NMDAR Hypofunction Animal Models of Schizophrenia

The N-methyl-d-aspartate receptor (NMDAR) hypofunction hypothesis has been proposed to help understand the etiology and pathophysiology of schizophrenia. This hypothesis was based on early observations that NMDAR antagonists could induce a full range of symptoms of schizophrenia in normal human subjects. Accumulating evidence in humans and animal studies points to NMDAR hypofunctionality as a convergence point for various symptoms of schizophrenia. Here we review animal models of NMDAR hypofunction generated by pharmacological and genetic approaches, and how they relate to the pathophysiology of schizophrenia. In addition, we discuss the limitations of animal models of NMDAR hypofunction and their potential utility for therapeutic applications.


INTRODUCTION
Schizophrenia is a debilitating psychiatric disease that affects ∼1% of the world population and places a major socio-economic burden (Blot et al., 2013). Patients exhibit positive and negative symptoms, as well as cognitive impairments, which typically emerge at early adolescence and worsen over time (Krystal et al., 1994;Morgan and Curran, 2006;Javitt, 2007;Howes et al., 2015). Both clinical and basic research suggest that schizophrenia is a neurodevelopmental disorder involving a variety of susceptibility genes, environmental risk factors, and epigenetic alterations (Cardno et al., 1999;Cannon et al., 2002;Shi et al., 2008;Walsh et al., 2008;Jaaro-Peled et al., 2009;Singh and O'Reilly, 2009), but it is unclear how these factors may contribute to the development of schizophrenic symptoms (Lewis and Gonzalez-Burgos, 2008;Faludi and Mirnics, 2011). Moreover, limited understanding on the pathophysiological mechanisms of schizophrenic brain has hindered the development of effective treatment for this disease. Current therapeutics is limited to antipsychotics (APDs) that mainly reduce positive symptoms but are ineffective for treating persistent negative or cognitive symptoms, which are often present before the onset of positive symptoms and lead to long-term functional impairments in schizophrenia patients (Howes et al., 2012;Citrome, 2014).
The glutamate hypothesis first emerged in the 1980s, and the N-methyl-D-aspartate receptor (NMDAR) hypofunction model was proposed afterwards upon observing that NMDAR antagonists could recapitulate a full range of positive, negative, and cognitive symptoms of schizophrenia in normal human subjects (Anis et al., 1983;Krystal et al., 1994). Consistent with this initial observation, findings from clinical, pharmacologic, and genetic studies suggested that NMDAR hypofunction may be one of the pathophysiological mechanism for schizophrenia (Goff and Coyle, 2001;Moghaddam, 2003;Jones et al., 2011;Javitt et al., 2012;Lin et al., 2012). NMDARs are ionotropic glutamate receptors that are comprised of the two obligatory NR1 subunits and two NR2 and/or NR3 subunits . They function as a "coincidence detector" of pre-and postsynaptic activity and have crucial roles in glutamatergic neurotransmission, local rhythmic activity, and synaptic plasticity (Collingridge et al., 1988;Olney and Farber, 1995;Jensen and Lisman, 1996;Fellin et al., 2009). Thus, NMDARs are known to modulate cognition, memory, and higher-order brain functions (Moghaddam et al., 1997;Adams and Moghaddam, 1998;Palmer et al., 2008;Collingridge et al., 2013). In the last several decades, animal models of NMDAR hypofunction have been widely utilized to study the neurobiology of schizophrenia, as well as to test drugs for treating symptoms of schizophrenia.
Here we review current literatures on animal models of NMDAR hypofunction using pharmacological and genetic approaches to induce NMDAR hypofunctionality.

PHARMACOLOGICAL APPROACHES TO INDUCE NMDAR HYPOFUNCTION
Phencyclidine Phencyclidine (PCP) is an abused drug that acts as a non-competitive blocker of NMDARs at lower doses. PCP administration was found to produce hallucinogenic activity in healthy human subjects (Itil et al., 1967). This had led to further investigation of the pathophysiological mechanisms for PCPinduced behavioral changes, particularly through inhibition of NMDARs (Carlsson and Carlsson, 1990;Johnson and Jones, 1990;Javitt and Zukin, 1991;Olney and Farber, 1995). To date, a wide range of animal studies have been conducted with different PCP administration regimen (acute, subchronic, or chronic), dosage, and co-administration with other drugs. The effects of PCP in animals at the molecular, electrophysiological, and behavioral levels were assessed by a variety of approaches.
In rodents, behaviors associated with positive symptoms of schizophrenia are assayed as novelty-induced hyperlocomotion in the open field test. Negative symptom-like behavioral alterations correlate with social withdrawal and anhedonia, which are commonly evaluated by social interaction, sucrose preference and forced swim tests (FST) (Sams-Dodd, 1996). Cognitive impairments are assessed using a variety of behavioral assays including radial arm maze, Morris water maze, Y-maze, T-maze, novel object recognition test, avoidance task learning test, attentional set-shifting task, modified hole board task, object and object-in-context recognition memory task, fivechoice serial-reaction time test, and contextual fear conditioning. Moreover, the endophenotype corresponding to sensorimotor gating deficits in schizophrenics is detected as impaired prepulse inhibition (PPI) of acoustic startle response (McKibben et al., 2010). Acute administration of PCP induces a full spectrum of behavioral changes associated with positive, negative and cognitive symptoms of schizophrenia. Positive symptomsrelated behaviors exhibited in acute PCP animal models include increased stereotypic behavior and ataxia, but there are conflicting reports regarding changes in locomotor activity. Negative symptoms-related behaviors are reflected as reduced social interaction, while cognitive deficits range from impaired latent learning, deficits in attention and cognitive flexibility to decreased sensorimotor gating (Sturgeon et al., 1979;Nabeshima et al., 1986;Mansbach and Geyer, 1989;Sams-Dodd, 1995, 1996Martinez et al., 1999;Noda et al., 2001;Abdul-Monim et al., 2003;Egerton et al., 2005). These behavioral changes are accompanied by increased brain metabolism in areas of cortex, basal ganglia, and thalamus, decreased parvalbumin (PV) mRNA expression in the hippocampus and dorsal reticular nucleus of the thalamus, and altered zinc finger protein 225 (zif268) mRNA expression in the infralimbic cortex (Martinez et al., 1999;Egerton et al., 2005). Another study showed that acute PCP administration during early postnatal time period results in loss of PV-containing neurons from the primary somatosensory, motor, and retrosplenial cortices (Wang et al., 2008).
