|Home | About | Journals | Submit | Contact Us | Français|
There is a long-standing paradox that N-methyl-D-aspartate receptors (NMDARs) can both promote neuronal health and kill neurons. Recent studies show that NMDAR-induced responses depend on the receptor location: stimulation of synaptic NMDARs, acting primarily through nuclear Ca2+ signaling, leads to the build-up of a neuroprotective ‘shield’, whereas stimulation of extrasynaptic NMDARs promotes cell death. These differences result from the activation of distinct genomic programmes and opposing actions on intracellular signalling pathways. Perturbations in the balance between synaptic and extrasynaptic NMDAR activity contribute to neuronal dysfunction in acute ischaemia and Huntington’s disease and could be a common theme in the aetiology of neurodegenerative diseases. Neuroprotective therapies should aim to both enhance the effect of synaptic activity and disrupt extrasynaptic NMDAR-dependent death signalling.
NMDARs (N-methyl-D-aspartate (NMDA) receptors) are cation channels that are gated by the neurotransmitter glutamate. They are essential mediators of many forms of synaptic plasticity and also mediate aspects of development and synaptic transmission1, 2. In addition to their synaptic localization, NMDARs are found at extrasynaptic and perisynaptic sites, which can be defined immunohistochemically (e.g. distal from the post-synaptic density, PSD) and electrophysiologically (e.g. out of reach of normal synaptic glutamate release, see Box 1). In immature hippocampal neurons, extrasynaptic NMDARs can represent up to three quarters of all NMDARs3. Although the proportion of synaptically located NMDARs increases with development, a significant population of NMDARs remain extrasynaptic in adulthood3-6. The exact location of extrasynaptic NMDARs had not been thoroughly addressed until recently6. Only around one quarter of extrasynaptic NMDARs in adult hippocampal slices are classified as being perisynaptic (that is, within 100 nm of the PSD), the rest being localized on dendrites or on non-perisynaptic parts of the spine. Of the dendritically localized extrasynaptic NMDARs, around one third are adjacent to glia-like processes, and around half are adjacent to axons. The physiological function of extrasynaptic NMDARs is not fully understood, but their activation by glutamate spillover may contribute to long-term depression 7-9. This Review will focus on the emerging role of extrasynaptic NMDARs in pathological scenarios, in contrast with the demonstrated survival-promoting effects of synaptic NMDAR activity.
Synaptic N-methyl-D-aspartate receptors (NMDARs) are classically defined as functional receptors that are activated by glutamate released during low frequency synaptic events 116, 150-153. This could include low-frequency evoked events, spontaneous synaptic events and action potential-independent synaptic release of single quanta of glutamate. Extrasynaptic NMDARs — that is, receptors that are not activated during low-frequency synaptic events — can be found at various locations 6, including on the cell body, the dendritic shaft, the neck of the dendritic spine and also adjacent to the post-synaptic density (often referred to as perisynaptic) (see the figure, which shows examples of immuno-electron-micrograph (EM) images of extrasynaptic NMDARs on the dendritic shaft (A) and dendritic spine (B) in the hippocampal CA1 area of an adult rat ; the NMDARs are labelled with immunogold (A) or immunoperoxidase (B) using an NR1 antibody). Notably, in cultured neurons a proportion of NMDARs have the capacity to move into and out of synaptic sites 113, 151, 154, particularly GluN2B-containing NMDARs in young neurons 113 (although other studies have concluded that the mobility of GluN2B- and GluN2A-containing NMDARs is similar 151). However, such NMDAR mobility was not observed in acute slices, raising the question as to whether mobility is more restricted in vivo 116.
To study signalling associated with synaptic and extrasynaptic populations of NMDARs, the receptors must be isolated for electrophysiological characterization, and stimulation paradigms should be able to preferentially activate either pool. Preferential activation of synaptic NMDARs in cultured hippocampal or cortical neurons can be initiated via pharmacological removal of the GABAA receptor-mediated inhibition from the network. This induces bursts of firing activity and results in tetrodoxin (TTX)-sensitive NMDAR-dependent Ca2+ transients 50. A large portion of NMDARs are not activated by this stimulation paradigm and are presumed to be predominantly extrasynaptic NMDARs 38. Other techniques, such as the stimulation of afferents, are more suited to slice preparations.
The general principle applied in studies of extrasynaptic NMDARs and the resultant downstream signalling is to irreversibly block synaptic receptors before bath-applying the agonist (glutamate or NMDA), so that the agonist will activate only the extrasynaptic receptors. Synaptic NMDARs can be blocked by selectively inducing synaptic glutamate release under conditions that activate synaptic NMDARs, but in the presence of the open channel blocker MK-801 (blockade is essentially irreversible over the timescale of most experiments). Brief pulses of high K+ or episodes of firing activity, or simply allowing spontaneous action potential-dependent synaptic events can, in the presence of MK-801, induce open channel blockade of synaptic NMDARs 38, 43, 115, 155. A slightly more stringent approach involves only blocking those NMDARs that are exposed to glutamate upon release of single quanta of glutamate 43. The isolation of extrasynaptic NMDAR currents in hippocampal slice CA1 neurons can involve a similar strategy: MK-801 open-channel blockade of synaptic NMDARs activated by low-frequency stimulation of the stratum radiatum, followed by local uncaging of glutamate to activate extrasynaptic NMDARs 116. Note that strategies aimed at selectively blocking synaptic NMDARs require confirmation that little or no activation of extrasynaptic NMDARs takes place over the timeframe of synaptic stimulation – such activation could be caused by astrocytic glutamate release or elevated ambient glutamate due to nearby cellular damage.
