Baseline factors defining immune complex detection in systemic lupus erythematosus: development of a three-factor score in a retrospective cohort study

Introduction

Systemic lupus erythematosus (SLE) is an autoimmune disorder characterized by diverse autoantibodies and multiorgan involvement. In its pathogenesis, endogenous nucleic acids released from dying cells bind to anti-double-stranded DNA (dsDNA) or anti–single-stranded RNA antibodies, leading to the formation of immune complexes (ICs). These ICs are internalized by plasmacytoid dendritic cells and monocytes, inducing type I interferon (IFN), which plays a central role in SLE pathogenesis. Type I IFN further promotes B-cell survival and differentiation through the induction of B-cell activating factor (BAFF), resulting in further antibody production (1). In addition to their immunostimulatory effects, ICs may deposit directly in tissues such as the skin and kidneys, activate complement, and cause organ damage (2, 3).

Glucocorticoids (GCs) have long served as the cornerstone of SLE therapy; however, minimizing GC exposure is now regarded as essential in modern SLE management. Recent EULAR recommendations emphasize reducing GC doses to ≤ 5 mg/day and discontinuing them whenever feasible, supported by increased use of hydroxychloroquine, immunosuppressive agents, and biologic therapies targeting type I IFN and BAFF (4–6). Because even low-dose GC therapy (<5 mg/day) may contribute to irreversible organ damage (5), tapering and discontinuation of GCs should be considered once clinical remission, as defined by the Definitions of Remission in SLE (DORIS) criteria, is achieved (4, 5, 7).

Previously, we reported that 28% of patients with active SLE achieved clinical remission off corticosteroids (CR off C) in the era of biologics, and that IC detection (≥ 1.5 μg/mL) was the sole independent predictor of CR off C (6). Subsequent real-world experience has further suggested an association between IC elevation (>3.0 μg/mL) and the efficacy of belimumab (8). As the C1q-based IC assay is no longer commercially available in Japan, clarifying clinical and serological features associated with IC detection may help identify patients likely to have detectable ICs. Therefore, we aimed to (1) determine the baseline factors defining IC detection in patients with active SLE and (2) develop an exploratory score to estimate IC detection and evaluate its association with achievement of CR off C, as IC detection did in our previous study (6).

Patients and methods

Patient cohorts

We retrospectively identified patients with SLE treated at our tertiary rheumatology center and constructed two cohorts from this population for different purposes. The evaluation cohort was used to determine the proportion of patients achieving CR off C, whereas the derivation cohort was used to identify predictors of IC detection. The evaluation cohort was not designed as a validation cohort for the derivation cohort.

Patients were eligible for inclusion if they satisfied all of the following criteria: After applying the above criteria, the evaluation cohort included patients who were followed for ≥ 1 year, regardless of IC measurement. The derivation cohort included patients who met the same criteria and had circulating ICs measured, irrespective of the follow-up duration. Patients with rheumatoid arthritis overlap (rhupus) were excluded from both cohorts, whereas those with overlap syndromes such as Sjögren's syndrome, systemic sclerosis, or mixed connective tissue disease were included.

• diagnosis of SLE according to the 1982 revised or 1997 updated American College of Rheumatology classification criteria (9, 10);
• visited our institution between April 2012 and December 2023; and
• active disease requiring initiation or escalation of GC therapy.

These cohorts were not mutually exclusive; some patients were included in both cohorts depending on follow-up duration and availability of IC measurement.

Achievement of GC-free clinical remission

First, we evaluated the achievement of CR off C in the evaluation cohort as of December 31, 2024. Because the physician's global assessment was not routinely recorded in earlier periods, remission off treatment as defined by the DORIS criteria could not be assessed (11). Therefore, we used CR off C as the outcome, defined as the absence of clinical activity on the SLE Disease Activity Index 2000 (SLEDAI-2K) while receiving no corticosteroids, as previously described (12). Serological activity (elevated anti-dsDNA antibodies or hypocomplementemia) and the use of immunosuppressants or biologic agents were permitted under this definition. For SLEDAI-2K scoring, anti-dsDNA antibodies measured by either enzyme-linked immunosorbent assay (ELISA) using the MESACUP DNA-II TEST “ds” kit, with a reference range of ≤ 12 IU/mL (Medical & Biological Laboratories Co., Ltd., Tokyo, Japan), or anti-DNA antibody measured by radioimmunoassay (RIA), with a reference range of <6.0 IU/mL (DENIS Pharma K.K., Tokyo, Japan), were accepted instead of anti-DNA antibody measured by the Farr assay. Aldolase was accepted as a surrogate for creatine kinase when evaluating muscle involvement. If serological data were not available immediately before enrollment, the closest available values were used.

