Immune Cells Feed Breast Cancer Tumors as AI and DNA Testing Reshape Cancer Care

Five new studies from Memorial Sloan Kettering Cancer Center are rewriting some of the most stubborn rules in oncology, and the findings read like a blueprint for how cancer medicine will be practiced in the coming decade. Published across Science Advances, Science Translational Medicine, Clinical Cancer Research, and Cell Reports Methods, the research spans tumor metabolism, radiation resistance, antibody-drug conjugates, precision oncology for rare gastrointestinal cancers, and an artificial intelligence system capable of watching life assemble itself cell by cell. Together, they share a single unifying idea: cancer cannot be understood, or defeated, by studying tumor cells in isolation. Whether the target is a nutrient-hungry breast tumor, a malignant seed drifting through cerebrospinal fluid, or a developing embryo, the surrounding ecosystem matters just as much as the cancer itself.

The most provocative of the new findings concerns triple-negative breast cancer, one of the most aggressive and least treatable forms of the disease. For years, drug developers have tried to starve triple-negative tumors by blocking glutamine, the amino acid these cancer cells rely on as their primary fuel. The strategy made elegant biochemical sense, yet it repeatedly stumbled in clinical trials. A team at MSK’s Sloan Kettering Institute, overseen by senior authors Justin Perry and Kayvan Keshari and led by postdoctoral researcher Nancy Santiappillai, has now explained why. Using advanced spatial metabolomics, a technique that maps the chemical activity of tissues with fine resolution, the researchers discovered that tumor-associated macrophages, immune cells that normally patrol and clean up within tumors, quietly sabotage the starvation strategy. Adapted to the oxygen-poor interior of the tumor, these macrophages rewire their own metabolism and release ornithine, an alternative nutrient that triple-negative breast cancer cells happily consume when glutamine runs dry. Even more striking, the macrophages engaged in clearing dead cells from the tumor release additional ornithine, effectively accelerating the very growth they were thought to suppress.

The therapeutic implications are considerable. When the team disrupted macrophage metabolism in mice using a drug called 6-AN, tumor growth slowed significantly in one triple-negative breast cancer model but not in another, a result that captures the humbling diversity of real tumors. In the responsive model, something remarkable happened: the macrophages flipped from a pro-tumor state into an anti-tumor one, and the number of cancer-killing CD8+ T cells inside the tumor rose. Targeting the metabolism of immune cells, in other words, may simultaneously cut off a hidden nutrient supply and awaken the immune system. But the effect was selective. Only tumors whose cells express high levels of two enzymes, OAT and PYCR1, in the proline synthesis pathway were vulnerable. Those enzymes help cancer cells tolerate redox stress, the damaging accumulation of unstable molecules, and enable construction of the extracellular matrix that tumors need to grow. An analysis of two large patient databases reinforced the clinical relevance: triple-negative breast cancer patients with high levels of those enzymes tended to have worse survival outcomes, tying the laboratory mechanism directly to human prognosis.

Elsewhere in the body, MSK researchers tackled one of oncology’s most feared complications: leptomeningeal metastasis, in which cancer spreads into the cerebrospinal fluid bathing the brain and spinal cord. Proton craniospinal irradiation, a form of radiation therapy, offers one of the few treatment options, but many cancers resist it. A team led by neuro-oncologist Adrienne Boire analyzed cerebrospinal fluid from patients both with and without leptomeningeal metastases, before and during radiation. The comparison revealed that cancer cells produce a signaling protein called CXCL1, which is linked to more aggressive disease and poorer responses to radiation. CXCL1 exerts its effects through a receptor called CXCR2, and that receptor became the team’s target. In mouse models, blocking CXCR2 with genetic tools or drugs suppressed the growth of leptomeningeal metastases. The combination of radiation with CXCR2 inhibition outperformed either treatment alone, raising the possibility that patients could receive lower radiation doses and thereby avoid some of the therapy’s most harmful side effects.

The translational pipeline moved with unusual speed. MSK has already opened a clinical trial testing radiation combined with SX-682, a CXCR2 inhibitor, in a collaborative effort among neuro-oncologist Jessica Wilcox, medical oncologist Monica Chen, thoracic medical oncologist Jark Jeng, and radiation oncologist Yao Yu. For a disease that has historically carried a prognosis measured in weeks, the leap from cerebrospinal fluid biomarker to active clinical trial represents the kind of rapid bench-to-bedside movement that precision oncology promises but rarely delivers so cleanly. If the trial confirms the preclinical results, patients with leptomeningeal disease could gain a strategy that sensitizes their tumors to radiation while sparing healthy tissue.

