1H NMR metabolomics analysis of glioblastoma subtypes: correlation between metabolomics and gene expression characteristics

Abstract

Glioblastoma multiforme (GBM) is the most common form of malignant glioma, characterized by unpredictable clinical behaviors that suggest distinct molecular subtypes. With the tumor metabolic phenotype being one of the hallmarks of cancer, we have set upon to investigate whether GBMs show differences in their metabolic profiles. (1)H NMR analysis was performed on metabolite extracts from a selection of nine glioblastoma cell lines. Analysis was performed directly on spectral data and on relative concentrations of metabolites obtained from spectra using a multivariate regression method developed in this work. Both qualitative and quantitative sample clustering have shown that cell lines can be divided into four groups for which the most significantly different metabolites have been determined. Analysis shows that some of the major cancer metabolic markers (such as choline, lactate, and glutamine) have significantly dissimilar concentrations in different GBM groups. The obtained lists of metabolic markers for subgroups were correlated with gene expression data for the same cell lines. Metabolic analysis generally agrees with gene expression measurements, and in several cases, we have shown in detail how the metabolic results can be correlated with the analysis of gene expression. Combined gene expression and metabolomics analysis have shown differential expression of transporters of metabolic markers in these cells as well as some of the major metabolic pathways leading to accumulation of metabolites. Obtained lists of marker metabolites can be leveraged for subtype determination in glioblastomas.

FIGURE 1.

NMR spectra of five biological replicates for nine glioblastoma cell lines studied in this work. The good consistency among replicates is apparent from spectral traces. Only the spectral region between 0.5 and 9 ppm is shown. Spectral points in the region between 2.1 and 2.2 ppm contain residual hydrogen-containing solvent and are therefore removed, as well as the region of 4.5–5 ppm, which is affected by water suppression.

FIGURE 2.

Spectra of metabolites used for multivariate linear regression analysis of glioblastoma spectra. Forty-one metabolites used in the analysis included all metabolites previously determined in NMR measurements of hydrophilic glioblastoma samples as well as samples of other cell lines. One-dimensional spectra of all 41 metabolites are shown in this figure along with the outline of the average spectrum for glioblastoma cell lines. Complete spectra of all metabolites were used in multivariate linear regression analysis.

FIGURE 3.

Principal components analysis of spectral data for nine GBM cell lines. Metabolites were independently extracted and measured for five biological replicates corresponding to nine cell line types. The grouping of several cell types is apparent. Comparison between PCA results of spectral and quantified metabolic data is shown in supplemental Figs. 1 and 2.

FIGURE 4.

FKM and HCL clustering of spectral data. A, HCL result for cell samples. B, FKM determined membership values for each measurement, where red represents the membership value of 1, and dark blue corresponds to membership of 0. Higher membership value indicates stronger belonging to a cluster. FKM was calculated with m = 1.8. Comparison between FKM clustering of spectral and quantified metabolic data is shown in supplemental Fig. 3.

FIGURE 5.

HCL clustering of quantitative metabolite data obtained using Levenberg-Marquardt multivariate linear regression method with spectral measurements for 41 metabolites from metabolomics databases with metabolite values normalized (divided by standard deviation and mean-centered). Sample types are grouped similarly, based on these quantitative metabolic data, as they were with spectral data in Figs. 3 and 4.

FIGURE 6.

HCL clustering of metabolites selected as most differentially concentrated between the four groups. The feature selection was performed from quantitative metabolite data using SAM analysis. Prior to SAM analysis, metabolite values were normalized (divided by standard deviation and mean-centered). SAM Δ value was 0.2 with zero median number of false significant metabolites. Outlined are sample and metabolite groups, which show the metabolites that are most significant for separation of samples for each group from all the other groups.

FIGURE 7.

Molecular footprints of the nine GBM cell lines assessed in the current study. Selected transcripts of interest, EGFR, PDGFRA, NF1, and IDH1, were amplified in the GBM cell models by RT-PCR.

FIGURE 8.

Direct connection network between gene expression and metabolites that are overconcentrated in group 1. A, genes directly involved in the transport of choline (particularly SCL44A1) are overexpressed in group 1 cells. In the figure, genes that are overexpressed in Group 1 relative to the other groups are shown in red, and the ones that are underexpressed in group 1 are shown in blue. B, relation between inositol and its transporter genes (SLC5A3 and SLC2A13) as well as enzyme involved in its digestion (inositol transferase). In subplots B1–B4, coloring is used to describe gene expression difference across four cell line types. Subplot B1 shows expression of genes in group 1 relative to groups 2, 3 and 4; subplot B2 shows expression of genes in group 2 relative to groups 1, 3, and 4; subplot B3 shows expression of genes in group 3 relative to groups 1, 2, and 4; and subplot B4 shows expression of genes in group 4 relative to groups 1, 2, and 3. Inositol is overconcentrated in group 1, and this can be related to the overexpression of its transporters as well as underexpression of the digestion enzyme. CALCA, calcitonin-related polypeptide alpha; CNTF, ciliary neurotrophic factor; NISCH, nischarin; NTS, neurotensin; APP, amyloid beta (A4) precursor protein; CDP-DG, cytidine diphosphate-diacylglycerol.

FIGURE 9.

Direct connection network between gene expression and metabolites that are overconcentrated in some of the groups. Included are l-glutamine (overconcentrated in groups 1 and 3), l-glutamate (overconcentrated in groups 1, 2, and 3), l-aspartate (overconcentrated in groups 1, 2, and 3), and citrate (overconcentrated in groups 1, 2, and 3) with metabolite overconcentration highlighted in red. Genes are colored based on their expression in one group relative to all the others, i.e. panel 1 is expression in 1 relative to 2, 3, and 4; panel 2 is expression in 2 relative to 1, 3, and 4; panel 3 is expression in 3 relative to 1, 2, and 4, and panel 4 is expression in 4 relative to 1, 2, and 3. According to the changes in metabolite concentration and gene expression, we are hypothesizing major transporters for each metabolite listed, and this is outlined with red arrows on the graph. PC, phosphorylcholine.

FIGURE 10.

Partial representation of triacylglycerol and glycerophospholipid metabolism related to synthesis and digestion of glycerol 3-phosphate. Glycerol 3-phosphate is overconcentrated in group 4 (highlighted in red). The gene colors represent relative expression levels in: panel 1, 2, group 1 relative to groups 2, 3, and 4 and group 2 relative to groups 1, 3, and 4 (same relative expression); panel 3, group 3 relative to groups 1, 2, and 4; and panel 4, group 4 relative to groups 1, 2, and 3, where red shows overexpression and blue shows underexpression. The branch of the triacylglycerol pathway is circled. The presented genes are: a, monoglyceridase; b, 2-lysophosphatidylcholine acylhydrolase; c, glycerate kinase; d, glycerophosphate transacylase. Glycerol 3-phosphate is overconcentrated in group 4 cells, and this can be related to up-regulation of 2-lysophosphatidylcholine acylhydrolase and glycerate kinase, both involved in glycerol 3-phosphate synthesis, and down-regulation of glycerophosphate transacylase, which is involved in its digestion. LPA, lysophosphatidic acid.