D-Serine is a physiological co-agonist of NMDARs (N-methyl-D-aspartate receptors) required for neurotransmission, synaptic plasticity and neurotoxicity. There is no consensus, however, on the relative roles of neurons and astrocytes in D-serine signalling. The effects of D-serine had been attributed to its role as a gliotransmitter specifically produced and released by astrocytes. In contrast, recent studies indicate that neurons regulate their own NMDARs by releasing D-serine via plasma membrane transporters and depolarization-sensitive pathways. Only a minority of astrocytes contain authentic D-serine, whereas neuronal D-serine accounts for up to 90% of the total D-serine pool. Neuronal and glial D-serine production requires astrocytic L-serine generated by a 3-phosphoglycerate dehydrogenase-dependent pathway. These findings support a model whereby astrocyte-derived L-serine shuttles to neurons to fuel the synthesis of D-serine by serine racemase. We incorporate these new findings in a revised model of serine dynamics, called the glia–neuron serine shuttle, which highlights the role of glia–neuron cross-talk for optimal NMDAR activity and brain development.
D-Serine and NMDARs (N-methyl-D-aspartate receptors)
NMDARs are crucial for neuronal homoeostasis and are involved in a wide range of brain processes, including normal neurotransmission, synaptic plasticity, and learning and memory . Altered NMDAR activity is implicated in diverse conditions, such as aging, neurodegeneration and neuropsychiatric disorders . NMDARs display complex regulatory mechanisms and are unique in their requirement of two ligands to operate. Whereas glutamate binds to the NR2 subunit, a co-agonist (glycine or D-serine) interacts with the NR1 subunit .
D-Serine, a D-amino acid enriched in the brain , is now appreciated as a major co-agonist of NMDARs. Two different approaches have been employed to access the role of D-serine in regulating NMDARs. The first is a D-serine-depletion method pioneered by the Snyder group, which employs D-amino acid oxidase enzyme to selectively destroy D-serine . Enzymatic removal of D-serine decreases synaptic NMDAR potentials as well as the expression of NMDAR-dependent LTP (long-term potentiation) at the CA3–CA1 synapses and other brain regions [4–10]. Conversely, removal of glycine with the glycine oxidase enzyme does not affect LTP at the hippocampal CA1 or synaptic NMDAR potentials , indicating that D-serine is the major NMDAR co-agonist under these experimental conditions. Likewise, depletion of endogenous D-serine prevents NMDAR-mediated excitotoxicity in cell cultures and hippocampal slices, indicating that D-serine is the dominant NMDAR co-agonist involved in neurotoxicity [11–13].
Another approach to investigate the role of D-serine is to study mice with targeted deletion of the D-serine biosynthetic enzyme SR (serine racemase). This genetic approach avoids the caveats associated with enzymatic depletion of D-serine, which sometimes fails to completely eliminate the co-agonist or display non-specific effects due to contaminants in the enzyme preparations. SR-KO (knockout) mice exhibit altered NMDAR transmission and decreased expression of LTP in hippocampal CA1, dentate gyrus and lateral amygdala [14–16]. The deficit in NMDAR-dependent LTP observed in SR-KO mice depends on the intensity of stimulation. Stronger presynaptic stimuli appear to release glycine and compensate for D-serine deficits in the lateral amygdala of SR-KO mice .
SR-KO mice are less susceptible to Aβ (amyloid β-peptide)-elicited neurotoxicity in vivo , and display lower infarct volume upon middle cerebral artery occlusion . These mice also exhibit NMDAR hypofunction phenotypes that are reminiscent of schizophrenia, such as impairments in pre-pulse inhibition and deficits in spatial memory , lower hippocampal volume and diminished number of dendritic spines . Altogether, studies using D-serine-removal strategies and SR-KO mice support the notion that D-serine is a major or even the dominant NMDAR co-agonist in the forebrain.
Which classes of NMDARs are activated by D-serine? To address this question, the Oliet and Mothet group treated hippocampal slices with glycine oxidase and D-amino acid oxidase enzymes to eliminate glycine or D-serine respectively . D-Serine appears to be the main ligand of NR2a-containing synaptic NMDARs at the hippocampal CA1, whereas glycine actions are more restricted to extra-synaptic receptors . However, other studies indicate that glycine substantially overlaps or compensates for D-serine deficits at synaptic NMDARs when SR-KO mice are used [9,16]. Bolshakov and colleagues demonstrated recently that the identity of the co-agonist is also determined by the synaptic activity level . Ambient D-serine mediates tonic activation of synaptic NMDARs in the amygdala, whereas stronger afferent stimulation appears to release glycine from glia, which is also required for synaptic NMDAR activation in this region . In the retina, Diamond and co-workers demonstrated that D-serine contributes to ambient co-agonist site occupancy, whereas glycine release appears to be required for activation of synaptic NMDARs in evoked synaptic events . Additional studies are required to precisely determine the relative roles of these co-agonists in stimulating the different types of NMDARs.
