N, O-Bis(trimethylsilyl) trifluoroacetamide (BSTFA) with 1% trimethylchlorosilane (TMCS), L-norvaline, L-norleucine, methanol, methoxyamine hydrochloride, and anhydrous pyridine, were purchased from Sigma-Aldrich.

Several C57BL/6 mice (8 weeks old) were purchased from the Shanghai Laboratory Animal Center (Shanghai, China). The PH mouse model was established through chronic hypoxia and OVA assistance as previously described [13, 14, 15]. Briefly, the PH mouse model was as follows: the hypoxia + C75 group received C75 (200 µg/kg/week, dissolved in 0.5% dimethylsulfoxide (DMSO)) for 5 weeks; the hypoxia group was intraperitoneally injected with the same amount of 0.5% DMSO without C75 each week; and the control group received 0.5% DMSO injection and was raised in a normoxic environment [13, 14, 15]. All animal experiments were performed with the approval of the Shanghai Jiao Tong University Institutional Animal Care and Use Committee.

The serum was collected and frozen at –80 ℃ until gaschromatography-mass spectrometry (GC/MS) analyses. Blood samples from four mice were randomly selected from each group. Serum samples were thawed and vortexed, 20 µL of which was mixed with 80 µL of cold methanol and an internal standard (5 µg/mL L-norvaline). The mixture was vortexed and maintained at –20 ℃ overnight. Following centrifugation, 70 µL of supernatant and 10 µL of L-norleucine (50 µg/mL) in a glass vial were evaporated to dryness under a nitrogen stream. Methoxyamine hydrochloride (50 µL, 20 mg/mL) in pyridine was subsequently added to the residue. 50 µL of BSTFA (with 1% TMCS) was added to the mixture and then derivatized at 70 °C prior to GC-MS metabolomics analyses. At the same time, a blank derivatization sample needs to be prepared in the process of sample preparation and GC/MS analyses. Quality control (QC) samples were pooled from all samples and they were prepared and analysed with the same procedure as the experimental samples.

Agilent gas chromatography-mass spectrometry (7890A/5975C) and an OPTIMA® 5 MS Accent fused silica capillary column (30 m $\times$ 0.25 mm $\times$ 0.25 µm, MACHEREY-NAGEL, Düren, German) were used to perform metabolomics on the derivatives. Helium passed through the column as the carrier gas and the constant flow rate was 1 mL/min. The gradient heating procedure was 60 °C for 1 min, 60 ℃ to 240 °C at a rate of 12 °C/min, 240 °C to 320 °C at a rate of 40 °C/min and finally maintained at 320 °C for 4 min. The solvent delay time was set to 5.4 min. The injection volume was 1 µL in nonshunt mode. The temperatures of the injector, transfer line, and electron impaction source were set to 250 °C, 260 °C, and 230 °C, respectively. The electron ionization energy was set at 70 eV. The data was collected in full-scan mode (m/z 50–600). The samples were annalysed in a random order.

The original GC-MS data were processed with reference to previously published protocols [16]. The data were normalized against total peak abundance and the normalized data were imported into a SIMCA-P (version 14.1, Umetrics, Umea, Sweden), in which the data were preprocessed by unit variance [17] scaling and mean centering before performing principal component analyses (PCA), partial least-squares discriminant analyses (PLS-DA), and orthogonal partial least-squares discriminant analyses (OPLS-DA). The potential differential metabolites were identified by combining variable influence on projection (VIP) values of the OPLS-DA model (VIP $>$1) and p values of Student’s t test (p 0.05) on the normalized peak areas. The false discovery rate (FDR) was calculated by the Benjamini-Hochberg method. Fold change was calculated as a binary logarithm of the average mass response (normalized peak area) ratio between the two groups.

