Pyruvate Kinase Inhibits Proliferation during Postnatal Cerebellar Neurogenesis and Suppresses Medulloblastoma Formation

Increased aerobic glycolysis, a common feature of proliferating cells during developmental and malignant growth (1, 2), has been proposed as a target for anticancer therapy. CGNPs are transit amplifying cells that proliferate in the postnatal brain (3, 4) that utilize aerobic glycolysis during normal brain development (5). CGNPs are also cells-of-origin for SHH-subgroup medulloblastoma (6, 7). Medulloblastomas coopt developmentally regulated programs of CGNPs (6, 7), including the metabolic phenotype of increased aerobic glycolysis (5, 8). Understanding how aerobic glycolysis supports CGNP proliferation and medulloblastoma tumorigenesis is essential for designing metabolically directed medulloblastoma therapies.

During postnatal brain development, CGNPs proliferate in the cerebellum in response to locally secreted Sonic Hedgehog (SHH), generating the largest neuron population in the mammalian brain (9–11). In synchrony with increased CGNP proliferation, SHH signaling induces hexokinase-2 (HK2) and aerobic glycolysis (5, 8). Approximately 30% of medulloblastoma patients show SHH pathway activation (12, 13) and mice with activating SHH pathway mutations develop spontaneous medulloblastomas that recapitulate the human disease (14, 15). These tumors arise from CGNPs and, like their cells-of-origin, upregulate HK2 and aerobic glycolysis (5, 8). CGNP proliferation and SHH-driven tumorigenesis in mice typify aerobic glycolysis supporting tumorigenesis arising from developmental growth.

We have previously shown that conditional deletion of Hk2 in the developing brain blocks SHH-induced aerobic glycolysis, disrupts CGNP differentiation, and reduces medulloblastoma growth, extending the survival of medulloblastoma-prone mice (5). These findings demonstrate the importance of aerobic glycolysis in development and cancer, suggesting that blocking glycolysis through HK2 inhibition may produce a clinically significant antitumor effect. Development of selective HK2 inhibitors for anticancer therapy, however, has been problematic (16). Whether targeting glycolysis downstream of HK2 would similarly inhibit tumor growth is unknown. To determine whether the distal portion of the glycolytic pathway is required for the proliferation of normal and transformed CGNPs, we analyzed the effect of disrupting pyruvate kinase (PK) during cerebellar neurogenesis and medulloblastoma formation.

While HK2 initiates glycolysis, PK catalyzes the final step, converting phosphoenolpyruvate (PEP) and ADP into pyruvate and ATP (17). Pkm is the pyruvate kinase expressed in the brain (18). Alternative splicing of a single exon from Pkm gives rise to either Pkm1, which includes exon 9, or Pkm2, where exon 9 replaces exon 10 (19). PKM1 is constitutively active and typically expressed in differentiated cells, while PKM2 has a regulated activity and is commonly expressed in cancer (20–24). PKM2 can be activated by fructose-1,6-bisphosphate, an upstream glycolytic intermediate (23, 24) and inactivated by interaction with tyrosine-phosphorylated proteins responding to growth factor receptor signaling (20, 21, 25). While the regulated activity of PKM2 may allow proliferating cells to adjust their metabolism to dynamic requirements (2), the specific benefit of low PKM2 activity has not been identified.

PKM2 is expressed in diverse cancers, suggesting a role in cancer growth. Xenograft tumors engineered to overexpress either isoform showed that PKM2 expression confers a relative growth advantage (26). Other studies have reported nonmetabolic, growth-promoting functions such as transcriptional regulation (27, 28) and histone phosphorylation (29, 30). However, isoform-specific deletion studies show that PKM2 is not required for tumor formation. Aged mice with isoform-specific Pkm2 deletion develop liver cancer (31) and breast cancers are accelerated in Brca1-deleted mice with codeletion of Pkm2 (32). These studies identify cancers in which Pkm2 deletion promotes tumor growth. However, neither study showed an effect of Pkm2 deletion on tumor glycolysis, potentially because PKM2 activity may be fully downregulated in established tumors. To identify metabolic processes linking Pkm2 to growth, it is necessary to analyze Pkm2 deletion in normal proliferative cells that, unlike cancer cells, are not locked into a proliferative state. CGNPs present an ideal opportunity to study Pkm2 function because these cells can modulate both aerobic glycolysis and proliferation, and are also vulnerable to malignant transformation in medulloblastoma tumorigenesis.

