Abstract
Glioblastoma (GBM) is increasingly recognized as a disease involving dysfunctional cellular metabolism. GBMs are known to be complex heterogeneous systems containing multiple distinct cell populations and are supported by an aberrant network of blood vessels. A better understanding of GBM metabolism, its variation with respect to the tumor microenvironment, and resulting regional changes in chemical composition is required. This may shed light on the observed heterogeneous drug distribution, which cannot be fully described by limited or uneven disruption of the blood–brain barrier. In this work, we used mass spectrometry imaging (MSI) to map metabolites and lipids in patient-derived xenograft models of GBM. A data analysis workflow revealed that distinctive spectral signatures were detected from different regions of the intracranial tumor model. A series of long-chain acylcarnitines were identified and detected with increased intensity at the tumor edge. A 3D MSI dataset demonstrated that these molecules were observed throughout the entire tumor/normal interface and were not confined to a single plane. mRNA sequencing demonstrated that hallmark genes related to fatty acid metabolism were highly expressed in samples with higher acylcarnitine content. These data suggest that cells in the core and the edge of the tumor undergo different fatty acid metabolism, resulting in different chemical environments within the tumor. This may influence drug distribution through changes in tissue drug affinity or transport and constitute an important consideration for therapeutic strategies in the treatment of GBM.
Significance: GBM tumors exhibit a metabolic gradient that should be taken into consideration when designing therapeutic strategies for treatment.
See related commentary by Tan and Weljie, p. 1231
Introduction
Glioblastoma (GBM) is a highly malignant and invasive form of brain tumor (1). Typical treatment for GBM only minimally delays the course of the disease and as such there is a substantial need for new therapies and therapeutic targets. The lack of progress in developing therapeutics for GBM is due to a combination of factors including limited and heterogeneous drug delivery across both the blood–brain and the aberrant blood–tumor barriers (BBB and BTB; refs. 2–4). In a previous study by this group, the relationship between drug distribution and drug response in GBM patient-derived xenograft (PDX) models was examined in relation to the breakdown of the BBB and contrast enhancement by MRI (5). A mutation that occurs in 50% of GBMs is amplification of the EGFR (6). The EGFR tyrosine kinase inhibitor erlotinib, which has been tested as a potential treatment for GBM, was mapped in PDX models and was correlated with multimodality data including phospho-proteomics and mRNA sequencing. This demonstrated that, due to a combination of limitations imposed by the BBB and tissue characteristics, the concentration of erlotinib measured in the intracranial tumor, was insufficient to successfully inhibit EGFR tyrosine kinase signaling. During this study an interesting distribution of global lipid signal was highlighted by stimulated Raman scattering microscopy, with higher drug concentrations detected in regions of lower lipid content (5). Regional variation in lipid content caused by differences in metabolism and cellular content could play a role in drug distribution, and hence this relationship, along with an observed decrease in erlotinib concentration measured at the tumor edge is the subject of the study described herein.
Tumor cells are surrounded by the tumor microenvironment; defined as the nontumor cells, molecules, and blood vessels that surround and feed a tumor cell (7). To fulfill the demands of biomass formation, proliferating cells increase their uptake of nutrients and reprogram glycolysis (8). The metabolism of each tumor cell is therefore dependent on the availability of nutrients in the microenvironment. The molecular composition of the microenvironment for each tumor cell varies depending on its location in the tumor mass and proximity to blood vessels (9). Cancer cells are programmed to increase the uptake of glucose, but a smaller proportion of this is oxidized in the Krebs cycle to produce ATP; instead cancer cells use glucose for anabolic processes (10). Fatty acids are an additional energy source, and can be metabolized to produce ATP providing twice as much energy as carbohydrates (10). Recent reports have suggested that ATP production via fatty acid oxidation (FAO) is required for the respiration and proliferation of malignant glioma cells (11); however, much about how different cells within tumor volumes utilize different metabolic pathways remains unknown. While the GBM microenvironment shares some commonalities with other solid tumors, such as immune cells, blood vessels, and the extracellular matrix, the tumor microenvironments of brain tumors also contain microglia and astrocytes and a substantially different metabolic milieu compared with tumors in other organs (7). In addition, a number of metabolic phenotypes exist among GBM; it is therefore necessary to understand intra- and intertumor metabolic heterogeneity.