We first discuss the background to the field of excitotoxicity, the dichotomous nature of NMDAR signalling, and the evidence that receptor location plays a role in this dichotomy. We then outline the molecular mechanisms that underlie synaptic NMDAR-dependent neuroprotection and the various pathways that are coupled preferentially to extrasynaptic NMDARs. After discussing the potential basis for differential signalling by synaptic versus extrasynaptic NMDARs we review recent studies that address the relevance of synaptic versus extrasynaptic NMDAR signalling in acute and chronic neurological disorders, focusing on stroke and Huntington’s disease. Finally we examine prospects for therapies designed to selectively target extrasynaptic NMDAR signalling while sparing synaptic signalling.
The neurotoxicity of sustained glutamate exposure to retinal neurons was first reported over 50 years ago10, at around the same time as the excitatory properties of glutamate were being characterized11. Later, Olney showed that glutamate toxicity was not restricted to retinal neurons, and coined the term “excitotoxicity” 12. Choi subsequently demonstrated that Ca2+ entry was a key mediator of glutamate excitotoxicity, and that the NMDAR was the primary source of toxic Ca2+ influx 13-15. Shortly after, it was proposed that Ca2+ entry through NMDARs was particularly effective at killing neurons compared to entry through other channels 16. At the same time, evidence was accumulating that implicated glutamate toxicity and NMDAR activity in acute neurological trauma such as stroke, as well as circumstantial evidence pointing to a role for glutamate toxicity and NMDAR activity in chronic neurodegenerative diseases, including Huntington’s disease 17-19. Since then, the work of many laboratories have advanced our understanding of how glutamate dyshomeostasis, ionic imbalance and abnormal NMDAR activity can contribute to such disorders 19-24.
The destructive effects of NMDAR activity are in striking contrast to the observation that the survival of several types of neuron is dependent on synaptic NMDAR activity and function25-27. Elimination of NMDAR activity in vivo causes widespread apoptosis and enhances trauma-induced injury in developing neurons28-32. In the adult CNS, NMDAR blockade exacerbates neuron loss when applied after traumatic brain injury (TBI) or during ongoing neurodegeneration33, and impairs the survival of newborn neurons in the adult dentate gyrus34. Thus, NMDARs are important for cell survival but can also be harmful and kill neurons. For many years it was thought that the degree of Ca2+ entry though NMDARs was solely responsible for these differences in cellular outcome: moderate levels of NMDAR activity were considered to be beneficial for neurons, whereas excessive activation of NMDARs, with subsequent ‘Ca2+ overload’ of the neurons, was deleterious. According to this model, neuronal responses to NMDAR activity followed a bell-shaped curve, where both too much and too little was potentially harmful35-37.
This view may be subject to revision in light of recent work that has revealed an alternative explanation for the ‘NMDAR paradox’, namely that it is the location of the NMDAR that also influences whether it is coupled to pro-death or pro-survival signals. According to this new model, synaptic NMDARs are neuroprotective, whereas extrasynaptic NMDARs preferentially initiate cell death pathways. Thus, the fate of neurons is not determined solely by the degree of overall NMDAR activity but by the extent to which synaptic and extrasynaptic NMDARs are activated. This may be best described with an ‘χ-shaped’ graph in which an ascending curve that represents increased neuroprotection due to increased synaptic NMDAR activity is superimposed on a descending curve that illustrates the progressive decrease in neuroprotection due to increasing extrasynaptic NMDAR activity (Figure 1). According to this concept, it is not the ‘Ca2+-overload’ per se that is the sole determinant of toxicity but rather the Ca2+ flow though NMDARs located outside the synapse that is particularly harmful to neurons. Synaptic NMDAR activity is by nature phasic and extrasynaptic activation is achieved by chronic agonism, but when equivalent overall Ca2+ concentrations are triggered by synaptic and extrasynaptic NMDARs, they cause strikingly different downstream events38. Ca2+ influx evoked by intense synaptic NMDAR activation is not only well tolerated by hippocampal neurons (i.e. it causes no perturbations to mitochondrial membrane potential), it is also the trigger for genomic processes that render neurons more resistant to apoptotic and oxidative insults38 (see below). In sharp contrast, comparable Ca2+ transients induced by activation of extrasynaptic NMDARs — either on their own or in the presence of synaptic NMDAR activation — trigger mitochondrial dysfunction and cell death38.
The distinct synaptic versus extrasynaptic NMDAR signalling pathways that lead to survival and death, respectively, were initially characterized in hippocampal neurons, and subsequently were confirmed to exist in cortical neurons39. Recent studies have shown that a selective activation of extrasynaptic NMDARs in hippocampal neurons triggers the same amount of cell death as activation of all NMDARs (synaptic and extrasynaptic) despite the fact that extrasynaptic NMDARs elicit a smaller initial Ca2+ elevation40, highlighting the importance of extrasynaptic NMDARs in excitotoxicity. Moreover, several studies have demonstrated that certain effects of extrasynaptic NMDAR signalling dominate over effects of synaptic signalling38, 41-43. However, under chronic conditions of NMDAR activation, the balance between synaptic and extrasynaptic NMDAR activity is important 44 and can influence the point at which escalating levels of bath-applied NMDA or glutamate become neurotoxic. For example, bath application of low concentrations of NMDA can induce firing activity and so preferentially activate synaptic pathways and activity-dependent neuroprotection, whereas higher levels of NMDA suppress firing, resulting in extrasynaptic pathways dominating 45.
According to the theory of neuronal health46, neurons can exist in a spectrum of states ranging from fully functioning and resilient at one extreme to dysfunctional and vulnerable to insults at the other extreme. The position of the neuron within this spectrum is influenced by a multitude of signals and stimuli, but is known to be shifted towards health and robustness by synaptic NMDAR activity. An episode of synaptic NMDAR activity promotes neuroprotection that lasts beyond the duration of the episode and continues after most signalling pathways are no longer active47. This long-lasting form of ‘acquired neuroprotection’ predominantly results from changes in gene expression that have effects at multiple levels within the cell (Figure 2) such as enhancement of mitochondrial health, boosting of antioxidant defences and suppression of caspase activation, all of which in turn serve to preserve neuronal function and improve viability in the face of insults.
