Asthma is one of the most prevalent pediatric chronic diseases worldwide and causes significant burden on patients, family, society and healthcare systems [1–3]. Characteristic clinical features are chronic airway inflammation and variable expiratory airflow obstruction presenting as cough, wheeze, chest tightness and dyspnea usually in response to specific triggers, e.g. viral infections and exposure to allergens [4].

Despite similar clinical manifestation, the pathobiology and course of the disease is very heterogeneous [5]. Many factors contribute to asthma development, among them genetics, epigenetics [6], prenatal influences as maternal smoking, viral respiratory tract infections in early life [7] and unfavorable environmental exposures to cigarette smoke, air pollution, allergens, or mold [8, 9]. There is particular need for non-invasive and simple to assess biomarkers for early disease detection and characterization, especially in children. While exhaled nitric oxide is considered as an established non-invasively accessible biomarker [10, 11], it mainly reflects the level of eosinophilic airway inflammation. For a more comprehensive analysis, volatile organic compounds (VOCs) in exhaled breath have gained considerable interest as biomarkers for lung diseases, especially in asthma [12–14].

Breath collection is non-invasive and only requires tidal breathing which makes it particularly attractive for the use in children even at a young age. A number of studies on breath VOCs in adult and pediatric asthma have been published in the last decade and are considered in recent reviews [12–14]. For example, Dallinga et al showed that exhaled VOCs can distinguish asthmatic from non-asthmatic children [15], and there is evidence that breath VOCs identify distinct inflammation phenotypes [16] or predict exacerbations in asthmatic children [17, 18]. Despite a large body of literature with a number of VOCs potentially discriminating between healthy controls and asthma patients or reflecting disease activity or treatment, there is currently no breath VOC biomarker or biomarker pattern that supports physicians in the diagnosis, treatment, and phenotyping of disease or in preventing exacerbations.

Here, we collected and analyzed breath VOCs from a subgroup of children of the All Age Asthma cohort (ALLIANCE), a multicenter prospective observational cohort recruiting children with pre-school wheeze and children and adults with asthma [19].

We hypothesized that VOCs or VOC patterns are distinct between children with asthma or pre-school wheeze compared to healthy children and contribute to identify extrinsic risk factors, and underlying disease mechanisms. Additionally, we investigated if VOCs were linked to inflammatory phenotypes, clinical features such as lung function, asthma treatment and exacerbation rate.

2.1. Study design

The ALLIANCE cohort of the German Center for Lung Research (DZL) is a prospective multicenter asthma cohort [19]. For this study, we collected 182 breath samples from 142 children (51 children with asthma, 55 children with wheeze and 36 healthy controls) at two pediatric specialist centers (Hannover, Munich) from October 2016 until spring 2020 (figure 1 and table S1). In a subgroup of 40 patients, a second breath collection was performed after one or two years (asthma n = 20, wheeze n = 20). The study was conducted in accordance with the principles embodied in the Declaration of Helsinki and in accordance with local statutory requirements. The study was registered at ClinicalTrials.gov (NCT02496468) and approved by all local ethics committees. All parents of study participants <18 years as well as study participants ⩾8 years gave their written informed consent.

Figure 1. Study population. Breath VOCs were analyzed from 142 subjects. 40 subjects were sampled on two visits, which were at least one year apart. The Hannover site had a focus on wheezers and started the recruitment for VOC sampling later during the study. The one patient with asthma from Hannover was planned to be included as wheezer but the visit could not be scheduled before the 6th birthday, therefore based on the definition criteria, this subject was considered as child with asthma.

Download figure:

Standard image High-resolution image

Children aged 6 months to 5 years were included if they had at least two episodes of wheeze during the past 12 months ('pre-school wheezer') as indicated by the parents in the respective questionnaire. Children ⩾6 years were included based on doctor-diagnosed asthma according to the Global Initiative for Asthma (GINA) guidelines and German guidelines [4, 20]. We also recruited healthy control subjects who had never been diagnosed with asthma or pre-school wheeze. Further inclusion and exclusion criteria are specified in the supplement or have been published [19]. Laboratory tests included differential blood count, and specific immunoglobulin E against 36 allergens measured by Euroline™ (Euroimmun, Germany). All children performed fractional exhaled nitric oxide (FeNO) measurements and spirometry from age 4 years onwards and VOC measurements from 3 years onwards. For the breath and near-subject room air collection we used an in-house developed breath collection device [21]. Further details regarding all procedures and study design are presented in the online supplement.

2.2. Analysis by gas chromatography–mass spectrometry (GC–MS)

GC-analysis of the samples was performed at Fraunhofer ITEM, Hannover, within eight days of initial breath collection. Turbo Mass Software 5.4 (PerkinElmer, USA) was used for automated identification of individual VOCs based on retention time of specific masses, comparing values with reference compounds and the NIST database (National Institute of Standards and Technology Mass Spectral Search Program Version 2.2 (NIST, USA)) [21, 22]. A total of 158 VOCs were quantified using peak height of specific target ions which most commonly matched with m/z signals of highest intensity in the respective VOC mass spectrum (total ion content (TIC)) [22]. Further details regarding measurement and analysis are available in the online supplement.

