Early memory deficits and extensive brain network disorganization in the App^NL-F/MAPT double knock-in mouse model of familial Alzheimer’s disease

Abstract

A critical challenge in current research on Alzheimer’s disease (AD) is to clarify the relationship between network dysfunction and the emergence of subtle memory deficits in itspreclinical stage. The AppNL-F/MAPT double knock-in (dKI) model with humanized β-amyloid peptide (Aβ) and tau was used to investigate both memory and network dysfunctions at an early stage. Young male dKI mice (2 to 6 months) were tested in three tasks taxing different aspects of recognition memory affected in preclinical AD. An early deficit first appeared in the object-place association task at the age of 4 months, when increased levels of β-CTF and Aβ were detected in both the hippocampus and the medial temporal cortex, and tau pathology was found only in the medial temporal cortex. Object-place task-dependent c-Fos activation was then analyzed in 22 subregions across the medial prefrontal cortex, claustrum, retrosplenial cortex, and medial temporal lobe. Increased c-Fos activation was detected in the entorhinal cortex and the claustrum of dKI mice. During recall, network efficiency was reduced across cingulate regions with a major disruption of information flow through the retrosplenial cortex. Our findings suggest that early perirhinal-entorhinal pathology is associated with abnormal activity which may spread to downstream regions such as the claustrum, the medial prefrontal cortex and ultimately the key retrosplenial hub which relays information from frontal to temporal lobes. The similarity between our findings and those reported in preclinical stages of AD suggests that the AppNL-F/MAPT dKI model has a high potential for providing key insights into preclinical AD.

Introduction

The sporadic form of Alzheimer’s disease (AD) represents the vast majority of late onset patients. Less than 1 % of the cases develop an early onset autosomal dominant AD (ADAD) form due to the presence of mutations in amyloid precursor protein (APP), presenilin 1 (PSEN1) or presenilin 2 (PSEN2) genes [7]. However, the relative similarity in the temporal sequence of cognitive symptoms and AD biomarkers in ADAD and sporadic AD suggests common steps in disease progression [[35], [26]]. The current challenge is to identify predictive biochemical and functional markers to ultimately develop prevention strategies [76].

Early signs of cognitive decline and β-amyloid peptide (Aβ) and tau pathologies have been associated to functional network perturbations in elderly and prodromal stages of AD Myers et al. [[48], [73], [54]]. Using functional magnetic resonance imaging (fMRI), alterations in resting state functional connectivity (FC) has been frequently reported in amnestic mild cognitive impairment (aMCI), a potential prodromal stage of AD [[56], [22]]. More recently, research on these FC alterations has benefited from the use of graph theory to evaluate the topology of functional networks [42]. In aMCI patients, some studies reported functional network perturbations within and between default mode network (DMN) and the medial temporal lobe (MTL) with reduced strength of cortical hubs [[21], [87], [42]]. Others have reported opposing or null results [[24], [27]]. This variability in FC outcomes has been proposed to reflect heterogeneity of aMCI staging [[36], [46]]. These findings highlight the need to investigate network alterations in earlier preclinical stages of AD Sperling et al., [76]. The recent development of highly demanding tasks taxing specific aspects of recognition memory, such as associative memory and mnemonic discrimination tasks, have revealed subtle memory deficits in asymptomatic populations at risk for AD and presymptomatic ADAD patients [[52], [75], [34]]. Reduced performance in these tasks have been linked to early signs of AD pathology in cognitively normal elderly [[62], [94], [43]]. Stark et al. [77] showed that in elderly reduced performance in an object-based mnemonic discrimination task was found only in individuals that were also impaired in a delayed word-list recall task, Interestingly, the same study reported that MCI patients cumulated mnemonic discrimination and traditional object recognition impairments. More recently, accelerated forgetting in a variety of recognition memory tasks was also detected in asymptomatic ADAD patients [89]. Thus, the emergence of deficits in mnemonic discrimination, associative memory and long-term recognition memory without impairment in traditional, low-demanding recognition paradigms may provide an early signal in the preclinical stage of AD. Interestingly, fMRI studies show that performances in such cognitively demanding tasks depend on specific interactions across DMN and MTL structures sensitive to aging and early AD [[61], [30], [60], [13]]. Evaluating FC changes related to early impairments in these tasks may achieve a clearer understanding of neural networks first affected in preclinical AD [25].

As a complementary approach, Alzheimer’s research benefits from new powerful tools with the humanized knock-in mouse models created to avoid several artificial phenotypes found in transgenic models [[69], [31]]. Among these models, those with slow onset of AD pathology provide a unique opportunity to capture comprehensively discrete molecular to functional perturbations associated to preclinical stages [[74], [70]]. The App^NL-F/MAPT double knock-in (dKI) model expresses APP with two ADAD mutations known to increase Aβ production and all six isoforms of human tau [67]. As such, the dKI mice would model ADAD rather than sporadic AD. Amyloid deposition and tau pathology of these mice was moderate at the age of 24 months [67]. Thus, the aim of this study was to use this slow development model to investigate regional dysfunctions and brain FC changes associated with the earliest deficit detected in a battery of three object exploration tasks. Each task preferentially tested one of the three aspects of recognition memory presumably affected in prodromal stages of AD, e.i., associative memory with the object place (OP) task, mnemonic discrimination with the pattern separation (PS) task or delayed recognition memory with the long-term novel object recognition task. At the age of onset of the earliest enduring deficit which was detected in the OP task, early signs of neuropathology were assessed using AD-relevant biochemical markers of APP metabolism and tau phosphorylation in the hippocampus and the medial temporal cortex (MTC). Post-mortem mapping of activity-regulated expression of immediate early gene c-fos has been widely accepted to identify populations of neurons activated during encoding and retrieval phases of different aspects of recognition memory in rodents (e.g., [[1], [5], [79]]. Immunohistochemical imaging of nuclear accumulation of the c-Fos protein product in activated cells provides several advantages over fMRI imaging, such as single-cell resolution and the ability to examine activity in behaving rodents. In this context, this approach is widely considered as a fair indicator of learning and memory driven neural activity associated to synaptic plasticity. Thus, c-Fos protein expression was quantified following OP encoding and OP recall in separate groups of wild type (WT) and dKI mice. Task-induced c-Fos expression was first analyzed in a large set of regions encompassing primarily the DMN/MTL to detect eventual trial- and/or genotype-specific regional differences. The c-Fos neuroimaging approach provides a single snapshot of cumulated activation over several minutes that precludes functional connectivity estimation based on the within subject co-variance computation typically used in fMRI. However, recent progress has been made in transferring imaging-based network analyses on c-Fos data repetition using a between subject covariance approach [[90], [79], [83]]. Thus, functional networks were investigated in WT and dKI mice during each phase of the OP task using across-subjects correlations, and. FC network properties and hubs were then further assessed using graph theory techniques.

