Metabolomic alterations associated with Behçet’s disease

For metabolomics and lipidomics profiling, 24 BD patients and 25 gender- and age-matched healthy controls (HC) (without a personal or family history of autoimmune diseases) were enrolled from Peking Union Medical College Hospital (PUMCH) between March 2014 and November 2014. For further validation, an independent cohort of BD (n = 25), rheumatoid arthritis (RA) (n = 12), systemic lupus erythematosus (SLE) (n = 12), Takayasu’s arteritis (TA) (n = 15), and Crohn’s disease (CD) (n = 15) patients, and 19 HC were enrolled from March 2014 to July 2018. All BD patients fulfilled the 1990 International Study Group BD criteria or the new International Criteria for Behçet’s Disease (ICBD) [8, 9]. RA, SLE, TA and CD patients fulfilled their respective diagnostic and classification criteria [10‐13]. All participants underwent a clinical evaluation, and hospital records were reviewed. The following data were collected: disease duration, clinical manifestations, erythrocyte sedimentation rate (ESR)/C-reactive protein (CRP) level, and treatment. BD disease activities were evaluated according to the BD Current Activity Form 2006 (BDCAF 2006; http://​www.​behcetdiseasesoc​iety.​org/​behcetwsData/​Uploads/​files/​BehcetsDiseaseAc​tivityForm.​pdf).

This study was carried out in accordance with the recommendations of the institutional committee for the Protection of Human Subjects from PUMCH. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the institutional committee for the Protection of Human Subjects from PUMCH. All methods were performed in accordance with the relevant guidelines and regulations.

Sterile siliconized 0.6-mL Eppendorf tubes were used for sample preparation, and 25 μL serum was added to the tubes followed by 100 μL cold chloroform/methanol (2/1) containing lipid standards at predetermined concentrations as described previously [14]. The mixture was vortexed for 30 s at room temperature and then centrifuged at 13,000×g for 5 min to separate the polar and nonpolar species. Upper and lower phases were collected separately and transferred to new tubes. The white interphase was discarded. The collected samples were dried using a speed vacuum. The pellet of the upper phase, which primarily contained polar metabolites, was resuspended in 200 μL 50% methanol for metabolomic profiling. The lower phase was resuspended in 200 μL isopropanol/acetonitrile/water (50/25/25) for lipidomic analysis.

The chromatographic and mass spectrometric parameters were used as described previously [15]. The ultra-performance liquid chromatography (UPLC) column eluent was introduced directly into the mass spectrometer by electrospray. For metabolomic profiling, a 2-μL sample was injected onto a reverse-phase ACQUITY BEH C₁₈ 50 × 2.1 mm 1.7-μm column (Waters Corp., Milford, MA) using an ACQUITY UPLC system (Waters Corp., Milford, MA). For lipidomics profiling, an Acquity CSH C₁₈ 50 × 2.1 mm 1.7-μm column (Waters Corp., Milford, MA) was used in the UPLC-quadrupole time-of-flight mass spectrometry (UPLC-QTOF-MS^E) analysis. MS^E is a technique by which both the precursor and fragment mass spectra are acquired by alternating between high and low collision energy during a single chromatographic run. Mass spectrometric analysis was performed on a XEVO G2 QTOF (Waters) operating in both positive and negative modes. Accurate mass was maintained by introducing the LockSpray interface of sulfadimethoxine (311.0814 [M + H]+ or 309.0658 [M − H]−) at a concentration of 250 pg/μL in 50% aqueous ACN at a rate of 150 μL/min. For biomarker validation, a 2-μL sample was injected into a reverse-phase ACQUITY HSS T3 C18 100 × 2.1 mm 1.7-μm column (Waters Corp., Milford, MA) and analyzed using the consistent UPLC-QTOF-MS system. The mobile phase consisted of acetonitrile (A) and water containing 0.1% (v/v) formic acid (B), while the gradient elution program (0.0–18.0 min, 10%–95% A; 18.1–20.0 min, 100% A) was applied for favorable separation. The flow rate was set at 0.4 mL/min. The column temperature was 40 °C. All chromatograms and mass spectrometric data were acquired in centroid mode using the MassLynx software (Waters Corp., Milford, MA).

