Cytokine and autoantibody clusters interaction in systemic lupus erythematosus

Evidence supports the existence of different subphenotypes in systemic lupus erythematosus (SLE) and the pivotal role of cytokines and autoantibodies, which interact in a highly complex network. Thus, understanding how these complex nonlinear processes are connected and observed in real-life settings is a major challenge. Cluster approaches may assist in the identification of these subphenotypes, which represent such a phenomenon, and may contribute to the development of personalized medicine. Therefore, the relationship between autoantibody and cytokine clusters in SLE was analyzed. This was an exploratory study in which 67 consecutive women with established SLE were assessed. Clinical characteristics including disease activity, a 14-autoantibody profile, and a panel of 15 serum cytokines were measured simultaneously. Mixed-cluster methodology and bivariate analyses were used to define autoantibody and cytokine clusters and to identify associations between them and related variables. First, three clusters of autoantibodies were defined: (1) neutral, (2) antiphospholipid antibodies (APLA)-dominant, and (3) anti-dsDNA/ENA-dominant. Second, eight cytokines showed levels above the threshold thus making possible to find 4 clusters: (1) neutral, (2) chemotactic, (3) G-CSF dominant, and (4) IFNα/Pro-inflammatory. Furthermore, the disease activity was associated with cytokine clusters, which, in turn, were associated with autoantibody clusters. Finally, when all biomarkers were included, three clusters were found: (1) neutral, (2) chemotactic/APLA, and (3) IFN/dsDNA, which were also associated with disease activity. These results support the existence of three SLE cytokine-autoantibody driven subphenotypes. They encourage the practice of personalized medicine, and support proof-of-concept studies.


Background
Systemic lupus erythematosus (SLE) is a heterogeneous systemic autoimmune disease (AD) characterized by a wide range of clinical and serological manifestations and a high disease burden [1]. SLE pathophysiology encompasses several mechanisms, such as T cell and B-cell abnormalities, impaired apoptotic debris clearance, autoantibody production, and abnormal cytokine secretion [2]. The diversity in clinical expression associated with different autoantibodies among patients supports the existence of different subphenotypes although similar treatment is given to almost all the patients with diverse effectiveness [2].
Autoantibodies are essential biomarkers for the diagnosis and classification of ADs, and several are known to be pivotal in the ADs pathophysiology [3]. In SLE they can form immune complexes, which may be deposited in tissues, and activate a direct immune response against a specific organ [4]. Autoantibodies are usually found long before symptom onset [5].
Autoimmune diseases evince similar immunopathogenic mechanisms (i.e., the autoimmune tautology) [18]. This explains the fact that one AD may coexist with one or more ADs (i.e., polyautoimmunity) [19], and that one AD may carry several autoantibodies with diverse specificity. Polyautoimmunity has been observed in up to 40% of patients with SLE [20,21]. In addition, non-lupus autoantibodies are observed frequently in SLE patients. Rheumatoid factor (RF) and anti-cyclic citrullinated peptide (CCP) antibodies are present in 42 and 5.6% respectively, but only 6.4% of the patients meet the criteria for rhupus [22]. Antiphospholipid antibodies (APLA) may be present in 54% of patients although antiphospholipid syndrome (APS) develops in only 10% of SLE patients [23]. In euthyroid patients with SLE, anti-thyroid peroxidase antibodies (TPOAb) and anti-thyroglobulin antibodies (TgAb) are observed in 21 and 10% of patients respectively, but confirmed autoimmune hypothyroidism is diagnosed in 12% [24].
Cytokines play an essential role in the pathophysiology of SLE. Thus, a number of cytokine-targeted therapies which have shown promising results have been developed, particularly in some subphenotypes of the disease [25]. However, the immune system shows a wide variation at both intra-and inter-individual levels. These differences among individuals, which may explain the differences observed among patients, have been called "immunotypes" [26].
Since human biology is a complex set of interacting components that work together to produce an outcome, a system approach may elucidate these interactions [27]. Systems biology of human disease, also known as systems medicine or network medicine, aims at identifying the main components of a system and at measuring how they change when the system is disturbed [26]. Since understanding the connections of the nonlinear complex processes of cytokines and autoantibodies in real-life settings is a major challenge, we analyzed the simultaneous relationship between them in patients with established SLE.

