Local glycolysis fuels actomyosin contraction during axonal retraction

Brain wiring is an essential developmental process in which neurons project axons that connect to their proper target to form functional circuits. Axonal guidance is mediated by a complex orchestration of attractive and repulsive cues that drive the growth cone to its final destination (Lokmane and Garel, 2014; Chédotal and Richards, 2010). The physiological importance of this developmental process is highlighted by several neurological disorders characterized by aberrant neuronal connections (Van Battum et al., 2015). Several mutations in genes involved in axonal guidance have indeed been found in human neurological disorders (Engle, 2010; Van Battum et al., 2015). In addition, non-genetic factors such as inflammation, maternal stress, or prenatal drug exposition can also lead to aberrant neuronal connectivity (Tortoriello et al., 2014; Lautarescu et al., 2020; Rasmussen et al., 2019).

Growth cones at the tip of the axons are dynamic structures that express receptors for the guidance cues, which for the main part are chemotrophic factors that induce growth toward or away from the source, such as the secreted netrins and slits or the membrane-bound ephrins and semaphorins (Ye et al., 2019; Seiradake et al., 2016; Blockus and Chédotal, 2016). Other factors have been identified that contribute to the complexity of axonal guidance and brain wiring. For example, cannabinoids and lysolipids function as repulsive cues that induce growth cone collapse and retraction of the axon (Gaffuri et al., 2012; Berghuis et al., 2007; Mulder et al., 2008; Saez et al., 2020; Wu et al., 2010; Ye et al., 2019; Guy and Kamiguchi, 2021; Quarta et al., 2017; Roland et al., 2014). Guidance factors activate several signaling pathways that cause cytoskeletal and membrane remodeling, resulting in an array of morphological changes, such as growth, arrest, collapse, retraction, turning, or branching. This motility of the growth cone is possible through the dynamic F-actin networks and myosin motors that allow mechanotransduction of the signals necessary for axonal pathfinding (McCormick and Gupton, 2020). Therefore, the shape of the growth cones is diverse and depends on the type of neurons, developmental stage, and environment; it can range from a simple single thin filipodium to a sizable display of filipodia and lamellipodia (Ye et al., 2019).

The endocannabinoid system is involved in axonal fasciculation and guidance by chemorepulsion of axonal growth cones during corticogenesis (Gaffuri et al., 2012; Berghuis et al., 2007; Mulder et al., 2008; Saez et al., 2020; Wu et al., 2010). The endocannabinoid 2-arachidonoglycerol (2-AG) is secreted from the subventricular zone (SVZ) and the cortical plate during embryonic development (Maccarrone et al., 2014), which prevents growing corticofugal axons expressing the cannabinoid receptor 1 (CB1) to enter the SVZ from the intermediate zone (Roland et al., 2014). Activation of CB1 at the growth cone activates the Ras homolog gene family member A (RHOA) and the Rho-associated coiled-coil-containing protein kinase (ROCK), resulting in axonal retraction mediated by actomyosin contractility (Berghuis et al., 2007; Roland et al., 2014; Backer et al., 2018). Other cues also induce axonal retraction through the RHOA/ROCK/actomyosin pathway, such as activation of the Eph/ephrin reverse signaling (Takeuchi et al., 2015) or various lysolipids (Quarta et al., 2017; Obara et al., 2011; Zhang et al., 2003). Therefore, the contractile actomyosin complex is a major player in growth cone collapse and retraction, acting downstream of repulsive guidance cues and RHOA/ROCK (Wahl et al., 2000; Wylie and Chantler, 2003; Gallo, 2006; Brown et al., 2009; Murray et al., 2010; Roland et al., 2014).

Non-muscular myosin II (NMII) is the motor component of the actomyosin complex. NMII is formed by three pairs of chains: two heavy chains containing the ATPase catalytic site and the F-actin binding domain, two regulatory light chains (RLCs), and two essential light chains that stabilize the structure (Garrido-Casado et al., 2021). Phosphorylation of RLCs downstream of ROCK causes a change in NMII conformation, self-association into bipolar filaments, and binding to F-actin forming the actomyosin complex. This contractile machinery exerts its mechanical tasks by consuming energy by the NMII ATPase catalytic site (Geeves, 2016). The three isoforms of NMII are expressed in neurons and the axons, but only NMIIA, which is enriched in the shaft and central domain, is thought to be responsible for axonal retraction (Miller and Suter, 2018). Accordingly, it was previously shown that in response to cannabinoids, NMII is phosphorylated, therefore active in the axonal region adjacent to the growth cone (Roland et al., 2014). However, the source of the ATP that fuels NMII contractility and axonal retraction is not known.

Neuronal bioenergetics strongly depends on mitochondrial metabolism (Kim et al., 2019; Yellen, 2018). Despite that, local glycolysis provides ATP for several active processes that require a transient and fast increase in energy (Díaz-García et al., 2017; Ivannikov et al., 2010; Zala et al., 2017; Yellen, 2018). Compared with mitochondrial oxidative phosphorylation, glycolytic enzymes can be localized close to ATP-consuming effectors. Stimulation of neuronal excitability immediately triggers a transient increase in aerobic glycolysis in response to activity of Na⁺/K⁺ pump activity (Díaz-García et al., 2017; Díaz-García and Yellen, 2019; Meyer et al., 2022). Also, a tight coupling between ATP production by glycolytic enzymes and consumption occurs for processive motors, such as kinesin and dynein for fast axonal transport of vesicles (Zala et al., 2013; Hinckelmann et al., 2016), and for the proton pumps at synaptic vesicles (Ishida et al., 2009; Ikemoto et al., 2003). In these cases, glycolytic enzymes form complexes with the ATP-consuming proteins and ATP is provided by direct channeling without an exchange with the cytosolic ATP pool. Assembly of proteins participating in the same functional pathway in complexes or super-complexes was selected during evolution and present multiple advantages, such as counteracting the dispersion of enzymes and substrates in the cell and allowing fine regulation of activity (Lyon et al., 2021).

