Schizophrenia-related cognitive dysfunction in the Cyclin-D2 knockout mouse model of ventral hippocampal hyperactivity

Elevated activity at the output stage of the anterior hippocampus has been described as a physiological endophenotype of schizophrenia, and its development maps onto the transition from the prodromal to the psychotic state. Interventions that halt the spreading glutamatergic over-activity in this region and thereby the development of overt schizophrenia could be promising therapies. However, animal models with high construct validity to support such pre-clinical development are scarce. The Cyclin-D2 knockout (CD2-KO) mouse model shows a hippocampal parvalbumin-interneuron dysfunction, and its pattern of hippocampal over-activity shares similarities with that seen in prodromal patients. Conducting a comprehensive phenotyping of CD2-KO mice, we found that they displayed novelty-induced hyperlocomotion (a rodent correlate of positive symptoms of schizophrenia), that was largely resistant against D1- and D2-dopamine-receptor antagonism, but responsive to the mGluR2/3-agonist LY379268. In the negative symptom domain, CD2-KO mice showed transiently reduced sucrose-preference (anhedonia), but enhanced interaction with novel mice and objects, as well as normal nest building and incentive motivation. Also, unconditioned anxiety, perseveration, and motor-impulsivity were unaltered. However, in the cognitive domain, CD2-knockouts showed reduced executive function in assays of rule-shift and rule-reversal learning, and also an impairment in working memory, that was resistant against LY379268-treatment. In contrast, sustained attention and forms of spatial and object-related memory that are mediated by short-term habituation of stimulus-specific attention were intact. Our results suggest that CD2-KO mice are a valuable model in translational research targeted at the pharmacoresistant cognitive symptom domain in causal relation to hippocampal over-activity in the prodrome-to-psychosis transition.


Introduction
The re-creation of endophenotypes of psychiatric diseases in animal models has been proposed as a promising strategy to meet the current challenge of identifying rodent models with predictive validity for drug discovery 1 . In schizophrenia, hyperactivity of the output stages of the anterior hippocampus represents a physiological endophenotype that is unique in predicting disease progression from the prodromal state to overt schizophrenia (psychosis) 2,3 . More precisely, baseline metabolic hyperactivity in the anterior CA1 region is elevated in prodromal patients that will develop psychosis in the near future, compared to individuals that do not. The development of hyperactivity in the anterior subiculum, the output stage of the hippocampus that is excited by CA1, in turn, coincides with the onset of psychosis and is accompanied by atrophy of the anterior hippocampus, which might manifest the disease state irreversibly 2 .
Therefore, interventions that ameliorate anterior hippocampal hyperactivity in the prodromal state and thereby halt further disease progression could be a promising therapeutic strategy 4 .
To support this research, rodent models that display hyperactivity of the ventral hippocampus (vHC, which is homologous to the human anterior hippocampus) are desired. In rats, rodent correlates of the positive symptom domain were induced by direct electrical or chemical stimulation of the vHC [5][6][7] . Alternatively, developmental manipulations such as pre-natal methylazoxymethanol acetate (MAM) injections 8,9 , which lead to vHC hyperactivity later (but also prefrontal aberrations), have been used. It is well established, that such vHC hyperactivity entails a hyper-dopaminergic state of the mesolimbic projection 5,8 , and thereby deficits of latent 9,10 and prepulse inhibition 6,7,9 , and enhanced novelty-and amphetamine-induced hyperlocomotion 5,7,8 .
Other studies focused largely on the contribution of a hypofunction of inhibitory interneurons to vHC hyperactivity. It had been suggested previously, that a homeostatic downregulation of the GABAergic output from parvalbumin-positive (PV) interneurons of the CA1region of the hippocampus might lead to an effective disinhibition of the pyramidal neurons of CA1, which would then over-activate the excitatory projection neurons of the ventral subiculum, that project to multiple target areas in the brain that are relevant for affective and cognitive functions 11 . Chemogenetic silencing of vHC interneurons in mice 12 as well as pharmacological vHC disinhibition with picrotoxin in rats 13 could indeed provoke some impairments in the positive and the cognitive symptom domains (especially hyperlocomotion and deficits of spatial short-term memory).
A different set of studies focused on the Cyclin-D2 knockout mouse model (Ccnd2 −/− , or CD2-KO) 14 that displays a reduced number of PV-cells in the hippocampus (20-40 % reduction), particularly in CA1, and some areas of the neocortex (10-30 %), but not the medial prefrontal cortex 15,16 . While the deletion of Ccnd2 affects cell development in the subventricular zone of the medial ganglionic eminence where PV-cells originate from, somatostatin-positive interneurons that derive from the same region, are not decreased in their numbers in the adult CD2-KO brain 16 . Cerebral blood volume (CBV) measurements in resting-state functional magnetic resonance imaging demonstrated that these animals show hyperactivity of the dentate gyrus, CA2/CA3 and CA1 of the vHC 15 but not the ventral subiculum; a pattern that is more akin to the prodromal state (see above) in which patients already present with cognitive symptoms. However, schizophrenia-related cognitive deficits have not yet been systematically examined in this mouse line (see Table 1 for a summary of previous behavioural assessments of Ccnd2 −/− mice) nor in most other rodent models of ventral hippocampal hyperactivity (for exceptions see ref. [13][14][15][16][17]. Most relevant to schizophrenia is the finding, that CD2-KO mice are less efficient in their search strategies in the water-maze and in spatial forms of cognitive flexibility, while basic spatial learning is intact [18][19][20][21] .
Importantly, CD2-KO mice also show a lack of adult neurogenesis in the dentate gyrus 20 , which adds to its translational value given that indicators of reduced neurogenesis were also found in post-mortem brains from patients with schizophrenia 22 . Furthermore, PV-cell hypofunction could be contributing to the neurogenesis deficit, since the innervation of new-born cells by PVinterneurons is critical for their survival and development 23 . In reverse, new-born granule cells activate strong dentate gyrus-feedback-and CA3-feedforward inhibition 24,25 , implying that reduced neurogenesis alone might lead to hippocampal hyperactivity. Importantly though, physiological and some behavioural aberrations in the CD2-KO mouse, could be rescued by implantation of GABAergic precursor cells specifically into the ventral hippocampus, pin-pointing this part of the hippocampus and its over-activity as the primary cause of those abnormalities 15 . This suggests that the Ccnd2 −/− mouse model might be well-suited to study the role of ventral hippocampal hyperactivity in the causation of schizophrenia-related deficits. To that end, we have subjected Ccnd2 −/− mice to a large battery of behavioural tests in the positive, negative, and cognitive symptom domains.