Methods

Animals

Founders of the App^NL-F/MAPT dKI and wild type (WT) mouse lines were established through the interbreeding of heterozygous single knock-in App^NL-F and MAPT mice obtained from the RIKEN BioResource Center (Japan) [[66], [31]]. The dKI line was backcrossed with C57BL/6J mice (Janvier Laboratories, France) every-three generation. The App^NL-F gene contains a humanized Aβ fragment with Beyreuther/Iberian and Swedish mutations, and the human MAPT gene led to the expression of all six isoforms of tau. Mice were group-housed with food and water ad libitum, and nesting material (room at 23 °C ± 1 °C; 12/12-hour light/dark cycle; lights on at 8.00 AM). Procedures were in compliance with rules of the European Community Council Directive 2010–63 and French Department of Agriculture Directive 2013–118 and approved by the local review board (CREMEAS: APAFIS#9848; Animal facilities: H 67–482-13).

Behavioral testing

Male mice were single-housed for 1 week before testing and a tunnel (15 cm long; 4 cm Ø) was provided into their new cage. This tunnel was subsequently used to transport them back and forth from their cage to the open field (92cmx92cmx50cm, Ugo Basile, Italy; center 15 lx and 45 ± 5 dB). Sets of objects (duplicates or triplicates), object positions and procedures were specific to each task based on previous experiments. During the preliminary phenotyping, mice received a 10-min habituation trial on the first day and two 10-min trials with a 5-min inter trial interval (ITI) on the second day. Whatever the task, acquisition trials lasted 10 min and retention trials lasted 8 min. In the long-term object recognition task, the mice first explored two identical objects on the first day, and were tested 24 h later with a novel object replacing one of the familiar objects. In the mnemonic discrimination task (also named behavioral pattern separation (PS) task), the mice began with two identical objects made of Lego bricks, then one familiar object was replaced by a novel object with a slightly different pattern of Lego brick rearrangement during the 5 min ITI [14]. In the OP task, there was two different objects during acquisition and then two copies of one of these objects after the 5-min ITI. For the easy tasks, the short-term object recognition task followed the same procedure than its long-term version with a shorter 5-min ITI. The object location task (OL) began with two identical objects, then one object was moved 55 cm apart during the 5-min ITI. In the OPc-Fos study, the whole procedure was adapted with a 3-h ITI and two 10-min trials to equalize duration of acquisition and retention trials before death. The mice were left in a dimly lit, quiet room (9 lx, 35 ± 5 dB) adjacent to the testing room for 90 min post-acquisition in the Acquisition groups, between acquisition and retention trials and again for 90 min post-retention in the Retention groups.

Object exploration was scored min per min during the whole retention session. However, retention performance was evaluated only during the first 6 min of exploration in all tasks, whereas only the first 4 min were considered in the OP task. In previous studies, maximal performance of C57/BL6J mice (dKI’s genetic background) was observed after 4–5 min before it faded away due to the tendency of the mice to equalize exploration of the two objects. A memory index (MI) calculated as follow:

$$\mathrm{M}\mathrm{e}\mathrm{m}\mathrm{o}\mathrm{r}\mathrm{y}\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{x}=\frac{\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{t}\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{l}\mathrm{a}\mathrm{c}\mathrm{e}\mathrm{d}\mathrm{o}\mathrm{b}\mathrm{j}\mathrm{e}\mathrm{c}\mathrm{t}-\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{t}\mathrm{u}\mathrm{n}\mathrm{c}\mathrm{h}\mathrm{a}\mathrm{n}\mathrm{g}\mathrm{e}\mathrm{d}\mathrm{o}\mathrm{b}\mathrm{j}\mathrm{e}\mathrm{c}\mathrm{t}}{\mathrm{T}\mathrm{i}\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{t}\mathrm{b}\mathrm{o}\mathrm{t}\mathrm{h}\mathrm{o}\mathrm{b}\mathrm{j}\mathrm{e}\mathrm{c}\mathrm{t}\mathrm{s}}-.5$$

The MI measures the proportion of time exploring the novel object, where 0 indicates chance level of exploration (hence the “- 0.5” part of the equation). Therefore, a MI of 0.5 indicates pure exploration of novelty, -0.5 pure exploration of the fixed object and 0 equal exploration of both objects hence chance level. Bevioural results were analyzed using a two sample t-test (WT vs dKI) and a one sample t-test (against chance level 0).

Western blot and Elisa assays

Western blot and Elisa experiments were made on a new cohort of dKI and WT male mice (n = 8/group). Entire left and right hippocampi and bilateral portions of the MTC (encompassing entorhinal, perirhinal and postrhinal cortices) were snap frozen in liquid nitrogen before storage at − 80 °C. Tissues were homogenized in ice-cold radioimmunoprecipitation assay buffer containing protease inhibitor cocktail, phenylmethylsulfonyl fluoride (Sigma-Aldrich), and phosphatase inhibitor cocktail (PhosStop, Roche Life Science). After centrifugation, supernatants were aliquoted and protein concentration was measured using the Bio-Rad Protein Assay. Brains extracts of Tg2576 and Thy-tau22 mice [71] were prepared in parallel as a positive control. For each mouse, one hippocampus and one subsample of the MTC portions were used for immunoblot analysis and the other ones for ELISA.

