Calcium signals tune AMPK activity and mitochondrial homeostasis in dendrites of developing neurons

Dendrite morphogenesis is a pivotal step for the establishment of neural circuit connectivity in the developing brain. It is widely accepted that calcium (Ca²⁺) signals evoked by neuronal activity play important roles in dendrite development. Ca²⁺ influx via voltage-gated calcium channels and NMDA receptors triggers activation of Ca²⁺/calmodulin-dependent kinases (CaMKs) that positively and negatively regulate dendrite growth through transcriptional control and local remodeling of the cytoskeleton (Konur and Ghosh, 2005; Redmond et al., 2002; Wong and Ghosh, 2002).

Rapid extension and branching of dendrites in developing neurons is accompanied by extensive cytoskeletal and membrane remodeling and intracellular transport, which increase the energetic burden, particularly in growing dendritic branches. We have previously shown that mitochondria are actively transported to growing dendritic arbors to maintain local ATP levels necessary for the regulation of actin dynamics (Fukumitsu et al., 2015). In addition to producing the majority of cellular ATP, mitochondria also participate in various processes indispensable for the development and function of dendrites, including Ca²⁺ homeostasis, lipid flux and reactive oxygen species (ROS) generation (Rangaraju et al., 2019; Sheng and Cai, 2012). It has recently been demonstrated that the gradual increase in the size and activity of mitochondria during neuronal differentiation correlates with the rate of neuronal maturation, including dendrite arborization (Iwata et al., 2023). The abnormal morphology of dendritic mitochondria is one of the pathogenic hallmarks of neurodevelopmental disorders including autism spectrum disorder (Anitha et al., 2023), supporting the notion that constant delivery of functional mitochondria throughout a large volume of dendrites is crucial for maintenance as well as differentiation of neurons (Pekkurnaz Gulcin and Wang, 2022; Schon and Przedborski, 2011).

Mitochondrial homeostasis in postmitotic neurons is highly dependent on dynamic fission and fusion events. Mitochondrial fission is particularly important for the biogenesis of new mitochondria and for the removal of damaged or aged mitochondria through mitophagy (Youle et al., 2012). Mitochondrial fission is triggered by the spiral assembly of the evolutionally conserved GTPase dynamin-related protein 1 (Drp1, also known as Dnm1l) around the constriction site on the mitochondrial surface. During fission, Drp1 is recruited to the outer mitochondrial membrane by one of its receptors, such as mitochondrial fission factor (MFF) (Gandre-Babbe and van der Bliek, 2008; Palmer et al., 2011; Yoon et al., 2003). Several groups, including us, have shown that the inhibition of mitochondrial fission leads to abnormal elongation and dysfunction of mitochondria, resulting in aberrant dendrite formation in vivo and in vitro (Fukumitsu et al., 2016; Ishihara et al., 2009; Li et al., 2004). In contrast, excessive mitochondrial fission also leads to reduced respiratory activity and abnormal distribution of mitochondria, which impairs normal development of dendrites (Chen et al., 2007; Wang et al., 2009). However, it remains unclear how the proper balance of mitochondrial fission is adaptively regulated in developing neurons, the shape, volume and activity states of which are dynamically changing.

The balance between fission and fusion should be finely tuned in response to changes in the cellular and subcellular metabolic state. Previous studies have highlighted the role of AMP-activated protein kinase (AMPK) as an energy sensor that sustains the metabolic state by regulating mitochondrial dynamics and functions (Herzig and Shaw, 2018). AMPK regulates several aspects of mitochondrial homeostasis by direct phosphorylation of key factors, including MFF, peroxisome proliferator-activated receptor-coactivator 1 (PGC1α, encoded by Ppargc1a), armadillo repeat-containing protein 10 (ARMC10), unc-51 like autophagy activating kinases 1 and 2 (ULK1 and ULK2) and mitochondrial fission regulator 1-like protein (MTFR1L) (Chen et al., 2019; Ducommun et al., 2015; Egan et al., 2011; Jäger et al., 2007; Kim et al., 2011; Tilokani et al., 2022; Toyama et al., 2016). AMPK is activated by direct allosteric binding of AMP, facilitating phosphorylation at Thr172 by the upstream kinase LKB1 (or STK11) in response to cellular ATP shortage (Hawley et al., 2003). Additionally, AMPK can be directly phosphorylated by the Ca²⁺-sensitive kinase CaMKK2 (also known as CaMKKβ) in response to an elevated intracellular Ca²⁺ concentration (Hawley et al., 2005; Hurley et al., 2005). In the central nervous system, AMPK has been involved in both the prevention and progression of neurodegenerative diseases, including Alzheimer's disease (AD). AMPK is activated by metabolic and oxidative stresses in early pathogenesis of AD, improving mitochondrial functions to restore energy balance (Muraleedharan and Dasgupta, 2021). In contrast, AMPK overactivation has been implicated in mitochondrial fragmentation and degradation during the progression of AD (Lee et al., 2022; Mairet-Coello et al., 2013). However, AMPK function in the differentiating neurons during normal brain development is less understood.

In this study, we investigated the impact of neuronal activity on mitochondrial dynamics in differentiating mouse hippocampal neurons. We demonstrate that neuronal activity finely tunes the CaMKK2-AMPK pathway in growing dendrites of immature neurons. Physiological levels of AMPK activity induce mitochondrial fission and mitophagy via phosphorylation of MFF and ULK1, which are crucial in removing dysfunctional mitochondrial components and maintaining functional mitochondria in growing dendrites. Thus, we provide for the first time a causal link between activity-dependent dendritic arbor development and AMPK-dependent mitochondrial quality control.

We previously established a long-term live-imaging protocol of dissociated hippocampal neurons and observed the gradual formation of dendritic arbors, with dynamic extension and retraction, in the first week of culture (Wu et al., 2015). Here, we confirmed that synapses had not yet formed at this stage, consistent with previous studies (Renger et al., 2001) (Fig. 1A). To visualize neuronal activity in developing dendrites, we transfected a plasmid expressing GCaMP6s in cultured hippocampal neurons and performed live imaging at 5 days in vitro (DIV). It has been shown that developing dendrites are activated by ambient glutamate spontaneously released from nearby immature axon terminals and that this non-synaptic activity is crucial for dendritic outgrowth (Andreae et al., 2015). Consistently, we observed intermittent Ca²⁺ transients in dendrites and cell soma, which occurred sporadically in individual neurons (Fig. 1B; Fig. S1A). These Ca²⁺ transients were abolished by combined treatment with a blocker of voltage-gated sodium channels (tetrodotoxin or TTX) and an NMDA receptor antagonist [D-(−)-2-amino-5-phosphonopentanoic acid or APV] (TTX+APV treatment) (Fig. 1B; Movie 1). Dendrites exhibited dynamic extension and retraction in drug-treated and untreated neurons, but growth retardation gradually became evident, with a significant decrease in total dendritic length and the number of branches at 48 h after TTX+APV treatment (Fig. 1C-E; Movie 2).

