Glycerophosphatidylcholine PC(36:1) absence and 3′-phosphoadenylate (pAp) accumulation are hallmarks of the human glioma metabolome

Glioma is the most prevalent malignant brain tumor. A comprehensive analysis of the glioma metabolome is still lacking. This study aims to explore new special metabolites in glioma tissues. A non-targeted human glioma metabolomics was performed by UPLC-Q-TOF/MS. The gene expressions of 18 enzymes associated with 3’-phosphoadenylate (pAp) metabolism was examined by qRT-PCR. Those enzymes cover the primary metabolic pathway of pAp. We identified 15 new metabolites (13 lipids and 2 nucleotides) that were significantly different between the glioma and control tissues. Glycerophosphatidylcholine [PC(36:1)] content was high and pAp content was significantly low in the control brain (p  < 0.01). In glioma tissues, PC(36:1) was not detected and pAp content was significantly increased. The gene expressions of 3′-nucleotidases (Inositol monophosphatase (IMPAD-1) and 3′(2′),5′-bisphosphate nucleotidase 1(BPNT-1)) were dramatically down-regulated. Meanwhile, the gene expression of 8 sulfotransferases (SULT), 2 phosphoadenosine phosphosulfate synthases (PAPSS-1 and PAPSS-2) and L-aminoadipate-semialdehyde dehydrogenase-phosphopante-theinyl transferase (AASDHPPT) were up-regulated. PC(36:1) absence and pAp accumulation are the most noticeable metabolic aberration in glioma. The dramatic down-regulation of IMPAD-1 and BPNT-1 are the primary cause for pAp dramatic accumulation. Our findings suggest that differential metabolites discovered in glioma could be used as potentially novel therapeutic targets or diagnostic biomarkers and that abnormal metabolism of lipids and nucleotides play roles in the pathogenesis of glioma.

the possibility that glutamate dehydrogenase2(GDH2)specific inhibition may serve as a viable therapeutic strategy for gliomas with IDH mutations 10 .
To date, apart from 2-HG, many metabolites associated with glioma have been reported [11][12][13][14] . Nevertheless, the number of significant metabolites discovered in glioma tissues is limited. Therefore, in this study, we prepared the polar and non-polar solvent extracts obtained from glioma and healthy control tissues and performed a non-targeted tissue metabolomics analysis. Non-targeted analysis of biomolecules still is an important trend in the field of "omics" including proteomics, genomics, transcriptomics and metabolomics 15 . In order to get unbiased metabolite profiling of glioma, we used the analysis procedure of discovery set and validation set 16 .
We used UPLC-Q-TOF/MS to compare the metabolomics profiling of glioma tissues and control brain parenchyma. Further, we examined the gene expressions of 18 enzymes associated with pAp metabolism by quantitative real-time polymerase chain reaction (qRT-PCR). These 18 enzymes cover primary metabolic pathways of pAp.

Results
Metabolomics profiling of glioma and control tissue in the discovery set. The discovery set included 13 control specimens and 33 glioma specimens. Analysis of PCA and OPLS-DA in discovery set is shown in Fig. 1.
The representative base peak intensity chromatograms and PCA analysis of the glioma and control tissues are shown in Fig. 1A,B. In order to explore the inherent grouping between glioma patients and healthy control subjects, PCA was used to map the samples based on their spectral profile without previous knowledge of the class. Score plot demonstrated that the metabolic profiles of glioma patients deviated from the control samples suggesting that significant biochemical changes must have occurred in glioma tissues. Next, an OPLS-DA model was employed to refine the separation results obtained by PCA. Indeed, the score plot resulting from the OPLS-DA model showed a superior separation between the glioma and control tissues (Fig. 1C). The corresponding S-plot indicated that the differential metabolites with VIP value ≥ 1 were responsible for discriminating glioma and control samples (Fig. 1D).
The relative abundances of the identified metabolites in the discovery set were shown in Fig. 2. Among the discovered metabolites, we observed two distinct classes: purine nucleotide(pAp and Appr > p) and lipids (glycerolipids, sterol lipids, prenol lipids, phospholipids and sphingolipids). Our results demonstrated that Glycerophosphatidylcholine PC(36:1) was not detected in the glioma tissues, but its level was significantly high in the control specimens (p < 0.01, Fig. 2 Table 1). Similarly, the levels of pAp and ADP-ribose 1″-2″ cyclic phosphate(Appr > p) were obviously higher in glioma tissues (p < 0.01, Fig. 2, Table 1).