To understand the association between tumor metabolism and drug efficacy, the spatial relationships between drugs and biomolecules is needed, both within the tumor as well as in the surrounding microenvironment. Matrix-assisted laser desorption/ionization (MALDI) mass spectrometry imaging (MSI) is a label-free technique capable of spatial detection and mapping of hundreds to thousands of molecules (12). MALDI MSI has been used in numerous metabolomic and lipidomic studies of cancerous tissues (13–15), and has also been used for the simultaneous analysis of drugs and drug metabolites with endogenous biomolecules, and potential markers of therapeutic response (16, 17). High mass resolution MALDI MSI of proteins and metabolites in a mouse model of GBM was reported by Dilillo and colleagues (18), where they were able to map numerous proteins and metabolites via different sample preparation techniques, and found metabolic differences between healthy and tumor brain tissues (18).
In this article we used MALDI Fourier transform ion cyclotron resonance (FT ICR) MSI to image lipids and metabolites in four different PDX models of GBM. Differences between normal brain and tumor tissue were apparent, as well as distinctive subregions within the tumor. A series of long chain acylcarnitines was detected with higher intensity in regions corresponding to the tumor edge. We subsequently investigated the distribution of two drugs, erlotinib and AZD1775, in the same PDX models, and mapped them relative to markers of metabolism and tumor vasculature.
Materials and Methods
Animals and tissue sample preparation
All studies involving animals were reviewed and approved by the Institutional Animal Care and Use Committee at Mayo Clinic. For full details of PDX model preparation see Supplementary Data. Briefly, xenografts were established in female athymic nude mice. A subset of mice was dosed with a single dose of erlotinib (100 mg/kg) or AZD1775 (200 mg/kg). Mice were euthanized (dosed animals 2 hours postdose) and brains were harvested and flash frozen on dry ice. Tissues were stored at −80°C prior to analysis. Tissues were placed at −25°C 1 hour before use, and tissue sections were prepared using a Microm HM550 Cryostat (Thermo Fisher Scientific) with the microtome chamber chilled at −21°C and the specimen holder at −19°C. Tissue sections of 12 μm thickness were thaw mounted onto ITO-coated microscopic slides (Bruker) for MALDI MSI and onto glass slides for either liquid extraction surface analysis (LESA) MS-MS or hematoxylin and eosin (H&E) staining. H&E microscopy images were acquired using a Zeiss Observer Z.1 microscope equipped with 20x Plan-APOCHROMAT lens and AxioCam MR3 camera. The mounted tissue sections were dried for 15 minutes in a desiccator prior to matrix application. 2,5-dihydroxybenzoic acid matrix (160 mg/mL solution in methanol/0.1% TFA/DMSO 69.3:29.7:1 v/v/v) was deposited using a TM-sprayer (HTX Imaging) with the following conditions: flow rate, 0.09 mL/minute, spray nozzle velocity, 1,200 mm/minute, spray nozzle temperature, 75°C, nitrogen gas pressure of 10 psi, and track spacing of 2 mm and 4 passes. To investigate regional ion suppression, deuterium-labeled palmitoylcarnitine was coated over a series of tissue sections from intracranial GBM12 PDX models. A solution of palmitoyl-l-carnitine-(n-methyl-D3) (20 μmol/L in 50/50 methanol/water v/v; Sigma-Aldrich) was deposited using the TM-Sprayer with a flow rate of 0.09 mL/minute, spray nozzle velocity of 1,200 mm/minute, spray nozzle temperature of 75°C, nitrogen gas pressure of 10 psi, and track spacing 2 mm and 2 passes. A solution of erlotinib-(D6) (1 μg/mL, in methanol), was coated over a further set of tissue specimens under the same TM-Sprayer conditions. The concentration of internal standard and spray parameters were optimized so that it could be detected without saturating the signal.
MALDI FT-ICR MSI
MSI data were acquired using a 9.4 Tesla SolariX XR FT-ICR MS (Bruker Daltonics) externally calibrated in electrospray ionization positive ion mode using a tuning mix solution (Agilent Technologies). MALDI MS images were acquired with a pixel step size of 100 μm. Spectra were acquired in positive ion mode with 250 laser shots accumulated at each location. Spectra were collected in the range m/z 100–3,000 with isolation of ions using the constant accumulation of selected ions mode, with Q1 set to 500 and an isolation window of 260. Online calibration of spectra to heme (m/z 616.17668, Δppm tolerance 200, and intensity threshold 500) was applied. The laser focus was set to “small,” the laser energy was set to 35% (arbitrary scale) with a laser frequency of 1,000 Hz. MALDI MS images were acquired and viewed using FlexImaging 5.0 (Bruker Daltonics).