Compared with acute exposure to PCP, subchronic and chronic exposure to PCP induces a more overt and sustained schizophrenia-like state in both rodents and in humans. As it provides a wider time frame to study long-lasting changes related to symptoms of schizophrenia, subchronic, and chronic PCP treated animal models are more commonly used to study the NMDAR hypofunctionality (Rainey and Crowder, 1975;Allen and Young, 1978;Cosgrove and Newell, 1991). In particular, these models have been applied to study the effects of NMDAR blockade on negative and cognitive symptoms-related behaviors, as well as to investigate the potential neurochemical and neuroanatomical basis underlying these behavioral changes (Jentsch and Roth, 1999;Lewis and Levitt, 2002; Table 1).

Ketamine
Ketamine was first synthesized as a PCP derivative back in the 1960s in an effort to minimize the side effects and neurotoxicity of PCP (Maddox et al., 1965). As a noncompetitive NMDAR antagonist, ketamine acts by specifically binding to a site near the channel pore (Roth et al., 2013). Ketamine was found to be safe and effective even with repeated administration with minimal side effects and was soon used as a clinical anesthetic  (Domino et al., 1965;Corssen and Domino, 1966;Li and Vlisides, 2016). Ketamine administration has often been used in animal studies as a preclinical model to test the effects of APDs and novel compounds since atypical APDs are found to be effective in blocking ketamine's behavioral effects in both humans and rodents (Jentsch and Roth, 1999;Krystal et al., 1999Krystal et al., , 2005Becker et al., 2003;Lees et al., 2004;Gilmour et al., 2009;Neill et al., 2010). However, apart from the blockade of NMDARs, ketamine can act on a wide-range of targets in various cellular processes (Li and Vlisides, 2016). Acute ketamine administration in animals results in cognitive deficits including reduced sensorimotor gating, spatial learning and memory impairments, along with changes in theta and gamma band activity (Verma and Moghaddam, 1996;de Bruin et al., 1999;Ehrlichman et al., 2009;Kittelberger et al., 2012;Szlachta et al., 2017;Coronel-Oliveros and Pacheco-Calderon, 2018). In contrast to cortical hypodopaminergic state observed in schizophrenia, these cognitive deficits are accompanied by increased DA release in the PFC (Verma and Moghaddam, 1996). Other studies also reported that cognitive deficits are associated with increased frontal cortical blood flow (Ingvar and Franzen, 1974;Vollenweider et al., 1997). Moreover, acute ketamine administration induces hyperlocomotor activity and stereotypic behaviors, which might be attributed to increased DA and serotonin turnover in the striatum and cortex (Irifune et al., 1991;Chatterjee et al., 2011Chatterjee et al., , 2012Coronel-Oliveros and Pacheco-Calderon, 2018). However, there are inconsistent reports as to the effect of acute ketamine administration on negative symptoms-related behaviors based on social interaction tests (Silvestre et al., 1997;Chatterjee et al., 2011;Coronel-Oliveros and Pacheco-Calderon, 2018). In fact, ketamine is known to have anti-depressant effects at low doses, which may oppose the induction of negative symptoms-related behaviors in animals (Zanos and Gould, 2018).
Subchronic ketamine administration, however, can induce negative symptoms-related behaviors, in addition to positive symptoms-related behaviors. These include increased time spent immobile during the forced swim test (FST), hyperlocomotor activity, and stereotypy (Becker et al., 2003;Chatterjee et al., 2011Chatterjee et al., , 2012. Hyperlocomotor activity exhibited after a subchronic dose of ketamine is attributed to increased DA and serotonin levels in the cortex and striatum, which is thought to result from alternations in gene expression of DA and serotonin receptors (Chatterjee et al., 2011(Chatterjee et al., , 2012. On the other hand, negative symptom-related behavior in both acute and subchronic ketamine administration is associated with decreased levels of glycine in the striatum, hippocampus, and cortex (Chatterjee et al., 2012). This is consistent with previous findings from schizophrenia patients that linked reduced levels of glycine to negative symptoms of schizophrenia and increased levels of glycine for the treatment of negative symptoms (Javitt, 2010).
Subchronic ketamine administration also induce cognitive deficits, such as reversal learning and long-term spatial memory impairments (Featherstone et al., 2012;Szlachta et al., 2017). In addition, chronic ketamine administration results in decreased PV interneuron density in the hippocampus (Keilhoff et al., 2004;Kittelberger et al., 2012). Fast-spiking PV interneurons are important for encoding and storage of information required for working memory, and dysfunction in fast-spiking PV interneurons are known to cause cognitive deficits, potentially through disrupting theta and gamma oscillations (Bartos et al., 2007). Indeed, subchronic ketamine administration induce cognitive deficits that are accompanied by alterations in theta and gamma oscillation from the hippocampus and PFC (Featherstone et al., 2012). Moreover, audioradiographic imaging studies have reported dysconnectivity between PFC and hippocampus in both acute and subchronic ketamine animal models (Dawson et al., 2013).
Overall, ketamine can induce behavioral alterations associated with schizophrenia symptoms but with less excitotoxicity than PCP. When comparing treatment regimens, subchronic ketamine treatment models are more widely used, since it can induce persistent negative and cognitive symptoms-related behaviors with hallmark features of dysfunction in neuronal circuitry (Lahti et al., 2001;Morgan et al., 2009). However, behavioral deficits and cortical dysfunctions cannot be attributed to changes only in the glutamatergic system as it may involve changes in other neurotransmitter systems such as DA and serotonin (Grace, 2012;

MK-801
Also known as dizocilpine, MK-801 is a non-competitive NMDAR antagonist that blocks the channel in a use-and voltage-dependent manner (Foster and Fagg, 1987;Huettner and Bean, 1988). Like ketamine, MK-801 seems to preferentially act on GABAergic interneurons. This selective action of MK-801 on interneurons reduces inhibitory influence on excitatory pyramidal neurons, leading to hyperexcitation in the PFC neuronal circuit (Yonezawa et al., 1998;Homayoun and Moghaddam, 2007). However, more recent studies showed that local MK-801 infusion into the PFC does not directly produce disinhibition (Gratton et al., 1987;Suzuki et al., 2002;Jodo et al., 2005), suggesting that hyper-glutamatergia in the PFC may require MK-801's effects in other brain regions such as the hippocampus (Jodo, 2013;Nakazawa et al., 2017). Moreover, MK-801 has substantially longer action of NMDAR blockade in rodents and higher specificity for NMDARs than ketamine Pinault, 2008;Hakami et al., 2009).
