NMR Metabolomic Analysis of Exhaled Breath ... - ACS Publications

Nov 13, 2014 - Carolina Vitale,. ‡ ... Exhaled breath condensate (EBC) is a noninvasive matrix to .... breathe tidally through a mouthpiece into a t...
0 downloads 0 Views 2MB Size
Article pubs.acs.org/jpr

NMR Metabolomic Analysis of Exhaled Breath Condensate of Asthmatic Patients at Two Different Temperatures Andrea Motta,*,†,# Debora Paris,†,# Maria D’Amato,‡,# Dominique Melck,†,# Cecilia Calabrese,§,# Carolina Vitale,‡ Anna A. Stanziola,‡ Gaetano Corso,∥ Matteo Sofia,‡,▽ and Mauro Maniscalco‡,⊥ †

Institute of Biomolecular Chemistry, National Research Council, Via Campi Flegrei 34, Pozzuoli (Naples) 80078, Italy Department of Respiratory Medicine, A.O. Dei Colli, University of Naples Federico II, Via Leonardo Bianchi 1, Naples 80131, Italy § Department of Cardiovascular and Respiratory Sciences, A.O. Dei Colli, Second University of Naples SUN, Via Leonardo Bianchi 1, Naples 80131, Italy ∥ Department of Clinical and Experimental Medicine, University of Foggia, Viale Luigi Pinto, Foggia 71122, Italy ⊥ Section of Respiratory Medicine, Hospital S. Maria della Pietà, Via San Rocco 9, Casoria (Naples) 80026, Italy ‡

S Supporting Information *

ABSTRACT: Exhaled breath condensate (EBC) collection is a noninvasive method to investigate lung diseases. EBC is usually collected with commercial/custom-made condensers, but the optimal condensing temperature is often unknown. As such, the physical and chemical properties of exhaled metabolites should be considered when setting the temperature, therefore requiring validation and standardization of the collecting procedure. EBC is frequently used in nuclear magnetic resonance (NMR)-based metabolomics, which unambiguously recognizes different pulmonary pathological states. Here we applied NMR-based metabolomics to asthmatic and healthy EBC samples collected with two commercial condensers operating at −27.3 and −4.8 °C. Thirty-five mild asthmatic patients and 35 healthy subjects were included in the study, while blind validation was obtained from 20 asthmatic and 20 healthy different subjects not included in the primary analysis. We initially analyzed the samples separately and assessed the within-day, between-day, and technical repeatabilities. Next, samples were interchanged, and, finally, all samples were analyzed together, disregarding the condensing temperature. Partial least-squares discriminant analysis of NMR spectra correctly classified samples, without any influence from the temperature. Input variables were either integral bucket areas (spectral bucketing) or metabolite concentrations (targeted profiling). We always obtained strong regression models (95%), with high average-quality parameters for spectral profiling (R2 = 0.84 and Q2 = 0.78) and targeted profiling (R2 = 0.91 and Q2 = 0.87). In particular, although targeted profiling clustering is better than spectral profiling, all models reproduced the relative metabolite variations responsible for class differentiation. This warrants that cross comparisons are reliable and that NMR-based metabolomics could attenuate some specific problems linked to standardization of EBC collection. KEYWORDS: asthma, exhaled breath condensate, metabolomics, nuclear magnetic resonance, principal component analysis



INTRODUCTION Exhaled breath condensate (EBC) is a noninvasive matrix to access the lung epithelial lining fluid without any discomfort or risk for patients. 1,2 EBC can provide complementary information to bronchoalveolar lavage and sputum induction, offering several potential applications for diagnosis and longterm follow-up.3 Volatile breath metabolites are condensed when the vapor contacts a surface whose temperature is below the saturation temperature of the vapor.4 However, optimal condensation requires that the physical and chemical properties of all exhaled compounds should be considered when setting the temperature. This is almost impossible because, in theory, each constituent requires a specific temperature, which for many volatile metabolites is still unknown. Although to a lesser © 2014 American Chemical Society

extent, other factors like design characteristics and mechanical features also affect the efficiency of condensate collection in commercial and custom-made devices.3 Therefore, each step of collection and processing procedures requires validation and standardization. Notwithstanding the above considerations, EBC is definitely being used in metabolomics studies. Nuclear magnetic resonance (NMR)-based metabolomics of EBC unambiguously recognizes biomarkers that separate children with asthma or adults with chronic obstructive pulmonary disease (COPD) from healthy subjects,5−8 discriminates patients with stable Received: October 7, 2014 Published: November 13, 2014 6107

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Table 1. Characteristics of the Study Populationa,b healthy subjects N age (y)c sex (F/M) BMI (kg/m2)

35 33 ± 1.1 16/19 21.8 ± 1.5

FEV1 (% pred)

106.9 ± 2.9

FVC (% pred)

110.1 ± 2.6

inhaled corticosteroids (y/n) short-acting ß2-agonists (y/n)

mild asthma

healthy subjects (validation set)

Anthropometric Data 35 35 ± 1.2 18/17 21.0 ± 1.6 Lung Function 79.3 ± 2.2d p < 0.0007 83.4 ± 1.6d p < 0.0009 Treatment Ongoinge 31/4 29/6

mild asthma (validation set)

20 29 ± 0.8 12/8 19.8 ± 0.5

20 31 ± 1.2 12/8 20.1 ± 0.8

106.4 ± 1.9

77.1 ± 3.3d p < 0.001 82.4 ± 2.0d p < 0.0006

116.1 ± 2.7

18/2 17/3

Data are expressed as n or mean ± SD. One-way ANOVA and unpaired t tests were used for comparing groups. Significance was defined as p < 0.05. bBMI, body mass index; FEV1, forced expiratory volume in one second; FVC, forced vital capacity. cAge range: healthy subjects: 24−44 years; mild asthma: 30−53 years; healthy subjects (validation test): 23−50; mild asthma (validation set): 25−46 years. dCompared with healthy subjects. e Inhaled budesonide at a dose of 400 μg once daily and inhaled beclomethasone at a dose of 200 μg b.i.d. were given via a spacer device. Inhaled salbutamol at a dose of 200 μg b.i.d. was given using a metered dose inhaler, as needed. a

cystic fibrosis (CF) from those with unstable CF,9 discriminates primary ciliary dyskinesia from CF,10 and efficiently investigates smoking-related lung diseases.11 In this study, we applied NMR-based metabolomics to asthmatic and healthy EBC samples using two different commercial condensers, ECoScreen 12 and TURBO− DECCS,13 operating at −27.3 and −4.8 °C, respectively. We initially assessed the within-day, between-day, and technical repeatabilities of all samples. Next, we independently analyzed the samples obtained with each condenser. Subsequently, samples were interchanged comparing healthy and asthmatic subjects collected at different temperature. Finally, all samples were analyzed together, independently from the condensing temperature. To discriminate different EBC samples through NMR spectra, we carried out a multivariate statistical data analysis using projection methods, and input variables were either integral bucket areas (spectral bucketing) or metabolite concentrations (targeted profiling). Although higher clustering and better quality parameters were observed in targeted profiling models, both approaches yielded reliable and robust regression models and identified the same metabolites as responsible for healthy and asthmatic class separation. In particular, patients and controls were always separated with no influence from the condensing temperature, suggesting that NMR-based metabolomics analysis of EBC is potentially independent from the collecting device. Therefore, our approach could attenuate some specific problems linked to standardization of the EBC collection, and NMR-based metabolomics of EBC could become a general method in the diagnosis and follow-up of pulmonary pathologies.



