Carboxyhemoglobin tracks acute arterial desaturation in healthy volunteers: a pooled analysis of four controlled desaturation studies Authors: Juyeon Oh, Kyung Mi Kim, Hyeonseok Lee, Ji-Yeon Bang, Eun-Kyung Lee, Byung-Moon Choi, Gyu-Jeong Noh Published: 2026 ABSTRACT Acute arterial hypoxemia frequently occurs in perioperative and critical care settings; however, the dynamic changes in endogenous biomarkers under these conditions remain poorly characterized. This study aimed to identify which among carboxyhemoglobin (COHb), methemoglobin (MetHb), and lactate best tracks the severity of acute arterial desaturation, and to explore potential ethnic differences in their responses. Data were retrospectively analyzed from four controlled desaturation studies involving 48 healthy volunteers (32 Korean and 16 African). Participants inhaled a hypoxic gas mixture of air and nitrogen to maintain arterial oxygen saturation (SaO₂) between 70% and 100%. Arterial blood samples were collected at each target SaO₂ level to measure COHb, MetHb, and lactate concentrations using CO-oximetry. Linear mixed-effects models (LMMs) were used to quantify the association between SaO₂ and each biomarker while accounting for repeated measurements. A total of 1,194 arterial samples from 48 participants were analyzed. Across SaO₂ ranges of 70–80%, 80–90%, and 90–100%, COHb exhibited the largest relative change among the three biomarkers (P < 0.001). LMM analysis showed a significant association between decreasing SaO₂ and increasing COHb (β₁=–0.0123, P < 0.001). In contrast, no significant associations were observed for MetHb (β₁=0.0003, P = 0.677) or lactate (β₁=–0.0011, P = 0.175). In subgroup analyses, Korean participants demonstrated significantly greater COHb elevations than African participants with increasing severity of arterial desaturation (P < 0.001). Among the evaluated endogenous biomarkers, COHb most closely tracked the severity of acute arterial desaturation. The greater COHb response observed in Korean participants is hypothesis-generating and suggests possible ethnic variability in physiological responses to controlled arterial desaturation.Clinical trial registration: Clinical Research Information Service (CRIS), Republic of Korea (http://cris.nih.go.kr) KCT0008376, KCT0008746, KCT0009938, and KCT0010039.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-54648-0. Introduction Hypoxemia and hypoxia are closely related yet distinct physiological phenomena. Hypoxemia refers to reduced arterial oxygen content or partial pressure (PaO₂), indicating inadequate oxygenation of the blood, whereas hypoxia denotes insufficient oxygen availability at the tissue or cellular level, where oxygen delivery fails to meet metabolic demand1,2. Clinically, however, “hypoxia” is often used to describe an arterial oxygen saturation (SaO₂) below 90%, corresponding approximately to a PaO₂ of 60 mmHg3. Acute hypoxia is frequently encountered in perioperative, critical care, and physiological settings. In addition, transient hypoxia is routinely experienced during breath-hold–related activities such as voluntary apnea and freediving, as well as in certain sports-related conditions, in which short periods of arterial oxygen desaturation occur. These activities have been extensively used as physiological models of brief hypoxic exposure and may provide useful context for interpreting biomarker responses to short-duration desaturation4,5. Although the degree of acute hypoxia can be quantitatively assessed by arterial blood gas analysis, which measures SaO₂, the dynamic changes in endogenous biochemical markers during acute hypoxia remain poorly characterized. Among potential biomarkers responsive to hypoxic stress, carboxyhemoglobin (COHb), methemoglobin (MetHb), and lactate were selected because each reflects a distinct oxygen-sensitive physiological pathway. COHb represents endogenous carbon monoxide production from heme degradation, a process closely linked to oxygen-dependent heme metabolism6. MetHb reflects oxidative modification of hemoglobin iron, which may be influenced by oxidative stress during hypoxia7. Lactate is a classical marker of impaired oxidative metabolism and enhanced anaerobic glycolysis under conditions of insufficient oxygen availability8. Supporting this concept, elevated COHb levels have been observed in mountain climbers exposed to hypobaric hypoxia9, while increased MetHb concentrations have been reported in populations residing at high altitudes10. Lactate has also been proposed as a biochemical marker of tissue hypoxia8,11. However, the relative responsiveness of these biomarkers during acute hypoxia in humans has not been systematically investigated. According to the International Organization for Standardization (ISO) standards for pulse oximeter performance evaluation, controlled desaturation studies are recommended across an SaO₂ range of 70–100%12. The U.S. Food and Drug Administration (FDA) further requires the inclusion of participants with dark skin pigmentation to ensure accuracy across ethnic groups13. Beyond skin pigmentation, the accuracy of pulse oximetry is influenced by a range of physiological and technical factors, including the measurement site (e.g., finger versus earlobe sensors), subject motion, sensor positioning, peripheral perfusion, vasomotor tone, and cold-induced vasoconstriction14. Controlled desaturation protocols are designed to minimize these confounders by employing standardized probe placement, stable participant posture, controlled ambient temperature, and continuous reference measurements from arterial blood gas analysis14. These controlled desaturation protocols provide a unique framework for examining endogenous biochemical responses to acute, short-term oxygen desaturation in healthy individuals. The present study aimed to determine which among COHb, MetHb, and lactate most accurately reflects the severity of acute arterial desaturation, induced