CAR-T Reaches the Synovium: COMPARE Phase 1 in ACPA-Positive RA Save

Treatment-refractory rheumatoid arthritis (RA) is more common than we sometimes admit. Roughly 11% of patients fail multiple b/tsDMARD classes, and for the ACPA-positive subset, rituximab has long been the fallback. It works, but it rarely produces durable disease-free intervals. CD19 CAR T-cell therapy was always going to be the question that followed the SLE data. Until EULAR 2026, that question had no prospective answer. The plenary session changed that.

Albach and colleagues from the Charité group in Berlin presented the phase 1 results of the COMPARE trial (OP008), the first investigator-initiated, prospective study of CD19 CAR-T cell therapy in treatment-refractory, ACPA-positive RA. Six patients received a single infusion of mivocabtagene autoleucel, a fully human CD19 CAR-T product (miv-cel, KYV-101), after standard lymphodepletion and discontinuation of all DMARDs. The phase 2 is already enrolling, randomising ten patients head-to-head against rituximab with cross-over to miv-cel for non-responders after 24 weeks.

The cohort was sick by every measure. Median age 52, median disease duration 10 years, one to seven previous b/tsDMARD classes including rituximab in part of the cohort. Baseline DAS28-CRP ranged from 4.1 to 6.4, with ACPA titres into the high hundreds and one patient above 1300 U/mL. Inclusion required failure of at least two b/tsDMARD classes plus sonographic synovitis. The trial enrolled the hard end of refractory RA, not the borderline cases.

Safety came out clean. All six patients developed CRS, grade 1 in four and grade 2 in two. All received tocilizumab and dexamethasone by protocol, not in response to symptoms. No ICANS. Early neutropenia was universal and managed with G-CSF, and five of six patients had mild hypogammaglobulinemia between 5 and 7 g/L. Notably, follow-up now extends to 52 weeks in the first two patients with no severe infections, malignancies, or unexpected late events.

Efficacy was striking for a phase 1. DAS28-CRP fell across all six patients into the low-disease-activity or remission zone by week 24, with ACR50 and ACR70 climbing progressively through week 36. The serology was where the data turned heads. ACPA seroconversion was sustained in four of six patients, RF-IgM in five of six, with RF-IgA falling in parallel. For a disease in which chronically elevated autoantibody titres typically persist even after rituximab, that depth of clearance is not what we have come to expect.

The tissue and imaging data carried the heaviest weight. Albach and colleagues showed paired bone marrow and synovial biopsies, before and after infusion. Post-infusion samples were essentially empty of CD19+ B cells in both compartments. This is the first direct demonstration that CD19 CAR-T reaches the synovium in RA, a question the field has been wanting to answer for two years. 68Ga-FAPI PET/CT showed imaging-confirmed reduction of fibroblast activation by week 22, and ultrasound documented collapse of synovial vascularity by week 16. The imaging is starting to corroborate what the clinical scores already say.

The B-cell biology closed the loop. Reconstitution between weeks 36 and 52 was dominated by naïve and transitional B cells, with marked contraction of memory subsets. And in the patients with available samples, CCP4-tetramer staining showed near-complete elimination of CCP4-reactive memory B cells, the autoreactive compartment specifically, not bulk B cells. Mechanistically, this is the cleanest evidence yet that CD19 CAR-T does in RA what it appears to do in lupus: deletes the pathogenic clones and rebuilds the repertoire with a more naïve, less autoreactive profile.

For the practising rheumatologist, the immediate clinical implication remains narrow. Six patients is a phase 1 cohort, and CAR-T programs are not within reach of most centres yet. For the very specific patient, ACPA-positive RA that has failed multiple b/tsDMARD classes with active sonographic synovitis and a serologic burden resistant to every B-cell depleter we have, there is now prospective interventional data with a randomised comparator already enrolling. Whether that turns into a referable option or stays a proof of concept depends on the phase 2 readout.