Ca2+ signalling is a major mediator of the dialogue between the synapse and cell nucleus. It has been known for over 20 years that seizure activity triggers transcriptional responses in neurons 48, as do physiological patterns of synaptic NMDAR activity, such as those that promote synaptic potentiation 49. Ca2+ entry through synaptic NMDARs is augmented by release from internal stores, after which Ca2+ is transduced to the cell soma where it invades the nucleus 50. Nuclear Ca2+ is one of the most potent activators of neuronal gene expression. Transcriptome analyses have revealed that in hippocampal neurons nearly 200 genes are controlled by nuclear Ca2+ signalling42, 51.
One important target of nuclear Ca2+ is the nuclear Ca2+/calmodulin-dependent protein (CaM) kinase IV and the transcription factor cyclic-AMP response element binding protein (CREB, Figure 3a–pathways A-D)52-58. CREB is the prototypical signal-regulated transcription factor, whose role in neuronal survival, as well as in several other processes (including synaptic plasticity, addiction, neurogenesis, and learning and memory), is well documented59-62,63. In hippocampal neurons, CREB-dependent gene expression is causally linked to the long-lasting phase of activity-dependent neuroprotection against apoptotic and excitotoxic insults47, 64. This form of acquired neuroprotection is dependent on nuclear Ca2+ signalling47, which is consistent with the known role for nuclear Ca2+ in CREB activation50, 57, 65. In recent years, the genomic programmes that underlie acquired neuroprotection have been uncovered42, 51. Among the pool of nuclear Ca2+-regulated genes, a core set of ten genes, termed Activity-regulated inhibitors of death (AID) genes have been identified, which have been shown to provide neurons with a broad-spectrum neuroprotective ‘shield’, both in cell culture and in animal models of neurodegeneration51. Some of the AID genes, which include the genes encoding activating transcription factor 3 (Atf3), B-cell translocation gene 2 (Btg2), B-cell lymphoma 6 (Bcl6), growth arrest and DNA damage induced gene 45 beta (GADD45beta), growth arrest and DNA damage induced gene 45 gamma (GADD45gamma), Inhibin beta-A (Inhba), Interferon activated gene 202B (Ifi202B), neuronal PAS domain protein 4 (Npas4), nerve growth factor IB (also known as nuclear receptor subfamily 4, group A, member 1) (Nr4a1), and serine protease inhibitor B2 (Serpinb2), seem to be potential CREB target genes and may provide neuroprotection through a common process that renders mitochondria more resistant to cellular stress and toxic insults (Figure 4a)42, 51, 66, 67. Another target of synaptic NMDAR and nuclear Ca2+-CREB signalling is the gene that encodes the neurotrophin brain-derived neurotrophic factor (Bdnf)38, 51, 68, 69. BDNF has neuroprotective properties70 and can rescue neurons from NMDAR blockade-induced neuronal death71.
Although CREB is a key mediator of nuclear Ca2+-dependent neuroprotection, other transcription factors implicated in mediating NMDAR-dependent survival may also be partly controlled by nuclear Ca2+, such as nuclear factor of activated T-cells (NFAT72). By contrast, NMDAR-dependent activation of other neuroprotective factors such as nuclear factor I, subtype A (NFI-A) might occur independently of nuclear Ca2+ elevation (its activation being dependent on the extracellular signal-regulated kinase 1/2 (ERK1/2) pathway and nitric oxide (NO) synthesis73).
A second mechanism through which synaptic NMDAR activity restricts the apoptotic potential of a neuron involves the transcriptional suppression of core components of the intrinsic apoptosis cascade (Figure 4a)66, 67. Synaptic NMDAR activity suppresses the expression of the pro-apoptotic Bcl-2 homology domain3 (BH3)-only member gene Puma in vitro and in vivo66, 67. Puma expression is both sufficient to induce cell death and necessary for apoptotic death in response to NMDAR blockade67. As with activation of CREB, suppression of Puma is specific to synaptic NMDAR signalling, as bath activation of all NMDARs (synaptic and extrasynaptic) fails to influence Puma expression. Suppression of Puma inhibits the apoptotic cascade upstream of cytochrome c release, but auxiliary protective mechanisms also exist downstream, as the genes encoding the core apoptosome components apoptotic protease activating factor 1 (Apaf-1) and pro-caspase-9 are also subject to transcriptional suppression by synaptic NMDAR activity66, 67. The net effect of all these transcriptional changes is that the activation of caspase-9 and downstream executioner caspases is very limited, apoptosis is inhibited, and the neurons remain viable and electrically active 67.
Synaptic NMDAR activity also suppresses the expression and/or activity of important pro-death transcription factors (Figure 3b, Figure 4a). The forkhead box O (FOXO) class of transcription factors can promote neuronal death following excitotoxic injury, trophic factor withdrawal and oxidative stress74-77. Known pro-death FOXO target genes include the BH3-only genes Noxa, Bcl2-like protein 11 (Bcl2l11, also known as Bim) and Puma, Fas ligand (Fasl) and thioredoxin-interacting protein (Txnip). In conditions of trophic deprivation, in which FOXOs are constitutively located in the nucleus, synaptic NMDAR activity promotes sustained activation of the Akt pathway, leading to the phosphorylation and nuclear export of FOXO and the subsequent inactivation of FOXO target genes, which includes Foxo1 (Foxo1 itself was recently identified as a FOXO target gene43, 78, Figure 3b–pathway A). Furthermore, synaptic NMDAR activity induces a long-lasting inhibition of FOXO nuclear import77 (Figure 3b–pathway D). This long-lasting blockade was shown to require nuclear CaM kinase IV activation and de novo gene expression, although the exact mechanism by which this leads to blockade of FOXO nuclear import remains unclear. Thus, synaptic NMDAR activity both promotes the nuclear export of FOXOs and the transcriptional suppression of Foxo1 and places a long-lasting blockade on FOXO nuclear translocation (Figure 3b). p53 is another pro-apoptotic transcription factor that is subject to negative transcriptional regulation by synaptic NMDAR activity and that may contribute to the suppression of Apaf-166, although p53-independent routes to Apaf-1 expression also exist67.