Chronic MK-801 treatment also results in dysfunctional GABAergic interneurons as indicated by reduced PV interneuron density in the hippocampus (Braun et al., 2007). Similarly to acute treatment, chronic and subchronic MK-801 treatment leads to cognitive deficits (Latysheva and Rayevsky, 2003;Stefani and Moghaddam, 2005;Rujescu et al., 2006;Li et al., 2011;Xiu et al., 2014;Liu et al., 2017;Unal et al., 2018). In addition, chronic MK-801 treatment leads to sensorimotor gating and social interaction deficits, although there are controversial results on locomotor activity (Latysheva and Rayevsky, 2003;Eyjolfsson et al., 2006;Xiu et al., 2014;Unal et al., 2018). Another study showed that chronic MK-801 induce increased anxiety-like behavior and working memory deficits, which was accompanied by degenerative changes of myelin sheaths, decreased white matter and corpus callosum volume (Xiu et al., 2014). Early postnatal chronic MK-801 treatment model reported reduced locomotor activity, decreased rearing behavior, exploratory behavior, increased anxiety-like behavior as well as learning impairments (Latysheva and Rayevsky, 2003; Table 3). Results from MK-801 administration in awake rats suggest that NMDAR inhibition causes cortical excitation by disinhibition of pyramidal neurons in the PFC (Homayoun and Moghaddam, 2007). In addition, studies consistently report that chronic MK-801 administration leads to increased dopaminergic and serotonergic activity in the frontal cortex, nucleus accumbens and striatum (Loscher et al., 1991). As discussed earlier, an increased dopaminergic activity in the frontal cortex by chronic MK-801 administration may not be consistent with the cortical hypodopaminergic state in schizophrenia patients, which further points to the limitations in the use of pharmacological rodent models to recapitulate the dysregulation of DA systems.

Knockout or Knockdown of NR1 Subunits
Based on genetic analyses, polymorphisms exist in both coding and promoter regions of glutamate receptor ionotropic NMDA type subunits (GRIN), impacting both NMDAR transcript levels and/or functions. Single nucleotide or dinucleotide-repeated polymorphisms of the NMDAR subunit genes, such as NR1 (GRIN1), NR2A (GRIN2A), and NR2B (GRIN2B), increase susceptibility to schizophrenia (Ohtsuki et al., 2001;Rice et al., 2001;Miyatake et al., 2002;Itokawa et al., 2003;Iwayama-Shigeno Qin et al., 2005;Martucci et al., 2006;Tang et al., 2006;Zhao et al., 2006). Moreover, postmortem and genetic studies from schizophrenia patients support that abnormalities in glycine modulatory sites on the NMDARs may contribute to the pathophysiology of schizophrenia (Coyle, 2004). Amongst various NMDAR subunits, NR1 is essential for formation and synaptic expression of NMDARs . A complete loss of NR1 subunit in mice [NR1 knockout (KO)] is neonatally lethal, suggesting that NMDARs play a crucial role in early development (Forrest et al., 1994). To overcome neonatal lethality, a line of NR1 knockdown (KD) mice was generated by ectopic transgenic expression of one of the NR1 splice variants (NR1-1a) in the NR1 KO mice. This line of NR1 KD mice exhibits altered dendritic differentiation, branching, and somatosensory pattern, accompanied by increased axonal arborizations with faster and prematurely developed projection neurons of the corpus callosum (Iwasato et al., 1997;Lee et al., 2005). However, the average lifespan of this NR1 KD mice was dependent on the level of transgene expression. This limits the usage of these mice for further behavioral, molecular, and structural studies in adult mice. Another NR1 KD mice were created by inserting a neomycin cassette into an intronic region of the GRIN1 gene, leading to significantly lower levels of NR1 expression (Mohn et al., 1999). These mice are viable and exhibit a full range of behavioral phenotypes associated with schizophrenia, such as hyperlocomotor activity, stereotypy, self-injury, decreased anxiety-related behavior, reduced nest building, impaired social and sexual interactions, abnormal evoked response potentials (ERPs), cognitive inflexibility, abnormal selective attention, with spatial cognitive and sensorimotor gating deficits (Mohn et al., 1999;Duncan et al., 2004;Moy et al., 2006;Bickel et al., 2007;Dzirasa et al., 2009;Halene et al., 2009). These NR1 KD mice also exhibit reduced synapse-number in an age-dependent manner, decreased NMDA currents, reduced 2-deoxyglucose (DG) uptake in areas of the neocortex, increased amplitudes of auditory and visual ERPs, attenuated cortical and hippocampal theta-gamma phase coupling, increased dendritic length, and synapse-specific reductions in 14-3-3ε and disrupted in schizophrenia 1 (DISC1) protein expression (Duncan et al., 2002;Dzirasa et al., 2009;Halene et al., 2009;Ramsey et al., 2011). Others have also developed hypomorphic NR1 subunit mice with single or double point mutations in NMDAR glycine binding sites, GRIN1 D481 and GRIN1 K483 . Hypomorphic NR1 mice with single point mutation in D481 (GRIN1 D481N ) are viable and exhibit increased startle reactivity, deficits in spatial recognition, spatial reference learning and memory, reduced sociability, anxiety, and sensitivity to NMDA-induced seizures, with impaired hippocampal LTP (Kew et al., 2000;Labrie et al., 2008). In contrast, GRIN1 K483Q mice are perinatally lethal (Kew et al., 2000). Subsequently, Ballard et al. developed the compound heterozygote mice with point mutations in both glycine binding sites. These mice (GRIN1 D481N/K483Q ) are viable and display hyperlocomotor activity, stereotypy, disrupted nesting behavior, spatial learning, and sensorimotor gating deficits. These behavioral changes are accompanied by striatal dopaminergic and serotonergic hyperfunction. Interestingly, deficits in hippocampal LTP are rescued by D-serine administration (Ballard et al., 2002).