under controlled, normocapnic conditions in healthy volunteers, and secondarily, to explore potential ethnic differences in their physiological responses. Methods Study design and setting This retrospective study analyzed pooled data from four controlled desaturation trials conducted at Asan Medical Center, Seoul, Korea. Each study was approved by the Institutional Review Board (2022 − 1657, 2023 − 0282, 2024 − 1200, and 2024 − 1311) and prospectively registered in the Clinical Research Information Service (CRIS; http://cris.nih.go.kr) prior to participant enrollment. The respective registration identifiers and dates were: KCT0008376 (February 21, 2023), KCT0008746 (August 28, 2023), KCT0009938 (November 22, 2024), and KCT0010039 (December 13, 2024). All trials were overseen by the same principal investigator (Byung-Moon Choi). All methods were performed in accordance with relevant guidelines and regulations, and the study was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants before enrollment. Volunteer population Participants were screened using a standardized questionnaire to confirm self-reported good health. No laboratory testing beyond arterial blood gas analysis was performed for eligibility assessment. The inclusion criteria were as follows: (1) age between 20 and 50 years; (2) fractional COHb in total hemoglobin (tHb) < 3%; (3) fractional MetHb in tHb < 2%; and (4) tHb concentration > 10 g/dL. Exclusion criteria included a known history of respiratory or cardiovascular disease, smoking, pregnancy, history of syncope, diabetes mellitus, dermatologic conditions potentially affecting SpO₂ measurements (e.g., onychomycosis), or a body mass index ≥ 35 kg/m². All participants were instructed to abstain from alcohol consumption for at least 3 days before each study day, and adherence was confirmed verbally on the morning of the study. Skin complexion was classified as light, medium, or dark according to the Fitzpatrick skin phototype scale15. Procedure for the controlled desaturation study The controlled desaturation procedure was performed in accordance with the ISO 80601-2-61:2017 standard12, which defines the recommended methodology for human arterial desaturation testing during pulse oximeter performance evaluation. Participants were positioned semi-supine (approximately 25° head-up) and breathed a hypoxic gas mixture of air and nitrogen via a closed breathing circuit. To prevent ambient air contamination, a nose clip was applied, and the gas mixture was delivered exclusively through a rubber mouthpiece. When necessary, carbon dioxide was added to the inspired gas mixture to maintain normocapnia and prevent respiratory alkalosis resulting from hyperventilation during severe arterial desaturation. Participants were instructed to maintain consistent tidal volumes and respiratory rates throughout the procedure. Each study comprised two rounds of induced arterial desaturation separated by a washout period of at least five minutes, during which participants removed the mouthpiece and nose clip and breathed room air. The detailed desaturation protocol has been described previously16,17. Measurements of COHb, MetHb, and Lactate A 20-gauge catheter was inserted into the radial artery for arterial blood sampling. An initial baseline sample was obtained while participants breathed room air to confirm eligibility according to the inclusion criteria. During controlled desaturation, arterial blood samples were collected while targeting stable SaO₂ levels between 70% and 100%, the analytical range specified by ISO 80601-2-61:2017 for pulse oximeter accuracy evaluation. Real-time guidance of sampling was based on pulse oximeter (SpO₂) readings, whereas the reference SaO₂ used in all analyses was measured retrospectively by CO-oximetry. Because the present study evaluates endogenous biomarker responses rather than pulse oximeter accuracy, no a priori exclusion was applied to samples whose retrospective CO-oximetry SaO₂ fell slightly below the 70% target. A clinically validated pulse oximeter (OxiMax® N-600x, Medtronic, Galway, Ireland) was used to monitor SaO₂ and guide sampling at each target saturation level. A total of 12 arterial samples were collected during each desaturation round, yielding 24 samples per participant. Sampling was distributed to achieve a balanced number of samples across the SaO₂ ranges of 70–80%, 80–90%, and 90–100%. SaO₂, PaO₂, COHb, MetHb, and lactate concentrations were measured using a CO-oximeter (ABL90 FLEX; Radiometer Medical A/S, Copenhagen, Denmark). All arterial samples were collected into dedicated arterial blood-gas syringes and analyzed immediately at the bedside on the CO-oximeter, minimizing pre-analytical changes to the measured biomarkers (in particular, in vitro lactate generation by blood cells). Arterial pH, PaCO₂, base excess of blood (BE(B)), bicarbonate (HCO₃⁻), and total hemoglobin (tHb) were also recorded from each sample. The device was inspected and calibrated by an authorized service provider within one week before each study session. Calculation of relative changes in biomarkers As SaO₂ levels decreased during arterial desaturation, endogenous biomarker concentrations tended to increase. To facilitate interpretation and enable direct comparison among biomarkers, all measurements were expressed as relative changes from baseline, defined as the value obtained while participants breathed room air before mixed-gas inhalation. For example, if the baseline COHb level was 0.8% and the corresponding value during arterial desaturation was 1.0%, the relative change was expressed as 125% (= 1.0 / 0.8 × 100). Temporal changes in biomarker levels were thus presented as relative changes from baseline. A biomarker was considered responsive to acute arterial desaturation if its level increased to ≥ 125% of the baseline value within the SaO₂ range of 70–80%. The responsiveness of COHb, MetHb, and lactate was evaluated across 96 