Glutamatergic regulation of cognition and functional brain connectivity: insights from pharmacological, genetic and translational schizophrenia research

Abstract

The pharmacological modulation of glutamatergic neurotransmission to improve cognitive function has been a focus of intensive research, particularly in relation to the cognitive deficits seen in schizophrenia. Despite this effort, there has been little success in the clinical use of glutamatergic compounds as procognitive drugs. Here, we review a selection of the drugs used to modulate glutamatergic signalling and how they impact on cognitive function in rodents and humans. We highlight how glutamatergic dysfunction, and NMDA receptor hypofunction in particular, is a key mechanism contributing to the cognitive deficits observed in schizophrenia and outline some of the glutamatergic targets that have been tested as putative procognitive targets for this disorder. Using translational research in this area as a leading exemplar, namely, models of NMDA receptor hypofunction, we discuss how the study of functional brain network connectivity can provide new insight into how the glutamatergic system impacts on cognitive function. Future studies characterizing functional brain network connectivity will increase our understanding of how glutamatergic compounds regulate cognition and could contribute to the future success of glutamatergic drug validation.

Linked Articles

This article is part of a themed section on Pharmacology of Cognition: a Panacea for Neuropsychiatric Disease? To view the other articles in this section visit http://onlinelibrary.wiley.com/doi/10.1111/bph.v174.19/issuetoc

Abbreviations

ASST

attentional set‐shifting task

5‐CSRTT

5‐choice serial reaction time task

BOLD

blood oxygen level‐dependent

CIQ

(3‐chlorophenyl) [3,4‐dihydro‐6,7‐dimethoxy‐1‐[(4‐methoxyphenoxy)methyl]‐2(1H)‐isoquinolinyl]methanone

CPT

continuous performance task

CX516

6‐[(piperidin‐1‐yl)carbonyl]quinoxaline

DLPFC

dorsolateral prefrontal cortex

EST

patients with established schizophrenia

FC

functional connectivity

FES

first episode schizophrenia

fMRI

functional MRI

HC

healthy control

LC

locus coeruleus

LY2140023

(−)‐(1R,4S,5S,6S)‐4‐amino‐2‐sulfonylbicyclo[3.1.0]hexane‐4,6‐dicarboxylic acid

mGlu recepor

metabotropic glutamate receptor

MMN

mismatch negativity

NAM

negative allosteric modulation

NOR

novel object recognition

Org 25935

2‐([(1R,2S)‐6‐methoxy‐1‐phenyl‐1,2,3,4‐tetrahydronaphthalen‐2‐yl]methyl‐methylamino)acetic acid

PAM

positive allosteric modulator

PCP

phencyclidine

PFC

prefrontal cortex

Ro 25‐6981

(αR,βS)‐α‐(4‐hydroxyphenyl)‐β‐methyl‐4‐(phenylmethyl)‐1‐piperidinepropanol maleate

RT

response time

SAR218645

(S)‐2‐(1,1‐dimethyl‐indan‐5‐yloxymethyl)‐2,3‐dihydro‐oxazolo[3,2‐a]pyrimidin‐7‐one

SZ

schizophrenia

TUNL

trial‐unique, delayed nonmatching to location

The glutamatergic system

Glutamatergic synapses in the CNS, responsible for fast excitatory neurotransmission, play a critical role in a broad range of cognitive functions. The structure of glutamate synapses and the molecular mechanisms underlying glutamatergic neurotransmission have previously been reviewed by others in detail (Sanz‐Clemente et al., 2013; Sudhof, 2013; Volk et al., 2015). In mature glutamatergic synapses, a vast array of proteins are involved in the packaging of glutamate into synaptic vesicles, the localization of these vesicles to presynaptic active zones and the docking and release of the contents of these vesicles into the synaptic cleft (Sudhof, 2013). Post‐synaptically, glutamate acts at both ionotropic glutamate receptors (Glu receptors), including the α‐amino‐3‐hydroxy‐5‐methyl‐4‐isoxazolepropionic acid (AMPA), kainate and NMDA receptors, and metabotropic G‐protein coupled glutamate (mGlu) receptors to cause depolarisation in the post‐synaptic neuron. mGlu receptors are also present on the synaptic bouton that play an important role in the regulation of glutamate release as autoreceptors (see Niswender and Conn, 2010 for review). The synaptic concentration of glutamate is also regulated by glutamate uptake, into both neurons and glial cells, mediated by a range of glutamate transporters (for review, see Vandenberg and Ryan, 2013) and the cystine/glutamate antiporter (System Xc, Bridges et al., 2012). Much research has been dedicated to elucidate the roles of these molecular components in the regulation of cognitive and brain function, in part due to the proposed central involvement of glutamate system dysfunction in a broad range of brain disorders with prominent cognitive deficits including schizophrenia (SZ) (Coyle, 2006), bipolar disorder (McCloud et al., 2015), major depressive disorder (deWilde et al., 2015), autism spectrum disorders (Volk et al., 2015) and Alzheimer's disease (Lin et al., 2014).

The role of the NMDA receptor in cognition

NMDA receptors are tetrameric structures assembled from two obligatory GluN1 subunits (formerly NR1) and two GluN2 (GluN2A and GluN2D, formerly NR2A to NR2D) subunits. NMDA receptors may also contain GluN3 subunits, which are particularly abundant during early life and appear to have a role in limiting synapse maturation. The persistence of GluN3A‐containing NMDA receptors into adulthood may contribute to the synaptic dysfunction in psychiatric disorders (Perez‐Otano et al., 2016), while the potential role of GluN3B is yet to be elucidated.

Differences in NMDA receptor subunit composition results in functional and pharmacological diversity, as exemplified by the differing pharmacology of GluN2A‐ versus GluN2B‐containing NMDA receptors (Smith et al., 2011). The contribution of the different NMDA receptor subtypes to cognition is relatively poorly defined. However, recent studies in genetically modified mice have proved useful in further elucidating the complex relationship that exists between NMDA receptors with specific subunit compositions, their cellular and brain region localization and distinct cognitive functions (Table 1). In addition to these studies a multitude of pharmacological studies conducted in rodents, non‐human primates and human participants have characterized the role of the NMDA receptor in cognition. Here, we briefly review the general insights gained from these pharmacological studies.

Table 1.

Cognitive deficits reported in genetic mouse models targeting the NMDA receptor

NMDA receptor gene	Mouse model	Cognitive deficit	Reference
GluN1	GluN1 hypomorphic mice	Impaired spontaneous alternation	Gregory et al., 2013
	GluN1 hypomorphic mice	Reduced spontaneous alternation	Barkus et al., 2012
		Impaired short‐term object recognition memory
		Impaired spatial reference memory
		Impaired learning in a visual discrimination task
	Ablation of GluN1 in cortical excitatory neurons of mPFC and SSCTX	Impaired short‐term object recognition	Rompala et al., 2013
		Normal spatial working memory
	Ablation of GluN1 in HP (DG and CA1)	Impaired in spatial reversal learning (water maze)	Taylor et al., 2014
		Not impaired in spatial discrimination (water maze)
	GluN1 knockout in Parvalbumin expressing interneurons	No deficit in cognitive flexibility, working memory or attentional processing	Bygrave et al., 2016
GluN2A	GluN2A KO mice	Impaired spatial working memory	Bannerman et al., 2008
		No impairment in long term spatial reference memory
	GluN2A KO mice	Impaired extra‐dimensional set shifting	Marquardt et al., 2014
		Not impaired in discrimination or reversal learning
	GluN2A KO in HP and CTX	Impaired reversal learning	Thompson et al., 2015
GluN2B	GluN2B KO in principal neurons of the postnatal forebrain	Impairment in spatial working memory, spatial reference memory, impaired recognition memory, performance deficits in simple Morris water maze and visual discrimination tasks	von Engelhardt et al., 2008
	GluN2B KO in HP (CA1 and DG)	Impaired spatial working memory and reversal learning	von Engelhardt et al., 2008
		No impairment in spatial reference memory
GluN2C	GluN2C KO mice	Deficit in fear conditioning and spatial working memory	Hillman et al., 2011
		No impairment in spatial reference memory
GluN2D	GluN2D KO mice	Impaired social memory	Yamamoto et al., 2017
		No impairment in novel object recognition
GluN3A	GluN3A KO mice	Improved spatial learning and enhanced object recognition memory	Mohamad et al., 2013
GluN3B	GluN3B KO mice	No difference in spatial reference memory and fear conditioning	Niemann et al., 2007

CA1, cornu ammonis 1 subfield; CTX, cortex; DG, dentate gyrus; HP, hippocampus; KO, knock‐out; mPFC, medial prefrontal cortex; SSCTX, somatosensory cortex.

