Abstract

   The maintenance of metabolic homeostasis relies on the ability to
   flexibly transit between catabolic and anabolic states in response to
   insulin signaling. Here we show insulin-activated ATM is a critical
   mediator of this process, facilitating the swift transition between
   catabolic-and-anabolic fates of glucose by regulating the functional
   status of PKM2 and HIF1α. In Ataxia-Telangiectasia (A-T), these
   mechanisms are disrupted, resulting in intrinsic insulin resistance and
   glucose intolerance. Consequently, cells exhibit a compensatory
   dependence on glutamine as an alternative metabolite for energy
   metabolism. Cerebellar degeneration, a hallmark of A-T, is
   characterized by the pronounced vulnerability of Purkinje cells,
   attributed to their unexpected sensitivity to insulin. Supplementation
   with α-ketoglutarate, the α-keto acid backbone of glutamine, has
   demonstrated potentials in alleviating glutamine dependence and
   attenuating Purkinje cell degeneration. These findings suggest that
   peripheral metabolic deficiencies may contribute to sustained
   neurodegenerative changes in A-T, underscoring the importance of
   screening, monitoring and addressing these metabolic disruptions in
   patients.

   Subject terms: Mechanisms of disease, Metabolic syndrome, Metabolism,
   Neuroscience
     __________________________________________________________________

   Insulin-activated ataxia-telangiectasia mutated (ATM) regulates glucose
   metabolism. Here the authors report that its disruption in a mouse
   model of ataxia-telangiectasia leads to insulin resistance, glutamine
   dependence, and selective Purkinje cell degeneration, while
   α-Ketoglutarate supplementation shows promise in mitigating
   neurodegeneration.

Introduction

   Ataxia-telangiectasia (A-T) is a rare autosomal recessive disorder
   caused by loss-of-function mutations in the ataxia-telangiectasia
   mutated (ATM) gene. While cerebellar ataxia is the hallmark clinical
   feature of A-T, a significant proportion of patients also exhibit a
   broad spectrum of chronic metabolic syndromes^[38]1,[39]2. This
   included growth retardation, muscular atrophy and unexpected weight
   loss^[40]3. Among aging A-T patients, additional metabolic
   comorbidities, including dyslipidemia, glucose intolerance,
   nonalcoholic fatty liver diseases and diabetes, have also been
   reported^[41]4–[42]9. Notably, a family case study has shown that
   insulin resistance is associated with the worsening of cerebellar and
   motor function^[43]7. More broadly, cerebellar ataxia is also a common
   clinical feature observed in over 150 inherited metabolic
   disorders^[44]10.

   Building upon the growing understanding of the interaction between the
   brain and the periphery, a fundamental question that remains unresolved
   is whether metabolic disruptions originating from the periphery can
   exert lasting effects on brain function via altering the profiles of
   endocrine signals derived from the peripheral system^[45]11. The
   metabolic manifestations observed in individuals with A-T suggest a
   plausible role for endocrine dysregulation in these processes^[46]12.
   Among the endocrine signals frequently disrupted in A-T, insulin stands
   out, with numerous reports documenting diminished insulin sensitivity
   or insulin resistance in A-T patients^[47]7,[48]8,[49]13. Under normal
   physiological conditions, insulin serves as a key anabolic hormone,
   promoting the biosynthesis and storage of macronutrients such as
   glycogen, protein, and lipids. Simultaneously, insulin inhibits the
   breakdown of these macromolecules, thereby preventing any futile
   metabolic cycles^[50]14. Previous studies have identified the ATM
   kinase as a downstream mediator of insulin signaling, facilitating
   insulin-stimulated glucose metabolism^[51]15–[52]17. However, several
   critical questions remain unanswered: the precise mechanism by which
   insulin activates ATM, whether ATM plays a role in coordinating the
   dynamic transitions between catabolic and anabolic states in
   insulin-sensitive cells, and why these metabolic disturbances primarily
   contribute to ataxic symptoms in A-T, but not cognitive and psychiatric
   conditions that are also commonly related to cerebellar
   damage^[53]18,[54]19.

   Here, we show that deficiencies in peripheral metabolism play a
   significant role in the development of sustained cerebellar changes in
   individuals with A-T. Specifically, our findings suggest that the
   activation of ATM by insulin triggers a unique phosphoproteome network
   that is crucial for transitioning the glucose metabolic fate from
   catabolic to anabolic. When ATM is lacking, extensive metabolic
   reprogramming occurs, leading to a compensatory reliance on amino
   acids, especially glutamine, as an alternative fuel source for
   mitochondria. These alterations make insulin-sensitive Purkinje cells,
   particularly those expressing Zebrin-II/ALDOC which are anatomically
   abundant in the posterior and flocculonodular lobes of the vermal
   cerebellum, susceptible to the metabolic changes associated with ATM
   deficiency. Such an anatomical-specific vulnerability thereby
   contributes to the development of ataxic and motor-related symptoms in
   A-T. Our study demonstrates that supplementing with α-ketoglutarate,
   the α-keto acid of glutamine, can serve as an alternative metabolite to
   mitigate systemic glutamine dependence, glucose intolerance and
   simultaneously address concerns related to ammonia neurotoxicity.

Results

ATM deficiency leads to systemic insulin resistance and metabolic
inflexibility

   To comprehend the impact of ATM deficiency on whole-body metabolic
   status, a comprehensive metabolic analysis was conducted in
   Atm-knockout (KO) mice at ages comparable to adolescence (3 months old)
   and early adulthood (6 months old) in humans^[55]20. While discernible
   differences in physical parameters were absent among the younger mice,
   evident stunted growth and reduced body weight were observed in the
   6-month-old Atm-KOs (Fig. [56]1a, b, Supplementary Fig. [57]1a).
   Post-weaning mice are typically fed on carbohydrate-rich diets, akin to
   common human diets, indicating that carbohydrate is a primary source of
   carbon for biomass synthesis and growth^[58]21. The efficiency of
   assimilating digested and absorbed carbohydrates, i.e., glucose, was
   evaluated by Kraft test, which simultaneously evaluates glucose
   tolerance (intravenous glucose tolerance test, IGTT) and insulin
   response^[59]22. While the 3-month-old mice exhibited nearly normal
   glucose tolerance, elevated plasma insulin profiles however, suggested
   pre-diabetic insulin resistance (Fig. [60]1c, d). In contrast, majority
   of 6-month-old Atm-KO mice were glucose intolerant complicated with
   compromised insulin responses, hinting a progression towards type 2
   diabetes (Fig. [61]1c, d). The HOMA-IR (Homeostatic Model Assessment
   for Insulin Resistance) index calculation further confirmed that Atm-KO
   mice, irrespective of age, were insulin resistant
   (HOMA-IR > 2.2)^[62]23 (Fig. [63]1d). These observations therefore
   hinted an incompetence of carbohydrate utilization. Further elaboration
   by the indirect calorimetry test revealed that while the average food
   intake and locomotor activities remained similar among all the groups,
   persistent, non-fluctuating respiratory quotient (RQ) values ranging
   between 0.8 and 0.9 (indicative of protein catabolism) in both light
   and dark cycles were found among the 6-month-old Atm-KO mice
   (Fig. [64]1e). This contrasted with the profound ability to switch from
   protein/lipid metabolism to carbohydrate catabolism (RQ = 0.9–1) during
   the dark cycles observed in other experimental groups (Fig. [65]1e).
   Such a lack of metabolic flexibility due to insulin resistance and
   diabetic complication gradually depleted the adipose reserves
   (Fig. [66]1f) and skeletal muscle masses (Fig. [67]1g). Notably,
   abnormal accumulation of oil droplets in the liver (Fig. [68]1h) and an
   elevated fed-stage serum triglyceride (TG) levels (i.e.,
   hyperlipidaemia) were found (Supplementary Fig. [69]1b). These
   observations suggested a gradual development of cachexic-like
   phenotype^[70]24 among the diabetic 6-month-old Atm-KO mice.

Fig. 1. ATM deficiency gives rise to a hypercatabolic phenotype.

   [71]Fig. 1
   [72]Open in a new tab

   a Summary on body weight differences (n = 20, one-way ANOVA). b
   Representative images of mice showing body length differences (n = 18),
   quantification presented in Supplementary Fig. [73]1a. c Left: results
   of IGTTs (n = 12, two-way ANOVA). Right: area under the curves (AUCs)
   (n = 12, one-way ANOVA). d Left: plasma insulin levels were assessed
   during fasting and 2 h following the IGTT (n = 12; two-way ANOVA).
   Right: HOMA-IR index data (n = 12; one-way ANOVA). e Left:
   Chronological variations in indirect calorimetry measurements depicting
   alterations in respiratory quotient, food consumption, and locomotor
   activity. Right: Changes observed between light-and-dark cycles (n = 8;
   two-tailed unpaired t-test for RQ; one-way ANOVA for the rest). f Left:
   visual representation of mouse truncal fat deposits. Right: Percentage
   wet weight of fat pads over total body weight (n = 16, one-way ANOVA).
   g Images of hindlimbs and wet weights for major skeletal muscles
   [n = 16, two tailed unpaired t-test for all except for GST where
   two-tailed Mann Whitney test was applied]. h Representative images of
   liver oil red O staining. Alterations in lipid droplet density and area
   are presented (n = 80 images, 16 mice per group, two-tailed
   Mann-Whitney test). i Schematic diagram detailing the daily treatment
   regimen of streptozotocin (STZ) (40 mg/kg body weight, i.p.) and
   glargine (50 Units/kg body weight, s.c.). j Temporal variations in the
   indirect calorimetry readings on alterations in respiratory quotient
   (n = 6). Quantification presented in Supplementary Fig. [74]1d. k
   Outcomes of IGTTs following the STZ treatment paradigm depicted in
   Fig. [75]1i (n = 12, two-way ANOVA). AUCs presented in Supplementary
   Fig. [76]1f. l Fed serum triglyceride levels assessed before and after
   the STZ treatment regimen (n = 12, two-way ANOVA). m Representative
   images of liver oil red O staining following the STZ treatment regimen.
   Quantification presented in Supplementary Fig. [77]1g. N presents
   biological replicates. Values represent the mean ± s.d. Source data are
   provided as a Source Data file.

   Previous study revealed ATM mediates insulin action on enhancing
   glucose uptake and assimilation^[78]16, this prompted us to investigate
   whether the metabolic characteristics observed in Atm-KO mice are
   contributed by defective insulin actions. An artificial model of
   insulin deficiency induced by streptozotocin (STZ) was employed in
   1-month old young mice (Fig. [79]1i)^[80]25, an age well precede any
   obvious metabolic disturbances in concern had emerged (Fig. [81]1a–h).
   Subsequently, these mice were administered daily equal dosages of
   either long-lasting insulin (glargine) or a control vehicle for the
   following two weeks (Fig. [82]1i). Indirect calorimetry results
   revealed that among the WT mice, glargine administration restored
   systemic carbohydrate utilization, primarily during the dark cycle when
   food was consumed (Fig. [83]1j, Supplementary Fig. [84]1c, d). This
   treatment also averted the onset of glucose intolerance, elevation of
   serum triglyceride levels, and accumulation of oil droplets in liver
   emerged in the WT mice in vehicle treatment (Fig. [85]1k-m,
   Supplementary Fig. [86]1e–g). In contrast, the same glargine
   administration program failed to improve these metabolic abnormalities
   in Atm-KO mice, indicating that ATM is crucial for insulin’s systemic
   metabolic effect and the loss of ATM leads to these metabolic
   abnormalities.

Cerebellar Purkinje cells are insulin sensitive and functionally associated
with peripheral insulin resistance

   We next investigated if metabolic and endocrine disruptions related to
   insulin-ATM signaling originating from the periphery may contribute to
   cerebellar degeneration and ataxia in A-T via the altered endocrine
   signals of insulin. To confirm the exceptional relevance of such
   signaling pathway in this unique brain region, we first examined the
   spatiotemporal transcriptomic profiles of various brain regions
   throughout the human lifespan. Gene expression levels of ATM in the
   cerebellar cortex consistently champed over other brain regions during
   both prenatal and postnatal periods (Supplementary Fig. [87]2a)^[88]26.
   Notably, a similar expression pattern was also found for insulin
   receptor (INSR) (Supplementary Fig. 2a). Subsequent immunohistochemical
   analyses with antibodies targeting insulin receptor isoforms (INSRα/β)
   and insulin receptor substrates (IRS1/2) in 6-month-old Atm-WT mice
   unveiled a notable presence of INSRβ and IRS1 signals within Purkinje
   cells (IP3R1 + ) (Supplementary Fig. [89]2b), hinting they are
   responsive to insulin originated from the peripheral circulation that
   gained entry into the central nervous system via an active transport
   mechanism^[90]27,[91]28. In the in vivo settings, temporal dynamics of
   the insulin response following a surge in blood glucose introduced
   through IGTT was recorded (Fig. [92]2a). As typically expected, the
   duration of heightened plasma insulin levels lasted for approximately
   2 h before returning to near the baseline. This therefore, also led to
   the detection of insulin in the cerebrospinal fluid (CSF). Notably,
   elevated basal and saturated CSF insulin concentrations were detected
   in all Atm-KO mice before and at 2 h after the initiation of IGTT
   (Fig. [93]2a) which mirrors a phenomenon found in other pre-diabetic
   models^[94]29. With reference to this physiological time duration of a
   systemic insulin response, 2-hour treatment of glargine to ex vivo
   cerebellar tissue cultures was conducted, which further validated the
   responsiveness to insulin in this brain region (Fig. [95]2b).
   Functionally, behavioral evaluations on the motor and coordination
   functions, alongside the corresponding HOMA-IR index, demonstrated
   inverse correlations (Fig. [96]2c). This prompted us to examine the
   cerebellar tissues harvested at 2 h post-IGTT from 6-month-old mice.
   Those from the Atm-KO group revealed signs of insulin resistance,
   characterized by diminished tyrosine autophosphorylation of INSRβ in
   conjunction with heightened serine phosphorylation on IRS1
   (Fig. [97]2d, Supplementary Fig. [98]2c). Previously, insulin was found
   to promptly activate ATM in cellular models^[99]16, such response was
   also replicated in our ex vivo cerebellar tissues model (Fig. [100]2e),
   via an unexpected mechanism associated with oxidative stress and
   dimerization of ATM (Fig. [101]2e, Supplementary Fig. [102]2d, e). Such
   effect was effectively abolished by ethyl pyruvate, a stable scavenger
   of hydrogen peroxide (H[2]O[2]) (Fig. [103]2f)^[104]30. While the
   tyrosyl phosphorylation cascade on the insulin receptor traditionally
   mediates the insulin signaling, emerging evidence indicates that the
   hormone can also initiate an alternative H[2]O[2]-dependent response
   via the actions of NAD(P)H oxidase (NOX)^[105]31,[106]32. Notably
   expressed in the cerebellum (Supplementary Fig. [107]2f), NOX4 has been
   linked to A-T, although its association with insulin signaling was
   previously unestablished^[108]33. In the current investigation, the
   inhibition of NOX4 by its specific antagonist GLX7013114 or shRNA
   effectively impeded insulin-induced dimerization and activation of ATM
   (Fig. [109]2f, Supplementary Fig. [110]2g), confirming the involvement
   of an oxidative stress-related mechanism in this process.

Fig. 2. Insulin-activated ATM modulates key regulators of aerobic glycolysis
in the cerebellum.

   [111]Fig. 2
   [112]Open in a new tab

   a Plasma and cerebrospinal fluid insulin levels assessed during the
   IGTT (n = 12; two-way ANOVA). b Representative immunoblot images on key
   insulin signaling components. Quantification presented on the right
   (n = 10; two-tailed unpaired t-test). c Summary on behavioral
   performances in the open-field and rotarod paradigms (n = 12, one-way
   ANOVA for all except for Rotarod, constant 15 r.p.m test where
   Kruskal-Wallis test was used). Correlations between HOMA-IR index and
   performances (n = 48, Pearson’s correlation). d Representative
   immunoblot images on key insulin signaling components (n = 16).
   Quantification presented in Supplementary Fig. [113]2c. e, f
   Representative immunoblots on ATM status (n = 8, one-way ANOVA). g PCA
   plot of phosphoproteome profiles (n = 4 for all except n = 3 for
   KO-Glargine group). h Statistical significance versus changes in
   magnitude of phosphopeptides ( | log2 fold change | >1) (Limma with
   Benjamini-Hochberg correction) is shown. i Cell component analysis and
   j. functional pathway enrichment of phosphopeptides enriched in
   “WT+Glargine” group referencing the STRING database (Benjamini–Hochberg
   correction) and KEGG database on the Enrichr^[114]172 (Fisher exact
   test with correction), respectively. k Protein-protein interaction
   network analysis and l kinase enrichment analysis of phosphopeptides
   enriched in “WT + Glargine” group referencing the STRING database
   (Benjamini–Hochberg correction) and kinase enrichment database on the
   Enrichr^[115]13 (Fisher exact test with correction), respectively. m
   Visualization of phosphorylation motif of phosphopeptides detected in
   “WT + Glargine” group via the pLogo platform (Binominal probability
   with Bonferroni correction)^[116]152. n Relative intensities of
   pPKM2(T328) and pHIF1A(T163) phosphopeptides (n = 4 for all except
   “KO-Glargine” where n = 3, one-way ANOVA). o Representative time-lapse
   images of GFP-tagged PKM2 and HIF1α transient dynamics in cortical
   neurons pre-transfected with AAV-Synapsin-mCherry-C1-WPRE and
   pEGFP-C1-PKM2/pg-HIF-1alpha-EGFP. Quantifications presented on the
   right (n = 10, two-way ANOVA). p, q Measurements of (p). PKM activity
   and (q). degree of HRE binding (n = 16, two-way ANOVA). r
   Representative images of PKM2 (n = 8). s, t Representative immunoblots
   on s, nuclear localization (n = 12, one-way ANOVA) and t structural
   status (n = 12, two-tailed unpaired t-test) of PKM2. Unless otherwise
   specified, all ex vivo glargine and TEPP-46 treatments were performed
   in 100 nM for 2 h. Values represent the mean ± s.d. N presents
   biological replicates. Source data are provided as a Source Data file.