To understand how NMDAR hypofunction affects behaviors at cellular and circuit levels, various animal models with brain region-and/or cell-type specific deletion of NR1 have been created. Forebrain excitatory neuron-specific NR1 KO mice exhibit impairments in spatial memory, encoding and flexible expression of non-spatial memory with slower acquisition of trace fear conditioning. These mice also exhibit degraded coactivation of CA1 place cells during exploration and lack NMDAR-mediated synaptic currents and LTP in CA1 synapses (McHugh et al., 1996;Tsien et al., 1996;Huerta et al., 2000;Rondi-Reig et al., 2001). Dentate-gyrus (DG) granule cellsspecific NR1 KO mice display deficits in the process of pattern separation, which is accompanied by reduced firing rate from cornu ammonis 3 (CA3) pyramidal cells during context-specific modulation. These mice also exhibit deficits in spatial working memory accompanied by impaired LTP in both medial and lateral perforant path inputs to the DG (McHugh et al., 2007;Niewoehner et al., 2007). CA1-specific NR1 KD rats were created by injecting NR1 antisense RNA into the CA1 region of the hippocampus. These rats exhibit impairments in avoidance task learning and memory formation (Cheli et al., 2002(Cheli et al., , 2006. Similarly, CA3 pyramidal cells-specific NR1 KO mice exhibit deficits in associative memory recall accompanied by impaired NMDAR-dependent LTP at the commissural/associational (C/A) pathway (Nakazawa et al., 2002). Another CA3-specific NR1 KO mice, created by an adeno-associated virus (AAV)-induced deletion of the NR1 gene, exhibit increased impulsive behavior, learning impairments, and decreased social approach behavior (Rajji et al., 2006;Finlay et al., 2015).
Mice with deletion of NR1 from cortical and hippocampal GABAergic interneurons during early postnatal development exhibit hyperlocomotor activity, anhedonia-like and anxietylike behaviors, mating and nest-building deficits, as well as social memory, spatial working memory, and prepulse inhibition deficits. These changes are accompanied by increased firing of cortical excitatory neurons with reduced neuronal synchrony. Overall, these findings support that NMDAR hypofunction in corticolimbic GABAergic interneurons during early postnatal development can lead to the development of schizophrenia-related behavioral phenotypes (Belforte et al., 2010). Furthermore, PV-specific NR1 KO mice were generated by targeted deletion of NR1 subunits from a subset of GABAergic interneurons. These mice exhibit deficits in spatial working, short-and long-term recognition memory, which are correlated with profound changes in neural activity related to cognition including increased gamma oscillation power and decreased theta oscillation power of local field potentials (LFP) in the hippocampus (Korotkova et al., 2010). Moreover, other groups reported that the PV-specific NR1 KO mice display blunted MK-801-induced hyperlocomotor activity, suggesting that NMDARs in PV interneurons may be the site of MK-801 action (Belforte et al., 2010;Carlen et al., 2012). However, a more recent study reported that these mice are not protected against behavioral effects of MK-801, as MK-801 administration leads to increased stereotypy and pronounced catalepsy, which confound the locomotor readout (Bygrave et al., 2016). NR1 deletion in forebrain pyramidal neurons also results in behavioral changes associated with schizophrenia. The forebrain pyramidal cells-specific NR1 KO mice display hyperlocomotor activity and decreased self-care, as well as social and cognitive impairments (Tatard-Leitman et al., 2015). In addition, these mice exhibit decreased expression of dopamine D2 receptor (D2R) and G-protein-regulated inward-rectifier potassium channel 2 (GIRK2) in the forebrain, increased baseline gamma power and pyramidal cell excitability (Tatard-Leitman et al., 2015). However, mPFC and sensory cortex pyramidal cellsspecific NR1 KO mice only exhibit PPI and short-term memory impairments, suggesting that deletion of NR1 subunit in pyramidal neurons of broader forebrain regions may be required to induce a full range of symptoms of schizophrenia (Kehrer et al., 2008). Taken together, these studies demonstrate that NR1mediated deficits in either pyramidal or GABAergic neurons could cause an imbalance of excitation and inhibition in the cortical neural circuit, leading to development of behavioral phenotypes related to schizophrenia ( Table 4).

Knockout of NR2A or NR2B Subunits
Amongst four NR2 subunits (A-D), the NR2A and NR2B subunits are known to predominantly function in the forebrain. They impart different characteristics on functional NMDARs, with NR2A-containing NMDARs having more rapid kinetics than NR2B-containing NMDARs (Kutsuwada et al., 1992;Monyer et al., 1994;Loftis and Janowsky, 2003). A switch of NR2B to NR2A subunit expression in the forebrain and sensory systems is linked with the timing of critical period for sensory plasticity, making NR2B and NR2A subunits particularly known for its involvement in postnatal brain development (Monyer et al., 1994;Sheng et al., 1994;Yashiro and Philpot, 2008). At birth, NR2B is the predominant subunit, whereas NR2A subunit expression begins around postnatal day 3 (P3), after which there is a gradual increase expression of NR2A subunits (Monyer et al., 1994;Sheng et al., 1994;Zhong et al., 1995). Furthermore, NR2A and NR2B subunits are known to interact differentially with binding partners in the postsynaptic density (PSD) and may potentially activate different downstream signaling pathways according to changes in NMDAR subunit composition during development (Massey et al., 2004). Thus, various genetic animal models with the loss of NR2A or NR2B subunits have been studied to understand how expression and function of these subunits may contribute to NMDAR hypofunctionality.
As suggested by its early expression in development, homozygous NR2B KO mice are perinatally lethal. These mice were examined shortly after birth in hippocampal slices. The loss of NR2B abolishes synaptic NMDA responses and long term depression (LTD) in the CA1 hippocampal neurons, suggesting that NR2B expression has an essential role in synaptic plasticity (Kutsuwada et al., 1996). Since homozygous NR2B KO mice died perinatally, the heterozygous NR2B mutant mice were generated to study the effect of NR2B deletion in adulthood. One study found diminished NMDAR excitatory postsynaptic currents (EPSCs) and LTP in the fimbrial (Fim)-CA3 synapse when recorded from hippocampal slices of the heterozygous NR2B mutant mice (Ito et al., 1997). A separate study reported that these mice exhibit enhanced startle responses to acoustic stimuli (Takeuchi et al., 2001).
