High-altitude pulmonary edema (HAPE) is defined as a form of noncardiogenic pulmonary edema that occurs in non-acclimatized healthy individuals who rapidly ascend to high altitude above 2500 m1. The reported incidence of HAPE ranges from an estimated 0.01% of skiers traveling from the plains to altitudes of 2500 m, to 15.5% of soldiers rapidly transported to altitudes of 3355 and 5940 m2,3. HAPE is the most common cause of death related to high altitude exposure. If left untreated, it can lead to a mortality rate of up to 50%4. The development of HAPE is initiated by the increased pressure in the pulmonary capillaries during prolonged altitude exposure. The increased pressure causes capillary stress failure and fluid leakage to the lung interstitia and alveoli, which contributes to edema formation5. Early symptoms include dyspnea with exercise, dry cough and loss of performance. Advanced cases can progress to dyspnea at rest, cyanosis, rales and pink frothy sputum4. Clinical diagnosis of HAPE relies on symptoms or complaints of patient, exam findings, electrocardiogram, X-ray radiograph and computed tomography (CT) of the lungs2. However, early diagnosis of HAPE is difficult due to the increased heterogeneity in clinical symptoms and lack of clinical experience4. For this reason, looking for characteristic diagnostic laboratory findings would enable the development of HAPE diagnosis. Although many genes, metabolites and pathways partly contribute to HAPE pathogenesis6, there are no routine laboratory examinations that can be performed to determine HAPE susceptibility. Therefore, it is of great clinical value to explore the pathogenesis of HAPE and find potential biomarkers in predicting HAPE.

Metabolomics is a technology to systematically analyze all metabolites in a biological system, which could provide insights into dysregulated metabolic pathways linked to disease processes7. Unique metabolites identified by metabolomics could monitor the pathology stage and become potential indicators in the discrimination or diagnosis of various diseases8,9. Some studies based on metabolomics approaches have been attempted in HAPE, including clinical and animal studies10,11,12. High-resolution proton nuclear magnetic resonance (1H NMR)-based metabolomics was first applied to analyze altered metabolic patterns of blood plasma from HAPE patients10. In addition, another study used ultra-performance liquid chromatography-mass spectrometry (UPLC-MS) to investigate plasma metabolite profiling from HAPE patients, which provided valuable reference indices for HAPE11. Recently, a rat model of HAPE was established to clarify the metabolic changes and preliminarily screen biomarkers with metabolomics12. Therefore, applying metabolomic approaches may not only identify biomarkers for HAPE susceptibility, but also provide further insight into the physiologic pathways affecting HAPE.

As commonly employed technologies for non-targeted metabolomics, gas chromatography-mass spectrometry (GC-MS) and LC-MS can be combined to acquire a more comprehensive analysis of the metabolic data due to their complementary nature13. In this study, serum samples from HAPE patients and healthy controls were collected, and metabolomic methods based on GC-MS and LC-MS were utilized to identify the different metabolite profiles. We aimed to identify the aberrant metabolic pathways in HAPE and seek potential adjuvant biomarkers that could improve the diagnosis of and monitoring ability for HAPE.

Materials and methods

Study design

Schematic diagram of the study design was illustrated in Fig. 1. First, serum samples from HAPE patients and healthy controls were collected. Second, the differential metabolites between the two groups were identified based on GC-MS and LC-MS analysis. Then, the dysregulated metabolic pathways in HAPE were explored. In addition, the correlation analysis and receiver operating characteristic (ROC) curve analysis were further performed to screen biomarkers for HAPE.

Fig. 1
figure 1

Schematic diagram of the study design. Serum samples from HAPE patients (n = 24) and healthy controls (n = 21) were analyzed by GC-MS and LC-MS to profile differential metabolites and explore dysregulated metabolic pathways. The correlation analysis and ROC curve analysis were further performed to screen biomarkers for HAPE.