Structural identification of differential metabolites was performed as follows. The AMDIS software (version 2.70, NIST, Gaithersburg, MD, USA) was applied to deconvolute mass spectra from raw GC-MS data, and the purified mass spectra were automatically matched with an author-constructed standard library including retention time and mass spectra, Golm Metabolome Database, and Agilent Fiehn GC/MS Metabolomics RTL Library, by which the metabolites were structurally confirmed.

The online software MetaboAnalyst 4.0 (http://www.metaboanalyst.ca) was used to analyse the pathways of differential metabolites. Biosynthesis of essential amino acids was included as mouse metabolism is involved in the intestinal flora.

The data were analysed retrospectively and were obtained from six patients with PH (PH group) and six relative controls (patients without PH and heart disease) of Shanghai Children’s Hospital, Shanghai, China, between 2018 and 2021. All patients underwent blood tandem mass spectrometry during hospitalization. Recruited patients in the PH group were diagnosed with PH according to Doppler echocardiography and clinical presentation [18]. Echocardiographic signs suggesting right atrial enlargement, right ventricular hypertrophy, pulmonary artery diameter expansion, or congenital heart disease were used to assess the probability of PH in addition to clinical manifestations. Patients with other potential causes of lung disease, such as pneumonia and connective tissue disease were excluded from this study.

The results of clinical data are represented as the median (interquartile ranges, IQRs). Comparisons among groups were evaluated by a nonparametric method. Statistical analyses were performed using SPSS software, version 21.0 (IBM Corp. Inc., Armonk, NY, USA) and graphed with GraphPad Prism version 7 (GraphPad Software, Inc., San Diego, CA, USA). For univariate statistical analyses, the normalized data derived from GC/MS analyses were calculated by a two-tailed Student’s t test in Excel 2007 (Microsoft, Redmond, WA, USA). p 0.05 was considered a significant difference.

GC/MS-based metabolomics was used to profile the serum metabolome of C57BL mice from 3 groups. PCA was established to show the metabolic difference between the control and hypoxia groups. The PCA results showed that there was a significant difference between the control and hypoxia groups (Fig. 1a) and the current model was reliable (R${}^2$X = 0.569). Then, we confirmed the results by a PLS-DA (R${}^2$X = 0.645, R${}^2$Y = 0.999, Q${}^2$ = 0.868) assay (Fig. 1b). An OPLS-DA assay (R${}^2$Y = 1, Q${}^2$ = 0.887) was used to analyse the differential metabolites. The OPLS-DA results showed a significant separation between the control and hypoxia groups (Fig. 1c). However, the PCA score plots (Fig. 1d) showed that there was no tendency to separate the hypoxia group from the hypoxia + C75 group (R${}^2$X = 0.476). When we discarded an outlier of the hypoxia + C75 group, the effective models of PLS-DA (R${}^2$X = 0.452, R${}^2$Y = 0.994, Q${}^2$ = 0.739) (Fig. 1e) and OPLS-DA (R${}^2$X = 0.867, R${}^2$Y= 1, Q${}^2$ = 0.999) (Fig. 1f) were built and the two models were clearly reliable. Both PLS-DA and OPLS-DA score plots showed that the hypoxia and hypoxia + C75 groups were distinctly separated. The two models revealed a significant metabolic difference between the hypoxia and hypoxia + C75 groups.

Fig. 1.

The metabolic differences in hypoxia-induced PH mice serum. (a–c) The PCA, PLS-DA, OPLS-DA plots between the control and hypoxia group. (d) The PCA plots between the hypoxia and hypoxia + C75 group. (e,f) The PLS-DA, OPLS-DA plots between the hypoxia and hypoxia + C75 group after discarding an outlier of hypoxia + C75 group. PCA, principle component analyses; PLS-DA, partial least-squares discriminant analyses; OPLS-DA orthogonal partial least-squares discriminant analyses.

