New Single-Cell Framework Maps How Myeloid Cells Shape Cancer Immunity

A team of researchers at Sun Yat-Sen University Cancer Center has unveiled a new computational framework that captures how tumor-infiltrating myeloid cells change their functional identities across many types of cancer. The framework, called the myeloid damage response index, or MDRi, moves beyond the long-standing habit of describing these immune cells by simple abundance counts or rigid polarization labels such as M1 and M2 macrophages. Instead, it measures coordinated transcriptional programs that reflect what myeloid cells are actually doing inside tumors: sensing tissue injury, driving inflammatory damage, promoting resolution and repair, and presenting antigen in the context of interferon signaling. The work, published in Cancer Immunology, Immunotherapy, offers an open-source toolkit that other investigators can apply to their own datasets immediately.

The central premise of the study is that macrophages and monocytes inside tumors exist in a continuum of functional states that cannot be reduced to a single binary. To quantify this continuum, the researchers built the MDRi around three core programs: an injury program capturing stress responses such as iron and heme handling, hypoxia, inflammatory chemokines and danger-sensing pathways; a resolution program reflecting tissue repair, efferocytosis, lipid processing and resident-like macrophage biology; and a combined antigen-presentation and interferon program, abbreviated APC/IFN, that brings together major histocompatibility complex activity and interferon-stimulated genes. From the injury and resolution programs they also derived an injury-resolution axis, a single metric that places each cell along a spectrum from active damage to active healing.

Constructing and validating a framework of this kind required an unusually broad evidence base. The team established MDRi in an immune checkpoint blockade-treated multi-cancer atlas containing 47,750 myeloid cells drawn from 192 samples spanning eight cancer types. They then tested the framework separately in an independent multi-cancer myeloid dataset to confirm that the identified programs were not artifacts of a particular cohort or sequencing batch. Because the antigen-presentation and interferon-response gene sets were deliberately constructed to be non-overlapping, agreement between these two signatures provided gene-independent support for the validity of the APC/IFN dimension, an important safeguard against circular reasoning in signature-based immunology.

One of the more technically interesting aspects of the study concerns the geometry of myeloid cell states. Using trajectory inference, the researchers found that macrophage and monocyte states organized along related but non-identical functional dimensions, and root-sensitivity analyses revealed a subtle but important caveat: the connectivity of the states was stable regardless of where the trajectory was anchored, but the inferred directionality of pseudotime depended on the choice of root. The authors interpret this as evidence that MDRi captures genuine transcriptional topology rather than a universal developmental sequence, a deliberately cautious conclusion that resists over-interpreting pseudotime as a maturation timeline in tumor myeloid biology.

The framework was then put to the test against clinical data. In exploratory analyses of checkpoint blockade-treated patients, the researchers observed that post-treatment non-responders showed concurrent elevation of injury, resolution and APC/IFN scores, suggesting a globally activated but functionally ambivalent myeloid compartment. However, when the analyses were adjusted for cancer type and treatment regimen, the data did not support an independent predictive effect of MDRi scores on response. Complementary analyses of T and natural killer cells indicated that immune differences associated with treatment response were actually more evident in pretreatment samples, hinting that the pretreatment immune landscape, rather than treatment-induced myeloid changes, may carry the stronger predictive signal.

Prognostic analysis in bulk tumor data added another layer of context dependence. The researchers projected MDRi programs onto patient-level cohorts from The Cancer Genome Atlas and fitted joint multivariable Cox models containing the injury, resolution and APC/IFN scores. The derived injury-resolution axis was excluded from these joint models because it is mathematically dependent on its two components, a statistically transparent choice. The results showed MDR injury acting as an adverse factor in selected cancers, while resolution and APC/IFN displayed associations that varied by cancer type and clinical endpoint. In other words, the same myeloid program can be associated with better outcomes in one tumor type and worse outcomes in another, which is precisely the kind of context dependence the framework was designed to expose.

Because single-cell atlases are far less common than bulk transcriptomic cohorts, a framework is only as useful as its portability. The researchers benchmarked MDRi against established immune-deconvolution methods such as MCP-counter, TIMER and xCell, as well as against published tumor-associated macrophage signatures. The comparison demonstrated partial but non-uniform overlap, meaning that MDRi captures myeloid-state information that is related to, but not fully explained by, conventional deconvolution scores. This positions the index as a complementary rather than redundant measurement, adding functional granularity that abundance-based methods cannot provide.

The spatial dimension of myeloid biology received particular attention. Using Visium spatial transcriptomics in nasopharyngeal carcinoma, the team mapped MDRi programs onto hematoxylin and eosin stained tissue sections, revealing that injury, resolution and APC/IFN programs occupy focal tissue niches rather than being uniformly distributed. A parallel multi-cancer Xenium analysis at single-cell spatial resolution across cervical cancer, glioblastoma, lung cancer and melanoma further revealed platform- and cancer-dependent spatial distributions of the MDRi-related programs, including cancer-specific estimates of how APC/IFN-high myeloid cells position themselves near tumor cells and how checkpoint interactions vary in their vicinity. These findings suggest that the functional identity of myeloid cells is shaped not only by cancer type but by precise anatomical microenvironment.

To make the framework accessible, the researchers released MDRi Explorer, an open-source Shiny application that implements scoring, reference comparison, survival analysis and benchmark visualization. Users can apply MDRi to their own transcriptomic data, compare their results against built-in references, and explore cancer-specific score distributions and survival associations from TCGA. The authors are careful to frame the contribution appropriately: MDRi is presented as a reusable, hypothesis-generating framework for investigating context-dependent myeloid functional organization, not as a universal prognostic signature or a clinically validated predictor. That restraint is notable in a field where signature-based tools are often oversold.

The broader significance of the work lies in its refusal to flatten myeloid biology into a single number. By decomposing the tumor myeloid compartment into injury, resolution and antigen-presentation/interferon dimensions, and by documenting honestly where those dimensions matter, where they do not, and how their meaning shifts across cancers, platforms and treatment settings, the study provides the immunology community with a map that is as much about uncertainty as about discovery. For researchers designing myeloid-targeted therapies, the message is that the same cell population may be a friend in one tumor and an enemy in another, and that any intervention must be interpreted against the local context of tissue damage, repair and antigen presentation.

Subject of Research: A single-cell-derived myeloid damage response index quantifying context-dependent myeloid functional states across cancers

Article Title: A single-cell-informed framework maps context-dependent myeloid damage-response states across cancers

Article References: Ding, R., Zheng, W., Long, Z., Cao, Z., Liang, J., & Quan, Q. (2026). A single-cell-informed framework maps context-dependent myeloid damage-response states across cancers. Cancer Immunology, Immunotherapy. https://doi.org/10.1007/s00262-026-04543-4

Image Credits: AI Generated

DOI: 10.1007/s00262-026-04543-4

Keywords: tumor-associated macrophages, myeloid cells, single-cell RNA sequencing, spatial transcriptomics, pan-cancer, immune checkpoint blockade, immune deconvolution, antigen presentation, interferon signaling, TCGA, myeloid damage response index, tumor microenvironment