desaturation rounds performed in 48 volunteers. The 125% responsiveness cutoff, together with the inter-range thresholds used for the sample-size calculation (a > 20% relative increase in median COHb from SaO₂ 90–100% to 80–90%, and a > 5% increase from 80 to 90% to 70–80%), were specified a priori, derived from the median COHb increases observed in pilot data from a previous controlled desaturation study17. These cutoffs are pragmatic operational definitions rather than thresholds derived from established physiological or clinical criteria. To assess whether the comparative ranking of biomarkers depends on this specific cutoff, a sensitivity analysis was conducted using alternative responsiveness thresholds ranging from ≥ 110% to ≥ 150%. Statistical analysis The primary endpoint was to identify the biomarker that most accurately reflects the severity of acute arterial desaturation across the SaO₂ range of 70–100%. SaO₂ values were stratified into three ranges: 90–100%, 80–90%, and 70–80%. A biomarker was considered significant if its relative value increased by > 20% as SaO₂ decreased from 90 to 100% to 80–90%, and by > 5% when SaO₂ decreased from 80 to 90% to 70–80%. Based on data from a prior controlled desaturation study17, we estimated that 45 participants would be required to meet these thresholds. As each controlled desaturation study conducted by our group typically includes 12 volunteers, data from four independent studies were combined to achieve the target sample size. The R code used for the sample size estimation is provided in the Supplementary Material 1. Data normality was assessed using the Shapiro–Wilk test. Normally distributed variables are reported as mean ± standard deviation (SD), and non-normally distributed variables as median (25th–75th percentile). Comparisons among the three endogenous biomarkers (COHb, MetHb, and lactate) were performed using one-way analysis of variance (ANOVA), followed by Dunn’s post hoc test for multiple comparisons. Within each biomarker group, changes across SaO₂ levels were analyzed using repeated-measures ANOVA with Holm–Šidák correction for multiple testing. Differences between ethnic groups were assessed using the Mann–Whitney rank-sum test. To further quantify the relationship between SaO₂ and each biomarker while accounting for repeated measurements within individuals, linear mixed-effects models were fitted with participant ID included as both a random intercept and a random slope for SaO2. The general model specification was: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Y_{ij}}={\beta _0}+{b_{0i}}+({\beta _1}+{b_{1i}})Sa{O_{2,ij}}+{\varepsilon _{ij}}$$\end{document} , where Yij means biomarker value for individual i at measurement j, β0 (fixed intercept) represents the expected biomarker value when SaO₂ is zero (a theoretical baseline used for model fitting rather than a physiologic state), b0i is a subject-specific random intercept accounting for within-individual correlation, β1 (fixed slope) represents the expected change in the biomarker associated with a 1% change in SaO₂, b1i represent a subject-specific random slope, capturing individual deviations from the population-level effects, and εij is the residual error. A statistically significant β1 indicates a statistically significant association between SaO₂ and the biomarker. Models were fitted using the lme function in the nlme package in R. To address possible carry-over of COHb between the two desaturation rounds, additional analyses were performed in which Round (1 vs. 2) was added as a fixed effect, the SaO₂ × Round interaction was tested by likelihood-ratio test, and a sensitivity analysis was restricted to Round 1 only. To assess the potential influence of acid-base shifts on the SaO₂–COHb association via Bohr and Haldane effects, the linear mixed-effects model for COHb was additionally fitted with pH, PaCO₂ and BE(B) included as covariates. A two-sided P-value of < 0.05 was considered statistically significant. All analyses were conducted using SigmaStat software (version 3.5 for Windows; Systat Software, Inc., Chicago, IL, USA) or R (version 4.5.2, R Foundation for Statistical Computing, Vienna, Austria). Results A total of 48 volunteers participated in four controlled desaturation studies, with 12 subjects enrolled in each, yielding a total of 96 controlled desaturation rounds. No participants withdrew during the study. Demographic characteristics are summarized in Table 1. A total of 1,194 arterial blood samples were obtained and analyzed. The actually achieved SaO₂ range across all samples was 68.0–99.2%; two samples (0.17%) had a CO-oximetry SaO₂ between 68% and 70%, marginally below the ISO 70% lower bound, and were retained for analysis. Individual SaO₂ and PaO₂ trajectories during controlled arterial desaturation are shown in Fig. 1. The median (25%–75%) duration of active hypoxic gas inhalation per round was 10.8 min (10.0–11.9 min). Figure 2 illustrates relative changes in endogenous biomarker levels throughout controlled arterial desaturation. COHb demonstrated a clear upward trend with increasing severity of arterial desaturation, whereas MetHb and lactate levels remained relatively unchanged. Comparison across SaO₂ ranges of 68–80%, 80–90%, and 90–100% revealed significant differences among the three biomarkers (Table 2). COHb exhibited greater relative changes than MetHb or lactate within equivalent SaO₂ ranges and more distinctly reflected the severity of arterial desaturation. Moreover, COHb’s responsiveness to acute arterial desaturation was statistically superior to that of the other two biomarkers. A sensitivity analysis using alternative responsiveness thresholds ranging from ≥ 110% to ≥ 150% yielded qualitatively identical conclusions, with COHb consistently exhibiting the highest proportion of responsive rounds at every threshold tested (Supplementary Material 2). Linear mixed-effects modeling showed that COHb had a significant