Ultrastructural evidence for glutamatergic dysregulation in schizophrenia

Abstract

The aim of this paper is to summarize ultrastructural evidence for glutamatergic dysregulation in several linked regions in postmortem schizophrenia brain. Following a brief summary of glutamate circuitry and how synapses are identified at the electron microscopic (EM) level, we will review EM pathology in the cortex and basal ganglia. We will include the effects of antipsychotic drugs and the relation of treatment response. We will discuss how these findings support or confirm other postmortem findings as well as imaging results. Briefly, synaptic and mitochondrial density in anterior cingulate cortex was decreased in schizophrenia, versus normal controls (NCs), in a selective layer specific pattern. In dorsal striatum, increases in excitatory synaptic density were detected in caudate matrix, a compartment associated with cognitive and motor function, and in the putamen patches, a region associated with limbic function and in the core of the nucleus accumbens. Patients who were treatment resistant or untreated had significantly elevated numbers of excitatory synapses in limbic striatal areas in comparison to NCs and responders. Protein levels of vGLUT2, found in subcortical glutamatergic neurons, were increased in the nucleus accumbens in schizophrenia. At the EM level, schizophrenia subjects had an increase in density of excitatory synapses in several areas of the basal ganglia. In the substantia nigra, the protein levels of vGLUT2 were elevated in untreated patients compared to NCs. The density of inhibitory synapses was decreased in schizophrenia versus NCs. In schizophrenia, glutamatergic synapses are differentially affected depending on the brain region, treatment status, and treatment response.

1. Introduction

Evidence from in vivo imaging, postmortem studies and animal models of schizophrenia implicate the glutamatergic system in schizophrenia (Coyle, 2006; Goff and Coyle, 2001; Javitt, 2004; Krystal, 2008). The aim of this paper is to provide ultrastructural evidence of glutamatergic dysregulation in schizophrenia and discuss how these findings relate to other postmortem findings and imaging studies. It is noteworthy that quantitative ultrastructural studies of postmortem brain in schizophrenia are rarely conducted outside our lab (Aganova and Uranova, 1992; Uranova and Levitt, 1987; Kolomeets and Uranova, 1999; Kolomeets et al.,2005, 2007). Although cohort size is small in these ultrastructural studies, the results are consistent with those of other studies using methodologies that permit larger cohorts. Psychotomimetic agents, acting on glutamate receptors, produce psychosis in healthy controls and exacerbate psychosis in people with schizophrenia (Itil et al., 1967; Lahti et al., 1995). Glutamate levels in the basal ganglia are elevated in both in vivo and postmortem studies of schizophrenia. Moreover, in experimental animals, psychotomimetics increase striatal neuronal activity, which is reversed by antipsychotic drugs (APDs) (White et al., 1995).

1.1. Glutamate circuitry in the basal ganglia and connecting structures

A simplified version of pertinent connections of the basal ganglia is shown in Figure 1A. Diverse regions of the cerebral cortex including limbic areas (such as orbitofrontal and anterior cingulate cortices), associative and cognitive areas (such as dorsolateral prefrontal cortex, DLPFC), and motor and premotor cortex project to the caudate nucleus and putamen (Alexander et al., 1986; Chikama et al., 1997; Eblen and Graybiel, 1995; Goldman-Rakic, 1994; Goldman-Rakic and Selemon, 1986; Haber et al., 1995, 2000; Kunishio and Haber, 1994; Parent and Hazrati, 1995; Smith et al., 1998) in a topographical pattern (Haber, 2000). The vast majority of striatal synapses are glutamatergic and form with the dendritic spines of medium spiny neurons (the major striatal cell type) (Kemp and Powell, 1971). The outputs of the striatum and substantia nigra pars reticulata are all GABAergic. However, the ventral tegmental area contains a population of neurons that are glutamatergic (Root et al., 2016). The subthalamic nucleus (Benarroch 2008) and the pedunculopontine nucleus (Charara et al. 1996) send glutamatergic projections to the substantia nigra/ ventral tegmental area.

Figure 1.

A) Simplified drawing illustrating glutamatergic connectivity between cortex, basal ganglia and thalamus. The cortex receives glutamatergic inputs from the thalamus; illustrated here is the connection between medial dorsal thalamus and layer III of granular cortex. Glutamatergic neurons in the cortex project to the caudate nucleus, the putamen and the nucleus accumbens. Layers III, V,VI project to the dorsal striatum (caudate, putamen). Specifically, layer III projects to the matrix and layers V,VI project to the patches. The subthalamic nucleus (STN) and the pedunculopontine nucleus (PPN) send glutamate projections to the substantia nigra, ventral tegmental area. B) Schematic drawing illustrating synaptic morphology. Several spines emerge from the dendrite and one is shown enlarged. Synaptic types are defined by the morphology of the postsynaptic density (PSD) and their target (dendrite, spine, axon initial segment, or soma). Type I synapses are mostly excitatory, and have been classically called “asymmetric”, based on a thick PSD in comparison to a nonexistent density on the presynaptic side. Type II synapses are mostly inhibitory, and have been classically called “symmetric” based on a thin PSD more comparable in size to the non-existent density on the presynaptic side. Axospinous synapses are depicted. AT, axon terminal.

