When faced with cancer the body mounts an immune response, and tumor-infiltrating lymphocytes – primarily B cells and T cells – attack the invading disease. Quantifying the spatial architecture of the immune cells, such as their location and the space between them, can aid in predicting patient outcome and response to treatment. But there’s a hitch.

“When you look at immune architecture on a high-definition pathology slide, you don’t really get a sense of the characteristics of a population of different immune cells because we don’t have the ability to visually distinguish a B cell from a T cell,” says Cristian Barrera, a doctoral candidate in the Wallace H. Coulter Department of Biomedical Engineering at Emory University and Georgia Tech.

Barrera and colleagues from several universities and medical centers developed a computational pathology approach, called PhenoTIL, that uses machine learning to identify clusters of immune cells and capture spatial interplay and features associated with tumor rejection. Using artificial intelligence, they identified tumor-infiltrating lymphocyte (TIL) clusters in routine pathology images for nearly 1,800 patients with non-small cell lung cancer (NSCLC).

“We discovered that only some of the TIL clusters are strongly associated with outcome and treatment response,” says Barrera, lead author on a recent article about the study findings in the journal npj Precision Oncology. 

The retrospective study included patients with early and late-stage NSCLC from the United States and Europe who had received many different treatment combinations, including chemotherapy and immunotherapy. 

“We showed that certain PhenoTIL clusters are associated with outcomes in lung cancer patients treated with multiple types of chemotherapies and also with checkpoint blockade. The approach was validated on Checkmate 057 with partners from Bristol Myers Squibb,” says study senior author Anant Madabhushi, PhD, the Robert W. Woodruff Professor of Biomedical Engineering at Emory University School of Medicine and Georgia Institute of Technology College of Engineering, a member of the Cancer Immunology research program at Winship Cancer Institute of Emory University and Research Career Scientist at the Atlanta Veterans Administration Medical Center. 

The findings are significant because TILs are not uniform and don’t contribute equally to patient outcome. Therefore, the sheer number of TILs isn’t always reflective of disease outcome. The PhenoTIL clusters appear to distinguish activated T cells – where the cancer-fighting action occurs – from exhausted or by-stander T cells.

“The PhenoTIL clusters were found in all patients in the study, though not all of them appear to have had activated immune cells. That is telling. Perhaps patients with exhausted or bystander T cells are not going to respond to treatment,” says Madabhushi.

Clinicians could potentially use PhenoTIL signatures to better predict treatment outcomes and aid in personalized treatment decision-making, especially regarding the selection of chemotherapy, immunotherapy and combination therapies. The next step for the research team is to move PhenoTIL into prospective clinical trials for validation of outcome in the context of combination therapies for lung cancer. The team is also exploring the utility of this approach in other cancer types.