A bioinformatic inquiry of the EAAT2 interactome in postmortem and neuropsychiatric datasets

Abstract

Altered expression and localization of the glutamate transporter EAAT2 is found in schizophrenia and other neuropsychiatric (major depression, MDD) and neurological disorders (amyotrophic lateral sclerosis, ALS). However, the EAAT2 interactome, the network of proteins that physically or functionally interact with EAAT2 to support its activity, has yet to be characterized in severe mental illness. We compiled a list of “core” EAAT2 interacting proteins. Using Kaleidoscope, an R-shiny application, we data mined publically available postmortem transcriptome datasets to determine whether components of the EAAT2 interactome are differentially expressed in schizophrenia and, using Reactome, identify which interactome-associated biological pathways are altered.

Overall, these “look up” studies highlight region-specific, primarily frontal cortex (dorsolateral prefrontal cortex and anterior cingulate cortex), changes in the EAAT2 interactome and implicate altered metabolism pathways in schizophrenia. Pathway analyses also suggest that perturbation of components of the EAAT2 interactome in animal models of antipsychotic administration impact metabolism. Similar changes in metabolism pathways are seen in ALS, in addition to altered expression of many components of the EAAT2 interactome. However, although EAAT2 expression is altered in a postmortem MDD dataset, few other components of the EAAT2 interactome are changed. Thus, “look up” studies suggest region- and disease-relevant biological pathways related to the EAAT2 interactome that implicate glutamate reuptake perturbations in schizophrenia, while providing a useful tool to exploit “omics” datasets.

Introduction

Excitatory amino acid transporter 2 (EAAT2) is the primary glutamate transporter in the brain, responsible for uptake of approximately 90% of glutamate in synaptosome preparations (Tanaka et al., 1997). The glutamate transporter protein complex, the EAAT2 interactome, or “glutamosome” (Gegelashvili and Bjerrum, 2019; Shan et al., 2012), describes a complex of proteins that co-localizes, physically interacts or functionally couples with EAAT2 to support the metabolic demands of glutamate transport (Robinson and Jackson, 2016). The clustering of EAAT2 and its interacting proteins is also hypothesized to contribute to the formation of “glutamate microdomains” (Shan et al., 2012). In these specialized extracellular regions, pools of glutamate are tightly regulated, facilitating excitatory neurotransmission by modulation of extrasynaptic glutamate receptors (McCullumsmith et al., 2014).

A number of studies have identified EAAT2 interacting proteins including various Na⁺/K⁺ATPase subunits (Rose et al., 2009; Shan et al., 2014), astrocytic SNARE protein vesicle associated membrane protein 3 (VAMP3) (Li et al., 2015) and presenilin 1 (Zoltowska et al., 2018). EAAT2b, a PDZ-domain containing isoform of EAAT2, interacts with NMDA receptors through postsynaptic density-95 protein (Gonzalez-Gonzalez et al., 2009). Exploratory approaches have identified from dozens to over a hundred EAAT2-associated proteins, in different species and tissue substrates, in samples that are typically enriched for EAAT2 and then analyzed using shotgun proteomic methods (Foster et al., 2018; Genda et al., 2011; Piniella et al., 2018; Shan et al., 2014). EAAT2 co-compartmentalizes with proteins involved in diverse biological processes, supporting functional coupling of EAAT2 and energy metabolism (glycolysis, oxidative phosphorylation) (Genda et al., 2011; Jackson et al., 2015; Shan et al., 2014), metabolic pathways and excitatory signaling, as reviewed (Robinson and Jackson, 2016).

