Breathomics in controlled asthma: identification of nonanoic and propanoic acids as volatile organic compounds

Graphical abstract

Overview of the study. VOC: volatile organic compound.

Abstract

Background

There is growing interest in identifying novel, non-invasive biomarkers reflecting endogenous inflammatory processes in asthma. This study aimed to evaluate the presence of volatile organic compounds (VOCs) in exhaled breath from patients with clinically controlled asthma and assess how tobacco exposure influences their expression.

Methods

Exhaled breath samples from 120 clinically controlled asthma patients and 89 healthy controls were collected using BioVOC breath samplers. Samples were analysed by gas chromatography-mass spectrometry, assessing five previously characterised VOCs: hexanal, heptanal, nonanal, propanoic acid and nonanoic acid.

Results

Compared to healthy controls, asthma patients exhibited a lower frequency of propanoic acid exhalation (25.0% versus 53.9%; p<0.001) and higher frequencies of nonanoic acid (30.8% versus 15.7%; p=0.019). These differences persisted after adjusting for smoking status. Stratified analysis revealed reduced propanoic acid exhalation in both smoking and non-smoking asthma subgroups compared to their respective controls (21.0% versus 55.6% and 29.3% versus 51.4%, respectively; p<0.001). Additionally, asthma current and former smokers had significantly increased detection of nonanoic acid compared to controls (33.9% versus 11.1%; p=0.0359). Multivariate analysis identified propanoic acid as a protective factor against asthma (OR 0.2 (95% CI 0.1–0.4); p<0.001), whereas nonanoic acid significantly increased asthma risk (OR 4.5 (95% CI 1.8–12.6); p=0.003).

Conclusions

Exhaled propanoic and nonanoic acids may serve as complementary non-invasive biomarkers for monitoring controlled asthma, independently of tobacco exposure. VOC analysis has promising potential to improve asthma management, therapeutic monitoring and patient stratification.

Shareable abstract

Controlled asthma shows distinct VOC signatures: lower propanoic acid and higher nonanoic acid in exhaled breath, independent of tobacco exposure https://bit.ly/4k5sCog

Introduction

Asthma is the second most prevalent chronic respiratory disease worldwide [1], primarily characterised by a chronic inflammation of the airways. Diagnosis typically relies on spirometry, including a positive bronchodilator challenge, variable airway obstruction or bronchial hyperresponsiveness. Additionally, the assessment of type 2 inflammatory biomarkers can enhance diagnostic precision and help define asthma phenotypes [2]. Airway eosinophilia, a hallmark biomarker, is usually assessed through invasive or minimally invasive procedures such as bronchoscopy or induced sputum analysis. As a non-invasive alternative, the fractional exhaled nitric oxide (F_ENO) test allows indirect evaluation of airway inflammation [3]. However, F_ENO accuracy is affected by factors such as smoking, corticosteroid use, dietary nitrates and other eosinophilic conditions [4]. Therefore, identifying robust, non-invasive biomarkers with minimal susceptibility to environmental and physiological confounders remains essential to improve diagnostic accuracy, disease monitoring and tailored asthma management. While no biomarker is entirely free from external influences, certain volatile organic compounds (VOCs) have shown greater stability under controlled sampling conditions and less confounding by ambient contamination or occupational exposures [5].

VOCs in exhaled breath reflect metabolic activity at the cellular and tissue levels, and are closely linked to airway inflammation processes [6, 7]. Studies have demonstrated that VOC profiles may serve as promising biomarkers in several respiratory diseases, including asthma. VOCs can be analysed by electronic nose (e-Nose) technology and gas chromatography-mass spectrometry (GC-MS). The e-Nose provides general VOC patterns known as “respiratory footprints” [8], but this method is limited by low specificity, environmental interference and poor reproducibility [9]. In contrast, GC-MS, considered the gold standard, accurately identifies individual VOCs in complex biological matrices [10], offering detailed insights into specific cellular and enzymatic metabolism pathways [11].