Full size image

Study participants

The study protocol was approved by the Ethics Committee of The 940 Hospital of Joint Logistics Support Force of Chinese PLA (registration number: 2021KYLL152) and all the participants have provided written informed consent. All procedures involving the human subjects were carried out in accordance with the recommendations of the Helsinki Declaration. A total of 24 patients with HAPE, who developed the disease after travelling from the lowlands to Ngari Prefecture, Tibet (4300 m), were enrolled from People’s Hospital of Rutog County between August 2021 and March 2022. The HAPE patients enrolled in this study did not receive any medication therapy before disease assessment and sample collection. The diagnosis of HAPE was based on standard criteria, including cough, dyspnea, cyanosis at rest, absence of infection, and the presence of pulmonary rales2. In addition, HAPE was confirmed by chest radiographic findings of infiltrates consistent with pulmonary edema. A total of 21 healthy controls participated in this study with the same method of ascent. Both groups were male, aged between 20 and 30 years, and travelled to Ngari Prefecture by aeroplane. All the participants were clinically evaluated by medical workers, and clinical indices, including temperature, heart rate (HR), respiratory rate (RR), blood pressure (BP) and pulse oxygen saturation (SpO2), were recorded. Besides, participants completed a questionnaire for acute mountain sickness (AMS) evaluation according to Lake Louise Score (LLS) from the four rated symptoms: headache, gastrointestinal symptoms, fatigue and/or weakness and dizziness/light-headedness14.

Sample collection and preparation

Fasting venous blood from all the subjects was collected in vacuum negative-pressure blood collection tubes and coagulated for 30 min at room temperature. The serum was separated by centrifugation (3500 rpm, 10 min) and stored at -80 ℃ for subsequent metabolomics analysis.

Prior to GC-MS analysis, 450 µL of methanol-acetonitrile (v: v = 2:1; Thermo Fisher, MA, USA) including L-2-chlorophenylalanine (2 µg/mL; Hengchuang Biotechnology, Shanghai, China) was added to 150 µL of serum. After vigorous shaking for 1 min and incubation on ice for 10 min, the mixture was stored at -40℃ for 30 min and then centrifuged at 12,000 rpm for 10 min at 4 ℃. 150 µL of supernatant was evaporated to dryness and the dried sample was dissolved in 80 µL of methoxyamine hydrochloride in pyridine (15 mg/mL; Macklin, Shanghai, China) and incubated for 60 min at 37 ℃. The mixture was added 50 µL of BSTFA (TCI, Shanghai, China) with 1% TMCS, 20 µL of hexane (CNW, Germany) and 10 µL of internal standard (NU-CHEK, MN, USA), and then derivatized for 60 min at 70 ℃. Quality control (QC) samples pooled from all samples were prepared with the same procedure.

Before performing LC-MS analysis, 450 µL of methanol-acetonitrile (v: v = 2:1) including L-2-chlorophenylalanine (2 µg/mL) was also added to 150 µL of serum. The mixture was vortexed for 1 min and incubated on ice for 10 min, and then stored at -40 ℃ for 2 h. Following centrifugation (12000 rpm, 10 min, 4 ℃), 150 µL of supernatant was filtered through a 0.22 μm organic phase pinhole filter, and then transferred to vials for further analysis. QC samples were prepared by mixing the extraction solutions of all samples in equal volumes.

GC-MS analysis

GC- MS analysis was performed according to a previous publication with modifications15. The prepared samples were analyzed by an Agilent 7890B/5977A GC-MS system (Agilent Technologies Inc., CA, USA). The separation of derivatives was achieved using a DB-5MS fused-silica capillary column (30 m × 0.25 mm × 0.25 μm; Agilent J & W Scientific, CA, USA). Helium (> 99.999%) was used as a carrier gas at a constant flow rate of 1 mL/min through the column. The injector temperature was maintained at 260 ℃, and the injection volume was 1 µL. The initial oven temperature was held at 60 °C for 30 s, ramped to 210 °C at a rate of 8 °C/min, ramped to 270 °C at 15 °C/min, ramped to 305 °C at 20 °C/min, and finally held at 305 °C for 5 min. The temperature of MS quadrupole and ion source (electron impact) was set to 150 and 230 ℃, respectively. The collision energy was 70 eV. The data were collected in full scan mode (m/z 50–500).

