Disrupted place cell remapping and impaired grid cells in knock-in model of Alzheimer’s disease

Summary

Patients with Alzheimer’s disease (AD) suffer from spatial memory impairment and wandering behavior, but brain circuit mechanisms causing such symptoms remain largely unclear. In healthy brains, spatially-tuned hippocampal place cells and entorhinal grid cells exhibit distinct spike patterns in different environments, a circuit function called “remapping.” We tested remapping in amyloid precursor protein knock-in (APP-KI) mice with impaired spatial memory. CA1 neurons, including place cells, showed disrupted remapping, although their spatial tuning was only mildly diminished. Medial entorhinal cortex (MEC) neurons severely lost their spatial tuning and grid cells were almost absent. Fast gamma oscillatory coupling between the MEC and CA1 was also impaired. Surprisingly, mild disruption of MEC grid cells emerged in younger APP-KI mice, although animals’ spatial memory and CA1 remapping remained intact. These results point to the remapping impairment in the hippocampus, possibly linked to grid cell disruption, as circuit mechanisms underlying spatial memory impairment in AD.

Graphical Abstract

In Brief

Place cells in the hippocampus exhibit distinct spike patterns in different environments, a circuit function called “remapping.” Jun et al show that remapping is disrupted in the APP knock-in mouse model, suggesting that remapping impairment is a circuit mechanism underlying spatial memory impairment in Alzheimer’s disease.

Introduction

Spatial memory impairment such as wandering behavior is one of the most troublesome symptoms in Alzheimer’s disease (AD), occurring in more than 60% of AD patients (Hope et al., 1994). Despite recent molecular and cellular findings in AD research, it is still largely unclear how deterioration of brain circuit function causes spatial memory loss in AD. In healthy brains of both humans and rodents, grid cells in the medial entorhinal cortex (MEC) and place cells in the hippocampus form a neural circuit that is critical for spatial memory and navigation (Scoville and Milner, 1957; O’Keefe and Dostrovsky, 1971; Morris et al., 1982; Ekstrom et al., 2003; Fyhn et al., 2004; Steffenach et al., 2005; Doeller et al., 2010). In this circuit, neurons in layers 2 and 3 of MEC, containing grid cells, send their axons to the CA1 region of the hippocampus that contains spatially-tuned place cells (Witter and Amaral, 2004; Moser et al., 2015; Nilssen et al., 2019). Grid cells and place cells have an ability to store distinct firing patterns corresponding to different environments, a function called “remapping” (Muller and Kubie, 1987; Fyhn et al., 2007; Alme et al., 2014; Kyle et al., 2015). Because remapping of place cells and grid cells can provide subjects with distinct combinatorial codes for multiple environments, remapping is thought to provide the circuit mechanism for the pattern separation of spatial memory (Colgin et al., 2008; Yassa and Stark, 2011). So far, the causal relationship between remapping and behavioral pattern separation is not clearly established in healthy subjects (but see (Jeffery et al., 2003)). No previous studies investigated remapping of place cells and grid cells in AD subjects.

In AD patients, pathophysiology starts with amyloid-β (Aβ) plaque deposition from the expression of APP, which may slowly lead to Tau pathology and deterioration of brain circuit functions and cellular degeneration in the course of more than 20 years (Sasaguri et al., 2017). The deterioration of circuit functions, as well as cellular degeneration occurring markedly in the entorhinal cortex and the hippocampus, would ultimately lead to memory impairment, with which patients are diagnosed with AD (Van Hoesen et al., 1991; Gomez-Isla et al., 1996; Khan et al., 2014). To replicate the slow and cumulative effect of pathological APP on brain circuit functions in mice, we recently generated a knock-in mouse model by manipulating the endogenous murine APP gene into the mutated human APP^NL-F-G gene (Saito et al., 2014). This homozygous knock-in manipulation causes Aβ deposition starting at 4-mo in the hippocampus and spatial memory impairment in a Y-maze task at 6-mo (Saito et al., 2014). Recent studies start to show that, in several transgenic AD models, transgenes were randomly inserted into the loci of endogenous genes, thus disrupting them (Tosh et al., 2017; Gamache et al., 2019). Therefore, pathophysiology observed in these transgenic AD models may not solely derive from the transgenic expression of APP or Tau, but from the “knock out” of endogenous genes unrelated to AD. These findings may indicate that phenotypes found in previous studies that used transgenic models may need to be revisited with another model that assures no endogenous genes are disrupted. In our knock-in model, the mutated APP gene is inserted into the mouse APP locus without disrupting other genes. Thus, pathophysiology in our APP-KI mice is more directly attributable to the mutated APP gene. Using APP-KI mice, we investigated the remapping of place cells and grid cells underlying the spatial memory impairments.

Results

Impaired spatial memory in old APP-KI mice

In this study, we specifically asked two questions: (1) Is remapping of hippocampal place cells deteriorated in AD subjects? (2) If so, is the place cell deterioration linked to impairment of entorhinal grid cells? Alternatively, does place cell deterioration occur autonomously inside the hippocampus and independent from grid cells? To this end, we investigated the remapping of hippocampal place cells and entorhinal grid cells in homozygous APP^NL-G-F mice (hereafter termed as APP-KI mice). These mice start to show Aβ deposition at 4-mo in the hippocampus and spatial memory impairment in a Y-maze task at 6-mo (Saito et al., 2014). Using this emergence of spatial memory impairment at 6-mo as a border, we defined a young period (3–5 mo, termed as “young APP-KI mice”) and an old period in APP-KI mice (7–13 mo, termed as “old APP-KI mice”).

We first tested if old APP-KI mice of 12–13 months of age show impairments in distinguishing spatially distinct environments, as in the case for AD patients’ spatial memory impairment. We used a context discrimination task, where mice were trained to distinguish two contexts having distinct sets of visual and olfactory sensory cues (Context A and Context B, Fig. 1A; see Methods) (McHugh et al., 2007). Because the APP-KI mouse is on the C57BL/6 background, we used 12-mo C57BL/6 mice as a control (referred to as WT mice). Mice were initially trained to induce a fear response in Context A for three days where a mild electrical foot shock was given 192 seconds after animals were brought into the box. (Fig. 1A, n = 8 WT mice and n = 8 APP-KI mice). Mice were then familiarized with another context, Context B, for two days. During the familiarization, there were no foot shocks. Mice were then trained with Context A and Context B for 5 days. During the 5-day training, both WT and APP-KI mice learned to freeze for more than 50% of time during the initial 3 min exposure to Context A, where a foot shock was again given 192 seconds after the exposure (Fig. 1B). By contrast, across the 5-day training, WT mice did not develop freezing in Context B where no foot shock was given (Fig. 1C). APP-KI mice, however, developed freezing in Context B to a similar extent as observed in Context A, presumably due to over-generalization between two contexts. To assess a degree of spatial discrimination between Context A and Context B, we defined Discrimination Index (DI) by calculating the difference of percent time frozen between Context A and Context B divided by a sum (Fig. 1D; see Method). The DI obtained from WT mice increased during the 5-day training, whereas DI did not improve in APP-KI mice. The DI of APP-KI mice was significantly lower between Day 3 to Day 5 compared to WT mice (p<0.01 or better, Fisher’s LSD post-hoc test after two-way ANOVA; p<0.001 for genotype comparison; see Supplementary Table S2 for statistics details hereafter). These results demonstrate that old APP-KI mice were impaired in context discrimination.