Animals
Cyclin-D2 knockout mice (Ccnd2 -/-, CD2-KO; 18 males, 12 females) 14 and wild-type littermate controls (19 males, 9 females) were generated by crossing heterozygous Ccnd2 -/+ mice. Animal numbers used for each test are stated in the respective figure panels, where appropriate, and in Supplementary Tables 1 and 3. The average age of the cohort at the time of each test is given in Supplementary Table 2. Sizes of same-gender samegenotype subgroups were chosen to be around 10 according to common standards in behavioural neuroscience 26 and, where possible, based on our own biometric estimates using prior data from the same tests 27 . If logistically possible, sub-group sizes of a single gender and genotype were increased to up to 19 to improve statistical power. Animals were group-housed in Type II-Long individually ventilated cages (Greenline, Tecniplast, G), enriched with sawdust, sizzle-nest TM , and cardboard houses (Datesand, UK), and subjected to a 13 h light / 11 h dark cycle. All experiments conformed to the German Animal Rights Law (Tierschutzgesetz) 2013 and were approved by the Federal Ethical Review Committee (Regierungsprädisium Tübingen) of Baden-Württemberg.

Behavioural tests
Detailed protocols for all behavioural tests are given in Supplementary Methods and the majority of protocols were as previously described 27 . Light levels were set to 100 lux for all tasks in open arenas. In brief, in the novel-object recognition test, mice were exposed to two identical objects, A1 and A2, in the sample phase (10 min) andafter a delay of 2 minto an identical copy of the familiar object, A3, and a novel object, B1, in the test phase (5 min). Direct physical interaction between the snout or whiskers with the object were scored manually by an observer blind to genotype, and the extent of novel-object recognition was determined according the ratio of interaction with B1 to interaction with B1 and A3 combined. Spatial novelty preference testing in the Y-maze was conducted analogously: animals explored the start arm and a pseudo-randomly selected goal arm in the sample phase (5 min), while they had access to all arms (including the previously unvisited novel arm) in the test phase (2 min) that followed after a 1 min-delay period during which the sawdust in the maze was mixed to remove olfactory landmarks. In the T-maze test of working memory, 10 trials were conducted per day in fooddeprived (85-90 % of free-feeding weight) animals. Each trial consisted of a sample run with a forced choice into one of the goal arms, and a choice run in which the previously unvisited arm was baited and its visit was counted as a correct choice. Animals were trained up over 5 days in a blocked paradigm with a delay between sample and choice run of 5 s and an inter-trial interval (ITI) of 3-10 min. Subsequently, the protocol was altered every second day starting with a massed design with a delay of 1 s and an ITI of 25 s, followed by protocols in a blocked design with delays of 5, 20, and 60 s. The assessment of spatial reference learning, rule-shift and rule-reversal were conducted in a plus-maze with transparent walls where the start arm (North or South) was chosen pseudorandomly and one of the goal arms (East or West) was baited according to the respective rule and its visit counted as a correct choice. 10 trials were given per day with a delay of 5-10 min between them. Animals were first trained up on an egocentric rule (turn left or turn right). After achieving the performance criterion of 85 % correct choices in 20 consecutive trials, animals were switched to an allo-centric rule (go West or go East) to assess spatial rule-shift learning. After reaching the same 85 % performance criterion the allo-centric rule was reversed (e.g. from go West to go East, or vice versa) to assess reversal learning.
Unconditioned anxiety was measured on the elevated plus-maze, according to the preference with which animals explored the two open vs the two closed arms. Novelty-induced locomotion was assessed by letting animals explore a new Typ-III-cage (Tecniplast, G) with clean sawdust for 2 h. Movement was video-tracked using ANY-maze (San Diego Instruments, CA, USA) in the Ymaze, the elevated plus-maze, and during novelty-induced locomotion. Reciprocal social interaction was assessed in similar cages but with darker walls (Tecniplast), into which a younger adult stimulus mouse of the same sex and strain had been introduced. Social interactions (sniffing, taxing with snout or whiskers, but not aggression or sexual behaviour) were scored offline manually and blind to genotype over a period of 16 min in 2 min intervals.