Respiratory Medicine. Subjects gave informed written consent, and the study was approved by the Clinical Research Ethics Committee. Patients, classified according to Global Initiative for Asthma (GINA) guidelines,14 had a history of bronchial asthma for at least 6 months prior to collection and no acute exacerbation (episodes of progressive increase in shortness of breath, cough, wheezing, or chest tightness or combination of these symptoms) within the past 1 month. Asthma severity was also assessed by a disease severity score (DSS) that included daytime, nighttime, and exercise-induced symptoms, frequency of asthmatic exacerbation, and the use of antiasthma medications.15 Patients with significant comorbidities were excluded. Controller and rescue therapies for asthma remained unchanged for 3 months before the study. Controls had no history of asthma, atopic disease, respiratory disorders, or other diseases, were nonactive smokers, and were not exposed to passive smoking. EBC samples were collected using ECoScreen and TURBO− DECCS condensers in a crossover trial during two sampling periods. Subjects were allocated randomly to two different groups, and the sequential order of the condensers was also randomized. Our randomization design was in line with previous reports addressing comparison of different condensation temperatures in a sequential manner.13 Power Analysis

Ideally, power calculations should be based on the expected changes in biomarker concentrations and on their variability. However, before analysis, biomarkers and their concentration changes that could determine class separation were unknown. Therefore, we used our results to backward evaluate the power of our analysis. By changing the parameters 1-α from 95 to 99.9% and 1-β from 80 to 99.9% and using for Unexposed and Exposed subjects the accuracy percentages obtained for our validation tests (see Results section), we obtained, for 1-α = 95% and 1-β = 80%, a number of subjects corresponding to 12 ± 2 healthy subjects and 13 ± 3 asthmatic patients, while for 1α = 99.9% and 1-β = 99.9%, we obtained 17 ± 3 healthy subjects and 18 ± 3 asthmatic patients, which are smaller than those used in our study, 35 healthy subjects and 35 asthmatic patients. It should be emphasized that usually 1-α = 95% and 1β is at least equal to 80% and that a value of 99.9% represents

MATERIALS AND METHODS

Subjects

A study population of 35 mild asthmatic patients and 35 agematched healthy controls was included in the primary analysis. To validate the model, we tested 20 healthy subjects and 20 mild asthmatic patients, not included in the primary analysis, blindly. Demographic characteristics of all subjects are shown in Table 1. Asthmatic patients were from the Department of Respiratory Medicine, Federico II University, Naples, while healthy subjects were recruited from staff at the Department of 6108

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Pioltello (MI), Italy) tested against particles approximately 0.3 μm in size] for respiratory protection were applied to the oneway valve of the mouthpiece condensers used for the whole set of experiments. NMR spectra of condensed room air from both devices were devoid of signals, confirming the absence of air pollutants (data not shown). To reduce the risk of contamination by inhaling hospital air, we sampled controls before starting work, and, as for patients, after staying 45 min in the greenhouse of the Department of Respiratory Medicine, which was shown to be contaminant free as previously described for the collecting room.

an extreme requirement. By considering the number of exposed, risk of disease among exposed, number of unexposed, and risk of disease among nonexposed as the parameters derived from our test on healthy subjects and asthmatic patients, we obtained approximation of the model of 100% in both 1-α = 95 and 99.9%. For projection methods like PLS-DA, there are no standardized methods for evaluating the power of the analysis. Therefore, we consider our study as a pilot study for which no a priori power analysis was possible. The data obtained could be used for an a posteriori power analysis, but the current methods appear not sufficiently robust to warrant a satisfactory analysis for PLS-DA. The permutation and the validation tests done within the PLS-DA confirmed the existence and validity of the model and avoided the overfitting problem (see later).

NMR Sample Preparation

EBC samples were rapidly defrosted. To provide a field frequency lock, we added 70 μL of a 2H2O solution [containing 0.1 mM sodium 3-trimethylsilyl [2,2,3,3-2H4] propionate (TSP) as a chemical shift reference for 1H spectra and sodium azide, a bacteriostatic agent, at 3 mM] to 630 μL of EBC reaching 700 μL of total volume.

EBC Sampling

ECoScreen (Jäger, Hö chberg, Germany) and TURBO− DECCS (ItalChill, Parma, Italy) collection of EBC was achieved as previously described.6,9 Condensing temperature was measured before and after each sample collection, obtaining, as average values, −27.3 ± 1.2 °C for ECoScreen and −4.8 ± 0.8 °C for TURBO−DECCS. The room temperature remained constant (24 ± 1.0 °C) throughout the sampling period. For ECoScreen, subjects were asked to breathe tidally through a mouthpiece into a two-way nonrebreathing valve, which also served as a saliva trap, for 15 min at the average collecting temperature of −27.3 °C, wearing a nose-clip. The same procedure was followed with the TURBO−DECCS condenser but at the average collecting temperature of −4.8 °C. The subjects collected their EBC all in the same room in 1 week. There was a clear trend toward increasing EBC sample volumes with decreasing collection temperatures.13 EBC samples were centrifuged for 1 min at 1000g immediately after collection so that all of the water droplets were driven to the bottom of the flask. They were immediately transferred into 5 mL glass vials, closed with 20 mm butyl rubber lined with PTFE septa, and crimped with perforated aluminum seals. Before sealing, volatile substances were removed from samples by a gentle nitrogen gas flow for 3 min. Samples were not dried out to avoid their precipitation, with a possible loss of nonvolatile compounds, or formation of aggregates upon dissolving the dried condensate for NMR measurements. Sealed samples were then frozen in liquid nitrogen so as to immediately “quench” metabolism and preserve the metabolite concentrations. The samples were stored at −80 °C until NMR analysis, which was performed within 1 week from sampling time. The salivary contamination of the samples was tested by measuring their α-amylase activity.16 The colorimetric reaction (Infinity amylase reagent, Sigma, Milan, Italy) is based on the hydrolysis of starch. The absorption was detected at 540 nm with a spectrophotometer (HP8453E Agilent Technologies Italia S.p.A., Cernusco s/N (MI), Italy). The limit of detection was 0.078 U/mL that corresponds to a 1000-fold dilution of saliva. None of the EBC samples contained detectable levels of amylase activity. Possible air contaminants in the EBC collecting room17 were monitored using a dedicated sampling pump for air monitoring (Zambelli Ego Plus TT, Zambelli S.r.l. Bareggio (MI), Italy), working at a flow rate of 8 L/min and tidal volume (500 mL) into both condensers, so as to simulate human breath.18 The pump was connected to the condensers outlet for 15 min, and special filters [3 M Particulate Filters P100 (3 M Italia S.p.A.

NMR Spectroscopy Measurements

All spectra were recorded on a 600 MHz Bruker Avance III spectrometer (Bruker BioSpin, Rheinstetten, Germany) equipped with a CryoProbe. 1D 1H NMR spectra were collected at 300 K with the excitation sculpting pulse sequence19 to suppress the water resonance. We used a double-pulsed field gradient echo, with a soft square pulse of 4 ms at the water resonance frequency and the gradient pulses of 1 ms each in duration, adding 128 transients of 64k complex points, with an acquisition time of 4 s/transient. Time-domain data were all zerofilled to 128k complex points, and prior to Fourier transformation, an exponential multiplication of 0.6 Hz was applied. 2D clean total correlation spectroscopy (TOCSY) spectra20 were recorded using a standard pulse sequence, incorporating the excitation sculpting sequence for water suppression. In general, 320 equally spaced evolution-time period t1 values were acquired, averaging four transients of 2048 points. Time-domain data matrices were all zero-filled to 4096 points in both dimensions, and prior to Fourier transformation a Lorentz-to-Gauss window with different parameters was applied for both t1 and t2 dimensions for all of the experiments. Both homonuclear 1D and 2D spectra were referred to 0.10 mM TSP, assumed to resonate at δ 0.00. For the natural abundance of 2D 1H−13C heteronuclear single quantum coherence (HSQC) spectra, we used an echo− antiecho phase-sensitive pulse sequence using adiabatic pulses for decoupling.21 128 equally spaced evolution time period t1 values were acquired, averaging 48 transients of 2048 points and using GARP4 for decoupling. The final data matrix was zero-filled to 4096 in both dimensions and apodized before Fourier transformation by a shifted cosine window function in t2 and in t1. Linear prediction was also applied to extend the data to twice their length in t1. Spectra were referred to the lactate doublet (βCH3) resonating at 1.33 ppm for 1H and 20.76 ppm for 13C. Spectral and Statistical Analysis

