Behavioural and molecular characterisation of the Dlg2 haploinsufficiency rat model of genetic risk for psychiatric disorder

Abstract Genetic studies implicate disruption to the DLG2 gene in copy number variants as increasing risk for schizophrenia, autism spectrum disorders and intellectual disability. To investigate psychiatric endophenotypes associated with DLG2 haploinsufficiency (and concomitant PSD‐93 protein reduction) a novel clinically relevant Dlg2 +/− rat was assessed for abnormalities in anxiety, sensorimotor gating, hedonic reactions, social behaviour, and locomotor response to the N‐Methyl‐D‐aspartic acid receptor antagonist phencyclidine. Dlg gene and protein expression were also investigated to assess model validity. Reductions in PSD‐93 messenger RNA and protein were observed in the absence of compensation by other related genes or proteins. Behaviourally Dlg2 +/− rats show a potentiated locomotor response to phencyclidine, as is typical of psychotic disorder models, in the absence of deficits in the other behavioural phenotypes assessed here. This shows that the behavioural effects of Dlg2 haploinsufficiency may specifically relate to psychosis vulnerability but are subtle, and partially dissimilar to behavioural deficits previously reported in Dlg2 +/− mouse models demonstrating issues surrounding the comparison of models with different aetiology and species. Intact performance on many of the behavioural domains assessed here, such as anxiety and reward processing, will remove these as confounds when continuing investigation into this model using more complex cognitive tasks.


| INTRODUCTION
The DLG2 gene locus is linked to multiple psychiatric disorders. Point mutations in promotor regions have been associated with autism, 1 schizophrenia and intellectual disability. 2 Copy number variants encompassing complete deletion of one copy of DLG2 result in increased risk for schizophrenia, 3 autism, 4 bipolar disorder 5 and epilepsy. 6 Such clinical evidence highlights the potential importance of DLG2 in the psychopathologies common to a broad range of disorders. DLG2 encodes the postsynaptic scaffold protein PSD-93 (also known as Chapsyn-110) in the membrane associated guanylate kinase (MAGUK) family. These proteins are responsible for anchoring and organising the numerous protein complexes required for development and plasticity at the synapse, in particular the NMDA receptor, [7][8][9][10] AMPA receptor, 11 potassium ion channels 8,12,13 and neuroligin 1-3. 7 Previous investigation into specific behavioural endophenotypes driven by Dlg2 disruption has used mouse models comprising both homozygous and heterozygous genetic lesions. Assessing homozygous knockdown models investigates the function of PSD-93 and potential compensation mechanisms in its absence. In contrast, heterozygous models are faithful to the deletions seen in human psychiatric patients and allow investigation into endophenotypes which could increase disease risk under these circumstances. Dlg2 À/À mice have shown abnormal social behaviours including impaired social preference, 14 increased repetitive behaviours and hypoactivity in response to novelty. 14,15 Deficits in cognitive flexibility and attention have also been shown, 16 aligning with similar deficits in human carriers of mutations to the DLG2 coding region.
On some behavioural tasks the phenotypes of homozygous and heterozygous models align. Yoo et al. 14 found that self-grooming appeared to increase with Dlg2 dosage, with the heterozygotes showing increased grooming relative to wild-types and the homozygotes showing more grooming than the heterozygotes. Winkler et al. 15 found a similar effect with reductions to Dlg2 causing more severe impairments to motor learning and coordination. However, in these investigations altered social behaviour 14,15 and hypoactivity in response to novelty 14,15 seen in homozygous mutants were not found in heterozygotes. From a clinical perspective this could indicate that Dlg2 heterozygosity does not increase psychiatric risk via altering social capabilities or habituation to stimuli, despite homozygous models indicating a role of PSD-93 in normal performance of these behaviours. Phenotypic differences between homozygous and heterozygous mouse models led us to focusing on assessment of a heterozygous rat model with the aim of isolating processes that precipitate disease arising from Dlg2 haploinsufficiency.
A range of psychiatric endophenotypes remain to be tested in heterozygous Dlg2 models, including faulty sensorimotor gating, anhedonia, and locomotor response to pharmacological challenge.
This work presents the first molecular and behavioural characterisation of a rat model generated using CRISPR/Cas9 gene editing technology which contains only one copy of the Dlg2 gene.
Here we assess whether the expected biological changes (reduced Dlg2 mRNA and protein expression) occurred in the heterozygous (+/À) model, and whether there is evidence of compensation for possible Dlg2 reduction from other MAGUK family members or related proteins, as has been shown in cortical neurons with complete PSD93 knockdown. 17 Behavioural consequences of Dlg2 haploinsufficiency was assessed by comparing Dlg2 +/À rats and wildtype littermates on a battery of psychiatric-relevant translational tests.
These included tests of anxiety, social behaviour, and anhedonia, key behavioural domains disrupted across disorders Dlg2 is implicated in. Regulating unwanted or unnecessary sensory inputs was assessed using sensorimotor gating, deficiencies of which characterise patients with schizophrenia 18,19 and autism, 20 in addition to rodent models of these conditions. 21 Hyperlocomotion in response to PCP (an NMDAR-antagonist) was also assessed as acute administration of PCP produces a transient psychosis-like phenotype 22,23 which may be exaggerated in a rodent model with potential NMDAR alteration of function 24 and an underlying genetic propensity towards psychosis.
Dlg2 +/À rats demonstrated selective reductions in mRNA and protein expression that was not compensated for by increases in the remainder of the Dlg family. There was also an absence of gross behavioural deficits related to anxiety, hedonic reactions, social behaviour, and sensorimotor gating in the model; however, Dlg2 +/À rats demonstrated a potentiated hyperlocomotion phenotype in response to PCP administration compared to wild-types. Together, these confirm the biological validity of the current model selectively relevant to reduced expression of Dlg2 and a possible concomitant alteration of NMDAR function 24 and furthermore suggests that clinically relevant reductions to Dlg2 expression will have more subtle effects than homozygous knockout mouse models.

