Edinburgh Research Explorer Learning and reaction times in mouse touchscreen tests are differentially impacted by mutations in genes encoding postsynaptic interacting proteins

The postsynaptic terminal of vertebrate excitatory synapses contains a highly conserved multiprotein complex that comprises neurotransmitter receptors, cell-adhesion molecules, scaffold proteins and enzymes, which are essential for brain signalling and plasticity underlying behaviour. Increasingly, mutations in genes that encode postsynaptic proteins belonging to the PSD-95 protein complex, continue to be identified in neurodevelopmental disorders (NDDs) such as autism spectrum disorder, intellectual disability and epilepsy. These disorders are highly heterogeneous, sharing genetic aetiology and comorbid cognitive and behavioural symptoms. Here, by using genetically engineered mice and innovative touchscreen-based cognitive testing, we sought to investigate whether loss-of-function mutations in genes encoding key interactors of the PSD-95 protein complex display shared phenotypes in associative learning, updating of learned associations and reaction times. Our genetic dissection of mice with loss-of-function mutations in Syngap1 , Nlgn3 , Dlgap1 , Dlgap2 and Shank2 showed that distinct components of the PSD-95 protein complex differentially regulate learning, cognitive flexibility and reaction times in cognitive processing. These data provide insights for understanding how human mutations in these genes lead to the manifestation of diverse and complex phenotypes in NDDs.


| INTRODUCTION
The postsynaptic terminal of excitatory synapses in vertebrate species contains a highly conserved set of proteins, including neurotransmitter receptors, cell-adhesion molecules, scaffold proteins and enzymes that are tightly organised into multiprotein complexes -the signalling machinery essential for synaptic transmission and plasticity underlying the regulation of behaviour. [1][2][3][4][5] These multiprotein complexes are organised into a hierarchy, and the most abundant postsynaptic supercomplex at vertebrate excitatory synapses is formed by PSD-95. [5][6][7][8] Through its multiple protein-protein binding domains, PSD-95 is a central organiser at the postsynaptic density (PSD) of excitatory synapses, directly anchoring the N-methyl-D-aspartate subtype of glutamate receptor (NMDAR) at the membrane and assembling a network of proteins around the NMDAR to enable synaptic signalling. 9,10 These interactors include cell adhesion molecules, such as neuroligins, numerous scaffold proteins, including DLGAP/GKAP and Shank, and various downstream cytoplasmic proteins, such as SynGAP, a GTPaseactivating protein (GAP) for Ras. [11][12][13][14][15] A large-scale mouse genetic screen of loss-of-function mutations in postsynaptic proteins showed that mutations in PSD-95 and its close interacting proteins had the strongest phenotypes in synaptic electrophysiology and behaviour, indicating that PSD-95 protein complexes are critical components of the postsynaptic terminal of excitatory synapses. 16,17 While many studies have investigated changes in measures of synaptic signalling and plasticity following mutations in genes encoding postsynaptic proteins, we know less about their roles in complex cognitive behaviour, especially given physiological phenotypes do not always map directly to distinct behavioural measures (e.g., impaired long-term potentiation does not always predict learning performance). 18 Increasing evidence demonstrates that human genetic disorders of cognition, which include neurodevelopmental disorders (NDDs) such as autism spectrum disorder (ASD), intellectual disability (ID), attention deficit hyperactivity disorder (ADHD) and epilepsy, converge on mutations in the postsynaptic proteome, particularly the PSD-95 protein complex. 1,5,9,19 For example, human mutations in SYNGAP1, NLGN3, DLGAP1, DLGAP2 and SHANK2 have been documented in NDDs. [20][21][22][23][24][25][26][27][28] NDDs are highly heterogeneous, but share aetiology (overlapping gene mutations) and comorbid cognitive and behavioural symptoms (impaired cognition, communication, adaptive behaviour and psychomotor skills). 29,30 A diagnosis of a combination of ASD, ID and epilepsy is commonly reported in individual patients. [31][32][33] Towards unravelling this genetic and phenotypic complexity, mice with genetically engineered mutations in genes encoding postsynaptic proteins provide valuable models to understand the impact of discrete mutations on the symptom profile in a mammalian organism. 34 Furthermore, the development of innovative behavioural tools, such as the touchscreen cognitive battery, has enabled the measurements of more complex cognitive behaviours disrupted in NDDs in rodent models. [35][36][37][38][39][40] The combination of these genetic and behavioural testing tools provides opportunities for unravelling the genetic basis of complex behaviours and disease. We have previously shown that mice lacking the Dlg4 gene, which encodes PSD-95, show robust impairments in simple associative learning, 37 whereas PSD-95 heterozygous mice display enhanced performance in the pairwise visual discrimination and reversal learning touchscreen tests. 41 Previous work by us and others has also examined mice carrying mutations in NMDAR subunits in these same behavioural tests and shown that substitution of the GRIN2B intracellular C-terminal domain with GRIN2A, 38 complete loss of GRIN2A 42 or loss of GRIN2B-containing NMDARs on GABAergic interneurons 43 impaired visual discrimination, but did not impact flexibility in reversal learning. These data provide tantalising evidence that distinct molecular components of the NMDAR-PSD-95 protein complex are differentially required for regulating discrimination and reversal learning.
To investigate whether gene mutations encoding proteins found in the postsynaptic NMDAR-PSD-95 multi-protein complex, which directly or indirectly physically interact with each other, display shared phenotypes in associative learning, updating of learned associations and response latencies, here we have used touchscreen-based assays (pairwise visual discrimination and reversal learning) to analyse the performance of mice with loss-of-function mutations in Syngap1, Nlgn3, Dlgap1, Dlgap2 or Shank2. These tasks allowed us to measure the ability to acquire information about the environment and modify behaviour in response to feedback when demands changed, which are processes that shape goal-directed decision making and more complex forms of cognition. Behavioural analysis was collaboratively undertaken across two laboratory sites, Cambridge (UK) and Melbourne (Australia), assessing female Syngap1, Dlgap1, Dlgap2, Shank2 mutant mice and male Nlgn3 mutant mice, respectively (see Materials and Methods). Our results indicate that these distinct components of the NMDAR-PSD-95 protein complex differentially regulate learning, cognitive flexibility and reaction times in cognitive processing. These data provide insights for understanding how human mutations in these genes lead to the manifestation of diverse and complex phenotypes in NDDs.