Western blot. For immunoblot analysis, 30 µg (APP) or 20 µg (tau) were loaded on 4–20 % precast gel (Tris-tricine 15–16 % precast gel: Mini-Protean TGX precast gels, Bio-Rad). After electrophoresis and transfer to nitrocellulose membranes (trans-Blot Turbo System, Bio-Rad), these were incubated first with 5 % skimmed milk (1 h, room temperature), then with primary antibodies overnight (2 % bovine serum albumin, tris-buffered saline 0.05 % Tween 20 (Sigma-Aldrich), 4 °C). After washes, membranes were incubated with horseradish peroxidase conjugated anti-mouse or anti-rabbit immunoglobulins (Jackson Immunoresearch) and enhanced ECL chemiluminescence detection kit (Thermo Fisher Scientific). They were finally re-probed with anti-actin antibody. Primary antibodies were rabbit polyclonal anti-APP C terminus (Sigma-Aldrich), rabbit polyclonal anti-tau (B19, generously gifted by JP Brion, ULB, Belgium), mouse monoclonal anti-phopho-Tau Thr181 (AT270, ThermoScientific), mouse monoclonal anti-actin (Sigma-Aldrich) and rabbit polyclonal anti-actin (Sigma-Aldrich). Secondary antibodies were peroxidase-conjugated AffiniPure goat anti-mouse and goat anti-rabbit (Jackson Immunoresearch). Band intensity (ChemiDoc Imaging system, Bio-Rad) quantified by densitometry analysis (ImageJ) were used to calculate ratios of total Thr181phosphorylated tau protein to total tau protein and total APP β-CTF to α-CTF APP-cleaved fragments for each mouse. Data were analyzed using a two sample t-test.

Sandwich ELISA. Total levels of human and mouse Aβ40 were determined by using the Human Aβ (1–40) assay kits which recognized with the same specificity the Aβ (1–40) in both species. Total levels of human and mouse Aβ42 were determined by using the Human Amyloidβ (1–42) (FL) assay kit for dKI mice and the Mouse/Rat Amyloidβ (1–42) assay kit for WT mice. All kits were purchased from IBL International and used following the manufacturer instructions. All dosages were done in duplicate and the signal was normalized to the protein concentration for each sample. Data were analyzed using a two sample t-test.

Mice perfusion and immunohistochemistry

Deeply anesthetized mice (sodium pentobarbital, 105 mg/kg intraperitoneally) were transcardially perfused with 0.1 % heparin 0.1 M phosphate-buffered saline (PBS) followed by 4 % paraformaldehyde (PFA) in phosphate buffer (PB; 0.1 M, pH7.4, 4 °C). Brains were postfixed 24 h in PFA, cryoprotected 48 h in 20 % saccharose PB, and stored at − 80 °C. Forty µm coronal cryostat sections were collected every 4th section for the mPFC, CLA, DH, MEC, and POC and every 6th section for the RSC, LEC, and PRC (Table 1). Sections were processed as follow at room temperature. Three PBS washes followed by a 30-min incubation in 1 % H₂O₂, an ultra-pure water wash and three more in PBS. A 45-min blocking incubation was done in 5 % natal goat serum (NGS) diluted in PBS-0.5 % triton, followed by a 2-day incubation with rabbit-anti-cFos (1/15000 in 2 % NGS, Synaptic Systems), two PBS washes and a 2-h incubation with biotinylated mouse-anti-rabbit (1/500 in 2 % NGS, Vector Laboratories) interrupted with two PBS washes. Finally, a 45-min incubation in the avidin/biotin solution (Vector Laboratories) was followed by three PBS washes and a last PB wash. Sections were revealed with a 10-min incubation in 3,3‐diaminobenzidine (Vector Laboratories). Images of whole sections were taken at 20x magnification using a Hamamatsu NanoZoomer S60 digital slide scanner (Hamamatsu Photonics K.K., Japan).

Table 1.

List of regions of interest (ROIs). ROIs were grouped in seven areas of interest: medial prefrontal cortex (mPFC, violet), claustrum (CLA, blue), dorsal hippocampus (DH, pink), retrosplenial cortex (RSC, red) and three areas for the medial temporal cortex (MTC, yellow), peri/postrhinal cortex, LEC and MEC. Color codes are used for community graphs in Fig. 4.

Region of interest	Acronym
Prelimbic cortex	PRL
Cingulate cortex	CG1
Infralimbic cortex	IL
Rostral claustrum	rCLA
Caudal claustrum	cCLA
Corpus ammoniss 1	CA1
Corpus ammoniss 3	CA3
Dentate gyrus suprapyramidal blade	DG_sp
Dentate gyrus infrapyramidal blade	DG_ip
Rostral retrosplenial cortex dysgranular (RSD)	rRSCd
Rostral retrosplenial cortex granular (RSG)	rRSCg
Caudal retrosplenial cortex dysgranular (RSD)	cRSCd
Caudal retrosplenial cortex granular (RSG)	cRSCg
Rostral perirhinal cortex dorsal (Ect)	rPRCd
Rostral perirhinal cortex ventral (PRh)	rPRCv
Caudal perirhinal cortex dorsal (Ect)	cPRCd
Caudal perirhinal cortex ventral (PRh)	cPRCv
Postrhinal cortex (Ect,PRh)	POC
Rostral lateral entorhinal cortex (DLEnt,DIEnt,VIent)	rLEC
Caudal lateral entorhinal cortex (DLEnt,DIEnt,VIent)	cLEC
Medial entorhinal cortex (MEnt)	MEC
Caudal medial entorhinal cortex (CEnt)	CMEC

Ex-vivo c-Fos imaging

The expression of c-Fos was quantified in 22 regions of interest (ROIs) (see Table 1) anatomically defined according to Franklin and Paxinos [23]. These regions were chosen a priori based on their relevance to early AD pathology and associative memory processing. Images were transformed into 8-bit grayscale and a grayscale threshold was set at a consistent level for each region by an experimenter blind to group condition (ImageJ, National Institute of Health, Bethesda, MD). Only c-Fos positive nuclei with a grayscale intensity above the threshold and an area between 25 and 300 µm² were counted over. At least three brain sections were processed per ROI. As variance strongly varied between regions, violating one of the requirement of ANOVA, mean c-Fos density was calculated for each ROI as the quantity of c-Fos marked nuclei per mm² normalized to the WT-Acquisition group. ROIs were grouped into seven areas anatomically and functionally justified to reduce the number of comparisons and, thereby, restricted Type I errors. Three-way ANOVAs compared test-phases (acquisition or retention), genotypes (WT or dKI) and regions for each area. Simple effects were examined with a Tukey post-hoc whenever an interaction was significant.

