An Increased Copy Number of Glycine Decarboxylase (GLDC) Associated with Psychosis Reduces Extracellular Glycine and Impairs NMDA Receptor Function

Abstract

Glycine is an obligatory co-agonist at excitatory NMDA receptors in the brain, especially in the dentate gyrus, which has been postulated to be crucial for the development of psychotic associations and memories with psychotic content. Drugs modulating glycine levels are in clinical development for improving cognition in schizophrenia. However, the functional relevance of the regulation of glycine metabolism by endogenous enzymes is unclear. Using a chromosome-engineered allelic series in mice, we report that a triplication of the gene encoding the glycine-catabolizing enzyme glycine decarboxylase (GLDC) - as found on a small supernumerary marker chromosome in patients with psychosis - reduces extracellular glycine levels as determined by optical fluorescence resonance energy transfer (FRET) in dentate gyrus (DG) and suppresses long-term potentiation (LTP) in mPP-DG synapses but not in CA3-CA1 synapses, reduces the activity of biochemical pathways implicated in schizophrenia and mitochondrial bioenergetics, and displays deficits in schizophrenia-like behaviors which are in part known to be dependent on the activity of the dentate gyrus, e.g., prepulse inhibition, startle habituation, latent inhibition, working memory, sociability and social preference. Our results demonstrate that Gldc negatively regulates long-term synaptic plasticity in the dentate gyrus in mice, suggesting that an increase in GLDC copy number possibly contributes to the development of psychosis in humans.

INTRODUCTION

The genetics of schizophrenia are complex, with common genetic variants contributing small effect sizes and rare variants, e.g., genomic copy number variants (CVNs) having large effect sizes(1). Exonic mutations that can help identify pathophysiologically important pathways are extremely rare. In a recent meta-analysis including 24,248 cases and 97,322 controls, rare coding variants in 10 genes were found to increase the risk for schizophrenia(2). Two of these genes were the glutamate receptor subunit genes GRIN2A and GRIA3. Genome-wide association studies identified 270 loci with 130 identified genes associated with schizophrenia(3). Interestingly, 30% of the 130 risk genes encode proteins that cluster around the pre- and postsynaptic components of the glutamatergic synapse and would affect NMDA receptor transmission(4). Moreover, in CNVs associated with schizophrenia genes encoding proteins localized in postsynaptic signaling complexes have been found to be enriched(5). For glutamate to open the NMDA receptor channel, a recognition site on the NR1 channel must be occupied by D-serine or its precursor, glycine. However, the pathophysiological role of these co-agonists in human disease is largely unexplored. When glycine was added to ongoing antipsychotic medication, negative and cognitive symptoms were reduced(6), suggesting a role of glycine in the pathophysiology of schizophrenia and potentially in the therapy of negative and cognitive symptoms. However, little is known about how glycine levels are regulated in the brain. Glycine is catabolized mainly by the glycine cleavage system (GCS) with glycine decarboxylase (GLDC) being the rate-limiting enzyme. In adult brain, GLDC is expressed in astrocytes, but not in neurons(7–9). Loss of function mutations in GCS system proteins including GLDC increase glycine levels are associated with non-ketotic encephalopathy(10–12).

We previously described a mother and son with a CNV in the 9p24.1 chromosomal region encompassing 14 genes, containing full duplication of 13 genes, and a full triplication of the gene for glycine decarboxylase (GLDC), a glycine-catabolizing enzyme(13,14). The structural basis of this CNV was identified to be a small supernumerary marker chromosome(15,16). The patients suffered from psychosis, the symptoms of which were attenuated by treatment with glycine and D-cycloserine, a partial agonist at the NR1 glycine site(14). We used gene targeting, CRISPR-Cas9-mediated recombination and trans-allelic recombination to generate an allelic series of mice to study the effects of the CNV identified in the patients, to analyze first the effects of the entire CNV and then of an increased copy number of the Gldc gene. The CNV leads to schizophrenia-like biochemical and behavioral phenotypes, and to changes in the expression of three genes (Arl3, mir-137 and srr) that have been shown to be associated with schizophrenia by GWAS, indicating that dysfunction of excitatory neurotransmission is an essential proximal event in the pathophysiology of schizophrenia and not merely a byproduct of other changes. Moreover, as Gldc is expressed in the brain in astrocytes but not in neurons(8,9,17), we provide evidence for a potential etiologically primarily non-neuronal (astrocytic) mechanism as a part of the pathophysiology of schizophrenia. Our findings link a mutation found in patients with psychosis to a dysregulation of the NMDA receptor and represent evidence that the enzyme glycine decarboxylase is a pathophysiologically relevant negative regulator of excitatory neurotransmission.

METHODS

Chromosome-engineered mouse models

We developed mouse models with 1–4 copies of 9p24.1 genes using gene targeting, trans-allelic recombination in the germline, and selective breeding to investigate pathophysiological mechanisms underlying the 9p24.1 copy number variants (Fig. 1A). Furthermore, to identify which gene(s) in this CNV contribute to pathophysiology, we developed shorter CNVs using gene editing with CRISPR/Cas-9 and trans-allelic recombination), generating mice with additional copies of Gldc (9R mice) and mice with additional copies of the other 9p24.1 genes (9L mice) (Fig. 3A) (see also Supplementary Methods). Mice were maintained on a C57BL/6J background. The mice were typically housed in groups of four to five per cage in standard conditions (23 ± 1 °C, 50 ± 5% humidity) in a 12-h light-dark cycle (lights on from 07.00–19.00). Food pellets and water were provided ad libitum. Male and female mice approximately 2–4 months of age were used in experiments. Control mice were derived from heterozgous crosses.

Fig. 1. Schematic representation of the generation of the 9p24.1 CNV mice.

A. Generation of the 9p24.1 deletion (Del), duplication (Dup) and triplication (Trip) alleles by selective breeding with targeted loxP sites in trans and a Hprt-Cre transgene, via trans-allelic recombination. B. Comparative genomic hybridization for the duplication (green) and deletion (red) alleles using an Agilent 1M array. C. Representative western blots and quantification of GLDC protein expression in hippocampus of 9p24.1 deletion (Del=1c9LR), duplication (Dup=3c9LR) and triplication (Trip=4c9LR) mice compared to WT (WT=2c9LR) (expressed as percentage of WT, unpaired two-sided Student’s t-test; 4c9LR vs WT; t(5)=3.269, *p<0.05, 1c9LR vs WT; t(5)=6.540, **p<0.01, n=3–4 for each group). D. Golgi staining was performed to determine dendritic spine density in dentate gyrus, suggesting a reduced density in triplication (Trip) mice (n=3–7 dendrites counted in 3 mice per genotype, One-way ANOVA followed by Bonferroni post hoc test, F(3,12)=4.624; p=0.032; 4c9LR vs WT; t=3.341, *p<0.05). E. Reduced and oxidized glutathione (GSH and GSSG, respectively) were measured in prefrontal cortex using HPLC–electrochemical detection. The GSH/GSSG ratio was calculated to indicate redox state and oxidative stress. F. Latent inhibition to conditioned freezing revealed absence of latent inhibition in 9p24.1 deletion (Del) and triplication (Trip) mice, i.e., the percentage time spent freezing during presentation of the tone was not different in mice that were pre-exposed to the tone [30 tones (20 sec, 70 dB, 2,800 Hz) with 30 sec intervals] versus non-pre-exposed mice, whereas pre-exposed WT and 9p24.1 duplication (Dup) mice were freezing less than non-pre-exposed mice (n=4–5 per group, One-way ANOVA followed by Bonferroni’s post-hoc test, F(7,23)=6.344; p=0.0003; WT-pre vs non-pre exposed; t=2.867, *p<0.05, Dup-pre vs non-pre exposed; t= 3.953, **p<0.01). G. Water T-maze non- matching to place task to test working memory. Mice were trained to alternate the arms visited in a water T-maze with typically >70% accuracy. On test days, different delay intervals between forced trials and choice trials were applied (15 s, 25 s and 45 s) (n=5–6/group, Two-Way ANOVA revealed main effect of genotype (F(3,16)= 3.882, p<0.05) and of delay (F(2,32)= 13.48, p<0.001) followed by Bonferroni posthoc test revealed at 45 sec vs 15sec; Dup, t= 4.126, ***p<0.001, and Trip, t= 3.713, **p<0.01). See also Fig. S1.