Breast cancer treatment, meanwhile, may be on the verge of its own diagnostic rethink. Trastuzumab deruxtecan, known as T-DXd or Enhertu, is an antibody-drug conjugate that uses an antibody to deliver chemotherapy directly into tumor cells, and it has already transformed outcomes in metastatic breast cancer. Yet its use is currently guided almost entirely by a single measurement: how much HER2 protein is visible on the surface of tumor cells. Approval was first limited to patients with high HER2 levels and later expanded to certain patients with lower levels, largely on the strength of MSK-led research. Now breast medical oncologists Joshua Drago, Nicholas Mai, and Sarat Chandarlapaty have identified a second, hidden indicator of benefit: mutations in the ERBB2 gene, which encodes the HER2 protein. In a study of 272 women with metastatic breast cancer lacking high HER2 levels, roughly 7 percent carried ERBB2 mutations, and those patients benefited from T-DXd for nearly twice as long as those without mutations, an average of 11 months compared with 6.2 months. In Chandarlapaty’s laboratory, cells carrying the mutation took up the drug faster and were killed more efficiently even when the amount of HER2 on the cell surface was identical.

The finding carries a direct message for clinical practice. Standard HER2 testing alone, the study suggests, may miss a group of patients who uniquely benefit from one of the most effective drugs available. Drago argues that genetic testing should be performed on the tumors of all patients with metastatic cancer, a stance that, if widely adopted, could reshape molecular diagnostics for breast cancer and push testing beyond protein-level measurements into the genome itself. It is a reminder that in modern oncology, the answer to whether a drug will work may be hiding in a mutation no one thought to check.

Genomic testing also anchors the fourth study, which offers hope to patients with biliary tract cancers, malignancies of the bile duct and gallbladder that are often diagnosed late and have few effective options once chemotherapy and immunotherapy fail. Gastrointestinal medical oncologist James J. Harding led an analysis of tumors from more than 1,200 patients with previously treated advanced disease, using MSK-IMPACT, a test that detects DNA changes across hundreds of cancer-associated genes. The team found that actionable DNA alterations were common across biliary tract cancer subtypes, identifying therapeutic targets including RAS alterations, MTAP deletion, and amplification of MDM2 and MET. Crucially, patients whose tumors harbored changes susceptible to targeted therapies lived longer without disease progression when matched to a drug aimed at their specific alteration than when treated with chemotherapy alone. By comparing tumor samples taken before treatment and after progression, the researchers also uncovered several mechanisms of resistance, insights that could guide the design of future combination therapies for this hard-to-treat disease.

The fifth study leaves human tumors behind entirely and looks at how a body builds itself. A team at the Sloan Kettering Institute has unveiled an artificial intelligence system called Twin Attention that can identify, track, and analyze individual cells in a developing embryo with striking accuracy. The system works by examining pairs of 3D snapshots of a growing embryo, learning to recognize each cell from its position and its relationships with neighboring cells, and assigning every cell a unique numerical fingerprint that captures not just location but tissue context. Demonstrated in the roundworm C. elegans, Twin Attention identified individual cells with roughly 93 to 97 percent accuracy as the embryo grew from a handful of cells to more than 500. Its most demanding test came in screening hundreds of embryos whose genes had been selectively silenced. A manual analysis of 700 such embryos would have consumed hundreds of hours; the AI system completed the task in a few hours, flagging developmental defects across the majority of genes tested and pinpointing 29 genes whose disruption delayed gastrulation, the pivotal step in which cells migrate into the embryo’s interior. Co-corresponding author Anthony Santella, a senior research scientist specializing in computer vision, says the tool could enable large-scale developmental screens that would otherwise be impossible, with applications in understanding how tissues form and malfunction in human disease. From starving tumors to decoding embryos, the through-line of this research is unmistakable: context is destiny, in biology as in medicine.

Subject of Research: Cancer immunometabolism, targeted therapy biomarkers, and AI-based developmental cell tracking

Article Title: MSK Research Highlights, October 9, 2026

Article References: MSK Research Highlights, October 9, 2026. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: triple-negative breast cancer, tumor-associated macrophages, glutamine metabolism, ornithine, spatial metabolomics, leptomeningeal metastasis, CXCL1, CXCR2 inhibitor, trastuzumab deruxtecan, ERBB2 mutations, biliary tract cancer, MSK-IMPACT