What defines IC detection?

Second, in the derivation cohort, we examined baseline clinical and immunological variables associated with IC detection, defined as circulating IC levels ≥ 1.5 μg/mL measured using a C1q solid-phase enzyme immunoassay (Fujirebio Holdings, Inc., Tokyo, Japan). Candidate variables included sex; age (dichotomized at the median value); time of enrollment (earlier vs. later than the median enrollment date); GC status (GC-naïve vs. GC-escalation); initial prednisolone-equivalent dose (≥ median vs. < median dose); total SLEDAI-2K score; each individual SLEDAI-2K item (13); and initial therapy with mycophenolate mofetil (MMF) or intravenous cyclophosphamide (IVCY). Immunological variables included anti-dsDNA antibody (ELISA), anti-DNA antibody (RIA), anti-RNP, anti-Sm, anti-SSA, and anti-SSB antibodies measured by immunoprecipitation; lupus anticoagulant measured by dilute Russell viper venom time or phospholipid neutralization; anti-cardiolipin IgG and anti-β2-glycoprotein I IgG; direct Coombs test (DCT); and complement levels with reference intervals of 73–138 mg/dL for C3, 11–31 mg/dL for C4, and 25–48 U/mL for CH50. These variables were screened to identify independent factors associated with IC detection for development of the predictive score.

Exploratory assessment of the association between the three-factor score and GC-free clinical remission

Finally, we explored the association between the three-factor score and GC-free clinical remission in the evaluation cohort.

Statistical analysis

For continuous variables, normally distributed data were presented as the mean ± standard deviation (SD), and non-normally distributed data as the median with interquartile range (IQR). The Student's t-test and Wilcoxon rank-sum test were used to compare normally and non-normally distributed data, respectively. Fisher's exact test was applied for categorical variables. Baseline factors with a potential association with IC detection in univariate logistic regression (p < 0.15) were considered candidates for multivariate analysis, given the absence of prior reference studies. For continuous variables whose positivity was significantly associated with IC detection, optimal cutoff values were determined using receiver operating characteristic (ROC) curve analysis with the Youden index. Values outside the assay's sensitivity range were replaced with the nearest measurable values (e.g., a value of 9.9 was substituted for “<10 IU/mL” for anti-dsDNA antibody), as such extreme values were unlikely to serve as optimal thresholds. When complete separation was observed (i.e., all male cases had detectable IC), Firth's penalized likelihood logistic regression was applied in both univariate and multivariate analyses to ensure stable estimation (14). To construct a predictive score for IC detection, independent factors were weighted according to the approximate magnitude of their β coefficients. The discriminatory performance of the predictive score for IC detection was assessed using the area under the receiver operating characteristic curve (AUC). Internal validation of the IC-detection model was performed using bootstrap resampling (1000 iterations) to assess the stability of the estimated discrimination. Achievement of CR off C was evaluated using Kaplan–Meier analysis. The association between the predictive score for IC detection and achievement of CR off C was assessed using the log-rank test, and crude hazard ratios (HRs) with 95% confidence intervals (CIs) were estimated using univariate Cox proportional hazards models. All analyses were performed using R (version 4.4.1; R Foundation for Statistical computing, Vienna, Austria). A p-value < 0.05 was considered statistically significant.

Results

Patient cohorts

One hundred and twenty-four patients were screened; 50 were excluded, and 74 were included in the study population (Figure 1). Reasons for exclusion were failure to meet the classification criteria (n = 2), absence of disease activity (n = 36), follow-up <1 year and no IC measurement (n = 9), and rhupus syndrome (n = 3). The included patients were categorized into three groups: followed for ≥ 1 year without IC measurement (n = 13), followed for ≥ 1 year with IC measurement (n = 57), and IC measured but followed for <1 year (n = 4). Accordingly, the evaluation cohort comprised patients followed ≥ 1 year regardless of IC measurement (n = 70), whereas the derivation cohort comprised patients with IC measured regardless of follow-up duration (n = 61). Of these, 57 patients were included in both cohorts. The derivation cohort was used to identify predictors of IC detection, whereas the evaluation cohort was used to assess CR off C.