Fig. 3. Generation and characterization of shorter complementary CNVs from the 9p24.1 CNV (9LR) using CRISPR/Cas-9 genome editing.

A. A loxP site was placed between the Uhrf2 and Gldc genes using CRISPR/Cas-9, and trans-allelic recombination was used as described in Methods to generate duplication alleles of Gldc alone (9R fragment; duplication=3c9R, triplication=4c9R) or of all the genes in the 9LR genomic segments (9L fragment; duplication=3cL, triplication=4cL). B. Representative western blots and quantification of GLDC protein in hippocampus of mice- with 4 copies of GLDC alone (4c9R), 4 copies of all other 12 genes of the 9p24.1 CNV (4c9L) and 4 copies of the entire 9p24.1 CNV(4c9LR) compared to WT (n=5 for each group, One-way ANOVA followed by Bonferroni test, F(3,37)=18.77; ***p<0.001, WT vs 4c9LR (t=6.07; ***p<0.001), WT vs 4c9R (t=5.24;***p<0.001). C. GLDC enzyme activity in hippocampal tissues of WT, 4c9LR, 4c9R and 3c9R (n=3–6, One-way ANOVA followed by Bonferroni test, F(3,16)=13.53; ***p=0.0001), WT vs 4c9LR (t=5.827; **p<0.01), WT vs 4c9R (t=3.366; *p<0.05). D, E. Qualitative fluorescence microscopy images of GLDC protein colocalization with S100beta (D) or Map2 (E) in CA1, CA3 and dentate gyrus regions of hippocampus, respectively, in WT and 4c9R mice. F. BDNF mRNA expression is reduced in whole hippocampus of 4c9R mice compared to WT (One-way ANOVA followed by Bonferroni test; F(2,17)=7.669; **p<0.01), WT vs 4c9R (t=2.729; *p<0.05). G. miR137 mRNA expression is increased in whole hippocampus of 4c9R mice compared to WT (One-way ANOVA followed by Bonferroni test, F(2,12)=5.683; *p<0.05, WT vs 4c9R (t=3.343, *p<0.05). H. Pyroxd2 mRNA expression is increased in whole hippocampus of 4c9LR and 4c9R mice (One-way ANOVA followed by Bonferroni test, F(2,12)=468.4; ***p<0.0001, 4c9LR vs WT (t=24.20; ***p<0.001) and WT vs 4c9R (t=28.33; ***p<0.001). Data are mean ± SEM, (n=5). I. Basal extracellular glycine levels are decreased in dentate gyrus but not in CA1 as measured by GlyFS R0/Rmax ratio from 4c9R mice compared to WT (unpaired two-sided Student’s t-test, t(22) = 3.59, **p<0.01 at DG, WT n = 13, 4c9R n = 11). LTP is suppressed at the mPP-DG synapses (J, K, L and M) but not at the CA3-CA1 synapses (N, O and P) in 4c9R mice. Summary of LTP experiments (HFS protocol) from four WT mice (n = 5 slices) and four 4c9R mice (n = 8 slices), as well as four 4c9R mice in the presence of 10 μM D-serine in perfusing medium (n = 6 slices) (L and M). LTP was induced by a 1-s train of 100 Hz stimulation. Insets in J, K and L (left and right) are the averages of 40 fEPSPs recorded before (1) and 20 fEPSPs recorded 40 min after (2) the induction (at arrow) of LTP in slices from WT and 4c9R mice. Asterisk (*) indicates significant differences between the three groups (M) (F(2,16) = 5.297, P = 0.017, one-way ANOVA, Bonferroni comparisons * P < 0.05 for 4c9R mice versus WT as well as 4c9R mice (D-serine) versus 4c9R mice. However, LTP is not suppressed in CA3-CA1 synapses in 4c9R mice (N, O and P, t(13) = 0.562, p = 0.584, unpaired two-sided Student’s t-test). Summary of LTP experiments from four WT mice (n = 8 slices) and four 4c9R mice (n = 7 slices). Insets in N and O are the averages of 40 fEPSPs recorded before (1) and 20 fEPSPs recorded 40 min after (2) the induction (at arrow) of LTP in slices from WT and 4c9R mice. All values represent the mean ± SEM. See also Fig. S3.

Behavioral experiments

Mice were tested in conventional, validated behavioral assays. These include tests for prepulse inhibition(18), startle habituation(18), latent inhibition(19), three chamber social interaction(20), Y-maze test(21), water T-maze test(22), novel object recognition test(19), forced swim test(19), fear conditioning test(19), elevated plus maze test(23), light/dark choice test(23), novelty suppressed feeding test(24), and sucrose preference test(25). For additional information, see Supplementary Methods.

Molecular experiments

Mouse tissues were subjected to various physiological, biochemical and molecular analyses using the following methods: Western blots(19) following a cellular fractionation protocol(26), glycine decarboxylase (GLDC) enzyme activity assay, glycine levels in acute hippocampal slices using ratiometric optical FRET sensor GlyFS(27), mitochondrial respiration measurements using seahorse assay(28), hippocampal slice electrophysiology(26), glutathione (GSH) measurements(20), RNA sequencing and analysis(29), KEGG enrichment pathway analysis, Golgi staining and dendritic spine analysis, quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR)(20), and immunofluorescence, confocal imaging and image analysis(19). For additional information, see Supplementary Methods.

Statistical Analysis:

The statistical analysis for the results of behavioral tests, Western blots, biochemical assays and qRT-PCR was performed using SigmaPlot 11.0 (Systat Software, USA) or Prism 5 or Prism 7 (GraphPad Software, USA). Results were analyzed using two-sided Student t-test, or one-way ANOVA (when comparing >2 groups) with post-hoc test Bonferroni’s multiple comparisons method or two-way repeated-measures ANOVA, followed by multiple comparisons with Bonferroni’s correction where appropriate (i.e., only when a significant [p < 0.05] main effect was detected). Two-tailed levels of significance were used and for all tests alpha value set at < 0.05. For electrophysiology, statistical analyses were performed in Excel (Microsoft), Origin Pro (OriginLab Corporation) and MATLAB (Mathworks). Data samples were tested for normality using the Shapiro-Wilk test.