Before analysis, within-day, between-day, and technical repeatabilities and detection limit were assessed at −4.8 and −27.3 °C following published protocols.9,11 Within-day repeatability was checked collecting two EBC samples for each subject (at times 0 and 6 h) within the same day at both temperatures, for a total of eight subjects (four healthy controls, four asthmatic patients). Each spectrum was subdivided in six 6109

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

bucketing indicated that a better discrimination could be obtained with buckets of 0.04 ppm, without including too many artifacts to describe the same signal. This is because signals are essentially isolated in our spectra. Furthermore, no pH dependence of resonances was observed; therefore, no spectral alignment was required in all studied samples. To validate the bucketing procedure, we also applied targeted profiling.24 Resonances were all assigned by 2D experiments (see above), and an iterative program based on LAOCN326 was used for spectral simulation of the experimental EBC spectra. Essentially, a pattern of Lorentzian lines for each metabolite was generated and superimposed in a trial-and-error fit to the experimental EBC spectra, therefore simulating resonance overlap. Peaks were quantified by referring to the concentration of the internal TSP reference signal therefore determining the concentration of individual compounds. To discriminate different EBC samples through NMR spectra, we carried out a multivariate statistical data analysis using projection methods as implemented in SIMCA-P+ 12 package (Umetrics, Umeå, Sweden). Input variables were either integral bucket areas (spectral bucketing) or metabolite concentrations (targeted profiling). All variables used for quantitative analysis were preprocessed with Pareto scaling, and separate models were built for both procedures. For spectral bucketing, each region was scaled to (1/Sk)1/2, where Sk is the standard deviation for the variable k, increasing the contribution of lower concentration metabolites to the models generated compared with models where no scaling is used. The input variables k for targeted profiling data were the percent error in concentration determination of each metabolite over all EBC samples.24 The model quality was evaluated by the goodness-of-fit parameter (R2) and the goodness-of-prediction parameter (Q2). For them, acceptable values must be ≥0.5, with |R2 − Q2| < 0.2 to 0.3.27 PCA was first applied to detect EBC metabolites trends and clusterings in an unsupervised (i.e., no prior group knowledge is used in the calculation) manner. However, to better identify clusterings, the spectral filtering orthogonal signal correction (OSC) routine28 together with the partial least-squares discriminant analysis (PLS-DA) was applied to reinforce classification. In addition, a permutation test (n = 300) was carried out to assess possible overfit of the model. Principal components are continuously added until the Q2 parameter (the goodness-of-prediction and the predicted variation) does not further improve and automatically avoids overparametrization or modeling of the noise. The final data were examined applying PLS-DA to discriminate between healthy subjects at different temperatures and healthy subjects and patients with asthma permuting all collecting temperatures. Demographic data are presented as mean ± SD after assessing for normality with D’Agostino−Pearson’s omnibus normality test. Normally distributed values were compared using the paired Student’s t test. If the normality test failed, the Wilcoxon signed-rank test was employed. Group differences were explored by one-way analysis of variance (ANOVA), followed by posthoc multiple comparisons according to Tukey’s test. Intraclass correlation analysis (ICC) was performed for each group to estimate the reliability of single measurements. Significance was defined as a value of p < 0.05.

regions (region 1: 8.8 to 6.8 ppm; region 2: 6.8 to 5.2 ppm; region 3: 4.5 to 3.5 ppm; region 4: 3.5 to 2.5 ppm; region 5: 2.5 to 1.5 ppm; region 6: 1.5 to 0.4 ppm), while the 5.2 to 4.5 ppm region was excluded because of the residual water resonance. As previously suggested, to avoid possible variation of metabolite concentrations due to differences in volume during EBC collection, the six regions were integrated and normalized to the total spectrum area.5 We obtained six parameters (the integrated fractional regions) for each spectrum, which, for eight selected spectra (subjects), at −4.8 and −27.3 °C gave 96 total values. The SD was within ±1.96 SD in 47 out of 48 samples at each temperature, which, according to the Bland− Altman test, indicates good within-day repeatability. For assessing between-day repeatability, which was expressed as intraclass correlation coefficient (ICC), three EBC samples, collected from the above eight subjects on days 1, 3, and 5, were analyzed with NMR spectroscopy at each temperature. The 4.4−0.4 ppm spectral area was integrated and normalized to the total spectrum area. The determined ICC was 0.99. Technical repeatability was assessed by acquiring NMR spectra of two different samples (one from healthy subjects and one from an asthmatic patient) 10 times consecutively at each temperature. The ICC for the 4.4−0.4 ppm spectral region was 0.99. The detection limit of our NMR technique was calculated by integrating 10 EBC spectra; they were normalized to the standard TSP (0.10 mM) internal reference, obtaining an average concentration of 0.12 ± 0.04 μM for the endogenous phenylalanine (signals at 7.3 to 7.0 ppm), among the lowest detected signals. Because NMR spectra show hundreds of resonances, the presence of discriminating elements (for example, signals originating from specific metabolites) in a series of spectra is often undetectable by visual inspection due to the inherent spectral complexity generated by line overlapping and is better highlighted by multivariate analysis, which carefully identifies hidden phenomena and trends in ensembles of spectra.22 The most often used statistical techniques are Principal Component Analysis (PCA) and Soft Independent Modeling of Class Analogy (SIMCA). Phased- and baseline-corrected 1H NMR spectra were automatically data reduced to 200 integral segments (“buckets”), each of 0.04 ppm, using the AMIX 3.6 software package (Bruker Biospin, Rheinstetten, Germany) between the 0.10−8.60 ppm spectral region. The residual water resonance region (5.20 to 4.50 ppm) was excluded, and each integrated region was normalized to the total spectrum area to avoid possible signal variation due to dilution of EBC samples, which could result in artificially increasing of the signals. The bucketing value was chosen as follows. Because EBC spectra present a limited number of peaks, with minimal overlap if compared with urine or serum spectra, the line width for singlets, doublets, triplets, and quartets was measured, obtaining, respectively, 10−15, 15−20, 30−40, and 40−50 Hz. Therefore, we evaluated different bucketing values (0.01, 0.02, 0.04, 0.08, 0.12, and 0.16 ppm), which include the found line width (corresponding to buckets ranging between 0.02 and 0.08 ppm). Compared with the selected 0.04-ppm bucket value, 0.12 and 0.16 ppm presented lower discrimination between classes because a single bucket covers more than one signal in the regions 2.3 to 2.1 and 1.4 to 1.0 ppm, therefore reducing the information content.23,24 Identically, lower discrimination was observed for 0.01 ppm buckets because there is no extra information gain if the bucketing resolution is higher than the intrinsic line width.25 A comparison of 0.02 and 0.04 ppm



RESULTS Figure 1 reports the representative NMR profiles (spectra) of EBC samples from healthy subjects and mild asthmatic patients 6110

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

temperatures (columns 3−6). Finally, all samples were considered together (column 7). Except for comparison of asthma samples (column 2), which we were unable to classify, the unsupervised PCA gave a sample classification of ca. 60% (samples’ classes correctly identified) for all of the analyses highlighted in Table 2 (data not shown). Application of OSCfiltered PLS-DA allowed stronger (95%) regression models for all comparisons but not asthma samples. Both PCA and PLSDA detected a single outlier, the same healthy subject detected at both temperatures, whose presence did not weaken the classification. Condensation at −4.8 °C and −27.3 °C Influences EBC Samples of Healthy Subjects but Not Those of Mild Asthmatic Patients

