BACE Inhibitor Treatment in Mice Induces Hyperactivity in a Seizure-Related Gene 6 Family Protein Dependent Manner but Does Not alter Spatial Learning and Reference Memory

BACE inhibitors, which decrease BACE1 (β-secretase 1) cleavage of the amyloid precursor protein, are a potential treatment for Alzheimer’s disease. Clinical trials using BACE inhibitors have reported a lack of positive effect on patient symptoms and, in some cases, have led to increased adverse events, cognitive worsening and hippocampal atrophy. A potential drawback of this strategy is the effect of BACE inhibition on other BACE1 substrates such as Seizure-related gene 6 (Sez6) family proteins which are known to have a role in neuronal function. BACE1 substrates are implicated in this effect and, potentially, in the cognitive decline in longer-term chronically treated patients. Protein concentrations were quantied using the DC Protein Assay kit (Bio-Rad, USA) and 40 µg of total protein were used for Western Blot analysis. Samples were boiled for 5 min at 95°C in Laemmli buffer. For the detection of Sez6L, Laemmli buffer without the disulde bridge reducing agent β-mercaptoethanol was used and samples were not boiled (13). Samples were separated on 7.5% Mini-PROTEAN TGX Stain-Free gels (Bio-Rad). Nitrocellulose or PVDF membranes (Bio-Rad) were incubated with primary antibody overnight at 4°C. After incubation with secondary antibody at room temperature for 1 hour, membranes were developed with Clarity ECL (Bio-Rad). The following antibodies were used: monoclonal Sez6 (1:100; see ref. (13)), monoclonal Sez6L (1:100; ref. (13)), polyclonal Sez6L2 (1:500; R&D Systems, AF4916), β actin (1: 1000; Sigma, A5441) and monoclonal APP (Y188; 1:6000; Abcam, ab32136), HRP coupled anti-rat, anti-rabbit, anti-goat and anti-mouse (Life Technologies). Western blot protein levels were normalized to total blot protein (23).


Introduction
BACE inhibitors, which decrease BACE1 (β-secretase) cleavage of the amyloid precursor protein (APP) and subsequently reduce neurotoxic amyloid-β (Aβ) levels, have been in clinical trials for the treatment of Alzheimer's disease. BACE inhibitor (BACEi) treatment has shown no positive effect on cognition in patients (1)(2)(3)(4) and has led to increased cognitive decline (2,3,5), more hippocampal atrophy (1,2,4,6) and increased adverse events including sleep disturbance, anxiety and weight loss (1,2,7). The causes of the negative outcomes associated with BACE inhibition have not been determined but may relate to altered proteolytic cleavage of the numerous non-APP BACE1 substrates in the brain (8-10), echoing issues seen with the use of γ-secretase inhibitors (11). The Seizure-related gene 6 (Sez6) family of proteins: Sez6, Sez6-like (Sez6L) and Sez6-like 2 (Sez6L2), are BACE1 substrates (10,12,13) with roles in neuronal development and function (14)(15)(16)(17)(18). Altered proteolytic cleavage of Sez6 contributes to the reversible alterations in synaptic plasticity seen in mice treated with BACEi (19,20). In this study, we have examined the effect of chronic BACEi treatment on mouse behaviour and dendritic spine properties and determined whether the observed effects are related to altered Sez6 family proteolytic cleavage by comparing wild-type (WT) and Sez6 family triple knockout (TKO) mice.

Mice and drug administration
Male mice with targeted deletions of the genes encoding Sez6, Sez6L and Sez6L2 (Sez6 TKO; (14)) and WT mice were maintained on a predominantly C57BL6 genetic background. Mice were housed under standard conditions and experiments were performed in the daytime. All animal experiments were approved by the Animal Ethics Committee at the University of Melbourne. BACE inhibitor Compound H (N-  (21)) was provided by Janssen Pharmaceutica (Beerse, Belgium). BACEi was dissolved in an emulsion of 30% (v/v) peanut oil in MilliQ water. This mixture was combined with irradiated powdered diet (SF14-156; Specialty Feeds, Australia) and made into a dough ball. The amount of BACEi equated to approximately 15mg/kg body weight (2g ball) or 30mg/kg body weight (4g ball). Mice in the 15 mg/kg/day BACEi group preferentially ate the dough ball and had access to regular chow. Mice in the 30mg/kg/day group had no or very limited access to regular chow (depending on the individual mouse weight) and all mice included in the study consumed the entirety or the majority of the dough ball each day.