Evidence from pharmacological studies in rodents

The impact of acute NMDA receptor antagonist administration

Acute administration of NMDA receptor antagonists, such as ketamine, phencyclidine (PCP) and dizocilpine, has been shown to negatively impact on domains of executive function in rodents, impairing cognitive flexibility (de Bruin et al., 2013; Gastambide et al., 2013) and disrupting attentional processing (Amitai and Markou, 2010; Barnes et al., 2016; Thomson et al., 2011). Acute NMDA receptor antagonist administration also negatively affects other cognitive domains impairing spatial reference learning and memory (for review, see Morris, 2013; Duan et al., 2013; Ihalainen et al., 2016), short‐term object recognition memory (Cloke and Winters, 2015; Rajagopal et al., 2016), associative memory (as assessed in the paired associates learning task, Kumar et al., 2015; Lins et al., 2015) and episodic learning and memory (Bast et al., 2005) in rodents. Acute NMDA receptor antagonist administration can also induce motivation deficits and motor impairments, which could potentially confound some of these cognitive measures (Noda et al., 2000). However, the impact of these drugs on cognitive performance can be assessed at doses and time points after administration where these confounding effects are absent. A key consideration in these acute NMDA receptor antagonist studies, as with all pharmacological studies involving drugs targeting the glutamatergic system, is the importance of the temporal effects of each compound. These effects may contribute to some of the different behavioural and neural effects observed between different studies. Another concern is the non‐selective pharmacology of the compounds used. For example, PCP also acts at nicotinic acetylcholine receptors (Fryer and Lukas, 1999) and dopamine (D₂) receptors (Seeman et al., 2009), while ketamine binds to a wide range of non‐glutamatergic targets (for review, see Mion and Villevielle, 2013). Thus, genetic studies and studies using more pharmacologically selective compounds have been instrumental in further supporting a key role for the NMDA receptor in cognition.

The impact of prolonged NMDA receptor hypofunction

A multitude of studies have characterized the effects of prolonged NMDA receptor hypofunction (induced by repeated, intermittent NMDA receptor antagonist treatment) on various cognitive functions in rodents. Cognitive testing in these studies is usually undertaken when animals are not experiencing the acute effects of these antagonists (i.e. when ‘drug free’). Thus, the cognitive deficits present are thought to result from the changes in plasticity that occur in the brain as a result of prolonged NMDA receptor hypofunction. These effects on plasticity include modifications in the function of non‐glutamatergic neurotransmitter systems (Jentsch et al., 1998; Lindefors et al., 1997), including the function of parvalbumin‐positive (PV+) GABAergic interneurons (Bygrave et al., 2016), changes in synaptic plasticity (Nomura et al., 2016), alterations in regional neuronal activity, including prefrontal cortex (PFC) hypofunction (Dawson et al., 2012), and altered brain network connectivity (see later discussion). In these studies, prolonged NMDA receptor antagonism has been shown to induce deficits in cognitive flexibility (Dawson et al., 2012; McLean et al., 2012), attentional processing (Thomson et al., 2011; Barnes et al., 2016), spatial reference learning and memory (Didriksen et al., 2007), working memory (Seillier and Giuffrida, 2009) and short‐term object recognition memory (Pyndt Jorgensen et al., 2015; Horiguchi et al., 2013; Rajagopal et al., 2016). While the translational relevance of some of these behavioural tests to aspects of human cognition is questionable (Kas et al., 2014; Pratt et al., 2012; Pryce and Seifritz, 2011), the overall findings indicate a central role for the NMDA receptor in a broad range of cognitive processes.

In addition to these studies, conducted in adult animals, the impact of pharmacologically‐induced NMDA receptor hypofunction at specific developmental time points (either in utero, during early postnatal development or during adolescence) on cognitive function in the fully developed animal has also been assessed (Broberg et al., 2008; Li et al., 2011; Zhao et al., 2014). These studies highlight the neurodevelopmental role of NMDA receptor activity, at defined epochs of brain development, in ‘setting up’ the brain for effective cognitive function. This area of research certainly warrants further systematic investigation.

Insights from studies using NMDA receptor subtype selective drugs

Pharmacological studies have also attempted to elucidate the role of specific NMDA receptor subtypes in cognition. For example, the distinct pharmacology of GluN2A versus GluN2B‐containing NMDA receptors has allowed the recent characterization of the role of these receptor subtypes in different cognitive functions. The GluN2A‐selective antagonist NVP‐AMM077 has recently been shown to decrease accuracy in a task assessing sustained attention (5‐choice serial reaction time task, 5‐CSRTT; Smith et al., 2011), but has little effect on location discrimination, paired‐associate learning and working memory (as assessed using the trial‐unique, delayed nonmatching to location (TUNL) task; Kumar et al., 2015). In contrast, GluN2B‐selective antagonists, such as Ro 25‐6981 and traxoprodil, appear to improve accuracy and processing speed in the 5‐CSRTT (Higgins et al., 2005; Smith et al., 2011). However, antagonism of GluN2B using ifenprodil has been shown to impair performance in response time (RT) in the 5‐CSRTT (Higgins et al., 2005). This conflicting finding, compared with the effects of other GluN2B antagonists, may be due to non‐selective actions of ifenprodil or its relative weak affinity for GluN2B‐containing NMDA receptors. Further NMDA receptor subtype‐specific effects are supported by the observation that traxoprodil administration impairs location discrimination but not working memory in the TUNL task (Kumar et al., 2015). In addition, Ro 25‐6981 ameliorates the effect of ketamine treatment on cognitive flexibility (assessed using the attentional set‐shifting task (ASST), Kos et al., 2011), supporting a primary role for GluN2B‐containing NMDA receptors in the effect of ketamine on cognitive flexibility.

There are a lack of studies reporting the cognitive impact of GluN2C and GluN2D‐selective compounds. While GluN2C/D‐selective compounds, such as the inhibitor DQP‐1105 (Acker et al., 2011), are available, their effects on cognition have not yet been tested. However, evidence for the role of GluN2C/D‐containing NMDA receptors in cognitive function is supported by studies conducted in genetically modified mice (Table 1). In addition, GluN2C/D‐containing NMDA receptor play an important role in the effects of the NMDA receptor antagonists ketamine and memantine (Kotermanski and Johnson, 2009), which may include their effects on cognitive functions. In addition, the absence of either GluN2C or GluN2D receptors has been shown to have different effects on the cortical oscillations induced by NMDA receptor blockade (Gupta et al., 2016; Sapkota et al., 2016). Other evidence supporting a role for GluN2C/D‐containing NMDA receptors in cognitive function comes from studies using positive allosteric modulators (PAMs), such as CIQ and D‐cycloserine, discussed later in this review.