In addition to apoptosis-suppressing effects, synaptic NMDAR activity also enhances antioxidant defences43 (Figure 2, Figure 4b), which contributes to neuroprotection against oxidative insults. NMDAR blockade in vivo triggers neuronal death that is associated with protein carbonylation (a marker of oxidative stress), which is suggestive of a role for physiological NMDAR activity in supporting antioxidant defences. Synaptic NMDAR activity also controls neuronal vulnerability to oxidative stress-induced death in vitro43. It triggers a number of changes to the thioredoxin-peroxiredoxin system that contribute to this neuroprotective effect. Specifically, synaptic activity enhances thioredoxin activity and facilitates the reduction of hyperoxidized peroxiredoxins, which are an important class of antioxidant enzymes. These effects are mediated by a coordinated programme of gene expression changes, some of which have been characterized43. These changes include the transcriptional suppression of the FOXO target gene Txnip, which is dynamically regulated by NMDAR activity in vitro and in vivo. Also, enhanced reduction of hyperoxidized peroxiredoxins is associated with transcriptional activation of two genes, Srxn1 (sulfiredoxin) and Sesn2 (sestrin 2). The products of these genes can mediate this reaction79-81, with sulfiredoxin induction likely to be more important82, 83. Trans-synaptic stimulation of synaptic NMDARs is crucial for boosting antioxidant defences, as chronic bath activation of all (synaptic and extrasynaptic) NMDARs failed to trigger the aforementioned transcriptional changes or protect against oxidative insults.
The gene expression changes (Figure 4b) do not account for all the antioxidant effects of synaptic NMDAR activity, raising the possibility that other antioxidant systems, such as the glutathione system, may also be subject to control by synaptic NMDAR activity.
NMDAR-dependent neuron death is mediated by a variety of pathways, the relative contribution of which may depend on the cell type, the developmental stage or intensity of the excitotoxic insult20, 84. Recent studies have found that certain pro-death pathways or events are preferentially activated by extrasynaptic NMDARs compared to synaptic ones (Figure 5). As discussed below, in some cases this involves direct antagonism of synaptic NMDAR-activated survival pathways, whereas in others it involves the triggering of events/pathways that are simply not activated by synaptic NMDARs.
Bath activation of NMDARs has long been known to be a poor activator of CRE-dependent gene expression in hippocampal neurons, compared to Ca2+ entry though voltage-gated Ca2+ channels (although activation of the serum response element (SRE) was equally good)85. This could be explained by the fact that bath activation of NMDARs induces CREB-activating phosphorylation of CREB at serine-133, followed shortly afterwards by a strong CREB-dephosphorylation signal 86, 87. By contrast, activation of voltage-gated Ca2+ channels induces sustained CREB phosphorylation 86. The fact that NMDARs seemed to be poor inducers of CRE-dependent gene expression when they were bath-activated was at apparent odds with the fact that NMDARs were known to induce activation of CRE-dependent gene expression following electrical activity 49, 88. These historical observations can be explained by the observation that synaptic NMDAR activity promotes sustained CREB phosphorylation, whereas extrasynaptic NMDARs are coupled to a dominant CREB dephosphorylating signalling pathway 38, 89 (Figure 3a). When all (synaptic and extrasynaptic) NMDARs are activated by bath-applied glutamate or NMDA, the extrasynaptic pathway dominates, resulting in weak CREB activation 38, 85, 86. The only exception is when low levels of bath-applied NMDA induce firing activity, in which case synaptic signalling dominates and CREB activation is relatively strong 45.
Recently, insight into the mechanistic basis of this differential signalling was uncovered and found to be centred on Jacob, a binding partner of the neuronal Ca2+ binding protein caldendrin90. Jacob, when localized to the nucleus, causes CREB dephosphorylation and also promotes a loss of synaptic contacts. Jacob contains a nuclear localization signal and its nuclear import requires importin-α binding. This binding and nuclear import is prevented by caldendrin, which binds Jacob in a Ca2+-dependent manner. Synaptic NMDAR activity was shown to promote caldendrin-dependent retention of Jacob outside of the nucleus, whereas extrasynaptic NMDAR activity promoted the nuclear accumulation of Jacob, with subsequent effects on neuronal survival 90 (Figure 3a–pathway F). As the coupling of extrasynaptic NMDA receptors to a CREB shut-off pathway is developmentally regulated and absent in very young neurons 89, it will be of interest to know whether there is a developmental switch that activates the Jacob pathway.
The ERK1/2 pathway, which can support activation of CREB and inactivate the pro-death protein Bcl-2-associated death promoter (BAD) has been implicated in NMDAR-dependent neuroprotection,26. Synaptic NMDAR activity promotes sustained ERK activity47, 91, whereas activation of all NMDARs by bath application of NMDA results in ERK activation that is followed by inactivation91-93. The basis for this bi-directional control of ERKs turned out to be due to the opposing actions of synaptic versus extrasynaptic NMDARs41. Synaptic NMDARs couple to activation, whereas extrasynaptic NMDARs promote ERK dephosphorylation and subsequent inactivation in both hippocampal41 and cortical neurons39 (Figure 3a–pathway E). Consistent with this, extrasynaptic NMDARs strongly inactivate Ras (upstream of ERK) 93. Thus, synaptic and extrasynaptic NMDARs are mutually antagonistic with regard to Ras-ERK signalling. However, ERK1/2 inactivation by extrasynaptic NMDAR is stimulus-specific: although extrasynaptic NMDARs exert an ERK1/2 shut-off signal that is dominant over that triggered by activation of synaptic NMDARs, it does not antagonize ERK1/2 activity induced by BDNF94.
