Identifying Biomarkers for Chronic Obstructive Pulmonary Disease in the Salivary Metabolome Authors: Denisa Asandei, Keiron O’Shea, Rachel Paes de Araujo, Adrian Mironas, Ricardo Da Costa, Timothy Asibey-Berko, Subhamay Ghosh, Chuan Lu, Scott O’Rourke, Luis A J Mur, Keir E Lewis Published: 2026 ABSTRACT PurposeWe compared salivary metabolomic signatures from patients with stable COPD across a range of severity of airflow obstruction with healthy controls.Patients and MethodsIn this exploratory study, 47 people with COPD and 48 age-matched, healthy controls, provided saliva that was assessed by flow infusion electrospray mass spectrometry (FIE-MS). Spectra were interrogated using an open-source library DIMEpy package.ResultsFour potential biomarkers identified the presence of COPD with a sensitivity of 73% and specificity of 72%. Six metabolites predicted the level of airflow obstruction, FEV1% in the COPD cohort (P < 0.001, R2 > 0.3, AUC > 0.7), whilst a range of multivariate approaches targeted six metabolites linked to COPD stage of severity (P < 0.001, AUC > 0.7). Identification of the metabolites suggested changes in pterin biosynthesis, lipid processing, nucleotide metabolism and melatonin in COPD patients.ConclusionThis proof-of-concept study shows metabolic fingerprinting of saliva samples is feasible and can differentiate patients with COPD from people (including smokers) without COPD and correlates with COPD severity as defined by level of airflow obstruction. Metabolic fingerprinting also offers insights into the metabolic pathways involved in COPD aetiology. Introduction Chronic obstructive pulmonary disease (COPD) is the fourth leading cause of death worldwide, causing 3.5 million deaths in 2021, approximately 5% of all global deaths.1 With the increase in smoking in low-middle income countries (LMICs) and increasing age in developed countries, the global prevalence is predicted to rise.2 COPD is characterized by fixed airflow obstruction, that is driven by inflammation and is usually progressive. The inflammation is caused by inhalation of external noxious agents, including cigarette smoke3 but also air pollution4 biomass fuel and occupational dusts. Childhood infections, airways hypersensitivity and allergens, abnormal lung development in early life are increasingly recognised. These external factors interact with host factors (in people with genetic susceptibility) in a complex way over time causing chronic airways inflammation, lung damage, poorly reversible airways obstruction and clinical manifestations.5 Symptoms of COPD include cough (usually productive), wheeze, breathlessness and fatigue. A diagnosis of COPD should be considered in people with appropriate symptoms and exposure to risk factors but the only acceptable current test to confirm it is post-bronchodilator forced spirometry confirming airflow obstruction, currently Forced Expiratory Volume in 1 second (FEV1) to Forced Vital Capacity (FVC) ratio <0.7.6,7 However, there are many problems with spirometry: many patients struggle with the forced manouevres, it takes over 15 minutes to perform properly, it is an aerosol‑generating procedure so can be compromised in respiratory pandemics; moreover, controversies exist over lower limits of normal especially in elderly and non‑Caucasians. Post-bronchodilator airways obstruction can occur in asthma and bronchiectasis so is not specific to COPD. Spirometry tends to overestimate lung obstruction in people with mixed obstructive and restrictive lung disease, where correction for total lung capacity improves accuracy.8 Spirometry does not pick up early inflammatory or emphysematous changes (seen on CT scans especially in Black male smokers9 or in people or smokers with mild airflow obstruction.10 People with chronic respiratory symptoms but with normal spirometry have been shown to be at increased clinical risk of hospitalisations and death, suggesting spirometry does not measure true COPD disease state.11 FEV1 is hard for non-specialists to perform and interpret and probably most importantly spirometry availability is poor in LMICs and sub-optimal even in developed countries especially in primary