LC-MS analysis

LC-MS analysis was modified from a previous publication16. ACQUITY UPLC I-Class system coupled with VION IMS quadrupole-time-of-flight (Q-TOF)-MS (Waters Corporation, MA, USA) was used to analyze the metabolic profiling in both electrospray ionization source (ESI) positive and ESI negative ion modes. The extracts were separated by an ACQUITY UPLC BEH C18 column (1.7 μm, 2.1 × 100 mm). Water and acetonitrile-methanol (v: v = 2:3), both containing 0.1% formic acid (Thermo Fisher, MA, USA) were used as mobile phases A and B, respectively. A linear gradient elution was performed with the following program: 0 min, 5% B; 4 min, 30% B; 8 min, 50% B; 10 min, 80% B; 14 min, 100% B held until 15 min; and 15.1 min, 5% B held until 16 min. The flow rate was 0.35 mL/min and the column temperature was 45 ℃. The injection volume was 3 µL. Data acquisition was performed in full scan mode (m/z 100–1200) combined with MSE mode, including two independent scans with different collision energies (CE) that can be alternatively acquired during the analysis. The MS parameters employed during the analysis were set as follows: a low-energy scan (CE 4 eV), and a high-energy scan (CE 20–45 eV) to fragment the ions. Argon (99.999%) was used as collision-induced dissociation gas. A scan time of 0.2 s was used, followed by an interscan delay of 0.02 s. The capillary voltage was set to 2.5 kV, and the cone voltage was maintained at 40 V. The source temperature was set to 115 ℃, and the desolvation gas temperature was maintained at 450 ℃. The desolvation gas flow rate was 900 L/h.

Data preprocessing

The raw GC-MS data were processed using MS-DIAL software, which performs peak detection, peak identification, MS2Dec deconvolution, characterization, peak alignment, wave filtering, and missing value interpolation. Metabolite characterization was achieved using LUG database, and a data matrix was generated as a result. In each sample, all peak signal intensities were segmented and normalized according to the internal standards with relative standard deviation (RSD) greater than 0.3 after screening. After the data was normalized, redundancy removal and peak merging were carried out to obtain the final data matrix.

The raw LC-MS data were processed by Progenesis QI V2.3 software (Nonlinear Dynamics, Newcastle, UK) for baseline filtering, peak identification, integral, retention time correction, peak alignment, and normalization. Compounds were identified based on precise mass-to-charge ratio (m/z), secondary fragments, and isotopic distribution via comprehensively using the Human Metabolome Database (HMDB), Metlin, Lipidmaps, and self-built databases. The extracted data were then further processed by removing any peaks with a missing value (ion intensity = 0) in more than 50% in groups, and replacing zero value by half of the minimum value. In addition, compounds with scores below 36 (out of 80) points were considered inaccurate and excluded. The final data matrix was generated by combining the positive and negative ion data.

Statistical analysis

The normalized data were imported into SIMCA software (version 14.1, Umea, Sweden) to carry out principal component analysis (PCA), partial least-squares discriminant analysis (PLS-DA), and orthogonal partial least-squares discriminant analysis (OPLS-DA). To prevent overfitting, 7-round cross-validation and 200 response permutation testing (RPT) were performed to evaluate the quality of the model. Variable importance of projection (VIP) values obtained from the OPLS-DA model were used to rank the overall contribution of each variable to group discrimination. A T test was used to verify the metabolites of difference between two groups were significant. Fold change (FC) was calculated as average normalized peak intensity ratio between two groups. The variables with VIP > 1, P < 0.05 and |log2FC| > 2 were identified as differential metabolites. Spearman rank correlation was used to analyze the correlation coefficient between differential metabolites and clinical indices. The ROC curve analysis was performed to evaluate the accuracy of biomarker screening.

SPSS 22.0 software (IBM, IL, USA) was used for data analysis. Continuous variables in accordance with a normal distribution were expressed as the mean ± standard deviation (SD) while the non-normally distributed variables were expressed as the median with interquartile ranges (IQR). Measurement data were tested for normal distribution and homogeneity of variance. A T test was used for normally distributed data, and a nonparametric test was used for non-normally distributed data. A P value of < 0.05 was considered statistically significant.