The glutamate hypothesis of schizophrenia suggests a global perturbation of the glutamatergic transmission system in the pathophysiology of this disorder, including NMDA receptor dysfunction and dysregulation of EAAT2 expression and localization (Krystal et al., 2017; O’Donovan et al., 2017). Evidence of changes in components of the human EAAT2 interactome were found in the dorsolateral prefrontal cortex (DLPFC) in subjects with schizophrenia (Shan et al., 2014). Significant increases in expression of metabolic enzyme aconitase 1 and EAAT2b were found in mitochondrial and extrasynaptic DLPFC fractions, respectively, in schizophrenia. Hexokinase expression, the initial enzyme of glycolysis, was significantly reduced in the mitochondrial fraction relative to the extrasynaptic fraction (Shan et al., 2014). In an earlier study, hexokinase attachment to the outer membrane of mitochondria was reported to be reduced in the parietal cortex in schizophrenia (Regenold et al., 2012). Dysregulation of the localization and expression of EAAT2 and EAAT2 interacting proteins, particularly those associated with energy metabolism, are evident in schizophrenia. However, the EAAT2 interactome has yet to be studied in severe mental illness.

In this study we 1) Identify a “core” list of EAAT2 interacting proteins found in at least two studies that have examined the EAAT2 interactome; 2) Identify biological pathways associated with this core interactome; 3) Exploit the wealth of publically available postmortem transcriptomic data to determine whether the components of the EAAT2 interactome are altered in schizophrenia and whether such changes extend to other neuropsychiatric and neurological disorders; 4) We confirm whether changes in this core interactome are found in a proteomic dataset in schizophrenia and 5) Finally, we examine whether the EAAT2 interactome associated biological pathways implicated in schizophrenia are similarly altered in a mouse model of antipsychotic administration. Overall, we apply a bioinformatic approach to identify changes in components of the core EAAT2 interactome in schizophrenia and present a workflow that can be utilized to examine other biological networks in postmortem neuropsychiatric research.

Materials and Methods

1.1. Identifying key components of the EAAT2 interactome

The core EAAT2 interactome list was compiled from a number of discovery-based studies that have published EAAT2 interacting protein data. Proteins identified in at least two of these studies are considered core components of the EAAT2 interactome. In total, four studies that characterize EAAT2 interacting proteins using different methods were identified (Table 1) and the published lists of EAAT2 interacting proteins were used for our in silico analysis. In human DLPFC, EAAT2 interacting proteins were identified following EAAT2 immunoisolation and mass spectrometry analysis of proteins in the EAAT2-enriched immunoprecipitate (Shan et al., 2014). A similar strategy was initially deployed in rat cortex (Genda et al., 2011). Seventy-four EAAT2 interacting proteins were identified in at least 2 of the 3 experiments carried out to characterize the interactome in the rat brain. A proximity based biotinylation assay (BioID) was used to identify proteins that interact with EAAT2 in HT22 cells (Piniella et al., 2018). Finally, gliosomes were prepared from the forebrains of mice treated with a pyridizine-based compound that increases translation of EAAT2 (Foster et al., 2018). Proteins with altered expression following treatment were identified by mass spectrometry. We compiled a list of 24 proteins found to interact or be associated with EAAT2 in at least 2 of the 4 EAAT2 interactome studies.

Table 1.

Summary of studies that characterize EAAT2 interacting proteins. DLPFC dorsolateral prefrontal cortex, GLT1 rodent EAAT2 homolog, IP immunoprecipitation, LC-MS/MS liquid chromatography coupled to mass spectrometry.

Study	Tissue	Methods	No. proteins
Shan et al. (2014)	Postmortem DLPFC	EAAT2 immunoisolation; Gel bands above 50kDa excised for LC-MS/MS	62
Genda et al. (2011)	Rat cortex	EAAT2 immunoisolation 1)All IP protein gel bands 2)20 fractions of gel lane labelled by anti-GLT1 antibody 3)GLT1 IP of supernatant fraction, 20 fractions of gel lane labelled by anti-GLT1 antibody; LC-MS/MS	74 proteins identified in at least 2 of 3 experiments
Foster et al. (2018)	Mouse forebrain	Mice administered pyridizine derivative LDN/OSU-0215111 Gliosome fraction; LC-MS/MS and microarray	157
Pinella et al. (2018)	GLT-1 transfected HT22 cells	Proximity based biotinylation assay (BioID)	24

Protein accession numbers were converted to gene symbols using the Uniprot Retrieve/ID mapping tool (https://www.uniprot.org/uploadlists). Na⁺/K⁺ ATPase consists of multiple subunits encoded by more than one gene. Several of these subunits were identified in different interactome studies, however, subunit ATP1A3 was common to 2 studies and is the isoform/gene symbol searched here. Proteins with broad descriptors in the published protein lists e.g. synapsin was converted to more than one gene symbol e.g. synapsin 1 (SYN1) and synapsin 2 (SYN2). This reduced stringency of protein identification and allowed for greater comparison of common proteins across studies using different species and methodologies. The full list of protein names and corresponding gene symbols is listed in Figure 2A.