Insulin-activated ATM modulates key regulators of aerobic glycolysis in the
cerebellum

   ATM, a serine/threonine kinase conventionally associated with nuclear
   DNA damage responses, showcases more than 700 phosphorylation targets
   in response to DNA damage stressors^[117]34. In contrast, when serving
   as a cytosolic stress sensor, ATM engages a distinct array of
   substrates^[118]35. To explore the cellular consequences downstream of
   insulin-activated ATM, quantitative label-free phosphoproteomics
   analysis was performed in cerebellar tissues collected from both Atm-WT
   and Atm-KO mice, which identified a total of 2320 phosphosites that
   could be reproducibly quantified (Supplementary Dataset [119]1). Among
   which, 983 (i.e., 42.3%) phosphosites differ significantly among all
   the experimental groups (Supplementary Fig. [120]3a–f, Supplementary
   Dataset [121]1). This change contrasted with the total proteomic
   analysis conducted in parallel, which only 7.5% (i.e., 236/3150) of the
   total proteins detected were different significantly (Supplementary
   Fig. [122]4, Supplementary Dataset [123]2). Such huge difference
   indicated that most changes detected at the phosphoproteome level were
   independent from the total protein level changes. Focusing on the
   phosphoproteome profiles, principal component analysis (PCA) and
   hierarchical clustering analysis revealed that glargine-treated groups
   were distinctly separated, in contrast to the vehicle-treated groups
   (Fig. [124]2g, Supplementary Fig. [125]3a). The data quality was
   validated, which revealed that despite these observations, the number
   of phosphosites quantified (except 1 sample) (Supplementary
   Fig. [126]3b), so as the sample coefficients of variation among all
   treatment groups (Supplementary Fig. [127]3c) were highly similar. In
   the subsequent differential expression analysis, while the
   vehicle-treated groups (i.e., WT-Glargine vs KO-Glargine) failed to
   reveal any meaningful differences in their phosphoproteome profiles
   (Supplementary Fig. [128]3f), comparisons made between
   insulin-stimulated versus unstimulated groups in either Atm-WT or
   Atm-KO samples revealed obvious differences in the phosphoproteome
   profiles (Supplementary Fig. [129]3d-e). Specifically, among the Atm-WT
   samples, pathway analyses with the proteins derived from unique
   upregulated phosphosites in glargine-treated group were implicated
   widely in a set of central carbon metabolic pathways (i.e.,
   glycolysis/gluconeogenesis, pentose phosphate pathway, fructose and
   mannose metabolism and central carbon metabolism in cancer)
   (Supplementary Fig. [130]3d). Such pattern was however not observed
   among the Atm-KO samples (Supplementary Fig. [131]3e). Notably, a
   number of conventional protein phosphorylation events known to be
   triggered by insulin were detected in both glargine-treated Atm-WT and
   Atm-KO samples groups, which included pTSC2 at S939^[132]36; pARAF at
   S580^[133]37; pPFKL at S775^[134]38; pPDHA1 at S293^[135]39; pACLY at
   S455^[136]38 and pMDH1 at S241^[137]38 (Supplementary Fig. [138]3g).
   These hinted that ATM is likely regulating certain downstream actions
   of insulin, but not all.

   Based on these initial findings, we then directly compared the
   phosphoproteome profiles between the two insulin-treated groups (i.e.,
   Atm-WT + Glargine vs. Atm-KO + Glargine), which revealed a total of 79
   differentially changed phosphopeptides (Fig. [139]2h). Cellular
   component analysis with the 54 phosphopeptides belonged to 40 proteins
   differentially enriched in insulin-treated Atm-WT group revealed their
   predominant localization in various neuronal compartments
   (Fig. [140]2i). This finding not only supports the notion that
   phosphorylation events primarily occur in neurons within the cerebellum
   but also suggests a preference for subcellular enrichment in “neuronal
   cell body”, “somatodendritic compartment” and “neuron projection
   cytoplasm,” indicating that the cytoplasm is a key site for
   phosphorylation events where dimeric activated ATM can be
   found^[141]40. Functionally, KEGG pathway enrichment analysis revealed
   their potential roles in aerobic glycolysis (i.e.,
   “glycolysis/gluconeogenesis”, “metabolism in cancer”, and “HIF-1
   signaling pathway”) and RNA splicing (i.e., “Spliceosome” and “RNA
   degradation”) (Fig. [142]2j). Despite the apparent lack of overlap in
   biological functions between these two sets of pathways, a more
   in-depth examination of the protein-protein association network using
   STRING has revealed interconnectedness among proteins enriched in them
   (Fig. [143]2k). Of particular note is the presence of the protein
   HNRNPA1 (Heterogeneous nuclear ribonucleoprotein A1) in the network, a
   key RNA-binding protein that regulates the alternative splicing of PKM
   pre-mRNA, thereby favoring the formation of PKM2 mRNA over that of
   PKM1^[144]41,[145]42. This suggests that the enrichment of
   phosphoproteins related to RNA splicing may further regulate the
   effectiveness and capacity of aerobic glycolysis in Atm-WT neurons
   exposed to glargine (i.e., insulin)—a metabolic status that supplies
   metabolites for synaptic formation and neurite outgrowth during brain
   development^[146]43. In the contrary, phosphopeptides enriched in
   insulin-treated Atm-KO group (i.e., 20 phosphopeptides belonged to 19
   proteins) failed to be yielded into any biological function
   (Supplementary Fig. [147]3h).

   The above observation suggested that such changes are unique to an
   activated insulin-ATM signaling network. This notion was supported
   further by the kinase enrichment analysis^[148]44. With the
   phosphopeptides enriched in the insulin-treated Atm-WT group, ATM
   emerged the top kinases predicted, displaying as well an interrelating
   network with a group of kinases associated with insulin signaling
   (CSNK2A1^[149]45, PRKCB^[150]46, CSNK1D^[151]47, PRKDC^[152]48,
   CSNK1E^[153]49) (Fig. [154]2l). Such pattern was not observed with
   phosphopeptides enriched from the insulin-treated Atm-KO group
   (Supplementary Fig. [155]3i). Moreover, the total number of kinases
   enriched in the Atm-WT group exceeded that in the Atm-KO, with the
   majority (18/23) of uniquely identified kinases in the former playing
   roles in insulin signaling (CSNK2A1^[156]45, PRKCB^[157]46,
   MAPK14^[158]50, MAPK12^[159]51, MAPK10^[160]52, RPS6KA3^[161]53,
   MAPK9^[162]54, MAPK8^[163]55, MAPK7^[164]56, MKNK1^[165]57,
   MAPK1^[166]58, SGK1^[167]59, PRKACB^[168]60). (Supplementary
   Fig. [169]3j). At the substrate level, a substantial proportion (i.e.,
   21/54) of phosphopeptides enriched in insulin-treated Atm-WT group were
   modulated at SQ/TQ residues (Fig. [170]2m)—the substrate targeting
   motif of the ATM kinase^[171]61. Notably, these hits included PKM2 and
   HIF1α (Fig. [172]2n)—the common regulators of aerobic glycolysis
   effect^[173]62,[174]63. The necessity of ATM in mediating the insulin
   effect on PKM2 and HIF1α SQ/TQ phosphorylation was further validated by
   a gene knockdown analysis (Supplementary Fig. [175]2h). In the presence
   of ATM, acute glargine exposure resulted in robust nuclear
   translocation of PKM2 (Figs. [176]2o, r, s), in the expenses of its
   cytosolic role as a metabolic enzyme (Fig. [177]2p). Such changes were
   likely attributed to the loss in its tetramer quaternary structure upon
   SQ/TQ site phosphorylation (Fig. [178]2t, Supplementary Fig. [179]2h).
   Similarly, a delayed effect on nuclear localization and DNA binding
   effect of HIF1α was observed (Fig. [180]2o, [181]q, Supplementary
   Fig. [182]2h). Such a slower response could be in part due the inherent
   instability of HIF1α in the cytoplasm under normoxic condition^[183]64,
   and additional time is needed for the protein to build up in the
   system. Complementing these phenomena, in silico docking simulations
   further supported that phosphorylation of the conserved T163 residue on
   HIF1α is predicted to enhance its interaction with the R281 residue on
   its nuclear binding partner ARNT (Aryl hydrocarbon receptor nuclear
   translocator) (Supplementary Fig. [184]2i); whereas for PKM2, docking
   studies using a PKM2 monomer structure demonstrated that
   phosphorylation at T328 may also enhance its interactions with its
   partner nuclear importing factor importin-α3 (Supplementary
   Fig. [185]2j). Together, this observation demonstrated that
   insulin-activated ATM modulates the protein stability as well as
   nuclear activities of these key regulators of aerobic glycolysis in the
   cerebellum.

Defective insulin-ATM signaling results in glycolytic insufficiency and a
shift towards glutamine dependence in the cerebellum

   With these findings, it is then crucial to comprehend the molecular
   details of how ATM mediates the metabolic impact of insulin and the
   repercussions on the cerebellar metabolic balance in the absence of it.
   Global metabolic profiling analysis of mouse cerebellar tissues
   collected at 3 and 6 months of age unveiled a common set of perturbed
   metabolites (Supplementary Fig. [186]5a, b). Those related to
   “methionine metabolism” and “glycine and serine metabolism” were
   consistently elevated in the absence of ATM (Supplementary
   Fig. [187]5c), while the commonly downregulated metabolites were
   implicated mostly in the central glycolytic network (e.g., Warburg
   effect, gluconeogenesis, glycolysis, fructose and mannose degradation),
   suggestive of a reduced glycolytic activity (Fig. [188]3a). Also,
   metabolites belonged to multiple lipid biosynthesis pathways (e.g.,
   phospholipid biosynthesis, de novo triacylglycerol biosynthesis,
   inositol phosphate metabolism, sphingolipid metabolism, glycerolipid
   metabolism, glycerol phosphate shuttle, cardiolipin biosynthesis,
   phosphatidylcholine biosynthesis, phosphatidylethanolamine
   biosynthesis) which reflects the anabolic capacity of lipid
   macromolecules, were diminished as well (Fig. [189]3a). For those take
   part in the mitochondrial bioenergetic networks (e.g., carnitine
   synthesis and mitochondrial electron transport chain) were also reduced
   in the absence of ATM (Fig. [190]3a). Matching with these findings,
   bulk transcriptomic profiling of cerebellar tissues harvested from
   6-month-old mice revealed that differentially downregulated genes were
   highly involved in glycolysis, adipogenesis (i.e., lipid anabolism),
   and myogenesis (i.e., protein anabolism) (Supplementary Fig. [191]5d–f,
   Supplementary Dataset [192]3). Subsequently, an integrated
   gene-metabolite interaction network analysis consistently underscored
   such metabolic disturbances in the cerebellum under ATM deficiency
   (Fig. [193]3b).

Fig. 3. ATM deficiency disrupts glycolytic capacity and reprograms to
glutamine dependence in the cerebellum.

   [194]Fig. 3
   [195]Open in a new tab

   a Metabolite set enrichment analysis of commonly downregulated
   metabolites enriched in cerebellar tissues of 3- and 6-mo Atm-KO mice
   (n = 6, Globaltest with Bonferroni correction^[196]173). b Integrated
   gene-metabolite network analysis in cerebellar tissues harvested from
   6-mo Atm-WT and KO mice. Significantly changed KEGG metabolic genes and
   metabolites were mapped to Reactome pathways (n = 4 for transcriptome;
   n = 6 for metabolome, Globaltest with Bonferroni correction^[197]173).
   c Pie chart illustrating the changes in proportions of glucose,
   glutamine and fatty-acid dependence in acute ex vivo cerebellar culture
   (n = 3). Original data is shown in Supplementary Fig. [198]5g. d Left:
   schematics of ^13C[6]-U-glucose carbon flow. Right: mass isotopologue
   analysis of lactate, glycerol-3-phosphate (G3P), α-ketoglutarate (α-KG)
   in ex vivo cerebellar cultures [n = 16, one-way ANOVA for all except
   for (1) “M + 0 Lactate” where Kruskal-Wallis test was used]. e Left:
   schematics of ^13C[5]-U-glutamine carbon flow. Right: mass isotopologue
   analysis of γ-Aminobutyric acid (GABA), oxaloacetate (OAA), palmitate
   in ex vivo cerebellar cultures [n = 16, one-way ANOVA for all except
   for (1) “M + 0 GABA”, (2) “M + 4 OAA”, (4) “M + 0 GABA” and (6) “M + 0
   and M + 2 Palmitate” where Kruskal-Wallis test was used]. f
   Representative immunoblots on importin α3-PKM2 interaction and PKM2
   protein quaternary structure status (n = 16). Quantifications presented
   in Supplementary Fig. [199]5i. g Representative immunoblot images on
   subcellular localization of PKM2 (n = 16). Quantification presented in
   Supplementary Fig. [200]5j. h Schematics of possible fates of PKM2
   cytosolic dimer/monomers. i Representative blot images of
   RNA-immunoprecipitation assay for the interaction between PKM2 and the
   IRES region of the c-Myc mRNA harvested from human patient fibroblasts
   (n = 10). Quantification presented in Supplementary Fig. [201]6b. j
   Representative immunoblots showing levels of c-Myc protein in human
   patient fibroblasts (n = 12). Quantification presented in Supplementary
   Fig. [202]6c. k Schematics of human c-Myc consensus binding site
   sequences. Representative ChIP-qPCR blots showing levels of c-Myc
   binding to glutaminase-2 (GLS2) promoter in human patient fibroblasts
   (n = 10). Quantification presented in Supplementary Fig. [203]7d.
   Unless otherwise specified, all ex vivo glargine and TEPP-46 treatments
   were performed in 100 nM for 2 h. Values represent the mean ± s.d. N
   presents biological replicates. Source data are provided as a Source
   Data file.

   These similar findings, together with the physiological observations at
   systemic level (Fig. [204]1), collectively highlighted an inherent
   incompetence in utilizing glucose as a source of fuel for energy and
   carbons for macromolecule biosynthesis. To validate these notions,
   mitochondrial fuel-dependence analysis was conducted in ex vivo
   cerebellar cultures, which revealed a prominent shift from glucose to
   glutamine as the primary source of respiratory fuel (Fig. [205]3c,
   Supplementary Fig. [206]5g). Further isotopologue tracing analysis of
   heavily labelled glucose (^13C[6]-U-glucose) demonstrated that even
   under unstimulated condition, glucose flux towards the formation of
   glycerol-3-phosphate (G3P, M + 3) and α-ketoglutarate (α-KG) (M + 2,
   M + 4 from 2^nd round TCA) were already compromised in Atm-KO; except
   that towards lactate production (M + 3) [Fig. [207]3d(1-3)]. Upon a
   short 2-hour exposure to glargine, Atm-WT group promptly redirected
   glucose carbons towards various anabolic fates [i.e., G3P (M + 3),
   riboluse-5-phosphate (M + 5), serine (M + 3), sphinganine (M + 2) and
   glycine biosynthesis (M + 2) [Fig. [208]3d(2), Supplementary
   Fig. [209]5h)] while reducing the relative commitment to the formation
   of catabolic products such as α-KG (M + 2, M + 4) [Fig. [210]3d(3)].
   Notably, this response was partially counteracted by TEPP-46, a
   tetrameric PKM2 stabilizer [Fig. [211]3d(4-5), Supplementary
   Fig. [212]5h]. In contrast, glargine had little effect on glucose fate
   in Atm-KO group [Fig. [213]3d(1-3)]. With regard to the heightened
   glutamine dependence instead, fate tracing by heavily labelled
   glutamine (^13C[5]-U-glutamine) revealed Atm-KO exhibited inherently
   reduced its contributions to GABA biosynthesis [Fig. [214]3e(1)],
   whereas its oxidative fate in mitochondria, as indicated by elevated
   levels of M + 2 and M + 4 oxaloacetic acid (OAA), was enhanced under
   all treatment conditions as compared to the Atm-WT counterparts
   [Fig. [215]3e(2)]. Contrary to this, in Atm-WTs, baseline
   glutamine-mediated GABA synthesis was high but reduced sharply upon
   glargine stimulation [Fig. [216]3e(1)]. This change was attributed to
   the shift of glutamine fate towards its reductive anabolic fate as
   evident by the increased levels of cytosolic M + 3 OAA and heavily
   labelled palmitate [Fig. [217]3e(2-3)]. Such anabolic effects of
   glargine were effectively prohibited by TEPP-46 [Fig. [218]3e(4-6)].

   These findings hinted that the decrease in the active tetrameric
   configuration of PKM2, brought on by the insulin-ATM signaling, is a
   downstream effector response regulating the catabolic versus anabolic
   fates of glucose and glutamine in cerebellum. Further supporting this
   notion, immunoblotting confirmed an enhanced PKM2-importin-α3 binding
   in Atm-WT following glargine treatment, leading to predominant signals
   of monomeric and nuclear-localized PKM2 forms (Fig. [219]3f, g,
   Supplementary Fig. [220]5i, j). Conversely, in Atm-KOs, PKM2 existed
   predominantly as cytosolic, low pyruvate kinase metabolic activity
   dimers instead (Fig. [221]3f, g, Supplementary Fig. [222]5i-j). In this
   status, dimeric PKM2 may directly facilitate internal ribosome entry
   site (IRES)-dependent translation of c-Myc (Supplementary
   Fig. [223]6a)^[224]65—a pivotal regulator of glutamine dependence
   (Fig. [225]3h)^[226]66. In human A-T patient fibroblasts,
   RNA-immunoprecipitation assays unveiled an increased interaction
   between PKM2 and the c-Myc mRNA IRES region, hence an elevated c-Myc
   protein level was detected (Fig. [227]3i, j, Supplementary
   Fig. [228]6b, c). Glutaminase-2 (GLS2), the enzyme crucially induced in
   A-T cerebellar tissues^[229]67, promotes glutamine dependence as the
   major bioenergetic fuel in the mitochondira^[230]68. Notably, common
   c-Myc-binding sites were identified at both mouse and human GLS2
   promoters (Supplementary Fig. [231]7a–c). Chromatin immunoprecipitation
   (ChIP)-qPCR analysis focusing on the -1930 to -1401 base pair region
   upstream of the human GLS2 promoter revealed augmented c-Myc binding in
   A-T cells (Fig. [232]3k, Supplementary Fig. [233]7d). Collectively,
   these data not only signify a significant metabolic reconfiguration,
   but also a diminished insulin-driven anabolic switching effect when ATM
   is lost. At the molecular level, these metabolic shifts were
   orchestrated by the predominant accumulation of dimeric PKM2 in the
   cytosolic compartment.