care . There is increasing need for a non-invasive, simple (ideally point-of-care) objective diagnostic test for COPD. To date no blood, sputum nor urine biomarkers have been reliably shown to diagnose COPD or severity of airflow obstruction although sputum and serum markers (eg blood eosinophils) predict response to certain treatments. Metabolomics is the scientific study of biological processes within an organism, cell or tissue. It allows for the systematic quantification and identification of metabolites sourced from a biological specimen and can give a comprehensive targeted or global assessment of endogenous metabolites.12 The potential of metabolomics to define clinically relevant biomarkers is being increasingly acknowledged.13 Metabolomics represents a promising technology in the discovery of biomarkers from complex source materials especially in a complex disease such as COPD, with multiple causes, multiple inflammatory and multiple repair pathways.5 Metabolomics have been applied to respiratory diseases for more than 10 years mainly concentrating on bronchoalveolar lavages, blood and exhaled breath condensates as reviwed by Gea in 202314 and Torelli 2024.15 Direct infusion mass spectroscopy (DIMS) has become a popular platform for high-throughput metabolomic analyses.16 Using DIMS, samples are directly introduced to the mass spectrometer negating time-costly chromatography and increasing sensitivity. While mostly composed of water (99%), the chemical composition of saliva is known to changes considerably in response to a variety of different physiological states, stimuli, insults and stressors.17,18 Salivary metabolomics has been used to detect and monitor not only oral diseases such as local oral cancer,19 systemic cancers,20 and lung cancer.21 Sputum metabolomics have also been applied to other non-respiratory diseases including Alzheimer’s22 and gestational diabetes.23 Saliva is an attractive biofluid for respiratory disease testing as it is a non-invasive and even more accessible than sputum. Changes in salivary metabolites may reflect lung disease in the lower respiratory broncho-epithelium and lung parenchyma but we could find no studies applying metabolomic approaches to sputum or saliva. However, certainly salivary metabolomes are also influenced by other factors such as chewing stimulation24 smoking status,25 food, periodontal disease and even gender, so any sputum or salivary metabolomic patterns in COPD need to be carefully assessed. Aims In this proof-of-concept exploratory study, we wanted to see if saliva was a viable biofluid for studying metabolomics and to compare patterns in people with COPD and healthy controls. Materials and Methods Participants The study was approved by our institutional ethics board (16/WA/0036) and subjects provided written informed consent prior to enrolment. 47 patients with stable COPD, attending Prince Philip Hospital clinics with no change in symptoms or medications for at least 3 months) and 48 controls (spouses, hospital and university staff) who had no respiratory symptoms and no known lung disease were recruited. Spirometry (Vitalograph Alpha R, Buckinghamshire, UK) was performed on patients at study enrolment. Smoking status was self-reported and validated with exhaled carbon monoxide (eCO) (Pico+TM, Smokerlyzer, Bedfont Scientific Ltd, Rochester, UK). A smoker was defined as anyone with eCO >10 parts per million (ppm) or admitting to current smoking. An ex-smoker was defined as more than 100 cigarettes in a lifetime but none in the last 6 months and eCO <10ppm. A non-smoker is defined as having smoked less than 100 cigarettes in their lifetime and none for 6 month and eCO<10ppm. Table 1 describes the groups.Table 1Characteristics of Both Groups (95 participants)CharacteristicCOPD n=47Control n=48p-valueAge (years)66.1 (12.3)55.7 (17.4)0.01% Male68%55%0.4BMI27.6 (7.5)29.3 (6.7)0.26Smoking status% Current smokers23%22%0.86% Ex-smokers64%40%0.17% Never smokers11%38%0.003FEV1% Predicted52.7 (±17)NANotes: In the COPD group, 12 were classified as moderate airways obstruction, 13 severe and 5 were very severe according to GOLD 2026 guidelines.Abbreviation: