ATP synthase c-subunit leak causes aberrant cellular metabolism in Fragile X syndrome

Summary

Loss of the gene (Fmr1) encoding Fragile X mental retardation protein (FMRP) causes increased mRNA translation and aberrant synaptic development. We find neurons of the Fmr1^−/y mouse have a mitochondrial inner membrane leak contributing to a “leak metabolism”. In human Fragile X syndrome (FXS) fibroblasts and in Fmr1^−/y mouse neurons, closure of the ATP synthase leak channel by mild depletion of its c-subunit or pharmacological inhibition normalizes stimulus-induced and constitutive mRNA translation rate, decreases lactate and key glycolytic and TCA cycle enzyme levels, and triggers synapse maturation. FMRP regulates leak closure in WT, but not FX synapses, by stimulus dependent ATP synthase β subunit translation; this increases the ratio of ATP synthase enzyme to its c-subunit, enhancing ATP production efficiency and synaptic growth. In contrast, in FXS, inability to close developmental c-subunit leak prevents stimulus dependent synaptic maturation. Therefore, ATP synthase c-subunit leak closure encourages development and attenuates autistic behaviors.

Graphical Abstract

In Brief

Lack of FMRP in Fragile X neurons is associated with a leak in the ATP synthase, whose blockade normalizes cellular and behavioral disease phenotypes.

Introduction

FXS is a devastating X-linked genetic disorder and the most common inherited cause of intellectual disability (Dolen et al., 2010; Wijetunge et al., 2012). It results from a CGG repeat expansion within the Fmr1 gene that leads to loss of expression of FMRP. The FX phenotype is characterized by constitutively increased mRNA translation rates, morphological immaturity of synapses and dendritic spines (Dolen et al., 2007), aberrant synaptic plasticity (Bear et al., 2004), excitotoxicity (Dolen et al., 2010; Dolen et al., 2007), and enhanced excitability (Brown et al., 2010; Deng et al., 2013; El-Hassar et al., 2019; McCullagh et al., 2020; Strumbos et al., 2010; Zhang et al., 2012). The structural and functional deficiencies in FX neurons correlate with loss of normal learning patterns and an increase in autistic behaviors (Santos et al., 2014) in rodents and humans (Zoghbi and Bear, 2012). Despite the fact that FMRP has been well described as an RNA-binding protein (Darnell, 2011; Darnell et al., 2011), there have been to date no effective therapies for the disorder.

Prior work found that depletion of Fmr1 and its homologue Fxr2 reduces fat deposits in mutant mice and leads to higher food intake, increased oxygen consumption and CO₂ production, suggesting uncoupled oxidative phosphorylation (Lumaban and Nelson, 2015). In human autism spectrum disorder patients elevated lactate levels, indicative of glycolysis driven by mitochondrial dysfunction, has been described (Dhillon et al., 2011; Goh et al., 2014). Hippocampal neurons of the Fmr1 CGG repeat knock in (KI) mouse (FMRP levels decreased by 42.6%), have small mitochondria with reduced mobility, high O₂ uptake, and a large proton leak (Kaplan et al., 2012). In a recent report, Drosophila fmr1 was found to alter the metabolome of the flies such that the dfmr1 mutants had decreased carbohydrate and lipid stores, were hypersensitive to starvation stress, hyperphagic and had reduced NAD⁺/NADH ratio (Weisz et al., 2018). This was related to a defect in mitochondria producing an increase in maximal respiratory capacity. A recent report also found that, in Fmr^1−/y mouse brains, electron transport complexes run at high rates even though ATP production is low, suggesting inner membrane inefficiency (H⁺ ion leak) (D’Antoni et al., 2020). Our group’s recent report shows that mitochondria from the Fmr1^−/y mouse brains have inefficient thermogenic respiration due to a futile coenzyme Q-regulated proton leak, leading to synaptic spine and behavioral abnormalities (Griffiths et al., 2020). Although these findings suggest that FMRP and related proteins might be required in general for normal developmental mitochondrial function, they also highlight the possibility that FX mitochondria are uncoupled, that they have an inner membrane leak that makes the mitochondria inefficient, burning fuel instead of saving energy.

The mitochondrial abnormalities could be emblematic of neuronal immaturity. Several findings indicate that FX neurons are underdeveloped (Bassell and Warren, 2008; Bhattacharya et al., 2012; Pyronneau et al., 2017). They have redundant postsynaptic spines with a thin, long filopodial shape that has failed to change to a mushroom shape characteristic of mature synapses (Bagni and Greenough, 2005). FX mitochondria are also small (Kaplan et al., 2012). The possibility that this structural immaturity throughout brain development is related to the delay of a metabolic shift during development is suggested by reports on cardiac development in the early WT embryo. At embryonic day (E) 9.5 profound inner mitochondrial membrane leak accompanies a lack of respiration (Hom et al., 2011). The leak closes between E11.5 and E13.5 at the onset of oxidative phosphorylation (Hom et al., 2011) and respiration becomes coupled to phosphorylation. This normal developmental process is related to closure of the cell death channel that spans the mitochondrial inner membrane, known as the permeability transition pore (mPTP), and development is hastened by the mPTP inhibitor cyclosporine A (CsA) (Beutner et al., 2017; Beutner et al., 2014).

We have reported previously that certain proteins and pharmacological reagents regulate directly mitochondrial inner membrane efficiency by binding to the ATP synthase enzymatic (F₁) portion (Alavian et al., 2015; Alavian et al., 2011; Chen et al., 2019; Chen et al., 2011). Our work indicates that a leak that regulates inner membrane ATP production efficiency resides within the membrane-embedded c-subunit ring of the ATP synthase (Alavian et al., 2014). We have further suggested that the c-subunit ring may form or contribute significantly to the CsA regulated mPTP (Alavian et al., 2014; Bernardi and Di Lisa, 2014; Bonora et al., 2013; Mnatsakanyan et al., 2019; Neginskaya et al., 2019; Rasola and Bernardi, 2014). We have shown recently that the mPTP is aberrantly active in Fmr1^−/y mitochondria (Griffiths et al., 2020).

We now suggest that the mitochondrial inner membrane leak of FX neurons and cells is caused by abnormally high levels of ATP synthase c-subunit. We find that the c-subunit leak causes persistence of an immature metabolic phenotype associated specifically with mitochondrial leak, termed here “leak metabolism”. The c-subunit leak aberrantly elevates protein synthesis; a decrease in c-subunit level or specific pharmacological inhibition of ATP synthase leak reduces protein synthesis rates and decreases the leak metabolism in the neurons. In keeping with the c-subunit leak as causative of abnormally delayed neuronal development, we find that inhibition of the ATP synthase leak allows for the maturation of synapses and normalizes autistic behaviors in a mouse model of FX.

Results

Mitochondrial morphology and function are altered in FX mouse neurons

Mitochondria are necessary for normal synapse formation and thus mitochondria themselves undergo developmental plasticity; large changes in mitochondrial structure and function occur during development (Brandt et al., 2017; McCarron et al., 2013). We therefore examined mitochondria in Fmr1^−/y neurons and compared them to mitochondria of WT neurons of the same age. Electron micrographs taken of hippocampal CA1 brain slices prepared from 2-month-old mice revealed that presynaptic vesicle pools were smaller in Fmr1^−/y synapses than those of WT synapses as previously described (Figs. 1A, B, D) (Klemmer et al., 2011). The mitochondria within Fmr1^−/y synapses also had a decreased area compared to WT synapses (Figs 1A, B and C), but the ratio of mitochondrial area to synapse area was similar in Fmr1^−/y vs. WT synapses (Figs. 1E–G), suggesting that mitochondrial size was scaled to the smaller synaptic size in Fmr1^−/y. The mitochondria of Fmr1^−/y synapses had no apparent disturbance of cristae structure, although there was a marked increase in matrix density relative to that of controls (Figs. 1E, F, H), possibly reflecting a high protein amount within the matrix compartment.

Figure 1. Fmr1^−/y mitochondria have an inner mitochondrial membrane leak.

(A, B) Representative electron microscopy images of brain slices show synapses from the CA1 region of the mouse hippocampus (WT and Fmr1^−/y). (C) Group data show a decrease in mitochondrial area in Fmr1^−/y compared to WT (N=15 for Fmr1^−/y and N=17 micrographs for WT, *p=0.043). (D) The vesicle number in presynaptic boutons is reduced in Fmr1^−/y (N=11 micrographs of each condition; *p=0.0143). (E, F) Example electron micrographs of synapses from the CA1 region of the mouse hippocampus slices (WT and Fmr1^−/y). (G) Mitochondrial area/synapse area is unchanged comparing WT to Fmr1^−/y (N= 13 micrographs for WT and 23 for Fmr1^−/y). (H) Group data show electron density is increased in Fmr1^−/y compared to WT (N=23 micrographs for Fmr1^−/y and 13 for WT; *p=0.0114). (I) Representative images of mitochondrial membrane potential indicator TMRM fluorescence in isolated cortical neurons. (J) Group data show TMRM intensity is reduced in FX compared to WT (N=20–24 neurons each condition, 4 independent cultures; ****p<0.0001). (K) Cytosolic ATP levels of isolated cortical neurons are reduced in FX compared to WT (N=3 wells for each, *p<0.05). (L) Illustration of method of measurement of bath H⁺ ion concentration using the H⁺ sensitive indicator ACMA. SMVs are illustrated with mitochondrial ATP synthase F₁ facing toward the bath. ATP hydrolysis causes H⁺ ion sequestration into the lumen of the vesicles. Vesicles are impermeant to the pH indicator. Red arrows show paths of H⁺ pumping (at the side of the c-subunit) and H⁺ leak (through the center of the c-subunit) (M) Lack of sequestration of H⁺ ions into Fmr1^−/y SMVs during ATP hydrolysis by the ATP synthase (N=3 samples per condition). (N) Representative patch clamp recordings of SMVs of Fmr1^−/y and WT at the indicated holding potential. Black trace indicates open channel. Red trace shows the relatively closed channel in the same recording after Dex exposure. (O) Current voltage relationship for the group of SMV recordings shown in (P). (P) Group data of peak conductances of the independent recordings (N=6 for WT and WT+Dex; N=8 for Fmr1^−/y and Fmr1^−/y +Dex) measured from 0 pA (*p=0.0449 comparing WT to WT+Dex; **p=0.0064 comparing Fmr1^−/y to Fmr1^−/y+Dex). Linear current voltage relationship was assumed for calculation of peak conductance. In Figs. 1C, D, G, H, J, K, unpaired two-tailed Student’s t-test was used. In Fig. 1M, two-way repeated measures ANOVA followed by Sidak’s multiple comparison test was used. In Fig. 1P, paired two-way Student’s t-test was used. Data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001).