Study population
This was a cross-sectional analytical study of 67 consecutive women with SLE. The subjects have been systematically followed at the Center for Autoimmune Diseases Research (CREA) in Bogota, Colombia. All the subjects fulfilled the 1997 update of the American College of Rheumatology (ACR) classification criteria for SLE [28]. The patient socio-demographic and cumulative clinical and laboratory data were obtained by interview, standardized report form, physical examination and chart review as previously reported [23,29]. The data were collected in an electronic and secure database.

Clinical variables
Clinical and laboratory variables were registered as present or absent at any time during the course of the disease as previously reported [23]. Other manifestations such as polyautoimmunity [19,30] and current pharmacological treatment were also assessed.
Current disease activity was measured using the Systemic Lupus Activity Questionnaire (SLAQ), a wellknown Patient-Reported Outcome (PRO) tool, which presents an adequate performance in large community-based cohorts [31][32][33]. As SLAQ is unavailable in Spanish, a linguistic validation was done. Working independently, two English proficient physicians (JBC, MR) translated the original US English version into Spanish [32]. Afterwards, they worked together to obtain a single Spanish version. A mother tongue professional translator independently back-translated this version into an English one. Lastly, the physicians compared the two versions to produce a second Spanish version. If there was disagreement, a third English-proficient physician (YR) decided which was the best version. Finally, a definite Spanish form was acquired and used with the patients (see Additional file 1).

Laboratory measurements
Serum samples were obtained during a state of fasting. A total of 14 autoantibodies were evaluated in the sera of patients. Detection of IgM RF, IgG anti-CCP third-generation (CCP3), IgM and IgG anti-cardiolipin antibodies (ACA), IgM and IgG anti-β2glycoprotein-1 (β2GP1) antibodies, IgG anti-double-stranded DNA (dsDNA) antibodies, IgG TgAb, and TPOAb were all quantified by the Enzyme-Linked-Immunosorbent Assay (ELISA) as previously reported [24]. Antinuclear antibodies (ANAs) were evaluated by using an indirect immunofluorescence assay. Positive ANA were considered from dilution 1/80. Negative and positive controls, provided by the manufacturer, were analyzed in parallel. Anti-Ro, anti-La, anti-RNP, and anti-Sm were also evaluated by ELISA. All the assay kits were from Inova Diagnostics, Inc. (San Diego, CA, USA).
Concentration of 15 human cytokines (IL-2, IL-10,  IL-6, IL-8, IL-9, IL-13, IL-12/23p40, G-CSF, IFNγ, IFNα,  IL-4, IL-1β, TNFα, IL-5, IL-17A) in serum samples from  patients was assessed by Cytometric Bead Array (CBA,  Becton-Dickinson Biosciences, San Diego, CA, USA). The test was done in accordance with the manufacturer's protocols. Briefly, 50 µL of assay beads and 50 µL of the sample under study or standard were added to each sample tube. The samples were incubated at room temperature in the dark for 1 h. Next, the samples were washed with 1 mL of wash buffer, centrifuged, and the resulting pellet was resuspended in 50 µL of PE-labeled antibodies (Detection Reagent). The samples were further incubated for 2 h, washed again, and centrifuged. After discarding the supernatant, the pellet was resuspended in 300 µL of wash buffer and analyzed on the same day in a FAC-SCanto II ™ flow cytometer (BD Bioscience ™ ). Before the analysis, the cytometer was standardized using calibration beads in accordance with the manufacturer's protocol. For each cytokine, a standard curve was assessed, and concentration of each cytokine was calculated as an interpolation of the standard curve using the FCAP Array ™ Software (BD Bioscience ™ ). Results were considered positive when the assay results were above a threshold value, which was confirmed in healthy individuals in whom evidence of acute or chronic disease including autoimmune, cardiovascular, or metabolic was not detected (Table 3) [34,35].