NA, not available. Table 1Characteristics of Both Groups (95 participants)CharacteristicCOPD n=47Control n=48p-valueAge (years)66.1 (12.3)55.7 (17.4)0.01% Male68%55%0.4BMI27.6 (7.5)29.3 (6.7)0.26Smoking status% Current smokers23%22%0.86% Ex-smokers64%40%0.17% Never smokers11%38%0.003FEV1% Predicted52.7 (±17)NANotes: In the COPD group, 12 were classified as moderate airways obstruction, 13 severe and 5 were very severe according to GOLD 2026 guidelines.Abbreviation: NA, not available. Characteristics of Both Groups (95 participants) Notes: In the COPD group, 12 were classified as moderate airways obstruction, 13 severe and 5 were very severe according to GOLD 2026 guidelines. Abbreviation: NA, not available. Saliva Sample Collection and Preparation The protocol was adapted from our previous work on sputum metabolomics.26 Particpants were asked to cough into sterile, 50 mL polypropylene tubes (Greiner Bio-One Ltd., UK). These spontaneous sputum/saliva samples were transferred to a sterile 2 mL microcentrifuge tube to which 7.5 mg of ≤ 160 µM (particle size) acetone-washed glass beads were added. To these, 380 µL of a solvent mix of high-performance liquid chromatography (HPLC) grade methanol (MeOH) and chloroform (CHCl3), in a ratio of 4:1 was added. To homogenise the mixture and disrupt the cells, saliva samples were placed on a vortex mixer for 5 seconds. Following this, samples were shaken for 20 min at room temperature and then stored at −20 °C for 20 min to allow protein precipitation. Samples were centrifuged (Thermo Scientific Heraeus Pico 21 microcentrifuge) at 21,000 x g for 6 min at 4 °C and then transferred to a sterile 2 mL microcentrifuge tube. From this, 100 µL were transferred to an MS vial. FIE-MS Analysis and Data Processing Untargeted metabolite profiling was undertaken using a Q Exactive plus mass analyser instrument with UHPLC system (Thermo Fisher Scientific©, Bremen, Germany). Data acquisition for each saliva sample was done by alternating the positive and negative ionisation modes, throughout four different scan ranges (15–110 m/z, 100–220 m/z, 210–510 m/z, 500–1200 m/z; All metabolic processing was performed using the open‑source DIMEpy package (https://www.github.com/AberystwythSystemsBiology/dimepy) in a sample-independent manner. Raw mzML files were acquired in positive mode and converted to centroided raw values. Simple quality control was employed, removing samples where the total ion count (TIC) was greater than the 3 the mean absolute error (MAE) of all summed spectra. Spectra were binned to 0.125 m/z to reduce the potential of spectrum errors arising from ion concentration fluctuations and machine drift. Value imputation was used to fill missing binned values to facilitate an unbiased comparison of univariate and multivariate statistical analyses. To remove bias of ion intensities between measurements, spectra were transformed through TIC and log normalised. Each spectrum was Pareto scaled before being exported to a CSV file for metabolic analysis. Metabolic Feature Selection and Analysis Explanatory metabolite features were identified using a combination of Welch’s t-test, ANOVA, random forests (RF), partial least squares discriminant analysis (PLS-DA) and principal component analysis (PCA). Receiving operating characteristic (ROC) curves and micro-averaged area under curve (AUC) values generated from cross-validated support vector machines (SVMs) were used to evaluate the predictive performance of selected features. In instances where patient sample numbers were insufficient for 10-fold cross validation, “leave one out” cross validation was used to evaluate the predictive performance of our metabolite models. Spearman’s rank correlation coefficients between selected explanatory variables were calculated, where variables with a high correlation (r ≥ 0.9) were considered to belong to a metabolite cluster indicative of the presence of possible ion types of a single compound. For signal annotation, accurate mass values were extracted from bins and programmatically queried using DIMEdb (https://dimedb.ibers.aber.ac.uk/). Once probable ion derivatives were removed, metabolites were