VOCs have recently been investigated in several respiratory conditions, e.g. cystic fibrosis [12], COPD [5, 13], asthma [14] and lung cancer [15]. In asthma, VOC analysis has demonstrated potential for diagnosis, phenotypic and endotype classification [16], prediction of exacerbations, assessment of adherence, therapeutic guidance, and monitoring of treatment response [17, 18]. van der Schee et al. [18] identified a VOC “respiratory footprint” in eosinophilic asthma using e-Nose technology, with diagnostic performance comparable to F_ENO sputum eosinophilia. Similarly, Schleich et al. [19], using GC-MS, identified hexane and 2-hexanone as key compounds, achieving diagnostic accuracy comparable to traditional inflammatory markers, especially when combined with F_ENO and peripheral eosinophilia (area under the curve 0.9). VOCs have also been correlated with medication adherence, such as urinary concentrations of salbutamol, an oral corticosteroid [20, 21]. Tobacco exposure has been shown to influence VOC profiles, with compounds such as nonanal associated with smoking habits in healthy individuals [22].

This study aims to analyse five VOCs (hexanal, heptanal, nonanal, propanoic acid and nonanoic acid) that have been previously characterised by our research group [5, 22], in patients with controlled asthma. These VOCs have demonstrated involvement in lipid metabolism, oxidative stress and inflammation-related processes in the airways [5], supporting their biological plausibility as respiratory biomarkers. Additionally, we aim to assess the influence of tobacco exposure on their expression, and explore their potential utility in asthma monitoring and therapeutic decision making.

Material and methods

This was a multicentre, cross-sectional study with historical controls conducted across five Asthma Units located in Madrid (Spain). Participants were recruited consecutively by non-probabilistic sampling over a 1-year period. Controls were selected from a previously characterised cohort described by Jareño-Esteban et al. [22].

Asthma severity was classified according the treatment intensity required to maintain control [23], which was assessed using the Asthma Control Test (ACT).

Inclusion criteria were: age ≥18 years; physician-confirmed diagnosis of stable asthma, with no exacerbations in the past month and an ACT score ≥20 points; and signed informed consent for participation in the study.

Exclusion criteria were: presence of any another chronic respiratory disease (e.g. COPD according to Global Obstructive Lung Disease criteria [24]); acute asthma exacerbation or respiratory tract infection within the previous 4 weeks; use of medications that could interfere with breath VOC analysis, including biological therapies or allergen-specific immunotherapy within the past 6 months, or systemic corticosteroids within the past month; current or past diagnosis of malignancy, or presence of active neoplastic disease; clinically significant cardiovascular disease (e.g. ischaemic heart disease); occupational exposure to agents affecting respiratory health; or lack of informed consent.

Participants were stratified into current, former and never-smokers according to criteria from the US Centers for Disease Control and Prevention, and smoking history was quantified using the pack-year index (PYI) [25].

Ethical considerations

The study adhered to the provisions of Organic Law 03/2018 on Data Protection and Law 41/2002 on Patients’ Autonomy and Rights and Information Management and Clinical Documentation Obligations. Approval was obtained from the Ethics and Clinical Research Committees of all participating hospitals.

VOC selection and analysis

Five VOCs (hexanal, heptanal, nonanal, propanoic acid and nonanoic acid) were selected for analysis using GC-MS. The selected VOCs were chosen based on prior validation by our research group in studies involving tobacco exposure and respiratory disease [5, 22, 26]. Those VOCs are well-characterised products of lipid peroxidation, particularly derived from ω-3 and ω-9 fatty acid metabolism. These molecules were also selected for their detection in exhaled breath and their minimal confounding by tobacco smoke or environmental pollution. Their biological plausibility and prior reproducibility in controlled populations support their relevance as candidate biomarkers for asthma.

Laboratory procedures

Exhaled air samples were collected from asthma and healthy controls following standardised procedures: after 1 h of rest, fasting and abstinence from smoking. Samples were obtained using BioVOC breath sampler chambers (Markes International, Bridgend, UK), obtaining a fixed volume of exhaled breath through three consecutive forced expiration manoeuvres, enabling VOC pre-concentration. Collected samples were then transferred into a thermal desorption tube internally coated with three adsorbent materials: Tenax TA, graphitised carbon black and carbonised molecular sieves (Markes International). Tubes were sealed with DiffLok caps (Markes International) until further analysis.