Starting from PCA sample clustering, a supervised PLS-DA analysis based on spectral bucketing, with the OSC spectral filtering, was carried out to identify specific metabolic changes and better define classification. The model showed a strong regression (95%, p < 0.001) between healthy groups obtained at −4.8 and −27.3 °C (Figure 2A). The scores of EBC samples (each point represents a patient) at −4.8 °C (black squares) are clearly separated from those at −27.3 °C (red triangles) and are described by R2 = 0.83 and Q2 = 0.81. The clear separation originates from metabolic differences and is not influenced by other variable parameters. No within-class or between-class discrimination due to age, sex, order of sampling, and time and carryover effects was observed at both temperatures. Although EBC spectra present a limited number of signals (Figure 1), the use of constant buckets may still complicate the analysis because more than one bucket can represent the same compound, or some buckets may include more than one metabolite. The metabolite concentrations (targeted profiling), obtained after spectral simulation (see Materials and Methods), were also used, obtaining a strong regression model (95%; p < 0.0001). Figure 2B depicts the OSC-filtered PLS-DA analysis of direct compound profiling (R2 = 0.91 and Q2 = 0.90); with respect to bucketing, it shows a more compact clustering. However, the plots reporting the contribution of each bucket to the overall model and of each compound for the targeted profiling model (not shown) suggested that both approaches well describe the EBC analysis of healthy subjects at −4.8 and −27.3 °C. We also determined R2 and Q2 values of each variable, for which a value of one implies complete data correlation and 100% prediction for a given data set. In general, Q2 < R2, but a large difference between them, with a small Q2, is undesirable, and a negative Q2 indicates that the model is not predictive.27 The bucketing model (Figure 2A) presents relatively good R2 and Q2 values for all variables, implying that all buckets fit the model relatively well and are reasonably well-predicted using cross-validation. Accordingly, artifacts from solvent suppression, uneven baseline, peak overlap, and chemical shift variation do not influence our EBC spectral analysis. Higher R2 and Q2 values were observed for concentration-based profiling (Figure 2B), essentially because all of the buckets (spectral regions) of a metabolite become a single variable, therefore drastically reducing possible artifacts. The metabolites responsible for between-class differences observed for healthy subjects at −4.8 and −27.3 °C were identified in the corresponding loadings plot of the direct compound profiling and are, in order of importance, succinate, hippurate, pyruvate, methanol, saturated fatty acids (SFA), Val, Phe, trimethylamine, acetate, urocanic acid, ethanol, 4OH-

Figure 1. NMR spectra of EBC samples. Representative 1-D 1H spectra of (A) a healthy subject (HS) at −4.8 °C, (B) a HS at −27.3 °C, (C) a mild asthmatic patient at −4.8 °C, and (D) at −27.3 °C. Only the 4.5 to 0.5 ppm region is represented.

collected at −4.8 °C with TURBO−DECCS and at −27.3 °C with ECoScreen. All 1D profiles at −4.8 °C (spectra 1A and 1C) and at −27.3 °C (spectra 1B and 1D) are characterized by sharp lines (resonances), which were assigned to single metabolites by resorting to 2D 1H−1H TOCSY and 1H−13C HSQC experiments (not shown). No carbohydrate signals were observed, therefore ruling out the presence of saliva.6 A comparison between spectra of healthy subjects (Figures 1A,B) indicates that the spectrum at −4.8 °C reproduces all main resonances of the −27.3 °C spectrum, although less intense. However, some minor peaks are observed at −4.8 °C (open circles in Figure 1A) and not at −27.3 °C (Figure 1B), namely, signals from Val (0.96 ppm), isobutyrate (1.14 ppm), Ala (1.46 ppm), Pro (2.04 ppm), pyruvate (2.39 ppm), succinate (2.42 ppm), trimetylamine (2.91 ppm), and creatine (3.02 and 3.93 ppm). On the contrary, choline moiety (3.43 and 3.21 ppm) and Met (1.97 ppm) signals (open squares in Figure 1B) were observed only at −27.3 °C and not at −4.8 °C. Spectra of mild asthmatic patients (Figure 1C,D) appear to be more uniform, essentially presenting two major differences, namely, pyruvate (2.39 ppm) and trimetylamine (2.91 ppm), labeled with open circles in Figure 1C. Comparison between the condensers was achieved as follows (Table 2). We separately compared healthy and asthmatic samples collected at −4.8 and −27.3 °C (columns 1 and 2); then, we allowed for all possible permutations between the 6111

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Table 2. EBC Metabolites Inducing Class Separation between Mild Asthma Patients and Healthy Subject Controls at −4.8 and −27.3 °C Collecting Temperaturesa,b 1

2

3

4

5

6

7

HS (−4.8 °C) vs HS (−27.3 °C)c

asthma (−4.8 °C) vs asthma (−27.3 °C)d

HS vs asthma (−4.8 °C)e

HS vs asthma (−27.3 °C)e

HS (−4.8 °C) vs asthma (−27.3 °C)e

HS (−27.3 °C) vs asthma (−4.8 °C)e

HS vs asthma (all)e

succinate hippurate

↑ ↑

− −

SFA Val

↓ ↓

pyruvate methanol SFA Val Phe

↑ ↑ ↓ ↓ ↑

− − − − −

↓ ↓ ↑ ↓ ↓

trimethylamine acetate urocanic acid ethanol 4OHphenylacetate adenosine propionate formate choline/Pcholine lactate

↑ ↓ ↓ ↑ ↓

− − − − −

adenosine hippurate Ala formate urocanic acid Pro ●acetate ethanol methanol Ile

↓ ↓ ↓ ↓

− − − −





↑ ↓ ↑ ↑ ↑

propionate 4OHphenylacetate Val ●acetate SFA Pro Tyr

↑ ↑

uracil urocanic acid

↑ ↓

Phe ■succinate

↑ ↑

SFA Val

↓ ↓

↓ ↑ ↓ ↑ ↓

■succinate SFA Phe hippurate trimethylamine

↓ ↓ ↑ ↓ ↓

Val propionate SFA methanol uracil

↓ ↑ ↓ ↑ ↑

↑ ↑ ↓ ↓ ↓

Arg trans-aconitate Phe

↑ ↑ ↑

Val Tyr

↓ ↓

Pro formate isobutyrate urocanic acid

↑ ↓ ↑ ↓

propionate Pro formate hippurate urocanic acid isobutyrate Phe

↑ ↑

a

Total comparison, including all samples, is reported as the rightmost column. bMetabolites, in order of importance, responsible for between-class differences were identified in the loadings plots of the direct compound profiling. c↑, Higher in HS at −4.8 °C (lower in HS at −27.3 °C). ↓, lower in HS at −4.8 °C (higher in HS at −27.3 °C). dNo separation was obtained for asthma samples at −4.8 and −27.3 °C with both PCA and PLS-DA. e Metabolites that present opposite behavior at different temperature are indicated as follows: ●acetate and ■succinate. trans-Aconitate is tentatively assigned. ↓, Lower in asthma (higher in HS); ↑, higher in asthma (lower in HS).

Analogously, both classes (asthmatics, blue triangles; healthy subjects, red triangles) are separated at −27.3 °C using spectral bucketing (95%, p < 0.003, Figure 4A), with R2 = 0.86 and Q2 = 0.79, and targeted profiling (95%, p < 0.0005, Figure 4B), the latter again showing better clustering and quality parameters (R2 = 0.91 and Q2 = 0.87). The metabolites responsible for asthmatics and controls separation at −4.8 and −27.3 °C, identified in the loadings plots (not shown) corresponding to the targeted profiling scores plots of Figure 3B and 4B, were SFA, Val, adenosine, hippurate, Ala, formate, urocanic acid, Pro, acetate, ethanol, methanol, and Ile at −4.8 °C (Table 2, column 3) and propionate, 4OH-phenylacetate, Val, acetate, SFA, Pro, Tyr, Arg, trans-aconitate, and Phe at −27.3 °C (Table 2, column 4). For the spectral bucketing, the same metabolites were identified, therefore confirming the robust correspondence between bucketing and targeted profiling for EBC samples. Models were tested blindly by using a sample set obtained from subjects not included in the primary analysis (20 healthy subjects and mild 20 asthmatic patients; Table 1). For the spectral bucketing, the −4.8 °C model correctly identified 19 out of 20 healthy subjects and 18 out of 20 asthmatic patients (93% accuracy), with a specificity of 95% and a sensitivity of 90%. The −27.3 °C model correctly identified 18 out of 20 healthy subjects and 17 out of 20 asthmatic patients (88% accuracy), with a specificity of 90% and a sensitivity of 85%. The targeted profiling models at −4.8 and −27.3 °C correctly identified 19 out of 20 healthy subjects and 19 out of 20 asthmatic patients (95% accuracy), with a specificity of 95% and a sensitivity of 95%.