Zebrin-II/ALDOC-positive Purkinje cells, reliant on glycolysis, are more
susceptible to metabolic challenges imposed by ATM deficiency

   On the top of exhibiting a high demand for energy, neurons in the
   maturing brain, in general, also relies on essential macromolecules
   like lipid- and amino acid- derivatives for proper growth and
   maintenance of synaptic functions^[234]69. Extended from the
   observation of reduced glycolytic-derived central carbon metabolites
   and neurotransmitter-related metabolites (e.g., GABA, serine,
   choline-reaction intermediates) (Fig. [235]3), targeted lipidomic
   analysis in cerebellar tissues revealed that the abundance of several
   species of fatty acids, phospholipids and sphingolipids was reduced in
   Atm-KO as well (Fig. [236]4a), hinting potential impacts to tissue and
   cell physiologies. Immunoblotting analyses further supported this
   notion, revealing signal reductions in markers of Purkinje cells
   (calbindin) but not those of granule neurons (NeuN) (Fig. [237]4b,
   Supplementary Fig. [238]8a). Further immunohistology examinations
   confirmed specific losses in IP3R1+ Purkinje cell number, soma
   diameter, and climbing fiber connections (VGlut2) (Fig. [239]4c–e,
   Supplementary Fig. [240]8b), suggesting an impaired dynamics of
   synaptogenesis and synaptic stabilization process in Atm-KO^[241]70.
   Notably, these changes were more prominently found within vermal
   posterior [i.e., Lobule (L) 6-L9] and flocculonodular lobes (i.e., L10)
   (Fig. [242]4c–e, Supplementary Fig. [243]8b), regions within the
   cerebellum that concern with whole-body posture, locomotion and
   balance^[244]71,[245]72.

Fig. 4. ALDOC-positive Purkinje cells reliant on glycolysis are selectively
more vulnerable to metabolic reprogramming due to ATM deficiency.

   [246]Fig. 4
   [247]Open in a new tab

   a Alterations in lipid-related metabolites (n = 5, two-tailed unpaired
   t-test). b Representative immunoblots of Purkinje cell (Calbindin,
   ALDOC) and granule neurons (NeuN) and GABAergic synapses (GAD67, VGAT)
   markers (n = 12). Quantifications presented in Supplementary
   Fig. [248]8a. c Schematics unveiling the characteristic foliation
   arrangement of cerebellar vermis. d Representative images depicting
   specific alterations in Purkinje cell density (n = 12). e Schematics of
   the gross anterior-posterior anatomy of the cerebellum. Key regions
   include SIM (simple lobule), C1 (crus 1), C2 (crus 2), PM (paramedian
   lobule), FL (flocculus), and PFL (paraflocculus). Quantifications and
   representative images demonstrating alterations in Purkinje cell
   (IP3R1 + ) density and the extent of climbing fiber connections
   (VGlut2) are shown (n = 16, two-tailed unpaired t-test). f
   Representative images and quantifications of ALDOC+ signals across
   distinct lobes (n = 16, two-tailed unpaired t-test). g–k Referencing a
   published dataset^[249]76, g UMAP plot of and h relative expression
   levels of ALDOC different cell types is shown. i Relative cell ratios
   of ALDOC+ and ALDOC- Purkinje cells. j DEG profiles in ALDOC+ Purkinje
   cells (Wilcoxon rank-sum test with Bonferroni correction). k Pathway
   enrichment of down-regulated DEGs in ALDOC+ Purkinje cells conducted on
   Enrichr^[250]172 (Fisher exact test with correction). l–o Referencing a
   published dataset^[251]163, l UMAP plot and t-SNE plots reveal the
   relative abundance of Purkinje cells in different cerebellar regions. m
   Relative expression level of ALDOC and n, cell ratio of ALDOC+ and
   ALDOC- Purkinje cells. o Gene set enrichment analysis of KEGG metabolic
   genes expressed in Purkinje cells found in different lobe regions
   (Weighted Kolmogorov–Smirnov test with Benjamini-Hochberg correction).
   p, q Referencing a published dataset^[252]162, p differential
   expression levels of HIF1α target genes and q the relative expression
   level of ALDOC in A-T versus NC samples (Limma with Benjamini-Hochberg
   correction). r Representative ChIP-qPCR blots illustrating the levels
   of HIF1α bound to Aldoc promoter in the acute ex vivo cerebellar
   culture exposed to 100 nM glargine (n = 8). Quantifications presented
   in Supplementary Fig. [253]S11f. Values represent the mean ± s.d. N
   presents biological replicates. Source data are provided as a Source
   Data file.

   Indeed, the innate properties of Purkinje cells are not uniform
   throughout different cerebellar lobules and can be sub-classified based
   on their Zebrin-II/ALDOC expression patterns^[254]73. Levels of ALDOC,
   a brain-specific isozyme of glycolytic aldolase highly expressed in a
   subpopulation of Purkinje cells located within the vermal posterior and
   flocculonodular lobes^[255]74, were more profoundly reduced in Atm-KO
   (Fig. [256]4c-e). This change was further validated by a similar
   pattern of losses in numbers of ALDOC+ Purkinje cells located at lobule
   L7-L10 (Fig. [257]4f, Supplementary Fig. [258]9). Intriguingly, as also
   a glycolytic enzyme^[259]75, these changes in ALDOC correlated with the
   pattern of loss in pIRS1 (Y612) signals, a marker of active insulin
   response (Supplementary Fig. [260]S10). The human relevance of such
   ALDOC changes was alternatively validated in an A-T cerebellar
   single-cell transcriptomic dataset^[261]76 (Fig. [262]4g–k,
   Supplementary Fig. [263]11d, e). Further comparison of single-cell
   transcriptome profiles of mouse Purkinje cells extracted from anterior
   (AL) and posterior/flocculonodular lobes (PL) of the vermal region not
   only validated that Purkinje cells within the PL were predominantly
   ALDOC+ (Fig. [264]4m, n). Additional investigations unveiled that this
   subset of Purkinje cells situated within the vermal PL region exhibited
   relatively heightened expression of genes associated with
   “glycolysis/gluconeogenesis,” “pentose phosphate pathway,” “fructose
   metabolism,” and “oxidative phosphorylation,” suggesting these cells by
   default have relatively greater activities and reliance in the central
   carbon metabolism network (Fig. [265]4o).

   The aldolase reaction is canonically situated at the center of the
   glycolytic pathway, implicated in enhancing lipid and cholesterol
   biosynthesis^[266]77. Here, by ectopic expression of ALDOC in HT22
   neuronal cells, this not only enhanced their glycolytic capabilities
   (Supplementary Fig. [267]11a-b), but also the production of
   glucose-derived triglyceride precursors like DHAP and G3P, as well as
   various species of fatty acids, phospholipids and sphingolipids
   (Supplementary Fig. [268]11c), thus validating the interrelationships
   among ALDOC level, glycolysis and lipid anabolism. Previous studies
   indeed revealed that Aldoc is a downstream target of
   HIF1α^[269]78–[270]80. This suggests that the losses of chronic
   insulin-activated ATM effects on HIF1α stabilization and subsequent
   nuclear localization in Atm-KO may compromise the expression of Aldoc
   as well. Accordingly, targeted gene expression analysis of 314 HIF1α
   targets extracted from the ChEA database revealed predominant
   reductions in A-T cerebellar tissues, including that of ALDOC as well
   (Fig. [271]4p-q). Further in silico promoter analysis revealed a
   conserved HIF1α binding region in both the mouse and human ALDOC gene
   (Supplementary Fig. [272]12). ChIP-qPCR then revealed a progressive
   increase in the association of HIF1α with the evolutionarily conserved
   hypoxia response element (HRE) positioned at bases -1308 to -1299
   within the Aldoc gene promoter, only in Atm-WT acute cerebellar tissue
   cultures after a prolonged exposure to glargine ( > 4 h) (Fig. [273]4r,
   Supplementary Fig. [274]11f). These findings imply that the levels of
   ALDOC and its impact on glycolytic capabilities are contingent upon a
   more chronic effect of insulin that is known to be crucial for normal
   growth and development during childhood and early
   adolescence^[275]81,[276]82—life stages that the maturation of the
   cerebellum is also the most robust^[277]83. These together explain how
   Zebrin-II/ALDOC-positive Purkinje cells are selectively more vulnerable
   under ATM deficiency, and how specific anatomical degeneration due to
   the loss of these cells results in motor-related phenotypes in A-T.

Nuclear-localized PKM2 co-activates HIF1α the aerobic glycolysis regulator to
reshape a metabolic landscape that favors glucose anabolic fate towards lipid
biogenesis

   Our data so far revealed that in the absence of ATM, insulin-sensitive
   cells fail to translate the hormonal signal into an instantaneous
   metabolic transition that fosters anabolic reactions that are
   particularly important for general growth and development during early
   childhood to early puberty^[278]84. During these life stages,
   hyperinsulinemia is considered as part of the physiological responses
   against reduced insulin sensitivity elicited by growth
   hormones^[279]85–[280]87. Under other circumstances, however,
   hyperinsulinemia is considered as pathological and it is a hallmark of
   prediabetes^[281]88, correlated to elevated risks of overweight and
   obesity in ordinary individuals^[282]89,[283]90. To investigate whether
   this pathophysiological outcome is otherwise stemmed from the sustained
   activation of ATM, six-month-old Atm-WT and Atm-KO mice were
   administered with daily doses of glargine for a duration of four weeks
   (in vivo half-life = 12–13.5 h)^[284]91,[285]92 to simulate a
   persistent hyperinsulinemia condition. This was either conducted alone
   or in combination with the blood-brain barrier-permeable PKM2 activator
   TEPP-46^[286]93 (Fig. [287]5a). After completing the treatment
   paradigm, majority of WT mice administered with glargine alone or in
   conjunction with TEPP-46 exhibited only slight indications of impaired
   glucose disposal (Supplementary Fig. [288]13a). However, indications of
   compensatory hyperinsulinemia and reduced insulin sensitivity were much
   evident and obvious, denoted by elevated levels of fasting plasma
   insulin and HOMA-IR values exceeding 2.2 after the IGTT (Supplementary
   Fig. [289]13b-c). Importantly, the marked increases in body weight
   observed in mice after receiving glargine alone were alleviated in
   those co-administered the TEPP-46 (Fig. [290]5b), suggesting that in
   the presence of a metabolically functional PKM2 tetramer, the
   pro-anabolic effect of chronic insulin is abolished. Conversely, given
   that the lack of ATM can directly nullify the anabolic effect of
   insulin as well (Figs. [291]2–[292]3), Atm-KO mice of this age were
   inherently insulin-insensitive and could even consider pre-diabetic or
   even diabetic prior to the commencement of this treatment regimen
   [Fig. [293]1c, d, Supplementary Fig. [294]13a–c (Atm-KO + vehicle
   group)]. Therefore, subsequent and sustained exposure to chronic
   glargine, whether administered alone or in conjunction with TEPP-46
   over the four-week period, did not result in discernible variances in
   glucose disposal, fasting plasma insulin levels, HOMA-IR values, or
   body weight, as compared to their corresponding vehicle-treatment group
   (Fig. [295]5b and Supplementary Fig. [296]13b, c). Likewise, indirect
   calorimetry assessment revealed that in WT mice, extended exposure to
   glargine alone in the absence of TEPP-46 impeded their ability to
   efficiently adjust the utilization of metabolic substrates between
   light and dark cycles (Fig. [297]5c). The resulting prolonged
   carbohydrate utilization during both cycles (indicated by a RQ = 0.9–1)
   was likely sustained by increased standard laboratory chow intake (62%
   carbohydrate)^[298]94 (Supplementary Fig. [299]13d). This phenomenon
   further explains how more rapid increase in body weight was observed in
   these mice (Fig. [300]5b), as was linked to increased adipose tissue
   deposits (Fig. [301]5d) and the accumulation of large-sized oil
   droplets in the liver (Fig. [302]5e). Despite these changes, no obvious
   differences in the majority of the skeletal muscle mass were found
   (Supplementary Fig. [303]13e). In contrast to these findings in Atm-WT
   mice, Atm-KO mice failed to respond to glargine or when co-treated in
   combination with TEPP-46 (Fig. [304]5b-e). Their inherent insulin
   resistance and glucose intolerance results in sustained reliance on
   lipid and protein utilization (indicated by RQ = 0.7–0.8) throughout
   both light and dark cycles (Fig. [305]5c). Despite these mice also
   consumed a comparable amount of standard laboratory chow (62%
   carbohydrate)^[306]94 as the Atm-WT mice on vehicle (Supplementary
   Fig. [307]13d), the absence of the anabolic impact of insulin resulted
   in inefficient carbohydrate utilization, as evidenced by the lack of
   biologically meaningful changes in respiratory quotient measurements at
   times when food was heavily consumed during the dark cycles
   (Fig. [308]5c), so as the status of glucose intolerance reflected by
   the IGTT (Supplementary Fig. [309]13a). This inefficiency extended to
   limited lipogenic capacities (Fig. [310]5d) while heightening lipolysis
   instead, as indicated by their significantly lower adipose tissue mass
   and accumulation of relatively small-sized oil droplets in the
   liver^[311]95 (Fig. [312]5d, e). These findings underscore the
   essential role of ATM in enabling the transition between catabolic and
   anabolic metabolic states at the whole-body systemic level.

Fig. 5. Chronic activation of ATM upon sustained exposure to insulin promotes
glucose anabolic flux towards lipid biogenesis.

   [313]Fig. 5
   [314]Open in a new tab

   a Schematic illustrates the daily regimen of glargine (50 Unit/kg body
   weight, subcutaneous) ± TEPP-46 (50 mg/kg body weight, intraperitoneal)
   administration over a 4-week period. b Changes in mean body weight
   throughout the treatment paradigm (n = 16, two-way ANOVA). c
   Chronological variations in indirect calorimetry assessments
   highlighting alterations in respiratory quotient. Summary of changes
   observed between light and dark cycles (n = 12; two-tailed unpaired
   t-test). d Changes in percentage of wet weight in various fat pads
   (n = 16, one-way ANOVA for all except for “BAT” and “RETRO” groups
   where Kruskal-Wallis test was used). e Representative images of liver
   oil red O staining and variations in lipid droplet densities and areas
   (n = 40 images, Kruskal-Wallis test). f Schematics illustrate the
   transition between PKM2 tetramers, dimers, and monomers, and potential
   outcomes for nuclear monomers. g, h Representative immunoblots
   depicting the protein dynamics of PKM2 and HIF1α in g whole cell and h
   subcellular fractionated lysates harvested from cerebellar tissues
   (n = 12). Quantification presented in Supplementary Fig. [315]13f, g. i
   Assessment of HRE binding and PKM metabolic activities in tissues
   collected after the drug treatment paradigm (n = 16, One-way ANOVA). j
   Schematics illustrate HIF1α targets (i.e., green highlights) with roles
   as pivotal regulators of the glycolytic pathways. k Expression levels
   of HIF1α targets, as indicated in j, in cerebellar tissues (n = 12,
   one-way ANOVA for all except for “Agpat5” and “Pdk1” where
   Kruskal-Wallis test was used). l, m Alterations in 1, targeted
   lipid-related metabolite in harvested cerebellar tissues and m
   statistical analyses of levels of each metabolite (n = 5, one way ANOVA
   for all except for “acetyl-CoA”, “Sphingomyelin(d18:1/18:0)”, “CDP” and
   “acetylcholine” where Kruskal-Wallis test was used). n Summary of
   longitudinal behavioral performances in the open-field test and rotarod
   tasks before and after the administration of drug treatments (n = 16,
   two-way ANOVA). Values represent the mean ± s.d. N presents biological
   replicates. Source data are provided as a Source Data file.

   To underscore if these systemic peripheral metabolic alterations would
   also be reflected in the cerebellum, cerebellar tissues were also
   collected from these animals for further analysis. As already shown
   previously (Fig. [316]2), following extended insulin exposure,
   activated ATM phosphorylates and promotes nuclear localization of both
   PKM2 and HIF1α. The combined effects also enabled nuclear-localized
   PKM2 to engage with and function as a co-activator of HIF1α^[317]96 in
   Atm-WT cells (Fig. [318]5f–h, Supplementary Fig. [319]13f-g).
   Remarkably, in specimens collected from Atm-WT subjects receiving
   TEPP-46 co-treatment, PKM2 stabilization in its tetrameric quaternary
   structure and metabolic activities were anticipated (Fig. [320]5g–i).
   This stabilization effect also counteracted ATM-mediated SQ/TQ
   phosphorylation on PKM2, likely an outcome of stearic hinderance caused
   by forced tetramerization. Subsequently, TEPP-46 also abolished
   glargine-stimulated nuclear translocation of PKM2 and its interaction
   with HIF1α (Fig. [321]5g, h, Supplementary Fig. [322]13f-g). Consistent
   with the observation in peripheral tissues such as the adipose tissue,
   chronic glargine-induced nuclear localization of HIF1α in conjunction
   with nuclear monomeric PKM2, which facilitated a robust nuclear HRE
   binding in the central nervous system (i.e., cerebellum and cerebral
   cortex) (Fig. [323]5i). Consequently, the expression levels of HIF1α
   target genes, which encompass genes encoding proteins and enzymes that
   facilitate the utilization of glucose carbons for fatty acids,
   phospholipids and sphingolipids biosynthesis, were also upregulated
   (Fig. [324]5j, k and Supplementary Fig. [325]14). The lipogenic impacts
   in the cerebellum were corroborated through targeted lipidomic
   assessment in the Atm-WT samples (Fig. [326]5l-m). Significantly, in
   the context of human A-T, lipogenic pathways were consistently
   compromised, as reflected from the bulk transcriptome data of human
   cerebellar samples (Supplementary Fig. [327]15). Specifically, this set
   of lipogenic pathways were also relatively more compromised ALDOC+ but
   not ALDOC- Purkinje cells in A-T brains (Fig. [328]4j–k, Supplementary
   Fig. [329]11d, e). Among the Atm-KOs, the absence of molecular
   responses suggested that the administration of glargine or TEPP-46 did
   not yield in any beneficial or detrimental effects on their behavioral
   performance (Fig. [330]5n). These findings together align with the
   earlier work indicating that defective lipid dyshomeostasis is a
   prevailing pathological mechanism in inherited cerebellar
   ataxia^[331]97.

