Selective behavioural impairments in mice heterozygous for the cross disorder psychiatric risk gene DLG2

Abstract Mutations affecting DLG2 are emerging as a genetic risk factor associated with neurodevelopmental psychiatric disorders including schizophrenia, autism spectrum disorder, and bipolar disorder. Discs large homolog 2 (DLG2) is a member of the membrane‐associated guanylate kinase protein superfamily of scaffold proteins, a component of the post‐synaptic density in excitatory neurons and regulator of synaptic function and plasticity. It remains an important question whether and how haploinsuffiency of DLG2 contributes to impairments in basic behavioural and cognitive functions that may underlie symptomatic domains in patients that cross diagnostic boundaries. Using a heterozygous Dlg2 mouse model we examined the impact of reduced Dlg2 expression on functions commonly impaired in neurodevelopmental psychiatric disorders including motor co‐ordination and learning, pre‐pulse inhibition and habituation to novel stimuli. The heterozygous Dlg2 mice exhibited behavioural impairments in long‐term motor learning and long‐term habituation to a novel context, but not motor co‐ordination, initial responses to a novel context, PPI of acoustic startle or anxiety. We additionally showed evidence for the reduced regulation of the synaptic plasticity‐associated protein cFos in the motor cortex during motor learning. The sensitivity of selective behavioural and cognitive functions, particularly those dependent on synaptic plasticity, to reduced expression of DLG2 give further credence for DLG2 playing a critical role in specific brain functions but also a mechanistic understanding of symptom expression shared across psychiatric disorders.

membrane-associated guanylate kinase protein superfamily of scaffold proteins, a component of the post-synaptic density in excitatory neurons and regulator of synaptic function and plasticity. It remains an important question whether and how haploinsuffiency of DLG2 contributes to impairments in basic behavioural and cognitive functions that may underlie symptomatic domains in patients that cross diagnostic boundaries. Using a heterozygous Dlg2 mouse model we examined the impact of reduced Dlg2 expression on functions commonly impaired in neurodevelopmental psychiatric disorders including motor co-ordination and learning, pre-pulse inhibition and habituation to novel stimuli. The heterozygous Dlg2 mice exhibited behavioural impairments in long-term motor learning and long-term habituation to a novel context, but not motor co-ordination, initial responses to a novel context, PPI of acoustic startle or anxiety. We additionally showed evidence for the reduced regulation of the synaptic plasticity-associated protein cFos in the motor cortex during motor learning.
The sensitivity of selective behavioural and cognitive functions, particularly those dependent on synaptic plasticity, to reduced expression of DLG2 give further credence for DLG2 playing a critical role in specific brain functions but also a mechanistic understanding of symptom expression shared across psychiatric disorders.

