Novel Approach for Analysis of Bronchoalveolar Lavage Fluid (BALF

Jul 19, 2014 - The first group included subjects who did not receive an inhaled or systemic corticosteroid for over 4 weeks prior to the exam (NSBA gr...
0 downloads 0 Views 4MB Size
Download PDF
Article pubs.acs.org/jpr

Novel Approach for Analysis of Bronchoalveolar Lavage Fluid (BALF) Using HPLC-QTOF-MS-Based Lipidomics: Lipid Levels in Asthmatics and Corticosteroid-Treated Asthmatic Patients Yun Pyo Kang,† Won Jun Lee,† Ji Yeon Hong,† Sae Bom Lee,† Jeong Hill Park,† Donghak Kim,‡ Sunghyouk Park,† Choon-Sik Park,§ Sung-Woo Park,*,§ and Sung Won Kwon*,† †

College of Pharmacy, Seoul National University, 1 Gwanak-ro, Gwanak-gu, Seoul 151-742, Republic of Korea Department of Biological Sciences, Konkuk University, Hwayang-dong, Gwangjin-gu, Seoul 143-701, Republic of Korea § Division of Respiratory and Allergy, Department of Internal Medicine, Soonchunhyang University Bucheon Hospital, 1174 Jung-dong, Wonmi-gu, Bucheon 420-767, Republic of Korea ‡

S Supporting Information *

ABSTRACT: To better understand the respiratory lipid phenotypes of asthma, we developed a novel method for lipid profiling of bronchoalveolar lavage fluid (BALF) using HPLC-QTOF-MS with an internal spectral library and high-throughput lipid-identifying software. The method was applied to BALF from 38 asthmatic patients (18 patients with nonsteroid treated bronchial asthma [NSBA] and 20 patients with steroid treated bronchial asthma [SBA]) and 13 healthy subjects (NC). We identified 69 lipids, which were categorized into one of six lipid classes: lysophosphatidylcholine (LPC), phosphatidylcholine (PC), phosphatidylglycerol (PG), phosphatidylserine (PS), sphingomyelin (SM) and triglyceride (TG). Compared with the NC group, the individual quantity levels of the six classes of lipids were significantly higher in the NSBA subjects. In the SBA subjects, the PC, PG, PS, SM, and TG levels were similar to the levels observed in the NC group. Using differentially expressed lipid species (p value < 0.05, FDR < 0.1 and VIP score of PLS-DA > 1), 34 lipid biomarker candidates with high prediction performance between asthmatics and controls were identified (AUROC > 0.9). These novel findings revealed specific characteristics of lipid phenotypes in asthmatic patients and suggested the importance of future research on the relationship between lipid levels and asthma. KEYWORDS: lysophosphatidylcholine, phosphatidylcholine, phosphatidylserine, phosphatidylglycerol, sphingomyelin, triglyceride, bronchoalveolar lavage fluid

■

INTRODUCTION

The collection of BALF is the most reliable way to sample the fluid lining the lower respiratory tract,7,8 and this fluid is suitable for use in physiologically relevant asthma studies. Bronchoalveolar cells are the direct pathogenic site of asthma, and the lipids in BALF are similar to those in the surfactants generated by alveolar type II cells.7,9 Additionally, as the BALF surfactants facilitate the small airway patency,9 there is an urgent need to understand the roles of the various lipids in BALF. However, a precise process to identify biomarkers for global lipid metabolism of the pulmonary surfactants in the BALF of asthmatics has not been fully elucidated. Therefore, the global profiling of lipids would provide a platform to study asthma pathogenesis and assess clinical applications to determine a way to reduce the aggravation of asthma caused by surfactant dysfunction.

