BIOANALYSISPAPER
Single cell RNA and T cell receptor (TCR) sequencing, which resolves immune activity cell by cell, from a 14 patient Novartis trial shows that sabatolimab, an antibody against TIM3, a checkpoint receptor expressed on T cells and NK cells,
A patient with a rare T-cell leukemia entered a 14-person Novartis trial with two drugs she had no reason to expect much from. She left in complete remission 23 months later, and the single-cell map of her blood now shows why: sabatolimab, an antibody against an immune checkpoint called TIM3, rewired her immune response through natural killer (NK) cells, fast-acting innate immune cells, and cytotoxic CD4 T cells, rather than the exhausted CD8 T cells that PD-1 drugs are designed to revive.
The data come from a phase Ib trial of sabatolimab plus decitabine, an epigenetic drug that reactivates silenced tumor-suppressor genes, in 14 patients with acute myeloid leukemia (AML, fast-progressing blood cancer) or higher-risk myelodysplastic syndrome (MDS, a precursor condition in which the bone marrow fails to make healthy blood cells), reported in Cancer Immunology Research. Six of those patients also had paired single-cell RNA and T-cell receptor sequencing on bone marrow and blood, the level of resolution that lets researchers see which immune cell type is doing the actual work.
PD-1 and CTLA-4 inhibitors, the checkpoint drugs most readers have heard of, work by releasing exhausted CD8 T cells, the "killer" T cells that have been silenced by chronic tumor exposure. TIM3 is expressed much more broadly, on NK cells, myeloid cells, and T cells alike, and sabatolimab's effect concentrated on cytotoxic NK subsets and on type I interferon signaling, an antiviral program that also primes tumor killing, not on the canonical CD8 exhaustion program. Fewer than 1% of bone marrow CD8 T cells in the cohort even showed a canonical exhaustion phenotype, which is part of why the canonical drugs have struggled in this disease.
The T-LGLL patient is the cleanest case. T-LGLL, or T-cell large granular lymphocytic leukemia, is itself a clonal expansion of cytotoxic T cells, a slow-growing immune cancer sitting on top of an acute blood cancer. She had pre-existing CD4+ T-LGLL, and on sabatolimab plus decitabine more than 20% of her circulating lymphocytes were T-LGLL cells whose T-cell receptors recognized her own blast cells. She cleared her AML and stayed in complete response for 23 months. The authors, working with the ASH 2025 abstract on the same anti-TIM3 biology, argue the responder profile generalizes: patients whose anti-tumor activity runs through NK cells or cytotoxic CD4 T cells, rather than exhausted CD8s, are the population to enrich for.
The honest counterweight is the cohort size. Fourteen patients, with the deep single-cell and TCR work on only six of them, is the kind of evidence that opens a question rather than closes one. The T-LGLL responder is one person; the cytotoxic CD4 expansion in other responders is a trend, not a verdict. The STIMULUS-AML1 phase II primary results presented by Amer Zeidan at EHA 2025 are the larger trial context for sabatolimab in higher-risk MDS, and the program-level signal there is what Novartis acted on.
That program-level signal is the part the wire will not make sense of without help. Novartis has publicly discontinued sabatolimab in MDS, framing its ASH and SABCS 2025 data around other pipeline assets rather than a continued MDS bet. The mechanism paper and the discontinuation news point in opposite directions and both are true: a small trial produced a clean biological finding, and a large trial produced a result Novartis decided was not enough to carry the program forward. Sabatolimab's Adis Insight drug profile records its original positioning for patients unfit for hematopoietic stem cell transplant, a population with limited alternatives, which is what made the discontinuation decision notable rather than routine.
The constructive next step is also a question the field now has to answer. If the active anti-tumor cell in this mechanism is an NK cell or a cytotoxic CD4 T cell rather than an exhausted CD8, then the standard immunotherapy playbook does not apply: measure PD-L1, look for high mutational burden, prioritize tumors with T-cell infiltrates. The next trial design question is who to pre-select, and how to measure NK- or CD4-driven anti-tumor activity in a screening assay, before enrollment rather than after.
For now, the takeaway is portable. There is a third checkpoint axis, it runs through innate-like and helper T cells rather than the famous exhausted CD8s, and a single dramatic responder in a 14-patient trial is the cleanest case so far that the field should be designing around that axis. The work left to do is not "does anti-TIM3 work" but "in which patients, and on what measurement."