Introduction
Systemic lupus erythematosus (SLE) is a complex, chronic autoimmune disease with diverse signs and symptoms that commonly affect multiple organs and tissues. SLE has an unpredictable course, with periods of flares and remissions. High-titer autoantibodies targeting nuclear antigens, including DNA, RNA, histones and ribonucleoproteins (RNP), are a defining feature of SLE. Prior to diagnosis with SLE, patients gradually accumulate new autoantibodies, and have an average of three (from Ro, La, antiphospholipid (APL), antinuclear antibody (ANA), dsDNA, Smith, and RNP) at diagnosis [
1]. Many patients likely have additional autoantibodies, as >100 autoantigens have been described in SLE [
2]. Levels of autoantibodies fluctuate with disease activity and are associated with specific organ involvement in SLE [
3]. Autoantibodies can directly cause pathology in SLE, as a human anti-DNA monoclonal antibody was capable of initiating early-stage lupus nephritis (LN) in severe combined immunodeficiency (SCID) mice [
4].
Ten to twenty percent of SLE patients have disease onset in childhood or adolescence. Pediatric SLE (pSLE) patients often initially present with more acute and severe disease than adults [
5], including a higher frequency of LN observed at presentation [
6,
7]. LN is one of the primary causes of morbidity and mortality in pSLE [
8]. Clinicians regularly evaluate urinary parameters, including hematuria, pyuria, cellular casts and proteinuria, to aid in the diagnosis and monitoring of LN. However, these metrics have low accuracy, especially in the context of monitoring for renal flare [
9]. Candidate biomarkers for LN in pSLE include antibodies against dsDNA [
3,
10], complement C3 and C4 levels [
10], urine mRNAs [
11], urine chemokines [
12,
13], and urine proteins/peptides [
14,
15]. While measurement of anti-dsDNA and complement C3 and C4 levels are commonly available clinical laboratory tests, only 50 % of LN patients display a decrease in C3 and C4 or increase in anti-dsDNA antibodies concurrent with a flare [
9,
16]. While multiple factors influence the development of LN, including complement, autoantibodies, environment, and genetics [
17], the majority of these approaches only measure single analytes, and may not capture the clinical heterogeneity in SLE.
Autoantigen microarrays allow highly multiplexed measurement of serum autoantibodies that recognize purified or recombinant protein and nucleic acid-containing autoantigens. Our group has developed microarrays to measure autoantibodies targeting known autoantigens [
18,
19], cytokines and chemokines [
20], and modified peptides [
21]. This platform enables the characterization of multiple autoantibodies in parallel, while using microliter amounts of patient sera. To our knowledge, autoantigen microarrays have yet to be used to identify autoantibodies associated with pSLE or predictive of pSLE LN. An advantage of using highly multiplexed experimental platforms is that they can be used to identify multianalyte signatures or scores associated with clinical features of SLE. For example, gene expression microarrays were used to identify the interferon (IFN) signature, associated with active and severe forms of SLE, and protein microarrays were used to establish the chemokine score, associated with disease activity and predictive of flares in SLE [
22‐
25].
In-depth knowledge of the diverse profiles of autoantibodies present in the serum of pSLE patients will increase our understanding of SLE, and aid in disease diagnosis and prognosis. There is significant interest in identifying autoantibody profiles that are associated with LN and predictive of renal flares, with a goal to enable preemptive treatment. In the current study, we utilized autoantigen microarrays to identify novel serum autoantibodies associated with pSLE. This included anti-B cell-activating factor (BAFF), which we found was associated with active disease. We also identified autoantibodies associated with proliferative nephritis in pSLE, including autoantibodies against aggrecan and collagens IV and X. Using autoantibody measurements in combination with clinical information, we developed a nephritis score that identified patients who have proliferative nephritis with high accuracy. In the future, a similar score could be used to enhance the diagnosis and monitoring of proliferative nephritis in pSLE patients.