Mouse models with deletion of NR2B subunits in certain brain regions and/or cell types were further studied to understand the role of NR2B subunits in cognitive functions. A forebrain pyramidal cells-specific NR2B KO mice were created by genetic deletion of NR2B beginning at early postnatal development using the calcium 2+ /calmodulindependent protein kinase II-alpha (CamKII-α) promoter. These mice display hyperlocomotor activity and exaggerated depressant effect (Badanich et al., 2011). Similarly, another study done on these forebrain pyramidal cells-specific NR2B KO mice reports impairment in spatial and non-spatial learning and memory, whereas hippocampal pyramidal cells-specific NR2B KO mice only display selective, spatial working memory deficits. Nevertheless, both forebrain and hippocampal pyramidal cells-specific NR2B KO mice exhibit decreased NMDAR-mediated EPSCs with accelerated decay kinetics and reduced cellular LTP in the hippocampus (von Engelhardt et al., 2008).
In contrast to NR2B KO mice, NR2A KO mice are viable. They exhibit hyperlocomotor activity and cognitive dysfunctions without any sensorimotor gating deficits (Sprengel et al., 1998;Brigman et al., 2008). The cognitive deficits exhibited in NR2A KO mice include spatial learning, fear coding, and discrimination learning impairments accompanied by impaired hippocampal LTP. When recorded from hippocampal slices generated from the NR2A KO mice, there was reduced NMDAR-mediated EPSCs and LTP in the CA3-CA1 synapse (Ito et al., 1997). Overall, the changes displayed in both genetic mutants of NR2A and NR2B subunits highlight the crucial role that both subunits play in synaptic plasticity and cognitive functions (Sakimura et al., 1995;Kiyama et al., 1998;Moriya et al., 2000). Due to overlapping roles of NR2A and NR2B subunits in learning and memory, NR2A and NR2B mutant mice would serve as great models to study the pathophysiology of cognitive symptoms of schizophrenia (Table 4).

Overexpression or Re-expression of NMDAR Subunits
Multiple studies have also examined behavioral and molecular consequences upon overexpressing different NMDAR subunit(s). In one study, overexpression of the NR1 subunit into the hippocampus of wild type mice via AAV-mediated delivery led to increased fear memory and neurogenesis as well as delayed onset of severe seizures (Kalev-Zylinska et al., 2009). Other studies reported that overexpression of NR2B subunits in the forebrain of a wild type mice improve learning and memory and enhance NMDAR-dependent synaptic potentiation (Tang et al., 1999(Tang et al., , 2001. In addition, transgenic mice with forebrain pyramidal cells-specific overexpression of NR2B postnatally exhibit enhanced social recognition  Forrest et al., 1994;McHugh et al., 1996McHugh et al., , 2007Tsien et al., 1996;Iwasato et al., 1997;Mohn et al., 1999;Huerta et al., 2000;Kew et al., 2000;Rondi-Reig et al., 2001;Ballard et al., 2002;Cheli et al., 2002Cheli et al., , 2006Duncan et al., 2002Duncan et al., , 2004Nakazawa et al., 2002;Lee et al., 2005;Moy et al., 2006;Rajji et al., 2006;Bickel et al., 2007;Niewoehner et al., 2007;Labrie et al., 2008;Dzirasa et al., 2009;Halene et al., 2009;Belforte et al., 2010;Korotkova et al., 2010;Ramsey et al., 2011;Carlen et al., 2012;Rompala et al., 2013;Finlay et al., Carlsson and Lindqvist, 1963;Sakimura et al., 1995;Ito et al., 1997;Kiyama et al., 1998;Sprengel et al., 1998;Moriya et al., 2000;Brigman et al., 2008 (Continued) Frontiers in Molecular Neuroscience | www.frontiersin.org  Kutsuwada et al., 1996;Ito et al., 1997;Takeuchi et al., 2001;von Engelhardt et al., 2008;Badanich et Gassmann et al., 1995;Erickson et al., 1997;Gerlai et al., 2000;Stefansson et al., 2002;Flames et al., 2004;Golub et al., 2004;Thuret et al., 2004;Gu et al., 2005;Roy et al., 2007;Ago et al., 2008;Chen et al., 2008;Barros et al., 2009;Fisahn et al., 2009;Feng et al., 2010;Pitcher et al., 2011;Yu et al., 2015 (Continued) Frontiers in Molecular Neuroscience | www.frontiersin.org • Weak defects in working memory in 14-3-3ε het mice • Reduction in the NMDAR-mediated synaptic currents in CA1 pyramidal neurons in 14-3-3 FKO mice • Protein levels of NR1 and NR2A subunits are reduced in the PSD fraction of 14-3-3 FKO mice • Impaired LTP at hippocampal CA3-CA1 synapses in 14-3-3 FKO mice • Altered synaptic connectivity in the prefrontal cortical synapses in 14-3-3 FKO mice • Altered synaptic neurotransmission and plasticity in 14-3-3ζ KO mice • Abnormal neuronal migration and axonal guidance defects in the hippocampus in 14-3-3ζ KO mice • Increased levels of DA in striatum in 14-3-3ζ KO mice Ikeda et al., 2008;Cheah et al., 2012;Ramshaw et al., 2013;Qiao et al., 2014;Foote et al., 2015;Xu et al., 2015 memory for different strains and animal species (Jacobs and Tsien, 2012). Another group found that overexpression of NR2B subunit in aged wild type mice enhances longterm spatial memory. On the other hand, transgenic mice with forebrain-specific NR2A overexpression display longterm memory deficits, suggesting that subtle shifts in subunit compositions can have major effects on native receptor function and highlights how different subunit compositions can produce receptors with different functional properties (Madden, 2002;Cui et al., 2013). Behavioral deficits in genetic mutants of NMDAR subunit(s) could also be rescued by restoring NMDAR subunit expression levels. In the NR1-KD mice, the hypomorphic insertion mutation could be excised in the Cre recombinase-dependent manner to restore NR1 in a temporal-regulated fashion. This allowed to investigate behavioral phenotypic rescue at different stages of development upon global restoration of NMDARs (Mohn et al., 1999;Mielnik et al., 2017). The mice with inducible NR1 rescue during adolescence or in adulthood achieved similar levels of functional recovery in some of cognitive deficits and negative symptoms-related behavioral changes, with cortically-mediated behaviors completely or nearly rescued. However, subcortically-mediated behaviors, such as hyperlocomotor activity and anxiety-related behaviors were only partially rescued. This suggested that higherorder brain functions are potentially more amenable to treatment in adulthood and unencumbered by critical period (Mielnik et al., 2017). Overexpressing NR2B in aged wild type mice was able to rescue age-associated impairments in hippocampal-dependent spatial memory. This was accompanied by enhanced LTP and increased NMDAR-mediated EPSPs from hippocampal slices generated from these aged mice with increased NR2B subunit expression, suggesting that increasing NR2B expression in aged animals can also enhance memory and synaptic transmission (Brim et al., 2013). These studies further point to how re-expression or overexpression of NMDAR subunits rescue behavior deficits associated to symptoms of schizophrenia, suggesting that enhancing levels of NMDAR subunits may ameliorate NMDAR hypofunctionality. However, limitations in fully restoring NMDAR subunits or rescuing behavioral phenotypes associated with subcortically-mediated behaviors still remain to be a challenge and may require further studies.