Synaptic NMDAR activity inactivates pro-death FOXOs at multiple levels (see above and Figure 3b), whereas extrasynaptic NMDAR activity does the opposite. Chronic activation of all (synaptic and extrasynaptic NMDARs) does not result in FOXO export from the nucleus, because activation of Akt is not sustained43 (Sustained activation of the Akt pathway leads to the phosphorylation and nuclear export of FOXO). More recently, it was shown that extrasynaptic NMDAR-evoked signals not only fail to induce FOXO nuclear export, but instead promote the nuclear import of FOXOs that are normally sequestered in the cytoplasm77 (Figure 3b–pathway C). This activation of FOXOs subsequently contributes to NMDAR-dependent neuronal death. Thus, the activation status of FOXOs is highly sensitive to the balance between synaptic and extrasynaptic NMDAR activity (Figure 3b).
Activation of extrasynaptic NMDARs strongly activate calpains, whereas synaptic NMDAR activity does not95. In agreement with this, the classical calpain substrate fodrin and Na+/Ca2+ exchanger 3 (NCX3)95 are cleaved following extrasynaptic, but not synaptic, NMDAR activity. NCX3 is known to be cleaved by calpains in animal models of stroke, leading to delayed Ca2+ deregulation and neuronal death96. Another substrate for extrasynaptic NMDAR-activated calpains is striatal enriched tyrosine phosphatase (STEP), which is cleaved into a form that cannot interact with its normal substrates. One such substrate is p38 mitogen-activated protein (MAP) kinase, which is negatively regulated by STEP, is enriched at extrasynaptic membrane fractions, and is known to contribute to death induced by chronic NMDA or glutamate exposure in cerebellar granule cells and cortical neurons95, 97-99. Inhibition of calpain-mediated STEP cleavage is sufficient to inhibit NMDAR-dependent p38 activation and can protect neurons from glutamate toxicity and oxygen-glucose deprivation95. Thus calpain — coupled to STEP cleavage and subsequent p38 activation — represents another pro-death cascade that selectively activated by extrasynaptic NMDARs (Figure 5). One important caveat to the role of p38 in excitotoxicity is that its influence on survival and death is heavily context-dependent. For example, it can promote neuroprotection through activation of myocyte enhancer factor 2 (MEF2) 100 and indeed can be activated by neuroprotective, synaptic NMDAR activity 99, 101. Thus, its contribution to neuronal death in excitotoxicity may rely on interactions with other events that are triggered by the excitotoxic insult.
The observation that p38 is disinhibited by extrasynaptic NMDAR activity through STEP cleavage also raises the question as to why extrasynaptic NMDARs fail to activate ERK1/2, as STEP is also a powerful negative regulator of ERKs95, 102. It is possible that other, dominant ERK-inactivating signals may be triggered in parallel (such as those discussed above), overriding any effects of STEP cleavage.
As a result of these or other differences in signalling, extrasynaptic NMDAR activity induces a very different programme of gene expression compared to synaptic NMDARs42. As new gene expression contributes to at least some forms of excitotoxic neuronal death, these changes are likely to be functionally important. Although the identity of such genes remains to be elucidated, they may include FOXO target genes or other factors targeted by extrasynaptic signals. For example, Clca1, the gene that encodes Ca2+ -activated chloride channel regulator 1, is selectively activated by extrasynaptic NMDAR activity in cell culture and in a stroke model, and its overexpression is sufficient to kill neurons42, 103, although the mechanism by which its expression is induced is unclear.
In addition to differentially controlling specific transcription-regulating events, synaptic and extrasynaptic NMDARs may also more globally affect signal processing towards and within the cell nucleus by regulating nuclear architecture. A structural analysis of hippocampal nuclei using a novel algorithm for three-dimensional reconstruction uncovered that - in contrast to what is commonly thought and what is depicted in virtually all textbooks - many cell nuclei are not round or near-spherical organelles but have a very complex, highly infolded shape 104. The structure of the nucleus is dynamic and signal-regulated, with synaptic and extrasynaptic NMDARs having opposing effects: stimulation of synaptic NMDARs increases the number of infolded nuclei, whereas the infolded shape is rapidly lost following Ca2+ flux through extrasynaptic NMDARs. Thus, neurons translate NMDAR signals into changes in nuclear geometry, providing a means for an activity-dependent modulation of nucleo-cytoplasmic exchange of molecules and ions 104. The changes in nuclear shape and consequent rearrangements of the nuclear architecture and spatial re-organization of chromosome territories may affect chromatin structures and accessibility and could have profound effects on gene expression and genome stability. In addition, infolded and near-spherical nuclei have similar volumes but infolded nuclei have a larger surface and larger number of nuclear pore complexes 104. Given that actively transcribed DNA segments may be located near nuclear pore complexes 105, 106, a possible increase in the number of nuclear pore complexes in infolded nuclei following synaptic NMDAR activation may enhance the capacity of the nucleus for transcription-coupled anchoring of promoter regions to nuclear pore complexes. Synaptic NMDAR-induced structural plasticity of the nucleus may therefore represent an adaptation to a metabolically more active state with an increased demand for gene expression.
The differences in signalling between synaptic and extrasynaptic NMDARs could theoretically be due to one of three factors, or a combination thereof. Firstly, they could be due to differences in the make-up of the NMDAR signalling complexes at extrasynaptic compared to synaptic locations. The components that make up the NMDAR signalling complex in the post-synaptic density are the subject of intense study107, but the molecules that associate with extrasynaptic NMDARs have been poorly characterized. Some proteins that are found in synapses and that are linked to synaptic NMDARs, such as the membrane-associated guanylate kinases (MAGUKs) postsynaptic density protein (PSD)-95 and PSD-93, discs, large homolog 1 (Drosophila) (DLG1, also known as SAP97, and discs, large homolog 3 (Drosophila) (DLG3, also known as SAP102), are also linked to extrasynaptic NMDARs. In 3-week old hippocampal cultures, around 20% of extrasynaptic NMDARs were found to be associated with PSD-95 and a further 20% with SAP1026. Association of extrasynaptic GluN1 NMDAR subunits with MAGUKs was also observed by double-labelling immuno-gold electron microscopy in adult hippocampal slices 6. This is of particular interest since MAGUKs have been implicated in coupling the NMDAR to downstream death pathways such as excessive nitric oxide production 108, 109. Other studies suggest the possibility of a specific association of extrasynaptic NMDARs with certain proteins. For example, it has been postulated that, as caldendrin is localised to the post-synaptic density, only trans-synaptic activation of synaptic NMDARs trigger local Ca2+ transients that are high enough to promote caldendrin-mediated Jacob retention90. Alternatively, the preferential localization of p38 (but not ERK1/2) at extrasynaptic membranes may explain its preferential activation by extrasynaptic NMDAR-dependent STEP cleavage95. It is likely that other, yet to be identified proteins associate differentially with synaptic/extrasynaptic NMDARs.
