Amidst the continued struggle to treat non-small-cell lung cancer, a new study led by Stanford University scientists suggests that a patient’s response to immunotherapy may hinge on how immune cells cluster around tumors. Their results reveal that spatial arrangements of certain immune cells within tumors can serve as powerful predictors of treatment response, surpassing existing biomarker tests.

Lung cancer leads global cancer mortality, and non-small-cell variants make up more than 80% of cases. Immune checkpoint inhibitors have transformed therapy yet help only 27–45% of recipients.

Reliable predictive biomarkers for immunotherapy response have eluded clinicians, who currently rely on PD-L1 immunohistochemistry, tumor mutational burden, and microsatellite stability tests, each offering modest predictive performance across trials and are prone to inconsistency.

In the study, “Single-cell multimodal analysis reveals tumor microenvironment predictive of treatment response in non-small-cell lung cancer,” published in Science Advances, researchers combined multiplex immunofluorescence, RNA sequencing, and deep-learning–driven histology to pinpoint tumor-immune architectures that correlate with patient outcomes.

Tissue samples from 132 non-small-cell lung cancer patients treated at Stanford Medical Center formed the foundation of the work. Among these, 50 patients underwent intensive multiplex immunofluorescence (mIF) imaging, while whole-slide histology images were available for 115 and RNA sequencing data for 122. Across all modalities, more than 45 million cells were profiled.

Multiplex immunofluorescence captured 33 protein markers across 255 tissue cores, yielding spatial coordinates for 1.5 million cells. Unsupervised clustering grouped local neighborhoods into eight phenotypes. A deep-learning model called NucSegAI, trained on 2.2 million nuclei and fine-tuned on 30 lung slides, mapped 45.6 million cells on 119 whole-slide histology images.

RNA-seq deconvolution estimated immune cell fractions, and gene-set enrichment linked spatial patterns to signaling pathways. A cytotoxic T lymphocyte (CTL) score summarized the fraction of cytotoxic T cell (Tc)–enriched neighborhoods per patient.

Among 34 patients who received anti–PD-1/PD-L1 therapy, responders carried 2.5 times more Tc cells and 6.5 times higher Tc-enriched neighborhoods than nonresponders. These tumors also exhibited stronger spatial interactions between Tc and dendritic cells (Dc), monocytes, and tumor cells.

Patients in the upper half of the CTL score distribution enjoyed significantly longer progression-free survival, while macrophage-dominant neighborhoods flagged early relapse.

Former smokers with non-small-cell lung cancer tended to carry a quieter immune footprint in their tumors.

Researchers at Stanford University found that tumor samples from former smokers contained significantly fewer regions where immune cells directly neighbored cancer cells. Never smokers, in contrast, had denser clusters of lymphocytes adjacent to tumor cells.

Researchers conclude that integrating spatial immunology with routine pathology can sharpen patient selection for costly immune checkpoint drugs and spare others toxic exposure.

Wider adoption may hinge on software such as NucSegAI that extracts similar insights from standard H&E slides, promising a streamlined path toward precision lung-cancer care.

© 2025 Science X Network