Animal care and experimental treatments were approved by the Medical Science Research Ethics Committee of Qinghai University (No. 2021-06, approved 21 June 2021).

A total of 64 male C57BL/6 mice, aged 5 weeks and weighing 16–17 g, were purchased from Hunan Slake Jingda Experimental Animal Co., Ltd. Mice were raised in an environment with a temperature of 18–23°C, and provided with sufficient food and water. After 1 week of adaptive feeding, the animals were randomly divided into a normal control group (NC, n = 24) and high-fat diet group (HFD, n = 40). HFD groups were fed a high-fat diet, while the NC group were fed with a normal diet. The feed source came from D12492, Beijing Keao Xieli Feed Co., Ltd. The feed inclued Casein, L-cystine, Corn Starch, Maltodextrin 10, Sucrose, Celluse, Soybean Oil, Lard, Mineral Mix S10026^*, Dicalcium phosphate, Calcium Carbonate, Potassium Citrate (1 H2O), Vitamin Mix, V10001, Choline Bitartrate, and FD&C Blue Dye # 1. The NC group was fed a standard diet (3.18 kcal/g, 4.0 g% fat, 67.3% carbohydrate, 19.1% protein). The HFD was fed a high-fat diet (5.24 kcal/g, 34.9% fat, 26.3% carbohydrate, 26.2% protein). After 8 weeks (13 weeks of age), the mice were fasted for 12 h and weighed, and the weight of the HFD group exceeded the average weight of the NC group by 10% as the criterion for judging the successful modeling of obese mice. After successful modeling, 20 male C57BL/6J obese mice and 20 normal feed mice were randomly divided into four groups: high altitude obese group (HOB), high altitude normal weight group (HN), low altitude obese group LOB (LOB), and low altitude normal weight group (LN). HOB and LOB groups were placed in a low-pressure oxygen chamber (simulated altitude of 5,000 m, atmospheric pressure of 425 mmHg, oxygen partial pressure of 72.5 mmHg, and oxygen concentration of 13.83%), LOB and LN groups were kept in a laboratory environment (altitude of 2,261 m, mean atmospheric pressure of 574 mmHg, oxygen partial pressure of 120.1 mmHg, and oxygen concentration of 17.5%) for 4 weeks. Body weight and food consumption were measured weekly; fecal gut microbiota and fecal metabolomics analysis were measured after 30 days.

Sterile plastic cassettes were used to collect fresh stool samples (5 g) from each group. The fecal samples were quickly placed in an ice box, transported to the laboratory with 2 h, and then stored at −80 in a refrigerator for further processing and testing.

Total genomic DNA was extracted from the samples using the acetyltrimethylammonium bromide (CTAB) method (Attitalla, 2011). DNA concentration and purity were analyzed using 1% agarose gels. According to the concentration, the DNA samples were diluted to 1 ng/μl using sterile water. An equal volume of 1 × loading buffer (containing SYBR green) was mixed with the PCR products, and electrophoresis was performed on a 2% agarose gel to visualize the PCR products. PCR products were mixed in equidensity ratios. Then, the PCR products were purified with a Qiagen Gel Extraction Kit (Qiagen, Germany). The DNA purity and concentration were analyzed by measuring the optical density (OD) at wavelengths of 260 and 280 nm with a NanoPhotometer R spectrophotometer (Implen, Munich, Germany) and then calculating the OD260:OD280 ratio. The DNA concentrations were measured with the Qubit R dsDNA Assay Kit in a Qubit R 2.0 Fluorometer (Life Technologies, Camarillo, CA, United States) (Du et al., 2022).

Based on many previous studies (Nagayama et al., 2020; Yu et al., 2020; Song et al., 2021), we selected the V3–V4 region to study the microbiome through second-generation sequencing. 16S rRNA gene sequencing was performed by Novogene Bioinformatics Technology Co., Ltd., China. DNA samples were diluted to a concentration of 1 ng/μl in sterile water and then PCR amplified with the 515F/806R primer set (515F: 5′GTGCCAGCMGCCGCGGTAA-3′, 806R: 5′-XXXXXXGG ACTACHVGGGTATCTAAT-3′). Sequencing libraries were generated using the TruSeq R DNA PCR-Free Sample Preparation Kit (Illumina, USA) following the manufacturer's recommendations, and index codes were added. Library quality was assessed with a Qubit@ 2.0 Fluorometer (Thermo Scientific) and Agilent Bioanalyzer 2100 system. Finally, the library was sequenced on the Illumina NovaSeq platform, and 250-bp paired-end reads were generated.