Achievement of GC-free clinical remission

Patient characteristics of the evaluation cohort are summarized in Table 1. The median age was 42 years, and 87.1% of the patients were female. More than 75% of patients were GC-naïve, and over one-quarter had lupus nephritis. The median SLEDAI-2K score was 10. During follow-up, hydroxychloroquine, immunosuppressive agents, and biologics were used at least once in 82.9%, 78.6%, and 52.9% of patients, respectively. Twenty-four patients (34.3%) achieved GC-free clinical remission during a median follow-up of 4.4 years (1,616 days; IQR 975.75–2275.3). Among the 24 achievers, the median time to remission was 3.5 years (1,295.5 days; IQR 995.25–1803.3) (Figure 2).

Table 1. Patient characteristics of the evaluation cohort (n = 70).

IQR, interquartile range; GC, glucocorticoid; SLEDAI-2K, SLE disease activity index 2000; CR off C, clinical remission off corticosteroids; HCQ, hydroxychloroquine; IS, immunosuppressive drugs; MMF, mycophenolate mofetil; CNI, calcineurin inhibitors; AZ, azathioprine; IC, immune complex.

What defines IC detection?

Patient characteristics of the derivation cohort are shown in Table 2, stratified by the presence or absence of IC detection. In the IC group, male sex, GC-naïve status, high-dose GC, and initial therapy with MMF were more frequent than in the non-IC group. The SLEDAI-2K score was also significantly higher in the IC group. Among individual SLEDAI-2K items, proteinuria, pyuria, rash, mucosal ulcers, and low complement were significantly more common in the IC group. Although hematuria and leukopenia tended toward higher prevalence in the IC group, the difference did not reach statistical significance. Seizure, visual disturbance, and cranial nerve disorder were absent, while psychosis, organic brain syndrome, lupus headache, cerebrovascular accident, vasculitis, and urinary casts were infrequent; none showed an apparent association with IC detection (data not shown). Among immunological variables, anti-dsDNA antibody (ELISA), anti-cardiolipin antibody, and low C3, C4, and CH50 levels were more frequent in the IC group. The anti-SSA/Ro antibody and a positive DCT also tended to be more frequent in the IC group.

Table 2. Patient characteristics of the derivation cohort (n = 61).

SD, standard deviation; SLEDAI-2K, SLE disease activity index 2000; IQR, interquartile range; MMF, mycophenolate mofetil; IVCY, intravenous cyclophosphamide; GC, glucocorticoid; ELISA, enzyme-linked immunosorbent assay; RIA, radioimmunoassay; IC, immune complex.

The IC group was defined as patients with detectable circulating ICs, using a cut-off value of ≥ 1.5 μg/mL by C1q solid-phase enzyme immunoassay. The non-IC group was defined as patients without detectable circulating ICs.

Recent enrollment indicates cases enrolled after May 2, 2017.

Before analyzing predictors of IC detection using logistic regression, we determined optimal cutoff values for continuous variables that were significantly associated with detectable IC (Figure 3). This analysis included anti-dsDNA (ELISA), anti-cardiolipin antibody, C3, C4, and CH50. We also evaluated anti-DNA antibody (RIA), as it is considered clinically useful in Japan despite not reaching statistical significance (p = 0.155). ROC analysis identified optimal cutoff values as follows: anti-dsDNA (ELISA) ≥ 57.5 IU/mL; anti-DNA (RIA) ≥ 62 IU/mL; and anti-cardiolipin antibody ≥ 10.5 IU/mL. In the univariate logistic regression analysis (Table 3A), variables potentially associated with IC detection (p < 0.15) included: male sex, GC-naïve status, high-dose GC, initial therapy with MMF, hematuria, proteinuria, pyuria, rash, mucosal ulcers, pleurisy, leukopenia, anti-dsDNA antibody (≥ 57.5 IU/mL), anti-DNA antibody (≥ 62 IU/mL), anti-Ro/SSA antibody, anti-cardiolipin antibody (≥ 10.5 IU/mL), DCT positivity, low C3 (<68.5 mg/dL), C4 (<10.5 mg/dL), and CH50 (<30.15 U/mL) (Table 3A). These 19 variables were considered candidates for multivariate analysis. Given the sample size (29 IC-positive and 32 IC-negative cases), we limited the number of variables in the multivariate model to three, following the one-in-ten rule.