Data availability:

The data that support the findings of this study are available in the supplementary material of this article. RNA sequencing data are deposited in the NCBI GEO database with the following accession number: GSE230871.

Supplementary data:

The Supplementary data include Figs. S1 to S5, Tables S1 to S4 and the Supplementary Methods.

RESULTS

9p24.1 CNV (4c9LR) mice display schizophrenia-like phenotypes

A small supernumerary marker chromosome carrying 14 genes from the 9p24.1 chromosomal region has been identified in two patients with psychosis(14–16). To assess the neurobiological significance of this marker chromosome, we generated deletion and duplication alleles for the mouse homologs of the genes on this chromosome (Rln1, Plgrkt, Cd274, Pdcd1lg2, Ric1, Emrp1, Mlana, 9930021J03Rik, Ranbp6, Il33, Trpd52l3, Uhrf2, Gldc) (for details see Methods, and Fig. 1A). The duplication and deletion alleles were confirmed by comparative genomic hybridization (Fig. 1B) and allowed us to breed mice with 1–4 copies of the genomic segment (hereafter called 9LR): deletion (1c9LR), wildtype (2c9RL=WT), duplication (3c9LR) and “triplication” (= homozygous duplication) (4c9LR) mice. Western blots revealed that mice with 1 copy of the 9LR segment expressed significantly less GLDC protein and mice with 4 copies expressed significantly (approximately threefold) more GLDC protein, one of the proteins in the 9LR segment (Fig. 1C). We then addressed the question whether the copy number variation leads to changes similar to some of those seen in patients with schizophrenia, such as reduction in dendritic spine density(30), signs of oxidative stress(31), latent inhibition deficits(32,33), and working memory deficits(34). In Golgi staining the dendritic spine density in dentate gyrus was significantly reduced in 4c9LR mice but not in 1c9LR and 3c9LR mice compared to WT (Fig. 1D). However, none of the copy number variant mice (1c9LR, 3c9LR and 4c9LR) displayed any change in the levels of reduced and oxidized glutathione (GSH and GSSG, respectively) or in the GSH/GSSG ratio in prefrontal cortex, providing unexpectedly little evidence of oxidative stress(35) (Fig. 1E). Notably, mice with 1 and 4 copies of the 9LR segment (1c9LR and 4c9LR) displayed a deficit in latent inhibition to conditioned freezing, whereas in mice with 2 (WT) and 3 copies of 9LR segment latent inhibition was present (Fig. 1F), indicating that latent inhibition, which is thought to help filter out irrelevant stimuli and to focus on the most important information in the environment, is sensitive to copy number of the 9LR segment. Since working memory is a core aspect of cognitive impairments in schizophrenia patients(34,36), we studied working memory deficits using rewarded alternations in a water T maze paradigm(37). The introduction of a 45 second delay between forced and choice trials reduced the percentage of correct responses in mice with 3 and 4 copies of 9LR (3c9LR and 4c9LR mice) (Fig. 1G), indicating that an increase in the copy number of the 9LR segment results in a working memory deficit. 4c9LR mice showed no difference to WT in hippocampal-independent delay conditioning (Fig. S1A), but increased freezing in hippocampal-dependent trace fear conditioning (TFC) (Fig. S1B) and contextual conditioning (Fig. S1C). As the patients with the 9p24.1 marker chromosome had symptoms of both psychotic and mood disorders, we also assessed additional behavioral parameters in these mice. A reduction in sucrose preference in 4c9LR mice indicated anhedonia (Fig. S1D). Novelty-suppressed feeding (Fig. S1E), motor activity across the light/dark cycle (Fig. S1F), and anxiety-like behavior in the light/dark test (Fig. S1G) and in the elevated plus maze (FIG. S1H) were unaltered. Motor activity in a familiar open field test was increased in 1c9LR and 4c9LR mice after group-housing but not after single housing (Fig. S1I). In the forced swim test, immobility was increased in 3c9LR and 4c9LR mice after 6 weeks of single housing isolation stress but not after group housing (Fig. S1J), indicating that social isolation stress may play a role, which is similar to observations in the Ank3+/− model for bipolar disorder(25).

RNA sequencing of mice with 1–4 copies of the marker chromosome genes reveals dysregulation of pathways relevant for schizophrenia

RNA sequencing revealed that expression of the genes in the 9p24.1 cytoband correlated with copy number in both hippocampus (top panel, HPC) and prefrontal cortex (bottom panel, mPFC) (Fig. S2A). PCA scatter plots show tissue-level variation as strongest among known batch variables (Fig. S2B, C, D). Differentially expressed genes (DEGs) were also correlated with genotype as the largest number of DEGs were observed in triplication mice in the HPC and in duplication mice in the mPFC (Fig. 2A–D, Tables S1a–S1f and Fig. S2E–H). KEGG-pathway analysis revealed long-term potentiation and dopaminergic synapse as downregulated pathways, while ECM-receptor interaction and cell adhesion pathways were upregulated pathways in HPC of triplication mice (Fig. 2E, F). For other genotypes, see Tables S2a–S2r. Significant enrichment was observed for DEGs from 9p24.1 CNVs with rare coding variants among genes linked to autism and neurodevelopmental disorders (NDD) from a recent large-scale sequencing study(38) (Fig. 2G and Table S3). Furthermore, Pyroxd2 (pyridine nucleotide-disulphide oxidoreductase domain 2), a gene encoding a protein involved in mitochondrial-respiration(39), was upregulated in all mutant genotypes in both the HPC and the mPFC (Fig. 2H, I). Mutations or deletion of PYROXD2 in humans leads to increased oxidative-stress and mitochondrial dysfunction(40). Furthermore, the expression of Arl3 (ADP-ribosylation factor-like GTPase 3), which has been associated with schizophrenia by common variant effects in a genome wide association study (GWAS)(41), was reduced in duplication and triplication genotypes in both HPC and mPFC (Fig. 2J, K).

Fig. 2. Gene expression changes in hippocampal and prefrontal cortex with KEGG enrichment pathways, and associated enrichment with TADA-ASD and NDD gene sets.