Metabolomics profiling of glioma and control tissue in the validation set.
To validate the observed findings in discovery set, 7 control brain parenchyma samples, 26 glioma specimens were collected and analyzed under the same analytical procedures (Fig. 3).
The relative content of the 15 distinctive metabolites in the validation set is shown in Table 2. In addition, statistical analysis demonstrated that the content difference of each metabolite between the gliomas (both WHO grade III-IV gliomas and the WHO grade II) and control specimens was significant (P < 0.01, Table 2). On the other hand, no statistically significant difference was observed between the WHO grades II and III-IV gliomas (P > 0.05).
Taken together, results obtained from both discovery and validation sets demonstrated that the PC(36:1) was an abundant component of the control brain parenchyma, while pAp content was significantly low. In contrast, PC(36:1) was not detected in glioma tissues and the pAp content was remarkably increased. The PC(36:1) absence and pAp accumulation were the most obvious characteristics in glioma metabolomes. Secondly, the abundance of PC(38:4) mass was relatively rich in glioma, but the variation degree of PC(38:4) was lower than that of PC(36:1) (Fig. 2, Table 2).
Gene expression of enzymes associated with pAp metabolism. Metabolomics profiling of glioma tissues demonstrated a notably higher pAp content (Fig. 2, Table 2). pAp is a toxic by-product 17 . In order to illuminate the mechanisms that led to the accumulation of pAp within glioma tissues, 18 enzyme genes related to pAp metabolism were analyzed by qRT-PCR. A comprehensive list of the analyzed genes was mentioned in Table 3.
The gene expression profiling of 18 enzymes (using GAPDH as internal control) are shown in Fig. 4.