Table 3. Univariate (A) and multivariate (B) analyses of factors associated with IC detection (derivation cohort, n = 61).

CI, confidence interval; GC, glucocorticoid; RIA, radioimmunoassay; anti-β2GPⅠ, anti-β2 glycoprotein I antibody; MMF, mycophenolate mofetil; IVCY, intravenous cyclophosphamide.

Firth-adjusted logistic regression model.

Anti-dsDNA antibodies were measured by enzyme-linked immunosorbent assay (ELISA).

In the multivariate analysis (Table 3B), we initially focused on five variables that were highly significant in the univariate analysis (p < 0.001): high anti-dsDNA antibody (ELISA), high anti-DNA antibody (RIA), and low C3, C4, and CH50 levels. We tested four models sequentially. First, analysis of low C3, C4, and CH50 (model 1) showed that low C3 and low CH50 remained significant, whereas low C4 was not. Next, comparison of high anti-dsDNA antibody (ELISA) and anti-DNA antibody (RIA) (model 2) showed that only high anti-dsDNA antibody (ELISA) remained significant. Third, a combination of high anti-dsDNA (ELISA) with low C3 and CH50 levels (model 3) showed that high anti-dsDNA and low C3 remained significant, whereas low CH50 did not. Finally, analysis of low C3 and high anti-dsDNA (model 4) identified both as significant and independent predictors of IC detection. When we tested models incorporating each of the remaining 12 factors, excluding male sex and initial therapy with MMF, alongside the two independent predictors (low C3 and high anti-dsDNA antibody), none of the remaining factors were significant (data not shown). In contrast, male sex remained significant even after adjustment with the two independent predictors in the Firth-adjusted model (model 5). Initial therapy with MMF also remained significantly associated with IC detection in a similar model including anti-dsDNA and C3 (model 6; Table 3B). In model 5, the β coefficients for male sex (3.54), high anti-dsDNA antibody levels (2.99), and low C3 levels (2.57) were of comparable magnitude, supporting the use of a simplified scoring system in which one point was assigned to each predictor. Although initial therapy with MMF showed an association with IC detection, it was not included in the final score because treatment decisions may reflect underlying disease severity and physician preference, which could introduce confounding. Additionally, when all four variables (male sex, anti-dsDNA antibody, C3, and initial therapy with MMF) were included in a single model, all remained significantly associated with IC detection (data not shown). The three-factor score based on male sex, anti-dsDNA antibody ≥ 57.5 IU/mL, and C3 < 68.5 mg/dL showed the following proportions of detectable IC: 6.3% (1/16) for a score of 0; 11.8% (2/17) for a score of 1; 91.7% (22/24) for a score of 2; and 100% (4/4) for a score of 3 (Figure 4). The three-factor score showed excellent discrimination for predicting IC detection (AUC 0.93, 95% CI 0.86–1.00). Bootstrap validation with 1,000 resamples yielded similar performance (95% CI 0.86–0.99).

Association between the three-factor score and CR off C

When evaluating the association between the three-factor score and CR off C, higher scores were significantly associated with achieving CR off C (log-rank p = 0.04). In the Cox proportional hazards model, each additional factor was associated with a crude HR of 1.71 (95% CI: 1.04–2.80, p = 0.034) (Figure 5).

Discussion

In this retrospective cohort study of active SLE, 34% of patients achieved GC-free clinical remission over a median follow-up period of 4.4 years. Most patients received hydroxychloroquine, immunosuppressants, and biologics at least once during follow-up (83%, 79%, and 53%, respectively), often in combination, suggesting that GC-free clinical remission is achievable in one-third of patients with SLE in the biologic era. Our previous study indicated that IC detection independently predicted GC-free clinical remission (6). In the present study, we identified three baseline factors associated with IC detection: male sex, elevated anti-dsDNA antibody levels, and reduced C3 levels. The resulting three-factor score demonstrated high discriminatory performance for IC detection (AUC 0.93) and was also associated with GC-free clinical remission in an exploratory analysis. Given that the C1q-based IC assay is no longer available in Japan, this simple score may serve as a potential surrogate for identifying patients likely to have detectable ICs and for generating hypotheses about the relationship between the three-factor score and GC-free clinical remission.

