Expression of LDHA and LDHB are mainly restricted to specific glioblastoma areas
We have previously showed that regional expression of specific molecules define GBM hallmarks5,27. Regional expression of the lactate dehydrogenase isoforms was first analyzed in a patient-derived xenograft mouse model by immunohistochemistry. Both LDH types were predominantly express on invasive cells compared to healthy tissue, however LDHA was mainly expressed in the central hypoxic area of the tumor and in some invasive cells, while LDHB was mainly expressed in peripheral tumor areas (Fig 1a). LDHA and LDHB did not colocalize in tumor cells, showing a distinct spatially-restricted expression pattern (Fig 1a). Next, we embedded spheroids invading in a 3D collagen matrix in paraffin, a model that recapitulates GBM oxygenation regional heterogeneity28, to analyse LDHA and LDHB expression in coronal paraffin sections (Fig 1b). LDHA was found highly expressed in the central hypoxic area but also in some single invasive cells, while LDHB was highly expressed at spheroid borders and invasive areas (Fig 1c). Of note, only few cells expressed both enzymes. In silico analysis on a single-cell RNAseq database from invasive and central GBM areas29 demonstrated a similar pattern, with preferential expression of LDHA in the central area and LDHB in the peripheral area albeit mRNA of both enzymes were found in a portion of invasive tumor cells (Supplementary Fig 1a). When using data extracted from the IVYGAP project, distinct regional LDHA and LDHB expression was observed, as well as a positive correlation between LDHA and the hypoxic transcription factor HIF1a (Supplementary Fig 1b). When analyzed in the GBM-TCGA data set, LDHA was found to be a marker of poor prognosis. LDHB expression was, on the contrary, linked to favourable prognosis (Supplementary Fig 1c). In summary, LDHA and LDHB expression is reciprocally restricted to hypoxic or peripheral areas.
Lactate modulates glioblastoma invasion by fuelling energy metabolism pathways
On physiological conditions, brain tissue is exposed to low level of oxygen, i.e. physiological hypoxia, that ranges from 0.5 to 7 % depending on the distance from blood vessels30. Local O2 concentration can be even lower in brain tumors, reaching around 0.1% in the core area. We, therefore, investigated whether stem-like cells (P3 or BL13 cells) in culture express LDHA and/or LDHB, and whether expression of these enzymes can be modified by incubating cells at 0.1% O2. We observed that LDHA expression was upregulated under hypoxic conditions (3x fold induction at 72 hours), while LDHB expression did not change (Fig 2a and Supplementary Fig 2a). Lactate production was also increased under hypoxia 0.1%, correlating with higher LDH activity in P3 and in BL13 cells (Fig 2b-d and Supplementary Fig 2b-d). Lactate has been reported to sustain tumor growth via monocarboxylate transporters (MCT1/4)31. To this aim, we tested the effect of lactate on tumor cell proliferation and invasion. To rule out a lactate-induced cytotoxic effect, lactate was added to the medium of P3 cells and cytotoxicity was measured. Lactate concentrations starting at 30 mM induced cytotoxicity at 24 hours (Supplementary Fig 2e). The effect of non-cytotoxic lactate concentrations was then tested and spheroid growth and invasion measured. Spheroid growth was significantly inhibited by 20 mM lactate (Fig 2e), while cell invasion increased in the presence of lactate from 10 mM onwards (Fig 2f). Lactate treatment also induced changes in cell morphology, promoting an elongated cell shape, reminiscent of a mesenchymal phenotype (Supplementary Fig 2f). In absence of glucose, lactate by itself induced a strong effect on cell invasion (Fig 2g). To rule out an effect of acidification, we tested cell invasion at the same pH of lactate treatment (pH 6.8), and no effect was observed (Fig 2g). Rotenone, a respiratory chain complex I inhibitor, completely blocked lactate-induced invasion. This suggests that lactate is fuelling mitochondria activity to promote invasion (Fig 2g). Pyruvate, the main LDH substrate was then tested in invasion experiments, and invasion rates was increased in comparison to control conditions but remained lower than under lactate stimulation (Supplementary Fig 2g). Of note, the respiration capacity of P3 cells was similar in complete medium, with glucose or lactate (Fig 2h). Lactate consumption by GBM mitochondrial activity was followed by [13C3]lactate infusion in glucose-starved P3 cells, wherein labelled TCA intermediates (citrate, oxoglutarate) and amino acids (alanine, glutamate, glutamine) were detected in the endometabolome after 1 hour but also in the exometabolome (Fig 2i and Supplementary Fig 3a-c). Intracellular abundance of lactate and pyruvate was found unchanged along the experiment (Figure 2i and Supplementary Figure 3a), but the amount of [13C3]lactate and [13C3]pyruvate immediately reached its maximum level (Supplementary Figure 3b). The appearance of [13C2]citrate from the beginning indicated a fast reaction from acetyl-CoA, and higher isotopologues increased quickly after, indicating an active TCA cycle (Supplementary Figure 3c). These results shed light on the pathways involved in lactate metabolism of GBM cells and is a direct demonstration that lactate is used as an energy source via the TCA cycle. Only a slight increase or almost no labelling of amino-acids not associated with the TCA cycle or gluconeogenesis was observed and, thus, this was not included in the analysis.