putatively annotated to MSI level 2, based upon physicochemical properties and spectral similarities. We were left with two unidentified metabolites which were taken forward for evaluation using a cross-validated SVM. A combination of ANOVA, random forests, PLS-DA and SVM-derived AUCs were used to build a metabolic model capable of classifying GOLD Stages of severity of airflow obstruction. Results Initial assessments using PCA suggested some discrimination between the salivary metabolomes from COPD patients compared to controls (Figure 1a). A cross-validated accuracy of 72% was achieved, with 33 out of the 47 COPD patients and 38 out of 48 Control samples - achieving a total sensitivity of 73% and a specificity of 72%. Each targeted metabolite that could successfully discriminate between COPD and Controls is individually displayed as micro-averaged cross-validated receiver operating characteristic curves (ROC) (Figure 1b).Figure 1(a) PCA score plot of our metabolic model and (b) Cross validated ROC for selected putatively identified metabolites that were able to successfully discriminate between COPD and Controls.A scatter plot and four receiver operating characteristic curves comparing COPD and Control samples.The image A showing a scatter plot with legend entries, COPD and Control. The x-axis label is PC 1 (48.9 percent), with tick labels from negative 2.0 to 2.0. The y-axis label is PC 2 (19.5 percent), with tick labels from negative 2 to 2. COPD points and Control points are distributed across the plot, with many COPD points between PC 1 about negative 1.8 to 0.8 and PC 2 about negative 2.2 to 1.8 and many Control points between PC 1 about negative 1.5 to 2.1 and PC 2 about negative 2.2 to 2.5. The image B showing four receiver operating characteristic plots titled Isoxanthopterin, 3,5-Dihydroxydecanoyl-CoA, PA(20:5(6E,8Z,11Z,14Z,17Z)-OH(5)/8:0) and LysoPE(18:3(9Z,12Z,15Z)/0:0). Each plot has x-axis label Specificity with range 0.0 to 1.0 and y-axis label Sensitivity with range 0.0 to 1.0. Each plot includes a diagonal reference line and a stepped curve. Micro-averaged AUC values shown are 0.74, 0.73, 0.75 and 0.77, respectively.Abbreviations: PCA, Principal Components Analysis; ROC, Receiver Operator Curve. (a) PCA score plot of our metabolic model and (b) Cross validated ROC for selected putatively identified metabolites that were able to successfully discriminate between COPD and Controls. The main sources of variation were identified by sequential analyses whereby the outputs were statistically significant (P < 0.05) using Welch’s t-test, RF, PLS-DA variable in projection (VIP) scores and receiving operating characteristic (ROC) curves under curve (AUC) values (Table 2).Table 2Putatively Identified Metabolite Table of Selected Metabolic Features That Demonstrated Superior Discriminant Performance Between COPD and Our Control GroupPutative IDMass-to-ion (m/z)Relative Change in COPD vs ControlAUCp-valuePLS-DA VIP ScoreRF Feature ScorePA(20:5(6E,8Z,11Z,14Z,17Z)-OH(5)/8:0)677.25Increase0.748< 0.0051.5210.007LysoPE(18:3(9Z,12Z,15Z)/0:0)496.25Decrease0.772< 0.0051.1620.006Isoxanthopterin536.125Decrease0.737< 0.0051.1250.0063,5-Dihydroxydecanoyl-CoA475.625Decrease0.732< 0.0051.1070.004Notes: The m/z ratio is the mass of the ion divided by the charge of the ion (expressed in Daltons per charge number).Abbreviations: AUC, area under the curve; PLS-DA, Partial Least Squares Discriminant Analysis; VIP, Variable Importance in Projection; RF, Random Forest. Table 2Putatively Identified Metabolite Table of Selected Metabolic Features That Demonstrated Superior Discriminant Performance Between COPD and Our Control GroupPutative IDMass-to-ion (m/z)Relative Change in COPD vs ControlAUCp-valuePLS-DA VIP ScoreRF Feature ScorePA(20:5(6E,8Z,11Z,14Z,17Z)-OH(5)/8:0)677.25Increase0.748< 0.0051.5210.007LysoPE(18:3(9Z,12Z,15Z)/0:0)496.25Decrease0.772< 0.0051.1620.006Isoxanthopterin536.125Decrease0.737< 0.0051.1250.0063,5-Dihydroxydecanoyl-CoA475.625Decrease0.732< 0.0051.1070.004Notes: The m/z ratio is the mass of the ion divided by the charge of the ion (expressed in Daltons per charge number).Abbreviations: AUC, area under the