In Vivo Real-Time Metabolic Isotopic Tracing via Breath Analysis upon Deuterated Water Ingestion: A Proof-of-Concept Study Authors: Zhihong Yin, Kapil Dev Singh, Zhifeng Tang, Mélina Richard, Jiafa Zeng, Urs Frey, Pablo Sinues, Xue Li Published: 2026 ABSTRACT Stable isotope tracing with deuterium oxide (D2O) offers a powerful approach for probing metabolic dynamics, yet noninvasive and real-time tools for monitoring in vivo deuterium (D) incorporation remain scarce. This study introduces secondary electrospray ionization–high-resolution mass spectrometry (SESI-HRMS) as a noninvasive, real-time approach for tracking D incorporation in human exhaled breath metabolites. A total of 11 healthy volunteers were enrolled across two independent study centers. Each participant provided up to 20 breath samples across fasting, post-D2O intake, and postprandial phases, yielding 213 samples in total. SESI-HRMS enabled the detection of 2,691 deuterium/hydrogen (D/H) pairs and revealed distinct metabolic trajectories associated with D labeling. Short-chain fatty acids (SCFAs) exhibited biphasic kinetics consistent with rapid orogastric exchange followed by microbial fermentation. Transient increases in pyruvaldehyde were associated with glycemic fluctuations, while dynamic alterations in medium-chain acylcarnitines suggested shifts in mitochondrial energy metabolism. Notably, SESI-HRMS unambiguously resolved D-labeled peaks from natural 13C isotopologues, demonstrating high analytical specificity for stable isotope tracing. Collectively, these findings demonstrate the feasibility of SESI-HRMS real-time breath analysis for noninvasive investigation of D-labeled metabolic dynamics in humans. Introduction Monitoring biomolecule levels and metabolic fluxes is crucial for elucidating metabolic mechanisms in biological systems and their alterations in response to disease and dietary interventions. Stable isotope tracing techniques excel at this objective, with deuterium (D) standing out due to its global labeling capacity and ease of administration. , D can be incorporated into nearly all biomolecules through D–H exchange, thereby enabling safe and cost-effective metabolic labeling. The natural abundance of D is exceedingly low (∼0.0115%), which minimizes interference from the natural isotopic background and enables highly sensitive detection of D-labeled compounds. Deuterium oxide (D2O) introduces D into biological systems directly via oral administration. After ingestion, it is absorbed within 75–120 min and rapidly equilibrates with body water pools, such as plasma, urine, and saliva. D2O labeling has become a powerful tool for metabolic studies, including monitoring lipid − and carbohydrate synthesis and turnover, as well as protein dynamics. , Despite its promise, most current implementations of D-based metabolic tracing rely on biofluid or tissue sampling followed by liquid- or gas chromatography–mass spectrometry (LC–MS or GC–MS), limiting their suitability for noninvasive, high-frequency, or real-time monitoring. More recently, deuterium metabolic imaging (DMI) has further highlighted the broader potential of D-based metabolic interrogation in vivo by combining D-labeled substrates with magnetic resonance spectroscopic imaging to generate spatially resolved metabolic maps within tissues. In parallel, real-time breath analysis has emerged as a compelling noninvasive approach for monitoring metabolic dynamics, with exhaled volatile and semivolatile metabolites providing access to systemic metabolic processes. Within the broader landscape, breath-based isotopic tracing may complement DMI by offering a distinct type of metabolic readout. Whereas DMI emphasizes spatially resolved mapping of labeled metabolism within tissues, breath analysis is well suited for real-time and repeated monitoring of dynamic metabolic changes reflected in exhaled metabolites. Isotope-based breath studies have already shown considerable clinical potential, as exemplified by the widely used 13C-urea breath test for Helicobacter pylori diagnosis. However, breath-based D2O labeling for metabolite-resolved monitoring of metabolic dynamics remains largely unexplored. Pioneering work using flowing afterglow mass spectrometry (FA-MS) has demonstrated that D2O can enable real-time and noninvasive tracking of total body water in vivo through exhaled breath analysis. − However, the limited selectivity and low resolution of FA-MS prevent the detection of trace-level metabolites in breath and