Data processing and statistical analysis
MSI data were analyzed using SCiLS Lab Software (Bruker Daltonics). Data were normalized to total ion intensity and were reduced to a peak list of approximately 500 m/z most relevant values. This peak list was created by correlating each spectrum with different peaks at different positions in the image, based on the orthogonal matching pursuit, and automatically finding the most correlated peaks. Bisecting k-means clustering (using the correlation distance metric) was used to provide segmented images showing regions of spectral similarity. Regions of interest (corresponding to tumor core and tumor edge) were defined on the basis of the resulting segmentation map and ROC analysis was performed to find ions that discriminated between tumor core and tumor edge. The m/z values with an AUC > 0.75 were searched against publicly available databases [LIPID MAPS and the Human Metabolome Database (HMDB); refs. 19, 20]. Ions were preliminarily identified if assignments were associated with Δppm < 2. Further characterization of ions preliminarily identified as acylcarnitines was performed by MS-MS.
LESA MS-MS
Serial sections to those analyzed by MALDI MSI were used for LESA MS-MS structural confirmation. Adjacent to the tissue on the glass slide were placed 0.5 μL dried droplets of standards of myristoyl-l-carnitine, palmitoyl-l-carnitine, and stearoyl-l-carnitine (10 μmol/L, in water; Sigma-Aldrich Inc.). LESA was performed using the TriVersa NanoMate (Advion) system coupled via nano-electrospray ionization (nano-ESI) to an amaZon Speed Ion Trap Mass Spectrometer (Bruker Daltonics). Sampling locations were selected within the tumor edge region (as determined from H&E staining of a serial section) or from dried droplets of each standard. The extraction/ESI solvent system was (60/40 methanol/0.1% formic acid v/v). Solvent (3 μL) was aspirated from a reservoir into a conductive pipette tip. The tip relocated to a predefined position 1 mm above the sample and dispensed 0.7 μL solvent onto the surface. The liquid microjunction was maintained for 5 seconds to allow soluble analytes to dissolve before 1 μL was reaspirated; this extraction process was repeated once and then injected into the mass spectrometer via nano-ESI, with a gas pressure of 0.3 psi and a tip voltage of 1.4 kV. Mass spectra were acquired for 5 minutes, in the positive ion mode, with an m/z range of 100–1,100 and the ion trap target tuned to m/z 400. After an initial 30 second acquisition of full scan MS, MS-MS was performed for the protonated species of each acylcarnitine as follows: myristoylcarnitine parent mass was m/z 372.3, palmitoylcarnitine parent mass was m/z 400.3, and stearoylcarnitine parent mass was m/z 428.3, all three species were fragmented with an isolation window of 2.0 and reaction amplitude 1.7.
RNA sequencing and analysis
ArcturusXT LCM System was used to capture cells from intracranial tumor section. Total RNA was isolated using RNAeasy Mini Kit (Qiagen), and then processed through the RNA Clean & Concentrator-5 Kit (Zymo Research) according to the manufacturer's protocols. Total RNA (50 ng) was used to generate whole-transcriptome libraries for RNA sequencing using Illumina's TruSeq RNA Sample Prep. Poly(A) mRNA selection was performed using oligo(dT) magnetic beads, and libraries were enriched using the TruSeq PCR Master Mix and primer cocktail. Amplified products were cleaned and quantified using the Agilent Bioanalyzer and Invitrogen Qubit. The clustered flowcell was sequenced on the Illumina HiSeq 4000 for paired 100-bp reads using Illumina's TruSeq SBS Kit V3. Lane level fastq files were appended together if they were sequenced across multiple lanes. The mouse reads were filtered out using in-silico method after sequencing. We used xenome (21) tool to classify reads into human, mouse, both, unmapped, and ambiguous categories. Human-only category reads were kept for downstream analysis. These processed fastq files were then aligned with STAR aligner using reference gene annotation (GTF) file. Transcript abundance was quantified and normalized using Salmon in the unit of transcripts per million. Consensus clustering was performed using ConsensusClusterPlus R package and heatmaps were generated using R heatmap.2 package with Euclidean Distance and average clustering method. Gene set variation analysis (GSVA; ref. 22) was then performed to determine the change and variation in pathway activities of glycolysis and fatty acid metabolism gene sets. Those gene sets were downloaded from Molecular Signatures Database (MSigDB; ref. 23). Log transformation was applied and default parameter settings were followed for RNA-seq data in GSVA. mRNA sequencing data have been deposited in EMBL-EBI European Nucleotide Archive database (accession number PRJEB35120).