D-Serine production by SR appears to be regulated by glutamatergic transmission. SR is activated by Grip-1 (glutamate receptor-interacting protein 1), which increases D-serine synthesis following AMPAR (α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid receptor) stimulation . On the other hand, activation of NMDARs decreases D-serine synthesis by eliciting translocation of SR from the cytosol (where it normally resides) to the dendritic membrane . This mechanism seems to involve SR phosphorylation at Thr227 and atypical O-palmitoylation at still unidentified serine/threonine residues. Snyder and colleagues demonstrated that NMDAR activation also stimulates NO (nitric oxide) production which nitrosylates SR at Cys113 and decreases D-serine synthesis as well . SR inactivation following NMDAR (over)stimulation might be a feedback mechanism that abrogates D-serine synthesis in order to prevent or limit neurotoxicity at nearby synapses.
D-Serine: a gliotransmitter or a neuronal signalling molecule?
Pioneering research in the last two decades suggests a role for astrocytes in synaptic transmission by releasing glutamate, D-serine and other molecules as gliotransmitters . Preventing exocytotic gliotransmitter release from astrocytes by using a calcium clamp technique disrupts the NMDAR-dependent LTP in hippocampal CA1 . The notion that astrocytes influence neuronal activity by releasing transmitters, however, is still hotly debated . McCarthy and colleagues showed that LTP is not affected by astrocyte Ca2+ signalling, raising questions about the gliotransmitter hypothesis .
Initial localizations of D-serine in glia and its apparent exocytosis from astrocytic cultures indicated that D-serine might be a gliotransmitter [5,8,27–32]. However, recent studies using SR-KO mice as controls demonstrate conclusively that the majority of D-serine signalling originates from neurons, suggesting that D-serine-mediated gliotransmission may play a more limited role [33–36].
Using new antibodies, we demonstrated that SR is preferentially expressed by neurons in vivo, which also contain D-serine and release it upon membrane depolarization or via neutral amino acid transporters [11,33]. Mori and co-workers confirmed the preferential expression of SR in glutamatergic neurons by using SR-KO mice as controls, which ensured the specificity of the antibodies .
In order to obtain higher sensitivity and specificity for SR staining, we recently collaborated with Sol Snyder's group to analyse transgenic mice that express EGFP under SR promoter . Using GFP as a surrogate marker, the neuronal predominance of SR was confirmed, indicating that only a minor fraction of astrocytes (10–20%) express SR in the cortex and hippocampus . Experiments with cell-selective SR-KO mice carried out by Joe Coyle's group corroborate the predominant role of neurons in D-serine synthesis . A substantial decrease in SR expression in the cerebral cortex and hippocampus was observed in mice with selective KO of the neuronal SR, whereas astrocytic-specific KO lowers total brain SR by only 15% in these regions .
Bearing in mind potential artefacts stemming from non-specific binding of antibodies, we recently optimized the staining methodology for D-serine. We found that the neuronal D-serine pool is substantially larger than thought previously . Using SR-KOs as negative controls, only approximately 10% of the astrocytes appears to contain authentic D-serine in the cerebral cortex and hippocampus, whereas up to 90% of the glutamatergic neurons are positive for D-serine in these regions . Previous reports did not employ SR-KO mice controls to evaluate the specificity of D-serine staining, casting doubts on earlier demonstrations of predominant astrocytic localizations of D-serine.
New genetic models developed by Joe Coyle and colleagues also support a role of neuronal D-serine in regulating NMDARs [36,37]. Neuron-selective SR-KO mice display decreased LTP at hippocampal CA1 and this is associated with reduced dendritic arborization and fewer dendritic spines in pyramidal neurons [36,37]. Conversely, astrocyte-selective SR-KO mice display normal phenotypes . However, cell-selective SR-KO mice display only modest decreases in brain D-serine, precluding a precise determination of the relative contribution of neurons and astrocytes in D-serine dynamics.
Pathways for D-serine release from neurons
The realization that neurons are the main D-serine-harbouring cells in the nervous system led us to investigate possible pathways for neuronal D-serine release. Neurons release D-serine by non-vesicular mechanisms that include depolarization-sensitive pathways and plasma membrane transporters [9,33]. Asc-1, a neutral amino acid antiporter selectively expressed by neurons , seems to play a major role in D-serine dynamics. Targeted deletion of this transporter disrupts the high-affinity D-serine transport in the brain and causes severe neurodevelopmental problems . Asc-1 operates predominantly (albeit not exclusively) in an exchange mode whereby D-serine transport is coupled to the counter-transport of a neutral amino acid, typically L-serine, L-alanine or L-cysteine .