Insights from studies in genetically modified mice

A range of studies have used genetically modified mice to further determine the role of the different NMDA receptor subtypes in a range of cognitive functions (Table 1). These studies are often able to extend the understanding gained from relevant pharmacological studies, in part due to the greater assurance of NMDA receptor subtype specificity, but also by targeting either specific brain subsystems (GluN1; Rompala et al., 2013; Taylor et al., 2014) or cell populations (GluN1 in PV+ interneurons; Bygrave et al., 2016). A limitation of these genetic studies is the neurodevelopmental role of NMDA receptors in the development of effective cognitive function, as highlighted by the persistent effects of non‐selective NMDA receptor antagonists when selectively administered at specific developmental time points (Broberg et al., 2008; Li et al., 2011; Zhao et al., 2014). Thus, the observed effects may be very different from those elicited by acute pharmacological regulation of these receptor subtypes. Nevertheless, studies using both genetic and pharmacological approaches offer complementary strategies to further elucidate the role of specific NMDA receptors subtypes in cognition, with modern genetic approaches offering new levels of granular neural system, temporal and cell‐type resolution.

The NMDA receptor hypofunction hypothesis of schizophrenia

The ‘glutamate hypothesis of SZ’ posits that a dysfunctional glutamatergic system is a key pathophysiological mechanism contributing to the clinical symptoms seen in patients with SZ (Luby et al., 1959; Carlsson et al., 2000; Farber et al., 2002; Javitt, 2007; Javitt et al., 2012). The ‘NMDA receptor hypofunction hypothesis of SZ’, a more specific theory of the glutamate dysfunction hypothesis, has its origins in the observation that acute administration of NMDA receptor antagonists, such as ketamine and PCP, induces psychotic‐like symptoms (delusions and hallucinations) in healthy controls (HCs) that are similar to those seen in patients with SZ (Krystal et al., 1994; Abi‐Saab et al., 1998). In addition, individuals who chronically abuse PCP (presumably inducing repeated intermittent NMDA receptor hypoactivity) show deficits in executive function similar to those seen in patients with SZ (Cosgrove and Newell, 1991). These observations have led to the development of two distinct but overlapping models of the glutamate hypothesis of SZ: (i) the ‘prolonged NMDA receptor hypofunction model’ and (ii) the ‘acute NMDA receptor hypofunction model’.

The ‘prolonged NMDA receptor hypofunction model’ postulates that prolonged hypoactivation of the NMDA receptor induces multiple pathological mechanisms involved in the disorder (Coyle, 2006; Coyle et al., 2010; Moghaddam and Krystal, 2012; Javitt et al., 2012) and that NMDA receptor hypofunction may be the final pathophysiological pathway for the positive, negative and cognitive symptoms experienced by patients with SZ (Carlsson et al., 1999; Goff and Coyle, 2001; Coyle, 2006; Javitt, 2010; Balu and Coyle, 2015). The ‘acute NMDA receptor hypofunction model’ originates from clinical trials when ketamine was administered to HCs. Early findings found that changes in glutamatergic signalling could explain the psychotomimetic effects of ketamine and PCP in terms of the positive symptoms present in individuals with first episode SZ (FES) or first episode psychosis (Krystal et al., 1994; Krystal et al., 1999; Khlestova et al., 2016). While the findings from the acute ketamine administration model have particularly increased our understanding of the positive and negative symptoms seen in patients with SZ, evidence for translationally relevant alterations in cognitive functions is more limited. We describe some of these exemplary cognitive studies below and outline their translational alignment to observations made in patients with SZ.

Evidence from pharmacological studies in human participants

In HCs, acute ketamine administration has been shown to significantly affect a range of cognitive functions, with the effects being similar to those seen in patients with SZ. Low‐dose ketamine administration (100 ng·mL⁻¹ of plasma) affects contingency learning in HCs when assessed using a probabilistic learning task (Vinckier et al., 2016), with ketamine administration inducing misleading cue‐outcome associations. These findings contrast with those reported in an early study where the same dose of ketamine (100 ng·mL⁻¹ of plasma) failed to alter task performance (Corlett et al., 2006). However, in both studies similar effects of ketamine on blood oxygen level‐dependent (BOLD) responses (increased) were observed in regions of the PFC of participants undertaking the task. Deficits in contingency learning as well as increased BOLD responses in the PFC have also been reported in patients with SZ (Diaconescu et al., 2011). However, decreased BOLD responses in the PFC of patients with SZ during contingency learning have also been reported (Dowd et al., 2016). In addition, increased BOLD responses in a range of other brain regions have also been reported in SZ patients undertaking this task (Diaconescu et al., 2011; Park et al., 2015; White et al., 2015) that are different from those seen in ketamine functional MRI (fMRI) studies in HCs. It is important to note that different temporal effects of drug administration may contribute to some of the diverse findings in behavioural performance and neural function reported in these clinical studies.

Ketamine administration has also been shown to significantly affect working and declarative memory in HCs. Ketamine administration significantly reduced accuracy in the continuous performance task (CPT) and impaired both immediate and delayed recall in the Hopkins verbal learning task (Krystal et al., 2005). The effect of ketamine on working memory function was confirmed by another study (Honey et al., 2008). The effects of ketamine on working and declarative memory tasks are similar to those seen in patients with SZ (Blokland et al., 2017; Green, 2016). However, the impact of ketamine on BOLD responses during working memory and declarative memory tasks in HCs is more difficult to corroborate with the fMRI findings seen between SZ patients and HC, where both increased and reduced BOLD responses are observed in patients with SZ when compared with HCs (Brown and Thompson, 2010; Dauvermann et al., 2014). While Anticevic et al. (2012) and Driesen et al. (2013) reported reduced BOLD responses in the dorsolateral PFC (DLPFC) and precuneus in HCs treated with ketamine and SZ patients during working memory performance, the findings by Honey et al. (2008) (increased BOLD responses in the basal ganglia and thalamus of HC treated with ketamine during the task) are more difficult to align with the observations made in patients with SZ.

Ketamine administration has also been shown to affect attentional processing, when assessed using both visual and auditory tasks. In HCs, ketamine was found to significantly reduce the response time (RT) to target stimuli in a visual oddball task assessing attentional processing (Watson et al., 2009), an effect that is similar to the reduced visual processing speed seen in patients with SZ (Urban et al., 2008). Similarly, in HCs, ketamine (0.24 mg·kg⁻¹) also affects attentional processing when assessed using an auditory processing task, increasing the number of false alarms during the task (Umbricht et al., 2000). In addition, in this study, ketamine administration was also found to reduce the peak amplitude of the mismatch negativity (MMN) signal, an aspect of the event‐related potential detected using EEG that is indicative of the arrival of an odd stimulus in a sequence of stimuli. Similar cognitive effects were independently observed when using both a low and high dose of ketamine, with only the higher dose inducing significant deficits in MMN (Heekeren et al., 2008). The deficits in auditory attentional processing and MMN seen during ketamine administration are similar to those reported in patients with SZ (Milovan et al., 2004). Summary outlines of these studies are provided in Table 2.

Table 2.