PKM2 is induced by SHH in CGNPs and expressed in the ND2:SmoA1 mouse model of medulloblastoma (8), suggesting a role in SHH-driven growth. PKM2 upregulation has also been noted in MYC-amplified group 3 medulloblastomas (33), suggesting a role for PKM2 in tumor growth across human medulloblastoma subgroups. Here, we analyzed PKM-mediated regulation of CGNP glucose metabolism, proliferation, and tumorigenesis. Like Hk2 deletion, Pkm2 deletion decreased CGNP conversion of glucose to lactate. However, unlike Hk2 deletion, Pkm2 deletion increased SHH-driven proliferation, and accelerated medulloblastoma formation in mice expressing a constitutively active Smo allele. The differential effects of impeding aerobic glycolysis through deletion of Hk2 or Pkm2 identify the glycolytic steps upstream of PK as critical to supporting proliferation during development as well as contributing to tumorigenesis.

All wild-type and genetically engineered mice were maintained on the C57/Bl6 background with at least 5 backcrosses. ND2:SmoA1 mice were provided by Dr. James Olson (Fred Hutchinson Cancer Research Center, Seattle, WA). Math1-Cre mice were provided by Dr. David Rowitch (University of California San Francisco, San Francisco, CA) and Robert Wechsler-Reya (Sanford-Burnham Medical Research Institute, La Jolla, CA). Eva Anton (University of North Carolina at Chapel Hill, Chapel Hill, NC) provided hGFAP-Cre mice. Pkm2^fl/fl mice were generously shared by Dr. Matthew Vander Heiden (Massachusetts Institute of Technology, Cambridge, MA). Medulloblastoma-prone mice were monitored daily for abnormalities of head shape and movement. At the onset of tumor symptoms, such as weight loss, ataxia, and impaired movement, animals were sacrificed and survival time to onset of symptoms was considered the event-free survival.

For 5-ethynyl-2′-deoxyuridine (EdU) experiments, mice were intraperitoneally injected with EdU (#A10044, Life Technologies) at 40 mg/kg in 50 μL of HBSS and dissected 24 hours later. Brains were fixed in 4% paraformaldehyde in 1× PBS for 24 hours at 4°C, then processed for histology. All animal handling and protocols were carried out in accordance with established NIH practices and approved under University of North Carolina Institutional Animal Care and Use Committee #13-121.0 and 15-306.0.

CGNPs were isolated and cultured as previously described (5, 34). Briefly, cerebella were dissected from P5 pups, dissociated, and allowed to adhere to coated culture wells in DMEM/F12 (#11320, Life Technologies) with 25 mmol/L KCl, supplemented with FBS and N2. After 4 hours, media were replaced with identical serum-free media. Cells were maintained in 0.5 μg/mL SHH (#464SH, R&D Systems) or vehicle (0.1% BSA in 1× PBS). Where indicated, 1 μmol/L vismodegib (#S1082, Selleck Chemicals) was added to cultures after the first 24 hours, with cells harvested 24 hours after drug treatment. In vitro CGNP proliferation was measured by EdU incorporation after a 1-hour exposure to 20 μmol/L EdU. EdU was visualized following the manufacturer's protocol (#C10337, Life Technologies). Cell counts were performed using Leica-Metamorph software (Molecular Devices). No cell lines were use in these studies.

¹H NMR and liquid chromatography-mass spectrometry (LC-MS) acquisition methods are described in detail in the Supplementary Materials and Methods. Briefly, CGNPs of each genotype were cultured in at least three replicate wells. Explanted CGNPs were maintained in [1,6-¹³C] glucose media for 24 hours followed by media sampling and cell extraction for metabolomic analysis. We normalized to cell number at 24 hours to account for differences in growth rates by counting cells in a replicate set of wells cultured in parallel.

For the enzymatic measurement of lactate, media from at least three replicate wells were sampled at the specified time points and lactate was quantified using the L-Lactate Assay Kit (#1200011002, Eton Bioscience) as per manufacturer's protocol.

For PK activity assays, cells from at least three replicate wells per condition, or whole cerebella from at least three replicate mice per genotype, were lysed and processed for the colorimetric assay per manufacturer's protocol (#K709-100, BioVision Inc.).

Mouse brain and tumor tissues were processed for IHC as described previously (35) using antibodies listed in Supplementary Materials and Methods. EdU was visualized per manufacturer's protocol. After IHC and EdU staining, where indicated, nuclei were counterstained with 200 ng/mL 4′6-diamidino-2-phenylindole (DAPI; #D1306, Thermo Fisher Scientific) in 1× PBS for 5 minutes. Stained slides were digitally acquired using an Aperio ScanScope XT (Aperio). Where indicated, maximum intensity projections of stained tissue sections were acquired on a Zeiss LSM 780 confocal microscope with a Plan-Apochromat 20× objective (NA 0.8). To quantify EdU, proliferating cell nuclear antigen (PCNA), and p27^Kip1-positive cells in the EGL, the EGL region was manually annotated on each section, which was then subjected to automated cell counting using Tissue Studio (Definiens) for fluorescent slides. For premalignant lesion analysis, the entire cerebellum was annotated and used for cell counts.

