AI-driven μPharma platform accurately predicts leukemia drug sensitivity

Genomic testing has become a staple of modern cancer treatment, allowing physicians to identify a tumor’s underlying cause and address the root of the issue. However, genomic testing does not always lead to actionable genomic markers or a drug recommendation. For cancers with few clear targets, such as T-cell acute lymphoblastic leukemia (T-ALL), oncologists often do not have enough information to tailor treatments to individual children. In these cases, aggressive treatment is favored, which has a high likelihood of success but can lead to long-term treatment-related morbidity. These outcomes can include secondary cancers, cognitive deficits, cardiac toxicity and endocrine disorders, underscoring the need for options to better inform treatment approaches.

Pharmacotyping is a viable alternative when genetic testing is not informative. This process involves measuring tumor cells’ drug sensitivities in a lab to identify therapeutic vulnerabilities, even. The result is a therapeutic strategy finely tuned to the tumor. However, pharmacotyping is labor-intensive, often lacks single-cell detail, making broad implementation in clinical settings very challenging.

In light of this, researchers from St. Jude and the University of Utah designed μPharma, an AI-driven platform to accurately predict how T-ALL cases will respond to treatment at single-cell resolution. “Pharmacotyping has tremendous potential for guiding therapy, but we have to make traditional methods faster, more accurate and easier to use,” said co-corresponding author Jun J. Yang, PhD, Department of Pharmacy & Pharmaceutical Sciences vice chair. “This is why our collaborative team created μPharma, and I’m very excited about its potential application in drug development and clinical trials for pediatric cancer.”

Published in Med, μPharma uses a digital microfluidic immunofluorescence assay to identify biomarkers of drug sensitivity within cells. These biomarkers are then used to predict if a drug is a viable option. This approach reduces the need for prolonged incubation times and a large number of patient tumor cells, which are standard for conventional pharmacotyping approaches.

“T-ALL affects both children and adults, and outcomes remain challenging, but if we can rapidly and accurately monitor the sensitivity of cancer cells and tailor treatment appropriately, we believe it can provide better and more therapeutic options for this aggressive cancer,” said co-corresponding author Alphonsus Ng, PhD, University of Utah. “That possibility is extremely exciting for us.”

Platform reduces pharmacotyping from weeks to hours

μPharma has three main components. Through the power of microfluidics, electric fields first organize the assay within nanoscale droplets containing as few as 1,500 immobilized cells. Fluorescence scanning then rapidly measures the expression of drug targets at the single-cell level along with their spatial distribution and cell shape features.

Next, these features are cross-checked against a drug-response database built specifically to identify patterns that indicate drug sensitivity. Finally, a machine learning model trained on these features predicts the sensitivity of the tumor to the drug.

As proof of concept, the researchers used μPharma to assess drug sensitivity in T-ALL specimens. They successfully validated the protein pLCK as a biomarker predictive of dasatinib sensitivity and identified another protein, pBCL2, as a previously unreported biomarker predictive of venetoclax sensitivity. The combined assay took just four hours.

“The ability to reduce the multiday pharmacotyping process down to potentially hours is a significant innovation,” said Yang. “The automated microfluidics approach also reduces sample input and eliminates all manual steps, which are prone to error.”

While the findings are preclinical, the impact that μPharma could have on pharmacotyping is significant. Rapid drug sensitivity screening offers the optimal treatment strategy, while also potentially reducing conventional cytotoxic chemotherapeutics, which are associated with long-term morbidity.

“These results are exciting because they demonstrate that we can rapidly and reliably predict drug sensitivity without lengthy tests,” said lead co-corresponding author Yue Lu, PhD, University of Utah. “This could significantly speed up clinical decision-making. Next, we plan to validate our platform clinically, expand its use to additional drugs and disease types, and continue working towards real-world patient care.”