K E Y W O R D S
autism spectrum disorder, bipolar disorder, habituation, learning, PSD93, psychosis, schizophrenia 1 | INTRODUCTION Schizophrenia (SZ), autism spectrum disorder (ASD), and bipolar disorder (BD) are often comorbid neurodevelopmental psychiatric disorders, 1,2 with overlapping symptoms, including cognitive, neurobehavioural and motor dysfunction. [3][4][5] There is also a substantial portion of shared genetic risk, [6][7][8] concordant with the correlation between functional disabilities and risk genes associated with developmental disorders. 9,10 Genetic analysis approaches, including genome wide association studies, studies of common variants as well as copy number variant (CNV) analysis, and exome sequencing for rarer but more penetrant mutations, combined with subsequent gene set enrichment analysis have been important for identifying the biological pathways underlying disorder pathophysiology and common targets for clinical interventions. 11 This approach has uncovered an enrichment of genes involved in synaptic plasticity and the postsynaptic density (PSD) in SZ. 6,[12][13][14][15][16][17] Convergence of mutations in genes important for synaptic formation, plasticity, and elimination have also been found in ASD, [18][19][20][21][22] with meta-analyses identifying postsynaptic complexes as one of three functionally interconnected pathways associated with ASD risk genes. 23,24 Discs large homolog 2 (DLG2, also known as post-synaptic density protein-93 [PSD-93] and chapsyn-110), is part of the membrane associated guanylate kinase (MAGUK) protein superfamily of scaffold proteins and is enriched at the PSD. 25 MAGUKs primarily bind and stabilise proteins at synapses, and regulate the synaptic localization of glutamate receptors, key for synaptic transmission and synaptic plasticity. 26 Both de novo CNVs and SNVs implicate DLG2 in increasing risk of SZ, 6,15,[27][28][29] ASD, [30][31][32] and potentially BD. 33 Mutations in DLG2 affecting its function and/or expression are plausibly related to the impairments of learning and memory underlying in the cognitive deficits seen in SZ, as well as altered synaptic plasticity linked to abnormal sociability and motor deficits reminiscent of ASD. 22,[34][35][36][37] Indeed, homozygous knockout mice (Dlg2 À/À ) showed no impairment in simple forms of learning, but showed deficits in complex learning, cognitive flexibility and attention comparable to similar deficits in humans carrying mutations disrupting the coding region of DLG2. 38 Furthermore, Dlg2 À/À mice displayed abnormal social behaviours, 39,40 increased repetitive behaviours and hypoactivity in response to novelty, 40 as well as defective long-term potentiation (LTP) 41 and aberrant excitatory synaptic transmission in the dorsal striatum. 40 Significantly, the functions of Dlg2 and other members of the Dlg family are dissociable both at the behavioural and synaptic levels. 26,38,41,42 Most genetic lesions observed in humans are deletions within or of DLG2 6,15,[27][28][29] and are heterozygous. 32,43 Little attention has been paid to investigating behavioural phenotypes associated with heterozygous Dlg2 models, arguably more translationally relevant than homozygous knockouts. Winkler et al. 39 identified a Dlg2 dose-dependent impairment of motor learning in mice. There has been no systematic investigation of reduced Dlg2 gene dosage on potentially core endophenotypes associated with SZ, ASD, BD, and other neuropsychiatric disorders including motor function, pre-pulse inhibition (PPI) of startle responses and habituation to a novel stimulus. [44][45][46][47][48][49][50][51][52][53] In this study we investigated motor co-ordination and motor learning, habituation to a novel context, PPI of the acoustic startle response (ASR) and anxiety in heterozygous Dlg2 adult male mice. Using cFos levels as a proxy marker for synaptic activity and plasticity, 54 we also tested the hypothesis that in the heterozygous Dlg2 mice impaired behavioural function would be correlated with reduced activity-dependent cFos activation.

| Behavioural analysis
Data from three separate cohorts male of mice is presented in this manuscript, with Cohort 1 and Cohort 2 undergoing more than one behavioural test (Table S1). At start of testing mice were 8-12 weeks of age. All tasks were completed prior to lights off. All mice were habituated to the experimenter for 1 week before testing, and to procedural rooms for 5 min before training and testing.

| Motor learning and performance
Two-day (2d) rotarod motor learning and motor performance task. Mice were habituated to the experimental room for 5 min prior to the first trial. Each mouse received 5-min trials on the rotarod (3 cm drum separated by flanges into 5 Â 5.7 cm individual lanes, fall height 16 cm, 47,600, Ugo Basile, Italy). Motor training comprised of 5 trials across 2 days (Day 1, two trials, Day 2, three trials) during which the rotarod was accelerated incrementally from 5-50 rpm over 5 min. Animals were returned to home cages in the testing room between trials. Following completion of a trial for the whole cohort (approx. ITI 30 mins) the next trial was started. Starting on Day 3, to assess motor performance, mice then completed two 5-min trials at a fixed speed (5-45 rpm in 5 rpm increments) over 5 days, with an inter-trial interval of 20-30 min during which the mice were returned to their home cage in the test room. The mean latency to fall for two trials per speed was calculated.
Three day (3d) rotarod motor learning task. Mice were habituated to the experimental room for 5 min prior to the first trial. Each mouse received 7 trials a day for 3 consecutive days. Each trial comprised of the rotarod accelerating incrementally from 4-40 rpm over 5 min, with an inter-trial interval of 5 min during which the mice were returned to their home cage in the testing room. For the assessment of cortical activity using cFos IHC, experimental mice were trained on the accelerating rotarod under the same conditions. However, mice were single caged in their housing (not test) room for 1.5 h after the second trial on Day 1. Control mice were singled caged in the holding room for 1.5 h.
In all tasks, mice were placed on the rod facing away from the experimenter. Rotarod training order was counterbalanced for genotype. Latency to fall (s), defined as the first fall from the rod or 1 full rotation on the rod, was measured for each mouse per trial. Consolidation of motor learning was analysed by comparing latency to fall between for the last trial on Day 1 (D1T7) to the first trial on Day 2 (D2T1), = latency to fall(s) (D2T1/(D2T1 + D1T7). The startle response (startle amplitude) to the first three pulses at 105 and 120 dB were averaged and analysed as an index of emotional reactivity before appreciable habituation. 56 To measure habituation of the startle response, responses to the first five pulse-alone trials at each intensity were measured. For each animal the absolute PPI unconfounded by the initial startle response was calculated as the percentage reduction in startle amplitude in pre-pulse (PPI) compared to pulse alone trials (105 dB and 120 dB). [%PPI = 100 Â (ASRstartle pulse alone À ASRprepulse + startle pulse)/ASRstartle pulse alone].