Discussion
Glioma is a common malignant brain tumor (1). To date, although many potential biomarkers have been found, there are still no effective therapeutic or diagnostic targets in clinical practice 9,[19][20][21] .
This study revealed obvious variation of 8 novel phospholipids in glioma and identified their precise molecular structures (Tables 1 and 2 3) and lyso-PE(18:0)] in glioma were higher than that in control samples. On the contrary, glycerophosphatidylcholine PC(36:1) was not detected in glioma tissues, but was abundant in the control brain parenchyma. Limited information is avaiblable for the aberrant phospholipid metabolisms in glioma tissues 19,22 . Gliomas harboring the IDH1-R132H mutation demonstrated altered phospholipid metabolism characterized by decreased phosphoethanolamine levels and increased glycerophosphocholine levels 19 . However, Pavithra Viswanath demonstrated both phosphoethanolamine and phosphocholine levels were reduced in IDHmut glioma cells and in IDHmut gliomas in orthotopic tumor   ratio could be a negative predictive marker for bevacizumab efficacy 23 . Those gliomas may represent a malignant phenotype that could resist the anti-VEGF treatment. Therefore, the metabolic changes of phospholipid metabolites can serve as useful biomarkers. The PC(36:1) was a very abundant composition in control brain parenchyma, which consists of one chain of vaccenic acid at the C-1 position and one chain of stearic acid at the C-2 position (HMDB 2018). It is a key component of the lipid bilayer of cells, as well as being involved in metabolism and signaling. PC (36:1) also is a common component in the plasma 24 . Therefore, the "complete" absence of PC(36:1) in glioma tissues is indeed a noticeable abnormality. The experiment result did not demonstrate a relationship between PC(36:1) absence and the glioma grade, it seems that PC(36:1) absence could result from a specific "mutation" or specific metabolic reprogramming that occurs in gliomas. However, the exact mechanisms of PC(36:1) absence in glioma tissues need to be explored in future studies.
In this study we observed that the content of DG(33:3) was increased in glioma. The analysis of neutral glycerolipid content proved useful in distinguishing gliomas from metastatic brain tumors by magnetic resonance spectroscopy (MRS) 25 . Chemically, eight lipid categories are reported, namely, fatty acyls (FA), glycerolipids, saccharolipids, polyketides, sterol lipids, prenol lipids, sphingolipids and phospholipids (PLs) 15 . Our experiment demonstrated that 5 lipid categories (13 lipids species), namely, glycerolipids, prenol lipids, cholesterol lipids, phospholipids (including PC and PE), sphingolipids (Ceramide and Sphingomyelin) were abnormal in glioma tissues. Therefore, it is plausible to speculate that abnormal metabolism of lipids could be a metabolomic phenotype in molecular pathology of glioma.
Lipidomics is an exciting new area 26 . It has been widely applied in several disciplines, including neuroblastoma 27 , mammalian cancer 28 , prostate cancer 29 , tetrahydrocannabinol addiction 30 , Alzheimer's Disease 31 and drug development 32 .
The experiment revealed that the level of pAp was significantly increased in the examined gliomas, whereas the content of pAp was very low in the control brain parenchyma. The VIP and fold change of pAp content were 7.12 and more than 500 times, respectively (Fig. 2, Table 2). pAp is a toxic by-product. It is a strong inhibitor for many enzymes. pAp inhibits PARP-1 [poly-(ADP-ribose) polymerase-1], which in turn results in abolished DNA  repair, destruction of the chromatin structure, alteration in DNA methylation and cell death 17,33 . pAp can also inhibit nucleoside diphosphate kinase and oligoribonucleases. Deficiency of nuclease function results in the accumulation of immature ribosomal 5.8S RNA, which leads to the impairment of ribosome biogenesis 34,35 .
In mammalian cells, pAp primarily produced by the transfer of a sulfate group from PAPS to various acceptor molecules. This sulfate transfer reaction is catalyzed by a number of sulfotransferases (p-2 in Fig. 4). Apart from sulfur metabolism, pAp can also be generated during the transfer of the 4-phosphopantetheine group from CoA to acyl carrier protein of FAS, catalyzed by AASDHPPT protein in fatty acid synthesis 17 (p-1 in Fig. 4). pAp is recycled into AMP and Pi by 3′-Nucleotidases 18,36,37 (p-3 in Fig. 4). The exact mechanism of pAp level increase in glioma tissues is not clear till now. Therefore, we carried out the gene expressions of 18 enzymes associated with pAp metabolism to uncover the cause at transcription level. The up-regulation of phosphopantetheine transferase (AASDHPPT) and sulfotransferases (NDST2, HS3ST2, CHST3, CHST7, CHST12, CHST1, TPST1 and TPST2) were directly responsible for the increase of pAp production. In humans both PAPSS1 and PAPSS2 possess ATP sulfurylase and APS kinase activity to synthesize PAPS 38,39 . As an active sulfate donor, PAPS is required by all sulfotransferases 39 . Therefore, PAPSS-1 and PAPSS-2 provided abundant PAPS substrate for sulfotransferases which leads to the increase of pAp production in gliomas.
The gene expressions of IMPAD-1 (gPAPP) and BPNT-1 were remarkably down-regulated in glioma, especially the former (more than 90% reduction). IMPAD-1 hydrolyzes the Golgi pAp yielding AMP and phosphate,  Table 3. Catalytic pathways of 18 enzymes: 1. P-1: activation pathway of FACcyt in fatty acid synthesis, P-2: sulphonation reaction pathway, P-3: pAp hydrolysis pathway and P-4: PAPS synthesis pathway. 2. Human genome encodes one phosphopantetheine transferase (AASDHPPT), two 3′-nucleotidases (IMPAD1 and BPNT-1), two PAPS synthase (PAPSS1 and PAPSS2) and a number of sulfotransferases (SULT). A total of 13 sulfotransferases divided into 4 types according to substrates and products of enzymes. Refer to Table 3 BPNT-1 hydrolyzes cytoplasm pAp to AMP and phosphate 18,36,37 . Therefore, it is reasonable to speculate that the deficiencies of IMPAD-1 and BPNT-1 function would lead to a lower degradation rate of pAp and hence promote its accumulation in glioma. Although Bpnt-1 and IMPAD-1 share a common substrate, their distinct subcellular localization suggests that they play unique roles in the cell 18,37 . Loss of BPNT-1 function results in liver failure,whole-body edema and death, while the levels of bisphosphorylated nucleotides, pAp and PAPS, were dramatically elevated in the BPNT-1 mutant liver and neurons 18,36 . Data showed that both gene expression of IMPAD1 and Bpnt-1 exist in human brain, the relative gene expression level of IMPAD1 is higher than that of BPNT-1 in human brain tissues (uniPROT 2018). Inhibition of BPNT-1 causes selective neuronal dysfunction 36 . IMPAD1 inactivation in mice and human produces neonatal lethality, lung abnormalities resembling atelectasis, dwarfism or chondrodysplasia with abnormal joint formation and impaired proteoglycan sulfation 18,36,37,40 . A recent study demonstrated that deletion of BPNT-1 leads to accumulation of pAp, causing iron deficiency anemia 41 . To sum up, pAp accumulation may be a cause for molecular pathology of glioma.
Finally, the content differences of metabolites samples with different WHO grades did not observed. We speculated that the lacking difference maybe relate to specimen location in glioma tissues. The specimen pieces were commonly taken from the area without necrosis or hemorrhage of glioma in order to obtain histology diagnosis, whereas necrosis or hemorrhage is general characteristics of high -grade glioma. Further, this observation needs more case researches in the future.