the resolution of complex molecular mixtures, posing a key barrier to advancing D-based breathomics. Building on our prior animal study, here we extend D-resolved secondary electrospray ionization–high-resolution mass spectrometry (SESI-HRMS) for real-time and noninvasive isotopic tracing of breath metabolites following D2O ingestion in humans across fasting and postprandial states. Results and Discussion A total of 11 healthy volunteers were enrolled across two study centers. In the discovery cohort, 5 healthy volunteers (3 males and 2 females; mean age, 35.2 ± 10.8 years) were recruited in Switzerland from July to September 2022. The validation cohort consisted of 6 participants (5 males and 1 female; mean age, 25.3 ± 5.7 years), who were recruited in China in September 2023. Participant characteristics, including age, sex, weight, body mass index (BMI), and D2O dosing, are shown in Table . All participants performed 20 breath measurements (4 pre- and 16 post-D2O ingestion), except for one subject, who provided 13 measurements (1 pre- and 12 post-D2O ingestion), yielding a total of 213 breath measurements. 1Participant Characteristics Group/Subject No. Sex Age (years) Weight (kg) BMI (kg/m 2 ) Dose D 2 O (mL) Group 1: Discovery cohort - Switzerland (CH) CH-1M3477.827.5754.46CH-2M4671.522.3250.05CH-3F406121.1142.70CH-4M2962.821.2343.96CH-5F275820.5540.60 Group 2: Validation cohort - China (CN) CN-1M265217.9936.40CN-2M318829.7561.60CN-3M246920.6048.30CN-4M246118.8342.70CN-5M236423.5144.80CN-6F245020.2835.00 Participant Characteristics Upon preprocessing the raw mass spectra, we generated a data matrix of 213 × 6,305 (samples × m/z features). Figure A shows an example set of 20 breath mass spectra from a representative participant in the m/z region 88.0473–88.0521, corresponding to the M + 1 peak region of deprotonated butyric acid. , Consistent with our previous animal study, we detected a distinct D incorporation peak that appeared after D2O ingestion, whose relative intensity (normalized to the corresponding 13C isotopic peak) varied over time. This observation underscored the robustness and sensitivity of our approach for detecting even minute D incorporation, and the dynamic changes in peak intensity may reflect ongoing metabolic processes in vivo. Notably, the high resolution of SESI-HRMS enabled the unambiguous separation of D-labeled peaks from the naturally occurring 13C isotopic peak, allowing true D incorporation to be distinguished from 13C isotopic interference. Real-time tracking of D incorporation in exhaled butyric acid. (A) High-resolution M + 1 mass spectrum of exhaled deprotonated butyric acid [C4H7O2]− from a representative subject, demonstrating 13C and D-labeled peaks. Experimental peak positions align closely with exact m/z values (dashed lines). (B) Temporal profile of the deprotonated butyric acid D/H ratio for the same participant as in panel A. Dashed lines indicate the time points of D2O and yogurt ingestion. (C) Individual D/H ratio temporal profiles of deprotonated butyric acid across the study cohort (N = 11). The y-axis is adjusted to 0–1.2 × Y 3 (where Y 3 denotes the third-highest value) to improve visualization of the time-course changes; the full-scale profiles are shown in Figure S1. Blue lines represent the discovery cohort (N = 5), and red lines represent the validation cohort (N = 6). H = hydrogen; D = deuterium; D2O = deuterium oxide. Building on these observations, we next examined the temporal profile of the isotopic ratio ([C4H6DO2]−/[C4H7O2]−) of deprotonated butyric acid. Figure B shows the corresponding time trace from the same participant as in Figure A. The ratio remained at background levels during the four data points preceding D2O ingestion. Immediately after D2O ingestion, the ratio peaked within 5 min and declined by nearly 70% at 15 min. Given that complete absorption of orally ingested D2O and its equilibration with body fluids generally require approximately 75–120 min, this rapid decrease most likely reflects clearance of residual D2O in the oral cavity. , This characteristic “spike-then-drop” pattern was consistently observed across all individuals in both cohorts (Figure S1). At the group level (Figure C), the overall temporal pattern remained broadly similar across individuals. Over the two hours after D2O ingestion, the D/H ratio rose rapidly, followed by a pronounced decline and subsequent stabilization. After yogurt consumption, the ratio increased again among most participants. Some