phenylacetate, adenosine, propionate, formate, choline/phosphorylcholine, and lactate (Table 2, column 1). For the bucketing procedure, in the variables of importance plot (VIP, not shown), we identified buckets corresponding to chemical shifts of the same metabolites, confirming a robust correspondence between bucketing and targeted profiling for EBC samples. Unsupervised PCA of mild asthmatic EBC samples collected at both temperatures gave a classification of 12 and 14% (samples’ classes correctly identified) for spectral bucketing and targeted profiling, respectively (data not shown), which did not improve using PLS-DA with the OSC spectral filter. In fact, we obtained R2 = 0.13 and Q2 = 0.13 for the bucketing procedure and R2 = 0.21 and Q2 = 0.22 for the targeted profiling, and the absence of class separation in the Supplementary Figure S1 in the Supporting Information indicates that asthmatic EBC samples are not significantly affected by the condensing temperature. EBC of Mild Asthma and Healthy Subjects Can Be Discriminated at −4.8 and −27.3 °C

As expected, mild asthmatic EBC samples can be distinguished from those of healthy subjects at −4.8 and −27.3 °C, using both spectral bucketing and targeted profiling. Statistical analysis of spectral bucketing data with the OSC-filtered PLSDA yielded a strong regression model (95%, p < 0.004) for samples collected at −4.8 °C (Figure 3A). The scores of asthmatic patients (gray squares) are clearly separated from healthy subjects (black squares), presenting R2 = 0.84 and Q2 = 0.79. In the targeted profiling OSC-filtered PLS-DA model (95%, p < 0.0003), the scores are better clustered (Figure 3B), showing higher R2 (= 0.90) and Q2 (= 0.84) parameters. 6112

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Figure 2. Partial least-squares-discriminant analysis (PLS-DA) of EBC of healthy subjects (HS) and mild asthmatic patients. Score plots showing the degree of separation of the model between HS at −27.3 °C (red triangles) and −4.8 °C (black squares), whose NMR data were analyzed by (A) spectral bucketing (quality parameters R2 = 0.83 and Q2 = 0.81) and (B) targeted profiling (R2 = 0.91 and Q2 = 0.90). Both models showed a strong regression (95%), but the targeted profiling data are well clustered with little difference within the subjects, except for an outlier that does not weaken classification. The labels t[1] and t[2] along the axes represent the scores (the first two partial least-squares components) of the model, which are sufficient to build a satisfactory classification model.

EBC of Mild Asthma and Healthy Subjects can Always be Discriminated, Independently of the Collection Temperature

data with the OSC-filtered PLS-DA yielded a strong regression model (95%, p < 0.009) comprising the scores of healthy subjects (black squares) collected at −4.8 °C and those of asthmatic patients (blue triangles) collected at −27.3 °C (Figure 5A). Asthmatic patients are clearly separated from healthy subjects, presenting high R2 (= 0.87) and Q2 (= 0.76) parameters, indicating that the model is powerful and reliable. The targeted profiling model (95%, p < 0.0007) shows a clear separation (Figure 5C), with better clustering for each class and

We also evaluated the possibility of interchanging healthy and mild asthmatic samples collected at −4.8 and −27.3 °C using both the spectral bucketing and the targeted profiling approaches. Asthmatic EBC samples can always be distinguished from those of healthy subjects, independently from the collecting temperatures. Statistical analysis of spectral bucketing 6113

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Figure 3. Partial least-squares discriminant analysis (PLS-DA) of EBC of healthy subjects (HS) and mild asthmatic patients. Score plots showing the degree of separation of the model between HS (black squares) and asthmatic patients (gray squares) at −4.8 °C, whose NMR data were analyzed by (A) spectral bucketing and (B) targeted profiling. The models showed a strong regression (95%), with the following quality parameters for (A) R2 = 0.84 and Q2 = 0.79 and (B) R2 = 0.90 and Q2 = 0.84. The labels t[1] and t[2] along the axes represent the scores (the first two partial least-squares components) of the model.

presents R2 = 0.92 and Q2 = 0.85. The corresponding loadings plot indicates that healthy subjects at −4.8 °C and asthmatic patients at −27.3 °C can be separated by uracil, urocanic acid, succinate, SFA, Phe, hippurate, trimethylamine, Val, and Tyr (Table 2, column 5), which are also found for the spectral bucketing. The scores of healthy subjects (red triangles) collected at −27.3 °C and those of asthmatic patients (gray squares) collected at −4.8 °C are also well-classified (Figure 5B) in the spectral bucketing model (95%, p < 0.008), showing

R2 = 0.83 and Q2 = 0.80. Clustering is enhanced in the targeted profiling model of Figure 5D (R2 = 0.89 and Q2 = 0.89), with Phe, succinate, Val, propionate, SFA, methanol, uracil, Pro, formate, isobutyrate, and urocanic acid as separating metabolites (Table 2, column 6). These results indicate that classes are well-separated independently from the collection temperature of the condensing device and that, although targeted profiles yields a stronger classification, spectral bucketing is also able to build satisfactory regression models. 6114

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Figure 4. Partial least-squares-discriminant analysis (PLS-DA) of EBC of healthy subjects (HS) and mild asthmatic patients. Score plots showing the degree of separation of the model between HS (red triangles) and asthmatic patients (blue triangles) at −27.3 °C, whose NMR data were analyzed by (A) spectral bucketing and (B) targeted profiling. The models showed a strong regression (95%), with the following quality parameters for (A) R2 = 0.86 and Q2 = 0.79 and (B) R2 = 0.91 and Q2 = 0.87. The labels t[1] and t[2] along the axes represent the scores (the first two partial least-squares components) of the model.

We finally compared all healthy subjects and asthmatic patients at both collecting temperatures. OSC-filtered PLS-DA of spectral bucketing data yielded a strong regression model (95%, p < 0.0019), with Figure 6A depicting the scores of healthy samples (n = 70, blue triangles at −27.3 °C, and black squares at −4.8 °C) and those of asthmatic samples (n = 70, gray squares at −4.8 °C, and red triangles at −27.3 °C). The scores of EBC healthy samples are clearly separated from asthmatic patients and are described by R2 = 0.81 and Q2 = 0.75, which confirm a powerful and reliable model. A better

regression is obtained with the targeted profiling model (Figure 6B), described by R2 = 0.91 and Q2 = 0.87. The associated loadings plot indicates that classes can be separated by SFA, Val, propionate, Pro, formate, hyppurate, urocanic acid, isobutyrate, and Phe (Table 2, column 7), also identified in the spectral bucketing. To validate the total spectral bucketing model, a sample set obtained from subjects not included in the primary analysis (20 healthy subjects and 20 asthmatic patients) was tested blindly (Table 1). The model correctly identified 18 of 20 healthy subjects and 18 of 20 asthmatic patients (90% 6115

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Figure 5. Partial least-squares discriminant analysis (PLS-DA) of EBC of healthy subjects (HS) and mild asthmatic patients. Score plots showing the degree of separation of the model between HS (black squares) at −4.8 °C and asthmatic patients (blue triangles) at −27.3 °C, using (A) spectral bucketing and (C) targeted profiling and between HS (red triangles) at −27.3 °C and asthmatic patients (gray squares) at −4.8 °C using (B) spectral bucketing and (D) targeted profiling. The models showed a strong regression (95%), with the following quality parameters (A) R2 = 0.87, Q2 = 0.76; (B) R2 = 0.83, Q2 = 0.80; (C) R2 = 0.92, Q2 = 0.85; and (D) R2 = 0.89, Q2 = 0.89. The labels t[1] and t[2] along the axes represent the scores (the first two partial least-squares components) of the model.