When contractile actomyosin is engaged in axonal retraction, it requires a rapid and continuous supply of ATP. While little is known about the energy source for actomyosin activity, several observations suggest that glycolysis may be associated with contractile forces in various cellular contexts. In 1988, George et al. observed that pharmacological inhibition of glycolysis blocked the shortening of amputated axons (George et al., 1988). Stress fibers, which contain actomyosin, are enriched in the key glycolytic enzyme glyceraldehyde 3-phosphate dehydrogenase (GAPDH; Schmitz and Bereiter-Hahn, 2002), which can bind to F-actin (Waingeh et al., 2006). Other pieces of evidence show that the glycolytic enzyme pyruvate kinase PKM2 phosphorylates NMII during cytokinesis (Jiang et al., 2014) and that the glucose transporter GLUT1 is structurally linked to actomyosin at tight junctions (Salvi et al., 2021). Recently, it was shown that five representative enzymes from the investment and payoff phases of glycolysis are present at the growth cone, mainly in the central domain and filipodia, in sensory neurons (Ketschek et al., 2021). These data indicate a possible functional link between NMII activity and glycolysis, which may provide ATP locally and on demand for actomyosin contractility. The presence and putative importance of such a close coupling has not been demonstrated yet. Therefore, here, we tested this hypothesis and studied the energetic metabolism of actomyosin contraction during axonal retraction.

Axonal dynamics was monitored in neurons expressing Lifeact-mCherry, which is a red fluorescent probe that binds to F-actin, in a media containing glucose and pyruvate, as previously described (Roland et al., 2014). As expected, addition of the synthetic cannabinoid WIN 55,212-2 (WIN) to growing axons activated CB1-induced growth cone collapse and axonal retraction in <2 min (Fig. 1 A and Fig. S1). We used oligomycin to inhibit the mitochondrial ATP synthase and 2-deoxyglucose (2DG), a non-metabolizable glucose analogue to block glycolysis. While inhibiting mitochondria or glycolysis by addition of oligomycin or 2DG caused growth cone collapse induced by WIN, only 2DG restrained axonal retraction (Fig. 1 A and Fig. S1). Simultaneous addition of oligomycin and 2DG to the medium blocked WIN-induced retraction similarly to 2DG inhibition (Fig. 1 A). These results suggested that only glycolysis is required for axonal retraction induced by WIN. Furthermore, axonal retraction was completely blocked when the assay was performed in neurons incubated for 1 h in media containing pyruvate as a unique energetic substrate, while retraction was normal when incubated for 1 h with glucose (Fig. 1 B). Similar results were obtained by blocking the two metabolic pathways 20 min before inducing axonal retraction and by using additional inhibitors, iodoacetate for glycolysis and sodium azide for mitochondrial respiration (Fig. 1 C). The physiological CB1 ligand endocannabinoid 2-AG induced retraction of smaller amplitude compared with WIN, similarly to what was previously described (Roland et al., 2014), but only the addition of 2DG inhibitor blocked axonal retraction (Fig. 1 D). To substantiate these observations, we tested the effects of physiological lysophosphatidylinositol (LPI), the endogenous agonist of G protein-coupled receptor 55 ligand, in cortical neurons and observed comparable results (Fig. 1 E). Finally, we tested direct activation using calpeptin, which activates RHOA downstream of CB1 allowing by-passing the heterogeneity of the receptor expression in neurons after transfection. Again, 2DG, but not oligomycin, blocked axonal retraction (Fig. 1 F). These results show that glycolysis, but not mitochondria, is required for axonal retraction independently of the stimulus.

Glycolysis may provide ATP for axonal retraction by two distinct mechanisms: from the cytosolic pool or by local production in a multienzyme complex where the ATP is directly transferred from the ATP-generating glycolytic enzymes to the active site of NMII (Fig. 2 F). To test which mechanism fuels actomyosin contraction in response to cannabinoids in embryonic rat hippocampal cultures, we used a Seahorse analyzer to simultaneously quantify oxygen consumption rate (OCR) and extracellular acidification rate (ECAR), which measure respiration and glycolysis, respectively. As expected, within 5 min we observed a reduction in the OCR with oligomycin treatment and a reduction in the ECAR with 2DG treatment (Fig. 2 A). Analysis of the effects of these two drugs on the total cellular ATP levels for 30 min showed: (1) significant depletion at all time points triggered by oligomycin (P < 0.01 or P < 0.001, two-way ANOVA); (2) transient change with 2DG, significant compared with the control at 4 min after treatment (P < 0.01, two-way ANOVA) but not at 2, 8, 16, and 32 min; and (3) complete consumption after 15 min treatment with the two drugs (Fig. 2 A). This indicated that in the neuronal cultures, the majority of cellular ATP was produced by mitochondria and not by glycolysis and that exhaustion of ATP levels needed inhibition of both metabolic pathways. To gain information on the local subcellular distribution of ATP at the growth cone and the adjacent axonal shaft under metabolic inhibition and axonal retraction, we used the ratiometric ATP/ADP probe Perceval. We have successfully used this reporter in axons depleted from mitochondria and showed that the sensitivity of this probe was sufficient to monitor metabolic perturbation in small structures (Zala et al., 2013; Hinckelmann et al., 2016). We acquired confocal microscopy images at rest and 2 min after drug treatments of hippocampal neurons transfected with Perceval. Growth cones and adjacent axons were binarized to obtain a mask in which the ratio of ATP to ADP was calculated (Fig. 2 B). As expected, and in agreement with the results of total ATP measures (Fig. 2 A), oligomycin inhibition of mitochondrial ATP generation led to a decrease in ATP and an increase in ADP signals resulting in a decrease of the ATP/ADP ratio at the growth cone and adjacent shaft, which was quantified by the shift of the distribution of pixels towards lower intensities. In contrast, inhibition of glycolysis by 2DG did not affect ATP, ADP signals, or the ratio, resulting in a similar pixel distribution. These results indicate that actomyosin contractility is not correlated with the cytoplasmic ATP/ADP ratio and that glycolysis inhibition does not block actomyosin contractility through modification of ATP levels.