From functional connectivity to functional networks

Functional connectivity (FC) was assessed for each Genotype-Phase group by computing between subject inter-regional spearman correlations. Correlation matrices were elaborated to visualize all possible pairwise inter-regional correlations. To assess global FC strength the mean r was calculated with retained, near-zeroed, and absolute valued negative correlations respectively. In the near-zeroed case, negative correlations were reduced to a value of 0.006, the smallest positive correlation observed.

From each correlation matrix, a functional network was generated as a fully connected weighted graph with edges weights reflecting inter-regional Spearman correlation strengths and nodes reflecting regions. Various algorithm variants exist to handle negative correlations and there is no obvious criterion to choose one variant over others. As negative correlations are weak and pose difficulties of interpretation when dealing with many graph theory techniques, the near-zeroed value option was retained to construct our functional networks [83]. Graph construction and graph analysis were done through the igraph [18] package on R R Core Team, 2017 [59]. For all network metrics, confidence intervals were computed through bootstrapping. This involves resampling subjects with replacement 1000 times, each time regenerating a functional network, then recalculating the estimate of interest. The 95 % quartile of the bootstrap distribution was taken as the 95 % confidence interval. Confidence intervals for the difference were used to test between genotype differences [92]. Groups were considered different to a p < 0.05 if the 95 % confidence interval for the difference ≥ 0, and to a p < 0.01 if the 99 % confidence interval for the difference ≥ 0.

Community analysis

How well can a network allow information to flow along network links will depend on how wide individual links are, and also on how links are disposed and aligned to form pipelines between the nodes that must communicate, without too many steps and bottlenecks. Networks with high modularity, Q, have strong connections between nodes within communities and relatively weaker connections between nodes of different communities.

$$Q=\frac{1}{2m}\sum _{i\ne j\in N}\left[w_{\mathit{ij}}-\frac{s_i{s}_j}{2m}\right]\delta \left(c_i,c_j\right)$$

where N denotes the set of all nodes, $m$ denotes the total number of edges, $w(i,j)$ denotes the edge weight between a node $i$ and another node $j$, $s$ denotes the sum of a node’s edge weights, $c$ denotes the community to which a node belongs, and $\delta (c_i,c_j)$ indicates if the compared nodes are in the same community ($\delta (c_i,c_j)$ is 1 if $c_i=c_j$, and 0 otherwise). Communities were detected in each bootstrap through finding the maximum modularity across all possible community partitions. This computation was done through the cluster_optimal function of the igraph package [10]. This computationally expensive detection method of evaluating all possible community partitions for modularity maximization was feasible due to the relatively small number of nodes in our network. Allegiance matrices were used to assess community stability across bootstraps, by depicting the percentage of bootstraps (n = 1000) that contain any given pair of regions within the same community.

Information flow

Nodal strength measures the degree to which each specific region can exchange information directly with all other regions of the network, and indirect communication can be measured using nodal efficiency. Nodal strength, $\mathit{s}$, is traditionally calculated as the sum of a node’s edge weights. In a fully connected network this is directly proportional to the average of a node’s edge weights. The average,${\mathit{s}}_{\mathit{a}\mathit{v}\mathit{e}\mathit{r}\mathit{a}\mathit{g}\mathit{e}},$ was used for better comparison with nodal efficiency.

$$s_{\mathit{average}}\left(i\right)=\frac{1}{n-1}\sum _{j\in N}{w}_{\mathit{ij}}$$

where N denotes the set of all nodes, $n$ denotes the total number of nodes, and $w_{\mathit{ij}}$ denotes the edge weight between a node $i$ and another node $j$.

For efficiency metrics, edge lengths were first computed as inverted edge weights. Nodal efficiency, ${\mathit{E}}_{\mathit{nodal}}$, was calculated as the average inverse shortest path length between a region and all other regions of the network.

$$E_{\mathit{nodal}}\left(i\right)=\frac{1}{n-1}\sum _{j\in N}\frac{1}{d_{\mathit{ij}}}$$

where $d_{\mathit{ij}}$ denotes the length of the shortest path (lowest sum of edge lengths) between a node $i$ and another node $j$. Global efficiency, ${\mathit{E}}_{\mathit{global}},$ was calculated as the average inverse shortest path length of the network.

$$E_{\mathit{global}}=\frac{1}{n\left(n-1\right)}\sum _{i\ne j\in N}\frac{1}{d_{\mathit{ij}}}$$

Results

Behavioural phenotyping reveals early alterations in object-place associative memory

A preliminary experiment was conducted to detect which form of recognition memory was first affected in young dKI male mice. Different cohorts of WT and dKI were tested in a battery of high demanding memory tasks at 2, 4, and 6 months of age (Fig. 1). All groups succeeded in the long-term object recognition task (Fig. 1A). A potential deficit in the PS task remained inconclusive until the age of 6 month (two sample t-test WT vs dKI t(17) = 2.364, p = 0.030; Fig.1B). Note that performances of dKI mice were highly variable and never differed from chance (one sample t-test vs chance for each dKI group t < 0.996, p > 0.055). As shown in Fig. 1C, a robust deficit in the OP associative memory task was found in 4- and 6-month old dKI mice (WT vs dKI 4mo t(20) = 2.325, p = 0.0313; vs chance dKI t(8) = 0.2086, p = 0.8399; WT vs dKI 6mo t(17) = 2.132, p = 0.0479; vs chance dKI t(10) = 0.7196, p = 0.4883). An independent cohort of 4-month old mice confirmed this impairment (WT vs dKI t(20) = 2.85, p = 0.0098; vs chance dKI t(10) = 0.3095, p = 0.763; Fig. 1D). Both genotypes showed intact performance in a short-term novel object recognition task (WT vs dKI t(19) = 0.0934, p = 0.9265; vs chance: t > 3.94, p < 0.003; Fig. 1E) and an object location tasks (WT vs dKI t(20) = 1.052, p = 0.9173; vs chance t > 3.03, ps < 0.013; Fig. 1F). Thus, the OP deficit of dKI mice was due to specific perturbations in associative memory rather than any single impairment in object recognition or spatial recognition.