Impact of NMDA receptor antagonist administration on cognitive functions in human controls

Study (year)	Experimental group/patient group		Control group/control matching criteria		Study design/Drug administration	Task design for cognitive function	Main findings – effects of drug on cognition and/or neural response measures during cognition
	N (M : F)	Mean age in years (SD)	N (M : F)	Mean age in years (SD)
Contingency learning
Vinckier et al., 2016	N/A		HCs 21 (11:10) 28.7 (±3.2)		Double‐blind, placebo‐controlled, randomized, within‐subjects study design Two separate drug challenge sessions (placebo/active drug): Ketamine: low‐dose bolus and i.v. injection Placebo: Saline	fMRI: Probabilistic learning task – parametric modulation. GLM: Separation of categorical regressors for cue and outcome onsets GLM: Outcome onsets modulated by two computational variables (ROI analysis)	Effect of ketamine on cognitive performance: Ketamine decreases optimization of outcomes given misleading unexpected outcomes Effect of ketamine on BOLD response: Altered BOLD response in fronto‐parietal regions based on contingency learning, mostly in regions of the cerebellum, MFG, DLPFC and inferior parietal cortex in ketamine when compared to placebo (ROI analysis)
Corlett et al., 2006	N/A		HCs 15 (8:7) 29 (±7)		Double‐blind, placebo‐controlled, randomized, within‐subjects study design Two separate drug challenge sessions (placebo/active drug): Ketamine: low‐dose bolus and i.v. Injection while in scanner Ketamine: high‐dose bolus and i.v. injection outside of Placebo: Saline	fMRI: Associative learning task. Associative relationships Prediction error	Effect of ketamine on cognitive performance: No difference in behavioural performance between ketamine and placebo Effect of ketamine on BOLD response: Increased BOLD response in the right PFC in response to expected stimuli in ketamine when compared to placebo (prediction error)
Working memory and declarative memory
Krystal et al., 2005	N/A		HCs – Amphetamine group (14 study completers) HCs – Ketamine group (13 study completers) 27 (16:11) 16 Male 33(±8.9) 11 Female 28(±5.2)		Double‐blind, randomized, placebo‐controlled, study design Amphetamine group: 1) Infusion of 0.25 mg·kg−1 followed by saline, 2) Ketamine infusion of 0.23 mg·kg−1, 3) Amphetamine placebo (saline), 4) Ketamine, 5) Placebo amphetamine, 6) Placebo ketamine. Ketamine group: Same idea as above but swapped between amphetamine and ketamine	CPT HVLT (6 versions) PANSS: Cognitive symptom score.	Effect of ketamine on cognitive performance: Significant effect in accuracy on CPT for ketamine but not for amphetamine. For HVLT immediate recall, there was a significant interactive effect of ketamine with amphetamine when compared to placebo. 2. For HVLT delayed recall, there was a significant interaction between ketamine and amphetamine.
Honey et al., 2008	N/A		HCs 15 (8:7) 29 (±7)		Double‐blind, placebo‐controlled, randomized, within‐subjects study design Two separate drug challenge sessions (placebo/active drug): Ketamine: low‐dose bolus and i.v. Injection while in scanner Ketamine: high‐dose bolus and i.v. injection outside of Placebo: Saline	fMRI: Working memory (N‐Back task). Button presses for targets and distractors. CPT (Button press for targets but not for distractors) Sentence completion task (Button press for task completion) Verbal self‐monitoring task (Sub‐vocalization during sentence completion and incompletion)	Effect of ketamine on cognitive performance: Significant effect of ketamine for RT compared to placebo. Significant effect of ketamine for RT compared to placebo. Effect of ketamine on BOLD response; Increased BOLD response of basal ganglia and thalamus after ketamine across all working memory load conditions with a strong effect for 2‐Back (ROI analysis) No effect of ketamine observed. No effect of ketamine observed. No effect of ketamine observed.
Anticevic et al., 2012	N/A		HC – Ketamine group 19 (10:9) 27.5 (±6.3)		Double‐blind, placebo‐controlled, randomized, within‐subjects study design Three separate drug challenge sessions (placebo/active drug): Ketamine: i.v. via initial bolus 0.23 mg·kg−1 over 1 min, followed by subsequent infusion (0.58 mg·kg−1 over 1 h) Placebo: 1 saline injection	fMRI (FC): Delayed spatial working memory task	Effect of ketamine on cognitive performance: Decreased accuracy for working memory versus control trials under ketamine Effect of ketamine on BOLD response; Ketamine attenuated task‐based activations of the DLPFC and precuneus for the working memory task and task‐based deactivations for the DMN. Effect of ketamine on FC; Seed‐based FC for the seed regions as part of the fronto‐parietal and the DMN: Significant modulation of the task‐based FC (delay part of the working memory task) and DMN under ketamine
Driesen et al., 2013	N/A		HCs – Ketamine group 22 (14:8) not reported		Double‐blind, placebo‐controlled, randomized, within‐subjects study design Three separate drug challenge sessions (placebo/active drug): Ketamine: i.v. via initial bolus 0.23 mg·kg−1 over 1 min, followed by subsequent infusion (0.58 mg·kg−1 over 1 h) Placebo: 1 saline injection	fMRI (FC): Spatial ‘2‐Back’ and ‘4‐Back’ conditions. Seed‐based cross‐correlation Global‐based connectivity	Effect of ketamine on cognitive performance: Decreased accuracy for working memory versus control trials under ketamine Effect of ketamine on BOLD response; Ketamine attenuated task‐based activations of the DLPFC and precuneus for the working memory task and task‐based deactivations for the DMN. Effect of ketamine on FC; Decreased FC between right DLPFC and MFG, IFG under ketamine when compared to placebo Decreased FC within the left DLPFC
Braun et al., 2016	N/A		HCs – Dextromethorphan group 37 (30:7) 25.3(±4.2)		Double‐blind, placebo‐controlled, randomized, cross‐over study design Two separate drug challenge sessions (placebo/active drug): Dextromethorphan: 120 mg in capsule form Placebo: capsule	fMRI (FC): N‐Back working memory task (0‐Back and 2‐Back conditions), Button presses for the target stimuli	Effect of dextromethorphan on cognitive performance: No differences for accuracy or RT in the working memory task between dextromethorphan versus placebo Effect of dextromethorphan on FC; FC for 270 regions in terms of network flexibility: Increased network flexibility under dextromethorphan when compared to placebo
Visual attentional processing
Watson et al., 2009	N/A		HC 23 (15:8) 24.55 (±2.59)		Double‐blind, placebo‐controlled study design Three separate drug challenge sessions: Saline (placebo) Ketamine (0.23 mg·kg−1 and infusion rate: 0.58 mg·kg−1·h−1) Thiopental (1.5 mg·kg−1 and infusion rate: 40mcg·kg−1·h−1)	EEG/ERP: 3‐stimulus visual oddball task	Effect of ketamine on thiopental on cognitive performance: Decreased target RT after thiopental and ketamine when compared to placebo; stronger effect in thiopental Effect of ketamine and thiopental on ERPs: Decreased target P3b amplitude at electrode Pz after ketamine versus placebo Decreased target P3b amplitude at electrode Pz after thiopental versus placebo No difference in target P3b amplitude between ketamine and thiopental Significant correlations between changes in P3b amplitude and target RT after thiopental but not ketamine at electrode Pz
Auditory attentional processing
Umbricht et al., 2000	N/A		HCs 20 (14:6) 24.6(±2.9)		Double‐blind, placebo‐controlled study design Two separate drug challenge sessions (placebo/active drug): Ketamine: (0.24 mg·kg−1 and infusion rate: 0.0 mg·kg−1·h−1) Placebo: Physiological sodium chloride solution and 5% glucose	EEG/ERP Visual AXE‐CPT during the auditory test paradigm (MMN)	Effects of ketamine on cognitive performance: Decreased correct detection of hits after ketamine administration compared to baseline Increased false alarms after ketamine administration compared to baseline and placebo conditions Effects of ketamine on auditory ERPs: Decreased peak amplitudes of MMN after ketamine in pitch‐deviance condition and duration‐deviance condition compared to baseline condition and placebo condition respectively. Increased N1 peak amplitude after ketamine administration compared to placebo
Heekeren et al., 2008	N/A		HCs 15 (9:6) 38 (not reported)		Randomized, double‐blind, cross‐over study design Low and high dose of the 5HT2A agonist DMT Low and high dose of S‐ketamine	EEG/ERP Visual AXE‐CPT during the auditory test paradigm (MMN)	Effects of ketamine on cognitive performance: Low‐doses and high‐doses of DMT and S‐ketamine impair behavioural performance during the AXE‐CPT Effects of ketamine on auditory ERPs: Decrease in generation of MMN after S‐ketamine > DMT Trend decrease in the frequency‐deviant‐induced MMN at electrode Fz after low‐dose S‐ketamine Decreased duration‐deviant MMN at electrodes Fz, F3, F4 after low‐dose and high‐dose S‐ketamine