curve; PLS-DA, Partial Least Squares Discriminant Analysis; VIP, Variable Importance in Projection; RF, Random Forest. Putatively Identified Metabolite Table of Selected Metabolic Features That Demonstrated Superior Discriminant Performance Between COPD and Our Control Group Notes: The m/z ratio is the mass of the ion divided by the charge of the ion (expressed in Daltons per charge number). Abbreviations: AUC, area under the curve; PLS-DA, Partial Least Squares Discriminant Analysis; VIP, Variable Importance in Projection; RF, Random Forest. Next the salivary metabolomes were assessed to identify metabolites that varied with the degree of airflow obstruction. Linear regression identified 6 salivary metabolites with significant associations for FEV1% predicted. (Figure 2). These suggest significant increases of the nucleotide‑associated metabolites adenosine monophosphate, 2′-Deoxy-5′-inosinic acid (dIMP) and the small‑chain fatty acid, acetic acid with higher FEV1 values. However, melatonin, tetrahydrobiopterin and melibiose increased with progressively lower levels of FEV1.Figure 2Correlation and linear regression analysis of metabolite intensities from FI-MS analysed saliva.Six scatter plots showing processed intensity versus forced expiratory volume in one second percent predicted.Image A: Scatter plot ′Adenosine monophosphate′ with ′Processed Intensity′ (-3 to 3) and ′FEV1 percent Pred′ (20 to 120). Line rises from 45 at -3 to 70 at 3. Image B: ′dIMP′ plot with ′Processed Intensity′ (-3 to 4) and ′FEV1 percent Pred′. Line rises from 50 at -3 to 75 at 4. Image C: ′Melatonin radical′ plot with ′Processed Intensity′ (-4 to 4) and ′FEV1 percent Pred′. Line falls from 70 at -4 to 30 at 4. Image D: ′Tetrahydrobiopterin′ plot with ′Processed Intensity′ (-2 to 4) and ′FEV1 percent Pred′. Line falls from 70 at -2 to 45 at 4. Image E: ′Acetic acid′ plot with ′Processed Intensity′ (-3 to 3) and ′FEV1 percent Pred′. Line rises from 50 at -3 to 75 at 3. Image F: ′Melibiose′ plot with ′Processed Intensity′ (-2 to 6) and ′FEV1 percent Pred′. Line falls from 80 at -2 to 45 at 6.Notes: Each point in the plot represents and individual measurement of a sample against its predicted FEV1%, all P<0.05. Adenosine Monophosphate: R2 value 0.414, p <0.001. dMP: R2 value 0.344, p =0.006. Melatonin Radical: R2 value −0.402, p =0.001. Tetrahydrobiopterin: R2 value −0.308, p=0.015. Acetic Acid: R2 value 0.335, p =0.008. Melibiose: R2 value −0.432, p <0.001.Abbreviation: FI-MS, Flow Infusion Electrospray Mass Spectrometry. Correlation and linear regression analysis of metabolite intensities from FI-MS analysed saliva. Given the association with FEV1, subsequently, the salivary metabolome was interrogated for features that could be linked to discrete GOLD categories. Any discriminating metabolites were tested against the GOLD stage, using a unidimensional approach through ANOVA. A total of five metabolites could be significantly associated with GOLD stage (Figure 3), two (melibiose and melatonin) had been previously correlated with FEV1 measurements but three – tetrahydrocortisol, 2-dodecenal and n-acetylcaderine – decreased with increasing COPD severity/airflow obstruction.Figure 3Box and whisker plots of putatively identified metabolites that successfully discriminated between GOLD-severity classification in COPD.Five box and whisker plots showing processed intensity across GOLD stages for five metabolites.Image displays five box and whisker plots for Melibiose, Melatonin, Tetrahydrocortisol, 2-Dodecenal and N-Acetylcadaverine. The x-axis is labeled GOLD stages: 1 (Mild), 2 (Moderate), 3 (Severe), 4 (Very Severe) and the y-axis is labeled Processed Intensity. Melibiose: y-axis from -3 to 7; medians: ~0 at 1, ~1 at 2, ~4 at 3, ~3 at 4; outliers near 6 at 1 and 6.5 at 2. Melatonin: y-axis from -3 to 3; medians: ~-2 at 1, ~-1 at 2, ~0 at 3, ~0.5 at 4; outliers above 2 at 2. Tetrahydrocortisol: y-axis from -3 to 4; medians: ~2 at 1, ~1 at 2, ~0 at 3, ~2 at 4; outlier near -2 at 1. 