Double LDHA/B KO impairs in vitro and in vivo GBM growth and invasion
First, we transduced P3 and BL13 cells with lentiviral Crispr-Cas9 constructs to fully eliminate LDHA and/or LDHB expressions (sgControl, sgLDHA, sgLDHB, and sgLDHA/B). Knock-out (KO) was validated by Western-blot blot in single LDHA or LDHB KO cells, and in double LDHA/B KO cells at 21 or 0.1% O2 (Fig 3a and Supplementary Fig 4a). Next, we performed functional experiments to further validate knockout strategy. Lactate secretion was only abolished on double LDHA/B KO cells (Fig 3b). LDH activity was decreased in LDHA, slightly decreased in LDHB single KO cells, but massively decreased in double KO cells (Supplementary Fig 4b). LDHB activity was not detected in LDHB or LDHA/B KO cells (Supplementary Fig 4c). Next, we transduced the lactate-sensitive FRET biosensor into all P3 cells to detect intracellular lactate levels (Supplementary Fig 4d)32. Basal lactate concentrations were higher in control, LDHA or LDHB KO cells than in double LDHA/B KO cells (Fig 3c). Adding oxamate, a potent LDH inhibitor that also induces lactate efflux by MCT trans-acceleration33, followed by a MCT blocker cocktail (diclofenac and AR-C155858) unveil that basal lactate concentrations are higher in control, LDHA or LDHB KO cells compared to double LDHA/B KO cells, are result explained by decreased lactate production (Fig 3c). Spheroid growth was next monitored over one week at 1% O2 and showed that double LDHA/B KO strongly reduced spheroid growth, unlike LDHA KO that moderately decreased spheroid growth and LDHB KO that had no effect (Fig 3d, left panel). This was associated with a strong ethidium homodimer-1 staining for visualization of apoptotic events (Fig 3d, right panel). This cell death signature was also supported by a high Annexin-V staining in double KO cells detected by cytometry (Supplementary Fig 4e).
The invasion capacity was analyzed in all cell lines and a moderate decrease in invasion was observed in LDHA KO in both P3 (Fig 3e) and BL13 cells (Supplementary Fig 4f). Nevertheless, LDHB KO P3 but not LDHB KO BL13 cells had higher invasive capacities (Fig. 3e and Supplementary Fig 4f). For LDHA/B KO cells, under 0.1 % O2, the invasive capacity dropped by 75% for P3 cells (Fig 3e), and by 50% for BL13 cells (Supplementary Fig 4f).
Spheroids from single and double LDHA/B KO cells were injected into immunodeficient mice to evaluate in vivo tumor development and survival (Fig 3f). No morphological differences were observed between control and single LDH KO (LDHA KO or LDHB KO) tumors (Fig 3g). However, double LDHA/B KO tumors were much smaller and less invasive than control tumors (Fig 3h), which correlated with an increase in mouse survival, in both, the P3 (Fig 3i) and the BL13 model (Supplementary Fig 4i). Only a small increase in mouse survival was observed in the LDHA KO group, and, surprisingly, a drastic decrease in survival in LDHB KO group (Fig 3i), the latter was most likely due to haemorrhages at the tumor site. Vascular endothelial growth factor (VEGF), the main inducer of neoangiogenesis by tumor cells, was quantified by ELISA and an increase was only observed in LDHB KO cells (Supplementary Fig 4g). The VEGF inhibitor bevacizumab (bev) was then injected in mice bearing single or double LDHA/LDHB KO P3 tumors. An increase in survival was observed in mice harbouring bev-treated LDKB KO tumors and treated with bev, similar to control animals (Supplementary Fig 4h). Increased survival was also seen in mice with double LDHA/B KO tumors, while no difference was observed in mice with control or LDHA KO tumors (Supplementary Fig 4h).