| Nlgn3 cohort (Melbourne, Australia)
A cohort of male Nlgn3 mutant mice was used for behavioural analysis in the present study. Nlgn3 loss-of-function mice on C57BL/6J background were bred in-house from a colony established with heterozygous male and female breeding founders obtained from Prof. Nils Brose (Max Planck Institute for Experimental Medicine). Details of the mutation and generation of the mice has been described previously. 45 Mice were backcrossed for more than 10 generations to C57BL/6.
Nlgn3 is an X-linked gene, therefore heterozygous females were bred with WT males to generate hemizygous Nlgn3 −/Y mice and WT littermatched controls. We specifically elected not to breed male Nlgn3 −/Y mice to minimise potential confounds, including those associated with previous reports of aggressive behaviour. As it is not possible to generate both male and female Nlgn3 homozygous null mutant mice and littermate-matched WT offspring from the same litter, and due to the additional logistical demands faced in generating two separate breeding regimes to generate male and female Nlgn3 mutant mice, only male mice were utilised for the current study.
Mice were held in a designated animal holding area within the Melbourne Brain Centre, which is a specific pathogen-free facility.

| Food restriction, habituation and pretraining in the touchscreen chambers
Prior to touchscreen testing, mice were food restricted and had their weights gradually reduced to 85%-90% of their free feeding weights over at least 3 days as previously described. 35 Weights were maintained at approximately this level throughout the whole experiment. In experiments conducted in Cambridge, the 85%-90% goal weights for each mouse were scaled up over time using standard strain weight curves to allow for normal growth. During food restriction, water was available ad libitum. Testing was carried out during the dark active phase of the light cycle.
Mice were subsequently trained through several stages of pretraining to acquire operant conditioning to nose-poke stimuli displayed on the touchscreen in order to obtain a reward. 35  Initiate" stage, except if a mouse touched the opposite side of the screen to the stimulus (i.e., the blank side), this resulted in a 5 s "time out" (during which the stimulus was removed, the house light was switched on and no reward was given) to encourage selective responding to the stimulus. After the "time out," a relatively short 5 s "correction ITI" began, and then, the mouse was able to initiate a "correction trial" (CT; a repetition of the preceding trial to which an incorrect response was made). CTs were repeated until a response to the stimulus (correct response) was made. Criterion for this stage was obtaining a response accuracy of ≥75% (23/30 trials) within 40 min over two consecutive sessions (Cambridge) or ≥70% (21/30 trials) within 60 min over two consecutive sessions (Melbourne).