Experimental timeline
Mice were singly housed from 8 weeks of age. BACEi Compound H was delivered in diet between 8 and 16 weeks of age. Behavioural testing began after 4 weeks of BACEi treatment and was conducted during the remaining 4 weeks of BACEi administration. Mice had at least one day of rest in between behavioural tests and 30mg/kg/day cohorts underwent testing in the following order: elevated open eld, context fear conditioning and extinction, light/dark box and Morris water maze. On the nal day of BACEi treatment, mice were killed with pentobarbitone sodium (Lethabarb, Virbac Australia) and the brain was dissected out and divided into hemispheres, one of which was placed into Golgi-Cox solution for dendritic spine analysis and the other collected for protein analysis.

Behavioural testing Elevated open eld
Mice were placed in the centre of the eld, a test arena (75 × 100 cm) without walls positioned 60 cm above the ground, and allowed to roam freely for 3 minutes. Overhead lighting was switched off and two spotlights on either side of the eld shone directly onto the eld to create an aversive environment. Videos were obtained and analysed using TopScan Lite (CleverSys Inc., USA).

Context fear conditioning and extinction
The fear conditioning chamber was a 25 × 30 × 24 cm plexiglass chamber (CleverSys Inc.) containing a stainless steel shock grid oor. A light cue and visual cues on the chamber walls were used. During the training session, mice explored the chamber for 3 minutes before receiving 3 × 0.8 mA shocks of 2 seconds duration with 30 seconds between each shock. Mice remained in the chamber for an additional 60 seconds before being transferred back to their home cage. Freezing time was measured for the 3 minute pre-shock and 60 second post-shock periods. Mice were placed in the chamber 24 hours later and daily thereafter for a further 7 days. In these test and extinction sessions, mice were allowed to explore freely for 3 minutes without receiving a shock. Fear behaviour (percentage of time spent freezing) was recorded by FreezeScan software (CleverSys Inc.) and analysed by the investigator.

Light dark box
Automated locomotor systems (27.5 × 27.5 × 13.5 cm; Med Associates Inc., USA) were used with black Perspex inserts (13.5 × 27.5 cm) to create a dark area in one half of the locomotor activity cell. The light half had a brightness of 950 lux. Mice were placed in the dark side to start the test and locomotor activity over 10 min was recorded and analysed by Activity Monitor software (Med Associates Inc.).

Morris water maze
The maze was 1.2 meter in diameter and the 10 cm diameter 'escape' platform was submerged 1 cm under the surface of the water (20°C) which was made opaque with white, nontoxic paint. Overhead lighting was switched off and an uplight was placed on top of one of the 3D spatial cues. The arrangement of prominent geometric images, 3D spatial cues and lighting in the region around the pool is known to enable spatial learning in the maze (e.g. (16)). The investigator was hidden from sight during testing. Each mouse underwent 4 trials per day, starting at one of 4 locations equally spaced around the pool. Each trial lasted a maximum of 120 seconds and at the end of each trial the mouse was placed or allowed to stay on the platform for 20 seconds. Acquisition training went for 7 days, during which the location of the platform was kept constant. An automated tracking system (TopScan Lite, CleverySys Inc.) recorded and analysed swim paths. The day after acquisition training was completed, the probe trial took place. In the 30 second probe trial, the platform was removed and each mouse was placed into the pool in a location directly opposite from where the hidden platform had been located. Time spent in the 'target' quadrant was used to determine whether mice had learnt the task. After the probe trial, mice began 4 days of reversal training where the hidden platform was moved to a different quadrant. Reversal training trials were conducted as described for acquisition training and at the completion of training each mouse completed a 30 second probe trial.

Locomotor cells
Mice were acclimatised to the test room for 1 hour immediately prior to testing and were tested for locomotor activity under low lighting conditions (10 lux). Mice were placed in the centre of the locomotor cell (27.5 × 27.5 × 13.5 cm; Med Associates Inc.) and locomotor activity in the horizontal and vertical planes was recorded over a 30 minute period.

Golgi-Cox impregnation
Modi ed Golgi-Cox impregnation of neurons was performed using the FD RapidGolgi Stain kit (FD NeuroTechnologies Inc., USA). One brain hemisphere from each mouse was xed, sectioned at 100 µm and developed. Neurons were analysed for dendritic spine properties as described by Risher et al. (22), with the slight modi cation that spines with a head width greater than 0.5 µm (rather than 0.6 µm) were classed as mushroom spines, as previously published (16). Approximately 20 µm lengths of oblique secondary dendrite from neurons in the stratum radiatum of CA1 pyramidal neurons or 30-40 µm lengths of basal secondary dendrites from layer V pyramidal neurons in the somatosensory cortex were selected for analysis. Z-stack images were captured using bright eld on a Zeiss Axio Imager M2 (Zeiss, Germany) using Stereo Investigator (MBF Bioscience, USA). The investigator was blind to the genotype and treatment groups during both imaging and analysis.