Conclusion
PC(36:1) absence and pAp accumulation were the most obvious metabolic abnormality in glioma. The gene expressions IMPAD-1 and BPNT-1 were dramatic down-regulated, which were the most primary causes for the pAp accumulation. PC(36:1) absence and pAp accumulation can be new clues to explore the molecular pathogenesis of glioma, further they may be used as potential therapeutic targets or biomarkers. Our findings suggest that abnormal metabolism of lipids and nucleotides play roles in the pathogenesis of glioma. To the best of our knowledge, this is the first study that reports the PC(36:1) absence and pAp accumulation in human glioma tissues.

Methods
Participant selection. 59 adult glioma patients and 20 control subjects were enrolled in this study. All of participants did not suffer hepatic disease, adiposis and endocrine disease (diabetes, thyroid disease, adrenal cortex disease and so on).
Tumor samples were obtained from gliomas during craniotomy. Gliomas were diagnosed and graded according to the WHO classification system. Control brain tissues were obtained from grossly normal brains of control subjects undergoing various surgical procedures that included cerebral injury and cerebral hemorrhage. Tissue specimens for the metabolomics study were immediately placed in cryogenic vials and snap-frozen in liquid nitrogen until component extractions. Other portions of the glioma and control brain tissues were stored in liquid nitrogen until RNA analysis.
The protocol for this study was revised and approved by the ethics committee of the First Hospital of Jilin University, Changchun, China. A written informed consent was obtained from all participants. All experimental procedures were carried out in accordance with the approved guidelines.

Sample preparation of polar and non-polar solvent extracts.
A two-step extraction procedure was carried out as described previously with some modification 42 . A total of 1500 μL chilled dichloromethanemethanol (3:1, v/v) was homogenized with 50 mg brain tissue in an ice-bath with a homogenizer (IKA, Staufen, Germany) for 30 seconds. The resulting homogenate was centrifuged at 12000 rpm for 25 minutes at 4 °C. Next, the supernatant was transferred to a fresh tube and used as the non-polar solvent extract. The residue was homogenized with 1500 μL methanol-water (1:1, v/v) as mentioned above, the supernatant was transferred to another fresh tube and used as the polar solvent extract. Subsequently, both fractions were dried under nitrogen, and stored at −80 °C until liquid chromatography/mass spectrometry (LC/MS) analysis.

