Bevacizumab treatment induces metabolic adaptation toward anaerobic metabolism in glioblastomas

In recent years, various -omics and deep sequencing technologies have provided important new insight into molecular sub-classifications of GBMs that have led to better prognostic information. However, less is known about the mechanisms of adaptability of tumor cells to changes in the microenvironment and how these changes affect the tumor cell populations. Such changes may either arise due to alterations in the microenvironment during tumor progression, or as a response to therapy. During the last few years, the vascular endothelial growth factor (VEGF) inhibitor bevacizumab has been extensively applied for the treatment of recurrent GBMs. Yet, recent results from two double-blind, placebo-controlled phase III trials, AVAglio [6] and RTOG 0825 [15], show that bevacizumab treatment does not increase overall survival, but leads to a minor increase in progression-free survival. GBMs are characterized by their diffuse infiltrative growth as well as extensive angiogenesis where VEGF has been shown to play a major role [45]. Thus, from a theoretical viewpoint, blocking the formation of new blood vessels should be beneficial for GBM patients and, indeed, dramatic effects are seen, exemplified by a strong reduction in contrast enhancement (CE) as well as vessel diffusion and perfusion, assessed by magnetic resonance imaging (MRI) following treatment [2, 8, 9, 12, 54, 57]. However, the exact mechanism by which GBMs evade bevacizumab treatment still remains enigmatic. Nevertheless, during the last decade intensive research has led to new insight into putative escape mechanisms. This includes: (i) escape via different modes of vascularization mediated by sprouting angiogenesis, vasculogenesis, vessel co-option and vascular mimicry (reviewed in [32, 51]); (ii) recruitment of pro-angiogenic myelogenic cells from the circulation [44]; (iii) secretion of numerous alternative pro-angiogenic factors such as bFGF, PIGF, Ephrins, PDGF-C, SDF-1α (reviewed in [33] and others [21, 31, 35]. Also, increased glioma cell invasion has been seen following suppressed VEGF signaling [8, 27, 47, 48]. For instance, a relative hostile hypoxic microenvironment induced by anti-angiogenic therapy may lead to an up-regulation of pro-invasive programs characterized by c-MET/HGF-induced mesenchymal transformation [24, 43], β1-integrin up-regulation [23] and increased MMP expression [13]. To add to this complexity, we have previously shown in a human GBM-derived xenograft model (that recapitulates the genotype and phenotype of the corresponding patient tumors) that vascular remodeling induced by anti-VEGF treatment leads to a more hypoxic tumor microenvironment which appears to favor metabolic changes in the tumors toward glycolysis, and where the net result is increased tumor cell invasion into the normal brain [27]. To study how bevacizumab affects the GBM metabolism, we performed ¹³C₆-glucose injections into tumor-bearing animals, followed by LC–MS based metabolomic analysis to determine in detail metabolic flux and steady-state metabolites in tumor extracts after treatment. To assess tumor growth, we also used complementary in vivo imaging methods including MRI and PET. Moreover, to determine if the observed changes in metabolism were caused by an adaptive response or clonal selection events within the tumors, we performed a flow cytometric ploidy analysis and multiparameter phenotyping on treated and non-treated tumors.

We have previously shown that bevacizumab induces higher levels of lactate in P3 tumors, suggesting an increase in glycolytic activity [27]. We therefore performed a detailed metabolic flux analysis of the P3 xenografts. First, we infused ¹³C₆-glucose i.v into normal rats and collected blood samples at 15, 20, 60 and 120 min after infusion. The blood samples were then subjected to LC–MS analysis where both ¹³C-labeled and unlabeled glucose were quantified. These initial studies revealed a rather quick utilization of glucose in the rats (Supplementary Fig. 2). Based on these results, we harvested the brain 15 and 120 min after ¹³C₆-glucose injection for metabolomic analysis. Tissues were collected from the main tumor mass as well as from the contralateral normal brain (Fig. 1). For metabolic analysis, metabolites were extracted from tumors and contralateral brain samples and quantified by LC–MS ion counts. The data of the metabolite M + 0 mass isotopologue and of M + 6 glucose isotopologue were then normalized and analyzed by principal components analysis (PCA). Clustering of samples between brain and tumor, and between treated and control animals were evident 15 min after labeled glucose injection (Fig. 3a, left panel). Distinctions between treated and control groups in brain or tumor samples were, however, less evident 120 min after injection (Fig. 3a, right panel).

A detailed analysis of metabolites 15 min after the injection of labeled ¹³C glucose revealed increased levels of labeled glucose (m + 6) in the tumors of bevacizumab-treated animals (Fig. 3b; left panel), suggesting an increased glucose flux after treatment. In agreement with this (increased glucose consumption), unlabeled ¹²C glucose and pyruvate were lower in the tumors of bevacizumab-treated animals compared to control tumors. At the same time total lactate levels (both labeled and unlabeled) were significantly increased after treatment, indicating an uncoupling of glycolysis from oxidative phosphorylation in favor of lactate production (Fig. 3b; left panel). An increased production of labeled lactate was also observed in the healthy brain which may suggest an effect of bevacizumab on normal brain cells or a contamination of the tissue with infiltrating tumor cells.