| Animals
Dlg2 heterozygous rats were generated on a Long Evans Hooded background by Horizon Discovery (Pennsylvania, USA) using CRISPR/ Cas9 gene editing technology. Successful founders generated by Horizon Discovery had a 7 bp deletion (782933-782939 in the genomic sequence) in exon 5 which caused a frame shift and generation of an early stop codon in exon 6. Confirmation of successful nonhomologous end joining activity was assessed by PCR and sequenced by Horizon Discovery, UK. Selected heterozygous founders were send to Charles River (Margate, UK) and bred to produce experimental colonies by breeding male heterozygous rats were bred with female wild-types resulting in a Mendelian distribution of wild-type and heterozygous pups. A more detailed description of the generation of this rat line can be found in Supplement 1 of Griesius et al. 24 Animals were housed in groups from two to four in standard cages (l Â w Â h: 50 cm Â 32 cm Â 21 cm) in rooms with a temperature between 19 and 23 C maintained on a 12 h light-dark cycle.
Cages had sawdust and paper nesting and environmental enrichment (wooden chews and cardboard tubes). Food and water were given adlib while conducting all tasks except the lick microstructure assessment where rats were maintained at 85-95% of their free-

| Tissue extraction
Rodents used for qPCR and Western blot were culled at 2-4 months old by inhalation of slowly rising CO 2 concentration for 8 min (administered by Home Cage Culling Chamber, Clinipath Equipment Ltd, UK) 2 weeks after completion of the same set of behavioural experiments focusing on reward learning which are not reported here. Brains were extracted from the skull and gross dissected by partitioning the cerebellum and rostral-most part of cortex (prefrontal cortex). The hippocampus and posterior cortex were then dissected. Extracted brain regions were flash frozen on dry ice and stored at -80 C prior to use.