Figure 1. Impaired spatial memory and mildly diminished spatial tuning of hippocampal CA1 neurons in old (7–13 mo) APP-KI mice.

(A) A context discrimination task. Mice were trained to induce fear response in Context A with a foot shock (Fear Conditioning A), familiarized with Context B (Familiarization), and then trained with Context A with a foot shock and Context B without foot shock (Discrimination Training, see text). (B) Percent time frozen in Context A for WT and APP-KI mice during the 5-day discrimination training. (C) Percent time frozen in Context B. (D) Discrimination Index calculated from (A − B) / (A + B) for percent time frozen. (E) A 64-channel recording device was targeted to CA1. (F) Sagittal sections with anti-Aβ immunostaining from 12-mo WT and APP-KI mouse (left), percent area with Aβ plaques in CA1 (middle, n = 8 sections from two mice in each group), and a section with Nissl staining showing a representative recording position in CA1 (right, red arrow). A, anterior; P, posterior; D, dorsal; V, ventral. (G) 1×1 m open field for testing spatial tuning of CA1 neurons. (H) Eight representative CA1 neurons in WT mice. Top: Red dots and gray lines denote spike position and animal trace in the 1×1 m open field, respectively. Bottom: Firing rate map. Color is scaled with maximum firing rate (Hz) shown at top right of each rate map. Spatial information (SI) score is shown at top left. (I) Eight representative CA1 neurons as in (H), but from APP-KI mice. (J) Cumulative distribution plot for spatial information score of CA1 neurons in WT mice (n=105) and APP-KI mice (n=113). (K) Top, distribution of spatial information scores calculated from CA1 neurons (real data). Bottom, distribution of shuffled data. Red dashed line indicates 95th percentile value (chance level, 0.63). 50% of CA1 neurons in WT mice and 31% of CA1 neurons in APP-KI mice passed the chance level and were defined as place cells. (L) Left: Mean spatial information score of CA1 neurons in WT mice (n=105) and APP-KI mice (n=113). Right: Percentage of neurons defined as place cells. See Table S2 for statistics details.

Hippocampal CA1 neurons showed severely disrupted remapping, but mildly diminished spatial tuning in old APP-KI mice.

Next, we investigated place cells in old APP-KI mice at 7–13 months of age. We implanted a recording device containing 64 channels of electrodes in APP-KI mice and recorded place cells (Fig. 1E–F and S1; no change observed within this age period, Fig. S2C). No difference in animals’ running speed was observed between APP-KI and WT mice (Fig. S1C). In this paper we took an “analyze-all-the-cells” approach because we focused on addressing how the whole population of neurons becomes deteriorated in APP-KI mice. To avoid biased sampling toward spatially tuned cells that can be fewer in APP-KI mice, we collected all recorded principal neurons (termed as “CA1 neurons” hereafter) in our data pool (n=105 neurons from n = 10 APP-KI mice; n=113 neurons from n = 10 WT mice; Supplementary Table S1). No difference was observed in the distribution of putative principal neurons and interneurons between WT and APP-KI mice (Fig. S1B). Neural activity of CA1 neurons was first recorded in a 1×1 m open field to evaluate their spatial tuning and place cell properties (Fig. 1G). A substantial amount of CA1 neurons in old WT mice showed spatially tuned spike activities (Fig. 1H). CA1 neurons in APP-KI mice also showed spatially tuned spikes, but exhibited less pronounced firing peaks (Fig. 1I). The degree of spatial tuning for individual CA1 neurons was quantitatively assessed using the spatial information score (Skaggs et al., 1993) (Fig. 1J–L). The analysis revealed that APP-KI mice had neurons with less spatial tuning (p<0.001), with lower mean spatial information than those from the CA1 neurons recorded in WT mice (p<0.001). However, mean firing rates, field size and firing field number of CA1 neurons did not differ between WT and APP-KI mice, suggesting the spatial tuning of CA1 neurons in APP-KI mice was only mildly deteriorated (Fig. S1D). We then defined place cells using the spatial information score (Langston et al., 2010). Neurons were classified as place cells if their spatial information score was larger than 0.63, which was the 95th percentile of a distribution of spatial information score based on shuffled data from 7–13 mo WT mice (Fig. 1J–K). This revealed that the proportion of place cells decreased from 50% in WT mice to 31% in APP-KI mice (p<0.001; Fig 1K–L). Among these place cells, we did not observe any differences in their spike properties recorded from the open field task between APP-KI and WT mice (Fig. S1E).

To investigate the remapping of CA1 neurons in APP-KI mice, we then trained animals to run back and forth in two 1m linear tracks (Tracks A and B) with distinct colors and textures (Fig. 2A, see experimental protocol). In a daily session, animals made 10 laps for each track in the following series: Track A, Track B, Track B and then back again in Track A. Similar to the emergence of behavioral spatial discrimination during the 5-day training (Fig. 1), CA1 neurons in WT mice showed remapping across Track A and Track B on the fifth day of the training, with distinct firing patterns between the two tracks (Fig. 2B, also known as “global remapping” (Leutgeb et al., 2005b)). However, with the same training condition, CA1 neurons in APP-KI mice showed unchanged firing patterns between Track A and Track B (Fig. 2C). Population vector (PVec) correlation analysis (Leutgeb et al., 2005b) was used to assess the degree of remapping between Track A and Track B (Fig. 2D–E). A PVec correlation of 0 indicates strong remapping, whereas a PVec correlation of 1 indicates identical firing patterns (that is, no remapping at all) (Leutgeb et al., 2005a). Conventionally, remapping has been assessed by calculating PVec using spatially tuned cells (such as place cells or grid cells). In this study, to capture the overall picture of the pathophysiology, we first used all recorded CA1 neurons to assess remapping in both WT and APP-KI mice (referred to as “population remapping” hereafter). PVec correlation analysis from all recorded CA1 neurons revealed that WT mice showed remapping with low PVec correlation between Track A and Track B (Fig. 2E). However, APP-KI mice exhibited PVec with significantly higher correlation, indicating that the remapping of CA1 neurons was disrupted in APP-KI mice (p<0.001, Fig. 2E). Mean PVec correlation in APP-KI mice was significantly higher than that of WT mice (p<0.001; Fig. 2F). We did not observe any difference in the field size, number of fields, firing rate or peak firing rate in the linear track recordings between WT and APP-KI mice (Fig. S2A). Disrupted remapping was observed independently of the direction of animals’ runs in the tracks (Fig. S2B). We also assessed the degree of remapping in individual neurons using the spatial correlation score between Track A and Track B (Leutgeb et al., 2005b), confirming that CA1 neurons showed disrupted remapping in APP-KI mice compared to that from WT mice (Fig. 2G). By contrast, activity patterns of CA1 neurons between the two recordings in Track A were similar, with no difference in PVec correlations between WT and APP-KI mice, indicating that stability of spatial tuning in CA1 neurons was spared in APP-KI mice (p>0.05, Fig. 2E; p>0.05, Fig. 2F).