Fig. 1.

Behavioral and neuropathological characterization of dKI mice. Upper graphs A,B,C: phenotyping of independent cohorts of male mice at 2 months (WT n = 10, dKI n = 10), 4 months (WT n = 12, dKI n = 9) and at 6 months (WT n = 8, dKI n = 11). Two month-old WT (n = 3) and dKI (n = 1) mice were added to confirm WT’s efficiency in object-place (OP) memory. (A) Both genotypes succeeded in the long-term object recognition task (NOR). dKI and WT groups differed (B) at 6 months in the behavioral pattern separation task and (C) at 4 and 6 months in the OP task. Middle graphs D,E,F: A separate cohort (n = 11/genotype) of 4-month old male mice (D) confirmed the OP deficit and succeeded in short-term (E) object recognition and (F) object location. Lower graphs G, H, I and J: Hippocampal and medial temporal cortex (MTC) neuropathology was investigated in a separate cohort of 4-month old male dKI and WT mice (n = 8/genotype). (G-H) In dKI mice, the pTau Thr181/total tau ratio increased only in the MTC, whereas the β-CTF/α-CTF ratio increased in both hippocampus and MTC. (I) Representative blots for total pTau Thr181 and tau proteins, and APP-cleaved fragments. (J) In the same cohort, Aβ Elisa analyses showed that the Aβ42/Aβ40 ratio of dKI mice also increased in these two regions. Data presented as mean (±SEM). *p < 0.05, ** p < 0.01, *** p < 0.001 and **** p < 0.0001 genotype difference; #p < 0.05 different from chance.

Tau pathology in the medial temporal cortex, but not the hippocampus

We investigated early AD pathophysiological changes in 4-month-old WT and dKI male mice using western blotting (Fig. 1G-J). Although WT and dKI mice expressed similar levels of total tau proteins (Fig. 1 G-H left), the level of Thr181 phosphorylated tau and the proportion of Thr181 phosphorylated tau over total tau was moderately, but significantly increased in dKI mice only in the MTC (WT vs dKI t(14) = 2.818, p = 0.0137). The β-CTF/α-CTF ratio was higher in dKI mice than in WT mice in both the hippocampus and the MTC (Fig. 1 G-H right: WT vs dKI hippocampus t(14) = 6.991, p < 0.0001; WT vs dKI MTC t(20) = 4.004, p = 0.0013, respectively; Fig. 1H). To further confirm amyloidogenic processing of APP in dKI mice, Elisa quantification of the Aβ40 and Aβ42 species was performed on the same cohort. In both the hippocampus and the MTC, Aβ40 levels were decreased while Aβ42 levels were increased in dKI mice compared to WT mice (supplementary Fig. 1, respectively; WT vs dKI: hippocampus Aβ40 t(14) = 8.087, p < 0.0001; hippocampus Aβ42 t(14) = 5.293, p = 0.0001; MTC Aβ40 t(14) = 3.298, p = 0.0053; MTC Aβ42 t(14) = 4.648, p = 0.0004). As shown in Fig. 1J, this led to increased Aβ42/Aβ40 ratio in both regions of the dKI mice (WT vs dKI: hippocampus t(14) = 10.510, p < 0.0001 and MTC t(14) = 5.662, p < 0.0001). Thus, OP deficits occur in conjunction with an increase in Aβ42 production and subtle changes in tau phosphorylation state, akin to an early preclinical AD stage with an initial tau pathology restricted to the MTC [37].

Increased c-Fos expression in the claustrum and entorhinal cortex of dKI mice

We next evaluated encoding versus recall dependent c-Fos expression changes in WT and dKI mice during the acquisition phase or during the retention phase (Fig. 2A) of the OP task. The ITI was extended to 3 h to maximize isolation of specific c-Fos activation for each phase. Again, dKI mice were unable to detect the novel OP association (two sample t-test WT vs dKI t(26) = 3.14, p = 0.004; one sample t-test vs chance: dKI t(13) = 0.139, p = 0.891 and WT t(13) = 3.93, p = 0.002; Fig. 2B). The c-Fos expression was then evaluated in 22 regions of interest (ROIs; see Table 1) grouped as parts of the mPFC, CLA, DH, LEC, RSC, MEC and PRC/POC areas (Fig. 2D-J). We show examples of WT and dKI c-Fos immunostaining (Fig. 2C), c-Fos detections (Supplementary Fig. 2) and a summary table with all c-Fos densities (Supplementary Table 1). We first assessed test-phase and genotype dependent changes in c-Fos counts normalized to the WT-Acquisition group (see Methods). Regardless of genotype, all areas showed higher levels of c-Fos expression in the acquisition phase (LEC phase effect: F(1, 96) = 5.22, p = 0.025; all other ROIs’ F > 25.6, p < 0.001), except the DH (phase effect: F(1, 192) = 1.85, p = 0.175; region × phase effect: F(3, 192) = 5.32, p = 0.002; post-hoc Tukey: p ≥ 0.052). It is noteworthy that dKI mice showed increased c-Fos expression in the CLA and the LEC regardless of the test phase (genotype effect: CLA F(1, 96) = 8.29, p = 0.005; LEC F(1, 96) = 10.36, p = 0.001, respectively). In the MEC, an increased c-Fos expression was only observed during the acquisition phase (genotype effect: F(1, 96) = 25.64, p < 0.001; genotype × phase effect: F(1, 96) = 7.46, p = 0.007; post-hoc Tukey: WT vs dKI *** p < 0.001).

Fig. 2.