Statistical analyses
The mixed-cluster methodology proposed by Lebart et al. [36] was used to find groups of patients with similar autoantibody and cytokine profiles. In short, cluster analysis seeks groups of individuals with similar values across several variables. The number of groups is algorithmically determined and consolidated in two steps: first, a hierarchical cluster analysis is done based on Ward's distance, for which the number of clusters is determined by means of the between-cluster inertia gain criterion. Second, the cluster membership for each individual is consolidated using a k-means algorithm on the centroids of each cluster. In the end, a categorical variable in which each individual is assigned to one and only one of the clusters derived is obtained [36]. Afterwards, a description of each cluster is developed by studying the distribution of each of the original variables used for clustering in each of the derived groups. This determine the composition and relation of the original variables and the clusters obtained.
This clustering method was used to obtain autoantibody clusters (named profiles from here on) based on the 14 autoantibodies, and cytokine profiles based on the 15 cytokines measured. Cytokines and autoantibodies with frequencies under 5% were excluded from the cluster analysis, since variables with low frequencies tend to generate clusters of patients with such atypical results exclusively. To assess associations between abovementioned profiles and other variables, we used the Chi square and Kruskall-Wallis tests. Statistical analyses were done using R version 3.3.2.

Ethics
This research was carried out in accordance with Resolution number 008430 of 1993 issued by the Ministry of Health of the Republic of Colombia and was classified as minimal risk research. The Ethics Committee of Universidad del Rosario approved the present project.

Autoantibody clusters
Three clusters of autoantibodies were defined ( Fig. 1): (1) neutral, in which the frequency of specific autoantibodies other than ANAs was below 40%; (2) APLA-dominant, which showed a greater than 50% frequency of ACA-IgG/ IgM, anti-dsDNA, and anti-RNP; and (3) anti-dsDNA/ ENA-dominant, which presented a high frequency of anti-dsDNA, anti-RNP, and anti-Sm antibodies. Table 4 shows the distribution of autoantibodies within clusters.
No association between autoantibody clusters and disease activity was found.