Sucrose preference was assessed overnight in animals that were single-housed on the first day of the testing sequence, on which they were presented with two bottles of drinking water. Over the 4 subsequent nights one of the two bottles contained either 1 % w/v (2 nights) or 10 % w/ v (2 nights) sucrose in drinking water, while only the water bottle remained in the cages during the day and for the sixth night. During this last night, animals received a Nestlet™ (Datesand) of 2.3-2.6 g, and nests were scored according to an established nest building rating system 28 , next morning.
For the 5-choice-serial reaction time task (5-CSRTT) mice were trained up in operant chambers (Med Associates Inc., VT, USA) as previously described 27 over 4 stages with increasing attentional difficulty until mice reached a consistent performance (≥80 % accuracy, ≤50 % omissions, ≥30 trials) on the stage 4 baseline protocol where the stimulus duration (SD) was 4 s and the inter-trial interval (ITI) time was 5 s. Subsequently, various protocols to challenge either sustained attention (by reduction of the SD to 1 s or distraction, or both) or impulse control (by extension of the ITI to 9 s) were applied over 2 consecutive days, followed by 1-3 d of baseline training before the next challenge. Afterwards, mice were trained on a simplified version of this task where only hole 2 or 4 was illuminated in each trial until mice showed an accuracy of ≥80 % in two consecutive sessions. Subsequently, the rule was shifted from a cued to a spatial version, whereby only hole 2 (or, in other mice, hole 4) was rewarded irrespective if hole 2 or 4 was illuminated in a given trial. To assess rule-shift learning in this operant paradigm, mice were trained until they had reached a criterion of 80 % accuracy across trials in which the correct hole was unlit on 3 consecutive days.
Condensed milk diluted with drinking water was used as reward on rewarded maze tasks, strawberry milk was used in operant box paradigms. Experimenters were blind to genotype in all instances of manual scoring (nest building, social interaction, novel-object recognition) and in all maze-based assays.

Behavioural pharmacology
To assess the effects of the D1R-antagonist SCH23390, the D2R-antagonist raclopride and the mGluR2/3-agonist LY379268 on novelty-induced locomotion in CD2-KO mice, each drug was injected i.p. at an injection volume of 10 ml/kg before an 1 h exposure in a novel open-field, identical to those used for the prior assessment of noveltyinduced locomotion (see above). S(-)raclopride-( + )tartrate salt (Sigma, G) was diluted in 0.3 % TWEEN80/ saline and injected 60 min before the start of the run. SCH23390 hydrochloride and LY379268 disodium salt (Tocris, G) were diluted in saline and injected 30 min or 45 min, respectively, before the run. For dose-response assessments of these drugs, separate adult male C57BL/6 cohorts were used in a between-subject design (4-8 mice per subgroup); either raclopride, SCH23390 or LY379268 were injected 30 min, 30 min, or 45 min, respectively, before a 30 min exposure to the novel open-field, followed by injection of 0.2 mg/kg MK-801 maleate (in saline) 27 and another 60 min of locomotor monitoring. For the Tmaze test, LY379268 was injected 50-60 min before testing commenced. Additionally, amphetamine-induced locomotion was tested by injection of D-amphetamine hemisulfate (Sigma; diluted in saline), 30 min after an acclimatisation period in the novel open-field, immediately followed by another hour of post-injection exposure. Latin-square designs with a fixed decreasing order of doses, counterbalanced for starting dose and genotype, were used for all within-subject studies.

Analysis
Within-subject data was analysed by repeated-measures ANOVA, and between-subject data with univariate ANOVA (subsequently termed "ANOVA" only), one-way ANOVA in cases of one-factorial design with more than 2 groups, and t-tests, orin cases of non-parametric datathe Mann-Whitney-U-Test (termed MWU) in cases of individual comparisons between two groups only, as appropriate. All graphs display mean ± s.e.m. Each experiment was conducted once in the cohorts of the stated sizes (n); for the confirmation of the effects of raclopride and LY379268 on novelty-induced locomotion, one between-and one within-subject-study were conducted each.

Sustained novelty-induced hyperlocomotion in CD2-KO mice
Novelty-induced hyperlocomotion was assessed over a 2 h exploration time. A repeated-measures ANOVA over this period split into 24 5-minute intervals revealed a significant effect of genotype (P genotype < 0.0005) indicating that KO mice were hyperlocomotive (Fig. 1a, b; see Supplementary Table S1 for statistical details on this and subsequent results including n-, F-and t-values). Additionally, the number of rotations, an indicator of stereotypies, was significantly elevated in knockouts of both genders (P genotype = 0.001; ANOVA, Fig. 1c).

