Polyglucosan neurotoxicity caused by glycogen branching enzyme deficiency can be reversed by inhibition of glycogen synthase


Address correspondence and reprint requests to Or Kakhlon, Department of Neurology, Hadassah-Hebrew University Medical Center, Ein Kerem, Jerusalem 91120 Israel. E-mail: ork@hadassah.org.il


Uncontrolled elongation of glycogen chains, not adequately balanced by their branching, leads to the formation of an insoluble, presumably neurotoxic, form of glycogen called polyglucosan. To test the suspected pathogenicity of polyglucosans in neurological glycogenoses, we have modeled the typical glycogenosis Adult Polyglucosan Body Disease (APBD) by suppressing glycogen branching enzyme 1 (GBE1, EC expression using lentiviruses harboring short hairpin RNA (shRNA). GBE1 suppression in embryonic cortical neurons led to polyglucosan accumulation and associated apoptosis, which were reversible by rapamycin or starvation treatments. Further analysis revealed that rapamycin and starvation led to phosphorylation and inactivation of glycogen synthase (GS, EC, dephosphorylated and activated in the GBE1-suppressed neurons. These protective effects of rapamycin and starvation were reversed by overexpression of phosphorylation site mutant GS only if its glycogen binding site was intact. While rapamycin and starvation induce autophagy, autophagic maturation was not required for their corrective effects, which prevailed even if autophagic flux was inhibited by vinblastine. Furthermore, polyglucosans were not observed in any compartment along the autophagic pathway. Our data suggest that glycogen branching enzyme repression in glycogenoses can cause pathogenic polyglucosan buildup, which might be corrected by GS inhibition.


Knockdown of glycogen branching enzyme in neurons led to accumulation of an insoluble form of glycogen called polyglucosan, to apoptosis and to activation of glycogen synthase. These effects were reversed by glycogen synthase inhibition through starvation and rapamycin treatments, suggesting a potential therapeutic value of glycogen synthase inhibition for treating glycogen storage disorders.

Abbreviations used



5-aminoimidazole-4-carboxamide ribonucleotide


AMP kinase


adult polyglucosan body disease




glycogen branching enzyme 1


glycogen storage disease type IV


glycogen synthase


glycogen synthase kinase


high-throughput screening


lafora disease


multivesicular bodies




polyglucosan bodies


prtotein targeting to glycogen


sodium dodecyl sulfate polyacrylamide gel electrophoresis


short helical RNA

The brain has extremely high energy demand, but relatively restricted energy storage capacity. Moreover, although neurons have the highest energy requirement in the brain, they do not store or turn over glycogen because of glycogen synthase (GS) inactivation and glycogen phosphorylase deficiency (Benarroch 2010). Rather than storing their own fuel, neurons rely on astrocytes, the brain's energy sensors. Astrocytes can respond to insulin by increasing glycogen stores (Heni et al. 2011) and to neuronal neurotransmitters, secreted during energy deficit, by furnishing neurons with energy substrates. These substrates are produced by glycogen mobilization, or by glycolysis, both down-modulated in neurons (Belanger et al. 2011). It is still debated which energy substrate is delivered to neurons by astrocytes. Recent computational studies maintain that both lactate, the long-accepted form of energy transported to neurons (Belanger et al. 2011; Choi et al. 2012), and glucose (DiNuzzo et al. 2010; Chander and Chakravarthy 2012), which also dictates neuronal lactate uptake (Calvetti and Somersalo 2012), fuel neurons.

As a consequence of their inactivated glycogen metabolism, the only form of glycogen that neurons synthesize is the abnormal polyglucosan, associated with various neurological disorders. Polyglucosan is amylopectin-like, poorly branched, and insoluble. It is the major constituent of insoluble deposits called polyglucosan bodies (PBs), also containing damaging ubiquitinated proteins and long-lived advanced glycated end-products (Cavanagh 1999). Polyglucosan forms whenever glycogen branching activity cannot keep pace with glycogen synthetase activity. Thus, in APBD (Lossos et al. 1991), a variant of glycogen branching enzyme 1 (GBE1) deficiency (glycogen storage disease type IV (Chen and Bucherell 1995), PBs accumulate because of GBE1 deficiency (Magoulas et al. 2012). On the other hand, in Lafora disease (LD) (Andrade et al. 2007), excessive glycogen phosphorylation (Striano et al. 2008; Turnbull et al. 2010), or insufficient GS degradation (Vilchez et al. 2007), leads to excessive glycogen elongation and accumulation of PBs called Lafora bodies. PB accumulation causes neuronal-selective apoptosis which spares astrocytes (Cavanagh 1999; Magistretti and Allaman 2007; Vilchez et al. 2007). Despite the neurotoxic potential of PBs, neurons cannot dispose of them, even though they are endowed with robust rapamycin-inducible autophagy and microtubular transport, which can clear insoluble inclusion bodies.

Our study is aimed at elucidating why polyglucosan accumulation is pathogenic in neurological glycogenoses (represented by APBD) and which measures might circumvent polyglucosan neurotoxicity. Using transduction of neurons with lentiviruses harboring shRNA directed against GBE1, we generated a GBE1-knocked down neuronal model of APBD, which featured increased GS activity, polyglucosan accumulation and associated apoptosis. We then attempted to clear the PBs and reverse apoptosis by rapamycin and starvation, able to phosphorylate and inactivate GS. We further showed that these protective effects depend on GS phosphorylation/inactivation and not on autophagy, which is also stimulated by rapamycin and starvation and which can degrade glycogen (Kotoulas et al. 2004). Our results suggest that future therapeutic endeavors in neurological glycogenoses should be based on down-modulation of the ratio between glycogen synthesis and branching via manipulation of GS or GBE1 activities, or on direct polyglucosan degradation.



