This cross-sectional study assessed eligible patients with obese asthma (n = 11) and lean asthma (n = 22) who were recruited from the Asthma Clinic of West China Hospital, Sichuan University between August 2015 and February 2016. The obese asthma group was defined as having obesity with BMI ≥ 30.0 kg/m², and the lean asthma group was defined as having normal weight between 18.5 and 25 kg/m² according to the World Health Organization.21 Asthma was based on a physician's diagnosis with a history of variable respiratory symptoms (e.g., wheeze, shortness of breath, chest tightness and cough) and either spirometric testing with a 12% or 200 mL improvement in forced expiratory volume in 1 second (FEV1) after 400 mcg ventolin (GSK, Avda de Extremadura, Spain) or a 20% or greater decrease in FEV1 with standard doses of methacholine (PD20FEV1 < 12.8 µmol) according to the Global Initiative for Asthma (GINA) guidelines1 and the National Asthma Education and Prevention Program (NAEPP).22 Only stable asthmatic patients, who had no exacerbation, respiratory tract infection or change in maintenance therapy in the previous 4 weeks, were included. Excluding criteria included unexplained weight change during the past 3 months, the existence of a previous diagnose of chronic respiratory disease and severe systemic disease such as lung cancer, bronchiectasis, heart disease, hypertension, diabetes or psychiatric disorders; medication use that may alter systemic inflammation and metabolism (e.g., hormonal contraceptives or statins); pregnant women; contraindication to spirometry or inability to complete study procedures. All patients were refrained from food intake on the day of sample collection in the same Asthma Clinic. The Clinical Research and Ethics Committee of West China Hospital (No. 2014-30) approved the study protocol, and written informed consent was obtained from all subjects.

All patients underwent assessment of obesity, demographics, medication use and asthma status using formatted questionnaires. Obesity was assessed by BMI, waist-to-hip ratio (WHR) and body composition. Height (cm), weight (0.1 kg), and waist (cm) and hip (cm) circumferences were recorded; body composition such as fat mass (FM), visceral fat area (VFA) and percent body fat (PBF) was measured by the body composition analyzer (InBody S10, Body Composition Analyzer; Biospace Co. Ltd., Seoul, Korea) according to the InBody S10 user's manual. Asthma control was evaluated using the Asthma Control Test23 (ACT), and the Asthma Quality of Life Questionnaire24 (AQLQ) was used to evaluate the impact of asthma in 4 life domains. Atopy was determined by skin prick testing (SPT) as previously described.25 We measured exhaled nitric oxide (FeNO) before spirometric testing using a NIOX analyzer (Aerocrine, Solna, Sweden), according to American Thoracic Society/European Respiratory Society (ATS/ERS) recommendations. Spirometry (MedGraphics Corp., St. Paul, MN, USA) was performed in accordance with the ATS/ERS guidelines.26

Sputum induction and processing were performed based on standard methods as described in our previous studies.27 Sputum was induced after the pre-treatment of 400 mcg salbutamol (GSK) using 4.5% saline atomized by an ultrasonic nebulizer (Cumulus; HEYER Medical AG, Bad Ems, Germany). If FEV1 was ≤ 40% of predicted, sputum was induced with 0.9% saline after it was deemed safe by the supervising physician. Sputum plugs were collected and an aliquot of 200 μL of sputum plug was quick-frozen immediately by liquid nitrogen and stored at −80°C for subsequent metabolism analysis. Further, a volume of 1% dithiothreitol (SPUTOLYSIN Reagent; Calbiochem^®, San Diego, CA, USA) was added equal to 4 times the remaining sputum w at 1,500 rpm for 10 minutes, and the sputum supernatant was aspirated and stored immediately at −80°C for subsequent detection. Total and differential cell counts were performed with centrifugation-smear (CYTOPRO 7620; WESCOR^®, Inc., South Logan, UT, USA) and staining preparation by well-trained Chinese and Australia lab researchers.