All participants were recruited from hospitals within the same autonomous region (Madrid, Spain), and thus shared similar environmental and urban air quality conditions. Additionally, none of the subjects reported occupational exposure to inhaled toxins or industrial pollutants. To control for ambient VOC interference, environmental samples were collected simultaneously from the same room using identical BioVOC samplers. This allowed us to determine endogenous origin using an exhaled/ambient ratio >1, a well-established threshold in breathomics research [17]. The limit of detection and quantification for each compound were established according to International Council on Harmonisation Q2 validation guidelines [27].

Sample analysis was centralised at a single reference laboratory within 24 h of collection to avoid inter-site variability. VOCs were separated and identified by thermal desorption followed by GC-MS, applying a standard calibration method previously validated by Jareño-Esteban et al. [22].

Statistical analysis

Demographic and clinical characteristics were described using mean and standard deviation for continuous data. Normality was assessed using the Shapiro–Wilk and Kolmogorov–Smirnov tests. Levene's test was used to evaluate homogeneity of variances. Parametric tests (ANOVA or t-test) were applied to normally distributed, homoscedastic variables; non-parametric tests (Kruskal–Wallis or Mann–Whitney U-test) were used otherwise. Categorical variables were compared using Chi-squared or Fisher's exact tests, as appropriate.

To ensure a robust and stable multivariate model, we applied a three-step variable selection strategy combining classical and machine learning methods. First, univariate analyses were performed for all candidate variables. Next, we applied both stepwise forward and backward selection methods, as well as the Boruta feature selection algorithm. Boruta is a wrapper method based on random forest classification that introduces shadow features (randomised versions of the original variables) to determine the true importance of each predictor. Variables are categorised as confirmed, tentative or rejected depending on whether their importance exceeds that of the most relevant shadow feature. For the final logistic regression model, we retained only those variables that were identified as relevant by at least two of the three methods, thereby enhancing the model's reliability and minimising the risk of overfitting. Prior to conducting the multivariate logistic regression, multicollinearity among predictors was assessed using the Variance Inflation Factor (VIF). All VIF values were <2, indicating the absence of significant multicollinearity.

Missing data handling was not required in this study. Participants with incomplete clinical or VOC data were excluded during the initial data validation phase and were not included in the final analysis. As a result, all variables used in the statistical models had complete cases only. This approach ensured the integrity of the dataset and obviated the need for imputation techniques.

To address the risk of type I error due to multiple testing, Bonferroni correction was applied when appropriate, particularly in pairwise comparisons between groups.

Statistical significance was defined as a p-value <0.05 for all analyses. Data processing and analyses were performed using R (www.r-project.org).

Results

Comparison of demographic and clinical characteristics between controlled asthma patients and the healthy control group

A total of 120 patients with controlled asthma (mild n=38, moderate n=44 and severe n=38) were compared with those of the control group (table 1). Asthma patients were significantly older (64.2±15.9 versus 49.3±9.7 years) and predominantly women (71.7% versus 52.8%).

TABLE 1.

Comparison of demographic and clinical characteristics between controlled asthma patients and healthy controls

	Total (n=209)	Controlled asthma (n=120)	Healthy controls (n=89)	p-value
Women	133 (63.6)	86 (71.7)	47 (52.8)	0.008
Age (years)	57.9±15.4	64.2±15.9	49.3±9.7	<0.001
≥57 years	111 (53.1)	89 (74.2)	22 (24.7)	<0.001
Smoking habit				0.081
Current smoker	54 (25.8)	24 (20.0)	30 (33.7)	0.025
Former smoker	62 (29.7)	38 (31.7)	24 (27.0)	0.462
Never-smoker	93 (44.5)	58 (48.3)	35 (39.3)	0.195
Tobacco exposure
Current+former	116 (55.5)	62 (51.7)	54 (60.7)	0.248
Pack-year index	11.7±17.7	7.5±13.4	17.4±21.0	<0.001
Volatile organic compound
Hexanal	57 (27.3)	33 (27.5)	24 (27.0)	1.000
Heptanal	68 (32.5)	36 (30.0)	32 (36.0)	0.448
Nonanal	90 (43.1)	52 (43.3)	38 (42.7)	1.000
Propanoic acid	78 (37.3)	30 (25.0)	48 (53.9)	<0.001
Nonanoic acid	51 (24.4)	37 (30.8)	14 (15.7)	0.019

Data are presented as n (%) or mean±sd, unless otherwise stated.