Supplementation of α-ketoglutarate (α-KG), the α-keto acid of glutamine,
mitigates metabolic challenges associated with ATM deficiency

   Our in vivo and in vitro analyses suggest that in ATM-deficient
   conditions, glycolytic insufficiency triggers a compensatory dependence
   on amino acids, particularly glutamine, as an alternative mitochondrial
   carbon source (Fig. [332]3c–e). While the major source of glutamine is
   skeletal muscles^[333]98, constant breakdown of proteins to sustain its
   increased demand as a fuel by the body can result in a reduction in
   lean muscle mass, further heightening the risk and progression of
   insulin resistance and other metabolic complications^[334]99. To
   mitigate this hypercatabolic effect, one straightforward strategy may
   involve the dietary supplementation of glutamine. While short-term
   supplementation over a period of 2 weeks showed initial
   promise^[335]100, concerns over ammonia (NH[3]) neurotoxicity^[336]101,
   particularly if such interventions are prolonged for the management of
   genetic disorders, may limit its potential.

   α-KG, the α-ketoacid derived from glutamine/glutamate, plays a pivotal
   role as the nexus between cellular carbon and nitrogen
   metabolism^[337]102. As a linear keto acid, this metabolite can be
   directly absorbed in the stomach and small intestine^[338]103,[339]104.
   α-KG in circulation could then traverse the blood-brain barrier through
   both carrier-mediated and passive diffusion
   mechanisms^[340]105,[341]106. At the target cell surface, the
   transmembrane transportation of a conjugated form of α-KG as divalent
   anions can be efficiently facilitated by sodium-dependent dicarboxylate
   transporters^[342]107—NaDC1/SLC13A2 and NaDC3/SLC13A3. Indeed, gene
   expression levels of them were induced in cerebellar tissues of A-T
   (Fig. [343]6a). Once inside the cell, α-KG has the capacity to
   transform into glutamate through a transamination reaction facilitated
   by glutamic-oxaloacetic transaminase 1-like 1 (GOT1L1), even in
   scenarios where the major GOT1 enzyme is suppressed (Fig. [344]6a).
   This conversion also generates cytosolic OAA, which may enter the
   mitochondria via specific carriers encoded by SLC25A34 and
   SLC25A35^[345]108 to bolster the TCA cycle. Concurrently, the glutamate
   produced during this process can undergo two metabolic fates: either it
   could be transformed into GABA by the cytosolic glutamate
   decarboxylase-1 (GAD1) or enter the mitochondria via glutamate
   carrier-2 (GC2/SLC25A18) where expression of these two genes both
   trended upward in A-T cerebellar tissues as well (Fig. [346]6a). Once
   inside the mitochondria, glutamate may deaminate back to α-KG by
   glutamate dehydrogenase (GLUD2) and enter the TCA cycle (Fig. [347]6a).
   The ammonia liberated from this deamination reaction can be utilized by
   the cytosolic glutamate-ammonia ligase (GLUL), which converts glutamate
   into glutamine to support GABA synthesis (Fig. [348]6a). In this
   sequence, exogenous α-KG could effectively serve as a substitute for
   endogenous muscle-derived glutamine in generating not only a carbon
   source for sustaining the TCA cycle (Fig. [349]6a), but also a balanced
   molarity of ammonia for GABA synthesis, as supported further by
   subsequent isotopologue tracing experiments (Fig. [350]6b-c). In ex
   vivo cerebellar cultures derived from Atm-KO mice, the introduction of
   heavily labelled disodium-conjugated 1,2,3,4-^13C[4]-α-KG effectively
   competed against the involvement of heavily labelled
   ^13C[5]-U-glutamine as mitochondrial fuel (Fig. [351]6c), suggesting
   the potential capability of α-KG to surrogate the exhaustion of
   endogenous glutamine derived from skeletal muscles or
   neurotransmitters. Significantly, tissue ammonia level was reduced as
   well (Fig. [352]6d), affirming that α-KG represents a less hazardous
   but a more efficient substitute for metabolic supplementation.

Fig. 6. α-Ketoglutarate supplementation mitigates endogenous glutamine
wastage and insulin resistance in Atm-KO mice.

   [353]Fig. 6
   [354]Open in a new tab

   a Diagram illustrates the metabolic fates of α-KG or glutamine.
   Expression level changes in genes involved in the key metabolic
   reactions in human A-T cerebellar tissues (from [355]GSE61019) compared
   to non-diseased controls. Values presented in brackets represent
   Log[2](fold change) followed by Log[10](adjusted p-values). b
   Schematics of ^13C[5]-U-glutamine and 1,2,3,4-^13C[4]-α-KG carbon flow.
   c Isotopologue analysis of GABA, succinate, citrate, and OAA in acute
   ex vivo cerebellar cultures. The relative distributions of heavy
   isotopologues over the total amount of the respective metabolite are
   depicted (n = 8, two-tailed unpaired t-test for all except for “M + 4
   citrate” where two-tailed Mann-Whitney test was used). d Tissue ammonia
   levels in acute ex vivo cerebellar cultures (n = 8, two-tailed unpaired
   t-test). e Schematics of a dietary regimen involving CaAKG
   supplementation (2% w/w) or vehicle alongside a standard laboratory
   diet. f Survival analysis throughout the treatment paradigm (n = 20,
   two-way ANOVA). g Representative mouse images depicting body length
   differences and their quantifications after the treatment paradigm
   (n = 16, one-way ANOVA). h Summary of the average wet weights of major
   skeletal muscles (n = 8, one-way ANOVA). i Chronological variations in
   indirect calorimetry readings demonstrating alterations in respiratory
   quotient. Overview of changes between light and dark cycles (n = 26,
   two-tailed Mann-Whitney test for all except for WT-V where two-tailed
   unpaired t-test was used). j Outcomes of IGTT (n = 26, two-way ANOVA)
   and calculations of the AUCs (n = 26, one-way ANOVA). k Assessment of
   plasma insulin levels after overnight fast and at 2 h after the IGTT
   (n = 26, two-way ANOVA). Calculation of the HOMA-IR index shown on the
   right (n = 26, one-way ANOVA). l, m Overview of behavioral performances
   in the l open-field and m rotarod paradigms (n = 26, Kruskal-Wallis
   test). Correlations between the HOMA-IR index and performances (n = 78,
   Pearson’s correlation). n Representative images of cerebellar vermis
   anterior (L3) and posterior (L8) lobes. Quantifications of IP3R1+
   Purkinje cell numbers per mm^2 area is shown (n = 12, one-way ANOVA for
   all except for L6 where Kruskal-Wallis test was used). Values represent
   the mean ± s.d. N presents biological replicates. Source data are
   provided as a Source Data file.

   In the in vivo context, a small pilot study utilized young Atm-KO mice
   at postnatal day 25 (P25) was conducted to first evaluate the safety
   and tolerability of 2% calcium α-KG (CaAKG), an alternate form of
   α-KG^[356]109,[357]110. This pilot study encompassed a 60-day period
   during which 2% CaAKG was either included in soften food pellets or
   administered intraperitoneally (10 mg/kg/day) (Supplementary
   Fig. [358]16a). Initial results from the IGTT conducted at P25
   indicated no significant differences in baseline blood glucose
   clearance abilities before the treatment (Supplementary Fig. [359]16b).
   However, by P85, following the completion of the treatment regimen,
   Atm-KO mice treated with the vehicle exhibited notably impaired blood
   glucose tolerance as compared to vehicle-treated Atm-WT mice
   (Supplementary Fig. [360]16b). Comparative assessments revealed that
   mice receiving 2% CaAKG through dietary incorporation over the same
   duration demonstrated more superior blood glucose control outcomes as
   compared to those received intraperitoneal injections (Supplementary
   Fig. [361]16b). Moreover, metabolic profiling conducted at P85 also
   revealed a more prominent improvement in the switching fuel dependence
   among the mice undergone dietary supplementation with 2% CaAKG, as
   evidenced by the bigger variance in RQ values between the light and
   dark cycle s (Supplementary Fig. [362]16c). Noteworthy, parameters such
   as food intake (Supplementary Fig. [363]16d) and locomotor activity
   levels remained relatively consistent across all experimental groups
   (Supplementary Fig. [364]16e). Based on the outcome of this pilot,
   subsequent experiments which extended the treatment duration to 185
   days via the dietary supplementation approach was implemented
   (Fig. [365]6e). Analysis of survival rates revealed a significant
   reduction in mortality among Atm-KO mice receiving 2% CaAKG from the
   diet (Fig. [366]6f), an improvement likely associated with enhanced
   overall growth, as evidenced by increased body length (Fig. [367]6g)
   and greater preservation of lean skeletal muscle mass^[368]111
   (Fig. [369]6h). As an α-keto acid (RQ > 1), CaAKG overcame the
   challenges related to inefficient utilization of carbohydrates (i.e.,
   absence of RQ switching to 0.9–1.0) (Fig. [370]6i) via serving as a
   biological fuel itself. This treatment concurrently alleviated the
   reliance on protein (RQ = 0.8), particularly when CaAKG was consumed
   along food at multiple times within a day to sustain this effect
   (Fig. [371]6i, Supplementary Fig. [372]16f-g). The consequent
   preservation in lean muscle mass, a largest tissue in the body, which
   is also sensitive to insulin, not only improved the overall glucose
   tolerance (Fig. [373]6j), fasting hyperinsulinemia status
   (Fig. [374]6k), but also the HOMA-IR status in comparison to the
   vehicle-treated Atm-KO counterparts (Fig. [375]6k). These endocrine
   profile improvements also significantly alleviated the degree of
   deterioration of coordination and motor function in these mice
   (Fig. [376]6l-m). The systematic enhancements in metabolic health that
   initiates from the peripheral, coupled with the direct impact of CaAKG
   on glutamine dependence, better preserved the number of insulin-sensing
   Purkinje cells (Fig. [377]6n, Supplementary Fig. [378]17) and their
   synaptic connections in the posterior and flocculonodular lobes of the
   vermal cerebellum (Supplementary Fig. [379]18).

Discussion

   Mammalian cells, unlike unicellular organisms, lack the intrinsic
   ability to independently uptake extracellular nutrients but instead
   depend on the presence of extracellular growth factor signals to
   regulate nutrient absorption and subsequent metabolic
   processes^[380]112,[381]113. Insulin, a crucial signal rapidly
   initiated in response to elevated postprandial blood sugar levels,
   plays a significant role in this regulatory network^[382]114. ATM is
   traditionally recognized as a pivotal component in the DNA damage
   response pathway^[383]115, yet recent investigations from our
   laboratory and others have also underscored the critical involvement of
   the kinase in the central carbon metabolism^[384]116. This dual role
   suggests that ATM may serve as a significant regulator that instigates
   metabolic reprogramming in response to physiological
   stressors^[385]67,[386]117,[387]118. Our work here highlights the
   pivotal involvement of ATM in translating acute and chronic insulin
   hormonal signals into appropriate anabolic responses by initiating a
   specific phospho-protein network. Importantly, the linkage between
   insulin and ATM also facilitates an important crosstalk between the
   peripheral and central nervous system. Noteworthy downstream components
   of this network include PKM2 and HIF1α, key regulators of aerobic
   glycolsis^[388]62,[389]114 that allows the diversion of glycolytic
   intermediates to facilitate the biosynthesis of various
   macromolecules^[390]119.

   The enzyme PKM assumes a pivotal role in the enzymatic conversion of
   phosphoenolpyruvate to pyruvic acid at the terminal step of glycolysis,
   thereby exercising critical regulation over the fate of glucose
   carbons^[391]120. Our investigations have unveiled that in individuals
   subjected to physiological duration of insulin stimulation, activation
   of ATM leads to instant phosphorylation and destabilization of PKM2
   tetramers. Consequently, this protein quaternary structure loss results
   in diminished pyruvate kinase activity, prompting a diversion of
   glucose carbons towards biomass generation in the cytoplasm instead of
   committing to mitochondrial oxidation, a phenomenon also reported by
   others as well^[392]121. In the context of chronic insulin exposure,
   however, the sustained activation of ATM further influences the
   behavior of PKM2. In conjunction with a delayed accumulation and
   nuclear translocation of ATM-phosphorylated HIF1α, PKM2 co-activates
   HIF1α^[393]96 to instigate a metabolic program favoring lipid
   biogenesis. While this phenomenon could be physiologically meaningful
   during mid-childhood^[394]84 to early puberty^[395]85–[396]87 for
   speeding overall growth and brain maturation, it is more commonly
   associated with prediabetes during other life stages^[397]88,
   contributing to body weight gain and excessive adipogenesis.
   Conversely, in the context of A-T, the deficiency of ATM resulted in
   PKM2 existing as a metabolically low-activity dimer in the cytoplasm,
   rather than the high-activity tetramer. This could be attributed to
   inherently depleted levels of its allosteric effectors, such as
   fructose-1,6-bisphosphate and serine, in these cells, which is also a
   side-effect of glycolytic insufficiency^[398]122,[399]123. The lack of
   ATM also impedes both the immediate and protracted anabolic responses
   evoked by insulin. This deficit potentially accounts for the
   manifestation of inherent insulin insensitivity, stunt and even
   neurodegeneration in individuals afflicted with the disease.

   Our findings show that HIF1α is a target of ATM resulting from
   prolonged, rather than short-term, insulin exposure. Therefore, under
   normal insulin response durations, we do not anticipate the
   stabilization and nuclear activities of HIF1α. However, when insulin
   response is extended—either as part of the normal growth and
   development process typically seen from mid-childhood to early
   puberty^[400]84 or in pathological conditions like pre-diabetes—the
   stabilization and nuclear activities of HIF1α emerge. This supports a
   transcriptomic landscape that further facilitates aerobic glycolysis to
   stimulate lipogenesis, aiding in the biosynthesis of membrane-related
   phospholipids or triglycerides for storage purposes. At molecular
   level, HIF1α is a transcription factor being continuously produced but
   swiftly degraded under normoxic conditions unless subjected to
   post-translational modifications^[401]124. The degradation process
   involves hydroxylation and destabilization of HIF1α by α-KG-dependent
   prolyl hydroxylases (PHDs) or asparaginyl hydroxylases such as factor
   inhibiting HIF-1 (FIH-1)^[402]125. Therefore, it is possible as well
   the supplementation of α-KG could instead engender a heightened risk of
   further diminishing the residual HIF1α activity and exacerbate the
   metabolic defects in ATM deficiency. Noteworthy, in addition to α-KG,
   the activities of PHDs/FIH1 are also dependent on the co-existence of
   other co-factors, including Fe (II) ions, oxygen, and
   ascorbate^[403]126. Moreover, it is also salient to note the
   distinctive capacity of PHD3 to augment instead of impeding PKM2
   function as a coactivator of HIF1α^[404]96. This knowledge collectively
   suggests the presence of a complex regulatory framework^[405]127, which
   emphasize that the regulation of PHD activities within the
   intracellular environment is reliant on a diverse range of factors,
   rather than being exclusively governed by α-KG. Indeed, HIF1α levels
   and nuclear activities in tissue lysates harvested from the test
   animals after α-KG supplement failed to reveal any notable differences
   as compared to the vehicle-treated animals (Supplementary
   Fig. [406]16h-i). A comparable observation was also noted in
   fibroblasts from A-T patients following chronic α-KG supplement
   (Supplementary Fig. [407]16i). These indeed were likely attributable to
   the active integration of exogenous α-KG into the cellular metabolic
   network, rather than being channeled to activate the PHDs that support
   HIF1α degradation. This notion becomes evident when both isocitrate
   dehydrogenase-1 (IDH1) (Inhibitor: Compound 13/IDH1i)^[408]128 and
   aspartate aminotransferase (GOT1) (Inhibitor: Compound 2c)^[409]129—the
   2 pivotal cytosolic enzymes that facilitate α-KG participation into the
   metabolic network—were inhibited, which resulted in reductions of HIF1α
   nuclear activities (Supplementary Fig. [410]16i).

   In A-T, the progressive degeneration of cerebellar Purkinje neurons and
   the consequent manifestation of ataxia delineate the most severe facets
   of the pathological condition. Purkinje cells are distinguished by
   their substantial physical dimensions, GABAergic properties, and
   intricate synaptic interconnections, which all these factors attribute
   to their elevated metabolic requisites relative to other neuronal
   populations within the brain^[411]130. Consequently, they are notably
   predisposed to perturbations in fuel metabolism^[412]67. Consistent
   with prior research^[413]131, the examination of spatial and temporal
   expression levels of ATM and INSR throughout the entire human lifespan
   has revealed their heightened levels within the cerebellar cortex as
   compared to other brain regions, starting from the late embryonic stage
   (i.e., 200 days postnatally). Our current study underscores the insulin
   sensitivity of Purkinje neurons and highlights that ALDOC-positive
   Purkinje cells, primarily localized anatomically in the posterior and
   flocculonodular regions of the vermal cerebellum, exhibit increased
   vulnerability in the absence of ATM. Damage to these specific
   cerebellar areas is commonly associated with postural ataxia and a
   range of impairments in visual tracking, oculomotor control, spatial
   cognition, and language skills frequently reported in A-T
   patients^[414]132. ALDOC, a member of the fructose-bisphosphate
   aldolase enzyme family akin to the common ALDOA and ALDOB^[415]133, is
   linked to enhanced glycolytic capacity and dependence upon ectopic
   overexpression, which is consistent with the transcriptome profiles of
   ALDOC-positive Purkinje cells. Furthermore, our findings suggest that
   chronic and sustained actions of insulin, a physiological endocrine
   change that occurs during mid-childhood and puberty^[416]83, could
   induce ALDOC expression via HIF1α, suggesting it could be an important
   contributor to cerebellar maturation and development during these life
   stages. A malfunctioning insulin-dependent metabolic network in
   ALDOC-positive Purkinje cells may therefore play a role in amplifying
   their susceptibility. This vulnerability in selective cerebellar
   regions in part elucidates how the ataxic phenotype but not cognitive
   or psychiatric impairments, emerges as a dominant symptom stemming from
   cerebellar degeneration in A-T.