Cytokine clusters and disease activity
There was a significant association between cytokine clusters and disease activity. (p = 0.022; Fig. 3a,). The distribution of autoantibody clusters differed between neutral cytokine and IFNα/Pro-inflammatory clusters (p = 0.031; Fig. 3b). Tables 6 and 7 show the distribution of cytokines by autoantibody clusters, and autoantibodies by cytokine clusters respectively.
A cluster characterized by the presence of anti Sm/RNP antibodies has been consistently reported [7,10,14,15], and in some reports, it has been associated with anti-dsDNA, thus yielding a Sm/RNP/dsDNA cluster [8,12]. Likewise, a Ro/La cluster has also been reported [9,10,15], and in some reports, it has been associated with anti-dsDNA [8,11,14]. These findings are supported by the cross-reactivity and similarity of anti-Sm and anti-RNP [39], and the induction of anti-Ro and anti-La by common ribonucleoproteins [40]. It is noteworthy that some authors have found one cluster with positivity for 3 or 4 ENA (with or without anti-dsDNA) [7,11,12]. This evidence is similar to our Cluster 3 in which a predominance of anti-dsDNA/ENA was observed. Anti-Sm and anti-La antibodies were virtually absent in neutral and APLA autoantibody clusters (Fig. 1). Autoantibody clusters did not show an association with disease activity [12,14]. This could be due to the measurement method, in which a well-known PRO questionnaire (i.e., SLAQ) [41] was used in contrast to physician-based indexes (i.e., SLEDAI) used in other cohorts. Nevertheless, there is no serologic test that reliably measures disease activity in SLE [42].
Second, four cytokine clusters were obtained, namely (1) neutral, (2) chemotactic, (3) G-CSF dominant, and 4) IFNα/Pro-inflammatory. To our knowledge, this is the first report on a serum cytokine cluster analysis in patients with SLE. The composition of the third and fourth cytokine clusters were validated by an external bioinformatic analysis which confirmed biological relationships among cytokines (Additional file 3).
The neutral cytokine cluster displayed a low frequency of cytokines that was below 25% (Fig. 2). The chemotactic  cytokine cluster showed a marked expression of IL-8 followed by lower frequencies of IL-12/23p40, IL-6, and G-CSF. IL-8 is a chemotactic cytokine, particularly involved in recruitment of neutrophils, which induces shape transformation, the 'respiratory burst' , and the release of granule contents [43]. Increased levels have been seen in SLE patients and they appear to be influenced by anti-dsDNA (Additional file 4: Table S2). High levels of IL-6 have also been found in SLE [44,45]. Both IL-6 and IL-8 have been shown to be up-regulated by endothelial cells treated with IgG APLA in vitro [46]. IL-8 has been associated with pregnancy morbidity in patients with SLE [35]. The anti-dsDNA antibody up-regulates IL-8 gene expression and elicits activation-induced cell death of human polymorphonuclear neutrophils [47], and the release of IL-8 [48]. The third cytokine cluster was named G-CSF dominant, since a particularly high frequency of G-CSF was seen although IL-12/23p40 was rather frequent. G-CSF is an essential growth factor for the differentiation of hematopoietic stem cells into granulocytes, particularly neutrophils. Synthetic G-CSF preparations (e.g., filgrastim, pegfilgrastim, lenograstim) are available to treat neutropenia [49]. Data is scarce regarding G-CSF in SLE (Additional file 4: Table S2). Furthermore, IL-12 and IL-23 are mainly pro-inflammatory cytokines that share a common structural unit and receptors. IL-12 consists of two subunits: p35 and p40, whereas IL-23 is comprised of subunits p19 and p40. The two share the p40 subunit, which interacts with the same membrane receptor [50]. Although available assays for p19, p35, and p70 (which includes subunits IL-12 p35 and p40) exist, we measured IL-12/23p40. IL-12 is pivotal for Th1 differentiation [51] and has been found to be higher in SLE patients [52]; its implication for physiopathology remains under investigation (Additional file 4: Table S2). In addition, IL-23 plays a role in the development of Th17 cells, and promotes IL-17 secretion [53]; clinical trials with anti-IL-12/23 are underway [50].
The last cytokine cluster revealed high levels of diverse cytokines, including G-CSF, IL-12/23p40, IL-17A, and IL-10. Nonetheless, IFNα and TNFα were the most frequent cytokines. Thus, the cluster was named IFNα/Proinflammatory. IFNα belongs to the Type I IFN family and is mainly secreted by plasmacytoid dendritic cells (pDC). Type I IFNs promote autoimmunity due to the activation of B-cell responses, maturation of monocytes into DC, and NETosis promotion [4,54]. One of the SLE hallmarks is its IFN I signature, which is dysregulated when compared to healthy controls (Additional file 4: Table S2). Recent clinical trials in SLE with a Type I IFN blockade (i.e., sifalimumab, anifrolumab) have shown promising results [55,56].
TNFα is a pro-inflammatory cytokine secreted by monocytes, macrophages, T cells, neutrophils, and mast cells. It promotes lymphocyte recruitment and inflammatory responses. However, it becomes immunosuppressive with chronic exposure [57]. Increased levels, which correlated with disease activity, have been found in SLE patients (Additional file 4: Table S2). A negative feedback loop between Type I IFN and TNFα has been suggested: when Type I IFN prevails, SLE may occur. TNF inhibits

