|Home | About | Journals | Submit | Contact Us | Français|
A major goal of biomedical research has been the identification of molecular mechanisms that can enhance memory. Here we report a novel signaling pathway that regulates the conversion from short- to long-term memory. The mTOR complex 2 (mTORC2), which contains the key regulatory protein Rictor (Rapamycin-Insensitive Companion of mTOR), was discovered only recently, and little is known about its physiological role. We show that conditional deletion of rictor in the postnatal murine forebrain greatly reduces mTORC2 activity and selectively impairs both long-term memory (LTM) and the late (but not the early) phase of hippocampal long-term potentiation (LTP). Actin polymerization is reduced in the hippocampus of mTORC2-deficient mice and its restoration rescues both L-LTP and LTM. More importantly, a compound that selectively promotes mTORC2 activity converts early-LTP into late-LTP and enhances LTM. These findings indicate that mTORC2 could be a novel therapeutic target for the treatment of cognitive dysfunction.
How memories are stored in the brain is a question that has intrigued mankind over many generations. While neuroscientists have already made great strides indentifying key brain regions and relevant neuronal circuits, many questions regarding the specialized molecular and neuronal mechanisms underlying memory formation remain unanswered. Post-translational modifications of synaptic proteins can explain transient changes in synaptic efficacy, such as short-term memory (STM) and the early phase of LTP (E-LTP, lasting 1-3 hours); but new protein synthesis is required for long-lasting ones, such as LTM and the late phase of LTP (L-LTP, lasting several hours)1-7. Changes in actin dynamics that mediate structural changes at synapses are also necessary for L-LTP and for LTM storage8-10. However, relatively little is known about the molecular mechanisms that underlie these processes.
The evolutionarily conserved mammalian target of rapamycin (mTOR) forms two complexes11-13. The first, mTORC1, consist of mTOR, Raptor and mLST8 (GβL), is sensitive to rapamycin, and is thought to regulate mRNA translation3. Although significant progress has been made in the identification of the mTORC1 pathway and understanding its function in cells and in vivo, much less is known about the second complex, mTORC2. mTORC2 is largely insensitive to rapamycin and contains the core components mTOR, mSIN1, mLST8 and Rictor (Rapamycin-Insensitive Companion of mTOR). Rictor is a defining component of mTORC2 and its interaction with mSIN1 appears to be required for mTORC2 stability and function14. Rictor is associated to membranes and is thought to regulate the actin cytoskeleton, but the precise molecular mechanism behind this effect remains unclear11,12,16. In addition, while little is known about mTORC2’s up-stream regulation, we are beginning to understand its downstream regulation and effectors: mTORC2 phosphorylates AGC kinases at conserved motifs, including Akt at the hydrophobic motif (HM) site (Ser-473), the best characterized read out of mTORC2 activity11,12.
The key component of mTORC2, Rictor, is important for embryonic development as mice lacking rictor die in early embryogenesis15,16. Rictor is highly expressed in the brain, notably in neurons16, and it seems to play a crucial role in various aspects of brain development and function. For example, genetic deletion of rictor in developing neurons disrupts normal brain development, resulting in smaller brains and neurons, as well as increased levels of monoamine transmitters and manifestations of cerebral malfunction suggestive of schizophrenia and anxiety-like behaviors17,18.
Given that i) actin polymerization is critically required for memory consolidation8-10, ii) mTORC2 appears to regulate the actin cytoskeleton11,12,19, and iii) mTORC2’s activity is altered in conditions associated with memory loss, such as aging and several cognitive disorders, including Huntington’s disease, Parkinsonism, Alzheimer-type dementia and Autism Spectrum Disorders20-25, we decided to investigate its potential role in memory formation, specifically in sustained changes in synaptic efficacy (LTP) in hippocampal slices, and in behavioral tests of memory. Our results show that through regulation of actin polymerization, mTORC2 is an essential component of memory consolidation. Briefly, we report here a selective impairment in L-LTP and LTM in mice and flies deficient in TORC2 signaling. Moreover, we have identified the up-stream synaptic events which activate mTORC2 in the brain and unraveled the detail downstream molecular mechanism by which mTORC2 regulates L-LTP and LTM, namely regulation of actin polymerization. Finally, a small molecule activator of mTORC2 and actin polymerization facilitates both L-LTP and LTM, further demonstrating that mTORC2 is a new type of molecular switch that controls the consolidation of a short-term memory process into a long-term one.