interindividual variability was observed at baseline, which may partly reflect differences in fasting status. Participants were instructed to fast for at least eight hours, but fasting duration beyond this minimum was not further specified. In addition, subtle differences were also observed across individuals in the magnitude and temporal profile of the later response. Given that short-chain fatty acids (SCFAs) originate primarily from gut microbial metabolism, , these differences may reflect variation in gut microbial composition and metabolic activity among participants. Although yogurt intake was standardized to 300 g and the products showed broadly comparable macronutrient profiles (approximately 10% carbohydrates, 3% protein, and 10% fat, predominantly saturated), the yogurt used in the two cohorts was purchased locally. As a result, compositional differences, including microbial content, may have contributed to the variability observed between cohorts. To systematically characterize the dynamic profiles of D-enriched molecular features across fasting and feeding states, we identified isotopologue pairs differing by n × 1.0063 Da (i.e., mass difference between H and D, where n = 1, 2, 3, ...) and calculated their signal intensity ratios. This procedure yielded 2,691 isotopic D/H pairs, including 2,372 detected in positive ion mode and 319 in negative ion mode. Based on the BIC profile, which reached a plateau at approximately K = 10 (Figure S2), noise-enhanced von Mises–Fisher mixture (NAvMix) clustering identified ten temporal trajectory clusters comprising 1,026 isotopologue pairs (Figures S3 and S4), each cluster representing a group of isotopologue pairs that shared a similar response profile across the experimental time course. We first selected positive response clusters showing increased D-labeling signals during the fasting and/or postprandial phases. The dominant temporal characteristics of these selected clusters are summarized in Table S1. Among these, we focused subsequent analyses on three representative clusters (Clusters #4, #7, and #10, 315 isotopologue pairs in total), which captured distinct positive response patterns (Figure A): a biphasic response spanning the post-D2O and postprandial phases (Cluster #4), a sustained positive postprandial increase (Cluster #7), and a sharp transient postprandial surge (Cluster #10). The 315 isotopologue pairs assigned to these three representative clusters were subsequently mapped to the Human Metabolome Database (HMDB), and the corresponding putative annotations are provided in Supplementary Table 1. Cluster-based temporal profiles of D/H ratios following D2O and yogurt intake. (A) Standardized mean D/H ratio trajectories for three representative positive response clusters selected from the ten-cluster NAvMix model. Each panel represents the standardized mean D/H ratio of all isotopologue pairs within the corresponding cluster, normalized by the mean and standard deviation of the −4 time point. (B–D) Representative D/H ratio temporal profiles for putatively annotated features characteristic of each cluster: deprotonated lactic acid [C3H5O3]− for Cluster #4 (B), deprotonated pyruvaldehyde [C3H3O2]− for Cluster #7 (C), and protonated hydroxy-propionylcarnitine [C10H20NO5]+ for Cluster #10 (D). Blue lines denote the discovery cohort (N = 5), and red lines denote the validation cohort (N = 6). Cluster #4 (n = 105) represented the biphasic response spanning the post-D2O and postprandial phases. Although its temporal profile resembled that of Cluster #5, Cluster #4 maintained D/H ratios above baseline after the initial peak, whereas Cluster #5 returned more closely toward baseline (Figure A and Figure S4). This more sustained post-D2O response motivated our focus on Cluster #4 for subsequent analyses. The D/H ratio increased immediately following D2O intake, declined rapidly, and rose again approximately 1–1.5 h after yogurt ingestion. Putative metabolite annotations (Supplementary Table 1) suggested enrichment of SCFA-related features, including propionic, valeric/isovaleric and caproic acids, together with other small organic acids such as lactic acid (Figure B). This compositional bias was further supported by RefMet-based enrichment analysis, which identified a significant over representation of oxygenated hydrocarbons (Fisher’s exact test, p < 0.001; Figure S5). The shared biphasic temporal pattern observed across these putatively annotated metabolites suggests a common underlying