Non-invasive metabolic imaging of brain tumors in the era of precision medicine

Introduction

Primary brain tumors in children and adults remain one of the most challenging forms of cancer to diagnose and treat. They are broadly classified into glial and non-glial tumors¹. Glial tumors are more common in adults while non-glial tumors such as medulloblastomas are more frequent in children¹. The clinical management of these patients relies heavily on non-invasive imaging. Magnetic resonance imaging (MRI), with or without contrast enhancement remains the current standard of care for initial evaluation. MRI can also guide tissue biopsy to establish diagnosis, permit the assessment of disease progression and evaluate the effectiveness of therapies.

While there have been significant strides in understanding the pathology of primary brain tumors, there remains a wide gap that must be bridged between our understanding of the biology of brain tumors and non-invasive imaging to guide diagnosis, treatment, and follow-up. In the era of personalized medicine we are beginning to appreciate the genetic complexity of primary brain tumors. This information can generate novel therapeutic windows to identify actionable targets and targeted therapies. However, the challenge remains in identifying, stratifying and monitoring patients with advanced imaging, and leveraging the information gathered from next generation sequencing to develop molecular imaging modalities.

The best example of molecular imaging in the clinical management of cancer comes from the field of cancer metabolism. Altered tumor metabolism can be imaged by two principle techniques. The first is PET imaging, which is based on labeling biologic substrates with radionuclides such as ¹¹C or ¹⁸F. PET imaging is highly sensitive because of its ability to detect low concentrations of radiolabeled substrates. The most common form of PET imaging in cancer takes advantage of the Warburg effect (the anaerobic metabolism of cancer cells, generating high lactate levels even in the presence of oxygen), and is based on labeling a glucose analogue with ¹⁸F (2-[¹⁸F] fluoro-2-deoxyglucose or ¹⁸FDG). Single-voxel MR spectroscopy (MRS) and multi-voxel MR spectroscopic imaging (MRSI) constitute the other technique that enables detailed detection of cellular metabolic activity by characterizing the chemical and molecular composition of tumor and tumor microenvironment using radiofrequency signals generated by nuclear spins of magnetic resonance active nuclei including ¹H, ³¹P and ¹³C.

A fundamental concept emerging in cancer biology is that oncogenes directly reprogram cellular metabolism to enable tumor cells to grow, proliferate and survive^{2, 3}. Many of the brain tumor oncogenes influence specific metabolic pathways including glucose, amino acid and fatty acid metabolism as discussed below. In this article, we evaluate (1) the recent genomic revolution in brain tumor biology, (2) the rewiring of metabolic pathways from genetic alterations, and (3) the translation of metabolic reprograming to non-invasive tumor imaging in patients.

Glial tumors in adults and children are genomically distinct

The World Health Organization defines two major groups of glial tumors: astrocytic tumors and oligodendroglial tumors. Additionally, histologic criteria are used to define tumor grade: low-grade (grade I and grade II) and high-grade (grade III and grade IV, or glioblastoma multiforme (GBM))¹. Glial tumors in adults and children are morphologically comparable and were thought to have similar biology. However, next generation sequencing studies strikingly reveal that gliomas in adults and children are genetically different. In adults, more than 90% of GBM show genetic alterations that converge on the PI3K-AKT/mTOR pathway. This includes enhanced activation of receptor tyrosine kinase such as EGFR (~60%), PDGFRA (~10%), FGFR (~3%) and MET (~1%) that signal mainly through the PI3K AKT/mTOR pathway^{4, 5}. Other genetic alterations in this pathway include PTEN deletion (~40%) and mutations in PI3K (~20%) (Table 1). Therefore the PI3K-AKT/mTOR pathway is one of the primary drivers of adult GBM. In contrast, high-grade gliomas in children are characterized by epigenetic mutations. Histone H3 K27M and G34R/V mutations are noted in 60% of pediatric glioblastomas and H3K27M mutations occur in more than 80% of brain stem gliomas (termed diffuse intrinsic pontine gliomas)^6–9 (Table 1).