All reagents were from Sigma-Aldrich (Rehovot, Israel), except for radionuclides and vinblastine, which were from Perkin-Elmer (Waltham, MA, USA) and Teva (Netanya, Israel), respectively. 5-Aminoimidazole-4-carboxamide ribonucleotide (AICAR) was a kind gift from Dr. Ann Saada-Reisch (Hadassah-Hebrew University Medical Center). Periodic acid-Schiff reagent kit was from Merck (Darmstadt, Germany).


Monoclonal antibodies to GBE1 were from Abnova (Taipei, Taiwan), those to tubulins and GS from Abcam (Cambridge, UK), those to LC3 from Nanotools (Teningen, Germany), those to phospho-GS (Ser641) from Cell Signaling Technology (Danvers, MA, USA) and those to c-myc (9E10) from the hybridoma bank. The monoclonal antibody to glycogen was prepared as previously described (Nakamura-Tsuruta et al. 2012) and used for immunofluorescence (Wang et al. 2013). Secondary antibodies, conjugated either to Dylight 549 (for immunofluorescence and flow cytometry) or to peroxidase (for immunoblotting), were from KPL (Gaithersburg, MD, USA). Antibodies to LC3 and c-myc used for immunofluorescence and flow cytometry were conjugated to the FluoProbes 647H (ex. 653 nm, em. 674 nm) and Phycoerythrin (PE)-Cy5 (ex. 488 nm, em. 679 nm), respectively, using the Lightning-Link conjugation kit from Innova Biosciences (Cambridge, UK). All antibodies were used according to manufacturers’ instructions.

Neuronal model

Primary neurons from E18 rat cortex were obtained from Brainbits (Springfield, IL, USA) and cultured according to the manufacturer's instructions. The content of these cultures was practically neuronal (96%), as confirmed by flow cytometry using antibodies against the respective neuron and glia markers β-3 tubulin and glial fibrillary acidic protein (Figure S1). Neuronal transfection was done by electroporation using Microporator MP-100 (NanoEn Tek, Seoul, Korea) and took place after lentiviral transduction.

Patient cells

Lymphoblasts and fibroblasts from a 65 years old female APBD patient were used with the understanding and written consent of the patient and following the ARRIVE guidelines. Derivation of these cells was approved by the Hadassah Institutional Review Board according to The Code of Ethics of the World Medical Association (Declaration of Helsinki), printed in the British Medical Journal (18 July 1964).

Suppression of GBE1 expression

After 5 days in culture, neurons were transduced with SMARTvector 2.0 lentiviral particles (Thermo Scientific, Lafayette, CO, USA) bearing shRNA against GBE1, or a non-targeting control. Transduction was done at a titer of 40 multiplicities of infection in half the volume of growth medium. 12 h post-transduction, neurons were washed twice with phosphate-buffered saline and the medium was replaced with half preconditioned/half fresh medium. Neurons were assayed 4 days after lentiviral transduction. Lentiviral constructs included a turboGFP reporter gene to enable detection of transduced cells.

Real-time Reverse Transcription PCR

Total RNA was purified from cells using the PureLink kit (Invitrogen, Carlsbad, CA, USA) and reverse-transcribed using the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems, Foster City, CA, USA). GBE1 mRNA content, normalized to the endogenous control Hprt, was determined by real-time RT-PCR using the pre-designed TaqMan Gene Expression system and the StepOnePlus thermocycler and analyzer (Applied Biosystems). Site directed mutagenesis was performed on mouse GS1 cDNA using the Quick Change kit (Stratagene, Santa Clara, CA, USA).

ATP assay

ATP was determined by a luciferase-based assay (Sigma-Aldrich). Luminescence was read with a 10-second integration time in the luminescence mode of DTX 880 Multimode Detector (Beckman Coulter, Indianapolis, IN, USA). Phosphofructokinase activity was measured using BioVision's (Milpitas, CA, USA) specialized kit.

GBE1 and GS activity assays

GBE1 activity was assayed based on (Lossos et al. 1991). This assay is based on the difference in the rate of incorporation of 14C-glucose-1-phosphate to endogenous glycogen between samples where glycogen branching enzyme (GBE) is denatured by boiling and untreated samples. The lower GBE activity, the lower this difference. GS activity was determined based on the rate of 14C-UDP-glucose incorporation to exogenous glycogen (Akman et al. 2011).


For indirect immunofluorescence, neurons were fixed with 3% paraformaldehyde, permeabilized with 0.2% Triton X-100, stained with antibodies, and analyzed by confocal microscopy using a Zeiss LSM 710 microscope (Jena, Germany) with a 60X/1.35 NA PlanApochromat oil-immersion lens. In Fig. 5a, neurons were permeabilized by saponin before fixation, using 5 min exposure to 0.5% saponin in 80 mM Piperanzine-N-N′-bis[2-ethane]sulfonic acid pH 6.8, 1 mM CaCl2, 5 mM EGTA. LC3-positive structures were quantified by the Image-Pro software (Media Cybernetics, Bethesda, MD). Electron microscopy was performed as described in (Zahavi et al., 2011), with uranyl-ethanol staining of the thin sections.