Asthma is one of the most common chronic respiratory diseases and is characterized by airway inflammation, hyperresponsiveness, airway obstruction, and remodeling.1,2 Many previous asthma studies have investigated the related lipids using a variety of samples, such as alveolar epithelial cell, serum, sputum, and bronchoalveolar lavage fluid (BALF). These studies mostly dealt with phosphatidylcholine (PC), phospholipase A2 (PLA2), lysophosphatidylcholine (LPC), and arachidonic acid (AA), and the results have provided insights regarding the pathogenesis of asthma. A previous study reported that PC can be hydrolyzed by PLA2 to produce LPC and AA, both of which are associated with asthma.3 LPC is known to induce alterations in surfactant properties, lung resistance, and capillary permeability.4−6 AA enables the production of lipids with strong vasoreactivity and bronchoreactivity, such as eicosanoids, through its metabolism by proinflammatory cells.3,6 © 2014 American Chemical Society

Received: February 28, 2014 Published: July 19, 2014 3919

dx.doi.org/10.1021/pr5002059 | J. Proteome Res. 2014, 13, 3919−3929

Journal of Proteome Research

Article

Fiberoptic Bronchoscopy

Lipids are one of the major categories of biological metabolites that make up cellular membranes; they also store energy and can act as signaling molecules.10,11 Lipid species are categorized by diverse classes, such as lysophospholipids, glycerophospholipids, sphingolipids, and triglycerides.12 Lipidomics is the study of lipids in biological systems, and this approach was used herein to determine whether such lipids may be used as potential biomarkers for quantitative and qualitative analytical technologies.12 Several lipidomic analytical methodologies have been developed using various mass spectrometric techniques, such as electron impact mass spectrometry (EI-MS), atmospheric pressure chemical ionization mass spectrometry (APCI-MS), matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS), and electrospray ionization mass spectrometry (ESI-MS).12,13 Among such methodologies, liquid chromatography coupled to ESI−high-resolution mass spectrometry (LC-ESI-HRMS) has been particularly useful in gathering quantitative and qualitative data for various lipid species.14−17 In this study, we developed a global lipid profiling method for BALF using high-performance liquid chromatography with quadrupole time-of-flight mass spectrometry (HPLC-QTOFMS). By combining a class-specific internal spectral library (derived from authentic standards) with a commercially available online database, we rapidly and easily identified 69 individual lipid species belonging to six lipid classes (LPC, PC, PG, PS, SM, and TG). The developed method was applied to a total of 51 BALF samples from asthmatic patients (n = 38) and normal controls (n = 13). The lipid profile was used to discriminate between asthmatic patients and normal controls and allowed for the further observation of corticosteroid effects on the asthmatic patients.

■

BAL was performed for each subject using fiberoptic bronchoscopy (Olympus B2-10; Olympus, Tokyo, Japan). A lavage was performed using three to four 50 mL aliquots of warm saline instilled at the right-middle lobe. The BAL procedure has been previously described.20 Aliquots of the cellfree BAL fluid were stored at −80 °C. Complications were not identified in any subject during the BAL procedure. Chemicals

The lipid standards were purchased from Avanti Polar Lipids (Alabaster, AL), as described in Supplementary Table S1 in the Supporting Information. HPLC-grade acetonitrile, 2-propanol, chloroform (CHCl3), methanol (MeOH), and water were purchased from JT Baker (Philipsburg, NJ). LC−MS-grade formic acid and ammonium acetate were purchased from Sigma-Aldrich (St. Louis, MO). Sample Preparation

The internal standard (IS) mixture contained the following lipid species that are composed of C17:0 fatty acid chains that are nonexistent in BALF:21 LPC (17:0/0:0), PC (17:0/17:0), phosphatidylserine (17:0/17:0), and phosphatidylglycerol (17:0/17:0) at 20 μg/mL and triglyceride (17:0/17:0/17:0) at 10 μg/mL, which were prepared in chloroform/methanol (2:1 v/v). Twenty microliters of the IS mixture was added to each 400 μL BALF sample. Six hundred microliters of a chloroform/methanol mixture (2:1 v/v) was then added to the sample, and this sample mixture was vortexed and incubated for 20 min at room temperature, followed by centrifugation at 16 000g for 10 min. Finally, the lower lipid extract layer was carefully separated and analyzed using HPLC-QTOF-MS. HPLC-QTOF-MS Conditions

