The assay kits for superoxide dismutase (SOD), malondialdehyde (MDA) and lactic acid were purchased from Jiancheng Bioengineering Institute (Nanjing, China). Formic acid was obtained from Fluka (Buchs, Switzerland). Acetone, citrate, and ammonium formate were obtained from Sigma-Aldrich (St. Louis, MO, United States). HPLC-grade methanol and acetonitrile (ACN) were purchased from Merk (Darmstadt, Germany). Valine, sodium succinate, phenylalanine, methionine, uric acid, arachidonic acid, linoleic acid, oleic acid, and palmitic acid were purchased from Shanghai Jingchun Reagent Co. Ltd. (Shanghai, China).

Male BALB/c mice, 6- to 8-week-old, weighing 18–22 g, were obtained from the Experimental Animal Center of Army Military Medical University. Mice were housed at 22 ± 2°C and 60 ± 10% relative humidity in a specific pathogen-free environment, with a 12/12 h light/dark cycle and ad libitum access to food and water. Thirty BALB/c mice were randomly divided into normoxic control (H0) group, acute hypoxic (H1) group, and acute repetitive hypoxia for four times (HPC) group.

Repeated hypoxic exposure were performed at room temperature (20 ± 1°C). The animal model of HPC was established according to our method described previously (Liu et al., 2017). After weighted, a mouse was placed in a 125-ml jar, which was sealed airtight with a rubber stopper and smeared with Vaseline. The mouse was taken out of the jar immediately following the appearance of the first asthmoid respiration (a sign of the hypoxia tolerance limit) (Wang et al., 2008); this was the first instance of hypoxia exposure. A maximum of 15 s was allowed for recovery under normoxic conditions. Subsequently, the mouse was immediately moved to another new jar with the same volume fresh in order to duplicate a progressive hypoxic environment three more times; the time of hypoxia tolerance in each mouse (from the beginning of the first airtight exposure to the final asthmoid respiration) was recorded. The H1 group was subjected to hypoxia only one time, and the H0 group did not undergo the hypoxic treatment. According to the following formula, the standard tolerance time was computed: T = t/(v -w)/0.94 × 100 (T, standard tolerance time; t, hypoxia tolerance time; v, jar volume; w, mouse weight).

At the end of the experiment, the mouse was sacrificed by decapitation. The hippocampus was isolated immediately on ice, frozen using liquid nitrogen, and stored at -80°C until analysis. All the animal experimental procedures and handling were approved by the Administrative Committee of Experimental Animal Care and Use of Army Military Medical University.

The antioxidative activity and the magnitude of anaerobic metabolism were detected with the levels of SOD, MDA, and lactic acid. All measurements were performed using the double antibody sandwich enzyme-linked immunosorbent assay (ELISA) kits according to the manufacture’s instructions.

Each hippocampal tissues sample was weighed, and then homogenized by a bead mill homogenizer (Bullet Blender Blue, Next Advance). The tissue was transferred to a chilled safe-lock microcentrifuge tube. A mass of chilled stainless steel beads equal to the mass of the tissue was added to the tube. Ten-fold (w/v) ice-cold 80% methanol was then added to the tissue and beads. Samples were mixed in the Bullet Blender Blue for 2 min at a speed of eight. The extracts were centrifuged at 12,000 rpm for 10 min at 4°C, and all supernatant were lyophilized and then stored at -80°C until analysis. Samples were resuspended using 100 μl 80% methanol. In order to take into account the signal drift of the mass spectrometer over the run time, the hippocampus samples were injected into UHPLC-QTOFMS system in a random run order.

UHPLC-QTOFMS analysis was performed on Agilent 1290 Infinity LC system coupled with Agilent 6530 Accurate-Mass Quadrupole Time-of-Flight (Q-TOF) mass spectrometer. An acquity UHPLC HSS T3 C18 column (2.1 mm × 100 mm, 1.8 μm, Waters, Milford, Ireland) was used for hippocampal sample separation. The other detailed UHPLC-QTOFMS protocols were carried out according to our previously published paper (Liao et al., 2016).

The data were pre-processed according to our previous work with slight modification (Liao et al., 2016). The raw data from UHPLC-QTOFMS were converted to mzData format by using Agilent MassHunter Qualitative software. XCMS package¹ based R software was used to extract features in these data. The XCMS major parameters full width at half maximum (fwhm), bandwidth (bw), and snthresh were set as 10, 10, and 5, respectively. Other parameters in XCMS were default settings. The variables presenting in at least 80% of either group were extracted. The resulting dataset was normalized using the intensity of the internal standard, and then the ion peaks generated by the internal standard were removed. The resulting matrix (observation, retention time, m/z, intensity) were introduced into SIMCA-P 13.0 software (Umetrics, Umeå, Sweden) for principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA), in which the data was mean centered and Pareto-scaled for both models. The quality of PCA and PLS-DA models was assessed with the relevant R2 and Q2 discussed elsewhere (Yin et al., 2009). One-way ANOVA was performed successively to reveal the statistical differences of the variables among different groups.