Brain fractionation
Brain hemispheres from mice treated with 30mg/kg/day BACEi were collected into 1ml ice cold Dulbecco's Phosphate Buffered Saline (DPBS, Life Technologies) containing protein inhibitor tablets (cOmplete ULTRA Tablets, Mini, EDTA-free EASYpack Protease Inhibitor Cocktail, Sigma), brie y stored on ice, transferred to 1ml ice cold diethylamine (DEA) buffer (50mM NaCl, 2mM EDTA, 0.2% DEA) for homogenization then underwent fractionation as described in Kuhn et al. 2012 (12), outlined below. Brain hemispheres from mice treated with 15mg/kg/day BACEi were collected into 1mL ice cold MilliQ water, brie y stored on ice until all brains were collected, homogenized in MilliQ water and homogenates were stored at -20°C. Brain homogenates were thawed on ice and 500 µL of 3 × DEA buffer was added before samples underwent fractionation. Homogenates from all samples were neutralized with 1M Tris (pH 7.6) and centrifuged at 14000 RPM at 4°C for 10 minutes. The supernatant was collected and spun at 65000 RPM at 4°C for 30 minutes in an ultracentrifuge (Beckman Coulter Inc., USA); this supernatant was collected and stored at -20°C as the 'soluble fraction'. The pellet from the initial 14000 RPM spin was washed with DPBS, centrifuged at 14000 RPM then resuspended in 200µl STE buffer (150mM NaCl, 50mM Tris, 2mM EDTA, 1% Triton X-100) before being centrifuged again at 14000 RPM at 4°C for 10 minutes; the supernatant from this spin was collected and stored as the 'membrane fraction'.

Meso Scale Discovery detection of Aβ40
The concentration of Aβ40 in the membrane fraction of brains from BACEi and vehicle treated mice was determined using the V-PLEX Aβ40 Peptide Panel (Meso Scale Diagnostics, USA) as per manufacturer's instructions. Two technical replicates of each sample were run. Concentration in pg/mL was converted to pg/µg using the total protein concentration of the samples as determined by the DC Protein Assay kit (see above).

Cocaine treatment
Mice were administered either 20 mg/kg cocaine or saline by intraperitoneal injection immediately prior to undergoing testing on the elevated open eld and locomotor cell (as described above). Mice were tested rst on the elevated open eld then in the locomotor cell with at least 3 days break between tests.

Statistical analyses
Context fear conditioning and Morris water maze data were analysed using a two-way ANOVA with repeated measures and the elevated open eld, light/dark box and locomotor cell data were analysed using a 2-way ANOVA. Aβ40 concentration was analysed using a 2-way ANOVA and normalized Western blot protein levels were analysed using an unpaired t-test. As indicated by Box's test of equality of covariance matrices, Morris water maze pathlength and latency data were transformed with a log 10 transformation. If Mauchly's test of sphericity was determined to be signi cant, the Greenhouse-Geisser correction was used (applicable to Morris water maze and context fear conditioning data). Hippocampal and cortical spines were analysed using a 2-way nested ANOVA. The Bonferroni post-hoc test was used for all relevant analyses. All data are represented as mean ± standard error of the mean (SEM), except for the Morris water maze probe trials which are represented as mean ± 95% con dence interval (as per Rogers et al. (24)). Results were considered statistically signi cant when a p < 0.05 was obtained. Data were analysed in GraphPad Prism 7 (GraphPad Software Inc., USA), Minitab 18 or IBM SPSS Statistics 25 (IBM Corporation, USA).

