Dysfunctional cerebellar Purkinje cells contribute to autism-like behaviour in Shank2-deficient mice

Loss-of-function mutations in the gene encoding the postsynaptic scaffolding protein SHANK2 are a highly penetrant cause of autism spectrum disorders (ASD) involving cerebellum-related motor problems. Recent studies have implicated cerebellar pathology in the aetiology of ASD. Here we evaluate the possibility that cerebellar Purkinje cells (PCs) represent a critical locus of ASD-like pathophysiology in mice lacking Shank2. Absence of Shank2 impairs both PC intrinsic plasticity and induction of long-term potentiation at the parallel fibre to PC synapse. Moreover, inhibitory input onto PCs is significantly enhanced, most prominently in the posterior lobe where simple spike (SS) regularity is most affected. Using PC-specific Shank2 knockouts, we replicate alterations of SS regularity in vivo and establish cerebellar dependence of ASD-like behavioural phenotypes in motor learning and social interaction. These data highlight the importance of Shank2 for PC function, and support a model by which cerebellar pathology is prominent in certain forms of ASD.

A utism spectrum disorders (ASD) are neurodevelopmental disease entities primarily defined by deficits in social interaction and repetitive behaviour 1 . In addition, individuals with autism often suffer from motor skill deficiencies 2 , many of which manifest early in the disease 3 . The aetiology of ASD is complex with reported pathophysiological alterations encompassing multiple brain regions, including the cerebellum 1 . Cerebellum-related motor symptoms of ASD patients have been observed through impairments in eye-blink conditioning 4,5 , eye movement abnormalities 6,7 , general motor learning deficits 8,9 , as well as balance and postural difficulties 10,11 . Patients with cerebellar lesions emerging later in development are often diagnosed with cerebellar cognitive affective syndrome, a condition characterized by deficits in language, executive function and emotion regulation which overlaps considerably with symptoms in ASD 12 . Anatomical evidence for cerebellar involvement in ASD includes a decrease in the number of Purkinje cells (PCs) by post-mortem brain histopathological examination 13,14 and functional connectivity between the cerebellum, and frontoparietal and sensorimotor regions in resting-state fMRI studies of ASD 15 . Moreover, the cerebellum is among the most prominent brain regions demonstrating high co-expression of ASD-associated genes 16 .
Emerging data indicate that neurodevelopmental disorders including ASD result from dysfunctional synaptic networks 17,18 . The postsynaptic density (PSD) in particular represents a critically important proteomic hub for a considerable proportion of mutations causing neurodevelopmental diseases, including ASD 19 . A prominent example is the Shank family of postsynaptic scaffolding proteins, which has gained wide attention because of their strong link to ASD [20][21][22][23][24] . To date, two studies have independently reported generating Shank2 knockout (KO) mice with ASD-like behaviour and abnormal hippocampal processing 25,26 . However, beyond the forebrain, Shank2 is also highly expressed in cerebellar PCs 27,28 . Moreover, patients with SHANK2-related ASD exhibit motor impairments consistent with cerebellar dysfunction 29 . However, the causal influence of cerebellar dysfunction on Shank2-related models of ASD has never been established.
We therefore used both global germ-line Shank2 KO (Shank2 À / À ) and PC-specific Shank2 KO (L7-Shank2 À / À ) mouse models to investigate the causal influence of Shank2 on cerebellar function and ASD-related behaviours. Notably, Shank2 À / À mice have impairments in plasticity at the parallel fibre (PF) to PC synapse, increased inhibitory input onto PCs and significant irregularities in PC simple spike (SS) activity. Moreover, L7-Shank2 À / À mice show deficits in social interaction and exhibit task-specific repetitive behaviour. Together, these results provide novel insight into the pathophysiological mechanisms by which SHANK2 mutations cause impairments in cerebellar function that may contribute to ASD.
Expression of Shank2 in L7-Shank2 À / À mice. To explore the behavioural impact of the lack of Shank2 in PCs, we generated a PC-specific KO of Shank2 (see Methods section) using the floxed version of the Shank2 À / À mutants 25 and the L7-vector 32 . Immunocytochemical analysis with the SA5193 rabbit primary SHANK2 antibody 25,36 confirmed that SHANK2 was specifically deleted in PCs in the PC-specific L7-Shank2 À / À mice, but not in WT littermates, whereas it was ubiquitously deleted in the global Shank2 À / À (Fig. 5). Importantly, PC-specific deletion of Shank2 had no discernible impact on cellular zebrin identity or on the zonal patterns of zebrin staining across cerebellar modules ( Supplementary Fig. 4).