Results
Metabolic differences between tumor core and tumor edge in intracranial GBM12 PDX models
Metabolomic MALDI MSI was performed on brain tissue sections from three different GBM12 PDX animals (Fig. 1). A segmentation map produced via bisecting k-means clustering (as described in Materials and Methods section) detected differences in mass spectra between gray and white matter in the brain, between tumor and healthy brain tissue, and between two different regions of the GBM tumor (see Fig. 1A and B). The tumor region was delineated by standard histopathology on a serial section of tissue. The segmentation map produced with k = 8 indicates two metabolically different regions of the tumor, one in the core of the tumor and one at the outer edge, labeled in white and blue, respectively. ROC analysis was applied to the reduced peak list to elucidate which ions in the mass spectra contribute to this difference. Ions with an ROC AUC > 0.75 were searched against LIPID MAPS and HMDB and candidate identities were proposed (for annotated peaklists see Supplementary Data File S1). A number of ions highlighted via this analysis were identified as long chain acylcarnitines. The Δppm associated with these assignments were all <2, which can be expected from high mass resolution FT ICR MS data. Hypothesis testing by a t test indicated that the difference in average intensity of each acylcarnitine was associated with a P value in most cases <0.001 and in all cases <0.05 (see Table 1). A total of seven different acylcarnitines were identified through this analysis, all demonstrating a similar spatial distribution within the tumor. MALDI ion images of three of the most abundant long chain acylcarnitines: myristoylcarnitine, palmitoylcarnitine, and stearoylcarnitine, are displayed in Fig. 1C–E. All three species were detected with higher intensity in the tumor edge compared with the tumor core. Structural confirmation of these ions was performed by LESA MS-MS (see Supplementary Results; Supplementary Fig. S1). Incorporation of the internal standard palmitoylcarnitine-(N-methyl-d3), according to the method proposed by Taylor and colleagues (24), into the matrix provided evidence to suggest that increased intensity of acylcarnitine was not observed because of differential ion suppression (see Supplementary Results; Supplementary Fig. S2A–S2D).
MALDI MSI analysis of intracranial GBM12 PDX model. A, H&E-stained serial tissue section demonstrating highly cellular tumor region. B, Segmentation map of MALDI MSI data produced via bisecting k-means clustering (k = 8), whereby different clusters were labeled with different colors, and two separate clusters are indicated within the tumor region, indicating spectral differences between these regions. C–E, MALDI MSI ion images of three different long chain acylcarnitines. Myristoylcarnitine, m/z 410.2666, detected as [M+K]+ (C); palmitoylcarnitine, m/z 400.3422, detected as [M+H]+ (D); stearoylcarnitine, m/z 428.3734, detected as [M+H]+ (E).
Summary of acylcarnitines detected from GBM12 PDX models, with associated m/z measured, AUC measured by ROC analysis between tumor core and tumor edge, the P value measured by t test for difference in average intensity of the ion between tumor core and tumor edge, the proposed identity of this ion, and the Δppm associated with this assignment.
3D MALDI MSI demonstrates that acylcarnitines are detected with increased intensity at the entire infiltrative tumor edge
Sequential coronal tissue sections were taken in increments of 160 μm throughout the tumor containing portion of an independent GBM12 PDX animal. These 22 sections were imaged using MALDI MSI to investigate the distribution of acylcarnitines throughout the entire tumor volume. All acylcarnitine species identified in this work were found to have the same distribution within tissue (see Supplementary Fig. S3), therefore example images of palmitoylcarnitine are presented here as a proxy for all long chain acylcarnitines. Acylcarnitines were detected with increased intensity at the edge of the tumor and at the interface between tumor and normal tissue, see Fig. 2A and B. These data demonstrate that this phenomenon was not peculiar to a single anomalous plane within the tumor of one animal. 3D reconstructions of MALDI ion images show the distribution of acylcarnitine within the tumor volume and demonstrate the location of the tumor relative to the vasculature and blood, via overlay with ion images of heme (Fig. 2C–E; ref. 25). A further set of images was acquired from 6 animals of the same GBM12 PDX model, tissue sections were acquired from roughly the same depth within the tumor (Supplementary Fig. S4). Similar distributions of acylcarnitines were measured in all 6 animals. Acylcarnitines were only detected with higher intensity at the interface between tumor and normal tissue, and not detected in portions of the tumor adjacent to the outer edge of the brain.