Pharmacological inhibitors of mTORC2 are not available, and mice lacking rictor in the developing brain show abnormal brain development. To circumvent this problem, we conditionally deleted rictor in the postnatal forebrain by crossing “floxed” rictor mice16 with the α subunit of calcium/calmodulin-dependent protein kinase II (αCaMKII)-Cre mice26, generating rictor forebrain-specific knockout mice (rictor fb-KO mice; see Methods and Supplementary Fig. 10). Because the αCaMKII promoter is inactive before birth27, this manipulation diminishes possible developmental defects caused by the loss of rictor.
Rictor fb-KO mice are viable and develop normally. They show neither gross brain abnormalities nor changes in the expression of several synaptic markers (Supplementary Fig. 1). mTORC2-mediated phosphorylation of Akt at Ser473 (an established readout of mTORC2 activity11,12) was greatly reduced in CA1 and amygdala (Fig. 1a-b), but was normal in the midbrain (Fig. 1c) of rictor fb-KO mice. By contrast, in mTORC2-deficient mice, mTORC1-mediated phosphorylation of S6K1 at Thr389 (a well-established readout of mTORC1 activity28) remained unchanged in CA1, amygdala or midbrain (Fig. 1a-c). Thus, conditional deletion of rictor selectively reduces mTORC2 activity in forebrain neurons.
To investigate the role of mTORC2 in synaptic function, we first showed that mTORC2 is activated in CA1 by either glutamate [via NMDA receptor (NMDAR)] or neurotrophins (Supplementary Fig. 2). Next, to determine whether short-term or long-term changes in synaptic potency alter mTORC2 activity, we compared the effects of one train of tetanic stimulation (100 Hz for 1s), which usually induces only short-lasting E-LTP, with that of four such trains (which typically induce a long-lasting L-LTP)1. Only an L-LTP-inducing stimulation consistently activated mTORC2 in CA1 neurons of control (Fig. 1d-f) but not rictor fb-KO mice (Fig. 1g-h). Hence, mTORC2 is engaged selectively in long-lasting synaptic changes in synaptic strength.
We then examined whether mTORC2 deficiency affects either E-LTP or L-LTP. Whereas a single train of tetanic stimulation generated a similar E-LTP in slices from rictor fb-KO and control littermates (Fig. 1i), four trains elicited a normal L-LTP in control littermates slices, but not in rictor fb-KO slices (Fig. 1j). Several tests showed that the impaired L-LTP in mTORC2-deficient slices cannot be attributed to defective basal synaptic transmission (Supplementary Fig. 3). Thus, reducing mTORC2 activity prevents the conversion of E-LTP into L-LTP.
Since L-LTP-inducing stimulation increases mTORC2 activity, we then investigated whether mTORC2 is activated as a result of behavioral learning. Contextual fear conditioning, induced by pairing a context (conditioned stimulus; CS) with a foot shock (unconditioned stimulus; US), resulted in sharp temporarily increase in mTORC2 activity and phosphorylation of the p21-activated kinase PAK (a key regulator of actin cytoskeleton dynamics29,30) 15 min after training (Fig. 2a-b). In contrast, the shock-alone (US) and the context-alone (CS) failed to increase mTORC2 activity (Fig. 2c). Thus, hippocampal mTORC2 is selectively activated by behavioral learning (CS+US).