Although the overall smoking distribution was similar between groups, current smoking was significantly less frequent among asthma patients (20.0% versus 33.7%) and asthma patients had a lower mean PYI (7.5±13.4 versus 17.4±21.0).

Pulmonary function tests in asthma patients showed (pre-bronchodilator) mean±sd values were: forced vital capacity (FVC) 3.2±1.2 L, corresponding to 10±22.0% of predicted; forced expiratory volume in 1 s (FEV₁) 2.3±0.98 L, 91.0±20.0% of predicted; FEV₁/FVC ratio 72.0±10.0%. Post-bronchodilator mean±sd values were: FVC 3.3±1.2 L, 101±19% of predicted; FEV₁ 2.4±1.1 L, 96.0±19.0% of predicted; FEV₁/FVC ratio 73.0±11.0%. Mean±sd F_ENO was 30.0±34.0 ppb.

Regarding asthma treatment, 74.2% received inhaled long-acting β-agonists and inhaled corticosteroids (14.2% low dose, 26.7% medium dose and 33.3% high dose), 25.8% only low inhaled corticosteroid dose, 77.1% leukotriene receptor antagonists, 58.3% anticholinergics, 4.2% macrolides and 14.6% chronic oral corticosteroids.

VOC analysis revealed lower detection of exhaled propanoic acid (25.0% versus 53.9%) and higher nonanoic acid (30.8% versus 15.7%) in asthma patients compared to healthy controls. No significant differences were found for the other VOCs.

Subgroup analysis (current+former versus never-smokers) showed consistent findings (table 2). The predominance of woman and older age persisted regardless of smoking status. This pattern persisted in both current+former and never-smokers with asthma (propanoic acid: p<0.001; nonanoic acid: p=0.0359).

TABLE 2.

Demographic and clinical characteristics between controlled asthma patients and healthy controls stratified by tobacco exposure

	Controlled asthma		Healthy controls		p-value
	Current+former smokers (n=62)	Never-smokers (n=58)	Current+former smokers (n=54)	Never-smokers (n=35)
Women	40 (64.5)	46 (79.3)	23 (42.6)	24 (68.6)	<0.001
Age (years)	62.4±13.0	66.2±18.3	50.2±9.6	48.1±9.7	<0.001
≥57 years	44 (71.0)	45 (77.6)	15 (27.8)	7 (20.0)	<0.001
Hexanal	18 (29.0)	15 (25.9)	16 (29.6)	8 (22.9)	0.885
Heptanal	18 (29.0)	18 (31.0)	20 (37.0)	12 (34.3)	0.812
Nonanal	29 (46.8)	23 (39.7)	27 (50.0)	11 (31.4)	0.306
Propanoic acid	13 (21.0)	17 (29.3)	30 (55.6)	18 (51.4)	<0.001
Nonanoic acid	21 (33.9)	16 (27.6)	6 (11.1)	8 (22.9)	0.0359

Data are presented as n (%) or mean±sd, unless otherwise stated.

Factors associated with controlled asthma

A multivariate regression analysis was performed. Variables were selected by univariate analysis (table 3), stepwise forward/backward analysis (table 4) and the Boruta method (figure 1), resulting in a final set of nine variables: gender, age, tobacco exposure, PYI, hexanal, heptanal, nonanal, propanoic acid and nonanoic acid. All variables except hexanal, heptanal and nonanal showed significant associations with controlled asthma.

TABLE 3.