Next, we tested for the impact of cannabinoid signaling on cellular energetic metabolism. Two previous studies reported opposite effects of cannabinoids: an increase in mitochondrial respiration in rat brains and a decrease in mitochondrial respiration and ATP levels in mice brain and primary neurons (Hebert-Chatelain et al., 2016; Costa and Colleoni, 2000). We tested the effect of WIN in human HEK293 cells stably expressing CB1 and observed a reduction of ∼10% in mitochondrial respiration and a small but not significant increase in glycolysis at the whole-cell level (Fig. 2 C). Similar results were obtained by treatment of hippocampal neurons with calpeptin (Fig. 2 D). These results show that WIN and calpeptin do not inhibit glycolysis to an extent that could explain the lack of axonal retraction. In addition, they also show the absence of a substantial increase in glycolysis at the whole-cell level, which is in favor of the energetic coupling hypothesis. In addition, calpeptin did not induce a modification of ATP and ADP levels or the ATP/ADP ratio at the growth cone and near the axonal shaft (Fig. 2 E). These data also show that the extent of axonal retraction is not correlated to the total or the local ATP levels (Fig. 1 A and Fig. 2 A). A reduction of 50% in total ATP was observed in the first minutes of oligomycin treatment compared to <10% in ATP with 2DG treatment, but only 2DG inhibited completely axonal retraction. Moreover, a blockade of mitochondrial functions or glycolysis before induction of retraction gave similar results. Considering these results, we could expect that repulsive cues would induce glycolysis to increase ATP levels necessary for axonal retraction. However, in our conditions, the activation of the CB1/RHOA repulsive pathway slightly reduced mitochondrial respiration but did not significantly affect either global glycolysis rate or local ATP levels. In conclusion, these results show that simultaneous generation of new ATP by glycolysis is necessary to fuel CB1/RHOA-mediated actomyosin contraction and axonal retraction. This is in favor of the second proposed mechanism where new synthesized ATP is directly transferred from glycolytic enzymes to the active site of NMII without exchange with the cytoplasm (Fig. 2 F).

Glycolysis is a 10-enzyme catabolic pathway from which two enzymes from the investment phase (aldolase and triosephosphate isomerase [TPI1]) and three enzymes from the payoff phase (GAPDH, enolase, and pyruvate kinase [PK]) were extensively quantified at the growth cones of sensory neurons (Ketschek et al., 2021; see diagram in Fig. S2). The ATP-producing payoff phase comprises five enzymes. The key enzymes are GAPDH, which is the first enzyme and frequently used as a surrogate marker for the entire pathway, and phosphoglycerate kinase (PGK) and PK, which catalyze the reactions that generate ATP. We performed a retraction assay using hippocampal neurons that expressed GAPDH-eGFP to follow the distribution and dynamics of GAPDH (Fig. 3 A and Video 1). We observed the presence of GAPDH-eGFP at the growth cone and distal axon during the growth phase, collapse, and retraction, which colocalized with F-actin (visualized by Lifeact-mCherry). This localization was confirmed by confocal microscopy on endogenous GAPDH and F-actin (Fig. 3 B). We showed previously that induction of retraction increased the amount of phosphorylated RLC in the axonal shaft (Roland et al., 2014). Therefore, we investigated the colocalization of NMII, GAPDH, and PGK with F-actin, which is enriched at the growth cone, and with microtubules, which are present in the axonal shaft and the central domain of the growth cone. For this, hippocampal cultures were treated for 5 min with DMSO (control or growth conditions) or calpeptin (retraction conditions; Fig. 3 C), fixed, and processed for immunofluorescence. NMII was enriched in the central and transition domains of the growth cone and in the axonal shaft in control and retraction conditions. GAPDH and PGK shared the same localization, colocalizing with actin at the growth cone and with microtubules in the axonal shaft (Fig. 3 C). Next, we asked if the remaining three enzymes of the payoff phase (phosphoglycerate mutase, enolase, and PK) were also present at the growth cone and adjacent axonal shaft and colocalized with F-actin. Indeed, all enzymes colocalized with F-actin and were detected at the growth cone and adjacent axonal shaft both in growth and retraction conditions (Fig. 3 C and Fig. S2). In summary, the five enzymes of the payoff phase of glycolysis are located at the growth cone and adjacent axonal shaft and colocalize with F-actin during axonal retraction.

To confirm the requirement of glycolysis for axonal retraction, we used RNA interference to reduce expression of GAPDH and consequently, to impair the downstream steps of glycolysis, which are responsible for ATP production. We tested the efficacy of two different small interfering RNAs (siRNAs) in decreasing GAPDH expression at the growth cone. For this, hippocampal neurons were cotransfected with siRNA (against GAPDH or a universal control) and a plasmid expressing LifeAct-mCherry for visualization of transfected growth cones, fixed, and immunostained with an anti-GAPDH antibody. GAPDH expression was evaluated by measuring the mean fluorescence intensity at the growth cone region defined by F-actin, as detected by LifeAct-mCherry fluorescence (Fig. 4 A). A significant drop but not a full extinction in GAPDH signal intensity was detected with both siRNA-GAPDH as compared with control siRNA, indicating a decrease in GAPDH expression (Fig. 4 B). Next, we tested the effect of siRNAs in the axonal retraction assay and observed a significant reduction in the amplitude of retraction with the two siRNA-GAPDH as compared with the control siRNA (Fig. 4, C and D). The partial effect of siRNAs relative to drug inhibition may be the result of residual GAPDH expression as detected in Fig. 4, A and B. Interestingly, the siRNA had also an effect on the axonal growth (Fig. 4, C and E). These results show that GAPDH expression at the growth cone and near shaft is necessary for axonal retraction.