negative association with SaO₂ (β₁ = − 0.0123, 95% CI − 0.0130 to − 0.0116, P < 0.001), indicating that COHb rose by an estimated 0.012% for each 1% reduction in SaO₂ (Fig. 3). To assess potential carry-over between the two desaturation rounds, Round (1 vs. 2) was added to the model as a fixed effect; the SaO₂ slope was essentially unchanged (β₁=−0.0119, P < 0.001) and the Round main effect was not statistically significant (P = 0.74). The SaO₂ × Round interaction was likewise not significant (likelihood-ratio P = 0.36), and a sensitivity analysis restricted to Round 1 yielded a virtually identical slope (β₁=−0.0117, P < 0.001). Together, these analyses indicate that any between-round carry-over of COHb did not materially bias the SaO₂–COHb slope estimate. Acid–base parameters and total hemoglobin across the three SaO₂ ranges are summarized in Table 3: deeper arterial desaturation was accompanied by a small but statistically significant respiratory alkalosis and a slight increase in BE, whereas HCO₃⁻ and tHb did not change. When pH, PaCO₂ and BE(B) were added simultaneously to the linear mixed-effects model for COHb as covariates, the SaO₂–COHb slope (β₁ = −0.0123) attenuated by only approximately 7% and remained highly statistically significant (P < 0.001), indicating that the SaO₂–COHb association is robust to adjustment for the measured acid–base parameters. In contrast, MetHb showed no significant association with SaO₂ (β₁ = 0.0003, 95% CI − 0.0011 to 0.0017, P = 0.677), and lactate likewise demonstrated no significant relationship (β₁ = − 0.0011, 95% CI − 0.0027 to 0.0005, P = 0.175). A subgroup analysis evaluating ethnic differences in COHb responses demonstrated that Korean volunteers exhibited significantly greater relative increases in COHb than African volunteers as the severity of arterial desaturation increased (Table 4). Baseline COHb measured during room-air breathing was, however, higher in African than in Korean participants (median [IQR]: 0.80 [0.70–0.90]% vs. 0.60 [0.60–0.80]%; Mann-Whitney P = 0.002), and baseline total hemoglobin (tHb) was lower in the African group (12.4 ± 1.6 vs. 14.7 ± 1.4 g/dL). Sex distribution also differed markedly between the two ethnic groups: among the 16 African participants, 15 were female and 1 was male, whereas among the 32 Korean participants, 10 were female and 22 were male (Fisher exact test P < 0.001). In a sex-adjusted linear mixed-effects model, the steeper SaO₂–COHb slope in Korean than in African participants persisted (SaO₂ × ethnicity interaction P < 0.001), although this strong design imbalance limits full statistical separation of sex and ethnicity effects. A formal SaO₂ × Sex interaction was also tested and was statistically significant (likelihood-ratio P = 0.007), with male participants showing a slightly steeper SaO₂–COHb slope than females; however, given the marked sex–ethnicity imbalance described above, this apparent sex effect cannot be cleanly disentangled from the ethnicity effect with the current dataset. Table 1Volunteer characteristics.All (N = 48)Study 1 (n = 12)Study 2 (n = 12)Study 3 (n = 12)Study 4 (n = 12)Age, yr27.0 (25.0‒29.5)26.0 (25.0‒30.0)28.0 (25.0‒30.0)26.5 (26.0‒29.5)26.5 (25.0‒29.5)Weight, kg69.5 (64.9‒72.9)69.6 (67.1‒72.6)68.6 (61.1‒71.8)70.3 (63.0‒77.5)70.6 (63.4‒72.1)Height, cm169.6 ± 9.8170.5 ± 10.2165.3 ± 10.8171.8 ± 9.3170.9 ± 8.5Male/Female23/256/65/76/66/6Sex by ethnicityKorean (M/F), n22/106/25/35/36/2African (M/F), n1/150/40/41/30/4African subgroup composition (Malawian / Angolan / Ugandan), n7/6/31/2/12/2/02/1/12/2/0ComplexionLight/medium/dark, n8/24/160/8/40/8/44/4/44/4/4EthnicityKorean/African, n32/168/48/48/48/4COHb, %0.7 (0.6‒0.8)0.7 (0.6‒0.8)0.8 (0.6‒0.9)0.6 (0.5‒0.8)0.8 (0.6‒0.8)MetHb, %0.9 (0.7‒0.9)0.7 (0.6‒0.9)0.9 (0.8‒1.0)0.9 (0.7‒1.1)0.9 (0.7‒1.1)Lactate, mmol/L 0.8 (0.5‒0.9)0.7 (0.6‒0.9)0.8 (0.6‒1.2)0.7 (0.6‒0.9)0.8 (0.6‒0.9)Data are presented as mean ± standard deviation, median (25–75%), or count, as appropriate. COHb: carboxyhemoglobin, MetHb: methemoglobin. Table 1Volunteer characteristics.All (N = 48)Study 1 (n = 12)Study 2 (n = 12)Study 3 (n = 12)Study 4 (n = 12)Age, yr27.0 (25.0‒29.5)26.0 (25.0‒30.0)28.0 (25.0‒30.0)26.5 (26.0‒29.5)26.5 (25.0‒29.5)Weight, kg69.5 (64.9‒72.9)69.6 (67.1‒72.6)68.6 (61.1‒71.8)70.3 (63.0‒77.5)70.6 (63.4‒72.1)Height, cm169.6 ± 9.8170.5 ± 10.2165.3 ± 10.8171.8 ± 9.3170.9 ± 8.5Male/Female23/256/65/76/66/6Sex by ethnicityKorean (M/F), n22/106/25/35/36/2African (M/F), n1/150/40/41/30/4African subgroup composition (Malawian / Angolan / Ugandan), n7/6/31/2/12/2/02/1/12/2/0ComplexionLight/medium/dark, n8/24/160/8/40/8/44/4/44/4/4EthnicityKorean/African, n32/168/48/48/48/4COHb, %0.7 (0.6‒0.8)0.7 (0.6‒0.8)0.8 (0.6‒0.9)0.6 (0.5‒0.8)0.8 (0.6‒0.8)MetHb, %0.9 (0.7‒0.9)0.7 (0.6‒0.9)0.9 (0.8‒1.0)0.9 (0.7‒1.1)0.9 (0.7‒1.1)Lactate, mmol/L 0.8 (0.5‒0.9)0.7 (0.6‒0.9)0.8 (0.6‒1.2)0.7 (0.6‒0.9)0.8 (0.6‒0.9)Data are presented as mean ± standard deviation, median (25–75%), or count, as appropriate. COHb: carboxyhemoglobin, MetHb: methemoglobin. Volunteer characteristics. Data are presented as mean ± standard deviation, median (25–75%), or count, as appropriate. COHb: carboxyhemoglobin, MetHb: methemoglobin. Table 2Absolute and relative carboxyhemoglobin (COHb), methemoglobin (MetHb), and lactate levels across different arterial oxygen saturation (SaO₂) ranges and their responsiveness to acute arterial desaturation.COHb (% / %)MetHb (% / %)Lactate (mmol/L / %)P-value90% ≤ SaO2 < 100% (n = 382)0.8 (0.7–0.9) /112.5 (100.0‒122.2)0.9 (0.7–1.0) /100.0† (100.0‒116.7)0.7 (0.5–0.9) /100.0† (80.0‒100.0)< 0.00180% ≤ SaO2 < 90% (n = 371)0.9 (0.8–1.0) /133.3* (125.0‒150.0)0.9 (0.8–1.0) /100.0† (100.0‒116.7)0.7 (0.5–0.9) /100.0† (80.0‒112.5)< 0.00168% ≤ SaO2 < 80% (n = 393)1.0 (0.9–1.1) /142.9* (133.3‒150.0)0.9 (0.8–1.0) /100.0† (91.7‒116.7)0.7 (0.5–0.9) /100.0† (77.1‒117.8)< 0.001Responsiveness to arterial desaturation90 (93.8)28‡ (29.2)24‡ (25)< 0.001Data are presented as median (25‒75%) or count (percentage), as appropriate. n: number of arterial blood samples. Relative values were calculated with reference to the measurements obtained under room-air breathing before the first arterial desaturation (baseline = 100%). Responsiveness to arterial desaturation was evaluated in 96 desaturation rounds from 48 volunteers, defined as an increase of ≥ 125% from the baseline value at an SaO₂ range of 70–80%. Baseline samples (N = 48) obtained while breathing room air were excluded from the analysis.