A, B. The bar graphs show the number of differentially expressed genes (DEG) upregulated (red) and down regulated (green) in hippocampus and prefrontal cortex of triplication mice with the Adj. P value <0.1. The volcano plots show the distribution of log2(fold change) and associated negative log₁₀(padj) for hippocampus (C) and prefrontal cortex (D) of triplication (Trip) mice as compared to WT, with top 30 DEGs highlighted in red. The KEGG pathway classification of genes into enriched pathway terms (E. green, down-regulated pathways; F. red, up-regulated pathways) in hippocampus of triplication mice. The KEGG pathway enrichment cut-offs of significance from FDR < 0.1 to p-value <0.05 was criteria. G. DEG enrichment for gene sets previously published with associations with neurological phenotype; Autism Spectrum Disorder (ASD): Transmission and de novo association with autism spectrum disorder; Neurodevelopmental Disorder-associated genes (NDD): Neurodevelopmental Disorder-associated genes(38). The top two among the DEGs are, Pyroxd2 showed upregulation in all the mutant genotypes in both H. Hippocampus, and I. prefrontal cortex, and Arl3 showed down-regulation in duplication and triplication genotypes at both J. Hippocampus and K. prefrontal cortex. The data presented as 10–90 percentile, (Hippocampus: n=7–8/group, prefrontal cortex: n=8/group), One-Way ANOVA followed by Bonferroni test, *p<0.05, **p<0.01, ***p<0.001. See also Fig. S2.

Increase of Gldc copy number results in dysregulation of BNDF, microRNA-137 and the mitochondrial protein Pyroxd2

To identify the gene(s) underlying the observed phenotype, we generated mice with two smaller, complementary CNVs, one including only Gldc (3c9R and 4c9R), and the other one including the 13 remaining genes of the 9LR segment (3c9L and 4c9L, for details see Methods and Fig. 3A). To determine whether the additional Gldc copies result in increased expression of GLDC protein and increased enzyme activity, we performed Western blots and enzyme assays with hippocampal tissue, a brain region that has been linked to psychosis(42). We found that GLDC expression and activity are indeed significantly increased in 4c9R mice, to a similar degree as in 4c9LR mice (Fig. 3B, C). In contrast, and in line with expectations, GLDC expression in 4c9L was unchanged (Fig. 3B). Previous reports indicate that GLDC is expressed in astrocytes but not in neurons(8,9). We questioned whether this also holds true in mice with additional copies of the Gldc gene and increased expression of the GLDC protein and qualitative images of immunofluorescence staining in CA1, CA3, and DG show a prominent colocalization of GLDC with astrocytic marker S100β in both WT and 4c9R genotypes (Fig. 3D) but not with the neuronal marker Map2 (Fig. 3E). Especially in the DG, staining for GLDC may be increased in 4c9R mice compared to WT. These findings suggest that the increased expression of GLDC in 4c9R mice does not alter the cellular distribution. We hypothesized that the increase of the Gldc copy number in the 4c9R mice would lead to dysregulation of the expression of genes linked to schizophrenia. BDNF has been found to be reduced in the brain of patients with schizophrenia(43). Using qRT-PCR, we found that hippocampal BDNF mRNA was reduced in 4c9R mice (Fig. 3F). MicroRNA miR-137 has been reported to be associated with an increased risk for developing schizophrenia in GWAS studies(41) and its dysregulation has been shown to impair synaptic plasticity(44) in hippocampus. miR-137 expression was increased in hippocampus of 4c9R mice but not in 4c9LR mice (Fig. 3G), providing another link between 4 copies of Gldc and schizophrenia risk genes identified by GWAS. Thus, the triplication of the Gldc gene alone is necessary and sufficient to cause changes in the expression of schizophrenia-related genes. In line with previous transcriptomics results (Fig. 2H), qRT-PCR showed an approximately two-fold increase in Pyroxd2 mRNA expression in 4c9RL and 4c9R mice (Fig. 3H).

4 copies of Gldc result in reduced extracellular glycine levels in dentate gyrus but not in CA1 and diminished long-term potentiation at the perforant pathway to DG synapses but not at CA3-CA1 synapses

Next, we tested the hypothesis that the increased GLDC activity in 4c9R mice (Fig. 3C) leads to reduced extracellular glycine levels in the hippocampus, resulting in NMDA receptor dysfunction. Extracellular glycine levels were assessed by anchoring the fluorescent glycine sensor GlyFS, via a biotin-streptavidin linker, in the extracellular space of the CA1 (st r) or DG regions (st m) of biotinylated acute hippocampal slices(27). The GlyFS fluorescence intensity ratio (R₀/R_MAX) was used to estimate the resting glycine levels. In the DG, a significant decrease in glycine levels was observed in 4c9R mice compared to WT, whereas in the CA1 subregion there was no difference (Fig. 3I). It should be noted that although the reported GlyFS fluorescence ratio is not a linear readout of the extracellular glycine concentration it can be used to estimate the latter as described before(27). We obtained 6.2 μM and 1.5 μM as estimates for the extracellular resting glycine concentrations for the DG of WT and 4c9R mice, respectively, using the previously reported GlyFS properties(27). Furthermore, we characterized the activity-dependent mechanisms that control extracellular glycine levels. Because glycine is involved in NMDAR-dependent synaptic plasticity(45), we focused on plasticity-inducing stimuli. In both 4c9R mice and WT mice, high-frequency stimulation of the Schaffer collateral CA3-CA1 synapses (Fig. S3A, B) but not of the perforant pathway-DG synapses (Fig. S3C, D) resulted in increased GlyFS-reported extracellular glycine levels.

Overall, the experiments using the recently developed GlyFS revealed that an increased copy number of Gldc, which we have shown to increase GLDC activity, results in a reduction of extracellular glycine levels in the DG but not in CA1. As it has been shown previously that while in CA1 D-serine is the only physiologically relevant co-agonist at the NMDA receptor, and in the DG glycine is the major NMDA receptor co-agonist(45,46), we hypothesized that the observed decreased extracellular glycine levels in the DG would translate into selective NMDA receptor hypofunction in the DG but not in CA1. To test this hypothesis, we assayed NMDAR-dependent long-term synaptic potentiation (LTP) by stimulating the medial-perforant-pathway (mPP) or the Schaffer-collaterals-fibers pathway (SCF) and recording field excitatory postsynaptic potentials (fEPSPs) in the DG (st-m) or CA1 (st-r) regions in hippocampal slices from 4c9R and WT mice respectively. The magnitude of tetanic stimulation-induced LTP was significantly reduced in mPP-DG pathway in slices from 4c9R mice compared to WT and reduced LTP could be rescued by the exogenously added NMDAR glycine site agonist D-serine in a saturating concentration of (10 μM) in perfusing medium (Fig. 3J, K, L, M) (F(2,16) = 5.297, P = 0.017, one-way ANOVA, Bonferroni test *P < 0.05 for 4c9R mice versus WT and 4c9R mice (in the presence of D-serine) versus 4c9R mice. This indicates that the observed LTP deficits in the DG could be due to diminished glycine site occupancy resulting in diminished NMDAR activation in the 4c9R mice. The magnitude of LTP was not altered at the CA3-CA1 synapses (Fig. 3N, O, P). Notably, the theta-burst LTP was also diminished in the mPP-DG pathway but not at the CA3-CA1 synapses in 4c9R mice (Fig. S3E–J). Thus, LTP deficits were not specific to the LTP induction protocol used. Overall, our electrophysiological studies revealed that NMDA receptor-dependent synaptic plasticity was significantly reduced in 4c9R mice at the mPP-DG synapse but not at the CA3-CA1 synapse, which could be explained by our finding that extracellular glycine concentrations in the DG are reduced (Fig. 3I).