Metabolic pathway analysis

Metabolic pathway analysis of identified differential metabolites was performed by MetaboAnalyst (version 6.0, https://www.metaboanalyst.ca/). A P value of < 0.1 was applied to select important target pathways. Schematic overview of the metabolic pathways was drew based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database (https://www.kegg.jp/)17.

Results

Clinical characteristics of participants

The study included 24 HAPE patients and 21 healthy controls, and the clinical characteristics of the two groups were summarized in Table 1. There was no significant difference in age distribution, body mass index (BMI) and systolic blood pressure (SBP) between the control and HAPE groups. Patients with HAPE had higher temperature, HR, RR, diastolic blood pressure (DBP) and LLS but lower SpO2 than control subjects.

Table 1 Clinical characteristics of study participants.
Full size table

Metabolic profiling of HAPE and control samples

The serum samples from the two groups were analyzed with GC-MS and LC-MS analysis. PCA analysis was performed on experimental samples and QC samples to monitor the robustness of sample preparation and the stability of instrument analysis (Fig. 2A, B). There were no obvious outliers in the samples and QC samples were tightly clustered both in GC-MS and LC-MS analysis, indicating good stability of the detection condition. Additionally, obvious separation trends between HAPE patients and healthy controls were observed in this unsupervised analysis. PLS-DA analysis was performed on samples to determine the overall metabolic differences between the two groups (Fig. 2C, D). The PLS-DA models showed that the HAPE group could be distinguished from the control group with good fitness (R2Y = 0.916, Q2 = 0.785 in GC-MS; R2Y = 0.991, Q2 = 0.925 in LC-MS).

Fig. 2
figure 2

Score plots of PCA and PLS-DA for the control and HAPE groups in serum GC-MS and LC-MS. (A) PCA score plot of GC-MS. (B) PCA score plot of LC-MS. (C) PLS-DA score plot of GC-MS. (D) PLS-DA score plot of LC-MS.

Full size image

Identification of differential serum metabolites in HAPE

The OPLS-DA model was used to assist in the selection of differential metabolites. As is shown in Fig. 3A and B, there was a noticeable separation between the control and HAPE groups from the OPLS-DA models (R2Y = 0.916, Q2 = 0.834 in GC-MS; R2Y = 0.991, Q2 = 0.924 in LC-MS). The permutation tests of the OPLS-DA models indicated that the original models were reliable for differential metabolites screening (Fig. 3C, D). The VIP value obtained from the OPLS-DA models was applied to filter the significant differential metabolites. On the basis of the combination of VIP > 1, P < 0.05 and |log2FC| > 2, a total of 119 differential metabolites (31 metabolites from GC-MS and 88 metabolites from LC-MS) were identified as potential biomarkers, among which 95 metabolites were decreased and 24 were increased in HAPE metabolites (Table 2). These metabolites consisted of organic acids, lipids, nucleosides, benzenoids, phenylpropanoids and polyketides. Volcano plots of the metabolites achieved from GC-MS and LC-MS were produced to visually display the changed metabolites (Fig. 4A, C). In addition, heatmaps were constructed to effectively visualize the intensity levels of the differential metabolites among the two groups (Fig. 4B, D).

Fig. 3
figure 3

Score plots and Validation plots of OPLS-DA for the control and HAPE groups in serum GC-MS and LC-MS. (A) OPLS-DA score plot of GC-MS. (B) OPLS-DA score plot of LC-MS. (C) Validation plot of GC-MS. (D) Validation plot of LC-MS.

Full size image
Table 2 The differential serum metabolites identified by the combination of GC-MS and LC-MS.
Full size table
Fig. 4
figure 4

Volcano plots and heatmaps of serum differential metabolites between the control and HAPE groups obtained from GC-MS and LC-MS analysis. (A) Volcano plot of differential metabolites obtained from GC-MS. (B) Heatmap of differential metabolites obtained from GC-MS. (C) Volcano plot of differential metabolites obtained from LC-MS. (D) Heatmap of differential metabolites obtained from LC-MS.

