Combined transcriptomic and lipidomic analysis reveals aberrant lipid metabolism in central nervous system hemangioblastomas

Peritumoral cysts are commonly detected in the central nervous system tumors, especially hemangioblastomas (HBs). However, the molecular mechanisms driving their formation and propagation are still unknown. We conducted an integrated lipidomics and transcriptomics analysis on solid and cystic HB samples in order to elucidate the changes in the lipid profile and expression of lipid metabolism-related genes during cyst formation. Transcriptomic analysis revealed differential expression of several genes between the solid and cystic HBs, and those associated with lipid metabolism, such as ADCY4, MGLL, ACOT2, DGKG, SHC1 and LPAR2, were markedly dysregulated in the cystic HBs. The lipidomic analysis further showed a significant reduction in the abundance of triacylglycerol, ceramide, lysophosphatidylcholine and lysophosphatidylethanolamine, and an increase in phosphatidylcholine and phosphatidylethanolamine levels in the cystic HBs. Furthermore, bioinformatics analysis revealed altered lipid biosynthesis, glycerophospholipid metabolism and phospholipase activity in the cystic HBs. Taken together, our findings indicate that cyst formation in HBs is related with aberrant lipid metabolism.


Results
Clinical features of HBs patients. Eleven patients including 6 males and 5 females with mean age of 47.9 years (31-70 years) were recruited in the study. The mean duration of symptoms, including dizziness, headache and vomiting, was 5.15 months (range 0.3-12 months). Two patients had genetic and/or clinical evidence of VHL disease, while the others had a sporadic disease. The general features of the patients are summarized in Table 1. Based on the Gd-enhanced MR images and intraoperative findings, 8 HBs showed a macroscopic cystic pattern (Fig. 1a-c), while solid pattern was observed in 3 HBs (Fig. 1d-f). HE and immunohistochemical staining showed extensive vascularization and presence of stromal cells in the cystic HBs (Fig. 1g-l). Numerous CD31 + thin-walled microvessels were seen, indicating reactive angiogenesis (Fig. 1h). All tumors were positive www.nature.com/scientificreports/ for neuron-specific enolase (NSE) and negative for GFAP (Table S2). In addition, Ki67 staining demonstrated mitosis in some cells, and an overall low proliferation rates of 1-30% (Table S2). Immunohistochemical staining results are summarized in Table S2.

Solid and cystic HBs show different gene expression profiles.
To determine the lipidomic alterations between solid and cystic HBs, we assessed the expression of genes involved in lipid metabolism (Fig. 2a), and identified 3334 DEGs between the two groups on the basis of fold change ≥ or ≤ 2 and p = 0.05 (Fig. 2b), of which 2190 were up-regulated and 1144 were down-regulated in the cystic HBs (Supplementary Table S3). Furthermore, supervised hierarchical cluster analysis showed that the DEGs were able to distinguish between the cystic and solid tumors (Fig. 2c).
The DEGs between solid and cystic HBs are related to lipid metabolism. The DEGs were functionally annotated using GO analysis (Supplementary Table S4). The significantly enriched BP terms among the upregulated genes in the cystic group were "blood vessel development", "vasculature development" and "blood vessel morphogenesis", "synaptic transmission", "nervous system development" and "cell-cell signaling" in the solid group (Fig. 3a). In addition, ten pathways related with lipid metabolism were up-regulated and 9 were down-regulated in the cystic group (Fig. 3b). KEGG pathway mapping was also conducted for the DEGs to categorize their functions, and the top 10 significant pathways in the up-and downregulated genes are shown in  Table S5. Among the pathways related to lipid metabolism, "Phospholipase D signaling pathway" (28 genes; fold enrichment = 1.56; p < 0.05) was strongly enriched in the cystic group, whereas "Phosphatidylinositol signaling system" (12 genes; fold enrichment = 2.04; p < 0.01), "Regulation of lipolysis in adipocytes" (8 genes; fold enrichment = 1.89; p < 0.05), "Sphingolipid signaling pathway" (13 genes; fold enrichment = 1.75; p < 0.05), "Biosynthesis of unsaturated fatty acids" (4 genes; fold enrichment = 1.42; p < 0.05) and "Fatty acid elongation" (4 genes; fold enrichment = 1.30; p < 0.05) were enriched in the solid group (Fig. 3d). Furthermore, the transcriptomic data revealed distinct expression patterns of the genes related to lipid metabo- www.nature.com/scientificreports/ lism between the solid and cystic HBs (Fig. 3e). Finally, the expression of 6 DEGs related to lipid metabolism, including ADCY4, MGLL, ACOT2, DGKG, SHC1 and LPAR2, were confirmed by qRT-PCR (Fig. 2d). The physiological function of SCH1 is regulation of vascular endothelial functions 21 . To determine the functional significance of SHC1 in the cystic HBs, the migration ability of HUVECs was examined with SHC1 overexpression. The microscopic pictures revealed the presence of migration characteristics with SHC1-OE (Fig. 2e). The migration ability of HUVECs was also significantly enhanced by SHC1 overexpression compared to control (Fig. 2e). Taken together, cyst formation in HBs is accompanied by dysregulation of genes associated with lipid metabolism.
Solid and cystic HBs have distinct lipidomes. The lipidomic profiles of solid and cystic HBs were analyzed by UPLC-Q-TOF-MS, and the orthogonal partial least-squares discriminant analysis (OPLS-DA) plots showed significant metabolic differences between the two (Fig. 4a,b). While the relative abundance of TG, Cer, LysoPC and LysoPE was significantly reduced in the cystic HBs, that of PC and PE were increased (*p < 0.05, **p < 0.01, Fig. 4c). However, the abundance of DG, PA, PG, PI, PS, SM, CE and MG were similar in both groups (p > 0.05, Fig. 4c). In addition, 19 lipid compounds were significantly different between the cystic and solid HBs . Although the total TG levels were significantly decreased in the cystic HBs (**p < 0.01, Fig. 4c), that of individual TGs were similar in both, and the potential target "Glycerophospholipid metabolism" pathway was strongly enriched in the cystic group (6 hits; p = 4.39*10 -6 ).

