Skip to main content

Dx Uses Cancer Cells' Electrical Properties to Measure Efficacy of Chemotherapy

Premium

NEW YORK – Researchers at Rutgers University have developed a point-of-care diagnostic device that uses multifrequency impedance spectroscopy and supervised machine learning to measure the effect of chemotherapeutic agents on cancer cells. According to a new study published in Microsystems and Nanoengineering, the device rapidly assesses the sensitivity of tumor cells to the drugs they're exposed to without the need for staining or labeling the cells.

The impetus behind the device's design was to create a diagnostic that could assess the efficacy of a targeted anti-cancer therapeutic approach being developed by Joseph Bertino, a resident researcher at Rutgers Cancer Institute of New Jersey, professor at Rutgers Robert Wood Johnson Medical School, and co-author of the study on the diagnostic device. Bertino and his team have developed and are testing a method that targets cancer cells by binding chemotherapy drugs to an antibody that targets matriptase, a membrane-bound protease expressed on the surface of certain tumor cells. In patients with cancers expressing matriptase, the conjugation of the matriptase-targeting antibody to the chemotherapeutic agent would target the drug more precisely to tumor cells and spare healthy cells.

In an email, Bertino noted that this approach would work for a high percentage of patients with solid tumors, as many of them express matriptase, and some B-cell malignancies.

"The way the device works is it takes tumor cells that have been treated or that have been mixed in with the antibody-coupled chemotherapeutic agent. And if the cells are expressing the matriptase they'll be sensitive to the drug — they will die. Otherwise the cell will remain alive," said Mehdi Javanmard, an associate professor at Rutgers' School of Engineering and the study's senior author. "And so, we use this impedance cytometer that works by measuring the electrical properties of the cells one by one to assess whether a cell is alive or dead, which tells us whether it's responded positively or not."

Javanmard noted that such a device could be useful in a number of contexts — whether in clinical trials to aid pharmaceutical companies in assessing the efficacies of therapies they are developing, or in the clinic to help a doctor rapidly assess whether a given patient is likely to respond to a therapy.

"You take the tumor cells obtained from a tumor biopsy, or even circulating tumor cells that have been isolated. You can mix them with the drug in advance, and if the cancer cells are drug resistant, then they will remain alive," he added. "If they're, not they'll be dead. And this device would be able to assess the degree of resistance or sensitivity to the drug."

The current gold standard for automated cell viability analysis is Beckman Coulter's Vi-Cell instrument, which uses the trypan blue dye exclusion method to perform its analysis, the researchers wrote in their study. But staining the cells limits the ability to perform any subsequent characterization or downstream molecular analysis on the cells and requires the use of optical instrumentation to assess cell viability. It also requires 0.5 to 1 mL of sample volume. 

Other recent studies have shown that certain microfluidic techniques and optical coherence tomography (OCT) could also be used to track cell death. But although OCT is label free and can quantitatively track cell death, the researchers noted, it would also require the use of optical instrumentation, making it less compatible with the needs of point-of-care diagnosis.

Using a cell's electrical properties to assess its viability — as the device developed by Javanmard and his team does — obviates the need for labels, stains, and optical instruments, making this approach less expensive than currently available methods. And the use of multiple frequencies at once results in higher classification accuracy. Further, this device requires as little as 20 μL of sample volume.

The team also developed a machine-learning algorithm in order to essentially teach the device to make determinations of when a cell was alive or dead, and then trained the artificial intelligence on a labeled training data set. The data used for training consisted of features extracted from 100 percent live and 100 percent dead cells, based on data from the Beckman Coulter Vi-Cell instrument, with features from live cells labeled as 1 and features from dead cells labeled as 0.

To evaluate the robustness and accuracy of the classifier algorithm, the researchers then tested it using three different tumor cell test samples — one containing 90 percent live cells, one with 50 percent live cells, and the third with 82 percent live. They reported accuracy as high as 95.9 percent, a true-positive rate as high as 95 percent, and a true negative rate as high as 97 percent from their various experiments with phase changes at different frequencies.

The researchers are now planning to test the device using clinical samples in order to continue training the AI. "For training, you would probably want to focus on a given cancer type. So if you're working with lymphoma, for example, you would probably want to train [the AI] based on lymphoma cells, and then if you're going to look at breast cancer, then you would probably want to do a separate set for that," Javanmard said. "There's an interest in seeing whether you could train [the device] using all of these different [tumor] types and then just make it a generic one-size-fits-all tool, but that would probably be more challenging."

Javanmard also noted that tumor heterogeneity could make it harder for the device to determine cell viability but said that added rounds of supervised machine learning on patient samples would likely help to overcome this problem.

"We would take an array of tumor cells from tumor samples. In one case we would kill 100 percent of them, and in the other case we would want 100 percent of them to be alive. We would run them through the device to essentially train the machine learning classifier of what the drug-resistant cell and what the drug-sensitive cell look like," he added. "And so, we train the classifier using different features — cell diameter, cell volume, cell shape, cell dielectric properties in a different membrane. To handle the issue of heterogeneity when working with patient samples, we would need to really train the classifier with large datasets in advance."

There's still a lot of work for Javanmard and his colleagues to do in the development of the device, but they already have an idea of how they'd like to roll it out for use in the clinic once the classifier is fully trained. They envision releasing an intermediate version of the diagnostic and then what Javanmard called a "long-term version, down the road."

In the nearer term, a doctor using the device would take a tumor biopsy from a patient, disassociate the cells and resuspend them in fluid, treat those suspended cells with a chemotherapeutic agent, and then inject that mixture into the device cartridge. In their study, the researchers noted that there are various protocols and kits commercially available to dissociate cells from human tissues and tumors into suspension.

Once the cartridge is inserted into the device, the reader would analyze the viability of the cells after they've interacted with the chemotherapy, displaying the answer on a screen. "It would make an assessment of what percent of the cells are drug sensitive and what percent are drug-resistant — [whether] 80 percent of these cells are drug sensitive and 20 percent of them are resistant to the drug, or 95 percent are drug sensitive and 5 percent are drug resistant, so [the clinician can assess whether it's] likely that this patient will respond positively to the therapy," Javanmard said.

In the longer term, the team would like to develop a version of the device that would work with liquid biopsy samples, isolating circulating tumor cells, then mixing them with various drugs, and feeding them into the device to assess drug sensitivity and resistance. This would be less invasive for patients, Javanmard noted.

He added that eventually, he and his colleagues would also like to build a closed feedback system feature into the device so that a clinician would be able to test combination chemotherapy regimens or cycle through different chemotherapy drugs.

"Drug resistance is a big problem — one single chemotherapeutic agent may not be sufficient," he said. "But being able to cycle through and adaptably determine a combination of drugs would be optimal."

In order to get the device from the lab to the bench once the team is finished with its development, it has to identify commercial partners, build a diagnostic that would be ready for manufacture, perform a validation trial involving hundreds of patients, and then go to the US Food and Drug Administration for approval. And approvals can take years, assuming the FDA is convinced that the device is better than the standard of care and doesn't negatively affect patients. The group currently doesn't have any corporate partnerships that Javanmard could discuss.

Currently, he estimated that it'll probably be about three to five years before clinicians see the intermediate version of the device in their offices. Bertino also said that human trials on his targeted therapeutic approach are still a few years away.