To examine whether neuronal activity affects mitochondrial dynamics in growing dendrites, we transfected a plasmid encoding Mito-DsRed into neurons that were then treated with or without TTX+APV. Under both conditions, mitochondria were distributed throughout the dendritic arbors, and a significant fraction of them moved along the dendritic shaft (Fig. S1B), as previously reported by us and others (Faits et al., 2016; Fukumitsu et al., 2015; Loss and Stephenson, 2017; Spronsen et al., 2013). Inhibition of neuronal activity did not affect mitochondrial motility in dendrites: neither the fraction of motile mitochondria nor their average speed was altered by TTX+APV treatment (Fig. S1B,C). There were no apparent changes in the density and distribution of mitochondria in dendrites in the drug-treated neurons (Fig. S1D). However, we found a significant increase in the length of dendritic mitochondria in the drug-treated neurons [median (interquartile range or IQR)=2.0 (1.2-2.9) µm in control versus 2.5 (1.6-3.8) µm in TTX+APV neurons; Fig. 1F,H-J]. In the axon, however, the length of mitochondria was much shorter, as previously reported (Lewis et al., 2018), and was unaffected by TTX+APV treatment [median (IQR)=1.2 (0.91-1.9) µm in control versus 1.2 (0.92-1.7) µm in TTX+APV neurons; Fig. 1G,K,L]. As mitochondrial morphology is affected by the balance between fission and fusion, we analyzed the frequency of fission and fusion in dendrites by high-resolution time-lapse imaging (Fig. S1E; Movie 3). We found that inhibition of neuronal activity significantly suppressed mitochondrial fission but had little or no effect on fusion (Fig. 1M). Conversely, acute glutamate treatment led to an immediate fission of mitochondria in the cell soma and dendrites, suggesting that enhanced neuronal activity facilitated mitochondrial fission (Fig. S1F-J). Taken together, these data suggest that neuronal activity is involved in the regulation of mitochondrial fission during dendritic outgrowth in developing neurons.

It has been shown that AMPK mediates mitochondrial fission in response to energy stress in neurons and non-neuronal cells (Toyama et al., 2016). We therefore examined whether AMPK is involved in activity-dependent mitochondrial fission in developing hippocampal neurons. AMPK is a heterotrimeric serine/threonine kinase composed of one catalytic (α1/α2), one regulatory (β1/β2) and one AMP/ATP-binding (γ1/γ2/γ3) subunit. Both AMPKα1 (encoded by Prkaa1) and AMPKα2 (encoded by Prkaa2) were expressed in cultured hippocampal neurons, as assessed by western blotting using isoform-specific antibodies (Fig. 2A). Specifically, AMPKα1 expression was high in earlier stages and declined by DIV10, whereas AMPKα2 expression started at a level comparable with that of AMPKα1, and then gradually increased, peaking at DIV10. The abundance of active AMPKα phosphorylated at Thr172 (detected by an antibody that recognizes both phosphorylated isoforms, p-AMPKα) increased during dendritic outgrowth in parallel with AMPKα2 expression (Fig. 2A).

To investigate the role of AMPK in mitochondrial dynamics and dendritic formation, we knocked down AMPK subunits in cultured neurons by expressing short hairpin RNAs (shRNAs) against AMPKα1 and AMPKα2 (Fig. S2A). At DIV3, we transfected dissociated hippocampal neurons with AMPKα1 shRNA or AMPKα2 shRNA plasmids together with a volume marker GFP construct and observed dendritic morphology at DIV5. AMPKα2 shRNA significantly reduced the size and complexity of dendritic branches, reminiscent of those of the activity-deprived neurons (Fig. 2B-D and Fig. S2B-E, compared with Fig. 1). Concomitant exogenous expression of an AMPKα2 mutant cDNA (AMPKα2 rescue) that was resistant to AMPKα2 shRNA largely prevented the dendritic defects, discounting any major off-target effects of the shRNA (Fig. 2B-D; Fig. S2C). In contrast, AMPKα1 shRNA produced much weaker effects on dendritic morphology. Furthermore, simultaneous knockdown of AMPKα1 and AMPKα2 was not phenotypically different from AMPKα2 knockdown, suggesting that AMPKα2 was the predominant isoform regulating dendritic morphogenesis in our culture (Fig. S2C-E). We further confirmed that repression of AMPKα2 using CRISPR interference (CRISPRi) significantly delayed dendritic outgrowth at DIV5 (Fig. S2F-I). We thus focused on AMPKα2 in subsequent experiments.

To confirm the requirement of AMPKα2 for dendritic formation in vivo, we introduced the AMPKα2 shRNA#1 together with GFP into CA1 pyramidal neurons at embryonic day (E) 14.5 by in utero electroporation. Compared with control neurons, AMPKα2-depleted neurons showed significantly reduced length and complexity of the apical dendrite at postnatal day (P) 5, supporting the notion that AMPKα2 functions in dendritic outgrowth of hippocampal pyramidal neurons (Fig. 2E-H). In contrast, the dendritic hypotrophy was less evident at P10, suggesting that dendritic growth retardation in AMPKα2 deficiency is reverted in the second postnatal week in vivo. We attempted to analyze mitochondrial morphology in hippocampal neurons in vivo, but owing to technical difficulty in oxygen inhalation by P5 pups, we were unable to maintain normal mitochondrial morphology in neurons in the P5 brain.

Returning to cultured cells, we next observed mitochondria in AMPKα2-depleted neurons by co-transfecting plasmids for Mito-DsRed and GFP. The AMPKα2 knockdown significantly increased mitochondrial length in dendrites [median (IQR)=2.1 (1.3-3.1) µm in control; 2.6 (1.6-4.1) µm in AMPKα2 knockdown; 2.0 (1.3-3.1) µm in rescue; Fig. 3A,C-E], but not in the axons at DIV5 [median (IQR)=1.2 (0.88-1.8) µm in control; 1.1 (0.82-1.7) µm in AMPKα2 knockdown; Fig. 3B,F,G]. There was no apparent change in the density and distribution of mitochondria in the hypotrophic dendrites, whereas the total mitochondrial mass appeared significantly decreased in AMPKα2-depleted neurons (Fig. S3A,B). AMPKα2 CRISPRi also induced abnormal elongation of mitochondria in dendrites (Fig. S3C-F). These results prompted us to use time-lapse imaging to analyze the fission/fusion frequency of dendritic mitochondria in the neurons transfected with AMPKα2 shRNA. As in the case for neuronal activity inhibition, AMPKα2 knockdown specifically lowered the frequency of fission but had little or no effect on fusion in dendrites (Fig. 3H).