Three-dimensional MALDI MSI of intracranial GBM12 PDX model. A, MALDI MSI ion images of m/z 400.3421 corresponding to palmitoylcarnitine in sequential coronal tissue sections taken in increments of 160 μm in the z-dimension throughout the tumor-containing portion of the mouse brain. B, Serial tissue sections of those used for MALDI MSI stained with H&E, indicating the highly cellular tumor region. C–E, 3D reconstructions of images shown in A, shown from different elevations (C and D) and overlaid with ions of heme (m/z 616.1766), demonstrating the distribution of palmitoylcarnitine relative to the vasculature (E). Red lines in C and D indicate the outer boundary of tissue sections, whereas in E, red represents heme signature.
Distinctive and different distribution and relative abundance of acylcarnitines are observed in four different GBM PDX models
To establish whether this phenomenon was observed in other models of GBM, tissues from four different GBM PDX models, GBM12, GBM22, GBM39, and GBM108 (n = 2), were prepared and analyzed by MALDI MSI, see Fig. 3. Tumor regions were delineated using a combination of H&E and MALDI MSI data segmentation; see Fig. 3A and B. Acylcarnitines were detected in all four models, with the highest relative intensity detected in GBM12, followed by GBM22, GBM39, and GBM108 in order of decreasing intensity, as shown in Fig. 3C. In all 8 animals of the four models, acylcarnitines were detected with higher intensity at the tumor edge. In some cases, localized spots of higher intensity were detected in central regions of the tumors, for example in GBM22.
MALDI MSI of GBM12, GBM22, GBM39, and GBM108 intracranial PDX models, A, H&E-stained serial sections. B and C, Segmentation map of MALDI MSI data, produced via bisecting k-means clustering (k = 8; B), and MALDI MSI ion images of palmitoylcarnitine (m/z 400.3420) in the four GBM PDX models (n = 2) showing the relative distribution of long chain acylcarnitines (C).
Acylcarnitine and ATP demonstrate inverse distributions in four different GBM PDX models
The distribution of other ions that might act as functional markers was investigated in relation to acylcarnitine. Heme, a cofactor of hemoglobin, ATP, and palmitoylcarnitine were visualized alongside H&E images of serial sections of the four GBM models studied, see Fig. 4. Ion images of heme allow visualization of the brain vasculature and blood—indicating the location, size, and density of blood vessels relative to anatomic features of the brain and the tumor (see Supplementary Fig. S5). The highest relative intensity of heme was detected from GBM108 PDX models, which was expected because of the highly vascular nature of GBM108. Higher intensity of heme was detected from the VEGF+ GBM108 PDX (see Fig. 4). ATP (assignment associated with Δppm = 0.01) was detected with higher intensity in tumor tissue compared with normal tissue, in all models. In normal brain tissue, higher intensity of ATP was detected in the hippocampus and white matter tracts, consistent with previous findings and supporting the identity of these ions (26). A side-by-side comparison of images of ATP and palmitoylcarnitine suggest an inverse relationship within the tumor, whereby high intensity of acylcarnitine is detected in regions of low intensity ATP and vice versa. Regions of higher acylcarnitine intensity in the tumor cores (for example in GBM22 and GBM108 VEGF− seem to correspond with regions of higher heme intensity), raising the possibility that the vasculature plays a role in the distribution of acylcarnitine. In cases of hypervascularization as in VEGF+, the distribution of heme and acylcarnitine were visually more different.
MALDI MSI of GBM12, GBM22, GBM39, and GBM108 intracranial PDX models with H&E-stained serial sections. Comparison of distribution of heme (m/z 616.1766, red) indicating the localization of the vasculature, ATP (m/z 508.0030, green), and palmitoylcarnitine (m/z 400.3420, blue).