Full size image

Metabolic pathway analysis

In order to better understand the metabolic changes related to HAPE, metabolic pathways of differential metabolites were analyzed based on network enrichment and topology analysis, which were focused on pyrimidine metabolism, citrate cycle, sulfur metabolism, phenylalanine metabolism and purine metabolism (Fig. 5A, Table S1). A schematic scheme of metabolic pathways based on KEGG database was presented to visualize the interaction between the differential metabolites (Fig. 5B). The metabolites related to pyrimidine metabolism (thymine, uracil and orotic acid), citrate cycle (cis-aconitic acid and succinic acid) and sulfur metabolism (adenosine phosphosulfate, phenol sulfate, 3-methoxyphenol sulfate, 3-acetylphenol sulfate, benzyl sulfate) were all decreased in HAPE patients. In phenylalanine metabolism, serylphenylalanine and phenylacetaldehyde were significantly decreased, while prolylphenylalanine was increased in the HAPE group. In regard to purine metabolism, 7-methylhypoxanthine, theophylline and its degradation product (1,3-dimethyluric acid) were markedly increased but adenosine monophosphate was reduced in patients with HAPE.

Fig. 5
figure 5

Metabolic pathway analysis of serum differential metabolites between the control and HAPE groups. (A) Metabolic pathway analysis was performed using MetaboAnalyst with differential metabolites. (B) Schematic scheme of the disturbed metabolic pathways associated with HAPE. Metabolites marked in red represent up-regulation and blue represent down-regulation in the HAPE group compared with the control group.

Full size image

Correlation analysis between the differential metabolites and clinical indices

To screen out differential metabolites directly related to clinical indices, a Spearman correlation analysis between 119 differential metabolites and 4 clinical indices (including SpO2, HR, RR and LLS) was conducted. From the correlation heatmap (Fig. 6), it can be seen that 39 differential metabolites have a high number of significant differences (P < 0.001) with these 4 clinical indices, and the 39 differential metabolites were identified as potential biomarkers related to HAPE. The detailed information of correlation analysis can be achieved from Table S2. Among the 39 metabolites, 4 metabolites were negatively correlated with SpO2 while positively correlated with HR, RR and LLS. And the correlation between the other 35 metabolites and clinical indices was opposite to the above results.

Fig. 6
figure 6

Correlation analysis of serum different metabolites and clinical indices (HR, RR, SpO2 and LLS). Boxes in heatmap marked in orange represent positive correlation and blue negative correlation. (***P < 0.001)

Full size image

Biomarker analysis

The ROC curve analysis was used to further investigate the value of the 39 potential biomarkers on the diagnosis of HAPE. The area under curve (AUC) was calculated and 7 metabolites with AUC ≥ 0.95 were summarized in Table 3. The ROC curves and level boxplots of these 7 metabolites were shown in Fig. 7. Among the 7 biomarkers, the levels of S-nitroso-N-acetylcysteine and threo-hydroxyaspartic acid were significantly increased in HAPE patients, while aminocaproic acid, emodin, 6-hydroxynicotinic acid, 3-acetylphenol sulfate and cis-aconitic acid were decreased. The AUC values of S-nitroso-N-acetylcysteine, aminocaproic acid, emodin, threo-hydroxyaspartic acid, 6-hydroxynicotinic acid, 3-acetylphenol sulfate and cis-aconitic acid were 0.982, 0.966, 0.966, 0.966, 0.962, 0.962 and 0.950, respectively. The sensitivity and specificity of S-nitroso-N-acetylcysteine, aminocaproic acid, emodin, threo-hydroxyaspartic acid, 6-hydroxynicotinic acid, 3-acetylphenol sulfate and cis-aconitic acid were 0.958 and 0.952, 0.917 and 0.952, 1.000 and 0.952, 0.917 and 0.952, 0.917 and 0.857, 0.917 and 0.905, and 0.958 and 0.952, respectively, which displayed high diagnostic accuracy for HAPE.

Table 3 Seven differential serum metabolites with AUC ≥ 0.95 between the control and HAPE groups.
Full size table
Fig. 7
figure 7

ROC curves and level boxplots of 7 differential serum metabolites with AUC ≥ 0.95 between the control and HAPE groups. (A) S-nitroso-N-acetylcysteine. (B) Aminocaproic acid. (C) Emodin. (D) Threo-hydroxyaspartic acid. (E) 6-Hydroxynicotinic acid. (F) 3-Acetylphenol sulfate. (G) cis-Aconitic acid.