Metabolic switch in double LDHA/B KO cells under hypoxia delineates vulnerabilities
To trace metabolic fluxes at 0.1% O2, cells were starved and infused with [13C6]glucose for 24 or 48 hours. Principal component analysis (PCA), defining global RNA regulation, showed that metabolism of double LDHA/B KO cells switched after 48 hours at 0.1% O2 when compared to control or single KO tumors (Fig 4a). Exometabolome of double LDHA/B KO cells strongly differed from other cells, by a decrease in glucose consumption and an absence of lactate production (Fig 4b). In addition, pyruvate and its derivates (acetate and formate) were only secreted in double LDHA/B KO cells (Fig 4b).
RNA sequencing was then performed on the same cells under 21 or 0.1 % O2, and confronted with metabolomics data. Metabolograms were used to integrate transcriptomics and metabolomics data as previously described34. The changes of each metabolite/transcript pair were determined and average metabolome/transcriptome variations of a metabolic pathways were visualized by comparing two conditions (Fig 4c). Based on global results showed previously (Fig 4a), metabolograms were built by analyzing total metabolite abundances in P3 control cells at 21% and while adapting at 0.1% O2 (Fig 4c, left panels, and supplementary Fig 5), and compared to P3 control and P3 double LDHA/B KO cells at 21% (Fig 4c, middle panels, and supplementary Fig 5) or at 0.1% O2 (Fig 4c, right panels, and supplementary Fig 5). When changes in consensus gene expression were analyzed, differences in glycolysis or oxidative phosphorylation for all comparisons were seen, but only negligible variations for amino acid synthesis (Fig 4c). Significant variations in the majority of pathways from the metabolome perspective were observed for most of the comparisons (Fig 4c). For P3 control cells, the majority of transcripts were related to glycolysis and showed increases in expression at 0.1% O2 when compared to 21% O2 (18 up, 12 down), whereas the majority of metabolites significantly decreased at 0.1% O2 (0 up, 4 down) (Fig 4c, left panels). Metabolograms were heterogeneous when comparing changes between P3 control and double LDHA/B KO cells at (1) 21% and (2) at 0.1% O2 (Fig 4c, middle and right panels). P3 double LDHA/B KO cells had a strongly modified metabolism, as seen by the increase in all metabolites and transcripts related to glycolysis and oxidative phosphorylation, especially under 0.1% O2 (Fig 4c).