Raw mass spectrometric data were processed using Progenesis QI software (Nonlinear Dynamics, Durham, NC) to generate a data matrix that consisted of the retention time, m/z value, and the normalized peak area. Statistical analysis and putative ion identification on the postprocessed data were conducted using MetaboLyzer [16]. Statistically significant ions were putatively identified in MetaboLyzer, which utilizes the Human Metabolome Database (HMDB), LipidMaps, and the Kyoto Encyclopaedia of Genes and Genomes (KEGG) database [8] while accounting for possible adducts, H⁺, Na⁺, and NH4⁺ in the ESI⁺ mode, and H^– and Cl^– in the ESI^– mode. The m/z values were compared with the exact mass of small molecules in the databases, from which putative metabolites were identified with a mass error of 10 ppm or less. KEGG annotated pathways associated with these putative metabolites were also identified. Lipid ions were validated with the fragmentation in MS^E results based on their identifier fragments and retention time with the help of nonendogenous lipid standards from each lipid class and/or the comparison of tandem mass spectrometry (MS/MS) fragments with reference spectra provided in METLIN and LipidMaps databases. SIMCA-P+ (Umetrics, Umea, Sweden) was used for principal component analysis (PCA). The heat map displaying the relative levels of differential metabolites was generated by the Random Forests (RF) algorithm as explained in detail in previous studies [17]. MS data acquired in negative ion mode were employed for quantitation of arachidonic acid (AA) and linoleic acid (LA).

For the preparation of calibration samples, reference substances of rosmarinic acid, AA, and LA were purchased from Sigma-Aldrich Company (MO, USA). Rosmarinic acid was dissolved in methanol to produce the internal standard (IS) solution at the concentration of 1 μg/mL. The reference solutions of the two targeted metabolites, AA and LA, at the stock concentrations of 1/1000 (v/v) for each were serially diluted with the IS solution to produce a series of calibration standard solutions. All calibration standard solutions were sealed and stored at 4 °C until use.

For serum samples, 100 μL serum was added to the tube, followed by 400 μL cold methanol. The mixture was vortexed for 30 s at room temperature and then centrifuged at 13,000×g at 4 °C for 5 min to precipitate the protein. The supernatant was transferred to the Eppendorf tube, dried using a speed vacuum at 25 °C, and then dissolved in 100 μL of the IS solution for LC/MS analysis.

Experimental values are presented as mean ± SD. Statistical analysis was performed using GraphPad Prism (San Diego, CA). The significance of the metabolite level was determined using a two-tailed student t test. P values less than 0.05 were considered significant.

24 BD patients (15 men and 9 women) were all of the Han Chinese population (100%). Their mean age was, on average, 35.83 ± 11.96 years old. The median disease duration of BD was 120 months (range 13–379 months). The median age at diagnosis of BD was 28.5 years (range 17–60 years). All cases initially presented with oral ulcers. The interval between the onset of an oral ulcer and the BD diagnosis ranged from 1 to 379 months, with a median time of 71 months. The baseline clinical characteristics and medications of the BD patients are shown in Table 1. Twelve cases had been reviewed with a mean follow-up time of 7.67 ± 2.06 months. After being treated with glucocorticoids or immunosuppressants, 11 patients (91.7%) improved as measured by decreased ESR and CRP levels.

To characterize metabolomic alterations associated with BD, we analyzed the metabolomics of BD patients and HC using a UPLC-QTOF-MS approach. Unsupervised PCA plots were generated by SIMCA-P software, and differential analysis was carried out using Metabolyzer. The volcano plot showing the differential ions between HC and BD patients is shown in Fig. 1. Individuals and variables with similar profiles are grouped together in the plot. In this PCA plot, the purple and green spots, which represent individual pretreated BD patients and HC, respectively, form two segregated clusters. Statistical analysis of the metabolomics data from pretreated BD patients and the control group revealed differential ions between these two groups with statistical significance as shown in the volcano plot (Fig. 1b). In the volcano plot, the red dots represent ions that show significantly different (p value less than 0.05) levels between BD patients and HC. Putative molecules of these differential ions were designated by screening the accurate mass in metabolite databases, as stated in the Methods.