Abnormal social and repetitive behaviour in L7-Shank2 À / À mice. We next examined social and repetitive ASD-like behaviours in the PC-specific L7-Shank2 À / À mice. The three-chamber social interaction task is a widely used social interaction paradigm for evaluating ASD-like behaviour in mouse models of autism 25,26 . WT mice exhibited a normal preference for the chamber in which the stranger mouse (S1) was present, compared with the empty chamber (P ¼ 0.0002, MWU-test) (Fig. 8a). In contrast, L7-Shank2 À / À mice had no preference for either S1 or the empty chamber (P ¼ 0.7, MWU-test) (Fig. 8b). Comparing the preference index (stranger-empty) between WT and L7-Shank2 À / À mice revealed a significantly decreased preference of L7-Shank2 À / À mice for the stranger mouse (P ¼ 0.0213) (Fig. 8c), indicating their social interaction deficits. With the introduction of a second stranger in the previously empty chamber, WT mice again demonstrated an increased preference for the novel stranger (S2), this time compared with the familiar mouse (S1) (P ¼ 0.0001, MWU-test) (Fig. 8d), whereas L7-Shank2 À / À mice showed no preference for either the familiar or the novel stranger mice (P ¼ 0.1, MWU-test) (Fig. 8e). Comparing the preference index (S2 À S1) between WT and L7-Shank2 À / À confirmed the impairment of social interaction in L7-Shank2 À / À mice (P ¼ 0.0136) (Fig. 8f). Because of previously reported compulsive grooming 25 and jumping 26 in global Shank2 À / À mice, we next examined repetitive behaviour. We observed no significant differences between WT and L7-Shank2 À / À mice in the percentage of buried marbles in the marble-burying task (P ¼ 1.0) (Fig. 8g) or in the duration of grooming over a 15 min period (P ¼ 0.054) (Fig. 8h). However, the L7-Shank2 À / À mice did show an increase of repetitive behaviour in the T-maze through an increased perseveration (P ¼ 0.0023, MWU-test) (Fig. 8i). Finally, we observed no significant difference in anxiety (P ¼ 0.7, w 2 -test) or olfactory sensitivity (P ¼ 0.6) of L7-Shank2 À / À mice that could have potentially biased the social behaviour assessments ( Supplementary Fig. 7). Together, these results indicate that L7-Shank2 À / À mice exhibit ASD-like social impairments and task-specific repetitive behaviour.

Discussion
Severe loss-of-function mutations in SHANK2 have been firmly established as conferring a high genetic risk for ASD and intellectual disability 20,22,29 . ASD patients with disruptive SHANK2 mutations exhibit motor impairments, language delay and cerebellar dysfunction including dysmetria  ARTICLE and dysdiadochokinesis 29 . Considering the increasing evidence for cerebellar involvement in ASD 40 , we investigated anatomical, molecular and physiological consequences of global Shank2 ablation in the cerebellum. In addition, we analysed L7-Shank2 À / À mice with cerebellar PC-specific deletion of Shank2 to evaluate the extent to which the ASD-related behavioural findings in global Shank2 À / À mice can be attributed to cerebellar dysfunction.
In recent years, several genetic mouse models for ASD have been used for the study of cerebellar abnormalities. The first study to implement a PC-specific model related to ASD involved the deletion of Fmr1, the gene encoding the fragile X mental retardation 1 protein 41 . In this study, the authors reported eyeblink abnormalities and increased LTD in both global and PCspecific deletion of Fmr1. Furthermore, global Nlgn3-KO mutants exhibited deficits in cerebellum-related motor performance as assessed by the ErasmusLadder 42 . A more recent model examining the 15q11-13 duplication ASD syndrome demonstrated impaired cerebellar synaptic plasticity and motor learning deficits as assessed by eye-blink conditioning 43 . Perhaps the most definitive study implicating cerebellar dysfunction as aetiologic for ASD-like behaviour involved a PC-specific deletion of Tsc1 (ref. 44). This study was the first to demonstrate that PC-specific deletion of an ASD-associated gene results in ASDlike behaviour. Finally, a very recent study using multiple mouse models of syndromic ASD found a consistent pattern of impaired sensorimotor integration 45 . These studies have established the foundation by which the cerebellar synaptic pathophysiology underlying ASD can be mechanistically investigated 18 .