source or coordinated response. The initial spike likely reflects transient local D–H exchange with residual D2O in the oral and upper gastrointestinal compartments before systemic equilibration. This interpretation is also consistent with previously reported early deuterium kinetics in saliva and blood following D2O ingestion. − The delayed postprandial increase is compatible with microbial fermentation of dietary carbohydrates following food intake, a process known to produce lactic acid and SCFAs. For example, increased carbon availability has been shown to shift bifidobacterial metabolism toward acetate and lactic acid production. Cluster #7 (n = 134) displayed a sustained postprandial response following yogurt ingestion (Figure A). The D/H ratio remained near baseline during the baseline and early post-D2O phases, aside from a modest transient increase after D2O ingestion. After yogurt ingestion, the ratio rose to a pronounced postprandial peak, then declined and stabilized above baseline over the subsequent two hours. Putative annotations suggested that this cluster involved features matching several chemical classes, including prenol lipids, organooxygen compounds, fatty acyls, carboxylic acids and derivatives, together accounting for over half of the assigned features (68/134; Supplementary Table 1). Consistent with this chemical diversity, RefMet-based enrichment analysis did not identify any single chemical class reaching statistical significance (Figure S6), indicating substantial chemical heterogeneity within this cluster. Pyruvaldehyde provides an illustrative example at the level of an individual putatively annotated feature (Figure C). Pyruvaldehyde, also known as methylglyoxal, has been described as a glycolytic byproduct formed from glycerol-3-phosphate (G3P) and dihydroxyacetone phosphate (DHAP), and is normally counteracted by the glyoxalase system. This biochemical context may be compatible with the observed peak-decline pattern, although direct metabolite confirmation was not performed in the present study. Other putatively annotated features in this cluster included γ-aminobutyric acid (GABA), malondialdehyde, dimethylglycine, β-aminobutyric acids, and pyroglutamic acid. These candidate annotations are associated with processes such as protein intake, lipid peroxidation, one-carbon metabolism, amino acid catabolism, and glutathione turnover, respectively. In a broader physiological context, , these findings highlight the potential of D2O-based breath analysis to capture dynamic changes across chemically diverse metabolic features, pending further targeted validation. Cluster #10 (n = 76) shared a postprandial-dominant trajectory with Cluster #7 but exhibited a more pronounced postprandial response, characterized by a shaper surge immediately after yogurt ingestion followed by a decline toward baseline (Figure A). Putative annotations (Supplementary Table 1) and RefMet-based enrichment analysis suggested an over representation of short-chain acylcarnitine-related features in this cluster (Figure S7; p = 0.001), predominantly C3–C6 species, including oxo- and hydroxy-propionylcarnitines, as well as butyryl-, valeryl-, and hexanoyl-carnitines. Hydroxy-propionylcarnitine provides an illustrative example of this within-cluster trajectory (Figure D). Acylcarnitines have been widely associated with mitochondrial fuel handling and metabolic flexibility during fasting-to-feeding transitions. − In this context, the pronounced postprandial response observed here may be compatible with short-term changes in mitochondrial acyl-group handling as substrate availability shifts after yogurt ingestion. The predominance of C3–C6 acylcarnitine-related features may reflect contributions from multiple metabolic inputs, including branched-chain amino acid (BCAA) catabolism and gut microbial SCFA metabolism. − To directly compare isotopic incorporation across clusters, Figure A summarizes the distributions in the apparent number of exchanged D atoms (n). Cluster #4 was dominated by single-site D incorporation, with 86% of features (90/105) assigned to n = 1. In comparison, Clusters #7 and #10 showed broader multisite labeling distributions, with the most frequent assignment at n = 3, accounting for 36% of features in Cluster #7 (48/134) and 43% in Cluster #10 (33/76). Notably, Cluster #10 also contained a distinct high-D-incorporation subgroup, with a prominent peak at n = 19 (Figure A), driven predominantly by acylcarnitine-associated