KEGG pathway analysis results indicate the metabolic pathways associated with the differential metabolites. Prominent pathways with a false discovery rate (FDR)-corrected p value less than 0.25 are shown in Fig. 1c.

Since the major pathways in Fig. 1c pointed to lipid metabolism, we performed the lipidomic profiling analysis by UPLC-QTOF-MS^E. To address whether the treatments have an effect on the disease-associated metabolomics, serum lipidomic profiling was performed using samples from twelve diagnosed patients before and after treatment, as well as a healthy cohort of the same number. Statistically significant differential metabolites between HC and pretreatment BD patients were determined using Metabolyzer. PCA analysis results for the healthy, pretreatment, and post-treatment groups based on these differential metabolites are shown in Fig. 2a. This PCA plot shows that the post-treatment cluster (red circle with a dotted line) is located in the middle, between the pretreatment cluster (dark blue circle with a dotted line) and the healthy cluster (green circle with a dotted line), suggesting a drift of diseased data points towards the direction of the healthy group after treatment. RF algorithms were used to generate a heat map of these differential metabolites for these three groups (Fig. 2b).

From the list of the differential ions, lower levels of several ions with the putative identification of phosphatidylcholine (PC) were found in the BD patient group compared with the HC group. Verification of the three selected ions was performed through MS/MS, which confirmed the identity of these ions as PCs (Additional file 1). The scatter plots in Fig. 3a show the lower level of the three PCs detected in the serum of the pretreatment BD patients compared with those in the serum of HC, but no significant difference was seen between the pretreatment and post-treatment groups.

In addition to the above phospholipids, levels of several free fatty acids from pretreatment BD patients were found to be significantly different from those of HC. We observed markedly lower levels of several polyunsaturated fatty acids (PUFAs) in the HC group compared with the pretreatment BD group, including two omega-6 (n-6) fatty acids (Fig. 3b) LA (18:2n-6) and AA (20:4n-6), and oleic acid (OA), an n-9 PUFA (Additional file 2). In contrast to the small difference between pretreatment and post-treatment BD groups for PCs, a significantly lower level of PUFAs was found in the post-treatment BD group compared with the pretreatment BD group, indicating that the treatment effectively corrected the abnormal increases of these PUFAs in BD patients. Validations of these PUFAs by MS/MS are shown in Additional file 2 and Additional file 3.

Receiver operating characteristic (ROC) analysis along with sensitivities and specificities of the area under the curve (AUC) > 0.85, is shown in Table 2 and in Additional file 4. ROC curves showed that AA was the most efficient diagnostic performance (AUC = 0.9495), compared with PCs and LA (Table 2 and Additional file 4). The sensitivity of PC(34:3), PC(40:8), LA, and AA in diagnosis of BD were comparable (0.96%, 0.88%, 0.9474%, and 0.9474%, respectively), but the specificity of PC(40:8) and AA were higher than PC(34:3) and LA (Table 2).

To verify these findings, the concentrations of LA and AA were further determined using reference standards in an independent cohort containing BD, RA, SLE, TA, and CD patients and HC. As shown in Fig. 4 and Additional file 5, the serum levels of LA and AA in BD patients were significantly higher than those in HC (p = 2.35 × 10⁻³ and p = 6.1 × 10⁻⁶). ROC curves were also made to indicate their diagnostic efficiency (Table 2 and Additional file 4). The serum levels of LA and AA in patients with RA, SLE, and TA were comparable with those in HC (Fig. 4). In addition, the levels of LA and AA in BD patients were significantly higher than they were in disease controls, suggesting that LA and AA might serve as specific markers for BD. Intriguingly, the serum levels of LA and AA in patients with CD were significantly lower than those in HC (p = 0.016 and p = 0.002, respectively) which need further investigation.