Insoluble Aβ overexpression in an App knock-in mouse model alters microstructure and gamma oscillations in the prefrontal cortex, and social and anxiety-related behaviours

We studied two new App knock-in mice models of Alzheimer’s disease (AppNL-F and AppNL-G-F), which generate elevated levels of Aβ40 and Aβ42 without the confounds associated with APP overexpression. This enabled us to assess changes in anxiety-related and social behaviours, and neural alterations potentially underlying such changes, driven specifically by Aβ accumulation. AppNL-G-F knock-in mice exhibited subtle deficits in tasks assessing social memory, but not in social motivation tasks. In anxiety-assessing tasks, AppNL-G-F knock-in mice exhibited: 1) increased thigmotaxis in the Open Field (OF), yet; 2) reduced closed-arm, and increased open-arm, time in the Elevated Plus Maze (EPM). Their ostensibly-anxiogenic OF profile, yet ostensibly-anxiolytic EPM profile, could hint at altered cortical mechanisms affecting decision-making (e.g. ‘disinhibition’), rather than simple core deficits in emotional motivation. Consistent with this possibility, alterations in microstructure, glutamatergic-dependent gamma oscillations, and glutamatergic gene expression were all observed in the prefrontal cortex, but not the amygdala, of AppNL-G-F knock-in mice. Thus, insoluble Aβ overexpression drives prefrontal cortical alterations, potentially underlying changes in social and anxiety-related behavioural tasks. Highlights AppNL-G-F KI mice displayed differing anxiety behaviours in two tests of anxiety. Prefrontal gamma was no longer NMDA-receptor dependent in AppNL-G-F KI mice. Prefrontal expression of Grin2b was reduced in AppNL-G-F KI mice. DTI found structural alterations in the hippocampus and prefrontal cortex in AppNL-G-F KI mice.