We next studied memory storage in two forms of Pavlovian conditioning, contextual and auditory fear conditioning. Contextual fear conditioning involves both the hippocampus and amygdala whereas auditory fear conditioning, in which the foot shock (US) is paired with a tone (CS), requires only the amygdala31. When mice were subsequently exposed to the same CS, fear responses (“freezing”) were taken as an index of the strength of the CS-US association. Rictor fb-KO mice and control littermates showed similar “freezing” behavior before training (naïve) and 2 hr after training, when their STM was measured (Fig. 2d-e). However, when examined 24 hr after training, both contextual and auditory LTM were significantly impaired in mTORC2-deficient mice (Fig. 2d-e). The less pronounced change in auditory fear LTM vs. contextual fear LTM in rictor fb-KO mice may be explained by the smaller reduction in mTORC2 activity in the amygdala (compare Fig. 1a vs. Fig. 1b). Spatial LTM was also deficient in rictor fb-KO mice when tested in the Morris water maze, where animals use visual cues to find a hidden platform in a circular pool32. Compared to controls, rictor fb-KO mice took significantly longer to find the hidden platform (Fig. 2f), and in the probe test, performed on day 7 in the absence of the platform, they failed to remember the platform location (target quadrant; Fig. 2g). The impaired spatial LTM was probably not caused by deficient visual or motor function since control and rictor fb-KO mice performed similarly when the platform was visible (Supplementary Fig. 4) and showed no significant difference in swimming speed (19.8 ± 0.6 vs. 19.6 ± 0.5 cm/s for control and rictor fb-KO mice). Hence, mTORC2 selectively fosters long-term memory processes.
Because TORC2 is evolutionarily conserved11,12, we also wondered whether its function in memory formation is maintained across the animal phyla. To this end, we studied olfactory memory in wild-type controls (Canton-S) and Drosophila TORC2 (dTORC2)-deficient fruit flies. As expected, in the brain of rictor mutant flies (rictorΔ1)33 dTORC2-mediated phosphorylation of the hydrophobic motif of Drosophila Akt (dAkt) at Ser505 (an established readout of dTORC2 activity34) was greatly reduced (Fig. 3a). As in mammals, LTM in Drosophila is protein-synthesis dependent35 and it is generated after “spaced” training (5 training sessions with a 15-min rest interval between each; Fig. 3b). By contrast, “massed training” (5 training sessions with no rest intervals) fails to elicit LTM but rather induces the anesthesia-resistant, protein-synthesis independent memory (ARM; Fig. 3b)35. Although responses to olfactory stimuli or electric shocks did not differ significantly between rictorΔ1 and control flies (Fig. 3c-e), in rictorΔ1 flies LTM was blocked, whereas the short-lasting ARM was unaffected (Fig. 3f). Thus, both in fruit flies and mice TORC2 promotes LTM storage.
We also probed the molecular mechanism by which mTORC2 regulates L-LTP and LTM by first testing whether mTORC2 deficiency impairs actin dynamics in CA1 neurons in vivo. Actin exists in two forms: monomeric globular actin (G-actin) and polymerized filamentous actin (F-actin) composed of aggregated G-actin. The transition between these two forms is controlled by synaptic activity8,9. The ratio of F-actin to G-actin, which reflects the balance between actin polymerization and depolymerization, was significantly reduced in CA1 of rictor fb-KO mice (Fig. 4a-b). Since Rho-GTPases have been identified as key intracellular signaling molecules that regulate actin dynamics at synapses36, we measured the activity of Rho-GTPases in CA1 of mTORC2-deficient mice. Rac1 (Ras-related C3 botulinum toxin substrate 1) and Cdc42 (cell division cycle 42), two major Rho GTPases, induce actin polymerization by promoting PAK and Cofilin phosphorylation8. Rac1-GTPase activity (but not Cdc42 activity) and the phosphorylation of PAK and Cofilin were greatly diminished in CA1 neurons of rictor fb-KO mice (Fig. 4c-f). Moreover, we found that the Rac1-specific guanine nucleotide-exchange factor (GEF) Tiam1 links Rictor (mTORC2) to Rac1 signaling (Fig. 4g-k). Finally, dendritic spine density in CA1 pyramidal neurons was significantly reduced in rictor fb-KO mice (Fig. 4l). Thus, in the adult hippocampus, mTORC2 regulates actin dynamics-mediated changes in synaptic potency and architecture.