1.1. Core EAAT2 interactome pathway analysis.

The EAAT2 interactome gene list was searched in Gene Ontology (http://geneontology.org/) and GOslim Biological Processes, Molecular Function and Cellular Components were generated. The top 10 findings with p-value <0.05 and false discovery rate (FDR) <0.05 were assembled. To generate biological pathways associated with the EAAT2 interactome, the gene list was searched in Reactome v70 (https://reactome.org/). The top findings, with a minimum of 4 entities (genes), p-value <0.05, FDR <0.05 were generated. Pathways are organized under their top-level hierarchical identifier e.g. metabolism. Pathways are then identified at multiple sub-levels. To reduce the number of redundant pathways whilst still allowing for meaningful interpretation of biological pathways, some sub-level pathways are contracted.

Finally, using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database (v11.0) https://string-db.org/, we generated an EAAT2 protein network (search term SLC1A2; homo sapiens). This network is based on evidence from multiple active interaction sources (text mining, experiments, databases, co-expression, neighborhood, gene fusion and co-occurrence). A network of 20 first shell interactors with an interaction score of medium confidence (0.400) was generated.

1.2. EAAT2 interacting protein “look up” studies

“Look-up” studies of the EAAT2 core interactome genes were conducted in Kaleidoscope (http://bpg-n.utoledo.edu:3838/CDRL/Kaleidoscope/). Kaleidoscope is an R shiny application that collates publicly available postmortem transcriptome data from neuropsychiatric and neurological disorders (Moody et al., 2019; Sullivan et al., 2019c). We searched schizophrenia, amyotrophic lateral sclerosis (ALS) and major depression datasets, as altered EAAT2 expression was reported in these illnesses (Lauriat and McInnes, 2007; O’Donovan et al., 2017). The EAAT2 interactome gene list was searched in a multi-region microarray dataset from subjects with schizophrenia (n=21) and controls (n=19) (Roussos et al., 2012). The brain regions searched are DLPFC (BA46), anterior cingulate cortex (ACC; BA32), mediotemporal area (MTA; BA21) and temporopolar area (TPA; BA 38). Over 70 postmortem schizophrenia transcriptome studies have been carried out in different tissues and using different methodologies. This includes well-powered RNAseq studies of the DLPFC using tissue from hundreds of subjects (Fromer et al., 2016; Jaffe et al., 2018) and a meta-analysis of multiple GSE datasets from frontal and parietal cortex regions (Gandal et al., 2018). We selected the dataset from Roussos et al. as a representative dataset that analyzed multiple brain regions from schizophrenia subjects. We also searched RNAseq (GSE57821) data from forebrain neurons derived from induced pluripotent stem cells (iPSCs) from a schizophrenia patient with a disrupted in schizophrenia (DISC1) mutation (Wen et al., 2014). NeuroLINCS RNAseq data (http://neurolincs.org/) from motor neurons derived from iPSCs from a subject with ALS was also searched (Thompson, 2017). iPSCs contain the donor subjects entire genetic background. They are a particularly useful model for the study of neuropsychiatric and neurological disorders as they recapitulate the genetic risk that contributes to the development of these disorders (Soliman et al., 2017). In conjunction with postmortem tissue, which is often obtained from the endpoint of disease, iPSCs are a powerful tool for modelling the development of pathological features in human disease. The DISC1 mutation is a rare, highly penetrant 4-base pair frame shift mutation that is associated with the development of schizophrenia in the affected individual (Wen et al., 2014). This model offers insight into the genetic susceptibility and the associated synaptic dysfunction that underlies the onset of schizophrenia in the affected individuals.

