Immune checkpoint inhibitors (ICI) offer significant promise for non-small cell lung cancer (NSCLC), but efficacy is limited to a subset of patients. Existing biomarkers, including tumor mutational burden (TMB) and PD-L1 Immunochemistry, offer limited insights. Predictive testing that shows high TMB or PD-L1-positive results have about 40% positive predictive value at best, according to Kenneth O’Byrne, MD, FRCPI, FRACP, Professor of Medical Oncology and Clinical Director of the Cancer and Ageing Research Program at Greenslopes Private Hospital, Queensland University of Technology & The University of Queensland Translational Research Institute, in Brisbane, Australia
During a 2024 World Conference on Lung Cancer session titled Predicting the Future: Novel Pathology Assessments and Imaging Biomarkers, Dr. O’Byrne presented “Deciphering Single-cell Resolved Tumor Microenvironment Profiles Using Spatial Metabolic Mapping.” He discussed key findings that could expedite the development of predictive biomarkers for ICI therapies.
“In order to improve prediction of response, we really need to develop a better understanding of the tumor microenvironment,” Prof. O’Byrne said. “The complexity involved has a massive impact on how patients perform with immune checkpoint inhibitors.”
Traditionally, hot tumors—tumors with high concentrations of inflammatory cells—have been thought of as more responsive. However, Prof. O’Byrne said details from a single-cell analysis suggest that the presence of inflammatory cells alone is not predictive of response to therapy.
Hot tumors highly infiltrated by inflammatory cells may be responsive to immune checkpoint inhibitors, while hot tumors with inflammatory cells that are largely clustered around the exterior of the tumor and excluded from the tumor microenvironment (TME) may not be responsive. The number of inflammatory cells is similar, but the biological activity can be very different, Prof. O’Byrne said.
Researchers used a multimodal approach to look for signatures typical of response and non-response in a retrospective cohort of NSCLC tissue samples from 45 patients treated with immune checkpoint inhibitors.
A 10-step analysis combined multiplexed immunofluorescence staining with deep learning-based analysis to classify cells into 15 distinct types. The cells were then further categorized using artificial intelligence-driven unsupervised clustering. Assigning single cells to tumor or TME regions allowed researchers to compute more than 1,000 distinct spatial features and compare features between responsive and non-responsive tumors.
The unsupervised AI analysis revealed 43 distinct cell subsets, Prof. O’Byrne said. The subsets were distinguished primarily by different metabolic and activation states. The key differentially expressed proteins included oxidative phosphorylation markers CS, SDHA, and ATPAS and metabolic enzymes HK1, GLUT1, and LDHA.
The group observed direct links between metabolic state, effector functions, and tissue localization. Metabolically active lymphocytes showed elevated levels of PD-1, MHC class I and II, and CD44 positivity. Treg cells emerged as a predictor for resistance to immune checkpoint inhibition, echoing findings from other studies.
An analysis of cellular neighborhoods identified 10 distinct neighborhoods, including a macrophage-mixed tumor phenotype that was associated with response to therapy. The association is not surprising as it is already known that macrophages can differentiate tissues they penetrate and tend to exhibit anti-tumor activity once inside tumor cells, according to Prof. O’Byrne.
Unsupervised clustering of tumor cells revealed three predominant metabolic states: OXPHOS–positive, OXPHOS–negative, and PPP–positive.
The PPP-positive state was characterized by upregulated ASCT2, pNRF2, and G6PD. These cells exhibited a higher rate of proliferation and upregulation of CD44, a tumor stemness marker. In tumors with a high content of PPP-positive tumor cells, more than 40%, were resistant to PD-1 blockade and showed reduced overall survival rates.
“If you were PPP high, you did not have a response to therapy and if you were PPP low, you did have a response to therapy,” Prof. O’Byrne said. “I’m emphasizing that this is method development, and we’ve got some interesting signals. So the data, in terms of what we found, need to be interpreted with caution.”