Increased GLDC expression impairs synaptic plasticity-related signaling pathways

We next examined BDNF protein levels and the activity of downstream pathways in 4c9R mice compared to WT. In hippocampal synaptoneurosomal fractions we detected a significantly reduced amount of BDNF protein (Fig. 4A). The ratios of phosphorylated, i.e., activated forms of Akt (pAkt-S473 and pAkt-T308) to total Akt were reduced (Fig. 4B and C), with no change in total Akt (Fig. 4D). In addition, the ratio of phosphorylated mammalian target of rapamycin (p-mTOR-S2448) to total mTOR, which is phosphorylated by Akt, was reduced (Fig. 4E), with no changes in total mTOR (Fig. 4H). When examining the downstream target of mTOR signaling, P70S6K, we did not find any changes in the ratio of phosphorylated P-P70S6K to total P70S6K or in total P70S6K, a protein involved in ribosome biogenesis and protein synthesis(47) (Fig. 4F and I). However, when we analyzed a downstream target of Akt, CREB, which regulates gene expression in activity-dependent synaptic plasticity, the ratio of phosphorylated CREB (P-CREB) to total CREB protein was reduced in the hippocampus of 4c9R mice, with no change in total CREB. (Fig. 4G and J). Thus, there was a dampening of both transcriptional regulation through the transcription factor CREB and translational regulation through the protein kinase mTOR activity downstream of the BDNF-Akt pathway in 4c9R mice, similar to disruptions in BDNF signaling in schizophrenia. These findings demonstrate multilevel effects of increased GLDC expression at the biochemical level.

Fig. 4. Akt/mTOR signaling and expression of BDNF in synaptoneurosomal fractions prepared from whole hippocampus in 4c9R mice.

Western blot images and quantification of proteins: A. BDNF (4c9R vs WT; t(6)= 2.208; *p=0.044), B. S473 p-Akt/ Akt (4c9R vs WT; t(6)= 2.743; *p=0.0336), C. T308 p-Akt/ Akt (4c9R vs WT; t(6)= 2.488; *p=0.047), D. Total Akt, E. S2448 p-mTOR/ mTOR (4c9R vs WT; t(6)= 4.413; **p=0.0045), F. p-P70S6K/ P70S6K and G. p-CREB/ CREB (4c9R vs WT; t(6)= 3.195; *p=0.0187), H. Total mTOR, I. Total P70S6K, J. Total CREB. All data presented as mean ± SEM, (n=4 each group, unpaired two-sided Student t-test).

4 copies of Gldc reduce dendritic spine density and lead to subregionally differential expression of GLDC and H3K4me3 histone methylation and deficits in mitochondrial bioenergetics

A reduced spine density is one of the most consistently observed neuropathological alterations in schizophrenia(48,49). The percentage of mature mushroom spines, which have been linked to memory functions(50,51), was significantly reduced in 4c9R mice compared to WT (Fig. 5A), and the percentage of stubby spines increased (Fig. 5B). The percentage of thin spines was not altered (Fig. 5C). The total spine density (protrusions) was reduced in 4c9R mice (Fig. 5D) compared to WT. In line with this observation, the microRNA mir132, a regulator of dendritic spine structure(52) which is downregulated in peripheral blood of patients with schizophrenia(53), was reduced in hippocampus of both genotypes with 4 copies of Gldc, 4c9R and 4c9LR (Fig. 5E).

Fig. 5. Effect of 4 copies of Gldc on dendritic spine morphology, mitochondrial bioenergetics and histone methylation in dentate gyrus and CA1.

Images of dendritic branches with spines from dentate gyrus from WT and 4c9R mice. Dendritic spine quantification shows a reduced percentage mushroom type spines (A) (n= 14–15 spines, unpaired two sided Student t-test, t(27)=2.639, *p<0.05), and an increased percentage of stubby type spines (B) (n= 14–15 spines, unpaired two sided Student t-test, t(27)=2.058, *p<0.05). C. There was no change in the percentage of thin spines. D. The total number of spines (protrusions/um) on secondary dendritic branches is reduced in dentate gyrus of 4c9R mice compared to WT (four mice each group, n= 14–15 spines, unpaired two sided Student t-test, t(27)=2.094, *p<0.05). E. Lower amounts of microRNA miR132 in hippocampus of 4c9Rmice and 4c9LR mice (n=5, One-way ANOVA followed by Bonferroni test, F(2,12)=9.524; **p<0.01; WT vs 4c9LR (t=4.337; **p<0.01) and WT vs 4c9R (t=2.588; *p<0.05)). F-H. Hippocampal sub-regional expression of GLDC and H3K4me3 proteins in CA1 and dentate gyrus of 9c4R mice. F. GLDC expression in CA1 and dentate gyrus in 4c9R mice compared to WT (n=4, 3 male+ 1 female mice per group, One-way ANOVA followed by Bonferroni test, F(2,13)=22.73; ***p<0.0001, DG; t=6.718, ***p<0.001). G, H. The histone methylation marker H3K4me3 is increased in CA1 sub-region of hippocampus of 4c9R mice (G) (4c9R vs WT, unpaired two-sided student t-test; t(6)=6.687, ***p<0.001) and decreased in dentate gyrus sub-region of hippocampus of 4c9R mice (H) (WT vs 4c9R, unpaired two-sided student t-test; t(6)=3.055; *p<0.05, n=4, 3 male+ 1 female mice per group). I.-K. Seahorse mitochondrial stress test. The OCR measurements show the characteristic responses to mitochondrial inhibitors and the uncoupler FCCP. Effect of 4 copies of Gldc on mitochondrial oxygen consumption rate (OCR) in tissue punches from hippocampal subregions. I. No change in CA1, J. Decreased basal OCR in DG (One-way ANOVA followed by Bonferroni test, F(7,16)=18.54, ***p<0.0001; 4c9R vs WT, t=2.962, *p<0.05), and reduced FCCP response (One-way ANOVA followed by Bonferroni test; F(7,16)=18.54, ***p<0.0001; 4c9R vs WT, t=4.209, **p<0.01). K. CA1 vs DG basal OCR (Two-way ANOVA revealed main effect of 4 copies of GLDC (F (3, 48) = 6.965, ***p<0.001) followed by Bonferroni test, DG vs CA1 basal OCR in 4c9R, t=5.426, **p<0.01). For all groups, 4 males and 2 female mice (n=6) were used. All data are presented as mean ± SEM. See also Fig. S4.

In view of our findings that the additional copies of Gldc affect extracellular glycine levels differentially in DG and CA1 (Fig. 3I), we investigated whether there are also other hippocampal subregional differences. The 4c9R mice displayed significantly higher GLDC expression in DG but not in CA1 (Fig. 5F), which could explain why extracellular glycine was reduced in 4c9R mice in DG but not CA1. BDNF protein expression was decreased in 4c9R mice in both CA1 and DG compared to WT (Fig. S4E, F). In view of a report that in pluripotent stem cells the activated glycine cleavage system catabolizes glycine to fuel H3K4me3 trimethylation(54), we investigated whether additional copies of Gldc would increase H3K4me3 trimethylation in the hippocampus. The H3K4me3 trimethylation was significantly increased in CA1 but significantly decreased in DG in 4c9R mice (Fig. 5G, H). Decreased levels of H3K4me3 associated with decreased GAD1 expression have been reported in prefrontal cortex from schizophrenia patients(55), indicating that altered levels of H3K4me3 may play a role in the pathophysiology of schizophrenia. The functional significance of the differential modulation of H3K4me3 in DG and CA1 is currently unclear.