Fatty acid metabolism upregulated in GBM PDX models with high acylcarnitine content and inverse relation with glycolysis
To further probe the metabolic processes taking place in the tumors, and relate them to the content of acylcarnitines and ATP levels, analysis by mRNA sequencing was performed. On the basis of the heatmaps highlighting the activated genes within pathways for fatty acid metabolism and glycolysis (Supplementary Data Files S2 and S3), a score was calculated for each GBM PDX model, and is represented in the GSVA plot (see Fig. 5). The levels of palmitoylcarnitine and ATP (measured by MALDI FT ICR MSI; Fig. 5A and B) are displayed alongside GSVA plots for fatty acid metabolism and glycolysis, respectively (Fig. 5C and D), for each specimen. In this blinded experiment, transcriptome data from the mice was eliminated to analyze only the human (tumor) gene expression. Then validation of samples' relationship was performed using consensus clustering, and heatmaps were used to visualize the similarity between the samples with respect to fatty acid metabolism and glycolysis, using MSigDB gene sets for those pathways. A list of upregulated genes was selected to obtain the GSVA score. Hallmark genes related to fatty acid metabolism were found to have higher expression in the samples with higher measured levels of acylcarnitine (GBM12; see Fig. 5A and C). This relationship was further probed by investigating transcript expression of specific components of the carnitine cycle (CPT1a, CPT2, CRAT, and SLC25A20). Normalized gene expression was plotted for CPT1a, CPT2, CRAT, and SLC25A20, for each GBM PDX model (see Supplementary Fig. S6A and S2B). It was found that the expression levels of these genes were broadly in agreement with measured levels of acylcarnitine, with GBM12 demonstrating the highest expression of CPT1a and SLC25A20. CPT2 demonstrated little difference in expression between the different GBM models. GBM108 demonstrated the lowest expression of CPT1a, CRAT, and SLC25A20. Hallmark genes associated with glycolysis were most highly expressed in GBM108 VEGF−, and one of the samples of GBM39. This was in agreement with higher levels of ATP in these specimens, as measured by MALDI FT-ICR MSI (see Fig. 5B and D).
Comparison of MALDI FT-ICR MSI and transcriptomics data. A and B, Graphs depicting average (mean) ion intensity of m/z 400.3420 (A) and 508.0030 (B), respectively, over the tumor region in each PDX model of GBM, detected by MALDI FT-ICR MSI. Error bars, 1 SD of the mean. C and D, GSVA plot of HALLMARK genes for fatty acid metabolism in the same set of GBM PDX models (C) and GSVA plot of HALLMARK genes for glycolysis (D). In general, models associated with high levels of acylcarnitine by MALDI MSI have higher expression of hallmark genes for fatty acid metabolism (GBM12) and models/samples with higher levels of ATP measured by MALDI MSI have higher expression of hallmark genes for glycolysis (GBM108 VEGF− and GBM39 2).
Drug distribution relative to structural and functional markers was visualized by MALDI MSI
The distribution of two different drugs, erlotinib and AZD1775, was investigated in three different PDX models of GBM (Fig. 6), MS images of both drugs are displayed alongside structural and functional markers: palmitoylcarnitine, ATP, and heme. Both drugs were detected with higher intensity in tumor regions of all the GBM models. Erlotinib, an EGFR inhibitor, was detected with lower intensity at the tumor edge, characterized by a dark ring around the tumor, in GBM12 models. Palmitoylcarnitine was detected with higher intensity in this region, demonstrating a somewhat inverse distribution (see overlay images in Supplementary Fig. S7). AZD1775, a WEE1 kinase inhibitor (with a key role on the Cdk1 protein that regulates the cell cycle at different checkpoints), was detected with a relatively homogeneous distribution throughout the tumor region, with drug detected at the tumor edge with approximately the same intensity as the tumor core. Incorporation of an internal standard (D6-erlotinib) indicated that there was no significant difference in desorption/ionization efficiency between the tumor edge and the tumor core (see Supplementary Fig. S8A–S8E). This indicates that the absence of erlotinib in the edges of the tumor was not due to ion suppression in that area. The molecular structures and physicochemical properties of erlotinib and AZD1775 are summarized in Supplementary Fig. S9. The pH-dependent solubility of erlotinib is established (27), where more acidic pH results in maximum efficacy, because of increased solubility of erlotinib (28). Therefore, the pH variation found in GBM might affect the solubility of erlotinib, leading to a heterogeneous distribution of the drug in the tumor. The distribution of heme varied between the three different animals/PDX models. Erlotinib and AZD1775 were found to distribute differently within GBM PDX tumors, which may be linked to their different pH sensitivities.