The cutoff value of 57.5 IU/mL for anti-dsDNA antibodies to predict IC detection was 4.8 times higher than the upper limit of the reference value (12 IU/mL). This finding suggests that a relatively high anti-dsDNA antibody burden may be required for detectable circulating IC formation in SLE. In the classical pathway of complement activation, C1q activation requires the close apposition of two or more immunoglobulin Fc regions within ICs (15). Alternatively, larger ICs, formed only when both dsDNA antigen and anti-dsDNA antibodies are abundant, may structurally facilitate Fc clustering, thereby promoting C1q binding and activation (16). Notably, half of our patients received biologic treatment. Belimumab and anifrolumab have been shown to be more effective in patients with elevated anti-dsDNA antibodies or hypocomplementemia (17, 18), which may partly explain why elevated anti-dsDNA antibodies and hypocomplementemia were associated with CR off C in the present cohort. Although an earlier study raised concerns that ELISA-based anti-dsDNA antibody testing was not equivalent to the Farr assay (19), our data support the clinical utility of ELISA-based anti-dsDNA antibody measurement for estimating IC detection. This discrepancy may reflect improvements in the diagnostic performance of contemporary ELISA-based assays, which may now approach that of the Farr assay (20).

In the pathogenesis of SLE, autoantibodies serve as precursors to IC formation, whereas low complement levels reflect IC consumption (16). Therefore, the association of elevated anti-dsDNA antibody levels and reduced C3 levels with IC detection is biologically expected and should not be interpreted as evidence of a novel mechanism. Rather, the purpose of the score was to combine routinely available variables into a simple exploratory surrogate for IC detection in settings where direct C1q-based IC measurement is unavailable. We initially hypothesized that antibodies other than anti-dsDNA antibodies might also contribute to IC formation and complement activation. For example, anti-SSA antibodies have been linked to an enhanced type I IFN gene signature accompanied by hypocomplementemia, whereas anti-RNP antibodies have been associated with this signature without hypocomplementemia (21). However, in our multivariate analysis, the associations of anti-SSA antibodies, anti-cardiolipin antibodies, and DCT positivity with IC detection were confounded by elevated anti-dsDNA antibody levels and low C3 levels.

Previous studies have shown that men with SLE are more likely to develop nephritis, accumulate damage more rapidly, and have a poorer prognosis compared with women (22, 23). However, our previous findings paradoxically suggested that men with SLE had more favorable outcomes with respect to GC-free clinical remission in the biologic era (6). This may be explained by the fact that men with SLE are more likely to develop IC–mediated disease, such as lupus nephritis, which is characterized by substantial IC deposition (2). In our derivation cohort, all eight men had detectable IC. In the evaluation cohort, five of nine men (56%) achieved GC-free clinical remission, compared with 19 of 61 women (31%). Notably, male sex remained an independent predictor of IC detection even after adjustment for the two serological predictors. These findings may partly explain why inclusion of male sex in the three-factor score contributed to the association between the score and GC-free clinical remission.

Before interpreting the present results, several limitations should be acknowledged. First, our findings were derived from a retrospective cohort with a small sample size, which may have introduced bias. Second, multiple statistical analyses may have led to overfitting. Third, the assumption that IC detection predicts GC-free clinical remission has not yet been externally validated. Fourth, the derivation and evaluation cohorts substantially overlapped; 57 of 61 patients in the derivation cohort were also included in the evaluation cohort. Therefore, the association between the three-factor score and CR off C should not be interpreted as a validation analysis of the score, but rather as an exploratory analysis requiring confirmation in an independent cohort. Finally, treatment variables such as the initial use of MMF were associated with IC detection in exploratory analyses but were not incorporated into the predictive score, as treatment decisions may reflect disease severity and physician preference. Despite these limitations, our findings, based on more than a decade of real-world data, provide useful information for estimating IC detection in SLE. Further studies in larger, independent SLE cohorts are warranted to confirm these findings.

In conclusion, we identified three factors that characterize IC detection in SLE: male sex, elevated anti-dsDNA antibody levels, and reduced C3 levels. The resulting three-factor score provides a simple exploratory tool for identifying patients likely to have detectable ICs and for assessing their likelihood of achieving GC-free clinical remission.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Generative AI was used only for language editing and improvement of clarity. The authors take full responsibility for the content of the manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Ola Grimsholm, Medical University of Vienna, Austria

Reviewed by: Ana Kozmar, University Hospital Centre Zagreb, Croatia

Yuzhou Gan, Peking University People’s Hospital, China