2-Dodecenal: y-axis from -3 to 3; medians: ~-1 at 1, ~-1 at 2, ~-2 at 3, ~-1.5 at 4; outliers near 2 at 1 and 3 at 2. N-Acetylcadaverine: y-axis from -3 to 6; medians: ~2.5 at 1, ~2 at 2, ~0 at 3, ~2 at 4; outliers near -2 and -0.5 at 1 and 6 at 4.Notes: Mild = FEV1 >80% predicted; Moderate = 50% ≤ FEV1 < 80% predicted; Severe = 30% ≤ FEV1 < 50% predicted; Very Severe = FEV1 < 30% predicted.Abbreviation: GOLD, Global initiative for Obstructive Lung Disease. Box and whisker plots of putatively identified metabolites that successfully discriminated between GOLD-severity classification in COPD. Discussion There have been many studies aiming to define new, specific biomarkers for the diagnosis of COPD but none have entered routine clinical practice. In this study, we explored saliva as a promising biofluid for the newer methods of metabolomic analysis and its potential to be used to differentiate COPD and also severity of airflow obstruction. Saliva proved to be a suitable biofluid to distinguish people with COPD and controls. This parallels other studies which have used nuclear magnetic resonance (NMR)-based metabolomics on exhaled breath condensate (EBC) to separate COPD patients from other groups, particularly asthmatics.27 They found increases in very low molecular weight volatiles such as methanol and ethanol could be linked to COPD in EBC. Another group was able to differentiate asthma patients from COPD by NMR-based metabolomics of EBC samples. They found that COPD patients, in comparison to those with asthma, have had increased methanol and ethanol levels. However, when Laurentiis et al compared EBC and saliva samples of COPD and controls from healthy and laryngectomized subjects, they demonstrated that both sample types were distinct and the EBC was not contaminated by saliva which may be why our study targeted different metabolites.28 Furthermore, none of the metabolites that we targeted in our study were noted in previous systemic metabolomic assessments of COPD which did not consider saliva.29,30 Our discriminatory metabolites included isoxanthopterin (2-amino-4,7-dihydroxypteridine), a pterin-class derivative, that is commonly found in human biofluids. As part of the pterin metabolism, it links purine and folate (vitamin B9) metabolism. Folate and its fully chemically reduced form – tetrahydrofolate (THF) – is an important cofactor in the biosynthesis of nucleic and amino acids.31 Changes in pterins, and possibly folate metabolism, have been associated with malignant diseases and neopterin levels are indicators of cellular immune activation and oxidative stress.32 Changes in pterins have not previously been associated with COPD but could reflect changes in oxidative stress that feature in its pathology.33 In the case of isoxanthopterin, this has shown to increase in the urine of stomach cancer patients but changes have not been reported previously with COPD.34 We also noted significant increases in certain lipids, lysophosphatidylethanolamine (lysoPE) and the oxidized phosphatidic acid (PA(20:5(5Z,8Z,11Z,14Z,17Z)/20:4(5Z,8Z,11Z,14Z)-OH(16R)). Lysophospholipids are usually generated by phospholipase A, whilst PAs can be derived by oxidative stress, so these increases could reflect inflammatory/oxidative events that are features of COPD.35 (S)-Hydroxydecanoyl-CoA is involved in the b-oxidation of fatty acid and this change in COPD patients could be indicative of altered wider lipid processing. To further assess the diagnostic potential of our study, we identified salivary metabolites that could be correlated with measurements of FEV1. In people with COPD, a lower FEV1 was positively correlated with nucleotide metabolism (adenosine monophosphate, dIMP) and fatty acid metabolism (as indicated by the short‑chain fatty acid, acetate). Pterin biosynthesis was again targeted with increased THF with lower FEV1 values. Interestingly, THF is a co-factor in the synthesis of melatonin which also showed a significant negative correlation with FEV1, which could again reflect oxidative changes.36 Indeed, pharmaceutical treatment with melatonin has been shown to reduce the severity of COPD symptoms as well as improving their abnormal sleep patterns possibly by inhibiting airway MUC5AC production via suppression of MAPK signaling in human airway epithelial cells.37,38 As GOLD