Secondly, receptor subunit composition could play a role in the differences in signalling between synaptic and extrasynaptic NMDARs. The cytoplasmic C-termini of the GluN2 subunits have undergone considerable sequence divergence through evolution, which potentially allows for differential association of signalling molecules. During development GluN2B is the predominant subunit at all synaptic and extrasynaptic locations in forebrain neurons, and GluN2A expression in the rodent CNS begins at the end of the first post-natal week110, 111. In the adult forebrain both the GluN2A and GluN2B subunits are prevalent. GluN2A becomes incorporated into synaptic NMDARs 112 and has been reported to become enriched at synapses compared to extrasynaptic locations, with GluN2B more abundant at extrasynaptic locations113, 114. No studies, however, have demonstrated a clear-cut spatial demarcation. Indeed, GluN2A subunits have been identified at extrasynaptic sites in cultured neurons115 and the subunit compositions of synaptic and extrasynaptic NMDARs in 3-week-old acute hippocampal slices have been found to be very similar116. Moreover, immunofluorescence and immuno-gold studies have also identified both GluN2A and GluN2B at extrasynaptic locations6.
Nevertheless, subunit differences could have a contributing effect if GluN2A- and GluN2B-containing NMDARs are coupled differently to pro-survival and pro-death pathways. Indeed, GluN2B or GluN2A-containing NMDARs have been proposed to promote neuronal death and neuronal survival, respectively117. However, the specificity of the GluN2A-specific antagonist used in this study117 has been questioned, particularly when employed in vivo or in non-steady state conditions118-122. Moreover, it has been shown that GluN2B- and GluN2A-containing NMDARs can both promote survival and death43, 114, 123. Whether they do so with differing efficiencies, however, remains unclear as a direct comparison has yet to be made. Indeed, attempting to do so would be challenging given the lack of a sufficiently selective GluN2A-specific antagonist. Also, the fact that GluN2B and GluN2A are expressed at different relative levels in development can cloud the issue, as developmental changes in NMDAR signalling can be misinterpreted as being due to a switch in GluN2 subunit when there are in fact many other potential explanations. Another important issue is that receptors can exist in a tri-heteromeric form that contains both a GluN2A and a GluN2B subunit 124, meaning that 2A- and 2B-dependent signals can theoretically be transduced from the same channel. Whether the GluN2 subunit (and particularly its C-terminus) influences whether an activated NMDAR will promote either survival or death will require more sophisticated approaches in order to be fully resolved.
A final potential factor contributing to differential signalling could be the way in which synaptic and extrasynaptic NMDAR pools are activated. Synaptic NMDARs are generally transiently and intensely activated by the trans-synaptic release of glutamate, while extrasynaptic NMDARs are typically activated chronically, by elevated levels of ambient glutamate. Subtle differences in the spatial properties of intracellular Ca2+ transients evoked by these stimuli could differentially affect signalling, even if the overall Ca2+ load were similar. If trans-synaptic versus chronic activation of NMDARs is a contributing factor in differential synaptic vs. extrasynaptic NMDAR signalling, this raises the question as to whether synaptic NMDARs can contribute to excitotoxic neuronal death under certain circumstances (e.g. chronic activation). Against this idea is the observation that prior blockade of synaptic NMDARs was not protective against NMDA-induced death 40. Nevertheless, it is a possibility that under certain conditions where extrasynaptic NMDARs are chronically active, that chronic activation of synaptic NMDARs may contribute to a toxic Ca2+ load. Conversely, it is unclear whether extrasynaptic NMDARs can contribute to neuroprotective signalling under certain circumstances. For example, if strong synaptic activity results in the transient trans-synaptic activation of a portion of perisynaptic or extrasynaptic receptors (as has been shown 125), does this contribute to neuroprotective signalling or antagonize those promoted by synaptic NMDAR activation? It is also not clear what proportion of a cell’s extrasynaptic NMDAR population need to be activated in order to promote pro-death signalling.
Another, related question is whether extrasynaptic NMDARs at different locations (dendritic, perisynaptic, cell body) signal differently. Studies have shown NMDARs demonstrating mobility between synaptic and extrasynaptic locations 113, 151, 154 (see Box 1), and it remains to be determined whether these receptors alter their signalling properties upon switching their location. One possibility is that an NMDAR moving (for example) between an extrasynaptic to a synaptic site, encounters a different biochemical environment and so associates with a different complement of downstream signalling molecules. However, the extent to which NMDARs are mobile in vivo remains to be clarified in the light of recent observations that synaptic and extrasynaptic NMDARs from stable pools in acute slices 116.
Addressing the above questions will enhance our knowledge about the molecular basis for differential synaptic vs. extrasynaptic NMDAR signalling, and enhance our understanding of the precise conditions that determine the balance between protective and destructive NMDAR signalling.
Inappropriate levels of NMDAR activity can contribute to neuronal death or dysfunction in a number of paradigms, including acute excitotoxic insults (e.g. stroke, TBI) and certain neurodegenerative diseases, such as HD and AD (reviewed in23, 24, 126). Recent work has specifically implicated extrasynaptic NMDAR activation in the aetio-pathophysiology associated with an acute and a chronic form of neuronal dysfunction (ischaemic stroke and HD, respectively).