Results
Chronic BACEi treatment decreases Aβ40 and Sez6 family protein ectodomain levels in the mouse brain and decreases mushroom spine density in the cortex.
Mice treated daily for 2 months in total with ~ 30mg Compound H per kg body weight (structure in Fig. 1A) had a substantial decrease in Aβ40 concentration in brain tissue compared to vehicle treated mice, equating to a 67 and 77% decrease in WT and Sez6 TKO brains respectively (Fig. 1B). This level of BACE inhibition markedly decreased the levels of soluble Sez6 family protein ectodomains in WT samples (Fig. 1C), particularly Sez6 and Sez6L (95% and 96% reduction respectively), previously shown to be prominent BACE1 substrates (12,13), and Sez6L2 (36% reduction). Chronic BACE inhibition did not result in changes to dendritic spine number or morphology in the hippocampus (Fig. 1D) however it decreased mushroom spine density in the somatosensory cortex in both WT and Sez6 TKO mice (Fig. 1E). There was a tendency for overall spine density to be reduced in BACEi vs. vehicle-treated WT brains but this was not signi cant at an α of 0.05 (Fig. 1E). There was no effect of BACEi treatment on spine length, width or length:width ratio in the hippocampus or cortex ( Supplementary Fig. 1A, B).
Although there was a higher level of Aβ40 in vehicle treated Sez6 TKO compared to WT samples ( Supplementary Fig. 1C), this was not apparent in a separate 15mg/kg/day cohort using a different sequential extraction protocol ( Supplementary Fig. 1D). There was no signi cant difference in full-length APP levels between genotypes in either cohort as determined by Western blot (Supplementary Fig. 1E, F).
BACEi treatment does not alter spatial memory or cognitive exibility in mice Mice underwent behavioural testing between weeks 4 to 8 of daily BACEi treatment. In the Morris water maze, mice treated with 30mg/kg/day Compound H learnt the platform location during the acquisition phase in the same manner as vehicle treated mice (Fig. 2Ai, ii) and during the probe trials (testing memory of the platform location) they spent the same amount of time in the target quadrant as vehicle treated controls (Fig. 2Aiii). Sez6 TKO mice displayed de cits in cognitive exibility, taking a longer time to locate the hidden platform during the reversal phase (Fig. 2Aii) as previously described (16). However, within genotypes, BACEi treatment did not signi cantly change the pathlength or latency to nd the platform in the reversal phase (Fig. 2Ai, ii); BACEi treated WT mice tended to take longer to nd the platform than vehicle treated WT mice on the rst day of reversal training but this was not signi cant at an α of 0.05 (Fig. 2Aii). During context fear conditioning, Sez6 TKO mice displayed enhanced freezing (Fig. 2B) in accordance with previous results (16). However, within genotypes, 30mg/kg/day BACEi treatment did not affect learning or extinction associated with context fear conditioning (Fig. 2B). BACEi and vehicle treated mice spent the same amount of time in the light zone in the light/dark box, a test of anxiety (Fig. 2Ci, ii) and there was additionally no effect of genotype in this test.
BACEi treatment increases hyperactivity on the elevated open eld in a Sez6 family protein dependent manner.
On the elevated open eld, a test of anxiety and exploration (25), WT mice treated with 30mg/kg/day Compound H moved signi cantly more than vehicle treated WT mice (Fig. 3A), representing an increase of ~ 24%. Vehicle treated Sez6 TKO mice moved less than vehicle treated WTs, in agreement with previous results (16). Sez6 TKO mice treated with 30mg/kg/day BACEi moved the same amount as vehicle treated Sez6 TKO mice (Fig. 3A). Similar results were seen with a lower dose of 15mg/kg/day BACEi: WT mice moved ~ 45% more when treated with BACEi while the movement of BACEi treated Sez6 TKO mice was unchanged by drug treatment (Fig. 3B). The signi cant genotype × drug interaction (p = 0.045) suggests that the hyperactivity caused by BACEi treatment is dependent on decreased BACE1 mediated ectodomain shedding of Sez6 family proteins. Sez6 TKO mice are known to exhibit motor de cits which affect their motor learning and co-ordination (16), so to support this interpretation it was important to determine whether Sez6 TKOs were capable of responding to a psychomotor stimulant by substantially increasing their distance travelled on the elevated open eld. In response to 20mg/kg cocaine treatment, both WT and Sez6 TKO mice showed a considerable increase in movement on the elevated open eld (110% and 254% respectively; Fig. 3C) and in locomotor cells (187% and 281%; Fig. 3D) indicating that the differential response of WT and Sez6 TKO mice to BACEi treatment is not solely due to de cits in motor capabilities.