severity is based on FEV1, it is unsurprising that some of these metabolites (melibiose and melatonin) also discriminate according to GOLD classification. The raised levels of salivary long-chain fatty aldehyde 2-dodecenal, could be a further indication of altered fatty acid metabolism in more severe airflow obstruction (more lung damage). Similarly, tetrahydrocortisol could reflect altered steroid metabolism that could be a product of cellular turnover in a more damaged lung or raised because (inhaled steroid) drug processing. A more novel observation is represented by N-acetylcadaverine, a component of polyamine metabolism. This pathway has many roles in cell proliferation but has been tentatively linked to COPD through the generation of nitric oxide.39 Pulik et al recently applied untargeted proteomic and metabolomic profiling to sputa from 26 smokers with COPD and 16 smokers without COPD). After quality control, 1180 proteins, 187 metabolites and 1234 lipids were retained. Using similar statistical approaches (PCA, PLS-DA) consistent cross-omics patterns were observed. COPD was characterized by upregulation of antioxidants and broad lipid remodeling also involving phosphatidylcholines, sphingolipids and eicosanoids. They also concluded that the multi-omic profiling of induced sputum can capture coherent airway-localized molecular signatures. However, their sputum samples were artificially induced (the method is not described) and smoking status was self-reported. They also emphasized the exploratory nature of their work and need for further validation40 Maniscalco et al applied metabolomics approaches comparing asthma and COPD. However, they applied nuclear magnetic resonance (NMR) testing and not to spontaneous sputum but exhaled breath condensates that require specialist collections. They excluded 15 subjects with co-morbidities (including diabetes which often co-exists in people with COPD) and they lost a further 8 samples that were unsuitable for EBC analysis. They built a training cohort of 20 samples from asthma and 32 with COPD and importantly, enrolled a second cohort for an external blind validation comprising 13 asthma and 20 COPD patients. Their two COPD cohorts were significantly older and had significantly more smokers and pack-years. The model based on their training set correctly identified 12 of 13 asthma patients and 19 of 20 COPD subjects, showing a sensitivity (true positive rate) of 92.3% and a specificity (true negative rate) of 95.0%. The discriminating metabolites were confirmed to be in COPD increased ethanol and methanol and decreased formate and acetone/acetoin. The AUC of the ROC for the total model of Figure 1b was 0.96.41 Freund studied novel sputum biomarkers in 29 current smokers including 14 people with COPD. They tested nanoparticles finding higher concentrations and smaller particle size correlated directly with sputum (IL-1β, IL-6 and TNFα) and serum (IL-8 and lower IL-10) markers of inflammation. Interestingly, they also found nanoparticle concentrations correlated with higher symptoms cores on the COPD Assessment Test. They had smaller numbers of COPD than us and used induced sputum and although there was signal heterogeneity, they proved the concept that sputum can be a reliable biofluid for novel tests.42 We note that none of the biomarkers in Figure 1 add additional value comparing against FEV1. Tt is possible that the biomarkers in Figure 1 record the presence of clinical COPD (an inflammatory state) whereas Figure 2 biomarkers only correlate with one aspect of COPD (degree of airflow obstruction) a measure of tissue destruction. This is similar to the principal that the current “clinical” GOLD phenotype (A, B, E) that determines treatment is not dependent FEV1 or that the key symptoms of COPD such as breathlessness nor cough correlate that well with the degree of airflow obstruction. A weakness of our study was that we did not have spirometry in the controls so there may be some, especially in the smoker/ex-smoker cohort who have asymptomatic airways obstruction. We did not have ethics permission to screen asymptomatic people with additional tests. It is important to