1.2. Electron microscopy in human postmortem brain

Electron microscopy in postmortem human brain is difficult due to issues with obtaining tissue with very short postmortem intervals. Our data presented in this review have been previously published and details of each cohort of subjects are found in our articles (Kung et al., 1998; Roberts and Knickman, 2002; Roberts et al., 2005a,b, 2012, 2015; McCollum et al., 2015; Mabry et al., 2019. Human postmortem brain tissue was obtained from the Maryland and Alabama Brain Collections, with family permission and with IRB approved protocols. Briefly, tissue was dissected from the fresh brain and stored until used in a fixative solution of 1% glutaraldehyde and 4% paraformaldehyde in 0.1M phosphate buffer (pH 7.4). Subjects with schizophrenia were matched to the control subjects for age, gender, postmortem interval and race when possible. Controls had no history of central nervous system or neurological diseases. Antipsychotic drug treatment, duration of illness, predominant symptomology, treatment response, and other medical details were obtained from hospital charts, autopsy reports and family interviews. The diagnosis of schizophrenia was made by two research psychiatrists according to the DSM criteria (DSM-III-DSM-V) at the time of the diagnosis.

1.3. Synaptic ultrastructure

Synaptic types are categorized by the morphology of the postsynaptic density and their target (dendrite, spine, axon initial segment, or soma) (Figure 1B). Synapses are defined by having three criteria: an axon terminal with vesicles at the synapse, a postsynaptic density, and parallel pre and postsynaptic membranes (Figure 1B). Traditionally, excitatory and inhibitory synapses can be identified by the morphology of the postsynaptic density (Gray, 1969). Synapses with thick postsynaptic densities are called “asymmetric” and usually identify excitatory synapses, while synapses with thin postsynaptic densities are called “symmetric” and identify inhibitory synapses. These terms assume no presynaptic density. The vast majority of glutamatergic axon terminals form synapses with dendritic spines (Harris and Kater, 1994; Arellano et al., 2007). Glutamate receptors and other molecules needed for glutamate function (Krueger-Burg et al. 2017) are localized to synapses with thick postsynaptic densities. For example, PSD95 is a scaffolding protein located in the postsynaptic density of excitatory synapses (Cho et al., 1992; Sheng & Hoogenraad, 2007). Gephyrin is a postsynaptic scaffolding protein found exclusively at GABAergic and glycinergic synapses (Craig et al., 1996).

Tract tracing, alone or in combination with immunohistochemistry, is used in animal models to identify projections and circuitry within the brain. Thus, particular projections can be identified by the combination of the postsynaptic target and symmetry of the synapse. For example, parvalbumin containing chandelier cells form symmetric synapses on the axon initial segment of cortical pyramidal cells (Lewis et al., 2002). Glutamatergic pyramidal cells from layer V cortex project to the striatum and form asymmetric axospinous synapses on medium spiny neurons (Kemp and Powell, 1971). Anatomical studies in human brain use the results of circuitry from the animal literature to make educated predictions as to the origin of synapses based on morphology.

1.3.1. Methods for counting synapses

Synapses are counted using two or three-dimensional techniques. Two dimensional techniques measure the density of synapses per unit area and are obtained from a method called simple profile counting. Three-dimensional reconstruction or stereological methods measure the density of synapses per unit volume (Geinisman et al., 1996). Stereological techniques were used to count synapses in the cortex (Figure 2A) and striatum (Figures 3&4), while simple profile counts were used in the substantia nigra. Examples of synapses are shown for each region in corresponding figures.

Figure 2.

Anterior Cingulate Cortex. Electron micrographs and graphs are modified from Roberts et al., 2015 as indicated. A,A’) Representative working images (magnification=15,000) used for synapse counting. Each is a montage stitched together from eight micrographs; A and A’ are taken from serial sections. Arrows indicate synapses (identified at higher magnification). Modified from Figure 2. B) An excitatory axospinous synapse (arrow) forms on a spine (sp) emerging from a dendrite (den) from a subject with schizophrenia. m, mitochondria. C) An inhibitory axospinous synapse (arrow) forms on a spine from a subject with schizophrenia. Modified from Fig 4D. D) A dendrite is postsynaptic to two excitatory synapses from a NC case. Modified from Fig 4B. E) Representative electron micrograph of a pyramidal cell, outlined in black, from a subject with schizophrenia. Modified from Fig 8B. F) Total synaptic density is decreased in layers III-VI combined in schizophrenia, due to a selective loss of axospinous synapses. Modified from Fig 5A. G) The number of mitochondria per neuron is decreased in layers III-VI combined in schizophrenia, due to a selective loss in the deep layers. Modified from Fig 3B. H) Mitochondria are reduced in number in axon terminals forming asymmetric (excitatory) synapses in layer III and in axon terminals forming symmetric (inhibitory) synapses in layers V,VI. Modified from Fig7B. Scale bars: A, E=5μm; B-D=500nm. Significance: *, p<0.05; **, p<0.01; ***, p<0.005.