Serum and PBMCs samples were collected using standardized methods provided in the Methods section in the supplement file. Serum and PBMCs samples were quick-frozen by liquid nitrogen and stored at −80°C prior to metabolism analysis.

C-reactive protein (CRP), leptin and adiponectin in serum was measured by sandwich ELISA (R&D SYSTEMS, Inc. Minneapolis, MN, USA). Interleukin (IL)-1β, IL-4, IL-5, IL-6, IL-8, IL-13 and tumor necrosis factor (TNF)-α in sputum supernatant as local inflammation were detected by a Luminex-based MILLIPLEX^® MAP Human Cytokine/Chemokine Magnetic bead Panel Kit (EMD Millipore Corporation, Billerica, MA, USA), and were analyzed using the software of Milliplex Analyst 5.1 as it has been demonstrated that the Luminex-based xMAP^® panel can be used for multi-analyte profiling of sputum using the routinely applied method of sputum processing with dithiothreitol.2829

Sputum and peripheral blood metabolites were analyzed according to the procedure as described in the previously published metabolomics profiling methods.3031 Briefly, the thawed sputum and serum were extracted with methanol following vortex for 30 seconds. Metabolites were extracted from PBMCs using extraction liquid (Vmethanol:Vchloroform = 3:1) following vortex for 30 seconds. The resuspended cells were snap-freezed in liquid nitrogen for 5 minutes, were thawed at room temperature, and were ultrasonized for 10 minutes in 4°C water bath. The freeze-thaw cycle is repeated 3 times. L-2-chlorophenylalanine (1 mg/mL stock in dH₂O; Shanghai Hengbai Biotech Co Ltd, Shanghai, China) was added as an internal standard. The samples were centrifuged at 12,000 rpm for 15 minutes at 4°C. Supernatant was transferred to a 2-mL GC-MS glass vial and was vacuum-dried at room temperature. The derivatized samples were obtained after the residue was incubated with methoxyamine hydrochloride (20 mg/mL in pyridine) for 30 minutes at 80°C, were bis-(trimethylsilyl)-trifluoroacetamide (BSTFA) plus 1% (v/v) trimethylchlorosilane (TMCS) (REGIS Technologies, Morton Grove, IL, USA) for 2 hours at 70°C subsequently, and were cooled to room temperature prior to GC-MS analysis.

Metabolomic profiling was identified by GC-MS analysis, which was performed using an Agilent 7890 gas chromatograph system coupled with a Pegasus HT time-of-flight mass spectrometer. The system utilized a DB-5MS capillary column coated with 5% diphenyl cross-linked with 95% dimethylpolysiloxane (30 m × 250 μm inner diameter, 0.25 μm film thickness; J&W Scientific, Folsom, CA, USA). An aliquot of 1 μL of the analyte was injected in a splitless mode. Helium was used as the carrier gas, the front inlet purge flow was 3 mL/min, and the gas flow rate through the column was 1 mL/min. The initial temperature was kept at 50°C for 1 min and then raised to 290°C at a rate of 10°C/min, where it remained for 10 minutes. The injection, transfer line and ion source temperatures were 280, 270, and 220°C, respectively. The energy was −70 eV in an electron impact mode. The mass spectrometry data were acquired in a full-scan mode with the m/z range of 33-600 (85-600 in PBMCs samples) at a rate of 20 spectra per second after a solvent delay of 366 seconds in sputum samples, 295.8 seconds in serum samples and 460 seconds in PBMCs samples, respectively.