The EAAT2 interactome gene list was also searched in an RNAseq dataset (GSE102556) of the DLFPC from male and female subjects with major depression (n=26) and controls (n=22) (Labonte et al., 2017). The subset of EAAT2 interactome genes with a ≥1.2 fold change (FC) or ≤ −1.2 FC in the “look up studies” were searched in Reactome. We present the top-level hierarchical biological pathways associated with the EAAT2 interactome in each disorder, which have a minimum of 3 entities (genes), and p-value cutoff <0.05.

Finally, we compiled a list of proteins that were differentially expressed in proteomic studies of postmortem schizophrenia (Focking et al., 2015; Martins-de-Souza et al., 2010a; Martins-de-Souza et al., 2010b; Saia-Cereda et al., 2017; Schubert et al., 2015; Velasquez et al., 2019; Wesseling et al., 2013) to determine whether the core EAAT2 interacting proteins have altered protein expression in schizophrenia. Although not independently validated, this (non-exhaustive) list describes protein changes in several different brain region or sub-regions in schizophrenia.

1.3. EAAT2 interacting protein “look up” of antipsychotic studies

To address the effects of antipsychotic medication on components of the EAAT2 interactome, we analyzed a dataset generated in an animal model of antipsychotic administration (Rizig et al., 2012). Microarray analysis of mouse forebrain following four weeks of clozapine or haloperidol administration (GSE6511) or 12 weeks of clozapine or haloperidol administration (GSE6467) in animal drinking water was searched in Kaleidoscope. The EAAT2 interacting targets that are significantly altered at ≥1.2 FC or ≤ −1.2 FC were identified and pathway analysis was carried out to determine the effect of medication on EAAT2 interactome associated biological pathways.

Results

1.1. EAAT2 “core” interacting proteins

The bioinformatic workflow used to identify and probe the EAAT2 interactome in schizophrenia is described in Figure 1. We examined the EAAT2 interactome characterized in four different studies (Table 1) and assembled a list of “core” EAAT2 interacting proteins that were identified in at least two of these studies. The EAAT2 interactome was studied in rat cortex (Genda et al., 2011), human DLPFC (Shan et al., 2014), mouse forebrain gliosomes following administration of a drug that increases EAAT2 translation (Foster et al., 2018) and in murine hippocampal neuronal (HT22) cells transfected with GLT-1 (Piniella et al., 2018). In addition, different methodologies were applied to identify EAAT2 interacting proteins, including immunoisolation coupled to mass-spectrometry and BioID. Despite species and methodological differences, a core interactome was identified and is listed in Figure 2A. The number of components of the EAAT2 interactome contributed by the four different interactome studies are shown in Figure 2B.

Figure 1.

Bioinformatic workflow. 1) A list of core EAAT2 interacting proteins was compiled from published EAAT2 interactome datasets. 2) The core EAAT2 interacting protein list is searched in Gene Ontology to generate molecular function, cellular component and biological processes and in Reactome to generate associated biological pathways. Using STRING, a predicted EAAT2 protein interaction network is generated. 3) “Look-up” studies are conducted in Kaleidoscope to determine whether components of the core EAAT2 interactome are altered in different disorders or in response to antipsychotic medication. 4) Biological pathways were generated from those components that are altered in the “look-up” studies. ACC anterior cingulate cortex, DLPFC dorsolateral prefrontal cortex, FC frontal cortex, GLT1 rodent excitatory amino acid transporter 2 homologue, MTA mediotemporal area, MNs motor neurons, PNs pyramidal neurons, PFC prefrontal cortex, TPA temporopolar area.

Figure 2.

The EAAT2 interactome. A) List of “core” EAAT2 interacting proteins, identified in at least 2 studies of the EAAT2 interactome. B) The number of common proteins identified in each of the published EAAT2 interactome studies. C) Gene ontology analysis of the list of EAAT2 interacting proteins. D) Biological pathway analysis of core EAAT2 interacting proteins. E) The network of EAAT2 interacting proteins identified by STRING analysis.