Introduction
Alzheimer's disease (AD) is classically associated with declining cognitive function (Scheltens, et al., 2016). However, this is only one aspect of the behavioural changes associated with AD. Other behavioural changes include reduced social engagement and increased anxiety. Although social aspects of AD have remained underexplored, social withdrawal is present up to 5 years prior to a clinical cognitive diagnosis (Jost and Grossberg, 1995). AD patients with larger social networks (the number of people with which one has meaningful contact with) have slower cognitive decline, compared to AD patients with small social networks (Bennett, et al., 2006).
Social factors may modulate the rate of disease pathology and cognitive decline, but crucially, also the chance of developing AD (Kuiper, et al., 2015). Studies have found that for the elderly that identify as lonely, the risk of developing AD was nearly doubled (Wilson, et al., 2007). Recent evidence from the Lancet Commission Report highlights that social isolation constitutes a 2.3% risk of developing AD (Livingston, et al., 2017). Thus, social factors not only modulate the risk of developing dementia, but also disease progression. Together, it can be inferred that changes in social motivation of the individual as a result of AD pathology may be a factor in disease progression.
Anxiety in AD is relatively common, with up to 71% of patients reporting anxiety concerns (Ferretti, et al., 2001,Teri, et al., 1999. Up to 6% had anxiety that reached the diagnostic criteria of generalised anxiety disorder of Diagnostic and Statistical Manual of Mental Disorders (Ferretti, et al., 2001). Anxiety behaviours may also predict conversion to AD. Mild cognitive impairment (MCI) is thought to be a precursor condition to AD. 83.3% of MCI patients that also exhibited anxiety symptoms converted to AD within a three year follow up period compared to 40.9% of persons with MCI but without anxiety (Palmer, et al., 2007). The latter suggests that anxiety is associated with early phases of AD. Neurodegeneration in brain structures in early AD may explain the increase in anxiety. Extensive literature has linked the amygdala to anxiety behaviours (Shin and Liberzon, 2010). In very mild and mild AD, amygdala atrophy was similar to that of the hippocampus and was correlated to anxiety (Poulin, et al., 2011), suggesting that amygdala degeneration could explain AD-related anxiety. Together, non-cognitive behavioural changes in AD can occur early in disease progression and have clear modulatory effects on disease progression. Currently, we have few insights into the biological mechanisms underpinning anxiety and social withdrawal in AD, highlighting the need to find AD mice that model these behaviours.
Broadly speaking, pathologies underpinning AD tend to group into two domains; those of amylogenic pathways and those of tau, although the two are not mutually exclusively (Bloom, 2014). The animal models have tended to focus on only one of these pathology pathways, although the majority of these model the amyloid aspects of AD through amyloid-β peptide (Aβ) plaques. Despite many transgenic mouse models of AD existing, previous generations of Aβ mice have achieved their Aβ overexpression by also overexpressing amyloid precursor protein (APP). Over time, it has become clear the APP overexpression alone can introduce confounds that make it difficult to dissociate the causal effects of Aβ compared to APP. Mice overexpressing human wild-type APP have cognitive impairments in the Morris watermaze and object recognition test, increased tau hyperphosphorylation and reduced GluA1, GluA2, GluN1, GluN2A, GluN2B, phosphorylated CaMKII and PSD95 (Simon, et al., 2009). This was also associated with reduced cell density in the pyramidal layer of CA1 (Simon, et al., 2009). Others have found that other Aβ mice (the APP23) produced higher amounts of APP fragments, including C-terminal fragment β /α and APP intracellular domain (Saito, et al., 2014). The phenotypes associated with APP overexpression alone encapsulate what are known to be ADspecific pathologies, underlining the potential risks of using APP overexpression models to make AD interpretations.
Recently, a new generation of AD mouse models have become available that achieves Aβ pathologies without the overexpression of APP (Saito, et al., 2014). This has been achieved by humanising the murine Aβ sequence through a knock-in (KI) strategy. Up to three familial mutations have been knocked-in to generate two AD mouse models. The App NL-F mouse contains the Swedish KM670/671NL mutation and the Iberian I716F mutation which results in elevated total Aβ and elevated Aβ 42 /Aβ 40 ratio (Saito, et al., 2014). The App NL-G-F mouse also contains the Swedish and Iberian mutations, but with the additional Artic mutation E693G which promotes aggressive Aβ oligomerisation and amyloidosis (Saito, et al., 2014). In App NL-G-F mice, Aβ plaques are saturated by 9 months of age, but by contrast, App NL-F mice exhibit relatively few plaques. Both models have approximately similar amounts of soluble Aβ species (Saito, et al., 2014). Hence, these two mouse models are useful to be able to dissociate the effects of soluble Aβ and plaque-based Aβ in the generation of AD-related pathologies.
Within the current study, we have sought to explore whether these new App KI mouse models display alterations in behaviours related to social motivation, social memory and anxiety. We subsequently used electrophysiology, diffusion tensor MRI (DTI) and quantitative PCR (qPCR) to determine putative biological mechanisms that may explain, with brain region specificity, the underlying cause of behavioural impairments.

Ethics
All procedures were approved by the Durham University and University of York Animal Ethical and Welfare Review Boards and were performed under UK Home Office Project and Personal Licenses in accordance with the Animals (Scientific Procedures) Act 1986.

Mice
Full details of the animals can be found elsewhere (Saito, et al., 2014). Upon arrival at Durham, mice were backcrossed once to the C57BL/6J line, which was the same background strain as the previous institution where they were bred. Homozygote App NL-F and App NL-G-F and wild-type littermate mice were bred in house by heterozygote pairings and were weaned at P21 and ear biopsied for genotyping.
Genotyping protocols can be found elsewhere (Saito, et al., 2014). All mice were housed with littermates.
Testing procedures can be found elsewhere (Dachtler, et al., 2014). All trials were recorded by Any-maze (Stoelting, Dublin, Ireland) tracking software. Mice undertook the following test in this order: open field, elevated plus maze, social approach, social recognition and social olfactory discrimination. In brief, mice were placed into a 44 cm 2 white Perspex arena and allowed free ambulation for 30 mins. The floor of the arena was divided by Any-maze into an outer zone of 8 cm and a centre zone of 17.5 cm 2 . Time spent and entries into these zones, along with distance travelled were measured. The elevated plus maze and social approach, social recognition, and social olfaction discrimination was run as in (Dachtler, et al., 2014), except that novel conspecifics and soiled bedding was sex matched with adult mice to the test subject. At the conclusion of testing, mice were either used for electrophysiology, killed by perfusion fixation with 4% paraformaldehyde in phosphate buffered saline (PBS) or killed by cervical dislocation for molecular biology.

