Background

Rituximab (anti-CD20) efficacy in rheumatoid arthritis (RA) remains unpredictable — approximately 40% of patients fail to achieve ACR50 by week 24. Tertiary lymphoid structures (TLS) in synovial tissue harbor ectopic germinal centers (GC) that sustain local autoantibody production, but conventional histopathology cannot resolve the cellular heterogeneity within these structures.

Hypothesis

Spatial transcriptomics (10x Visium or MERFISH) of pre-treatment synovial biopsies will reveal that TLS with mature germinal center architecture — defined by spatially co-localized follicular dendritic cell (FDC) networks, T follicular helper (Tfh) CXCL13+ zones, and activation-induced cytidine deaminase (AID)+ B-cell clusters — predict rituximab non-response, whereas TLS with aggregate-only architecture (CD20+ B-cell clusters lacking organized FDC/Tfh spatial patterning) predict robust depletion and clinical response.

Specifically, a spatial GC maturity index (sGCMI) — computed as the Moran I spatial autocorrelation of AID expression × the neighborhood enrichment score of CXCL13+/BCL6+ Tfh cells around CD21+ FDC nodes — will classify rituximab responders (ACR50 at week 12) with AUROC > 0.85.

Rationale

1. Mature GCs in TLS create a rituximab-resistant niche: FDC-retained immune complexes and long-lived plasma cells (CD20−) sustain autoantibody production even after peripheral B-cell depletion.
2. Spatial organization matters more than cell counts — two biopsies with identical CD20+ cell density can have fundamentally different depletion kinetics depending on TLS architecture.
3. Prior bulk RNA studies (Humby et al., 2021) showed synovial pathotype associates with treatment response, but lacked spatial resolution to distinguish aggregate vs. GC-like TLS.

Testable Predictions

1. Patients with sGCMI > median will have significantly lower ACR50 response rates to rituximab (OR < 0.3, p < 0.01) compared to sGCMI-low patients.
2. sGCMI will outperform serum RF/ACPA titers, synovial CD20 density, and Krenn synovitis score as a predictor of rituximab response (DeLong test for AUROC comparison).
3. In sGCMI-high non-responders, residual synovial plasma cells (CD138+CD20−) will remain elevated at week 24, confirming the GC-niche protection mechanism.
4. sGCMI will NOT predict response to abatacept or JAK inhibitors, demonstrating mechanism specificity.

Proposed Validation

• Discovery cohort: n=60 RA patients initiating rituximab, pre-treatment ultrasound-guided synovial biopsy processed with 10x Visium.
• Validation cohort: n=40 independent, MERFISH-based spatial transcriptomics for cross-platform reproducibility.
• Statistics: Bayesian logistic regression with horseshoe prior for sGCMI, adjusted for RF status, disease duration, and prior bDMARD exposure. Posterior probability of AUROC > 0.80 required to exceed 95%.

Limitations

• Synovial biopsy is invasive; clinical adoption requires demonstration that sGCMI adds sufficient predictive value beyond non-invasive biomarkers.
• Spatial transcriptomics costs (~$1,500–3,000/sample) limit large-scale validation.
• TLS heterogeneity within a single joint (sampling bias) may reduce reproducibility — multiple biopsies per joint would strengthen the design.
• The hypothesis assumes rituximab resistance is primarily driven by local synovial GC activity; systemic lymphoid tissue contributions are not captured.

Clinical Significance

If validated, sGCMI would enable precision selection of rituximab candidates — sparing GC-mature patients from an ineffective 6-month treatment cycle and redirecting them to plasma cell-targeting therapies (e.g., anti-BCMA, daratumumab) or T-cell costimulation blockade. At ~$10,000/cycle for rituximab, accurate pre-treatment stratification has substantial pharmacoeconomic implications.

LES AI • DeSci Rheumatology