Cerebral organoids derived from patients with schizophrenia display a reduced basal oxygen consumption rates (OCR)(56). We found that in CA1 basal respiration, respiration after inhibition of ATP-linked respiration by oligomycin and FCCP (carbonyl cyanide-p-trifluoromethoxyphenyl-hydrazon)-stimulated maximal respiration were not altered in 4c9R mice in CA1 (Fig. 5I). However, in DG basal OCR and FCCP-stimulated maximal OCR were significantly reduced in 4c9R mice compared to WT (Fig. 5J). Non-mitochondrial respiration after addition of rotenone and antimycin C was unchanged between genotypes in both CA1 and DG (Fig. 5I, J). Our results thus indicate reduced basal and maximal mitochondrial respiration in 4c9R mice. Interestingly, when comparing CA1 to DG within a genotype, basal respiration was lower in DG compared to CA1 in 4c9R mice but no difference in basal respiration was observed between the two regions in WT mice (Fig. 5K). In addition, maximal respiration was significantly reduced in 4c9R mice compared to WT in DG (Fig. S4A) but not in CA1 (Fig. S4B). Spare/reserve capacity was not altered in DG or CA1 of 4cR mice (Fig. S4C, D).

4 copies of Gldc lead to deficits in sensorimotor gating, attentional processes, cognitive and social behaviors

Finally, we examined whether 3 or 4 copies of Gldc in the 3c9R and 4c9R mice are necessary and sufficient for the behavioral phenotypes previously seen in the 4c9LR mice which have 4 copies of the entire 9p24.1 segment (9LR). To this end, we performed behavioral assays pertaining to positive symptoms, negative symptoms and cognitive symptoms using mice with an increased copy number of Gldc only (3c9R, 4c9R) (see Methods and Fig. 3A), mice with an increased copy number of the other 9p24.1 genes excluding Gldc (3c9L, 4c9L) (see Methods and Fig. 3A), and the 4c9LR mice with four copies of all genes of the entire 9p24.1 segment as positive controls.

When testing latent inhibition (LI) to conditioned freezing, WT, 3c9L, and 4c9L mice pre-exposed to a tone before fear conditioning froze less than mice not pre-exposed to the tone, i.e., they displayed the LI effect. In contrast, in 4c9LR, 3c9R and 4c9R mice, the LI effect was absent (Fig. 6A). As the LI effect is dependent on DG activity(19), reduced glycine levels in the DG might contribute to the LI deficit. The experiments show that at least 3 copies of Gldc are necessary and sufficient for expression of the LI deficit, and an increase in the copy number of the other 9p24.1 genes does not affect LI.

Fig. 6. Effect of Gldc copy number variation on startle habituation, latent inhibition, prepulse inhibition and working memory.

A. Latent inhibition to conditioned freezing. Bar graph comparing the percent freezing between two conditions, i.e., pre-exposed (white) and non-pre-exposed (matted) to a tone, within genotype for WT, 4c9RL, 4c9R, 3c9R, 4c9L and 3c9L mice (n=16 per bar (8 males, 8 females), One-way repeated measures ANOVA showed significant main effect of pre-exposure (F(11,165) =5.510, ***p<0.0001) indicating that mice not pre-exposed to a tone froze longer than mice pre-exposed to a tone. However, Bonferroni test showed significant difference in WT (t= 4.140, ***p<0.001), 4c9L (t= 3.338, **p<0.01) and 3c9L (t= 4.198, ***p<0.001), but not in 4c9LR, 4c9R and 3c9R mice. B. Habituation to acoustic startle. The startle responses for mice in response to the first and last startle pulses of each session are shown for each genotype (WT, 4cLR, 4cR, 3cR, 4cL, 3cL). Two way repeated measures ANOVA showed significant main effect of habituation process (F(1,66) =84.04, ***p<0.001) and significant main effect of genotype (F(5,66) =34.66, ***p<0.001) followed by main effect of interaction (F(5,66) =9.278, ***p<0.001). Bonferroni test revealed that there was significant habituation in WT (t= 5.594, ***p<0.001), 4c9L (t=7.825, ***p<0.001) and 3c9L (t= 5.757, ***p<0.001) (n=10 with 5 male and 5 female mice). C, D. Line graph comparing prepulse inhibition to acoustic startle (PPI) at different prepulse intensities with SOA 120ms in WT, 4c9LR, 4c9R, 3c9R, 4c9L and 3c9L mice. At 120ms SOA, the two-way repeated measures ANOVA showed significant main effect of genotype (F(5,96) =3.441, **p<0.01), significant main effect of prepulse intensity (F(3,96) =112.7, ***p<0.001) on PPI and a significant interaction between prepulse intensity and genotype (F(15,96) =2.641, ***p<0.001). Further the Bonferroni test showed significant difference between WT vs 4c9R at 73dB prepulse intensity (C) (t= 3.028, *p<0.05) but not with the other genotype (D), (n=10 with 5 male and 5 female mice. The inset in C. shows the one-way ANOVA analysis for 73dB only at 120ms SAO, revealing a significant reduction of percent PPI in 4c9LR and 4c9R compared to WT (One-way ANOVA followed by Bonferroni test, F(2,52)=4.270, *p<0.05; WT vs 4c9LR, t= 2.319, *p<0.05 and WT vs 4c9R, t= 2.708, *p<0.05). E, F. In the Y-maze spontaneous alternation test, the percentage of spontaneous alternations (E) (One-way ANOVA showed significant main effect of genotype on spontaneous alternations (F(5,53) =12.12, ***p<0.0001) and Bonferroni test showed significant difference between WT vs 4c9LR (t= 5.462, ***p<0.0001), WT vs 4c9R (t= 4.040, ***p<0.0001) and WT vs 3c9R (t= 2.702, *p<0.05), and the number of entries made into the arms (F) is shown for all genotypes (n=8–11, percent alternations). G. Water T-maze test. Bar graph showing percent correct choices made during test session comparing between genotypes at each delay duration mice (n=8–10). A the two-way repeated measures ANOVA showed significant main effect of delay durations (F(2,82) =18.78, ***p<0.0001) and significant main effect of genotype on correct choice (F(5,82) =5.126, ***p<0.0001). At 15 seconds delay, 4c9R mice had significantly reduced performance compared to WT mice (t= 2.810, *p<0.05). At 25- and 45-seconds delay, both 4c9LR (t=3.211, **p<0.01 and t= 3.211, **p<0.01) and 4c9R mice (t= 3.011, **p<0.01 and t= 3.011, **p<0.01) had significantly reduced performance compared to WT mice, respectively). H, I, J. Social novelty preference test. H. Bar graph showing comparison of time spent with empty object (E) and time spent with stranger mouse-S1, an index of sociability (Two-way repeated measures ANOVA showed significant main effect of sociability, F(1,52) =27.27, ***p<0.0001), further Bonferroni test revealed significant differences between exploration of E vs S1 mice in WT (t= 3.030, *p<0.05), 4c9L (t= 2.765, *p<0.05) and in 3c9L (t= 4.318, ***p<0.001). I. Bar graph showing the novelty preference index comparison between different genotypes (One-way ANOVA showed a significant difference in novelty preference, F(5,57) =7.947, ***p<0.001) between 4c9LR vs WT (t= 3.567, **p<0.01), 4c9R vs WT (t= 3.745, **p<0.01) and 3c9R vs WT (t= 4.350, ***p<0.001). n=8–10 (4–5 males and 4–5 females). J. Bar graph showing comparison of time spent with stranger mouse-S1 (now familiar) and stranger mouse-S2 (Novel), an index of social novelty (Two-way repeated measures ANOVA showed a significant main effect of genotype on social novelty interactions, F(5,52) =3.08, *p<0.05) and a significant interaction between genotypes and social novelty interaction (F(5,52) =8.154, ***p<0.001). Further, Bonferroni test revealed significant interaction differences between S1 vs S2 mouse in WT (t= 3.142, *p<0.05), 3c9R (t= 3.422, **p<0.01), 4c9L (t= 3.480, **p<0.01) and in 3c9L (t= 3.195, *p<0.05). All data presented as mean ± SEM.

