I’d always filed the joint pain from cancer immunotherapy under “rough patch.” The drug overshoots, a knee swells, steroids settle it, and everyone moves on. I never asked the obvious question, which is why so many of those knees swell up again.
A team at MD Anderson and Yale went looking for the answer inside the joint. In a paper in Cancer Immunology Research, Synat Keam, Roza Nurieva, Sang Taek Kim and colleagues drew joint fluid from six patients with checkpoint-inhibitor arthritis twice, once at the first flare and again when it returned. Many of the most expanded T cell clones in the second sample had already been caught in the first, and they carried a stronger inflammatory signature the second time around.
These relapses aren’t rare. MD Anderson’s release says roughly 20 to 50 percent of patients who develop this arthritis go on to have more than one flare. At those odds I can’t keep calling it a side effect that passes. For a large share of patients it behaves more like a condition the drug installs.
of patients with checkpoint-inhibitor arthritis flare more than once
- Lower bound
- 20
- Upper bound
- 50
Matching barcodes in the knee
The group put the fluid through single-cell RNA sequencing, immune-receptor sequencing and flow cytometry, and compared it against joint fluid from six cancer-free osteoarthritis patients. Two populations kept turning up. One was effector CD8 T cells, the immune system’s killers. The other was a specialized group of CD4 T cells carrying high levels of PD-1 and CXCL13, a chemical signal that pulls other immune cells into the tissue.
But wait: how do you know a cell at the second flare belongs to the same family as one from the first, and isn’t just a lookalike? This is my favorite trick in immunology. Each T cell builds its receptor through a one-time genetic shuffle, and every daughter cell inherits that exact sequence, so the receptor works as a barcode for the whole family. In the team’s ACR Convergence 2025 abstract, 93 percent of the top 10 expanded effector CD8 clones were shared between the first and second flare. For the PD-1-high, CXCL13-high CD4 cells it was 50 percent. Receptor sequencing covered only three of the six patients, so this is a handful of joints, but the overlap is hard to miss.
| Group | Value (percent) |
|---|---|
| Effector CD8 clones | 93 |
| PD-1-high, CXCL13-high CD4 clones | 50 |
The barcodes show a lineage, not one cell camping out. Matching sequences mean the same clonal families expanded at both flares. They don’t prove that any single cell stayed in the knee the whole time. The authors’ explanation, which the release sums up as arthritis that “behaves like an immune-memory disease,” is that the immune system is answering the same target a second time.
That second answer is louder. According to the paper’s abstract, the dominant clones carried type 1 inflammatory cytokine signatures, including interferon gamma, TNF and IL-21, and these were strongest at the second flare. The clones were also wired into dense signaling with one another and with myeloid cells and B cells through chemokines such as CXCL9, CXCL10 and CXCL13. The release adds an odder detail. During the recurrent flares, regulatory T cells, whose job is to quiet an immune response, were producing inflammatory molecules too. In these six patients, at least, the brakes were feeding the fire.
The same memory, pointed at the knee
Nurieva states the irony herself in the ScienceDaily summary and the release: “Cancer immunotherapy works by creating a lasting immune response against tumors, but that same persistence may also contribute to recurring inflammatory side effects that can be debilitating for patients.”
Durability is what checkpoint inhibitors are sold on. You take the brakes off T cells, they learn the tumor, and that memory keeps protecting you long after the last infusion. There’s no way to order only the memory you want. When released T cells turn on joint tissue, this work suggests they can learn that target too. “Immune-related adverse event,” the phrase in these drugs’ safety tables, makes the whole thing sound like weather that blows through.
Rheumatologists have seen the clinical version for years. In a Johns Hopkins cohort reported by Braaten and colleagues, 70 percent of patients who came back for follow-up still had active arthritis 3 months after stopping immunotherapy, and 48 percent still did at 6 months. Longer treatment, combination immunotherapy and multiple other immune side effects all made improvement less likely. At EULAR 2026, a Leeds team following more than 200 patients with checkpoint-inhibitor arthritis or joint pain reported that about 57 percent had persistent symptoms at a median follow-up of 16 months. The same team says joint pain without inflammatory arthritis may affect up to six times as many patients as the arthritis itself.
The cellular trail runs back further. In 2020, investigators at HSS and Brigham and Women’s described an unusual population of cytotoxic, proliferating CD38-high CD8 T cells in 10 patients with this arthritis. Nobody has shown those are the cells the MD Anderson group tracked. Both findings, though, point to an aggressive killer T cell compartment expanding in an inflamed joint after these drugs.
Calming the knee without calling off the tumor attack
The quick fix for an inflamed joint is broad immune suppression, and that’s exactly what you’d hesitate to give someone whose cancer treatment is their immune system.
The warning people usually cite comes from a different organ. In a 2018 retrospective study in Cancer, Alexander Faje’s group at Massachusetts General Hospital reviewed melanoma patients who developed pituitary inflammation on ipilimumab. Patients given high-dose steroids had a median overall survival of 23.3 months, while the low-dose group’s median hadn’t been reached (hazard ratio 0.24). That’s an association in a different side effect, and the Hopkins arthritis cohort found that the drugs used for the arthritis didn’t appear to affect tumor response. The worry persists anyway, which is why a narrower tool would be worth having.
The authors want to build that tool. If effector CD8 cells and PD-1-high, CXCL13-high CD4 cells drive the relapses, MD Anderson says they could flag high-risk patients as biomarkers and guide targeted therapies that suppress the arthritis while preserving the anti-tumor response. For now that’s a hypothesis. This was a cell-mapping study of six people. It tests no therapy, and the release concedes that nobody yet knows whether these cells can reliably predict a future flare.
On the money: the listed funders are federal agencies, cancer and rheumatology foundations, family funds and MD Anderson itself, and no drugmaker appears among them.
What the study changes is the word “unlucky.” A patient who flares twice may be meeting the same immune families twice. If I were starting a checkpoint inhibitor, I’d ask for a rheumatologist’s name before the first infusion, and I’d treat a first swollen joint as the start of a file, not the end of an episode.
Sources
- Keam et al., Cancer Immunology Research (2026)
- Keam et al., ACR Convergence 2025 abstract
- MD Anderson release via Newswise
- ScienceDaily summary (2026)
- Johns Hopkins Rheumatology, Braaten et al. (2020)
- RheumatologyLive, Leeds cohort at EULAR 2026
- HSS, unique immune cells in checkpoint inhibitor-induced arthritis (2020)
- Healio, Faje et al. in Cancer (2018)