A NOVEL MODELING APPROACH TO ELUCIDATE THE ROLE OF AUTOANTIBODIES IN COMPLEMENT ACTIVATION IN SLE

PV228 / #365 Poster Topic: AS23 - SLE-Diagnosis, Manifestations, & Outcomes Background/Purpose In SLE, autoantibodies (ANAs) can promote pathogenesis by forming immune complexes (ICs) that activate complement. While antibodies to DNA (anti-DNA) are known to be associated with complement levels, the role of other ANAs in activating complement is less clear. To elucidate better serological biomarkers in the context of novel therapies to decrease immunoglobulin levels or B cells, we modeled the relationship between autoantibodies (anti-DNA, other ANAs, anti-C1q) and complement. Methods Adult SLE patients (SLICC or ACR/EULAR criteria) were enrolled during routine clinic visits from June 2020 to June 2024. At each visit, treating rheumatologists scored the PGA and SLEDAI, medications were recorded, and autoantibodies were measured. Autoantibodies including anti-DNA, anti-RNA-binding proteins (RBPs), and anti-C1q were measured by ELISA. Complement activation was defined as (1) low C3, (2) low C4, (3) low C3 and low C4, and (4) low C3 or low C4. Potential predictors of complement activation were modeled in 4 steps: (1) anti-DNA; (2) anti-DNA, anti-RBPs (Ro-52, Ro-60, Sm, La, U1RNP, RNP-70), and anti-C1q; (3) anti-DNA, anti-RBPs, anti-C1q, and medications; and (4) anti-DNA, anti-RBPs, anti-C1q, and disease activity. To identify linear and possible nonlinear relationships between predictors and complement activation, we considered both generalized linear models (GLMs; specifically, logistic regression with LASSO regularization) and decision tree models, each with continuous predictors. Models were trained and tuned on 80% of patients and evaluated on the remaining 20%. Results The study included 526 visits in 257 patients (mean age 42 years; mean disease duration 13 years; 88% female; 58% Black, 29% White; 6% Hispanic). Almost one-quarter of visits had low C3 or C4; anti-DNA was positive at 40% of visits. In Lasso regression models, the presence of anti-DNA accurately predicted complement levels (AUC: 0.71-0.79; Table 1); model performance improved with the inclusion of anti-RBPs and anti-C1q (AUC: 0.73-0.84). The inclusion of medications or disease activity led to limited improvement in model performance. Across outcomes of complement activation, anti-DNA and anti-C1q were consistently associated with low complement (Figure 1). For the outcome of low C3, a 1 standard deviation increase in anti-DNA levels increased the odds of having low C3 by approximately 123%; a 1 standard deviation increase in anti-C1q levels was associated with an 82% increase in the odds of having low C3. Results were similar for low C4. The results of the decision tree models (Table 1) align with the findings from Lasso logistic regression models. For both low C3 and low C4, the decision tree consistently selected anti-DNA and anti-C1q as the primary splitting variables, further affirming their predictive power. Table 1. Model performance of serologies, medications, and disease activity to predict low complement. Figure 1. Coefficients of anti-DNA, anti-RBPs and anti-C1q on low C3 and low C4. Error bars indicate 95% confidence intervals obtained via bootstrapping (1000 resamples) the development set. Conclusions These results support the important role of anti-DNA antibodies in complement activation as reflected in levels of C3 and/or C4; the effects of other ANAs in the model were less marked, perhaps reflecting a more limited ability of these antibodies to form ICs that activate complement. The association of anti-C1q with low C3 and/or C4 is consistent with a role of this antibody in activating complement and suggests the value of assaying anti-C1q in studies on therapies that can impact autoantibody levels.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.