GOslim analysis of the core EAAT2 interactome (Figure 2C) identifies metabolism-related biological processes and purine nucleotide and catalytic activity related molecular functions. The top 10 components (p<0.05) were identified. The top biological pathways (Figure 2D) associated with the EAAT2 interactome are metabolism-related, including energy metabolism pathways “glycolysis” and “the citric acid cycle”. The immune system, signal transduction and plasticity related pathways are also implicated.

Using STRING, we generated a 20-node protein network based on archived EAAT2 interactions (Figure 2E). However, few of the proteins identified in the STRING network are identified in the core EAAT2 interacting protein list. Thus, direct biochemical assays identify different interactors than the coexpression and text mining data sources primarily used to generate the STRING network. Common proteins from STRING network analysis and the EAAT2 interactome list are limited to EAAT2 and ribosomal protein family members.

1.2. EAAT2 “core” interacting proteins in disease

Using the list of core EAAT2 interactome proteins, we searched Kaleidoscope for the corresponding genes in postmortem microarray datasets from subjects with schizophrenia. The dataset assays gene expression in four postmortem brain regions: DLPFC, ACC, MTA and TPA. The greatest number of EAAT2 interactome gene changes (FC ≥ 1.2 or ≤ −1.2) were found in the ACC (13/32 genes) and DLPFC (10/32 genes) with relatively few changes found in the MTA (5/32) and TPA (3/32). Using the subset of EAAT2 interactome genes that surpassed the cutoff, biological pathway analysis identified “metabolism” and “neuronal system” genes across most brain regions (Figure 3A). Not enough genes were identified to carry out pathway analysis for the TPA. Interestingly, “carbohydrate metabolism” and “glycolysis” was identified in all regions except the DLPFC, where “metabolism of amino acids” was the main metabolic pathway found. “Metabolism of Proteins-Translation” and “Metabolism of RNA” pathways were unique to the DLPFC. The relevance of the biological pathways in the MTA and TPA is limited by the small number of genes input into the analysis. All pathways had a minimum “entities found” cutoff of 3 genes, except MTA (2 entities; SCZ analysis), DLPFC Fm (2 entities; MDD analysis) and antipsychotic analysis (2 entities). However, this data suggests that alterations in the EAAT2 interactome in frontal cortex regions DLPFC and ACC may be particularly relevant to schizophrenia. All pathways identified in postmortem brain regions were also found in DISC1-PNs. Additionally, “vesicle-mediated transport” was found in DISC1-PNs only.

Figure 3.

A) Biological pathways associated with components of the EAAT2 interactome that are altered in disease or following antipsychotic administration. B) EAAT2 interacting genes that are altered by +/− 1.2 FC and included in biological pathway analysis. Additional genes, altered +/− 1.15 FC are also shown. ACC anterior cingulate cortex, ALS amyotrophic lateral sclerosis, CLZ clozapine, DLPFC dorsolateral prefrontal cortex, DISC1-PNs pyramidal neurons, FC fold change, Fm female, HAL haloperidol, iPSC-MNs motor neurons derived from induced pluripotent stem cells, M male, MDD major depressive disorder, MTA mediotemporal area, TPA temporopolar area.

Changes in EAAT2 interactome genes were compared and contrasted in other disorders with reported dysregulation of EAAT2 expression including ALS and major depression (Lauriat and McInnes, 2007; O’Donovan et al., 2017; Parkin et al., 2018). EAAT2 expression was altered 2 FC in motor neurons derived from iPSCs in ALS and other EAAT2 interacting genes were also altered by at least 1.2 FC (Figure 3B). Following pathway analysis, “metabolism” and “metabolism of proteins”, were implicated in ALS (Figure 3A). However, despite gene expression changes of over 1.2 FC in EAAT2 in female and male depressed subjects in the DLPFC, too few changes in EAAT2 interacting genes were found in major depression to conduct pathway analysis, suggesting that EAAT2 interactome changes are less relevant in this disorder.

Using published postmortem proteomic datasets, we determined that 42% (10/24) of the proteins identified in the “core” EAAT2 interactome were also altered in postmortem schizophrenia, whereas 27% (53/196) of the proteins identified in the individual EAAT2 interactome studies had altered protein expression. This suggests that proteins that form the “core” EAAT2 interactome show greater perturbation in protein expression in schizophrenia.