Cultured cells and whole cerebella were lysed by homogenization in RIPA buffer containing protease inhibitor cocktail, NaF, and sodium orthovanadate. Protein concentrations were quantified using the bicinchoninic acid (BCA) method (#23229, Thermo Fisher Scientific) and equal concentrations of protein were resolved on SDS-polyacrylamide gels followed by transfer onto polyvinylidene difluoride membranes. Immunologic analysis was performed on the SNAP i.d. Protein Detection System (Millipore) as per manufacturer's protocol with antibodies listed in Supplementary Materials and Methods. Western blots were developed using the enhanced chemiluminescent SuperSignal West Femto Maximum Sensitivity Substrate (#34095, Thermo Fisher Scientific) and digitized using the C-DiGit blot scanner (LI-COR Biosciences). Quantification was performed using Image Studio Lite software (LI-COR Biosciences).

DNA was lysed from toe cuttings and used to generate Pkm2 PCR products. Pkm2 PCR primers were TAGGGCAGGACCAAAGGATTCCCT (forward) and CTGGCCCAGAGCCACTCACTCTTG (reverse). PCR products were extracted from a 0.8% agarose gel using the MinElute Gel Extraction Kit (#28604, Qiagen), cloned by TOPO TA Cloning (#K4575-J10, Life Technologies) and sequenced from the M13F and M13R primer sites by Sanger sequencing (Eton Bioscience).

All patient samples were obtained with consent as outlined by individual institutional review boards. Written informed consent was obtained at the time of surgical resection. Deidentified medulloblastoma tissues were obtained from Queensland Children's Tumor Bank (Brisbane, Australia), German Cancer Research Center (Heidelberg, Germany), Sanford-Burnham Medical Research Institute (La Jolla, CA), and Hospital for Sick Children (Toronto, Ontario, Canada).

Library construction, sequencing, and alignment of RNA-seq data are described in detail in the Supplementary Materials and Methods. Transcriptome aligned binary alignment files (bams) were processed using RNA-Seq by Expectation Maximization (RSEM) software (36). The reference for RSEM was built using the same GRCh37-lite assembly and GRCh37.75 GTF as used by STAR aligner (37). This was followed by the calculation of PKM isoform level FPKM using RSEM with default parameters in 97 human tumors. The accession number for the RNA-seq data is European Genome-phenome Archive: EGAD00001001899.

PKM2 FPKM expression was converted to a Z-score and used to segregate SHH patients into a high or low category with the criteria of being greater than 0.5 or smaller than –0.5 respectively. A Cox proportional hazard model was used to determine the presence of a survival difference.

We found dichotomous Pkm isoform expression in the postnatal brain that varied with differentiation state. Differentiated neurons throughout the brain expressed PKM1 and not PKM2 (Fig. 1A). In contrast, neural progenitors of the cerebellar external granule layer (EGL), hippocampus (HC), and subventricular zone/rostral migratory stream (SVZ/RMS) expressed PKM2 and not PKM1 (Fig. 1B).

The temporal expression of PKM2 in the cerebellum coincided with postnatal neurogenesis. CGNP proliferation peaks at postnatal day 7 (P7) and ends by P15, as CGNPs progressively exit the cell cycle, migrate to the internal granule layer (IGL), and differentiate into cerebellar granule neurons (CGN; refs. 3, 4, 7). Western blot analysis of whole cerebellum lysates showed that PKM2, like the CGNP proliferation marker cyclin-D2 (CCND2), decreased between P7 and P15, while PKM1 conversely increased (Fig. 1C). PKM2 was not limited to proliferating progenitors: both proliferating, PCNA⁺ CGNPs in the outer layer of the EGL (oEGL), and differentiating, PCNA^– CGNPs of the inner EGL (iEGL) expressed PKM2 (Fig. 1D). Similarly, in the SVZ/RMS, both PCNA⁺ and PCNA^– progenitors expressed PKM2 (Fig. 1E). Thus, PKM2 marked undifferentiated brain progenitors, whether proliferating or quiescent, while PKM1 marked differentiated neurons.

To determine whether medulloblastomas, like CGNPs, exclusively expressed PKM2, we analyzed PKM isoform expression in transgenic, medulloblastoma-prone ND2:SmoA1 mice. ND2:SmoA1 mice express a mutant, constitutively active allele of Smo, driven by the NeuroD2 promoter. In CGNPs of ND2:SmoA1 mice, SHH-pathway activation prolongs proliferation beyond P15. These mice develop medulloblastoma with incomplete penetrance after a variable latency (14). During this latent period, CGNPs continue to proliferate in premalignant lesions within the EGL, generating progeny that differentiate, undergo apoptosis, or remain proliferative.