features. Among the 16 carnitine-assigned features in cluster #10 (Supplementary Table 1), 12 were observed at n = 19, whereas the remaining four were detected at n = 13 (1 feature), n = 11 (1 feature), and n = 7 (2 features). To assess the reliability of the dominant 19 D assignments, we selected eight putative 19D-labeled acylcarnitine features with sufficient signal quality for isotopic pattern validation. For these eight features (Table S2), measured isotopic patterns showed close agreement with simulated patterns in both the 12C and corresponding 13C windows (Figure B; Figure S8). Pearson correlation analysis across breath samples further supported these assignments, with median r values of 0.956 for the 12C window, 0.858 for the 13C window, and 0.958 for the combined fit (Figure C). Mechanistically, the high D counts observed for acylcarnitine-related features may be with multisite incorporation of water-derived hydrogens into acyl-CoA and carnitine metabolism through β-oxidation associated reactions and NAD­(P)­H-dependent reductions. For a C10 acyl moiety, four β-oxidation cycles would involve approximately 16–20 potentially labelable positions, consistent with the n ≈ 19 isotopologue observed here. , By contrast, Cluster #4 was characterized mainly by single-site D–H exchange (∼86% 1-D incorporation), consistent with its enrichment in organic acids and rapid, nonenzymatic D–H exchange of carboxylic OH groups. Together, these comparisons help distinguish shallow exchange from multisite metabolic incorporation and suggest that D2O-based labeling may capture distinct patterns of postprandial metabolic turnover across chemically and physiologically diverse metabolite pools. Deuterium incorporation patterns and isotope-pattern validation of high-level D-incorporation features. (A) Distribution of the apparent number of incorporated deuterium atoms (n) across three representative time-course clusters (Clusters #4, #7, and #10). Histograms represent the fraction of features within each cluster that exhibit a given level of D incorporation. The y-axis is truncated (45–85%) to improve visualization of differences among lower-frequency incorporation levels. (B) Isotope-envelope comparison for one putative 19D-labeled acylcarnitine ([C10HD19O5N]+; m/z: 253.2529) in a representative breath sample. Experimental profiles are compared with exact-line theoretical profiles simulated at a resolution of 280,000 for the 12C and corresponding 13C windows, with intensities normalized to the 12C apex. Additional putative 19D-labeled acylcarnitine comparisons are shown in Figure S8. (C) Pearson correlation analysis comparing measured and theoretical isotope patterns for the eight selected putative 19D-labeled acylcarnitine features across breath samples. Only samples in which both the 12C and corresponding 13C centroids were detected within ± 5 ppm of the theoretical m/z values were considered evaluable. To obtain a systematic view of the biology captured by D2O-based isotopic labeling combined with real-time SESI-HRMS breath analysis, we performed pathway enrichment analysis (Figure ). Out of a total of 2,691 isotopic D/H pairs, 315 isotopologue pairs assigned to Clusters #4, #7, and #10 were selected as the significant set, with the remaining pairs used as background for pathway enrichment analysis based on Fisher’s exact test. The analysis revealed significant over-representation (p < 0.05) of pathways associated with central carbon and amino-acid metabolism, including glycine/serine/threonine metabolism, butanoate metabolism, valine/leucine/isoleucine biosynthesis and degradation, propanoate metabolism, and pyruvate metabolism (Figure ; Supplementary Table 2; Figure S9). At the network level, these enriched pathways appeared to converge on a limited number of highly connected metabolic hubs, including pyruvate metabolism and SCFA-related pathways. This convergence suggests that breath metabolomics may capture coordinated, system-level metabolic changes across the organism, rather than merely isolated biochemical reactions. Upon intake, D2O rapidly equilibrates with the body-water hydrogen pool, enabling solvent-mediated D–H exchange at labile sites and incorporation into newly formed C–H (C–D) bonds during enzyme-catalyzed reactions. − Pathways with high turnover and abundant exchangeable or transferable hydrogens may therefore be more susceptible to D labeling and may be detected earlier in D2O-based tracing readouts. The observed enrichment of