BOLD, blood oxygen level‐dependent; CPT, continuous performance task; DLPFC, dorsolateral prefrontal cortex; DMN, default‐mode network; DMT, dimethyltryptamine; EEG, electroenchephalogram; ERP, event‐related potential; FC, functional connectivity; fMRI, functional magnetic resonance imaging; HVLT, Hopkins verbal learning test; IFG, inferior frontal gyrus; i.v., intravenous; MFG, middle frontal gyrus; N/A, not applicable; PANSS, positive and negative symptoms scale; ROI, region of interest; RT, response time.

The impact of NMDA receptor co‐agonists and partial agonists on cognition

Much research has been dedicated to elucidate the procognitive potential of activating NMDA receptors, with positive modulation rather than agonism (which risks inducing excitotoxicity) being a key area of research. The glycine site on the NMDA receptor provides an attractive drug target because of its positive modulatory effects on NMDA receptor signalling. The amino acid derivatives D‐serine and D‐cycloserine act as partial agonists at this site (Mothet et al., 2000; Watson et al., 1990). The therapeutic potential of NMDA receptor‐positive modulators in SZ patients has been reviewed previously (Kantrowitz and Javitt, 2010; Balu and Coyle, 2015; Goff, 2012). Here, we highlight and discuss recent findings of rodent and human studies featuring these drugs with a particular focus on cognition and, where applicable, summarize neuroimaging findings from rodent and human studies.

Pharmacological studies in rodents

In rodent studies, both D‐serine and D‐cycloserine have been shown to have procognitive effects. For example, D‐cycloserine improves short‐term object recognition, potentiates contextual and cued fear extinction learning in rats (Sugiyama et al., 2015; Walker et al., 2002) and improves spatial learning in aged rats (Baxter et al., 1994). In addition, intrahippocampal D‐cycloserine administration has been shown to reverse dizocilpine‐induced impairments in working memory performance in the radial arm maze (Kawabe et al., 1998), suggesting that D‐cycloserine administration can reduce the impact of acute NMDA receptor hypofunction on working memory. Similar effects have been shown for D‐serine, which improves working memory performance in the T‐maze alternation test and enhances novel object recognition, while also reversing the long‐term memory deficits induced by dizocilpine (Bado et al., 2011). In addition, the recently developed tetrapeptide rapastinel (also known as GLYX‐13), a partial agonist of the NMDA receptor glycine site, has also been shown to improve learning and memory in young and aged rats (Burgdorf et al., 2011) as well as restoring object recognition in mouse models of acute and prolonged NMDA receptor hypofunction (Rajagopal et al., 2016).

D‐cycloserine demonstrates greatest efficacy at NMDA receptors containing GluN2C subunits (Ogden et al., 2014), suggesting a central role for this NMDA receptor subtype in its procognitive effects. Potentiation of GluN2C/D‐containing NMDA receptors using CIQ has been shown to facilitate fear learning and extinction in mice (Ogden et al., 2014) and reverses the deficit in working memory (spontaneous alternation test) in mice following acute dizocilpine administration (Suryavanshi et al., 2014). A role for GluN2C subunit‐containing NMDA receptors in working memory is further supported by observations in GluN2C knockout mice (Hillman et al., 2011, Table 1). Interestingly, these mice do not show deficits in spatial reference memory, which contrasts with the ability of D‐cycloserine to improve spatial reference memory in aged rats (Baxter et al., 1994). This suggests that the activity of D‐cycloserine at other NMDA receptor subtypes may be more important for its effects on spatial reference memory or that developmental adaptations prevent the effect of GluN2C knockout on spatial reference memory in GluN2C knockout mice.

Evidence from pharmacological studies in humans

In contrast to the results of rodent studies, which support the procognitive potential of NMDA receptor partial agonism, the evidence from studies in humans, including studies in patients with SZ, is less persuasive. In HCs, glycine administration does not improve general cognitive performance on the CogState test battery (Neumeister et al., 2006) or in a visual attention task (O'Neill et al., 2011). Furthermore, D‐cycloserine administration does not improve motor sequence learning in HCs (Gunthner et al., 2016). In contrast, initial studies undertaken in a small sample of 12 HC males support a significant effect of the glycine transporter inhibitor Org 25935 on verbal learning and delayed recall, but not on any of the other cognitive tests employed (D'Souza et al., 2012). More recently, Org 25935 administration was also shown not to improve performance in a visuo‐spatial task, a working memory task or a verbal memory task in HCs (Christmas et al., 2014). This suggests that the procognitive potential of these compounds in HCs may be limited. Given that NMDA receptor function may be optimal in HCs, it is not surprising that these compounds fail to significantly improve cognitive performance in these studies. Despite these findings, one might still predict that these compounds would have procognitive potential in patients with brain disorders thought to involve NMDA receptor hypofunction, such as patients with SZ. However, studies investigating the procognitive potential of drugs that positively modulate NMDA receptor activity in patients with SZ are negative overall. For example, D‐cycloserine adjunctive treatment does not improve composite scores of general cognitive function or most individual cognitive domain scores when assessed using standardized neuropsychological batteries, in patients with established SZ (EST) (Buchanan et al., 2007; Goff et al., 2005; Goff et al., 2008; Weiser et al., 2012; Cain et al., 2014). D‐cycloserine treatment also fails to improve performance in the CPT and working memory tasks in patients with EST (Duncan et al., 2004). However, there are also positive findings where D‐cycloserine improved cognitive performance following a cognitive remediation programme in patients with EST (Cain et al., 2014) and D‐cycloserine has been shown to facilitate fear extinction therapies in people with anxiety disorders (Norberg et al., 2008). These findings suggest that D‐cycloserine may yet hold therapeutic value as it can potentiate the efficacy of cognitive behavioural therapies, at least in some cognitive domains. Despite this suggestion, overall findings from recent meta‐analysis do not support the procognitive efficacy of compounds potentiating NMDA receptor activity in SZ. While Tsai and Lin (2010) found a positive impact of NMDA receptor enhancing agents (D‐cycloserine, glycine and sarcosine) on cognitive symptoms in patients with SZ (assessed using the positive and negative symptoms scale cognitive subscale), two more recent meta‐analyses found no effect (Choi et al., 2013; Iwata et al., 2015). Choi et al. (2013) found that D‐cycloserine, D‐serine and the AMPA receptor PAM CX516, when used as adjunctive treatments, did not significantly improve function in five cognitive domains (Choi et al., 2013). Furthermore, Iwata et al. (2015) found that NMDA receptor glycine site drugs had no significant effect in eight cognitive domains. Thus, any procognitive effects of NMDA receptor glycine site modulators in patients with SZ are yet to be robustly established. One reason for the disparity between preclinical and clinical studies may be the testing of these compounds as adjunctive treatments in patients, as their procognitive efficacy has not been tested in the context of prolonged antipsychotic administration preclinically (a summary of the information from these studies is presented in Table 3). Furthermore, clinical efficacy has only been tested in patients with EST, and whether these drugs would be beneficial during earlier stages of the disease, such as in FES patients, has not yet been adequately tested. The temporal relevance of NMDA receptor hypofunction and thus the procognitive potential of enhancing NMDA receptor signalling over the time course of disease progression need to be much more clearly defined. Finally, the relatively short treatment duration used in some clinical trials, typically between 4 and 24 weeks (Choi et al., 2013) with one study at a more lengthy 36 weeks (Iwata et al., 2015; Table 3) may also contribute to the overall negative findings.