We found that CGNPs in premalignant lesions of P60 ND2:SmoA1 mice expressed PKM2 and not PKM1 (Fig. 1F). Similarly, medulloblastomas derived from these premalignant lesions consistently and homogeneously expressed PKM2, while adjacent, normal brain expressed PKM1 (Fig. 1G). We compared the expression of PKM1 and PKM2 mRNA in human medulloblastoma samples by RNA-seq. As in the mouse model, we found exclusive expression of PKM2, with minimal PKM1 in all four medulloblastoma subgroups (Fig. 1H). Together, our studies of mouse and human tumors show that developmental PKM2 expression persists throughout medulloblastoma tumorigenesis.

To examine the effect of SHH signaling on PKM1/2 expression, we isolated CGNPs and compared PKM isoform expression in the presence or absence of SHH ligand. We labeled S-phase cells by adding EdU to CGNP cultures 1 hour before fixation. EdU⁺ and PKM2⁺ cells were more numerous in SHH-treated wells (Fig. 2A; cyan arrows) compared with SHH-deprived controls (Fig. 2B; arrows). In SHH-treated wells, PKM2 was detected in both EdU⁺ and EdU^– cells (Fig. 2A and B; white arrowheads). Consistent with the absence of PKM1 in the EGL, no CGNPs in either condition showed detectable PKM1; rare cells that were PKM1⁺ did not have morphology characteristic of CGNPs (Fig. 2C and D). These data show that SHH sustained PKM2 expression in CGNPs, whereas PKM1 expression remained low during the initiation of differentiation in SHH-deprived CGNPs.

To determine the effect of SHH inhibition on PKM expression, we analyzed SHH-treated CGNPs exposed to the Smo inhibitor vismodegib (38) by quantitative PKM Western blot. As an indicator of SHH-driven proliferation, we also measured CCND2 expression. SHH-treated CGNP explants demonstrated 5.5-fold more CCND2 compared with SHH-deprived CGNPs, and 2-fold more PKM2 (Fig. 2E). PKM1, which likely originated from rare cells detected by IHC (Fig. 2C and D), was also significantly increased SHH-treated explants. Vismodegib decreased CCND2 by 30% but did not significantly alter PKM1 or PKM2 expression (Fig. 2E). Thus, both PKM2 and CCND2 were upregulated by SHH, but compared with CcnD2, Pkm was markedly less responsive to changes in SHH signaling. These data suggest that Pkm, unlike CcnD2 (11), is not a direct SHH target.

Tyrosine kinase signaling, and specifically insulin signaling, can downregulate PKM2 catalytic activity in diverse cell lines (21, 39, 40). While CGNPs are typically cultured with insulin-rich N2 supplement to improve cell viability (41), we have shown that CGNPs can be cultured without N2 for 24 hours, reducing IGFR activation without compromising survival (5). We maintained CGNPs in the presence or absence of SHH or N2, to determine the effect of SHH/IGFR costimulation on Pkm expression and PK activity in CGNPs.

We found that SHH/IGFR costimulation was required for CGNP aerobic glycolysis, as N2 deprivation reduced lactate production (Fig. 2F) in a manner comparable with SHH-withdrawal (5). We have shown that growth factor stimulation induces aerobic glycolysis through a complex program that depends on many genes, including N-myc and Hk2 (5). To determine whether this program may include changes in PK activity and PKM expression, we analyzed CGNPs cultured with or without SHH or N2. N2-deprivation potently inhibited IGF signaling, resulting in a 70% reduction in AKT phosphorylation, and more modestly reduced PKM2 expression by 30% (Supplementary Fig. S1). By measuring the conversion of exogenous PEP to pyruvate in CGNP lysates, we found that CGNPs maintained with SHH+N2 showed significantly higher PK activity compared with either SHH without N2 or N2 without SHH (Fig. 2G and H). These findings show that CGNP PK activity and PKM2 expression were responsive to extracellular signaling, and were maximal with SHH/IGFR coactivation, thus varying directly with glycolysis. While reduced PK activity has been correlated with growth factor stimulation in cancer cells (20, 25, 26), in CGNPs, developmentally relevant growth factors increased PKM2 expression, PK activity, and aerobic glycolysis.

To determine the developmental significance of PKM in CGNPs, we examined the effect of conditionally deleting the Pkm2-specific exon 10 (19). We crossed Math1-Cre mice, which express Cre recombinase in CGNPs (42, 43), with Pkm2^fl mice that harbor loxP sites flanking Pkm exon 10 (32). The resulting Math1-Cre;Pkm2^fl/fl mice (Pkm2^cKO) were viable and fertile without neurologic deficits. Pkm2^cKO cerebellum showed normal foliation and cellular organization, without PKM2 or PKM1 in the EGL (Fig. 3A and B). PKM1 expression was increased in IGL of Pkm2^cKO mice, which is populated by CGNs that derive from CGNPs and inherit the deletion of exon 10 (Fig. 3A). Thus, Pkm2 deletion in the Math1 lineage did not induce compensatory PKM1 in progenitors but did increase PKM1 expression by terminally differentiated neurons. Consistent with these changes, Western blot analysis showed reduced PKM2 and increased PKM1 in P7 Pkm2^cKO cerebella compared with no-Cre littermate controls (Fig. 3C). In whole cerebellum lysates, PK activity was equivalent in Pkm2^cKO and littermate controls (Fig. 3D). In CGNPs, however, Pkm2 deletion blocked all detectable PKM expression.