Paired-end reads were assigned to samples based on their unique barcodes and were truncated by removing the barcode and primer sequences. Paired-end reads were merged using FLASH (V1.2.7) (Magoč and Salzberg, 2011), which is a very fast and accurate analysis tool that was designed to merge paired-end reads when at least some of the reads overlapped with the read generated at the opposite end of the same DNA fragment. The splicing sequences are called raw tags. Quality filtering of raw tags was performed under specific filtering conditions to obtain high-quality cleantags (Bokulich et al., 2013) according to the QIIME (V1.9.1) (Caporaso et al., 2010) quality control process. The tags were compared with those in the reference database (Silva138 database) using the UCHIME (Attitalla, 2011) algorithm to detect chimeric sequences, and the chimeric sequences were removed (Haas et al., 2011). Then, effective tags were finally obtained.

The Kruskal–Wallis test was used to investigate significant differences in operational taxonomic units (OTUs) and the abundance-based coverage estimator (ACE), Chao 1, Simpson's, and Shannon's indexes among the four groups. To correct for type I errors, we applied the Bonferroni method for multiple comparisons between two groups. Differences in microbial community abundances between the obesity group and the control group were analyzed using the Wilcoxon rank sum test, and the significance of these differences was assessed using the false discovery rate (FDR). Principal coordinate analysis (PCoA) was performed using the WGCNA package, stat packages, and ggplot2 package in R software (v 2.15.3) to compare the similarity of community structures. Multiresponse permutation procedures (MRPPs) (O'Reilly and Mielke, 1980) were used to analyze differences in the microbial community structure between groups.

The supernatant of the stool sample was taken after grinding and centrifugation. The QC samples were prepared by mixing the supernatants of all samples in equal quantities. UHPLC-MS/MS analyses were performed using a Vanquish UHPLC system (ThermoFisher, Germany) coupled with an Orbitrap Q Exactive TM HF mass spectrometer (Thermo Fisher, Germany). Samples were injected onto a Hypesil Goldcolumn (100 × 2.1 mm, 1.9 μm) using a17 min linear gradient at a flow rate of 0.2 mL/min. The eluents for the positive polarity mode were eluent A (0.1% FA in Water) and eluent B (Methanol). The eluents for the negative polarity mode were eluent A (5 mM ammonium acetate, pH9.0) and eluent B (Methanol). The solvent gradient was set as follows: 2% B, 1.5 min; 2–85% B, 3 min; 85–100% B, 10 min; 100–2% B, 10.1 min; 2% B, 12 min. Q Exactive TM HF mass spectrometer was operated in positive/negative polarity mode with spray voltage of 3.5 kV, capillary temperature of 320°C, sheath gas flow rate of 35 psi and aux gas flow rate of 10 L/min, S-lens RF level of 60, and Aux gas heater temperature of 350°C.

The raw data files generated by UHPLC-MS/MS were processed using the Compound Discoverer 3.1 (CD3.1, ThermoFisher) to perform peak alignment, peak picking, and quantitation for each metabolite, and these metabolites were annotated using the KEGG database, HMDB database, and LIPIDMaps database. Positive and negative data were combined to obtain a combined data set, which was imported into the SIMCA-P+14.0 software. Principal component analysis (PCA) was first used to observe the overall distribution of each sample and the stability of the whole analysis process. Then, OPLS-DA were used to distinguish the overall differences of metabolic profiles between groups and screen differential metabolites between groups. The differentially expressed metabolites should meet the importance of the first principal component variable in the OPLS-DA model at projection VIP > 1 and P < 0.05. Volcano plot (FC > 1.2 or FC < 0.833), P used Benjamin and Hochberg to check (FDR). Metabolic pathway analysis was performed using MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/) software and for mapping. The data were subjected to multivariate (PCA) and orthogonal OPLS-DA and univariate [fold change (FC) and T-tests] analysis. Mean metabolite concentrations in each group were used to calculate FC values, Final identification of differential metabolites by VIP, FC and P.

