A total of 100 AOSD patients (74 active and 26 inactive) and 60 healthy controls (HCs) were included in the training and validation sets. An additional independent set consisting of 21 patients with sepsis, 25 AOSD patients and 18 HCs was used to determine the specificity of miRNAs signature in AOSD patients. All AOSD patients fulfilled Yamaguchi's criteria after exclusion of those with infectious, neoplastic and autoimmune disorders (19). All sepsis patients fulfilled the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) (20). All HCs subjects were recruited from age- and sex-matched volunteers with no history of autoimmune, rheumatic, or other diseases. Information on demographic and clinical data was entered into a database along with the laboratory test results. The AOSD disease activity of each patient was assessed using a modified Pouchot score (21). The experimental design was approved by the Ethics Committee of Ruijin Hospital (identifier 2016-62) and Huashan hospital (identifier 2017-338), Shanghai China, and all the participants provided informed consent. All plasma and serum samples were stored at −80°C immediately after collection.

Data of the miRNA sequencing from the PBMC of AOSD patients and HCs were acquired as previously described (22). PBMC from patients with AOSD (N = 3) without any treatment and sex- and age-matched HCs (N = 3) were used. In the discovery stage, 482 miRNAs were detected for the purpose of finding the upregulated and downregulated miRNAs.

Target genes of the top 5 upregulated miRNAs were predicted and further analyzed to obtain information about biological functions, pathways and networks by using the web-based bioinformatics tool QIAGEN's Ingenuity Pathway Analysis (IPA; Ingenuity Systems, http://www.INGENUITY.com).

Total RNAs were extracted from 200 uL of plasma using the QIAzol miRNeasy Serum/Plasma kit (Qiagen, Valencia, CA, USA) according to the manufacturer's instructions. Caenorhabditis elegans miRNA (cel-miR-39: 5′-UCACCGGGUGUAA-AUCAGCUUG-3′) was used as a spiked-in control (Qiagen, Valencia, CA, USA). 5.25 × 10⁸ copy of cel-miR-39 was added to each denatured sample after combining the plasma sample with 1 mL QIAzol Lysis Reagent. Total RNAs were eluted with 14 µL of RNase-free water and stored at −80 °C.

The cDNA was prepared with the miScript II RT kit (Qiagen) following the manufacturer's instructions. Plasma miRNAs were quantified using miScript SYBR Green PCR Kit (Qiagen) as previously described (10). The relative expression levels of miRNAs were qualified using the equation: amount of target miRNA expression = 2^−ΔCt, ΔCt = Ct plasma miRNAs- Ct cel-miR-39. The results were magnified 1,000 times.

Serum levels of TNF-α, IL-1β, IL-6, and IL-18 were measured by the Meso Scale Discovery electrochemiluminescence assay (MSD, Rockville, MD, USA) according to manufacturer's instructions.

All data were statistically analyzed using the SPSS version 20.0 software (SPSS Inc., Chicago, IL, USA). Quantitative data were expressed as the means ± SD. Data between two groups with a Gaussian distribution were analyzed using an unpaired t-test, while non-parametric data were assessed using the Mann-Whitney U test. Data among three groups or more were analyzed using ANOVA (one-way analysis of variance) or Wilcoxon rank-sum test. Receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) were used to assess the sensitivity and specificity of miRNA biomarkers for the diagnosis of AOSD. We used three different methods (logistic regression, k-nearest neighbor and support vector machine) to find the best combination of the 5 selected miRNAs (miR-142-5p, miR-101-3p, miR-29c-3p, miR-29a-3p, and miR-141-3p) to diagnosis of AOSD. In the training stage, 109 samples (68 AOSD, 41 HCs) were included. In the validation stage, 51 samples (32 AOSD, 19 HCs) were used to verify the model obtained before by means of ROC curve and AUC index. Spearman correlation analysis was performed to test whether the expression levels of 5 miRNAs were correlated with clinical manifestations. P < 0.05 was considered statistically significant.