During the breeding of Pkm2^cKO mice, we noted an unanticipated tendency for germline recombination of the Pkm2^fl allele. This recombination event was detected as a smaller than expected PCR product when primers flanking Pkm exon 10 were used to amplify genomic DNA. While these primers were separated by 605 bp in the Pkm2^fl/fl allele and 560 bp in the Pkm2^+/+ allele, we found pups in several litters that generated 220-bp PCR products consistent with the recombined allele (Supplementary Fig. S2A). Sequencing of this band demonstrated that intron 9–10 and 10–11 were brought together, separated by a single loxP site, consistent with Cre-mediated excision of exon 10 from Pkm (Supplementary Fig. S2B; ref. 32). Similar germline recombination has been observed when Pkm2^fl mice were crossed with other nongermline Cre drivers (M. Vander Heiden, unpublished data), suggesting that Pkm exon 10 deletion is positively selected in the germline. The recombined allele (Pkm2^null) was heritable and Pkm2^null/null mice were viable and fertile without neurologic abnormalities, consistent with the published Pkm2-null phenotype (31). Pkm2-null mice have been reported to develop spontaneous hepatocellular carcinoma in the second year of life (31); however, we did not observe any spontaneous tumor formation in our Pkm2^null/null aged up to one year. In Pkm2^null/null mice, PKM2 was absent from all brain cells, while PKM1 was appropriately limited to differentiated neurons (Supplementary Fig. S2C and S2D). Like Pkm2^cKO mice, Pkm2^null/null mice showed normal brain anatomy.

Although Pkm2^cKO cerebella had normal organization, quantitative analysis showed that CGNP proliferation was increased compared with PKM2-intact littermate controls. We analyzed proliferation dynamics in vivo, by injecting EdU at P6 into Pkm2^cKO and Pkm2^fl/fl no-Cre littermates. We harvested cerebella 24 hours after EdU injection and quantified the number of EdU⁺ cells in the EGL that expressed the CGNP differentiation marker p27^Kip1. Both genotypes included differentiating, EdU⁺p27^Kip1+ CGNPs and proliferating, EdU⁺p27^Kip1– CGNPs (Fig. 4A and B). While EdU labeling was similar between genotypes (Fig. 4C), we noted a trend toward reduced p27^Kip1 labeling in Pkm2^cKOCGNPs (Fig. 4D). Importantly, EdU⁺ CGNPs that were p27^Kip1– were more numerous in the EGL of Pkm2^cKO mice (Fig. 4E). In vitro studies provided additional evidence of increased proliferation with Pkm2 deletion, as CGNPs explanted from Pkm2^cKO mice showed a higher fraction of EdU⁺ cells after a 1-hour pulse, compared with CGNPs explanted from littermate controls (Fig. 4F).

Consistent with increased proliferation of Pkm2^cKO CGNPs, analysis of equivalent midline cerebellar sections showed that Pkm2^cKO cerebella contained significantly more cells (CGNPs and CGNs) compared with littermate controls (Fig. 4G). The ratio of EGL:IGL was not significantly different between genotypes, however (Fig. 4H), indicating that Pkm2 deletion did not prevent CGNPs from progressing through their normal differentiation trajectory. Thus, Pkm2 deletion blocks all PKM expression in CGNPs, with the effects of increasing proliferation and reducing cell-cycle exit, without altering cell fate.

We used a nonbiased, metabolomic approach to determine the effect of Pkm2 deletion and consequent loss of PKM function on CGNP metabolism. We isolated CGNPs from P5 Pkm2^cKO mice and littermate controls and cultured them in three replicate wells for 24 hours in media with [1,6-¹³C] glucose replacing unlabeled glucose. We analyzed media metabolites from each replicate well for each condition at 0 and 24 hours using NMR spectroscopy followed by orthogonal partial least squares discriminant analysis to identify species with metabolic rates that varied consistently across genotypes (Supplementary Table S1).