Diffusion Tensor MRI
8 wild-type (4 male (9.02 months ± 0.63), 4 female (9.43 months ± 0.30)) and 8 App NL-G-F homozygotes (4 male (8.88 months ± 0.22), 4 female (9.89 months ± 0.21)) were used in for MR imaging. Image acquisition has been described elsewhere (Pervolaraki, et al., 2017). MR imaging was performed on a vertical 9.4 Tesla spectrometer (Bruker AVANCE II NMR, Ettlingen, Germany). During imaging, the samples were placed in custom-built MR-compatible tubes containing Fomblin Y The ex vivo mouse brain 3D diffusion-weighted images were reconstructed from the Bruker binary file using DSI Studio (http://dsi-studio.labsolver.org). Direction Encoded Colour Map (DEC) images were generated by combining the information from the primary eigenvectors, diffusion images and the fractional anisotropy (FA).
Regions were extracted by manually segmenting orbitofrontal cortex (OFC), anterior cingulate cortex (ACC), hippocampal and whole amygdalar regions using a mouse brain atlas (Franklin and Paxinos, 2008). We chose to focus on the aforementioned brain regions as these have been linked to social behaviour (Eleftheria Pervolaraki, et al., 2018). Additionally, the hippocampus , the amygdala (Davis, 1992,Davis, et al., 1994 and prefrontal cortical regions (including the ACC) (Davidson, 2002,Etkin, et al., 2011 have all been linked to anxiety. Extraction of FA, mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD) was performed within selected segmented brain regions, with 3 100 μm sections (one anterior and one posterior to the segmented section) extracted.
Full atlas-based description of the segmentation can be found in .
All salts were obtained from BDH Chemicals (Poole, UK) or Sigma-Aldrich (Poole, UK) and KA and CPP were obtained from BioTechne (Minnesota, USA).
Extracellular field recordings were obtained using borosilicate micropipettes (Harvard Apparatus) filled with ACSF and had resistances of 2-5 (MΩ). Recordings were bandpass filtered at 0.1Hz to 300Hz. Power spectra were derived from Fourier transform analysis of 60 s epochs of data and results were presented as mean ± SEM.

RNA Extraction and qPCR
For molecular biology, mice underwent cervical dislocation, with the brain extracted and placed into a mouse brain blocker (David Kopf Instruments, Tujunga, USA). The olfactory bulbs were removed, followed by a section of tissue from Bregma 3.56 to 2.58 mm, and from Bregma -0.82 to -2.80 mm. The tissue containing prefrontal tissue was snap frozen. The posterior section was then further dissected with a scalpel to remove the amygdalar region, which was then snap frozen and stored at -80°C until use. 6 wild-type (4 male, 2 female (10.26 months ± 0.41)) and 6 App NL-G-F homozygotes (3 male, 3 female (9.49 months ± 0.13)) were used for qPCR. Brain tissue of extracted regions (amygdala and prefrontal cortex) were homogenised by TissueRuptor drill (Qiagen, Manchester, UK), with ~90 mg used for RNA extraction.

Instructions were followed as per the Bio-Rad Aurum Total RNA Fatty and Fibrous
Tissue Kit (Bio-Rad, Hemel Hempsted, UK, Cat# 7326830) and our previously optimised protocol (E. . RNA quantity and quality were confirmed by Nanodrop spectrophotometer. cDNA was generated by the iScript cDNA Synthesis Kit (Bio-Rad, UK), with 1 µg of RNA used per reaction.
Primers (Table 1) were designed using Primer3 after identifying the gene sequence on NCBI. Primers (Integrated DNA Technologies, Leuven, Belgium) were tested for specificity and run conditions optimised by PCR using whole brain cDNA. Plates were run with 10 µl per reaction, with 1 µl of cDNA, Bio-Rad SYBR Green (cat# 1725121) and 300 nM primers. Samples were run in triplicate using the protocol of 95C for 3 min, followed by 95C for 10 s, 60C for 30 s repeated 35 times. Gene expression was imaged using a Bio-Rad CFX Connect and analysed using Bio-Rad Connect Manager, and quantified using the 2^ddCt method against the housekeeping gene Gapdh which did not differ between the genotypes.

