitosis, Turek said.

He noted that in clinical practice, a patient will receive a single biopsy, and a lab will provide testing and analysis within 48 hours, fast enough to inform a physician about drugs to which the patient may be resistant.

Resistance to chemo

The technology is being developed to address a vexing problem with chemotherapy: Less than half the patients diagnosed with cancer respond favorably to chemotherapy because different cancers have various degrees of chemoresistance, the researchers said.

"Predicting response to chemotherapies within all cancer types has been fraught with confusion," Michael Childress, an associate professor of comparative oncology at the Purdue University College of Veterinary Medicine, said in an interview. He has been providing clinical outcome data used to refine and continue to train the predictive algorithm within the group's binary classifier. Including additional clinical outcomes for dogs and people receiving chemotherapies improves the classifier, he said.

Childress noted that for their work, the heterogeneity of dog tumors is among the most important comparisons with human tumors. "That kind of heterogeneity helps you differentiate poor responders from good responders," and it can be applied to testing human responses, he added.

Older methods for predicting patient response to chemotherapy broke up tumors into individual cells and regrew them as two-dimensional cell cultures, Turek said. However, the phenotype of the tumor is very dependent upon the 3D microenvironment, he noted. "If you break up a tumor and culture the cells in culture plates, the phenotype may be different from that of the tissue in the patient," Turek said.

Although existing two-dimensional tests have been limited in their ability to achieve improvement in the selection of chemotherapy, there have been "tremendous efforts in genomics to provide the answer to predicting patient response to therapy," he said. "Biodynamic imaging is a tissue phenomic approach. We know the destination, or patient response, but we cannot tell you how we got here. Ultimately, as genomic and phenomic libraries are assembled, we should be able to build a bridge between genotype and phenotype."