| Locomotor activity and habituation
The locomotor activity of individual mice was conducted in clear Perspex activity boxes (CeNeS Cognition, Cambridge, UK, 21 cm Â 36 cm Â 20 cm; 2 IR beams 3 cm from box end, 1 cm from box floor) in the dark. For each mouse, individual beam breaks were recorded (using a custom BBC BASIC V6 programme with additional ARACHNID interfacing, Campden Instruments, Loughborough, UK) in 5 min bins over a 2 h session for 5 consecutive days. Mice were assigned an activity box for the duration of experiment. Twelve mice were run concurrently and the order/time of testing was conserved across days. Each batch of 12 mice was counterbalanced for genotype. The total number of beam breaks per day was recorded and the total number of beam breaks in the first 6 bins (30 min quartile, Q1-Q4) each day was calculated to measure between-session and within-session activity, respectively. Changes in between-session activity = beam breaks (Day 5/[Day1 + Day 5]).

| Imaging
M1 was imaged as a Z stack using a Zeiss LSM 900 confocal microscope running ZEN 2 (blue) software at 20Â magnification and analysis was conducted in Fiji (1.52 g, https://imagej.nih.gov/). Exposure time was constant between sections. The average intensity projection was analysed, taken from the central focal plane and 2 adjacent focal planes in either direction, with an interplane interval of 1 μm. The area of M1 59 was measured using the freehand tool and cFos + cells were manually counted. The average number of cFos + cells/mm 2 per mouse was calculated by dividing the mean number of counted cells by the area measured. Per animal 10-12 sections were counted.