Flow cytometry and quantification of apoptosis

Apoptosis was quantified using the FC 500 flow cytometer (Beckman Coulter). In Figs 3-5 and 7 and Figure S2, neurons were co-stained with Invitrogen's Annexin-V-PE, and 7-Amino-actinomycin D (7-AAD), for detection of apoptotic and necrotic cells, respectively. Cells (10 000) were measured using the 488 nm excitation line and PE (FL-2) and 7-AAD (FL-4) emission detectors. FL-1/FL-2 compensation [all cells analyzed were green fluorescence protein (GFP) positives] and FL-2/FL-4 compensation were set during acquisition using single dyes controls and confirmed post-acquisition by the FC 500 QuickComp feature. In Fig. 8, neurons transduced and transfected with c-myc-tagged GS constructs were permeabilized and co-stained with Annexin-V-PE, PE-Cy5-conjugated anti-myc antibody (FL3) and the TO-PRO-3 DNA dye (FL4, Invitrogen). Gated FL1 (GFP) and FL3 (c-myc tagged GS) positive cells were measured using the 488 nm laser for the GFP (FL1), PE (FL2), and PE-Cy5 (FL3) emission detectors and the 635 nm laser for the TO-PRO-3 (FL4) emission detector. Compensations between all combinations of FL-1 to FL-4 emissions were set prior to gating and confirmed as above. For microscopic detection of apoptosis, cells were stained with Annexin V-Pacific Blue and 7-AAD.


The significance of differences (< 0.05) between pairs of treatments and among multiple treatments was confirmed by Student's t-test and one-way anova with Tukey post hoc test, respectively.


Generation and characterization of a neuronal model for APBD

To reproduce polyglucosan deposition, the hallmark of APBD, and ensuing neurotoxicity in cultured neurons, we generated a neuronal GBE1 knockdown model. This model is based on the observation that GBE1 levels (Assereto et al. 2007) and activities (Lossos et al. 1991) are reduced in APBD, which is, in fact, a clinical variant of glycogen storage disease type IV. GBE1 suppression in our model was confirmed at the mRNA and enzyme activity levels (Fig. 1a), and at the protein level by immunoblotting (Fig. 1b) and indirect immunofluorescence (Fig. 1c). This reduction is similar in degree to that observed in lymphoblasts from an APBD patient homozygous for the GBE1 Y329S mutation (Fig. 1b).

Figure 1.

Repression of glycogen branching enzyme 1 (GBE1) expression in primary neurons. (a) E18 rat embryo cortical neurons were transduced with lentiviral particles (Dharmacon) containing short helical RNA (shRNA) targeted against GBE1 (shGBE1), or a non-targeting shRNA (nt). The lentiviral construct also encoded for a turboGFP reporter gene to enable detection of transduced cells. GBE1 activity (white) and mRNA levels (gray) were determined by a specific assay (Lossos et al. 1991), based on incorporation of 14C-glucose-1-phosphate into glycogen, and by real-time RT-PCR, respectively. Both GBE1 activity and mRNA levels were significantly reduced in GBE1-knocked down neurons (shGBE1), compared to control (nt) neurons. (b) Left panel, Cell lysates (20 μg protein) of lymphoblasts derived from healthy control subjects, and subjects homozygous to the GBE1 Y329S mutation were subjected to sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and immunoblotting using anti-GBE1 and anti-α-tubulin antibodies. The position of the molecular weight markers is shown to the right. Right panel, Cell lysates of shGBE1 and nt-transduced neurons were subjected to SDS-PAGE and immunoblotting as in the left panel. (c) Confocal images of primary neurons transduced as in (a) with nt, or shGBE1 lentiviral particles as indicated, all encoding for turboGFP. Neurons were paraformaldehyde fixed, permeabilized with Triton X-100, and stained with a monoclonal antibody against GBE1. Green channel, GFP; red channel, GBE1. A merge of the two channels is also shown. Scale bars = 20 μm (upper panel), 5 μm (lower panel). As in all confocal figures, confocal laser power and gain and offset settings were identical between the treatments.

We next tested the ability of our model to reproduce polyglucosan accumulation and neurotoxicity. As our indirect immunofluorescence data show, no glycogen is detected in control neurons transduced with lentiviral particles containing non-targeting shRNA (Fig. 2a). In contrast, GBE1-knocked down neurons stained for glycogen show punctate cytosolic structures, similar to those observed in neurons in which GS activity had been induced (Vilchez et al. 2007, Fig. 2b). These structures are purportedly polyglucosans, being larger than soluble glycogen and showing typical fibrillar structure in the EM (Fig. 6d). Unfortunately, we were not able to extract enough glycogen to further confirm the polyglucosan nature of these deposits by iodine adsorption. Interestingly, neurons transduced with relatively lower level of shGBE1 lentiviruses (as documented by the lower GFP expression) also had lower levels of polyglucosan and lesser cell rounding, typical of the early stages of apoptosis (Fig. 2c). This observation suggests that polyglucosan accumulation is responsible for neuronal damage in our model.

Figure 2.

Confocal images of E18 rat embryo cortical neurons transduced with the lentiviral particles bearing non-targeting control (a), or glycogen branching enzyme 1 (GBE1) shRNA (b) constructs as in Fig. 1. Neurons were paraformaldehyde fixed, permeabilized with Triton X-100, and stained with a monoclonal antibody against glycogen. Green channel, GFP; red channel, glycogen. A merge of the two channels is also shown. (c) shGBE1-transduced neurons treated as in (b) but also showing a neuron transduced with relatively less viral particles (arrow), compared to other neurons (arrow head). The DIC channel is also shown. Scale bars = 10 μm (a), 2 μm (b) and 5 μm (c).