* P < 0.001 vs. SaO₂ range of 90–100%, † P < 0.001 vs. relative COHb at each SaO₂ range, ‡ P < 0.001 vs. relative COHb for responsiveness to acute arterial desaturation. Table 2Absolute and relative carboxyhemoglobin (COHb), methemoglobin (MetHb), and lactate levels across different arterial oxygen saturation (SaO₂) ranges and their responsiveness to acute arterial desaturation.COHb (% / %)MetHb (% / %)Lactate (mmol/L / %)P-value90% ≤ SaO2 < 100% (n = 382)0.8 (0.7–0.9) /112.5 (100.0‒122.2)0.9 (0.7–1.0) /100.0† (100.0‒116.7)0.7 (0.5–0.9) /100.0† (80.0‒100.0)< 0.00180% ≤ SaO2 < 90% (n = 371)0.9 (0.8–1.0) /133.3* (125.0‒150.0)0.9 (0.8–1.0) /100.0† (100.0‒116.7)0.7 (0.5–0.9) /100.0† (80.0‒112.5)< 0.00168% ≤ SaO2 < 80% (n = 393)1.0 (0.9–1.1) /142.9* (133.3‒150.0)0.9 (0.8–1.0) /100.0† (91.7‒116.7)0.7 (0.5–0.9) /100.0† (77.1‒117.8)< 0.001Responsiveness to arterial desaturation90 (93.8)28‡ (29.2)24‡ (25)< 0.001Data are presented as median (25‒75%) or count (percentage), as appropriate. n: number of arterial blood samples. Relative values were calculated with reference to the measurements obtained under room-air breathing before the first arterial desaturation (baseline = 100%). Responsiveness to arterial desaturation was evaluated in 96 desaturation rounds from 48 volunteers, defined as an increase of ≥ 125% from the baseline value at an SaO₂ range of 70–80%. Baseline samples (N = 48) obtained while breathing room air were excluded from the analysis.* P < 0.001 vs. SaO₂ range of 90–100%, † P < 0.001 vs. relative COHb at each SaO₂ range, ‡ P < 0.001 vs. relative COHb for responsiveness to acute arterial desaturation. Absolute and relative carboxyhemoglobin (COHb), methemoglobin (MetHb), and lactate levels across different arterial oxygen saturation (SaO₂) ranges and their responsiveness to acute arterial desaturation. 0.8 (0.7–0.9) / 112.5 (100.0‒122.2) 0.9 (0.7–1.0) / 100.0† (100.0‒116.7) 0.7 (0.5–0.9) / 100.0† (80.0‒100.0) 0.9 (0.8–1.0) / 133.3* (125.0‒150.0) 0.9 (0.8–1.0) / 100.0† (100.0‒116.7) 0.7 (0.5–0.9) / 100.0† (80.0‒112.5) 1.0 (0.9–1.1) / 142.9* (133.3‒150.0) 0.9 (0.8–1.0) / 100.0† (91.7‒116.7) 0.7 (0.5–0.9) / 100.0† (77.1‒117.8) Data are presented as median (25‒75%) or count (percentage), as appropriate. n: number of arterial blood samples. Relative values were calculated with reference to the measurements obtained under room-air breathing before the first arterial desaturation (baseline = 100%). Responsiveness to arterial desaturation was evaluated in 96 desaturation rounds from 48 volunteers, defined as an increase of ≥ 125% from the baseline value at an SaO₂ range of 70–80%. Baseline samples (N = 48) obtained while breathing room air were excluded from the analysis. * P < 0.001 vs. SaO₂ range of 90–100%, † P < 0.001 vs. relative COHb at each SaO₂ range, ‡ P < 0.001 vs. relative COHb for responsiveness to acute arterial desaturation. Table 3Arterial acid–base parameters and total hemoglobin across SaO₂ ranges.90% ≤ SaO2 < 100%(n = 382)80% ≤ SaO2 < 90%(n = 371)68% ≤ SaO2 < 80%(n = 393)P-valuepH7.371 ± 0.0227.390 ± 0.0257.399 ± 0.024< 0.001PaCO₂, mmHg45.8 ± 3.243.7 ± 3.543.0 ± 3.1< 0.001BE(B), mmol/L1.19 ± 1.341.36 ± 1.371.65 ± 1.29< 0.001HCO₃⁻, mmol/L26.5 ± 1.326.3 ± 1.326.5 ± 1.20.124tHb, g/dL13.6 ± 1.713.6 ± 1.713.6 ± 1.80.933Data are mean ± SD. P-values are from Kruskal–Wallis test across the three SaO₂ ranges. BE(B): base excess of blood; HCO₃⁻: bicarbonate; tHb: total hemoglobin. Table 3Arterial acid–base parameters and total hemoglobin across SaO₂ ranges.90% ≤ SaO2 < 100%(n = 382)80% ≤ SaO2 < 90%(n = 371)68% ≤ SaO2 < 80%(n = 393)P-valuepH7.371 ± 0.0227.390 ± 0.0257.399 ± 0.024< 0.001PaCO₂, mmHg45.8 ± 3.243.7 ± 3.543.0 ± 3.1< 0.001BE(B), mmol/L1.19 ± 1.341.36 ± 1.371.65 ± 1.29< 0.001HCO₃⁻, mmol/L26.5 ± 1.326.3 ± 1.326.5 ± 1.20.124tHb, g/dL13.6 ± 1.713.6 ± 1.713.6 ± 1.80.933Data are mean ± SD. P-values are from Kruskal–Wallis test across the three SaO₂ ranges. BE(B): base excess of blood; HCO₃⁻: bicarbonate; tHb: total hemoglobin. Arterial acid–base parameters and total hemoglobin across SaO₂ ranges. Data are mean ± SD. P-values are from Kruskal–Wallis test across the three SaO₂ ranges. BE(B): base excess of blood; HCO₃⁻: bicarbonate; tHb: total hemoglobin. Table 4Comparison of relative carboxyhemoglobin (COHb) responses to acute arterial desaturation between Korean and African volunteers.Korean (np = 32)African (np = 16)P-value90% ≤ SaO2 < 100% (ns=261/ns=121)111.1 (100.0‒120.0)112.5 (110.0‒122.9)0.02280% ≤ SaO2 < 90% (ns=245/ns=126)133.3* (125.0‒150.0)128.6* (125.0‒137.5)0.00268% ≤ SaO2 < 80% (ns=262/ns=131)150.0* (133.3‒166.7)137.5* (125.0‒144.4)< 0.001Data are presented as median (25‒75%). SaO₂: arterial oxygen saturation, np: number of participants, ns: number of arterial blood samples. Relative values were calculated with reference to the measurements obtained under room-air breathing prior to the first arterial desaturation (baseline = 100%). Baseline samples (ns = 48) obtained while breathing room air were excluded from the analysis.