MATERIALS AND METHODS

In a random order, the lipid extracts were separated using a 1260 HPLC system (Agilent, CA) and a HPLC column (Brownlee SPP C18, 2.7 μm, 2.1 mm × 75 mm PerkinElmer, Branchburg, NJ) at 50 °C. The solvent and gradient conditions from a previous study were modified to achieve separation of a wide range of lipid species.22 We applied mobile phase A, which included water (1% 1 M ammonium acetate and 0.1% formic acid), and mobile phase B, which included acetonitrile/2propanol (1:1 v/v, 1% 1 M ammonium acetate and 0.1% formic acid). The gradient elution began with 65% phase A, followed by 20% phase A for 8 min, then 0% phase A for 14 min; it was then maintained for 14 min. Flow rate was 0.4 mL/min. The separated lipid peaks were analyzed using an Agilent 6530 QTOF-MS (Agilent, CA) equipped with an electrospray ion source with Agilent Jet Stream Technology. The QTOF-MS was operated in ESI-positive mode. Ion source parameters were as follows: gas temperature, 350 °C; sheath gas temperature, 400 °C; gas flow, 11 L/min; sheath gas flow, 12 L/min; nebulizer pressure, 20 psi. Scan source parameters were as follows: capillary voltage, 4000 V; nozzle voltage, 0 V; fragmentor voltage, 170 V; and skimmer voltage, 60 V. The m/z accuracy was calibrated using a lock-mass system with a reference solution composed of purine and HP-0921 (m/z; 121.0509 and 922.0097), and the data were collected in the range of 50 to 1500 m/z, with a 1.03 spectra/s acquisition speed for the MS and MS/MS modes. For MS/MS fragment ion acquisition, five fragmentation energies (10, 20, 30, 40, and 50 eV) were applied.

Subjects

This study was performed using 38 asthma patients and 13 healthy subjects. The asthma patients were recruited from Soonchunhyang University Bucheon Hospital, Republic of Korea. We enrolled patients with mild-to-moderate asthma, as described by the Global Initiative for Asthma (GINA) guidelines.18 Mild to moderate persistent asthma was defined as having postbronchodilator forced expiratory volume in 1 s (post-BD FEV1) at 1).

The raw data were exported in mzData format and imported into MZmine 2.10, which is an open-source LC−MS data processing software.23 As recommended by the online tutorial (http://mzmine.sourceforege.net), a peak table that listed the m/z value, retention time, and peak area was generated using a series of processes (peak detection, chromatographic deconvolution, isotopic-peak grouping, gap filtering, and peak-list alignment). The aligned peak was then further identified as described in the results. After normalization to BAL recovery (recovered volume/instilled volume), as previously described,24 the identified lipid species were relatively quantified by normalizing with relevant internal standards because the ionization rates were similar between the lipid classes;25 the data were then further transformed to the class composition scale. The class composition scale describes the mole percentage of lipid species that compose each corresponding lipid class. Both individual quantity and class composition-scaled data were converted to CSV file format to be uploaded in MetaboAnalyst (http://www.metaboanalyst.ca/ MetaboAnalyst) and processed in accordance with the tutorial.26,27 The variables containing a missing value over 20% in each population were eliminated, and the remaining values were replaced by one-half of the minimum positive value in the data set. Afterward, the data were log-transformed and autoscaled, which were mean-centered and divided by the standard deviation of each variable. The data set was then used for further statistical analysis and the development of potential diagnostic marker candidates. Statistical Analysis and Multivariate Analysis

All of the univariate and multivariate statistical analyses were performed using MetaboAnalyst.26,27 Differentially expressed lipids (DELs) from the combined experimental groups were discovered using a nonparametric Kruskal−Wallis test, followed by a Wilcoxon rank sum test to examine “between-group” differences. During the analysis, only features with a p value less than 0.05 and a false discovery rate (FDR) less than 0.1 were considered significant. Additionally, a supervised method, partial least square discriminant analysis (PLS-DA), was performed to search for

Lipid Identification

High-throughput lipid identification was performed using lipid extract MS/MS spectra, an internal spectral library, and the online database SimLipid, version 3.0 (PremierBiosoft International, Palo Alto, CA) for a range of different fragmentation energies (10, 20, 30, 40, and 50 eV). Detailed descriptions of the identified lipid nomenclature are provided in the Supporting Information. 3921

dx.doi.org/10.1021/pr5002059 | J. Proteome Res. 2014, 13, 3919−3929

Journal of Proteome Research

Article

Figure 1. Lipid identification procedure.