Quantitative data with normal distribution between the two groups were expressed by mean±standard deviation and analyzed by student t test. The quantitation sequencing data with non-normal distribution were expressed by median ± quartile range (QR) and analyzed by Wilcoxon rank sum test. Association analysis of gut microbiota and metabolites was carried out using spearman correlation. The SPSS 22.0 version was used for statistical analysis. P < 0.05 was considered as statistically significant different.

The results showed that there was no significant difference in body weight between the two groups at baseline (P > 0.05) after 8 weeks of feeding, and the body weight gain value of the high-fat diet group was higher than that of the normal control group in the same feeding environment. When the weight of the two groups was compared at the same time, there was a significant statistical difference in the weight of the two groups at each time point (P < 0.05) (Table 1).

Twenty mice were randomly selected from the two groups with a high-fat diet (LOB and HOB), as were 20 mice from the normal diet groups. the weight change of mice after 4 weeks was shown in Table 2; the results showed that at the same altitude, there were statistical differences in body weight between the two groups between the normoxic group (2,261 m) and the hypoxic group (5,000 m) (P < 0.05). The comparison between the obese groups at different altitudes showed that the weight gain of the HOB was lower than that of the LOB, and the difference between the two groups was statistically significant (P < 0.05).

Mice heights were not different between altitudes and body weights (P > 0.05), and mice weights were statistically different between groups at the same altitude (P < 0.05). The body weight of the normoxic group was higher than that of the hypoxic group (Table 3).

α-diversity analysis by Shannon, Simpson, Choa1, and ACE index showed that there were differences in α-diversity between the groups (P < 0.05). Further intergroup comparisons showed that in the same oxygen groups, Shannon, Simpson, Chao1, and ACE indices were significantly different in the HOB and HN groups (P < 0.05). Simpson index was significantly different in the LOB and LN groups (P < 0.05), suggesting that gut microbiota diversity was lower in the obese group than in the control group. In the same body weight groups, Chao1 and ACE indices were significantly different in HOB and LOB groups (P < 0.05, Table 4), suggesting that gut microbiota diversity was lower in the hypoxic obese group compared to the control group. All these results suggested that hypoxia and obesity affect the diversity of the gut microbiota. PcoA was used to analyze the β-diversity of fecal samples among the four groups, and the results showed that the gut microbiota composition of normal weight mice was similar; the gut microbiota composition of obese mice was significantly different at different oxygen levels (P < 0.05, Figure 1). Using MRPPs based on Bray-Curtis distance, there was a small difference between samples within each group and a large difference between groups (P < 0.05, Table 5).

Overall, 25 phylum, 38 classes, 92 orders, 139 families, 230 genera, and 117 species were detected in the bacterial microbiome communities of the four group samples. Composition at the phylum level: In the obese group, Firmicutes, Bacteroidota, and Desulfobacterota dominated, and their proportions were HOB group (67.96%, 15.85%, 12.68%) and LOB group (72.38%, 17.09%, 6.34%). Firmicutes/Bacteroidota in the HOB group was higher than in the LOB group. Normal body groups were dominated by Firmicutes and Bacteroidota, and their proportions were HN (47.61%, 45.14%) and LN (51.52%, 42.70%) (Figure 2A). At the genus level, the LN group was dominated by Muribaculacea (35.02%), Faecalibaculum (6.13%), and Dubosiell (22.90%), while the LOB group was dominated by Muribaculaceae (10.26%), Faecalibaculum (28.17%), and Lactobacillus (16.44%). The HN group was dominated by Muribaculaceae (34.82%) and Dubosiella (9.04%) and the HOB group by Romboutsia (21.01%) and Faecalibaculum (14.07%). Among them, the hypoxic obesity group was significantly dominated by Romboutsia (Figure 2B).

