Mesenchymal BTICs were more invasive than proneural BTICs
We analyzed 8 different primary BTIC lines and their respective differentiated counterparts (TCs) (Suppl. Table 2). Four BTIC lines each were classified as proneural (BTIC-7, BTIC-8, BTIC-17, BTIC-18) and mesenchymal (BTIC-10, BTIC-11, BTIC-12, BTIC-13) according to the Verhaak classification (5) based on microarray analyses. Classification into molecular subgroups was confirmed using next generation sequencing and expression of proneural and mesenchymal markers in immunocytochemistry.
Invasion of tumor cells into the surrounding healthy tissue is one major hallmark of glioblastoma. As we hypothesized that proneural and mesenchymal cells might differ in their invasiveness, we explored invasion of BTICs and TCs on organotypic brain tumor slices (OBSCs). Therefore, green-fluorescent mesenchymal and red-fluorescent proneural BTICs and TCs were monitored over 14 days under the fluorescent microscope. Mesenchymal BTICs and TCs were significantly more invasive than proneural BTICs and TCs (Fig. 1), with the maximum migratory distance per day being about 18-times higher in mesenchymal than proneural cells. Other basal cell characteristics including proliferation, migration and expression of signaling molecules did not differ significantly among the lines, when explored under low-glucose conditions (5.5 mM) and normoxia (Suppl. Fig. 1).
Proneural BTICs were more responsive to OXPHOS inhibition
Previous assays had already indicated significant heterogeneity between BTICs and TCs treated with metabolic agents (11). To explore this observation further, we comprehensively assessed the response of several BTICs and TCs to metformin, which was used as a model substance to inhibit OXPHOS.
The effects of metformin on proliferation were explored using the CyQuant Direct Cell Proliferation Assay 48 and 96 hours after treatment with various concentrations of metformin. Treatment with metformin led to a dose- and time-dependent inhibition of proliferation among all investigated BTICs and TCs (Suppl. Fig. 2). Interestingly, proneural BTICs were significantly more responsive to metformin than mesenchymal BTICs and TCs, especially at lower concentrations of metformin (Fig. 2A).
Next, we evaluated, whether the observed differences in proliferation might be due to apoptosis, as demonstrated previously (29). In our model, 1 µM staurosporine reliably induced apoptosis, as indicated by increased expression of cleaved caspase 3 in BTICs and TCs (Suppl. Fig. 3). In contrast, we did not observe a significant increase in caspase cleavage after treatment with increasing doses of metformin in most of the BTICs. We did observe a dose-dependent activation of AMPK and inhibition of mTOR and STAT3 (Fig. 2B and Suppl. Fig. 3), leading to increased autophagy (Suppl. Fig. 3L-S). To confirm results from Western Blot analyses, we used a vector containing the human LC3B gene fused to green fluorescent protein (GFP) to visualize autophagosome formation (exemplarily shown for BTIC-8, Suppl. Fig. 3T).
The impact of increasing doses of metformin on migration was investigated in spheroid migration assays over 16, 24, 40 and 48 hours. Whereas high-, but not low-dose metformin inhibited migration of most BTICs and TCs, one mesenchymal cell line (BTIC-13) did not change its migration behavior regardless of the dose of metformin (Suppl. Fig. 4). When comparing the group of proneural and mesenchymal BTICs, proneural BTICs were significantly more responsive to high-dose metformin (Fig. 2C). Early time points were examined to distinguish migration from proliferation. Treatment with metformin led to a dose-dependent inhibition of cellular oxygen consumption with different basal oxygen consumption of BTIC-18 (Fig. 2D) and TC-18 (Fig. 2E.)
Oxygen consumption did not correlate with molecular subtype
We investigated routine and FCCP stimulated cellular respiration using high-resolution respirometry. In line with published data, metformin severely inhibited complex-I-dependent-respiration (Fig. 3A), leading to comparable effects of 0.5 µM rotenone.
BTIC-18, which had shown high sensitivity to metformin in proliferation assays, respired markedly more than other BTICs at baseline and after use of the uncoupling agent FCCP. However, endogenous and FCCP stimulated respiration varied widely among other BTICs and TCs (Fig. 3B, C). Consequently, there were no significant general differences in cellular respiration between the group of proneural and mesenchymal BTICs and TCs (Fig. 3D, E).
Next, we hypothesized that differential response to OXPHOS inhibitors may be due to different abilities to use or activate glycolytic rescue mechanisms. Therefore, we explored extracellular metabolites. Interestingly, mesenchymal cells consumed far more glucose than proneural cells (Fig. 4A and 4B) and showed increased extracellular lactate levels. Treatment with metformin led to increased consumption of glucose and production of lactate. We also performed Seahorse analyses to gain deeper insights into extracellular acidification rates (ECAR) and glycolytic reserve capacity. The ECAR and the glycolytic reserve capacity were significantly higher in mesenchymal than proneural BTICs (Fig. 4C, D), which may however also be attributed to CO2 production by dehydrogenases.