Immunostaining
Staining for amyloid plaques can be found elsewhere (Ly, et al., 2011). Tissue previously scanned by MRI was washed and immersed in 30% sucrose for >48 hrs.
Tissue was cryosectioned at 30 µm, followed by immersion in 88% formic acid for 15 mins. Endogenous peroxidase activity was quenched with H 2 O 2 for 30 mins, followed by blocking in 2% bovine serum albumin (BSA) (Thermo Fisher Scientific) for 1 hr at room temperature. Sections were then incubated with 1:500 6E10 (BioLegend, San Diego, CA, USA) overnight at 4°C, followed by 1:500 biotinylated anti-mouse (Vector Labs, Peterborough, UK) for 2 hrs at room temperature. Sections were then reacted with Vectastain ABC kit (Vector Labs) followed by diaminobenzidine (DAB) treatment (Vector Labs) prior to mounting on slides for imaging by light microscopy.

Data Analysis
All data are expressed as mean ± standard error of the mean (SEM). To assess the variance between genotypes within a single brain structure across hemispheres, data was analysed by within subject repeated measures two-way ANOVAs or unpaired T-tests. To correct for multiple comparisons, we employed the Benjamini-Hochberg Procedure, with false discovery rate set to 0.4 (corrected P values stated).
Behaviour was analysed with ANOVAs, followed by tests of simple main effects. To test for discrimination against chance, one sample T-tests were used set against chance (0.5). Non-significant statistical results, particularly hemisphere comparisons, can be found in Supplemental Materials. Statistical testing and graphs were made using GraphPad Prism version 6 and SPSS v22.

App NL-G-F KI mice have changes in emotional behaviours
To assess the potential role of Aβ in emotional behaviour, we undertook two behavioural paradigms that are widely used to probe anxiety: the open field and the elevated plus maze. In the open field, mice were allowed to freely ambulate for 30 mins, during which we measured the distance they travelled in 5 min blocks. We To explore whether social motivation was altered in App KI mice, we examined sociability in the three-chambered social approach test  which exploits the preference of a mouse to explore a novel mouse enclosed within a wire cage compared to an identical empty cage. All genotypes showed clear preference for exploring the cage containing the novel mouse, and this did not differ between the genotypes ( Fig. 2A: genotype F (2, 55) = 1.71, p = 0.19, genotype x discrimination F (2, 55) <1, p = 0.576, genotype x sex F (1, 55) = 1.98, p = 0.149). Next, we tested whether App KI mice were able to show social novelty preference for exploring a second novel conspecific compared to the previous explored conspecific (Fig. 2B). We found a significant interaction between genotype and sex for social novelty recognition (F (2, 55) = 3.39, p = 0.041), although there was no main effect of genotype (F (2, 55) = 2.31, p = 0.11) or genotype x discrimination (F (1, 55) = 1.17, p = 0.32).
To further explore this sex effect, we specifically examined whether the discrimination ratios for social novelty preference for each genotypic sex were above chance. For both males (Fig. 2C) and females (Fig. 2D), only wild-type mice displayed discrimination that was significantly above chance, with male App NL-F and App NL-G-F KI mice showing greater variation in performance (see Fig. 2 legend for statistics).
Finally, we tested whether App KI mice showed motivation to explore a social smell (soiled bedding) compared to a non-social smell (clean bedding). Similar to the social novelty preference, we observed that discrimination between the social and non-social olfactory stimulus differed by sex and genotype ( Supplementary Fig. 2: genotype x discrimination x sex F (2, 55) = 4.31, p = 0.019, genotype x discrimination F (2, 55) <1, p = 0.619, genotype x sex F (1, 55) = 2.01, p = 0.144). To further investigate the source of difference, we examined the discrimination ratios separated by sex (Fig. 2E,F). For males, all genotypes show significant preference for exploring the social smell compared to chance. However, although female wild-type and App NL-F KI mice showed preference for the social smell, female App NL-G-F KI mice were not significantly different from chance (see Figure legend for statistics).