The Metastat method was used to analyze the differences in microbiota abundance composition of the top 10 microbiota at the genus level between the groups. For the same body weight groups, the HN vs. LN groups demonstrated that Dubosiella, Romboutsia, and Turicibacter were significantly different (P<0.05), and the abundance of Romboutsia and Turicibacter was higher in the HN group. In the HOB vs. LOB group, Faecalibaculum, Romboutsia, Lactobacillus, Turicibacter, and Erysipelatoclostridium were significantly different (P<0.05), the abundance of Romboutsia was higher in the HOB group, and Romboutsia was significantly dominant in the hypoxia group. At the same oxygen level, in the LOB and LN groups, Muribaculaceae, Faecalibaculum, Dubosiella, Lactobacillus, and A2 showed significant differences (P<0.05), and Faecalibaculum and Lactobacillus had higher abundance in LOB. In the HOB and HN groups, Muribaculaceae, Faecalibaculum, Romboutsia, Lactobacillus, Turicibacter, and A2 showed differences (P<0.05), and Faecalibaculum and Romboutsia were more abundant in HOB (Table 6), suggesting that Faecalibaculum, Lactobacillus, and A2 were predominant in the obese groups, while Romboutsia and Turicibacter were predominant in hypoxia groups.

The PCA and clustering heatmap illustrate the differences between the sample distributions. The PCA map shows that the metabolites are clustered within groups and separated between groups (Figure 3A), and the clustering heatmap can roughly indicate that metabolites of the same body weight have similar expression patterns (Figure 3B).

OPLS-DA model validation (Figure 4) revealed significant differences between the distribution of samples, suggesting a significant difference of fecal metabolites between the two groups, The model parameters are shown in Table 7, suggesting that the model was reliable. For the same body weight groups, i n HOB vs. LOB, a total of 736 different metabolites (P < 0.05, VIP > 1) were detected (Figure 5A), while 64 differential metabolites were successfully annotated, the majority of which were dominated by lipid compounds. KEGG pathway enrichment analysis of a total of 64 differential metabolites revealed that 60 differential metabolites were involved in pathway enrichment, involving a total of 26 metabolic pathways, of which seven pathways were significantly enriched (P < 0.05), namely arginine biosynthesis (P = 0.000), alanine, aspartate and glutamate metabolism (P = 0.001), pyrimidine metabolism (P = 0.004), D-Glutamine and D-glutamate metabolism (P = 0.004), purine metabolism (P = 0.005), steroid hormone biosynthesis (P = 0.010), and arginine and proline metabolism (P = 0.030) (Figure 6A). The enriched differential metabolites are shown in Table 8. In HN vs LN, a total of 737 differential metabolites were detected (P < 0.05, VIP > 1) (Figure 5B); 209 differential metabolites were successfully annotated. KEGG enrichment pathway analysis of a total of 209 differential metabolites revealed that 61 differential metabolites were involved in pathway enrichment, involving a total of 26 metabolic pathways, of which seven pathways were significantly enriched (P < 0.05), involving a total of 34 metabolic pathways, of which four pathways were significantly enriched (P < 0.05), namely histinide metabolism (P = 0.000), arginine metabolism (P = 0.002), primine metabolism (P = 0.023), and purine metabolism (P = 0.034) (Figure 6B); the enriched differential metabolites are shown in Table 9.

For the same hypoxia level groups, in HOB vs. HN, a total of 705 differential metabolites (P < 0.05, VIP > 1) were detected (Figure 5C), while 121 differential metabolites were successfully annotated. The KEGG enrichment pathway analysis of a total of 121 differential metabolites revealed that 78 differential metabolites were involved in pathway enrichment, involving a total of 26 metabolic pathways, of which six pathways were significantly enriched (P < 0.05), namely steroid hoemone biosynthesis (P = 0.000), alainine, aspartate and glutamate metabolism (P = 0.015), arginine biosynthesis (P = 0.029), arachidonic acid metabolism (P = 0.030), butanoate metabolism (P = 0.032), and tyrosine metabolism (P = 0.044) (Figure 6C), the enriched differential metabolites are listed in Table 10. In LOB vs. LN, a total of 785 differential metabolites (P < 0.05, VIP > 1) were detected (Figure 5D); 143 differential metabolites were successfully annotated. The KEGG enrichment pathway analysis of a total of 143 differential metabolites revealed that 74 differential metabolites were involved in pathway enrichment, involving a total of 19 metabolic pathways, of which five pathways were significantly enriched (P < 0.05): pyrimidine metabolism (P = 0.002), pantothenate and CoA biosynthesis (P = 0.009), purine metabolism (P = 0.018), steroid hormone biosynthesis (P = 0.033), butanoate metabolism (P = 0.048), and tyrosine metabolism (P = 0.044) (Figure 6D). The enriched differential metabolites are shown in Table 11. These results suggest that hypoxia mainly affects nucleotide and amino acid metabolism, whereas obesity affects lipid and amino acid metabolism.