temperature of TURBO−DECCS, NMR resonances are less intense (i.e., corresponding metabolites show lower concentration) for both healthy and asthmatic subjects (Figure 1). The intensity reduction is not a generalized scaling-down effect because some extra peaks are observed at both temperatures. Differences are stronger for healthy (spectra 1A and 1B) than for asthmatic (spectra 1C and 1D) subjects, and new peaks are more numerous at −4.8 °C than at −27.3 °C in both classes, confirming that the optimal collection temperature is not the lowest but is strictly dependent on the physicochemical properties of exhaled compounds. We started investigating the healthy group collected at −4.8 and −27.3 °C. By using spectral bucketing and targeted profiling at both temperatures, we identified two different subject groups, which separated because of succinate, hippurate, pyruvate, methanol, SFA, Val, Phe, trimethylamine, acetate, urocanic acid, ethanol, 4OH-phenylacetate, adenosine, propionate, formate, choline/phosphorylcholine, and lactate (Table 2, column 1). The diverse profiles observed for healthy subjects do not represent “different healthy phenotypes” because EBC samples are from the same cohort of subjects, and differences in metabolite concentrations are only due to the condensation temperature. Therefore, the previously described metabolites are very sensitive to the condensing temperature, and care should be taken when comparing their concentration in samples of healthy subjects collected at different temperatures. On the contrary, no classification model was obtained for asthmatic patients, suggesting that their sampling at −4.8 and −27.3 °C does not significantly alter the metabolic profiles.

accuracy) with a specificity of 90% and a sensitivity of 90%. The targeted profiling model correctly identified 19 out of 20 healthy subjects and 19 out of 20 asthmatic patients (95% accuracy), with a specificity and a sensitivity of 95%. Correlations

We found no correlation between metabolomic data and the anthropometric and clinical data as well as the pharmacological treatment reported in Table 1.



DISCUSSION NMR-based metabolomics of EBC has shown a high degree of accuracy in diagnosing respiratory diseases.5−11 However, the lack of standardization of collection techniques as well as of a universal collecting device29 makes EBC often unsuitable in large multicenter studies because absolute values from different devices may not be directly comparable. The condensate composition mainly depends on the cooling temperatures,30,31 and highly variable and contrasting results have been reported for EBC metabolites. Different levels of eotaxin, cysteinylleukotriene, 8-isoprostane, ammonia, hydrogen peroxide, albumin, mucin, nitrogen oxides (NOx), total protein, and pH32−38 as well as EBC volume31 have been reported. By NMR-based metabolomics we investigated EBC samples from the same cohort of subjects collected with TURBO− DECCS and ECoScreen condensers. Those condensers were previously shown to be reproducible,13,38 but a direct comparison using EBC from the same subjects has never been reported. Healthy and mild asthmatic subjects showed full within-day, between-day, and technical repeatabilities at both −4.8 and −27.3 °C. Because of the higher condensing 6116

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

Figure 6. Partial least-squares discriminant analysis (PLS-DA) of EBC of healthy subjects (HS) and mild asthmatic patients. Score plots showing the degree of separation of the model between HS at −27.3 °C (black triangles) and at −4.8 °C (black squares) and asthmatic patients at −27.3 °C (blue triangles) and at −4.8 °C (gray squares) using (A) spectral bucketing, and (B) targeted profiling. Both models showed a strong regression (95%), with the following quality parameters (A): R2 = 0.81 and Q2 = 0.75 and (B) R2 = 0.91 and Q2 = 0.87. The labels t[1] and t[2] along the axes represent the scores (the first two partial least-squares components) of the model.

subjects collected at −27.3 °C and asthmatic patients collected at -4.8 °C. In the first model classes were separated by uracil, urocanic acid, succinate, SFA, Phe, hippurate, trimethylamine, Val, and Tyr (column 5), while in the second one classes were differentiated by Phe, succinate, Val, propionate, SFA, methanol, uracil, Pro, formate, isobutyrate, and urocanic acid (column 6). Strong PLS-DA regression models with both approaches efficaciously described the comparison of all healthy subjects and asthmatic patients collected at both temperatures, and groups were separated by SFA, Val, propionate, Pro, formate, hyppurate, urocanic acid, isobutyrate, and Phe (column 7).

Asthmatic patients are clearly separated from healthy subjects in both models at −4.8 and −27.3 °C. At −4.8 °C, SFA, Val, adenosine, hippurate, Ala, formate, urocanic acid, Pro, acetate, ethanol, methanol, and Ile (Table 2, column 3) are separating biomarkers. At −27.3 °C, between-class separation is achieved through propionate, 4OH-phenylacetate, Val, acetate, SFA, Pro, Tyr, Arg, trans-aconitate, and Phe (column 4). By interchanging the samples, we verified that mild asthmatic patients can always be distinguished from healthy subjects, independently from the collecting temperatures. Strong PLS-DA regression models from spectral bucketing and targeting profiling data efficaciously described healthy subjects collected at −4.8 °C and asthmatic patients collected at −27.3 °C as well as healthy 6117

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

urocanic acid decrease, while propionate, Pro, isobutyrate, and Phe increase. This warrants that cross data comparisons are reliable, independently from the collecting temperature, and this is of importance as differentiating metabolites are directly linked to the mechanisms of the pathology. In particular, urocanic acid appears to be an appropriate asthma biomarker. It originates from histidine by histidine ammonialyase, but histidine is also converted to histamine by histidine decarboxylase. Histamine is a potent mediator of acute inflammation, and its presence is increased in the asthmatic airway. Therefore, in asthma the histidine−histamine conversion may well be favored over the histidine-urocanic acid pathway. It can be hypothesized that monitoring the histidine conversion pathways may become a way of controlling asthma severity. We are aware that the present study relies on a limited number of subjects and that our data should also be verified at other temperatures and on other condensers. However, they strongly suggest that NMR-based metabolomics is essentially independent from the EBC condensing temperature and that EBC profile is determined by the disease itself, as the asthmatic samples collected at −4.8 and −27.3 °C were undifferentiated (Supplementary Figure S1 in the Supporting Information). Mixing data at different temperatures could initially produce simple yes/no responses (i.e., asthma or not, stable or exacerbating), but we believe that enlarging the subject cohorts and comparing different respiratory diseases will improve models and the scoring accuracy. Finally, our data consolidates NMR-based “breathomics” as a new noninvasive holistic approach for the assessment of respiratory diseases, providing the physician with an extra diagnostic tool.

Taken together, the previously described data indicate that models at different temperatures, independently from the spectral analysis, correctly classify asthmatic and healthy subjects, and this warrants the same diagnostic value to samples collected with different condensers or at different temperatures. It is concluded that NMR profiles of EBC collected at different temperatures and with different condensers may safely be interchanged for diagnostics. Interestingly, although PCA pattern recognition techniques are more stable for targeted profiling data with respect to spectral bucketing analysis,24 both approaches yield reliable and robust regression models for EBC NMR spectra. EBC spectra are relatively simple, and both spectral bucketing and targeted profiling approaches work efficaciously. This ensures that most of the buckets contain a single resonance but does not avoid splitting of resonances over multiple buckets, which may include spectral noise in the buckets, and PCA has been reported to be sensitive to spectral noise.39 This, together with the overlaps observed in the 2.3 to 2.1 and 1.4 to 1.0 ppm regions, may explain the higher clustering and better R2 and Q2 values observed in targeted profiling models. However, both approaches identified the same metabolites responsible for class separation, suggesting a strong correspondence between bucketing and targeted profiling for EBC samples. Asthma represents a heterogeneous syndrome with many clinical classifications,40 often overlapping with other respiratory diseases.41,42 Tests for asthma using blood or urine have been proposed,43−46 but they are not currently used in the clinical practice. NMR-based metabolomics of urine identified 23 metabolites discriminating stable asthmatics from healthy subjects and exacerbated from stable asthmatics, both with 94% accuracy. In both models, metabolites involved in the TCA cycle significantly increased due to the greater effort to breathe during exacerbation or hypoxic stress originating from poor oxygenation.47 Nontargeted LC−MS of urine classified a range of atopic asthma phenotypes by essentially relying on reduction of urocanic acid, methylimidazolacetic acid, and an Ile-Pro fragment, which were also shown to correlate to lung function.48 Analysis of our models (Table 2, columns 3−7) indicates that most metabolites are also found in the urine model,47,48 namely, adenosine, hippurate, Ala, formate, urocanic acid, Ile, Pro, Tyr, trans-aconitate, Phe, succinate, trimethylamine, and isobutyrate. Except for urocanic acid, Ile, Pro, Tyr, and trimethylamine, which all decrease, all of the other metabolites increase. At −4.8 °C (column 3), we observed an increase in Ala, Pro, ethanol, methanol, and Ile, while SFA, Val, adenosine, hippurate, formate, urocanic acid, and acetate decrease. At −27.3 °C (column 4) propionate, 4OH-phenylacetate, acetate, Pro, Arg, trans-aconitate, and Phe increase, while Val, SFA, and Tyr decrease. Interestingly, in asthmatic patients the acetate concentration decreases at −4.8 °C and increases at −27.3 °C: this might reflect the fact that metabolites in healthy subjects are differently affected by the collecting temperature, as their separation relies upon difference in 17 metabolites (column 1). Succinate behaves similarly in the mixed models (Table 2, columns 5 and 6), where concentration differences due to temperature are expected to be more pronounced. Notwithstanding the different ranking (columns 5−7), metabolites are able to separate mild asthmatic from healthy subjects. In particular, the all-sample model (column 7) reproduces the relative concentrations observed in the other comparisons: in asthma SFA, Val, formate, hippurate, and