To examine mitochondrial function at an earlier stage of development, we prepared cortical neurons from mice at postnatal day (P) 1–2 and studied them at the time when synapses are forming rapidly (DIV14). Mitochondrial inner membrane polarization is indicative of appropriate use of the electron transport chain. Using the membrane potential indicator tetramethylrhodamine methyl ester (TMRM), we determined that FX mitochondria generated less than half the membrane potential of the mitochondria of WT neurons (Fig. 1I, J). Lack of membrane potential can arise from rapid ATP production caused by inability to keep up with increased energy demand or from depolarization caused by a mitochondrial inner membrane leak. In either case, ATP levels may be decreased. In fact, ATP levels in FX neurons were slightly decreased compared to WT control neurons (Fig. 1K), suggesting that FX neurons fall behind in their combined mitochondrial and glycolytic ATP production.

Fmr1^−/y mitochondria have a large conductance inner membrane leak

In the setting of increased energy demand, H⁺ ions move across the mitochondrial ATP synthase and cause a conformational change in the ATP synthase molecular complex that results in ATP production. The enzyme can also hydrolyze ATP to provide energy to translocate H⁺ ions across the inner mitochondrial membrane in reverse mode. The 9-amino-6-chloro-2-methoxyacridine (ACMA) assay measures the ATP synthase enzymatic rate using this reverse mode (Fig. 1L) (Caviston et al., 1998). To perform this assay, mitochondrial inner membrane remnants that are relatively enriched in ATP synthase (submitochondrial vesicles, SMVs) are prepared (Chan et al., 1970; Chen et al., 2004; Ko et al., 2003). In these SMVs, the F₁ or enzymatic portion of the ATP synthase is exposed to the medium, which contains a membrane impermeant H⁺ indicator. SMVs prepared from WT brain readily sequestered H⁺ ions in response to the addition of ATP to the bath. In contrast, Fmr1^−/y SMVs failed to sequester H⁺ ions in response to ATP addition (Fig. 1M), indicating either failure of the enzyme to translocate H⁺, or a H⁺ leak in the SMV membranes. To detect directly an inner mitochondrial membrane leak, we analyzed patch clamp recordings of Fmr1^−/y brain SMVs. We found a large multi-conductance, voltage dependent channel activity not present in WT brain SMVs (Fig. 1N–P), indicating that lack of sequestration of H⁺ ions in the ACMA assay was due to loss of the H⁺ ions through an open channel in the SMV membranes.

Our previous reports demonstrated an effect of an ATP synthase modulator, Dexpramipexole (Dex) on inner mitochondrial membrane leak channel activity (Alavian et al., 2015). It had been described previously that Dex, which is the R(+) enantiomer of the widely used Parkinson’s drug Pramipexole, has no significant dopaminergic efficacy, is not a CsA analog, and that it readily crosses the blood brain barrier (Bozik, 2009; Bozik, 2010; Cudkowicz et al., 2011). We reported that radiolabeled Dex binds to ATP synthase F_O subunit b and F₁ subunit OSCP, closing an inner membrane leak in patch clamp recordings, enhancing ATP production efficiency and decreasing cell death and oxygen consumption with modest potency (Alavian et al., 2015). To test if ATP synthase leak was the cause of the open channel in Fmr1^−/y SMVs, we applied Dex during the recordings. Dex decreased small conductance activity of the WT SMVs and decreased large conductance channel activity in the Fmr1^−/y SMVs as measured by change in peak conductance (Fig. 1N and P). This suggests that at least some of the leak in Fmr1^−/y mitochondrial inner membranes is produced by increased activity of the ATP synthase leak channel, although a resistant fraction of the leak could be produced by uncomplexed c-subunit or by other channels.

ATP synthase contains a highly regulated, multi-conductance channel that is contained within the c-subunit ring of the main membrane bound (F_O) portion (Alavian et al., 2014; Mnatsakanyan et al., 2019). Pathological opening of the channel may occur upon conformational change of the ATP synthase including loss of an inhibitory structure in its cavity (Gerle, 2016; Gu et al., 2019; Mnatsakanyan and Jonas, 2020; Vlasov et al., 2019), separation of the F₁ from the F_O (Alavian et al., 2014), or loss of F₁ (Chen et al., 2019). We hypothesized based on these previous findings and the channel recording data that the ratio of c-subunit (F_O) to F₁ might be altered such that Fmr1^−/y mitochondria could have an unopposed (uncoupled to F₁) c-subunit that would readily leak ions across the inner mitochondrial membrane in a voltage-dependent manner. We found, in Fmr1^−/y, that ATP synthase β-subunit (a major part of the soluble enzyme or F₁ component) level was markedly elevated above that measured in control mitochondria (Fig. 2A), but that the level of c-subunit (F_O) was elevated to an even greater degree (Fig. 2A). To determine if this resulted in free c-subunit in the mitochondrial membrane, we subjected mitochondrial proteins to non-denaturing Native-PAGE electrophoresis (Fig. 2B). These immunoblots showed that, although total amounts of assembled ATP synthase monomer plus dimer were not significantly elevated in Fmr1^−/y compared to WT, level of free c subunit was markedly elevated compared to WT controls. We analyzed the blots in two ways. In the same blot we compared the level of Fmr1^−/y fully assembled ATP synthase (monomer plus dimer) to the WT controls and the level of Fmr1^−/y free c-subunit to the WT controls (left set of histograms). In the right histogram, we calculated the ratio of free c-subunit to its own assembled ATP synthase within the same lane. Both analyses showed that free c-subunit is markedly elevated in Fmr1^−/y compared to WT controls.

Figure 2. Expression of ATP synthase subunits causing inner membrane leak is increased in Fmr1^−/y mitochondria.

(A) c- and β- subunit protein expression levels are higher in Fmr1^−/y brain mitochondria compared to those of WT (N=8 independent samples for each, **p=0.0083 and *p=0.031). (B) An example of three independent experiments of non-denaturing Native-PAGE electrophoresis of isolated brain mitochondria. Immunoblotting performed with anti-c-subunit antibody. The abundance of the free c-subunit is higher in Fmr1^−/y mitochondria compared to WT. Left set of histograms show the level of fully assembled ATP synthase (monomer plus dimer) and the level of free c-subunit in Fmr1^−/y as a percent of WT control. Right histograms: Ratio of free c-subunit to its own assembled ATP synthase within the same lane. N=3 for each condition. (C) FMRP immunoprecipitation from isolated synaptosomes pulls down ATP synthase β-subunit mRNA but not ATP synthase c-subunit mRNA (ATP5G2) as detected by RT-PCR (shown is one of N=3 independent immunoprecipitation experiments). All experiments in this figure used unpaired Student’s t-test, data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001).

We next carried out experiments to determine if the increase in β and c subunit levels in Fmr1^−/y mitochondria can be attributed directly to FMRP-regulated mRNA translation. Many of the nuclear-encoded mRNAs for components of the ATP synthase have been found to bind FMRP in crosslinking immunoprecipitation experiments, and mRNA for the ATP synthase β subunit is the 87^th most represented on the list of over 800 mRNAs (Darnell et al., 2011). To confirm the binding of FMRP to ATP synthase β-subunit mRNA, we immunoprecipitated FMRP from synaptoneurosomal (hereafter synaptosomal) lysates and performed PCR on the cDNA synthesized from the immunoprecipitate. We confirmed that FMRP binds β subunit mRNA. By RT-PCR, ATP5G2 was found to be by far the dominant c-subunit gene but we failed to detect binding of FMRP to any of the three c-subunit gene products (only ATP5G2 is shown in Fig. 2C). In Fmr1^−/y synapses, qRT-PCR of the synaptosomes (Suppl. Fig. 1), performed to detect total expression of ATP synthase β subunit and ATP5G2 c-subunit mRNA, showed that these mRNAs were elevated compared to those of WT synapses. Therefore, although loss of FMRP may lead to enhanced transcription of both ATP synthase β and c-subunit mRNAs, in contrast only β subunit, but not c-subunit, translation is likely to be regulated by FMRP. Taken together, these results suggest a role for FMRP in ATP synthase assembly during activity dependent synaptic development (see Fig. 5D).