Similarly, low-grade gliomas in adults and children, although histologically similar, have different genetic drivers. More than 70% of intermediate-grade (grade II and III) gliomas (including astrocytomas and oligodendrogliomas) in adults bear mutations in isocitrate dehydrogenase (IDH) 1/2 (Table 1). IDH mutant gliomas tend to occur in younger adults and are associated with a favorable prognosis relative to their wild-type counterparts¹⁰ (Table 1). IDH 1/2 mutant gliomas that progress to GBM are termed secondary GBM. These IDH 1/2 mutant GBM constitute ~5% of adult GBM and show a significantly favorable prognosis compared to IDH wild-type tumors (Table 1). Low-grade gliomas in children and young adults such as pilocytic astrocytomas show constitutive activation of BRAF. Oncogenic fusion of the kinase domain of BRAF with KIAA1549 is seen in 30–70% of pilocytic astrocytomas (with varying frequency depending on cortical versus cerebellar tumor location)¹⁰. Likewise, activating BRAF point mutations (V600E) are observed in other low-grade pediatric brain tumors like pleomorphic xanthoastrocytomas (70%), gangliogliomas (20%) and at lower frequencies in pilocytic astrocytomas and diffuse astrocytomas¹⁰. Thus, the genetic drivers in childhood and adult gliomas are distinct (Table 1).

Precision medicine: next generation sequencing enables prognostic sub-classification of brain tumors and defines therapeutic targets

Another important insight from these molecular studies is the sub-classification of tumors into distinct groups that can inform clinical management of patients beyond histologic grade. For example, grade IV medulloblastomas (MB), which are typically pediatric tumors can be sub-grouped into four molecular subtypes: SHH, WNT, Group 3 and Group 4¹¹ (Table 1). Of these, WNT tumors constitute ~10% of all cases, have WNT-pathway associated gene expression patterns and bear the best prognosis with a 5-year survival of ~95%. SHH tumors represent ~30% of MB and are characterized by constitutive activation of the SHH pathway and bear an intermediate prognosis with ~75% survival. Group 3 tumors comprising ~25% of cases show alterations in c-MYC and bear the worst prognosis with ~50% survival. Group 4 tumors represent ~35% of cases, demonstrate intermediate prognosis with ~75% survival and are poorly characterized¹¹. Similarly, adult glioblastomas can be sub-grouped into four categories: proneural, mesenchymal, neural and classic. Of these, the proneural subtype is associated with IDH 1/2 mutations and have the best prognosis⁴.

Indeed, next generation sequencing has refined our understanding and approach to caring for adults and children with brain tumors. These studies have opened up novel avenues for the development of targeted therapies. For example, vismodegib, an antagonist of the SHH pathway, is effective in patients with SHH-MB but not in patients with other MB subtypes¹². Similarly, inhibition of the BRAF V600E mutation with vemurafenib has demonstrated efficacy in childhood brain tumors that bear this mutation^13–15. Moreover, pharmacological inhibitors of mutant IDH1 show promise in IDH1 mutant but not in IDH wild-type glioma animal models¹⁶. Thus, integrating the molecular characteristics of these tumors has become integral in the management of brain tumor patients. These molecular features are currently assessed on tissue samples obtained after biopsy or tumor resection. However, this is not always feasible, as some brain tumors such as pontine and brain stem tumors are difficult to biopsy due to surgical challenges associated with tumor location. Further, the direct evaluation of the efficacy of targeted therapies with longitudinal, serial biopsies is not practical. Conventional MRI, the current standard, is inadequate to evaluate molecular alterations in brain tumors. Thus, non-invasive metabolic imaging of brain tumors is now emerging as a means to assess some of these molecular and metabolic alterations.

Glucose metabolism: oncogenic reprograming and imaging

Glycolysis is the pathway by which six-carbon glucose is metabolized into three-carbon pyruvate (Figure 1). In aerobic conditions, pyruvate can be oxidized to acetyl-coenzyme A, which enters the mitochondrial tricarboxylic acid (TCA) cycle. Brain tumors exhibit the Warburg effect and divert pyruvate away from the mitochondria. Warburg effect is thought to be advantageous to tumor cells for several reasons, such as enhanced production of glycolytic intermediaries to support macromolecular synthesis, generation of reducing equivalents such as NAD+, secreted lactate that alters the microenvironment to enable tumor invasion and escape from immune cells, and as an adaptation to lowered oxygen levels in hypoxic environments^17–20.

The Warburg effect in brain tumor cells is controlled at several levels, including regulation by the PI3K/AKT/mTOR pathway, BRAF activation and MYC induction. For example, phosphorylated AKT enhances glycolysis by increasing expression of glucose transporters and by activating glycolytic enzymes such as hexokinase-2 (HK2) and phosphofructokinase-1 (PFK-1)^21–23. Similarly, Activated mTORC1 alters glucose metabolism and glycolytic gene expression by regulating c-Myc activity in EGFRvIII-driven GBM²⁴. BRAFV600E and MYC can also stimulate the transcription networks related to glycolysis including glucose transporters and glycolytic enzymes^{25, 26, 27}. TP53, which is frequently mutated in astrocytic tumors, can also regulate glycolysis by influencing glycolytic enzymes and glucose transporters²⁸ (Figure 1). Thus, imaging glucose metabolism is a very attractive target in brain tumors.