Table 8 Distribution of autoantibodies and cytokines in integrative clusters
Data correspond to number of patients (%) RF rheumatoid factor, CCP3 anti-cyclic citrullinated peptide third-generation, TPOAb anti-thyroperoxidase antibody, TgAb anti-thyroglobulin antibody, dsDNA antidouble stranded DNA antibodies, ACA anticardiolipin antibody, β2GP1 β2 glycoprotein-1, ANAs antinuclear antibodies, IL interleukin, G-CSF granulocyte colonystimulating factor, IFN interferon, TNF tumor necrosis factor, ACR American College of Rheumatology * Statistically significant after Bonferroni correction a Renal criteria was defined as active urinary sediment, or proteinuria > 500 mg/24 h or positive renal biopsy [23,24]   the development of pDCs and their production of Type I IFN [58]. IL-17A belongs to the IL-17 family. It is largely produced by Th17 cells although it is secreted by several immune cells [59]. IL-17A has been widely studied in autoimmunity and is thought to play a pivotal role in SLE physiopathology (Additional file 4: Table S2). IL-10 is an immunomodulatory cytokine secreted by several cell populations although it exerts an essential role in B cell processes. Thus, it may promote hyperactivity of the B-cell compartment, thus leading to increased autoantibody production. Due to its dual function (i.e., B-cell stimulation/antibody production, and T-cell inflammatory response reduction), its role in SLE is not fully understood (Additional file 4: Table S2). IL-6 is a multifunctional cytokine secreted by several cells of both the innate and adaptive immune systems as well as by nonimmune cells such as fibroblasts [60]. Increased serum levels are found in SLE and appear to be associated with joint involvement [44,61,62], and disease activity [52]. A few monoclonal antibodies (e.g., tocilizumab, sarilumab) are current treatment options for different ADs [4]. Nevertheless, randomized clinical trials for SLE are lacking. A summary of the role of IL-6 in SLE is shown in Additional file 4: Table S2.
IL-2 is predominantly produced by Tregs. Deficiency in IL-2 secretion is involved in the pathogenesis of SLE through the impairment of Treg growth and survival [64]. Recent evidence has shown that IL-2 secretion is impaired by high levels of IL-23 and IL-23R [65]. This may explain the absence of IL-2 in our clusters due to the high frequency of IL-12/23p40 in patients (Additional file 4: Table S2). IL-4 is secreted by several immune cells, particularly basophils. Since it stimulates B-cells, it may be involved in antibody production and SLE pathophysiology (Additional file 4: Table S2). IL-9 is a pleiotropic cytokine, produced by an ample variety of immune cells including mast cells, NKT cells, Th2, Th17, Treg, and the recently described Th9 [66]. It is considered a proliferative cytokine, which can induce the production of IL-6, mainly by mast cells. It enhances B-cell production of IL-4, IgE, and IgG1 and promotes isotype switching. In contrast, IL-9 secreted by Treg induces tolerance [66]. Some data on the implications of IL-9 for SLE have been described (Additional file 4: Table S2). IFNγ is mainly produced by T and NK cells. It is important for T cell differentiation and B-cell isotype switching [67]. Its role in SLE is described in Additional file 4: Table S2. Data regarding IL-5 and IL-13 in SLE patients is scarce. IL-5 is pivotal for eosinophil differentiation [68], whereas IL-13 is an immunomodulatory cytokine secreted by Th2 CD4 T-cells (Additional file 4: Table S2).
Regarding the interaction between autoantibodies and cytokines, the absence of autoantibodies was associated with a low frequency of cytokines (Table 6). Disease activity, in turn, was lower in the neutral cytokine cluster (Fig. 3a). The enhancement of antibody production and activation of autoreactive B cells may be favored by a Th2 environment [51,69]. Some inflammatory cytokines (e.g., TNFα, G-CSF) were less common in the neutral autoantibody cluster although statistical significance was not reached ( Table 6). IL-10 was significantly lower in the neutral autoantibody cluster than in the APLA-dominant and dsDNA/ENA dominant clusters (Table 6). Abnormally increased IL-10 synthesis seems contributing to the spontaneous hyperactivity of the B cell compartment, so that it can directly result in autoantibody production by committed plasma cells, circulating immune complexes formation, and eventually in tissue and organ damage [70]. IL-5 was absent in the neutral autoantibody cluster, which may be in line with the reported worse renal outcomes associated with elevated IL-5 urine concentration [68]. Additionally, anti-dsDNA antibodies were less frequent within the neutral cytokine cluster (Table 4), which highlights the involvement of these autoantibodies in SLE pathophysiology. The IFNα/Pro-Inflammatory cluster revealed a significant low frequency of the neutral autoantibody cluster (Fig. 3b). This finding is in line with the capability of IFNs and IL-17 to induce antibody secretion [54,71]. These results support the fact that SLE disease activity is mediated by cytokine secretion [2] and the potential role of autoantibodies in the enhancement of cytokine production [48,72].