   These brain-specific observations, alongside their association with the
   broader systemic changes, suggest that a metabolic-cantered strategy
   could potentially ameliorate or prevent certain symptoms, contingent
   upon the timing of intervention. Our investigation supports that the
   early-life introduction of dietary supplementation with α-KG as an
   innovative approach to enhance growth, bolster lean muscle mass, and
   uphold insulin sensitivity in the context of ATM deficiency. α-KG,
   situated at the nexus between carbon and nitrogen metabolism within the
   TCA cycle, targets the primary sites of metabolic dysregulation. Given
   the convergence of α-KG metabolism and that of glutamine, as well as
   the heightened presence of cell surface receptors for α-KG in A-T
   suggest a potential preference for α-KG in Atm-deficient cells and
   tissues. Moreover, the metabolism of α-KG also circumvents the
   requirement for glutamine deamination, a process that may result in
   neurotoxic ammonium accumulation in ATM-deficient cells. Instead, α-KG
   metabolism likely achieves nitrogen equilibrium, via acting as a
   principal amino-group receptor in the transamination reactions, the
   first catabolic step of most amino acids^[417]134. With reference to
   previous reports showing its capacity to enhance longevity in aged mice
   and mitigate systemic inflammation while devoid of documented adverse
   outcomes^[418]109, α-KG emerges as a promising metabolite with
   potential therapeutic implications. Prolonged α-KG supplementation has
   exhibited favorable outcomes on insulin sensitivity in obese rats
   through the augmentation of nitric oxide production in endothelial
   cells and the inhibition of hepatic gluconeogenesis via diverse
   pathways^[419]110,[420]135, which may explain how glucose intolerance
   is improved in A-T mice after α-KG supplementation. Future
   investigations should delve into whether these effects of α-KG can
   mitigate the peripheral metabolic complications linked to ATM
   deficiency and conceivably in other conditions characterized by similar
   metabolic perturbations.

   Collectively, our findings emphasize the central importance of ATM
   activation as a previously unrecognized key regulator of both cellular
   catabolism and anabolism. ATM functions as a finely tuned switch that
   facilitates the communication between the endocrine insulin cues and
   the intricate intracellular metabolic framework. This work provides
   novel mechanistic understandings regarding how disturbances in
   homeostatic insulin action could result in either over- or under-active
   ATM responses, both of which may trigger persistent metabolic
   irregularities. Moreover, it also suggests that peripheral metabolic
   deficiencies is an overlooked contributing risk factor accelerating
   sustained neurodegenerative changes and ataxia in A-T. Our results also
   indicate that α-KG could be considered as a potential therapeutic
   target for managing human A-T and other disorders characterized by
   similar metabolic challenges. It is also noted that while the
   widespread metabolic perturbations inherent to the disease were
   consistently observed across various tissues sampled from the Atm-KO
   mouse model, it is crucial to recognize that such analyses may have
   limitations in fully elucidating possible metabolic differences that
   could have pre-existed among distinct cell subtypes (e.g., neurons
   versus glia) located within different regions of the tissue.
   Furthermore, the clarity on whether a dosage-dependent effect exists,
   particularly in patients with mutations leading to residual ATM
   proteins, remains uncertain^[421]136. Addressing these nuances will
   necessitate dedicated investigations in future studies.

Methods

Reagents, antibodies, open reading frame plasmids

   Unless otherwise specified, all chemicals and reagents were purchased
   from Sigma-Aldrich. Details of antibodies, special reagents, assay
   kits, sequence-based reagents and analytical software, so as sequences
   for oligos and a list of unique reagents are provided in Supplementary
   Table [422]1. Unique reagents generated in this study will be made
   available upon reasonable request to the lead contact with a completed
   Institutional Materials Transfer Agreement.

Animal maintenance, tissue harvest and primary cortical neuronal culture

Animal maintenance and sample characteristics

   For all experiments, unless otherwise stated, no inclusion or exclusion
   criteria were applied other than genotype, HOMR-IR status and age.
   Animals were housed under a 12 h light/dark cycle at room temperature
   (22  ±  2 °C) and constant humidity levels at around 50–70%, with food
   and water provided ad libitum in a specific pathogen free (SPF)
   environment. Within these criteria, were randomly chosen among the
   available colonies of both sexes. B6;129S4-Atmtm1Bal/J (Atm^+/+,
   Atm + /− or Atm^−/−) mice were obtained from The Jackson Laboratory.
   They were maintained and bred in the Laboratory Animal Service Centre
   of the Chinese University of Hong Kong. Heterozygous
   B6;129S4-Atmtm1Bal/J were crossbred to produce Atm^-/- (i.e., Atm-KO)
   for the study. The Atm^+/+ (i.e., Atm-WT) littermates were used as
   controls for all experiments. All in vivo physiological and behavioral
   experiments and ex vivo tissue culture experiments involved subjects
   from both sexes to ensure that any potential existence of sex-biased
   effect would be observed. An exception arose in the context of the
   study involving STZ injection. It is noteworthy that STZ, a
   pharmacological agent utilized to induce diabetes in laboratory mice,
   as documented in the literature^[423]25, exhibits differential
   sensitivity in female mice due to potential protective effects
   conferred by estrogen^[424]137. Consequently, owing to the inherent
   complexities introduced by the estrous cycle in fertile female mice and
   the disparate STZ dosages required to elicit pancreatic β-cell toxicity
   (with female subjects typically necessitating higher STZ doses than
   their male counterparts), historical precedents have favored the
   utilization of male animals in STZ-induced diabetic mouse
   studies^[425]138,[426]139. In alignment with this sex-biased
   experimental paradigm, we also adopted a similar approach for this
   assay to ensure consistency and comparability with existing literature
   and methodologies. For the in vitro primary cortical neuronal-based
   analysis, cells were mixed cultured from tissues of both sexes. All
   protocols were approved by the Animal Ethics Experimentation Committee
   (AEEC) at CUHK (and their care was in accordance with the institutional
   and Hong Kong guidelines) (Ref.: 20-012-GRF; 20-104-NSF; 23-064-NSF).

Brain tissue harvesting

   Brain tissue isolation was performed in adult mice unless otherwise
   specified. Mice were first anesthetized by an intraperitoneal
   administration of 1.25% (vol/vol) avertin at a dosage of 30 ml per kg
   body weight. The heart of each mouse was then exposed, the left chamber
   was catheterized, and the right atrium was opened. Chilled
   physiological saline was perfused transcardially for 3 min to remove
   blood from the body. After perfusion, the cranial bones were opened,
   cerebral cortex and cerebellar tissues of the whole brain were
   collected. Other non-brain tissues, including multiple fat pads,
   hindlimb skeletal muscles and liver were also collected, snap-frozen in
   liquid nitrogen and stored at −80 °C before use.

Cortical neuronal culture

   Embryonic day 16 embryos of both sexes were collected in ice-cold
   PBS-glucose, and the cortical lobes were dissected out. Meninges were
   removed under a dissection microscope, and the cortices were placed in
   1× trypsin solution and placed in DMEM in 10% (vol/vol) fetal bovine
   serum to inactivate the trypsin, followed by transfer to neurobasal
   media supplemented with B-27, penicillin–streptomycin (1×) and
   L-glutamine (2 mM; GlutaMAX, Invitrogen). Tissue was triturated ten
   times through a wide-opening 5 ml pipette and allowed to settle to the
   bottom of a 15 ml conical tube. Cells that remained afloat in solution
   were retained, while pellets at the bottom were removed. Surviving
   cells were counted with trypan blue to identify dead cells and were
   plated on poly-L-lysine-coated (0.05 mg ml–1) glass coverslips. Unless
   otherwise specified, cells were plated in 24-well plates at 50,000
   cells per well and allowed to mature for over 7–10 days in vitro before
   transfection or lentiviral transduction, or over DIV14 before drug
   treatment experiments. Samples were randomly chosen for different
   treatments.

Human patient fibroblast culture

   Lines of primary human skin fibroblasts from A-T patients and their
   respective controls were obtained from the NIA Aging Cell Repository of
   the Coriell Institute for Medical Research and maintained in DMEM/10%
   FBS.

Lentivirus production and transduction

   Human embryonic kidney 293FT cells (Invitrogen) were transfected using
   Lipofectamine 2000 (Invitrogen) with the expression of two helper
   plasmids: psPAX2 and pMD2.G. 10 µg transfer vector, 5 µg pMD2.G, and
   5 µg psPAX2 of DNA were used per 10 cm plate. 48 h after transfection,
   the supernatants of four plates were pooled, centrifuged at 780 g for
   5 min, filtered through a 0.45 µm pore size filter, and further
   centrifuged at 76,000 × g for 2 h. The resulting pellet was resuspended
   in 100 µl of PBS. Lentivirus titration was performed with a p24 ELISA
   (Clontech).

IGTT and plasma insulin test

   Mice were fasted for 16 h before intraperitoneal injection with
   2 mg per kg body weight of glucose. Blood samples were taken from the
   tail vein at 0, 20, 40, 60, 80, 100 and 120 min. Glucose levels were
   measured using an Accu-Chek glucose meter (Roche Diagnostics). Plasma
   insulin levels were measured via ELISA assays using an ultrasensitive
   mouse insulin kit (Crystal Chem). The HOMA-IR index was calculated
   using the following formula: fasting plasma glucose
   (mmol l^–1) × fasting insulin (mIU l^–1)/22.5.

Collection of CSF from the cisterna magna followed by CSF insulin test

   CSF was collected from the cisterna magna of mice before and after the
   IGTT. Glass capillary tubes (Sutter Instrument; borosilicate glass
   B100-75-10) were prepared on a Sutter P-87 flaming micropipette puller
   (heat box set at 300 and the pressure index set at 330) and trimmed so
   that the inner diameters of the tapered tips were about 0.5 mm.

   Mice were anesthetized by intraperitoneal administration of 1.25%
   (vol/vol) avertin at a dosage of 30 ml per kg body weight. For each
   individual mouse, the skin near the neck was first shaved, and then the
   body was placed prone on the stereotaxic instrument with direct contact
   to a heating pad. Once the head was secured with the head adaptors, the
   surgical site was swabbed with 10% povidone iodine, followed by 70%
   ethanol, and a sagittal incision of the skin was made inferior to the
   occiput. Under a dissection microscope, subcutaneous tissue and muscles
   (m. biventer cervicis and m. rectus capitis dorsalis major) were
   separated by blunt dissection with forceps. Then, the mouse was laid
   down so that the head was at a nearly 135° angle to the body. Under the
   dissection microscope, the dura mater of the cisterna was blotted dry
   with a sterile cotton swab and penetrated with a capillary tube to
   reach the cisterna magna. When a notable change in the resistance to
   the capillary tube occurred following insertion, the CSF was collected
   into the capillary tube.

   The capillary tube was carefully removed, and CSF was ejected from the
   capillary tube with a syringe into a 1.5-ml tube, and frozen
   immediately on dry ice until further assays. After CSF sampling,
   muscles were realigned, and the skin was sutured. About 1 ml of 0.9%
   NaCl was injected to prevent dehydration. The mice were kept in a 37 °C
   incubator until full recovery. At 3 h after full recovery, mice were
   subjected to the IGTT as described above, and an additional round of
   CSF sampling was performed immediately after the IGTT. Once CSF samples
   were collected, mice were immediately sacrificed for collection of
   other tissue samples. Finally, CSF insulin levels were measured by
   ELISA using an ultrasensitive mouse insulin kit following the
   manufacturer’s protocol (Crystal Chem).

Behavioral tests

   All mice were individually housed. All behavioral tests were performed
   during the dark phase of the circadian cycle between 19:00 and 23:00.
   All behavioral testing began by allowing the mice to habituate in the
   test rooms for 2 h before any tests. Experiments were performed blinded
   to the age and the HOMA-IR status. Using an overhead camera,
   experimental mice were also subjected to an open-field test and a
   rotarod test.

Open-field test

   The open-field test was used for evaluation of anxiety and locomotion.
   Rodents show distinct aversions to large, brightly lit, open and
   unknown environments. It is assumed that they have been
   phylogenetically conditioned to see these types of environments as
   dangerous. In the experiments, mice were placed in the center of an
   open-field arena 50 cm (length) × 50 cm (width) × 38 cm (height) that
   was made from white high-density and non-porous plastic. Free and
   uninterrupted movement of the mouse was allowed for 5 min and movements
   were video taped. Locomotor activity was measured using the number of
   crossed grids, while exploratory activity was measured using the number
   of rearing on the hind feet. The total travel distance and time spent
   in the outer versus the inner zone areas of the field were computed
   using a Smart 3.0 video tracking system (Panlab, Harvard Apparatus).

Rotarod test

   To assess the acquisition of skilled behavior in mice, we first
   modified the standard rotarod test to emphasize the learning aspect of
   the test and minimize other factors. A rotarod machine with automatic
   timers and falling sensors (MK-660D, Muromachi-Kikai) was used. The
   mouse was placed on a drum. Before training sessions, the mice were
   habituated for 1 min immediately before the session. Animals were given
   two training trials (inter-trial interval: 2 h) with the rotarod
   adjusted to accelerate from 4 r.p.m. to 40 r.p.m. over a 5 min period
   each day. Latency to fall was measured. After a week of training, mice
   were tested using the rotarod adjusted to maintain a constant speed of
   15 r.p.m. for the entire 5 min test period. The latency of the mice to
   fall off the rod or take one revolution was measured. Trials were
   repeated four times with inter-trial intervals of 30 min over a single
   day.

In vivo drug treatment paradigms

In vivo streptozotocin (STZ) ± glargine treatment paradigm

   At around 5 days prior to initiating the experiment, four independent
   cohorts of male littermate Atm-WT or Atm-KO mice were obtained from the
   breeding program as mentioned above and they were kept on standard
   laboratory diet (Teklad 2918) briefly during the post-weaning period
   until reaching 1 month old. A repeated (5x) low doses of STZ paradigm
   was then introduced. Before all STZ injections, food was refrained from
   animals for 4 h. At time prior to injection, STZ was freshly prepared
   in 50 mM sodium citrate buffer (pH 4.5) to a final concentration of
   4 mg/ml. For each mouse, STZ working solution was injected
   intraperitoneally at 40 mg/kg (1.0 ml/100 g) in all experimental
   animals for 5 consecutive days. After each injection, mice were allowed
   to free access food and 10% sucrose water for recovery; except on day 6
   and onwards, 10% sucrose water was replaced back as regular water. On
   Day 6, glargine (50 Unit/kg body weight, s.c. injection, once a day)
   was performed for the upcoming 14 days until P49^[427]140,[428]141.
   Both IGTT and serum triglycerides were evaluated at time before STZ
   administration (P19) as well as on P49 after the entire drug treatment
   paradigm. Indirect calorimetry assessment was performed for 24 h after
   the dose of glargine injection on P49 prior tissue harvesting for other
   experiments.

In vivo Glargine ± TEPP-46 treatment paradigm

   This part was performed in accordance with a previous published study
   with slight modifications^[429]140,[430]141. Independent cohorts of
   littermate Atm-WT or Atm-KO mice of both sexes were obtained from the
   breeding program as mentioned above and they were kept on a standard
   laboratory diet (Teklad 2918) until 6 months old. At this age, mice
   were treated with either saline or a long- and slow-acting insulin
   reagent, glargine (50 Unit/kg body weight, s.c. injection, once a day),
   for 4 weeks. One glargine-treated WT group was co-administered with
   TEPP-46 (50 mg/kg body weight, i.p. injection, once a day) as
   well^[431]93. Mice were housed individually at the regular 20–22 °C,
   allowed to feed on the regular chow diet (Teklad 2918) ad libitum, and
   kept at 12 h light/dark cycles. Before the treatment starts, IGTT test
   on blood glucose level and plasma insulin measurement was performed
   after an overnight (16 h) fast with a glucose meter (Abbott Diabetes
   Care, Inc.) as to ensure similar baseline values in subjects within the
   Atm-WT or Atm-KO group. All mice were inspected daily to monitor their
   general health conditions, body weight change and food consumption.

In vivo 2% (w/w) CaAKG feeding paradigm

   Independent cohorts of littermate Atm-WT or Atm-KO mice were obtained
   from the breeding program as mentioned above until weaning at
   post-natal day 25 (P25). All mice were housed on a 12 h light/dark
   cycle and kept at 20-22°C. Treatment was started right after weaning.

   CaAKG-treated animals were subjected to a 2-month (pilot study) or
   6-month long 2% (w/w) CaAKG supplement on the regular mouse diet
   (Teklad 2918 Irradiated 18% protein and 6% fat diet) while the control
   groups were kept on the standard 2918 diet. Pure calcium 2-oxoglutarate
   (Carbosynth Company) was homogeneously mixed during manufacturing of
   the 2918 diet prior to irradiation and pelleting. The exact starting
   sample size could be found in the survival data. Mice were housed
   individually for food consumption recording and to prevent fighting and
   injuries. All lifespan and health span-related experiments were
   initiated at around 7 months of age in the 6-month long study. Baseline
   parameters such as body weight, sizes and glucose tolerance were
   evaluated, prior a fair partitioning of mice was done into different
   groups, i.e. for any given mouse in any given group, there are similar
   mice of both sexes in all other groups. This allows any outcome of the
   study to be more related to experiments or the treatment rather than
   the inherent property of a group. Mice were then inspected daily to
   monitor their general health conditions, body weight change and food
   consumption.

In vivo intraperitoneal (i.p.) administration of 2% (w/w) CaAKG treatment
paradigm

   The volume of injection was the lowest volume possible and not exceeded
   the current recommended guidelines [i.e., Needle Gauge: 25–27 g;
   Volume: < 10 ml/kg (for a 25 g mouse, maximum volume would be
   0.25 ml)]. Animals were gently removed from the cage and restrain
   appropriately in the head-down position. A daily injection
   (10 mg/kg/day) was performed at the lower right quadrant of the abdomen
   to avoid damage to the urinary bladder, cecum and other abdominal
   organs. Since the entire injection program had lasted for 60 days,
   injection was performed on alterative side of the lower abdomen
   differed from the side injected on the day before. Once after
   injection, all animals were placed back into their corresponding cages
   and observe for any complications (e.g. bleeding at injection site).

Indirect calorimetry

   Mice after undergoing through various in vivo treatment paradigms at
   various ages were individually placed in the registration chambers of
   the calorimetry system and allowed to adapt for 24 h prior the
   recording. After adaptation, the volume of carbon dioxide production
   (VCO[2]) and the volume of oxygen consumption (VO[2]) were recorded for
   at least 24 h using the Oxylet calorimeter system (Pan Lab/Harvard
   Instruments). The analysis of the respiratory exchange ratio (RER) was
   performed using the Metabolism software (Pan Lab/Harvard
   Instruments)^[432]142.