Neural habituation is thought to act as an essential filter for our complex sensory environment and is disrupted in schizophrenia(57,58). We compared startle response amplitudes to startle pulses alone presented at the beginning and at the end of a prepulse inhibition (PPI) session. We found that while WT, 3c9L an 4c9L mice displayed a reduction in the response to the last startle pulses compared to the first startle pulses, i.e., startle habituation, whereas 3c9R, 4c9R and 4c9LR mice responded indistinguishably to the first and last startle pulses, i.e., displayed a startle habituation deficit (Fig. 6B). When assessing PPI to an acoustic startle stimulus, a measure of sensorimotor gating, PPI was reduced in 4c9LR (but not in 4c9R, 3c9L and 4c9L) mice at 73dB prepulse intensity with 60ms stimulus onset asynchrony (SOA) (Fig. S4G, H). At 120ms SOA, at a 73 dB prepulse level 4c9R (but not 4c9LR, 3c9R, 3c9L and 4c9L) mice displayed reduced PPI (Fig. 6C, D), indicating that 4 copies of Gldc lead to a PPI deficit. The response in the absence of a startle stimulus during the PPI session was not altered in any genotype compared to WT (Fig. S4I). In summary, the results of startle experiments reveal that extra copies of the Gldc gene are necessary and sufficient for both PPI deficits and startle habituation deficits.

The Y maze spontaneous alternation paradigm was used to assess working memory. Mice with 3 or 4 copies of Gldc, i.e., 3c9R, 4c9R, and 4c9LR mice (but not 3c9L or 4c9L mice) displayed a reduction in the percentage of spontaneous alternations (Fig. 6E), with the number of entries remaining unaltered between genotypes, suggesting no motivational alterations or differences in locomotor activity (Fig. 6F). In the non-matching to place water T-maze task, where mice were trained to alternate between two trials by rewarding correct alternations, in the testing phase intra-trial delays of 15 sec, 25 sec, and 45 sec were introduced that make the task more difficult. When compared to WT, 4c9LR mice and 4c9R mice displayed a reduction in the percentage of correct response rates. At 15 seconds delay, 4c9R mice had a significantly reduced performance compared to WT, and at 25- and 45-seconds delays, both 4c9LR and 4c9R mice had a significantly reduced performance compared to WT (Fig. 6G). To further clarify the effect of delays within genotype, we compared the within genotype performance between different delay durations (Fig. S5I). In 4c9LR mice, performance was significantly reduced at 25 and 45 seconds compared to 15 seconds, whereas in 4c9R and 3c9R mice, as the performance was low for all the delay durations including at 15 seconds delay, there was no difference between any delays. 3c9L and 4c9L mice performed indistinguishable from WT mice. Our results show that 4 copies of Gldc are necessary and sufficient for disruption of working memory both in a spontaneous alternation task and in a rewarded alternation task. The other 12 genes of the 9p24.1 CNV do not appear to contribute to this phenotype.

Schizophrenia frequently involves poor social functioning (59,60). We assessed whether mice with 9p24.1 copy number variation display differences in sociability and social novelty in the three-chamber social interaction test. In the sociability test, while WT, 3c9L and 4c9L mice spent more time with a stranger mouse under a cup (S1) rather than with an empty cup (E), the 4c9LR, 3c9R, and 4c9R mice spent almost equal time with S1 and E (Fig. 6H). In the subsequent social novelty test, while WT, 3c9L, and 4c9L mice spent more time exploring the novel stranger mouse (S2) compared to the now familiar mouse (S1), 4cLR and 4cR mice spent the same time with S1 and S2 mice, while 3c9R mice spent even significantly more time with S1 rather with S2 (Fig. 6J). The novelty preference index was reduced in 4c9LR, 3c9R, and 4c9R mice, compared to WT, but not in 3c9L and 4c9L mice (Fig. 6I). Additional behavioral characterization revealed no major genotypic differences in fear conditioning, forced swim test, open field test, and the novel object recognition test (Fig. S5A–H).

DISCUSSION

This study found that 4 genomic copies of Gldc, as seen in two patients and resulting in GLDC overexpression, are sufficient to induce a schizophrenia-like phenotype. The primary molecular mechanism appears to be that an increased GLDC activity leads to reduction of extracellular glycine in DG but not in CA1, resulting in reduced synaptic excitatory neurotransmission, which is dependent on NMDA receptors, as evidenced by a LTP deficit in DG but not in CA1. This NMDA receptor hypofunction is apparently also sufficient to lower BDNF expression, and the activity of BDNF-dependent biochemical pathways involved in synaptic plasticity, and to impair mitochondrial respiration. Moreover, the behavioral picture with working memory and other deficits (PPI, startle habituation, latent inhibition, and social interaction), may be considered to constitute a schizophrenia-like behavioral phenotype in mice. In patients with schizophrenia, BDNF is deceased in the postmortem brain(43), and deficits in PPI(61,62), startle habituation(57,58), latent inhibition(32,33),(63), social functioning(59,60) and working memory(64) have been reported. A reduced spine density is one of the most consistently observed neuropathological alterations in schizophrenia(48,49).

The LTP suppression phenotype in DG might underly at least some of the observed behavioral deficits, e.g., the latent inhibition and working memory deficits(19,65). While GLDC overexpression in our 4c9R model is under the control of the endogenous promoter and thus global, our study identifies a specific vulnerability of dentate gyrus changes in the activity of GLDC.