Full size image

Discussion

In this study, the serum metabolites of the control and HAPE groups were compared and analyzed by using metabolomics with the combination of GC-MS and LC-MS. A total of 119 metabolites with significant differences were identified as potential biomarkers. Based on metabolic pathway analysis, these metabolites were found to be primarily involved in pyrimidine metabolism, citrate cycle, sulfur metabolism, phenylalanine metabolism, purine metabolism, fatty acid metabolism, tryptophan metabolism and histidine metabolism. Additionally, 7 biomarkers, specifically S-nitroso-N-acetylcysteine, aminocaproic acid, emodin, threo-hydroxyaspartic acid, 6-hydroxynicotinic acid, 3-acetylphenol sulfate and cis-aconitic acid, were obtained for the diagnosis of HAPE after correlation analysis and ROC analysis.

The involvement of differential metabolites in pyrimidine and purine metabolism supported that nucleotide metabolism was disturbed in HAPE patients. For pyrimidine metabolism, the levels of orotic acid, uracil and thymine in serum were lower in HAPE patients than healthy controls. Orotic acid, as an intermediate of pyrimidine de novo synthesis, is released from the mitochondrial dihydroorotate dehydrogenase (DHODH) for conversion to uridine monophosphate (UMP) by the cytoplasmic UMP synthase. UMP can be used to synthesize uracil and thymine18,19. DHODH relies on ubiquinone that is provided by the electron transport chain (ETC) complex III, thereby forming a functional link between mitochondrial ETC and pyrimidine biosynthesis20. Therefore, we hypothesized that the reduced levels of orotic acid, uracil and thymine might be related to decreased DHODH activity owing to oxygen deficiency when exposed to high altitude. It has been verified that hypoxia-induced deprivation of pyrimidine (deoxy)nucleosides might be causative in cell cycle arrest19. In addition, cell cycle arrest of lung alveolar epithelial cells can exacerbate lung injury through promoting inflammation and restricting tissue repair21,22. Consequently, it is reasonable to suppose that disturbed pyrimidine metabolism may provoke lung injury by inducing cell cycle arrest in HAPE patients. For purine metabolism, our results were consistent with a previous study that perturbed purine metabolism is implicated in HAPE11. Our results revealed that the levels of theophylline, 1,3-dimethyluric acid and 7-methylhypoxanthine were increased while adenosine monophosphate (AMP) was decreased in HAPE subjects relative to healthy controls. Theophylline is a xanthine derivative and widely used for the treatment of asthma and chronic obstructive pulmonary disease (COPD), which can be converted to 1,3-dimethyluric acid after hydroxylation23. Hypoxanthine is a metabolite involved in purine biosynthesis, the level of which is elevated in HAPE and COPD patients owing to tissue hypoxia11,24. Elevated hypoxanthine concentrations may result in the increase of theophylline, 1,3-dimethyluric acid and 7-methylhypoxanthine. However, the level of AMP was reduced in our study, which seemed different from previous works25,26. We hypothesized that decreased AMP might be related to reduced generation of adenosine under hypoxic condition. And the decrease of adenosine and AMP may result in pulmonary vascular endothelial dysfunction and increase vascular resistance, prompting pulmonary hypertension (PH)27. It has been reported that perturbed purine metabolism is associated with HAPE and hypoxia-induced PH11,27,28. Therefore, targeting the altered purine metabolism (including critical metabolites and enzymes) may provide new perspectives for pathophysiology and treatment of HAPE.