Normal amphetamine-induced hyperlocomotion in CD2-KO mice
In a similar locomotor activity paradigm, injection of 2 mg/kg amphetamine after an acclimatisation period of 30 min resulted in a significant subsequent increase of locomotion relative to the vehicle and 1 mg/kg amphetamine conditions in both genotypes (P dose < 0.0005, P genotype = 0.015, Pgenotype*dose = 0.957, ANOVA; betweensubject design; Fig. 1d, e). Univariate ANOVA within each dose revealed that the higher locomotion of CD2knockouts relative to controls seen qualitatively at each dose, only reached significance after vehicle (P = 0.007) but not after amphetamine injections (P > 0.1). Likewise, when expressing locomotor activity post-amphetamine relative to locomotor activity before injection there was no difference between genotypes (P = 0.759), while the effect of dose remained (P < 0.0005, ANOVA). This suggests that CD2-KO mice are similarly sensitive to amphetamine as controls; if anything, amphetamine aligns the wildtype with the knockout levels of locomotor activity (Fig. 1f).

Incomplete reduction of novelty-induced locomotion by the D2/3 R antagonist raclopride
Signalling through dopamine-receptors of the D2-type (D2R) in the nucleus accumbens mediates the level of spontaneous locomotor drive in rodents 29 , and antagonism of these receptors is the suggested mechanism of the antipsychotic action of neuroleptics. We therefore tested, if the novelty-induced hyperlocomotion in CD2-KO mice is indeed responsive to the D2R-antagonist raclopride. We first confirmed in a dose-response study in a separate group of male C57BL/6 J mice, that the doses 0.33, 1 and 3 mg/kg raclopride were effective in reducing hyperlocomotion induced either by novelty or by the NMDA-receptor blocker MK-801 (Fig. 1g). Using the two lowest effective doses 0.33 and 1 mg/kg, we firstly performed a between-subject study in CD2-KO mice and controls. While both doses of raclopride significantly reduced locomotion (distance moved) within both genotypes, surprisingly, knockout mice remained significantly hyperlocomotive compared to wildtype controls injected with the same doses (P < 0.0005, ANOVA; P < 0.05 for ANOVAs at each dose, Fig. 1h). That means, although 0.33 and 1 mg/kg could reduce locomotion in knockouts to the level of vehicle-injected wildtypes, these doses did not alleviate the effect of genotype within each dose. Also, 0.33 mg/kg raclopride could significantly reduce the number of rotations in wildtype mice compared to vehicle (P = 0.004, Tukey post-hoc test), but not in knockout mice (P = 0.646), suggesting that the drug is even less effective in the latter (Supplementary Table S1).
To investigate this further, a subset of mice were continued in a within-subject regime with injections of vehicle and the doses 0.11, 0.33, 1 and 3 mg/kg. Again, at all doses, knockout mice remained hyperlocomotive compared to wildtype controls (P < 0.0005, repeatedmeasures ANOVA; P < 0.05 for univariate ANOVAs). Across all 5 doses a significant dose*genotype interaction was apparent (P = 0.033, Fig. 1i), indicating a differential efficacy of raclopride. Expressing the amount of locomotion as percentage of the locomotion each mouse displayed in its vehicle-trial revealed that the higher doses, 1 and 3 mg/kg, reduced novelty-induced locomotion significantly less in knockout mice compared to wildtype controls (P < 0.05, ANOVA, Fig. 1j), supporting the view that the drug is less effective in CD2 knockouts. This suggests that, while D2Rs may mediate baseline locomotion, there is an additional D2R-independent mechanism of increasing locomotor drive in CD2-knockouts. This is reminiscent of the hyperlocomotion produced by electrical (20 Hz) stimulation of the ventral hippocampus in rats, which is mediated by D1-rather than D2receptors 5 . at the stated doses on novelty-induced locomotion, plotting total distance moved in 1 h. (j, n, r) Within-subject assessment of raclopride (j), SCH23390 (n), and LY379268 (r) at the stated doses on novelty-induced locomotion, plotting distance moved in 1 h (as %) relative to the vehicle (0) value in each animal. (l, p) Same data as in (m, q), within-subject assessment of SCH23390 (l) and LY379268 (p) at the stated doses on noveltyinduced locomotion, plotting distance moved in 5 min intervals. Wildtype controls, grey; CD2-KO, black; f, females, m, males; numbers in brackets denote n for each group. All doses stated in mg/kg. * P < 0.05, ** P < 0.01, *** P ≤ 0.001. For trends, P values are stated as numbers. Center values represent mean, Error bars, s.e.m.