While GLDC has not been associated with schizophrenia by GWAS, as outline before, NMDA receptors and glutamatergic synapses have, so that our 4c9R model supports pathways implicated by GWAS. Moreover, the expression of three genes associated with schizophrenia in GWAS(41) is altered. miR-137(66), which is involved in the regulation of synaptogenesis and synaptic plasticity(44,67), cortical morphology and cognition(51), is upregulated in 4c9R mice (Fig. 3G), and Arl3 (ADP-ribosylation factor-like GTPase 3)(41,68,69), is downregulated in 3c9LR and 4c9LR mice (Fig. 2J,K), indicating that the NMDA receptor hypofunction may be proximal to changes in the expression of these genes that have been linked to schizophrenia by GWAS. Moreover, it has previously been reported that in astrocytes from 4c9LR mice the expression of the srr gene encoding serine racemase, another gene linked to schizophrenia by GWAS(41), is increased. Furthermore, other susceptibility genes for schizophrenia also target and stimulate the activity of CREB, which regulates BNDF expression, providing another link to schizophrenia GWAS hits(70). In addition, the 9p24.1 triplication in 4c9LR mice resulted in enrichment of differentially expressed genes with rare coding variants linked to autism and neurodevelopmental disorders (NDD) from a recent large-scale sequencing study(38) in our transcriptomics analysis.

Transgenic mice (“high-GCS”) constitutively overexpressing GLDC from a CAG promoter (likely in multiple cell types) have previously been reported to have reduced levels of glycine in homogenates from cerebral cortex but not from striatum, and reduced levels of extracellular glycine in dialysate from striatum(71,72). This shows that extracellular glycine levels may be more sensitive to an increase in GLDC activity compared to glycine concentrations in the homogenate, which mostly represents intracellular glycine. An indication that additional copies of Gldc may reduce glycine levels came from studies in primary astrocyte cultures, in which the glycine concentration in the medium was decreased in cells from 4c9LR mice compared to WT on days 2 and 3, suggesting that the additional Gldc copies result in reduced glycine levels(7). In these cultures, expression of phosphoglycerate dehydrogenase (PHGDH), an enzyme involved in the biosynthesis of serine and glycine, was increased, as was expression of serine racemase (SR), both changes which might compensate for the increased catabolism of glycine(7). A previous study by Nagai et al. (2021) showed that specifically attenuating G_q-GPCR signaling in astrocytes by β-adrenergic receptor kinase 1 (iβARK) results in startle habituation deficits and impaired cognitive functions(73). Thus, the study by Nagai et al. (2021) and our findings provide converging evidence that dysfunction of astrocytes can at least in some cases drive the pathophysiology of schizophrenia.

We hypothesized that in our 4c9R mice glycine levels would be reduced in hippocampal homogenates. However, the values for 4c9R (1.721±0.46μM, n=4) and WT mice (1.745±0.30μM n=4) were not different. We then hypothesized that the extracellular glycine levels may be reduced in the 4c9R mice. Using the optical glycine FRET sensor GlyFS and two-photon excitation fluorescence microscopy(27), we found that extracellular glycine levels, which are relevant for modulation of NMDA receptor activity, are reduced in DG (but not in CA1), providing a mechanistic explanation for the LTP deficits at the mPP-DG synapse.

There are interesting parallels and differences between mice with 4 copies of Gldc (4c9R) which have a reduced level of glycine and serine racemase knockout (SRKO) mice, which have a reduced level of D-serine in the brain(18). SRKO mice display LTP deficits at the mPP-DG synapse(26), as do 4c9R mice (Fig. 3K), indicating that both glycine and D-serine may support LTP at this synapse. However, the SRKO mice display a LTP deficit at the CA3-CA1 synapse(74) while 4c9R mice do not (Fig. 3O). A reduction in DG glutamatergic output, which has recently been shown to result in CA3 hyperactivity in mice(75), has been postulated to result in increased CA3-pattern completion and loss of mnemonic functions specific to the DG, in particular pattern separation, and thus cognitive errors generating psychotic associations, resulting in memories with psychotic content(76). Our current results showing in mice that an increase in the copy number of the Gldc gene - as observed in patients with psychosis – leads to a reduction of extracellular glycine levels in the brain and suppression of LTP at the mPP-DG synapse, as well as deficits in cognitive functions (e.g., latent inhibition) that have previously been shown to be dependent on intact DG function(19) support the hypothesis that DG hypofunction, which is also present in SRKO mice(26), may play a relevant role in the pathophysiology of schizophrenia.

A striking similarity is that both models display reduced spine densities in the DG, reduced BDNF levels, and reduced activity of BDNF-dependent signaling pathways (Akt-mTOR and Akt-CREB) involved in synaptic plasticity(26) (Fig. 4). A remarkable difference is that while 4c9LR mice show prepulse inhibition and startle habituation deficits (Fig. 6B, C), this is not the case for SRKO mice(18), indicating that reduced D-serine and reduced glycine levels in the brain result in partially overlapping and partially distinct phenotypes. The convergence of deficits in glycine and D-serine on NMDA receptor function and the fact that serine racemase has been linked to schizophrenia in GWAS(41), strengthen the view that the additional copies of GLDC may play a relevant pathophysiological role in the patients harboring such copies on their marker chromosome.

The transcriptomics analysis (Fig. 2H,I) also revealed the unexpected finding that the Pyroxd2 (pyridine nucleotide-disulphide oxidoreductase domain 2) gene encoding a protein involved in mitochondrial-respiration(39) is upregulated in 1c9LR, 3c9LR and 4c9LR mice in both the HPC and the mPFC (Fig. 2H,I), and we also found that it is upregulated in 4c9R mice with 4 copies of Gldc.(Fig. 3H). The PYROXD2 protein is located in the mitochondrial inner membrane/matrix and interacts with mitochondrial complex IV subunit COX5B(39). As mutations or deletion of PYROXD2 in humans lead to increased oxidative-stress and mitochondrial dysfunction(40), we investigated mitochondrial respiration and found that mitochondrial respiration is attenuated in the DG of 4c9R mice (Fig. 5J), which is in line with a report studying organoids from patients with schizophrenia(56).

On a fundamental level, our study discovered that GLDC, which in the brain is expressed in astrocytes but not in neurons(17,77,78) modulates extracellular glycine levels in the DG but not in CA1, and that additional copies of Gldc reduce these glycine levels and thus attenuate LTP, which may underlie cognitive and/or other deficits. Targeting glycine metabolism by slowing its degradation by GLDC, perhaps in combination with an inhibition of glycine reuptake, may be a novel strategy to address the NMDA receptor hypofunction which is now widely recognized as a core pathophysiological deficit in schizophrenia that can drive psychosis(4).

There are some limitations to our study. First, the GLDC triplication is a very rare mutation, but we show that in the pathophysiology it is proximal to multiple genes and pathways linked to schizophrenia by GWAS. Second, in the CNS, glycine is not only a co-agonist at the NMDA receptor, it is also an agonist at inhibitory strychnine-sensitive glycine receptors(79) and at excitatory eGly (GluN1/GluN3A) receptors(80). The functions of these receptors are not well understood. We have not tested whether reduced glycine levels in mice with 4 copies of Gldc might also have an effect on these receptors and can thus formally not exclude potential contribution of these receptors to the observed phenotypes.