Given previous studies reported on the importance of SHANK2 in the regulation of neuronal plasticity 25,26 , we investigated both synaptic and intrinsic plasticity of cerebellar PCs (Fig. 4). Our results indicate that SHANK2 is crucial for PF-PC LTP, but not LTD. In addition, we show that SHANK2 is important for intrinsic plasticity of neuronal excitability 46 . In contrast to a recent study of the 15q11-13 duplication ASD syndrome in which PC synaptic plasticity deficits were limited to LTD 43 , our results now highlight LTP impairments as a candidate mechanism underlying the cerebellar pathophysiology of ASD. SHANK2 is a dedicated scaffolding protein, which has a major role in the regulation of glutamate receptor integration, synaptic transmission and plasticity 17,47 . Future molecular and functional studies will have to elucidate the exact mechanisms by which SHANK2 mediates plasticity in the PC, but it may well include a suboptimal integration of GluR subunits as the expression levels of GluA1 and GluA2 were both reduced in the Shank2 À / À mice. Since GluR subunit levels were analysed in synaptosomes from whole adult cerebella, it remains to be investigated to what extent these reductions result from specific changes at the PF to PC synapse and/or other cerebellar cortical synapses, and at what developmental stage they start to occur.
Because of the cerebellar physiological impairments and the previously reported motor hyperactivity in Shank2 À / À mice 25,26 , we examined activity levels during both baseline exploration and motor learning. To our surprise, we did not find motor performance abnormalities in the PC-specific L7-Shank2 À / À mice during various assessments including five separate locomotion and eye movement tests ( Supplementary Fig. 2).   (a,b) After 3 days of training, WT (n ¼ 9) but not L7-Shank2 À / À mice (n ¼ 6) (P ¼ 0.018) learned to increase their speed during a conditioned ErasmusLadder test using tone-cued rung displacements. (c,d) In vestibulo-ocular reflex (VOR) gain increase training, L7-Shank2 À / À mice (n ¼ 8) were not able to adapt their gain like WTs did (n ¼ 7) (P ¼ 0.006).
However, we did observe substantial impairments of cerebellar motor learning including conditioning of locomotion and eyeblink responses, as well as adaptation of compensatory eye movements (Fig. 6). It might seem counterintuitive that baseline motor performance can be intact while the capacity for motor learning is reduced, but this combination has been observed in many different mutant lines over the last few decades 48 . Most likely, it reflects the indispensable role of PC plasticity for the acquisition of new behaviours within relatively short periods of time as occurs during the experimental training paradigms (that is, in the order of hours), and the ability of the motor performance control system to compensate upstream and/or downstream of the affected synapse when prolonged adaptation periods are available as occurs during postnatal development (that is, in the order of weeks) 49 . The potential causality of the identified electrophysiological abnormalities as underlying the motor learning impairments is strengthened by our independent findings in another PC-specific mouse mutant (L7-PP2B À / À ), in which also both synaptic LTP and intrinsic plasticity were affected 32 . Together, these phenotypes point towards a PC-dependent contribution to the behavioural motor impairments frequently observed in ASD.
In addition to changes in plasticity, we also found that inhibition of PCs was enhanced in the global Shank2 À / À mutants. Since reduced inhibition of PCs increases regularity of SS activity 34 , we hypothesized that PCs of the global Shank2 À / À mutants should have a higher level of irregularity of SS firing (Fig. 3). Indeed, this hypothesis was not only consistent with the in vivo extracellular recordings in vermal lobule X of the global Shank2 À / À , but also confirmed in three different areas (Lobules I-V, Lobules IX-X and the flocculus) of the L7-Shank2 À / À mice (Fig. 7). Moreover, this correlation was also in line with the fact that the differences in sIPSCs, CV and CV2 had bigger effect sizes in the posterior lobules than the anterior lobules.
The increased frequency of inhibition in the global Shank2 À / À did not occur concomitantly with increased amplitude of sIPSCs postsynaptically at PCs, indicating that the observed effect could be of pre-synaptic origin. The PC irregularity in the L7-Shank2 À / À would then have to originate from a pre-synaptic effect of the postsynaptic absence of SHANK2. Indeed, recent evidence indicates the possibility of Shank-mediated transsynaptic signalling through transmembrane proteins affecting both post and pre-synaptic processes important for vesicle release probability 50 . This type of transsynaptic signalling could manipulate the inhibitory input to PCs either directly or indirectly, for example, through altered glutamate spillover from the CF to PC synapse 51 . Future research aimed at pinpointing the sites relevant to the effects described above should thus focus not only on PC-specific mouse mutants, but also on those in which their afferents are specifically affected 52,53 . In doing so, important consideration in studies implementing cell-specific deletions should be given to germ-line analyses, given the sensitivity of the L7-cre 44,54 and Shank2 lines (Supplementary  Table 1) 55,56 .