| Statistical analysis
Data was analysed using SPSS 23 (IBM Corporation, New York, USA), with an alpha level of p < 0.05 was regarded as significant throughout.
All data was tested for normality using the Shapiro-Wilks test. Out  3.1 | There was a significant reduction in Dlg2 mRNA in the cortex of heterozygotic mice There was no difference in the expression of Dlg2 in the cerebellum (t [9.301] = 0.547, p = 0.597, t-test) or hippocampus, (t (21) = À 0.238, p = 0.815, t-test) between WT and Dlg2 +/À mice ( Figure 1A,B). However, there was a reduction in mRNA in the cortex in the Dlg2 +/À mice (t (15) Figure 1C). The expression of Dlg1, Dlg3, or Dlg4 in the same brain regions was not altered in the Dlg2 +/À mice ( Figure S1).
3.2 | Dlg2 +/À mice exhibit impaired reactivity to an acoustic stimulus but intact PPI The average of the first three acoustic stimulus pulses at 120 dB were analysed as an index of emotional reactivity. Dlg2 +/À mice exhibited a reduced reactivity to the 120 dB auditory stimulus (t (40) Figure 2D). By using a lower intensity startle stimulus to ameliorate a potential confound of PPI measurement by the responses to the standard 120 dB acoustic startle stimulus in the mutant animals, we showed that both genotypes showing similar emotional reactivity and PPI to a 105 dB stimulus ( Figure S2). This suggests that failure to see an effect on absolute PPI in the Dlg2 +/À mice is not likely due to a ceiling or floor response effect masking modulation of startle responses. Thus, while the startle response to a loud acoustic stimuli in the Dlg2 +/À mice was reduced at least for a 120 dB startle stimulus, sensorimotor gating was intact in Dlg2 +/À mice.
3.3 | Dlg2 +/À mice demonstrate impaired between-session habituation to a novel context  F I G U R E 2 Reduced acoustic startle responses but intact PPI of acoustic startle in Dlg2 +/À mice. (A) Dlg2 +/À mice (HET) showed reduced emotional reactivity compared to WT as measured as the averaged startle response to the first three startle stimuli at 120 dB. (B) The difference in startle responses between the genotypes persisted across the first 6 pulses at 120 dB. (C) There was a monotonic reduction of the startle response by an increasing intensity prepulse stimulus 40 ms before the 120 dB startle stimulus in both WT and HET mice. This occurred despite a reduced startle response in the HETs without a pre-pulse (P120). (D) Both HET and WT mice exhibit a similar increase in startle response inhibition (% PPI) with pre-pulse intensity when expressed as a percentage of the 120 dB pulse-alone (P120). PP4P120, PP8P120, and PP16P120 = pre-pulse 4 dB, 8 dB and 16 dB above background noise (70 dB), respectively. Error bars are SEM. n = 25 (WT) 17 (HET). HET compared to WT, *p < 0.05 (ANOVA), **p < 0.01 (t-test (A), ANOVA (B) and (C)) The ratio of activity measured on Day 1 and Day 5 was calculated as a measure of between-session habituation. Both WT (t (17) = 10.967, p = 0.001, t-test) and Dlg2 +/À (t (15 = 4.817, p = 0.001, ttest) mice showed a decreased ratio of activity compared to chance performance (0.5) indicating between-session habituation in both genotypes. However, Dlg2 +/À mice demonstrated less change in activity ratio at Day 5 and thus less habituation to the context than WT mice (t (33) = À2.750, p = 0.010, t-test) ( Figure 3B). No differences were found for within-session habituation on any day ( Figure S3). Therefore, while the Dlg2 +/À mice appeared to show normal withinsession habituation and memory of the context exposure, they showed weaker habituation after 5 days.