Discussion
In this study, we analyzed a new pSLE cohort using autoantigen microarrays, and identified serum autoantibodies that are associated with clinical manifestations of pSLE. We identified BAFF as a novel autoantigen in pSLE and found that the presence of BAFF autoantibodies was associated with active disease. We found that autoantibodies to collagen IV, collagen X and aggrecan were associated with proliferative nephritis in pSLE. Using autoantibody measurements and clinical information, we created a combined signature capable of accurately identifying pSLE patients with proliferative nephritis.
This is the first report of a new pSLE cohort, which includes a repository of >1087 longitudinal serum samples from 122 pSLE patients, including 71 new-onset patients. Extensive clinical information was collected at each patient visit, covering up to 8 years of clinic visits. An exceptional feature of the cohort is that autoantigen microarray analysis has now been performed on >100 of the patient samples, making in-depth investigation of the role of autoantibodies in pSLE possible.
To our knowledge, this is the first description of using autoantigen microarrays to identify autoantibodies associated with pSLE or predictive of pSLE LN. Our analysis comparing serum IgG reactivity between pSLE patients and healthy controls identified candidate autoantigens that have not been reported in pSLE, including collagen type X, OGDC-E2, sp100, PL-12, SRP54 and BAFF (Fig.
1). Our group previously showed that BAFF autoantibodies are present in the serum of adult SLE patients, and that the autoantibodies were capable of blocking stimulation of the BAFF receptor [
20]. In agreement with these findings, we found significantly higher levels of BAFF autoantibodies in the serum of pSLE patients than in healthy controls (Fig.
2a). Further, we found that higher levels of IgG reactivity were significantly associated with elevated SLEDAI scores. This parallels our group’s report that the presence of BAFF autoantibodies was associated with activation of the type I IFN pathway in adult SLE patients [
20,
22]. These findings are important in light of the fact that BAFF is the target of a fully human, recombinant monoclonal antibody, belimumab, which was recently approved for the treatment of SLE [
42]. Further studies will be required to determine whether SLE patients with BAFF autoantibodies respond differently to belimumab, compared to patients without BAFF autoantibodies.
Unsupervised hierarchical clustering of the serum autoantibodies of pSLE patients and healthy controls revealed multiple distinct patient clusters (Fig.
1). The clusters corresponded to groups in which the predominant autoantibody reactivity was to Sm/RNP-related antigens, DNA/histone-related antigens, Ro/La, or RiboP, or was relatively low to most antigens. This clustering pattern bears similarities to autoantibody clusters previously identified in adult SLE [
43,
44] and pSLE [
10,
45,
46]. We compared SLEDAI scores and the frequency of proliferative nephritis between pSLE patients within each cluster. We found the low reactivity cluster had lower SLEDAI scores, and the DNA/histone-related antigens cluster had an increased frequency of nephritis, although neither reached statistical significance (data not shown). Other groups have observed a similar association between DNA/histone-related antigen clusters and nephritis or evidence of nephritis [
43,
44,
46]. We are currently investigating the association of these clusters with other clinical parameters, including evidence of an association between antibodies to mumps and measles viruses and the DNA/histone cluster.
To identify autoantibodies associated with proliferative nephritis, we compared the serum IgG reactivity of pSLE patients with biopsy-proven proliferative nephritis, and those with either biopsy-confirmed class II nephritis or no significant evidence of proliferative nephritis (Fig.
3). Similar to previous reports, we identified an association between dsDNA [
3,
10] and C1q [
10,
37,
38] autoantibodies with LN in pSLE (Figs.
3 and
4). Autoantibodies to dsDNA are used in the diagnosis of SLE, fluctuate with disease activity, and are associated with kidney involvement in pSLE [
3]. A positive correlation has been observed between glomerulonephritis and C1q autoantibodies in adult SLE [
37]. While C1q autoantibodies were found in pSLE patient serum, the same association with LN was not observed in one report [
47], but a significant association with LN was observed in more recent studies [
10,
48,
49]. Similar to previous reports, we observed an association between LN and autoantibodies to alpha-actinin and fibrinogen [
39‐
41]. Alpha-actinin is an actin-binding protein and member of the spectrin family of proteins. A fraction of dsDNA autoantibodies in the sera of SLE patients also bind alpha-actinin [
39]. Presence of these cross-reactive autoantibodies was associated with a higher frequency of glomerulonephritis [
39]. Patients in our cohort may also have had cross-reactive autoantibodies, as there was a relatively strong relationship between serum levels of anti-dsDNA and anti-alpha-actinin (R
2 = 0.594,
p <0.001).