We further documented the importance of AMPK activity in mitochondrial dynamics by pharmacological activation and inhibition. We found that dendritic mitochondria were significantly shortened after a 3-5 h treatment with the AMPK activator 5-aminoimidazole-4-carboxamide ribonucleotide (AICAR; 1 mM), underscoring the involvement of AMPK in the regulation of mitochondrial fission in hippocampal neurons (Fig. 3I-L). Collectively, these data suggest that AMPKα2 has crucial roles in mitochondrial fission and dendritic formation in hippocampal neurons, and that the two phenomena are mechanistically linked.

The striking phenotypic similarities between the modulation of neuronal activity and AMPK activity prompted us to examine whether AMPK mediates the activity-dependent control of mitochondrial dynamics in developing dendrites. In order to analyze the spatiotemporal dynamics of AMPK activity in developing hippocampal neurons, we transfected them with a plasmid encoding AMPKAR-EV, a highly sensitive fluorescence resonance energy transfer (FRET)-based biosensor for AMPK (Konagaya et al., 2017). AMPK phosphorylates the substrate peptide and promotes its binding to the FHA1 domain and conformational change in AMPKAR-EV. This leads to an increase in the FRET efficiency from super-enhanced cyan fluorescent protein (SECFP) to YPet. (Fig. S4A). We observed high basal FRET signals in the axon, which presumably reflected the activity of AMPK-related kinases such as SAD-A/B (also known as BRSK2 and BRSK1, respectively) (Barnes et al., 2007; Courchet et al., 2013; Sample et al., 2015). Indeed, AMPKα2 shRNA specifically suppressed FRET signals in the somatodendritic compartment but not in the axon (Fig. S4B), supporting the notion that the FRET signals in dendrites, but not in axons, reflect AMPK activity.

Consistent with previous reports, we observed a rapid increase in AMPK activity in the cell body and dendrites upon glutamate treatment (Connolly et al., 2014; Weisová et al., 2009); moreover, the AMPK activity was abrogated by AMPKα2 knockdown (Fig. S4C). Interestingly, we found an unambiguous periodic fluctuation of AMPK activity with a frequency of zero to five peaks in 3 min, in the somatodendritic compartment of untreated neurons. The basal fluctuation was strongly suppressed by co-transfection of AMPKα2 shRNA (Fig. 4A,B) or inhibition of neuronal activity by TTX+APV treatment (Fig. 4C,D), suggesting that neuronal activity caused dynamic changes in AMPK activity. To further prove the causal relationship of neuronal activity and AMPK activation, we performed simultaneous imaging of Ca²⁺ and AMPK activity using a red fluorescent Ca²⁺ indicator, jRGECO1a, together with AMPKAR-EV. We observed periodic Ca²⁺ spikes that highly synchronized with AMPK activity fluctuation in dendrites (Fig. 4E,F; Movie 4). The temporal cross-correlation analysis revealed the highest correlation at 5 s after Ca²⁺ influx, indicating that Ca²⁺ influx precedes AMPK activation (Fig. 4G). A negative-control FRET biosensor with a non-phosphorylatable PKA substrate motif (PKA-TA) (Kamioka et al., 2012) did not respond to Ca²⁺ transients, negating the possibility that FRET signals are influenced by indirect effects such as pH changes (Fig. S4D). Inhibition of neuronal activity by TTX+APV treatment suppressed both Ca²⁺ spikes and AMPK fluctuation (Fig. S4E; Movie 4). In contrast, AMPK knockdown suppressed only the AMPK fluctuation and did not alter the Ca²⁺ transient (Fig. S4F; Movie 5), indicating that AMPK fluctuation was induced by neuronal activity. The AMPK fluctuation synchronized with Ca²⁺ spikes was not observed in the axon (Fig. S4G; Movie 6).

Two upstream kinases, CaMKK2 and LKB1, regulate AMPK activity by phosphorylation (Hawley et al., 2003, 2005; Hurley et al., 2005). As CaMKK2 has been implicated in activity-dependent formation of dendrites (Wayman et al., 2004, 2006), we examined whether CaMKK2 is required for AMPK fluctuation in response to neuronal activity. Co-transfection of CaMKK2 shRNA or treatment with a CaMKK2 inhibitor (STO609, 10 µM) suppressed the periodic activation of AMPK (Fig. 4H-J). In contrast, the AMPK fluctuation was not affected by treatment with a LKB1 inhibitor (Pim1, 1 µM) (Fig. 4K,L). Taken together, these data suggest that neuronal activity dynamically regulates AMPKα2 activity via Ca²⁺/CaMKK2 in the somatodendritic region of developing hippocampal neurons.

Mitochondrial fission is involved in biogenesis and mitophagic degradation of mitochondria. It has recently been reported that the downstream pathways could be distinguished by the position of fission, either at the midzone or at the periphery (Kleele et al., 2021). In normal developing hippocampal neurons, we observed both midzone and peripheral fissions in growing dendrites, with the frequency of the midzone fission, which was implicated in mitochondrial biogenesis in the previous study (Kleele et al., 2021), being approximately 2.1-fold higher than that of peripheral fission (Fig. 5A,B). TTX+APV treatment hampered both midzone and peripheral fissions but had a stronger inhibitory effect on peripheral fission [12.8±6.97% (mean±s.e.m.) reduction in midzone fission versus 39.7±9.74% reduction in peripheral fission; Fig. 5B; Fig. S5A]. AMPKα2 knockdown displayed strikingly similar effects, with stronger suppression of the peripheral fission (35.4±4.86% reduction in midzone fission versus 59.0±8.33% reduction in peripheral fission; Fig. 5C, Fig. S5A).