| Dlg2 +/À mice displayed impaired motor learning on two accelerating rotarod protocols
In an initial experiment (Cohort 1), motor learning and function (performance) were measured using an accelerating (Days 1 and 2) and 3.5 | Dlg2 +/À mice lack increased neuronal activity in M1 following motor learning Using cFos expression as a proxy marker for neuronal activity, 54  ) were first trained and tested using an accelerating rotarod protocol (Days 1 and 2) followed by a fixed speed protocol (Days 3-8) to assess motor learning and function, respectively. Motor learning in Cohort 2 was examined across three days using an accelerating rotarod protocol with more trials per day (7 trials) then the Cohort 1. (B) Cohort 1 Dlg2 +/À mice exhibited a shorter latency to fall than WT mice at Trial 2 only (p = 0.004) on the accelerating rotarod trials. The first 2 trials were conducted on Day 1, remaining trials on Day 2 (C). In Cohort 1, the latency to fall for both genotypes decreased as the speed increased on the fixed rotarod task ( An impairment in motor skill learning but not motor co-ordination in Dlg2 +/À mice was shown in independent cohorts using two different rotarod training protocols. Using a sparse training protocol consisting of short training trials on an accelerating rotarod spread over 2 days, the Dlg2 +/À mice showed a markedly lower improvement in performance between the first two trials on Day 1 as shown by a smaller increase in the latency to fall compared to WT. In WT, the greatest improvement in performance is seen between trials 1 and 2 on Day 1 and is consistent with the typical pattern seen with learning. 64 Nevertheless the performance of the heterozygous mutants matched those in WT during subsequent trials on Day 2. The similarity in performance by the end of training and a formal assessment of motor function on the fixed speed rotarod after motor skill acquisition show that motor function in the Dlg2 +/À mice was unaffected per se, and that the motor learning deficit observed was not confounded by basic deficits in motor function. The pattern of deficit in the period of rapid learning observed on the first day, equating with within-session training, reflects an impact of Dlg2 heterozygosity on the "fast" phase of motor skill learning. 61,64,65 Smaller performance improvements seen in the Dlg2 +/À mice in early trials might also be explained by changes including altered attention, motivation or physiological "warm up" effects that are more specifically associated with withinsession. [66][67][68][69] However, using a more prolonged motor learning protocol (7 trials/day for 3 days) to better observe learning and memory function, Dlg2 +/À mice similarly showed deficits in early withinsession learning. But they also showed a performance deficit that persisted across at least 3 days of training. The mutants showed retention of performance gains between the last trial on Day 1 and the first on Day 2 indicating that between-session consolidation of motor skill memory was intact in the Dlg2 +/À mice. Thus overall, Dlg2 +/À mice show deficits in both "fast" within-session and "slow" betweensession motor skill learning, but intact motor function and memory consolidation.
The within-session motor learning was associated with an increase in cFos expression in M1 motor cortex in WT mice. The expression of cFos is commonly used as molecular marker of synaptic plasticity and learning. [70][71][72][73] The observed increase in expression demonstrates the activation M1 by the rotarod protocols we employed, which is consistent with the critical contribution of M1 to fast motor skill learning in rodents 61 and to fast and slow motor skill learning in both rodents and humans. 74 The failure to see a similar increase in cFos expression in the Dlg2 +/À mice correlates with their impairment in motor learning and implicate DLG2-associated plasticity mechanisms, particularly LTP 41 to support motor learning. An alternative explanation that the large difference in cortical cFos expression between the genotypes simply relates to locomotor activity is less likely because both groups show low levels of motor activity (i.e., low latencies to quit the rotating rod) at this non-asymptotic point in early training.
While the plasticity processes in motor cortex are associated with fast and slow learning, the loci and the contributing circuits, including subcortical structures, are distinct 61,74,75 and likely involve the selective engagement of neuronal subtypes or clusters. 76 Furthermore, the activated networks that represent the motor memory "engram" continue to be refined through both within-and between-session motor skill learning. The deficits in both fast and slow motor skill learning in the Dlg2 +/À mice are consistent with a broad role for DLG2 in motor leaning-associated plasticity processes in M1 (this study) and associated circuits more widely. 40 While the discrepancy we see in our two learning experiments regarding between-session learning, with Dlg2 +/À mice showing persistently lower performance between sessions only when trained using longer sessions over several days, cannot be explained by a difference in the task used (they are the same), it is possible that prolonged training engages different circuits such as cortico-striatal pathways involved in developing habitual responses. 75 We speculate that these circuits may also be more sensitive to the reduced DLG2 dosage in this heterozygous mouse model.
Deficits in between-session habituation to a novel context in the Dlg2 +/À mice manifested as a smaller decrease in locomotor activity after five consecutive days than WT. Repeated exposure to an unfamiliar stimulus, such as a context, results in reduced responses, including locomotor activity, over time as the stimulus becomes less novel. 77,78 Nevertheless, within-session habituation was similar in both genotypes, perhaps reflecting a greater impact of DLG2 heterozygosity on long-term than short-term habituation and the different molecular and plasticity mechanisms supporting them. 79 Notably long-term habituation requires protein synthesis for its production and maintenance 77,80 similar to NMDA receptor-mediated plasticity mechanisms regulating long-term associative memory. 81 Interestingly, habituation impairments have been reported in Gria1 knockout mice. 82 The authors proposed that deficits in habituation are central to psychological processes such as sensitization and aberrant salience that give rise to psychosis associated with SZ and other psychiatric conditions. Together with our data, this suggests disruptions that AMPAR-mediated signalling involving DLG2 at glutamatergic synapses is a mechanism underlying behavioural habituation deficits that are core for psychosis.