App NL-G-F KI mice have microstructural changes in the prefrontal cortex and hippocampus
Given that behavioural changes in the open field, elevated plus maze and social olfaction test are altered only in App NL-G-F KI mice, we decided to take this genotype forward for further analysis to determine the biological mechanisms that may explain these impairments. Our approach was to examine the integrity of brain regions associated with both social and anxiety behaviours (see Methods). These centred Given the amygdala has been widely associated with both social and anxiety behaviours (Adolphs, 2010,Davis, 1992,Davis, et al., 1994, we segmented the whole amgdalar region, in anterior and posterior planes, and separately, the basolateral nuclei (BLA), to determine whether structural alterations may explain the behavioural changes. For all measures (FA, MD, AD and RD), in both the anterior and posterior amygdala, plus the BLA, we did not observe any significant changes in tissue diffusion properties (Supplementary Table 1 for non-significant statistics and Supplementary Fig. 3).
Ventral hippocampal regions have been associated with anxiety and fear behaviours , whilst dorsal and ventral regions are increasingly being associated with social recognition (Hitti andSiegelbaum, 2014,Okuyama, et al., 2016). We therefore next examined the microstructure of the hippocampus along the anterior to posterior axis. FA of the anterior (Fig. 3A) and posterior (Fig. 3B) hippocampus did not differ between wild-types (Supplementary Table 2 for nonsignificant statistics). However, whilst MD in the anterior hippocampus (Fig. 3C) was similar between the genotypes F (1, 14) <1, p = 0.263), MD was significantly higher in the posterior hippocampus (Fig. 3D) in App NL-G-F KI mice (F (1, 14) = 12.18, p = 0.011).
We next segmented prefrontal cortical regions including the OFC and the ACC. In addition to roles for OFC and ACC regions in anxiety and social behaviours, resting state functional MRI (rsfMRI) has shown that the ACC was the most altered brain region in App NL-G-F KI mice. However, it is currently unknown as to whether structural changes in the ACC contributed to the rsfMRI result. In the OFC of App NL-G-F KI mice, there were no significant differences in FA ( genotype (t (14) <1, p = 0.37) but MD was significantly increased ( Fig. 4D; genotype (t (14) = 3.13, p = 0.021). We furthered explored prefrontal cortical changes by quantifying AD and RD. In the OFC, both AD and RD were significantly increased in App NL-G-F KI mice (F (1, 14) = 8.34, p = 0.032 (Fig. 4E) and F (1, 14) = 8.23, p = 0.042 (Fig.   4F), respectively). AD and RD were also significantly increased in the ACC of App NL-G-F KI mice (t (14) = 2.88, p = 0.037 (Fig. 4G) and t (14) = 3.19, p = 0.016 (Fig. 4H).
Finally, we performed amyloid plaque staining on corresponding tissue sections as was analysed for DTI. By 9 months of age in App NL-G-F KI mice, the OFC, ACC, amygdala and hippocampus all exhibit substantial plaque load ( Supplementary Fig.   4), indicating that the lack of DTI changes in the amygdala are not due to an absence of amyloid plaques.