Metabolites identification was conducted according our previous work (Liao et al., 2016). Briefly, the quasi-molecular ion peak was identified according to ESI⁺ and ESI^- mode scans of UHPLC-QTOFMS. Then, the most probable molecular formula was calculated using the Agilent MassHunter software and the structure information were obtained by searching free databases such as Metlin, HMDB, and KEGG. Finally, the commercial standard MS/MS spectrum was used to confirm the identified compound. In addition, pathway analysis and visualization was performed by using web-based MetaboAnalyst 3.0 (Xia and Wishart, 2011; Xia et al., 2012).

The time of hypoxia tolerance in mouse was recorded when the first asthmoid respiration appeared. As shown in Figure 1, the hypoxia tolerance of mice was prolonged with the increase in the time of hypoxia exposure, which suggested that the hypoxia preconditioned mouse model was successfully established.

Compared with the H0 group, the hippocampal activity of SOD significantly increased in the mice in the H1 and HPC groups, while the levels of MDA and lactic acid significantly increased in the H1 group and decreased in the HPC group (Figure 2).

The typical total ion current (TIC) chromatograms of UHPLC-QTOFMS are shown in Figure 3. The stability of UHPLC-QTOFMS system is very important to obtain valid biochemical information. To assess the stability, the coefficients of variation (CV) of internal standard was calculated. It showed the CV was less than 15%, suggested that the method was robust. 1215 and 919 ion peaks were obtained from positive and negative ion modes of UHPLC-QTOFMS, respectively.

After data normalization, an unsupervised PCA was conducted on the dataset, which a separated trend of inter-group was observed on the scores plots. The plots of unsupervised PCA and supervised PLS-DA scores obtained from the three groups were displayed in Figures 4, 5. The PLS-DA scores plots from positive and negative ion datasets showed that the H0, H1, and HPC groups could be separated distinctly. The PLS-DA models were validated by a permutation test (99 times). As shown in Figures 4C, 5C, model validation with permutation test generated intercepts of R2 = 0.219 and Q2 = -0.33 in positive ion dataset and R2 = 0.226 and Q2 = -0.289 in negative ion dataset, which meant that two the PLS-DA models were non-overfitting and reliable.

Metabolites of interest that significantly contributed to the separation of different groups were identified based on a threshold of variable importance in the projection (VIP) values (VIP > 1) generated from PLS-DA model. To select the potentially different metabolites, these discriminated metabolites were validated according to one-way ANOVA (p < 0.05). According to the above two different statistical methods, 34 significantly altered metabolites were identified in HPC (Table 1).

The relevant pathways and networks of HPC were revealed by MetaboAnalyst 3.0. According to a threshold of the impact-value (≥0.10), the potential target metabolic pathway of HPC were filtered, including phenylalanine, tyrosine and tryptophan biosynthesis, taurine and hypotaurine metabolism, phenylalanine metabolism, glutathione (GSH) metabolism, alanine, aspartate and glutamate metabolism, tyrosine metabolism, tryptophan metabolism, purine metabolism, citrate cycle, and glycerophospholipid metabolism were disrupted in HPC (Figure 6 and Table 2). In addition, further correlated pathways were constructed using the reference map obtained from KEGG (Figure 7).

It is well-known that SOD is an important antioxidant and MDA is an index of free radial induced lipid peroxidation, which are commonly used as indicators of oxidative damage. The results of the present study demonstrated that, compared with the H0 group, the hippocampal activity of SOD significantly increased in the mice in the H1 and HPC groups, while the content of MDA significantly increased in the H1 group and decreased in the HPC group. These alterations indicated that the antioxidant capacity of hippocampal tissue was enhanced with the formation of HPC. Therefore, it was hypothesized that HPC protected the hippocampal tissue against hypoxic injury partially by enhancing the antioxidant capacity.

Glutathione, a main scavenger of free radicals in the body, can be divided into reductive GSH and oxidized GSH (GSSG). Under physiological conditions, GSH is essential to maintain cellular biological functions and to remove excess free radicals in the body (Enns and Cowan, 2017). The results of the present study demonstrated that GSH was significantly decreased in the H1 group and increased in the HPC group, whereas GSSG showed the opposite trend. The mechanism may be associated with the reduced activity of GSH reductase (promote the conversion of GSSG to GSH, using NADPH as substrate) induced by acute hypoxia (Maiti et al., 2008; Coimbra-Costa et al., 2017). This leads to a loss of antioxidant capacity of the GSH system and increased oxidative stress during acute hypoxia. After HPC, the level of GSH was increased and GSSG was decreased, indicating the increased antioxidant activity. We speculate that the regulation of GSH metabolism may serve a principal role in HPC. To increase GSH levels in the brain might be a potential strategy against hypoxia-induced brain injury.