Acute cerebellar brain slice culture

   Mouse cerebellar brain slices were prepared in coronal orientation as
   previously reported with slight modifications^[433]143. With freshly
   harvested brain tissues from mice at different ages (For details,
   please refer to different figure panels), tissues were trimmed and that
   the top part of the cerebellum was discarded. Both hemispheres were
   then glued on the chuck with the median side facing up. Slices were
   then cut (0.4 mm) using a Vibroslice (VT1000S, Leica) in an ice-cold
   solution containing 64 mM NaCl, 2.5 mM KCl, 1.25 mM NaH[2]PO[4], 10 mM
   MgSO[4], 0.5 mM CaCl[2], 26 mM NaHCO[3], 10 mM glucose, and 120 mM
   sucrose. Slices were allowed to recover for 60 min prior initiation of
   acute drug treatments for whereas time points in artificial
   cerebrospinal fluid (aCSF) containing 126 mM NaCl, 2.5 mM KCl, 1.25 mM
   NaH[2]PO[4], 2 mM MgSO[4], 2 mM CaCl[2], 26 mM NaHCO[3], and 10 mM
   glucose. All solutions were saturated with 95% O[2]/5% CO[2]
   (volume/volume). For experiments where glargine and or TEPP-46 were
   treated to the slices, 100 nM glargine and 100 nM TEPP-46 was used (For
   details, please refer to different figure panels).

Frozen sections preparation, Immunofluorescence histochemistry and Oil Red O
staining

   10 µm cryo-sections of frozen mouse brains or livers were used.

Immunohistochemistry

   This was performed in accordance to a previously published study with
   slight modifications^[434]144. Samples were fixed with fresh 4%
   (wt/vol) paraformaldehyde for 10 min, washed and followed by
   permeabilization with 0.3% Triton-X100 in PBS for 10 min. After
   blocking with 5% (wt/vol) BSA in PBS for 1 h, primary antibodies were
   added and incubated overnight at 4 °C. The following day, coverslips
   were washed three times (10 min each) with PBS. After rinsing,
   secondary antibodies were applied for 1 h at room temperature followed
   by three additional washes with PBS. The coverslips were then inverted
   and mounted on glass slides with ProLong Gold Antifade Reagent (Life
   Technologies). All samples were examined and imaged on a upright
   fluorescent microscope (Nikon Ni-U) equipped with 10x and 20x
   objectives (CFI Plan Apochromat Lambda series) and Nikon DS-QI2 camera.
   Image acquisition and analysis were performed on the NIS-Elements
   software (Nikon). For visualizing the whole cerebellar structure,
   sequential images were stitched pairwise using the linear blending
   algorithm in Image J^[435]145.

Oil red O staining

   0.7% Oil Red O stock solution (m/v in absolute isopropanol) was
   prepared by diluting the stock solution with distilled water to obtain
   the working solution. Cryo-sections of frozen mouse livers were washed
   with 60% isopropanol (v/v in PBS) and stained in Oil Red O working
   solution for 10 min. Subsequently, stained sections were washed with
   60% isopropanol (v/v in PBS) to clear the background noise signals. All
   samples were examined and imaged on the brightfield function of the
   Nikon Ni-U upright fluorescent microscope equipped with 20x objective
   (CFI Plan Apochromat Lambda series) and Nikon DS-QI2 camera. The images
   were first bandpass-filtered to sharpen positive droplet-like signals
   for easier analysis, followed by automatic image thresholding and
   particle analysis in Image J.

Subcellular protein fractionation, co-immunoprecipitation, SDS–PAGE and
western blotting

   For nuclear and cytoplasmic fractionation, freshly harvested brain
   tissues or cell platelets were used. A subcellular protein
   fractionation kit for tissues (ThermoFisher) was used following the
   manufacturer’s protocol. For whole cell lysates, frozen or freshly
   harvested samples were homogenized in RIPA buffer (Millipore) with 1×
   complete protease inhibitor mixture (Roche) and 1× PhosSTOP phosphatase
   inhibitor mixture (Roche) on ice and centrifuged for 10 min at
   18,400 × g to remove large debris. The protein concentration of the
   supernatant was determined using a Bradford assay (Bio-Rad). For
   co-immunoprecipitation, 1 mg of the total cell lysate was first
   incubated with control IgG (Santa Cruz Biotechnology) for 30 min,
   precleared with 50 µl of Dynabeads Protein G (Invitrogen) and then
   incubated with various antibodies according to the suggested dilutions
   on the product datasheets overnight at 4 °C. Beads bound with immune
   complexes were collected using DynaMag-2 (Life Technologies) and washed
   three times before elution in 90 µl of buffer containing 0.2 M
   glycine-HCl, pH 2.5, which was neutralized with 10 µl of neutralization
   buffer (1 M Tris-HCl, pH 9.0). The eluates were subjected to 9–15%
   SDS–PAGE and western blot analysis.

   For western blotting, 100 µg of proteins derived of total cell lysates
   or subcellular fractionated lysates were separated by SDS–PAGE and
   transferred to polyvinylidene difluoride membranes. For the
   cross-linking experiments, cells were washed with ice-cold PBS three
   times and treated with 5 mM disuccinimidyl suberate (DSS, A39267,
   ThermoScientific, Waltham, MA) for 30 min at room temperature. The
   cross-linking reaction was stopped by adding the quenching solution
   (1 M Tris, PH 7.5) to the final concentration of 20 mM for 15
   min^[436]146. Then, cell lysates were used for WB. Following blocking,
   membranes were probed with various primary antibodies to determine
   different levels of protein expression. Immunoreactive antibody–antigen
   complexes were visualized using the SuperSignal West Femto
   Chemiluminescent Substrate (ThermoFisher). For the detection of
   dimerized ATM, proteins were denatured by the addition of 2% SDS
   without boiling and separated under nonreducing conditions at
   4 °C^[437]147. Full scan blots can be found in Source Data file.

Pyruvate kinase colorimetry assay

   This assay was performed according to the manufacturer’s protocol
   (Abcam). In brief, fresh tissues or cells were first extracted with 4
   volumes of assay buffer, centrifuged to obtain clear extract and
   assayed immediately. For the standard curve, pyruvate standard was
   diluted to 1 nmol/μl and a serial dilution was prepared. For reaction
   mix preparation, the assay buffer, substrate mix, enzyme mix and
   OxiRed^TM Probe were mixed and added to standard- or sample-containing
   wells, followed by colorimetry measurement at OD[570nm] for multiple
   time points. Pyruvate kinase (PK) activity was calculated as:
   [MATH: <mi mathvariant="normal">PK</mi><mspace
   width="0.16em"></mspace><mi
   mathvariant="normal">activity</mi><mo>=</mo><mrow><mo>(</mo><mrow><mi
   mathvariant="normal">B</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">x</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">Sample</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">Dilution</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">Factor</mi></mrow><mo>)</mo></mrow><mo>/</mo><mrow
   ><mo>[</mo><mrow><mrow><mo>(</mo><mrow><mi
   mathvariant="normal">T</mi><mn>2</mn><mo>−</mo><mi
   mathvariant="normal">T</mi><mn>1</mn></mrow><mo>)</mo></mrow><mspace
   width="0.25em"></mspace><mi mathvariant="normal">x</mi><mspace
   width="0.16em"></mspace><mi
   mathvariant="normal">V</mi></mrow><mo>]</mo></mrow><mrow><mo>(</mo><mro
   w><mi mathvariant="normal">in</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">nmol</mi><mo>/</mo><mi>min</mi><mo>/</mo><mi
   mathvariant="normal">ml</mi><mo>=</mo><mi
   mathvariant="normal">mU</mi><mo>/</mo><mi
   mathvariant="normal">mL</mi></mrow><mo>)</mo></mrow> :MATH]

   Where:

   B is the pyruvate amount from pyruvate standard curve (in nmol).

   T[1] is the time of the first reading (A[1]) (in min)

   T[2] is the time of the second reading (A[2]) (in min)

   V is the sample volume added into the reaction well (in mL).

   Unit definition: One unit of Pyruvate Kinase is the amount of enzyme
   that will transfer a phosphate group of PEP to ADP; yielding 1.0 μmol
   of pyruvate per minute at 25°C.

HIF1α transcription factor DNA binding activity assay

   This enzyme linked immunosorbent assay (ELISA) served as a replacement
   of the radioactive electrophoretic mobility shift assay (EMSA) and that
   the assay was performed by strictly adhering to the manufacturer’s
   protocol (Abcam).

Tissue ammonia colorimetry assay

   Ammonia Assay Kit purchased from Abcam was used for the estimation of
   ammonia content in cerebellar tissues collected. The assay was
   performed by strictly adhering to the manufacturer’s protocol.

Triglyceride (TG) ELISA assay

   Triglyceride (TG) ELISA Kit purchased from Sigma was used for the in
   vitro quantitative measurement of triglyceride concentrations in serum
   collected. The assay was performed by strictly adhering to the
   manufacturer’s protocol.

Live imaging of cytoplasmic redox status with Cyto-roGFP or CellROX dye

   Cyto-roGFP obtained from Addgene (plasmid no. 49435) senses redox
   changes in a cell^[438]148. The Cyto-roGFP biosensor was transfected
   into cortical neurons on DIV 9, and 24 h after transfection, neurons
   were exposed to different treatment conditions in neurobasal culture
   medium. Cultures were then imaged for 24 h at hourly intervals in a 95%
   air/5% CO2-gassed incubator using a Leica TCS SP8 confocal laser
   scanning platform, equipped with Leica HyD hybrid detector and
   visualized through a HC PL APO CS2 63 × (1.40 NA) oil-immersion
   objective. Image acquisition was controlled by LAS X. The
   redox-sensitive protein reporter has excitation maximum at 400 ± 15 nm
   and 484 ± 15 nm and an emission maximum at 525 ± 15 nm. The relative
   amplitudes of these peaks depend on the state of oxidation. With
   increased oxidation the 400 ± 15-nm excitation peak increases, while
   the 484 ± 15-nm peak decreases^[439]149. Importantly, the ratiometric
   nature of the analysis renders the results independent of the
   expression levels of the plasmid in any one cell. Data were collected
   with the Leica Application Suite X Microscope Software. The
   fluorescence excitation ratios were obtained by dividing integrated
   intensities obtained from manually selected portions of the imaged
   regions of intact whole cells collected using 400 ± 15-nm and
   480 ± 15 nm excitation filters after appropriate background correction.
   Background correction was performed by subtracting the intensity of a
   nearby cell-free region from the signal of the imaged cell.
   Alternatively cytoplasmic redox state was determined by the CellROX
   dye, which was used according to the manufacturer’s protocol.

Chromatin immunoprecipitation (ChIP)-PCR assay

   A total of 5 × 10^6 A-T patient fibroblasts (Fig. [440]3k) or acute
   cerebellar slice culture treated with or without 100 nM glargine for
   various time courses were cross-linked with 3.7% formaldehyde (Sigma)
   at room temperature for 10 min. Samples were incubated with 0.125 M
   glycine to terminate cross-linking, washed twice with PBS and lysed
   with SDS nuclear lysis buffer (1% SDS, 10 mM EDTA and 50 mM Tris-HCl,
   pH 8.1) for 10 min on ice. Sonicated lysates were diluted in ChIP
   dilution buffer (0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 167 mM NaCl
   and 16.7 mM Tris-HCl, pH 8.1) and incubated with 10 µg of rabbit IgG
   (Santa Cruz Biotechnology) and Dynabeads Protein G (Invitrogen)
   overnight at 4 °C with gentle shaking. The cleared supernatants were
   mixed with either 20 µl of c-MYC antibody for Fig. [441]3k, anti-HIF1α
   antibody for Fig. [442]4r or with pre-immune rabbit IgG as negative
   controls for both overnight at 4 °C. Antibody–protein–DNA complexes
   were co-precipitated with Dynabeads Protein G. Protein–DNA conjugates
   were eluted from the bead complexes with elution buffer (100 mM NaHCO3
   and 1% SDS) for 30 min. Cross-links were reversed in 5 M NaCl. RNA and
   protein were removed by incubation first with 10 µg DNase-free RNase-A
   at 37 °C for 1 h, and then with 20 µg proteinase K (Sigma) at 50 °C for
   4 h. DNA was recovered by phenol–chloroform extraction and ethanol
   precipitation. A DNA fragment encompassing the c-MYC binding region of
   the human GLS2 promoter or HIF1α binding region of mouse Aldoc promotor
   was amplified using 35 cycles of PCR at 94 °C for 30 s, 55 °C for 30 s
   and 72 °C for 30 s. All amplified products were resolved on a 1.5%
   agarose gel. Primers are listed in Supplementary Table [443]1. Full
   scan blots can be found in Source Data file.

RNA immunoprecipitation (RIP)

   RIP was performed using the Magna RIP^TM RNA-binding protein
   immunoprecipitation kit (Millipore) following strictly the
   manufacturer’s guidelines. Human patient fibroblasts were treated with
   ice-cold PBS and Pierce^TM IP lysis buffer (ThermoFisher) supplemented
   with RNase inhibitors and a protease inhibitor cocktail. The
   supernatants were incubated with the anti-PKM2 antibody or control
   rabbit IgG with rotation at 4 °C overnight before further incubation
   with protein G bead for another night at the same temperature. After
   that protein-RNA precipitates were enriched and bounded RNAs were
   extracted by the proteinase K-chloroform method. Enrichment of the
   internal ribosome entry site region of the MYC mRNA was performed by
   reverse transcription followed by RT-PCR.

qPCR with reverse transcription analysis

   Total cellular RNA was purified from cerebellar tissues or cultured
   cells using a RNeasy mini kit (Qiagen) following the manufacturer’s
   protocol. For real-time qPCR, RNA was reverse-transcribed using a
   High-Capacity cDNA Reverse Transcription kit (Applied Biosystems)
   according to the manufacturer’s instructions. The resulting
   complementary DNA was analysed by qPCR with reverse transcription using
   SYBR Green PCR Master Mix (Applied Biosystems). All reactions were run
   on a LightCycler 480 Instrument II (Roche Diagnostics) with a 15 min
   hot start at 95 °C followed by 40 cycles of a 3-step thermocycling
   program: denaturation step of 15 s at 94 °C; annealing step of 30 s at
   55 °C; and extension step of 30 s at 70 °C. Melting curve analysis was
   performed at the end of every run to ensure that a single PCR product
   of the expected melting temperature was produced at a given well. A
   total of seven to nine biological replicates with four technical
   replicates were performed for each treatment group. Data were analysed
   using the comparative Ct method (ΔΔCt method). The primers (5′ to 3′)
   used were designed on Primer-Blast and are listed in Supplementary
   Table [444]1.

Conventional cell line and patient fibroblast culture, transfection and live
imaging

   Lines of primary human skin fibroblasts from A-T patients and their
   respective controls were obtained from the NIA Aging Cell Repository of
   the Coriell Institute for Medical Research and maintained in DMEM/10%
   FBS. Immortalized mouse hippocampal cells (HT22 cell line, ThermoFisher
   Scientific) were cultured in DMEM/10% (vol/vol) fetal bovine
   serum/penicillin–streptomycin medium (Gibco). Primary cortical neurons
   were isolated from E16 embryos of C57BL/6 J or Atm + /− (Atm-KO)
   pregnant mice and cultured as described above. DNA constructs were
   transfected with Lipofectamine 2000 or Lipofectamine LTX with Plus
   Reagent into HT22 cells and primary cortical neuronal cultures,
   respectively. Following the manufacturer’s protocol, 6 h after
   transfection, cells were refreshed with culture medium and further
   incubated for 48–72 h to allow recovery and ectopic expression prior
   harvesting or any drug treatments.

   For primary cortical neurons that were transfected with either
   pEGFP-C1-PKM2 (A gift from Axel Ullrich; Addgene plasmid #64698) or
   pg-HIF-1alpha-EGFP (A gift from Violaine Sée; Addgene plasmid # 87204);
   time lapse confocal microscopy was performed afterwards in various
   treatment conditions, as indicated in the text. Cells were incubated on
   the microscope state at 37 °C, 5% CO[2] or 20% O[2] and observed by
   confocal microscope using a Leica Sp8 with a 63 × 1.3 NA oil immersion
   objective. Excitation of EGFP was performed using an argon ion laser at
   488 nm. Emitted light was detected through a 505–550 nm bandpass filter
   from a 545 nm dichroic mirror. Excitation of the empty
   AAV-Synapsin-mCherry-C1-WPRE (A gift from Michael Courtney; Addgene
   plasmid # 159956) used as a control was performed using a green
   helium-neon laser (543 nm) and detected through both a 545-nm dichroic
   mirror and a 560-nm long pass filter. Data capture was carried out with
   the LAS X Life Science Microscope software. These experiments were
   performed three times and ~45 cells were analyzed for each construct.

Label-free phosphoproteome analysis

Protein extraction, digestion and clean up

   Following the specified treatment protocols and duration, cerebellar
   tissues were promptly collected and flash-frozen in liquid nitrogen.
   All specimens were subsequently stored at –80 °C prior to
   cryo-pulverization. The EasyPep Mini MS Sample Prep Kit by ThermoFisher
   was utilized for the ensuing protein extraction processes, albeit with
   minor adaptations. To elucidate, 5 mg of tissue samples underwent
   homogenization in 500 µl of lysis solution containing 1 µl of universal
   nucleases using Dounce homogenizers. This was followed by
   centrifugation at 16,000 g for 10 min, resulting in the retrieval of
   the supernatant for protein quantification. Subsequently, 500 µg of
   protein lysates were adjusted to 500 µl with lysis solution and then
   combined with 250 µl of reduction solution and 250 µl of alkylation
   solution at 95 °C for 30 min. Upon the samples’ cooling, 250 µl of
   reconstituted enzyme solution, comprising trypsin/lysyl endopeptidase
   (LysC), was introduced to the reduced and alkylated protein samples.
   The mixture was then subjected to overnight incubation with agitation
   at 300 r.p.m. at 37 °C to facilitate protein digestion. Upon completion
   of the incubation period, 250 µl of Digestion Stop Solution was added
   to the sample, followed by gentle mixing. After the completion of the
   digestion process, 300 µl of the digested proteins were subsequently
   introduced to a dry Peptide Clean-up column, followed by centrifugation
   at 1500 g for 2 min, with the subsequent discarding of the flowthrough.
   This procedure was iterated until all the lysates had been loaded onto
   the column. Subsequently, the column underwent a washing process
   involving washing buffers A and B to remove impurities, before the
   enriched proteins were eluted. The eluted peptides were then subjected
   to drying using a SpeedVac system (ThermoScientific) and subsequently
   stored at -80 °C until further processing, or for phosphopeptide
   enrichment.