Figure 2. Hippocampal CA1 neurons in old APP-KI mice showed disrupted remapping.

(A) Remapping was tested in 1-m linear Track A and Track B with distinct colors and textures. (B) Five representative CA1 neurons in WT mice recorded in Tracks A, B, B, and A. Red dots and gray lines at the top of each track denote spike position and animal trace in the 1 m linear tracks, respectively. Bottom color map denotes firing rate map. Color is scaled with maximum firing rate (Hz) shown at top right. Spatial correlation between Track A and Track B (SCab) is shown at top left. (C) Five representative CA1 neurons from APP-KI mice are shown as in (B). (D) The degree of remapping was assessed using population vector (PVec) correlation Track A and Track B (see text). (E–G) population remapping calculated using all recorded CA1 neurons. (E) Cumulative distribution plots for PVec correlation between Track A and Track B (left) and between two recordings in Track A (right). Lower PVec correlation of ~0 denotes stronger remapping (black arrow). (F) Left: Mean PVec correlation between Track A and Track B. Right: Mean PVec correlation between two recordings in Track A did not differ. (G) Cumulative distribution plots of spatial correlation between Track A and Track B (left) and between two recordings in Track A (right). (H) PVec correlation as in (E–F), but for CA1 neurons defined as place cells. (I) PVec correlation as in (E–F), but for CA1 neurons defined as non-spatial cells. See Table S2 for statistics details.

We then assessed remapping of place cells (Fig. 2H). PVec correlation between Tracks A and B confirms that neurons defined as place cells also have severe impairment in the remapping (Fig. 2H). Interestingly, the PVec correlation from two recordings in Track A were slightly lower in APP-KI mice (Fig. 2H, see Fig. S2D for spatial correlation), suggesting that the remapping impairment of place cells may be due to their reduced stability of place fields. In CA1 non-spatial cells, remapping observed in WT mice was also impaired, whereas the field stability remained intact (Fig. 2I).

Together, these results demonstrate that CA1 principal neurons from APP-KI mice exhibited disrupted remapping for different spatial environments, while the spatial tuning of CA1 neurons was only mildly deteriorated.

Diminished fast gamma oscillations, but not slow gamma oscillations in the hippocampal CA1 of old APP-KI mice

CA1 neurons in the hippocampus receive spatially-tuned information from hippocampal CA3 via the Schaeffer collateral as well as from the MEC via temporoammonic pathway (Witter and Amaral, 2004). We next asked whether these inputs contribute to disrupted remapping in CA1. We leveraged findings from previous healthy animal studies that identified two distinct types of gamma oscillations (Bragin et al., 1995; Colgin et al., 2009) (Fig. 3A). When fast gamma oscillations occur in the hippocampal CA1, the MEC also exhibits fast gamma oscillations (Colgin et al., 2009). The oscillations from these two brain regions are highly synchronized, suggesting that fast gamma oscillations support signal transfer from the MEC to CA1. By contrast, synchronization of slow gamma oscillations suggests signal transfer from CA3 to CA1 (Colgin et al., 2009). We thus analyzed gamma oscillations in CA1 of WT and APP-KI mice. Recordings in the open field, where animals have relatively constant running speed, were used for gamma oscillation analyses. In WT mice, we observed gamma episodes across the slow and fast gamma range (20 – 90 Hz; Fig. 3B). However, in APP-KI mice, gamma oscillations were observed only in low frequency ranges, and marginal gamma episodes were observed at a frequency range above 40 Hz. Comparison of oscillatory power normalized by gross 1–100 Hz frequency power (Deshmukh et al., 2010) showed reduction of power at 62 – 100 Hz range, and increase of power at 5 – 8 Hz (theta) and 14 – 20 Hz (theta harmonics) ranges in APP-KI mice (false discovery rate (FDR) corrected for 100 multiple comparisons at 1–100 Hz bins, q<0.05; Fig. 3C).

Figure 3. Fast gamma oscillations, but not slow gamma oscillations, were diminished in the hippocampal CA1 of old APP-KI mice.

(A) Fast and slow gamma oscillations that reflect inputs from MEC and CA3, respectively, were recorded in the CA1 during the open field task (see text). (B) A representative time-resolved spectrogram showing gamma oscillation episodes in a WT mouse (left) and APP-KI mouse (right). Corresponding raw local field potential traces are shown at the bottom. (C) Normalized power spectra from WT and APP-KI mice (n=10 mice each). Significant difference in power for each 1Hz bin is shown with green dot at the top (q<0.05, FDR corrected for multiple comparison). (D) Example cross-frequency coherence plots showing that oscillatory powers at 20–40 Hz and 40–80 Hz frequency (y-axis) were modulated by theta phase (x-axis) in a WT mouse (left), but the theta modulation of 40–80 Hz oscillatory power was absent in an APP-KI mouse (right). Dotted line denotes 40-Hz border between fast and slow gamma. (E) Occurrence rates of fast gamma episodes (top) and slow gamma episodes (bottom) for WT (n=10) and APP-KI mice (n=10). (F) Normalized power spectrograms averaged across all theta cycles in three representative WT mice (left) and APP-KI mice (right). Theta phase is shown at the bottom. Dotted line denotes 40-Hz border between fast and slow gammas. (G) Mean vector length for the distribution of fast gamma episodes (top) and slow gamma episodes (bottom) across theta phase. (H) Fast gamma phase distribution of spikes from example three CA1 neurons from WT mice (left) and APP-KI mice (right). Schematics of two cycles of gamma waveform are shown in the bottom. (I) Mean vector length for the distribution of spikes across fast gamma phase (top) and slow gamma phase (bottom) for WT CA1 neurons (n=104 cells) and APP-KI CA1 neurons (n=105 cells). See Table S2 for statistics details.

To test whether fast and slow gamma oscillations are deteriorated in CA1, we investigated temporal structures of gamma oscillations. Previous healthy animal studies showed that gamma oscillations occur mainly at the peak to falling phase of co-existing theta oscillations, a property termed theta-gamma coupling (Bragin et al., 1995; Colgin et al., 2009). The theta-gamma cross-frequency coupling correlates with animals’ memory performance and is suggested to be a mechanism for local information processing during memory demands (Tort et al., 2009; Axmacher et al., 2010). Cross-frequency coherence plots from WT mice revealed peaks at two gamma frequency ranges at 20 – 40 Hz and 40 – 90 Hz that were coupled to specific phases of theta oscillations (Fig. 3D). We thus defined slow gamma oscillations at 20 – 40 Hz frequency range, and fast gamma oscillations at 40 – 90 Hz range in our experiment. In APP-KI mice, only the slow gamma oscillations from 20 – 40 Hz range were coherent with co-existing theta oscillations (Fig. 3D). The occurrence rate of fast gamma oscillation was diminished in APP-KI mice (p<0.001, Fig. 3E), whereas the occurrence rate of slow gamma remained similar (p >0.05, Fig. 3E). Cross-frequency coupling spectrogram revealed fast and slow gamma oscillations occurring in the peak to falling phase of theta oscillations in WT mice, whereas fast gamma oscillations were significantly lost in APP-KI mice (Fig. 3F). The degree of theta-gamma cross-frequency coupling was assessed by measuring the phase-locking vector length (Colgin et al., 2009). The analysis revealed diminished theta-fast gamma coupling, while theta-slow gamma coupling was spared (p<0.05 for fast gamma; p>0.05 for slow gamma, Fig. 3G). To test whether spike timing is entrained by fast and slow gamma oscillations, we examined the temporal organization of spike activities during slow and fast gamma oscillations (Colgin et al., 2009). Phase-locking of spikes to fast gamma oscillations was diminished in APP-KI mice, but not for slow gamma oscillations (p<0.001 for fast gamma, p>0.05 for slow gamma; Fig. 3H–I and Fig. S3C). Altogether, our results demonstrate that fast gamma oscillations were disrupted in the hippocampal CA1 of APP-KI mice, while slow gamma oscillations were relatively spared. These results suggest that MEC→CA1 signal transfer via fast gamma oscillations is deteriorated in APP-KI mice, whereas slow gamma-mediated CA3→CA1 signal transfer remains relatively intact.