As described above, differential metabolites were identified through the combination of VIP values (VIP value $>$1) acquired from the OPLS-DA model and p values (p value 0.05) from Student’s t test on the normalized peak. There were 29 differential metabolites identified in the hypoxia group compared with the control group (Supplementary Table 1). The visualized heatmap between the control and hypoxia groups was shown in Fig. 2a. Fifteen metabolites were upregulated and 14 were downregulated in the hypoxia group. Among them, glycerol, cholesterol, docosahexaenoic acid, putrescine, glutamine, and other metabolites were significantly increased in the hypoxia group compared with the control group, while cholic acid, chenodeoxycholic acid and other metabolites were decreased.

Fig. 2.

Heatmap of the potential differential metabolites. (a) Representative image of the comparison between the control and hypoxia groups. (b) Representative image of the comparison between the hypoxia and hypoxia + C75 groups. Red represents the high relative level of each metabolite, and purple represents the low relative level of each metabolite.

Fifteen metabolites were identified in the hypoxia + C75 group compared with the hypoxia group (Supplementary Table 2). The visualized heatmap between the hypoxia and hypoxia + C75 groups were shown in Fig. 2b. Seven metabolites were upregulated and 8 were downregulated in the hypoxia + C75 group. We observed that cholic acid and chenodeoxycholic acid levels were increased in the hypoxia + C75 group compared to the hypoxia group, whereas the levels of glycerol, putrescine, and pyroglutamine were decreased.

Based on the above analyses, we found that fatty acids, polyamine, and glutamine were common differential metabolites. In addition, other common differential metabolites including Docosahexaenoic Acid (DHA), glycerol, cholic acid, chenodeoxycholic acid, glutamine, and putrescine were increased in the hypoxia group compared with the control group, and C75 treatment was partially rescued (Fig. 3).

Fig. 3.

The statistical diagram of serum differential metabolites including DHA, glycerol, cholic acid, chenodeoxycholic acid, glutamine, and putrescine shared by the three groups of mice. Values in the box plots are presented as the normalized peak areas of the metabolites in the three groups. DHA, Docosahexaenoic acid.

Metabolomics comprehensive treatment software MetaboAnalyst 4.0 (http://www.metaboanalyst.ca, McGill University, Sainte-Anne-de-Bellevue, Quebec, Canada) was used to analyse potential metabolic pathways. Pathway impact $\ge$0.10 was identified as the potential target pathway criterion. Phenylalanine, tyrosine, tryptophan biosynthesis, taurine, hypotaurine, arginine, proline, phenylalanine, glycerolipid, alanine, aspartate, glutamate, and tryptophan metabolism were identified and involved in PH metabolic disorder resulting from chronic hypoxia intervention (Fig. 4a). Nicotinate, nicotinamide, beta-alanine, glycerolipid, arginine, and proline metabolism were identified as changed metabolic pathways in the hypoxia + C75 group compared with the hypoxia group (Fig. 4b). Based on the above, we hypothesized that arginine, proline, and glycerolipid metabolism were related to PH.

Fig. 4.

Metabolic pathway analysis shows the pathways of significantly altered metabolites. (a) The results of metabolite set enrichment analyses of the control and hypoxia group. (b) The results of metabolite set enrichment analyses of the hypoxia and hypoxia + C75 group. The spot size represents the contribution of the metabolic pathway to the difference between the two groups, and the color represents the correlation between the metabolic pathway and difference between the two groups.

Next, we examined the levels of the differential metabolites mentioned above and other PH related metabolites [5, 8] in paediatric PH patients. We selected six PH children and six relative controls (patients without lung and heart disease) and examined the serum metabolite levels. Clinical information of the recruited patients was shown in Supplementary Table 3. We found that glutamine and caproyl carnitine levels were increased in PH paediatric patients. Palmitoyl carnitine was also increased in PH paediatric patients; however, the difference was not significant (Fig. 5).

Fig. 5.