Phosphopeptide enrichment

   Phosphopeptides were extracted from the digested peptides using the
   High-Select TiO2 Phosphopeptide Enrichment Kit (ThermoFisher), with
   slight modifications. Initially, the lyophilized peptides were fully
   dissolved in 150 µl of binding/equilibration buffer (pH<3). The
   suspended peptide solution was then transferred to a TiO2 spin tip,
   with a maximum of 150 µl loaded per transfer, and centrifuged at 1000 g
   for 5 min. This loading process was repeated until all the suspended
   peptides had been transferred to the spin tip. Following this, the spin
   tip underwent a series of washing steps, involving the
   binding/equilibration buffer, wash buffer, and LC-MS grade water, to
   eliminate impurities and non-phosphorylated peptides, thereby
   concentrating the phosphopeptides for subsequent analyses.
   Subsequently, the phosphorylated peptides were eluted, speed dried
   using the SpeedVac system (ThermoScientific) and then stored at −80 °C
   until further investigations were conducted.

Liquid chromatography-mass spectrometry analysis

   Both the complete tryptic-digested peptides and the enriched
   phosphopeptides were introduced into a Vanquish Neo LC system
   (ThermoFisher) for the purpose of peptide separation. The peptides
   underwent separation employing a trap-and-elute methodology utilizing a
   20 mm × 75 µm trap column (ThermoFisher) and an Aurura Ultimate XT
   25 cm × 75 µm C18 column (Ionopticks). The temperature of the
   separation column was maintained at 50 °C using an XT compatible heater
   controller (Ionopticks, Australia). Solvent A and B were constituted of
   0.1% formic acid in milli-Q water and 0.1% formic acid in acetonitrile,
   respectively. The gradient settings, with a flow rate of 300 nl/min,
   were programmed as follows: a linear increase from 2% B to 6% B over
   0–2 min, followed by a gradient from 6% B to 30% B within 2–77 min, and
   subsequently from 30% B to 90% B during 79–82 min, maintaining this
   composition until 87 min. The gradient then decreased from 90% B to 2%
   B at 87.1 min and was sustained until 90 min.

   Subsequent to elution, peptides were directly introduced into an
   Orbitrap Fusion Lumos Mass Spectrometer (ThermoFisher), where mass
   spectra data were obtained employing a data-dependent acquisition (DDA)
   approach within a scan range of 400–1500 m/z. The spray voltage was
   maintained at 2 kV, accompanied by an ion transfer tube temperature of
   300 °C. For MS1 analysis, the resolution was set at 60,000, with an AGC
   target of 4.00E5 and a maximum injection time of 50 ms. Following this,
   MS2 spectra were acquired utilizing a 3 s cycling time and a 40 s
   dynamic exclusion period. The orbitrap resolution for MS2 was set at
   15,000, with HCD collision energy of 30%, an AGC target of 5.00E4, and
   a maximum injection time of 22 ms.

MaxQuant quantification

   After the experimental procedures, mass spectrometric data analysis was
   performed using MaxQuant, a specialized quantitative proteomics
   software tailored for the analysis of extensive datasets obtained
   through high-resolution MS, following the previously published workflow
   and procedures^[445]150. This software accommodates various labeling
   methodologies and supports label-free quantification techniques.
   MaxQuant, freely accessible for download from the specified source,
   incorporates the andromeda search engine for peptide identification,
   quantification, and the viewer application for the examination of raw
   data^[446]150. The raw data files were analysed using MaxQuant to
   obtain phosphosite identifications and their respective label-free
   quantification values using the following parameters. Raw data were
   analyzed using MaxQuant’s (version 2.6.6.0) (Cox & Mann, 2008)
   Andromeda search engine in reversed decoy mode based on a Mus musculus
   reference proteome (Uniprot-FASTA, UP000000589, downloaded November
   2024) with a false discovery rate (FDR) of 0.01 at both peptide and
   protein levels, and identification of post-translational modifications
   were specified to identify site-specific protein phosphorylation in the
   corresponding samples. Digestion parameters were set to specific
   digestion with trypsin with a maximum number of 2 missed cleavage sites
   and a minimum peptide length of 7. Oxidation of methionine,
   amino-terminal acetylation and phosphorylation sites at serine,
   threonine and tyrosine (STY) were set as variable modifications;
   carbamidomethylation of cysteine was set as fixed modification, with a
   maximum number of 5 modifications per peptide. The tolerance window was
   set to 20 ppm (first search) and to 4.5 ppm (main search). Label-free
   quantification was set to a minimum ratio count of 2, re-quantification
   and match-between-runs was selected and 4 biological replicates per
   condition were analyzed. Resulting raw output protein group files
   generated from MaxQuant were processed using Phospho-Analyst to perform
   differential expression analysis to visualize the results preprocessed
   with MaxQuant^[447]151. The original data is deposited on the PRIDE
   Archive under the project accession number PXD062018.

Identification of significant phosphosites using phospho-analyst

   Subsequently, MaxQuant output tables, which contain information about
   quantification of proteins and PTMs, were used as input to
   Phospho-Analyst online platform for quality control and post-hoc
   analyses
   ([448]https://analyst-suites.org/apps/phospho-analyst/)^[449]151. For
   the phosphosite report, the input data were normalized based on the
   assumption that the majority of phosphosites do not change between the
   different conditions. Contaminant phosphosites, reverse sequences and
   phosphosites identified “only by site” were filtered out. In addition,
   phosphosites with localization probability < 0.75 have been removed as
   well. The phosphosite data was converted to log2 scale, samples were
   grouped by conditions and missing values were imputed using “Miss not
   At Random” (MNAR) method, which uses random draws from a left-shifted
   Gaussian distribution of 1.8 standard deviation apart with a width of
   0.3. Protein-wise linear models combined with empirical Bayes
   statistics were used for the differential expression analyses. The
   limma package from R Bioconductor via the software was used to generate
   a list of differentially expressed phosphosites for each pair-wise
   comparison. A cutoff of adjusted p-value of 0.05 (Benjamini-Hochberg
   method) along with a |log2 fold change| of 1 has been applied to
   determine significantly regulated phosphosites in each pairwise
   comparison. Visualization of phosphorylation motif was performed on the
   pLogo platform ([450]https://plogo.uconn.edu/)^[451]152.

   For the proteinGroup report, the data from MaxQuant were normalized
   based on the assumption that the majority of proteins do not change
   between different conditions. Contaminant proteins, reverse sequences
   and proteins identified “only by site” were filtered out. In addition,
   proteins that have been only identified by a single peptide and
   proteins not identified/quantified consistently in the same condition
   have been removed as well. The LFQ data was converted to log2 scale,
   samples were grouped by conditions and missing values were imputed
   using the “Miss not At Random” (MNAR) method, as described above.
   Protein-wise linear models combined with empirical Bayes statistics
   were used for the differential expression analyses. The limma package
   from R Bioconductor via the software was used to generate a list of
   differentially expressed proteins for each pair-wise comparison. A
   cutoff of adjusted p-value of 0.05 (Benjamini-Hochberg method) along
   with a |log2 fold change| of 1 has been applied to determine
   significantly regulated phosphosites in each pairwise comparison.

In silico docking

   The mouse heterodimeric HIF1α: ARNT (HIF1β) complex structure was
   obtained from the Protein Data Bank (PDB: 4ZPR)^[452]153. According to
   that, the human HIF1α and HIF1β structures were modelled with
   SWISS-MODEL algorithm^[453]154. For the protein structures of human
   apoenzyme PKM2 (PDB: 3BJT)^[454]155 and importin-α3 (PDB:
   6BVZ)^[455]156, they were directly extracted from the Protein Data Bank
   accordingly. To simulate the phosphorylation sites predicted form the
   SILAC experiment, targeted phosphate groups were added to the HIF1α and
   PKM2 using PyTMs plugin of the PyMOL software^[456]157. With the
   structures ready, HADDOCK2.4 was used to simulate the strengths of
   protein-protein interactions and distance between interacting
   residues^[457]158.

   To achieve a more definitive and accurate sampling of protein-protein
   interaction but not a random selection on binding structures based on
   conformation energies that fails to fit previous experimental findings,
   the following docking parameters were set prior the docking simulation.
   To drive the docking, ambiguous interaction restraints (i.e. the active
   site residues) derived from previous protein-protein interaction
   analysis of experimentally resolved structures were referenced.
   Residues were regarded as accessible if they fulfil the minimum 15% of
   relative solvent accessibility, whereas passive residues were defined
   if they are located within the 6.50 Å radius of active site residues.
   On the other hand, non-polar hydrogens were eliminated during the
   docking processes; and that the semi-flexible regions were
   automatically defined.

   During the simulation, a three-step docking procedure using the
   software default setting was conducted, characterized by an initial
   rigid docking, followed by semi-flexible docking and lastly a structure
   refinement step in the presence of water molecules. The resulting
   HADDOCK scores and predicted binding energies of the top three clusters
   were considered and compared. Cut-off distances of proton-acceptor
   (hydrogen bonds) and carbon-carbon (hydrophobic contacts) were set at
   2.5 Å and 3.9 Å, respectively. Interactions between amino acid residues
   located on the docking proteins were subsequently analysed by the
   Protein-Ligand Interaction Profiler (PLIP)^[458]159. The representative
   images were created on the PyMOL software.

In silico gene promotor analysis and targeted transcription factor binding
site identification

   The bindings of c-MYC and HIF1α transcription factors respectively on
   GLS2 and ALDOC genes were first analysed on the ChIP-Atlas and the
   gross genome binding regions were visualized on the IGV genome
   browser^[459]160,[460]161. Upon such confirmation, detailed mapping of
   the binding sites was performed with immediate promoter sequences
   extracted from the NCBI Genome Data Viewer using the using the JASPAR
   database, with a score threshold > 85 %. Conserved transcription factor
   binding sites found on the human and mouse homolog genes promotor were
   extracted for the downstream chromatin immunoprecipitation (ChIP)-PCR
   experiment, as described above.

Human microarray data mining

   A human expression array dataset (Gene Expression Omnibus accession no.
   [461]GSE61019), comparing samples of A-T and control cerebella, was
   analysed with GEO2R^[462]162. Raw data from each of the samples were
   extracted. All transcripts that reflected significant changes (adjusted
   P ≤ 0.05) by the GEO2R analyser were further extracted and analysed
   using GSEA with the KEGG pathway groupings of genes. Lipid
   metabolism-related pathway which were significant (Nom p-value < 0.05)
   and enriched in normal controls were selected and shown in
   Supplementary Fig. [463]15.

Single-nucleus RNA sequencing data mining and analysis

   The original and raw single-nuclei RNA sequencing data of human
   cerebellar cortex was downloaded from the Single Cell Portal
   (SCP1300)^[464]76. Similarly, the original and raw single-nuclei RNA
   sequencing data of mouse cerebellar cortex were downloaded from the GEO
   database ([465]GSE165371).

   DropletUtils was then used as the first filter to remove any cells with
   extremely low expression in each of the samples, with the false
   positive rate set as 0.01. Filtered cells was then merged into as an
   integrated dataset which was then analysed by the Seurat 4.0.5. Genes
   expressed in at least three nuclei were retained, and those outlier
   nuclei with a high ratio of mitochondrial encoded transcripts (i.e.,
   >20%, < 200 UMI) relative to the total RNA, and those potential
   doublets ( > 6000 UMI) were discarded in the subsequent analysis. Top
   30 dimensions and resolution with 0.5 as input were used to build the
   UMAP graphs, tSNE graphs and cell clusters. Major cerebellar cell types
   were annotated according to the published study^[466]163. For the
   identification of differentially expressed genes of the targeted
   Purkinje cell clusters located at the anterior and posterior lobes was
   calculated using the FindMarkers function in the Seurat 4.0.5 package
   with the cut off value set at p-value < 0.01 and min.pct = 0.25. Those
   enriched in target clusters were then clustered for pathway analysis
   using the EnrichR software ([467]https://maayanlab.cloud/Enrichr/).

Mouse cerebellar tissue bulk RNA sequencing

   Frozen cerebellar tissues from the transgenic mice were harvested and
   sent to the Novogene for total RNA extraction and RNA sequencing with
   the Illumina HiSeq X Ten platform. FASTQ files obtained were
   subsequently analysed by the FastQC (V. 0.11.9) of Babraham
   Bioinformatics, serving as a quality control assessment tool for these
   data. After assuring the quality of the data files, genome indexes were
   generated using the STAR software using. These indexes were generated
   with the mouse annotation and mouse genome assembly provided by the
   Gencode Release 28 (GRCm39, released in May 2021). The cleaned dataset
   was then aligned with the genome indexes using the STAR alignment
   software. FeatureCounts was then used to quantify the sequencing data.
   Data extraction, matrix construction and differential gene expression
   analysis were then performed using the DESeq2 package on R. Pathway
   analysis was performed with Enrich R. The original FASTQ files were
   deposited on GEO Omnibus ([468]GSE222655).

Untargeted metabolome analysis by capillary electrophoresis time-of-flight
mass spectrometry (CE-TOFMS) and liquid chromatography (LC)-TOFMS

   Metabolome analyses were performed in mouse cerebellar tissues using
   CE-TOFMS for both cationic and anionic metabolites on the basis of
   service purchased from Human Metabolome Technologies’ standard library.
   Samples were sent to HMT where their weights were first measured. For
   CE-TOFMS preparation, samples were mixed with 1500 µl of 50%
   acetonitrile in water (v/v) containing internal standards (10 µM) and
   homogenized by a homogenizer (1500 r.p.m., 120 sec x 1 times). The
   supernatant (400 µl) was then filtrated through 5-kDa cut off filter
   (ULTRAFREE-MC-PLHCC, HMT) to remove macromolecules. The filtrate was
   centrifugally concentrated and resuspended in 50 µl of ultrapure water
   immediately before measurement. Whereas for LC-TOFMS preparation,
   weighted samples were mixed with 300 µl of 1% formic acid in
   acetonitrile (v/v) containing internal standards (10 µM) and
   homogenized by a homogenizer (1500 r.p.m., 120 s × 2 times). The
   mixture was yet again homogenized after adding 100 µl of Milli-Q water
   and then centrifuged (2300 × g, 4 °C, 5 min). After the supernatant was
   collected, 300 µl of 1 % formic acid in acetonitrile (v/v) and 100 µl
   of MilliQ-water were added to the precipitate. The homogenization and
   centrifugation were performed as described above, and the supernatant
   was mixed with previously collected one. The mixed supernatant was
   filtrated through 3-kDA cut-off filter (NANOCEP 3 K OMEGA, PALL
   Corporation, Michigan, USA) to remove proteins and further filtrated
   through column (Hybrid SPE phospholipid 55261-U, Supelco, Bellefonte,
   PA, USA) to remove phospholipids. The filtrate was desiccated and
   resuspended in 200 µl of 50% isopropanol in Milli-Q water (v/v)
   immediately before the measurement.

CE-TOFMS measurement

   The compounds were measured in the Cation and Anion modes of CE-TOFMS
   based metabolome analysis in the following conditions. Samples were
   diluted in 2 folds for measurement, to improve analysis qualities of
   the CE-MS analysis.

   The parameters for the detection of cationic metabolites were
   established as follows: An Agilent CE-TOFMS system (Agilent
   Technologies Inc., Machine No. 12) equipped with a fused silica
   capillary (inner diameter: 50 µm, length: 80 cm) was utilized. For the
   analytical conditions, the Cation Buffer Solution (product number:
   H3301-1001) was employed as both the running and rinse buffer. Sample
   injection was performed under a pressure of 50 mbar for 5 s. The
   capillary electrophoresis (CE) voltage was set to +30 kV. In the mass
   spectrometry phase, ionization was conducted using the ESI
   (electrospray ionization) positive mode, with a capillary voltage of
   4000 V. The mass spectrometer was operated with a scan range of m/z
   50–1000. The sheath liquid used was HMT Sheath Liquid (product number:
   H3301-1020).

   The parameters for the detection of anionic metabolites were
   established as follows: An Agilent CE-TOFMS system (Agilent
   Technologies Inc., Machine No. 5) equipped with a fused silica
   capillary (inner diameter: 50 µm, length: 80 cm) was utilized. For the
   analytical conditions, the Anion Buffer Solution (product number:
   I3302-1023) was employed as both the running and rinse buffer. Sample
   injection was performed under a pressure of 50 mbar for 22 s. The
   capillary electrophoresis (CE) voltage was set to +30 kV. In the mass
   spectrometry phase, ionization was conducted using the ESI
   (electrospray ionization) negative mode, with a capillary voltage of
   3500 V. The mass spectrometer was operated with a scan range of m/z
   50–1000. The sheath liquid used was HMT Sheath Liquid (product number:
   H3301-1020).

LC-TOFMS measurement

   The compounds were analyzed in both Positive and Negative modes using
   LC-TOFMS-based metabolome analysis under optimized conditions. To
   enhance the quality of the CE-MS analysis, samples were diluted 1-fold
   prior to measurement.

   The analysis for cationic metabolites in Positive mode was conducted
   using an Agilent 1200 series RRLC system SL (Agilent Technologies Inc.)
   coupled with an ODS column (2 × 50 mm, 2 µm) and an Agilent LS/MSD TOF
   mass spectrometry system (Machine No. 9). The column temperature was
   maintained at 40 °C, and the mobile phase consisted of H2O with 0.1%
   formic acid (Mobile Phase A) and a mixture of isopropanol,
   acetonitrile, and water (65:30:5) containing 0.1% formic acid and 2 mM
   ammonium formate (Mobile Phase B). The flow rate was set at 0.3 mL/min,
   with a total run time of 20 min followed by a post-run time of 7.5 min.
   The gradient conditions included Mobile Phase B starting at 1% for the
   first 0.5 min, increasing linearly to 100% over 13.5 min, and held at
   100% for the remaining 6.5 min. Mass spectrometry was performed using
   ESI in Positive ionization mode, with a nebulizer pressure of 40 psi, a
   dry gas flow rate of 10 L/min, and a dry gas temperature of 350 °C. The
   capillary voltage was set to 4000 V, and the scan range was m/z
   100–1700. A sample injection volume of 1 µL was used for the analysis.