Incomplete reduction of novelty-induced locomotion by the D1R antagonist SCH23390
Therefore, we repeated this experiment with the dopamine D1-receptor (D1R) antagonist SCH23390. Assessing the efficacy of 0.011, 0.033, 0.1 and 0.3 mg/kg against novelty-and MK-801-induced locomotion in a separate cohort of C57BL6/J males revealed that the two intermediate doses 0.033 and 0.1 mg/kg had the strongest, yet statistically moderate, effects (between-subject design; Fig.  1k). Those two doses were tested against novelty-induced locomotion in CD2-KO mice and controls (within-subject design). While the drug significantly reduced locomotion dose-dependently in both genotypes, KO mice again remained hyperlocomotive relative to wildtype controls receiving the same dose (P genotype < 0.0005, P dose < 0.0005, P dose*genotype = 0.016, repeated-measures ANOVA; Fig.  1l-m). Expressing locomotor activity under each dose relative to distance moved after injection of vehicle confirmed that the drug was less effective in reducing hyperlocomotion in CD2-KO mice compared to wildtypes (P genotype = 0.02, repeated-measures ANOVA; Fig. 1n).
Similar to the results obtained with D2R-and D1Rspecific antagonists, the antipsychotic clozapine, which acts as antagonist on all dopamine receptors, could reduce novelty-induced hyperlocomotion only at a relatively sedative dose (3 mg/kg) but not at an intermediate dose (1 mg/kg) in CD2-KO mice (see Supplementary Figure 1).
These data suggest that a largely dopamine-independent component of locomotor drive is elevated in CD2knockouts.

Normalization of novelty-induced locomotion by the mGluR2/3 agonist LY379268
Agonists of the metabotropic glutamate-receptor type 2/ 3 (mGluR2/3) have been proposed to be able to scale down glutamatergic transmission 30 , and one compound from this family, LY379268, has been shown to normalize ventral hippocampal hyperactivity evoked by chronic ketamine application in mice 2 . In a separate cohort of adult male C57BL/6J mice, we found that a dose of 1 mg/kg LY379268 could reduce novelty-induced but not MK-801induced hyperlocomotion, while 3 mg/kg strongly reduced novelty-induced locomotion and moderately normalized MK-801-induced locomotion (Fig. 1o). Testing both doses in CD2-KO mice (within-subject design), we found that 3 mg/kg can completely normalize noveltyinduced locomotion in CD2-KO mice relative to wildtype controls receiving the same dose (P genotype = 0.941), while 1 mg/kg achieved a partial reduction (P genotype = 0.029 at 1 mg/kg and P genotype < 0.0005 at vehicle, ANOVA, Fig. 1p, q). Indeed, LY379268 appeared to reduce locomotion even stronger in knockouts than in wildtypes (P genotype*dose = 0.01) leading to significantly less relative locomotion at 3 mg/kg compared to wildtype controls (P genotype = 0.043, ANOVA; Fig. 1r). We confirmed the finding of a significant interaction between dose (vehicle vs 3 mg/kg) and genotype in a between-subject study in an extended cohort (P genotype*dose = 0.015, n: 15 KO, 20 WT; data not shown).

CD2-KO mice are impaired in various forms of cognitive flexibility
Impairments of executive function are a hallmark of schizophrenia, and reduced capacity to learn a rule-shift in a spatial associative learning task in the plus-maze has been demonstrated in Ccnd2 −/− mice previously 19 . Firstly, we sought to confirm this finding. KO mice acquired a spatial long-term memory according to an egocentric rule similarly well as wildtypes: the number of trials needed to reach the criterion of 85 % correct of the last consecutive 20 trials did not differ significantly between genotypes (P genotype = 0.232, t-test), and the learning curves were similar (P genotype = 0.367, repeatedmeasures ANOVA over the second half of the number of training sessions needed on average to reach criterion;  Table 3 for statistical analysis of this data and all subsequent datasets related to cognitive flexibility). This indicates, that CD2-KO mice can process spatial information and acquire a spatial memory largely normally. Once animals had reached criterion, the associative rule was shifted to an allo-centric one 19 . In this task, KO mice needed significantly more trials to reach criterion again (P genotype = 0.04, t-test) and average performance was lower compared to wildtypes (P genotype = 0.048, repeated-measures ANOVA over the second half of the training sessions needed on average to reach criterion; Fig. 2d-f).
Once mice had reached criterion again, rule-reversal learning was assessed: the allo-centric spatial location that was associated with a reward was switched. Similarly to the rule-shift, KO mice required more trials than controls to reach criterion again (P = 0.021, t-test) and their performance was inferior (P = 0.01, repeatedmeasures ANOVA over the second half of the number of training sessions needed on average to reach criterion). Note that no significant difference was seen when analysing all training sessions (P = 0.497; Supplementary Table 3a).
However, rule-shift learning is usually attributed to prefrontal rather than hippocampal dysfunction 31 . Therefore, we sought confirmation of our surprising finding by conducting a newly developed operant-box paradigm which implements a rule-shift within the visual domain and thereby resembles the Wisconsin cardsorting test (which can be indicative of prefrontal dysfunction in humans; Fig. 2j): mice that had been trained up in the 5-choice-serial-reaction-time task (see below) in which a rewarded nose-poke is indicated by a cue light, were subjected to a rule-shift whereby cue-lights were rendered meaningless while the fixed spatial location of the hole now indicated the correct choice. KO mice required significantly more trials to reach the criterion of 80 % accuracy in the choice of the unlit rewarded hole on three consecutive days (P = 0.001, t-test) and showed lower accuracy during training (P = 0.009, repeatedmeasures ANOVA over the second half of the training sessions needed on average to reach criterion, Fig. 2k, l).
Surprisingly, in both rule-shift assays, CD2-knockouts did not display a significantly higher number of perseverative errors. In contrast, their lower performance in the plus-maze was mainly due to erroneous choices that were never reinforced by the prior training, in line with a previous study 19 Table 3).