1.3. EAAT2 “core” interacting proteins in antipsychotic dataset

Following four weeks of haloperidol or clozapine administration, significant changes in components of the EAAT2 interactome were found in mouse forebrain, although EAAT2 expression was not altered (+/−1.2 FC) by either drug (Figure 3B). Fewer gene changes were found after 12 weeks of antipsychotic administration and gene expression was primarily downregulated at this time point. The biological pathway “metabolism”, one of the main pathways implicated in disease states, is identified following 12 weeks but not 4 weeks of chronic haloperidol and clozapine administration (Figure 3A). The difference in patterns and valence of gene expression at 12 weeks highlight the importance of chronic drug administration paradigms, to more closely represent chronic medication treatment courses for schizophrenia patients.

Discussion

EAAT2 buffers and transports glutamate, removing it from the synapse. As well as terminating the action of glutamate in the synapse and preventing potentially neurotoxic spillover, EAAT2 plays a role in shaping synaptic plasticity and is essential for maintaining normal brain function (Murphy-Royal et al., 2015; Tzingounis and Wadiche, 2007). Indeed, homozygous mice deficient in GLT1 experience lethal spontaneous seizures due to a lack of glutamate clearance and increased glutamate levels (Tanaka et al., 1997). Glutamate uptake is a high-energy demand process and the activity of EAAT2 is supported by a network of metabolic and structural proteins. Altered expression and localization of EAAT2 is implicated in neuropsychiatric and neurological disorders like schizophrenia and ALS (O’Donovan et al., 2017). However, the complex of interacting proteins that form the EAAT2 interactome is not well studied in these illnesses.

Using an in silico approach, we generated a list of EAAT2 interacting proteins that were identified in at least two studies that attempted to characterize the EAAT2 interactome (Foster et al., 2018; Genda et al., 2011; Piniella et al., 2018; Shan et al., 2014). Biological pathway analysis of this “core” list identified chemical transmission, immune, metabolism and bioenergetic-related pathways, which are functionally coupled to EAAT2 (Genda et al., 2011; Jackson et al., 2015). Interestingly, the EAAT2 protein-protein interaction network drawn using STRING, with most EAAT2 interactions derived from text mining and coexpression studies, does not contain many of the proteins identified by direct study of the EAAT2 interactome (Foster et al., 2018; Genda et al., 2011; Piniella et al., 2018; Shan et al., 2014). However, evidence of a functional link, derived from experimental data in STRING, implicates interaction between EAAT2 and ribosomal proteins (Wan et al., 2015). Ribosomal proteins were also identified in the “core” interactome list, suggesting a highly conserved interaction.

To identify whether expression of the “core” EAAT2 interacting proteins are perturbed in schizophrenia, we conducted “look up” studies in published postmortem transcriptome datasets. Although less pertinent to the study of protein interactions, curated transcriptome studies are publically available and contain data on all genes of interest, unlike proteomic datasets, which are typically limited to identifying, at most, a few thousand proteins. We utilized a microarray dataset that examines four different brain regions in schizophrenia (Roussos et al., 2012).

The greatest number of expression changes (± 1.2 FC) in components of the EAAT2 interactome in schizophrenia were found in the ACC and DLPFC, and in the iPSC DISC1-PN dataset. Postmortem studies have identified region-, lamina- and cell-level changes in EAAT2 and EAAT2 splice variants in the ACC (Katsel et al., 2011; O’Donovan et al., 2015). In the DLPFC, altered post-translational modification (glycosylation) but not total protein or gene expression of EAAT2 was reported (Bauer et al., 2008; Bauer et al., 2010; O’Donovan et al., 2015). EAAT2 expression changes have not been directly assayed in temporal areas MTA or TPA although reduced EAAT2 protein was found in the superior temporal gyrus (Shan et al., 2013). Pathway analysis suggests that the EAAT2 interactome is altered in a region-specific manner in schizophrenia, with dysfunction in a broader range of pathways in frontal cortex regions than temporal areas. “Transmission across chemical synapses” pathway was a commonly identified pathway across studies, as altered EAAT2 expression and activity in disease impacts glutamate transmission (Tzingounis and Wadiche, 2007). Altered functional coupling of glucose metabolism and EAAT2 is implicated in several brain regions in schizophrenia and in iPSC DISC1-PNs and MNs in schizophrenia and ALS. IPSCs generated from disease population often display extensively altered gene expression profiles compared to control populations (Wen et al., 2014).