* P < 0.001 vs. SaO₂ range of 90–100% within the same ethnic group. Table 4Comparison of relative carboxyhemoglobin (COHb) responses to acute arterial desaturation between Korean and African volunteers.Korean (np = 32)African (np = 16)P-value90% ≤ SaO2 < 100% (ns=261/ns=121)111.1 (100.0‒120.0)112.5 (110.0‒122.9)0.02280% ≤ SaO2 < 90% (ns=245/ns=126)133.3* (125.0‒150.0)128.6* (125.0‒137.5)0.00268% ≤ SaO2 < 80% (ns=262/ns=131)150.0* (133.3‒166.7)137.5* (125.0‒144.4)< 0.001Data are presented as median (25‒75%). SaO₂: arterial oxygen saturation, np: number of participants, ns: number of arterial blood samples. Relative values were calculated with reference to the measurements obtained under room-air breathing prior to the first arterial desaturation (baseline = 100%). Baseline samples (ns = 48) obtained while breathing room air were excluded from the analysis.* P < 0.001 vs. SaO₂ range of 90–100% within the same ethnic group. Comparison of relative carboxyhemoglobin (COHb) responses to acute arterial desaturation between Korean and African volunteers. Data are presented as median (25‒75%). SaO₂: arterial oxygen saturation, np: number of participants, ns: number of arterial blood samples. Relative values were calculated with reference to the measurements obtained under room-air breathing prior to the first arterial desaturation (baseline = 100%). Baseline samples (ns = 48) obtained while breathing room air were excluded from the analysis. * P < 0.001 vs. SaO₂ range of 90–100% within the same ethnic group. Fig. 1Changes in arterial oxygenation during two periods of controlled arterial desaturation separated by a washout interval. (A) Time course of arterial oxygen saturation (SaO₂) and (B) arterial oxygen tension (PaO₂) during repeated periods of controlled arterial desaturation. Each desaturation period was separated by a washout interval of at least 5 min, during which peripheral oxygen saturation measured by the reference pulse oximeter (SpO₂) was verified to have returned to 100% and remained stable for more than 1 min before the second desaturation period. No additional arterial sampling was performed during the washout interval. Changes in arterial oxygenation during two periods of controlled arterial desaturation separated by a washout interval. (A) Time course of arterial oxygen saturation (SaO₂) and (B) arterial oxygen tension (PaO₂) during repeated periods of controlled arterial desaturation. Each desaturation period was separated by a washout interval of at least 5 min, during which peripheral oxygen saturation measured by the reference pulse oximeter (SpO₂) was verified to have returned to 100% and remained stable for more than 1 min before the second desaturation period. No additional arterial sampling was performed during the washout interval. Fig. 2Relative changes in COHb, MetHb, and lactate levels during two periods of controlled arterial desaturation separated by a washout interval. (A) Carboxyhemoglobin (COHb), (B) methemoglobin (MetHb), and (C) lactate levels during repeated periods of controlled arterial desaturation. Each desaturation period was separated by a washout interval of at least 5 min, during which peripheral oxygen saturation measured by the reference pulse oximeter (SpO₂) was confirmed to have returned to 100% and remained stable for more than 1 min before the second desaturation period. Relative values were calculated with reference to baseline measurements obtained during room-air breathing before the first desaturation period (baseline = 100%). Relative changes in COHb, MetHb, and lactate levels during two periods of controlled arterial desaturation separated by a washout interval. (A) Carboxyhemoglobin (COHb), (B) methemoglobin (MetHb), and (C) lactate levels during repeated periods of controlled arterial desaturation. Each desaturation period was separated by a washout interval of at least 5 min, during which peripheral oxygen saturation measured by the reference pulse oximeter (SpO₂) was confirmed to have returned to 100% and remained stable for more than 1 min before the second desaturation period. Relative values were calculated with reference to baseline measurements obtained during room-air breathing before the first desaturation period (baseline = 100%). Fig. 3Relationships between SaO₂ and COHb, MetHb, and lactate concentrations. (A) COHb, (B) MetHb, and (C) lactate concentrations plotted against SaO₂. The red line represents the population-level fitted line based on the fixed effects from the linear mixed-effects model. For COHb, the estimated fixed effects were a fixed intercept (β₀) of 1.94 and a fixed slope (β₁) of − 0.0123. For MetHb, β₀ = 0.87 and β₁ = 0.0003. For lactate, β₀ = 0.86 and β₁ = − 0.0011. Relationships between SaO₂ and COHb, MetHb, and lactate concentrations. (A) COHb, (B) MetHb, and (C) lactate concentrations plotted against SaO₂. The red line represents the population-level fitted line based on the fixed effects from the linear mixed-effects model. For COHb, the estimated fixed effects were a fixed intercept (β₀) of 1.94 and a fixed slope (β₁) of − 0.0123. For MetHb, β₀ = 0.87 and β₁ = 0.0003. For lactate, β₀ = 0.86 and β₁ = − 0.0011. Discussion In this exploratory study, we examined dynamic changes in endogenous biomarkers during controlled acute arterial desaturation in healthy volunteers. Among COHb, MetHb, and lactate, COHb demonstrated the most pronounced and consistent increase as arterial oxygen saturation declined, indicating superior responsiveness to acute arterial hypoxemia. Furthermore, the relative increase in COHb was significantly greater in Korean than in African participants under comparable desaturation conditions. The precise mechanisms underlying the rapid rise in COHb during brief arterial desaturation remain uncertain, but several biologically plausible explanations exist. Under hypoxic conditions, hypoxia-inducible factor-1α (HIF-1α) is stabilized, enhancing the transcriptional activity of the HIF-1 complex and upregulating target genes such as heme oxygenase-1 (HO-1)18. HO catalyzes the rate-limiting conversion of heme to biliverdin, free iron, and carbon monoxide (CO)19. Increased HO-1 mRNA expression and CO production have been observed in hypoxic rat cardiomyocytes20. However, although HO-1 mRNA induction can occur within tens of minutes18 , functional HO-1 protein accumulation