Discussion
HBs are benign tumors that are frequently associated with peritumoral cysts and are found throughout the CNS and can be the source of significant neurological morbidity and mortality 22 . However, the mechanisms underlying cyst formation are still poorly understood, which has hindered the development of non-surgical therapeutic strategies. HBs are an ideal tumor model to investigate the molecular mechanisms of cyst formation within the nervous system. The altered lipid metabolism and its regulatory mechanism in HB genesis and progression is of critical importance, especially in the context of cyst formation. The accompanying lipidomic changes entail a complex interplay between metabolic and transcriptional signaling pathways. Therefore, we conducted a largescale systematic omics analysis to compare the lipidomes and transcriptomes of solid and cystic CNS HBs. Transcriptomic analysis revealed that several genes were differentially expressed between the cystic and solid HB samples. The cystic HBs were characterized by up-regulation of genes regulating angiogenesis, which is also a key driver of tumor growth and development 23 . Since HBs are highly vascularized tumors, they are susceptible to the VEGF-induced increase in vascular permeability compared to vessels in the healthy brain 11,18 . Overexpression of VEGF has been shown in HBs and other human tumors 24 . In addition, endothelial cells respond to VEGF-A, which promotes endothelial cell proliferation and migration 25 . The Notch pathway intersects with VEGF-A signaling to coordinate endothelial cell behaviors 26 . The DEGs identified in the RNA sequence analysis included several that are directly involved in cyst formation. AQP1, an integral plasma membrane protein with six transmembrane domains, is up-regulated in cystic HBs 27 as well as in stromal cells 28 . Additionally, previous study demonstrated epidermal growth factor receptor (EGFR) overexpression and activation in CNS HBs 29 . Expression of galectin-3 was correlated with the expression of VEGF in the development of hemangioblastoma 30 , which is a diagnostic marker and attractive molecular targets for CNS HBs. Furthermore, the DEGs were enriched in "Phospholipase D signaling pathway", and showed a decrease in "Regulation of lipolysis in adipocytes", "Sphingolipid signaling pathway", "Biosynthesis of unsaturated fatty acids" and "Fatty acid elongation", strongly suggesting that lipid metabolism plays an important role in cyst formation.
Fatty acids are critical components of cellular membrane biosynthesis. Most tumors show aberrant high rates fatty acid synthesis, elongation and desaturation that promote cancer cell proliferation 31  . In addition, the expression for fatty acid metabolizing enzymes like acyl-CoA thioesterases 2 (ACOT2) was down-regulated in cystic HBs, ACOT is involved in fatty acid degradation 35,36 , and its decreased levels affect the fatty acid reserves, which are alternative sources for energy production and GPs and SLs biosynthesis. A previous study showed that inhibition of fatty acid synthase (FASN) blocks HIF-1α/VEGF-A signaling in response to hypoxia, and suppresses neovascularization in glioma by upregulating VEGF165b 37 . In the present study however, FASN expression was similar in both cystic and solid HB, indicating that the higher fatty acid levels in cystic HBs are likely due to increased fatty acyls of SLs and SPs, which also promote neovascularization via VEGF signaling.
Neural cell membranes consist of glycerophospholipids that have both structural and functional roles 38 . Changes in glycerophospholipid levels affect cellular functions, endocytosis and exocytosis, cytoskeleton regulation and membrane fusion 39 . In addition, neural membrane glycerophospholipids are significantly altered in brain tumors, cognitive disorders, and neurodegenerative diseases such as Alzheimer's disease, autism and schizophrenia [40][41][42] . Our results suggest that aberrant levels of lipids required for glycerophospholipid metabolism likely drive cyst formation in HBs, which was also supported by the transcriptome data.
Phosphatidylinositol 3-kinase (PI3K) is a vital regulator of multiple signaling cascades and activates the downstream targets Akt/mTOR, which is activated by the growth factors VEGF 43 . Consistent with this, the genes involved in the PI3K signaling pathway were also dysregulated in our study. The cystic HBs showed increased expression of inositol polyphosphate-4-phosphatase type II B and phosphoinositide-3-kinase regulatory subunit 5, and decreased levels of inositol 1,4,5-trisphosphate receptor type 1, inositol polyphosphate-4-phosphatase type I A, myotubularin related protein 1, inositol polyphosphate-5-phosphatase J, phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit beta and diphosphoinositol pentakisphosphate kinase 1. In contrast, the total PI content was similar in both HB types, while the significantly different PI species were mostly at low abundance in the cystic HBs as per the lipidomics results. The occurrence of cystic HBs is frequently associated with molecular pathway changes in epidermal growth factor receptor and PI3K/Akt/mTOR pathways 44 . However, the exact role of PI3K signaling in cystic HBs formation still needs to be elucidated.
In conclusion, cyst formation in CNS tumors is associated with aberrant lipid biosynthesis, glycerophospholipid metabolism and phospholipase activity. Our results give novel insights into the cyst formation in HBs in the context of aberrant lipid metabolism.