Distribution of drugs (erlotinib, m/z 394.1770; AZD1775, m/z 501.2733) relative to palmitoylcarnitine (m/z 400.3421), ATP (m/z 508.0031), heme (m/z 616.1766), and H&E images of serial sections. Higher intensity of both drugs was detected in the tumor regions of all PDX models. In GBM12, lower erlotinib intensity was detected at the tumor edge, correlating with regions of high acylcarnitine content. A similar inverse relationship between palmitoylcarnitine and erlotinib was observed in GBM39. AZD1775 was detected relatively homogeneously throughout the tumor, including at the tumor edge, where acylcarnitine intensity was high. ATP was detected with increased intensity in tumor regions of all PDX models and with an inverse distribution to palmitoylcarnitine.
Discussion
The study described herein investigated spatially resolved metabolism in GBM PDX models. We describe the novel finding of increased intensity of long chain acylcarnitines at the tumor edge, relative to the tumor core. We observed the same distribution among seven different long chain acylcarnitines, in 8 different animals of the same GBM PDX model and similar distributions in a further 2 animals of four different GBM models. Differences in the relative abundance of acylcarnitines between the different GBM models was observed and raises the possibility that the tumor type or grade may contribute to these differences, although more extensive studies will be required. Metabolism was mapped in 2D sections and reconstructed in 3D images, demonstrating that the increase in intensity of acylcarnitines was detected at the entire interface between tumor and normal tissue, and was not detected at tumor edges, which extended to the edge of the brain. This suggests that the increase in acylcarnitine content delineates an interface between normal and tumor cells. Utilizing stable isotope internal standards, we suggest that the increase in acylcarnitine intensity at the tumor edge was not explained by ion suppression or technical artifact.
Long chain acylcarnitines are activated molecules involved in the transport of long chain fatty acids across mitochondrial membranes, after which, they can undergo β-oxidation to produce ATP. Glucose has been traditionally considered the primary energy source for the adult brain under normal conditions; however, there is evidence to suggest that fatty acids are also used by normal brain tissue (29). This is perhaps supported by our detection of acylcarnitines in the white matter tracts of the normal mouse brain. It has previously been demonstrated that human GBM tumors require FAO for respiration and proliferation and inhibition of FAO prolonged survival in a syngeneic mouse model of malignant glioma (11). Rewiring of metabolic programs, such as aerobic glycolysis and increased glutamine metabolism, are crucial for cancer cells to shed from a primary tumor, overcome the nutrient and energy deficit, and eventually survive and form metastases. However, the role of lipid metabolism in the aggressive phenotype of malignant cancers remains unclear. Our observation of different abundance of acylcarnitine at the tumor edge versus the core suggests that different regions of the tumor are undergoing different relative levels of FAO. Furthermore, analysis by RNA sequencing suggests that GBM PDX models with higher acylcarnitine content also demonstrate upregulated fatty acid metabolism.
Significant research on the properties and functions of acylcarnitines has focused on metabolism, that is, shuttling fatty acids for β-oxidation; however, there is a body of work demonstrating that acylcarnitines play roles in neuroprotection and signaling. Most of these studies have focused on l-carnitine and acetyl-l-carnitine and not the esterified derivatives, however, the carnitine portion, which is common to all three species, has antioxidant and antiapoptotic functions. The roles of palmitoyl-l-carnitine in the brain have also been studied, with palmitoyl-L-carnitine thought to influence membrane fluidity due to its amphiphilic nature and ability to react on membrane surfaces. Palmitoyl-L-carnitine has also been found to stimulate the expression of GAP-43, a protein that is involved in neuroplasticity and neurotransmission (30). It is possible that the accumulation of acylcarnitines at the tumor edge provides protection to tumor cells in this region, resulting in an overall survival benefit to the tumor. The proinflammatory properties of long chain acylcarnitines (31) were recently discussed in an article detailing the accumulation of long chain acylcarnitines at the periphery of a spinal cord injury site in rat. The striking distribution of acylcarnitines was correlated with the presence of microglia in tissue surrounding the injury site (32). It is possible that a similar inflammatory response is triggered by the presence of acylcarnitines in the GBM PDX tumor. The multiple roles played by acylcarnitines should be considered in future studies to elucidate the mechanism of action of acylcarnitine in GBM, beyond β-oxidation.