| Pairwise visual discrimination and reversal learning
Following successful completion of pretraining, mice were then tested in the pairwise visual discrimination task. 35 In this test, mice were presented with two stimuli, "Left diagonal" and "Right diagonal." 46 Stimuli were counterbalanced, so that each stimulus was equally designated as the correct (S+; rewarded) and incorrect (S−) across animals of all genotypes. Stimuli were presented spatially pseudorandomly on the screen, one in each window, and remained on the screen until mice made a response. Responses to S+ resulted in the removal of both stimuli and coincided with the reward tone, illumination of the reward magazine and delivery of reward, followed by a 20 s ITI.
Responses to S− resulted in stimulus removal, 5 s time-out signalled by house-light illumination and no reward delivery, followed by a 5 s correction ITI then repeated CTs until mice correctly responded to S+.
All sessions consisted of 30 first presentation trials per session (excluding CTs) except for the first session of visual discrimination testing in Cambridge, which was tested over 2 days in sub-sessions of 15 trials each session. When mice reached the visual discrimination learning criterion (≥80% correct on two consecutive sessions), mice were moved on to the reversal phase the following session. The reversal learning task was like visual discrimination except that S+ and S− were now reversed. To account for high perseveration in the early phase of reversal, which impacts the number of first presentation trials completed per session, the first two reversal learning sessions were split into sub-sessions of 15 trials per session. It should be noted that many mice struggled to complete the required 30 first presentation trials within a daily session from the start of reversal learning for several days. Therefore, if a mouse completed less than 23 trials per day, it was given seven trials or more, as required, on the next day, until the total sum of successive daily trials was 30. In Cambridge cohorts, if the number of trials was over 23 but below 30, the mice were given 31-37 trials on the next day, so that the sum of the first presentation trials in 2 days was 60. Therefore, for the analysis of reversal learning curves, compound sessions comprising usually 30, but in exceptional cases, 23-37 first presentation trials, were used rather than actual daily trials per session. For experiments conducted in Cambridge, animals were trained towards a reversal learning criterion that was the same as visual discrimination (≥80% correct on two consecutive sessions), with mice receiving a minimum of 19-20 compound sessions regardless of when they met this criterion. Some animals that did not attain the reversal criterion within 19 sessions were tested further. For experiments conducted in Melbourne, there was no set reversal learning criterion and all animals were tested for a maximum of 20 sessions of reversal. Therefore, for uniformity, we analysed reversal data per 19-20 compound sessions for all mouse cohorts.
Several parameters were calculated to assess performance during visual discrimination and reversal learning including trials (first presentation, i.e., excluding CTs), errors (incorrect choice on first presentation trials) and CTs. For quantitative assessment of perseverative behaviour during reversal learning, the ratio of CTs to errors (perseveration index) was calculated. Latencies to make correct and incorrect responses, as well as to collect rewards following a correct response were also evaluated. In visual discrimination, because individual mice reached criterion after variable numbers of sessions, we have only analysed latencies for the first 5-7 sessions of testing where all mice were represented. For reversal learning, we analysed latencies for 19-20 compound sessions.