Regional c-Fos activation during both phases of the object-place (OP) task. WT and dKI mice were tested in (A) the acquisition phase only (above, n = 12/group) or the entire task (below, n = 14/group) before being perfused to process their brain for c-Fos immunohistochemistry. (B) dKI mice tested with a 3 h-ITI confirmed the OP deficit (WT vs dKI, ** p < 0.01; vs chance, # p < 0.05). (C) Representative microphotographs of a whole section with a higher magnification hippocampal-prefrontal cortex inset for c-Fos immunostaining in WT and dKI mice (scale bars at 500 µm). (D-J) Graphs for c-Fos counts normalized to the WT-Acquisition group. For each area a 3-factor ANOVA was performed for test-phase (Acquisition-Lines, Retention-Grid), genotype (WT-green, dKI-purple) and ROI effects (test-phase and genotype effects; *p < 0.05, **p < 0.01, ***p < 0.001. (D-E,G-J) All areas in both genotypes showed increased in c-Fos activation during the acquisition phase compared to the retention phase, except in (F) the dorsal hippocampus. In dKI mice, c-Fos activity was higher within (E) the claustrum and (G) the lateral entorhinal cortex during both test-phase, and (I) within their medial entorhinal cortex only during the acquisition phase. Data presented as mean (±SEM). ROI prefixes – r,rostral;c,caudal; suffixes – d,dysgranular; g, granular; d, dorsal; v, ventral; ip, infrapyramidal; sp, suprapyramidal. Central schematics indicate locations of the 22 ROIs adapted from Allen Mouse Brain Atlas derived vector images [40]. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Computing functional couplings

From the c-Fos signal, we assessed FC by computing the inter-regional Spearman correlation coefficients (r) for each Genotype-Phase group. Correlation matrices were used to visualize all possible correlations within each group (Fig. 3A-B). We first assessed global FC strength by taking the mean r value of each matrix. Most FC couplings were positive, however we found a few weakly anti-correlated pairs of regions (supplemental Fig. 3).

Fig. 3.

Functional connectivity represented with correlation matrices. These matrices were based on inter-regional Spearman correlations for c-Fos expression during (A) acquisition (n = 12/genotype) and (B) retention (n = 14/genotype). Colors reflect correlation strength (scale, right). Global. (C) During acquisition There was no change in global FC strength whether it was assessed by taking the mean r value of each matrix with retained, near-zeroed (reduced to 0.006), and absolute valued negative correlations. (D) During retention however, there was a significant decrease in global FC strength with original and near-zeroed correlations, but not with absolute values. Bar graphs represent the mean bootstrap value, and the error bars represent the bootstrapped 95 % confidence interval. * -the 95 % CI for the difference ≥ 0.

During acquisition, we found no significant change in global FC strength between WT and dKI groups, for any of three ways to treat negative correlations in FC (Fig. 3C). During retention, there was a decrease in global FC strength in dKI mice with respect to WT. This decrease was significant when negative correlations were considered as disruptive or very weak, but not when their absolute value was taken (Fig. 3D). This indicates that two phenomena coexist: first, a reduction of positive inter-regional correlations, corresponding to decreased “cooperation” between some regions; second, an increase in absolute strength of inter-regional negative correlations, corresponding to increased “conflict” between some regions.

The dKI networks reveal consistent departures from WT community structure

Each functional connectivity matrix can be considered as the adjacency matrix of a weighted undirected network [65], and, as such, its organization can be assessed through the analysis of functional networks, using techniques arising from graph theory. From each matrix a functional network was generated as a fully connected weighted graph. Allegiance matrices were then computed to extract a robust consensus set of communities [6],Fig. 4).

Fig. 4.

Functional networks computed as fully connected weighted graphs. Regions are represented as nodes (circles) and inter-regional correlation strengths as edge weights (line thickness, correlations of r < 0.2 are not shown)). For visual examination of networks, nodes were placed so that they lie closer to nodes with which they are strongly connected (Fruchterman-Reingold layout algorithm), and were color coded according to a priori areas (Table 1). (A,B) In both WT groups there was a stable MTC/cRSC community (red area at the bottom right, 4A,B). (A) WT Acquisition also showed a stable DH/mPFC community (large red area at the top left). (B) In WT Retention, the mPFC shares community allegiance with the whole network (red area across the top), especially through the RSC (middle of the top). (C,D) Across both dKI groups, the DH regions were disengaged from frontal regions (mPFC: green area top left, 4C; mPFC/CLA orange square top left AD, 4D). The MEC/CMEC/POC regions displayed a consistently modified community allegiance in respect to the WT, associating more with DH/mPFC (red/orange areas at upper right) and sometimes less with the rest of the MTC (more orange/green areas at lower right). Acquisition: n = 12/genotype; retention: n = 14/genotype. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

A stable community was found in both task phases encompassing most of the MTC and the cRSC (Fig. 4A,B). The main difference between WT acquisition and retention groups was the differential involvement of the DH or mPFC with this MTC community. During acquisition, the DH shared a community primarily with the mPFC (Fig. 4A). During retention, the mPFC shared allegiance across the whole network (Fig. 4B). This global incorporation of the mPFC appeared to be driven largely through the RSC (Fig. 4B).

During acquisition, the DH community of dKI mice largely disengaged with the mPFC as compared to the WT-acquisition group (Fig. 4C). Moreover, the DH appeared to be more strongly aligned to the CMEC, MEC, and POC as compared to the rest of the MTC. During retention, both the mPFC and the DH disengaged with the MTC. This led to distinct communities containing mPFC/CLA, DH and most of the MTC (Fig. 4D). Similarly to the acquisition phase, the MEC/CMEC/POC regions were heavily recruited by the mPFC and DH communities at the cost of disengagement with the rest of the MTC community. The consistent recruitment of the MEC/CMEC/POC in both phases suggests a role for these regions as compensatory hubs in dKI mice.

During retention, dKI mice show severe losses in network efficiency

The general capacity for a network to sustain efficient flows is quantified in graph theory by metrics such as global efficiency [39]. It is important to note that even if a network has reduced global FC strength, its global efficiency may rest un-affected if there are well placed “hub” regions to facilitate indirect communication. To evaluate information flow at the regional level, nodal strength and nodal efficiency metrics were used. Regions with higher strength and efficiency can be considered to contribute more to their network. High direct connectivity of a region, such as high strength, is also shown to describe potential hub regions [83] which, as we have described, are needed to facilitate indirect information flow and maintain global efficiency.