GBE1-knocked down neurons also manifested apoptotic blebbing (Fig. 3a) and stained positive for Annexin V, a marker of apoptosis, and the DNA stain 7-AAD, a marker of necrosis, similar to neurons treated with the apoptosis-inducer sodium nitroprusside (Fig. 3b). Confocal staining of fixed neurons with Annexin V and 7-AAD (Fig. 3c) confirmed that increased apoptotic indices (scored in Fig. 3d) were not simply caused by increased susceptibility to flow cytometry sample preparation. These results suggest that polyglucosan accumulation in APBD may be pathogenic because it triggers apoptosis.

Figure 3.

(a) DIC images of rat embryo cortical neurons transduced with lentiviral particles bearing non-targeting control (nt, lower panel), or glycogen branching enzyme 1 (GBE1) shRNA (shGBE1, upper panel) constructs as in Fig. 1. Scale bars = 2 μm (upper panel), 10 μm (lower panel). (b) Neurons transduced with lentiviral particles bearing non-targeting control (nt) or GBE1 shRNA (shGBE1), or nt neurons treated with 1 mM of sodium nitroprusside for 24 h as a positive control for induction of apoptosis (SNP) were stained with Annexin V-Phycoerythrin (PE) and 7-Amino-actinomycin D (7-AAD) and analyzed by flow cytometry, as described in 'Methods'. Percentages of cells in the different quadrants are indicated. (c) Neurons transduced with GBE1 shRNA were paraformaldehyde fixed, co-labeled with the apoptotic dye Annexin V-Pacific Blue [An-polyglucosan bodies (PB)] and the necrotic DNAcounter-stain 7-AAD and analyzed by confocal microscopy (examples in two different magnifications are shown). Spectral separation between An-PB and 7-AAD is larger than between Annexin-V-PE and 7-AAD. Neurons positive for An-PB only were scored as undergoing apoptosis (white arrowheads), neurons positive for 7-AAD only were scored as necrotic (yellow arrowheads), and neurons positive for both An-PB and 7-AAD (green arrowheads) were scored as being at end-stage apoptosis. Apoptotic indices scored based on three independent experiments in which scoring was blinded to the treatment were similar to the indices obtained by flow cytometry in (d). (d) Apoptotic indices of neurons transduced with non-targeting (nt) and GBE1 shRNA lentiviral particles (shGBE1), or transduced with non-targeting lentiviral particles and treated with SNP. Neurons were scored as in (c). Data are based on three independent experiments. For all indices, scoring of the SNP and shGBE1 treatments were higher than the nt treatment but not significantly different between themselves. These data are a part of the more inclusive data shown in Fig. 6b.

The mTOR inhibitors rapamycin and starvation can reverse polyglucosan accumulation and damage

Having established and validated a neuronal model for APBD, we used it to test possible therapeutic avenues. As shown in Fig. 2, GBE1 knockdown led to accumulation of the polyglucosan form of glycogen, while glycogen was not observed at all in control neurons. This finding suggested that de novo glycogen synthesis by GS had to be activated by the GBE1 knockdown and that the adverse effects of this knockdown might be corrected by GS inhibition. We therefore opted to test GS inhibition as a strategy for correcting GBE knockdown-mediated damage. Using shRNA as a tool for GS inhibition was unsuccessful because the yield of viable neurons where both GBE1 and GS were knocked down was too low for performing experiments. Therefore, as an alternative tool for inhibiting GS, we capitalized on inhibition of mTOR, a major switch of cell anabolism which can activate GS via phosphorylative inhibition of glycogen synthase kinase3ß. Treatment of GBE1-knocked down neurons with the mTOR inhibitors rapamycin or starvation reduced polyglucosan accumulation (Fig. 4a) and counteracted its associated apoptosis (Fig. 4b and c, cf Fig. 3b).

Figure 4.

(a) Confocal images of E18 rat embryo cortical neurons transduced with GFP encoding shGBE1 lentiviral particles (see Fig. 1) and treated with DMSO, 100 nM rapamycin (Rap), or starvation medium (st, Hank's Buffered Salt Solution supplemented with 1 g/L glucose) for 4 day. Following the treatments, neurons were paraformaldehyde fixed, permeabilized with Triton X-100, and stained with a monoclonal antibody against LC3 coupled to the FluoProbes647H dye (ex. 653 nm, em. 674 nm) and a monoclonal IgM antibody against glycogen followed by a Dylight549-coupled secondary anti-IgM antibody. Green channel, GFP; blue channel, LC3; red channel, glycogen. Scale bars = 2 μm (upper panel), 10 μm (second and third panels). (b) Neurons treated as detailed and abbreviated in (c) were analyzed by flow cytometry as in Fig. 3b. (c) Apoptotic index analysis performed as in Fig. 3d, comparing neurons transduced with non-targeting vector and treated for 4 day with DMSO (nt), 100 nM rapamycin (nt/Rap), or starvation medium (nt/st); transduced with nt lentiviruses and treated with 1 mM SNP for 24 h (SNP); transduced with lentiviruses containing shRNA against glycogen branching enzyme 1 (GBE1) and treated for 4 day with DMSO (shGBE1), or 100 nM rapamycin (shGBE1/Rap); transduced with lentiviruses containing shRNA against GBE1 and treated for 4 day with starvation medium (shGBE1/st). For all indices, scoring of the SNP, shGBE1, shGBE1/Rap/3-MA, and shGBE1/st/3-MA treatments was higher than all other treatments which were not significantly different among themselves. To the left of the vertical line are the results from Fig. 3d, shown here again for comparison.

