The research team, led by Eun Hye Joo, Young Shin Song, and senior author Young Joo Park, assembled an unusually comprehensive dataset to interrogate this question. Sixteen fresh-frozen tumor specimens spanning the full aggressiveness spectrum, from conventional papillary thyroid carcinomas driven by BRAF V600E and follicular thyroid carcinomas driven by RAS, through to their anaplastic counterparts, underwent single-nucleus RNA sequencing, whole genome sequencing, and bulk whole transcriptome sequencing. Twenty additional formalin-fixed paraffin-embedded tissues were profiled with two complementary spatial transcriptomics platforms, 10x Visium and NanoString GeoMx DSP. The investigators then validated their findings against external datasets comprising 311 bulk transcriptomes and 94 single-cell RNA sequencing samples drawn from public repositories, alongside immunohistochemical analysis of 96 thyroid cancer tissues. This multimodal architecture allowed the team to cross-check every major conclusion across independent technical platforms, from whole-genome mutational profiling down to single-nucleus resolution cell atlases.

The genomic backdrop revealed a striking mutational stratification. Anaplastic tumors from both lineages almost universally harbored mutations in the TERT promoter and TP53, two alterations conspicuously absent from their less aggressive precursors. Aggressive and high-grade differentiated tumors consistently carried TERT promoter mutations, and the overall mutational burden and copy-number complexity climbed steadily with aggressiveness, peaking in anaplastic disease. Structural rearrangements, including gene fusions, proved more frequent in RAS-driven tumors than in their BRAF-mutant counterparts, while mitochondrial DNA alterations clustered predominantly in Complex I genes, with truncating mutations concentrated in the BRAF lineage. These patterns establish that the road to anaplasia is paved with accumulating genomic damage, but the nature of that damage differs depending on the initiating oncogene.

The study’s most consequential finding emerged from pseudotime trajectory analysis of tumor epithelial cells, conducted with the Monocle framework on more than 102,000 single nuclei. BRAF V600E-driven tumors followed what the authors describe as a gradual dedifferentiation trajectory, a slow erosion of thyroid identity accompanied by progressive immune pathway activation. RAS-driven tumors, by contrast, displayed abrupt transitions, leaping from a differentiated state into an aggressive one with little intermediate territory. These abrupt shifts were characterized by aneuploidy, activation of the epithelial–mesenchymal transition program, hypoxia signaling, and extensive extracellular matrix remodeling. In practical terms, a BRAF-mutant tumor appears to slide down a slope, whereas a RAS-mutant tumor appears to fall off a cliff. This distinction has real clinical weight, because it suggests the two subtypes may present different windows of opportunity for early intervention, and that monitoring strategies calibrated to one subtype may fail entirely for the other.

Copy-number inference performed with the CopyKAT algorithm on the single-nucleus data confirmed that aneuploid epithelial subpopulations expanded dramatically in the RAS-driven anaplastic tumors, providing a genomic correlate for their abrupt phenotypic jumps. Whole transcriptome scoring reinforced the picture: papillary tumors driven by BRAF showed lower thyroid differentiation scores but higher ERK activation than their follicular counterparts, while RAS-driven anaplastic carcinomas exhibited the lowest differentiation scores and highest ERK activity of any group studied. The ESTIMATE algorithm revealed a fascinating see-saw dynamic across progression. Aggressive differentiated tumors paradoxically showed reduced immune and stromal infiltration relative to non-aggressive ones, yet both lineages reversed this pattern completely upon transition to anaplastic carcinoma, with immune and stromal scores surging as epithelial content declined. The reversal was even more pronounced in BRAF-driven anaplastic tumors, hinting that these tumors cultivate a particularly inflammatory niche in their terminal phase.

Perhaps the most therapeutically tantalizing results concern cancer-associated fibroblasts, which emerged as central puppeteers in the dedifferentiation drama. Using CellPhoneDB ligand–receptor analysis across systematically stratified epithelial subtypes, the team uncovered mutation-specific communication circuits between fibroblasts and tumor cells. In BRAF V600E-mutant anaplastic tumors, integrin-based signaling predominated, with the fibroblasts anchoring epithelial cells through extracellular matrix contacts. In RAS-driven anaplastic tumors, an expanded repertoire of interactions appeared, including the PLAU–PLAUR urokinase axis, the TNFSF10–TNFRSF10B death receptor pathway, and the amphiregulin AREG–EGFR circuit. These were not speculative inferences confined to computational prediction. The team spatially validated the interactions using the Stopover tool on Visium data, calculating Jaccard indices for colocalized ligand and receptor expression within tumor regions, and confirmed protein-level expression of key players including L1CAM, ITGAV, ITGB1, PLAU, and PLAUR through immunohistochemistry on tissue microarrays. Crucially, the abundance of these epithelial subtypes correlated with poor clinical outcomes in the 311-patient validation cohort, analyzed through Cox proportional hazards modeling with mutation-stratified optimal cutoffs.