To our knowledge this was the first study investigating acoustic startle and PPI in Dlg2 mutants. PPI deficits are a common feature of a range of neuropsychiatric disorders serving as an endophenotype to identify risk genes in humans and to illuminate neural processes and circuits affected in genetic models in animals. 83 The Dlg2 +/À mice show no impairment in PPI. A decrease in reactivity to the standard 120 dB startle stimulus in the mutants is unlikely to account for the inability to detect a change in PPI because PPI was similarly unaffected using a 105 dB intensity startle stimulus where reactivity in the two genotypes was matched. The decreased startle reactivity is unlikely to be due to a gross motor deficit as motor co-ordination and initial locomotor activity an open field were unaffected in the Dlg2 +/À mice. Thus, whilst an explanation regarding the impaired ASRs needs to be elucidated, reduced dosage of Dlg2 does not appear to affect sensorimotor gating in this mouse model.
Our observations of unaffected motor performance and locomotor activity in the Dlg2 +/À mice can be viewed as broadly concordant with previous investigations using DLG2 models when considering gene dosage. Motor coordination and performance in the exon 6 Dlg2 À/À deletion model has been reported to be either unaffected 84 or mildly impaired, but with the impairment absent was in the heterozygous mice. 39 Both Dlg2 À/À 42 (exon 6) and Dlg2 Δ14/Δ14 (the same exon targeted as in the mice used in our study) deletion models show mild hypoactivity activity in novel open fields but not in familiar environments. 40 Together the data suggest that while subtle effects can be seen on motor function-related behaviours in deletion models, these deficits are absent when one copy of Dlg2 is present. The dosage effect may highlight a role for DLG2 in locomotor-related behaviour, including locomotor behaviours associated with novelty detection, but the role is redundant because DLG paralogs are able to compensate for a reduction of DLG2 function. This hypothesis may equally apply to another behavioural phenotype observed in genetic DLG2 models, that of altered social behaviours which is seen in DLG homozygotes 39,40 but not heterozygotes. 39 Thus, the utility of heterozygous models is not only in interrogating more accurate genetic risk models for human disease but also for identifying the essential, or selective functions, associated with the targeted gene. Additionally, it is important to note that we measured reduced expression of DLG2 in some brain regions (cortical) but not others (hippocampus and cerebellum). A differential impact on regional expression levels if recapitulated more widely, and are reflected at the functional protein level, may be another source of specificity and variability in observed functional deficits within and between genetic models.
The selectively of the impairments we see in Dlg2 +/À mice may reflect the impact of DLG haploinsufficiency on long-term synaptic plasticity mechanisms within functional circuitries. Deficits in long-term motor learning and habitation may be indicative of impairments in synaptic plasticity mechanisms in the cortico-striato-thalamo-cortical and cortico-cerebello-thalamo-cortico pathways underlying motor learning (as discussed earlier), and the medial temporal-midbrain-striatal pathways that mediate the novelty-to-familiarly transition of stimuli that manifest as habituation. 85 DLG2 is critical for LTP, which underlies long-term protein-synthesis dependent learning, but not AMPA and NMDA receptor-mediated synaptic transmission 41 that are extant for information processing and behavioural responses not dependent on LTP such as motor co-ordination, PPI, and anxiety. In comparison to DLG2 mutants, DLG4 (PSD-95) models show alterations in both glutamatergic long-term plasticity and basal transmission 41 and also more severe and wider ranging behavioural affects. 38 These observations lend further support for a role for DLG2 in functions specifically dependent on long-term plasticity. It is being established that the multiple paralogs of DLG2 have differential roles in regulating synaptic function. 86,87 Understanding the precise contribution of individual DLG2 variants to functional outputs will be an important goal for future studies and will help us understand the subtly and selectivity of the impact of specific DLG2 mutations on brain development, behaviour and function.

| CONCLUSIONS
In this paper, we present data indicating heterozygosity of Dlg2, which has been previously associated with a number of psychiatric diseases, results in selective behavioural deficits similarly associated with disease in Dlg2 +/À mice. The deficits in motor learning and long-term habituation of locomotor responses to a novel context that we observed in the mutant mice can broadly be considered as failures to adapt to the environment. These are also functions dependent on synaptic plasticity mechanisms, which DLG2 mediates.
The reduced cortical regulation of the plasticity-associated gene cFos during motor learning may indeed be a molecular signature for impairments in synaptic plasticity in DLG2 mutant models. Other functions including sensorimotor gating of an acoustic stimulus measured by PPI, motor co-ordination and anxiety were unaffected in this Dlg2 +/À mouse model. This suggests functional behavioural and cognitive domains associated with long-term adaptation to the environment that also depend on specific long-term plasticity mechanisms are particularly sensitive to haploinsufficiency of DLG2 across psychiatric disorders.

ACKNOWLEDGMENTS
This work was also supported by a Wellcome Trust Strategic Award 'DEFINE' grant no. 100202/Z/12/Z and core support from the Neuroscience and Mental Health Research Institute, Cardiff University, UK.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.