Figure 3.

Caudate Nucleus and Putamen. A) Synaptic density is shown for the caudate nucleus and putamen in normal controls (NC) and subjects with schizophrenia (SZ). All synapses combined (total), asymmetric axospinous (AS) and all asymmetric synapses (Asym) are increased in density in the caudate nucleus in schizophrenia. The putamen shows a similar pattern, but without reaching statistical significance. Modified from Figures 4A and 5A, Roberts et al., 2005a. B) A light microscopic image of a calbindin immunolabeled section from the striatum. Calbindin immunolabeling, which is much more robust in the matrix (M) than in the patches (P), was used to distinguish the two compartments. The trapezoids indicate regions within each compartment that were sampled for synaptic density. Small arrows indicate immunolabeled neurons. Modified from Figure 2, Roberts and Knickman, 2002. C) Total synaptic density in subjects with schizophrenia was significantly higher in the caudate matrix (CM all) and the putamen patches (PP all) than in controls. The density of asymmetric axospinous (AS) synapses was significantly higher in the putamen patches (PP AS) in schizophrenia; the same pattern, albeit only at a trend level, was found in the caudate matrix (CM AS). Modified from Figures 3A and 4A from Roberts et al., 2005b. D) The schizophrenia cohort was subdivided into treatment responders and treatment resistant/or off APDs. In the striatal patches, the treatment resistant/off APD cases had higher synaptic densities of total synapses, asymmetric axospinous synapses and all asymmetric synapses than controls and treatment responders; treatment responders had a higher density of asymmetric axodendritic synapses than controls and treatment resistant/off APD cases. Total Modified from Figure 6, Roberts et al., 2012. Significance: *, p<0.05; **, p<0.01.

Figure 4.

Nucleus Accumbens. A) Western blot analysis of vGLUT2 in control (NC), schizophrenia as a group (SZ) and divided by APD status at the time of death: on-drug (SZ-ON), and off-drug (SZ-OFF). B) Graphs showing optical density of vGLUT2 where the schizophrenia group has significantly higher protein levels of vGLUT2; A and B were modified from Figure 1C in McCollum and Roberts 2015. Electron micrographs and synapse quantification are modified from McCollum et al., 2015 as indicated. C) An axon terminal (AT) forms an asymmetric synapse (arrow) with a spine (sp); modified from Figure 1 D). An AT forms an asymmetric synapse (arrow) with a dendrite (DEN). E) A spine (SP) receives convergent input from axon terminals (AT) making asymmetric (black arrows) and symmetric (white arrow) synapses; modified from Figure 1B. F) An axon terminal (AT) forms a symmetric synapse (arrow) with a dendrite (DEN); modified from Figure 1C. G-I) Quantification of synaptic density in the core and shell. G,H) Density of all synapses (Total), total asymmetric (Total Asym) and total symmetric (Total Sym) synapses, total axospinous (Total Sp) and total axodendritic (Total Den) synapses. I-J) Density of synapse types: asymmetric axospinous (AS), asymmetric axodendritic (AD), symmetric axospinous (SS), and symmetric axodendritic (SD). K,L) Serial sections of an axon terminal forming an elaborate asymmetric multi-perforated synapse with a spine. Scale bars (B-D, K,L) = 500 nm.

2. Synaptic abnormalities in schizophrenia

2.1. Anterior Cingulate Cortex

The anterior cingulate cortex (ACC) is part of prefrontal cortex and is composed of distinct anatomical subregions, each with different functional properties (Peterson et al., 1999; Vogt et al., 1992). Human imaging studies indicate that the ACC is involved in mediating attention, executive functions, cognitive tasks (Koski and Petrides, 2001), response conflict (Barch et al., 2001; Kerns et al., 2004), response inhibition, error commission (Braver et al., 2001; Mathalon et al., 2002), and emotional processing (Bush et al., 2000).