Having shown that axonal retraction required glycolytic enzymes, we asked if these enzymes could form a molecular complex with NMII, which is the molecular motor and ATPase of actomyosin. Therefore, we performed NMII immunoprecipitation using an antibody that recognizes NMII heavy chains from protein extracts prepared from cortical cultures. Unsupervised analysis by mass spectrometry (MS) of proteins from NMII and GFP precipitate revealed 450 putative NMII-interacting proteins from 1,435 identified (Table S1). These proteins were selected based on a peptide sequencing coverage over 10% of the sequence, a higher peak area, and a higher NMII specificity (i.e., abundance in NMII precipitate as compared to GFP precipitate). Interestingly, glycolysis was in the top 10 functional terms identified by Gene Ontology (GO) analysis that were enriched in this data (Tables 1 and S1). As expected, heavy and light chains of NMII and actin were the most abundant proteins detected and increased by hundreds of fold in magnitude in NMII precipitate compared with GFP precipitate (Tables 2 and S1). Interestingly, effectors of CB1/RHOA/ROCK signaling and axonal guidance were specifically pulled-down with NMII (Tables 2 and S1). Two glycolytic enzymes of the investment phase (glucose-6-phosphate isomerase and TPI1, the fifth and last enzyme of this phase) and five of the payoff phase were also identified as putative NMII interactors by MS (Tables 2 and S1), and confirmed by western blot for GAPDH and the two ATP-producing enzymes, PGK and PK (Fig. 5 A). This suggests that production of ATP may occur locally, in a molecular complex, to fuel NMII activity and actomyosin contraction.

To visualize this putative complex in neurons, we used the proximity ligation assay (PLA), which detects two proteins at <40 nm distance, suggesting direct protein–protein interaction (Fredriksson et al., 2002; Alam, 2018). Glycolytic enzymes and actin are abundant proteins; furthermore, actin is present as monomers (G-actin) or polymers (F-actin) in the cell. Therefore, to visualize the proteins specifically associated with the cytoskeleton, neurons were permeabilized with detergent before fixation to wash out all unbound cytoplasmic proteins and keep intact F-actin, and microtubules and their associated proteins were stabilized by phalloidin and taxol. NMII-actin was detected close to F-actin in basal conditions and, as expected, the PLA signal increased significantly after a 5-min treatment with calpeptin (Fig. 5, B and D), indicating the formation of the actomyosin complex. GAPDH-actin was also proximal to actin and, similarly to NMII, recruitment of GAPDH to F-actin was increased by calpeptin treatment (Fig. 5, B and D). This increase was specific to actin cytoskeleton as PLA fluorescence signal between tubulin and GAPDH was not increased by calpeptin (Fig. 5 B). Finally, the couples NMII-GAPDH and NMII-PGK were also observed in close proximity in basal conditions and, most interestingly, a large increase was observed in both PLA fluorescence signals following calpeptin treatment (Fig. 5, B and D). To quantify the localization of this putative association at the growth cone and adjacent axonal shaft, we represented the position of the PLA NMII-PGK punctae relative to the distal tip (Fig. 5 C). Interestingly, the PLA punctae were regularly distributed from the tip of the growth cone to the adjacent shaft, with a mean of 8.7 µm from the axonal tip corresponding to the proximal part of the growth cone. This localization is consistent with the localization of phosphorylated NMII previously described during axonal retraction (Roland et al., 2014). In other terms, recruitment of GAPDH and PGK to NMII and F-actin was enhanced at the growth cone and adjacent axonal shaft under retraction conditions. Altogether, these results suggest the formation of a glycolytic–actomyosin complex during axonal retraction.

To test the hypothesis that glycolysis is required locally to power axonal retraction, we took advantage of microfluidic devices to isolate axonal growth cones from dendrites and cell bodies (Taylor et al., 2005). The polydimethylsiloxane (PDMS) microchip (Fig. 6 A) consists of two chambers (Fig. 6, A 1 and 3) that are connected by long channels (Fig. 6, A 2) in which axons can extend from the proximal chamber to the distal chamber. To prevent the diffusion of drugs from the growth cone and adjacent axonal shaft to the rest of the axons and cell bodies, we created a pressure gradient by reducing the volume of media in the distal chamber. This was confirmed by the injection of fluorescein in the distal chamber and absence of leakage into the microchannels (Fig. 6 B). Video microscopy was performed when axons started to grow into the distal chamber (Fig. 6, A 3). Axonal retraction was induced by injection of LPI in the distal chamber (Fig. 6 C). Inhibition of axonal retraction was observed when glycolysis was simultaneously blocked by 2DG injection in the growth cone chamber, but not by 2DG injection in the somatic chamber (Fig. 6 C). These results show that local glycolysis is required for axonal retraction. Altogether, our results support the model of energetic coupling between glycolytic enzymes and actomyosin contractility (Fig. 6 D).

Here, we investigated how energy was supplied to NMII, a processive motor that converts ATP into mechanical forces used for actomyosin contractility. We found, using complementary approaches, that glycolytic enzymes and actomyosin form a dynamic functional complex necessary for axonal retraction in response to physiological ligands (endocannabinoids and LPI). Using microfluidics, we confirmed that this energetic coupling between glycolysis and actomyosin is localized to the region encompassing the growth cone and near axonal shaft, which is consistent with the localization of phosphorylated NMII previously described during axonal retraction (Roland et al., 2014). This type of energetic coupling, where ATP is produced locally and on demand, was described before for other dynamic neuronal functions that need an instant and high supply of ATP (Kim et al., 2019; Zala et al., 2017; Yellen, 2018). These results contribute to the accumulating evidence showing that essential functions of a neuronal cell, such as firing (Díaz-García et al., 2017; Ivannikov et al., 2010; Ikemoto et al., 2003; Ishida et al., 2009), axonal transport of vesicles (Zala et al., 2013), and as shown here, retraction of processes, require simultaneous breakdown of glucose by glycolysis, the cellular pool of mitochondrial ATP not being consumed.

