Introduction
Behcet’s disease (BD) is a chronic, multisystem autoinflammatory disorder characterized by recurrent oral and genital ulcerations, uveitis, and skin lesions, as well as vascular, neurological, and gastrointestinal manifestations [
1,
2]. BD, also called Silk Road disease or Behcet’s syndrome, mainly occurs in countries along the ancient Silk Road from the Mediterranean Basin across Asia to Japan [
3]. BD is considered as one of the most common causes of uveitis and the primary cause of blindness [
4]. Our recent study involving 15 373 uveitis patients showed that BD accounted for 10.6% of cases [
5].
Although the etiology of BD remains unclear, genetic susceptibility, environmental factors, viral and bacterial infections, inflammation, and immune dysregulation are involved in its development [
6‐
9]. Immune dysfunction of both adaptive and innate immunity plays an essential role in the pathogenesis and progression of BD [
10]. The levels of pro-inflammatory and anti-inflammatory cytokines have been extensively studied in the serum and plasma of patients with BD [
11‐
13]. These inflammatory cytokines produced by immune cells can regulate or activate other immune cells, causing tissue damage. For example, T helper 17 (Th17) cells, which are the major subsets of CD4
+ T cells, are essential to the process of BD. The differentiation of human naïve CD4
+ T cells into Th17 cells is regulated by cytokines such as interleukin 6 (IL 6), transforming growth factor-β (TGF-β), interleukin 21 (IL 21), and interleukin 23 (IL 23) [
14,
15]. Several studies have demonstrated that cytokines could serve as potential drug targets for the treatment of BD or candidate biomarkers for the prediction of disease activity, severity, and prognosis [
16‐
18].
Evidences also indicate that some immune response-related proteins can regulate the secretion of inflammatory cytokines and differentiation of immune cells via the Janus kinase—signal transducer and activator of transcription (JAK-STAT), nuclear factor-κB (NF-κB), and P38 mitogen-activated protein kinase (P38-MAPK) signaling pathway [
19‐
21]. For example, increased tripartite motif-containing 21(TRIM21) can activate the NF-κB signaling pathway to promote the secretion of IL6, interleukin 1β (IL 1β), and IL 23 and induce the differentiation of Th17 cells in BD [
22]. However, the potential role of immune response-related proteins in immune and inflammatory function modulation in BD is less well studied.
The aim of this study was to investigate the expression profile of immune response-related proteins in the plasma of patients with BD and identify potential plasma biomarkers in BD.
Methods
Study population
Active BD patients ((training cohort
n = 27, validation cohort
n = 28) and healthy controls [HC (training cohort
n = 25, validation cohort
n = 28)] matched by age and sex were enrolled in the study from the First Affiliated Hospital of Zhengzhou University. BD was strictly diagnosed by rheumatologists according to the diagnostic criteria developed by the International Study Group for Behçet’s disease [
1]. BD activity was evaluated using the Behçet Disease Current Activity Form (BDCAF) [
23]. Uveitis was diagnosed by an ophthalmologist. Intraocular inflammation was evaluated according to the standardized uveitis nomenclature (SUN) working group classification [
24]. Detailed demographic information and clinical details of the BD patients are listed in Table
1 and Supplementary Table S
1.
Table 1
Clinical characteristics of Behcet’s disease (BD) and healthy controls (HC)
HC |
HC (N, %) | 25 | 12 (48.00) | 13 (52.00) | |
Age (mean (SD)) | 40.28 (8.11) | 41.67 (8.54) | 39.00 (7.81) | 0.423 |
BD |
BD (N, %) | 27 | 13 (48.10) | 14 (51.90) | |
Age (mean (SD)) | 33.37 (14.19) | 36.92 (17.70) | 30.07 (9.44) | 0.216 |
Treatment (%) | 8 (29.60) | 2 (15.40) | 6 (42.90) | 0.322 |
Disease duration (months) (median (Q1, Q3)) | 36.00 (21.50–74.00) | 36.00 (24.00–40.00) | 68.50 (20.25–75.00) | 0.593 |
Oral or Genital ulcers (%) | 27 (100.00) | 13 (100.00) | 14 (100.00) | |
Skin involvement (%) | 8 (29.60) | 4 (30.80) | 4 (28.60) | 1.000 |
Joint involvement (%) | 7 (25.90) | 4 (30.80) | 3 (21.40) | 0.909 |
Uveitis (%) | 14 (51.90) | 7 (53.80) | 7 (50.00) | 1.000 |
Vascular involvement (%) | 4 (14.80) | 0 (0.00) | 4 (28.60) | 0.122 |
Neurological involvement (%) | 2 (7.40) | 2 (15.40) | 0 (0.00) | 0.430 |
Gastrointestinal involvement (%) | 2 (7.40) | 1 (7.70) | 1 (7.10) | 1.000 |
Plasma collection
Fresh peripheral blood (10 ml) was collected in EDTA tubes, and plasma was isolated by centrifugation at 2000 g for 10 min, and then stored at − 80 °C until use.