We observed a significant decrease in SS frequency in the anterior lobules of global Shank2 À / À mice, but not in their posterior lobules, nor in the anterior or posterior lobules of L7-Shank2 À / À mice. We believe that this inconsistency may reflect the fact that the spontaneous SS firing frequency of PCs is probably largely due to their intrinsic properties rather than the synaptic efficacy of their inhibitory or excitatory inputs 33 . Indeed,  ARTICLE blocking inhibitory or excitatory synaptic input to PCs by deleting their GABA-A-gamma2 receptor-subunits or abolishing voltage gated calcium channels at their PF input primarily affects the regularity of SS firing, rather than their firing frequency 34,53 . Thus, the consistent irregularity of SS in PCs, particularly in the posterior lobe, of the Shank2 À / À mutants underlines the putative importance of precise SS regularity for behavioural output 48 . Although abnormalities in the anterior and posterior lobules have both been proposed as relevant sites of cerebellar pathology in ASD 57 , our converging data obtained in the posterior lobe suggest that the mechanisms governing the regularity of SS firing reveal a common biological vulnerability in the aetiology of ASD.
Here we report impaired social and task-specific repetitive behaviour due to the PC-specific deletion of Shank2 (Fig. 8). This result is particularly interesting as, to our knowledge, it is the first PC-specific mouse model for a non-syndromic form of autism in which ASD-like behaviour has been established. The impaired social behaviour, late-onset ataxia and reduced excitability of PCs previously observed in L7-Tsc1 mice 44 were due to the absence of a protein that inhibits mTOR signalling through which the translation of a wide variety of proteins is regulated. In contrast, here we show that disruption of the synapse through the absence of a single postsynaptic scaffolding protein in the PC is sufficient to show impaired ASD-related motor learning and social behavioural impairments. In addition to the social impairments, we found signs of enhanced repetitive behaviour in the T-maze paradigm, but not the marble-burying task or grooming tasks. Since the T-maze task reveals the level of cognitive inflexibility following decision-making over consecutive trials rather than the level of repetitious behaviour dominated by high-frequency motor activity that characterizes the other two tasks and that may well be confounded by deficits in cerebellar motor learning, these results highlight the importance of the PC synaptic function for ASD beyond the classically ascribed motor-related behaviour.
the Shank family of postsynaptic scaffolding proteins has many different interacting proteins in the PSD through which they could contribute to the functional establishment of regulatory mechanisms for plasticity. The translational challenge from synapse to behaviour brings about two main questions: How does an impaired PC mediate ASD-related behaviour? And how might PC impairments lead to abnormal brain function beyond the cerebellum with regard to neurodevelopmental critical periods? The first question has been extensively addressed by the accumulating evidence regarding the contribution of ASD-related cerebellar dysfunction to impaired motor learning, as apparent from the eye movement adaptation, ErasmusLadder and eye-blink conditioning findings examined here and by other investigators 45 . It is indeed possible that the increased inhibition and irregularity of SS firing, in addition to impaired cerebellar plasticity mechanisms, may contribute to social and repetitive behaviour-related phenotypes in ASD. We believe that the answer to how the cerebellum can essentially contribute to socially impaired behaviour could reside in various mechanisms. The idea that disruption of a certain brain area during development could affect the development and consequently the function of other inter-connected areas, also termed developmental diaschisis, has recently been put forward as a prime mechanism for the cerebellum in its ability to influence other cortical areas in critical developmental periods 58 . In the future, the latter hypothesis can, for example, be tested with PC-specific Shank2 ablation at different stages during development using inducible mouse models, as has recently been employed for other ASD-related genes 24,59 . These experiments will help to further elucidate the mechanisms by which differential genes, such as Shank2, regulate cerebellar function and ultimately ASD-like behaviour.

Methods
Experiments and analyses were performed with the experimenters blinded to the genotype. Mice used were global germ-line Shank2 À / À and their littermate WTs all bred on a mixed C57BL6/N and C57BL6/J background. The generation of these mice has previously been described in detail 25,26 . The L7-Shank2 À / À was generated by crossing Purkinje specific L7(Pcp2)-Cre 54 with Shank2 flxd/flxd (ref. 25). Genotyping was performed on postnatal day (P)7-10 using primers 1700 S (5 0 -TCCATGGTT TCGCGAGAGCG-3 0 ), 1842 AS (5 0 -TCCCTATT GGGACGCAGTGG-3 0 ) and 2394 AS (5 0 -CAGCATCATGACAATGTCTCCA-3 0 ). For all experiments, we used mice from both genders, unless indicated otherwise. The mice were individually housed with food and water available ad libitum and in 12:12 h light/dark cycles. All experiments were approved by local and national ethical committees.
Primary antibodies. The anti-SHANK2 SA5193 antibody has been characterized previously 25  Golgi stainings. Adult mouse cerebella were dissected and prepared using the FD Rapid GolgiStain Kit (NeuroTechnologies, Vilnius, Lithuania). Serial coronal sections of 150 mm were collected from WTs and global Shank2 À / À mice and Z-stack images were taken using an upright Axioscope (Carl Zeiss, Jena, Germany). Distal dendrites of PCs were traced for spine analysis.