| Quantitative polymerase chain reaction (qPCR)
Using the QIAGEN RNeasy kit, RNA was isolated from prefrontal cortex, hippocampal and cerebellar tissue from individual animals (Cohort 5

| Western blot
Western blot analysis was conducted on hippocampal, prefrontal cortex, posterior cortex and cerebellar tissue from WT (n = 12) and HET (n = 12) animals. Each tissue sample was lysed in Syn-PER lysis and extraction buffer (Thermo Fisher, UK) with mini protease inhibitor cocktail (Roche Diagnostics) and phosphatase inhibitor (Cell signalling, UK) according to description from manufacturer. After using BCA Assay kit to measure the total amount of protein in each sample, electrophoresis and blotting were carried out.   Anxiety on the EPM is associated with more stretch attend postures, fewer head dips and increased defecation in addition to reduced open arm exploration. 25 Faecal boli were also counted for each animal after each run, and for female rodents vaginal cytology was performed after testing on the maze to determine oestrus stage.

| Open field test
The open field apparatus was a 100 Â 100 cm square wooden arena  All stranger rats were habituated to the chambers for 10 min prior to testing. For analysis raw exploration times of the chambers (rodents directing their nose at the chamber at a distance of <2 cm) were used in addition to d2 discrimination ratio (Equation (1)). Discrimination ratio gives a readout of the difference in exploration time between the two stimuli without the confound of overall tendency to explore for long or short durations.

| Statistical analysis
All statistical analysis was performed using JASP Version 0.14.1 (JASP can be followed: a Bayes factor between 1 and 3 gives weak or anecdotal support to the null, a factor between 3 and 10 represents some supporting evidence, while a factor more than 10 indicates strong evidence for the null. Bayes factors were calculated for factorial ANOVAs in the way   Figure 1G and H shows that for Dlg4 there were no main effects of brain region (F (1, 12) Of the three proteins analysed only PSD-93 showed consistent decreases across all four brain regions in the Dlg2 +/À rats compared to wild-types ( Figure 2). Integrated densities were analysed using repeated measures ANOVA with within-subjects factor of brain region (prefrontal cortex, posterior cortex, hippocampus, cerebellumapart from NR1 where expression in cerebellum was negligible in all cases and thus this region was omitted from the analysis) and between-subjects factor of genotype. Example blots can be seen in Supplementary Figure S1. The reduction of PSD-93 in Dlg2+/À rats compared to wild-types across all brain regions is shown in Figure 2A. in Dlg2 +/À and wild-type rats. These were assessed across four brain regions: prefrontal cortex (PFC), posterior cortex (CX), hippocampus (HP) and cerebellum (CB). Cerebellar NR1 expression was too low for analysis thus is not reported. Data is shown as mean ± SEM integrated density plotted plus individual data points. n = 12 wild-type, 12 Dlg2 +/À . Across the prefrontal cortex, posterior cortex, hippocampus and cerebellum Dlg2 +/À rats showed a decrease in PSD-93 compared to wild-types, with no changes in PSD-95 or NR1 NMDA receptor subunit levels genotype Â brain region interaction (F (1.148, 19.514) = 0.902, p = 0.368, n 2 p = 0.050; BF exclusion = 8.238). Thus, it seems that Dlg2 haploinsufficiency does not have downstream effects on the expression of related proteins.

| Acoustic startle response (ASR) and pre-pulse inhibition (PPI)
Response to increasing amplitudes of startle stimuli were assessed using mixed ANOVA and mixed Bayesian ANOVA with withinsubjects factor of pulse intensity (70, 80, 90, 100, 110 and 120 dB) and between-subjects factors of genotype and sex from data acquired in the third block of trials from the startle sessions ( Figure 5A). There