During acquisition, strength and efficiency distributions were largely conserved between genotypes with strong emphasis on RSC and MTC regions (Fig. 5A,B). However, in the dKI mice there appeared to be additional involvement of mPFC, MEC and POC. Nevertheless, global efficiency was not severely affected in the dKI network (Fig. 5C). The unique significant drop in nodal strength was seen in the suprapyramidal blade of the DG (DG_sp) of dKI mice (Fig. 5D). As there was no drop in nodal efficiency (Fig. 5E), their DG_sp could still effectively communicate with the rest of the network indirectly presumably through the rRSC or MEC/CMEC/POC compensatory hubs.

Fig. 5.

Network organization of information flow. Node strength and nodal efficiency distributions was examined in WT and dKI mice for both test-phases (acquisition: n = 12/genotype; retention: n = 14/genotype). These distributions were visualized using necklace diagrams, where circle size reflects the within network normalized (A,F) strength² and (B,G) efficiency² normalized to the highest square value for each network. Global efficiency was then compared directly between genotypes to assess (C,H) network integration. (D,I) Nodal strength and (E,J) nodal efficiency were compared between genotypes to assess region dependent changes in direct and indirect information flow, respectively. Bar graphs and dot graphs represent the mean bootstrap value, and the error bars represent the bootstrapped 95 % confidence interval. * -the 95 % CI for the difference ≥ 0; ** -the 99 % CI for the difference ≥ 0.

During retention, heavy emphasis was placed on mPFC and RSC regions in the WT network (Fig. 5F,G). In the dKI network, this emphasis was largely lost and the distributions took on a more homogenous structure, though the rRSCg and mPFC subregions still appeared to be among the most involved. Global efficiency was significantly reduced in the dKI network (Fig. 5H), with severe drops in strength seen in both RSCd subregions, the PRL and the POC (Fig. 5I). Loss in nodal efficiency was even more prevalent, with additional sharp reductions in the CG1 and the cRSCg (Fig. 5J). Thus, retention phase dependent functional integration across the cingulate cortex appears severely disrupted in the dKI, an effect that could be linked to reduced hub strength of the RSCd.

Discussion

Our interest in initial stages of AD pathology prompted us to investigate task-activated brain network alterations in the new App^NL-F/MAPT dKI mouse model at the age of onset of the earliest deficit detected in a battery of task taxing three aspects of recognition memory presumably affected in the preclinical stage of AD. At the age of 4 months, when the first memory deficit appeared in the OP association task, markers of increased amyloid processing of APP were detected in both the hippocampus and the MTC of dKI mice, while an early sign of tau pathology was found in the MTC, but not in the hippocampus. The dKI mice showed an abnormal increase in c-Fos activation in MEC regions during the acquisition phase, and in LEC and claustrum regions during both phases of the task. An extensive analyses of network connectivity showed that internal communication between and within MTL regions is disorganized in dKI mice. In addition, reduced information flow between the MTL and interconnected regions was associated with a disruption of the pivotal retrospenial “hub”.

Object-place recognition deficit as an early marker of emerging AD neuropathology

Performance deficits of dKI mice in the OP task appear quite specific to the associative nature of the OP task because the same animals succeeded in tasks taxing single spatial or object modality recognition. This impairment was associated to a shift of APP processing towards the production of β-CTF and Aβ in both the dorsal hippocampus and the MTC, as expected in mice bearing a Swedish (NL) ADAD mutation favoring β-secretase cleavage of APP. However, only the MTC showed an increased phosphorylation of tau at Thr181, a preclinical marker of AD [80]. Other mouse models of AD also show early impairments of OP associative memory [[8], [29]]. In pre-plaque TgCRND8 mice, the emergence of OP deficits was specifically associated with the overproduction of β-CTF, but not Aβ, [29]. Thus, the abnormal level of β-CTF in the MTL, eventually combined with an increased proportion of p-tau Thr181 in the MTC, might have been sufficient to contribute to OP deficits in dKI mice. Within the MTC, the LEC is known to play a key role in OP associative memory in rodents [[91], [15], [38]]. In the elderly, reduced volume of its human homologue, the anterolateral EC, is correlated with performance in an OP task [94]. In the dKI mice, the LEC was the only MTL area showing an increased c-Fos signal during both phases of OP testing which might correspond to sustained neural activity and/or abnormal neural plasticity. The poor performance of dKI mice could be explained by a disruptive effect of LEC dysfunction on brain networks supporting OP associative memory. Thus, the cognitive and neuropathological state of 4-month old dKI mice is in line with the subtle OP deficits associated with early signs of AD in elderly [[30], [94]] and the general worsening of neuropsychological markers of EC and PRC functions detected>10 years before AD diagnosis [33]. The PS deficit was detected at the age of 6 months. As their pathology progresses, dKI mice show a mnemonic discrimination impairment similar to the one reported in normal ageing C57BL/6J mice [14]. In rodent models as well as in elderly and aMCI patients, PS deficit has been attributed to a functional disconnection between the EC and the hippocampus [[4], [11], [60]]. In light of these findings, the mnemonic discrimination deficit appearing in 6-month old dKI mice may reflect the spreading of neuropathological alterations from the LEC to the hippocampus as observed in single App^NL-F KI mice [57]. However, a more comprehensive characterization of amyloid and tau pathologies progression is necessary to support this hypothesis. In addition, forms of memory other than mnemonic discrimination and OP association memory could be impaired in the dKI mice. Current research is in progress to test other forms of recognition memory, such as episodic-like memory, and spatial navigation which are known to be affected in early stages of AD. All these findings strongly suggest that the dKI model recapitulate early impairment in subtle forms of recognition memory and specific vulnerability of the MTC to the sequential progression in AD pathology within the MTL as reported in preclinical stages of AD [37].