However, a caveat of this approach is its lack of specificity: As an anabolic switch, mTOR does not only activate GS and glycogenesis but also inhibits catabolic autophagy. Inhibition of mTOR, an established inhibitor of autophagy, is therefore also expected to induce autophagy. Indeed both rapamycin and starvation treatments induced autophagy, as shown by the accumulation of autophagosomes, detected by their marker LC3 (Fig. 4a). We therefore had to test the possibility that autophagy induction is epiphenomenal to GS inhibition as a mechanism by which mTOR inhibitors restored GBE knockdown damage.

Autophagy, induced by mTOR inhibitors, is dispensable for correcting GBE knockdown damage

Theoretically, autophagy could have led to sequestration and auto-lysosomal digestion of polyglucosan by acid maltase and thus mediate the observed beneficial effects of the mTOR inhibitors. Using saponin permeabilization, we showed that the LC3 positive structures accumulated by rapamycin and starvation treatments (Fig. 4a) are autophagic vesicles [membrane associated, rather than cytosolic, entities (Fig. 5a)]. The increased autophagic flux, induced by rapamycin and starvation, presents maximal number of autophagosomes to the autophagic maturation process in which these autophagosomes mature into degradative autolysosomes. Hence, LC3 positive autophagosomes accumulate, as autolysosomal degradation is unable to keep pace with the overwhelming autophagic flux (Fig. 5a and (Boland et al. 2008)). If autolysosomal degradation was responsible for polyglucosan clearance, rapamycin and starvation treatments would have maximized it.

Figure 5.

(a) Right panel: Confocal images of glycogen branching enzyme 1 (GBE1)-knocked down neurons (Fig. 1) treated for 4 day with DMSO (DMSO), 100 nM rapamycin alone (Rap), or with 1 μM vinblastine (Rap+Vin), or starvation medium (st). All neurons were saponin permeabilized, so as to stain only membrane associated antigens, and then paraformaldehyde fixed and stained with a monoclonal antibody against the autophagosomal marker LC3. Scale bars = 10 μm. Left panel: Average numbers of LC3 positive structures per cell were scored by the Image-Pro software. (b) Left panel: shGBE1-knocked down neurons were treated for 4 day with DMSO, or the indicated combinations of 100 nM rapamycin (Rap), starvation medium (st), the pepstatin A (5 μM) and E64D (10 μM) lysosomal protease inhibitors (PI) and 1 μM vinblastine (Vin). Neurons were then lysed and subjected to SDS-PAGE and immunoblotting using antibodies against LC3. Molecular weight markers for lipidated LC3 (LC3 II, 16 kD) and non-lipidated LC3 (LC3 I, 18 kD) are shown. Right panel: Autophagic flux was quantified in the DMSO, rapamycin and starvation treatments (the only treatments where the LC3I band was detectable in the immunoblots) by densitometric measurement of the LC3 II/LC3 I ratios based on three independent immunoblots. Values significantly higher than the DMSO and Rap/st treatments are denoted by single and double asterisks, respectively. (c) GBE1 knocked down neurons treated with rapamycin and vinblastine (Rap/Vin), or starvation medium and vinblastine (st/Vin) as in (a) were stained for glycogen (red channel). GFP fluorescence (green channel) is shown to confirm neuronal transduction. A merge of the red and green channels is also shown. (d) Left panel, GBE1 knocked down neurons treated with rapamycin and vinblastine (Rap/Vin), or starvation medium and vinblastine (st/Vin) as in (a) were analyzed by flow cytometry as in Fig. 3b. Right panel, Apoptotic index analysis of the treatments in the Left panel, performed as in Fig. 3d. (e) Neurons treated as indicated were lysed and subjected to SDS-PAGE and immunoblotting using antibodies against phosphor Ser641 glycogen synthase (GS) (pGS), or GS. Scale bar = 20 μm.

To test the dependence of polyglucosan purging on autophagic maturation and autolysosomal degradation, we blocked both processes pharmacologically (double knockdown of GBE1 and autophagy genes is unfeasible, as explained above) using vinblastine, an established inhibitor of autophagic maturation (Kochl et al. 2006). The vinblastine block was manifested by further increase in the number and intensity of LC3 positive vesicles (Fig. 5a) and in the ratio of lipidated to cytosolic LC3 (LC3II/LC3I, Fig. 5b), which is inversely correlated with autophagic flux. We confirmed vinblastine's effect by showing that it reproduced the block in autophagic flux induced by lysosomal protease inhibitors (Fig. 5b, cf lipidated LC3 (LC3 II) to cytosolic LC3 (LC3 I) ratio between ‘Rap+PI’ and ‘Rap+Vin’ lanes), and also blunted the sensitivity of rapamycin induced neurons to them (Fig. 5b, cf ‘Rap+Vin’ lane to ‘Rap+Vin+PI’ lane). Even though it blocked autophagic maturation, vinblastine did not reverse the rapamycin (Fig. 5c, upper panel), or starvation (Fig. 5c, lower panel)-mediated down-modulation of polyglucosan accumulation, or did it reverse their anti-apoptotic protection (Fig. 5d). Interestingly, in contrast to vinblastine, 3-MA, an inhibitor of autophagosome formation, did blunt the reversal of apoptosis by rapamycin and starvation (Figure S2a and b). However, we presume this reversal depended not on inhibition of autophagy, but on the 3-MA-mediated GS activation (Fig. 7a and (Das and Hollinger 2012)) through dephosphorylation (Figure S2c and (Das and Hollinger 2012)), or phosphofructokinase1 inhibition (Figure S2c), which would slightly increase glycogen-6-phosphate (G-6-P). This assumption is corroborated by the observation that, in contrast to 3-MA, vinblastine affected only autophagy and not GS phosphorylation (Fig. 5e). The data presented in Fig. 5 thus suggest that protection of GBE1-knocked down neurons by rapamycin or starvation was independent of autophagic maturation and autolysosomal degradation.