Interestingly, many of the glycolytic enzyme isoforms that are preferentially expressed during brain development in proliferating cells are elevated in brain tumors. For example, HK2 (yielding glucose-6-phosphate from glucose) and pyruvate kinase M2 isoform (PKM2, which converts phosphoenolpyruvate to pyruvate) are two enzyme isoforms that are expressed in the embryonic but not the adult brain^29–31. Brain tumors such as GBM and medulloblastomas preferentially express HK2 and PKM2^31–37. Gambhir and colleagues reported the generation of a PET imaging probe specific for PKM2 using a class of N, N-diarylsulfonamides (DASA) known to promote PKM2 tetramer formation. [¹¹C] labeled DASA-23 was taken up in orthotopic glioma models and corresponded to tumor-associated PKM2 expression³⁸ (Figure 1).

¹⁸FDG PET imaging was one of the earliest tools used to measure local glucose utilization in the brain. ¹⁸FDG, like glucose, is transported into the cell by glucose transporters and phosphorylated by hexokinase to 2-deoxyglucose-6-phosphate. This intermediate is unable to be further catabolized, and becomes trapped in the cell and is a measure of glucose uptake^39–41. However, ¹⁸FDG PET for the imaging of gliomas shows poor tumor-to-background contrast due to the physiologic high levels of glucose uptake in the brain (Figure 3). It also has limited specificity in distinguishing between tumor and non-malignant processes including infection and inflammation⁴². MRS can be used to monitor glucose metabolism in human subjects and glioma animal models using ¹³C-labeled glucose. Because MRS visualizes only paramagnetic atoms, this technique detects only ¹³C rather than the more abundant ¹²C. The conversion of the administered metabolite, such as ¹³C-labeled glucose, into its downstream products can be measured by simultaneously measuring the disappearance of ¹³C-glucose and the appearance of ¹³C into metabolic products such as ¹³C-labeled lactate or pyruvate. Using this technique, glycolysis can be evaluated in ex vivo tumor specimens from animal models and from GBM patients^{43, 44}. Intriguingly, mice implanted orthotopically with GBM cells derived directly from patients and infused with ¹³C-labeled glucose showed entry of glucose carbons into the TCA cycle with lower generation of lactate than expected⁴⁴.

However, ¹³C is difficult to image by MRS due to its low natural abundance (approximately 1%) and unfavorable gyromagnetic ratio, both of which result in low signal to noise ratios⁴⁵. To overcome these limitations, ¹³C-enriched probes can be “hyperpolarized” by exposure to microwaves at extremely low temperatures prior to tracer administration, using a technique known as dynamic nuclear polarization. This alters the Boltzmann distribution of ¹³C and can increase sensitivity by more than 10,000-fold⁴⁶. Because ¹³C-labeled probes remain hyperpolarized for a very short period of time, the probe must be infused immediately into the subject for MRS detection. The conversion of hyperpolarized [1-¹³C] pyruvate to [1-¹³C] lactate (Figure 1) as a result of rapid exchanges between the probe and a large tumor lactate pool arising from the Warburg effect is the most extensively characterized metabolic process, including the first human cancer study⁴⁷. In gliomas, hyperpolarized ¹³C MRS studies have been conducted in animal models. For example, the generation of [1-¹³C] lactate from [1-¹³C] pyruvate was detected in xenografts in rats, but not normal brain. Enhanced sensitivity in hyperpolarized MRS holds significant promise in evaluating glycolysis and other metabolic changes in human glioma patients. Moreover, high-field 7T MRI technique has now expanded the capability of MRS, as well as techniques utilizing chemical exchange saturation transfer (CEST) effect, ²³Na and ³¹P spectroscopy^{48, 49}. Compared to 3T field strengths in which the resonances of glutamate undergo strong coupling and overlap with glutamine and gamma-aminobutyric acid (GABA) peaks, localized ¹H MRS at 7T can be used to reproducibly measure cortical GABA and glutamate concentrations⁵⁰. Moreover, using ³¹P MRS the rate of ATP synthesis in the human brain can be quantified, and extra- versus intracellular pH differentiated⁵¹. Finally, the spectrum of metabolic studies may be further expanded by spectroscopic techniques with enhanced physiological and functional information based on CEST contrast, in which amide and hydroxyl protons from different amino acids, proteins and other molecules have been used to image glucose accumulation in tumor models (glucoCEST), to detect early tumor response in glioblastoma^52–54. Increasingly, combined PET/MRI systems may enable a multiparametric approach to non-invasively characterize brain tumors⁴⁹.