CD2-KO mice are unimpaired in object-related and spatial short-term habituation
Dysfunction of sustained and stimulus-specific attention (attentional control) have been related to schizophrenia 32,33 , and could explain deficits in other cognitive tests conducted in this study. We first assessed a form of spatial short-term memory that is largely mediated by short-term habituation of attention using the spatial novelty-preference test in the Y-maze 34 . In this test, habituation is measured as a preference for a novel as opposed to a familiar choice arm (or 'space'). In the sample phase, KO mice showed significant hyperlocomotion (P = 0.018, univariate ANOVA) but spent a similar amount of time in the to-be-familiar arm (P = 0.425) as wildtype mice, suggesting equal levels of exploration (Fig. 3a-c). In the test phase, mice of both genotypes explored the novel arm significantly more than the familiar arm, in terms of both entries and time (P < 0.0005, repeated-measures ANOVA, Fig. 3d, e), and there were no genotype-related differences in the preference ratios for the novel arm (P > 0.2, ANOVA; Fig. 3f). Additionally, preference ratios were significantly above chance level in both genotypes (P < 0.005, one-sample t-test).
Novel-object recognition represents an object-related version of the novelty-preference test of habituationdriven short-term memory (Fig. 3g) 33,35 . CD2-KO mice, compared to controls, showed strongly increased exploration of the objects in the sample phase and the test phase (P < 0.01; ANOVA; Figure 3h-j). However, mice of both genotypes showed significantly higher exploration of the novel compared to the familiar object in the test phase (P < 0.0005, repeated-measures ANOVA), and did not differ in their preference for the novel object (P > 0.4, ANOVA; Fig. 3k). In summary, this indicates, that CD2-KO mice are not impaired in short-term habituation of stimulus-specific attention.

CD2-KO mice are not impaired in sustained attention, motivation, perseveration or impulse control
To assess sustained attention, CD2-knockouts and wildtype controls were trained on the 5-choice serial reaction time task (5-CSRTT) through 4 stages of increasing difficulty, before being challenged with different protocols (see Supplementary Methods). Mice of both genotypes learned its basic operant cycle in the habituation training similarly well, in terms of required trials and obtained rewards (Supplementary Table 1a). In the baseline protocol of the 5-CSRTT, which was conducted with a stimulus duration (SD) of 4 s and an inter-trialinterval (ITI) of 5 s, knockout mice appeared significantly better in the primary measure of attention, % accuracy (P = 0.038; ANOVA), although in each of the gender groups alone there was no significant effect of genotype (P > 0.05, t-test; Fig. 3m). In a measure of inattentiveness, %omissions, female knockouts performed worse than female wildtype mice (P = 0.005, t-test), although, qualitatively, the opposite was observable in the male cohort, yielding a significant genotype*gender interaction (P = 0.006, ANOVA; Fig. 3n). The basic measure of motivation, reward latency, did not differ between genotypes (P > 0.3, ANOVA), but females, irrespective of genotype, had a longer reward latency than males (P gender < 0.05, P gender*genotype = 0.307, ANOVA; Fig. 3o). Additionally, perseveration, indicated by unrewarded repetition of nosepokes into previously rewarded holes, was not significantly higher in CD2-KO mice compared to controls (P = 0.303, ANOVA; Fig. 3p).
After baseline training, four challenge protocols were conducted, out of which three were targeted at sustained attention (shortening of the SD, distraction during the ITI, or both), and one at impulse control (increasing the ITI to 9 s). Within 60 sessions, 2 wildtype males and 7 of the 13 involved females (KO and WT) did not reach the criterion in the baseline stage to proceed to challenge protocols, which were therefore only conducted in males. Each challenge protocol was conducted over two consecutive days and was preceded and succeeded by sessions with the baseline protocol. For analysis of each parameter, the values from the two challenge days were averaged and a repeated-measures ANOVA was calculated by including the performance of the preceding baseline session.
Regarding impulse control, all challenges increased the primary measure of motor impulsivity, % premature responses (P < 0.0005 for all challenges except the exclusive distraction (P = 0.015), repeated-measures ANOVA). Similarly, all challenges decreased attentional accuracy (P < 0.0005 for all attention challenges, P = 0.001 for the impulsivity challenge). A trend of slightly better attention in CD2-KO mice at baseline (see above) was seen in the baseline sessions of most protocols (Fig. 3q-t). However, in knockout mice attentional accuracy was affected significantly stronger by the first (and therefore unexpected) attentional challenge (P challenge*genotype = 0.006), and a similar trend (P challenge*genotype = 0.066) was seen with the first distraction challenge (Fig. 3q, s). However, that effect was absent at the later combined SD-reduction/distraction challenge (Fig. 3t), suggesting it does not reflect a genuine, reproducible attentional deficit. In line with this interpretation, CD2-KO mice even displayed significantly lower omission rates than controls under both distractionchallenge conditions (P < 0.05, t-tests, Fig. 3s, t).
Finally, there were no significant effects of genotype nor genotype*challenge interactions (P > 0.05) detectable on any of the protocols in the measures reward latency, % premature, or %perseverative responses (Fig. 3q-t).
In summary, CD2-knockout mice appeared largely normal on all psychiatrically relevant parameters measured by the 5-CSRTT and were unimpaired in acquiring an associative memory during operant learning.