Table 3.

Impact of NMDA receptor coagonists and partial agonists on cognition

Study (year)	Experimental group/patient group phase of SZ		Control group/control matching criteria		Study design/drug administration	Task design for cognitive function	Main findings – effects of drug on cognition and/or neural response measures during cognition
	N (M : F)	Mean age in years (SD)	N (M : F)	Mean age in years (SD)
General cognition
Goff et al., 2005	EST‐ D‐cycloserine group (adjunctive treatment) 27 (24:3) 45.9(±7.4)   EST – placebo group 28 (20:8) 47.0(±8.6)		N/A		Randomized double‐blind, placebo‐controlled, parallel group study design 6 month trial D‐Cycloserine: 50 mg capsule at bedtime in adjunct with conventional antipsychotic medication Placebo: capsule at bedtime in adjunct with conventional antipsychotic medication	Cognitive battery: CVLT IQ estimation (Vocabulary, Information, Digit Span and Block Design) ANART Stroop Test Categories Finger Tapping WCST	Effect of D‐Cycloserine on cognitive performance: No difference between treatment groups on any cognitive task at weeks 8, 12 and 24 compared to baseline.
Neumeister et al., 2006	N/A		HCs 12 (8:4) 28.5(±10.5)		Double‐blind, randomized, balanced cross‐over study design Two separate drug challenge sessions (placebo/active drug): Glycine: 200 mg·kg−1 body weight for 45 min i.v. Placebo: saline	Neuropsychological testing outside of PET scanner: CogState computerized battery (i.e. attention, visual, verbal and working memory, executive function, speed of processing)	Effect of glycine on cognitive performance: No significant effect of glycine on any of the neuropsychological tests compared to placebo Effect of glycine on PET measures: Decreased whole‐brain cerebral metabolic rate of glucose (CMRGlu) in glycine treated HCs when compared to placebo Decreased rCMRGlu in cerebellum and DLPFC without whole‐brain correction (ROI analysis)
Buchanan et al., 2007	EST – D‐cycloserine group 53 (not reported) 44.4(±10.4) EST – glycine group (adjunctive treatment) 52 (not reported) 42.6(±10.8)   EST –placebo group 52 (not reported) 43.4(±11.4)		N/A		16 week double‐blind, double‐dummy, parallel group, randomized study design Three treatment groups (pill‐based): Active glycine + placebo D‐cycloserine Placebo glycine + active D‐cycloserine Placebo glycine + placebo D‐cycloserine	Battery: Processing speed Verbal fluency Motor speed Vigilance Auditory memory Visual spatial memory Auditory working memory Visuo‐spatial working memory Executive function	Effect of glycine on cognitive performance: No difference between glycine and placebo on the composite cognition summary score No significant glycine/placebo or D‐cycloserine/placebo differences
Liem‐Moolenaar et al. 2010	N/A		HCs 45 (45:0) 18–55   Four treatment groups of 15 subjects in each		Double‐blind, placebo‐controlled, four‐period cross‐over ascending dose study design Scopolamine 0.5 mg or placebo i.v. for 15 min and 0, 3, 10, 30 mg of R213129 or placebo oral administration	Adaptive tracking, finger tapping, Stroop test, VVLT	Effect of scopolamine on cognitive performance: Decrease in all parameters of VVLT with scopolamine when compared to placebo Decrease in all parameters of the Stroop test with scopolamine when compared to placebo
Weiser et al., 2012	EST – D‐serine group (adjunctive treatment) 97 (74:23) 39.39 (±12.0)   EST – placebo group 98 (70:28) 39.75(±12.3)		N/A		16 week double‐blind, randomized, placebo‐controlled study design Over course of study between 1.6 g.day‐1 and 2 g.day‐1 of D‐serine	Hebrew version of the MATRCIS with 10 subtests	Effect of D‐serine on cognitive performance: No effect of treatment group or group‐by‐time interaction for the composite score and the individual scores
Cain et al., 2014	EST – D‐Cycloserine group (adjunctive treatment) 18 (16:2) 48.8(±11.5)   EST – Placebo group 18 (15:3) 46.2(±13.3)		N/A		8 week single‐blind, randomized, placebo‐controlled study design D‐cycloserine 50 mg.week‐1 or placebo (capsule)	Cognitive Remediation between 3–5 timed weekly – Brain Fitness Programme Auditory discrimination training MATRICS neuropsychological test battery	Effect of D‐Cycloserine on cognitive performance: No difference in auditory discrimination training in D‐Cycloserine when compared to placebo at baseline Increase in auditory discrimination training in D‐Cycloserine when compared to placebo at each visit after baseline No difference in the composite sore or individual scores of the MATRICS in D‐Cycloserine when compared to placebo Increase in composite score and individual scores of some MATRICS tests in placebo when compared to D‐Cycloserine
Christmas et al., 2014	N/A		HCs – Org 25 935 group   16 (16:0) 23.8(± not reported)   HCs – placebo group 16 (16:0) 25.3(± not reported)		Double‐blind, randomized, placebo‐controlled, parallel group, single‐dose study design Org 25 935: Oral dose of 12 mg Matching placebo	Visuo‐spatial cognitive task (Manikin task) Digit span Verbal memory task	Effect of Org 25925 on cognitive performance: No difference between the groups in the Manikin task or any of the other tasks
Working memory
Duncan et al., 2004	EST – D‐Cycloserine group (adjunctive treatment) 10 (10:0) 48.7 (±5.1)   EST‐ placebo group 12(12:0) 54.5(±6.8)		N/A		4‐week double‐blind, randomized, parallel‐group study design D‐cycloserine: 50 mg (capsule) or placebo (capsule)	1. AXE‐CPT 2. Sternberg Short Term Memory Scanning Paradigm At baseline, after 2 weeks and 4 weeks In subgroup of 7 EST in the D‐cycloserine group and 8 in the placebo group	Effect of D‐cycloserine on cognitive performance: No differences in accuracy or RT on the CPT between D‐cycloserine and placebo at all three time points No differences in accuracy or RT on the Sternberg test between D‐cycloserine and placebo at all three time points

ANART, Adult North American Reading Test; CVLT, California Verbal Learning Test; CPT, continuous performance test; HC, healthy control; HVLT, Hopkins Verbal Learning Test; IQ, intelligence quotient; MATRICS, measurement and treatment research to improve cognition in schizophrenia; N/A, not applicable; PANSS, positive and negative symptoms scale; PET, positron emission tomography; WCST, Wisconsin Card Sorting Test; VVLT, visual verbal learning test.