During brain development, the precise arrangement of neuronal connections is established by neurite and axonal outgrowth and retraction. Retraction of axons occurs during development to eliminate improperly addressed or supernumerary projections, or in the adult in case of injury (Cowan et al., 1984; Pease and Segal, 2014; Gan and Lichtman, 1998; Luo and O’Leary, 2005). Growth cone pathfinding at the tip of axons is a succession of phases of growth, pause, collapse, retraction, and bifurcation (Dumoulin et al., 2021; Wen and Zheng, 2006). It is recognized that mitochondria are required for axonal growth and branching (Courchet et al., 2013; Smith and Gallo, 2018). However, it was recently found that glycolytic enzymes are localized at the growing cone and that glycolysis is necessary for sensory axonal extension (Ketschek et al., 2021; Courchet et al., 2013). This requirement for glycolysis may result from a more efficient energy channeling compared with mitochondria, resulting from a higher rate of ATP production due to the physical proximity of ATP production (Pfeiffer et al., 2001). Therefore, glycolysis could play a more important role in growth cone dynamics than previously assumed.

Physical interaction between glycolytic enzymes and the cytoskeleton has been known for 40 yr (Knull and Walsh, 1992; Shearwin et al., 1990; Masters, 1984), but its physiological significance may have been masked by the abundance of these proteins. Cytoskeleton architecture and contractility are essential for cellular division, migration, and differentiation. These are energy-demanding processes, and it is likely that the required ATP is provided by a functional interaction with glycolytic enzymes, as reported here. In agreement with this idea, a recent study showed that energy production by aerobic glycolysis in normal and cancer cells is regulated in response to the mechanical properties of the environment, which is sensed by the actomyosin cytoskeleton (Park et al., 2020). Activation of the RHOA/ROCK signaling pathway by multiple signals and receptors results in actomyosin contractility and cytoskeleton remodeling (Amano et al., 2010; Narumiya and Thumkeo, 2018; Amin et al., 2013). A number of these signals and effectors that activate RHOA, for example, ephrins and trio rho guanine nucleotide exchange factor (TRIO), are also implicated in axonal guidance (Backer et al., 2018; Shamah et al., 2001; Wahl et al., 2000). Also, several studies showed that a long-term activation of RHOA and ROCK induces glucose uptake and glycolysis (Wu et al., 2021; Yang et al., 2016; Begum et al., 2002; Furukawa et al., 2005). Therefore, an exciting hypothesis is that RHOA/ROCK may be a key hub that simultaneously triggers rapid energy-consuming structural changes and direct channeling of ATP generated by glycolysis. Interestingly, we found that several of these proteins interact with NMII by MS analysis.

Glycolytic enzymes can be associated in transient complexes by protein–protein interactions to form supramolecular complexes, allowing an efficient substrate channeling or tunneling (Zhang and Fernie, 2021; Sweetlove and Fernie, 2018). In our specific case, there are two potential mechanisms that could explain how glycolytic enzymes are recruited to the actomyosin complex. GAPDH and PGK interact with F-actin but also with the monomeric ß-actin (Havugimana et al., 2022; Go et al., 2021; Waingeh et al., 2006; Schmitz and Bereiter-Hahn, 2002). RHOA induces the formation of the actomyosin complex, which involves the activation of NMII and the assembly of actin filaments. Therefore, one possible mechanism is the hijacking of glycolytic enzymes by actin during F-actin polymerization. A second possible mechanism could involve post-translational modifications of glycolytic enzymes. Many post-translational modifications of GAPDH have been identified, which play a fundamental role in the diverse cellular functions of this moonlighting protein (Sirover, 2020, 2021).

One of the current hypotheses for the emergence of neurodevelopmental disorders is dysfunctional brain energetic metabolism (Oyarzábal et al., 2021; Kim et al., 2019). Therefore, understanding how energy is generated and consumed during axonal navigation can help predict the effect of defects in glycolysis or mitochondrial ATP in brain connectivity. We and others observed that blocking the endocannabinoid system by the specific CB1 receptor antagonist/inverse agonist AM281 by RHOA knockout or the NMII inhibitor blebbistatin affected axonal guidance with an increased number of misrouted axons in the SVZ (Roland et al., 2014; Cappello et al., 2012). We expect that a similar effect would be observed by decreasing glycolysis during embryonic development.

In conclusion, in this study, we presented evidence that a higher-order complex actomyosin-glycolysis at the growth cone is necessary for the translation of the repulsive signal of endocannabinoids or LPI and activation of RHOA into axonal retraction (Fig. 6 D). The ATP necessary for NMII contraction is generated directly on demand by the glycolytic enzymes associated with actomyosin.

Human embryonic kidney cells (HEK293, RRID:CVCL_0045, from CRL-1573; ATCC) stably expressing FLAG-CB1-GFP (Leterrier et al., 2006) were cultivated in DMEM (Thermo Fisher Scientific) with 10% FBS (Thermo Fisher Scientific). Hippocampal and cortical neuronal cultures were derived from combined male and female rat embryos at embryonic day 18 (E18; Sprague Dawley, RRID:RGD_734476, from Janvier Labs). The different brain structures were dissected in cold DMEM (Thermo Fisher Scientific) under a stereomicroscope and tissue was dissociated enzymatically in HBSS (Thermo Fisher Scientific) with trypsin (0.1%; Sigma-Aldrich) and DNase I (0.001%; Roche) for 15 min at 37°C. Trypsin was inactivated by addition of trypsin inhibitor (0.1%; Thermo Fisher Scientific). The cells were dissociated by delicate trituration with a pipette tip and the resulting suspension was centrifuged. The cells were resuspended in neurobasal media (Thermo Fisher Scientific) supplemented with 2% B-27 (Thermo Fisher Scientific), 1% penicillin/streptomycin (Thermo Fisher Scientific), and 2 mM L-glutamine (GlutaMAX, Thermo Fisher Scientific; NB+ media), and plated on the required surface previously coated with poly-D-lysine. Cortical neuronal cultures were kept in this media. For hippocampal neuronal cultures, media was changed after a period of 2 h to allow attachment to NB+ media conditioned for 24 h on a rat glial culture (NB+C media). Cultures were kept in a humidified incubator (95% air, 5% CO₂) at 37°C.