Figure 5. The time course of synaptic stimulation-induced changes in protein synthesis and EF2 phosphorylation is disrupted in Fmr1^−/y synaptosomes, normalized by ATP synthase leak inhibition.

(A) Representative immunoblots of synaptosomal samples harvested at the indicated time points before and after 0.2 μM D-serine stimulation. Top panels show puromycin incorporation; middle panels show p-EF2 and EF2; bottom panels show ATP synthase β subunit and protein loading control (GAPDH). CsA restores the normal pattern of response to stimulation in Fmr1^−/y synaptosomes. (B-D) Group data for experiments shown in (A): (B) for puromycin incorporation, (C) p-EF2/EF2 protein levels and (D) ATP synthase β subunit protein levels. One-way ANOVA followed by Tukey’s multiple comparisons test was used for all panels in the figure. Synaptosomes were prepared from at least three independent animals per condition. N=samples. Data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001)

FX neurons have a mitochondrial leak-dependent metabolic phenotype

Despite the leaky inner membrane, ATP levels were only slightly decreased in FX neurons compared to controls (Fig. 1K), suggesting that glycolytic ATP might contribute to the overall ATP level in the neuronal cytoplasm. To determine the metabolic phenotype of the neurons, we labeled newly synthesized proteins with puromycin and immunoprecipitated with an anti-puromycin antibody. We then evaluated the newly synthesized proteome by liquid chromatography with tandem mass spectrometry (LC/MS/MS) (Fig. 3A, B, Suppl.Table 1). This study revealed that FX neurons have an increase in certain glycolytic enzymes including hexokinase II, pyruvate kinase M2 variant and lactate dehydrogenase and also in enzymes required for TCA cycle and NAD⁺/NADH metabolism, including enzymes of the malate/aspartate shunt and isocitrate dehydrogenase (Dayton et al., 2016; Li et al., 2016; Roberts and Miyamoto, 2015; Vander Heiden et al., 2009; Zheng et al., 2016). Since these data suggested an increase in activity of glycolytic enzymes, we measured lactate levels and found that they were markedly increased in the FX culture media (Fig. 3 C). High glycolytic activity and lactate production, but also increases in TCA cycle enzymes are hallmark features of immature and developing cells (Fame et al., 2019). Although puromycin incorporation into newly translating proteins provided us with an estimate of the high translation rate of metabolic enzymes, it did not tell us about the steady state levels of these enzymes. To test if specific enzymes are elevated in FX, we performed immunoblots on WT and FX neuronal cultures and brain mitochondria. These showed that the levels of certain key enzymes are elevated in FX (Fig. 3 D–G), including hexokinase and pyruvate dehydrogenase, suggestive not only of enhanced glycolytic flux but also of enhanced TCA cycle function, supporting the idea that increased flux through the FX mitochondria is caused by the mitochondrial inner membrane leak. Because Dex closed the mitochondrial inner membrane leak in patch clamp recordings and we reported previously that Dex enhances the efficiency of oxidative phosphorylation (Alavian et al., 2015), we tested if Dex treatment of the FX neuronal cultures would eliminate the increased flux of the leak metabolic phenotype. We found that Dex effectively decreased the lactate levels in FX neurons and reversed the abnormally high levels of glycolytic and TCA enzymes (Fig. 3 C, B and D). These results are consistent with increased flux through glycolytic and TCA pathways caused by the inner membrane leak in FX. Dex normalizes the aberrant leak metabolism of FX neurons.

Figure 3. Metabolic profile of FX cortical neurons shows enhancement of glycolysis/TCA flux in FX compared to WT.

(A) Glycolysis and tricarboxylic acid (TCA) cycle schematics illustrating enzymes involved in both pathways. Enzymes increased in FX>WT in at least 2 of 3 independent cultures are labeled in green. (B) Averaged spectral counts of metabolic peptides expressed in WT and FX cortical neuron cultures. Puromycin immunoprecipitates were analyzed by LC/MS/MS after cortical neuronal cultures (DIV14) were exposed to puromycin for 15 minutes. Shown in blue are the enzymes decreased by Dex treatment in at least 2 out of 3 FX cultures (N=3 independent cultures of each condition). Metabolic peptides completely removed by Dex treatment are indicated above the graph in blue lettering. (C) Lactate levels are elevated in the culture media collected from FX primary neurons compared to those of WT. Exposure of FX neurons to Dex significantly decreases lactate levels in the media. (D) Representative immunoblots of WT and FX cortical cultures exposed to vehicle or Dex. (E) Quantification of blots shown in (D). At least 3 independent cultures were used. A set of 4 key enzymes is elevated in FX compared to WT. Dex treatment normalizes the protein levels of all enzymes in the set. (F) Representative immunoblots of WT and Fmr1^−/y mitochondria isolated from brain. (G) Quantification of blots shown in (F). At least three animals per condition. Glycolytic enzymes and pyruvate dehydrogenase protein levels are elevated in Fmr1^−/y mitochondrial fractions. In Figs. 3C and E, two-way ANOVA followed by Tukey’s multiple comparisons test was used. In Fig. 3G, unpaired two-tailed Student’s t-test was used. Data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001).

Genetic or pharmacological modulation of ATP synthase leak decreases abnormally elevated protein synthesis rates in FX

Previous work has shown that the anti-apoptotic protein Bcl-xL acts on the ATP synthase to close the mitochondrial inner membrane leak (Alavian et al., 2011; Chen et al., 2011) but whether this is tied to changes in protein synthesis rate was not known. Both translation of the reporter Dendra (Suppl. Fig. 2A, B), and puromycin incorporation assays (Fig. 4E–H) demonstrated that rates of mRNA translation in general are elevated in FX neurons compared to WT, as has been described previously (Bear et al., 2004; Dolen et al., 2007; Jacquemont et al., 2018; Muscas et al., 2019; Udagawa et al., 2013). We found that rates of Dendra translation or puromycin incorporation were similarly elevated in WT neurons by exposure to the selective Bcl-xL inhibitor ABT-737 (Suppl Fig. 2C–E), and in contrast these rates were reduced by Bcl-xL protein transfection into synapses (Suppl. Fig. 2F, G). These findings are consistent with the hypothesis that opening of the ATP synthase leak (Alavian et al., 2011) increases overall mRNA translation rates in developing neurons. The pharmacological reagents (Dex and CsA) and Bcl-xL all bind in the soluble portion (F₁) of the ATP synthase, not within the membrane embedded portion. Therefore, to determine if membrane embedded ATP synthase c-subunit leak channel directly modulates mRNA translation rate, we depleted c-subunit protein by siRNA or increased c-subunit protein level by overexpression in human FX fibroblasts and measured protein synthesis rates by puromycin incorporation. We found that mild depletion of c-subunit decreased protein synthesis rate (Fig. 4A) and, in contrast, c-subunit overexpression increased the rate of protein synthesis (Fig. 4B), directly implicating c-subunit leak in regulation of protein synthesis rate in human FX cells. In FX cortical neurons, the ATP synthase leak inhibitor Dex also decreased protein synthesis rates in a concentration dependent manner (Fig. 4C, D), consistent with Dex effects on decreasing the open probability of the ATP synthase leak channel. Application of Dex also decreased protein synthesis in brain slices (Fig. 4E, F). We then tested a well-known and more potent inhibitor of the ion channel conductance of the ATP synthase that also inhibits the mPTP, CsA (Baines et al., 2005; Nakagawa et al., 2005). CsA prevents binding of Cyclophilin D to ATP synthase subunit OSCP (Giorgio et al., 2009; Giorgio et al., 2013). In FX neuronal cultures, small concentrations of CsA readily reversed the aberrantly increased mRNA translation rate (Fig. 4G, H), but had no effect on mRNA translation in WT neurons (Fig. 4G, H), suggesting that the CsA sensitive leak is larger in FX compared to WT neuronal mitochondria (see Fig. 1N–P).

Figure 4. Inhibition of the mitochondrial inner membrane leak decreases protein synthesis in FX.

(A) c-subunit depletion in human FX fibroblasts decreases the rate of protein synthesis. Representative puromycin (top), c-subunit immunoblots (middle) and protein controls (bottom) are shown at left. Quantification of the immunoblots is shown at right. (B) c-subunit overexpression in human FX fibroblasts increases the rate of protein synthesis. Representative blots of puromycin and c-subunit are shown at left as in (A). Quantification of the immunoblots is shown at right. (C) Dex decreases protein synthesis rate in FX neurons. Representative blot of puromycin incorporation in FX cortical cultures is shown in the presence of different concentrations of Dex. Cultures were treated with Dex for 2–24 hr. (D) Group data for experiments shown in (C). The group data for WT translation rates are shown in (H). N=samples from at least 3 independent cultures. (E) Puromycin incorporation into WT and Fmr1^−/y after incubation of mouse brain slices with 10 μM Dex or vehicle for 2.5 hr. (F) Group data of experiments shown in (E). N=3–6 brain slices for each condition, at least three animals per condition. (G) Representative immunoblot of puromycin incorporation showing that the rate of protein synthesis in FX cortical neurons is increased over WT neurons; exposure to 0.2 μM CsA for 7 days reduced the rate of puromycin incorporation in FX neurons. (H) Quantification of the data shown in (G) N=samples from at least 3 independent cultures. (I-M) Increase in rosettes (indicating actively translating ribosomes) in Fmr1^−/y brains is normalized to WT level by CsA exposure. Representative electron micrographs of CA1 region of hippocampal brain slices of Fmr1^−/y mouse compared to WT (R, ribosomal rosettes; M, mitochondria). (K) shows a higher magnification of the actively translating ribosomes (rosettes) shown in (J). Parallel slices from the corresponding hemisphere were incubated in 0.2 μM CsA or vehicle for 2.5 hours. Scale bars as indicated. (N) Group data of 65–76 micrographs analyzed per condition. Figs. 4A, B used unpaired two-tailed Student’s t-test. Fig. 4D, one-way ANOVA followed by Tukey’s multiple comparisons test was used. For Figs. 4F, H, N, two-way ANOVA followed by Tukey’s multiple comparisons test was used. Data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001).