In an ultrastructural study of the ACC (Figure 2A–E), we found that all synapses combined were decreased in density by 28% in the schizophrenia cases compared to the NCs (Figure 2F) (Roberts et al., 2015). Asymmetric axospinous synapses (Figure 2B) were reduced in density by approximately 30% in the patients vs. NCs (Figure 2F). Symmetric axospinous synapses (Figure 2C) were reduced in density by 46% in the schizophrenia group vs. NCs. Axodendritic synapses (Figure 2D) showed no difference in density between the schizophrenia group and NCs. In the combined layers of the ACC, the number of mitochondria per neuronal somata (Figure 2E) in the schizophrenia cases was decreased by 43% in comparison to NCs (Figure 2G). This decrease was due to a selective loss of mitochondria from neurons in layers V/VI. In layer III, the density of mitochondria in axon terminals forming any type of synapse was lower in schizophrenia by 34% as compared to NCs (Figure 2H). This finding was due to a selective loss of mitochondria in terminals forming asymmetric axospinous synapses. In layer V,VI, the density of mitochondria in terminals forming symmetric synapses, both with spines and dendrites was lower in the schizophrenia group than NCs (Figure 2H).

Our findings showed that the ACC in schizophrenia has a decrease in number of excitatory synaptic connections, possibly arising from multiple sources, which may have effects on several downstream pathways (Roberts et al., 2015). Analysis of the vesicular glutamate transmitters (vGLUTs) yields important information about the location of the cell bodies of origin (Bellocchio et al., 2000; Vigneault et al., 2015). The amount of the vGLUTs is directly related to quantal release of glutamate and thus is a good measure of glutamate levels (Wilson et al., 2005). vGLUT1 mRNA is found in the cell bodies of origin of cortical glutamatergic neurons, and the protein is found in the terminals, while vGLUT2 mRNA is found in the cell bodies of origin of subcortical glutamatergic neurons, and the protein is found in the terminals (Vigneault et al., 2015). In the ACC, protein analysis of the vGLUTs (Oni-Orisan et al., 2007) demonstrated that levels of vGLUT1, but not vGLUT2, were reduced in schizophrenia, and not a cause of medication. Taken together, these data suggest that cortical-cortical connections are decreased in the ACC in schizophrenia.

Pyramidal neurons in layers V,VI project to the striatum, brainstem, or thalamus (Goldman and Nauta, 1977). The reduction in number of mitochondria in neurons in these layers suggests that neurons that project to these subcortical regions may have a compromised metabolism. Another downstream target is the hippocampus, where decreased numbers of excitatory axospinous synapses have been reported in schizophrenia cases (Kolomeets et al., 2005, 2007). Dopaminergic terminals form symmetric axospinous synapses (Kung et al., 1998), and this type of synapse is reduced in density in prefrontal cortex in schizophrenia, possibly reflecting a loss of dopamine input. In support of this finding are previous studies showing a decrease in density of dopaminergic fibers in deep cortical layers in the ACC (Benes et al., 1997), as well as other regions of prefrontal cortex (Akil et al., 1999).

Although a number of magnetic resonance spectroscopy (MRS) studies of the ACC have been conducted in schizophrenia, inconsistent results have been reported (Merritt et al., 2016). Such varying results are likely to be due to differences in stage of illness and medication status, magnet strength, sequence acquisitions and data post processing. However, consistent with postmortem findings, two recent studies conducted in first episode psychosis subjects using ultra-high field magnets [7 Tesla (T)] to obtain better spectral resolution reported a decrease in ACC glutamate (Reid et al., 2019, Wang et al., 2019). Fewer studies have evaluated ACC GABA levels in schizophrenia (Egerton et al., 2017). Wang and colleagues (Wang et al., 2019) reported decreases in both ACC glutamate and GABA levels in first episode psychosis subjects.

MRS studies in schizophrenia have also consistently reported decreased N-acetyl-aspartate levels in the ACC (Jessen et al., 2013; Reid et al., 2010, 2019). N-acetyl-aspartate is an amino acid that is primarily synthesized in neuronal mitochondria (Bates et al., 1996) and is metabolized in oligodendrocytes to acetate, a precursor of myelin lipid synthesis (Chakraborty et al., 2001). Thus, the amount of NAA detected in imaging studies may reflect neuronal integrity, the number of axon terminals and synapses, myelin integrity, and/or mitochondrial function (Bates et al., 1996; Chakraborty et al., 2001; Ledeen et al., 2006; Lentz et al., 2005; Moffett et al., 2007; Stork and Renshaw, 2005). Therefore, the results of the present study and those of Aganova and Uranova (1992) support that any of these mechanisms could underlie decreased NAA as seen in in vivo imaging studies. The decrease in density of axospinous synapses together with a lower proportion of remaining terminals that contain mitochondria and lower somatic mitochondrial counts, suggests a decrease in cortical synaptic efficiency. Overall, our changes suggest alterations in multiple cortical connections that may impact on cognitive functions subserved by the ACC. Cognitive tasks of inhibitory control are known to activate a network of regions that includes the ACC as its core region (Kouneiher et al., 2009). Again, consistent with postmortem findings, functional MRI (fMRI) studies in schizophrenia have repeatedly demonstrated altered ACC functional activation during cognitive control tasks (Kerns et al., 2005; Weiss et al., 2007; Minzenberg et al., 2009). In addition, an abnormal relationship between ACC glutamate and the blood oxygen level dependent signal have been reported in schizophrenia (Flakenberg et al., 2014; Cadena et al., 2018; Overbeek et al., 2019).