If L-LTP is impaired in mTORC2-deficient slices because actin polymerization is abnormally low, increasing the F-actin/G-actin ratio should convert the impaired, short-lasting LTP elicited by four tetanic trains into a normal L-LTP. We therefore predicted that jasplakinolide (JPK), a compound which directly promotes actin polymerization37, should restore the normal function. Indeed, a low concentration of JPK (50 nM), raised the low F-actin/G-actin ratio and restored L-LTP in rictor fb-KO slices (Fig. 5a-b), but had no effect on wild-type (WT) slices (Fig. 5a, c) or on baseline synaptic transmission in rictor fb-KO slices (Supplementary Fig. 5a). In addition, cytochalasin D, an inhibitor of actin polymerization, blocked L-LTP in WT slices (Fig. 5c), but had no effect on the short-lasting LTP evoked either by a single tetanic train in control slices (Supplementary Fig. 5b) or by repeated tetanic stimulation in mTORC2-deficient slices (Fig. 5b). The deficient L-LTP in mTORC2-deficient slices is therefore primarily caused by impaired actin polymerization.
To determine whether deficient actin dynamics underlie the impaired LTM in rictor fb-KO mice, we bilaterally infused JPK into the CA1 region (Supplementary Fig. 6), at a low dose (50 ng) that promoted F-actin polymerization only in rictor fb-KO mice (Supplementary Fig. 7a). JPK infused immediately after training boosted contextual LTM (Fig. 5d), but had no comparable effect on hippocampus-independent auditory LTM (Fig. 5e) in rictor fb-KO mice or on contextual LTM in WT mice (Fig. 5f). These pharmacogenetic rescue experiments provide strong evidence that deficient actin dynamics account, at least in part, for the impaired LTM in rictor fb-KO mice.
Given that i) a short-lasting LTP is evoked by either repeated tetanic stimulation in mTORC2-deficient slices (Fig. 1i) or by a single tetanic train in control slices (Fig. 1h) and ii) JPK restored the deficient L-LTP in mTORC2-deficient slices (Fig. 5b), we predicted that JPK would facilitate the induction of L-LTP in control slices. Combining JPK with a weak stimulation, that normally elicits only a short-lasting E-LTP, we found that indeed JPK lowers the threshold for the induction of L-LTP in WT slices (Fig. 5g).
Having shown that boosting actin polymerization converts short-lasting LTP into long-lasting LTP, we next wondered whether this JPK-facilitated L-LTP depended on new protein synthesis. We found that the sustained LTP induced by a single train at 100 Hz in combination with JPK was blocked by the protein synthesis inhibitor anisomycin (Supplementary Fig. 8a). Furthermore, JPK could not rescue the impaired L-LTP induced by four trains at 100 Hz in the presence of anisomycin (Supplementary Fig. 8b). These data indicate that the actin cytoskeleton-mediated facilitation of L-LTP depends on protein synthesis.
We then bilaterally-infused JPK or vehicle into CA1 of WT mice immediately after a weak Pavlovian fear conditioning training (a single pairing of a tone with a 1s, 0.7 mA foot-shock). This protocol generated only a relatively weak memory in vehicle-infused mice, as measured 24 hr after training (Fig. 5h). In contrast, in JPK-infused mice the same protocol induced a greatly enhanced contextual fear LTM (Fig. 5h). As expected, JPK had no effect on contextual fear STM (Fig. 5i) or hippocampus-independent auditory fear LTM (Supplementary Fig. 7b). Because JPK acts directly on actin itself by increasing its polymerization37, our data support the idea that actin polymerization is an essential mechanism for the consolidation of L-LTP and LTM.
We then reasoned that direct activation of mTORC2 signaling should convert short- to long-lasting memory processes (for both LTP and LTM). To test this hypothesis, we employed a small molecule (A-443654), which increases mTORC2-mediated phosphorylation of Akt at Ser473 (independently of mTORC138). We found that A-443654 promoted mTORC2 activity, PAK phosphorylation, as well as actin polymerization in WT slices (Fig. 6a-b), but not in mTORC2-deficient slices (Fig. 6d-e). More importantly, A-443654 converted a short-lasting E-LTP into a sustained L-LTP in WT slices (Fig. 6c), but it failed to do so in rictor fb-KO slices (Fig. 6f). Thus, the facilitated L-LTP induced by combining A-443654 and a single high frequency train is mediated by mTORC2.