| Data analysis
All statistical analyses were performed with a significance level of 0.05 (adjusted, if necessary, as described below) using GraphPad Prism 8 (GraphPad Software, Inc.). Throughout the text, numerical data are presented as the mean ± standard deviation. In graphs, data are presented as box-whisker plots or as the mean ± standard error of the mean.
Pairwise comparisons between mutants and WT mice (Syngap1 and Nlgn3 cohorts) were performed using the Student's independent In the present study, we assessed the effects of loss-of-function mutations on locomotor activity, operant pretraining, visual discrimination learning, reversal learning and reaction times. Because several parameters for each of these categories were measured to infer the overall effect, we adjusted p-values within each category for each mutant cohort for the family-wise error rate using the Holm-Šídák correction procedure. For example, to analyse significance of effects on reaction times, we stacked p-values for genotype, session and genotype × session effects for latencies to make correct and incorrect touches to the screen and to collect rewards during both visual discrimination and reversal tasks (3 × 3 × 2 = 18 p-values in total) and applied the Holm-Šídák correction, so that the effects were deemed significant only if their unadjusted p-value was below 0.0036-0.0051, depending on the cohort. Response and reward collection latencies in individual sessions were often right-skewed even after log 10 or square root transformations. Therefore, for between-genotype comparisons, median rather than mean latency values were used to represent central tendency measures that would be robust to the effect of outliers.
One Syngap1 +/− mouse failed to achieve the pairwise discrimination learning criterion after 40 daily sessions and was therefore excluded from subsequent testing. Additionally, as highlighted above in the reversal learning section, many mice struggled to complete 30 first presentation trials within a daily session from the start of reversal learning for several days ( Figure S1). The Kaplan-Meier survival analysis of the number of days required for the animals to complete 570 reversal trials equivalent to 19 sessions showed that genotype significantly affected "survival" curves in the Dlgap1 cohort (p = 0.015, log-rank Mantel-Cox test), with mutants requiring more days than WTs (p = 0.0038, log-rank test for trend). For other cohorts, "survival" curves were not statistically different at the chosen level of significance, although in Nlgn3 and Shank2 cohorts, WT mice tended to require more days to complete 570 reversal trials (p = 0.053 and 0.0615, respectively, log-rank Mantel-Cox test; Figure S1). Data from two Dlgap1 −/− mice and one Shank2 +/+ mouse were excluded from the reversal learning analysis because they performed only 459, 450 and 390 trials over 38, 40 and 33 test days, respectively.

| Spontaneous locomotor activity and exploratory behaviour during habituation to the touchscreen chambers
We measured parameters of spontaneous locomotor and exploratory behaviour when the mice were first exposed to the touchscreen chambers during the habituation stage of pre-training and observed signs of hyperactivity in several mutants (Figure 1

| Updating of learned associations in reversal learning
After mice achieved the criterion in the visual discrimination task, the reward contingency of S+ and S− stimuli was reversed to enable investigation of the capacity for reversal learning. Many mice struggled to complete 30 first presentation trials within a daily session from the start of reversal learning for several days (see Materials and Methods and Figure S2). Given this, we assessed differences in response accuracy (i.e., percentage of correct responses; Figure 3 Methods. Significant main effect of genotype is denoted as follows: # p < 0.05; ### p < 0.001. Significant genotype × compound session interaction effects (indicated as § § p < 0.01; § § § p < 0.001) were followed by post hoc Holm-Šidák multiple comparisons tests to reveal differences between mutant mice and WT littermates at individual sessions with significant effects being indicated as follows: *p < 0.05; **p < 0.01, ***p < 0.001. All original p values associated with the effects of genotype, session and genotype × compound session interaction were adjusted for multiple comparisons using the Holm-Šídák correction. +/− heterozygous, −/Y hemizygous, −/− homozygous, +/+ WT Analysis of the perseveration index across reversal learning provides a measure of an animal's tendency to display repetitive behaviour following an incorrect response. As expected, perseverative behaviour for all mice was higher during early reversal sessions and this progressively decreased across subsequent sessions (Figure 3(B)).