Since metabolograms are based on the total metabolite abundances, metabolite abundance was calculated after [13C6]glucose labelling using the fractional contribution and was incorporated into the central pathway of the carbon metabolic network map obtained from previous comparisons (Fig 4c). To explore the relationship between transcriptomic and metabolomic data, transcript levels of metabolism-related genes were incorporated into the central carbon metabolic network pathway map (Fig 4d and supplementary Fig 6a-b). At 0 h, the difference between P3 double LDHA/B KO versus control cells was low to non-significant (Supplementary Fig 6b). The detailed metabolic map depicts the increase in abundance for the majority of metabolites after [13C6]glucose labelling in double KO LDHA/B cells versus control cells for the glycolytic and the oxidative phosphorylation pathways (Fig 4d). Fourteen transcripts from the glycolytic pathway (HK1, HK2, GPI, PFKB3, PFKB4, ALDOC, TPI1, PGK1, ENO1, ENO2 and PKM), and from the Krebs cycle (ACO2, SUCLG1and SDHB) were upregulated, while 6 were downregulated (GAPDHS, LDHA, LDHB, DLAT, SDHD, CS) (Fig 4d). These findings indicate that, under hypoxia, double LDHA/B KO deregulates cell metabolism at many levels. Global transcripts at 0.1% O2 were increased in P3 control cells when compared to 21% O2 (Supplementary Fig 6c, upper panels), impacting metabolism, and in particular glycolysis (HK2, ENO1, ENO2, GAPDH, PKM, LDHA). At 21 % O2, major biological pathways are related to membrane and organelle dynamics during cell division and cell cycle regulation, at the contrary, in more unfavorable condition at 0.1% O2, pathways are involved to low oxygen adaptation and maintenance of energy production (Supplementary Fig 6c, upper panels). Hypoxic stress has been found to decrease the nucleotide pool35. Indeed, the abundance of nucleotides was lower while fractional contribution was enriched suggesting a continuous but limited turnover (Supplementary Fig 7). Under 0.1% O2, the transcription profiles were more enriched in double LDHA/B KO cells when compared to P3 sgControl cells (Supplementary Fig 6c, lower panels). This was also seen at 21% O2 (Supplementary Fig 6c, middle panels). The differences in abundance and isotopolog contribution between control and double KO LDHA/B cells were less pronounced in 21% O2. In hypoxia, the abundance in nucleotide monophosphates (AMP, GMP and UMP) was also higher in double KO LDHA/B cells (Supplementary Fig 7). The fractional contribution of purines and pyrimidines mainly consisted in m+5 labeling in double KO LDHA/B cells whereas m+5, m+7 and m+8 labeling was seen in control cells.
Double LDHA/B KO cells reorganize their mitochondrial respiratory chain to support growth under 0.1% O2 concentration
To corroborate the RNA sequencing results (Fig 4c-d and supplementary Fig 6a), we exposed P3 cells to 21 or 0.1% O2 and analyzed the subunits of the mitochondrial respiratory chain by Western blot. NDUFB8, SDHA, UQCRC2, COXII protein levels increased upon 0.1% O2 in double LDHA/B KO cells, but only NDUFB8 and COXII at 21% O2 (Fig 5a). Expression of the subunit V ATP5A complex did not change in high or low oxygen conditions (Fig 5a). For single LDHB or LDHA KO cells, only minor changes were observed (Fig 5a). Immunostaining of mitochondria networks revealed an increase in the mitochondrial mass and modifications in network shape, defined as aspect ratio, only in double LDHA/B KO cells (Fig 5b). Under uncoupling conditions, double LDHA/B KO cells possess higher respiratory capacity than the other cells (Fig 5c). Two in vivo strategies were employed to inhibit these metabolic adaptations. The double LDHA/B KO tumors were challenged either with the respiratory complex I inhibitor phenformin or by irradiation (Fig 5d). Phenformin injections did not improve survival of KO LDHA/B tumor bearing mice when compared to control tumors (Fig 5e), while irradiation massively increased survival in mice with double LDHA/B KO tumors (Fig 5e).
Use of anti-epileptic drug targeting LDH activity efficiently reduces GBM development
Mice were next treated with the anti-epileptic drug stiripentol, which inhibit LDH activity and is capable to cross the blood-brain barrier36. Importantly, this drug has already been used in preclinical therapy studies in vivo in mice26 (Fig 6a). We first validated LDH inhibition in vitro in P3 GBM cells by intracellular lactate recording with the FRET biosensor Laconic, finding that lactate level decreased and its production was inhibited a 50% after 24 hrs treatment with 500 µM of stiripentol (Fig 6b). This was correlated with a drastic decrease of basal and uncoupled respiration in pre-treated P3 cells (Fig 6c). Furthermore, stiripentol significantly reduced P3 spheroid proliferation (Fig 6d) and invasion at 0.1% O2 (Fig 6e). To test stiripentol anti-cancer properties in vivo, intraperitoneally injection of either 100 mg/kg of stiripentol (stiri) or 10 mg/kg of bevacizumab (bev) alone, or in combination (combo) in animals implanted with P3 cells were then carried-out (Fig 6f). Stiripentol treatment had some effect on mouse survival when administered alone but its combination with bevacizumab treatment significantly increased survival (Fig 6f). Immunohistochemistry analysis showed reduction in tumor size and invasion when treated with the combinatory regimen (Fig 6g).