App NL-G-F KI mice
Although some microstructure changes were detected in the hippocampus of App NL-G-F KI mice (anterior hippocampal AD and posterior hippocampal RD), these were relatively inconsistent compared to the alterations observed within the prefrontal cortex. Another study found the ACC as being the most significantly altered brain region in App NL-G-F KI mice as detected by rsfMRI (Latif-Hernandez, et al., 2017), suggesting that substantial changes are occurring in the ACC. As such, and given the purported importance of the ACC and amygdala to anxiety and social behaviours, we decided to contrast these two regions to examine whether prefrontal function is affected as a result of the DTI-derived microstructural alterations.
Gamma oscillations have been widely associated with roles in learning, coherence between brain regions facilitating information transfer (Buzsaki and Wang, 2012), social behaviour (Cao, et al., 2018) and anxiety behaviours within the medial prefrontal cortex (Adhikari, et al., 2010). Gamma has been shown to disrupted in other AD mouse models, such as the APP/PS1 (Klein, et al., 2016), the TAS10 overexpression model (Driver, et al., 2007) and in the entorhinal cortex of App NL-G-F KI mice (Nakazono, et al., 2017), and these impairments occur relatively early into Aβ pathology. Thus, gamma oscillations represent a useful target for exploring ADrelated pathology.
We generated gamma oscillations in brain slices using kainate and tested their dependency on NMDA receptors, which have previously been shown to modulate peak frequency through inhibitory postsynaptic currents (McNally, et al., 2011) by modifying recruitment of different interneuron subpopulations (Middleton, et al., 2008). Within the BLA of the amygdala, we found that in wild-types, peak amplitude and frequency of gamma oscillations were unaffected by the application of the broad spectrum antagonist NMDA receptor antagonist CPP ( Fig. 5Aiii; t (14) <1, p = 0.991 and t (14) <1, p = 0.656, respectively). We observed similar results in the App NL-G-F KI mice with peak amplitude and frequency unaffected by NMDA antagonism (Fig. 5Aiv;t (14) <1, p = 0.951 and t (14) <1, p = 0.709, respectively). Together, this suggests that gamma oscillations in the BLA in App NL-G-F KI mice are unaffected by Aβ deposition.
Next, we studied gamma oscillations within the ACC of the prefrontal cortex. In wildtypes, the frequency of gamma oscillations were unaffected by CPP ( Fig. 5Biii; t (14) = 1.91, p = 0.086). However, gamma peak amplitude was significantly reduced by CPP application (Fig. 5Biv; t (14) = 2.96, p = 0.014), suggesting that ACC gamma oscillations require NMDA receptors. In App NL-G-F KI mice, we found that unlike in wild-types, neither gamma frequency nor amplitude was altered by NMDA receptor antagonism (t (14) <1, p = 0.430 and t (14) <1, p = 0.404, respectively), suggesting that ACC gamma oscillations in App NL-G-F KI mice have lost their dependency upon NMDA receptors.
To further analyse the NMDA receptor-dependent alterations in the prefrontal cortex, we next analysed the mRNA expression of synaptic genes including NMDA receptors and pre and postsynaptic receptors relating to NMDA receptor function.
Together, the change in the dependency of NMDA receptor mediated gamma in the ACC of App NL-G-F KI mice could be related to reduced Grin2b or presynaptic release through Munc18-1.

Discussion
Within the current study, we have further characterised the behavioural profile of the new generation of App KI mice. Our results indicate that App NL-G-F KI mice have differing anxiety behaviours depending upon the paradigm used, and a mild impairment in social recognition and social olfactory discrimination. We further characterised the cellular correlates of these impairments, and found that App NL-G-F KI mice have structural alterations in the prefrontal cortex, changes to NMDAreceptor dependent gamma oscillations in the ACC, which may relate to a lack of

Grin2b.
The development of the App NL-F KI and App NL-G-F KI mice has allowed researchers to dissociate the amyloidogenic effects of Aβ deposition. At around 9 months of age,