MEC neurons in old APP-KI mice showed severe impairment in spatial tuning

Disrupted fast gamma oscillations observed in hippocampal CA1 suggest that neural activity in the MEC is deteriorated in APP-KI mice. To directly test this possibility, we next investigated spatial representation in the MEC of old APP-KI mice (7–13 mo). We implanted a recording device in the dorsal MEC, and performed unbiased recording of principal neurons in layers 2 and 3 (termed as “MEC neurons” hereafter, n=61 neurons from n = 13 WT mice; n= 65 neurons from n = 13 APP-KI mice; Fig. 4A–C and Fig. S4; no difference in the distribution of sampling positions between WT and APP-KI mice, Fig. S4B). To test for their spatial tuning, MEC neurons were recorded while animals were randomly foraging in the 1×1 m open field. Among the recorded MEC neurons from WT mice, a portion of cells exhibited spatially tuned multiple firing fields as grid cells (Fig. 4D). By contrast, most of the recorded MEC neurons in APP-KI mice showed scattered firing patterns around the open field (Fig. 4E). To evaluate their grid cell property, we analyzed the gridness score by computing the autocorrelation of their firing rate map (Sargolini et al., 2006) (Fig. 4D–F). A cumulative distribution curve of the gridness score for all recorded MEC neurons indicates that, compared to WT mice, APP-KI mice significantly lost a group of neurons with gridness score of more than ~0.4 (p<0.05, Fig. 4F). To define grid cells, we used a cut-off of gridness score at 0.41, which was the 95th percentile of the distribution of gridness score based on the shuffled WT data (Langston et al., 2010) (Fig. 4F and S4D). We found that, among recorded MEC neurons, 20% were grid cells in WT mice, whereas only 2% were grid cells in APP-KI mice (p<0.001, Fig. 4F–G). The scattered firing of APP-KI neurons resulted in significantly reduced spatial tuning (p< 0.001) and increased firing field area (p <0.01, Fig. 4G). However, we did not observe any difference in their mean or peak firing rates, suggesting that MEC neurons in APP-KI mice did not show hyperactivity reported previously in severe transgenic AD models (Busche et al., 2008) (Fig. 4G and S4D). The distribution of spatial information scores for MEC neurons indicated that APP-KI mice lost a significant portion of neurons with high spatial information score, suggesting that aperiodic spatially-tuned MEC neurons (Fyhn et al., 2004; Diehl et al., 2017; Hardcastle et al., 2017; Miao et al., 2017) were also impaired in APP-KI mice (Fig. 4H). These data indicate that spatial tuning of MEC neurons, including grid cells, was severely disrupted in APP-KI mice.

Figure 4. MEC neurons, including grid cells, showed impaired spatial tuning in old APP-KI mice.

(A) A 64-channel recording device was targeted to layers 2/3 of the dorsal part of medial entorhinal cortex (MEC). (B) Sagittal sections with anti-Aβ immunostaining for 12-mo WT and APP-KI mouse (left) and percent area with Aβ plaques in the MEC of WT and APP-KI mice (right, p<0.001, n = 8 sections from two mice in each group). (C) Sagittal sections with Nissl staining showing a representative recording position in the dorsal MEC (right, red arrow). Dotted lines denote the MEC. A, anterior; P, posterior; D, dorsal; V, ventral. (D) Eight representative MEC neurons from WT mice recorded in the 1×1 m open field. Top: Red dots and gray lines denote spike position and animal trace, respectively. Middle: Firing rate map. Color is scaled with maximum firing rate (Hz) shown at top right. Spatial information (SI) score is shown at top left. Bottom: Autocorrelograms of the firing rate maps shown for 2×2 m range. Color is scaled with correlation from r = −1 to 1. Gridness score (GS) is shown at top right. (E) Eight representative MEC neurons as in (D), but from APP-KI mice. (F) Cumulative distribution plot for gridness score of MEC neurons in WT mice (n=61) and APP-KI mice (n=65). A threshold of 0.41, obtained from 95^th percentile of shuffled WT data, was used to define grid cells. (G) From left to right: Percentage of neurons defined as grid cells, spatial information score, firing field size, and mean firing rate. (H) Cumulative distribution plots of spatial information score from all recorded MEC neurons from WT and APP-KI mice. A threshold of 0.43, obtained from 95^th percentile of shuffled WT data, was used to define spatially-tuned cells. See Table S2 for statistics details.

Remapping of MEC neurons in old APP-KI mice

We then tested remapping of MEC neurons. Previous healthy animal studies showed that MEC grid cells exhibit remapping across distinct environments (Fyhn et al., 2007; Diehl et al., 2017). The remapping of MEC grid cells occurs coherently with the remapping of hippocampal place cells. Consistent with the previous finding, we observed that MEC neurons in WT mice showed distinct grid fields between Track A and Track B (Fig. 5A). However, MEC neurons in APP-KI mice showed scattered firing along both Track A and Track B, showing marginal difference in the firing patterns between the two tracks (Fig. 5B). PVec correlation analysis from all recorded MEC neurons revealed that APP-KI mice show larger PVec correlations than that from WT mice, indicating that population remapping property of MEC neurons is reduced in APP-KI mice (p<0.001, Fig. 5C–D and S5). We also assessed spatial correlation of individual neurons between Track A and Track B, confirming that MEC neurons showed reduced remapping property in APP-KI (Fig. 5E). We did not observe a significant difference between WT and APP-KI mice in the stability of MEC neurons from two Track A recordings (p > 0.05, Fig. 5C–E).

Figure 5. Remapping of MEC neurons in old APP-KI mice.