Chroma TOF4.3X software of LECO Corporation and LECO-Fiehn Rtx5 database were used for raw peak exacting, the data baseline filtering and calibration of the baseline, peak alignment, deconvolution analysis, peak identification and integration of the peak area, missing values of raw data were filled up by half of the minimum value. In addition, internal standard normalization method was employed in this data analysis. The resulted 3-dimensional data involving the peak number, sample name and normalized peak area were fed to the SIMCA14 software package (Umetrics, Umea, Sweden) for principal component analysis (PCA) and orthogonal projections to latent structures-discriminate analysis (OPLS-DA). In order to obtain a higher level of group separation and get a better understanding of variables responsible for classification, supervised orthogonal projections to OPLS-DA were applied. The parameters for the classification from the software, R²Y and Q²Y, were used to evaluate the goodness-of-fit and goodness of prediction retrospectively. To further validate the model, the permutation tests (n = 200) was proceeded.

The LECO/Fiehn Metabolomics Library was used to identify the compounds with a similarity value for the compound identification accuracy, which is reliable if the similarity is < 700. To refine this analysis, the first principal component of variable importance projection (VIP) was obtained. The VIP values exceeding 1.0 were first selected as changed metabolites. The remaining variables were then assessed by Student's t test (P < 0.05), and variables were discarded with no significant change between 2 comparison groups. The identified differential metabolites were further searched on online database, Kyoto Encyclopedia of Genes and Genomes (KEGG), PubChem Compound, Chemical Entities of Biological Interest (ChEBI), Japan Chemical Substance Dictionary Web (NIKKAJI) and Chemical Abstracts Service (CAS), and the top altered pathways were identified by pathway topology enrichment analysis in MetaboAnalyst 3.0 (http://www.metaboanalyst.ca/) with the impact of each pathway calculated using the Relative-betweenness Centrality.32

As for demographic and clinical measurements, continuous variables are expressed as means ± standard deviations or medians (interquartile ranges), depending on their distribution. Categorical variables are reported as absolute numbers and percentages. Group comparisons for continuous data were performed using either unpaired Student's t-tests (parametric data) or the Wilcoxon rank sum test (non-parametric data), with categorical variables evaluated by the χ² test or Fisher's exact test. Pearson's (parametric data) or Spearman's (non-parametric) correlation was used to describe the specific correlation between differential metabolites and obesity or clinical and inflammation profiles, as previously described.33 A P value of less than 0.05 was deemed statistically significant. Statistical analysis was performed with the SPSS software (version 21.0; IBM Corp., Armonk, NY, USA).

The ACT scores were significantly lower in obese asthmatics than in lean subjects (P = 0.032). There was no statistical difference observed between the 2 groups in demographic data, smoking, quality of life, asthma duration, asthma medications, airway obstruction in FEV1%predicted and FEV1/ forced vital capacity (FVC), FeNO and IgE levels. Peripheral blood and sputum cell profiles were generally balanced between the 2 groups (Table 1). We also compared the differences in body composition between the 2 groups. Obese asthmatic subjects, as expected, had significantly higher BMI, FM, fat free mass (FFM), VFA and WHR than lean asthmatic subjects (all P < 0.05) (Table 2).

Obese asthmatic subjects had significantly higher leptin (P = 0.008) and lower adiponectin (P < 0.001) levels compared with lean asthmatic subjects, while the CRP level was comparable (P = 0.839). As for the inflammatory mediators in induced sputum supernatant, we found that IL-1β was significantly elevated in obese asthmatic subjects compared with lean asthmatic subjects (P = 0.047). However, there were no statistical differences in IL-6, IL-8, TNF-α. Interestingly, in comparison with lean asthmatic subjects, obese asthmatic subjects showed higher levels of IL-4 (53.5 [48.38–72.34] vs. 10.30 [0.60–29.26] pg/mL, P = 0.004), and IL-13 (3.92 [2.74–8.64] vs. 2.46 [1.96–3.04] pg/mL, P = 0.017) (Table 2).