Moreover, the NMDAR levels can be influenced by activity or experience. Multiple studies showed that both NR1 and NR2A subunit protein expressions increase in the hippocampus of 1, 2, and 3-month old rats following habituation to a new environment via open field test and following novel object recognition task (Baez et al., 2013;Cercato et al., 2016). Shanmugasundaram et al. reported that 6 h after training rats in a radial maze along 10 consecutive days, there is an increase in NR1 and NR2B protein levels in synaptosomal fractions from the hippocampus and an increase in synaptic NR1 and NR2A levels in the PFC (Shanmugasundaram et al., 2015). In conclusion, NMDAR levels are influenced by genetic overexpression, reexpression of NMDAR subunit(s) and are also experienceand activity-dependent.

Metabotropic Glutamate Receptors (mGluRs)
Glutamatergic neurotransmission is mediated by both metabotropic and ionotropic glutamate receptors. While ionotropic glutamate receptors such as NMDA and quisqualate/a-amino-3-hydroxy-5-methyl-5-isoxazolepropionic acid (AMPA) receptors are responsible for fast excitatory transmission, metabotropic receptors have a modulatory role (Kew and Kemp, 2005). Metabotropic glutamate receptors (mGluRs) are further subdivided into three groups, with Group I metabotropic glutamate receptors (mGluR1 and mGluR5) mainly localized postsynaptically and positively linked to phospholipase C (PLC), which is known to potentiate glutamate function at NMDARs. In contrast, Group II (mGluR2 and mGluR3) and III (mGluR4, mGluR6, mGluR7, and mGluR8) metabotropic glutamate receptors are primarily presynaptic and are negatively linked to adenylyl cyclase (AC), which is known for its role in modulating neurotransmitter release (Javitt, 2004). Moreover, Group I mGluRs induce the enhancement of NMDAR currents and are involved in the direct phosphorylation of the NMDARs (Cartmell et al., 2000;Pisani et al., 2001). In particular, mGluR5s at the PSDs are physically linked to NMDARs and are known to enhance NMDAR function, this makes mGluRs a leading target for novel therapeutics to treat cognitive symptoms of schizophrenia (Ehlers, 2002;Yang et al., 2004;Gray et al., 2009).
Animal model studies have reported that the excitatory glutamatergic neurotransmission through ionotropic and metabotropic glutamate receptors is necessary for the correct postnatal development of the GABAergic network, with mGluR5s having a fundamental role in the development of PV interneurons (Uzunova et al., 2014;Barnes et al., 2015). PV-specific mGluR5 KO mice display memory impairments, increased compulsive-like behaviors, abnormal sensorimotor gating, and altered responsiveness to stimulants. These behavioral changes are accompanied by reduced PV interneuron density, decreased inhibitory synaptic currents, as well as abnormal ERPs and brain oscillatory activity (Barnes et al., 2015) Another study showed that mGluR5 KO mice exhibit deficient developmental switch from NR2B-to NR2A-containing receptors at synapses onto hippocampal CA1 pyramidal neurons and pyramidal primary visual cortical neurons (Matta et al., 2011), suggesting that mGluR5 plays an important role in regulating age-dependent expression of NMDAR subunits.
In addition, pharmacological modulations of different mGluR subtypes may provide alternative mechanisms to normalize aberrant neurotransmission induced by NMDAR hypofunction (Moghaddam et al., 1997;Marino and Conn, 2002). A licensed drug originally discovered for the prevention of relapse in alcohol dependence, acamprosate, interferes with mGluR5-dependent glutamate release. Preclinical studies with acamprosate suggest that it normalizes glutamate release and NMDAR functions without altering the normal glutamatergic neurotransmission (De Witte et al., 2005). Other approaches have also been made to normalize glutamatergic neurotransmission by using glutamate release inhibitors, such as selective mGluR2/3 agonists (LY2140023), which alleviates positive and negative symptoms in schizophrenia patients during phase II clinical trials, but failed in phase III clinical trials (Patil et al., 2007;Li et al., 2015). Thus, far, mGluR-targeting compounds have yet to be clinically approved for schizophrenia patients Kinon et al., 2011;Hopkins, 2013). Further studies in different animal models may facilitate this process by establishing functional link between mGluRs and NMDAR hypofunctionality ( Table 4).
In animal model studies, the NRG1 homozygous null mutant mice are found to be embryonically lethal (Erickson et al., 1997). This led to the studies of viable, NRG1 heterozygous null mutant mice. These mice exhibit hyperlocomotor activity and spatial learning impairments (Gerlai et al., 2000). Another NRG1 hypomorphic mice were created by deleting the transmembrane domain of the NRG1 protein. These heterozygous mice exhibit hyperlocomotor activity and impaired sensorimotor gating, which was accompanied by decreased MK-801 binding from the forebrain, suggesting fewer functional NMDARs (Stefansson et al., 2002). Subsequently, studies using mutant ErbB4 or NRG1 mouse lines revealed that the NRG1-ErbB signaling is crucial for tangential migration of cortical GABAergic interneurons, as the number of GABAergic interneurons in the embryonic cortex is decreased in either the neuronal specific NRG1 KO mice or a line of ErbB4 null mice (Flames et al., 2004). Another NRG1 heterozygous mice were generated by targeted disruption of type III NRG1 (Chen et al., 2008), which is one of many NRG1 isoforms highly expressed throughout embryonic and postnatal brain development and are implicated in schizophrenia (Meyer et al., 1997;Anton et al., 2004;Longart et al., 2004). These NRG1 heterozygous mice exhibit impaired performance on delayed alternation task and sensorimotor gating deficits. They also display enlarged lateral ventricles and decreased dendritic spine density on hippocampal pyramidal neurons (Chen et al., 2008).