Discussion
The clearest effect of BACEi treatment on mouse behavior observed in this study, hyperactivity in an aversive environment, is due to reduced cleavage of Sez6 family proteins, adding support to the concept that interference with certain functions of one or more Sez6 family members contributes to the negative outcomes seen with BACE inhibition in patients. These effects may be mediated through accumulation of Sez6 family protein transmembrane isoforms and/or by the reduction of shed Sez6 family ectodomains. While the signi cance of BACEi associated hyperactivity in mice is unclear, heightened locomotor activity has been correlated with increased hippocampal activity in APP transgenic mice (reviewed by (26, 27)) and associated with dendritic spine instability (28), suggesting the hyperactivity may re ect the early stages of hippocampal dysfunction. The behavioural effects of BACEi treatment in mice presented here are similar to those reported for a conditional knockout line with BACE1 deletion in postnatal forebrain excitatory neurons (29), namely hyperactivity on the open eld, indicated by greater distance travelled, and no memory impairment in context fear conditioning or in the probe trial of the Morris water maze. Whilst no memory impairment in context fear conditioning was also observed in an adult conditional knockout with partial BACE1 deletion (30), both this study and a previous study of BACE inhibition in mice (19) report no hyperactivity on the open eld. Results from the former are in contrast to our ndings from the 15mg/kg/day BACEi cohort, with both protocols causing an ~ 50% decrease in Aβ40 brain levels in adult mice ((30), Supplementary Fig. 1D). The unwalled, elevated open eld used in this study is a slightly more aversive environment than a standard open eld protocol which may contribute to these con icting ndings. Delayed learning in the Morris water maze was observed in older BACE1 conditional knockouts (29) but not in BACEi treated mice in this study, suggesting that learning de cits might become apparent after a longer BACEi treatment period and/or in older mice. It is possible that more subtle cognitive changes occurring in human patients would not be easily seen in BACEi treated mice in our test paradigms; mouse touchscreen test batteries modelling patient cognitive examinations (31,32) could be incorporated into future preclinical studies.
BACEi treatment in mice is known to alter dendritic spine turnover (19) in a reversible and Sez6-dependent manner (20) therefore it was important to determine whether BACE inhibition in this study produced similar changes in spine number or classi cation to those seen in Sez6 knockout (15) and Sez6 TKO (16) mice. Similar to previous reports of BACEi-induced decreases in spine numbers on apical dendrites of cortical neurons (19,20,33), BACEi treatment reduced mushroom spine density on basal dendrites of pyramidal cortical neurons. In this case, the reduction was not Sez6 family member dependent, as a similar decrease was observed in neurons from BACEi treated WT and Sez6 TKO mice. This observation might indicate opposing effects of the shed ectodomains of Sez6L/Sez6L2 and Sez6 on mushroom spine density and this possibility could be explored further in the single gene knockout lines. It is also relevant that the effect of BACE inhibition on cognition was tested prior to taking the brains for dendritic spine analyses and, thus, mice had experienced environmental enrichment during behavioural testing. Enrichment alters spine properties including density (34), which may have masked any treatment-induced changes and the previously identi ed shift from mature to immature spines seen in behaviourally naïve Sez6 TKO mice (16). Nevertheless, our results reinforce the need to examine the effect of chronic BACEi treatment on spine density and morphology with particular attention given to mushroom spine types, noting that any spine changes will vary according to neuron and dendrite type. The recently reported hippocampal long-term potentiation de cit in mice with only partial adult BACE1 deletion (30) emphasizes the importance of understanding how synaptic plasticity may be affected in patients treated with BACE inhibitors.

Conclusions
When considering the future of BACE inhibition as a therapeutic approach for the treatment of Alzheimer's disease, it is important to rst identify the factors underlying the increased cognitive worsening seen in BACEi clinical trials. While lowering the dosage of BACE inhibitors will likely limit the negative effects of treatment on cognition (2, 3), modi cations which speci cally reduce the impact of treatment on key BACE1 substrates should be considered in order to maximise the potential bene t of this strategy. In this report, no signi cant de cits in memory or cognitive exibility in mice treated with BACEi were observed, suggesting that clinical results may relate to the age, disease state and/or duration of treatment. However, even in young mice, BACEi increased locomotor activity in an aversive environment and this effect depended on Sez6 family proteins, providing evidence that interfering with Sez6 protein functions contributes to the adverse effects seen with BACE inhibition.

Figure 2
BACEi treatment does not alter spatial memory or cognitive exibility in mice. A) In the Morris water maze (MWM), WT and Sez6 TKO mice swam an equivalent distance to reach the hidden platform during acquisition training (from day 2) and reversal training (i) and Sez6 TKOs took a longer time than WT mice to nd the hidden platform during reversal training (ii). 30mg/kg/day BACEi treatment did not signi cantly affect pathlength or latency within genotypes (i-ii), or the ability of mice to successfully learn the hidden platform location in the acquisition and reversal probe trials (iii) although a tendency for BACEi treated WT mice to take longer to nd the platform than vehicle treated WT mice on the rst day of reversal training is noted (ii). n=15-16 per group. Ai, Log10 of pathlength (mm) and Aii, Log10 of latency (seconds): 2-way RM ANOVA with acquisition and reversal analysed separately. interactions. All graphs show mean ± SEM except Aiii which shows mean ± 95% CI. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001.