Lactate production was significantly reduced in Pkm2^cKO CGNPs compared with controls (Fig. 5A). Lactate was the species with the largest change in metabolic rate in Pkm2^cKO CGNPs versus controls (Supplementary Table S1). Lactate production and arginine utilization were both significantly reduced in the Pkm2-deleted cells, but the change in arginine was 10-fold smaller and unlikely to be directly related to glucose metabolism. We confirmed that Pkm2^cKO CGNPs produced less lactate by measuring media lactate in replicate wells by colorimetric assay (Fig. 5B). The ¹³C fractional enrichment of lactate was also decreased in Pkm2^cKO CGNPs (Fig. 5C and D), showing that a smaller proportion of utilized glucose was converted to lactate.

To determine how Pkm2 deletion altered the intracellular disposition of glucose in CGNPs, we analyzed cell extracts by LC-MS. Diverse molecules detected by LC-MS detected did not show significantly different distribution between genotypes. We noted a trend toward increased choline in Pkm2^cKO CGNPs; this trend cannot be directly related to glucose metabolism, but is consistent with increased proliferation (Fig. 5E). We were able to detect ¹³C labeling in glutamate, which we used as a measure of flow through the Krebs cycle, and in the ribose component of ATP, ADP, and AMP, which we used as a measure of flow through the pentose phosphate pathway (PPP; Fig. 6).

Relative to Pkm2-intact controls, Pkm2^cKO CGNPs showed a small, but statistically significant decrease in glutamate with 3 or 4 ¹³C atoms (Fig. 6). The incorporation of multiple ¹³C atoms from [1,6-¹³C] glucose requires both conversion of pyruvate to acetyl-CoA by PDH and multiple turns of the Krebs cycle (Supplementary Fig. S3). The reduced glutamate with 3 or 4 ¹³C atoms in controls indicates that that PKM2 deletion marginally reduced labeling downstream of PDH, consistent with a shift in fate of glucose-derived carbons.

Both genotypes demonstrated ¹³C incorporation into ribose. LC-MS demonstrated labeled parent masses of ATP, ADP, and AMP that fragmented into unlabeled adenine and ¹³C-labeled ribose. These data demonstrate active flow through the PPP. However, we did not detect a robust difference in PPP flux between genotypes (Fig. 6). These data suggest that unidentified reactions in Pkm2^cKO CGNPs bypass the conversion of PEP to pyruvate, allowing the incorporation of ¹³C into glutamate and lactate at a reduced but detectable rate, while maintaining comparable flow through the PPP. In net, Pkm2 deletion redirected glucose-derived metabolites away from both lactate generation and retention in the Krebs cycle, without causing a large increase in PPP flux.

We examined the functional consequence of Pkm2 deletion on SHH-driven medulloblastoma tumorigenesis by crossing Pkm2^cKO and Pkm2^null/null mice with medulloblastoma-prone mouse lines. We generated tumor-prone mice using either the SmoM2 allele that induces rapidly developing progressive medulloblastoma with 100% incidence by P20, or with the ND2:SmoA1 allele that induces tumors more slowly. In contrast to the antitumor effect of Hk2 deletion, we found that Pkm2 deletion did not slow tumor progression in Math1-Cre;SmoM2;Pkm2^fl/null mice (Fig. 7A) or hGFAP-cre;SmoM2;Pkm2^fl/null mice (data not shown) compared with Pkm2-intact, littermate controls. In both of these models, however, rapid tumor growth limits the ability to determine whether PKM2 ablation accelerates disease. To address this question, we examined Pkm2 deletion in ND2:SmoA1 mice.

ND2:SmoA1 mice with Pkm2 deletion developed tumors more frequently than Pkm2-intact ND2:SmoA1 controls. We compared ND2:SmoA1 mice with two mutant copies of Pkm2, including Math1-Cre;ND2:SmoA1;Pkm2^fl/fl, Math1-Cre;ND2:SmoA1;Pkm2^fl/null, and ND2:SmoA1;Pkm2^null/null genotypes to control ND2:SmoA1 mice with two intact copies of Pkm2, including no-Cre;ND2:SmoA1;Pkm2^+/+, no-Cre;ND2:SmoA1;Pkm2^fl/+, and Math1-Cre;ND2:SmoA1;Pkm2^+/+ genotypes. We observed a trend toward increased tumor formation in Pkm2–deleted mice at P150, where the incidence was 37.5% with Pkm2 deletion versus 15.8% for controls (P = 0.245, Fisher exact test). By 300, tumor incidence was 100% in Pkm2-deleted mice versus 57.9% for controls (P = 0.004, Fisher exact test). Consistently, ND2:SmoA1 mice with Pkm2 deletion demonstrated significantly shorter progression-free survival (Fig. 7B).