pyruvate/lactate-related metabolites is consistent with rapid turnover of the pyruvate–lactate nodes and its dependence on NADH-mediated redox cycling, which differs substantially between fasting and postprandial states. − Through one-carbon metabolism, D tracers in serine may be transferred into NAD­(P)H and subsequently into malate, proline, and other downstream metabolites. − BCAA catabolism supplies carbon to acetyl-CoA and succinyl-CoA. BCAAs, their α-keto acids, and succinate-linked tricarboxylic acid (TCA) intermediates have been associated with differential substrate utilization across metabolic states. − SCFAs are key microbiota-derived fuels that influence host energy homeostasis and signaling. , Fasting promotes hepatic ketogenesis (β-hydroxybutyrate and acetoacetate), whereas feeding shifts metabolism toward carbohydrate utilization. Accordingly, changes in D2O labeling of SCFA-related features may reflect the combined effects of host fuel switching and the availability and oxidation of microbiota-derived SCFAs. − Network view of enriched metabolic pathways and mapped metabolites. The bigger, tan-colored nodes denote metabolic pathways identified as significantly enriched by Fisher’s exact test using isotopologue pairs from Clusters #4, #7, and #10 as the significant set and all remaining pairs as background. Pathway node size reflects the number of cluster-associated isotopic features mapped to each pathway. Remaining smaller nodes represent mapped metabolites. Red circles indicate compounds assigned to the isotopologue pairs belonging to Clusters #4, #7, or #10, whereas gray circles represent compounds assigned to the background features. Enrichment statistics and the ranked pathways are provided in Figure S9 and Supplementary Table 2. While this study demonstrates the utility of SESI-HRMS for real-time in vivo tracing of D-labeled metabolites after D2O administration, several limitations warrant consideration. First, the present study should be regarded as exploratory and hypothesis-generating, given the modest sample size (N = 11) and restriction to healthy young volunteers. These factors constrain statistical power for broader biological inference and limit the generalizability of our findings to broader and clinically relevant populations. Future work should therefore evaluate this approach in larger, more heterogeneous cohorts, and further refine its application across diverse complementary experimental systems. Second, our experimental protocol captured metabolic responses over a relatively short 5-h fasting-to-feeding window, which constrains the assessment of longer-term metabolic adaptations. Extending breath sampling over longer periods in future studies could enable characterization of slower kinetic processes and provide a clearer view of interindividual variability in D2O incorporation dynamics. Third, metabolite annotation in the current work remains putative because it relies primarily on accurate-mass MS data, while interpretations involving potential microbial contributions are further limited by the absence of direct microbiome characterization. Future studies combining multiomics readouts (e.g., microbiome profiling) with systematic MS/MS validation, and confirmation with chemical standards for key metabolites, should improve annotation confidence and mechanistic insight. , Expanding breath-metabolite libraries via community standards will further enhance translational readiness. , Conclusions In this proof-of-concept study, we show that D2O-based isotopic labeling combined with SESI-HRMS enables noninvasive, time-resolved assessment of metabolic dynamics through exhaled breath. High-resolution isotopic analysis revealed reproducible labeling dynamics across fasting-to-feeding transitions and linked D–H exchange features to central carbon, SCFA, and amino-acid metabolism. Together, these findings support breath-based deuterium tracing as a dynamic in vivo readout of nutritional and metabolic responses, with particular relevance for postprandial metabolic phenotyping, metabolic flexibility, and host-microbial metabolic interactions. With further targeted validation in larger and more diverse cohorts, this framework may provide a useful tool for investigating metabolic conditions marked by impaired postprandial control, altered carbon fluxes, or dysregulated lipid handling, particularly insulin resistance and early type 2 diabetes, and potentially related obesity-associated metabolic dysfunction. Supplementary Material