Uric acid is generated by enzyme xanthine oxidase (XO) which catalyzes hypoxanthine to xanthine and then to uric acid (Sinha et al., 2009). The results of the present study indicated that the level of uric acid was significantly increased in the H1 group and decreased in the HPC group, while the level of xanthine was gradually decreased. The mechanism may be associated with the reduction in ATP levels when increased adenine nucleotide turn over was coupled with the stimulation of XO (Hare and Johnson, 2003). In addition, uric acid is an important non-enzymatic antioxidant. The decreased level of uric acid in the HPC group may indicate a decreased level of oxidative stress.

Glutamate is a major excitatory neurotransmitter working at a variety of excitatory synapses in the nervous system and excessive extracellular glutamate concentrations are neurotoxic (Rothman and Olney, 1995). The results of the present study indicated that the level of glutamate was significantly increased in the H1 group and decreased in the HPC group, whereas the inhibitory neurotransmitter, including γ-aminobutyric acid (GABA), dopamine and adenosine, and taurine as the antagonist of excitatory amino acid gradually increased with the formation of HPC. Therefore, it was hypothesized that HPC ameliorates excitotoxicity partially by shifting from excitatory glutamate-mediated neurotransmission to inhibitory GABA-mediated neurotransmission (Dave et al., 2005; Yenari et al., 2008) and/or upregulating the levels of dopamine (Kim et al., 2004) and adenosine. Additionally, previous studies have demonstrated that taurine modulated neuronal activity, combated ER stress responses induced by glutamate or regulated the balance of intracellular and extracellular calcium ions, and protected neurons from ischemic and hypoxic injury, in addition to other functions (Tang et al., 2014; Prentice et al., 2017). Therefore, upregulating the abovementioned inhibitory neurotransmitter might be beneficial to the tolerance of mice to hypoxia.

The level of lactic acid was increased in HPC-induced mice, which suggested that anaerobic glycolysis was involved in HPC. Meanwhile, the levels of citrate and oxoglutaric acid were increased, which suggested that mitochondrial TCA cycle activity was inhibited in the HPC group. Reportedly, the activity of lactate dehydrogenase (LDH) was significantly increased and succinate dehydrogenase (SDH) was decreased in the hippocampus of the hypoxia-preconditioned mice. This result indirectly substantiated the results of the present study (Lu et al., 2005).

Inosine 5′-phosphate is a key substrate of the de novo synthesis of guanine nucleotides, which is catalyzed to xanthosine 5′-phosphate by inosine-5′-phosphate dehydrogenase with the nicotinamide adenine dinucleotide (NAD⁺)-dependent oxidation (inosine 5′-phosphate + NAD⁺ + H₂O ⇌xanthosine 5′-phosphate + NADH + H⁺). This process plays an important role in regulating the synthesis of DNA and RNA (Greenberg, 1951). Our data showed that the inosine 5′-phosphate was significantly increased and NADH was decreased in the HPC group, which indicated that the DNA and RNA synthesis were reduced with the formation of HPC. The purpose of these changes may be to reduce the energy consumption.

Our results indicated that lysophosphatidylcholines (LysoPCs) [(LysoPC (16:0) and LysoPC (18:0)] and free fatty acids (FFAs) (arachidonic acid, linoleic acid, oleic acid, and palmitic acid) are significantly increased in the H1 group and decreased in the HPC group, whereas PC demonstrated the opposite trend. These alterations may be putatively ascribed to the activation of phospholipase A2 (PLA2), which mediates the release of lysoPCs and specific fatty acids from PC. Reportedly, PLA2 was significantly increased under hypoxia, whereas decreased gradually to a normal level with the formation of HPC. This result indirectly substantiated the results of the present study (Liu et al., 2006). Our study indicated that the phospholipid metabolic pathway may serve a principal role in HPC.

Our result demonstrated that sphingosine was significantly decreased in the H1 and HPC groups, with the lowest levels in the HPC group. It is conceivable that this alteration may be ascribed to the activation of sphingosine kinase (SphK), which catalyzes the conversion of sphingosine to sphingosine phosphate-1. Reportedly, the activity of SphK2 was significantly increased with the formation of HPC. This result indirectly substantiated the result of the present study (Wacker et al., 2009, 2012). Thus, we speculate that the sphingolipid metabolic pathway may play an important role in HPC and the bioactive components of this pathway may be suitable preventive and therapeutic targets for protecting the brain tissue against hypoxic injury.