The technical achievement underlying these insights deserves emphasis. Single-nucleus RNA sequencing sidesteps the dissociation bias that plagues conventional single-cell protocols in fibrous, matrix-rich tumors, and the team layered on sophisticated deconvolution strategies, using CELLCODE surrogate proportion variables to estimate cell-type contributions in bulk transcriptomes and GraphST to deconvolve spatial spots against single-nucleus references. Trajectory tracts through physical tissue space were reconstructed with SPATA, allowing pathway activity to be scored along anatomical paths from differentiated tumor cores into anaplastic fronts using AUCell enrichment. Batch effects across the sixteen patients were corrected with Harmony, and mixed-effects models with cell type as a random effect cleanly separated the influence of differentiation and ERK signaling scores on cellular composition. Every analytical choice was vetted across four independent data modalities, a level of internal replication that is still rare in cancer single-cell studies.

The biological implications ripple outward in several directions. First, the finding that dedifferentiation trajectories are mutation-specific challenges the implicit assumption, embedded in much of the prior literature, that thyroid cancer progression follows a universal path. Previous single-cell studies largely pooled papillary tumors without stratifying by driver mutation, effectively averaging away the very differences this study illuminates. Second, the identification of actionable ligand–receptor pairs opens concrete therapeutic avenues. The AREG–EGFR axis and the urokinase system are both druggable, and the spatial restriction of these interactions to RAS-driven anaplastic tumors suggests that patient selection based on driver mutation could sharpen responses in future trials. The integrin signaling dominance in BRAF-mutant disease points toward matrix-targeting or anti-adhesion strategies for that subgroup. Third, the immune re-infiltration observed in both anaplastic lineages provides a mechanistic rationale for the emerging clinical interest in immune checkpoint inhibitors for advanced thyroid cancer, while explaining why such approaches have shown variable efficacy: the immune contexture differs by mutation, stage, and even within individual tumors.

The study also contextualizes why conventional therapies falter in advanced disease. Radioactive iodine uptake depends on functional thyroid differentiation, and both trajectories documented here culminate in the collapse of the thyroid differentiation program. The TDS metric, derived from the TCGA-THCA cohort, served as a quantitative spine for the entire analysis, and its decline along pseudotime tracked precisely with the loss of iodine-handling machinery. Mutation-specific kinase inhibitors have delivered clinical benefit in selected cases, but their efficacy wanes in dedifferentiated tumors, a pattern this study attributes at least in part to the fibroblast-mediated survival circuits that persist independent of the driver oncogene’s canonical downstream signaling.

Limitations remain. The single-nucleus cohort comprised sixteen tumors, necessarily small given the rarity of anaplastic specimens suitable for fresh-frozen multi-omics profiling, though the extensive external validation across hundreds of bulk and single-cell samples substantially mitigates this concern. The spatial platforms, Visium and GeoMx, resolve gene expression at spot or region level rather than true single-cell resolution, which the team compensated for through reference-based deconvolution. And as with all human tumor atlases, the trajectories described are inferential reconstructions rather than direct time-lapse observations of any single tumor’s evolution.

Nevertheless, the study delivers what the field has lacked: a unified, mutation-stratified map connecting the genomic events, epithelial state transitions, and microenvironmental rewiring that convert indolent thyroid cancers into killers. It reframes anaplastic transformation not as a single molecular event but as two distinct, mutation-scripted programs, each with its own pace, its own stromal conspirators, and its own vulnerabilities. For a disease where median survival in the anaplastic form remains measured in months, that map is not merely descriptive. It is a targeting system, pointing clinicians toward the right molecular door for the right patient, and reminding the field that in thyroid cancer, as in few other diseases, the first mutation a tumor acquires may determine the last chapter of its story.

TNBC remains one of the most aggressive forms of breast cancer, characterized by the absence of estrogen, progesterone, and HER2 receptors. This receptor-negative profile limits targeted treatment options, leaving chemotherapy as the primary systemic intervention. However, clinical outcomes to chemotherapy in TNBC are notoriously variable, suggesting underlying biological heterogeneity that has remained elusive until now. Recognizing this therapeutic challenge, the researchers sought a deeper comprehension of tumor-intrinsic and microenvironmental determinants driving response variability.

Leveraging fresh pre-treatment tumor biopsies from 101 TNBC patients, the investigators conducted single-cell RNA sequencing encompassing more than 427,000 individual cells. This comprehensive cellular atlas was complimented by spatial transcriptomic mapping of tumors from 44 patients, allowing the integration of gene expression data with cellular localization within the tumor architecture. A rigorous comparative analysis was performed against the Human Breast Cell Atlas, a reference database cataloging the normal breast tissue cellular milieu, enabling precise discrimination of malignant and non-malignant cell populations.