Crude and adjusted odds ratios for factors associated with controlled asthma

	Crude		Multivariate
	OR (95% CI)	p-value	OR (95% CI)	p-value
Demographic variables
Women	0.4 (0.2–0.8)	<0.001	0.8 (0.4–1.8)	0.554
≥57 years	8.7 (4.7–16.8)	<0.001	20.4 (8.8–52.7)	<0.001
Smoking-related variables
Tobacco exposure (current+former)	0.7 (0.4–1.2)	0.196	3.7 (1.3–10.8)	0.010
Pack-year index	0.9 (0.9–0.9)	<0.001	0.9 (0.9–0.9)	<0.001
Volatile organic compounds
Propanoic acid (yes)	0.3 (0.2–0.5)	<0.001	0.2 (0.1–0.4)	<0.001
Nonanoic acid (yes)	2.4 (1.2–4.9)	0.013	4.5 (1.8–12.6)	0.003
Hexanal (yes)	1.0 (0.6–1.9)	0.932	Not included#
Heptanal (yes)	0.8 (0.4–1.4)	0.364	Not included#
Nonanal (yes)	1.0 (0.6–1.8)	0.927	Not included#

^#: variables labelled as “Not included” were not retained in the final multivariate logistic regression model based on the variable selection criteria.

TABLE 4.

Selection of predictors associated with patient outcomes in each analysis step, based on p-values and Akaike Information Criterion

Item	Estimate	se	z-value	Pr (>|z|)
(Intercept)	4.104	1.007	4.077	<0.001
Age range	−0.010	0.016	−6.159	<0.001
Tobacco exposure	1.001	0.489	2.045	0.041
Pack-year index	0.069	0.017	4.084	<0.001
Propanoic acid	−1.492	0.389	−3.839	<0.001
Nonanoic acid	0.960	0.442	2.170	0.030

FIGURE 1.

Boruta plot showing variable importance in predicting the outcome of interest. Green boxes indicate confirmed important variables; red boxes represent unimportant variables; yellow boxes denote tentative variables for which the algorithm could not reach a conclusive decision; and blue boxes correspond to shadow features (randomised variables used as a baseline for assessing the relevance of the real features).

The multivariate regression analysis (figure 2 and table 3) confirmed results from the univariate analysis. Significant risk factors for controlled asthma were identified as age ≥57 years (OR 20.4 (95% CI 8.8–52.7); p<0.001), current or former smoker (OR 3.7 (95% CI 1.3–10.8); p=0.010) and exhaled nonanoic acid (OR 4.5 (95% CI 1.8–12.6); p=0.003). Conversely, protective factors were lower PYI (OR 0.9 (95% CI 0.9–0.9); p<0.001) and exhaled propanoic acid (OR 0.2 (95% CI 0.1–0.4); p<0.001).

FIGURE 2.

Forest plot illustrating predictors associated with controlled asthma. Multivariate model for asthma group.

Discussion

Asthma is a heterogeneous inflammatory disease. Identifying reliable, non-invasive biomarkers could greatly enhance diagnosis, monitoring, adherence evaluation and therapeutic decision making. Traditional diagnostic methods face limitations due to the influence of both external and internal factors on the assessment of airway inflammation [4]. In this context, the analysis of VOCs in exhaled breath, commonly referred as “breathomics”, has emerged as a promising approach. Despite technical challenges, including variability in sample procedures, analytical methodologies and environmental influences [28, 29], breathomics has demonstrated significant potential for clinical application.

Recent studies have underscored the value of exhaled VOCs as actionable biomarkers for asthma and other respiratory conditions. For instance, Shahbazi Khamas et al. [30] showed that distinct VOC profiles could accurately classify levels of asthma control in paediatric patients. These findings support the future development of point-of-care diagnostic tools based on exhaled metabolites.