CD2-KO mice show a working memory impairment that is resistant against LY379268
To assess spatial working memory, mice were trained on a rewarded alternation (delayed-non-match-to-place) task on the T-maze over 5 days in a blocked paradigm with a delay (intra-trial interval) of 5 s and an inter-trial interval of > 3 min. While female knockouts did not differ from controls in this initial training paradigm, male knockouts performed worse across those training sessions (females: P genotype = 0.497; males: P genotype = 0.005; both genders: P genotype = 0.013, P gender = 0.002, P genotype*gender = 0.123, P day = 0.001, repeated-measures ANOVA; Fig. 4a-c). Subsequently, mice were challenged by different alterations of the paradigm, each performed in two consecutive sessions: firstly, the delay was maximally reduced ("1 s") and the paradigm changed to massed trials with an inter-trial interval of 25 s, which had revealed subtle deficits in mice with PV-interneuron-specific NMDA receptor hypofunction previously 27,36 . Then, blocked paradigms with increasing delay (5 s, 20 s, 60 s) were conducted. Across paradigms and genders, knockout mice were impaired (P < 0.0005) and females of both genotypes performed significantly better than males (P < 0.0005; repeated-measures ANOVA). When assessing individual challenges by univariate ANOVAs across genders, all protocols except the 60s-delay revealed significantly reduced performance in knockouts (P < 0.05; Fig. 4d). However, when analysing females alone in the 60 s-protocol, again CD2-KOs performed significantly worse (P = 0.024, t-test), suggesting that the strongly reduced performance in male wildtype mice prevented the detectability of an effect of genotype across genders in this protocol (Fig. 4e).
We next examined, if the mGlu2/3R-agonist LY379268 was also effective against the working memory deficit in CD2-KO mice as it was against their hyperlocomotion. However, across doses, the lower performance of knockouts remained apparent (P genotype = 0.007, P > 0.05 for effects of dose, gender and all interactions, repeatedmeasures ANOVA, within-subject design, Fig. 4f).
These data imply that CD2-KO mice show a delayindependent impairment of working memory, which is resistant against treatment with mGluR2/3-agonists.

CD2-KO mice show normal unconditioned anxiety
During the testing of novelty-induced locomotion in an open field (see above), we found no effect of genotype on the average distance to the cage walls (P > 0.1 for all , percent omissions (n)), incentive motivation (latency to reward receptacle, (o)), and perseveration (percent perseverative responses into the correct hole, (p)) in the baseline protocol (4 s stimulus duration, SD; 5 s inter-trial-interval, ITI; average of the last 4 sessions before the challenges started). (q-t) Measures of attention (accuracy, left; %ommissions, second from left), impulsivity (percent premature responses relative to all initiated trials, middle), perseveration (percent perseverative responses, second from right) and motivation (reward latency, right) in various challenge protocols (right value in all subpanels, average of 2 consecutive sessions) relative to the baseline protocol immediately before the challenge (left value in all subpanels). Attention was challenged by a reduction of the SD to 1 s (q), a random distraction by switching of the house light for 0.5 s during the ITI (s), and a combination of SD-reduction to 1 s and random switching the house light off for 1 s (t). Impulsivity was challenged by increasing the waiting time (ITI) until the holes were illuminated to 9 s (r). Wildtype controls, grey; CD2-KO, black; f, females, m, males; numbers in brackets denote n for each group. # P < 0.1, * P < 0.05, ** P < 0.01, *** P ≤ 0.001. In within-subject data (line graphs), symbols on horizontal lines indicate an effect of the within-subject factor (not drawn in (q-t) for clarity), symbols on vertical lines indicate effects of genotype across conditions (ANOVA), symbols on the connecting lines indicate an interaction between both factors, and symbols below the data point of the challenge conditions (q-t) indicate a significant effect of genotype in the challenge data alone (t-test). The yellow lines indicate chance-level performance. Error bars, s.e.m.
factors, ANOVA, Fig. 5a), suggesting similar levels of unconditioned anxiety. Likewise, on the elevated plusmaze, animals of both genotypes spent significantly more time in the closed compared to the open arms (P arm < 0.0005, repeated-measures ANOVA, Fig. 5b). Accordingly, the preference ratio for the open arm (time spent in open arms divided by time spent in all arms) was not significantly different between genotypes (P = 0.488), but was significantly lower in females compared to males (P = 0.003, univariate ANOVA, Fig. 5c). Preference ratios calculated from the number of entries did not differ between any groups (P > 0.1, ANOVA, Fig. 5d). This indicates that anxiety is unaltered in CD2-knockouts.