During an ischaemic episode, extracellular glutamate accumulates due to synaptic release and impairment or reversal of uptake mechanisms22, 127, and neurons become depolarized, relieving the Mg2+ blockade of NMDARs. These events have the combined effect of causing strong activation of NMDARs128, possibly including those located outside the synapse. The consequences of reversed uptake can be studied in culture through the use of the transportable glutamate transport inhibitor L-trans-pyrrolidine-2,4-dicarboxylate (PDC), which causes reversed glutamate uptake through hetero-exchange129. In neuron/astrocyte co-cultures, PDC treatment causes glutamate efflux and excitotoxicity due to an activation of extrasynaptic NMDARs that mainly arises from reversed astrocytic uptake130. Interestingly, a significant proportion of extrasynaptic NMDARs in vivo are located adjacent to glia6, raising the possibility that astrocytic glutamate release may lead to activation of these receptors.
A recent study has demonstrated that extrasynaptic NMDAR activity may be selectively enhanced in ischaemia via an intracellular signalling cascade131. Ischaemia recruits death-associated protein kinase (DAPK1) directly to the C-terminal tail of GluN2B, leading to phosphorylation of the channel and increased single-channel conductance. The effect is specific to GluN2B-containing NMDARs and results in a selective enhancement of extrasynaptic NMDAR currents131, which is not observed in DAPK1-deficient neurons. Moreover, peptide-mediated disruption of DAPK1–NMDAR interactions was shown to protect neurons against oxygen-glucose deprivation in vitro and ischaemic damage in vivo131. Taken together, these studies indicate that extrasynaptic NMDAR activity contributes to ischaemic neuron loss.
Recent studies show that the relationship between extrasynaptic NMDAR signalling and mutant Huntingtin (mtHtt) may be a key factor in HD progression44, 132 (Figure 6a). One study focused on the presymptomatic YAC128 transgenic mouse, which expresses a mutant huntingtin (mtHtt) that contains 128 CAG repeats, as compared to a YAC18 control mouse which expresses Htt with a non-pathological CAG repeat length132. Although expression of mtHtt had no effect on synaptic NMDAR activity, YAC128 transgenic mice showed both an increase in extrasynaptic NMDAR current and in extrasynaptic NMDAR expression specifically in striatal medium spiny neurons. This enhanced extrasynaptic NMDAR activity was dominated by GluN2B-containing NMDARs and persisted in aged YAC128 mice132. Consistent with the previously mentioned coupling of extrasynaptic NMDAR activity to CREB dephosphorylation and inactivation 38, reduced CREB phosphorylation was observed in the striatum of YAC128 mice132.
In addition to the influence of mtHtt on extrasynaptic NMDAR function, a recent study has demonstrated that synaptic and extrasynaptic NMDAR signalling have distinct and opposing effects on the neurotoxicity of mtHtt itself. Synaptic NMDAR activity reduces mtHtt toxicity by increasing the formation of non-toxic mtHtt inclusions by a process involving the transcriptional up-regulation of the chaperonin TRiC (T complex-1 ring complex) subunit TCP1. TCP1 knock-down or blocking of synaptic activity reduced the formation of inclusions, rendering neurons vulnerable to mtHtt-induced neuronal death. This mtHtt-induced neuronal death is itself dependent on extrasynaptic NMDAR activity44, which is consistent with the upregulation of extrasynaptic NMDAR currents that have been observed in YAC128 mice132 and, indeed, with the large body of literature implicating excessive NMDAR-dependent activity in HD pathology23.
In addition to the enhancement of extrasynaptic NMDAR activity by mtHtt and in mediating mtHtt-induced excitotoxic death, extrasynaptic NMDARs may also serve as up-stream regulators of mtHtt toxicity44. Extrasynaptic NMDAR activity was found to support the expression of Rhes, a small GTPase that controls the sumoylation of mtHtt, a modification that promotes mtHtt toxicity and prevents its aggregation133.
Part of the toxic effects of mtHtt and excessive extrasynaptic NMDAR activity in HD pathology may be attributable to the suppression of pro-survival CREB target gene expression. As stated above, extrasynaptic NMDAR activity is coupled to a CREB-inactivating dephosphorylation signal38. Consistent with this, CREB phosphorylation levels are reduced in the striatum of YAC128 mice (where extrasynaptic NMDAR activity is elevated)132. Also, mtHtt can suppress CREB-mediated gene expression by sequestering the CREB coactivator CBP — an effect that contributes to mtHtt toxicity134, 135. A key neuroprotective CREB target gene whose underexpression is implicated in HD pathogenesis is the transcriptional coactivator peroxisome proliferator-activated receptor-γ coactivator 1a (PGC-1α)136. In neurons, PGC-1α regulates both mitochondrial density and antioxidant defences, and controls vulnerability to excitotoxic insults by these or other mechanisms101, 137, 138. MtHtt interferes with CREB-dependent PGC-1α expression, and this leads to mitochondrial dysfunction and metabolic defects139, 140. Indeed, PGC-1α expression is lower in the striatum of human HD patients, but not in the cerebellum or hippocampus, which is consistent with the brain areas associated with HD pathology. As a further exacerbating factor, the striatum seems to be particularly vulnerable to loss of PGC-1α expression141. Overexpression of PGC-1α protects both cortical and striatal neurons from mtHtt- and 3-nitropropanoic acid-induced cell death, and lentiviral transduction of PGC-1α into the striatum reduces striatal loss in a mouse model of HD44, 138-140.