In support of this conclusion, polyglucosans in rapamycin-treated or starved GBE-knocked down neurons were not observed in autophagosomes (Fig. 4a) and autolysosomes (Fig. 6c and Figure S3a), or were they observed in organelles partaking in the autophagic pathway: multivesicular body (Fig. 6a and Figure S3a) and amphisomes (the fusion product of autophagosomes and multivesicular bodies, Fig. 6b and c, sup. Fig. 3b). Apparent exosomes, formed by exocytotic diversion of the amphisome internal vesicles, also did not contain polyglucosans in rapamycin-treated (Fig. 6c) or starved (Figure S3c) GBE-knocked down neurons.

Figure 6.

shGBE1-transduced neurons were treated for 4 days with 100 nM rapamycin, fixed, dehydrated and analyzed by transmission EM. Micrographs show (a) a multivesicular body with internal vesicles (arrow); (b) an amphisome (circled), characterized by its content of electron-dense area juxtaposing intraluminal vesicles (arrow); (c) An autolysosome (arrowhead) and apparent exosomes (arrows in blowup i) near the PM. Different types of vesicles are shown in the framed area blown up in i. We presume the indicated vesicles are exosomes based on the fact that they are similar in size to the internal vesicles (white arrow) of the amphisome blown up in ii. (d) A micrograph showing an apoptotic neuron containing polyglucosans, based on their typical fibrillar ultrastructure (black arrowhead) shown in the blow up. Polyglucosans are typically also found in the nucleus (arrow), possibly as a result of apoptotic rupture of the nuclear envelope (white arrowhead). Scale bars = 200 nm (a and b), 500 nm (c), 2000 nm (d).

In contrast to their rapamycin-treated and starved counterparts, the untreated, apoptotic GBE1-knocked down neurons did contain non-membrane-associated polyglucosans smaller than 200 nm (Fig. 6d). This finding suggests that neurons that are not in contact with other cell types, as they are in the brain, might succumb to cell death once polyglucosan deposits appear and before they are able to grow in size to the > 1 μm PBs observed in APBD. This would explain why the average size of glycogen deposits (see also Figs 2 and 4) was smaller than in APBD affected brains, where PBs can occlude axons.

Correction of GBE knockdown damage by mTOR inhibitors is mediated by GS inhibition

We next aimed at confirming that the mode of action of rapamycin and starvation was through GS inhibition, as also observed in HepG2 cells (Varma et al. 2008), and implemented in rescue of a Pompe disease mouse model (Ashe et al. 2010). We show that both rapamycin treatment and starvation reduced GS activity both in control and in GBE1-knocked down neurons (Fig. 7a). GS basal activity, up-modulated 4-fold in GBE1-knocked down neurons, was lowered to control level when GBE1-knocked down neurons were treated with rapamycin, or starved (Fig. 7a). This action of both treatments was 3-MA reversible, suggesting that GS is rendered constitutively active by 3-MA, as corroborated by the observation that 3-MA rendered GS activity refractory to G-6-P stimulation. None of the treatments affected G-6-P-stimulated GS activity, suggesting it overrode GS phosphorylation state, in accordance with (Bouskila et al. 2010). GS protein levels were not affected by any of the treatments as our immunoblotting data show (Fig. 7a). To confirm that rapamycin and starvation mediated their effect via GS repression, we treated GBE1-knocked down neurons with the GS inactivator (AICAR) (Bultot et al. 2012). Like rapamycin and starvation treatments, AICAR action in GBE1-knocked down neurons led to reduction in GS basal activity without affecting the G-6-P-stimulated one (Fig. 7a), induction of autophagy (Fig. 7b, also observed in Lee et al. 2010), clearing of polyglucosans (Fig. 7b), and rescue of apoptosis (Fig. 7c). These results further corroborate our main concept that induction of autophagy is epiphenomenal to GS inactivation as a neuroprotective strategy against polyglucosan-mediated damage. Our observation that only GS activity but not its levels where modified in our APBD neuronal model is in apparent contrast with the results of (Valles-Ortega et al. 2011) who showed that in a mouse model of another neurological glycogenosis, LD, total GS levels were actually increased, while its activity did not significantly change. This apparent discrepancy demonstrates that increase in polyglucosan levels on its own does not necessarily lead to increase and accumulation of glycogen binding proteins in insoluble PB, as shown by the persistence of cytosolic GS (alongside GS in PB) in GBE1-knocked down neurons (Fig. 7d). Accumulation of glycogen binding proteins in PB is expected; however, if their levels are independently increased, as might be the case for GS when its E3 ubiquitin ligase malin is knocked down (Valles-Ortega et al. 2011).

Figure 7.