| Lick microstructure assessment
Repeated measures ANOVA and Bayesian repeated measures ANOVA were used to analyse consumption and lick cluster data with genotype as a between subject factor and sucrose concentration as a within-subject factor. As Figure 6A shows F I G U R E 6 The drinking behaviour of Dlg2 +/À and wild-type rats when presented with low and high concentrations of sucrose. Data is shown as mean ± SEM plus individual values (A) consumption (g) and (B) lick cluster size. n = 28 wild-type, 20 Dlg2 +/À . Consumption and lick cluster size to low and high sucrose solutions did not vary with genotype Figure 6B reveals that lick cluster size also varied with sucrose concentration, as both wild-type and Dlg2 +/À rats showed larger cluster sizes at 16% than to less palatable 4% (main effect of concentration, F (1, 46) = 56.074, p < 0.001, n 2 p = 0.549). Genotype had no effect on lick cluster size at either concentration (non-significant main effect of genotype, F (1, 46) = 0.642, p = 0.427, n 2 p = 0.014 nonsignificant genotype Â concentration interaction F (1, 46) = 0.002, p = 0.965, n 2 p = 0.000). Evidence for any genotype effect on lick cluster size is inconclusive (BF exclusion = 2.384) as is that for the genotype Â concentration interaction (BF exclusion = 2.341). Because evidence for impaired hedonic reactions would require that Dlg2 +/À s would have lower lick cluster size than wild-types Bayesian one-tailed independent samples t-tests were done on the lick cluster sizes for 4% and 16% conditions, finding evidence for the absence of this expected Dlg2 +/À less than wild-type effect in both instances (4% BF 01 = 7.887, 16% BF 01 = 5.735).
The variation in both lick cluster and consumption with concentration is expected and informs that the experiment successfully manipulated the hedonic properties of the stimuli. The lack of any reduction in lick cluster size for the Dlg2 +/À rats suggests there is no suggestion of an anhedonic response to palatable stimuli.

| Social preference test
Raw exploration times for conspecific and object in the social preference test are shown in Figure 7A. The discrimination ratio for social preference is shown in in Figure 7B. This was significantly different from 0 for the entire cohort (one-sample t-test t (53) = 9.988, p < 0.001, d = 1.359) reflecting the tendency to explore the conspecific more than the object. There were no genotype differences in discrimination ratio: non-significant main effect of genotype (F (1, 50) = 0.850, p = 0.361, n 2 p = 0.017; BF exclusion = 4.973). This reflects the fact that rats explored the conspecific more than the object as expected, yet Dlg2 haploinsufficiency has no influence on this tendency. F I G U R E 7 Dlg2 +/À and wild-type exploration times on the social preference task. Data is shown as mean ± SEM plus individual values (A) raw exploration and (B) d2 discrimination ratios. n = 26 wild-type, 32 Dlg2 +/À . There was no effect of genotype on rodents preference for exploring the unknown conspecific relative to the object in the social preference task F I G U R E 8 Locomotor activity in response to PCP injection in Dlg2 +/À and wild-type rats. Data is shown as mean ± SEM distance travelled plotted in 10-min bins. The dotted line just past 30 min denotes when the PCP injection occurred. n = 26 wild-type, 32 Dlg2 +/À . Dlg2 +/À rats demonstrated a locomotor response to PCP which was more sustained and exaggerated than wild-types baseline before PCP injection. Comparisons for genotype at 10 min