and enzymatic activity typically require several hours, making it unlikely to explain the rapid COHb elevation observed after only brief hypoxic exposure in this study. In contrast, heme oxygenase-2 (HO-2), a constitutively expressed isoform abundant in the brain, vascular endothelium, and carotid body, acts as an oxygen-dependent enzyme and O₂ sensor21 . HO-2 contributes to the acute regulation of endogenous CO signaling in response to fluctuating oxygen tensions22. While HO-1 may mediate longer-term adaptations, evidence suggests that HO-2 may drive immediate CO generation in the brain and cerebral vasculature during acute hypoxia23. Most endogenous CO is bound to hemoglobin, with approximately 10–15% distributed in extravascular compartments, primarily associated with myoglobin and other heme proteins24. Physiological studies indicate that CO readily equilibrates between tissue and blood compartments25, implying that transient changes in oxygen tension could influence this distribution. Although direct experimental confirmation is lacking, it is conceivable that acute hypoxia induces redistribution of endogenous CO toward hemoglobin binding sites, resulting in a transient rise in circulating COHb. This hypothesis offers a coherent physiological framework linking acute hypoxia, HO-2 activity, and CO dynamics, but further studies directly quantifying CO fluxes between tissue and blood are warranted to substantiate this mechanism. Beyond this redistribution hypothesis, several alternative explanations should also be considered. First, the analytical performance of the ABL90 FLEX co-oximeter at very low COHb concentrations is finite; according to the manufacturer’s specifications, the within-run repeatability (S₀) is approximately 0.07% and the total reproducibility (Sx) is 0.23–0.26% absolute COHb in the low concentration range, with a coefficient of variation of about 5% near 5% FCOHb. Although the SaO₂–COHb slope estimate (β₁ = −0.0123) is derived from 1,194 paired observations and is therefore well above the analytical noise floor on a population basis, the absolute change in COHb across the desaturation range (~ 0.2–0.4%) is itself small in clinical terms. Second, acute hypoxia is accompanied by physiological changes in ventilation and perfusion that may alter CO handling. Hyperventilation lowers PaCO₂ — confirmed in our cohort, where overall PaCO₂ decreased from 45.8 ± 3.2 mmHg at SaO₂ 90–100% to 43.0 ± 3.1 mmHg at SaO₂ 68–80% (P < 0.001) — and the resulting mild respiratory alkalosis (pH 7.371 to 7.399, P < 0.001) shifts the oxyhemoglobin dissociation curve. At the pulmonary level, hypoxic pulmonary vasoconstriction redistributes blood flow, while systemic hypoxia typically increases cardiac output and redirects regional perfusion toward vital organs. These ventilation–perfusion responses could plausibly modify the rate at which endogenous CO produced in peripheral tissues equilibrates with circulating hemoglobin. Third, deoxygenation of hemoglobin itself alters its conformational state and ligand-binding properties through the Bohr and Haldane effects26,27, which could shift the equilibrium between tissue-bound and hemoglobin-bound endogenous CO independently of any change in net CO production. Distinguishing the relative contribution of these mechanisms would require direct quantification of tissue and blood CO fluxes, which was beyond the scope of the present study. MetHb is a form of hemoglobin in which the iron within the heme group is oxidized from the ferrous (Fe²⁺) to the ferric (Fe³⁺) state28. Elevated MetHb levels are usually acquired, occurring when exposure to certain drugs or toxic agents accelerates the oxidation of hemoglobin28. Under normal physiological conditions, MetHb is continuously reduced back to functional hemoglobin by NADH–cytochrome b₅ reductase7. Because acute hypoxia exerts only a limited effect on hemoglobin oxidation, a rapid rise in MetHb is unlikely; therefore, MetHb may have limited utility as a biomarker reflecting short-term arterial desaturation. The lack of a clear lactate response to brief arterial desaturation is likely multifactorial. In particular, the absence of an increase in blood lactate is consistent with preserved systemic compensation and indicates that the present protocol induced arterial hypoxemia rather than overt tissue hypoxia. First, blood lactate accumulation is itself a delayed indicator of tissue metabolism, reflecting the temporal lag between intracellular lactate production and its appearance in the systemic circulation; this lag is likely longer than the ~ 10-minute hypoxic exposure used in the present study. Second, compensatory cardiovascular responses — including increases in heart rate and cardiac output — help preserve tissue oxygen delivery, thereby preventing a significant metabolic shift toward anaerobic glycolysis29. In addition, lactate functions not only as a byproduct of anaerobic metabolism but also as a key metabolic intermediate involved in diverse physiological pathways30, which may further account for its stable levels during the brief hypoxic exposure observed in this study. While the effect of skin pigmentation on pulse oximetry accuracy is well established31, ethnic differences in physiological responses to hypoxia have been less thoroughly explored. In a classic comparative study of ventilatory responses to hypoxia among four racial groups—Caucasian, Black, Asian, and Mestizo—the Asian group demonstrated a lower responsiveness than the Black group32. Moreover, polymorphisms in the hypoxia-inducible factor-1α (HIF-1α) gene have been reported to be associated with ethnicity33, suggesting potential genetic variability in hypoxia sensing or adaptation. Although the mechanism underlying the greater COHb increase in Asian compared with African participants remains unclear, our findings support the existence of ethnic variability in physiological responses to acute arterial desaturation. In