To determine whether Pkm2 deletion accelerated tumor growth through a cell-autonomous process, we separately compared the Math1-Cre;ND2:SmoA1;Pkm2^fl/null and ND2:SmoA1;Pkm2^null/null subgroups to Pkm2-intact ND2:SmoA1 controls. This analysis also showed significantly shorter survival for both Math1-Cre;ND2:SmoA1;Pkm2^fl/null (Fig. 7C) and ND2:SmoA1;Pkm2^null/null genotypes (Fig. 7D). We found no statistically significant difference between different Pkm2-deleted genotypes. Thus, the accelerating effect of Pkm2 deletion on SmoA1-driven tumorigenesis was cell autonomous.

Like Pkm2-deleted CGNPs, Pkm2-deleted medulloblastoma cells expressed neither PKM2 (Fig. 7E) or PKM1 (Fig. 7F). However, we noted scattered PKM1⁺ cells within Pkm2-deleted tumors (Fig. 7E″′). These cells were consistently PCNA^– (Supplementary Fig. S4) and were also found in Pkm2-intact tumors. These nonproliferative, PKM1⁺ cells may be entrapped neurons or differentiated progeny of tumor cells. Similar PKM1⁺ stromal cells were previously observed in Pkm2-deleted breast tumors (32).

The slow process of tumor formation in ND2:SmoA1 mice allowed us to analyze the effect of Pkm2 deletion on the growth of premalignant lesions. We counted PCNA⁺ cells in cerebella of ND2:SmoA1 mice with deleted or intact Pkm2, harvested at either P60 or P120 and normalized the number of PCNA⁺ cells to the total number of cells in each cerebellar section (Fig. 7G). In controls, the median fraction of PCNA⁺ cells in the cerebellum and the range of these values decreased over time, consistent with a dynamic balance between growth and growth suppression. In contrast, in Pkm2-deleted genotypes, the median and range of PCNA⁺ fractions increased over time, consistent with reduced growth suppression during the premalignant period. Altogether, Pkm2 deletion increased SHH-driven CGNP proliferation during development, accelerated tumorigenesis in premalignant lesions in Smo-mutant mice, and increased the incidence of tumor formation from these lesions.

The more rapid progression of mouse tumors with Pkm2 deletion was mirrored by a trend toward shorter survival in medulloblastoma patients with low PKM2 expression. We classified SHH-subgroup medulloblastoma patients as having either high or low PKM2–expressing tumors and compared their clinical outcomes. All patients were treated similarly with radiation and chemotherapy. While not statistically significant, we found that patients with low PKM2 expression trended toward shorter survival times (Fig. 7H). Together, these data confirm that PKM2 is not essential for tumor progression and suggest that low PKM2 expression, like Pkm2 deletion, enhances medulloblastoma tumor growth.

We analyzed the functional significance of pyruvate kinase isoform expression in the developing brain and in medulloblastoma. While aerobic glycolysis has been associated with proliferation in both development (5, 44) and cancer (1, 2), our data show that disrupting glycolysis at PK increases, rather than decreases both developmental proliferation and tumorigenesis. We show that within the neural lineage, Pkm is spliced into mutually exclusive expression patterns that correlate PK activity with differentiation state. In the postnatal brain, undifferentiated progenitors in the cerebellum, hippocampus, and SVZ expressed the less active PKM2 isoform, while differentiated neurons expressed the more active PKM1. The correlation of PKM2 with the undifferentiated state was maintained in medulloblastoma. The PKM1:PKM2 dichotomy persisted after conditional or germline deletion of Pkm2, which increased PKM1 in differentiated cells but not in progenitors. The lack of compensatory upregulation of PKM1 in brain progenitors contrasts with prior observations in MEFs and other tissues, where PKM1 expression increases on Pkm exon 10 deletion (31, 45). In Pkm2-deleted MEFs, increased PKM1 caused nucleotide scarcity and cell-cycle arrest. However, Pkm2^cKO CGNPs expressed neither PKM isoform, showed decreased conversion of glucose to lactate and increased proliferation. Pkm2 deletion also enhanced the growth of premalignant lesions and tumors in ND2:SmoA1 mice. These findings demonstrate that the absence of detectable PKM2 in Pkm2^cKOmice increased developmental proliferation and neural progenitor susceptibility to tumorigenesis.

The different effects of Pkm2 deletion in CGNPs and MEFs argue that proliferation is restricted by PKM1, rather than enhanced by PKM2. We propose that high PK activity inhibits SHH-driven proliferation and suppresses tumor formation. Consistent with this model, splicing Pkm to PKM2 has a permissive effect on proliferation and abrogating PKM expression is more permissive. The finding that SHH/IGFR cosignaling increased CGNP PK activity is not necessarily inconsistent with this model as growth factors are known to induce simultaneously both positive and negative regulation of proliferation in primary cells. For example, SHH activation induces Patched (Ptc) expression in CGNPs, which negatively regulates SHH signaling (46, 47). Activating PKM2 may be a similar inhibitory element within the normal regulation of SHH-induced growth in primary cells.