ASSOCIATED CONTENT

S Supporting Information *

Figure S1. PLS-DA of EBC of mild asthmatic patients at −27.3 and −4.8 °C using spectral bucketing and targeted profiling. This material is available free of charge via the Internet at http://pubs.acs.org.



AUTHOR INFORMATION

Corresponding Author

*Phone: +39 81 867 5228. Fax: +39 81 804 1770. E-mail: [email protected]. Author Contributions #

A. Motta, D. Paris, M. D’Amato, D. Melck, and C. Calabrese contributed equally to this work. Notes

The authors declare no competing financial interest. ▽ Matteo Sofia sadly passed away, untimely, in Naples on September 17, 2013. This study is dedicated to his curiosity, inspiration, and contribution to the field of pulmonary research.



ACKNOWLEDGMENTS We acknowledge the National Research Council for funding. We are grateful to patients for their important contribution to this study. We also thank Prof. Maria Triassi, Prof. Paolo Montuori, and Dr. Raffaele Corso (Department of Public Health, University of Naples Federico II, Naples) for use, advice, and expert assistance with the sampling pump for air monitoring and Dr. Giovanna Caccavo (Department of 6118

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

(19) Hwang, T.-L.; Shaka, A. J. Water suppression that works: excitation sculpting using arbitrary waveforms and pulse field gradients. J. Magn. Reson. 1995, 112, 275−279. (20) Griesinger, C.; Otting, G.; Wüthrich, K.; Ernst, R. R. Clean TOCSY for proton spin system identification in macromolecules. J. Am. Chem. Soc. 1988, 110, 7870−7872. (21) Kay, L. E.; Keifer, P.; Saarinen, T. Pure absorption gradient enhanced heteronuclear single quantum correlation spectroscopy with improved sensitivity. J. Am. Chem. Soc. 1992, 114, 10663−10665. (22) Lindon, J. C.; Holmes, E.; Nicholson, J. K. Pattern recognition methods and applications in biomedical magnetic resonance. Prog. Nucl. Magn. Reson. Spectrosc. 2001, 39, 1−40. (23) Spraul, M.; Neidig, P.; Klauck, U.; Kessler, P.; Holmes, E.; Nicholson, J. K.; Sweatman, B. C.; Salman, S. R.; Farrant, R. D.; Rahr, E.; Beddell, C. R.; Lindon, J. C. Automatic reduction of NMR spectroscopic data for statistical and pattern recognition classification of samples. J. Pharm. Biomed. Anal. 1994, 12, 1215−1225. (24) Weljie, A. M.; Newton, J.; Mercier, P.; Carlson, E.; Slupsky, C. M. Targeted profiling: quantitative analysis of 1H NMR metabolomics data. Anal. Chem. 2006, 78, 4430−4442. (25) Nicholson, J. K.; Foxall, P. J. D.; Spraul, M.; Farrant, R. D.; Lindon, J. C. 750 MHz 1H and 1H-13C NMR spectroscopy of human blood plasma. Anal. Chem. 1995, 67, 793−811. (26) Castellano, S.; Bothner-By, A. A. Analysis of NMR spectra by least squares. J. Chem. Phys. 1964, 41, 3863−3869. (27) Eriksson, L.; Johansson, E.; Kettaneh-Wold, N.; Trygg, J.; Wikström, C.; Wold. S. Multi- and Megavariate Data Analysis. Part I: Basic Principles and Applications, 2nd ed.; Umetrics AB: Umeå, Sweden, 2006. (28) Wold, S.; Antti, H.; Lindgren, F.; Ö hman, J. Orthogonal signal correction of near infrared spectra. Chemom. Int. Lab. Sys. 1998, 44, 175−185. (29) Rosias, P. P. R.; Robroeks, C. M.; Niemarkt, H. J.; Kester, A. D.; Vernooy, J. H.; Suykerbuyk, J.; Teunissen, J.; Heynens, J.; Hendriks, H. J.; Jöbsis, Q.; Dompeling, E. Breath condenser coatings affect the measurement of biomarkers in exhaled breath condensate. Eur. Respir. J. 2006, 28, 1036−1041. (30) Hoffmeyer, F.; Raulf-Heimsoth, M.; Harth, V.; Bünger, J.; Brü ning, T. Comparative analysis of selected exhaled breath biomarkers obtained with two different temperature-controlled devices. BMC Pulm. Med. 2009, 9, 48 DOI: 10.1186/1471-2466-9-48. (31) Czebe, K.; Barta, I.; Antus, B.; Valyon, M.; Horváth, I.; Kullmann, T. Influence of condensing equipment and temperature on exhaled breath condensate pH, total protein and leukotriene concentrations. Respir. Med. 2008, 102, 720−725. (32) Soyer, O. U.; Dizdar, E. A.; Keskin, O.; Lilly, C.; Kalayci, O. Comparison of two methods for exhaled breath condensate collection. Allergy 2006, 61, 1016−1018. (33) Leung, T. F.; Li, C. Y.; Yung, E.; Liu, E. K.; Lam, C. W.; Wong, G. W. Clinical and technical factors affecting pH and other biomarkers in exhaled breath condensate. Pediatr. Pulmonol. 2006, 41, 87−94. (34) Prieto, L.; Ferrer, A.; Palop, J.; Domenech, J.; Llusar, R.; Rojas, P. P. R. Differences in exhaled breath condensate pH measurements obtained with two commercial devices. Respir. Med. 2007, 101, 1715− 1720. (35) van Hoydonck, P. G.; Wuyts, W. A.; Vanaudenaerde, B. M.; Schouten, E. G.; Dupont, L. J.; Temme, E. H. Quantitative analysis of 8-isoprostane and hydrogen peroxide in exhaled breath condensate. Eur. Respir. J. 2004, 23, 189−192. (36) Wells, K.; Vaughan, J.; Pajewski, T. N.; Hom, S.; Ngamtrakulpanit, L.; Smith, A.; Nguyen, A.; Turner, R.; Hunt, J. Exhaled breath condensate pH assays are not influenced by oral ammonia. Thorax 2005, 60, 27−31. (37) Liu, J.; Conrad, D. H.; Chow, S.; Tran, V. H.; Yates, D. H.; Thomas, P. S. Collection devices influence the constituents of exhaled breath condensate. Eur. Respir. J. 2007, 30, 807−808. (38) Rosias, P. P. R.; Robroeks, C. M.; Kester, A.; den Hartog, G. J.; Wodzig, W. K.; Rijkers, G. T.; Zimmermann, L. J.; van Schayck, C. P.; Jöbsis, Q.; Dompeling, E. Biomarker reproducibility in exhaled breath

Respiratory Medicine, A.O. Dei Colli, University of Naples Federico II, Naples) for patients supervision.