The ErbB4 KO mice are also embryonically lethal (Gassmann et al., 1995). Thus, the heterozygous ErbB4 mutant mice were most widely used to study animal models of ErbB4 receptor. These mice mainly exhibit hyperlocomotor activity and reduced power of kainate-induced gamma oscillations, accompanied by reduced numbers of PV interneuron density in the hippocampus as well as reduced calbindin (CB) and GABAergic interneurons in the cortex (Stefansson et al., 2002;Flames et al., 2004;Fisahn et al., 2009). Other animal models with ErbB4 deletion in brain region and/or cell-type-specific manner have also been studied. Central nervous system (CNS)-specific ErbB4 KO mice exhibit delayed postnatal motor development, and disrupted cue utilization during Morris water maze (Golub et al., 2004). CNS-specific double KO (dKO) mice lacking both ErbB2 and ErbB4 at early embryonic stages display sensorimotor gating deficits, which is accompanied by decreased spine density in the cortex and hippocampus (Barros et al., 2009). Interestingly, another CNS-specific ErbB4 homozygous null mice did not exhibit any motor function impairments or any other behavioral deficits (Thuret et al., 2004). Moreover, expression of dominant negative ErbB4 in oligodendrocytes and myelination Schwann cells from E15 result in thinning of the myelin sheath of the corpus callosum, altered oligodendrocyte morphology, increased number of cells expressing differentiated oligodendrocytes as well as dopaminergic abnormalities, which is predicted to result from defective myelin (Roy et al., 2007; Table 4).
It is worth noting that most studies seem to suggest a link between increased NRG1-ErbB signaling and decreased NMDAR function in the PFC. Bath application of NRG1 reduces NMDAR-mediated currents in PFC pyramidal neurons (Gu et al., 2005). A postmortem tissue-stimulation study also reported that NRG1 stimulation suppresses NMDAR activation in the human PFC (Hahn et al., 2006). Conversely, blockade of NMDARs via chronic MK-801 administration in rats increases the expression of NRG1 and ErbB4 proteins (Feng et al., 2010). Moreover, others have found that NRG1-ErbB4 signaling may cause NMDAR hypofunction by either suppressing Src-mediated enhancement of synaptic NMDAR function or through upregulation of excitatory amino-acid carrier 1 (EAAC1) (Pitcher et al., 2011;Yu et al., 2015). Given that most of the animal models reviewed here were based on deficiency of the NRG1-ErbB4 signaling, caution must be exercised in interpreting these results in the context of NMDAR hypofunctionality.
Based on previous studies, 14-3-3 proteins play a crucial role in surface expression of NR2C subunit in cerebellar granule cells (Chen and Roche, 2009), and 14-3-3 family members are found to associate and colocalize with NMDARs (Taya et al., 2007).
We have generated the 14-3-3 functional KO (FKO) mice by transgenic expression of a dimeric fourteen-three-three peptide inhibitor (difopein), which antagonizes the binding of 14-3-3 proteins to their endogenous partners in an isoformindependent manner to disrupt 14-3-3 functions in the brain (Wang et al., 1999;Masters and Fu, 2001;Cao et al., 2010;Foote et al., 2015). Based on our behavioral analyses, 14-3-3 FKO mice exhibit a full range of endophenotypes associated with core symptoms of schizophrenia. These behavioral changes include hyperlocomotor activity, social interaction impairments, deficits in sensorimotor gating, working memory, and associative learning and memory. In 14-3-3 FKO mice, we also identified a significant reduction in the NMDAR-mediated synaptic currents in CA1 pyramidal neurons. The protein levels of NMDA receptors, particularly NR1 and NR2A subunits, are selectively reduced in the PSD fraction of 14-3-3 FKO mice. These observations thus provide in vivo evidence linking 14-3-3 dysfunction to NMDAR hypofunction in forebrain neurons (Qiao et al., 2014;Foote et al., 2015). In a recent study, we reported that AAV-mediated expression of the 14-3-3 inhibitor specifically within the hippocampus alone is sufficient to induce several behavioral deficits including hyperactivity, impaired associative learning and memory, and reduced sensorimotor gating. Conversely, selectively restoring the function of 14-3-3 in the forebrain of the 14-3-3 FKO mice attenuate hyperlocomotor activity of the 14-3-3 FKO mice. In addition, we show that postsynaptic NMDA receptor levels are regulated by acute 14-3-3 manipulations (Graham et al., 2019). Taken together, findings from these animal model studies directly link 14-3-3 inhibition and NMDAR hypofunction in specific forebrain regions to certain schizophrenia-associated behavioral deficits.
Previous genetic and postmortem studies have identified 14-3-3ζ (YWHAZ) and 14-3-3ε (YWHAE) as candidate risk genes for schizophrenia (Middleton et al., 2005;Wong et al., 2005;Ikeda et al., 2008). This led to animal model studies to determine whether the loss of 14-3-3ζ or 14-3-3ε isoform leads to behavioral changes related to symptoms of schizophrenia. The 14-3-3ζ KO mice display hyperlocomotor activity, sensorimotor gating and cognitive deficits. These mice also display alterations in synaptic neurotransmission and plasticity, abnormal neuronal migration and axonal guidance defects in the hippocampus, and increased levels of DA in the striatum (Cheah et al., 2012;Ramshaw et al., 2013;Xu et al., 2015). As 14-3-3ε KO mice are found to be neonatally lethal, 14-3-3ε heterozygous mice are studied for behavioral deficits related to schizophrenia. 14-3-3ε heterozygous mice display weak defects in working memory and enhanced anxiety-like behavior (Ikeda et al., 2008;Toyo-oka et al., 2014). Taken together, these results suggest a link between these 14-3-3 isoforms to schizophrenia. Further studies are required to understand whether NMDAR hypofunctionality may be mediating behavioral deficits associated with schizophrenia exhibited in 14-3-3 isoform-specific animal models (Table 4).