We also carried out electron microscopy to visualize the effects of CsA on the protein synthetic machinery in the CA1 region of hippocampal slices from Fmr1^−/y and WT mice. Slices were treated with CsA or vehicle for 2.5 hours then fixed for processing. Our studies showed a large increase in the number of actively translating ribosomal assemblies (rosettes) in Fmr1^−/y CA1 neurons, consistent with an enhancement in mRNA translation over that seen in WT neurons (Fig. 4I–N). Strikingly, CsA greatly reduced the number of rosettes in the contralateral half of the same coronal brain slice, suggesting that its effects on mitochondrial inner membrane leak rapidly decreased the rate of mRNA translation in Fmr1^−/y CA1 neurons.

Abnormally enhanced mitochondrial inner membrane leak prevents stimulus-dependent protein synthesis

FX neuronal synapses have abnormal stimulus dependent synaptic plasticity and ineffective synapse maturation (Pfeiffer and Huber, 2009; Sidorov et al., 2013). We considered that inefficient ATP production by mitochondria in FX synapses could compromise stimulation-dependent phosphorylation events of the protein synthesis machinery, thereby affecting synaptic growth. The translation of mRNAs for synaptic proteins during long term changes in synaptic plasticity is regulated by an increase in phosphorylation of elongation factor 2 (EF2) ~5 minutes after neuronal stimulation (Autry and Monteggia, 2012; Gildish et al., 2012; Park et al., 2008; Scheetz et al., 2000; Um et al., 2013). We found that stimulation of WT synaptosomes using the NMDA glutamate receptor co-agonist D-serine led to robust phosphorylation of EF2 at 5 minutes followed by increased protein synthesis at 30 minutes, as has been reported previously (Fig. 5A–C)(Scheetz et al., 2000). Strikingly, both EF2 phosphorylation and protein synthesis changes failed to occur following synaptic stimulation in Fmr1^−/y synapses (Fig. 5A–C). If the mitochondrial membrane leak were responsible for the failure of phosphorylation of EF2 during synaptic stimulation, then this specific phosphorylation event would be rescued by closing the leak. Indeed, application of low dose CsA completely rescued EF2 phosphorylation and restored the normal pattern of protein synthesis upon synaptic stimulation in Fmr1^−/y synapses (Fig. 5A–C), confirming that this phosphorylation event and subsequent changes in protein synthesis rate are sensitive to ATP synthase leak modulation.

One potential mechanism by which FMRP could regulate the leak across the inner mitochondrial membrane is by controlling the translation of FMRP-bound mRNA coding for the ATP synthase β subunit. A stimulus-induced increase in protein levels of β subunit would decrease the level of free c-subunit by assembling the full F₁F_O ATP synthase. To test this, we measured β subunit protein levels in the same stimulated synaptosomes that were used for determination of p-EF2 levels and puromycin incorporation (Fig. 5D). These studies showed that levels of ATP synthase β subunit protein are increased 30 minutes after stimulation of WT synaptosomes but are unchanged after stimulation of Fmr1^−/y synaptosomes. Moreover, CsA treatment of the Fmr1^−/y synaptosomes restored the normal stimulus-induced increase in β subunit protein (Fig. 5D). Comparing column 3 to column 2 (in the last panel of Fig. 5D) we find there is a markedly enhanced rate of de novo synthesis of β subunit protein at 30 minutes after peak EF2 phosphorylation, suggesting a rapid change in ATP synthase stoichiometry upon synaptic stimulation.

ATP synthase modulation enhances synaptic plasticity

A number of studies have reported that FXS produces abnormalities in dendritic spines, including an increase in the number of spines and delays in their morphological maturity (for review please see (He and Portera-Cailliau, 2013)). To determine if ATP synthase leak closure rescues dendritic spine abnormalities, we carried out analyses of dendritic spine morphology in WT and FX neuronal cultures exposed to Dex or vehicle. In vehicle-treated WT cultures (at DIV 20), mushroom (mature) spines contributed approximately 50% of the total spine count, whereas in the FX cultures, mushroom spines represented only 10%. Dex treatment caused a three-fold increase in the mature spines in FX cultures which increased the mature spines to 25% of the total spine count (Fig. 6A–F).

Figure 6. ATP synthase leak inhibition enhances synaptic plasticity.

(A-D) Representative micrographs of primary neurons. Insets show dendritic shafts. (E) Dendritic spines were categorized according to their morphology into mature and immature spines. Illustration depicts subtypes of spines analyzed. Histograms show FX neurons have a reduced percentage of mature spines and an increased percentage of immature spines compared to WT. The types of spines are graphed as a percent of the total number of spines counted per unit length. N=5 neurons per condition from at least 2 independent cultures. (F) 5 μM Dex treatment each day for 6 consecutive days (DIV 15–20) caused an increase in the percent of mature dendritic spines / total spines per unit length in FX neurons. Dex treatment had no effect on WT neuron spine density. N= 5 neurons per condition from at least 2 independent cultures. (G) Dex normalizes dendritic ATP levels in stimulated FX neurons. ATP levels were measured in neurons at DIV 20 using FRET-based ATP reporter ATeam YEMK. Neurons were stimulated for three min. with 10 μM D-serine and ATP values were recorded at 1 hour after stimulation. Histogram shows ATP values at 1 hour after stimulation as a percentage of WT at 1 hour after stimulation. N=3 neurons per condition; 15–30 ROIs measured per neuron. (H-K) Abnormal behavior in Fmr1^−/y mice is rescued by Dex. Two-month old mice were given 3 intraperitoneal injections of 10 mg/kg Dex or saline over 40 hours prior to behavioral testing. Repetitive behaviors (grooming and nestlet shredding) were normalized by Dex in Fmr1^−/y mice. Hyperactivity as measured by locomotor activity was normalized by Dex in Fmr1^−/y mice. In Fig. 6E, F, unpaired two-tailed Student’s t-test was used. For 6G, one-way ANOVA followed by Tukey’s multiple comparisons test was used. For Figs. 6H, I, J, K, two-way ANOVA followed by Tukey’s multiple comparisons test was used. Data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001)

Dex, in addition to CsA, increases the efficiency in ATP production by ATP synthase (Alavian et al., 2015), which should phosphorylate targets such as EF2 during synaptic stimulation. To determine if local ATP levels are affected by stimulation in FX synapses, we measured ATP levels in living neurons using an ATP-FRET construct (Imamura et al., 2009b). We found that at 1 hour after synaptic stimulation, ATP levels were decreased in FX synapses compared to WT. In contrast, treatment with Dex restored post-stimulation levels of ATP in FX synapses to those of WT (Fig. 6G).

Dex rescues autistic behaviors in the Fmr1^−/y mice

Developmental synaptic plasticity is required for normal mammalian behavior (Citri and Malenka, 2008). Fmr1^−/y mice exhibit abnormal behaviors that can be measured by various paradigms. For example, they are more likely to engage in repetitive behaviors including grooming and shredding their nestlets (Angoa-Perez et al., 2013; Kalueff et al., 2016; Kane et al., 2012; Silverman et al., 2010). Fmr1^−/y mice are also hyperactive compared to WT mice, as measured by overall locomotor activity (Baker et al., 2010; Dolan et al., 2013; Tranfaglia, 2011). To determine if ATP synthase leak closure rescues these behavioral abnormalities, we injected 2-month old mice with Dex over 2 days and tested their repetitive behaviors and locomotor activity. Although Dex did not significantly alter repetitive behaviors in WT mice, it markedly reduced the time spent grooming and the number of attempts made at grooming in Fmr1^−/y animals (Fig. 6H, I). Dex also decreased the abnormal nestlet shredding behavior measured in Fmr1^−/y mice, suggesting that inhibiting the ATP synthase leak normalizes certain types of autistic behaviors (Fig. 6J). Dex also markedly reduced the hyperactivity of Fmr1^−/y mice (Fig. 6K). The amelioration of these behavioral patterns by Dex suggests that ATP synthase leak inhibition is required for the development of normal mammalian behaviors.

Discussion

We have found that the mitochondrial inner membrane leak of FX neurons and cells is caused by abnormal levels of ATP synthase c-subunit. The c-subunit leak causes persistence of a mitochondrial leak metabolic phenotype characterized by high glycolytic flux, high lactate levels and increased levels of glycolytic and TCA enzymes. The leak also aberrantly elevates overall and specific protein synthesis; a decrease in c-subunit level or pharmacological inhibition of the ATP synthase leak reduces protein synthesis rates and decreases the leak metabolism enzymes. In Fmr1^−/y synapses stimulation-dependent protein synthesis is absent. This is correlated with a lack of stimulus induced EF2 phosphorylation and lack of synthesis of the ATP synthase β-subunit. These abnormalities are readily reversed by ATP synthase leak inhibitors, suggesting that leak closure is required for the ATP-dependent phosphorylation of EF2 adjacent to mitochondria. EF2 phosphorylation may regulate the change in subsets of proteins synthesized. We find in FX neurons that there is an overabundant synthesis of lactate and of enzymes supporting a high flux glycolytic/TCA cycle “leak” metabolism, indicative of metabolic immaturity. Consistent with the hypothesis that the c-subunit leak is also a major cause of synapse immaturity, we find that inhibition of the ATP synthase leak allows the maturation of synapses and normalizes autistic behaviors.