a previous modeling study using a turnover model to characterize the relationship between PaO₂ and SaO₂, ethnicity was identified as a significant covariate influencing the fractional turnover rate constant (kout)17. Simulations from that model indicated that, at PaO₂ levels of 25–40 mmHg, Africans exhibited slightly higher SaO₂ values than Koreans—results consistent with our present findings and reinforcing the hypothesis of ethnic variability in hypoxic response17. However, several caveats temper this interpretation. First, baseline COHb was significantly higher in African than in Korean participants and baseline tHb was lower in the African group, indicating intrinsic hematological differences between subgroups that may have shaped the relative-change response even though it was referenced to each individual’s own baseline. Second, the present sample (32 Korean, 16 African) was modest and was not designed a priori for inter-ethnic comparison. Third, sex distribution was strongly imbalanced between the two groups (15 of 16 African participants were female, whereas only 10 of 32 Korean participants were female), which prevented complete decoupling of sex and ethnicity effects in the statistical model. Additional unmeasured factors—including genetic variability affecting heme metabolism, differences in ventilatory chemosensitivity to hypoxia, dietary or smoking history, and environmental CO exposure—may also contribute. Accordingly, the observed inter-ethnic difference in COHb response should be regarded as exploratory and hypothesis-generating rather than as definitive evidence of biological variability in physiological adaptation to acute hypoxia. Several limitations of this study should be acknowledged. First, we were unable to provide a definitive mechanistic explanation for why COHb exhibited the most sensitive response to short-term acute arterial desaturation. Although we proposed a hypothesis involving the redistribution of endogenous CO between tissue and blood compartments, this remains speculative and unverified experimentally. While the limited responsiveness of MetHb and lactate could be reasonably attributed to their physiological mechanisms, the precise basis for the COHb response cannot yet be fully explained by existing evidence. Second, the observed ethnic differences were restricted to comparisons between Korean and African participants. The inclusion of individuals with dark skin pigmentation followed the U.S. FDA’s requirement for clinical accuracy testing of pulse oximeters, which mandates the participation of individuals with dark complexions13. However, participants from other ethnic backgrounds, such as Caucasians, were not included in this study; thus, potential variations in COHb responsiveness across a broader spectrum of ethnicities could not be evaluated. A further limitation concerns the linear mixed-effects model used as the primary analysis. The fixed intercept β₀ is parameterized at SaO₂ = 0% and therefore serves only as a mathematical anchor of the regression line rather than as a physiologically meaningful baseline. The linearity assumption was examined post hoc using continuous piecewise mixed-effects regression (knots at SaO₂ = 80% and 90%) and a generalized additive model with smoothing splines, and residual diagnostics (residual versus fitted values, normal quantile–quantile plot, and residual versus SaO₂ plot) (Supplementary Material 3); these analyses indicated some departure from strict linearity, with the steepest slope above SaO₂ = 90% and a flatter, near-saturating profile below 80%, while the residual diagnostics revealed no other major deviations from the model’s distributional assumptions. Nevertheless, the direction and statistical significance of the SaO₂–COHb association were unchanged, indicating that the linear approximation provides a reasonable population-averaged summary across the studied range. A final limitation concerns the external validity of the present findings. Healthy young volunteers undergoing brief (~ 10 min), normocapnic, controlled arterial desaturation do not represent the broader spectrum of clinical hypoxia, which often coexists with systemic inflammation, impaired tissue perfusion, metabolic acidosis, organ dysfunction, or oxygen deprivation lasting hours to days. Such pathological conditions are likely to engage additional and slower mechanisms — including transcriptional induction of HO-1, sustained anaerobic glycolysis, and lactate accumulation — that were neither expected nor observed within our 10-minute desaturation window. The present findings should therefore be interpreted as describing the endogenous biomarker response to short-term, controlled arterial desaturation in healthy volunteers, and direct extrapolation to critically ill patients with sustained or pathological hypoxia is not warranted. In conclusion, among the endogenous biomarkers assessed, COHb showed the strongest association with the degree of brief, controlled arterial desaturation in healthy volunteers. Because the mechanistic basis for this rapid response remains uncertain — possibly involving analytical variability of co-oximetry at low concentrations, deoxygenation-related shifts in CO partitioning, and ventilation–perfusion changes during hypoxia — these findings should be regarded as hypothesis-generating rather than as evidence that COHb is a validated clinical biomarker of tissue hypoxia. The greater COHb rise observed in Korean compared with African participants further suggests, with the caveats noted above, that ethnic variability may influence endogenous CO dynamics during arterial desaturation, although confirmation in larger, ethnically balanced cohorts is required. Together, these results provide a basis for future studies aimed at clarifying the biochemical pathways and assessing whether the observed COHb response carries clinical significance. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3