remember that to have a clinical diagnosis of COPD (warranting any intervention), people first need symptoms as well as a risk factor/noxious exposure to lung irritants – to consider spirometry only then.43 The hgroups had similar BMIs but the control group were significantly younger and had slightly more females but this was not significant. Groups were not matched for proportions of ex- and never smokers although the effects of previous smoking on metabolomes have not been reported. The groups had very similar proportions of current smokers which we felt was the most important potential confounder. This likely represents the higher levels of smokers in spouses of people with COPD than the general background population of similar age. All our COPD patients were prescribed combinations of inhaled steroids, long-acting beta agonists and long-acting muscarinic antagonists according to GOLD recommendations with no medication changes for at least 4 months, but in this early feasibility study, we did not record medication doses, duration of treatments type of inhaler nor timing of medication with respect to saliva sampling.43 None were prescribed antibiotics (including prophylactic) within 4 months of sampling. We did not compare sputum metabolites against blood or urinary metabolites nor other biomarkers. We did not have enough patients to note the effects nor severity of comorbidities. All patients were deemed clinically stable for at least 6 weeks, ie no symptoms of worsening breathlessness, cough nor increased sputum from their day-to-day symptoms. We did not standardize sampling for time of day (samples were taken between 0900 and 15.00 due to researcher availability in hospitals), and subjects were not fasted and the effects of this and any diurnal variation on sputum metabolomics are not known. We did not record time of last inhaler or last cigarette (although most smokers had done so within the previous 12 hours), record tooth-brushing (time of cleaning or type of toothpaste etc). Although exploratory, we were mindful to make external validation easier and certainly any real-world-feasibility test should have as few prerequisites as possible. We did not attempt any power calculations as this was the first stage exploratory study with untargeted high throughput, mass spectroscopy screening. Strengths of the study include sample size (for a feasibility study) and this is the only study we could find that validated smoking status biologically. We did not exclude people with any other illnesses and allowed any concomitant medications including various inhalers. Another strength is that we applied two statistical techniques (linear against FEV1% predicted and reporting patterns against standard clinical sub-grouping of severity of COPD / airways obstruction. Metabolomic profiling allows better understanding of the molecular pathophysiology of inflammatory airways diseases and could identify new treatment targets. This feasibility pilot study and putative biomarkers require validation with larger cohorts, subgroups according to clinical phenotype (GOLD A, B, E), radiological subtypes, recording of comorbidities and within the same patients comparing exacerbation versus stable states. A non-invasive biofluid and small number/combination of biomarkers could allow the development of bedside tests to diagnose and monitor this common disease more objectively. Conclusion This is an exploratory and hypothesis generating study; it is the first known study applying a high-throughput (non-targeted) metabolomics approach based on FIE-MS, coupled with powerful biomarker selection methods in saliva in stable COPD versus controls. It confirms the early feasibility of using the metabolic profile of saliva to discriminate between disease and healthy (asymptomatic) states and we have found novel differences in metabolites that correlate with the presence and degree of severity of airflow obstruction. Moreover, these metabolites are biologically plausible as novel non-invasive salivary biomarkers for the diagnosis and monitoring of COPD and by increasing our understanding metabolic pathways could even identify future treatment targets.