As the peripheral fission has been implicated in the mitophagy pathway, we asked whether AMPK-induced mitochondrial fission is required for mitophagic degradation of dendritic mitochondria. To this end, we used an antibody against the autophagy marker p62 (encoded by Sqstm1) to immunostain neurons transfected with the plasmid for Mito-DsRed with or without AMPKα2 shRNA. The p62 signals formed puncta, some of which were colocalized with mitochondria in the soma and dendrites in control neurons. By contrast, the number of p62 puncta and those colocalized with mitochondria were significantly decreased in AMPKα2 knockdown cells (Fig. 5D-G; Fig. S5B). We also analyzed the distribution of LC3 (Map1lc3b), an autophagosomal membrane protein that binds to p62 and initiates the autophagy pathway. Neurons were transfected with plasmids expressing mCherry-LC3 (Sakurai et al., 2010) and Mito-EGFP with or without AMPK shRNA. The number of LC3 puncta and those colocalized with mitochondria were significantly decreased in dendrites of AMPKα2 knockdown cells (Fig. S5C-E). These results suggest that AMPK-dependent mitochondrial fission is required for the initiation of mitophagy pathway.

We further tested whether AMPK-mediated fission is required for lysosomal degradation of mitochondria by using mt-Keima, a pH-sensitive dual-excitation ratiometric fluorescent protein targeted to mitochondria. Mt-Keima exhibits an excitation peak at 440 nm at the physiological pH of the mitochondria (pH 8.0), whereas it undergoes a shift to a longer-wavelength excitation around 586 nm in the acidic lysosomal environment (pH 4.5) (Katayama et al., 2011; Sun et al., 2017). We observed a marked decrease in the number of mitochondria undergoing mitophagic degradation in AMPK KD cells (Fig. S5F,G). Taken together, these data support that AMPK-induced fission is required for mitophagic flux in developing hippocampal neurons.

AMPK deletion also affected midzone fission, which has been associated with mitochondrial biogenesis. Consistently, the total volume of mitochondria was significantly decreased in AMPKα2 knockdown cells (Fig. S3B). These data suggest that AMPK-mediated mitochondrial fission is important for mitophagy initiation and mitochondria biogenesis during dendrite development.

AMPK phosphorylates various downstream substrates involved in many aspects of mitochondrial homeostasis. It regulates mitochondrial fission by phosphorylation of MFF, which promotes its association with Drp1 at the mitochondrial outer membrane (Ducommun et al., 2015; Toyama et al., 2016). AMPK is also known to phosphorylate the autophagy-initiating kinase ULK1 (Egan et al., 2011). We thus asked whether AMPK phosphorylates these substrates in response to neuronal activity.

Consistent with our observation that AMPK activity was dynamically regulated in synchrony with neuronal activity, AMPK phosphorylation rapidly increased within 5 min of glutamate treatment (Fig. 6A,C). In contrast, TTX+APV treatment reduced the basal level of AMPK phosphorylation. Concomitantly, MFF phosphorylation at Ser146 was increased by glutamate treatment, whereas it was repressed by TTX+APV treatment. Likewise, the level of ULK1 phosphorylation at Ser555 paralleled that of AMPK phosphorylation, which was altered by manipulation of the activity states in cultured neurons. We also confirmed that the phosphorylation levels of MFF and ULK1 were upregulated by AICAR treatment (Fig. 6B,D). AMPK has been shown to phosphorylate and activate PGC1α, the master regulator of mitochondrial biogenesis. However, we detected no obvious difference in the phosphorylation level of PGC1α by Phos-tag SDS-PAGE in response to treatment with TTX+APV, glutamate, AMPK inhibitor [Compound C (CC)] or AMPK activator (AICAR) (Fig. S6A). Taken together, these results suggest that AMPK activation by neuronal activity induces mitochondrial fission and mitophagy by phosphorylation of MFF and ULK1.

To further prove that AMPK-induced mitochondrial fission is required for dendritic outgrowth, we knocked down MFF by expressing shRNAs against MFF (Fig. S6B). The phenotype of MFF knockdown mimicked that of AMPK-deficient neurons, with significantly reduced dendritic size and complexity (Fig. 6E-G) and increased mitochondrial length (Fig. 6H-K). The dendritic hypotrophy and mitochondrial elongation were rescued by a phosphomimetic version of MFF (Ser129Asp/Ser146Asp, S2D) (Fig. S6C-I). Strikingly, MFF knockdown decreased the number of p62 puncta and those colocalized with mitochondria in dendrites, supporting the notion that mitochondrial fission mediated by MFF is crucial for mitophagy in growing dendrites (Fig. 6L-N). These data indicate that mitochondrial fission and consequent mitophagy are required for dendritic outgrowth in developing neurons.

We next asked whether AMPK-mediated mitochondrial fission and mitophagy are required for mitochondrial quality control. Using the fluorescent dye tetramethylrhodamine methyl ester (TMRM, 20 nM), we asked whether the mitochondrial membrane potential (ΔΨm) was affected in the elongated dendritic mitochondria in neurons that were defective in AMPK-mediated mitochondrial fission. We found a slight but significant decrease in basal TMRM fluorescence in individual mitochondria of AMPKα2 knockdown cells, suggesting lower mitochondrial function in AMPK-deficient dendrites (Fig. 7A,B). We observed intermittent flickers of TMRM fluorescence in some dendritic mitochondria in control neurons (Fig. 7C; Movie 7). TMRM flickering is caused by the opening of the mitochondrial permeability transition pore (mPTP), an essential process for maintaining mitochondrial homeostasis by preventing an overload of Ca²⁺ and ROS in the mitochondrial matrix induced by respiration activity (Bonora et al., 2022; Buckman and Reynolds, 2001). In AMPKα2 knockdown neurons, the frequency of TMRM flickering was significantly reduced, suggestive of lower respiratory activity of mitochondria in these cells (Fig. 7D).

We then analyzed mitochondrial ROS levels (mitoROS) using the fluorescent dye MitoSOX (500 nM). MitoROS is a byproduct of electron transport chain activity, which is increased by high respiration and triggers mPTP opening (Zorov et al., 2014). MitoSOX signals formed puncta in the mitochondrial matrix, which were significantly decreased in AMPKα2 knockdown cells (Fig. 7E,F). These results suggest that AMPK-mediated mitochondrial fission and mitophagy are required for the maintenance of functional mitochondria in growing dendrites of hippocampal neurons.

To meet the rapidly growing energy demands incurred by the extension of dendritic arbors, mitochondria are actively produced and transported throughout the expanding dendritic compartments. Here, we demonstrate that neuronal activity regulates mitochondrial homeostasis through the activation of the Ca²⁺-CaMKK2-AMPK axis to support dendritic outgrowth in developing neurons. AMPK deficiency prevented mitochondrial fission and mitophagy, likely lowering mitochondrial activity. Our observations suggest that AMPK-dependent regulation of local mitochondrial homeostasis is an important mechanism of activity-induced dendritic growth besides the well-known transcription-dependent mechanism.