Neuronal cultures were transfected between days in vitro (DIV) 4 and DIV8 with lipofectamine 2000 (Thermo Fisher Scientific) as previously described (Roland et al., 2014). Plasmids used were: GAPDH-eGFP (Dastoor and Dreyer, 2001), CB1-CFP (Roland et al., 2014), Lifeact-mCherry (Riedl et al., 2008; Roland et al., 2014), and Perceval (Berg et al., 2009). SiRNA targeting sequences were rat GAPDH (1) 5′-CGU​AUC​GGA​CGC​CUG​GUU​A-3′, rat GAPDH (2) 5′-CUU​CUC​UCG​AAU​ACC​AUC​A-3′, or Universal Negative Control #1 (guaranteeing a minimal reduction of 75%; Sigma-Aldrich).

Hippocampal neurons were plated in poly-D-lysine (Sigma-Aldrich) coated 18-mm coverslips in 12-well plates at a density of 50,000 cells per well and transfected with lipofectamine 2000 at DIV5–7, which corresponds to 2–3 d previous to axonal retraction assay. The coverslips were mounted in a Ludin chamber with prewarmed modified E4 medium containing glucose (Sigma-Aldrich) and/or pyruvate (Thermo Fisher Scientific; Roland et al., 2014). The chamber was placed in a microscope-incubator set at 37°C and time-lapse video-microscopy on 25–40 LifeAct-mCherry–positive growth cones was recorded with an inverted microscope (NikonN-STORM microscope [Nikon Instruments] equipped with an Ixon DU-897 camera [Andor] and controlled with NIS-Elements) at a frequency of 1 image/min with 20× objective lens plus an additional 1.5 magnification lens for quantification analysis. The drugs injected during the acquisition were calpeptin (Cytoskeleton, Inc.), WIN 55,212-2 (Tocris), 2-AG (Tocris), LPI (Sigma-Aldrich), oligomycin (Sigma-Aldrich), and 2DG (Sigma-Aldrich). For localization of GAPDH-eGFP during retraction, videos were acquired with a 100× objective. Compounds were injected during acquisitions and blind analyses of the videos were performed with Fiji (RRID:SCR_002285; Schindelin et al., 2012). Stacks were first corrected for movement with a stabilizer plug-in (https://www.cs.cmu.edu/∼kangli/code/Image_Stabilizer.html) and growth cone dynamics were analyzed with the KymoToolbox plug-in (Zala et al., 2013) and manual tracking.

Microchips composed of a piece of elastomer (PDMS) sealed on Flurodish Tissue Culture Dish with Cover Glass Bottom were manufactured and assembled according to previously published protocol (Taylor et al., 2005), with the difference that the channels connecting the two chambers were designed as diodes (Peyrin et al., 2011; Fig. 6 A). The PDMS resin and the curing agent were manually mixed in a mass ratio of 9:1, poured on the epoxy replicates, and placed in a vacuum chuck for 30–45 min to remove all air bubbles. The polymerization was achieved in an oven at 70°C for at least 2 h. The PDMS was then removed and cut with a scalpel to the dimensions of the glass slide. Four holes were perforated over the reservoirs with a 2-mm puncher. PDMS pieces with the motif pattern faced up and the open Petri dishes were introduced in an O₂ plasma cleaner to activate the PDMS and glass surface. Thereafter, the two surfaces were bound together by exercising a gentle pressure with forceps on the PDMS pieces in the Petri dishes. The chambers and channels were then immediately coated with a mix of poly-D-lysine (0.5 mg/ml) and laminin (10 mg/ml), incubated 3 h at 37°C, and washed three times with neurobasal medium for 10 min. 1 million cortical neurons in 10 μl neurobasal medium were plated in the distal chambers. After 2 h, when neurons have attached to the surface, neurobasal + B27 was added to the four reservoirs. Cultures were regularly checked between DIV4 and DIV7 to select the microchambers that had at least five axons entering the distal chamber for retraction assay. Prior to the assay and video recording, 20 μl of media was removed from the reservoir of the distal chamber to produce a gradient of pressure between the two chambers, thus preventing drugs from diffusing into the different channels. This was tested by injection of 10 μl of PBS containing fluorescein (∼0.02 g/L). Retraction assay was performed by adding 10 μl of 10 µM LPI in the growth cone chamber simultaneously to 150 mM 2DG in the same injection if required or simultaneously to 10 μl of 150 mM 2DG injection in the somatic chamber. Due to the dead volume of the reservoirs and chambers, the final concentration is estimated to be 2–5 µM for LPI and 30–75 mM for 2DG.