The role of AMPA receptors in cognition

AMPA receptors are heterotetramers formed from distinct subunits (GluA1–4) or as Ca²⁺‐permeable homotetramers composed of GluA1 subunits (for review, see Henley and Wilkinson, 2016). AMPA receptor expression and trafficking is a highly dynamic process, regulated by neuronal activity, and plays a central role in neuronal plasticity. PAMs of AMPA receptors typically potentiate the channel‐open state of the receptor upon glutamate activation, enhance LTP and have varying effects on long‐term depression (LTD), depending on the class of compound (Arai and Kessler, 2007). Promising results from early studies reported that PAMs improve cognitive function in human participants and in rodents. For example, administration of the AMPA receptor PAM CX516 improved associative and recognition memory performance in HCs (Ingvar et al., 1997). AMPA receptor PAMs have also been reported to improve cognitive performance in ageing healthy participants, in measures such as delayed recall performance (Lynch et al., 1997) and working memory (Wezenberg et al., 2007). These findings are corroborated in rodent studies, where administration of a benzamide AMPA receptor PAM improved performance in discriminative and spatial memory tasks in rats (Staubli et al., 1994). Rodent research has also shown the drug reverses cognitive deficits in subchronic PCP rodent models, relevant to SZ, in behaviours such as attentional set‐shifting (Broberg et al., 2009) and novel object recognition (Damgaard et al., 2010). However, CX516 was shown to be ineffective in improving deficits in cognitive flexibility and working memory seen in patients with SZ, when given as an adjuvant with antipsychotic drugs (Goff et al., 2008). Therefore, the therapeutic potential of AMPA receptor PAMs in SZ requires further investigation.

The role of metabotropic glutamate receptor subtypes 2 and 3 (mGlu2 and mGlu3) in cognition

Numerous studies have demonstrated the efficacy of mGlu_2/3 agonists and PAMs in reversing cognitive dysfunction in animal models. For example, mGlu_2/3 agonists improve acute PCP‐induced working memory deficits (Eglumetad; Moghaddam and Adams, 1998), and deficits in working memory and latent inhibition in GluN1 knockout mice (SAR218645; Griebel et al., 2016), and novel object recognition performance in a post‐weaning social isolation rat model (LY379268; Jones et al., 2011). In addition, the mGlu₂‐selective PAM was shown to improve cognitive flexibility in control rats (Nikiforuk et al., 2010). However, some studies have failed to reproduce these findings and also show that mGlu_2/3 agonists may actually worsen some aspects of cognition, including working memory, when given alone (Eglumetad in rodents and marmosets; Schlumberger et al., 2009; Spinelli et al., 2005) or have no significant effect in control animals (LY395756; Li et al., 2015). Therefore, these compounds may only be effective in improving cognition in states of glutamatergic dysfunction.

In human studies, mGlu_2/3 receptor agonists have provided some promising results, improving cognitive performance. For example, LY2140023 demonstrated encouraging results as a treatment for the positive and negative symptoms in patients with SZ (Patil et al., 2007), but ultimately, the drug failed to pass Phase III clinical trials (Adams et al., 2014). Unfortunately, the procognitive effects of this drug were not assessed in patients with SZ, and the putative procognitive effects of mGlu_2/3 PAMs in SZ are yet to be firmly established.

The role of the mGlu₅ receptor in cognition

In disorders thought to involve NMDA receptor hypofunction, such as SZ, drugs active at mGlu₅ receptors have been proposed as potential therapeutics, due to the close functional coupling between the two receptors and the ability of mGlu₅ activation to potentiate NMDA receptor activity (Awad et al., 2000; Pisani et al., 2001). A key focus of research has been on PAMs of the mGlu₅ receptor, drugs that act by binding to an allosteric site on the receptor to potentiate its activation by glutamate (CPPHA; Chen et al., 2008; CDPPB; Uslaner et al., 2009). In unimpaired (control) rodents, mGlu₅ PAMs have been shown to improve object recognition memory (ADX47273; Liu et al., 2008a; CDPPB; Uslaner et al., 2009), spatial learning (CDPPB and ADX47273; Ayala et al., 2009), contextual fear acquisition (DFPE; Gregory et al., 2013) and extinction learning (CDPPB; Cleva and Olive, 2011). In rodents, mGlu₅ PAMs have also been shown to limit the impact of NMDA receptor antagonists on cognition. For example, CDPPB reverses the dizocilpine‐induced deficits in NOR (Uslaner et al., 2009) and cognitive flexibility, assessed using the ASST (Darrah et al., 2008; LaCrosse et al., 2015). The mGlu₅ PAM ADX47273 has also been shown to decrease premature responding in the 5‐CSRTT test of impulsivity in trait‐impulsive rats and attenuates the increased impulsiveness in rats following dizocilpine administration (Isherwood et al., 2015). These procognitive effects are thought to be mediated by the ability of mGlu₅ PAMS to enhance synaptic plasticity, through both the enhancement of LTP and LTD (Ayala et al., 2009; Xu et al., 2013). However, while some of these effects may be dependent on the interaction of the mGlu₅ receptor with the NMDA receptor, others may be independent of this interaction (Rook et al., 2015).

There is also evidence of cognitive improvement with negative allosteric modulation (NAM) of mGlu₅ receptors. For example, CTEP reverses an inhibitory avoidance deficit in mice with 16p11.2 microdeletion (Tian et al., 2015). However, injection of the mGlu₅ antagonist MPEP into the lateral ventricles of control rats prior to training impairs working memory performance in the radial arm maze (Manahan‐Vaughan and Braunewell, 2005) and systemic pretreatment exacerbates dizocilpine‐induced deficits in spatial working memory (Homayoun et al., 2004). The mGlu₅ NAMs, basimglurant and MTEP, also impair performance in the 5‐CSRTT in control animals (Isherwood et al., 2015). Overall, these data suggest that an optimal level of mGlu₅ activity is required for effective cognition, and the preclinical research suggests that mGlu₅ receptors may be a promising therapeutic target to improve cognitive deficits, at least in some disorders. However, research in human subjects has yet to demonstrate an effect of mGlu₅‐selective drugs on cognition (Berry‐Kravis et al., 2016).

Glutamatergic regulation of functional brain network connectivity: insights from pharmacological studies targeting the NMDA receptor

The study of how drugs that target the glutamate system alter functional brain network connectivity to influence cognition is in its extreme infancy. However, recent data from studies characterizing the impact of NMDA receptor antagonists on brain network connectivity have provided new insight into the glutamatergic regulation of brain connectivity. In addition, these studies have highlighted the potential translational value of the analysis of brain network connectivity, with the reported effects appearing to be conserved between species (rodents, primates and humans) and across imaging modalities, in measures of brain network connectivity. Here, we provide a brief overview of the studies that have characterized the impact of NMDA receptor antagonists on functional brain network connectivity. The results highlight the potential future utility of this approach in studying other manipulations of the glutamatergic system, whether they are pharmacological or genetic, that are known to affect cognition.

Insights from rodent studies characterizing the impact of NMDA receptor antagonists on functional brain network connectivity

Acute treatment with a subanaesthetic dose of ketamine induces abnormal increases in functional brain network connectivity as analysed using [¹⁴C]‐2‐deoxyglucose functional brain imaging (Dawson et al., 2014). Ketamine treatment increases the number of functional connections and alters the topographic properties of functional brain networks to increase clustering between local brain regions. This suggests that subanaesthetic ketamine treatment affects cognition by promoting abnormally enhanced functional connectivity (FC) in local subsystems. From a neural subsystems perspective, this includes abnormally increased local FC between subfields of the PFC (Dawson et al., 2013), which parallels the PFC regional hyperconnectivity induced by subaneasthetic ketamine treatment in primates (Gopinath et al., 2016) as well as during resting‐state in humans (Anticevic et al., 2015). In contrast, subanaesthetic ketamine treatment in rats impairs long‐range connectivity, including, for example, decreased PFC FC to thalamic inputs (Dawson et al., 2013; 2014). This suggests that ketamine treatment both compromises the ability of the PFC to receive information from other neural subsystems and that enhanced local clustering within the PFC compromises the appropriate segregation of the information received at the local level. These two mechanisms may contribute to the impact of ketamine on PFC‐dependent cognitive processes in rodents (Nikiforuk and Popik, 2014; Nikiforuk et al., 2016). In addition, FC between neuromodulatory subsystems such as the dorsal raphe nucleus, the origin of serotonergic (5‐HT) innervation and the locus coeruleus (LC) and the origin of noradrenergic (NA) innervation to the PFC are abnormally enhanced by acute NMDA receptor blockade (Dawson et al., 2013, 2014). Both 5‐HT and NA are known to modulate PFC‐dependent cognitive processes (Berridge and Spencer, 2016; Clarke et al., 2007). Thus, the modification of the connectivity between neuromodulatory subsystems and the PFC may be a key mechanism contributing to the impact of acute NMDA receptor blockade on cognition, in addition to the local effects of ketamine in the PFC and other cognitive neural subsystems (e.g. hippocampus). The disruption of thalamo‐cortical connectivity is a major effect of acute NMDA receptor antagonist treatment, with the thalamic reticular nucleus being a particularly important target (for review, see Pratt and Morris, 2015). Interestingly, disrupted thalamocortical connectivity is found both in rodents treated with ketamine, when brain networks are analysed using [¹⁴C]‐2‐deoxyglucose, and in human participants treated with ketamine when analysed using resting‐state magnetoencephalography (Rivolta et al., 2015), supporting not only the conservation of alterations in FC across species but also across different imaging modalities. This conservation may be key to facilitating translation in the context of identifying procognitive drugs that target the glutamatergic system.