(A) Five representative MEC neurons from WT mice recorded in Tracks A, B, B, and A. Red dots and gray lines at the top of each track denote spike position and animal trace in the 1 m linear tracks, respectively. Bottom color map denotes firing rate map. Color is scaled with maximum firing rate (Hz) shown at top right. Spatial correlation between Track A and B (SCab) is shown at top left. (B) Five representative MEC neurons from APP-KI mice are shown as in (A). (C–E) population remapping calculated using all recorded MEC neurons. (C) Left: Cumulative distribution plots for PVec correlation between Track A and Track B. Right: Cumulative distribution plots for PVec correlation between two recordings in Track A. (D) Mean PVec correlation between Track A and Track B (left) and between two recordings in Track A. (E) Left: Cumulative distribution plots of spatial correlation between Tracks A and B (left) and between two recordings in Track A. (F) PVec correlation as in (C–D), but for MEC neurons defined as spatially tuned cells. (G) PVec correlation as in (C–D), but for MEC neurons defined as non-spatial cells. (H) Correlation between spatial tuning (spatial information score, x-axis) and remapping (spatial correlation for Tracks A vs Track B, y-axis). Each dot represents recorded CA1 neurons in WT mice (top) and APP-KI mice (bottom). (I) Correlation between spatial tuning and remapping as in (A), but for MEC neurons in WT mice (top) and APP-KI mice (bottom). Only MEC neurons in APP-KI mice showed significant correlation. See Table S2 for statistics details.

We next assessed remapping properties from each cell class of MEC neurons. Because grid cells were significantly disrupted and reduced in number from APP-KI mice (Fig. 4G and S4D), remapping analysis could not be performed for grid cells. Spatially tuned cells were also reduced (Fig. 4E), resulting with 12 spatially tuned cells recorded in APP-KI mice. Although less reliable from the limited sample number, remapping of spatially tuned cells in APP-KI mice were comparable to that in WT mice, suggesting intact remapping from surviving spatially tuned cells (Fig. 5F). By contrast, while MEC non-spatial cells showed discernible remapping in WT mice, remapping property of non-spatial cells was reduced in APP-KI mice (Fig. 5G).

To obtain insight into the causality between impairment of spatial tuning and remapping property in CA1 and MEC neurons, we measured the correlation between spatial information score and spatial correlation between Track A and Track B in individual neurons (Fig. 5H–I). In WT mice, neither CA1 neurons nor MEC neurons showed correlation between spatial tuning and remapping, suggesting that these two functional properties are independent in healthy WT mice (p>0.05, Spearman correlation, Fig. 5H–I). However, in APP-KI mice, we found that MEC neurons show significant negative correlation between spatial information score and spatial correlation (p<0.001 and r(67) = −0.49, Spearman correlation, Fig. 5I; also see Fig. S5D). By contrast, CA1 neurons in APP-KI mice did not show significant correlation (p>0.05, Fig. 5H), implying that the remapping disruption of individual CA1 neurons is independent from their spatial tuning property.

Fast gamma coupling between CA1 and MEC were impaired in old APP-KI mice

To test if the impairment of MEC neurons is accompanied by the deterioration of local gamma oscillations, we next examined gamma oscillations in the MEC (Fig. 6A). Because previous healthy animal studies found only fast gamma but not slow gamma oscillations in the MEC (Chrobak and Buzsaki, 1998; Colgin et al., 2009), we focused our analysis on fast gamma oscillations. We indeed observed only marginal slow gamma oscillations in the MEC of WT mice (Fig. 6B and 6E). No difference was observed for MEC slow gamma oscillations between WT and APP-KI mice (Fig. S6). The analysis of theta-fast gamma cross-frequency coupling revealed the absence of fast gamma oscillations coupled to theta oscillations in APP-KI mice (Fig. 6B). However, we did not see any difference in the normalized oscillatory power nor in the occurrence rate of fast gamma oscillations between WT and APP-KI mice (q>0.05, FDR corrected, Fig. 6C; p>0.05 for gamma occurrence, Fig. 6D). Theta-phase spectrograms showed tendency of fast gamma oscillations in APP-KI mice being weakly modulated by theta oscillations compared to WT mice (p=0.051, Fig. 6E–F). Although our observation revealed only a tendency of reduction, the tendency was consistent with our previous finding that the theta-fast gamma coupling was reduced in the MEC of 5-mo anaesthetized APP-KI mice (Nakazono et al., 2017). The spikes from MEC neurons in WT mice were phase-locked to local fast gamma oscillations, as reported in a previous study (Quilichini et al., 2010) (Fig. 6G–H). However, the spike-fast gamma phase-locking was markedly disrupted in APP-KI mice (p<0.01, Fig. 6G–H and S6B).

Figure 6. Impaired fast gamma oscillations in the MEC of old APP-KI mice.

(A) Fast gamma oscillations were recorded from the MEC during the open field task. (B) Example cross-frequency coherence plots showing that 40–100 Hz oscillatory power (y-axis) was modulated by theta phase (x-axis) in a WT mouse (left), but this modulation was absent in an APP-KI mouse (right). Dotted line denotes 40-Hz border between fast and slow gamma. (C) Normalized power spectra for WT and APP-KI mice (n=13 mice each). No difference was observed in the power spectra. (D) Occurrence rates of fast gamma episodes in WT and APP-KI mice. (E) Normalized power spectrograms averaged across all theta cycles in three representative WT mice (left) and APP-KI mice (right). Theta phase is shown at the bottom. Dotted line denotes 40-Hz border between fast and slow gammas. (F) Mean vector length for the distribution of fast gamma episodes across theta phase. (G) Fast gamma phase distribution of spikes from example three MEC neurons in WT (left) and APP-KI (right) mice. Schematics of two cycles of gamma waveform are shown in the bottom. (H) Mean vector length for the distribution of spikes across fast gamma phase for MEC neurons in WT (n=60 cells) and APP-KI (n=62 cells) mice. (I–L) Disrupted coupling of fast gamma oscillations between the MEC and CA1. (I) Fast gamma oscillations were recorded simultaneously in the MEC and hippocampal CA1 during the open field task. (J) Mean coherence plot of oscillatory activity as a function of frequency in WT and APP-KI mice (n=6 mice each). Significant difference in coherence for each 1Hz bin is shown with green dot at the top (q<0.05 at 58–75 Hz range, FDR corrected for multiple comparison). (K) Representative averaged local field potential traces from the MEC of WT mouse (top) and APP-KI mouse (bottom) during fast gamma episodes detected in CA1. t = 0 corresponds to the peak amplitudes of the fast gamma episodes in CA1. (L) MEC:CA1 power ratio during fast gamma episodes detected in CA1. See Table S2 for statistics details.

To directly test the idea that the impairment of fast gamma oscillations affects MEC→CA1 signal transfer in APP-KI mice, we implanted two bundles of electrodes and simultaneously recorded from the hippocampal CA1 and the MEC (n = 6 WT mice and n = 6 APP-KI mice) (Fig. 6I). To test the degree of synchronization between the gamma oscillations in the MEC and the CA1, we assessed coherence of oscillatory activity (Colgin et al., 2009; Igarashi et al., 2014). Analysis revealed that the gamma oscillations between the MEC and CA1 have weaker coherence at 58 – 75 Hz range in APP-KI mice (q<0.05, FDR corrected; Fig. 6J). We then collected oscillatory activity in the MEC when fast gamma oscillations were detected from the hippocampal CA1 (Colgin et al., 2009) (Fig. 6K). The MEC fast gamma oscillations were significantly desynchronized from the CA1 fast gamma oscillations in APP-KI mice. This was confirmed by the decreased MEC:CA1 ratio of fast gamma powers, where fast gamma power of CA1-gamma-triggered LFP episodes in the MEC was divided by the power of CA1 fast gamma episodes (Colgin et al., 2009) (p<0.05, see method) (Fig. 6L). The reduction of MEC-CA1 fast gamma coupling (Fig. 6K–L) without the reduction of MEC fast gamma power (Fig. 6C) implies that MEC fast gamma episodes tend to occur when CA1 fast gamma is absent. In agreement with this notion, a reduction of CA1/MEC fast gamma power ratio was also observed when MEC gamma was used for triggering (Fig. S6C). Altogether, these results demonstrate that the temporal coordination of the gamma oscillations between the MEC and the CA1 was diminished, suggesting that MEC→CA1 signal transfer mediated by the fast gamma oscillations is deteriorated in APP-KI mice.