Measurement of plasma proteins
The plasma levels of 92 immune response-related proteins were measured using a proximity extension assay (PEA, Olink Proteomics, Shanghai, China) [
25]. The data are presented as normalized protein expression (NPX) values on a log2 scale. Twelve proteins were excluded from downstream analysis with intra- and inter-assay coefficient of variance (%CV) and the frequency of missing values of more than 20% in each sample. One patient sample was excluded because of quality control failure (Supplementary Figure S
1a). In addition, an NPX value of less than 0 was replaced by the intragroup mean in some samples.
Data analysis and statistics methods
Principal component analysis (PCA) was performed using “FactoMiner” and “factoextra” R packages. Categorical variables are described as numbers (percentages) and compared using the chi-square test or Fisher’s exact test. Continuous variables are presented as median and interquartile range (IQR). Differences between two and three groups were compared using the non-parametric Mann–Whitney U test and Kruskal–Wallis test with Dunn’s correction, respectively. The results are presented in the form of tables or boxplots. Volcano and heatmap plots were drawn using the “ggpubr” and “pheatmap” packages. The correlated heatmap was plotted to visualize the Pearson’s correlations between differentially expressed proteins (DEPs) using the “ggcorrplot” package.
Gene ontology (GO) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using the R package “clusterProfiler” (version 3.18.1) [
26]. To further investigate the correlation between DEPs, a protein network interaction diagram (PPI) was constructed using the online tool STRING (version 11.5,
https://cn.string-db.org/).
Feature selection and prediction model creation
The recursive feature elimination (RFE) algorithm, which includes feature extraction, feature selection, and model training, was performed for the features selected based on the random forest (RF) with fivefold repeated cross-validation. All DEPs were used to train the prediction model, and the feature importance of the variables was calculated and ranked using accuracy and kappa metrics. An optimal subset of features was selected from all DEPs for the prediction model creation. To construct the prediction model, five algorithms were used based on the package “caret”: naive Bayes (NB), support vector machine (SVM), extreme gradient boosting (XGB), random forest (RF), and neural network (NNET). Receiver operating characteristic (ROC) analysis, recall (sensitivity), specificity, precision, accuracy, F1 score, and residual distribution were used to assess the prediction performance of the different models on the testing set. ROC curves were plotted using the “pROC” package.
Consensus clustering
To investigate the role of differentially expressed immune response-related proteins, different clinical phenotypes, sex, and age in BD patients, K-means consensus clustering with k from 2 to 7 was performed using the R package “ConsensusClusterPlus.” The clustering results were visualized using t-distributed stochastic neighbor embedding (tSNE) based on the “Rtsne” R package. All analyses were carried out using the R language, version 4.0.3.
Discussion
BD is a chronic, multisystem autoinflammatory disorder. The diagnosis of BD mainly relies on clinical symptoms. In this study, we examined the expression levels of immune response-related proteins in the plasma of patients with BD using the Olink Immune Response panel. The results demonstrated aberrant expression of immune response-related proteins profiles in BD patients. Potential biomarkers were identified by constructing predictive models using machine learning algorithms. We also constructed a novel molecular disease classification model to identify the subsets of BD.
The etiology of BD remains unknown. We measured the expression levels of immune response-related protein to investigate the immunopathogenesis of BD. A total of 43 DEPs were identified in the BD and HC groups. The results of GO and KEGG enrichment analyses highlighted that the NF-κB signaling pathway and Toll like receptor 9 (TLR9) signaling pathway are involved in the occurrence of BD. These results are consistent with those of a previous study. Verrou et al. performed RNA-sequencing analysis in peripheral blood mononuclear cells and found that the NF-κB signaling pathway is related to BD [
27]. Previous studies also reported that the NF-κB signaling pathway could protect T cells against CD95-mediated apoptosis in BD [
28]. The NF-κB signaling pathway is considered a typical pro-inflammatory pathway, and the activation of signaling pathways induces the production of various proinflammatory cytokines such as IL-6 and IL-8 [
29]. The NF-κB signaling pathway is also involved in the development of other rheumatic autoimmune diseases [
30]. Activation of the NF-κB signaling pathway induces chronic inflammation of the synovium in rheumatoid arthritis [
31].