■

RESULTS AND DISCUSSION

(Supplementary Figure S1F in the Supporting Information). Additionally, TG was detected over 20.5 min retention time. Although we successfully identified the specific classes, in the case of PC, PG, PS, and SM, it was difficult to identify the fatty acid moieties because the intensities of the related fragment ions were low (Figure 2A and Supplementary Figure S2B−D in the Supporting Information). In certain cases, the small peaks were observed with low-resolution and high mass-accuracy fluctuations, which prevented their exact identification (data not shown). Therefore, MS/MS spectra were generated over a series of different fragmentation energies (10−50 eV) to identify the fatty acids. During the acquisition of spectra, the m/z value with the maximum intensity of fragment ion was selected for identification. For example, as shown in Figure 2A, PC with two 16:0 fatty acid moieties was identified in BALF using m/z values from the parent and fragment peaks with maximum intensity values at different collision energies: [M + H]+ at 10 eV, [M-FA+H]+ and [HG+H]+ at 30 eV, and [FA+H]+ at 40 eV. Additional examples using other lipid species, such as LPC, PG, PS, and SM, are shown in Supplementary Figure S2 in the Supporting Information. In our analytical system, the detectable TG species in BALF coeluted as a mixture of isobaric compounds. The four [M-FA +H]+ ions corresponding to [M-14:0+H]+, [M-15:0+H]+, [M16:0+H]+, and [M-18:0+H]+ were detected at the same retention time. On the basis of accurate parent ion m/z values, this combination consisted of TG (15:0−15:0−18:0), TG (16:0−16:0−16:0), and TG (14:0−16:0−18:0) (Figure 2B). In this case, the relative quantities of the three isobaric compounds were unclear. As an alternative, the isobaric compound mixture was identified by “lipid class (number of carbons in the fatty acid moiety: number of double bonds in the fatty acid moiety)” (e.g., TG (48:0)). When the fatty acid moiety-related fragment ions of glycerophospholipids or sphingomyelin were not detected at a given fragmentation energy, the same annotation was used. Because it was difficult to simultaneously interpret multiple MS/MS scans, we used the commercial lipid-identifying software, SimLipid, to identify lipids within numerous MS/

Characteristics of Study Subjects

The subjects’ characteristics and BAL fluid cell profiles are shown in Table 1. Among the experimental groups, we did not observe significant differences in age, gender, atopic dermatitis frequency, lung function, and smoking or asthma duration. However, the number of blood eosinophils was significantly greater for the NSBA group compared with the SBA and NC groups. Additionally, significant differences in the eosinophil and macrophage percentages were observed. The NSBA eosinophil percentage was significantly higher than the control group. Conversely, the macrophage percentage in the NSBA group was lower than that of the control group. We did not detect significant differences between SBA and the other groups. Rapid, Accurate Lipid Identification Based on the Internal Spectral Library and the Online Database

The lipid identification procedure was performed, as shown in Figure 1. First, using an internal spectral library that contained authentic lipid standards, we defined the class-specific major adducts in the parent ions and structural characteristics in the fragment ions. We used the following authentic standards for internal spectral library development: LPC (17:0), PC (17:0/ 17:0), PG (17:0/17:0), PS (17:0/17:0), TG (17:0/17:0/17:0), and SM (d18:1/24:1) (Supplementary Table S1 and Figure S1 in the Supporting Information). The library was composed of MS/MS spectra for a series of fragmentation energies (10, 20, 30, 40, and 50 eV). The major ions for the six lipid classes were defined in positive ionization mode; for LPC, PC, PS, and SM, the major ions were [M + H]+; for PG, it was [M + Na]+; and for TG, it was [M+NH4]+ (Supplementary Table S1 in the Supporting Information). The different lipid classes generated specific fragment ions. For the LPC, PC, and SM classes, the fragment ion 184.07 corresponded to the polar headgroup of phosphocholine. Similarly, the fragment ion 195.00 corresponded to the polar headgroup of PG. We also observed a neutral 185.00 Da decrease for the PS-specific fragment ion. For TG, the [M-FA +H]+ ion was the most abundant in the MS/MS spectrum 3922

