Immunotherapy has fundamentally changed oncology, yet its primary challenge remains a stubborn binary: most patients do not respond to treatment. To bridge this gap, clinicians have relied on biomarkers like PD-L1 expression and Tumor Mutational Burden (TMB). However, as research progresses, it has become clear that these single-metric indicators are often insufficient. They act as static snapshots of a highly dynamic, multidimensional battle between the immune system and the tumor.
The Failure of Static Single-Protein Metrics
The most common biomarkers, such as PD-L1 protein levels, suffer from what Janis M. Taube et al. describe as "predictive leakage." In their review, Implications of the tumor immune microenvironment for staging and therapeutics, the authors explain that some patients who test negative for a marker still derive significant clinical benefit, while others who test positive do not respond at all.
This insufficiency is starkly illustrated by case reports, such as that by Qijun Wang et al., which details a patient with PD-L1 negative gastric cancer achieving complete remission—a result that standard metrics would have deemed impossible (Unexpected curative effect of PD-1 inhibitor in gastric cancer with brain metastasis: A case report). Similarly, Elizabeth Alwers and Jakob Nikolas Kather found that established molecular markers like KRAS mutations fail to predict how rectal tumors respond to neoadjuvant therapy, suggesting that the drivers of tumor growth are distinct from the mechanisms of treatment-induced death.
The Importance of Spatial and Cellular Contexture
The reason these markers fail is that they ignore the "immune contexture"—the specific spatial arrangement and functional state of cells within the tumor microenvironment (TME). Janis M. Taube et al. argue that the density and location of immune cells are more predictive than raw protein expression.
This spatial architecture is critical for modern therapies like antibody-drug conjugates. Barbara Pistilli et al. demonstrated in the ICARUS-BREAST01 trial that a tumor's physical organization is a better predictor of drug efficacy than simple target expression levels (Patritumab deruxtecan in HR+HER2− advanced breast cancer: a phase 2 trial). Furthermore, Jiekun Yang et al. identified that specific cell subtypes, such as mature regulatory dendritic cells (mregDCs), act as essential orchestrators of the immune response, suggesting that we must look at functional partnerships between cells rather than just their presence (Mature and migratory dendritic cells promote immune infiltration and response to anti-PD-1 checkpoint blockade in metastatic melanoma).
From Static Snapshots to Dynamic and Systemic Markers
A tumor is not a static entity; it evolves under the pressure of treatment. Jianhua Chen et al. found that tracking how a patient's blood-based TMB changes during treatment is far more accurate than a single baseline measurement (On-treatment blood TMB as predictors for camrelizumab plus chemotherapy in advanced lung squamous cell carcinoma: biomarker analysis of a phase III trial).
Moreover, the body's systemic state influences the TME. Alessandra Cesano et al. proposed a framework for "Cancer Immune Responsiveness" (CIR) that integrates germline genetics and lifestyle factors with the TME (Advances in the understanding and therapeutic manipulation of cancer immune responsiveness: a Society for Immunotherapy of Cancer (SITC) review). This systemic view is supported by Filippo Gustavo Dall’Olio et al., who showed that a patient’s history of viral infections—their "immune fitness"—is a strong predictor of survival (Antecedent viral immunization and efficacy of immune checkpoint blockade: an extensive serum antibody profile to predict outcomes in non-small cell lung cancer). Even psychological stress, captured through "psycho-biomarker" gene signatures, has been linked to immunotherapy outcomes by Qian Zuo et al. (Exploring the predictive “psycho-biomarkers” for checkpoint immunotherapy in cancer).
The Role of AI and Multi-omics
To handle this complexity, researchers are turning to artificial intelligence and multimodal data integration. Yi Yang et al. developed the SimTA framework, which combines serial CT scans (radiomics) with clinical data to outperform traditional methods in lung cancer (A multi-omics-based serial deep learning approach to predict clinical outcomes of single-agent anti-PD-1/PD-L1 immunotherapy in advanced stage non-small-cell lung cancer.).
Similarly, Zhen Han et al. used "pathomics" to extract predictive features from standard H&E biopsy slides, providing a low-cost way to capture spatial interactions that PD-L1 tests miss (Development and interpretation of a pathomics-driven ensemble model for predicting the response to immunotherapy in gastric cancer). By integrating tumor-intrinsic mutations with extrinsic immune signatures, Denise Lau et al. demonstrated that multimodal models can even identify "backdoor" immune responses, such as cytotoxic CD4+ T cells, that drive benefit when traditional pathways are hidden (Integration of tumor extrinsic and intrinsic features associates with immunotherapy response in non-small cell lung cancer).
Key Foundational Work
Implications of the tumor immune microenvironment for staging and therapeutics (2017) — contributed 1 concept elements to this direction.
Mechanistic modeling quantifies the influence of tumor growth kinetics on the response to anti-angiogenic treatment (2017) — contributed 1 concept elements to this direction.
Phase I/Ib study of olaparib and carboplatin in women with triple negative breast cancer (2017) — contributed 1 concept elements to this direction.