TLR9 signaling pathway is essential for the regulation of both innate and adaptive immunity, and it is also involved in the production of type I interferons (IFNs) [
32]. A recent study reported that dysregulation of TLR9 contributes to the production of IFN-γ and leads to fatal inflammatory disease in neonates [
33]. Activation of the TLR9 signaling pathway has been observed in patients with primary Sjögren’s syndrome based on single cell phosphorylation profiling [
34]. Additionally, in an experimental autoimmune encephalomyelitis (EAE) animal model of multiple sclerosis, pathogens have been attributed to TLR9-mediated innate immunity [
35]. Overall, our study further identified the activation of the NF-κB and TLR9 signaling pathways in the plasma of BD. Together, these studies indicate that NF-κB and TLR9 signaling pathways are involved in the immunopathogenesis of BD. The DEPs were also enriched in infections-related signaling pathways such as toxoplasmosis and Epstein-Barr virus infection. Although the role of toxoplasmosis and Epstein-Barr virus infection in the pathogenesis of BD is less well understood, some studies have demonstrated that toxoplasmosis and Epstein-Barr virus infection are the risk factors for other systemic immune diseases, including rheumatoid arthritis and systemic lupus erythematosus [
36,
37], and were associated with the activation of the NF-κB signaling pathway [
38,
39]. In addition, a recent plasma proteomic study in BD patients also revealed that several infection pathways, for example, pertussis, amoebiasis, and tuberculosis, were associated with the pathogenesis of BD [
40]. These pathways implicated the role of infection in the pathogenesis of BD.
IL-10, FCRL3, MASP1, NF2, FAM3B, and MGMT are potential candidate biomarkers for BD. The diagnosis of BD was made based on clinical symptoms [
41]. To the best of our knowledge, the current study is the first to use machine learning algorithms to identify the potential candidate biomarkers in BD.
IL-10 is an anti-inflammatory cytokine that can inhibit Th1 cytokine production and Th1 cell differentiation [
6]. Our results are consistent with the results reported by Aridogan et al., which described the elevated level of IL-10 in the serum of active BD [
42]. In addition, our previous study assessed the aqueous cytokine levels in BD and senile cataract patients. However, the expression level of IL-10 was not statistically significant, which might be because the intraocular inflammations of BD were in the inactive phase [
43]. Overall, the overexpression of IL 10 may represent a compensatory mechanism in response to chronic inflammation in BD. The overexpression of IL 10 may play an important role in dampening excessive inflammation by inhibiting IL 6, which is also highly expressed in our study [
44]. Another possible explanation is that IL-10 may have a dual role in immune responses. While IL 10 is generally considered to be anti-inflammatory, it can also promote inflammation under certain circumstances [
45]. For example, IL 10 has been shown to enhance the inflammatory response in some autoimmune diseases, such as systemic lupus erythematosus (SLE) [
46]. It is possible that IL 10 has a similar pro-inflammatory effect in BD.
FCRL3 is an orphan receptor, which is only expressed on the lymphocyte cell surface. it can inhibit the secretion of TNF-α, IL 1β, IL 6, and IL-8 by promoting the expression of IL 10 in multiple sclerosis [
47]. In addition, a single nucleotide polymorphism in the FCRL3 promoter region binding of the NF-κB is associated with rheumatoid arthritis, autoimmune thyroid disease, and systemic lupus erythematosus [
48]. Our previous study also found associations between a single nucleotide polymorphism of FCRL3 and BD susceptibility in the Chinese population [
49]. MASP1 is a serine protease involved in complement system. It is essential for defense against invading pathogens and altering host structures [
50]. NF2, FAM3B, and MGMT are primarily involved in regulating the tumor immune microenvironment [
51‐
53]. We reported, for the first time, a significant difference in the expression levels of NF2, FAM3B, and MGMT between BD and HC. However, further experiments are needed to explore the functional role of NF2, FAM3B, and MGMT in BD and other autoimmune diseases.
We report a novel molecular disease classification model for BD based on an unsupervised consensus clustering algorithm. DEPs, clinical phenotypes, sex, and age were used to construct the model. BD patients were divided into two subsets, cluster 1 with 14 patients and cluster 2 with 12 patients, characterized by distinct cytokine production profiles and disease activity. The characterization of cluster 1 was high disease activity and high TRIM5, SH2D1A, PIK3AP1, HCLS1, and DFFA expression. The characterization of cluster 2 showed low disease activity associated with a higher expression of CCL11. Our molecular disease model differed from the previous clinical classification model in that it is a novel immunophenotype for BD [
54]. This model provides insight into the immunopathogenesis of BD and might help further refine the classification and diagnosis of BD. Besides, TRIM5, EGLN1, SH2D1A, and DFFA were correlated with disease duration, which may explain the classification model.
Another interesting finding from our study was that PLXNA4 (plexin A4) is a DEP between BDU and BDNU, whereby PLXNA4 expression was down-regulated in BDU. A previous study indicated that cytokines could impair vascular integrity by downregulating the expression of PLXNA4 [
55]. This may explain the occurrence of retinal vasculitis in BDU.
Our study had some limitations. Most patients in our study previously received small doses of immunosuppressants; however, the effect of treatments was weak. Further clinical significance and function of candidate biomarkers need to be comprehensively investigated. It is undeniable that sample sizes are small in our study. We only compared the uveitis phenotype and without uveitis phenotype in BD patients. Further expanded experimental sample size and analysis of the relationship between immune response-related proteins and other phenotypes of BD patients will be necessary.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.