Non-targeted metabolomics analysis by UPLC-Q-TOF/MS. UPLC-Q-TOF/MS parameter.
Chromatographic separation was performed on an Acquity ultra performance liquid chromatography (UPLC) Bridged Ethyl Hybrid (BEH) C18 column (2.1 mm × 100 mm, 1.7 μm, Waters Corp., Milford, USA) using Waters Acquity TM UPLC system. The dried non-polar and polar solvent extracts was respectively dissolved in 500 μL of acetonitrile-water (1:1, v/v) and 500 μL of methanol-water (1:1, v/v), then combined, centrifuged at 13000 rpm for 20 min at 4 °C and 5 μL of the resulting mixture was injected into the UPLC system for analysis by BEH C18 column. The column was maintained at 40 °C and eluted at a flow rate of 0.45 mL/min, using a mobile phase of (A) 5% acetonitrile in water (by volume) and (B) 95% acetonitrile in water (by volume). The gradient program was optimized as follows: 0-1 min, 10% B to 50% B; 1-6 min, 50% B to 95% B; 6-8 min, 95% B to 98% B; 8-12 min, 98% B to 100% B and a final 12-14 min equilibration with 10% B. The eluent from the column was directed to the mass spectrometer without splitting. RNA Isolation and qRT-PCR analysis. The total RNA was extracted from 50 mg tissues using TRIzol Reagent (Invitrogen) according to the manufacturer protocol. RNA integrity was analyzed on 1.0% agarose gel and the quantity was determined using a NanoDrop 2000C Spectrophotometer (Thermo Scientific) according to the standard protocols. Subsequently, 1 ug RNA was reverse-transcribed with a PrimeScript ™ RT reagent Kit (TaKaRa) for cDNA synthesis and genomic DNA removal was performed in accordance with the manufacturer's protocols. The qPCRs were performed in triplicate sets with a SYBR premix Ex Taq ™ II kit (Tli RNaseH Plus; TaKaRa) in a real-time PCR detection system (TaKaRa) according to the provided instructions. Gene-specific primers were designed using primer-blast (http://www.ncbi.nlm.nih.gov/tools/primer-blast/). The primer sequences were listed in Table 4. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and TATA-box binding protein (TBP) were selected as endogenous controls. The primer sequences are as follows: GAPDH(138 bp), F: GCACCGTCAAGGCTGAGAAC, R: TGGTGAAGACGCCAGTGGA; TBP (132 bp), F: TGCACAGGAGCCAAGAGTGAA, R: CACATCACAGCTCCCCACCA. The specificity of amplification was assessed by dissociation curve analysis, and the relative abundance of genes was determined using the 2 −ΔΔCt method. The mRNA expression levels of 18 genes were compared between glioma tissues (discovery set and validation set, n = 59) and normal brain tissues (discovery set and validation set, n = 20), and all data were expressed as Mean ± standard deviation. The means between the two groups were compared using Student's t-test and p < 0.05 was considered statistically significant.

Statistical analysis.
Multivariate statistical analysis. The raw spectral data were analyzed with MassLynx Applications Manager Version 4.1 (Waters, Manchester, UK). Deconvolution, alignment and data reduction were performed to generate a list of retention times (RT) and mass pairs with a corresponding peak area for all the detected peaks from each file in the data set. The main parameters were set as following: RT range 0.5-14 min; mass range 50-1200; extracted ion chromatograms (XIC) at 0.02 Da mass window; automatic calculation of the peak width and peak-peak base-line noise; the use of raw data during the deconvolution procedure; marker intensity threshold (count) was set at 300; the mass tolerance was 0.02 Da; RT windows were 0.2 s; the noise elimination level was set at 6 to retain the isotopic peaks. The resulting UPLC-MS data were then transferred to SIMCA-P software package (version 12.0, Umetric, Umeå, Sweden). Principal component analysis (PCA) analysis was performed by the Markerlynx 4.1 software to discern the metabolic profiles of glioma and control tissues and the results were determined by the visual inspection of score plots. Supervised models were subsequently performed by orthogonal partial least squares discriminant analysis (OPLS-DA) to maximize the separation between the different classes and identify biomarkers associated with glioma. The results were visualized in the form of score plots and the potential biomarkers were selected based on the variable importance in the project (VIP) value and S-plot.

NO Gene
Primer Sequence (5′->3′) NO Gene Primer Sequence (5′->3′)  Mathematical statistics. Statistical analyses were performed using the statistical software package (SPSS for Windows, version 21.0; IBM-SPSS, Chicago, IL, USA). Box plots were performed using MetaboAnalyst 3.0 43 . The distribution of continuous variables was assessed using the Kolmogorov-Smirnov test. Normally distributed variables were presented as mean ± standard deviation (SD). Non-normally distributed variables were presented as median (IQR). Data were plotted as mean ± standard error of the mean (SEM). Kruskal-Wallistest, one-way ANOVA and independent sample t test were performed to determine the difference between the examined groups. A P value < 0.05 was considered to statistically significant.
Ethics approval and consent to participate. This study was reviewed and approved by the Ethics Committee of First Hospital of JiLin University, and all patients provided written informed consent. All experimental procedures were carried out in accordance with the approved guidelines.