Electron microscopy. Adult mice were transcardially perfused with fixative (2% paraformaldehyde, 2.5% glutaraldehyde, 1% saccharose in 0.1 M cacodylate buffer, pH 7.3) and their cerebella were dissected and post-fixed overnight at 4°C. After dehydration and staining with 2% uranyl acetate, the material was embedded in epoxy resin. Ultrathin sections were cut using an ultramicrotome (Ultracut UCT, Leica). After lead citrate staining, sections from WTs and global Shank2 À / À mice were examined using an electron microscope (JEM 1400 TEM, Jeol). For ultrastructural PSD analysis, spine synapses have been selected in the distal molecular layer where the PF-PC contacts greatly outnumber other types of synapses.
Biochemistry. Adult mouse cerebella were homogenized on ice in HEPESbuffered sucrose (320 mM sucrose, 5 mM HEPES, pH 7.4) containing protease inhibitor mixture (Roche, Mannheim, Germany). The homogenate was centrifuged (1,000g, 4°C) to remove cell debris and nuclei. The supernatant was further centrifuged (12,000g, 4°C) to obtain a pellet containing the cerebellar synaptosomes. Equal amounts of 10-20 mg protein per lane were separated by SDS-polyacrylamide gel electrophoresis and blotted onto polyvinylidene fluoride membranes using standard protocols. After incubation with primary antibodies (1:1,000 for anti-Shank1, anti-GluA1, anti-GluA2, anti-Nlgn3; 1:10,000 for anti-b3-Tubulin), immunoreactivity was visualized on X-ray film (GE Healthcare, Freiburg, Germany) using HRP-conjugated secondary antibodies (Dako; Hamburg, Germany) and the SuperSignal detection system (Thermo Scientific). For quantification, the films were scanned, the grey value of each band was analysed by ImageJ (the National Institutes of Health, Bethesda, MD, USA) and normalized to the grey value of b3-Tubulin.
Immunohistochemistry. Mouse brains were snap-frozen after removal without perfusion. Tissue was sectioned at 7 mm using a cryostat at À 20°C and a knife temperature of À 14°C. Sections were air-dried on superfrost glass and stored at À 80°C. For staining, sections were defrosted at room temperature (RT) for 60 min and subsequently washed with À 20°C MeOH for 3 min followed by 3 Â 10 min of PBS wash. To permeabilize membranes, sections were incubated for 60 min in 0.5% Triton X-100 in PBS at RT and washed 3 Â in PBS for 10 min. . This was concluded by a 2 Â 10 min PB wash. Following PBS washing, the sections were thionin-stained and permount-covered using standard protocols. Images were taken using an upright confocal microscope (LSM 700, Zeiss, Oberkochen, Germany) using 405 nm and 555 nm wavelengths. All immunohistological stainings have successfully been replicated on multiple occasions.
Ex vivo electrophysiology. Following decapitation of mice under isoflurane anaesthesia, the cerebellum was removed into an ice-cold 'slicing medium', containing (in mM) 240 sucrose, 2.5 KCL, 1.25 Na 2 HPO 4 , 2 MgSO 4 , 1 CaCl 2 , 26 NaHCO 3 and 10 D-Glucose that was carbogenated continuously (95% O 2 and 5% CO 2 ). Sagittal slices, 250 mm thick, of the cerebellar vermis were cut using a vibrotome (VT1200S, Leica) and put in carbogenated artificial cerebrospinal fluid (ACSF) containing (in mM): 124 NaCl, 5 KCL, 1.25 Na 2 HPO 4 , 2 MgSO 4 , 2 CaCl 2 , 26 NaHCO 3 and 20 D-Glucose, for at least 1 h at 34±1°C before the start of the experiment. Slice physiology was done at RT 21±1°C or 33±1°C as indicated in the Results section and in the presence of 100 mM picrotoxin except for the sIPSCs recordings. Whole-cell patch clamp recording were performed with an EPC9 amplifier (HEKA Electronics, Lambrecht, Germany). Recordings were excluded if the series (Rs) or input resistances (Ri) changed by 415% during the experiment, which was determined using a hyperpolarizing voltage step relative to the À 65 mV holding potential. Data analysis (rise times (10-90% for EPSC and APs)), decay time (tau) for EPSC and IPSC amplitudes, AP threshold (identified by steepest slope in membrane potential prior to AP) and AHP amplitude (minimal membrane potential relative to the AP threshold) were performed using Clampfit software (Molecular Devices). For whole-cell recordings PCs were visualized using an upright microscope (Axioskop 2 FS, Carl Zeiss) equipped with a Â 40 objective. Recording electrodes (3-5 MO, 1.65 mm outside diameter (OD) and 1.11 mm interior diameter (ID), World Precision Instruments, Sarasota, FL, USA) were filled with an intracellular solution containing (mM): 120 K-Gluconate, 9 KCL, 10 KOH, 4 NaCL, 10 HEPES, 28.5 Sucrose, 4 Na 2 ATP, 0.4 Na 3 GTP (pH 7.25-7.35 with an osmolarity of 295±5). Note that we adjusted the osmolarity using sucrose 46,60 . For the recording of sIPSCs, we used an intracellular solution containing (mM): 150 CsCl, 1.5 MgCl 2 , 0.5 EGTA, 4 Na 2 ATP, 0.4 Na 3 GTP, 10 HEPES, 5 QX314 (pH 7.25-7.35 with an osmolarity of 295 ± 5). For extracellular stimulation of PFs, similar patch electrodes were filled with ACSF and positioned in the upper third of the molecular layer lateral to the patched PC. The stimulation intensity was set to evoke an EPSC of 300 ± 50 pA (typically 3-6 mA stimulation intensity). For PF-PC transmission, we used various inter-stimulus intervals (50-200 ms) (see Fig. 2). For recordings of spontaneously occurring IPSCs (sIPSCs), we used the previously mentioned K þ -based internal and recorded their occurrence during at least 120 s.