The observations of increased glucose consumption were further substantiated by ¹⁸F-FDG micro-PET imaging showing an increased uptake of this radioactive glucose analog after treatment (Fig. 3b; right upper panel). As expected, concentrations of lactate measured in vivo by MRS showed a 17 % increase in the bevacizumab-treated animals (n = 15) compared to the control group (n = 17), confirming the results of the flux analysis (Fig. 3b; right lower panel).

We also assessed, after bevacizumab treatment, changes in expression of key enzymes associated with glycolysis and with the pentose phosphate pathway (PPP) in both the P3 and P13 xenografts. As shown in Fig. 3c, the glycolytic enzymes aldolase C (ALDOC), hexokinase 2 (HK2) and phosphofructokinase 1 (PFKP), as well as the PPP enzymes glucose-6-phosphate dehydrogenase (G6PD), phosphogluconate dehydrogenase (PGD), transaldolase 1 (TALDO1) and transketolase (TKT) were significantly up-regulated at the transcriptome level, yet with a stronger trend in the P3 tumors compared to P13 (Fig. 3c, left panels). Of note, the enzymes G6PD and PGD are both NADPH-producing enzymes in the oxidative arm of the PPP which may contribute to NADPH production in the tumors. Increased enzyme levels were further confirmed at the protein level for PDK1, PFKP, HK2 and TKT (Fig. 3c, middle and right panels).

In summary, the presented results provide definitive evidence that bevacizumab treatment leads to an up-regulation of glycolysis, concomitantly with an activation of the PPP pathway which may represent a major carbon source for energy and biomass production in the treated tumors.

Angiogenesis is a prominent feature in GBMs where an activation of the VEGF-A signal transduction pathway, mediated by autocrine and paracrine mechanisms, represents a prominent component of angiogenesis induction [45]. Based on promising phase II studies that suggested a reduction in tumor size and prolongation of progression-free survival with radiological response rates ranging between 21 and 61 % for recurrent GBMs [28, 56], two large phase III clinical trials were initiated (AVAglio and RTOG 0825). Despite minor differences, the two trials showed that bevacizumab treatment had no effect on overall patient survival, but with a somewhat longer progression-free survival [6, 15]. Several mechanisms have been proposed on how tumors may resist anti-angiogenic therapy (reviewed in [32, 33, 51], yet the mechanisms behind this refractoriness are unclear. Preclinical models of other cancers suggest that antiangiogenic therapy may cause a temporary vascular normalization leading to increased blood perfusion and improved oxygen delivery [25]. Initial studies on GBMs with cediranib, a pan-VEGF receptor tyrosine kinase inhibitor, showed structural normalization of blood vessels and reduced vessel leakage [2]. However, it was later found that only a small percentage of patients (7 out of 30) displayed a functional normalization of the vessels with increased blood perfusion [52]. We have previously shown in a preclinical model that characterizes human GBMs, that bevacizumab treatment causes an induction of hypoxia leading to reduced tumor perfusion [27]. The patient-derived xenograft models used in these studies have previously been shown to retain the histological features of human GBMs, including infiltrative growth, angiogenesis and necrosis, while maintaining the heterogeneity of tumor subpopulations and the genetic mutations of the primary tumor [16, 49, 58]. We therefore chose to use these models instead of syngeneic models or xenogeneic models established from cell lines that frequently show genetic drift and clonal selection resulting in the loss of key histological and genomic features that characterize human GBMs [30]. In the present work, our previous observations on tumor hypoxia are further substantiated by in vivo ¹⁸F-FMISO PET imaging showing a gradual increase in hypoxia following treatment (Fig. 2c). This phenomenon has also been observed in the clinic [11, 20] where it has been shown to have a predictive value on patient survival after treatment onset [50]. Even though strong induction of hypoxia has been observed by us and other groups, it cannot be ruled out that an increased perfusion may occur locally or in a subset of patients following treatment. Although it is presently not clear if this phenomenon also occurs in bevacizumab-treated patients, it will be interesting to determine the biomarker potential of oxygenation in GBM patients as has been suggested for cediranib [1].

To determine how GBMs adapt to bevacizumab treatment, we performed a detailed ¹³C₆-glucose metabolic flux analysis comparing untreated and treated tumors. Using glucose labeled with ¹³C on all the carbons it has been shown in glioma patients that glucose is cleaved glycolytically for lactate production, but also that an important fraction of glucose carbons enter via pyruvate into the TCA cycle under physiological conditions [36]. Thus, the oxidative mitochondrial metabolization of glucose is important in GBM for energy metabolism and to supply TCA-derived precursors for the synthesis of non-essential amino acids. We show that anti-angiogenic treatment leads to an increased influx of ¹³C₆-glucose into the tumors with a subsequent increase in LDH and ¹³C-labeled lactate levels, while key metabolites associated with the TCA cycle are reduced following treatment. These data provide conclusive evidence that bevacizumab leads to hypoxic tumors with an increase in glycolytic activity and decreased oxidative respiration. The LC–MS analysis also revealed reduced levels of l-glutamine. These observations are in agreement with previous reports indicating that under hypoxia glutamine is used via reductive glutamine carboxylation to fuel macromolecule synthesis, in particular lipogenesis [38].