Proteomic studies also broadly implicate altered energy metabolism pathways in neuropsychiatric disorders (Zuccoli et al., 2017). This includes findings of altered metabolism at in the DLPFC (Pennington et al., 2008; Prabakaran et al., 2004) and ACC (Beasley et al., 2006; Clark et al., 2006, 2007; Martins-de-Souza et al., 2010b) but not the postsynaptic density (Focking et al., 2015) in schizophrenia. However, EAAT2 protein expression was not significantly altered in these proteomic studies, in line with previous reports (Bauer et al., 2008). Thus, perturbation of EAAT2 interacting proteins and altered localization of EAAT2, rather than total protein expression changes, may be related to the pathophysiology of schizophrenia (O’Donovan et al., 2017; Shan et al., 2014).

PET studies suggest that EAAT2 activation induces glucose uptake in the brain (Zimmer et al., 2017). Perturbed glucose utilization is implicated in neuronal but not astrocyte cell populations in the DLPFC in schizophrenia (Sullivan et al., 2019a). Increased lactate, produced in astrocytes and shuttled to neurons in times of bioenergetic demand to support energy production in the TCA cycle, is also found (Dean et al., 2016; Sullivan et al., 2019b). Thus, the pathways associated with dysregulation of EAAT2, and components of its interactome, may reflect the bioenergetic deficits, primarily via glucose usage, found in schizophrenia.

Dysregulation of EAAT2 is found in other psychiatric and neurological disorders. In ALS, EAAT2 protein expression was decreased in the motor cortex and spinal cord and glutamate uptake was also reduced in the brain (Bristol and Rothstein, 1996; Fray et al., 1998; Rothstein et al., 1992; Rothstein et al., 1995). In “look up” studies of RNAseq analysis of MNs derived from iPSCs from a subject with ALS, EAAT2 gene expression and a number of EAAT2 interacting genes (Thompson, 2017) were altered. “Metabolism” pathways including “glycolysis” were identified in ALS. Deficits in energy metabolism is a feature of ALS, with motor neurons in this disorder being particularly susceptible to energetic stress (Vandoorne et al., 2018). Reduced EAAT2 gene and protein expression is reported in MDD hippocampal and frontal cortex (DLPFC, ACC) brain regions (Choudary et al., 2005; Medina et al., 2013; Miguel-Hidalgo et al., 2010), with increased levels in non-depressed subjects who die by suicide compared to depressed subject who die by suicide (Klempan et al., 2009; Sequeira et al., 2009). EAAT2 gene expression was reduced in the DLPFC in male and female subjects in the dataset reported here (Labonte et al., 2017) but few other gene changes in components of the EAAT2 interactome were found. Perturbations of the EAAT2 interactome may not be implicated in the neurobiology of depression.

Animal models of antipsychotic administration cannot recapitulate the effects of lifelong medication use in severe mental illness. However, transcriptome analysis of tissue from these animals offers a useful tool to understand the effects of antipsychotics on gene networks and biological pathways that may be altered following medication administration. Distinct differences in patterns of gene expression were found in mice administered the antipsychotics haloperidol or clozapine for four weeks compared to twelve weeks (Rizig et al., 2012). Haloperidol, a typical, first generation antipsychotic and clozapine, an atypical, second generation antipsychotic, have differences in mechanisms of action (Meltzer, 2013). However, EAAT2 gene expression was not significantly altered (+/− 1.2 FC) by either drug at any time point. Previous reports also found no change in frontal cortex expression of EAAT2 (GLT-1) in rats treated chronically (9 months) with haloperidol-decanoate (O’Donovan et al., 2015). At the 12-week time point, fewer changes in expression of EAAT2-associated genes were induced by antipsychotics, and no gene changes were common to both drugs. “Metabolism-the citric acid cycle” pathways are implicated, at 12-weeks only, by the gene changes induced by haloperidol. This suggests that metabolism pathways are also be influenced by chronic treatment with this antipsychotic. Models of antipsychotic administration which better represent chronic courses of treatment, and different classes and types of drugs are essential to understand the unique effects of these pharmacotherapies on the brain transcriptome.

