Background

Belimumab response in SLE remains unpredictable, with SRI-4 response rates of only 50–60% at 52 weeks. Standard flow cytometry analysis collapses high-dimensional immune cell phenotypes into predefined gates, discarding the geometric structure of the data. Topological data analysis (TDA), specifically persistent homology, captures multi-scale shape features — connected components (H₀), loops (H₁), and voids (H₂) — that are invisible to conventional statistical methods.

Hypothesis

Persistence diagrams computed from baseline peripheral blood immunophenotypic point clouds (≥30-parameter spectral flow cytometry) contain topological features — specifically H₁ loops in the B-cell/plasmablast/T-follicular helper subspace — that predict belimumab SRI-4 response at week 12 with AUROC >0.82, outperforming linear biomarker combinations (anti-dsDNA, C3/C4, BAFF levels) by ≥10 percentage points.

Mechanistic Rationale

BAFF blockade disrupts specific B-cell maturation trajectories. These trajectories form characteristic topological structures in phenotypic space: responsive patients should exhibit H₁ loops reflecting cyclic differentiation pathways (transitional → naïve → activated → plasmablast → memory → transitional) that are BAFF-dependent. Non-responders may show fragmented topology (fewer persistent H₁ features) indicating BAFF-independent survival niches or dominant extrafollicular pathways.

Proposed Methodology

1. Cohort: n=200 SLE patients initiating belimumab, ≥30-parameter spectral flow cytometry at baseline
2. TDA pipeline: Vietoris-Rips filtration on immunophenotypic coordinates (arcsinh-transformed, UMAP-aligned), persistence diagram computation via Ripser
3. Feature extraction: Persistence landscapes (PLs) at orders 1–5 for H₀, H₁, H₂; Betti curves; persistence entropy; bottleneck and Wasserstein distances from reference diagrams
4. Pharmacogenomic integration: FCGR3A V158F and TNFSF13B (BAFF) promoter variants as covariates in persistence landscape regression
5. Prediction model: Penalized logistic regression on vectorized PLs (L1 regularization) with nested 5-fold CV
6. Primary endpoint: SRI-4 at week 12; secondary: time to first flare over 52 weeks

Testable Predictions

• Baseline H₁ persistence in the B-cell subspace is significantly higher (permutation test, p<0.01) in week-12 responders vs non-responders
• Persistence entropy of the full immunophenotypic cloud at baseline correlates negatively with time-to-first-flare (Spearman ρ < −0.3)
• Addition of topological features to a model containing anti-dsDNA + C3 + BAFF improves net reclassification index (NRI) by >0.15
• FCGR3A V/V genotype modifies the association between H₁ persistence and response (interaction p<0.05)

Limitations

• Spectral flow cytometry with ≥30 parameters requires specialized equipment, limiting generalizability to resource-constrained settings
• Vietoris-Rips complexes on ~10⁵ cells are computationally expensive; subsampling strategies may lose rare populations
• Topological features are inherently non-interpretable without additional visualization (mapper graphs); clinician adoption may be limited
• Single-timepoint topology may miss dynamic trajectory information; serial sampling would strengthen the approach
• Confounding by corticosteroid dose and concurrent immunosuppressants must be addressed

Clinical Significance

Identifying belimumab non-responders at baseline would avoid 12+ weeks of ineffective therapy, reducing cumulative organ damage and healthcare costs. A topological biomarker signature could be computed from standard flow cytometry data (post-acquisition), requiring no additional patient burden. If validated, this framework generalizes to any biologic where target-dependent immune cell trajectory disruption is the mechanism.

RheumaAI Research • rheumai.xyz • DeSci Rheumatology