MRS can also be used to measure neuromatabolites linked to mitochrondial function and brain bioenergetics, such as glutathione, an antioxidant, and lactate, a byproduct of glycosis. In schizophrenia, reduced ACC glutathione has been reported by several (Kumar et al., 2018; Das et al., 2019), including one study at 7 T (Wang et al., 2019) suggesting a decreased protective buffer from oxidative stress. Interestingly, elevated lactate levels have been reported in several brain areas in schizophrenia (Hagihara et al., 2017; Sullivan et al., 2019), including the ACC (Rowland et al., 2016). Increased lactate levels, as well as decreased pH, are features of schizophrenia as well as other psychiatric illnesses (Hagihara et al., 2017). Importantly, these findings are also present in animal models of several psychiatric disorders and are thought to be an undelying pathology of illness rather than an artifact of postmortem interval (Hagihara et al., 2017; Sullivan et al., 2019). The finding of elevated lactate in schizophrenia suggests that, in the presence of impaired mitochondrial function, a shift to greater glycosis results in the accumulation of lactate. In addition, phosphorus MRS (³¹P MRS) allows the quantification of phosphorus metabolites, including energy related metabolites, such as adenosine triphosphate (ATP), phosphocreatine (PCr), inorganic phosphate (Pi), and nicotinamide adenine dinucleotide (NAD) signals. In the respiratory chain of the mitochondria, ATP is formed from adenosine diphosphate (ADP) and Pi, a process coupled to the reaction transferring the energy moiety of ATP to Cr to generate PCr. The reversible reaction can then generate ATP from ADP and PCr. Not only can these metabolites be measured, but also the rates of these chemical exchanges can be estimated using ³¹P MRS. In schizophrenia, a number of studies have reported abnormalities of these metabolites, as well as altered chemical exchange rates (Yuksel et al., 2015). Together, these imaging studies strongly implicate impaired mitochondrial function in schizophrenia.

2.2. Caudate nucleus and putamen

The dorsal striatum, including the caudate nucleus and putamen, processes motor, cognitive and limbic functions. It is also the region of the brain with the highest density of dopamine D₂ receptors, the target of antipsychotic drugs (Carlsson, 1988, 2006; Creese et al., 1976). The striatum receives robust projections from most regions of cortex, including the ACC (Kunishio and Haber, 1994). The majority of neurons in the striatum are medium spiny neurons and they receive the majority of inputs onto their spines; these inputs from cortex and thalamus form glutamatergic synapses (Kemp and Powell, 1971). We found higher synaptic density in the caudate nucleus in schizophrenia with similar trends in the putamen (Figure 3A) (Roberts et al., 2005a). This increase is due to a selective increase in asymmetric synapses onto spines, suggesting the presence of more spines receiving glutamatergic synapses.

The caudate nucleus and putamen are organized in a striosomal pattern (Graybiel and Ragsdale, 1978), which contains patches (or striosomes) imbedded in a larger matrix. These two anatomically distinct striatal compartments differ from each other in several ways including neurochemical composition (Gerfen et al.,1985; Graybiel and Ragsdale, 1978; Holt et al., 1997; Prensa et al., 1999), neuronal organization (Hirsch et al., 1989; Holt et al., 1997; Kubota and Kawaguchi, 1993; Penny et al., 1988), connectivity (Gerfen, 1989, 1992; Kincaid and Wilson, 1996; Ragsdale and Graybiel, 1991), and response to antipsychotic drugs (Bubser and Deutch, 2002). Layer III glutamatergic neurons of the cortex project to the matrix, while the glutamate neurons in layers V/VI project to the patch compartments (Gerfen, 1989).

Patches and matrix are easily distinguished by calbindin immunolabeling, which is present more abundantly in the matrix (Gerfen et al., 1985) (Figure 3B). The matrix compartment in associative regions preferentially receives inputs from the dorsolateral prefrontal cortex (Eblen and Graybiel, 1995), which processes higher cognitive functions, such as working memory (Goldman-Rakic, 1999). Motor and somatosensory cortices also project to the matrix in a topographical fashion (Flaherty and Graybiel, 1993). When considering the patch and matrix separately, total synaptic density was elevated in the caudate matrix and the putamen patches, with asymmetric axospinous synapses driving this increase (Figure 3C) (Roberts et al., 2005b). Thus, it would be expected that increased glutamatergic synaptic density in the striatal matrix could impact cognition, working memory, motor, and somatosensory functions, while increased glutamatergic inputs to the patches would impact limbic function.