ATP synthase leak closure regulates metabolic and synaptic maturation in an FMRP dependent manner.

Our findings highlight the complex metabolic scenario of FMRP-deficient neurons. Normally there is an increase in synaptic/neuronal activity during early neuronal development (Bailey et al., 2015; Watson et al., 2016) accompanied by a change in metabolism from glycolytic to oxidative phosphorylation (Fame et al., 2019; Zheng et al., 2016). Our findings suggest persistence of a glycolytic/TCA cycle leak metabolic phenotype at the time of synapse formation in the FX neuronal cultures. Although it has been previously shown and is well-accepted that FMRP binds to synaptic mRNAs, the specific mRNAs that regulate synapse development are not known and how they contribute to development in an FMRP dependent manner is not fully understood. The present results suggest that FMRP binds to ATP synthase β subunit mRNA to regulate the timing of metabolic maturation from a leak phenotype toward oxidative phosphorylation. It is not likely that the change in levels of glycolytic/TCA enzymes occurs by direct FMRP binding to the enzyme mRNAs. Only Hexokinase I is a possible FMRP target out of the group of metabolic enzymes that we find elevated by immunoblot (Darnell et al., 2011). Instead, when synaptic stimulation increases ATP synthase β subunit levels in an FMRP dependent manner and closes the c-subunit leak, this may change the probability of translation of a subset of mRNAs waiting near the mitochondria, thereby changing levels of metabolic enzymes required for efficient oxidative phosphorylation and decreasing those needed for the leak metabolism. Of course it is also possible that stimulation might release FMRP from a metabolic transcriptional regulator (Fame et al., 2019), thereby coordinating the decrease in glycolytic/TCA enzymes with the increase in ATP synthase enzyme components. Future analysis will discover which scenario is tractable.

ATP synthase c-subunit leak channel regulates the rate of protein synthesis in FX neurons and fibroblasts.

One of the key findings of this study is that the ATP synthase c-subunit leak channel level activity regulates the rate of protein synthesis. We demonstrate that knock down or overexpression of the c-subunit directly regulates the rate of protein synthesis measured by puromycin incorporation. Pharmacological reagents Dex and CsA that bind within the ATP synthase F₁ to reduce c-subunit leak channel activity (Alavian et al., 2015; Alavian et al., 2011; Chen et al., 2011; Giorgio et al., 2009; Szabo and Zoratti, 1991) also reduce overall protein synthesis. Finally, Bcl-xL, a protein that we have reported previously to bind to the F₁ (Alavian et al., 2011; Chen et al., 2011), reduces protein synthesis, suggesting that Bcl-xL or another endogenous ATP synthase leak modulator (Stefely and Pagliarini, 2017), could also assist in the stimulus dependent changes in protein synthesis (Suppl. Fig. 2F–H). We find that the reason the leak is so important for regulation of the rate of protein synthesis is because the increase in mitochondrial ATP produced by leak closure is used in phosphorylating local translation targets. We identified EF2, since its rapid phosphorylation after high intensity glutamate receptor stimulation (Scheetz et al., 2000) suggested it as a candidate to produce a change in synaptic plasticity. Indeed, we found that there was no phosphorylation of EF2 or change in rate of protein synthesis in Fmr1^−/y synapses after glutamate receptor stimulation. In contrast, the stimulus dependent phosphorylation event was rescued by closing the ATP synthase leak with CsA (Giorgio et al., 2009; Szabo and Zoratti, 1991).

Mitochondrial inner membrane leak closure decreases “leak” metabolism, favoring more efficient oxidative phosphorylation during synaptogenesis.

Immature cells prefer a metabolism favoring glycolytic production of ATP over mitochondrial ATP production (Warburg, 1956; Zheng et al., 2016), but the idea that a shift toward oxidative phosphorylation could occur during synaptogenesis has not been shown previously. Recent reports in cardiomyocytes support this developmental shift. Cardiomyocytes change their metabolism (from glycolytic to oxidative) over several days during embryonic development (Hom et al., 2011). The respiratory complexes aggregate into a “supercomplex” (Beutner et al., 2017). These events occur earlier in cells that have been exposed to CsA (Hom et al., 2011). Studies also support a similar scenario in early neuronal differentiation from stem cells (Fame et al., 2019; Zheng et al., 2016). Our previous work has suggested that oxidative changes in synapses during development are regulated by increases in expression of the anti-apoptotic protein Bcl-xL. Bcl-xL levels peak during periods of synaptogenesis in the developing brain, then remain elevated in adulthood (Krajewska et al., 2002). We have reported that overexpression of Bcl-xL enhances synapse formation and maturity in hippocampal neurons (Li et al., 2008a). Bcl-xL supports both mitochondrial biogenesis and movement of mitochondria closer to synaptic sites during synaptic enlargement (Berman et al., 2009; Li et al., 2008a). Accompanying this change is an improvement in the efficiency of oxidative phosphorylation as Bcl-xL interacts directly with the ATP synthase F₁ to improve enzymatic function and close the inner membrane leak (Alavian et al., 2011; Chen et al., 2011).

Unlike for cardiomyocytes as described above, at the same embryonic dates (about E9), oxygen consumption is high in the developing nervous system yet accompanied by high glycolytic flux. Different from the heart, oxygen consumption then actually decreases upon neuronal maturation as oxidative phosphorylation takes over as the main metabolic phenotype (Fame et al., 2019). Recent accounts have highlighted that metabolic phenotypes are much more complex than simply “glycolytic” vs. “oxidative” and that mitochondria are not always silent when glycolytic metabolism is favored (Li et al., 2016; Vander Heiden et al., 2009; Zheng et al., 2016). On the contrary, mitochondrial metabolism is often enhanced in that use of the TCA cycle for anabolism is increased; this includes upregulation of enzymes needed for lipid biosynthesis, protein synthesis and deoxy- and ribonucleic acid biosynthesis. Hallmarks of this state are the upregulation of enzymes involved in glutaminolysis (Wise et al., 2008), malate aspartate shuttle (Li et al., 2016) the re-supply of NAD⁺, alteration of the NAD⁺/NADH ratio (Magni et al., 2008) and changes in pyruvate metabolism (Dayton et al., 2016). Enzymes upregulated for this form of metabolism include lactate dehydrogenase, malate dehydrogenase, glutamate dehydrogenase, isocitrate dehydrogenase and PKM2 (Dayton et al., 2016). In our current study, the list of enzymes that are upregulated in FX vs. WT neurons and synapses includes all the pathways mentioned above. Most strikingly, these are rapidly downregulated in FX neurons following Dex treatment, suggesting high metabolic flexibility in early developing neurons and synapses. These results also suggest that inner membrane leak may be upregulated early in development to enhance electron transport so that NAD⁺ can be reformed from NADH. In addition, the TCA cycle may run faster in the presence of an inner membrane leak to synthesize components of developing cells/synapses. It is likely that this “leak” metabolic phenotype is advantageous to FX neurons since they require increased protein synthesis and may need enhanced lipid supply for membrane remodeling. This is supported by several recent reports including in dfmr1 mutant mitochondria in which it was described that mitochondria have significantly increased maximum electron transport system (ETS) capacity accompanied by high oxygen consumption, reduced carbohydrate and lipid stores and hyperphagia, suggestive of an inner mitochondrial membrane leak (Weisz et al., 2018). NAD⁺/NADH ratio was also significantly lower in the dfmr1 mutants relative to controls (Weisz et al., 2018), suggesting the requirement for upregulation of enzymes that resupply NAD⁺. A recent report on Fmr1^−/y mouse brains shows that electron transport complexes run at high rates even though mitochondrial ATP production is low, suggesting inner membrane inefficiency (D’Antoni et al., 2020). Our recent findings confirm that mitochondria from the Fmr1^−/y mouse brains have inefficient and membrane potential dependent enhancement in oxygen consumption caused by an inner membrane leak sensitive to coenzyme Q and CsA (Griffiths et al., 2020). Therefore, the “leak” metabolism of neuronal immaturity in our analysis of these examples is characterized by high glycolytic/TCA flux and high electron transport with an increase, not a decrease, in oxygen consumption.

Although we have been concentrating on the leak as the abnormality in this study, the ATP synthase is only one part of a complex inner mitochondrial membrane structure (Cogliati et al., 2016; Davies et al., 2012). Changes in cristae morphology at the onset of oxidative metabolism contribute to the enhanced efficiency of ATP production (Esparza-Perusquia et al., 2017). These alterations in mitochondrial inner membrane architecture may occur during normal synaptic maturation, suggesting other ways in which mitochondrial plasticity may be required for mature synapse formation.

STAR METHODS

RESOURCE AVAILABILITY

LEAD CONTACT

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Elizabeth A. Jonas (elizabeth.jonas@yale.edu).