Through this large-scale cellular deconstruction, TNBC tumors were stratified into four archetypal profiles based on cancer cell transcriptional signatures. Crucially, a coherent set of thirteen highly expressed, cancer-specific genes emerged as a transcriptional signature underpinning these archetypes. This gene panel reflects a coordinated regulatory program influencing tumor cell phenotypes and their crosstalk with the surrounding microenvironment. Such molecular stratification advances beyond traditional histopathological classifications, offering a granular lens into tumor heterogeneity.

Integral to their findings was the characterization of macrophage populations within the TNBC tumor microenvironment. Macrophages, versatile immune cells known for roles in phagocytosis and immune regulation, exhibited distinct subtypes with divergent associations to therapy response. The study identified 49 immune cell states consolidated into eight spatially consistent cell neighborhoods, each correlating with specific cancer archetypes and neoadjuvant chemotherapy outcomes. Notably, certain macrophage subsets displayed gene expression programs linked to either pro-tumoral or anti-tumoral functions, suggesting their pivotal role in modulating chemotherapy efficacy.

Prevailing TNBC research has often focused on T cells within the tumor immune milieu; however, this comprehensive study illuminates the critical influence of macrophage heterogeneity. The discovery of macrophage-associated transcriptional signatures coexisting with cancer cell states sheds light on intricate tumor-immune interactions that may drive differential drug sensitivities. These insights underscore macrophages as potential biomarkers and therapeutic targets, offering avenues for immunomodulatory strategies tailored to TNBC’s complex ecosystem.

To translate these biological insights into clinically actionable tools, the researchers developed a machine learning model informed by the 13-gene transcriptional signature. This predictive model demonstrated robust capacity to forecast patient-level responses to chemotherapy prior to treatment initiation, paving the way for precision oncology approaches. By anticipating therapeutic outcomes, clinicians could potentially refine treatment regimens, avoid unnecessary toxicity, and enhance patient survival.

The methodological innovation of integrating single-cell genomics with spatial transcriptomics exemplifies a paradigm shift in cancer biology. This approach captures both gene expression nuances and tissue architecture, enabling a multidimensional understanding of tumor biology—a necessity for deciphering TNBC’s notorious heterogeneity. The scale and depth of this dataset represent one of the largest single-cell genomic efforts conducted in TNBC to date, setting a new benchmark for future studies.

Looking ahead, these findings hold promise for transforming TNBC management by enabling personalized treatment strategies informed by tumor-specific cellular and molecular features. While prospective clinical validation is requisite before routine adoption, the identification of macrophage subtypes and the gene panel offers a biologically rational foundation for new diagnostics and therapeutic innovations, including macrophage-targeted therapies and combination immunochemotherapy.

Dr. Nicholas Navin, chair of Systems Biology at MD Anderson, emphasized the novelty of this work in dissecting gene-expression programs and immune cell architecture in TNBC. Similarly, Dr. Clinton Yam, associate professor of Breast Medical Oncology, highlighted the potential of these discoveries to revolutionize treatment prediction and patient care, marking a significant stride toward individualized breast cancer therapy with improved efficacy and reduced morbidity.

This study was made possible through extensive collaborations and funding support from prominent institutions including the NIH, NCI, CPRIT, and multiple philanthropic foundations. The comprehensive author disclosures and detailed findings are accessible through the Nature publication, underscoring the rigor and transparency underpinning this seminal work.

In conclusion, this expansive investigation unravels the layered complexity of TNBC’s tumor microenvironment and cancer cell heterogeneity, spotlighting macrophage diversity and a targeted gene expression signature as key determinants of chemotherapy response. By integrating cutting-edge single-cell technologies with sophisticated computational models, this research paves the way for precision medicine approaches that could markedly improve therapeutic outcomes and quality of life for patients battling triple-negative breast cancer.

Subject of Research: Triple-negative breast cancer tumor microenvironment characterization and chemotherapy response prediction

Article Title: A 13-gene transcriptional signature and macrophage subtypes predict chemotherapy response in triple-negative breast cancer

News Publication Date: May 13, 2026

Web References:

• The University of Texas MD Anderson Cancer Center: http://www.mdanderson.org
• Nature publication: https://www.nature.com/articles/s41586-026-10469-9
• Human Breast Cell Atlas: https://navinlabcode.github.io/HumanBreastCellAtlas.github.io/

References:
Navin, N., Yam, C., et al. (2026). Single-cell transcriptional profiling identifies macrophage subtypes associated with chemotherapy response in triple-negative breast cancer. Nature. https://doi.org/10.1038/s41586-026-10469-9

Image Credits: The University of Texas MD Anderson Cancer Center

Keywords: Triple-negative breast cancer, chemotherapy response, tumor microenvironment, single-cell analysis, spatial transcriptomics, macrophages, gene expression, transcriptional signature, machine learning, cancer genomics, immuno-oncology, personalized medicine