Together with Jean-Marie Billard, we have identified D-isoleucine as a selective Asc-1 substrate that elicits D-serine release by enhancing D-serine/amino acid exchange in neurons, but not in astrocytes, which lack Asc-1 . Using D-isoleucine as a tool, we found that Asc-1 modulates LTP at the hippocampal CA1, indicating that neurons regulate their own plasticity by Asc-1-mediated D-serine release . By using a D-serine biosensor, Marinesco and colleagues demonstrated recently that Asc-1 is a major pathway for D-serine release in vivo , but it is still not clear how Asc-1 is activated under physiological conditions. Additional studies with high-affinity Asc-1 blockers that do not activate amino acid hetero-exchange are required to evaluate the role of Asc-1 in D-serine dynamics. Furthermore, additional pathways for neuronal D-serine release to regulate NMDARs are also possible.
The serine shuttle
The serine shuttle model
SR has a low apparent affinity for L-serine (Km ~10 mM), indicating that the enzyme is not normally saturated with its substrate . The majority of brain L-serine comes from the glial metabolism of glucose via the PHGDH (3-phosphoglycerate dehydrogenase) pathway. PHGDH is an astrocytic enzyme that converts 3-phosphoglycerate into 3-phosphohydroxypyruvate, the committed step in L-serine biosynthesis . Expression of PHGDH is not detectable in adult neurons [35,45]. Furuya and colleagues demonstrated that targeted deletion of the Phgdh gene in mouse astrocytes lowers brain L-serine levels by 80%, indicating that astrocytes are the main source of brain L-serine . Phgdh-KO mice also display a drastic decrease in D-serine immunoreactivity in neurons, similar to that seen in SR-KO mice . The data are consistent with the notion that astrocytic L-serine shuttles to neurons to fuel the neuronal synthesis of D-serine [35,42] (Figure 1).
A possible pathway for L-serine export from astrocytes is the neutral amino acid transporter ASCT1, which seems to be mainly expressed by astrocytes . In neurons, L-serine is taken up by a still unidentified transporter and subsequently converted into D-serine by the predominantly neuronal SR [11,34–36]. Neuronal D-serine can be released via the Asc-1 transporter or depolarization-sensitive pathways to activate synaptic NMDARs [9,33,36] (Figure 1). Interestingly, Asc-1 transporter also releases glycine by hetero-exchange mechanisms, and this may also regulate NMDARs . Because L-serine can be converted into glycine by the serine hydroxymethyltransferase enzyme, it is possible that the serine shuttle from glia to neurons also provides glycine.
The α,β elimination activity of SR towards D-serine was thought previously to influence D-serine dynamics via limited degradation of neuronal D-serine. This reaction would limit the achievable intracellular D-serine concentration in neurons, allowing greater D-serine accumulation in astrocytes for they express little SR . However, recent data indicating that D-serine localizations mirror those of SR  suggest that the α,β elimination reaction does not play an important role in the relative distribution of D-serine in neurons and glia.
The serine shuttle model predicts that disruption of glial metabolism will disturb D-serine signalling as well. In this framework, neuronal dependence on astrocytic metabolism may underlie the profound inhibitory effect of glia toxins in NMDAR synaptic potentials and LTP [5,47]. Nevertheless, although only a small fraction of astrocytes contain authentic D-serine , our model does not exclude possible regulatory roles of glial D-serine on NMDARs. Glial D-serine may also be relevant in neurodegeneration. Aβ appears to increase SR expression in microglia, which may release D-serine and contribute to neurotoxicity via NMDAR activation .
The importance of the glia–neuron serine shuttle is by no means limited to the cellular mechanisms of D-serine production. Furuya and Watanabe  have shown that astrocyte-derived L-serine is key for neuronal survival and is required for the synthesis of essential metabolites, including sphingolipids and phosphatidylserine. Consistent with a major role for L-serine in brain development, patients harbouring mutations in the PHGDH gene have severe neurodevelopmental problems, including microcephaly, mental retardation and seizures .
Astrocytes had long been thought to be the only source of D-serine in the brain, but recent studies indicate that a large majority of D-serine comes from neurons. Although these studies do not exclude a role for D-serine as a potential gliotransmitter, it seems now unlikely that astrocytes play an important role as a direct source of D-serine. On the other hand, astrocytes export L-serine to neurons to serve as the substrate for D-serine biosynthesis. This new view of D-serine dynamics places the glia–neuron metabolic cross-talk as the ultimate determinant of D-serine levels. The glia–neuron serine shuttle is also crucial for normal brain development and underscores the importance of astrocytes for brain function. Future research exploring the mechanisms of L-serine export, D-serine release and the relative contributions of glycine and D-serine will clarify the role of the serine shuttle in NMDAR regulation.
5th Conference on Advances in Molecular Mechanisms Underlying Neurological Disorders: A joint Biochemical Society/European Society for Neurochemistry Focused Meeting held at the University of Bath, U.K., 23–26 June 2013. Organized and Edited by Marcus Rattray (University of Bradford, U.K.) and Rob Williams (University of Bath, U.K.).
This work was funded by grants from the Israel Science Foundation and the Legacy-Heritage fund.