OPLS-DA analysis showed that obese subjects were separated from lean asthmatics (Fig. 1). Description in full was included in the Results section in this article's supplement. The differential metabolites in induced sputum, serum and PBMCs between obese asthmatic and lean asthmatic subjects are listed in Table 3. As depicted in the heat map (Fig. 2), it indicated that, in induced sputum, there were significantly increased metabolites such as benzoic acid, 3-hydroxybutyric acid, hydrocinnamic acid, aspartic acid 2, xanthine, 4-aminobutyric acid 1, glutaric acid, gly-pro, d-glucoheptose 1, gluconic lactone 2, L-glutamic acid, phytosphingosine 2, beta-glutamic acid 1, pyrrole-2-carboxylic acid, pyrophosphate 3 and 3-aminopropionitrile 1, and reduced metabolites as indole-3-acetic acid and shikimic acid in subjects compared with lean asthmatic subjects. The most significant differential metabolites identified in sputum samples were further annotated on online database of KEGG (Fig. 3) with higher expression metabolites highlighted by red dots and lower expression labeled by blue dots. For the metabolic profile in serum, it showed an increase in valine, uric acid, N-Methy-DL-alanine and beta-glycerophosphoric acid, and a decrease in asparagine 1 and d-glyceric acid in subjects compared with lean asthmatic subjects, which was indicated in the heat map (Supplementary Fig. S1A). Meanwhile, PBMCs samples had an increase in 3-hydroxynorvaline 2 and a decrease in 3-hydroxybutyric acid, linolenic acid, isoleucine in the obese asthmatic patients comprised with lean asthmatic subjects (Supplementary Fig. S1B). Predictive accuracy of obese asthma and lean asthma using differential metabolites were assessed by calculating the areas under the receiver operating characteristic (ROC) curves for OPLS-DA model in SIMCA. The criterion parameter (YPredPS) retrieved from the current predictionset was used for thresholding and the ROC curve was created by plotting the true positive rate (TPR) versus the false positive rate {FPR = 1 − True Negative Rate (TNR)} at various threshold settings of the criterion parameter (YPredPS). In other words, the ROC curve visualizes the classifier's Sensitivity versus 1 − Specificity. Each point on the ROC curve represents a pair sensitivity/specificity values corresponding to a particular decision threshold. As results shown in Fig. 4, the predictive regression models for obese asthma using differential metabolites indicated an excellent accuracy for both induced sputum and PBMC with the area under the curve (AUC) of 0.966 and 0.916, respectively.

To analyze the most relevant pathways after identifying differential metabolites, the metabolic pathway topology enrichment analysis was applied with the impact value, and −log(p) were calculated to evaluate the importance of the pathways underlying the pathophysiology mechanism in obese asthma. As a result, the pathway topology analysis for induced sputum indicated 14 potentially metabolic pathways involved in the development of obese asthma (Fig. 5A), the most important of which were cyanoamino acid metabolism (P = 1.8 × 10⁻³; impact: 0.333), caffeine metabolism (P = 3.3 × 10⁻³; impact: 0.031), alanine, aspartate and glutamate metabolism (P = 4.5 × 10⁻³; impact: 0.103), phenylalanine, tyrosine and tryptophan biosynthesis (P = 5.8 × 10⁻³; impact: 0.098), pentose phosphate pathway (P = 8.4 × 10⁻³; impact: 0.043). For serum samples, 5 pathways were identified (Fig. 5B), 3 of which with the pathway impact higher than 0.01 were glyoxylate and dicarboxylate metabolism (P = 1.6 × 10⁻³; impact: 0.033), glycerolipid metabolism (P = 6.0 × 10⁻⁴; impact: 0.021) and pentose phosphate pathway (P = 6.0 × 10⁻⁴; impact: 0.022). Unfortunately, we did not uncover any divergent metabolic pathways in PBMCs samples between obese asthmatic and lean asthmatic subjects as most of the differential metabolites identified could not match corresponding metabolites in KEGG database.

Obesity-associated differential metabolites were validated using the confirmation of relationship between these differential metabolites and anthropometrical parameters and biomarkers (Leptin and adiponectin) of obesity (Fig. 6 and Supplementary Table S1). As a result, it indicated that most of identified metabolites significantly correlated to anthropometrical parameters and biomarkers of obesity. The relationships between anthropometrical parameters of obesity and differential metabolites were fully described in the “Results” section in this article's supplement.