(a) Lower panel: Neurons were treated as detailed and abbreviated in Figs. 4, 5. shGBE1 5-Aminoimidazole-4-carboxamide ribonucleotide (AICAR), glycogen branching enzyme 1 (GBE1)-knocked down neurons treated for 4 days with 1 mM AICAR. Then, 1 mg of neuronal tissue was homogenized in 50 μL and subjected to glycogen synthase (GS) activity assay as described in (Akman et al. 2011) with (gray bars) or without (black bars) 10 mM glucose-6-phosphate. Upper panel: Cell lysates derived from the same treatments were subjected to SDS-PAGE and immunoblotting using anti-GS and anti-α-tubulin antibodies. Molecular weight markers are shown to the right. (b) Confocal images of E18 rat embryo cortical neurons transduced with GFP encoding shGBE1 lentiviral particles (see Fig. 1) and treated for 4 days with 1 mM AICAR. Following the treatment, neurons were fixed and stained for LC3 and glycogen as described in Fig. 4. (c) GBE1 knocked down neurons treated for 4 day with 1 mM AICAR were stained with Annexin V-Phycoerythrin and 7-Amino-actinomycin D and analyzed by flow cytometry, as described in 'Methods'. Percentages of cells in the different quadrants are indicated. (d) GBE1 knocked down neurons were fixed as described in 'Methods' and stained with a GS monoclonal antibody and a monoclonal IgM antibody against glycogen followed by a Dylight549-coupled secondary anti-IgM antibody and a Dylight647-coupled secondary anti-IgG antibody. Blue channel, GS; Red channel, glycogen. A merge of the two channels is shown to the right. Scale bar = 2 μm.

Having demonstrated that mTOR inhibitors exerted their action via GS inhibition, we set out to find a molecular mechanism for this activity. As Fig. 8a, shows, treating neurons with both rapamycin and starvation restored the inhibitory GS phosphorylation at Ser641 (Skurat and Roach 1995), dephosphorylated by GBE knockdown. In parallel, we reconfirmed, as in Fig. 7a, that rapamycin and starvation also lowered GS activity, up-modulated in GBE1-knocked down neurons, while none of the treatments affected total GS levels. This GS inhibition also coincided with restoration of ATP to control levels. Depletion of ATP as a kinase substrate, or as an AMP source for the GS inactivator AMP-activated protein kinase (AMPK) (Bultot et al. 2012), can possibly explain why GS was less phosphorylated and responsive to G-6-P activation in the GBE1-knocked down neurons, which undergo ATP-consuming apoptosis (Fig. 3). Further support for the notion that rapamycin and starvation acted via GS inhibition can be found in our observation that over-expression of a S641A, S645A, S649A, S653A, S657A phosphorylation sites-mutated and activated GS (Skurat and Roach 1995) overturned their ameliorating effects on polyglucosan accumulation and apoptosis (Fig. 8b). On the other hand, the beneficial effects of rapamycin and starvation were refractory to over-expression of a dual-site mutated GS where the quintuple phosphorylation mutations were supplemented by the D450A, R461A, F466A glycogen binding site mutations (Baskaran et al. 2011). These observations demonstrate that only an active (dephosphorylated) and glycogen binding GS can reverse the beneficial effects of rapamycin and starvation, which deducibly act by GS inhibition.

Figure 8.

(a) Upper panel, Neurons treated as described and abbreviated in Fig. 4 (e) were lysed and subjected to SDS-PAGE and immunoblotting using antibodies against phospho-Ser641 glycogen synthase (GS) (pGS), or GS. Lower panel, Neurons treated as in Upper panel were assayed for GS activity as in Fig. 7 and for ATP as described in 'Methods'. (b) Left panel, Neurons treated as in (a) were transfected as described in 'Methods' with constructs encoding for c-myc tagged mouse GS1 with the following phosphorylation sites mutattions S641A, S645A, S649A, S653A, S657A (Δphos), or phosphorylation site mutations supplemented with the following glycogen binding mutations D450A, R461A, F466A (ΔphosΔbi). Transfected neurons treated as indicated were fixed as described in 'Methods' and stained with a c-myc antibody conjugated to Phycoerythrin-Cy5 and a monoclonal IgM antibody against glycogen followed by a Dylight549-coupled secondary anti-IgM antibody. Right panel. The transfected neurons treated as indicated were analyzed by flow cytometry as described in 'Methods'. Scale bars = 1 μm (upper and second panels), 10 μm (third and lower panels).


This study was aimed at providing a proof of principle for developing therapeutics against APBD and other glycogenoses resulting from PB accumulation. To test the pathomechanism of PB accumulation, we opted for a neuronal model where formation of polyglucosans can be induced and the consequences followed. This system consisted of naïve neurons where the GBE/GS activity ratio (inversely proportional to the rate of polyglucosan buildup) was down-modulated by transduction with lentiviruses harboring shRNA against GBE1. While the relatively low cell number of transduced neurons limits the range of experiments that can be carried out using this system (e.g., fractionation is not feasible), the system is conceptually preferred over neurons derived from GBE knockdown or knockout mice (Akman et al. 2011) where neuronal polyglucosans pre-exist and therefore the effects following their formation cannot be followed.