Citrate cycle, also known as the tricarboxylic acid (TCA) cycle or Krebs cycle, is one of the central metabolic pathways responsible for the production of energy. The levels of metabolites related to citrate cycle (including cis-aconitic acid and succinic acid) were decreased in HAPE subjects, suggesting a reduced flux of citrate cycle in HAPE. It can be hypothesized that the lung adapted to favour aerobic glycolysis rather than oxidative phosphorylation to reduce the local oxygen consumption under hypoxia condition29. However, some previous studies indicated that hypoxia could cause the accumulation of intermediates of citrate cycle30,31, which was inconsistent with our results. Thus, the metabolic changes of citrate cycle in HAPE need further investigation. The results of this study showed that cis-aconitic acid obtained a high AUC with good sensitivity and specificity, suggesting a good diagnostic effect. Under normal condition, cis-aconitic acid is converted to isocitrate by aconitase in citrate cycle. However, cis-aconitic acid can be converted to itaconate and this reaction is catalyzed by the enzyme cis-aconitate decarboxylase (CAD) when the citrate cycle is bypassed during energy production32. Itaconate can act as a regulator of inflammation and immunomodulatory functions, and recent studies have highlighted the potential of targeting itaconate as a therapeutic strategy for lung diseases33,34. Therefore, the role of itaconate in HAPE may provide a novel perspective for therapeutic interventions, which needs to be investigated further. Succinic acid plays diverse and important roles as an immunometabolite in innate immune response. Nevertheless, it remains controversial whether succinic acid can be considered as a harmful signal or a protective molecule in the inflammatory response. Some previous studies have reported that succinic acid induced inflammation through activation of succinate receptor 1 (SUCNR1), thereby worsening lung injury and accelerating the development of acute respiratory distress syndrome (ARDS)35,36. However, succinic acid can also act as a protective molecule under different inflammatory conditions37. Thus, whether succinic acid acts as a pro-inflammatory driver or an anti-inflammatory signal in HAPE should be explored in further studies. Fatty acid metabolism is another critical energy source and can interact with citrate cycle. This study found that the levels of metabolites associated with fatty acid metabolism were down-regulated in HAPE. It can be speculated that the reduced flux of citrate cycle leads to decreased generation of acetyl CoA and reducing intermediates, including nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FADH2), which are essential for fatty acid synthesis38. In addition, previous studies suggested that fatty acid metabolism affects multiple cellular pathways during PH29. However, a reported LC-MS study suggested that a significantly higher levels of fatty acid amides were observed in HAPE subjects relative to healthy controls11, which was inconsistent with our results. Therefore, the role of fatty acid metabolism in the development of HAPE requires further research in the future.

Hydrogen sulfide (H2S) is the most important metabolite of sulfur metabolism. It has been demonstrated that the level of H2S is decreased in response to hypoxia at high altitude, which may be used as a biomarker for assessing hypoxia exposure in people that ascend to high altitude39,40. In our study, the levels of metabolites related to sulfur metabolism (including adenosine phosphosulfate, phenol sulfate, 3-methoxyphenol sulfate, 3-acetylphenol sulfate, benzyl sulfate) were all decreased in HAPE patients as compared to controls. We hypothesized that changes in these metabolites might be associated with the reduced H2S level since H2S can be degraded to sulfate, which is further converted to these differential metabolites41. In addition, our results suggested that 3-acetylphenol sulfate could be a good biomarker for diagnosis of HAPE, which needs further experimental validation.

For amino acid metabolism, our study suggested that disturbed phenylalanine metabolism, tryptophan metabolism and histidine metabolism were implicated in HAPE. While previous metabolomics studies focused on other amino acid metabolism pathways, such as glutamate metabolism, tyrosine metabolism and valine, leucine and isoleucine biosynthesis11,12. In a rat model of pulmonary edema, the serum level of phenylalanine was significantly increased42. It can be explained that inflammation decreases the activity of phenylalanine hydroxylase and then impair the conversion of phenylalanine to tyrosine, resulting in phenylalanine accumulation43. Thus, changed levels of the downstream metabolites of phenylalanine in our study highlight dysbolism of phenylalanine in HAPE. Tryptophan is primarily metabolized through three main pathways: serotonin (5-hydroxytryptamine) pathway, kynurenine pathway and microbiome (indole pyruvic acid) pathway44. In our study, the levels of melatonin radical and indole-3-acetamide were increased while 2,5-pyridinedicarboxylic acid was reduced in HAPE patients. The results suggested that serotonin pathway and microbiome pathway are strengthened but kynurenine pathway is weakened in HAPE. Previous studies demonstrated that melatonin could attenuate pulmonary edema by preventing alveolar-capillary barrier dysfunctions via inhibiting inflammation and apoptosis45,46. This shows a kind of possibility that the increased melatonin radical level plays a protective role in HAPE. In addition, another endogenous metabolite of tryptophan, 5-methoxytryptophan, was significantly decreased in HAPE patients. Interestingly, 5-methoxytryptophan is a circulating anti-inflammatory molecule that can ameliorate endotoxin-induced acute lung injury47,48. Therefore, the therapeutic effect of 5-methoxytryptophan on HAPE is promising and needs experimental validation. The level of 3-methyl-L-histidine was reduced in HAPE patients relative to controls in our study, which is opposite to previous work10. Besides, it has been reported that serum 3-methyl-L-histidine level was increased in ARDS and COPD patients, reflecting skeletal muscle atrophy and fiber loss49,50. Although our result related to 3-methyl-L-histidine cannot be explained rationally, the role of histidine metabolism in HAPE is worth exploring.