Plasma and sera from 100 AOSD patients (74 active and 36 inactive) and 60 healthy controls (HCs) were collected. The clinical characteristics of these subjects in each group are detailed in Table 1. Among the 74 active AOSD patients in the qRT-PCR analysis, common manifestations included spiking fever (73, 100%), sore throat (46, 62.6%), evanescent rash (62, 83.8%), and arthralgia (31, 41.9%). Hepatomegaly, splenomegaly and lymphadenopathy were noted in 5 (6.8%), 36 (48.6.0%), and 58 (78.4%) patients, individually. There were no significant differences in terms of mean age or sex distribution between AOSD patients and HCs.

To identify miRNAs that were differentially expressed in AOSD and HCs, we per- formed miRNA deep sequencing from PBMC of 3 AOSD patients and 3 HCs. A volcano plot and a heat map (Figure 1) were used to visualize the distinction between the miRNA expressions. Missing data were abandoned or imputed using the nearest-neighbor method, and z-score and 0~1 normalization methods were further conducted on the miRNA expression data. A total of 482 miRNAs were identified in both groups (Figure 1A). We performed filtering between AOSD patients and HCs, using the following criteria: fold change ≥2.0 (or ≤ 0.5) and P ≤ 0.05. A hierarchical clustering of differentially expressed miRNAs was shown in Figure 1B, including 10 upregulated miRNAs and 10 downregulated miRNAs, indicating that the expression panels of miRNAs in PBMC of AOSD patients differ from those in HCs. Among the 10 candidate upregulated miRNAs, 5 plasma miRNAs (miR-142-5p, miR-101-3p, miR-141-3p miR-29a-3p, and miR-29c-3p) were significantly increased in AOSD patients (N = 20) compared with HCs (N = 12) (Supplementary Figure 1). We chose these 5 miRNAs for further analysis.

The five upregulated mature miRNAs were searched in the QIAGEN's Ingenuity Pathway Analysis (IPA) to identify the possible mRNA targets involved in the pathophysiology of AOSD. Three of the interested miRNAs (miR-142-5p, miR-101-3p, miR-29a-3p) were included in the IPA database. The network elaborating the interactions between miRNAs and mRNAs was generated by using the miRNAs Target Filter, filtered by activation of leukocytes, systemic autoimmune disease, viral infection and lymphoma. Confidence was restricted to experimentally observed and highly predicted while species was restricted to human beings (Supplementary Figure 2).

To confirm the results of miRNA sequencing and hypothesis whether circulating miRNAs secreted by PBMC be a diagnostic tool, we then measured the levels of 5 top upregulated miRNAs (miR-142-5p, miR-101-3p, miR-29c-3p, miR-29a-3p, and miR-141-3p) in plasma in a training set consisting of 68 AOSD patients and 41 HCs. Compared with HCs, we identified a significantly higher level of these 5 miRNAs in plasma from AOSD patients (Figures 2A–E, P < 0.001, P < 0.05, P < 0.0001, P < 0.0001, and P < 0.05, respectively).

We further accessed the sensitivity and specificity of these 5 miRNAs in diagnosing AOSD using ROC analysis. And then, we used logistic regression to determine the best combination of miRNAs to predict AOSD. Three classification algorithms were used to generate the predictive miRNA combinations: k-nearest neighbor (KNN, k = 3), support vector machine (SVM) and binary and multivariate logistic regression model (LR). We found that a linear combination of the expression levels of miR-142-5p, miR-101-3p, and miR-29a-3p generated the best predictive model to diagnose AOSD by using KNN, SVM, and LR. The ROC curves and diagnostic accuracies of three models were shown in Figures 2F–H, and sensitivity and specificity of the classification model were selected by maximizing the Youden index. Compared with the single miRNA, a 3-miRNA panel increased the AUC value to 0.8436, 0.7721, and 0.7758 by KNN, SVM and LR, respectively (Figures 2F–H). The sensitivity was 0.8824 and specificity was 0.8049 by KNN prediction. Taken together, our results suggest that this 3-miRNA panel can be adopted as a diagnostic tool for AOSD.