REFERENCES

(1) Horváth, I.; Hunt, J.; Barnes, P. J. ATS/ERS Task Force on Exhaled Breath Condensate. Exhaled breath condensate: methodological recommendations and unresolved questions. Eur. Respir. J. 2005, 26, 523−48. (2) Dompeling, E.; Rosias, P. P. R.; Jöbsis, Q. Exhaled Breath Condensate Sample Collection: Standards and Open Issues. In European Respiratory Monograph: Exhaled Biomarkers; Horvath, I., de Jongste, J. C., Eds.; Latimer Trend & Co: Plymouth, U.K., 2010; Vol. 49, pp 152−161. (3) Rosias, P. P. R. Methodological aspects of exhaled breath condensate collection and analysis. J. Breath Res. 2012, 6, 027102. (4) Atkins, P. W.; De Paula, J. Physical Chemistry, 9th ed.; W. H. Freeman: New York, 2011; pp 135−152. (5) Carraro, S.; Rezzi, S.; Reniero, F.; Héberger, K.; Giordano, G.; Zanconato, S.; Guillou, C.; Baraldi, E. Metabolomics applied to exhaled breath condensate in childhood asthma. Am. J. Respir. Crit. Care Med. 2007, 175, 986−990. (6) de Laurentiis, G.; Paris, D.; Melck, D.; Maniscalco, M.; Marsico, S.; Corso, G.; Motta, A.; Sofia, M. Metabonomic analysis of exhaled breath condensate in adults by nuclear magnetic resonance spectroscopy. Eur. Respir. J. 2008, 32, 1175−1183. (7) Sofia, M.; Maniscalco, M.; de Laurentiis, G.; Paris, D.; Melck, D.; Motta, A. Exploring airway diseases by NMR-based metabonomics: a review of application to exhaled breath condensate. J. Biomed. Biotechnol. 2011, 2011, 403260. (8) Motta, A.; Paris, D.; Melck, D.; de Laurentiis, P.; Maniscalco, M.; Sofia, M.; Montuschi, P. NMR-based metabolomics of exhaled breath condensate: methodological aspects. Eur. Respir. J. 2012, 39, 498−500. (9) Montuschi, P.; Paris, D.; Melck, D.; Lucidi, V.; Ciabattoni, G.; Raia, V.; Calabrese, C.; Bush, A.; Barnes, P. J.; Motta, A. NMR spectroscopy metabolomic profiling of exhaled breath condensate in patients with stable and unstable cystic fibrosis. Thorax 2012, 67, 222−228. (10) Montuschi, P.; Paris, D.; Montella, S.; Melck, D.; Mirra, V.; Santini, G.; Mores, N.; Montemitro, E.; Majo, F.; Lucidi, V.; Bush, A.; Motta, A.; Santamaria, F. Nuclear magnetic resonance-based metabolomics discriminates primary ciliary dyskinesia from cystic fibrosis. Am. J. Respir. Crit. Care Med. 2014, 190, 229−233. (11) de Laurentiis, G.; Paris, D.; Melck, D.; Montuschi, P.; Maniscalco, M.; Bianco, A.; Sofia, M.; Motta, A. Separating smokingrelated diseases using NMR-based metabolomics of exhaled breath condensate. J. Proteome Res. 2013, 12, 1502−1511. (12) Rosias, P. P. R.; Dompeling, E.; Hendriks, H. J.; Heijnens, J. W.; Donckerwolcke, R. A.; Jöbsis, Q. Exhaled breath condensate in children: pearls and pitfalls. Pediatr. Allergy Immunol. 2004, 15, 4−19. (13) Goldoni, M.; Caglieri, A.; Andreoli, R.; Poli, D.; Manini, P.; Vettori, M. V.; Corradi, M.; Mutti, A. Influence of condensation temperature on selected exhaled breath parameters. BMC Pulm. Med. 2005, 5, 10 DOI: 10.1186/1471-2466-5-10. (14) GINA Report, Global Strategy for Asthma Management and Prevention. GINA Website: http://www.ginasthma.org/uploads/ users/files/GINA_Report_2014.pdf. Updated May 2014. (15) Brooks, S. M.; Berstein, L.; Ragnuprasad, P. K.; Maccia, C. A.; Mieczkowski, L. Assessment of airway hyperresponsiveness in chronic stable asthma. J. Allergy Clin. Immunol. 1990, 85, 17−26. (16) Gaber, F.; Acevedo, F.; Delin, I.; Sundblad, B. M.; Palmberg, L.; Larsson, K.; Kumlin, M.; Dahlen, S. E. Saliva is one likely source of leukotriene B4 in exhaled breath condensate. Eur. Respir. J. 2006, 28, 1229−1235. (17) Izquierdo-Garcia, J. L.; Peces-Barba, G.; Ruiz-Cabello, J. Influence of ambient air on NMR-based metabolomics of exhaled breath condensates. Eur. Respir. J. 2012, 40, 1294−1296. (18) Marieb, E. N.; Hoehn, K. Human Anatomy and Physiology, 9th ed.; Prentice Hall Pearson: Boston, 2013; pp 804−843. 6119

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120

Journal of Proteome Research

Article

condensate collected with different condensers. Eur. Respir. J. 2008, 31, 934−942. (39) Halouska, S.; Powers, R. Negative impact of noise on the principal component analysis of NMR data. J. Magn. Reson. 2006, 178, 88−95. (40) Centers for Disease Control and Prevention (CDC). Vital signs: asthma prevalence, disease characteristics, and self-management education: United States, 2001−2009. MMWR Morb. Mortal Wkly Rep. 2011, 60, 547−552. (41) Adamko, D. J.; Sykes, B. D.; Rowe, B. H. The metabolomics of asthma. Novel diagnostic potential. Chest 2012, 141, 1295−1302. (42) Aaron, S. D.; Vandemheen, K. L.; Boulet, L. P.; McIvor, R. A.; Fitzgerald, J. M.; Hernandez, P.; Lemiere, C.; Sharma, S.; Field, S. K.; Alvarez, G. G.; Dales, R. E.; Doucette, S.; Fergusson, D. Canadian Respiratory Clinical Research Consortium. Overdiagnosis of asthma in obese and nonobese adults. CMAJ. 2008, 179, 1121−31. (43) Fens, N.; Zwinderman, A. H.; van der Schee, M. P.; de Nijs, S. B.; Dijkers, E.; Roldaan, A. C.; Cheung, D.; Bel, E. H.; Sterk, P. J. Exhaled breath profiling enables discrimination of chronic obstructive pulmonary disease and asthma. Am. J. Respir. Crit. Care Med. 2009, 180, 1076−1082. (44) Tauber, E.; Halmerbauer, G.; Frischer, T.; Gartner, C.; Horak, F., Jr.; Veiter, A.; Koller, D. Y.; Studnicka, M. Urinary eosinophil protein X in children: the relationship to asthma and atopy and normal values. Allergy 2000, 55, 647−52. (45) Severien, C.; Artlich, A.; Jonas, S.; Becher, G. Urinary excretion of leukotriene E4 and eosinophil protein X in children with atopic asthma. Eur. Respir. J. 2000, 16, 588−592. (46) Bafadhel, M.; Clark, T. W.; Reid, C.; Medina, M. J.; Batham, S.; Barer, M. R.; Nicholson, K. G.; Brightling, C. E. Procalcitonin and Creactive protein in hospitalized adult patients with communityacquired pneumonia or exacerbation of asthma or COPD. Chest 2011, 139, 1410−1418. (47) Saude, E. J.; Skappak, C. D.; Regush, S.; Cook, K.; Ben-Zvi, A.; Becker, A.; Moqbel, R.; Sykes, B. D.; Rowe, B. H.; Adamko, D. J. Metabolomic profiling of asthma: diagnostic utility of urine nuclear magnetic resonance spectroscopy. J. Allergy Clin. Immunol. 2011, 127, 757−764. (48) Mattarucchi, E.; Baraldi, E.; Guillou, C. Metabolomics applied to urine samples in childhood asthma; differentiation between asthma phenotypes and identification of relevant metabolites. Biomed. Chromatogr. 2012, 26, 89−94.

6120

dx.doi.org/10.1021/pr5010407 | J. Proteome Res. 2014, 13, 6107−6120