Fig. 7 describes the effect of GBE1 on GS activation. In GBE1-knocked down neurons, basal GS activity increased 3-4-fold and became refractory to G-6-P activation. The results of (Akman et al. 2011) in embryonic mouse muscle show a similar, albeit significantly less pronounced, trend on GS activity in GBE1 knockdown mice (1.4-fold increase in basal activity and change in G-6-P activity ratio from 0.27 in Gbe1+/+ to 0.6 in Gbe1−/−, as compared to a change from 0.27 in control to 1 in GBE1-knocked down neurons). In contrast, however, (Lamperti et al. 2009) showed a lack of GS activity in the heart and liver of a GBE1 deficient neonate, possibly attributable to low availability of free glucosyl ends in larger PBs. A possible explanation of GBE1 control over GS activity in relative values (the absolute values are expectedly lower in neurons) is that GBE1 suppression leads to ATP deficit (Fig. 8a), probably more pronounced in isolated neurons, which undergo ATP-consuming apoptosis (Fig. 3). This ATP deficit might inhibit kinases and AMPK and render GS both less phosphorylated (more active) and less responsive to G-6-P activation. Hence, the differences between the three studies might be explained in terms of different ATP availabilities. Another explanation is based on the rate of glycogen2 synthesis. Reduced rate of glycogen synthesis could theoretically increase glycolytic flux, generating more G-6-P and rendering GS less G-6-P responsive. Although we could not determine the rate of glycogen synthesis in neurons, we presume it would have been higher in GBE1-knocked down neurons, since they contained glycogen in contrast to control neurons. Therefore, we hypothesize that reduced rate of glycogen synthesis contributed to the decreased GS G-6-P responsiveness only in Gbe1−/− embryonic muscle (Akman et al. 2011) and not in GBE1-knocked down neurons. In general, the observation that GBE knockdown can increase GS activity, induce de novo formation of polyglucosans (Fig. 2) and therefore be rescued by GS inhibition is important therapeutically as GS inhibition is more feasible as a prospective treatment than restoration of aberrant GBE activity. However, therapeutic implementation of this principle is still far.

Our results show that GS inhibition reversed polyglucosan accumulation and ensuing apoptosis (Fig. 7). On the basis of these results, we propose a therapeutic strategy aimed at alleviating polyglucosan-mediated damage in neurological glycogenoses. This strategy is based either on increasing GBE/GS activity ratio, and thus arresting further buildup of polyglucosans, or, preferably, on polyglucosan degradation, which eliminates pre-existing polyglucosans. We observed pre-existing polyglucosans in an APBD patient skin fibroblasts grown in ketogenic medium, where no de novo polyglucosans could have formed. As Figure S4 shows, polyglucosan accumulation induced by glucose can be reduced by rapamycin, which could not remove pre-existing polyglucosans. These data are consistent with the data in Fig. 4 where rapamycin removed polyglucosans from GBE-knocked down neurons, when it was introduced in parallel to the shGBE lentivirus, so that it inhibited de novo synthesis of polyglucosans, by counteracting GS activation (Figs 7 and 8). Therefore, rapamycin and other GS inhibitors are only expected to ameliorate glycogenoses, or slow down their progress, rather than eradicate polyglucosans which underlie their pathogenesis. GS inhibition has already been shown effective in the treatment of a murine model of LD via prtotein targeting to glycogen knockout (Turnbull et al. 2011). This approach is also clinically practical since GS deficiency is relatively tolerable in both mice (Pederson et al. 2005; Douillard-Guilloux et al. 2010) and humans (Kollberg et al. 2007), especially if blood glucose is monitored closely to avoid hypoglycemia.

A complementary approach to GS inhibition, also aimed at increasing GBE/GS activity ratio, would be up-modulation of GBE1 activity. We currently employ high-throughput screening of small molecule libraries to identify molecules with acceptable pharmacokinetic and pharmacodynamic profiles and minimal side effects which reduce polyglucosan levels. These molecules might be novel GS inhibitors, or enhancers of endogenous GS inhibitors, such as AMPK. Alternatively, these molecules may be GBE1 stabilizers, or more potent drugs promoting polyglucosan degradation.

Our results (Figs 5 and 6) suggest that, while induced, autophagy was not responsible for the improvement of GBE knockdown damage mediated by GS inhibition. Nevertheless, it is still possible that glycogen-selective autophagy might be effective in removing polyglucosans and relieving their damage. Specifically, glycogen autophagy could be induced by laforin, associated with both polyglucosan [not soluble glycogen, (Chan et al. 2004; Tiberia et al. 2012)] and the ER, the initiation site for autophagosome biogenesis (Hayashi-Nishino et al. 2009). These two associations suggest that laforin may be a primer for selective glycogen autophagy, especially as it was also shown to induce cytoprotective autophagy (Aguado et al. 2010). We conjecture that glycogen autophagy might have a therapeutic potential in glycogenoses based on observations in non-neuronal cells, where autophagy was shown to sequester and metabolize glycogen (Kotoulas et al. 2004; Schiaffino et al. 2008). Thus, glycogen autophagy might benefit APBD and LD patients both by PB disposal and as energy source through glycogen degradation in astrocytes.

In conclusion, we propose that GS inhibition, GBE1 activation, polyglucosan degradation and possibly glycogen autophagy are all avenues worth pursuing for the treatment of APBD and other neurological glycogenoses.


This study was supported by a research grant from the APBD Research Foundation to which we are greatly indebted. O. K. and H. O. A. designed research; O. K., H. G., N. F. and Y. L. performed research; O. B., T. T., Y. L. and H. O. A. contributed vital new reagents and analytical tools; A. L. provided clinical advice and patient samples; O. K. and H. O. A. analyzed and interpreted data; O. K. performed statistical analysis; O. K., H. O. A. and S. D. wrote the manuscript. The authors declare no conflict of interests.