When considering treatment response and treatment resistance, the increase in total and asymmetric synapses in the patches was confined to the treatment resistant group (Figure 3D) (Roberts et al., 2012). Treatment responders had normal levels of synapses with the exception of an increase in asymmetric axodendritic synapses, typical of thalamic afferents (Sadikot et al., 1992a,b) in both the patch and matrix. This could be due to an increase in input of thalamic inputs or retraction of spines, but not the synapses onto them. The source of these extra axodendritic synapses could be determined by localizing vGLUT1 and vGLUT2 to identify cortical and thalamic inputs, respectively (Fremeau et al., 2001, 2004).

Although our cohort of treatment resistant and responsive patients was small, in vivo imaging studies have confirmed our results. Demjaha and colleagues (Demjaha et al., 2014) found that treatment resistance is associated with higher striatal glutamate levels and normal dopamine levels. MRS studies have shown increased glutamate in the striatum of individuals at high risk to develop schizophrenia, in patients with schizophrenia who were off medication and floridly psychotic, or drug naive at their first episode of psychosis (de la Fuente-Sandoval, 2009, 2011, 2012, 2013). Moreover, elevated levels of glutamate, located in the associative striatum in medication naive first episode patients, return to normal following antipsychotic treatment (de la Fuente-Sandoval, 2013). The associative region of the striatum is also the location of increased dopamine release in schizophrenia (Kegeles et al., 2010). These data suggest that increases in striatal glutamate predate the illness, are not a medication effect, and occur in the same region where increased dopamine release is found.

Disturbance in striatal functional connectivity is more pervasive in treatment-resistant patients (White et al., 2016). Prior to treatment, striatal functional connectivity patterns, as well as changes in striatal functional connectivity with treatment were associated with better treatment response (Cadena et al., 2019; Sarpal et al., 2015). In addition, caudate volumes measured prior to treatment were also associated with subsequent treatment response (Hutcherson et al., 2014). Thus, like in postmortem studies, imaging studies in schizophrenia found evidence of altered glutamatergic activity, as well as associations between striatal function, structure, and treatment response.

2.3. Nucleus Accumbens.

The nucleus accumbens integrates signaling from multiple regions of the brain, receiving input from areas including the prefrontal cortex, hippocampus, amygdala, thalamus, and midbrain. (Groenewegen and Trimble, 2007). Importantly, all of these areas have been associated with schizophrenia, making the nucleus accumbens a prime region for integrating multiple disrupted areas (Grace, 2000). In a recent study, we found that vGLUT2 protein levels in the nucleus accumbens were significantly elevated in the schizophrenia group (Figure 4A,B) (McCollum and Roberts, 2015). Protein levels of vGLUT2 did not differ between schizophrenia subjects on medication (SZ-ON) and schizophrenia subjects off medication (SZ-OFF) subjects. Protein levels of vGLUT1 were similar in control and schizophrenia subjects, with no difference between SZ-ON and SZ-OFF subjects (data not shown). Similar values in schizophrenia both on and off medication suggest that there is no effect of APDs. The selective increase in vGLUT2 protein implicates subcortical afferents. mRNA expression studies in other brain areas such as the temporal cortex (Uezato et al., 2009) and thalamus (Smith et al., 2001) also support an increase in glutamate expression in neurons in subcortical locations.

The nucleus accumbens is divided into a core and shell, the former being more similar to the dorsal striatum in neurochemical composition and connectivity. Using calbindin immunolabeling as a template, we were able to quantify synapses selectively in the core versus the shell (McCollum et al., 2015). Most synaptic morphologies (Figure 4C–F) were typical of dorsal striatum in both the core and shell. Within the core, there was an increased density of total synapses in the schizophrenia group (Figure 4G). This increased density was specific to asymmetric synapses, while symmetric synapses had a similar density between the two groups. The density of all axospinous synapses was increased at a trend level (p = 0.054). Further, when analyzing synapse subtypes, the increased density was found exclusively in asymmetric axospinous synapses (Figure 4I). These differences were unique to the core region; the shell had similar densities of all synaptic types between the two groups (Figure 4H–J).

Large, elaborate multi-perforated synapses (Figure 4K,L) were found in both groups with the same frequency. These enormous synapses have not been reported in other species to our knowledge or in other regions of the human brain. Increased spine volume, synapse length, and perforations are all associated with increased synaptic activity (Calverley et al., 1990; Hering and Sheng, 2001; Yuste and Bonhoeffer, 2001). Thus, these synapses could be a manifestation of the complex interconnectivity of the nucleus accumbens with many other brain regions. Since they were present in both groups, they may be a unique characteristic of the human nucleus accumbens, albeit independent of the disease state in schizophrenia.