MATERIALS AVAILABILITY

All unique/stable reagents generated in this study are available from the Lead Contact without restriction. For reagents please contact Elizabeth A. Jonas (lead contact) (elizabeth.jonas@yale.edu).

DATA AND CODE AVAILABILITY

The published article includes all datasets generated or analyzed during this study. Original data for the LC/MS/MS dataset are included in this manuscript in supplementary table 1. LC/MS/MS source files are available from Pawel Licznerski (pawel.licznerski@yale.edu) on request without restrictions.

EXPERIMENTAL MODEL AND SUBJECTDETAILS

Mice

Wild type (WT) (FVB.129P2-Pde6b⁺Tyr^c-ch/Ant) and Fmr1^−/y (FVB.129P2-Fmr1^tm1Cgr/J) mice were purchased from Jackson Laboratories (Bar Harbor, MA). All procedures were performed in accordance with the NIH Guidelines for the Care and Use of Laboratory Animals and approved by Yale University’s Institutional Animal Care and Use Committee (IACUC).

Human fibroblast lines.

Human FX and WT cell lines were kindly shared by Dr. Gary J. Bassell laboratory (Emory University, Atlanta, GA) and Dr. Elizabeth Berry-Kravis (Rush University Medical Center, Chicago, IL).

Cells were cultured in high glucose DMEM supplemented with 10% v/v FBS, 100 U/ml penicillin, 100 μg/ml streptomycin (all from Gibco).

METHOD DETAILS

Mice hippocampal primary cultures.

Primary hippocampal neurons were prepared from mouse embryos (WT or FX male and female pups at E19), as described previously (Beaudoin et al., 2012; Kaech and Banker, 2006; Li et al., 2008b). Briefly, after isolation of hippocampi from prenatal brains, neurons were dissociated and plated (0.15 × 10⁶ cells/35mm plate) in plating medium with 5% FBS. After 2–4 hr incubation, plating medium was changed to neurobasal medium supplemented with B-27, glutamine, and antibiotics (Invitrogen GIBCO life technologies, Carlsbad, CA). Neurons were grown at 37°C in a 5% CO₂ and 20% O₂ humidified incubator.

Mouse cortical primary cultures.

Cortical cultures were prepared from P0-P2 FVB (WT control) and FMRP KO pups as described in (Beaudoin et al., 2012).

Dendra translation indicator.

Plasmid coding for Dendra embedded in the 5’ and 3’ UTR of beta-actin was kindly shared by Dr. Deanna Benson (Icahn School of Medicine at Mount Sinai, New York, NY). Cells were transfected using Lipofectamine 2000 (Invitrogen) and experiments were performed at DIV14–21. When Dendra is translated in vitro, it emits green fluorescent light. Green fluorescence is photoconverted to red fluorescence by exposure to fluorescent light 400–490 nm for 2 min., after which the newly developing green fluorescence represents newly translated actin reporter. Measurements were obtained at 5 min. after photoconversion, using a Zeiss Axiovert 200 microscope and analyzed for fluorescence intensity (center of the soma) using ImageJ software.

ABT-737 treatment.

A stock solution of ABT-737 (Selleckchem, Houston, TX) was prepared in dimethyl sulfoxide (DMSO). ABT-737 (1μM) or the same volume of DMSO was added into the culture dishes for mRNA translation studies.

Cyclosporine A.

Cyclosporine A (CsA) was purchased from Cell Signaling and stock solution was prepared in dimethyl sulfoxide (DMSO). Primary neuronal cortical cultures at DIV 14–16 were treated with 0.2 to 0.5 μM CsA (final concentration) or DMSO as a control for 6 hours, harvested, lysed and processed for further analysis.

Dexpramipexole.

A dose of Dex for in vivo treatment was chosen by searching the literature and analyzing previous reports on Dex metabolism (Bozik et al., 2011; Cudkowicz et al., 2011; Muzzi et al., 2018). A stock solution of Dexpramipexole dihydrochloride (SIGMA-ALDRICH, St. Louis, MO) was prepared in sterile dH₂O and used at different concentrations described in the manuscript.

DNA plasmid transfections.

All neuronal cultures were transfected at day 5–7 DIV (days in vitro) using Lipofectamine 2000 reagent (Invitrogen), according to the manufacturer’s specifications. The same reagent was used to transfect human fibroblast lines. Vector for c-subunit expression was the same as previously used and described by (Alavian et al., 2014).

siRNA.

DsiRNA (Integrated DNA Technologies, USA) stocks were prepared according to the manufacturer’s protocol. Transfections of siRNA (final concentration: 25 pmol per well) were performed using Lipofectamine 2000 reagent (Invitrogen). Cell lysates were prepared 20–24 hours post transfection.

Control DsiRNA:

• Sense: CGUUAAUCGCGUAUAAUACGCGUAT

• Antisense: AUACGCGUAUUAUACGCGAUUAACGAC

c-subunit DsiRNA (hs.Ri.ATP5G1.13.1):

• Sense: CCAGUGAAUUCAUCUAAACAGCCTT

• Antisense: CGGGUCACUUAAGUAGAUUUGUCGGAA

Purification of recombinant Bcl-xL.

Flag-tagged Bcl-xL (Li et al., 2013) was immunoprecipitated from HEK293T cell lysates using the EZview Red ANTI-FLAG M2 Affinity Gel (Sigma-Aldrich) according to the manufacturer’s protocol. The purified protein samples were examined by western blot.

Puromycin incorporation for measurement of protein synthesis.

Puromycin (Puromycin dihydrochloride, SIGMA-ALDRICH) labeling was performed as previously described (Schmidt et al., 2009). Briefly, cells or synaptosomes were incubated in puromycin containing media (10 μg/ml) for 15 minutes, washed twice with cold PBS, lysed in 1xRIPA buffer (Cell Signaling) containing proteinase inhibitors (Roche, Indianapolis, Indiana) and phosphatase inhibitors PhosSTOP (Roche Diagnostics GmbH, Mannheim, Germany).

For in vivo studies adult (2 month or older) mice were injected (IP) with (30 mg/kg) puromycin (SIGMA ALDRICH). After 2.5 hr animals were sacrificed and brain and liver samples were harvested and homogenized in 1xRIPA buffer (Cell Signaling) containing proteinase inhibitors (Roche, Indianapolis, Indiana) and phosphatase inhibitors PhosSTOP (Roche Diagnostics GmbH, Mannheim, Germany) and processed for western blot analysis. Alternatively, adult mice were sacrificed and 200 micrometer brain slices containing prefrontal cortex were incubated in hippocampal recording buffer (with 95% O₂ and 5% CO₂) containing (in nmol): 125 mM NaCl, 25 mM NaHCO₃, 2.5 mM KCl, 25 mM glucose, 1.25 mM NaH₂PO₄, 1 mM MgCl₂, 2 mM CaCl₂ bubbled with 95%O₂, 5% CO₂, for 2.5 hr. with Dexpramipexole (final concentration 10 μM). For the last 15 minutes of incubation puromycin was added to the bath (final 10 μg/ml). Next, brain slices were homogenized in 1xRIPA buffer (Cell Signaling) containing proteinase inhibitors (Roche, Indianapolis, Indiana) and phosphatase inhibitors PhosSTOP (Roche Diagnostics GmbH, Mannheim, Germany) and processed for western blot analysis.

Immunoprecipitation of puromycin labelled peptides.

Isolated synaptosomes or P0–P2 FVB or FX DIV14–16 cortical cultures were treated for 15 min. with puromycin. 50–100 μg of protein from cell lysate, was incubated at 4°C overnight with 1μg of anti-puromycin antibody (Kerafast). Next, 50 μl of protein G agarose (Roche Diagnostics GmbH) was added to samples for an overnight incubation at 4°C. Beads were washed 3 times with 1x RIPA buffer (Cell Signaling) and processed for mass spectrometry analysis. For experiments with Dexpramipexole, cortical cultures were treated with 10μM Dex for 2–24 hours prior to puromycin treatment and then processed for western blot.

Co-immunoprecipitation of FMRP and the beta- and c-subunit of the ATP synthase.

WT and FX purified synaptosomal lysates were incubated at 4°C overnight with 10 μg of mouse anti-FMRP 7G1–1 (developed by Stephen T. Warren, this antibody was obtained from the Developmental Studies Hybridoma Bank developed under the auspices of NICHD and maintained by the University of Iowa, Department of Biological Sciences, Iowa City IA 55242) or 10 μg of mouse IgG. Next, immunoprecipitates were immobilized on A/G agarose beads (Santa-Cruz Biotechnology) at 4°C overnight and washed 4–5×15 minutes with 1x RIPA lysis buffer (Cell Signaling) with 40 U/ml protector RNase inhibitor (Roche) and 1x Complete EDTA-free protease inhibitor cocktail tablet (Roche). mRNA was purified from beads using the quickRNA purification kit (Qiagen) according to the manufacturer’s protocol. RNAs were then reverse transcribed using a cDNA reverse transcription kit (Biorad) according to the manual. Next, a standard PCR reaction was run using the following primers:

• GGCAAGATGGGGTATAGAGA; Map1b forward primer

• CCCACCTGCTTTGGTCTTTG; Map1b reverse primer

• AAGCTGGAGAACAACTTGGAC; Arc forward primer

• CCCCCAAGACTGATATTGCTGAG, Arc reverse primer

• Primer sequences above have been published by (Brown et al. 2010).