   The analysis of anionic metabolites in Negative mode was performed
   using an Agilent 1200 series RRLC system SL (Agilent Technologies Inc.)
   coupled with an ODS column (2 × 50 mm, 2 µm) and an Agilent LS/MSD TOF
   mass spectrometry system (Machine No. 9). The column temperature was
   maintained at 40 °C, with Mobile Phase A consisting of H2O containing
   0.1% formic acid and Mobile Phase B comprising a mixture of
   isopropanol, acetonitrile, and water (65:30:5) with 0.1% formic acid
   and 2 mM ammonium formate. The flow rate was set at 0.3 mL/min, with a
   total run time of 20 min and a post-run time of 7.5 min. The gradient
   conditions began with Mobile Phase B at 1% for the first 0.5 min,
   followed by a linear increase to 100% over 13.5 min, and held at 100%
   for the remaining 6.5 min. Mass spectrometry was conducted in ESI
   Negative ionization mode, with a nebulizer pressure of 40 psi, dry gas
   flow rate of 10 L/min, and dry gas temperature of 350 °C. The capillary
   voltage was set at 3500 V, and the scan range covered m/z 100–1700. A
   sample injection volume of 1 µL was used for the analysis.

   Peaks detected in both CE-TOFMS and LC-TOFMS were extracted using
   automatic integration software (MasterHands ver. 2.17.1.11 developed at
   Keio University) in order to obtain peak information including m/z,
   migration time (MT) in CE, retention time (RT) in LC, and peak area.
   The peak area was then converted to relative peak area by the following
   equation. the limit of peak detection was determined by a signal-noise
   ratio of at least 3.

Relative peak area = metabolite peak area/internal standard peak
area x sample amount

   Putative metabolites were then assigned from HMT’s standard library and
   Known-Unknown peak library on the basis of m/z and MT or RT. The
   tolerance was ± 0.5 min in MT and ± 0.3 min in RT, ± 10 ppm (CE-TOFMS)
   and ± 25 ppm (LC-TOFMS) in m/z. If several peaks were assigned the same
   candidate, the candidate was given the branch number.

Mass error (ppm) = (measured value − theoretical value)/measured value x 10^6

   Subsequent absolute quantification was performed in target metabolites.
   All the metabolite concentrations were calculated by normalizing the
   peak area of each metabolite with respect to the area of the internal
   standard and by using standard curves, which were obtained by
   single-point (100 µM or 50 µM) calibrations. Significantly changed
   metabolites (with Log[2]FC ± 0.5; P < 0.05) were enriched and analyzed
   by the Metabolite Set Enrichment Analysis (MSEA) or the Joint Pathway
   Analysis module (with KEGG metabolic gene expression data) on
   MetaboAnalyst
   ([469]https://www.metaboanalyst.ca/MetaboAnalyst/ModuleView.xhtml).

Lipid extraction and selected lipidomics analysis

   The following procedures were performed based on a published protocol
   with slight modification^[470]164. Lipid extraction procedures were
   performed as reported previously with minor modifications^[471]165.
   Briefly, 10 mg of frozen cerebellar tissues was deactivated by 900 μL
   mixture of chloroform: methanol (1:2) with 10% deionized water. Then
   the mixture was homogenized by OMNI Bead Ruptor (OMNI, USA). After
   homologenization at 4 °C for 1 h, 400 μL deionized water and 300 μL
   chloroform were added to the mixture, which the mixture was vortexed
   for 10 min followed by centrifugation at 9021 × g, 4 °C for 15 min. The
   resulting under layer organic phase material was then transferred to a
   sterile tube. Any unextracted residuals were again subjected to an
   repeated extraction step with another 500 μL chloroform. By combining
   the products extracted from the two extraction steps, the samples were
   then frozen dried in a SpeedVac (Genevac). Dried samples were stored at
   −80 °C until further analysis.

   Lipidomics approach was conducted by Exion UPLC-QTRAP 6500PLUS (Sxiex)
   via electrospray ionization (ESI) iron source under conditions as
   follows: curtain gas = 20, ion spray voltage = 5500 V, temperature =
   400 °C, ion source gas 1 = 35, ion source gas 2 = 35. Specifically,
   Triacylglycerol (TAG) and diacylglycerol (DAG) were detected using a
   reverse-phase high-performance liquid chromatography
   (HPLC)-electrospray ionization- mass spectrometry (MS) with Phenomenex
   Kinetex 2.6 μm C18 column (inner diameter 4.6 × 100 mm). The glyceride
   lipids were separated by isocratic elution mode with chloroform,
   methanol, 0.1 mol/L ammonium acetate (100: 100: 4) at a flow rate of
   160 μL/min for 20 min. Quantitative analyses of TAG and DAG were
   conducted by applying the Neutral Loss MS/MS technology, referencing
   the internal standards. Similarly, free cholesterol, sterols and
   corresponding esters were analyzed using HPLC tandem MS analysis
   through atmospheric pressure chemical ionization (APCI) mode and were
   quantified by referencing internal standards.

   Lipids with different polarities including phospholipids and
   sphingolipids, as well as free fatty acids were separated by NP-HPLC
   according to previously reported^[472]166 using Phenomenex Luna 3 μm
   silica column (inner diameter 150 × 2.0 mm). HPLC condition was set as
   follows: mobile phase A (chloroform: methanol: ammonium hydroxide,
   89.5:10:0.5), B (chloroform: methanol: ammonium hydroxide: water, 55:
   39: 0.5: 5.5). 95 % A was run for 5 min which was then linearly reduced
   to 60% in 7 min. After continuing for another 5 min, the condition was
   changed to 30% and was maintained for 15 min, then was changed back to
   initial gradient and kept for 5 min. Individual polar lipid species
   were quantified using multiple reaction monitoring transitions by
   referencing their internal standards.

Stable isotope labelled glucose, glutamine and competitive
alpha-ketoglutarate (α-KG) metabolite tracing

   Metabolic fate and catabolic flux of labelled glucose
   (^13C[6]-U-glucose) and glutamine (^13C[5]-U-glutamine) was performed
   as previously reported^[473]167. In primary cortical neurons,
   ^13C[6]-U-glucose or ^13C[5]-U-glutamine tracing followed by capillary
   electrophoresis-time of flight mass spectrometer (CE-TOF/MS) (Agilent
   Technologies) were performed. DIV14 primary cortical neurons were
   incubated in the glucose-free medium supplemented with 10 mM
   (^13C[6]-U-glucose) (Cambridge Isotope, CLM-1396) or glutamine-free
   medium supplemented with 2 mM ^13C[5]-U-glutamine (605166,
   Sigma-Aldrich) and collected at 2 h of incubation at isotopic-steady
   state for both glycolysis and TCA cycle investigations^[474]168.
   Similarly, metabolic fate and competitive metabolic flux analyses of
   glutamine and α-KG were performed as well. In acute cerebellar slice
   culture, 2 mM ^13C[5]-U-glutamine isotope alone or simultaneously with
   1,2,3,4-^13C[4]-α-KG (Cambridge Isotope, 6363-53-7) for 2 h prior
   harvesting. To ensure steady-state tracing, all tracing analyses were
   conducted for durations exceeding the suggested timeframes elucidated
   by Jang and colleagues, whereby the isotopic-steady state occurs within
   approximately 10 min for glycolysis and around 2 h for the
   tricarboxylic acid (TCA) cycle in in vitro^[475]168.

   At harvesting, cells or tissues were then washed twice with 10 mL of 5%
   mannitol aqueous solution, and subsequently incubated with 1 mL of
   methanol containing 25 µm internal standards (methionine sulfone,
   2-(N-morpholino) -ethanesulfonic acid (MES) and D-camphor-10-sulfonic
   acid) for 10 min. Four hundred microliters of the extracts were mixed
   with 200 µL Milli-Q water and 400 µL chloroform and centrifuged at
   10,000 g for 3 min at 4 °C. Subsequently, 400 µL of the aqueous
   solution was centrifugally filtered through a 5-kDa cut-off filter to
   remove proteins. The filtrate was centrifugally concentrated and
   dissolved in 50 µL of Milli-Q water that contained reference compounds
   (200 µm each of 3-aminopyrrolidine and trimesate) immediately before
   metabolome analysis.

   The relative concentrations of all the charged metabolites in samples
   were measured by CE-TOFMS, following the methods as previously
   reported^[476]169. In brief, a fused silica capillary (50 µm internal
   diameter × 100 cm) was used with 1 M formic acid as the electrolyte.
   Methanol: water (50% v/v) containing 0.1 µm hexakis
   (2,2-difluoroethoxy) phosphazene was delivered as the sheath liquid at
   10 µL min−1. Electrospray ionization (ESI)-TOFMS was performed in
   positive-ion mode, and the capillary voltage was set to 4 kV. Automatic
   recalibration of each acquired spectrum was achieved using the masses
   of the reference standards. [(13 C isotopic ion of a protonated
   methanol dimer (2 MeOH + H)]+, m/z 66.0632) and ([hexakis
   (2,2-difluoroethoxy) phosphazene + H]+, m/z 622.0290). Quantification
   was performed by comparing peak areas to calibration curves generated
   using internal standardization techniques with methionine
   sulfone^[477]169. To analyze anionic metabolites, a commercially
   available COSMO(+) (chemically coated with cationic polymer) capillary
   (50 µm internal diameter × 105 cm) (Nacalai Tesque) was used with a
   50 mm ammonium acetate solution (pH 8.5) as the electrolyte. Methanol;
   5 mm ammonium acetate (50% v/v) containing 0.1 µm hexakis
   (2,2-difluoroethoxy) phosphazene was delivered as the sheath liquid at
   10 µL min−1. ESI-TOFMS was performed in negative ion mode, and the
   capillary voltage was set to 3.5 kV. For anion analysis, trimesate and
   CAS were used as the reference and the internal standards,
   respectively^[478]170. Mole percent enrichment of isotopes, an index of
   isotopic enrichment of metabolites, was calculated as the percent of
   all atoms within the metabolite pool that are labelled according to the
   established formula^[479]171.

Metabolic fuel flux assays

   The Mito Fuel Flex Tests were performed on Seahorse XFe24 Bio-analyzer
   (Agilent). All assays were performed following manufacturer’s
   protocols. In brief, the test inhibits import of three major metabolic
   substrates (pyruvate, fatty acids, and/or glutamine) with mitochondrial
   pyruvate carrier inhibitor UK5099 (2 µM), carnitine
   palmitoyltransferase 1 A inhibitor etomoxir (4 µM), or glutaminase
   inhibitor BPTES (3 µM). This test determines cellular dependence on
   each of the metabolites to fuel mitochondrial metabolism by inhibiting
   the individual substrate import. Baseline OCR was monitored for 18 min
   followed by sequential inhibitor injections (i.e., Treatment 1 or
   Treatment 2) with OCR reading for 1 h following each treatment. The
   inhibitor treatment combinations and calculations are shown as below:
   [MATH: <mtable><mtr><mtd columnalign="left"><munder
   accentunder="true"><mrow><mi>M</mi><mi>e</mi><mi>t</mi><mi>a</mi><mi>b<
   /mi><mi>o</mi><mi>l</mi><mi>i</mi><mi>t</mi><mi>e</mi><mspace
   width="0.16em"></mspace><mi>t</mi><mi>e</mi><mi>s</mi><mi>t</mi></mrow>
   <mrow><mo stretchy="true">_</mo></mrow></munder></mtd><mtd
   columnalign="left"><munder
   accentunder="true"><mrow><mi>T</mi><mi>r</mi><mi>e</mi><mi>a</mi><mi>t<
   /mi><mi>m</mi><mi>e</mi><mi>n</mi><mi>t</mi><mspace
   width="0.25em"></mspace><mn>1</mn><mspace
   width="1.0em"></mspace><mi>T</mi><mi>r</mi><mi>e</mi><mi>a</mi><mi>t</m
   i><mi>m</mi><mi>e</mi><mi>n</mi><mi>t</mi><mspace
   width="0.25em"></mspace><mn>2</mn></mrow><mrow><mo
   stretchy="true">_</mo></mrow></munder></mtd></mtr><mtr><mtd
   columnalign="left"><mi>G</mi><mi>l</mi><mi>u</mi><mi>c</mi><mi>o</mi><m
   i>s</mi><mi>e</mi><mo>/</mo><mi
   mathvariant="normal">Pyruvate</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">dependence</mi></mtd><mtd columnalign="left"><mi
   mathvariant="normal">UK</mi><mn>5099</mn><mspace
   width="1.0em"></mspace><mi
   mathvariant="normal">Etomoxir</mi><mo>+</mo><mi mathvariant="normal">
   BPTES</mi></mtd></mtr><mtr><mtd
   columnalign="left"><mi>G</mi><mi>l</mi><mi>u</mi><mi>t</mi><mi>a</mi><m
   i>m</mi><mi>i</mi><mi>n</mi><mi>e</mi><mspace
   width="0.16em"></mspace><mi>d</mi><mi>e</mi><mi>p</mi><mi>e</mi><mi>n</
   mi><mi>d</mi><mi>e</mi><mi>n</mi><mi>c</mi><mi>e</mi></mtd><mtd
   columnalign="left"><mi mathvariant="normal">BPTES</mi><mspace
   width="1.0em"></mspace><mi
   mathvariant="normal">Etomoxir</mi><mo>+</mo><mi
   mathvariant="normal">UK</mi><mn>5099</mn></mtd></mtr><mtr><mtd
   columnalign="left"><mi mathvariant="normal">Fatty</mi><mspace
   width="0.16em"></mspace><mi mathvariant="normal">acid</mi><mspace
   width="0.16em"></mspace><mi
   mathvariant="normal">dependence</mi></mtd><mtd columnalign="left"><mi
   mathvariant="normal">Etomoxir</mi><mspace width="1.0em"></mspace><mi
   mathvariant="normal">BPTES</mi><mo>+</mo><mi
   mathvariant="normal">UK</mi><mn>5099</mn></mtd></mtr></mtable> :MATH]
   [MATH: <mi
   mathvariant="normal">Dependency</mi><mrow><mo>(</mo><mrow><mi>%</mi></m
   row><mo>)</mo></mrow><mo>=</mo><mrow><mo>(</mo><mrow><mrow><mo>[</mo><m
   row><mi mathvariant="normal">Baseline</mi><mspace
   width="0.16em"></mspace><mi mathvariant="normal">OCR</mi><mo>−</mo><mi
   mathvariant="normal">Target</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">inhibitor</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">OCR</mi></mrow><mo>]</mo></mrow><mo>/</mo><mrow><m
   o>[</mo><mrow><mi mathvariant="normal">Baseline</mi><mspace
   width="0.16em"></mspace><mi mathvariant="normal">OCR</mi><mo>−</mo><mi
   mathvariant="normal">All</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">inhibitors</mi><mspace width="0.16em"></mspace><mi
   mathvariant="normal">OCR</mi></mrow><mo>]</mo></mrow></mrow><mo>)</mo><
   /mrow><mo>×</mo><mn>100</mn><mi>%</mi><mo>.</mo> :MATH]

Glycolysis stress test

   The XF Glycolysis Stress Test (Seahorse Bioscience, Agilent) was used
   to assess glycolysis function in cells, which was conducted using the
   XFe24 Analyzer. By directly measuring ECAR, the kit provided a standard
   method to assess the following key parameters of glycolysis flux:
   glycolysis, glycolytic capacity and glycolytic reserve, in addition to
   non-glycolytic acidification. Cells were seeded at a density of 45,000
   cells/well. Cells were first incubated in pyruvate-free glycolytic
   assay medium for 1 h prior to the first injection of a saturated
   concentration of glucose (final concentration: 10 mM). The cells
   catabolize glucose into pyruvate via the glycolysis pathway, producing
   ATP, nicotinamide-adenine dinucleotide (reduced form), water and
   protons. The discharge of protons into surrounding medium leads to a
   sudden increase in ECAR, which was used to define the basal glycolytic
   capacity. The second injection was oligomycin (final concentration:
   1 µM), which may divert energy production to glycolysis by restricting
   mitochondrial ATP production. Consequently, the sharp increase in ECAR
   indicates the level of glycolytic capacity. The final injection was
   2-deoxy-glucose [2-DG; final concentration: 50 mM], which is a glucose
   analogue that inhibits glycolysis through competitive binding to
   glucose hexokinase; the first enzyme in the glycolytic pathway. The
   resulting decrease in ECAR confirmed that the ECAR produced in the
   experiment was caused by glycolysis. The gap between glycolytic
   capacity and glycolysis was defined as the glycolytic reserve. The ECAR
   prior to glucose injection is referred to as non-glycolytic
   acidification and may occur due to additional processes in the cell.

Quantification procedures and statistical analyses

   For each experiment, no statistical methods were used to predetermine
   sample sizes, but our sample sizes were similar to those reported in
   recent publications. An assessment of the normality of all datasets was
   conducted, with this analysis documented in the source data files. In
   instances where datasets exhibited non-normal distributions,
   non-parametric statistical tests were employed for further statistical
   evaluations. All samples were analyzed and data collected were blinded
   to the experimental conditions. All experiments were performed in at
   least three independent occasions. N-numbers presented represent
   biological replicates. Differences between two groups were analyzed
   using two-tailed unpaired Student’s t-test (parametric) or two-tailed
   Mann-Whitney test (nonparametric). Differences between three groups
   were analyzed using one-way ANOVA (parametric) or Kruskal-Wallis test
   (nonparametric). Two-way ANOVA was used to determine the effect of two
   nominal predictor variables. The potential association between variants
   was analyzed using Pearson correlation coefficient. Except
   omics-related analyses, all statistical analyses were performed using
   GraphPad Prism 10 software package. Exact P-value is provided in each
   figure panel except when it is smaller than 1e-15 (i.e., the lower
   limit of the software”, then “<1e-15” is labelled instead. P  <  0.05
   was considered to indicate statistical significance.

Reporting summary

   Further information on research design is available in the [480]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [481]Supplementary Information^ (11.1MB, pdf)
   [482]41467_2025_64360_MOESM2_ESM.pdf^ (56KB, pdf)

   Description of Additional Supplementary Files
   [483]Supplementary Dataset 1^ (4.7MB, xlsx)
   [484]Supplementary Dataset 2^ (8.7MB, xlsx)
   [485]Supplementary Dataset 3^ (3.4MB, xlsx)
   [486]Reporting Summary^ (95.1KB, pdf)
   [487]Transparent Peer Review file^ (1.2MB, pdf)

Source data

   [488]Source Data^ (35.3MB, xlsx)

Acknowledgements