Interestingly, we also observed reduced levels of l-cysteine and l-cystathione as well as glutathione, suggesting increased oxidative stress in the tumors following bevacizumab treatment. Since the NADP⁺/NADPH ratio in the tumor was unchanged following treatment, the reduction in glutathione may be related to reduced de novo synthesis, as suggested by the changes in the carboxyl cycle metabolites. Maintenance of NADPH status is supported by our finding that key PPP enzymes are up-regulated following treatment, suggesting that this pathway is engaged to maintain NADPH and biomass production and ensure continued tumor growth even at low oxygen levels.

A previous study by Kathagen et al. [26] highlighted the metabolic flexibility of GBM cells, describing a switch between glycolysis and PPP activity following changes in environmental oxygen levels. In apparent contrast to our results, the group reported a down-regulation of PPP enzymes in response to severe hypoxia. However, it should be noted that the study investigated the response to acute changes of oxygenation, while our results are based on chronic hypoxic stress in vivo as also confirmed by ¹⁸F-FMISO imaging. Thus, it is plausible that an initial down-regulation of the PPP in response to acute or short-term hypoxia is followed by increased PPP activity upon adaptation of the tumor cell to hypoxic conditions.

A key question is why the GBMs in our study turn to glycolysis following bevacizumab treatment that will give a growth advantage. In this context, a common property of many cancers is up-regulation of glycolysis, resulting in increased glucose consumption, frequently seen in clinical tumor imaging. It has therefore been suggested that a persistent turnover of glucose to lactate even under aerobic conditions (Warburg effect) is an efficient means to increase biomass production and maintain high proliferation rates [14]. The up-regulation of glycolysis and the induction of microenvironmental acidosis after bevacizumab treatment may lead to a cellular adaptation, on the one hand maintaining continued tumor growth and on the other hand promote invasion. Such adapted cells have a powerful growth advantage, which may promote unconstrained proliferation and invasion. Support for this view comes from intravital microscopy studies showing increased tumor cell invasion in areas with low pH levels in experimental models [10]. Also in our xenografts models, we observed an up-regulation of LDH and increased lactate levels suggesting an acidification of the microenvironment. Interestingly, single invasive LDH-positive tumor cells were observed even in highly oxygenated normal brain areas following treatment, a phenomenon that was also observed in patient biopsies taken at various time points after treatment (Fig. 5). Thus, our findings were also validated in human biopsies. Similar results were recently reported in another study [19] in which lactate levels were assessed in vivo using multivoxel ¹H MR spectroscopy. Elevated levels of lactate were reported in the tumor core. Yet, in the diffuse infiltrative areas, the levels of lactate were lower, possibly reflecting that the limited fraction of infiltrated cells in healthy tissue is insufficient to significantly influence the lactate level measured at the scale of MRS voxels (mmol). It should, however, be emphasized that under normoxia, the LDH-positive infiltrative cells, may convert lactate to pyruvate which may enter the TCA cycle. It is therefore still an open question if the infiltrative tumor cells are truly glycolytic.

To evaluate whether the changes observed after treatment could be explained by a selection of a new genetic clone and/or putative stem cell-like populations with increased resistance to treatment, we performed flow cytometric ploidy and multiparameter phenotypic marker studies on non-treated and treated tumors. We did not observe any changes in ploidy levels, genetic aberrations and stem-like associated cell membrane marker profiles following bevacizumab therapy, suggesting that the change in cellular behavior after treatment represents an adaptive metabolic response. Yet, further studies are warranted to delineate in detail to what extent clonal subpopulations within GBMs contribute to the changes in metabolic phenotypes observed.

Our findings of increased intratumoral hypoxia and glycolytic activity, leading to increased acidification of the microenvironment and invasiveness, suggest that therapies that target the metabolic adaptation mechanisms may provide a synergistic effect on bevacizumab therapy. Indeed, a number of preclinical studies have started to show reduced tumor progression when bevacizumab is combined with inhibitors targeting evasion mechanisms related to invasion and metabolism [5, 7, 22, 24, 29, 34, 37, 39, 42].

In conclusion, we show that bevacizumab treatment causes a metabolic shift in the tumors toward glycolysis where the cells depend less on oxidative respiration to produce ATP. This makes the recurrent tumor cell populations less dependent on angiogenesis. Moreover, this shift supports the notion that H(+) diffuses from the proximal tumor microenvironment into adjacent normal tissues where it causes tissue remodeling that permits local invasion. In this context a future challenge will be to develop therapeutic approaches that target adaptive metabolic responses in tumors as a result of therapy-induced perturbations.