• GCCAGAGACTATGCGGCGCAG; mATP5B forward primer

• GGACCTCTCTCATCAATAGG; mATP5B reverse primer

• GGCCTGTGTCTGCCTCCCTCC; mATP5G1 forward primer

• GGCGACCATCAAACAGAAGAG; mATP5G1 reverse primer

• GAGCACCTCTCAGCTGCTGAGTCG; mATP5G2 forward primer

• GGCCTCTGAGAGGGCAAAGCC; mATP5G2 reverse primer

• GTTCGCCTGCGCCAAGCTCGC; mATP5G3 forward primer

• GTGAAGGGTTTCAGCACCAG; mATP5G3 reverse primer

Western blot analysis.

Brain tissue or primary neuronal culture lysates or mitochondrial lysates were prepared using RIPA lysis buffer (Cell Signaling) containing proteinase inhibitors (Roche, Indianapolis, Indiana) and phosphatase inhibitors PhosSTOP (Roche Diagnostics GmbH, Mannheim, Germany). The protein concentration was measured using a BCA kit (Pierce, Rockford, Illinois). Then, protein samples were electrophoretically separated on an SDS-PAGE gel (4–20% gradient gel, Bio-Rad, USA) and transferred overnight to PVDF membranes (0.2 μm pores, Bio-Rad, USA). The membranes were incubated in 2% BSA (Tris-buffered saline (TBS), 0.1% Tween 20) for 1 h and then incubated at 4°C overnight with primary anti-Bcl-xL (1:1000, Cell Signaling), anti-puromycin (1:1000, Kerafast), anti-GAPDH (1:1000, Santa Cruz Biotechnology), anti-PSD-95 (1:1000, Cell Signaling), anti-p-EF2 (1:1000, Cell Signaling), anti-EF2 (1:1000, Santa Cruz Biotechnology), anti-ATP5g1/2/3 (1:1000, abcam), anti-ATPb (1:1000, abcam), anti-beta-actin (1:1000, Cell Signaling), anti-FMRP (1:1000, Cell Signaling). Antibodies for glycolytic enzyme detection were supplied by Glycolysis Antibody Sampler Kit #8337, Cell Signaling and used at 1:1000 dilution. After 3 × 15 min washes membranes were incubated for 1 h with secondary, horseradish peroxidase (HRP) conjugated antibodies (1:5000, Cell Signaling) and developed using a chemilluminescence kit (Pierce, Rockford, Illinois).

Blue Native Page Electrophoresis.

Protein complexes from 20 μg (verified by BCA assay) of mitochondria (per lane) were separated on Bis-Tris 3–12% Native gels. Samples were solubilized on ice for 20 minutes with 4 μg digitonin/μg protein. After separation the protein complexes were wet-transferred onto a polyvinylidene fluoride (PVDF) membrane, which was probed with anti-ATP5G1,2,3 antibody for ATP synthase c-subunit.

Isolation of mitochondria.

Mitochondria were isolated and purified from mouse brain as previously described (Ofengeim et al., 2012; Sacchetti et al., 2013). In brief, brain tissue was homogenized in isolation buffer (250mM sucrose, 20mM HEPES, 1mM EDTA, 0.5% BSA). After a series of centrifugations, the nuclear material, cytosolic fraction and the mitochondrial pellet containing synaptosomes were separated. Synaptosomes were disrupted by applying 1200 psi pressure for 10min and mitochondria were separated by ultracentrifugation (Brown et al., 2004). To isolate mitochondria from human fibroblasts, a Qproteome™ Mitochondria isolation kit was used (Qiagen, Cat.37612), according to the manufacturer’s protocol.

Isolation of SMVs from mouse brain.

FVB or FMRP KO mouse brain tissue (without cerebellum) was homogenized in ice-cold isolation buffer (250 mM sucrose, 20 mM Hepes (pH 7.2), 1 mM EDTA, and 0.5% BSA). After a brief centrifugation at 1,500g, the supernatants were centrifuged at high-speed (16,000×g) for 10 min at 4 °C. The crude pellets were re-suspended in isolation buffer and a pressure of 1,200psi was applied for 10min, followed by rapid decompression. The pure mitochondrial fraction was then pelleted in a ficoll density gradient by centrifugation and washed with isolation buffer. Non-ionic detergents (digitonin and Lubrol-PX) were used to further solubilize and stabilize membrane-bound protein complexes, and the sub mitochondrial vesicles (SMVs) were isolated by a final 2 hr ultracentrifugation (Chan et al., 1970; Sacchetti et al., 2013). Freshly prepared SMV protein amount was quantified using the Bradford protein assay

Preparation of synaptosomal fractions.

Synaptosomal fractions were prepared as previously described (Alavian et al., 2011). Briefly, cultured neurons were homogenized in isotonic mitochondrial buffer and centrifuged at 600 × g for 10 min at 4°C. The pellet, containing the nuclear and unbroken cells, was discarded and the supernatant, containing the mitochondrial, synaptosomal and cytosolic fractions, was centrifuged at 10,000g for 30 min at 4 °C. The supernatant, containing the cytosolic fraction, was separated from the pellet. The pellet containing the mitochondrial and synaptosomal fractions was resuspended in 100 μl isolation buffer and layered onto a 7.5–10% Ficoll gradient. After 30 min ultracentrifugation at 90,000g, 4 ° C, the mitochondrial pellet and the middle layer, containing synaptoneurosomes (hereafter called synaptosomes) were removed. The synaptosomal layer was resuspended in isolation buffer and centrifuged for 10 min at 20,000g, resulting in a crude synaptosomal pellet.

D-serine stimulation of synaptosomes.

Synaptosomal samples purified from FVB and FMRP KO mice and were stimulated with D-serine (final concentration 0.2 μM), supplied in extracellular (EC) recording buffer without magnesium (120mM NaCl, 4mM KCl, 2mM CaCl₂, 10mM HEPES, 10mM D-Glucose, pH=7.4). 1μg of puromycin was also added to measure protein synthesis. EC buffer lacking D-serine was added to unstimulated control samples. Samples were harvested and lysed in 1xRIPA at 3, 5, 15 and 30 minutes post stimulation.

Quantitative Real Time RT-PCR.

Total RNA was extracted from FVB and FX synaptosomes using RNeasy Plus Mini Kit according to the manufacturer’s protocol (Qiagen). Next, extracted RNA was reverse transcribed using Bio-Rad iScript first cDNA synthesis kit. TaqMan® Gene Expression Assays (Thermo Fisher Scientific, USA) was used to quantify mRNA levels. Data were analyzed using the 2^−ΔΔCT method using beta-actin as the normalizing endogenous control. The following probes were used: mATP5G1 (Mm02601566_g1), mATP5G2 (Mm00848143_g1), mATP5G3 (Mm01334541_g1), mATP5B (Mm01160389_g1), mActB (Mm02619580_g1).

Measurement of ATP levels in Figure 1.

Primary cortical neurons were seeded onto 96 well plates (0.015 × 10⁶ neurons/ well). After 1–2 weeks incubation, cells were treated as stated in relevant figure legends. ATP production was measured by using ATPlite™ Luminescence Assay System (PerkinElmer, Waltham, MA) according to the manufacturer’s protocol. Cells were washed with sterile PBS, lysed and incubated with substrate (luciferin) for 15 min. The reaction between ATP, luciferase and luciferin produced bioluminescence. ATP-induced-luminescence was measured with a VICTOR³ multilabel reader (PerkinElmer, Waltham, MA).

Measurement of mitochondrial potential (Δψ).

Mitochondrial membrane potential (Δψ) was measured using the fluorescent lipophilic cationic dye tetramethylrhodamine methyl ester (TMRM, Invitrogen, Molecular Probes, Carlsbad, CA, USA), which accumulates within mitochondria in a membrane potential-dependent manner. Primary hippocampal neurons were stained with 5 nM TMRM for 30 min at 37 °C in the dark. Images were taken using a Zeiss LSM 710 confocal scanning microscope and TMRM fluorescence densitometry was analyzed using ZEN software (Carl Zeiss Microscopy GmbH, Jena, Germany).

ACMA assay.

ACMA (9-amino-6-chloro-2-methoxyacridine, Sigma A5806) fluorescence quenching was measured as previously described (Alavian et al., 2011) with some modifications. In brief, 2μM ACMA and 50 μg of the isolated mouse brain submitochondrial vesicles (SMVs), were used. SMV suspension fluorescence was measured at 490 nm using a PerkinElmer VICTOR3 (PerkinElmer, Waltham, MA) multilabel plate reader.

Lactate assay.

Growth media were collected at DIV20 from WT and FX neuronal primary cultures treated with either 5μM Dex or vehicle daily from DIV15 to DIV20. Lactate levels in the growth medium samples were calculated using the L-Lactate assay Kit I (Eton Bioscience, Charlestown, MA) according to the manual provided.

SMV ion channel recordings.

SMV recordings were made by forming giga-ohm seals onto SMVs in intracellular solution (120 mM KCl, 8 mM NaCl, 0.5 mM EGTA, 10 mM, Hepes (pH 7.3)) using an Axopatch 200B amplifier (Axon Instruments) at room temperature (22–25 °C). Recording electrodes were pulled from borosilicate glass capillaries (WPI) with a final resistance in the range of 50–120 MΩ. Signals were filtered at 5 kHz using the amplifier circuitry. Data were analyzed using pClamp 10.0 software (Axon Instruments). Membrane currents under different experimental conditions were assessed by measuring peak membrane current (in pA). All current measurements were adjusted for the holding voltage assuming a linear current-voltage relationship: The resulting conductances are expressed in pS according to the equation G = I/V where G is conductance in pS, V is the membrane holding voltage in mV, and I is the peak membrane current in pA. Group data were quantified in terms of conductance. All population data were expressed as mean ± SEM.