Materials and methods
Human tissue samples collection. Surgically resected brain tissue samples were collected from 11 HB patients-8 cystic and 3 solid tumor cases-from the West China Hospital of Sichuan University. The following clinical information was also collected for each patient: age, gender, duration of illness, symptoms, tumor characteristics, location, size and VHL disease (Table 1). In vitro scratch wound assay. HUVECs and SHC1-OE cells were plated in equal numbers in 6-well tissue culture plates (3 × 10 5 cells/well) to achieve 90% confluence. Thereafter, a vertical wound was created using a 200 μL pipette tip within a certain area. Scraped cells were removed by washing the monolayer twice with PBS. The cells were incubated in DMEM containing with 1% FBS at 37 °C in 5% CO 2 for 72 h. The area of the cellfree wound was recorded with microscopy at 0, 24, 48 and 72 h. Measurement of scratch wound was assessed by using ImageJ software. RNA isolation and High-throughput sequencing. Total RNA was isolated from 3 samples each of solid and cystic tumors using the TRIzol reagent (Invitrogen, Carlsbad, CA, USA). The RNA concentration was measured using a NanoDrop ND-2000 instrument (Thermofisher Scientific), and its integrity was evaluated by denaturing agarose gel electrophoresis.

Statements. The study was approved by
Briefly, the rRNAs were removed from the total RNA mix using Ribo-Zero rRNA Removal Kit (Illumina, San Diego, CA, USA), and the RNA-seq libraries were constructed using the TruSeq Stranded Total RNA Library Prep Kit (Illumina, San Diego, CA, USA). Quality control of the libraries was performed using Agilent 2100 Bioanalyzer (Agilent, Santa Clara, CA) and sequenced using an Illumina HiSeq Sequencer with a 150 bp paired-end run.

Bioinformatics analysis. The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes
(KEGG) pathway enrichment analyses (https ://www.genom e.jp/kegg, accessed on December 13th 2018), were performed for the differentially expressed genes (DEGs) to interpret their biological functions 46,47 . The top 10 enriched GO terms and the significant KEGG pathways were identified on the basis of −log10 p-value.
Quantitative reverse transcription-PCR (qRT-PCR) analysis. Randomly selected genes were validated by qRT-PCR. Total RNA was reverse transcribed into cDNA using a kit (Invitrogen, Carlsbad, CA, USA), and q-PCR was performed on a Bio-rad CFX Connect Real-Time PCR System. The relative expression levels of target genes were normalized to gapdh using the 2 −∆∆Ct method. All primers are listed in Supplementary Table S1.
Accession numbers. RNA-seq data are available at the Gene Expression Omnibus (GEO) database repository under the accession number GSE148216.
Lipid extraction and LC-MS lipid metabolite analysis. Total lipids were extracted from ~ 10 mg frozen tissue using methyl-tert-butyl ether (MTBE) as previously described 48 . Briefly, 150 µl ice-cold methanol was added to each brain tissue sample (25-30 mg) followed by 450 µl MTBE, and then mixed gently. The samples were incubated at 4 °C for 10 min on an orbital shaker, and homogenized by ultra-sonication for another 10 min on an ice bath. To separate the lipid and aqueous phases, 300 μl of 25% methanol was added to the homogenates