(D)-2-Hydroxyglutarate: IDH1 mutant oncometabolite and MRS imaging

IDH enzymes generate α-ketoglutarate (α-KG) from isocitrate, and resultantly generate reducing equivalents. There are three cellular isoforms: IDH1, 2 and 3. IDH1 is present in the cytosol, while IDH2 and IDH3 are mitochondrial. Mutations in IDH1 and IDH2 are heterozygous and missense. IDH1 mutations are the most common in gliomas (more than 90%) and involve the catalytic site arginine at position 132, which is mutated to histidine (R132H), while IDH2 mutations are rare in gliomas⁵⁵. IDH mutations result in the production of the metabolite (D)-2-hydroxyglutarate ((D)-2HG) (Figure 2). (D)-2HG transforms immortalized human astrocytes, and one of its mechanisms of oncogenesis is related to its structural similarity to α-KG^56–59. α-KG serves as a critical cofactor for dioxygenase enzymes that regulate various cellular functions like histone and DNA demethylation. (D)-2HG can inhibit the function of many of these enzymes resulting in both DNA hypermethylation at CpG islands called G-CIMP (glioma-CpG island methylator phenotype) and histone hypermethylation^60–63.

As (D)-2HG is produced in IDH 1/2 mutant gliomas but not in wild-type gliomas, the non-invasive detection of (D)-2HG could serve as a biomarker for the diagnosis, treatment and surveillance of these tumors. Detection and quantification of (D)-2HG using in vivo MRS has been demonstrated in IDH1 mutant gliomas^64–66. One of the challenges of detecting (D)-2HG is its proximity to glutamine/glutamate and NAA resonance peaks⁶⁷. This requires optimizing MRS using techniques such as two-dimensional correlation spectroscopy or J-difference spectroscopy to enable the differentiation of overlapping metabolites^{64, 65}. Using these techniques, (D)-2HG can be detected in milimolar ranges (1.7–8.9 mM) in IDH1/2 mutant but not in wild-type gliomas. Further, the detection of (D)-2HG in vivo correlates with better prognosis, mirroring previously reported next generation sequencing data^{55, 68, 69} However, while the detection of (D)-2HG may be sensitive in the assessment of IDH1/2 mutations, the specificity may depend on the size and cellularity of the tumor. A recent study suggests that the sensitivity may be as low as 8% for small tumors (defined as <3.4 mL volume) in contrast to 91% sensitivity for larger tumors (> 8 mL in volume)⁷⁰. The enzymatic function of mutant IDH 1/2 to catalyze the conversion of α-KG to (D)-2HG, detectable in vivo, may be harnessed using hyperpolarized ¹³C MRS⁷¹ For example, the conversion of α-KG labeled with ¹³C at the 1^st carbon position (1-¹³C] α-KG) to [1-¹³C] (D)-2HG (Figure 2) can be detected in rats bearing IDH1 mutant but not IDH1 wild-type GBM xenografts⁷¹. While these observations are yet to be translated into the clinic, as the only imaging modality specific to IDH mutations, in vivo MRS (with or without hyperpolarization) represents an exciting foray into the use of a pharmacodynamic biomarker.

Metabolic reprograming and imaging amino acid metabolism

Many amino acids including glutamine, glutamate, methionine, aspartate and tyrosine demonstrate altered metabolism in brain tumors. The mechanisms used by brain tumor oncogenes to reprogram amino acid metabolism are just beginning to emerge.

Brain tumor oncogenes reprogram fatty acid metabolism

Cancer cells can engage in both de novo fatty acid synthesis and enhanced fatty acid oxidization. Synthesized fatty acids can serve as precursors for the generation of many biomolecules including phospholipids, sphingolipids, prostaglandin and cholesterol esters. Choline is an essential nutrient that is required for synthesis of phospholipids and the neurotransmitter acetylcholine. Choline levels are increased in cancer cells and are thought to reflect increased cell membrane turnover. Using ¹H and ³¹P MRS, the increase in choline-containing metabolites can be non-invasively detected in vivo in brain tumors¹⁰¹ (Figure 6).