Synaptic activity acts as a positive regulator of PGC-1α expression, both directly — via CREB activation — and indirectly — by inhibiting mtHtt-dependent suppression of PGC-1α (due to the aforementioned TCP1-dependent aggregation of mtHtt)44, 101. Moreover, synaptic activity enhances PGC-1α’s transactivating capacity through a post-translational, p38-dependent mechanism101. Taken together, the balance between synaptic and extrasynaptic NMDAR activity controls the toxicity of mtHtt and is itself controlled by mtHtt (Figure 6a). This reciprocal, feed-forward control may play a key role in HD progression and indicates that a selective blockade of extrasynaptic NMDARs and sparing of synaptic NMDAR activity offers a potentially attractive therapeutic strategy, as outlined below44, 132.
Despite evidence from animal studies that have implicated NMDARs in ischaemic brain damage, clinical trials of NMDAR antagonists for stroke and TBI have failed due to poor tolerance and efficacy25, 142. The important role of NMDARs in the CNS may mean that for many antagonists the maximum tolerated dose may not be therapeutically effective, as many unacceptable side-effects are not off-target effects but are due to blockade of synaptic NMDAR activity. Evidence that NMDAR activity can exert a neuroprotective effect has led to suggestions that it may even promote recovery in the post-reperfusion phase, and prevent delayed neuronal loss in the penumbra25, 143. Dichotomous signalling by NMDARs is also evident in juvenile TBI. Here, treatment with NMDAR antagonists reduces primary excitotoxic death but exacerbate secondary apoptosis, resulting in increased overall death30. In adult TBI, the NMDAR has also been proposed to rapidly switch between ‘destructive’ and ‘recovery’ roles144, 145. As such, antagonists may be protective in early stages of the insult progression but not later, highlighting the need for more specific anti-excitotoxic strategies in which the physiological and neuroprotective roles of NMDARs are considered. One promising strategy is the use of antagonists that preferentially block NMDARs in excitotoxic scenarios and of pathologically-activated therapeutics, which are activated only in the pathological state146. As an example of the former, memantine is a clinically well-tolerated NMDAR antagonist with pharmacological properties that are suited to the blockade of chronic NMDAR activity in pathological situations without interfering with normal synaptic function146.
Memantine is an open channel blocker with a fast off-rate. Its uncompetitive nature results in an effective blockade of chronic extrasynaptic NMDAR activity that is triggered by elevated levels of glutamate126. Interestingly, due to its fast off-rate and voltage-dependent binding properties memantine, when used at low doses, does not accumulate in the channel and so does not substantially interfere with normal synaptic NMDAR activity 126, 147. As such, it was proposed that low doses of memantine would not substantially affect the trans-synaptic activation of synaptic NMDARs while allowing blockade of chronic extrasynaptic NMDAR activation. This discrimination was recently demonstrated definitively in a study on autaptic hippocampal cultures148. This study showed that the relative selectivity for extrasynaptic NMDARs is sensitive to the concentration of memantine used, as increasing the levels of memantine can lead to blockade of synaptic NMDARs (Figure 6b). Note, however, that it has been shown that memantine also preferentially antagonizes GluN2C- and GluN2D-containing NMDARs over GluN2A- and GluN2B-containing NMDARs under physiological levels of Mg2+ 149. This should be taken into account when interpreting the effects of memantine in systems in which GluN2C/D levels are substantial.
Nevertheless, the usefulness of memantine in preferentially uncoupling the NMDAR from pro-death extrasynaptic NMDAR signalling has also been shown39, 43. For example, low doses of memantine not only block NMDAR-dependent neuronal death that is induced by toxic doses of NMDA without interfering with trans-synaptic NMDAR-dependent activation of neuroprotective pathways; it does so in a manner which does not induce neuronal degeneration in vivo, unlike ‘conventional’ antagonists such as MK-80143. The in vivo use of memantine to antagonize neurodegenerative pathways that were characterized in vitro has been elegantly demonstrated in the context of HD pathology44, 132. Treatment of YAC128 mice with low dose memantine (1 mg/kg) from 2 months of age restored striatal phospho-CREB levels to a level similar to that in wild-type animals, and prevented the motor learning deficits that are normally apparent at 4 months of age132. A similar dosing regime also prevented striatal loss in YAC128 mice at 12 months and prevented motor deficits44. The protective effects of low-dose memantine were in sharp contrast with the effects of a much higher dose (30 mg/kg), which blocks all NMDARs: an exacerbation of striatal cell loss and motor deficits44. Given that memantine is a well-tolerated drug that has already been approved for clinical use, it will be of great interest to see whether clinical trials of memantine for HD reveal a reduction or delay of symptoms.
The survival and well-being of neurons requires synaptic activity and a dialogue between the synapse and the nucleus. Ca2+ signals, initiated by synaptic NMDARs, propagate inward to the nucleus and control genomic programmes that lead to a decrease in mitochondrial vulnerability, reduced expression of pro-apoptotic factors including caspases, and a strengthening of antioxidant defence systems. This form of acquired neuroprotection increases the ‘neuronal survival reserve’ that enables neurons to cope with harmful conditions in order to fulfil their function as computational units within neural circuits. Extrasynaptic NMDARs antagonize the pro-survival signalling of synaptic NMDARs by disturbing the communication between the synapse and the nucleus at various levels. Cell death associated with neurodegenerative diseases (including AD and HD) may be partly due to an imbalance of synaptic and extrasynaptic NMDAR signalling caused by synapse loss, failure to transduce Ca2+ signals from the synapse to the nucleus, or by redistributions of NMDARs from synaptic to extrasynaptic site. The antagonistic signalling of synaptic and extrasynaptic NMDARs provides a novel conceptual basis for future developments of neuroprotective therapies.
We thank Dr Karen Bell and Dr. C. Peter Bengston for comments on the manuscript. Work in the authors’ laboratories is supported by the Wellcome Trust, the MRC, the BBSRC, the Royal Society (GEH) and by the Alexander von Humboldt Foundation, the ERC Advanced Grant, the Deutsche Forschungsgemeinschaft, the EU Network of Excellence NeuroNE, and the EU Project GRIPANNT (HB). GEH is a MRC Senior Non-Clinical Research Fellow. HB is a member of the Excellence Cluster CellNetworks at Heidelberg University.