Moffitt Study Identifies Distinct Tumor-Immune Ecologies That Predict Immunotherapy Response in Lung Cancer

Researchers at Moffitt Cancer Center have identified distinct spatial tumor–immune ecosystems that predict whether patients with advanced non–small cell lung cancer will benefit from immunotherapy. The findings, published in Cancer Research, show that analyzing how tumor and immune cells are organized and interact within the tumor microenvironment may predict disease progression more accurately than PD-L1 status alone. The paper was featured on the cover of the publicationand is also part of its Data Science special series.

Using multiplex imaging, spatial statistics and machine learning, investigators analyzed paired pre- and on-treatment biopsies from patients enrolled in a clinical trial combining the HDAC inhibitor vorinostat with the PD-1 inhibitor pembrolizumab. Rather than focusing on single markers, the team examined how immune cells and tumor cells were positioned relative to one another, defining spatial “neighborhoods” and broader tumor–immune ecologies.

Tumors from patients whose disease progressed were characterized by an immune-suppressive architecture before treatment began, including greater spatial clustering of FoxP3-positive regulatory T cells and PD-1–expressing immune cells near tumor cells. In contrast, tumors from patients with stable disease showed stronger colocalization of CD3- and CD8-positive effector T cells interacting with tumor cells. These findings suggest that response to immunotherapy may be influenced by preexisting spatial organization within the tumor microenvironment.

When researchers trained predictive models using these spatial ecosystem features, they achieved up to 87.5% accuracy in forecasting disease progression. By comparison, using PD-L1 expression alone resulted in approximately 63% predictive accuracy. The results support expanding lung cancer diagnostics beyond single-marker testing toward spatial biomarkers that may help guide more precise immunotherapy decisions.

Q&A with Sandhya Prabhakaran, Ph.D., co-author and applied research scientist and Alexander Anderson, Ph.D., chair, of the Integrated Mathematical Oncology Department at Moffitt. 

What was the main goal of this study?

The primary goal was to define and quantify distinct tumor–immune ecologies in non–small cell lung cancer and determine whether these spatial patterns can predict disease progression and response to immunotherapy.

What does it mean to view the tumor as an ecosystem?

It means treating the tumor microenvironment as a dynamic community of interacting cell types rather than isolated tumor or immune cells. Treatment response depends on how these cells are spatially organized and functionally interacting, not just whether specific markers are present.

What differences did you see between responders and non-responders?

Responders had tumors that were more immune-permissive before treatment. Non-responders had tumors with a suppressive spatial architecture that limited effective immune attack. These ecosystem patterns were largely present before therapy began.

Why did these spatial patterns predict response better than PD-L1 alone?

PD-L1 measures the presence of a single checkpoint molecule. The tumor–immune ecosystem reflects whether immune cells are properly positioned and capable of killing the tumor. Because treatment success depends on spatial cell interactions, ecosystem-based patterns were more predictive than PD-L1 alone.

How could this change clinical decision-making?

The findings support moving beyond single-marker testing toward ecosystem-based patient stratification. Patients with immune-permissive ecosystems may benefit from checkpoint inhibitors alone, while those with suppressive ecosystems could be directed toward combination therapies or clinical trials earlier.

How feasible is this approach in clinical practice?

While not yet routine, multiplex immunohistochemistry and digital pathology platforms are becoming more common. With further validation and streamlined computational tools, spatial immune profiling could become part of lung cancer diagnostics in the coming years.