If mTORC2 is involved in learning and memory, acute activation of mTORC2 should also promote LTM. Indeed, in WT mice, an intra-peritoneal (i.p.) injection of A-443654 increased the activity of both mTORC2 and PAK in the hippocampus (Supplementary Fig. 9). In addition, when WT mice were i.p-injected with either vehicle or A-443654 immediately after a weak Pavlovian fear conditioning training (a single pairing of a tone with a 1s, 0.7 mA foot-shock; Fig. 7a), we found that 24 hr later the A-443654-injected mice froze nearly twice as often as vehicle-injected controls, indicating that their contextual LTM was enhanced (Fig. 7b). By contrast, A-443654 failed to enhance contextual LTM in mTORC2-deficient mice (Fig. 7c), confirming the selectivity of A-443654. The facts that A-443654 was injected post-training and that contextual STM is not altered by A-443654 (Fig. 7d) argue against non-specific responses to fear. Taken together these pharmacogenetic data provide compelling evidence that A-443654’s enhancing effect on synaptic plasticity and behavioral learning is crucially dependent on mTORC2.
Changes in actin dynamics are required for long-lasting synaptic plasticity and memory consolidation8-10. Although changes in synaptic actin dynamics are thought to occur during learning, how the synaptic actin cytoskeleton controls memory storage remains poorly understood. Our findings reveal that the recently discovered mTOR complex 2 (mTORC2) bidirectionally controls the actin polymerization that is required for the conversion of a short-term synaptic process (E-LTP, STM) into a long-lasting one (L-LTP, LTM). Specifically, we found that genetic inhibition of mTORC2 activity blocks actin polymerization, actin regulatory signaling (Fig. 4) and selectively suppresses LTM and L-LTP (Fig. 1 and and2).2). Conversely, activation of mTORC2 by A-443654 promotes actin polymerization, actin signaling and enhanced L-LTP and LTM in WT mice (Fig. 6a-c and Fig. 7a-b) but not in mTORC2-deficient mice (Fig. 6d-f and Fig. 7c).
More importantly, restoring pharmacologically actin polymerization in mTORC2-deficient mice reverses the impaired L-LTP and LTM (Fig. 5b, 5d). We speculate that the stabilization of the actin cytoskeleton in mTORC2-deficient neurons leads to a morphological reorganization of the synapse which, in response to activity, could facilitate the trafficking and insertion of AMPA receptors clustered at the Postsynaptic Density (PSD). Consistent with this notion, structural plasticity was found to be altered in mTORC2-deficient hippocampal neurons (Fig. 4l). Alternatively, the restoration of actin dynamics in mTORC2-deficient synapses could induce a functional rather than a morphological change as AMPA receptors insertion in the hippocampus during LTP might occur independently of changes in spine shape39. Another interesting possibility is that changes in actin remodeling could regulate changes in gene expression at synapses which are required for L-LTP and LTM (see below).
Other lines of evidence further support the hypothesis that mTORC2 promotes long-term changes in synaptic strength by promoting actin polymerization. First, L-LTP induction is associated with an increase in the F-actin/G-actin ratio40-42 as well as with changes in synaptic morphology and actin signaling43. Second, inhibitors of actin polymerization block the late-phase of LTP, leaving the early phase of LTP intact44-46. Consistent with these data, only stimulation that induces a stable L-LTP reliably increases F-actin at spines45. Third, direct activation of actin polymerization by JPK converts E-LTP into L-LTP and enhances LTM (Fig. 5g-h). Fourth, the disruption of actin filaments in CA1 impairs the consolidation of contextual fear LTM47. Fifth, inhibition of actin polymerization and/or actin regulatory protein signaling in the lateral amygdala blocks auditory fear LTM but not STM48,49. Finally, mTORC2 is activated during learning (Fig. 2a-c), but only by protocols that induce late-LTP (Fig. 1d-f).