Mitochondrial fission is a prerequisite for the biogenesis of mitochondria and for the elimination of damaged components through the mitophagy pathway (Lewis et al., 2016; Twig et al., 2008). Recent studies have demonstrated that symmetric fission generating two equal daughter mitochondria leads to doubling of mitochondrial biomass, whereas asymmetric fission generates a smaller mitochondrion that is subjected to the mitophagy pathway (Kleele et al., 2021). We observed that both symmetric and asymmetric fissions in growing dendrites were downregulated by the inhibition of neuronal activity or AMPK, with stronger effects on asymmetric fission. We showed that neuronal activity induces phosphorylation/activation of two downstream effectors of AMPK, MFF and ULK1, which are the key regulators of mitochondrial fission and mitophagy, respectively. Mitochondria with defective fission and mitophagy in AMPK-depleted neurons exhibited multiple signatures of mitochondrial dysfunction, such as reduced ΔΨm and ROS production, suggesting that AMPK-mediated mitochondrial fission and mitophagy are required for the maintenance of healthy mitochondria in growing dendrites.

We also found that the inhibition of midzone fission was accompanied by the decrease in total biomass of dendritic mitochondria in AMPKα2 knockdown neurons, supporting the regulation of mitochondrial biogenesis by activity-induced AMPK signaling. In contrast, phosphorylation level of PGC1α, a known target of AMPK involved in the regulation of mitochondrial biogenesis, was not changed in this context. The precise molecular mechanism of the regulation of mitochondrial biogenesis by activity-induced AMPK signaling remains to be explored in future studies.

AMPK and AMPK-related kinases are expressed in both dendrites and axons of cortical neurons, but their activities seem differentially regulated in distinct subcellular compartments. Previous studies have shown that the activities of SAD-A/B and NUAK1 and their upstream kinase LKB1 are strongly biased in the axon and regulate axon specification and branching (Barnes et al., 2007; Courchet et al., 2013). The SAD kinases have been shown to regulate the fission/fusion balance of mitochondria through phosphorylation of tau in the axon (Di Meo et al., 2021; DuBoff et al., 2012). In contrast, we demonstrate that AMPK overactivation and suppression alter mitochondrial fission, and these changes are seen only in dendrites but not in the axon (Fig. 3). The activation of AMPK in dendrites and AMPK-related kinases in the axon appears to be regulated by distinct molecular mechanisms. SAD-A/B and NUAK1 have been shown to be constitutively activated in the axon by LKB1. In contrast, AMPK phosphorylation dynamically oscillates in dendrites, depending on Ca²⁺ influx and CaMKK2 activity, whereas it is independent of LKB1. AMPK activity was difficult to detect in the axon owing to the high basal FRET signals, presumably reflecting the high activity of SAD-A/B and/or NUAK1 in the axon. Nonetheless, neither neural activity nor AMPK overactivation enhanced fission of axonal mitochondria, suggesting that mitochondrial dynamics and transport are differentially regulated in the axon and dendrites by distinct AMPK-related kinases.

One interesting finding in the present study is that AMPK activity dynamically oscillates in synchrony with Ca²⁺ influx triggered by neuronal activity. Ca²⁺-dependent AMPK phosphorylation is rapidly downregulated in several tens of seconds, implying a regulatory loop involving a negative regulator of AMPK activity that counteracts CaMKK2-mediated phosphorylation. The molecular pathway of AMPK dephosphorylation is less understood, but a Ca²⁺-dependent phosphatase such as PP2A is a likely candidate, as PP2A has been implicated in AMPK dephosphorylation and inactivation in non-neuronal cells (Joseph et al., 2015). The dynamic oscillatory activation of AMPK would enable fine spatiotemporal tuning of mitochondrial dynamics in response to the local activity states in dendritic compartments.

Intriguingly, the CaMKK2-AMPK pathway regulating MFF and ULK2 activities has been implicated in the excessive mitochondrial fission and mitophagy in degenerating dendrites in AD models (Lee et al., 2022; Mairet-Coello et al., 2013). It is suggested that disruption of Ca²⁺ homeostasis overactivates the CaMKK2-AMPK pathway in these neurons. Thus, Ca²⁺-sensitive activation of AMPK could be a double-edged sword, promoting dendritic growth by regulating mitochondria turnover during normal development, while contributing to neurodegeneration by excessive mitochondrial degradation under pathological conditions.

It has been shown that genetic deletions of AMPK subunits exhibit little or no cell-autonomous defects in neuronal development in mice (Dzamko et al., 2010; Muraleedharan et al., 2020; Ramamurthy et al., 2014; Williams et al., 2011). Conditional deletion of AMPKα1/α2 (AMPKα1^−/−; AMPKα2^F/F; Emx1-Cre) exhibits no overt phenotypes in the polarization and survival of cortical neurons, except for the thinning of the cortical layer at P3 (Muraleedharan and Dasgupta, 2021; Williams et al., 2011). In contrast, acute treatment with AICAR or metformin induces rapid AMPK phosphorylation and disrupts axon formation in cultured cortical neurons (Amato et al., 2011). Thus, AMPK is thought to be dispensable for neuronal differentiation during normal development, but it exerts its effects when overactivated under stress conditions. In the present study, we show that defects in dendrite growth by acute AMPK inhibition are evident only in differentiating neurons in the first postnatal week, and that they recovered in later stages in vivo. It has also been reported that AMPK deficiency in newly born neurons significantly delays neuronal migration, but it recovers at later stages (Naito et al., 2020). It is therefore possible that the transient morphological changes might have been overlooked in previous studies. Mild and transient phenotypes by chronic inhibition of AMPK might be due to the robust multi-step regulation of mitochondrial dynamics. In mature neurons, long-term potentiation of synaptic activity causes mitochondrial fission via CaMKII and Drp1 phosphorylation (Divakaruni et al., 2018). The intracellular Ca²⁺ rise also induces a distinct type of mitochondrial fission initiated from the inner mitochondrial membrane (Cho et al., 2017). AMPK might predominantly mediate spontaneous activity before synaptic maturation, whereas other mechanisms may take over and control mitochondrial fission in more mature neurons. We observed a transient dendritic growth retardation by AMPK knockdown, but due to technical difficulties, we were unable to observe changes in mitochondrial dynamics in vivo. A recent report has demonstrated that the rate of increase in the length and activity of mitochondria is a determinant of morphological maturation of cortical neurons (Iwata et al., 2023). Further studies are needed to clarify whether AMPK mediates activity-dependent changes in mitochondrial dynamics in developing neurons in vivo.