Oncogenes such as EGFR can control fatty acid synthesis through the activation of the PI3K-AKT pathway by stabilizing enzymes such as sterol regulatory element-binding protein-1 (SREBP-1) that regulate lipogenesis¹⁰²(Figure 2). Moreover, PI3K-AKT activation in gliomas results in elevated levels of acetyl-CoA, which is a central precursor for fatty acid synthesis, by enabling the incorporation of carbons from acetyl-CoA into growing fatty acid chains^103–105. Acetyl-CoA can be derived from pyruvate and can also be synthesized from acetate by the enzyme Acyl-CoA Synthetase Short-Chain Family Member 2 (ACSS2)^{106, 107}. GBMs show increased expression of ACCS2, which results in increased acetyl-CoA pools that could be used both for oxidation within the TCA cycle and for lipid synthesis (Figure 2). MRS studies using ¹³C-labeled acetate suggest mitochondrial oxidation of acetate-derived acetyl-CoA in orthotopic brain tumor models and a human GBM patient¹⁰⁸. In small clinical studies, ¹¹C labeled acetate PET imaging has been found to have a sensitivity of 90% in detecting low and high-grade gliomas¹⁰⁹.

Tumor metabolism and advanced imaging of cellular proliferation, hypoxia, and angiogenesis

Additional key components of tumorigenesis driven by alterations in tumor cell energy metabolism may be directly assessed with molecular imaging. Altered tumor metabolism supports rapid cellular proliferation. Compared to normal proliferating cells, tumor cells have increased levels of thymidine kinase (TK-1). Radiolabeled nucleoside analogues such as 3’-deoxy-3’[(18)F]-fluorothymidine (¹⁸F-FLT) are transported into cells, phosphorylated by TK-1, and trapped intracellularly at an increased rate compared to normal cells due to increased tumor levels of TK-1^110–112. This process is dependent on breakdown of the blood-brain-barrier (BBB) since transport across the intact BBB is slow, and is therefore not as useful in non-enhancing primary brain tumors with intact BBB (e.g. low-grade glioma); it may also be limited in distinguishing moderately proliferative tumors driven by thymidine salvage from highly proliferative tumors relying on de novo thymidine synthesis^113–115.

Hypoxia is a critical feature of rapidly growing tumors, which develop adaptive responses through angiogenesis, enhanced glucose metabolism and optimized mitochondrial respiration that confers a survival advantage to hypoxic tumor cells. An important prognostic factor for resistance to antineoplastic treatments including systemic therapy and radiotherapy, hypoxia may be imaged with a number of PET tracers, of which ¹⁸F-fluoromisonodazole (¹⁸F-FMISO) is the most well-studied¹¹⁶, providing spatial delineation of hypoxia in brain tumors independent of BBB disruption and tumor perfusion¹¹⁷.

A closely related phenomenon, angiogenesis, is a hallmark of tumor growth that contributes to a hostile tumor microenvironment with low oxygen tensions and high interstitial fluid pressure that selects for a more malignant tumor phenotype¹¹⁸. A target structure for molecular imaging, α_vβ₃-integrin receptor, is highly expressed on activated endothelial cells during angiogenesis. Various ligands based on the tripeptide RGD (Arg-Gly-Asp), which binds with high affinity to the α_vβ₃-integrin receptor, have been developed for PET¹¹⁹. The glycosylated cyclic pentapeptide 18F-galacto-RGD has been successfully applied to patients with malignant gliomas, with high uptake in regions of rapid proliferation and tumor infiltration in contrast to normal brain¹²⁰.

Integrating metabolic imaging with clinical care in brain tumor patients

We have thus far examined how different genetic alterations in primary brain tumors reprogram metabolism and how some of these pathways can be translated into non-invasive imaging in preclinical animal models and patients. We now examine the potential for metabolic imaging to be assimilated into routine clinical care.

Conclusions

As we make significant strides in understanding the molecular basis of brain tumors, further work is needed to develop non-invasive tools for evaluating these alterations and integrating them into daily clinical practice. Altered cancer metabolism offers a unique window to integrate genomic information with advanced imaging, and the field has rapidly advanced with novel techniques that have been successfully implemented in preclinical models and patients. However, additional work is needed to standardize these metabolic imaging techniques for routine clinical use. In combination with other functional imaging modalities, these techniques may prove complementary to conventional MRI in characterizing tumor biology and metabolism for patient management. Future studies based on these oncogenic-driven metabolic pathways may enhance diagnosis, prognostication, treatment and surveillance, and ultimately improve outcome in patients with brain malignancies.