For CF stimulation, similar electrodes (filled with ACSF) were positioned near the patched PC soma in the surrounding granule layer. We selected those recordings in which CF stimuli elicited clear all-or-none responses and lacked the co-activation of PC axons (identifiable by backpropagating APs) for further analysis. For CF elimination, experimental tissue was prepared in a similar way for all age groups. We systematically scanned the granule cell layer to elicit CF responses and recorded PC responses (using an intracellular solution containing (in mM): 115 CsMeO 3 , 20 CsCl, 2.5 MgCl 2 10 HEPES, 0.6 EGTA, 4 Na 2 ATP, 0.4 Na 3 GTP, 10 Na-phosphocreatine) at À 20 mV holding potential to prevent voltage escape during the CF responses. For CF-PC transmission, we evaluated the pairedpulse ratio at 50 ms stimulus interval. To evaluate the complex spike waveforms, we analysed the amplitude of the Na þ -spike and the amplitude of the first Ca 2 þ -spikelet evoked during the LTD-tetanus 61 . Current clamp recordings were corrected offline for the calculated liquid junction potential ( À 10.2 mV).
The synaptic (LTP, LTD) and intrinsic plasticity protocols were recorded from lobules 5/6 and conducted as described previously 32,46 . In short, for synaptic plasticity all recordings were done in voltage-clamp, except for the tetanus, which consisted of single-pulsed PF-stimulation at 1 Hz for 5 min (LTP) or single-pulsed PF þ single-pulsed CF stimulation (5 ms interval) at 1 Hz for 5 min (LTD). We evaluated the synaptic plasticity by the change in PF-EPSC (baseline at 0.05 Hz) relative to the mean value calculated during the last 5 min pre-tetanus. For intrinsic plasticity, we utilized the PF-LTP tetanus (but without bias currents, that is, I ¼ 0 pA) and evaluated the impact on the number of APs evoked by 300 pA current injections during 500 ms (presented at 0.05 Hz).
Extracellular PC recordings. In vivo recordings were performed as recently described 33 . An immobilizing pedestal was fixed on the skull and a craniotomy (Ø 3 mm) was performed on the occipital bone. After recovery of 5 days, mice were head-fixed and body restrained for recordings. Single-unit recording was identified by the presence of a short SS pause (46 ms) after each complex spike. PCs were recorded from vermal lobules I-V and X using single barrel (2.0 mm OD, 1.16 mm ID, Harvard Apparatus, MA, USA) and double barrel (theta septum, 1.5 OD, 1.02 ID; World Precision Instruments) boroscilate glass pipettes. In vivo recordings were analysed offline using Spiketrain (Neurasmus BV, Rotterdam, The Netherlands, www.neurasmus.com) and custom scripts in MatLab (Mathworks, Natick, MA, USA). The CV is calculated by dividing the s.d. by the mean of the interspike intervals, whereas CV2 is calculated as 2 Â |ISI n þ 1 À ISI n |/ (ISI n þ 1 þ ISI n ).