2.3. Statistics

We computed the mean value of the 158 target VOCs for all technical replicate samples. Data was then log-transformed. We used the Mann–Whitney-U-Test (MWU) for univariate comparisons between patients (asthma, wheeze) and healthy controls and the Pearson correlation coefficient where appropriate. The false discovery rate was assessed by correcting raw p-values with the p.adjust function in R (using the Benjamini and Hochberg (BH) option). Corrected p-values are indicated as pBH. Site effects were accounted for by batch correction using the ComBat function of the R package sva to our data. For some analyses that were limited to data from one site we used both the original (uncorrected) and the batch corrected 'site-effect–free data' to test whether similar results were obtained.

Due to the exploratory nature of our analysis, we focused on the complete VOC dataset (n = 158 VOCs). However, to avoid missing important observations by the multiplicity correction we also ran the analysis with subsets of our 158 VOCs. A detailed description of these subsets is provided in the online supplement. In addition we dimensionally reduced the dataset using an unsupervised clustering approach called Cytomod [23]. Briefly, VOCs were clustered and assigned to co-varying modules based on the pairwise Pearson-correlations between log-transformed VOC values. For more details, please refer to the online supplement. Module scores reflecting the expression of each module in each participant (see online supplement) were compared between participant groups with different characteristics using MWU tests. P-values were adjusted for the number of modules tested. A multivariable logistic regression model was used to analyze the association between module 6 and disease status. We focused our analysis on module 6, because most VOCs considered as air pollutants were found in module 6.

3.1. Demographics

As shown in figure 1, we included 55 children with pre-school wheeze, 51 children with asthma and 36 healthy controls (table 1). Total IgE and FeNO were higher (p < 0.01) in asthmatics than in healthy controls while FEV1 and FEV1/FVC values were lower (p < 0.01). Blood eosinophils were higher in asthmatic children compared to controls (p = 0.01), and atopy was more prevalent in subjects with wheeze and asthma (table 1).

Table 1. Demographics.

  nAsthmanWheezenHealthyNumber of subjectsM/H5150/15530/253633/3Genderfemale/male5116/355520/353612/24Ageyears5112.9 ± 3.5***555.7 ± 1.3***369.7 ± 3.3Age at inclusionyears5110.8 ± 3.1552.7 ± 1.5***369.7 ± 3.3Atopy§% pos.468543583537FeNOppb4424.7 (11.1;36.9)**239.6 (5.1;19.9)249.1 (5.6;20.2)FEV1z-score49−0.53 (−1.09;0.23)***48−0.39 (−1.09;0.55)***360.24 (−0.24;1.26)FEV1/FVCz-score49− 0.57 (−1.34;0.15)***47− 0.28 (−1.18;0.87)360.16 (−0.67;1.12)Blood leucocytesCells μl−1467350 (6100;8300)427500 (6400;9200)**346675 (5500;7730)Blood eosinophilsCells μl−146358 (198;570)*42381 (260;582)**34201 (139;395)Blood neutrophilsCells μl−1463327 (2654;4300)423430 (2603;4304)343046 (2452;3696)Total IgEkU/L45232 (162;562)**41111 (35;302)34110 (41;240)Asthma medication:       SABAn (%)5119 (37%)5515 (27%)  Montelukastn (%)512 (4%)554 (7%)  ICSn (%)5110 (20%)5513 (24%)  LABA/ICSn (%)5126 (51%)558 (15%)  Wheezing episode (yes)$n (%)5113 (26%)5525 (45%)  ⩾1 exacerbation (yes)$n (%)512 (4%)555 (9%)  

M = Munich, H = Hannover.Mean ± SD or median (25%; 75% quartile) are presented. §: specific IgE ⩾0,7 against at least 1/36 food- and aeroallergens.Medication: number of children with respective treatment in the month before the VOC visit.Wheeze episode: symptoms of wheeze which required treatment with salbutamol for >2/7 d.Exacerbation: wheeze episode which required treatment with systemic steroids or admission to hospital.$: refers to 12 month prior to the VOC collection visit.***p < 0.001, **p < 0.01,*p < 0.05.

3.2. Basic data description, quality, and plausibility of VOC data

As expected, acetone and isoprene were major VOCs in breath of children, while disinfectants like 1-propanol, 2-propanol, ethanol, 2-phenoxyethanol, and 1-phenoxypropan-2-ol were the major VOCs detected in the room air samples. In line with the 'owncloud concept' [22] we found the expected correlations between near-subject room air and breath for VOCs, that are frequently used in lifestyle products (personal care, cosmetics, home care) or for environmental pollutants. Among these were geranyl acetone, siloxanes, as well as benzene (figure S1). Furthermore, patients exposed to second-hand smoke and in one patient to active smoking showed increased levels of cigarette smoke related VOCs (table S2) [21, 24]. Isoprene levels in exhaled breath significantly correlated with age (r = 0.36, p < 0.0001), as previously shown [25].