The increase in density of asymmetric axospinous synapses indicates an increase in glutamatergic-type inputs to the core. This is consistent with our vGLUT2 findings and in the synaptic data in the dorsal striatum described above. In agreement with this finding is evidence for an increased expression of excitatory postsynaptic density proteins after ketamine stimulation and in the NAcc in postmortem schizophrenia (Kajimoto et al., 2003; Iasevoli et al., 2007). Rodent models of schizophrenia also exhibit increased indices of glutamatergic-type input in the NAcc. Rats treated with psychotomimetics have increased mRNA expression of excitatory postsynaptic density genes (Das et al., 2019) and increased dendritic spine density in the NAcc (Flores et al., 2007; Robinson and Kolb 1997, 1999), both of which are tightly correlated with the density of asymmetric axospinous synapses (Li et al., 2003). A likely source of this elevated input in schizophrenia and its functional implications are discussed below.

One of the major targets of the cortex is the striatum. Psychotomimetics increase striatal spine density (Li et al., 2003; Flores et al., 2007; Robinson & Kolb, 1996, 1997) and increase striatal neuron firing (Dietz et al., 2009; White et al., 2005), linking psychosis to excessive striatal glutamate levels and spine number. In adult rodents, intermittent chronic exposure to PCP results in elaboration of the dendritic arbor and increased dendritic spine density in striatum (Robinson & Kolb, 1997, 1999). These changes are persistent, as they remain one month after treatment termination, and appear to be specific to brain regions that receive dopamine innervation because no differences in spine density between PCP and saline-treated groups are observed elsewhere (Field et al., 2011). These agents given to animals increase striatal glutamate levels and spine density on striatal neurons, linking psychosis with increased striatal glutamatergic synapses (Flores et al., 2007). Spine number is highly coupled to glutamatergic synapse number and vice versa.

The ventral striatum/ nucleus accumbens is a region which is typically activated during task tapping into reward processing (Daniel and Pollmann, 2014), and a large body of reward studies in schizophrenia have reported abnormal engagement of this region (Chase et al., 2018; Radua et al., 2015). In schizophrenia, abnormal prediction error signal in this region was correlated with glutamate levels measured in the substantia nigra, suggesting that glutamatergic dysfunction might contribute to abnormal reward activation (White et al., 2015).

2.5. Substantia Nigra/Ventral Tegmental Area

The substantia nigra/ventral tegmental area (SN/VTA) houses one of the largest clusters of dopaminergic cells in the brain (Hökfelt et al. 1973; Fallon et al. 1978).The SN/VTA is composed of distinct anatomical subregions, each with different connections and neurotransmitter phenotypes (McRitchie et al., 1996; Haber, 2000; Gorelova et al., 2012; Root et al.,2016). Elevated striatal dopamine via the nigrostriatal pathway is associated with psychosis (see review by Howes et al. 2009). The preponderance of evidence indicates that dopamine abnormalities are pathway specific and underlie different symptoms (for review see Weinstein et al. 2017). The SN/VTA in schizophrenia displays abnormal glutamate receptor density (Mueller et al. 2004), hyperplasia of mitochondria in axon terminals (Kolomeets and Uranova 1999), myelin pathology and cytoskeletal abnormalities (Walker et al. 2017). Moreover, dopaminergic neurons have larger nucleoli, increased transcriptional and synaptic activity (Williams et al. 2013. A number of studies indicate increased dopamine activity, protein, and/or mRNA levels (Toru et al. 1988; Howes et al. 2013; Williams et al. 2013; Schoonover et al. 2017), while others do not (Ichinose et al. 1994; Perez-Costas et al. 2012; Rice et al., 2016). The reason for these discrepancies could be methodology, location of the SN sampled or differences in patient populations.

We recently examined the SN/VTA in postmortem tissue using protein analysis (Schoonover et al., 2017), immunohistochemistry and electron microscopy (Mabry et al., 2018). We found that protein levels of vGLUT2 were different in a comparison of NCs, and patients on or off APDs (Figure 5A), (Schoonover et al., 2017). Specifically vGLUT2 levels were increased in off drug patients compared to NCs. While vGLUT1 levels were not altered in the schizophrenia group as a whole, the relationship of vGLUT1 with GAD67 was opposite in in the central region (on a mediolateral axis) of NCs versus the schizophrenia group (Figure 5B) (Mabry et al., 2018). The central region of the SN projects to the central, associative region of the striatum (Haber et al. 2000; Haber 2014) and this region exhibits an increase in dopamine release in schizophrenia (Kegeles et al. 2010). Thus, it appears that there is an abnormal relationship between glutamate and GABA in schizophrenia.