Finally, we investigated the distribution of two different drug compounds, both of which have been investigated in the possible therapeutic treatment of GBM, with different mechanisms of action and physicochemical properties, in three different GBM models. Erlotinib, whose solubility is known to be pH sensitive (27), was detected with higher intensity in the tumor core, and lower intensity at the tumor margin. Lower intensity of acylcarnitine was detected in GBM39 (relative to GBM12) and the inverse association between acylcarnitines and erlotinib was to a lesser degree. Conversely, AZD1775 was detected with a relatively homogeneous distribution throughout the tumor. It is well established that tumors generally have a lower pH than their surrounding tissue. A study using MRI-based pH measurement found that in rat glioma models, the tumor core was more acidic than the tumor edge (33). The pH of the extracellular space in tumors varies between 6.5 and 7.5 (34). Tumor acidosis is intricately linked with tumor metabolism, with the increased production and accumulation of H+ via increased lactic acid production and hydration of CO2 produced via oxidative metabolism (35). Acidic tumor environments have also been shown to promote FAO via the downregulation of acetyl-CoA carboxylase 2 (35). This raises the possibility that the observed distribution of erlotinib is influenced by spatial differences in tumor pH, as a result of variation in metabolism.
In summary, we have identified a class of molecules that are highly abundant at the proliferating/infiltrative edge of the tumor in four different PDX models of GBM. Our data demonstrate that the metabolomic composition of the tumor edge is different from that of the tumor core. We suggest that GBM cells at the infiltrative tumor edge undergo different relative levels of FAO, an energy production pathway that has recently been shown to be essential for the respiration and proliferation of glioma cells. The various roles of acylcarnitines beyond FAO should be considered in future work to determine the full implications of this characteristic distribution, and the possible survival advantage that acylcarnitines could provide.
Disclosure of Potential Conflicts of Interest
S. Santagata is a consultant (paid consultant) for Rarecyte. J.N. Sarkaria reports receiving a commercial research grant from Novartis, Basilea, Cavion, Curtana, Forma, AbbVie, Actuate, Boehringer Ingelheim, Bayer, Celgene, Cible, Mitochon, Genentech, Wayshine, Sanofi, Beigene, Lilly, Glaxo-Smith-Kline, Peloton, Glionova, and Bristol-Myers Squibb Pharmaceuticals. N.Y.R. Agar is a scientific advisor (paid consultant) for BayesianDx and has unpaid consultant/advisory board relationship with Bruker. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: E.C. Randall, S.S. Basu, N.L. Tran, W.F. Elmquist, N.Y.R. Agar
Development of methodology: E.C. Randall, B.G.C. Lopez, M.C. Haigis, J.N. Sarkaria
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.C. Randall, B.G.C. Lopez, N.L. Tran, J.N. Sarkaria
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.C. Randall, B.G.C. Lopez, S. Peng, M.S. Regan, W.M. Abdelmoula, H. Yoon, M.C. Haigis, J.N. Agar, N.L. Tran, W.F. Elmquist, N.Y.R. Agar
Writing, review, and/or revision of the manuscript: E.C. Randall, B.G.C. Lopez, S. Peng, S.S. Basu, S. Santagata, H. Yoon, J.N. Agar, N.L. Tran, W.F. Elmquist, F.M. White, J.N. Sarkaria, N.Y.R. Agar
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.Y.R. Agar
Study supervision: F.M. White, N.Y.R. Agar
Other (pathology review): S. Santagata
Acknowledgments
This work was funded by NIH U54 CA210180 MIT/Mayo Physical Science Oncology Center for Drug Distribution and Drug Efficacy in Brain Tumors, and by the Dana-Farber Cancer Institute PLGA Fund (9616692). N.Y.R. Agar receives support from the Ferenc Jolesz National Center for Image Guided Therapy NIH P41-EB-015898. N.Y.R. Agar also receives support from NIH R01CA201469. E.C. Randall received an NIH R25 (R25 CA-89017) Fellowship in partnership with the Ferenc Jolesz National Center for Image Guided Therapy at BWH (P41 EB015898). S.S. Basu receives support from NIH Training Grant T32 HL007627. The authors thank Dr. Catherine Rawlins, Daniel Donnelly, and Nicholas Schmitt for technical assistance with the MALDI FT ICR instrumentation.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Footnotes
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
Cancer Res 2020;80:1258–67
- Received February 22, 2019.
- Revision received July 12, 2019.
- Accepted November 13, 2019.
- Published first November 25, 2019.
- ©2019 American Association for Cancer Research.
References
Volume 80, Issue 6