CONCLUSION
Here we reviewed various animal models of NMDAR hypofunctionality created through pharmacological and genetic approaches. Animal models of schizophrenia have also been generated by using dopaminergic agonists, such as amphetamine and apomorphine, which primarily induce behavioral changes limited to hyperlocomotor activity, stereotypy, and sensorimotor gating deficits (Kokkinidis and Anisman, 1980;Sharp et al., 1987;Swerdlow et al., 1994;Swerdlow and Geyer, 1998;Marcotte et al., 2001;Jones et al., 2011). Similarly, direct infusion of GABA A receptor antagonist picrotoxin into the mPFC only causes sensorimotor gating deficits in rats (Japha and Koch, 1999;Marcotte et al., 2001). In comparison, as reviewed in this article, NMDAR antagonists induce a full range of behaviors corresponding to symptoms of schizophrenia, with robust changes in negative and cognitive symptoms-related behaviors. Moreover, some behavioral deficits induced by NMDAR antagonists are reversed by administration of APDs, including haloperidol and clozapine (Irifune et al., 1991;Noda et al., 1995;Verma and Moghaddam, 1996;Sams-Dodd, 1999;Abdul-Monim et al., 2006;Szlachta et al., 2017). Thus, administration of NMDAR antagonists provides a relatively valid animal model for studying symptomologies and pathophysiology of schizophrenia.
However, there are limitations in the use of NMDAR antagonists to model NMDAR hypofunctionality. First, we must ask whether pharmacologically induced alterations in the brain are a true representation of dysfunctional neural circuitry in schizophrenic patients. Olney et al. argued that since NMDAR antagonists cause NMDAR hypofunctionality throughout the brain, these animal models cannot target NMDARs in a neural circuit specific manner, thus may fail to precisely recapitulate schizophrenia pathophysiology (Olney et al., 1999). Second, schizophrenia is considered a neurodevelopmental disorder with deficits occurring during early brain development, yet a majority of animal studies have been focused on NMDAR antagonistinduced changes in adulthood (Harrison and Weinberger, 2005;Rapoport et al., 2005;Fatemi and Folsom, 2009;Powell, 2010). Further studies should be extended to address schizophrenia as a neurodevelopmental disorder by administering NMDAR antagonists at either prenatal or postnatal time period and examine their impacts on behavior and neural circuitry during development. Third, it is debatable whether some of the druginduced behavioral changes in rodents are direct or even true correspondence of symptoms exhibited by schizophrenia patients. For example, hyperlocomotor activity measured in the open field test may have low relevance to psychosis in human patients (Powell and Miyakawa, 2006;Moore, 2010). This should contribute to limitations in interpretation of results obtained from these animal models with currently used behavioral paradigms (Moore, 2010). Fourth, behavioral and neurophysiological phenotypes exhibited in animal models of NMDAR antagonists are not specifically corresponding to symptoms of schizophrenia. For example, subchronic ketamine treatment in rodents leads to increased aggression, which is not a schizophrenia-related behavioral phenotype (Becker et al., 2003). Other behavioral phenotypes in pharmacological models, such as social interaction deficits, hyper responsiveness to sensory stimuli as well as cognitive deficits, overlap with symptoms associated with autism spectrum disorder (ASD) (Crawley, 1999(Crawley, , 2012. Finally, NMDAR antagonists cause neurotoxicity, evidenced by neuronal vacuolization, neuronal necrosis, and other cytotoxic changes. This must be taken into consideration when interpreting results from those animal studies, especially for chronic models of NMDAR antagonist (Olney et al., 1989(Olney et al., , 1991Nakao et al., 2003). Neurotoxicity should also be considered by the researcher when deciding on the dosage, regimen, and the type of NMDAR antagonist proposed to be used depending on the aim of the study.
Compared to pharmacological models, genetic animal models have several advantages for studying NMDAR hypofunctionality in relation to schizophrenia. First, schizophrenia is a disease of common symptomatology with etiological heterogeneity caused by various genetic and environmental factors (Takahashi, 2013). Human genome wide association studies (GWAS) have identified NMDAR genes or genes encoding proteins that regulate NMDARs to be candidate risk genes for schizophrenia Purcell et al., 2014; Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014; Harrison, 2015). There is also evidence of reduced expression of some NMDAR-subunits in schizophrenia patients (Balu, 2016). Thus, genetic animal models of NMDAR hypofunction may have certain construct validity for schizophrenia. However, we must emphasize that NMDAR hypofunction itself does not guarantee the animal model of schizophrenia, since NMDAR hypofunction leads to different neurological and psychiatric conditions, which do not necessarily correspond to the symptoms of schizophrenia. Second, de novo mutations in schizophrenia patients are overrepresented in glutamatergic postsynaptic proteins comprising of NMDAR complexes, making genetic animal models particularly useful in dissecting the signaling pathways and molecular mechanisms leading to NMDAR hypofunction (Forrest et al., 1994;Fromer et al., 2014). As reviewed here, some of the genetic animal studies, such as NRG1/ErbB4 and 14-3-3 animal models, provided evidence to support NMDAR hypofunctionality as a potential convergence point for symptoms of schizophrenia (Snyder and Gao, 2013). Third, various genetic models have been created to either directly or indirectly downregulate NMDAR functions in a spatial and temporal controlled manner. Some of these models specifically examine the role of NMDAR functions in different cell types and/or brain regions, while others have studied the impact of NMDAR hypofunction at different neurodevelopment stages. These studies may help reveal "how" genetic and environmental factors (signaling pathways) lead to NMDAR hypofunctionality and behavioral deficits, as well as "where" (in which neural circuits) and 'when' these changes occur during neurodevelopment. Such information will be crucial for our understanding of the disease progress. In the future, some of the animal models may provide us useful platforms for identifying novel therapeutics for schizophrenia patients.

AUTHOR CONTRIBUTIONS
GL was responsible for drafting, writing, reviewing, and editing this manuscript. YZ was responsible for writing, reviewing, and editing the manuscript with suggestive ideas on formatting and outlining this manuscript.

FUNDING
This work was supported by a research grant from NIMH (MH115188A).