Our last analysis, which was integrative and in which all the biomarkers were included, showed three clusters which reinforced the results (Fig. 4a). The G-CSF cytokine cluster observed in the second analysis was distributed throughout the three integrative clusters. In addition, these three clusters were associated with the disease activity (Fig. 4b).
Our study supports the importance of individualized treatment of patients since both autoantibody and cytokine clusters were established in a cohort of SLE patients that also showed interaction and association with disease activity. The identification of SLE subphenotypes has been suggested previously [73] and is pivotal for the implementation of personalized medicine [74]. Our results depict the existence of different subphenotypes based on both diverse disease-specific and nonspecific autoantibodies and on easy-access molecules such as cytokines. A comprehensive assessment of multiple biomarkers, which is feasible with multiplex assay technologies [27], should offer the possibility of a novel taxonomy for SLE and the implementation of targeted therapies based on cytokine patterns (i.e., proof-of-concept studies).
Systems biology approaches have been applied to rheumatic diseases, in order to find novel biomarkers and therapeutic strategies. These efforts have focused on gene-level interactions and their relationship with clinical manifestations. For instance, Chiche et al. [75] found three individual IFN modules through transcriptional repertoire analysis using microarray technology, which showed an association with some clinical variables. Likewise, Bancherau et al. [6] described seven discrete groups of SLE patients based on their gene expression patterns and clinical disease severity. Reclassification of SLE patients based on the results of 'omics' studies has been proposed [76].
The possible shortcomings of our study must be acknowledged. The main objective was to evaluate simultaneously the relationship among cytokines, autoantibodies, and the disease activity at one point in time. Therefore, the lack of association between clusters and cumulative clinical characteristics was expected. Likewise, the effects of treatment on the modulation of cytokine/autoantibody levels were not taken into account. The main considered outcome was the activity of the disease. However, the results of this exploratory study should stimulate further longitudinal designs using larger groups of patients to fully describe these subtle complexities. Regarding the cytokine measurement method, previous reports on RA patients' sera showed that the correlation of CBA assays with ELISAbased methods for cytokine detection is moderately-high although a lower concentration for some cytokines (i.e., IL-2, TNFα, IL-10) may be detected [77]. Furthermore, CBA is not affected by the presence of RF in contrast to other multiplex technologies (e.g., Luminex) and allows the assessment of multiple biomarkers using relatively small sample volumes [77]. We are aware of the reported low positivity of IFNα levels in serum, which is rarely detectable by ELISA or bioassays, and the suggested assessment of gene expression monitoring [78]. However, our results yielded appropriate IFNα assessment by CBA and even the identification of an IFNα-related cluster. Measurements of IFNγ using CBA appear to be proper which supports a low concentration in our patients [77]. Although other techniques are used in research to measure cytokines, CBA is implemented in many clinical and regular laboratories. This is, therefore, a cost-effective and practical method. Another potential drawback of our study could be the method of disease activity quantification, SLAQ, a well-known PRO questionnaire [41], which was used under a non-structured validation. Since SLE may exhibit a disconnection between disease activity and patient perceived well-being, PROs may help empower patients on disease management. Numerous regulatory agencies encourage the use of PROs in clinical trials [79]. Note that our results showed an association between SLAQ scores and cytokine clusters. This fact is particularly interesting given that SLAQ does not include autoantibodies (e.g., anti-dsDNA) in its criteria in contrast to clinical indexes (e.g., SLEDAI). Thus, disease activity could be biased by anti-dsDNA. Instead, as shown herein, disease activity as portray by SLAQ may be due to others biomarkers such as cytokines. Another potential limitation of the present study is that the observed results may be due to chance alone or the moderate sample size. However, such a possibility would be unlikely given the highly significant results seen as well as their consistent direction and magnitude within the different analyses.

Conclusions
Systemic lupus erythematosus is a heterogeneous systemic AD with profound cytokine abnormalities. Multiple disease-specific and non-specific biomarkers are present in SLE patients. Clustering methods allow the identification of association among these markers and yield different subphenotypes. Additional systems medicine approaches are warranted in order to reveal the strength of these interactions, which should assist in the implementation of personalized medicine.