Our prior Hk2 deletion studies demonstrate that the initiation of glycolysis supports postnatal neurogenesis and sustains the malignant growth of medulloblastomas. In contrast, our current studies of Pkm2 deletion show that impeding the flow of glucose to lactate increases CGNP proliferation and accelerates medulloblastoma tumorigenesis. Together, these deletion studies show that the pathways through which glucose is metabolized significantly influence the proliferative behavior of undifferentiated cells. The divergent effects of Hk2 deletion versus Pkm2 deletion establish the glycolytic steps upstream of PK as critical to supporting the proliferative phenotype of neural progenitors and cognate tumors.

Between the reactions catalyzed by HK2 and PKM2 are several reversible reactions that permit glycolytic intermediates to be shunted toward biosynthetic processes. We and others have previously proposed that low flow downstream of PEP, achieved by preferentially splicing PKM to generate PKM2, may support growth by diverting glycolytic intermediates away from lactate generation and toward the synthesis of macromolecules (2, 48, 49). However, evidence for decreased lactate and increased anabolism with Pkm2 deletion was lacking in Pkm2-deleted breast tumors (32) and MEFs (45). Recent work found that the direct incorporation of glucose carbons is a minor component of the biomass of proliferating cancer cells (50). Working with primary neural progenitors limited the amount of material for analysis and thus the range of metabolites we could detect. However, we were able to detect ¹³C in extracellular lactate and in intracellular glutamate and ribose. Pkm2 deletion decreased the incorporation of glucose into lactate and glutamate, suggesting that that altering PEP metabolism redirects glucose carbons to different metabolic fates. Pkm2 deletion did not produce a detectable effect on the incorporation of glucose into ribose, suggesting that PPP flux is not the major destination of glucose carbons redirected by low PK activity.

Our ¹³C studies show that CGNPs lacking both PKM1 and PKM2 can metabolize glucose to pyruvate, raising the question of what alternative pathway metabolizes PEP. We were not able to detect changes in glycolytic intermediates that might indicate how PEP is processed in Pkm2^cKO CGNPs. However, prior studies have shown that PKM2-expressing cells with low PK activity can generate pyruvate by transferring the high-energy phosphate of PEP to the enzyme PGAM1 (51). This phosphorylation activates PGAM1, which may promote biosynthetic metabolism (52). If PEP-PGAM1 phosphotransfer occurs in CGNPs, it may mediate the growth-promoting effect of PKM2 splicing that is enhanced by Pkm2 deletion.

Our work suggests that Pkm functions as a tumor suppressor. Identifying Pkm as a tumor suppressor has profound implications for aerobic glycolysis and may be effectively targeted in cancer treatment. Inhibiting PKM function as an anticancer therapy would be counterproductive while drugs that increase PKM catalytic activity may be of limited therapeutic value if tumors can thrive with low PKM expression. The loss of tumor suppressor genes, however, may be targeted through alternative approaches that require defining the consequent changes in pathway regulation (53, 54). A detailed understanding of how low PK activity promotes proliferation may produce novel clinical strategies to treat tumor growth.

M.G. Vander Heiden has ownership interest (including patents) in Agios Pharmaceuticals and is a consultant/advisory board member for Agios Pharmaceuticals and Aeglea Biotherapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K. Tech, A.P. Tikunov, M.G. Vander Heiden, T.R. Gershon

Development of methodology: K. Tech, A.P. Tikunov, J. MacDonald, T.R. Gershon

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Tech, A.P. Tikunov, J. Meidinger, T. Fish, A.J. Mungall, R.A. Moore, S.J.M. Jones, M.A. Marra, M.D. Taylor, J. MacDonald

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Tech, A.P. Tikunov, H. Farooq, A.S. Morrissy, T. Fish, S.C. Green, Y. Li, Y. Ma, S.J.M. Jones, M.D. Taylor, J. MacDonald, T.R. Gershon

Writing, review, and/or revision of the manuscript: K. Tech, A.P. Tikunov, H. Farooq, T.R. Gershon

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Tech, J. Meidinger, H. Liu, Y. Li

Study supervision: K. Tech, T.R. Gershon

Other (provided unpublished (at the time) animals and had numerous discussions regarding data interpretation): M.G. Vander Heiden

We thank the UNC CGBID Histology Core, the UNC Tissue Pathology Laboratory Core and UNC UCRF, and the UNC Neuroscience Center Confocal and Multiphoton Imaging facilities for providing expertise in immunohistochemistry and confocal imaging.

UNC CGBID Histology Core was supported by P30 DK 034987 and the UNC Tissue Pathology Laboratory Core was supported by NCI CA016086. K. Tech, J.M. Macdonald, and T.R. Gershon were supported by the National Institute of Neurologic Disorder and Stroke (R01NS088219). K. Tech and T.R. Gershon were supported by the American Institute for Cancer Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.