We further validated the expression levels of 5 miRNAs in an independent validation study consisting of 32 AOSD patients and 19 HCs. The upregulation of 5 miRNAs in plasma of AOSD patients was validated in the validation set (Figures 3A–E). Moreover, we confirmed that the combination of miR-142-5p, miR-101-3p, and miR-29a-3p expression generated the best predictive model in the validation set. The AUC value for the 3-miRNA biomarker panel including miR-142-3p, miR-101-3p, and miR-29a-3p was 0.7854, 0.7706, and 0.7796 by KNN, SVM and LR analysis, respectively (Figures 3F–H). Furthermore, analysis of the data from all 160 samples in training and validation sets produced an AUC value of 0.8250, 0.7717, and 0.7777 by three classification algorithms (Figures 3F–H). And the combination of 3 miRNAs (miR-142-5p, miR-101-3p, miR-29a-3p) generates the best predictive model. Logistic regression produced a model: Logit(p) = −0.8581+0.0012 × miR-142-5p-0.0012 × miR-101-3p +0.0041 × miR-29a-3p (Figure 3H). In a conclusion, our data suggest that the 3-miRNA panel can serve as potential non-invasive biomarkers for AOSD diagnosis.

Furthermore, we compared the levels of 5 miRNAs in AOSD patients with diverse disease activity. The plasma levels of miR-142-5p, miR-101-3p, miR-29a-3p, miR-29c-3p, and miR-141-3p were significantly higher in AOSD patients with active disease (N = 74) compared with those with inactive disease (N = 26) (Figure 4, P < 0.001, P < 0.001, P < 0.001, P < 0.001, and P < 0.01, respectively). Moreover, the levels of 5 miRNAs were significantly higher compared with those in HCs. And there was no significant difference in the levels of 5 plasma miRNAs between AOSD patients with inactive disease and HCs. Taken together, our data indicate that plasma miRNA signature is associated with the disease activity of AOSD.

We next assessed the correlations between the levels of plasma miRNAs and clinical manifestations. Compared with AOSD patients without high spiking fever, the level of plasma miR-101-3p was significantly higher in AOSD patients with fever (Table 2, P = 0.0056). In addition, our results showed that patients with sore throat had higher level of miR-101-3p (Table 2, P = 0.0061). And in the presence of arthralgia, the level of miR-101-3p was significantly increased in AOSD patients (Table 2, P = 0.0484). Furthermore, the level of miR-29c-3p was significantly higher in patients with myalgia than in patients without myalgia (Table 2, P = 0.0143). Thus, miR-101-3p would be considered a potential biomarker to diagnosis AOSD patients timely, as Yamaguchi's criteria includes the clinical manifestations of high spiking fever and arthralgia.

It has been demonstrated that inflammatory cytokines play a critical role in the pathogenesis of AOSD, we next measured the correlation of plasma miRNAs levels with inflammatory cytokines, including IL-1β, IL-18, IL-6, and TNF-α. As shown in Figure 5, the level of plasma miR-101-3p was significantly positively correlated with TNF-α (r² = 0.2413, P = 0.0434) and IL-6 (r² = 0.2457, P = 0.0458) in serum. We found that other miRNA levels showed no significant correlation with IL-1β, IL6, IL-18, and TNF-α in AOSD patients (data were not shown). In a conclusion, the results indicate that miR-101-3p may play an important role in the expression levels of IL-6 and TNF-α.

It's very important and difficult to differentiate sepsis from AOSD in clinical practice. To determine the disease specificity of the 5 miRNAs identified in AOSD patients, we enrolled patients with sepsis (N = 21) as non-AOSD disease controls. Samples from an independent set consisting of AOSD patients (N = 25) and HCs (N = 18) were collected. The expression levels of 5 miRNAs (miR-142-5p, miR-101-3p, miR-29c-3p, miR-29a-3p, and miR-141-3p) were significantly higher in AOSD patients compared to that in sepsis patients (Figures 6A–E). We further accessed the sensitivity and specificity of 5 miRNAs in distinguishing AOSD from sepsis using ROC analysis. We found that 4-miRNA panel (miR-142-5p, miR-101-3p and miR-29c-3p and miR-141-3p) produced the best predictive model to separate AOSD from sepsis by using KNN, SVM, and LR. Compared with the single miRNA, a 4-miRNA panel increased the AUC value to 0.8448, 0.8248, and 0.8952 by KNN, SVM, and LR, respectively (Figures 6F–H). The diagnostic sensitivity of the panel for AOSD distinguishing from sepsis patients was 0.88 and the specificity was 0.8095 by KNN prediction, and SVM and LR analysis also showed high sensitivity and specificity (Figures 6F–H). Taken together, these data suggest that these 5 miRNAs have a potential value to distinguish AOSD patients from sepsis.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.