Apart from cis-aconitic acid and 3-acetylphenol sulfate mention above, our results also revealed that S-nitroso-N-acetylcysteine, aminocaproic acid, emodin, threo-hydroxyaspartic acid and 6-hydroxynicotinic acid had high diagnostic accuracy for HAPE. S-nitroso-N-acetylcysteine is an S-nitrosothiol that can act as a nitric oxide (NO) carrier and donor51. In the lungs, NO counterbalances vasoconstrictor stimuli in the physiological regulation of hypoxic pulmonary vasoconstriction52. It has been demonstrated that endothelial dysfunction with decreased NO availability plays a major role in HAPE52, hence the increasing serum concentration of S-nitroso-N-acetylcysteine may not exert protective effect on HAPE. Aminocaproic acid is an antifibrinolytic agent that has been used in the treatment or prevention of hemorrhagic diseases53. Thus, the reduced level of aminocaproic acid may induce or aggravate pulmonary hemorrhage in HAPE patients. Emodin exhibits multiple beneficial effects, including antibacterial, anti-inflammatory, antioxidant, and anti-hyperlipidemic effects54. Previous studies have suggested that emodin can ameliorate sepsis or acute pancreatitis associated lung injury through inhibiting inflammatory response and oxidative stress via different signaling pathways55,56. Our results indicated that serum emodin was decreased in HAPE subjects relative to controls. Therefore, supplement of emodin or Chinese medicinal herbs containing emodin may have protective effects against HAPE. Threo-hydroxyaspartic acid can induce neuronal damage of the substantia nigra, pars compacta in animal model57. The increased level of threo-hydroxyaspartic acid may lead to neurologic impairment in HAPE patients, which needs further research. The change of 6-hydroxynicotinic acid level suggested that nicotinate and nicotinamide metabolism was involved in HAPE. Under hypobaric hypoxia, the activation of nicotinamide adenine dinucleotide phosphate (NADPH) oxidase results in consumption of NADPH and redox imbalance58. Therefore, the interaction between metabolic pathways and redox reaction may have important implications for pathogenesis of HAPE.

There are some limitations in our study. First, the sample size was small and the identified biomarkers were not further verified. Follow-up studies using a larger sample size will be carried out to validate the biomarker potential. Second, it would have more clinical value if metabolomic profiles could be compared among healthy control, AMS patients and HAPE patients. Further population expansion with different health conditions is required for a more accurate optimization.

Conclusions

In conclusion, the serum metabolic profile of HAPE patients were analyzed using GC-MS and LC-MS analysis. A total of 119 differential metabolites were identified, which were primarily involved in pyrimidine metabolism, citrate cycle, sulfur metabolism, phenylalanine metabolism and purine metabolism. Based on correlation analysis and ROC analysis, 7 biomarkers (S-nitroso-N-acetylcysteine, aminocaproic acid, emodin, threo-hydroxyaspartic acid, 6-hydroxynicotinic acid, 3-acetylphenol sulfate and cis-aconitic acid) were screened out with high diagnostic accuracy. These findings could provide new insights into early detection and intervention of HAPE.