| DISCUSSION
This work presents the characterisation of which molecular and behavioural capabilities are spared and impaired in the Dlg2 heterozygous rat model; a model with direct clinical relevance to CNVs that increase risk for a variety of psychiatric conditions including schizophrenia, 3 autism 1 and intellectual disability. 2 The model is specific to Dlg2 and valid, with evidence that mRNA and protein expression of Dlg2 is reduced in the absence of changes to levels of other Dlgs. Behaviourally Dlg2 +/À rats performed comparably to wild-types on tests of anxiety, hedonic reactions, social behaviour, and sensorimotor gating.
When locomotor response to PCP challenge was assessed, Altered psychostimulant sensitivity has also been shown in other psychosis-relevant CNV rodent models, the 22q11.2 41 and 1q21.1 42 microdeletion mouse models. In the 22q11.2 mouse this manifested as an exaggerated locomotor response to PCP and ketamine. In the 1q21.1 mouse exaggerated locomotor behaviour was seen in response to amphetamine but was not significant with administration of PCP, yet PCP resulted in sensorimotor gating impairments. As administration of non-competitive NMDAR antagonists in healthy rodents and humans can induce a behavioural syndrome isomorphic to positive and negative symptoms of schizophrenia 22,43,44 and exacerbates positive and negative symptoms in schizophrenic patients [45][46][47] the finding of PCP sensitivity in these rodent CNV models may highlight a general psychosis susceptibility. Findings such as these contribute to the hypoglutamaterigic hypothesis of psychosis. 48 However, it is difficult to use these findings to determine which biological processes are altered in CNV rodent models. Acute PCP administration has been reported to activate serotonergic, glutamatergic, noraderenergic, cholinergic and neurotensinergic transmission in rodents and monkeys. [49][50][51] It will be informative to focus on investigating how individual behavioural alterations to NMDAR antagonists, and their timescales, may be subserved by particular neurotransmitter dysfunctions to identify mechanisms underpinning these in genetic disorder models.
Coherence between findings on homozygous mouse models, heterozygous mouse models and the CRISPR-Cas9 generated Dlg2 +/À rat were mixed. Findings of impaired social preference and hypoactivity in the open field test which are seen in Dlg2 À/À mice but not Dlg2 +/À mice 14,15 were also not seen in the Dlg2 +/À rat. Comparisons between complete and heterozygous gene knockout models allow a distinction to be made between knowledge about the function of a protein and processes which require complete PSD93 levels. The difference here implies that having some functional PSD93 might 'rescue' these phenotypes. This could be supported by PSD93 acting in tandem with PSD95 or other MAGUKs. Where social behaviour is concerned it has been shown that there is a similarity in social deficits in PSD95 +/À mice and PSD93 À/À mice, with increased expression of PSD93 in the hippocampus of PSD95 +/À mice implying that PSD93 is acting to compensate in this mouse model. 15 In Dlg2 +/À rodent models it could be that intact PSD95 in the presence of some PSD93 was sufficient to support intact social preference performance, meaning that while PSD93 has some role in social behaviours it is not so essential that genetic haploinsufficiency produces a gross deficit.
The increased self-grooming phenotype seen in homozygous and heterozygous mouse Dlg2 mutants 15 was not seen in the rat model. This comes with the caveat that grooming was assessed in the EPM and open-field test in the rat line, while Yoo et al 14 assessed grooming in a clean home cage. This comprised 20 min of habituation followed by 10 test minutes in which self-grooming was recorded. Dlg2 +/À and wild-type rats placed in the EPM or open field for 5 and 10 min, respectively, would not have habituated to the apparatus, as is crucial for anxiety tests, meaning that increased self-grooming may only be seen after the habituation period.
The observed lack of anxiety and pre-pulse inhibition phenotypes in the Dlg2 +/À rat were also found in Dlg2 +/À mice. 52 However, Dlg2 +/À mice demonstrated a subtle deficit of habituation to the acoustic stimulus in this facet of the sensorimotor gating task, which was not seen in the Dlg2 +/À rat. This may be due to differences in model generation, with the heterozygous mouse generated by introduction of a cassette upstream of the critical exon (14) on chromosome 7 while the heterozygous rat was generated by 7 bp deletion within the rat Dlg2 gene resulting in a frame shift and premature stop codon. This behavioural difference also points to potential caveats in comparisons across psychiatric risk models with different species backgrounds, which is also relevant when comparing homozygous and heterozygous mice with the rat model.
Another point of interest is the lack of interactions between sex and genotype in this work (see Supplement 1 for detailed data), which in the main was conducted on mixed sex cohorts. This is an important inclusion in the rodent modelling of psychiatric susceptibility literature which is often conducted on male-only cohorts, and thus runs the risk of limited translational generalisability. Critically, the results here relating to the genotype manipulation were unaffected by the sex of the animals.

| CONCLUSION
The Dlg2 +/À rat validly models a single copy deletion of Dlg2 including concomitant mRNA and protein decrease in the absence of obvious compensation. No gross behavioural deficits on tasks relevant to a broad spectrum of psychiatric phenotypes were found, except for exaggerated hyperlocomotion in response to PCP, an NMDAR-antagonist. A similar selectivity in phenotype is seen for other psychiatric CNV models such as the 1q21.1 mouse. 42 This demonstrates the behavioural subtlety of the model and highlights issues with drawing clinical conclusions from homozygous models. It also paves the way for investigation into more complex behavioural domains such as memory and learning without the concern of confounds from anxiety, hedonic processing, hearing, and social processing.