Mild disruption of MEC grid cells emerged in young APP-KI mice while spatial memory and CA1 remapping remained intact

The above results imply that the impairment of MEC neurons can be a primary deficit that induces other secondary impairments in old APP-KI mice. If this is the case, the impairment of MEC neurons may emerge in the earlier phase of AD progression. To test this possibility, we examined spatial memory and recorded MEC and CA1 neurons from young APP-KI mice of 3–5 months of age.

We first assessed the context discrimination of young APP-KI mice (Fig. 7A–C). Intriguingly, young APP-KI mice discriminated Context A and Context B, and did not develop freezing in Context B similar to control young WT mice (p> 0.05 for genotype, two-way ANOVA, n = 8 APP-KI and n =8 WT mice; Fig. 7A–B). Discrimination index did not differ between APP-KI and WT mice (p> 0.05 for genotype, two-way ANOVA). These data indicate that young APP-KI mice exhibited intact spatial memory.

Figure 7. CA1 neurons in young APP-KI mice showed intact spatial memory and CA1 remapping.

(A–C) The context discrimination task as in Fig. 1, but for young (3–5 mo) APP-KI and WT mice. Percent time frozen in Context A (A), percent time frozen in context B (B) and Discrimination Index (C) are shown as in Fig. 1. No significant difference was found in young APP-KI mice during the task. (D–H) Recording of CA1 neurons as in Fig 1 and Fig. 2, but from young APP-KI and WT mice. (D) A coronal section with anti-Aβ immunostaining from 3-mo APP-KI mouse (middle), percent area with Aβ plaques in the hippocampal CA1 (right, n = 8 sections from two mice in each group). (E) Two representative CA1 neurons from WT and APP-KI mice recorded in the open field, as in Fig. 1H–I. (F) Two representative CA1 neurons from young WT and APP-KI mice recorded in the linear tracks, as in Fig. 2B–C. (G) Spatial information score (left) and percent place cells in young mice (right). No significant difference was observed between young WT and APP-KI mice. Spatial information of 0.63 was used for defining place cells as in Fig. 1. (H) Population remapping calculated using all recorded CA1 neurons. Cumulative distribution plots for PVec correlation between Tracks A and B (left) and between two recordings in Track A (right). (I) PVec correlation as in (H), but for CA1 neurons defined as place cells. (J) PVec correlation as in (H), but for CA1 neurons defined as non-spatial cells. See Table S2 for statistics details.

The hippocampal CA1 region of young APP-KI mice already showed Aβ plaque formation (Fig. 7D). Our recordings from young APP-KI mice showed that CA1 neurons exhibit comparable spatial tuning and remapping (p>0.05 for all of spatial tuning, % place cells and population vector correlation, n = 35 neurons in n = 5 WT mice and n = 50 neurons in n = 6 APP-KI mice; Fig. 7E–H). Remapping property was spared in place cells, but was weakly diminished in non-spatial cells (Fig. 7I–J). These data suggest that Aβ plaque formation in CA1 does not affect spatial tuning nor remapping of CA1 neuronal population in young APP-KI mice.

We then assessed MEC neurons of young APP-KI mice. The MEC showed a moderate amount of Aβ plaques (Fig. 8A). As expected, our recording showed that impaired spatial tuning of MEC neurons existed concurrently with Aβ plaques (Fig. 8B–C). Spatial information was diminished (p<0.01; n = 42 neurons in n = 11 WT mice and n = 39 neurons in n = 8 APP-KI mice). The percentage of grid cells decreased to 7.7 % (p<0.05, Fig. 8C), a milder disruption compared to the reduction to 2% in 12-mo APP-KI mice. The population remapping of MEC neurons, as well as remapping of spatially tuned cells, were weakly reduced compared to young WT mice (p<0.01, Fig. 8D–F), whereas no reduction was discernible for non-spatial cells (p>0.05, Fig. 8G).

Figure 8. MEC neurons in young APP-KI mice showed mildly disrupted spatial tuning.

Recording of MEC neurons as in Fig. 4–5, but from young APP-KI and WT mice. (A) A sagittal section with anti-Aβ immunostaining from 3-mo APP-KI mouse (middle), percent area with Aβ plaques in the MEC (right, n = 8 sections from two mice in each group). (B) Two representative MEC neurons from WT and APP-KI mice recorded in the open field, as in Fig. 4D–E. (C) Young APP-KI mice showed reduced spatial information score (left) and percent grid cells than young WT mice. Gridness score of 0.41 was used for defining grid cells as in Fig. 4. (D) Two representative MEC neurons from WT and APP-KI mice recorded in the linear tracks, as in Fig. 5A–B. (E) Population remapping calculated using all recorded MEC neurons. Cumulative distribution plots for PVec correlation between Tracks A and B (left) and between two recordings in Track A (right). (F) PVec correlation as in (E), but for MEC neurons defined as spatially tuned cells. (G) PVec correlation as in (E), but for MEC neurons defined as non-spatial cells. (H) Time course plots of impairments found in the study. Impairment Index (II) was calculated for quantitatively plotting impairments in spatial memory, spatial tuning and remapping for CA1 (left) and MEC (right, II = 1 − APP-KI/WT, see Text). For comparison, a plot for spatial memory is shown in both CA1 and MEC panels. Increase of Aβ plaque is normalized and plotted so that percent Aβ plaque area in the same region of old APP-KI mice becomes 1. See Table S2 for statistics details.

Figure 8H summarizes the data obtained in our study (Fig. 8H). The age-dependent progression of deterioration in spatial memory, spatial tuning, and population remapping found in our results were plotted for CA1 and MEC. In these plots, we defined Impairment Index (II) as the following: each functional property (spatial tuning score for spatial tuning, PVec correlation for remapping and Discrimination Index for spatial memory) obtained from APP-KI mice was first normalized by that obtained from age-matched WT mice. This ratio was then subtracted from 1 (II = 1 − APP-KI / WT) so that II becomes 0 if there is no impairment and II becomes 1 if the functional property is completely impaired (see Method). Properties in APP-KI mice with significant reduction relative to WT mice found in the study were shown with asterisks (*p<0.01, **p<0.001). In these plots, the impairment of spatial memory was plotted in both CA1 and MEC plots. For Aβ plaque, percent areas with Aβ plaque in young APP-KI mice were plotted so that the values obtained from the same area in old mice become 1. These plots indicate that the spatial tuning impairment in the MEC emerges in young APP-KI mice, and suggest that this impairment is already close to their asymptote. By contrast, memory impairment and the CA1 remapping impairment emerged only in old APP-KI mice. These results demonstrate that the spatial tuning impairment of MEC neurons emerged earlier than memory impairment or CA1 remapping impairment in APP-KI mice.