For ATP measurement, hippocampal neurons were plated in 96-well plates with black borders seeded at 5,000 cells per well. At DIV7, the neurons were treated with 10 µM oligomycin (Sigma-Aldrich) and 30 mM 2DG (Sigma-Aldrich), CellTiter-Glo (Promega) was added, and photons were measured with a luminometer. For local ATP/ADP measurement, hippocampal neurons were transfected with Perceval and observed at DIV6–7 with a confocal Leica SP8 equipped with a 63× (NA1.4) oil immersion objective. Before acquisition, neurons were incubated for 30 min with 10 µM of the pH reporter SNARF-5F (Thermo Fisher Scientific). For Perceval signal measurement, excitation was set at 405 nm (ADP sensitive) and 496 nm (ATP sensitive) and detection of photomultiplier tube (PMT) detector was set at 520–560 nm. For pH, signal excitation was set at 550 nm and PMT was set at 575–585 nm for the isosbestic record and 615–635 nm for the pH-sensitive signal. pH calibration was acquired at pH 7.0, 7.2, 7.4, 7.6, and 7.8 pH with 10 µM Nigericin (Sigma-Aldrich) after Perceval acquisition. No change in pH was observed between baseline and after drug treatment, therefore no correction was applied in Perseval images.

Acidification rate (proportional to glycolysis) and OCR (respiration) were analyzed using a Seahorse XF analyzer (XFp; Proteigene/Agilent). For this, HEK293 cells were plated at 10,000 cells per well and hippocampal neurons were plated at 5,000 per well (XFp cell culture miniplates; Proteigene/Agilent) and analyzed 1 d after plating or at DIV5–7, respectively. Neuronal media were changed to XF Base Medium (Proteigene/Agilent) at pH 7.4 containing 10 M glucose (Sigma-Aldrich), 1.0 mM pyruvate (Thermo Fisher Scientific), and 0.5 mM glutamine (Sigma-Aldrich). Oligomycin (10 µM; Sigma-Aldrich), 2DG (30 mM; Sigma-Aldrich), WIN (100 nM; Tocris), calpeptin (0.2 U/ml, Cytoskeleon), and control DMSO (Sigma-Aldrich) were loaded at a port of the sensor cartridge (XFp extracellular flux cartridges; Proteigene/Agilent). OCR and ECAR were first measured for 30 min to establish the baseline and 5 min after drug injection. Values represent the percentage of baseline.

For immunofluorescence, hippocampal neurons plated at 50,000 cells per well in 12-well plates (DIV5) were washed with warm PBS and fixed at room temperature with a warm solution of PBS containing 4% PFA (Sigma-Aldrich) and 4% sucrose (VWR) for 10 min. Cells were washed twice with PBS and blocked for 1 h in a PBS solution containing 1% BSA and 0.1% Triton X-100 followed by 1 h incubation in the blocking solution containing the primary antibodies (1:200). Antibodies used were rabbit monoclonal anti-enolase-2, anti-PGAM1, anti-PKM1/2 anti-PKM2 and anti-GAPDH (Cell Signaling Technology), rabbit polyclonal anti-GAPDH (Sigma-Aldrich), and rabbit monoclonal anti-PGK1 (Abcam). Cells were washed three times with PBS (Thermo Fisher Scientific) and incubated 1 h with fluorescent secondary goat antibodies (1:200), ATTO484-phalloidin (1:200; Sigma-Aldrich). Coverslips were mounted with fluorodish (Sigma-Aldrich) on a glass support before three washes with PBS for 10 min.

For PLA staining, a permeabilization step was performed before fixation to remove the soluble proteins not bound to F-actin and microtubule cytoskeleton before applying the PLA protocol (Duolink in situ PLA; Sigma-Aldrich). Cells were quickly washed with warm PBS and incubated for 3 min in warm extraction solution (1% Triton X-100, 2% polyethylene glycol [MW 35 kD; Sigma-Aldrich]), 2 mM unlabeled phalloidin (Sigma-Aldrich), and 2 mM taxol (Sigma-Aldrich) in PEM buffer (100 mM PIPES, pH 6.9; 1 mM EGTA; 1 mM MgCl₂; Sigma-Aldrich). Cells were washed with a warm PEM buffer containing 2 mM unlabeled phalloidin and 2 mM taxol and fixed with a warm solution of 4% PFA and 4% sucrose in PEM buffer. Antibody used was 1:200, mouse monoclonal anti-actin and anti-ß-tubulin (EnCor), rabbit polyclonal anti-phospho-myosin (RLC; Rockland), human anti-NMII (Institut Curie), rabbit-anti GAPDH (Sigma-Aldrich), and mouse monoclonal anti-PGK1 (Abcam). Incubation with ATTO484-phalloidin (1:1,000; Sigma-Aldrich; for 10 min) was done after the PLA protocol and just before the coverslip mounting.

10 million cortical neurons were plated in 10-cm Petri dishes coated with poly-D-lysine (Sigma-Aldrich). At DIV3, neuronal media was removed, cells were washed with PBS and lysed with 1 ml of buffer containing 20 mM Tris-Cl (pH 7.4), 100 mM NaCl, 1% Triton X-100, 1 mM DTT, 10 mM EDTA, and 1% Halt Protease & Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific). Protein concentration was measured using BCA Protein Assay (Thermo Fisher Scientific). Immunoprecipitation was performed with 1 mg of total proteins in 1 ml and with Dynabeads Protein G coupled with human antibodies anti-NMII or anti-GFP (Institut Curie). Western immunoblotting was performed following standard procedures using 4–20% Mini-PROTEAN TGX Stain-Free Protein Gels and Trans-Blot Turbo Mini 0.2 µm polyvinylidene fluoride (PVDF) Transfer Packs membranes (Biorad). Primary antibodies for mouse anti-GAPDH (1:1,000; Cell Signaling Technology), rabbit anti-NMII (1:1,000; Institut Curie), rabbit anti-PGK1 (1:1,000; Abcam), and rabbit anti-PKM/2 (1:1,000; Cell Signaling Technology) were incubated overnight at 4°C or 1 h at room temperature. Horse anti-mouse or goat anti-rabbit HRP-coupled secondary antibodies (1:2,500; Cell Signaling Technology) were added (1 h at room temperature) and detected sequentially using SuperSignal West Pico PLUS Chemiluminescent Substrate luminescence solution (Thermo Fisher Scientific) as per the manufacturer’s instructions, and signal was detected with ChemiDoc Touch Imaging System (Biorad).