Brain Electron microscopy.

Brain slices from FVB (WT) and FMRP KO adult mice (over 2 month-old) were incubated in artificial cerebral spinal fluid (ACSF) containing (in mM): 125 mM NaCl, 25 mM NaHCO₃, 2.5 mM KCl, 25 mM glucose, 1.25 mM NaH₂PO₄, 1 mM MgCl₂, 2 mM CaCl₂ bubbled with 95%O₂, 5% CO₂ with vehicle or 0.2μM Cyclosporin A for 2 hr 45 min. Samples were fixed in 4% paraformaldehyde in 0.25M Hepes for 1 hour. Samples were rinsed in PBS and re-suspended in 10% gelatin, chilled and trimmed to smaller blocks and placed in cryoprotectant of 2.3M sucrose overnight on a rotor at 4°C. They were transferred to aluminum pins and frozen rapidly in liquid nitrogen. The frozen block was trimmed on a Leica Cryo-EMUC6 UltraCut and 65–75nm thick sections were collected using the Tokoyasu method. The frozen sections were collected on a drop of sucrose, thawed and placed on a nickel formvar/carbon coated grid and floated in a dish of PBS ready for immunolabeling. Grids were placed section side down on drops of 0.1M ammonium chloride to quench untreated aldehyde groups, then blocked for nonspecific binding on 1% fish skin gelatin in PBS. All grids were rinsed in PBS, fixed using 1% glutaraldehyde for 5mins, rinsed and transferred to a UA/methylcellulose drop, then dried for viewing. Samples were viewed FEI Tencai Biotwin TEM at 80Kv. Images were taken using Morada CCD and iTEM (Olympus) software. Analysis of electron micrographs was performed using ImageJ software (NIH).

Dendritic spine density analysis and ATP measurement.

Hippocampal neurons were isolated from E18.5–19 FMRP KO and WT mice using enzymatic digestion with Trypsin EDTA (Gibco) followed by mechanical trituration and then cultured in Neurobasal media with 1 X B27 supplement (Gibco). Half of the medium was replaced with fresh medium every week. Cells were transfected at DIV1 with the ATeam ATP reporter construct (Imamura et al., 2009a) (kindly provided by Dr. Imamura and Dr. Noji from Osaka University, Osaka, Japan) using Lipofectamine 3000 (Thermo Fisher) as per manufacturer’s recommendations. Cells were treated with 5 μm Dexpramipexole additively for 6 consecutive days from DIV 15 – DIV 20. At DIV 20 conditioned medium was collected for measurement of lactate levels.

For quantification of dendritic spines, cells were fixed at DIV 20 with 4% buffered formalin and immune-stained with anti-GFP antibody (Abcam, ab13970). Images were acquired with a Zeiss 880 Airyscan microscope. Dendritic spines were accessed visually on the basis on their morphology; mature spines were identified by their mushroom-like shape; all other spines were considered immature (see illustration in Fig. 6).

ATP levels were measured in neurons at DIV 20 using the well characterized FRET based ATP reporter ATeam YEMK (Imamura et al., 2009a) in a Hepes (10 mM) based buffer containing NaCl (125 mM), KCl (3 mM), CaCl₂ (2 mM), MgCl₂ (2 mM) and D-Glucose (5 mM). For LTP stimulation of neurons, 10 μM D-Serine was applied in the same buffer but lacking MgCl₂ for 5 min. FRET measurements were performed with a Zeiss 710 confocal microscope equipped with a controlled atmosphere cabinet at 25 °C. Measurements of changes in pH using a pH sensitive indicator were performed separately and did not show any significant changes in pH before or after stimulation during the times of acquisition of ATP signal.

Mass Spectrometry.

Following separation of protein complexes in one-dimension by SDS-PAGE, protein bands of interest were excised for bottom-up protein identification by LC/MS/MS. Gel bands were prepared as described (Glass et al., 2017). Briefly, excised gel bands in 1.5 Eppendorf tubes are washed 4 times; first with 500 μL 60% acetonitrile containing 0.1%TFA and then with 5% acetic acid, then with 250 μL 50% H₂O/50% acetonitrile followed by a 250 μL 50% CH₃CN/ 50 mM NH₄HCO₃, and a final wash with 250 μL 50% CH₃CN/10 mM NH₄HCO₃ prior to removal of wash and complete drying of gel pieces in a Speed Vac. 10 μL of a 0.1 mg/mL stock solution of trypsin (Promega Trypsin Gold MS grade) in 5mM acetic acid is freshly diluted into a 140 μL solution of 10mM NH₄HCO₃ to make the working digestion solution. 124 μL of the working digestion solution is added to the dried gels pieces (additional 10 mM NH₄HCO₃ was added to ensure gel pieces are completely submerged in the digestion solution) and incubated at 37 °C overnight. Sample is then stored at −20 °C until analysis. Tryptic peptides were separated on a nanoAcquity™ UPLC™ column (Waters) coupled to a Q-Exactive Plus mass spectrometer. High resolution tandem LC MS/MS data were collected by Higher-Energy Collisional Dissociation (HCD) with a 1.4 Da window followed by normalized collision energy of 32%. Resulting LC MS/MS data were analyzed and processed through Proteome Discoverer (v.2.2 and linked to MASCOT search engine v.2.4) and further integrated with Scaffold (v.4.8, Proteome Software Inc.).

Behavioral experiments.

Male FVB (WT) and Fmr1^−/y mice 2 months of age were used for all experiments. All animal procedures were in accordance with US National Institutes of Health standards and approved by the Yale University Institutional Animal Care and Use Committee. Prior to behavioral testing mice were handled individually by the investigator to decrease anxiety. Next, mice received 3 IP injections of Dex (10 mg/kg) over the course of 40 hours: two injections separated by 24 hour period and the third 16 hours after the second injection. Behavioral testing started 2–3 hours after the last (third) injection. For repetitive behaviors (grooming and nestlet shredding (Angoa-Perez et al., 2013; Silverman et al., 2010)) mice were placed in a new, empty home cage and their behavior was monitored during 10 minute sessions, video recorded and scored manually. Grooming was identified as body licking or stroking, scratching of the head or body with the two forelimbs. Attempts at grooming were defined as a total number of grooming events during the 10 minute session. For nestlet shredding mice were placed in a new empty home cage without bedding with one cotton nestlet and recorded for 10 minutes. Nestlet shreds were collected and weighed. Exploratory locomotion (total time moving, walking plus running) was assessed during a 5 minute session, recorded and scored manually (Baker et al., 2010; Dolan et al., 2013). The investigator was blinded as to the genetic variant during scoring.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis was performed using Prism 8 (GraphPad Software, San Diego, CA). Data are presented as mean ± SEM. Paired or unpaired Student’s two-tailed t test was used for two group comparisons. For multiple comparisons, one-way or two-way ANOVA test with Tukey post-hoc test was used. Statistical details and methods used in each experiment can be found in figures and in the figure legends. p<0.05 is considered statistically significant. p values are provided in figure legends (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001).

Supplementary Material

1

Supplementary Figure 1. Related to main Figure 2 and 5. qRT-PCR expression profile of c-subunit (ATP5G2) and β-subunit mRNA from synaptosomes. Shown is fold change of Fmr1^−/y over WT (N=4 and 3 different animals, *p<0.05). Unpaired two-tailed Student’s t-test was used. Data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001)

2

Supplementary Figure 2. Related to Figure 4. Protein synthesis rate is elevated in FX neurons. (A) Representative images of Dendra in isolated cortical neurons before, just after, and 5 minutes after photoconversion (see methods). (B) The rate of translation of Dendra during 5 min after photoconversion in WT and FX mouse neurons. (C) The rate of translation of Dendra in WT neurons exposed to vehicle (DMSO) or 1 μM Bcl-xL inhibitor, ABT-737 for 20 min. (D, E) Puromycin incorporation over 15 minutes in WT cortical neurons exposed to 1 μM ABT-737 or vehicle control (DMSO) for 1 hour prior to puromycin application. (F, G) Example of puromycin assay before and after transfection of recombinant Bcl-xL protein (0.045–0.79 mg/ml) into synaptosomes. Bcl-xL represses protein translation in FX but not in WT. (F) Group data for puromycin incorporation for WT control samples. (G) Group data for puromycin incorporation for FX samples from at least three different animals (*p=0.037). Unpaired two-tailed Student’s t-test. Data are represented as mean ± SEM. (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001)

3

Supplementary Table 1. Full dataset from LC/MS/MS experiments. Related to Figure 3. (N=3 for each condition: WT, FX, FX+Dex). For experiments with Dex, cortical cultures were treated with 10μM Dex for 2 hours prior to puromycin treatment and then processed for LC/MS/MS. Please refer to method section for details.

4

Supplementary Table 2. Oligonucleotide sequences used in Figure 2 and 4 and Suppl. Fig. 1.

Highlights.

1. ATP synthase c-subunit leak in Fragile X causes aberrant metabolism

2. Changes in ATP synthase component stoichiometry regulate protein synthesis rate

3. Inhibition of the leak normalizes synaptic spine morphology and Fragile X behavior