Discussion

In this study, using novel APP-KI mice, we for the first time investigated how the expression of mutated APP gene and resultant Aβ affect the remapping of neurons in CA1 and the MEC. In old APP-KI mice, we found that the spatial memory and remapping of CA1 neurons was disrupted. Spatial tuning of MEC neurons, including grid cells, was severely impaired. By contrast, in young APP-KI mice that do not show any spatial memory impairment, mild impairments of MEC neurons were found while CA1 neurons were still intact for both spatial tuning and remapping. These results are summarized in Fig. S8D. Thus, answers to our initial questions would be that, (1) yes, the remapping of CA1 place cells was disrupted, and (2) yes, the impairment of remapping in CA1 place cells accompanied the spatial tuning impairment of MEC neurons, including grid cells. The alternative hypothesis that the place cell deterioration occurs independently from impairments in MEC neuron is thus unlikely. Previous anatomical studies in AD patients showed that neurons in the entorhinal cortex exhibit the earliest neuronal degeneration (Van Hoesen et al., 1991; Gomez-Isla et al., 1996), but the functional deterioration of entorhinal neurons and its link to spatial memory impairment has been unclear. Although impairments in the spatial tuning of place cells or grid cells were reported (Cacucci et al., 2008; Kunz et al., 2015; Fu et al., 2017; Mably et al., 2017), no previous studies have investigated remapping in AD subjects. Our study provides the first circuit-level evidence that identifies the remapping disruption of hippocampal CA1 neurons, possibly linked to the impairment of MEC neurons, and as an underlying mechanism for memory impairment in AD (Fig. S8D).

What are causal relationships between the MEC spatial tuning impairment, CA1 remapping impairment, spatial memory impairments and Aβ plaque formations in the MEC and CA1 in our APP-KI mice? Previous studies in healthy animals showed that the manipulation of MEC neuronal activity modulates remapping in CA1 (Miao et al., 2015; Kanter et al., 2017), whereas another study showed that CA1 remapping can occur without inputs from the MEC (Schlesiger et al., 2018). Thus, the mechanism for remapping in the entorhinal-hippocampal circuit has not been clearly established in healthy subjects. Importantly, our recording from young APP-KI mice provides the following clues (Fig. 8F): (1) The spatial tuning impairment was found only in the MEC, but not in CA1; (2) Although Aβ plaques were already present weakly in CA1, spatial tuning and remapping in CA1 were not affected in young APP-KI mice; (3) Behavioral memory impairment did not occur even with the emergence of mild MEC spatial tuning impairment. Taken together, this line of evidence raises a possibility that the circuit dysfunction progresses in the direction of MEC→CA1 in the entorhinal-hippocampal circuit. Although speculative, the following steps may coherently explain our findings: (i) In young APP-KI mice, Aβ plaques induced by the knock-in expression of mutated APP may cause the impairment of spatial tuning of MEC neurons including grid cells (Fig. 8), presumably through the deterioration of network interaction in the MEC (Moser et al., 2014). (ii) As disease progresses in old APP-KI mice, MEC neurons with impairment in spatial tuning increase in number (Fig. 4), and remaining neurons with intact remapping property further becomes reduced (Fig. 5). Note that, because of the severe disruption of spatial tuning of MEC neurons, it was unclear from our result if remapping was truly disrupted in the MEC. The spatial tuning impairment in the MEC may result in the severe remapping impairment of CA1 neurons in old APP-KI mice (Fig. 2), even though the CA1 spatial tuning is relatively spared (Fig. 1). (iii) The remapping impairment of CA1 neurons may ultimately result in the behavioral impairment in spatial memory (Fig. 1). It is likely that the dysfunction in the CA1, rather than the dysfunction in the MEC, may be the direct cause for the spatial memory impairment. A pathophysiological progression in the opposite CA1→MEC direction, or the autonomous emergence of symptoms in the CA1, would be incompatible with the early emergence of MEC impairment observed in young APP-KI mice. Our results, however, do not exclude a possibility that another brain region unexamined in this study causes impairments observed in the MEC and CA1. Whatever the causality may be, our finding emphasizes the importance of the entorhinal cortex as a brain region that manifests spatial tuning impairment earlier than the hippocampus. In the AD research field, the hippocampus has been the main brain region of interest for decades. Together with a previous study showing diminished grid cell activity in humans at AD risk (Kunz et al., 2015), our results point to MEC spatial tuning impairment as a potential “functional biomarker” that may predict subsequent development of memory impairments in AD. Yet, discrepancy of pathophysiology can exist between APP-KI mice and AD patients. The combination of three mutations (Swedish, Iberian and Arctic mutations) harbored in APP-KI mice does not occur simultaneously in familial AD patients, and AD patients do not harbor homozygous mutations as used in this study. Importantly, the impairments found in our APP-KI model need to be thoroughly tested in AD patients.

Consistent with our previous findings showing disrupted fast gamma oscillations in the MEC of young APP-KI mice (Nakazono et al., 2017), we found that fast gamma oscillations were disrupted in old APP-KI mice in both the MEC and CA1. The fast gamma coupling between the MEC and CA1 is also diminished, suggesting impaired signal transfer in the MEC→CA1 circuit. Together with previous finding of impaired gamma oscillations in the hippocampus (Iaccarino et al., 2016; Mably et al., 2017), our results suggest that the reactivation of gamma oscillations may be used as a potential therapeutic method in AD (Nakazono et al., 2018). Interestingly, we observed a gap of fast and slow gamma oscillations around 40 Hz in our WT and APP-KI mice, compared to the border at 50 Hz in rats (Colgin et al., 2009). There may exist a species-specific border between slow and fast gamma oscillations in mice.

In healthy animals, neurons in the entorhinal-hippocampal circuit exhibit two forms of remapping known as “global remapping” and “rate remapping” (Leutgeb et al., 2005b). When animals move across environments with distinct features, entorhinal grid cells and hippocampal place cells show distinct patterns of firing (global remapping) (Muller and Kubie, 1987; Fyhn et al., 2007). By contrast, when there is only a slight change in an environment, hippocampal place cells show change in their firing rate, while keeping the position of firing field constant (rate remapping) (Leutgeb et al., 2005a; Leutgeb et al., 2007). Although it is likely that both global remapping and rate remapping are needed for high-precision spatial memory, disruption of global remapping would completely deprive AD patients of spatial information about which environment they are currently in. Our results showing the disruption of global remapping in the hippocampal CA1 of APP-KI mice clearly explain the devastating impairment of spatial memory in AD patients, who often exhibit the wandering symptom and cannot discriminate distinct features of surrounding environments. Future development of techniques to protect or reactivate spatial tuning and remapping in the entorhinal-hippocampal circuit may lead to therapeutic methods that can be used to slow the rate of spatial memory decline in AD patients.

STAR★METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Kei Igarashi (kei.igarashi@uci.edu).

Materials availability

This study did not generate new unique reagents.