Compensatory eye movements. Mice between 8 and 10 weeks of age were prepared for head-restrained recordings of compensatory eye movements. These types of recordings have been described in detail previously 32 . To head restrain the mice during the eye movement task, a small pedestal was attached using Optibond primer and adhesive (Kerr, Bioggio, Switzerland) under isoflurane anaesthesia in O 2 (induction with 4% and maintained at 1.5% concentration). After a recovery period of 2-3 days, mice were head-restrained by fixation using the pedestal in the experimental set-up. A round screen with a random dotted pattern ('drum') surrounded the mouse during the experiment. The OKR, VOR and the lightguided VOR (VVOR) were induced using a sinusoidal rotation of the drum in light (OKR), rotation of the table in the dark (VOR) or the rotation of the table (VVOR) in the light. The motor behaviour was assessed by rotating the table and/or drum at 0.1-1 Hz with a fixed 5°amplitude. To evaluate motor learning, a mismatch between visual and vestibular input was created. Rotating both the visual and vestibular stimuli in phase (at the same amplitude) induced a decrease of gain; rotating the drum at greater amplitude relative to the table induced the so-called phase reversal of the VOR (day 1, 5°phase difference; day 2, 7.5°; day 3-4, 10°). Rotating the visual and vestibular stimuli out of phase (at the same amplitude) induced the VOR gain increase. All training protocols were induced at 0.6 Hz with table rotation amplitude of 5°. For eye illumination during the experiments, two table-fixed infrared emitters (output 600 mW, dispersion angle 7°, peak wavelength 880 nm) and a third emitter, which produced the tracked corneal reflection, were mounted to the camera and aligned horizontally with the optical axis of the camera. Eye movements were recorded with eye-tracking software (ETL-200, ISCAN systems, Burlington, NA, USA). Gain and phase values of eye movements were calculated using Matlab.
Eye-blink conditioning. Mice of 12-15 weeks of age were prepared for headrestrained eye-blink conditioning 39 . In short, a small brass pedestal was attached to the skull using Optibond primer and adhesive and Charisma (Heraeus Kulzer, Armonk, NY, USA), under isoflurane anaesthesia. Three to five days after surgery, mice were habituated during two short (30-45 min) sessions on 2 days in a sound-and light-isolating chamber which houses the eye-blink set-up. During these sessions, mice were head-fixed and suspended over a foam cylindrical treadmill. While no stimuli were presented, a 27.5 gauge needle through which unconditional stimulus (US) air-puffs are delivered was positioned at 5 mm from the centre of the left cornea, a green LED (ø 5 mm) that delivers the CS was placed 5 cm in front of the mouse and an GMR magnetometer (NVE, Eden Prairie, MN, USA) was fixed above the left eye. During 5 subsequent acquisition training days, this sensor measured the distance of a miniscule magnet (1.5 Â 0.7 Â 0.5 mm) that was placed on the left lower eyelid with high accuracy, while 200 paired trials were presented, usually spaced 10±2 s apart, plus the time needed for a sporadic unstable eyelid to stabilize in open position. Each paired trial consisted of a 280 ms green LED CS, co-terminating with a 30 ms air puff (30 psi, through an MPPI-3 pressure injector; ASI, Eugene, OR, USA). All experiments were performed at approximately the same time of day by the same experimenter. Individual eye-blink traces were analysed using custom LabVIEW (National Instruments, Austin, TX, USA) or Matlab scripts. Trials with significant activity in a 500 ms pre-CS baseline period were regarded as invalid for further analysis. Valid trials were aligned by making the mean of their baseline activity zero, and the average amplitude of all post-US unconditioned blink responses was used to denote 100% eyelid closure. From this, the average eyelid closure as a percentage from baseline to full closure at the end of the CS-US interval was calculated over all valid trials. To calculate percentage of CRs, trials were judged to contain a CR if the eye closed for 45% between 50 and 250 ms after CS and the CR reached its peak after 100 ms.
ErasmusLadder. The ErasmusLadder (Noldus, Wageningen, Netherlands) is a fully automated system consisting of a horizontal ladder between two shelter boxes. It has 37 rungs on each side, spaced 15 mm apart, and attached to custom-made pressure sensors that are continuously monitored. To create a left-right alternating pattern, even rungs on one side and odd rungs on the other side are elevated by 6 mm. Prototype testing revealed that optimum forepaw displacement for mice is about 6 cm in a single step at medium high velocity, which is the distance between three consecutive elevated rungs and is defined as an efficient step in the text (for more details, see ref. 37). It is clear from previous studies that mice improve their walking efficiency over training sessions by increasing the number of efficient steps relative to steps of lower sizes 37 . In the current study, mice (male, 15-18 weeks old) were tested in six daily sessions consisting of two unperturbed sessions, one session with a fixed obstacle in the middle of the ladder and three perturbed sessions. During the first three sessions, mice were trained to walk between 2 shelter boxes for 50 trials each day. In the perturbed sessions, a sudden appearance of a rising rung on the right side of the mouse was used as the US. A 15 kHz tone was used as the CS and preceded the US by 200 ms in CS-US paired trials. CS-only trials and paired trials were randomly distributed among 50 trials. There were twice as many paired trials as CS-only trials.