| Reaction times during visual discrimination and reversal learning
In human discrimination tests, latencies to respond (reaction times) are taken as an index of processing speed which can vary with cognitive load. 47 Therefore, in addition to our key measures of learning, we examined latencies to make correct and incorrect responses, as well as to collect rewards. While most studies employing the touchscreen visual discrimination and reversal learning tasks commonly report latencies pooled across sessions for the whole task (task-level), we sought to assess latencies at both task ( Figures S4 and S5) and session-by-session levels (Figures 4 and 5) as we have previously seen  Figure S4A) and reversal learning (t 12 = 3.659, adjusted p = 0.0131, Figure S5A). Incorrect response latencies were also faster in Syngap1 +/− mice during both visual discrimination and reversal stages ( Figures S4C and S5C), but only the effect of genotype in reversal learning was significant following correction for multiple testing, (t 12 = 6.052, adjusted p = 0.00034, Figure S4B). Neither analyses revealed any significant differences in reward collection latencies in Syngap1 +/− mice ( Figures S4C and S5C).
In contrast, Nlgn3 −/Y mice displayed slower correct response latencies during acquisition of visual discrimination (Figure 4(A) and to collect rewards following a correct response are illustrated. Significant genotype × compound session interaction (indicated as § p < 0.05; § § p < 0.01; § § § p < 0.001) was followed by post hoc Holm-Šidák multiple comparisons tests to reveal differences between mutant mice and WT littermates at individual sessions with significant effects being indicated as follows: *p < 0.05; **p < 0.01, ***p < 0.001. Significant main effects of genotype are indicated as follows: # p < 0.05; ## p < 0.05. All original p values associated with the effects of genotype, session and genotype × compound session interaction were adjusted for multiple comparisons using the Holm-Šídák correction. +/− heterozygous, −/Y hemizygous, −/− homozygous, +/+ WT to make correct responses, but this decreased at a faster rate to be more comparable to WT littermates as reversal learning sessions progressed ( Figure 5(A)). Similarly, incorrect response latencies were also slower during both visual discrimination and reversal learning in (Figures 4(B) and 5(B) and Figures S4B and S5B). Tasklevel analysis showed this main effect of genotype was statistically significant during visual discrimination learning (t 30 = 3.624, adjusted p = 0.0066, Figure S5B) but not reversal learning, which narrowly missed the significance threshold after correction (F (1,30) = 9.157; adjusted p = 0.063). Neither analyses revealed any differences in reward collection latencies in Nlgn3 −/Y mice (Figures 4(C) and 5(C) and Figures S4C and S5C).  behaviour. 50 However, to the best of our knowledge, the profound reversal learning deficit in Syngap1 +/− mice we observed in the current study has not been previously reported.
In contrast, although Nlgn3 −/Y mice also initially displayed hyperactivity during habituation, like Syngap1 +/− animals, they exhibited slower response latencies than their WT littermates, with faster learning, which was most evident during the reversal stage. Increased locomotor activity in Nlgn3 −/Y animals has been reported previously. 52 Furthermore, although learning in the water maze has been shown to be essentially unperturbed, in the test for reversal learning when the escape platform was relocated, The touchscreen pairwise visual discrimination and reversal learning tests measure the ability to form and update stimulus-reward associations. One may speculate that the rate at which an animal forms the initial association might directly impact the rate at which that association can be flexibly updated. The gene mutations examined in the current study, interestingly, either had the same directional impact on visual discrimination acquisition and reversal learning (i.e., Syngap1 mutants were impaired on both; Nlgn3 and Dlgap2 were faster on both) or did not affect either parameter (Dlgap1 and Shank2 were normal on both). However, we have shown previously that this is not always the case (e.g., Dlg2 mutants showed normal discrimination but impaired reversal; Dlg3 mutants display enhanced discrimination and normal reversal 37 ; whereas GRIN2A 2B C-term mutants showed impaired discrimination and normal reversal). 38  and altered processing speed. [72][73][74] The touchscreen-based visual discrimination and reversal learning assays allow the measurement of associative learning, updating of learned associations and response latencies, reflecting speed of processing in these tests. Our data provides progress towards uncovering the complexities of genotypephenotype relationships, revealing diverse phenotypes that can result from mutations encoding proteins within the same synaptic multiprotein complexes. 4,7,37 In this context, our work reinforces the growing view that there is no singular "one size fits all" animal model of NDDs that would recapitulate the complex and diverse behavioural symptoms observed across patients; therefore collectively, multiple models are essential 75 for how we move forward in the diagnosis, management and treatment of NDDs.
It is now known that NMDAR-PSD-95 multi-protein complex consists of a family of complexes made from different combinations of postsynaptic proteins, and that they are differentially distributed into synapses in different regions of the brain. 7,8 Mapping the location of postsynaptic proteins at single-synapse resolution shows a high diversity of synapses arising from the differential spatial expression of proteins. 76,77 This could be important for interpreting how the mutations give rise to the range of behavioural phenotypes observed in this study. The common or convergent phenotypes could arise from the presence of different proteins in the same synapses, and the expression in different synapses could give rise to distinct phenotypes. It has also been shown that mutations in postsynaptic proteins change the spatial organisation of synapse types, known as synaptome reprogramming, and this may also modify the circuits required for behavioural responses. 76 The rodent touchscreen cognitive platform is increasingly recognised as a unique and valuable tool to dissect and model complex cognitive behaviours of clinical relevance. 37,39,78 Our present study extends previous work, 79 highlighting the robustness of using standardised rodent touchscreen assays across multiple laboratory sites to address concerns of reliability and replicability.