dx.doi.org/10.1021/pr5002059 | J. Proteome Res. 2014, 13, 3919−3929

Journal of Proteome Research

Article

Figure 2. MS/MS spectra plot for the different fragmentation energies used to identify PC (16:0/16:0) and TG (48:0). Exact MS/MS spectra for (A) PC (16:0/16:0) and (B) TG (48:0) acquired using an HPLC-QTOF-MS and a series of fragmentation energies (10−50 eV). For the MS/MS spectrum, the class-specific adducts from PC and TG were well detected as [M + H]+ and [M+NH4]+, respectively, at 10 eV. From the spectra, structural information-related fragment ions are shown in an overlay plot derived from different fragmentation energies. In PC (16:0/16:0), the fragment ion derived from loss of FA has a 496.34 m/z and the greatest intensity at 30 eV. Furthermore, the FA ion at 239.24 m/z was detected with energies greater than 20 eV; the greatest intensity was at 40 eV. In TG (48:0), four different m/z values were detected from fragment ion formed by losses of FAs. The fragment ions [M-18:0+H]+, [M-16:0+H]+, [M-15:0+H]+, and [M-14:0+H]+ corresponded to 524.48, 551.50, 565.53, and 579.53 m/z, respectively; the greatest intensities for the different fragmentation energies were as follows: [M-15:0+H]+ and [M-14:0+H]+ at 10 eV and [M18:0+H]+ and [M-16:0+H]+ at 20 eV. (PC, phosphatidylcholine; TG, triglyceride; FA, fatty acid).

minutes. Before using the HTP, we constructed suitable lipidmatching parameters based on the internal spectral library developed herein. The difference threshold between experimental and theoretical m/z values was 1, as shown in Supplementary Table S4 in the Supporting Information. Additionally, after applying the Kruskal−Wallis test followed by the Wilcoxon rank sum test and PLS-DA, a total of 34 lipid species (p value < 0.05, FDR < 0.1, and VIP > 1) showed both significant differences and discriminative ability between the NSBA and NC groups. As shown in Figure 5A, these lipids consisted of LPC (N = 3), PC (N = 20), PG (N = 1), PS (N = 2), SM (N = 4), and TG (N = 4), all of which were elevated in the NSBA group. For these lipids, we performed ROC curve analysis to confirm their suitability as lipid biomarker candidates. As shown in Figure 5B,C, the area under the ROC curve (AUROC) value was 0.903, with a 95% CI (0.673 to 1), indicating an excellent prediction of performance; the result was further validated using an 1000-iteration permutation test (p value = 0.004). Therefore, 34 lipid species in BALF were identified as reliable biomarker candidates to distinguish between the NSBA and NC groups.

Table 3. PLS-DA Results of the Cross Validation and Permutation Tests among the NSBA, SBA, and NC Groups permutation test (p value)

cross-validation individual quantity (normalized by I.S.) NSBA vs NC NSBA vs SBA SBA vs NC class composition (mole percentage) NSBA vs NC NSBA vs SBA SBA vs NC

accuracy

R2

Q2

accuracy

B/W

0.85 0.75 0.673

0.431 0.288 0.138

0.321 0.097 −0.102

0.007* 0.117 0.951

0.006* 0.112 0.952

0.75 0.706 0.66

0.319 0.854 0.269

0.013 −0.105 −0.016

0.309 0.373 0.559

0.228 0.329 0.564

separation pattern on the score plot; this model was successfully optimized and displayed acceptable performance during the validation process (Q2 = 0.321, R2 = 0.431, accuracy = 0.85, accuracy-based p value = 0.007, B/W-based p value = 0.006). As a result of the PLS-DA analysis, the VIP score was assigned to individual lipid species according to their impact on the intergroup separation. Of all comparison models, we elected to export only the VIP scores obtained from the validated and optimized PLS-DA model of the NSBA versus NC comparison, which used the individual quantity scale. Lipids with VIP scores greater than 1 were considered to have significant impact on the group separation. A total of 37 lipid

Correlational Analysis Revealed Distinctive Relationships of Inflammatory Cells and Lipid Biomarker Candidates between NSBA and NC groups