In contrast to the effects of acute NMDA receptor blockade, prolonged NMDA receptor hypofunction, as induced by subchronic PCP treatment, disturbs functional brain network connectivity in rodents (Dawson et al., 2012; 2014). At the global network scale, this results from a decreased number of functional connections in the brain network, decreased clustering and an increase in the number of functional connections that must be traversed to reach one brain region from another (a measure known as average pathlength, Dawson et al., 2014). These global alterations strongly parallel those reported in functional resting‐state brain networks of EST (Micheloyannis et al., 2006; Liu et al., 2008b), supporting the translational potential of these network analyses. Subchronic PCP treatment also results in decreased thalamic, hippocampal and PFC connectivity and induces a decrease in the functional integration between the hippocampus and PFC (Dawson et al., 2012), which could contribute to the cognitive deficits seen as a result of prolonged NMDA receptor hypofunction. Decreased hippocampal‐PFC resting‐state FC is also seen in patients with SZ (Kraguljac et al., 2017) and genetic rodent models relevant to the disorder (Dawson et al., 2015; Sigurdsson et al., 2010). Again, a central role for altered neuromodulatory system connectivity is indicated as a result of prolonged NMDA receptor hypofunction, with decreased PFC – LC connectivity supported (Dawson et al., 2012).

Insights from human studies characterizing brain network connectivity alterations induced by NMDA receptor antagonists

Graph theory approaches to characterizing altered brain network connectivity have been widely applied in relation to brain network connectivity in SZ patients (Micheloyannis et al., 2006; Liu et al., 2008a; van den Heuvel et al., 2010; Hadley et al., 2016) and in a range of other cognitive brain disorders. However, very few studies have applied graph theory network analysis in the context of pharmacologically‐induced NDMA receptor hypofunction in HCs. For example, using task‐free pharmacological MRI, Joules et al. (2015) report increased degree centrality, indicative of the number of functional connections that a given region has in the context of the brain network, for the basal ganglia and decreased centrality for cortical regions, including regions in the frontal cortex, following ketamine administration (Joules et al., 2015). No other graph theory measures were reported as a part of this study, and the authors themselves highlight the additional insight that the application of these additional measures could give. The data‐driven approach taken in this study highlights the value of using graph theory approaches to define alterations in network connectivity. The reduced PFC connectivity induced by ketamine administration in this study contrasts with the increased PFC connectivity reported using FC analysis in HCs by others (Anticevic et al., 2015) and the preclinical data that support the general enhancement of PFC connectivity following ketamine administration (Dawson et al., 2014). The reasons for this disparity remain unclear but may include the use of only one form of connectivity analysis in the study of Joules et al. (2015) (degree centrality) or that the regions of interest that were included in the analysis influenced the findings (as ketamine treatment has been shown to increase and decrease PFC connectivity to different brain regions; Dawson et al., 2013). For example, another recent study used a seed regional approach to characterizing the impact of ketamine on PFC‐hippocampal connectivity in HCs, which indicates a ketamine‐induced increase in the FC between these neural subsystems (Grimm et al., 2015). Interestingly, this investigation also confirmed similar effects in rodents using the same approach, further highlighting the translational value of measuring functional brain network connectivity.

As the application of graph theory methods in the context of NMDA receptor antagonist‐induced alterations in brain network FC is relatively limited, here, we also consider the effects on task‐based FC from fMRI studies. To date, the vast majority of these studies have been applied in the context of the influence of NMDA receptor antagonists during working memory tasks, which has potential translational confounds when attempting to compare the observed effects with those seen in resting‐state brain imaging data. However, overall, the effects reported seem to be similar to those found when brain imaging is undertaken at rest, and in animal models. For example, Driesen et al. (2013) found that acute ketamine administration significantly reduced FC between the DLPFC and middle frontal gyrus and impaired performance during working memory function in HCs. In a more recent study, dextromethophan led to increased FC within a brain network comprising 270 seed regions involving the DLPFC (Braun et al., 2016). The findings parallel the increased DLPFC FC seen during working memory function in patients with SZ (Siebenhuhner et al., 2013). Anticevic et al. (2012) also found that ketamine administration increased task‐based FC in the fronto‐parietal network and reduced task‐deactivated FC of the default mode network during a working memory task (Anticevic et al., 2012), which also appears to be similar to the effects seen in patients with SZ. To date, only these three studies have reported the effects of NMDA receptor antagonists on alterations in task‐based FC in HCs. More research into the impact of NMDA receptor antagonists, and other glutamatergic compounds, on working memory and brain network connectivity is needed in order to gain a better understanding of the effects of glutamatergic modulators on cognitive function and their role in SZ. Multi‐modal studies of cognitive function, with the simultaneous measurement of glutamatergic concentrations (using magnetic resonance spectroscopy) in combination with BOLD fMRI, may lead to greater insights into glutamatergic responses during cognitive functions in patients with SZ (for example, see Taylor et al., 2015).

Conclusion

The glutamatergic system plays a primary role in the regulation of multiple domains of cognition. Targeting glutamatergic neurotransmission offers hope for the treatment of cognitive deficits seen in patients with SZ and other brain disorders with pronounced cognitive deficits. Characterizing the effect of a modified glutamate system function on brain network connectivity offers new systems‐level insight into the mechanisms underlying the glutamatergic regulation of cognition. The study of how functional brain networks are modulated by glutamateric neurotransmission is in its extreme infancy. Here, we have outlined recent studies that have characterized the impact of NMDA receptor antagonists on brain network connectivity as a leading exemplar of the new insight that can be gained from the study of how the glutamate system modulates brain network connectivity and cognition. The use of this approach may provide results that have great translational value, as initial observations appear to be conserved across different species and imaging modalities. Future studies dedicated to investigating the effects of other procognitive compounds and modifications of glutamatergic system function, whether they are pharmacological or genetic, should be undertaken in order to further understand the mechanisms through which these manipulations elicit their effects on cognition.

Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY (Southan et al., 2016), and are permanently archived in the Concise Guide to PHARMACOLOGY 2015/16 (Alexander et al., 2015a,b).

Conflict of interest

The authors declare no conflicts of interest.

Dauvermann, M. R. , Lee, G. , and Dawson, N. (2017) Glutamatergic regulation of cognition and functional brain connectivity: insights from pharmacological, genetic and translational schizophrenia research. British Journal of Pharmacology, 174: 3136–3160. doi: 10.1111/bph.13919.