Mesenchymal BTICs showed increased expression of glycolytic genes
To identify genetic differences underlying differential metabolic patterns of proneural and mesenchymal BTICs, we performed a gene set enrichment analysis based on the mRNA expression of 36 published BTICs including those used for the present functional and metabolic assays (30). Data are deposited at the gene expression omnibus (GEO) functional genomics data repository under the accession numbers GSE51305 and GSE76990. BTICs were segregated into proneural and mesenchymal BTICs according to the gene signatures published by Verhaak et al.(5) Interestingly, using hallmark analyses, we found the metabolic hallmark glycolysis to differ between proneural and mesenchymal BTICs (Fig. 5A), with mesenchymal BTICs showing a more glycolytic transcriptome. Significantly changed genes of the hallmark glycolysis are depicted in Suppl. Table 3, taking into consideration that microarray data may not necessarily reflect protein expression.
Among the differentially expressed glycolytic genes, we found monocarboxylate transporter 4 (MCT4), which is important for the outward transport of lactate to be one interesting candidate for future research. MCT4 mRNA expression was significantly higher in mesenchymal than proneural BTICs on both the mRNA (Suppl. Table 3) and protein level (Fig. 5B). Using data from the Cancer Genome Atlas (TCGA), we were able to confirm that MCT4 (Fig. 5C), but not MCT1 expression (Fig. 5D), was significantly higher in mesenchymal than proneural GBs. Intracellular lactate levels were higher in BTIC-8 and BTIC-18 than in BTIC-11 and BTIC-13 (Suppl. Fig. 5A), and reduced after treatment with metformin.
To determine if MCT4 plays a major role in mediating the response to metformin in proneural versus mesenchymal BTICs, we inhibited MCT4 in several BTICs and explored their response to metformin. Silencing of MCT4 with siRNA was feasible (Suppl. Fig. 5B-E), but did not increase the inhibitory effects of metformin on tumor cell proliferation and migration (Suppl. Fig. 5F, G), even when used after glutamine withdrawal or under hypoxia (data not shown). Based on our results, although MCT4 may play a role in the different metabolic preferences of proneural and mesenchymal BTICs, sensitivity to metformin is likely based on several different genetic and metabolic alterations. We hence investigated glycolysis in proneural and mesenchymal BTICs in more detail.
Proneural BTICs redirected glucose to the PPP
Considering that expression of glycolytic genes may not necessarily translate into increased glycolysis, we next investigated activity of the key glycolytic enzymes hexokinase, glucose-6-phosphate dehydrogenase, and phosphofructokinase. Their expression in TCGA varied according to enzyme isoforms (Suppl. Fig. 6).
Mesenchymal BTICs showed increased activity of hexokinase and phosphofructokinase, whereas proneural BTICs showed increased activity of glucose-6-phosphate dehydrogenase (Fig. 6B-D, Suppl. Fig. 6E, K, O). We suspected that proneural BTICs metabolized more glucose via the pentose-phosphate pathway (PPP). To prove this assumption, we performed pyruvate and lactate tracing using [13C2-1,2]glucose as tracer substrate (Fig. 6A, E-J). Indeed, we were able to confirm increased flux through the PPP in BTIC-18 as indicated by the higher abundance of m+1 isotopologues for pyruvate and lactate.
Key findings could be related to clinical data
To validate our in vitro data, we compared our results with survival data from glioma patients. First, high expression of MCT-4 (Suppl. Fig. 7A), but not MCT-1 (Suppl. Fig. 7B) translates into inferior patient survival in the TCGA cohort. Although not reaching statistical significance, there is a trend for improved survival among patients with high expression of G6PDH (Suppl. Fig. 7C).
Second, using a meta-analytic approach, we retrieved available data from published studies on clinical responses to metformin, overall and according to subgroups. There are several observational studies on metformin in glioma, several (mostly ongoing) clinical trials and few suggested treatment regimens outside of clinical trials (Suppl. Table 4). When retrieving data from observational studies relating use of metformin to overall (OS) or progression-free survival (PFS), there is a trend for prolonged OS and PFS of patients with GBM on metformin in some of the studies (Suppl. Fig. 7B, C). Interestingly, there is only one study on high-grade gliomas that stratified for WHO grade and found a significantly improved OS and PFS among patients with WHO grade III glioma using metformin.