Step length and step time were defined as the distance and time between two consecutive touches from the right front limb. To estimate motor adaptation in CS-only trials, we calculated step speed (step length/step time) using only the steps within 1 s before and 1 s after the CS. The speed ratio during conditioning was defined as the speed post-CS divided by the speed pre-CS.
General behavioural analyses. Behavioural experiments were performed using L7-Shank2 À / À and WT littermate controls aged 8-16 weeks during the light period of their diurnal cycle. The mice used in the general behavioural experiments described here underwent multiple tests.
For the PhenoTyper test, mice were placed in a homecage-like apparatus (Noldus) with ad libitum access to food and water, and left to explore for 30 min. Locomotion was recorded using automated software (Noldus Ethovision XT 11) and distance and speed were calculated.
For the open field test, mice were placed in a novel circular, brightly-lit 110-cm-diameter open arena for 10 min. Locomotion was recorded using automated software (Noldus Ethovision XT 11) and total distance travelled, as well as average speed were calculated. During analysis the arena was subdivided into three concentric zones named the inner (25 cm), middle (15 cm) and outer zone (15 cm), and percentage of time in each zone was calculated.
For the three-chamber social interaction test, age-and gender-matched WT target subjects (Stranger 1 and 2) were habituated for 5 consecutive days before beginning of testing by being placed inside round metal-wired cages. On the test day, experimental mice were placed in the central chamber of a clear Plexiglas box (60 Â 35 cm) divided into three inter-connected chambers. After habituation for 5 min, an unfamiliar mouse (Stranger 1; S1) was introduced into a wire cage in one of the side-chambers and an empty wire cage in the other side-chamber. The dividers were then raised and the test mouse was allowed to freely explore all three chambers over a 5-min session. Next, the mouse remained in the chamber with stranger 1 for an additional 5-min session. Subsequently, a novel stranger mouse (Stranger 2 ) was placed in the previously empty wire cage and again the test mouse was left to explore for 5 min. Time spent in each chamber, as well as overall locomotion, was calculated using automated software (Noldus Ethovision XT 11). Preference indices were calculated by subtracting the time spent with the empty wire cage from the time spent with stranger 1 (S1 À E), and subtracting the time spent with stranger 1 from time spent with stranger 2 (S2-S1).
For the T-maze spontaneous alternation test, mice were placed at the base of a T-maze (arm length 50 cm) and were given the choice to freely explore either the right or left arm of the maze for 10 consecutive trials. A choice was assumed to be made when mice stepped into an arm with all four paws. At that moment, the gate to that arm was closed and the animal was allowed to explore the arm for 5 s.
Then, the mouse was gently placed back at the base of the T-maze for the next trial. When the mouse chose a similar arm at two consecutive trials, this was scored as number of repeats, indicative of repetitive behaviour.
For the grooming test, mice were removed from their homecage, received a single puff of water spray and placed in clean transparent cages (15 Â 15 Â 20 cm) under bright light for 15 min. Behaviour was recorded with a high-speed camera (30 Hz full frame rate), and time spent grooming was scored by two independent raters. All types of grooming-paw licking, nose and head self-grooming, body grooming, leg grooming and tail/genital grooming-were scored.
For the marble-burying test, Makrolon cages (50 Â 26 Â 18 cm) were filled with 4 cm of bedding material and 20 glass marbles, which were arranged in an equidistant 4 Â 5 grid. Animals were given access to the marbles for 30 min. Marbles 100% covered by bedding were scored as buried and marbles covered partially contributed 50% to the total score.
For the olfactory test, during the habituation phase a piece (1.25 g) of cookie was put into the subject's cage each day for 3 consecutive days, and checked for complete consumption the following day. For 24 h before the test phase, mice were completely food-deprived. Subjects were placed in a clean Makrolon cage (50 Â 26 Â 18 cm) with 4 cm of clean bedding, and allowed to habituate for 5 min. Then, a piece of cookie (1.25 g) was hidden in a random corner in the cage at 1 cm depth. Latency to find the cookie was recorded for a maximum of 15 mins (900 s).
Data analysis. In the text, mean ± s.d. values are presented, in the figures s.e.m. values are reported, and P values o0.05 are considered significantly different. Two-sided Student's t-tests were performed, unless stated otherwise. For a complete representation of the data, we have included a detailed overview of all statistics in Supplementary Table 1. In addition, since the L7-Cre line was shown to reveal germ-line deletions (see also ref. 44) and since the potential impact of cerebellar PCs on general cognitive tests is widely debated 62,63 , we have checked for heterozygous germ-line deletions in our L7-Shank2 mice and statistically excluded the possibility that they could have influenced our conclusions on the non-cerebellar paradigms (Supplementary Table 1).
Data availability. In addition to our Supplementary Data File, which contains data points that support the findings of this study, data sets are available from the corresponding authors upon request.