A systematic review of breathomics in adult asthma confirmed the field's diagnostic promise but highlighted limitations, including methodological heterogeneity, inconsistent identification of discriminatory VOCs across studies and the lack of validation in independent cohorts [31]. Our study directly addresses some of these gaps by evaluating five previously identified and validated compounds using GC-MS in a new, well-characterised population of controlled asthma patients. Notably, we demonstrate consistent associations of propanoic and nonanoic acids with asthma controls (propanoic acid was significantly less frequent and nonanoic acid more frequent in exhaled breath from asthma patients compared to healthy controls). These differences persisted regardless of smoking status, suggesting a specific VOC signature in controlled asthma. Additionally, although hexanal, heptanal and nonanal were among the initial candidate VOCs based on prior literature, they were not retained in the final multivariate model. This was due to inconsistent associations across selection methods and lack of statistical significance in adjusted analyses. While these compounds have been previously linked to oxidative stress and tobacco exposure, their absence in the final model suggests limited relevance in clinically controlled asthma or a lower discriminatory capacity under our study conditions. This highlights the importance of rigorous biomarker validation in context-specific clinical settings.

Propanoic acid emerged as a protective biomarker in asthma control (OR 0.2 (95% CI 0.1–0.4); p<0.001), consistent with its role in fatty acid metabolism [32], and this association remained independent of tobacco exposure. Conversely, nonanoic acid was positively associated with asthma, even in clinically controlled patients (OR 4.5 (95% CI 1.8–12.6); p=0.003), reinforcing its potential role in oxidative stress and persistent inflammation.

Both compounds are carboxylic acids linked to distinct biological processes in the respiratory system. They originate from lipid metabolism of fatty acids [32, 33] (ω-3 for propanoic acid and ω-9 for nonanoic acid), intrinsically connected to complete oxidation in highly oxidative environments and bacterial fermentation. These VOCs may reflect respiratory microbiota dysbiosis, epithelial cell damage or chronic cellular inflammation [6, 34–36]. Although specific biological mechanisms remain unclear, these compounds likely reflect pathophysiological processes relevant to asthma's inflammatory component. In line with this, recent studies have strengthened the clinical relevance of VOCs by identifying volatile biomarker signatures capable of predicting sputum eosinophilia with diagnostic accuracy comparable to conventional inflammatory markers [36].

Regarding tobacco exposure, VOC detection has been analysed clinically in healthy populations [22] and COPD patients [5]. Nonanal and hexanal were identified as discriminative VOCs associated with oxidative stress, aligning with previous pathophysiological frameworks [6] and consistent with findings from Callol-Sanchez et al. [15], where nonanoic acid emerged as indicative of molecular oxidative phenomena in lung cancer patients who smoked, although cancer patients were excluded from our study to avoid bias. These findings might explain the observed data in our study: detection and presence of propanoic and nonanoic acids in asthma patients controlled by treatment, acting through alternative inflammatory pathways, mainly lipid peroxidation [33]. However, while the action mechanisms of asthma medications (inhaled corticosteroids, bronchodilators and leukotriene antagonists) are well known, there is limited information about metabolic changes and their impact on VOC detection, potentially due to study difficulties, metabolic complexity and medication effects on outcomes. Generally, inhaled corticosteroids reduce isoprene and pentane levels by decreasing oxidative stress and lipid peroxidation. Bronchodilator treatment increases isoprene by reducing oxidative damage, while leukotriene antagonists decrease exhaled nitric oxide and volatile sulfides by reducing inflammation and energy metabolism [20, 35, 36]. Nevertheless, some studies have significantly correlated VOC detection with urinary salbutamol and systemic corticosteroids [20].

We recognise that our study has limitations, particularly due the use of historical controls, which may introduce selection or temporal biases. Furthermore, VOCs were analysed qualitatively rather than quantitatively, which limits the ability to measure absolute concentrations or identify dose–response relationships. The relatively advanced age of the study cohort (mean age 64 years) may limit the external validity of our findings in younger populations with asthma. Despite these limitations, our study presents important strengths: the rigorous internal and external validation of VOC biomarkers [22] in a reference centre, careful controlling for ambient VOC contamination, and a detailed evaluation of tobacco exposure, all of which reinforce the reliability and clinical relevance of our findings.

Conclusions

Propanoic and nonanoic acids appear to be associated with residual airway inflammation and oxidative stress in patients with clinically controlled asthma. Detecting these VOCs, overcoming current methodological challenges (e.g. using portable devices), may provide a valuable tool for monitoring inflammation and treatment adherence. Further research is necessary to validate their role as reliable biomarkers in asthma management.