Recent studies have demonstrated that cortical neurons in a mouse model of Huntington's Disease exhibit transient defects in dendritic morphology in the first postnatal week (Braz et al., 2022). Despite the finding that the retarded dendritic growth spontaneously reverted during the second week, the early defects were found to influence pathogenesis in adulthood, as therapeutic intervention during the first week forestalled disease symptoms in later life. Whether the transient growth defects by AMPK deficiency in early development impact mature neural circuits or whether AMPK has a distinct function in dendrites of mature neurons is one of many important questions that entail further study.

The present study highlights a new role of AMPK in mitochondrial homeostasis in normal development of neurons in addition to its major role in safeguarding neurons from energetic stress. Dynamic activation and inactivation of AMPK by neuronal activity are clearly important for the fine-tuning of mitochondrial homeostasis. Thus, future work should focus on the precise dynamics and mechanism of AMPK activation as well as inactivation for a new perspective on the physiology and pathology of AMPK signals.

All animals were treated in accordance with the guidelines of the Animal Experiment Committee of Kyoto University. ICR mice were kept in a 12 h/12 h dark/light cycle at 23±3°C and 50% humidity, with standard food and water provided ad libitum, in group housing of up to three animals per cage.

The reagents used in this study were as follows: APV (200 μM, Sigma-Aldrich, A5282), tetrodotoxin (TTX, 0.5 μM, Abcam, ab120054), AICAR (1-2 mM, AdipoGen Life Sciences, AG-CR1-0061), Compound C (dorsomorphin, 20 µM, Nacalai Tesque, 18768-04), L-glutamic acid (glutamate, 100 µM, Sigma-Aldrich, G1251), STO-609 (10 µM, MedChemExpress, HY-19805), Pim1/AKK1-IN-1 (1 µM, MedChemExpress, HY-10371), TMRM (20 nM, Fujifilm, 203-18041) and Mito-SOX (500 nM, Invitrogen, M36008).

pCAG-EGFP, pAAV-CAG-Mito-DsRed and pAAV-CAG-Mito-EGFP were constructed as previously described (Fukumitsu et al., 2015; Umeshima et al., 2007). For RNAi experiments, the target sequences for scramble control (5′-GACATTTCATCCGTTTAGTTA-3′), AMPKα1 sh#1 (5′-GGCACACCCTGGATGAATTAA-3′), AMPKα1 sh#2 (5′-GTTGTAAACCCCTATTATTTG-3′), AMPKα2 sh#1 (5′-GGTAGACAGTCGGAGCTATCT-3′), AMPKα2 sh#2 (5′-GACAATCGGAGAATAATGAAC-3′), CaMKK2 (5′-CCCTTTCATGGATGAACGAAT-3′), MFF sh#1 (5′-CTTCATTAAGACGTCAGATAA-3′) and MFF sh#2 (5′-GATCGTGGTTACAGGAAATAA-3′) (Lewis et al., 2018) were cloned into pBAsi-hH1 vector (Takara, 3220). pCAG-AMPKα1 (WT) and pCAG-AMPKα2 (WT) plasmids were created by PCR amplification of cDNAs for AMPKα1 (NCBI reference sequence: NM_001013367.3) and AMPKα2 (NCBI reference sequence: NM_178143.2) from a mouse brain cDNA library and inserted into the pCAGGS vector (RIKEN, RDB08938; Niwa et al., 1991). Resistant mutants of AMPKα1 and AMPKα2 that contained three silent mutations within the respective shRNA target sequences were generated using the Primestar mutagenesis protocol (Takara, R046A; https://catalog.takara-bio.co.jp/product/basic_info.php?unitid=U100005179). pPBbsr2-4031NES (AMPKAR-EV) and pPBbsr2-3601NES (PKA-TA) were a gift from Michiyuki Matsuda (Kyoto University, Kyoto, Japan). The pCAGplay-GCaMP6s was a gift from Yoshiaki Tagawa (Kagoshima University, Kagoshima, Japan). The pCAG-jRGECO1a plasmid was generated by subcloning the jRGECO1a sequence from pGP-CMV-NES-jRGECO1a (a gift from Masayuki Sakamoto, Kyoto University, Kyoto, Japan) into the pCAGGS vector. The pX458-CAG-dCas9KRAB-2A-GFP plasmid was generated by subcloning the CAG promoter from pCAGGS and the dCas9KRAB sequence from CRISPR-dCas9KRAB (a gift from Kuniya Abe, RIKEN BioResource Research Center, Tsukuba, Japan) into the pX458 vector (Addgene plasmid #48138). For CRISPRi experiments, the gRNA sequence for AMPKα2 (5′-TGCCGAAGGTGCCGACGCCCAGG-3′) was inserted into pX458-CAG-dCas9KRAB-2A-GFP. To generate a shRNA-resistant, phosphomimetic mutant of MFF (Ser129Asp/Ser146Asp, S2D), mouse MFF cDNA (NCBI Reference Sequence: NM_029409.3) was cloned from a mouse brain cDNA library and mutagenized using a PCR-based method. The mutant clone was tagged with myc at the N-terminus and inserted into the pCAGGS vector. pmCherry-LC3 (rat microtubule-associated protein 1 light chain 3, LC3, a marker of autophagosomes) was a gift from Ichiro Nakagawa (Kyoto University, Kyoto, Japan). pCAG-mt-Keima was generated by subcloning mt-Keima from mt-mKeima/pcDNA3 (RIKEN RDB15456) into the pCAGGS vector.

Primary cultures of hippocampal neurons were prepared as previously described with a few modifications (Wu et al., 2015). In brief, P0 mouse hippocampi were dissected, dissociated using Neuron Dissociation Solutions (Fujifilm, 291-78001) and plated on poly-D-lysine (Sigma-Aldrich, P6407)-coated coverslips at a density of ∼3.0-3.5×10⁵ cells/cm² in minimum essential medium (Gibco, 11095080) supplemented with 10% horse serum (Gibco, 26050070), 0.6% D-glucose, 1 mM sodium pyruvate (Sigma-Aldrich, S8636) and 1% penicillin-streptomycin. Three hours after plating, the medium was replaced with neurobasal medium (Gibco, 21103049) supplemented with 2% (v/v) B-27 (Gibco, 17504044) and 0.25% (v/v) GlutaMAX (Gibco, 35050061). All neurons were maintained at 37°C in 5% CO₂. 5 µM AraC (Sigma-Aldrich, C1768) was added at DIV3. Neurons were transfected at DIV3 using Lipofectamine 2000 (Thermo Fisher Scientific, 11668019) according to the manufacturer's instructions. The CRISPRi plasmid was transfected at DIV2.