According to the prevailing view of memory consolidation, LTM is distinguished from STM by its dependence on protein synthesis1-7. Consequently, all the “molecular switches” identified so far are transcription or translation factors that regulate gene expression (from CREB50 to eIF2α51 to Npas452). However, like protein synthesis, mTORC2-mediated actin polymerization determines whether synaptic and memory processes remain transient or become consolidated in the brain. The evolutionary conservation of this new model, indicated by our comparable findings in Drosophila (Fig. 3), also suggests that our findings may be relevant to the study of memory consolidation in higher mammals, including humans.
Whether actin-mediated changes in synaptic strength depend on, or are perhaps triggered by, changes in gene or protein expression is not immediately clear. Nevertheless, a step towards clarifying the link between actin polymerization and protein synthesis during L-LTP is the finding that actin polymerization, triggered by LTP-inducing stimulation, induces the synthesis of the PKMzeta53, a kinase necessary for the maintenance of L-LTP54. In agreement with these data, we found that the JPK-facilitated L-LTP induced by one tetanic train was blocked by anisomycin (Supplementary Fig. 8a). These results support the idea that actin polymerization is up-stream of protein synthesis. However, we cannot rule out the possibility that protein synthesis and actin polymerization are parallel processes during L-LTP and LTM. Another intriguing possibility is that changes in actin polymerization could directly affect changes in gene expression. For example, actin polymerization promotes the shuttling of the myocardin-related transcription factor (MRTF) protein MKL to the nucleus where it interacts with the Serum Response factor (SRF), thus inducing activity-dependent gene expression in neurons55,56. Alternatively, incorporation of G-actin into F-actin filaments could alter local translation at synapses, by modulation the trafficking of ribosomes, translation initiation factors, RNA bindings proteins or even specifics mRNAs57. If so, the facilitated L-LTP induced by promoting actin polymerization should be insensitive to transcriptional inhibitors.
Our results suggest that neurons have evolved a remarkable bimodal strategy that allows them to control L-LTP and LTM storage both temporally (through regulation of protein synthesis) and structurally (through control of actin dynamics). In this respect, given that mTOR regulates two key processes of L-LTP and LTM - namely mTORC1-mediated protein synthesis3,5 and mTORC2-mediated actin cytoskeleton dynamics - we propose that mTOR is a key regulator of memory consolidation, controlling “distinct” aspects, the temporal through mTORC1 and the structural through mTORC2.
Dysregulation of mTORC1 and mTORC2 signaling appears to have a crucial role in memory disorders, such as the cognitive deficit associated with Autism Spectrum Disorder (ASD). Interestingly, the activity of mTORC2 is altered in the brain of ASD-patients harboring mutations in PTEN and/or TSC1/2 (two upstream negative regulators of mTORC1)58,59. In addition, in PTEN and TSC2 ASD-mouse models, prolonged rapamycin treatment in vivo, which indeed ameliorates the ASD-like phenotypes and restores mTORC1 activity, also corrects the abnormal mTORC2 activity23,60. Our new findings that mTORC2 plays a crucial role in memory consolidation raises the intriguing possibility that the neurological dysfunction in ASD is caused by dysregulation of mTORC2 rather than mTORC1 signaling.
In conclusion, our study not only helps to define key basic cellular and molecular mechanisms of physiological learning and memory but it points towards a new therapeutic approach to the treatment of human memory dysfunction in cognitive disorders or even aging, in which mTORC2 activity is known to be abnormally low.
Methods and supplemental figure legends are available in the online supplementary material of the paper.
We thank M. Magnuson, I. Dragatsis and K. Tolias who generously provided rictor/floxed mice, αCaMKII-Cre mice and the Tiam1’s constructs, respectively. We also thank A. Placzek and W. Sossin for comments on an early version of the manuscript. This work was supported by funds to M. C-M (NIMH 096816, NINDS 076708, Searle award grant 09-SSP-211 and Whitehall award).