3.3. Potential confounders: site, age, gender

The median levels of most VOCs were different between study sites. We consider technical reasons, like instrument drift, as unlikely for this observation (details are presented in the online supplement) and adjusted the data as outlined in material and methods. To evaluate the impact of age we checked the correlation between VOC values and age using the site effect-free data for all subjects. Besides the already mentioned positive isoprene correlation, we only found a negative correlation for 1,2-propandiol and age (r = −0.27), with borderline significance (pBH = 0.068). No significant gender-specific differences for breath VOCs were observed (pBH > 0.05).

3.4. Reproducibility between visits

For 40 children (asthma n = 20, wheeze n = 20) we were able to collect a second sample in a follow-up visit one year after taking the initial sample. In three cases, the second sample was taken two years later. Initial and follow-up VOC values correlated significantly for 13 VOCs (pBH < 0.05, figure S2 and table S3). Most of these VOCs appear to be environment- and lifestyle-related. The correlation between the visits was not significant for naphthalene (figure S3, p = 0.06, pBH = 0.27).

3.5. Cluster analysis

In the unsupervised cluster analysis, we identified a total of 9 modules (figure 2), consisting of 3–39 VOCs. The VOCs contained in each module are listed in the supplement (table S4). Interestingly, all BTEX (benzene, toluene, ethylbenzene, xylenes) VOCs and other VOCs considered as air pollutants, like naphthalene, were found in module 6. A large number of VOCs in module 4 were substances used as fragrances and flavors typically found in many lifestyle and cleaning products.

Figure 2. Heatmap of VOC modules: VOCs are first clustered together based on pairwise Pearson correlations. A reliability score (i.e. the fraction of times that a given VOC is assigned to the same cluster) is calculated over 1000 permutations of participants and is used to assign VOCs to modules. The color bars on the left side depict the module membership for each VOC (a full list is given in supplement table S4), and the coloring in the heatmap represents the reliability score of each VOC.

Download figure:

Standard image High-resolution image 3.6. VOCs differ between children with asthma or pre-school wheeze and healthy controls

We found 24 significantly different VOCs (pBH < 0.05) between healthy controls and asthma patients and 23 significantly different (pBH < 0.05) VOCs between healthy controls and children with wheeze (table 2). There was a large overlap, with 13 VOCs having significantly different levels both in asthma patients and wheezers compared to healthy controls (table 2). As clinical characteristics differed between healthy controls and patients (table 1), we performed additional subgroup analyses which are described in detail in the online supplement. We did not find any evidence that these differences influenced the results.

Table 2. Differences between healthy controls, asthmatic children, and children with wheeze.

VOC Asthma vs. HealthyWheeze vs. Healthy      VOC Asthma vs. HealthyWheeze vs. Healthy      Lower in Diseased PatientsModule 4pMWUpBHpMWUpBHFragrance/FlavorFoodPersonal CareCleaningDisinfectantAir pollutantIncreased in Diseased PatientsModule 6pMWUpBHpMWUpBHFragrance/FlavorFoodPersonal CareCleaningDisinfectantAir pollutantunidentified C10H18 (a)x<0,001 <0,001 <0,001 0,007       Isoprene (b) <0,001 <0,001 0,3180,507      Geranyl acetatex<0,001 0,003 0,003 0,035 xxxx              Pentadecane <0,001 0,003 0,006 0,042 xxx   1-Phenoxypropan-2-olx0,005 0,034 0,0390,144  xxx Menthyl acetatex<0,001 0,005 0,001 0,028 xxxxx Cycloheptasiloxanex0,0210,0930,004 0,037   xx  Decanoic acid methyl esterx0,001 0,009 0,002 0,033 x  x  Benzoic acidx0,0220,093<0,001 0,008 xxxx  Decanoic acidx<0,001 0,009 0,008 0,054 xxxx  1-Nonenex0,0680,1790,004 0,037       unidentified VOC (a) <0,001 0,009 0,0640,213      1,2-Propanediolx0,230,400,001 0,028 xxxxxxNeryl acetatex0,002 0,019 0,0380,144xxxx  2-Ethylhexanol 0,008 0,047 0,005 0,038 xx   x2-Propanol 0,002 0,018 0,1280,302xxxxx Limonene 0,0510,1460,001 0,028 xxxx  1-Decanol 0,002 0,020 <0,001 0,007 xxxxx Naphthalenex<0,001 0,005 <0,001 0,007      xIonone 0,002 0,017 <0,001 0,007 xxxx              Citronellolx0,001 0,015 0,002 0,033 xxxx  Ethylbenzenex0,0700,1790,1590,328     xb-Linaloolx0,001 0,017 0,1290,302xxxxx Benzenex0,0460,1380,3440,537     xDecanal 0,003 0,023 0,0250,120xxxxx p-/m-Xylenex0,1440,2970,1510,328     xLinalyl acetatex0,005 0,031 0,1580,328xxxxx o-Xylenex0,2380,4110,1140,280     xDioctyl ether 0,003 0,025 0,0780,222x x