Spearman's correlation was used to investigate the relationship between gut microbiota and metabolites. The relationship between the most differentially expressed metabolites and the top 10 genera was analyzed in mice (Figure 7). In the same body weight groups, in HOB vs. LOB, we found that the abundance of Romboutsia was positively correlated with the fecal lipid compounds adrenosterone, desoxycortone, 2-methoxyestrone, and cortisol (P < 0.05), and with that of fecal nucleotide compounds guanosine, cAMP, adenosine, uridine, and inosin (P < 0.05), but negatively correlated with organic acids L-glutamine, fumaric acid, L-hydroxyproline, L-ornithine, L-glutamic acid, N-acetylornithine, and L-argininosuccinate (P < 0.05). Lactobaculum was positively correlated with the organic acids L-glutamine, fumaric acid, L-hydroxyproline, L-ornithine, L-glutamic acid, N-acetylornithine, and L-argininosuccinate (P < 0.05), but there was a negative correlation with nucleotides guanosine, thymidine, adenosine, uridine, and inosine (P < 0.05). This is the opposite of Romboutsia in relation to metabolites (Figure 7A). In HN vs. LN, Turicibacter was positively correlated with the nucleotide compounds adenosine, urocanic acid, and cytidine (P < 0.05) and Romboutsia was negatively correlated with the organic acid compounds allantoate and ornithine (P < 0.05). These results suggested that the microbial Romboutsia characterized by hypoxia showed a negative correlation with organic acid compounds and a positive correlation with nucleotide and lipid compounds (P < 0.05) (Figure 7B).

In the same oxygen level groups, in HOB vs. HN, Romboutsia was positively correlated with the fecal lipid compounds prostaglandin I2, estradiol, estriol, desoxycortone, prostaglandin F2α, and adrenosteron (P < 0.05), but negatively correlated with the organic acid L-argininosuccinate, succinic acid, and fumaric acid (P < 0.05). Muribaculaceae was negatively correlated with the fecal lipid compound prostaglandin I2, estradiol, estriol, desoxycortone, prostaglandin F2α, and adrenosteron (P < 0.05), but positively correlated with the organic acid succinic acid. Lactobacillus was positively correlated with fumaric acid, succinic acid, and L-argininosuccinat (P < 0.05). A2 was positively correlated with prostaglandin I2, estradiol, estriol, desoxycortone, prostaglandin F2α, and adrenosteron (P < 0.05), but negatively correlated with fumaric acid, succinic acid, and L-argininosuccinat (P < 0.05) (Figure 7C). In LOB vs. LN, Muribaculaceae was positively correlated with nucleotides 2′-deoxyadenosine 5′-monophosphate, dCMP, adenosine 5′-monophosphate, xanthine, and 2′-deoxyadenosin (P < 0.05); Lactobacillus was negatively correlated with 2′-deoxyadenosine 5′-monophosphate, dCMP, and adenosine 5′-monophosphat (P < 0.05). Faecalibaculum was negatively correlated with orotic acid, succinic acid, and pantothenic acid (P < 0.05), but positively correlated with uracil and cytidine-5′-monophosphat (P < 0.05). It was positively correlated with the fecal lipid compound estriol (P < 0.05) and with correlated with acetoacetat (P < 0.05). A2 was negatively correlated with 5′-adenylic acid, adenosine 5′-monophosphate, and deoxycytidin (P < 0.05) (Figure 7D). These results suggested that gut microbiota in the hypoxic group were positively correlated with lipid and nucleotide compounds and negatively correlated with organic acid compounds.