Figure 5.

The Substantia Nigra. A) Sample western blots of vGLUT2 protein in controls (NC) and the schizophrenia group divided by APD status at the time of death: on-drug (SZ-on), and off-drug (SZ-OFF). Graph showing between group differences in the optical density of vGLUT2; posthoc analysis showed that the SZ-off group has significantly higher protein levels of vGLUT2 than did the NCs. Blots and graphs were modified from Figures 1A and 2E, respectively, in Schoonover et al., 2016. B) Regional analysis of correlations between vGLUT1 and GAD67 for NCs and schizophrenia patients. A two correlation coefficient test showed a significant difference between groups (two-tailed, p<0.023); modified from Figure 5B in Mabry et al., 2018. C,D) Electron micrographs from the substantia nigra. D) modified from Figure 7C in Mabry et al., 2018. C,D Scale bars=500nm. E) Graph comparing the synaptic density per 100μm² of all synapses, asymmetric synapses and symmetric synapses; modified from Figure 8A in Mabry et al., 2018. Patients with schizophrenia had a trend toward fewer total synapses and had significantly fewer inhibitory synapses than controls.

Interestingly, we observed a wide variety of different synaptic morphologies that had postsynaptic densities that were characteristic of excitatory synapses, inhibitory synapses, or had PSDs of intermediate thickness (Figure 5C,D) (Mabry et al., 2018). Moreover, many synapses had presynaptic densities (Figure 5D). Quantification of synaptic density revealed a decrease in symmetric, or inhibitory synapses in patients, and a trend toward more synapses overall (Figure 5E). The neurochemical organization of the SN/VTA region is more complicated than other regions of the brain. Recent studies have shown that neurons in the SN/VTA contain multiple transmitters, and use more than one transmitter in a given synapse (Barker et al. 2016; Root et al. 2016), which could lead to the variability in synaptic morphologies. Immunohistochemical localization of glutamate and GABA at the electron microscopic level would be required to identify and parse out the function of the synapses with unusual morphologies. Our data shows an increase in glutamate from subcortical sources, an imbalance of excitatory and inhibitory relationships, and fewer inhibitory synapses in patients.

The investigation of the SN in schizophrenia patients using MRI and MRS spectroscopy imaging techniques has only begun recently (Bolding et al. 2013). Functional MRI reports show disturbed functional connectivity (Hadley et al 2014), which is correlated with increased dopamine synthesis capacity (Howes et al. 2013; Watanabe et al. 2014). MRS spectroscopy indicates that Glx, a combined measure of glutamate and glutamine, is elevated in schizophrenia, suggesting excessive glutamate release (White et al., 2015). Schizophrenia patients have a higher maximum signalling of neuromelanin, suggesting increased dopaminergic neuronal activity (Graham, 1779; Shibata et al. 2008; Watanabe et al. 2014). Moreover, there is a different distribution of neuromelanin density in the SN, suggesting heterogeneity in dopamine turnover within the SN. (Mabry et al., 2018). The SN and other areas of the basal ganglia display hyperactivity during cognitive tasks (Yoon et al. 2013), which is correlated with symptom severity (Yoon et al. 2014). Increased excitatory neurotransmission and/or decreased inhibitory neurotransmission in the SN could underlie these imaging findings.

3. Summary:

Glutamate release, as measured by the vGLUTs, and glutamatergic synapses, as measured by morphologically defined synapses, are differentially affected in different brain regions in schizophrenia. Overall decreases in synaptic density in the schizophrenia group in the ACC are consistent with the reduced neuropil hypothesis, which posits fewer cortical synaptic connections (Selemon and Goldman-Rakic, 1999). Fewer axospinous synapses in the ACC in schizophrenia are consistent with previously reported spine loss (Glantz and Lewis, 1999). Fewer mitochondria in specific populations of axon terminals and in the somata of neurons suggest abnormal energy levels in those structures. However, what happens in the cortex, appears to stay in the cortex, because there are multiple lines of evidence of increased glutamatergic release and density of excitatory synapses in all regions of the basal ganglia that have been studied. This evidence comes from ultrastructural studies, other postmortem methodologies and is confirmed with in vivo imaging studies.

Figure 6.

Summary diagram synthesizing the findings. Results from our ultrastructural, light microscopic and protein studies have shown very discrete changes in schizophrenia. In the anterior cingulate, decreases in excitatory and inhibitory synapses, and layer specific changes in mitochondrial density occur. In the caudate, putamen and nucleus accumbens increases in the number of excitatory synapses as well as levels of vGLUT2 were present in specific compartments. The substantia nigra showed an increase in vGLUT2 protein, a decrease in number of inhibitory synapses and an altered relationship between GABA and glutamate.