HEK293T cells (Riken BioResource Center Cell Bank, RCB2202) were cultured in DMEM (Gibco, 11965-092) supplemented with 10% fetal bovine serum (Sigma-Aldrich, F7524), 100 U/ml penicillin and 100 μg/ml streptomycin (Gibco, 15140-122) at 37°C in 5% CO₂. Cells were passaged every 2-3 days and maintained in plastic tissue culture treated dishes (IWAKI, 3010-060). shRNA constructs were transfected using Lipofectamine 2000.

Cells were harvested in RIPA lysis buffer [50 mM Tris HCl (pH 7.4), 150 mM sodium chloride, 0.25% (w/v) sodium deoxycholate, 1% (v/v) Nonidet P-40, 1 mM EDTA, and protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, 78442)]. Lysates were subjected to SDS-PAGE and immunoblotting. For Phos-tag SDS-PAGE, 100 μM Mn²⁺-Phos-tag was added to the acrylamide gel (Kinoshita et al., 2005).

The primary antibodies used were: mouse anti-GFP (1:5000, Thermo Fisher Scientific, A11120); rabbit anti-AMPKα1 (1:3000, Abcam, ab32047); rabbit anti-AMPKα2 (1:3000, Abcam, ab3760); rabbit anti-AMPKα (1:1000, Cell Signaling Technology, 2532); rabbit anti-phospho-AMPKα (Thr172) (1:1000, Cell Signaling Technology, 2535); rabbit anti-MFF (1:3000, Proteintech, 17090-1-AP); rabbit anti-phospho-MFF (Ser172/146) (1:1000, Affinity Biosciences, AF2365); rabbit anti-ULK1 (1:2000, Cell Signaling Technology, 8054); rabbit anti-phospho-ULK1 (Ser555) (1:1000, Cell Signaling Technology, 5869); rabbit anti-PGC1α (1:5000, Abcam, ab191838); and HRP-conjugated mouse anti-β-actin (1:10,000, Santa Cruz Biotechnology, sc-47778 HRP). The secondary antibodies used were: HRP-conjugated goat anti-mouse IgG (1:10,000, Bio-Rad, 170-6516) and HRP-conjugated goat anti-rabbit IgG (1:10,000, Bio-Rad, 170-6515).

Pregnant mice on day 14.5 of gestation were deeply anesthetized via the intra-abdominal injection of a mixture of medetomidine, midazolam and butorphanol. Plasmid DNA diluted with saline containing 0.01% Fast Green was injected into the lateral ventricle of mouse embryos with a glass needle. Four current pulses (amplitude, 40 V; duration, 50 ms; intervals, 950 ms) were delivered with a forceps-shaped electrode (CUY650P3, NepaGene) connected to an electroporator (CUY21, NepaGene). The concentrations of the plasmids used are as follows: pCAG-EGFP (0.2 µg/µl); shRNA-scramble control (3-6 µg/µl); shRNA- AMPKα2 (3-6 µg/µl); CRISPRi gRNA-empty (5 µg/µl); and CRISPRi gRNA- AMPKα2 (5 µg/µl).

For immunocytochemistry, cultured cells were fixed with 4% paraformaldehyde in PBS and permeabilized with PBS containing 0.25% Triton X-100 (PBS-T). Cells were then blocked with blocking solution (PBS with 2% BSA) and incubated with primary antibody at 4°C overnight. After washing with PBS, cells were incubated with the secondary antibodies at 4°C overnight. The antibodies used were mouse anti-PSD95 (1:200, Abcam, ab2723), rabbit anti-AMPKα2 (1:200, Proteintech, 18167-1-AP), goat anti-rabbit Alexa Fluor 647 (1:400, Invitrogen, A21244) and goat anti-mouse Alexa Fluor 647 (1:400, Invitrogen, A21236).

For immunohistochemistry, mice at P5 or P10 were perfused with PBS followed by 4% paraformaldehyde in phosphate buffer. Their brains were removed and postfixed for 2-3 h at 4°C. After washing with PBS, the brains were embedded in 3.5% low-melting agarose in PBS. The brains were then sectioned into 200 µm-thick sagittal slices with a vibratome (NLS-AT, Dosaka EM). The sections were permeabilized in PBS-T for 15 min and blocked with 2% skim milk in PBS-T for 30 min. The sections were incubated with a chick anti-GFP (1:2000, Invitrogen, A10262) primary antibody in blocking solution at 4°C overnight, followed by incubation with a goat anti-chicken Alexa Fluor 488 (1:400, Invitrogen, A11039) secondary antibody at 4°C overnight.

Images of the fixed samples were acquired with a laser scanning confocal microscope (FV1000, Olympus) equipped with UPlanSApo 20× (NA 0.75) and UPLSAPO 40× (NA 0.95) objectives, or a spinning-disk confocal microscope (Dragonfly, Andor) equipped with Nikon PLAN APO 40× (silicone oil, NA 1.25) and Apo TIRF 100× (oil immersion, NA 1.49) objectives.

For morphometric analysis, dendrites were traced using the ImageJ plugin Simple Neurite Tracer (SNT). Total dendrite length and the number of branches were calculated using MATLAB software. For quantification of dendritic mitochondrial densities (ratio of mitochondrial area to dendrite area), acquired images were binarized and the areas filled with Mito-DsRed signals and volume marker GFP signals were quantified using ImageJ.

For p62 analysis, hippocampal neurons transfected with GFP and Mito-DsRed were fixed at DIV5 and immunostained with guinea pig anti-p62 primary antibody (1:100, Progen, GP62-C) and goat anti-guinea Pig Alexa Fluor 633 secondary antibody (1:400, Invitrogen, A21105). For LC3 analysis, hippocampal neurons were transfected with Mito-EGFP and mCherry-LC3 at DIV3 and fixed at DIV5. The number of p62 and LC3 puncta in dendrites was counted using the ImageJ plugin Find Maxima. For colocalization analysis of p62 and LC3 puncta and mitochondria, Manders’ overlap coefficient was calculated from anti-p62 signals or mCherry-LC3 signals and Mito-DsRed signals in dendrites using the ImageJ plugin JaCoP. Surface rendering was done with Imaris software.

For mt-Keima analysis, hippocampal neurons were transfected with mt-Keima at DIV3 and subjected for live imaging at DIV5 using a spinning disk confocal microscope (Dragonfly, Andor) equipped with a Apo TIRF 100× (oil immersion, NA 1.49) objective at 37°C with 5% CO₂ flow with the following filters: excitation 488 nm/emission 575-625 nm (neutral) and excitation 561 nm/emission 575-625 nm (acidic).