CD2-KO mice show transient anhedonia in the sucrose preference test
Analysing the preference ratio for sucrose (consumed volume of liquid from sucrose bottle relative to total consumed liquid) over five nights, we found a strong effect of night (indicating sucrose concentration; P < 0.0005) but also a night*genotype interaction (P = 0.007) without an effect of genotype (P = 0.647, repeatedmeasures ANOVA), suggesting that both genotypes develop a strong preference for sucrose but with a different trajectory (Fig. 5e-g). This divergence, resulted largely from the first encounter with sucrose (night 2), where an interaction between genotype and bottlecontent (water vs sucrose, P = 0.001) on the consumed liquid volume, as well as a significantly reduced consumed volume of sucrose solution (P = 0.014) and sucrose preference in knockouts (P = 0.012, ANOVAs) were apparent. Such a phenotype of mild anhedonia at the beginning of sucrose-preference testing in CD2-KO mice was described previously 21 .

Nest building is normal in CD2-KO mice
We measured nest building as a compound indicator of hippocampus-dependent social and motivational functioning 28 . Neither the nest score nor the amount of untorn material (bits of > 0.1 g) were significantly different between genotypes (P > 0.2 for within-gender comparisons of both parameters, Mann-Whitney-U-test, Fig. 5h-j).

CD2-KO mice show enhanced reciprocal social interaction
To assess social interaction, test mice were left to interact with younger, same-sex C57BL/6 stimulus mice in novel cages. Surprisingly, the total interaction time and the number of interactions were significantly higher in CD2-KO mice compared to wildtypes (P genotype < 0.05, Performance (correct choices out of 10 trials, in %) over the 5 initial training days shows a significant effect of genotype across the total cohort (b), which is driven by the male subgroup (c, right), rather than the females (c, left). (d, e) Performance under 4 different challenge protocols conducted in the same order as plotted with 2 consecutive days per protocol (averaged for the display) shown for the whole cohort (d) and the genders separately (e); protocols differed by their design (massed vs blocked) and the delay between sample and choice trial as indicated on the x-axes. (f) After a re-training (identical to the previous one (b), data not shown) either vehicle (0), 1 or 3 mg/kg LY379268 was injected into CD2-KO mice and controls prior to testing sessions in the massed design (1 s delay). Performance (correct choices out of 10 trials, in %) is shown for both genders combined. Wildtype controls, grey; CD2-KO, black; f, females, m, males; numbers in brackets denote n for each group. * P < 0.05, ** P < 0.01. In within-subject data (line graphs), symbols above data points indicate an effect of group in the indicated condition, symbols on vertical lines indicate effects of genotype across conditions. The yellow lines indicate chance-level performance. Error bars, s.e.m. ANOVA; Fig. 5k, l). Males showed more interaction than females (P gender < 0.0005 for both parameters), and the increased interaction in CD2-KO mice reached significance only in the male subgroup in both parameters (P males < 0.02, P females > 0.1, t-tests). This difference was mainly driven by the early stages of the test, in which CD2-KO mice showed a sustained high level of interaction time (P interval*genotype = 0.003, repeated-measures ANOVA; Fig. 5n).

Discussion
We here did a comprehensive behavioural assessment of deficits across the three symptom domains of schizophrenia in the Cyclin-D2-KO mouse model of ventral hippocampal hyperactivity. It was shown previously, that this mouse line gains its construct validity as a model of schizophrenia from (1) a hypofunction of parvalbuminpositive interneurons, (2) reduced hippocampal volume, and (3) a lack of adult neurogenesis -three cellular pathologies observed in schizophrenia 2,15,16,[20][21][22]37 . It also displays a mesolimbic hyperdopaminergic state, behavioural hyperactivity, and elevated activity of hippocampal subfields CA1-3, all of which could be ameliorated by increasing inhibition in the ventral hippocampus in a previous study 15 . This prior work suggests that CD2-KOmice could provide a unique rodent model to understand the relationship between the endophenotype of hippocampal hyperactivity -analogously observed in patients (e-g) Paradigm and data from the sucrose preference test of anhedonia. (f) Absolute consumption volumes from the bottles with indicated sucrose concentration in drinking water over 5 nights, and (g) preference ratios (consumed volume of sucrose-water/total consumed liquid) over the 5 nights, with the yellow line indicating lack of preference. Note that both bottles contained water in the first night. (h-j) Paradigm and data for nest-building assessment according to (i) the qualitative nest score (1-5 rating scale with 5 indicating the best nests 28 ), and (j) left-over untorn material relative to weight of the original nestlet. (k-n) Paradigm and data for assessment of reciprocal social interaction with an unknown younger adult mouse of the same sex and strain. (l) Total time of social interactions and (m) total number of interactions over 16 min. (n) Interaction time in 2 min bins. Wildtype controls, grey; CD2-KO, black; f, females, m, males; numbers in brackets denote n for each group. * P < 0.05, ** P < 0.01, *** P ≤ 0.001. For trends, P-values are stated as numbers. Error bars, s.e.m.