In addition to autoantigens known to be associated with LN, we found multiple autoantigens that have not been previously associated with proliferative nephritis, including aggrecan, collagen IV and collagen X (Figs.
3 and
4). Aggrecan is the major proteoglycan in articular cartilage, and contributes to its resilience. Aggrecan autoantibodies have been described in SLE, rheumatoid arthritis, systemic sclerosis, Sjögren’s syndrome, and ankylosing spondylitis [
50]. Collagen IV is a structural protein that forms networks in the glomerular basement membrane. Autoantibodies to collagen IV have been reported in SLE, although their relation to LN was not discussed [
51]. Autoantibodies to collagen IV are more commonly associated with Goodpasture’s disease, a rare autoimmune disease in which autoantibodies damage the lung and kidneys. Collagen X is a homotrimeric short-chain collagen that is primarily expressed by chondrocytes, and is a structural component of articular cartilage. It has previously been described as an autoantigen in type 1 diabetes [
52]. Interestingly, we found that autoantibodies to collagen X were also higher in pSLE patients than in healthy controls (Fig.
1). We are currently evaluating whether aggrecan and collagens IV and X are similarly associated with LN in adult SLE.
Pathogenesis of LN is influenced by multiple pathways and factors, including complement, autoantibodies, environment, and genetics [
17]. We utilized a stepwise regression method, called LASSO, to generate a predicative model of proliferative nephritis based on multiple clinical and experimental variables (Fig.
5a and Figure S2 in Additional file
1). LASSO constrains the sum of the absolute values of the regression coefficients, shrinking the coefficients of redundant or uninformative variables to zero, resulting in a sparse model. In this way, LASSO models tend to be simplified, interpretable, and efficient. Our model correctly categorized 91 % (21/23) of patients in an independent test set with 100 % sensitivity and 87 % specificity (Fig.
5a). For comparison with other clinical measures, see Figures S1 and S3 in Additional file
1. Our analysis included samples from four new-onset pSLE patients who were suspected to have nephritis, but were not confirmed by biopsy due to inconclusive biopsy, bleeding risk, or heart involvement (Fig.
5a). Our results suggest that the patient with an inconclusive biopsy may not have needed cytotoxic therapy.
Current clinical methods perform poorly at monitoring renal progression, and there is a need for new methods that might allow preemptive management of renal flare. While our longitudinal study had a small sample size, we provide evidence that our nephritis score is predictive of renal flare (Fig.
5b). We found that nephritis scores increased to levels predictive of proliferative nephritis between 43 and 71 days before biopsy. We selected the controls of our longitudinal analysis carefully, including two who were suspected to have nephritis based on clinical measures and observation, but were found to have class II LN on biopsy. These patients would be the most difficult to separate from patients with proliferative nephritis, highlighting the strength of our approach. Interestingly, rebound peaks in the patients’ nephritis scores were observed immediately following biopsy. This appears to be due to the common clinical practice of titrating back cytotoxic treatment following a flare, and suggests that nephritis scores could be used to monitor and avoid similar instances of rebound inflammation. In the future, this approach could be used with a larger cohort to create a similar model to aid clinicians in diagnosing and monitoring proliferative nephritis in pSLE patients.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
DJH, VKD and IB wrote the manuscript. IB acquired, analyzed and interpreted patient clinical samples and information. IB and CL conceived, designed and acquired the microarray experiments. DJH, VKD and JVP designed, acquired, analyzed and interpreted the ELISA experiments. DJH and VKD designed, performed, and interpreted the LASSO analyses. JVP, CL and PJU designed experiments, interpreted findings, and critically revised the manuscript. All authors read and approved the final manuscript.