The therapeutic effects of ceftriaxone, a β-lactam antibiotic that stimulates EAAT2 transcription, has been extensively explored in animal models of ALS, Alzheimer’s disease, pain, addiction and others, as reviewed (Yimer et al., 2019). Novel small molecules that upregulate EAAT2 translation have also been developed and successfully tested in neurological models like Alzheimer’s disease (Takahashi et al., 2015b). However, the therapeutic potential of EAAT2-modulators in animal models has yet to translate directly to the clinic. In a randomized, double-blind placebo study testing efficacy and safety of ceftriaxone treatment in ALS, promising stage-2 efficacy was not maintained at stage-3 (Cudkowicz et al., 2014). Despite this, enthusiasm for EAAT2 as a potential target for treatment in neuropsychiatric and neurological disorders remains high (Blacker et al., 2019; Lin et al., 2012; Takahashi et al., 2015a; Yimer et al., 2019). Considering EAAT2 in the context of its protein interactome in future therapeutic strategies may also prove useful. For example, Na⁺/K⁺ ATPase, whose coupling to EAAT2 is required for glutamate transport (Rose et al., 2009), is a member of a highly druggable family of enzymes, ATPases, whose therapeutic potential have being explored in a range of different illnesses (Alevizopoulos et al., 2014; Aperia, 2007; Goldberg et al., 2018).

Limitations and Conclusions:

The postmortem transcriptomic and proteomic datasets probed in this study were carried out in largely region-level brain samples. They are not enriched for EAAT2 and do not specifically target EAAT2-associated interactions. Thus, changes in gene or protein expression may reflect changes in the broader physiological milieu, as well as in the EAAT2 interactome. Equally, altered expression of EAAT2 interactome components may not be detected in heterogeneous region-level samples (McCullumsmith and Meador-Woodruff, 2011). Biological pathways were generated with a relatively small number of input genes and this should be considered when interpreting the relevance of pathway identification.

Bioinformatic analysis is a powerful tool to mine the growing number of neuropsychiatric and postmortem “omics” datasets that are currently available. A strength of this approach is that it provides an overview of the main biological pathways that are associated with expression changes of components of the EAAT2 interactome. This approach also offers insight into the profound effects that chronic medication treatment can have on gene and protein expression and different biological processes in the brain. Examining mouse models of typical and atypical antipsychotic administration show the importance of studying the chronic effects of different classes of drugs on the brain.

However, there are also several limitations to the postmortem datasets that can currently be data mined. There is a relative paucity of proteomic, metabolomic or other functional “omic” datasets e.g. kinomics, relative to transcriptomic datasets. Most datasets analyze whole brain regions or even pool multiple different brain regions together for analysis, with the exception of some cell-level transcriptomic datasets (Arion et al., 2015; Arion et al., 2017; Enwright Iii et al., 2017). Detecting laminar-, cell subtype- or subcellular -specific expression changes remains a challenge. This strong bias towards gene expression data limits what can be determined about complex protein-protein interactions in disease states using this approach.

Despite the importance of EAAT2, which represents as much as 1% of total brain protein (Lehre and Danbolt, 1998), the EAAT2 interactome in disease has not yet been characterized. The predicted EAAT2 protein network, generated in STRING, does not reflect the EAAT2 protein interactions identified following direct biochemical assays designed to characterize the interactome. Thus, while “omics” data mining is a powerful tool, targeted biochemical assays are still necessary to understand the changes that occur in complex protein networks in disease. As additional “omics” datasets are generated and made publically available, the utility of in silico and “look-up” studies will expand.