App NL-G-F KI mice have nearly 100 times the amount of insoluble Aβ 42 compared to
App NL-F KI mice (Saito, et al., 2014). Thus, at 9 months, App NL-G-F KI mice are almost plaque saturated compared to App NL-F KI mice, which have few detectable plaques (Saito, et al., 2014). This enables the dissociation between whether soluble or plaque-based Aβ 42 results in changes to behavioural performance.
Within the current study, we observed a conflicting phenotype between increased Distance travelled in the open field is significantly greater in 6 month old App NL-G-F KI mice, whilst at 8 months (the current study) and 10 months of age, there were no significant differences (Latif-Hernandez, et al., 2017,Whyte, et al., 2018, suggesting hyperactivity is only detectable in younger mice. Like our study, Latif-Hernandez et al. (2017) also used the elevated plus maze to explore anxiety behaviours in App NL-G-F KI mice, except they compared to App NL KI mice. They found App NL-G-F KI mice spent more time in the open arms at 6 months of age (Latif-Hernandez, et al., 2017), which we replicated at 8 months of age. A very recent study has also found that 6-9 month old App NL-G-F KI mice spent more time in the elevated plus maze open arms (Sakakibara, et al., 2018). Together, independent studies suggest that App NL-G-F KI mice display an anxiolytic profile in the elevated plus maze. The contradiction between our results in the open field and elevated plus maze is curious but not unique. Tg2576 mice also show open field thigmotaxis yet increased time in the elevated plus maze open arms (Lalonde, et al., 2003). It is possible that the increased time spent in the open arms reflects a disinhibition phenotype. Disinhibition is a well-established, albeit less common, AD phenotype (Chung andCummings, 2000,Hart, et al., 2003), which could be manifested within the current study as a failure to inhibit the choice to enter the open arm. Given the lack of specific data to speak to a disinhibition hypothesis, it is clear that future anxiety testing in App NL-G-F KI mice will require multiple paradigms to clearly delineate anxiety from disinhibition. A benefit of our study is that we are the first to behaviourally compare App NL-F KI and App NL-G-F KI mice beyond the original Saito (2014) paper. Regarding the open field and elevated plus maze, genotypic differences were only found in App NL-G-F KI mice, suggesting behavioural modifications were primarily driven by insoluble Aβ. Currently, aged App NL-F KI mice have not been tested for anxiety behaviours, but it would be interesting to see if as these mice age, and plaque pathology increases, whether their behavioural phenotype becomes more like the App NL-G-F KI genotype.
Several other AD mouse models exhibit impairments in social behaviour, including APP swe /PS1 (Filali, et al., 2011) and Tg2576 mice (Deacon, et al., 2009). Latif-Hernandez et al. (2017 examined social behaviours in App NL-G-F KI mice and found that although not significant, the discrimination ratio for exploring a novel mouse compared to an empty cage, and for comparing between the previously explore mouse and a second novel mouse, had fallen to chance by 10 months of age (Latif-Hernandez, et al., 2017). We found that although App NL-F KI and App NL-G-F KI mice show robust preference for exploring a novel conspecific compared to an empty cage, social recognition discrimination was close to chance for both sexes. However, statistical analysis of the absolute times spent exploring each conspecific did not reveal a significant effect of genotype upon discrimination, suggesting that the overall detrimental effect upon social behaviour is likely to be mild. We did, however, find that social preference as measured by discriminating a social odour cue was only impaired in female App NL-G-F KI mice. A key difference between our study and Latif-Hernandez et al. (2017) is that they used only female mice. Our study, with mixed-sex groups, suggests that females show a greater impairment in social behaviour than males, which will be an important consideration for future studies.
The ACC and amygdala are brain regions well established with mediating anxiety/fear and social behaviours (Davidson, 2002,Davis, 1992,Davis, et al., 1994,Etkin, et al., 2011. Further, these areas are also susceptible to degeneration in early AD pathology (Huang, et al., 2002,Poulin, et al., 2011,Scheff and Price, 2001. Other transgenic AD mouse models, notably the APP/PS1 mouse, exhibit early amygdala pathology (Guo, et al., 2017,Lin, et al., 2015, whilst Aβ deposition in the PDAPP mouse begins in the cingulate cortex (Schenk, et al., 1999). Together, we hypothesised that structural and functional alterations within these regions could partly explain behavioural impairments.
Our analysis of microstructural integrity using DTI revealed that the prefrontal cortex of App NL-G-F KI mice, including the OFC and ACC, was substantially altered, and the functional consequences of these changes to the ACC was that gamma oscillations lost their dependency upon NMDA receptors, notably GluN2B receptors. Thus far, there have been limited investigations into the biological pathways altered in App NL-G-F KI mice. Using rsfMRI, Latif-Hernandez et al. (2017) found that the cingulate cortex was the most significantly altered brain region, with no visible impairment in the amygdala. Although further work is required to definitively describe the physiological pathways that explain the altered behaviour in App NL-G-F KI mice, our findings present an important step in this process. Although other brain regions likely contribute to changes in anxiety and social behaviours (gamma oscillations have also been found altered in the medial entorhinal cortex of App NL-G-F KI mice (Nakazono, et al., 2017)), the prefrontal cortex could represent a key region. Neural oscillations in the medial prefrontal cortex have clear links to anxiety behaviours, both in the open field and elevated plus maze (Adhikari, et al., 2010), and medial prefrontal cortex gamma is required for the expression of social novelty (Cao, et al., 2018). Additionally, the loss of GluN2B or its phosphorylation impairs social behaviour (Jacobs, et al., 2015,Wang, et al., 2011 and increases anxiety in the open field (Hanson, et al., 2014) and the elevated plus maze (Delawary, et al., 2010). Furthermore, NMDA receptors are necessary for gamma mediated by the goblet cell interneurons within the entorhinal cortex (Middleton, et al., 2008), with specific antagonism of GluN2B significantly reducing hippocampal gamma power (Hanson, et al., 2013). Together, insoluble Aβ overexpression drives prefrontal cortical alterations through gamma oscillatory impairments by reduced GluN2B expression. The lack significant changes to gamma, gene expression and microstructure in the amygdala suggests that it may not undergo pathological damage at the same rate as other regions. This indicates that social and anxiety behavioural impairments in App NL-G-F KI mice are driven by regions that do not include the amygdala, although further work will be required to corroborate this.