Distributed increase in c-Fos may reflect spreading of abnormal activity

Compared to their WT controls, dKI mice showed higher expression of c-Fos restricted to the acquisition phase within the MEC. The MEC is part of the network encoding the global spatial frameworks with minor influence of object features per se [[20], [81]]. Increased c-Fos activation in the MEC might be driven by abnormal neuronal activity associated with encoding of a new spatial configuration of the objects within the open-field. On the other hand, during the recall phase, novelty was limited to a non-spatial change with one object replacing another one. Alternatively, the MEC might be less prone to abnormal neural activity and/or plasticity than the LEC. Both the CLA and the LEC expressed high level of c-Fos regardless of test-phase. Well documented in mouse models of AD, early LEC hyperactivity within the LEC is triggered by local increase in amyloidogenic metabolites of APP and it is considered as a major factor driving propagation both amyloid and tau pathologies to its main outputs, e.g., the hippocampus [[93], [49], [63]]. With its reciprocal connections with prefrontal, cingulate and retrohippocampal areas, the CLA is thought to play a pivotal role in networks supporting cognitive functions [86]. As such, evidence for a dysfunctional CLA in the dKI mice is an original finding. In AD patients and models, only a few studies reported Aβ accumulation and neurodegeneration in this region [[58], [50], [28]]. Interestingly, an early study in Alzheimer's patients found neuronal loss restricted to a CLA subregion strongly connected with the EC [47]. Thus, our finding is coherent with the hypothesis of the EC exporting abnormal neural activity and neuropathology to the CLA and other limbic cortices in preclinical stages [9]. In addition, Avila and Perry [3] recently suggested that the CLA itself could also play a central role in spreading AD pathology in the brain.

Early-stage hyperactivity detected with fMRI has been proposed to reflect compensatory mechanisms which would first help maintaining and then worsen cognitive function through the spreading of neuropathological processes [17]. There is also in line with some indication for a reorganization of network information flow in the dKI mice. The community organization of functional networks during both phases show that the MEC and the POC are more integrated with mPFC and DH communities. This change in community allegiance may be a consequence of abnormal neural activity within the EC, much like artificial stimulation of brain structures can enforce specific functional networks [88]. In certain cases, this shift towards stronger outbound allegiance may help facilitate alternative communication between the MTC and interconnected regions, as seen with the maintenance of indirect information flow of the DG_sp during acquisition. The apparition of compensatory hubs may be one way through which hyperactivity could initially help memory processing during the preclinical AD.

PFC and RSC disrupted during object-place associative memory retention

The mPFC and RSC were the most heavily utilized regions in WT mice during the retention phase, but direct and indirect information flow through these regions was markedly reduced in dKI mice (Fig. 5F). RSC lesion is known to disrupt OP associative memory [53], and the mPFC-MTC communication was also shown to be essential [[15], [32]]. The RSC is likely to play a pivotal “hub” role in this communication through its strong structural connections with both the mPFC and the MTC [[82], [78], [45]]. In humans, the posterior cingulate cortex (PCC) and RSC, close equivalents to the rodent RSC [84], are also shown to support structural connectivity between the mPFC and MTL essential to associative memory networks [44]. In the community organization of WT mice, the hubness of the RSC during retrieval was supported by its ideal position to facilitate communication between the mPFC and MTL communities. The severe loss in strength of RSCd regions in dKI mice may therefore indicate an OP recall-dependent roadblock for the mPFC-MTL communication. Our results also corroborate those of human studies showing that mPFC and PCC/RSC regions are often disrupted when functional and structural connectivity are evaluated in aMCI patients, APOE ε4 carriers and presymptomatic ADAD patients [[12], [16], [85]]. Reduced connectivity in the PCC/RSC may constitute a typical FC marker of preclinical AD [36]. These regions are among the first to display amyloid accumulation [[35], [51]]. Moreover, initial increases in PCC amyloid deposition correlate with face-name associative memory deficits in subjective cognitive decline [68]. Emerging amyloid and tau neuropathologies in the mPFC/PCC/RSC and the MTC, may thus have a combined negative effect on FC in the cognitive network supporting associative memory recall.

Contrast with resting state fMRI in other mouse models of AD

Contrary to our results, studies evaluating the resting state networks of classical transgenic AD mouse models in their pre-plaque or early tau pathology stages have predominantly found cortical and hippocampal-cortical hyperconnectivity, while hypoconnectivity appears at later stages [2]. This discrepancy could be related to the type of model (transgenic versus KI) or the technique used to generate data (c-Fos expression versus haemodynamic response). Alternatively, it might also suggests that perturbations in resting state FC predict memory FC perturbations, but do not directly mirror them. Moreover, the mouse models used to evaluate pre-aggregate resting state FC presented either amyloid pathology [74] or tau pathology [19] but not both concurrently. Combined resting state fMRI and PET imaging in humans has shown that increased Aβ alone is associated with hyperconnectivity of the DMN while combined Aβ and Tau pathologies reveal hypoconnectivity [72]. Whether these contrasting results reflect differences between “resting state vs memory driven FC” or differences in “amyloid/tau pathological staging” will become more thoroughly understood once resting state fMRI will be performed in dKI mice. The identification of major hubs dysfunction in the present study will be useful for seed-based analyses of fMRI data. In addition, it would be interesting to test the validity of the hub dysfunctions identified in the present study using optogenetic or chemogenetic techniques in WT mice (blocking) or in dKI mice (activation) to induced OP deficits or alleviate OP impairments, respectively.

Conclusions

The dKI mouse model of AD was caught in a preclinical stage when AD-relevant recognition deficit just emerged and MTC tau pathology did not propagate yet to the hippocampus. Besides the EC dysfunctions associated to local neuropathology, evidence for abnormal CLA activity in this preclinical model should stimulate interest in its preclinical implication. The origin and functional consequences of this dysfunction needs to be further investigated using new molecular tools to dissect the functional contribution of the CLA’s inputs, especially the EC, as well as its impact on target regions, such as the mPFC, the DH and the RSC. The dKI model also recapitulates frontotemporal disconnection with a disrupted communication between cingulate areas and the MTL during the retention-phase. In addition, it would be interesting to test the validity of the hub dysfunctions identified in the present study using optogenetic or chemogenetic techniques in WT mice (blocking) or in dKI mice (activation) to induced OP deficits or alleviate OP impairments, respectively. Finally, the similarity between our findings in the dKI model and those reported in the earliest stages of the disease suggests that this model has a high potential for generating new discoveries on the earliest stages of AD.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: