Synopsis

Keywords: Tumors, Deuterium, glioma

Dynamic glucose-enhanced deuterium MRS (DGE ²H-MRS) coupled with Marchenko-Pastur PCA (MPPCA) denoising has been recently applied to immunocompetent mouse glioblastoma subtypes (GL261 and CT2A), demonstrating the ability to measure glucose metabolism through glycolysis and mitochondrial oxidation non-invasively, and its association with tumor proliferation. Here, we extend this approach to DGE Deuterium Metabolic Imaging (DGE-DMI) coupled with tensor MPPCA (tMPPCA) denoising, to map glucose fluxes in the same mouse models of glioblastoma. Our results demonstrate a strong association between glycolytic rates and MRI and histologic features of inter/intra-tumor heterogeneity.

INTRODUCTION

Glioblastoma (GBM) are aggressive brain tumors with poor prognosis, largely due to their intrinsic heterogeneity and lack of non-invasive methods to assess it [1]. To address this gap, dynamic glucose-enhanced deuterium spectroscopy (DGE 2H-MRS) was previously harnessed with Marchenko-Pastur Principal Component Analysis (MPPCA) denoising [2-4], demonstrating its ability to differentially quantify glucose turnover rates through glycolysis and mitochondrial oxidation in mouse glioblastoma, and modulation of these pathways fluxes in vivo according to tumor cell proliferation [5]. Here, we expand this approach to DGE Deuterium Metabolic Imaging (DGE-DMI) [6, 7] coupled with tensor MPPCA (tMPPCA) denoising [8]. Thus, we investigate for the first time the links between glucose metabolic fluxes and heterogeneity features of the tumor microenvironment and its surrounding tissues in mouse glioblastoma.

METHODS

GBM tumors in vivo and post-mortem
All animal experiments were preapproved by institutional and national authorities, and carried out according to European Directive 2010/63. GBM tumors were induced in immunocompetent C57BL6/j mice by intracranial stereotactic injection of 1E5 glioma cells (GL261, n=5; or CT2A, n=5) in the caudate nucleus [9]. The animals were monitored for tumor growth at 1 T (Bruker Icon) and studied 2-6 weeks post-inoculation at 9.4 T (Bruker BioSpec), using a customized 2H/1H transmit-receive surface coilset (NeosBiotec). Fasted mice (4-6h) were cannulated in the tail vein and placed in the animal holder (1-2% isoflurane in 29% O₂; rectal temperature 36-37ºC; breathing 60-80BPM. DGE-DMI was acquired with slice-FID CSI (10kHz SLR excitation pulse, 55º flip angle, 2.27 slice thickness, with final in-plane resolution 560×560 µm², 175ms TR, 256 complex points, BW=1749 Hz), following a 6,6′-²H₂-glucose i.v. bolus injection (2 g/Kg). Each animal further underwent T2-w (RARE: 3s TR, 40ms TE, Turbo-8, 1mm slice thickness, 70×70 µm² resolution) and DCE-T1 MRI (FLASH: 8º flip-angle, 16ms TR, 2.27mm slice thickness, 140×140 µm² resolution; Gd-DOTA i.v. bolus, 0.1 mmol/Kg), to assess tumor volume and tissue perfusion, respectively. Finally, animals were sacrificed, the brains were removed, washed in PBS and immersed in 4% PFA until paraffin embedding, 1 week later. Samples were sectioned for histology and digitalization.
Data analyses
DGE-DMI data were processed in MATLABR R2018b and jMRUI 6.0b. This included tMPPCA denoising [8] and spectral quantification using AMARES and DHO reference (natural abundance) [5], with quality control based on Cremer-Rao Lower Bounds (<50%) and SNR of the initial DHO peak (SNRi >3). Data were analyzed voxel-wise with kinetic modeling to generate glucose flux maps through glycolysis (glucose-to-lactate) and mitochondrial oxidation (glucose-to-glutamine-glutamate), as recently reported [5]; additionally, average concentration maps [10] were generated for glucose, lactate, and glutamine-glutamate. DCE T1-MRI data were analyzed with DCE@urLab [11], using the Extended Tofts model to derive tissue perfusion (k^trans). Histopathological analyses (H&E staining) were performed independently by an experimental pathologist, and semi-automated cell counting performed on immunostained sections (ki67) with QuPath 0.2.3, as before [5]. Groups were compared using the two-tailed Student’s t-test, and correlation analyses performed with the Pearson R coefficient.

RESULTS

DGE-DMI achieved good spatio-temporal resolutions in CT2A and GL261 cohorts (~3 µL/voxel, 12 min/spectrum) and spectral SNRi was increased by 3-fold (p<0.001) after tMPPCA denoising (Fig. 1). This improved the overall quality of the estimated glucose flux maps (Fig. 2), based on pixel detectability: +38.3±9.7% and +62.2±19.2% (p<0.01), for CT2A and GL261, respectively (Fig. 3). Despite the histopathologic differences in cell morphology and stromal-vascular phenotypes verified between the two glioblastoma models, tissue perfusion correlated pixel-wise with glycolysis rate, within each subject (0.585 < R < 0.824) (Fig. 4); whereas tumor volume (58.5±7.2 mm3) and total tissue cellularity correlated strongly across subjects with glucose consumption and glycolysis rates, in each cohort and in pooled cohorts (0.574 < R < 0.984) (Fig. 5). These associations were overall weaker without tMPPCA denoising (0.512 < R < 0.804 and 0.267 < R < 0.935), respectively), and not consistent for glucose, lactate, and glutamine-glutamate concentration maps.

DISCUSSION

DGE-DMI coupled with tMPPCA denoising generates glucose flux maps, which are associated with perfusion and histologic features of inter/intra-tumor heterogeneity in glioblastoma-bearing mouse brains. Importantly, such correlations were inconsistent for metabolite concentration maps (rather than pathway fluxes).

CONCLUSION

By probing the glioma microenvironment and its surrounding tissues, these results complement our previous findings with localized deuterium spectroscopy exclusively in the tumor region [5]. Given the application of glucose-enhanced DMI to glioblastoma patients [10], and to healthy subjects at 9.4 Tesla [12], this methodology is clinically translatable holding strong potential for precision oncology.

Acknowledgements

This work was supported by H2020-MSCA-IF-2018, ref. 844776 (RVS); and Champalimaud Foundation. The authors thank the Vivarium of the Champalimaud Centre for the Unknown, a research infrastructure of CONGENTO co-financed by Lisbon Regional Operational Programme (Lisboa2020), under the PORTUGAL 2020 Partnership Agreement, through the European Regional Development Fund (ERDF) and Fundação para a Ciência e Tecnologia (Portugal), under the project LISBOA-01-0145-FEDER-022170.