The magnetic beads were resuspended in 20 μl of 50 mM NH₄HCO₃ and digested overnight at 37°C with 0.5 µg trypsin (Promega). The supernatants were dried in a speed vacuum and dissolved in 0.1% formic acid for LC-MS/MS analysis. 200 ng of samples were analyzed on a timsTOF Pro 2 mass spectrometer (Bruker Daltonics) coupled to an Evosep one system (Evosep) operating with the 30SPD method developed by the manufacturer. Briefly, the method is based on a 44-min gradient with a C18 analytical column (0.15 × 150 mm, 1.9 µm beads) equilibrated at room temperature and operated at a flow rate of 500 nl/min. H₂O/0.1% formic acid (FA) was used as solvent A and acetonitrile/0.1% FA as solvent B. The timsTOF Pro 2 was operated in parallel accumulation-serial fragmentation (PASEF) mode over a 1.3-s cycle time. Mass spectra for MS and MS/MS scans were recorded between 100 and 1,700 m/z. Ion mobility was set to 0.75–1.25 V·s/cm² over a ramp time of 180 ms. Data-dependent acquisition was performed using six PASEF MS/MS scans per cycle with a near 100% duty cycle. Low m/z and singly charged ions were excluded from PASEF precursor selection by applying a filter in the m/z and ion mobility space. The dynamic exclusion was activated and set to 0.8 min, a target value of 16,000 was specified with an intensity threshold of 1,000. Collisional energy was ramped stepwise as a function of ion mobility. Precursor ions for MS/MS analysis were isolated with a 2-D window for m/z < 700 and 3-D for m/z > 700. Singly charged precursor ions were excluded with a polygon filter. MS raw files were analyzed by Peaks Online X software (build 1.6; Bioinformatics Solution, Inc.) using the Rattus norvegicus database (SwissProt release 2022_01, 8,147 entries). Parent mass tolerance was set to 20 ppm with fragment mass tolerance of 0.05 D. Specific tryptic cleavage was selected and a maximum of two missed cleavage was authorized. Half of the disulfide bridge was set as a fixed modification, whereas oxidation, acetylation of protein N-termini, and deamidation were set as possible variable modifications. The maximum number of variable modifications per peptide was limited to three. Identifications were filtered based on a 1% false discovery rate threshold at both peptide and protein group levels. The 450 putative NMII-interacting proteins were selected using the following cutoffs and criteria: peptide coverage in NMII precipitate >10%; −10LgP > 50; area of peaks in NMII precipitate >150; area of peaks in NMII precipitate > area in eGFP precipitate; and specificity index NMII > specificity index GFP. The online tools PANTHER Version 14 were used for GO and gene enrichment analyses (Ashburner et al., 2000; Gene Ontology Consortium, 2021; Mi et al., 2019).

Statistical analysis is described in each figure legend and was performed using jamovi software (Version 1.2). Two-tailed unpaired Student’s t test was used for comparing two groups and one-way ANOVA was used for comparing three or more groups; both were followed by a post-hoc analysis (Tukey and Bonferroni). Data distribution was assumed to be normal but this was not formally tested. P < 0.05 was considered to be statistically significant. Statistics and plotting were performed using R Studio or jamovi. All experiments were replicated at least two times; details are found in the legends of the figures.

Rat experimental procedures were performed in accordance with the Council of European Union directive of September 22, 2010 (2010/63/UE), and with the French decree of February 1, 2013 (n°2013-118).

Key resources given in Table 3.

This manuscript is accompanied by two supplementary figures, a video, and a supplemental table. Fig. S1 contains data supporting Fig. 1; it shows the collapse of growth cones after stimulation with WIN in control or under metabolic inhibition. Fig. S2 contains data supporting Fig. 3; it shows a schema of the glycolytic enzymes and immunofluorescences at growth cone of all the enzymes of the payoff phase. Video 1 contains data supporting Fig. 3 A; it shows the dynamic of a growth cone and its retraction after WIN. Table S1 supports Tables 1 and 2; it contains the results of the MS analysis (sheet 1), the 450 putative proteins interacting with NMII (sheet 2), and the GO term analysis (sheet 3).

This study did not generate new unique reagents.

Further information should be directed to and will be fulfilled by the lead contact, Diana Zala ([email protected]).

The data underlying all figures and tables are available in the published article and its online supplemental material. This study did not generate/analyze datasets/code.

We thank Julie Nguyen from the NeurImag facility at Institut of Psychiatry and Neurosciences of Paris, France, for the neuronal cultures and Erwan Boëdec from the Biochemistry and Biophysics facility for help with western blots at Institut of Psychiatry and Neurosciences of Paris. We thank Sandrine Moutel of the Recombinant Antibody Platform and TAb-IP Platform of Institut Curie for providing antibodies. We also thank the Proteomics facility of the Institut Jacques Monod, supported by the Région Ile-de-France, Université Paris Cité, and Centre National de la Recherche Scientifique, for assistance. We thank Derin Reha Ulusoy for preliminary experiments and Sonia Garel for scientific discussions.

This work was supported by a grant from DIM Cerveau et Pensé 2015 Neuroflux (D. Zala) and by a donation from Agilebio (R. Santos). R. Santos is a Centre National de la Recherche Scientifique researcher; L. Lokmane, Z. Lenkei, and D. Zala are Institut National de la Santé et de la Recherche Médicale researchers.

Author contributions: R. Santos conceptualized the study, designed, conducted and analyzed experiments, and wrote the paper. L. Lokmane conducted and analyzed experiments, and revised the paper. D. Ozdemir, C. Traoré, A. Agesillas, and C. Hakibilen conducted experiments. Z. Lenkei conceptualized and supervised the study, and revised the paper. D. Zala conceptualized and supervised the study, designed, conducted and analyzed experiments, and wrote the paper.