To determine the relationship between inflammatory cells and lipid biomarker candidates, we performed an additional correlational analysis. As profiling of BAL fluid cell is shown in Table 1, NSBA eosinophil percentage was significantly higher than the NC group. Conversely, the macrophage percentage in the NSBA group was lower than that of the 3925

dx.doi.org/10.1021/pr5002059 | J. Proteome Res. 2014, 13, 3919−3929

Journal of Proteome Research

Article

Figure 4. Score plot of PLS-DA for comparison of the NSBA and NC groups.

inflammatory response such as mast cell degranulation and eosinophils infiltration. S1P level increases in BALF of asthmatic patients and is involved in hyper-reactivity and remodeling of airway and acts as chemotactic signal for recruiting eosinophils.44−47 In this study, levels of eosinophil percentage and individual quantity of DELs that belong to SM have increased in NSBA group. As shown in Figure 6A, Spearman’s correlation test showed a positive correlation between eosinophil and SM (d34:0) only in NSBA group. The results support the findings of previous studies determining elevation of sphingolipids and eosinophil in response to altered metabolism of asthmatic patients by analyzing BALF, which represents respiratory conditions. Moreover, as shown in Figure 6B, macrophage percentage had strong positive correlation (correlation coefficient = 0.720, p value < 0.01) with PS (18:1/18:1) of NC group. In contrast with other cell types, nonapoptotic macrophage expresses PS on surface and removes apoptotic cells by phagocytosis.48 In general, PS is asymmetrically distributed in plasma bilayer and cells express PS in inner leaflet and increase PS expression externally during apoptosis.48 Meanwhile, macrophage percentage had no correlation with PS of NSBA group even though the concentration of PS (18:1/18:1) increased in NSBA group. From this result, we postulated that PS increment in BALF of NSBA may be due to other factors such as PS expressed in external membrane of apoptotic epithelial cells, which was induced by cooperation of T cells and eosinophils in asthmatic

NC group. It is well known that eosinophils and their mediators are associated with pathogenesis of asthma causing epithelial damage and aggravating inflammation.37,38 From Spearman’s correlation analysis, we found distinctive correlation of the inflammatory cells with lipid biomarker candidates between NSBA and NC groups. As shown in Figure 6A, eosinophil showed significant (p value < 0.05) positive correlation with PC (31:0) of NC group, whereas it showed strong positive correlation with SM (d34:0) and more various PC species such as PC (16:0−18:2), PC (38:2), and PC (32:0a) of NSBA group. Supportingly, previous studies demonstrated that a large number of activated eosinophils are present in BALF of asthmatic patients, and it stimulates PC secretion in alveolar type II cells.39−41 It is also known that SM and its metabolic products, ceramide and sphingosine-1-phosphate (S1P), are involved in inflammatory process of asthma. Ceramide is produced by hydrolysis of sphingomyelin through sphingomyelinase; then, it further converted to sphingosine and S1P by ceramidase and sphingosine-kinase.42 It was previously reported that both SM and sphingomyelinase increased in erythrocyte membrane of asthmatic patients. Upregulation of sphingomyelinase may catalyze more production of ceramide from SM.43 Supportingly, another previous study on ovalbumin-sensitized guinea pig model of asthma had shown that increased level of ceramide in airway epithelium of guinea pigs was closely related to respiratory abnormalities including reduction of latency period for cough and dyspnea.44 Ceramide mediates lung 3926

dx.doi.org/10.1021/pr5002059 | J. Proteome Res. 2014, 13, 3919−3929

Journal of Proteome Research

Article

Figure 5. Receiver operating characteristic (ROC) curve and BALF lipid biomarker candidates for the discrimination of the NSBA and NC groups. Discriminative and differentially expressed lipid species between the NSBA and NC groups (p value < 0.05, FDR < 0.1, and VIP > 1) are displayed in a box and whisker plot (A). Additionally, the ROC curve results of the significant lipid species (B) and the 1000-iteration permutation results (C) are displayed.

Figure 6. Spearman’s correlation test between lipid biomarker candidates and inflammatory cells. In the bar graph, Spearman’s correlation coefficient between lipid biomarker candidates and inflammatory cells such as eosinophils (A) and macrophage (B) in NSBA (red) and NC (blue) are displayed (significance: *, p value < 0.05; **, p value