We further explored whether there were significant correlations of differential metabolic profiling with clinical characteristics in all subjects with asthma (Fig. 6 and Supplementary Table S2). Interestingly, it found some differential metabolites such as aspartic acid 2, beta-glutamic acid 1, pyrrole-2-carboxylic acid, 4-aminobutyric acid 1, gluconic lactone 2 in induced sputum positively correlated to FEV1% and FEV1/FVC% reflecting airway obstruction (r < 0.45, P < 0.05). However, no differential metabolites in serum and PBMCs were statistically related with FEV1% or FEV1/FVC% except for isoleucine in PBMCs that positively correlated to FVC% (r = 0.37; P = 0.046). In induced sputum, phytosphingosine 2 had a liner correlation to ACT score (r = −0.508; P = 0.028) and glutaric acid was related with AQLQ score (r = 0.507; P = 0.027) when ACT and AQLQ were transformed into normal distribution parameter.

Asthma was well characterized by airway inflammation. Therefore, we observed relationships of differential metabolites with airway inflammation (Fig. 6 and Supplementary Table S3). Firstly, it was found that d-glucopheptose 1 was associated with the percentage of neutrophils in induced sputum (r = 0.536; P = 0.040). Secondly, it was indicated that differential metabolites in induced sputum were significantly related with airway inflammatory biomarkers. In terms of T_H2^high cytokines, the IL-4 level was inversely associated with shikimic acid (r = −0.598; P = 0.024), meanwhile IL-5 positively correlated to gly-pro (r = 0.651; P = 0.012) and IL-13 level to 3-hygroxybutyric acid (r = 0.593; P = 0.025). We also found that IL-1β levels correlated to d-glucoheptose 1 (r = 0.58; P = 0.029) and pyrophosphate 3 (r = 0.54; P = 0.047), and TNF-α levels to pyrophosphate 3 (r = 0.60; P = 0.024), and IL-8 levels to pyrophosphate 3 (r = 0.60; P = 0.023) and 3-aminopropionitrile 1 (r = 0.55; P = 0.04). Thirdly, for serum differential metabolites, it indicated d-glyceric acid was related with IL-13 (r = −0.48; P = 0.026), N-Methyl-DL-alanine with IL-1β (r = 0.52; P = 0.014), uric acid with IL-6 (r = 0.47; P = 0.026) and IL-8 (r = 0.48; P = 0.024). Fourthly, for differential metabolites in PBMCs, 3-Hydroxynorvaline 2 was related with IL-13 (r = −0.47; P = 0.048), 3-hydroxybutyric acid with IL-1β (r = −0.50; P = 0.036), and linolenic acid with IL-4 (r = −0.62; P = 0.006), IL-13 (r = −0.52; P = 0.027), and IL-1β (r = −0.51; P = 0.032).

Finally, we screened 18 potential metabolic signatures of obese asthma (Fig. 7). The differential metabolites from subjects with obese and lean asthma identified in OPLS-DA analysis were first validated, and their relationships with obesity assessments (BMI, fat, FFM, PBF, VFA, WHR, leptin or adiponectin) by correlation analysis were in 22 metabolites. Then, we further validated the 22 obesity-associated metabolites in relation to clinical and inflammation profiles of asthma by correlation analysis. Finally, there were 18 metabolic signatures associated both with obesity and asthma, which were pyrophosphate 3, d-glucoheptose 1, shikimic acid, aspartic acid, beta-glutamic acid 1, 4-aminobutyric acid 1, phytosphingosine 2, xanthine, 3-aminopropionitrile 1, 3-hydroxybutyric acid, gly-pro in sputum (n = 11), and valine, uric acid, d-glyceric acid, N-Methyl-DL-alanine in serum (n = 4), and linolenic acid, 3-hydroxynorvaline 2, isoleucine in PBMCs samples (n = 3).