For calcium imaging, cultured hippocampal neurons were transfected with pCAGplay-GCaMP6s plasmid at DIV3. Time-lapse images were acquired every 3 s for 5 min at DIV5 using a spinning-disc confocal microscope (CV1000, Yokogawa) with a UPlanFLN 40× objective (oil immersion, NA 1.3, Olympus) at 37°C with 5% CO₂ flow. For time-lapse imaging of mitochondrial dynamics, cultured hippocampal neurons were transfected on DIV3 with pCAG-EGFP and pAAV-CAG-mitoDsRed plasmids. Time-lapse images were acquired every 3 s for 20 min at DIV5 using a spinning-disc confocal microscope (CV1000, Yokogawa) with a UPLSApo 100× objective (oil immersion, NA 1.4, Olympus) at 37°C with 5% CO₂ flow. Mitochondrial movement was tracked using the ImageJ plugin MTrackJ. The mitochondrial fission and fusion events were counted manually in approximately 80-µm segments of apical dendrites of approximately the same diameter. Peripheral and midzone mitochondrial fissions were defined as events in which the relative position of the fission site along the length axis was 0-25% and 25-50%, respectively. The percentage of motile mitochondria was calculated by dividing the number of mitochondria that were displaced more than 4 µm from their original positions within 20 min by the total number of mitochondria. A continuous movement of mitochondria (>1.5 µm for >6 s) was defined as one moving event, and their speed was calculated using MATLAB software.

For FRET imaging, cultured hippocampal neurons transfected with AMPKAR-EV were imaged every 3 s for 3 min at DIV5 with an inverted epifluorescence microscope (IX83, Olympus) equipped with LUCPlanFLN 20× (NA 0.45) and UplanSApo 40× (silicone oil, NA 1.25) objectives. For dual imaging of FRET and calcium, cultured hippocampal neurons were co-transfected with AMPKAR-EV or PKA-TA and jRGECO1a, and imaged every 5 s for 5 min. Fluorescence images were acquired with the following filters and mirrors: excitation filters FF02-438/24-25 (for CFP/ YFP); a dichroic mirror FF458-Di02-25x36 (for CFP/YFP); emission filters FF01-483/32-25 (for CFP); FF01-542/27-25 (for YFP); and filter sets U-FMCHE (excitation 565-585 nm/dichroic mirror 595 nm/emission 600-690 nm) (for jRGECO1a).

For FRET ratio analysis, after background subtraction, the YFP/CFP fluorescence intensity ratio was calculated in the region of interest (ROI) of the proximal dendritic region. The FRET_baseline for each cell was calculated as the mean of the lower 10% of values across all time points. The FRET signal (YFP/CFP ratio)/FRET_baseline was plotted along the time axis, and the FRET peak frequency was quantified using the MATLAB findpeaks function with a threshold for peak heights greater than 0.05. The cross-correlation function was calculated followed methods described previously (Kunida et al., 2012) using MATLAB software.

For TMRM imaging, cultured hippocampal neurons transfected with Mito-EGFP were incubated with 20 nM TMRM 1-2 h prior to imaging. Time-lapse images were acquired every 5 s for 10 min at DIV5 with a spinning-disc confocal microscope (CV1000, Yokogawa) equipped with a UPLSApo 100× objective (oil immersion, NA 1.4, Olympus).

For TMRM intensity analysis, after background subtraction, integrated intensity of TMRM in the ROI of 4×4 pixels (0.1024 µm²) in individual mitochondria was measured. TMRM flickering, defined as a transient loss of TMRM intensity to the background level, was scored as the number of flickering events that occurred within 10 min in the soma and proximal dendrites, divided by the mitochondrial area.

For quantification of mitochondrial ROS, cultured hippocampal neurons transfected with Mito-EGFP at DIV3 were incubated with 500 nM Mito-SOX for 15 min at DIV5 and then the Mito-Sox was washed out. Time-lapse imaging was initiated 18-20 h after treatment using a spinning-disc confocal microscope (CV1000, Yokogawa) equipped with a UPLSApo 100× objective (oil immersion, NA 1.4, Olympus). Mito-SOX intensity in individual mitochondria [(F_mitochondria/F_background)/(mitochondrial area)] was quantified using ImageJ.

Statistical analyses were done using MATLAB and R. Anderson–Darling test and Bartlett's test were used to test for normality and to determine appropriate parametric or non-parametric tests. Wilcoxon rank-sum test was used for the analysis of the dendrite morphology (Figs 1D,E, 2G,H and 6F,G; Fig. S2H,I), mitochondrial length (Figs 1H-L, 3F,G,J-L and 6I-K; Figs S1H-J and S3D-F), mitochondrial speed (Fig. S1C), mitochondrial area (Fig. S3B), colocalization of p62 or LC3 and mitochondria (Figs 5F,G and 6M,N; Fig. S5D,E), mt-Keima (Fig. S5G), TMRM intensity and flickering events (Fig. 7B,D), and Mito-ROS (Fig. 7F). Kolmogorov–Smirnov test was used for AMPK FRET frequency analysis (Fig. 4B,D,I,J,L). Paired two-tailed t-test was used to analyze mitochondrial fission and fusion frequency before and after drug treatment in the same cells (Figs 1M and 5B). Unpaired two-tailed t-test was used for the analysis of mitochondrial fission and fusion frequency in different cells (Figs 3H and 5C), mitochondrial motility (Fig. S1B) and mitochondrial density in dendrites (Figs S1D and S3A). Kruskal–Wallis test with Bonferroni test was used for multiple comparisons of dendrite and mitochondrial morphology (Figs 2C,D and 3C-E; Figs S2D,E and S6D,E,G-I). Kruskal–Wallis test with Steel–Dwass test was used for multiple comparisons of western blotting (Fig. 6C,D).

We thank Dr Michiyuki Matsuda for the AMPKAR-EV plasmid; Dr Yoshiaki Tagawa for the GCaMP6s plasmid; Dr Masayuki Sakamoto for the jRGECO1a plasmid; Drs Kuniya Abe, Yusuke Kishi, Shinnosuke Suzuki and Ryuta Kinoshita for the CRISPR-dCas9KRAB plasmid; Drs Ichiro Nakagawa and Takashi Nozawa for the pmCherry-LC3 plasmid; Dr Atsushi Miyawaki for the mt-Keima plasmid; iCeMS Analysis Center at Kyoto University for technical assistance with imaging using a Dragonfly microscope; Drs Takaki Miyata, Mayumi Okamoto and Daniel Packwood for advice; and Dr James Hejna for critical reading of the manuscript.

This article has an associated ‘The people behind the papers’ interview with some of the authors.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201930.reviewer-comments.pdf