The serum levels of caproyl carnitine, glutamine, palmityl carnitine, and arginine in PH pediatric patients and their relative controls. Differences between groups were assessed by non-parametric statistical method. Bars represent 25 and 75 percentiles and ns means non-significance. (n = 6). PH group, pulmonary hypertension (PH) patients.

Excessive proliferation of PASMCs in PH needs to mobilize lipid metabolism [3, 19]. FAS is a key enzyme that produces long-chain fatty acids [12]. FAS inhibitors inhibited proliferation, increased apoptosis in hypoxic human PASMCs, and ameliorated the symptoms and pulmonary vascular remodelling of MCT-treated rats [11]. In this study, we observed that fat synthesis products, such as glycerol, cholesterol, and DHA, were significantly elevated; in contrast, fat breakdown products, such as cholic acid, chenodeoxycholic acid, and uric acid, were decreased in the hypoxia group. C75 treatment partially reversed these effects (Fig. 3), which was consistent with the above findings and showed that C75 treatment played a protective role in hypoxia-induced PH mice by inhibiting the synthesis of fatty acids.

Polyamines are involved in numerous physiological functions, such as cell growth, gene regulation, differentiation, and development [20]. Polyamines, including spermidine, spermine, and putrescine, are essential factors for eukaryotic cell growth [21, 22]. These polyamines were upregulated in cancer transformation and tumour progression [23]. Based on the above information, PH pathogenesis is similar to cancer [3, 11, 19]. In this study, we found that the putrescine level was elevated in the hypoxia group compared with its relative control, and C75 treatment mitigated the hypoxia-induced elevation (Fig. 3). Our results were in line with the following PH rat model and idiopathic pulmonary arterial hypertension (IPAH) patients [6, 24]. Hautbergue et al. [6] showed that putrescine levels were upregulated in the PH rat model, whereas arginine (which is the precursor of putrescine) was decreased. Moreover, He et al. [24] also reported that plasma spermine levels were higher in patients with IPAH than in healthy controls, and they validated this finding in an experimental rodent model. They also validated spermine function in vitro, and spermine promoted human PASMC proliferation and migration, which aggravated pulmonary vascular remodelling [24].

However, the arginine level in the PH group was higher than that in the control group, although there was no significant difference between the two groups; this result was inconsistent with the change in putrescine in this animal study and previous studies [6], which may be caused by a few clinical studies and sample differences between individuals from the two groups. Selection bias caused by the enrolment requirement could occur. The above results indicated that C75 treatment might play a protective role in hypoxia-induced PH mice by inhibiting the synthesis of putrescine.

It is well known that glutamine participates in the bioenergetics of tumour cell proliferation and supports cellular resistance to oxidative stress [25, 26]. The carbon and nitrogen provided by glutamine are vital for the homeostasis of fatty acids, amino acids, and nucleotides [27]. Glutamine is converted to glutamate by glutaminase (GLS) [27]. GLS is related to tumour progression and growth rate in rodent models, and tumour proliferation is inhibited by gene knockout or GLS inhibitors [28, 29, 30], suggesting that glutamine is involved in tumour cell proliferation. Previous studies reported that glutamine was significantly downregulated in PH patients [31, 32] and played an important role in scavenging reactive oxygen species (ROS) by producing glutathione, which was involved in tumour progression and PH severity [27, 33]. In this study, we noticed that glutamine levels were elevated in the hypoxia group compared with its relative control, and C75 treatment partially reversed this increase (Fig. 3). Change of glutamine level was not consistent with previous studies [34, 35], which may have resulted from the small sample size and the different models of PH. Additionally, previous study reported glutamine levels decreased in the RV tissue of model with PH. There may be some differences in the metabolomics of serum and RV tissue.

Furthermore, the changes in glutamine in the PH patients also do not correspond with those of the previous study [31]. The small number of clinical studies and sample differences between individuals might be potential reasons. Additionally, previous studies reported that glutamine levels decreased in the lung tissue of patients with PH [31], and there may be some differences between the metabolomics of serum and lung tissue.