Conclusion
The findings presented herein show a clear dichotomy of anxiety behaviours between two different paradigms. Contrasted to App NL-F KI mice, only App NL-G-F KI mice exhibited any change in anxiety behaviours, suggesting that plaque-based Aβ is responsible for these effects. We further postulate that microstructural integrity, gamma oscillatory function and Grin2b expression within the prefrontal cortex may, in part, explain these behavioural changes. Our data continues to highlight the importance of using AD models that do not have the confound of APP overexpression. Further studies will be required to continue to refine the mechanisms that explain anxiety impairments in App NL-G-F KI mice.

Figure 2
Social behaviours in App KI mice. A. All genotypes showed similar preference for exploring a novel, same-sexed novel conspecific (Stranger 1), compared to an empty cage. However, App KI mice displayed a weaker discrimination between the previously explored mouse (Stranger 1) and a second novel conspecific (Stranger 2), which differed by sex (B). To further investigate this sex difference, social recognition discrimination ratios were spilt by sex. Male (C) and female (D) App KI mice did not show discrimination that was significantly above chance, suggesting a social recognition impairment. Mice were then required to discriminate between a social smell (soiled bedding) or a non-social smell (clean bedding). Although all male genotypes showed preference for exploring the social cue (E), the discrimination ratio for female App NL-G-F KI mice was not above chance (F). ***=P<0.001, **=P<0.01, *=P<0.05. Error bars = s.e.m. Wild-type: male n = 9, female n = 11; App NL-F KI: male n = 10, female n = 9; App NL-G-F KI: male n = 10, female n = 9.

Figure 3
Alterations in fractional anisotropy (FA) and mean diffusivity (MD) in the hippocampus. DTI images of the hippocampus was segmented at two regions; anterior (Bregma -1.94 mm) and posterior (Bregma -3.28 mm). FA was not significantly altered between the wild-type and App NL-G-F KI mice in the anterior (A) and posterior (B) hippocampus. MD of the anterior hippocampus did not differ between the genotypes (C), however MD in the posterior hippocampus was significantly higher in App NL-G-F KI mice (D). Diffusivity was further characterised by examining axial (AD) and radial (RD) diffusivity. AD was significantly increased in the anterior hippocampus of App NL-G-F KI mice (E), but not in the posterior hippocampus (F). RD did not differ between the genotypes in the anterior hippocampus, but App NL-G-F KI mice had significantly increased RD posterior hippocampus compared to wildtypes. **=P<0.01, *=P<0.05. Error bars = s.e.m. Wild-type n = 8, App NL-G-F KI n = 8.

Figure 4
Alterations in fractional anisotropy (FA) and mean diffusivity (MD) in the orbitofrontal cortex and anterior cingulate cortex of the prefrontal cortex. DTI images were segmented for the orbitofrontal cortex at Bregma +2.58 mm and for the anterior cingulate cortex at Bregma +1.18 mm. FA was not significantly altered between the wild-type and App NL-G-F KI mice in the orbitofrontal cortex (A), however MD was significantly higher for App NL-G-F KI mice (B). Similarly, FA was did not differ between the genotypes for the anterior cingulate cortex (C), but MD was significantly higher in App NL-G-F KI mice (D). Axial (AD) and radial (RD) diffusivity was then analysed. In App NL-G-F KI mice, AD (E) and RD (F) were both significantly increased in the orbitofrontal cortex. AD (G) and RD (H) were also both increased in App NL-G-F KI mice in the anterior cingulate cortex. *=P<0.05. Error bars = s.e.m. Wild-type n = 8, App NL-G-F KI n = 8.  Gene expression within the frontal cortex and amygdala. A. Although App NL-G-F KI mice generally had higher expression of our selected genes within the amygdala, these were not significantly different to wild-types. B. Within the frontal cortex, expression of both Grin2b and Stxbp1 (Munc18-1) were significantly reduced in App NL-G-F KI mice. *=P<0.05. Error bars = s.e.m. Wild-type n = 6, App NL-G-F KI n = 6.