Data availability

Neurophysiological data and analytical codes are available upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

Mice were maintained in standard housing conditions on a 12h dark/light cycle with food and water provided ad libitum. All procedures were conducted in accordance with the guidelines of the National Institutes of Health and approved by the Institutional Animal Care and Use Committee at the University of California, Irvine. We previously generated App^{NL-G-F/NL-G-F} maintained on a C57BL/6 background (Saito et al., 2014). For all experiments, 28 App^{NL-G-F/NL-G-F} mice (21 males and 7 females) and 27 App^WT/WT mice (18 males and 9 females) between 3 months and 13 months of age were used. We noted no difference in our preliminary analyses between sexes. Detailed information about ages when the recording was performed is in Supplementary Table S1. Animals were housed in a reversed 12h dark/light cycle, and all testing occurred during the dark phase. If animals died or were sick during or before the final test, their recording positions could not be validated and therefore were removed from the data analysis.

METHODS DETAIL

SURGERY AND ELECTRODE PREPARATION

All mice received a custom-built 64-channel drive modified from Liang et al (Liang et al., 2017) targeting either CA1 or MEC. For a group of animals, eight tetrodes targeted CA1 of hippocampus and another eight tetrodes targeted the superficial layer of medial entorhinal cortex. Tetrodes were constructed from four twisted 17 μm polyimide-coated platinum-iridium (90%–10%) wires (California Fine Wire). The electrode tips were plated with gold to reduce electrode impedances to between 150 and 300 kΩ at 1 kHz. Animals were anesthetized with isoflurane (air flow: 0.8–1.0 l/min, 1% isoflurane, adjusted according to physiological condition). The mice received subcutaneous injections of buprenorphine at the start of the surgery. Depth of anesthesia was examined by testing tail and pinch reflexes as well as breathing. Upon induction of anesthesia, the animal was fixed in a Kopf stereotaxic frame for implantation. A craniotomy was performed centering at AP 2.5 mm ML 2.5 mm from Bregma for hippocampus (approximately 1 × 1mm) and 0.2–0.4mm anterior to transverse sinus and ML 3.5mm from midline for medial entorhinal cortex (approximately 1 × 1mm). The dura was carefully removed, and the tetrodes were implanted. A stainless-steel screw fixed to skull above the cerebellum served as the ground electrode. The recording drive was secured to the scratched skull using dental cement.

APPARATUS AND TRAINING PROCEDURES

Behavioral training started at least 5 days after the surgery and data collection was performed exactly on the fifth day of the training. All experiments were conducted in a room containing a 1m-square box made of black acryl and a polarizing white cue card (216mm × 280mm) and two 1-m long linear tracks. In the open field task, running was motivated in the 1m-square box with cookie crumbs thrown randomly into the enclosure. Each session lasted 10 to 20 min. Upon finishing the open field task, animals rested for 5 minutes and then were tested in the remapping task. The remapping task training required mice to run in 1-m long linear tracks. Two 1-m long linear tracks were used: (Track A) Black acryl enclosure with inner wall decorated with patterns of white cue cards across the track and black rubber floor. (Track B) White acryl enclosure with inner wall decorated with distinct patterns of black cue signs across the track and white sandpaper floor. During the remapping task, animals were trained to run in successive series of Track A → Track B → Track B → Track A. Two tracks were in the same room. Between the sessions, animals were given 5 minutes rest. Running was motivated by placing cookie crumbs on the end positions of the linear tracks. Each session lasted 5 to 10 min. On the linear tracks, the mice ran 10 full laps (back and forth).

DATA COLLECTION

LFPs were recorded single-ended using a ground in the skull above the cerebellum. One mouse without single-ended recording was excluded for LFP analyses. The tetrodes were connected to a multichannel, impedance matching, unity gain headstage (Neuralynx). The output of the headstage was conducted to a data acquisition system (Neuralynx). Unit activity was amplified by a factor of 3000–5000 and band-pass filtered from 600 Hz to 6000 Hz. Spike waveforms above a threshold set by the experimenter (~55 μV) were time-stamped and digitized at 32 kHz for 1 ms. LFP signals, 1 per tetrode, were recorded in the 0–475 Hz frequency band at a sampling rate of 2000 Hz. Notch filters were not applied. The recording system tracked the position of two light-emitting diodes (LEDs), one large and one small, on the head stage (sampling rate 50 Hz) by means of an overhead video camera.

DATA ANALYSIS

Unless indicated otherwise, analyses were performed using MATLAB codes kindly provided by Dr. Edvard Moser, or written by the authors (Igarashi et al., 2014)

SPIKE ANALYSIS

Spike Sorting and Cell Classification.

Spike sorting was performed offline using graphical cluster-cutting software, MClust by Dr. David Redish. Putative excitatory cells were distinguished from putative interneurons by spike peak-valley width and average rate (Bartho et al., 2004). All putative excitatory cells with spike peak-valley width of more than 230 μs and mean firing rates of more than 0.1 Hz (mean firing rate during all left and right cue sampling intervals) were included for further analysis. We used this conservative criterion only for selecting putative principal neurons, but not for selecting interneurons. It should be noted that neurons with spike peak-valley width of less than 230 μs contain a certain number of cells with low spike firing frequency, which could potentially be principal neurons.

Firing rate maps.

Position estimates were based on tracking of the LEDs on the head stage connected to the recording drive. For rate maps in the random foraging task, data were speed-filtered; only epochs with instantaneous running speeds of 1 cm/s or more were included.

To characterize firing fields, the position data were sorted into one-dimensional 1 cm bins for the linear tracks and 2.5 cm × 2.5 cm bins for the open field. For the open field map, the path was smoothed with a 21-sample boxcar window filter (400 ms; 10 samples on each side). Firing rate distributions were then determined by counting the number of spikes in each bin as well as the time spent per bin. Maps for number of spikes and time were smoothed individually using a boxcar average over the surrounding 5 × 5 bins. Weights were distributed as follows:

$$\text{box}=[0.00250.01250.02000.01250.0025;\dots 0.01250.06250.10000.06250.0125;\dots 0.02000.10000.16000.10000.0200;\dots 0.01250.06250.10000.06250.0125;\dots 0.00250.01250.02000.01250.0025;]$$

Spatial information score.

For each cell, the spatial information score in bits per spike was calculated from the recordings in the open field task as

$$\mathit{\text{Spatial Information Score}}=\sum _i{p}_i\frac{{\lambda }_i}{\lambda }{\text{log}}_2\frac{{\lambda }_i}{\lambda }$$

where λ_i is the mean firing rate of a unit in the ith bin, λ is the overall mean firing rate, and p_i is the probability of the animal being in the ith bin (occupancy in the ith bin / total recording time) (Skaggs et al., 1993). An adaptive smoothing method, introduced by Skaggs et al. (Skaggs et al., 1996), was used before the calculation of information scores (Henriksen et al., 2010).

Place/grid field area and number of fields.

A firing field in the open field was estimated as a contiguous region of at least 7.5 cm × 7.5 cm where the firing rate was above 20% of the peak rate. Low threshold for peak rate was 1 Hz. Additional fields were identified by deleting the detected field from the rate map and iterating the search for contiguous firing regions in the remaining part of the rate map until no additional fields were found. The cell’s peak rate was estimated as the highest firing rate observed in any bin of the smoothed rate map. For the linear tracks, a place field was defined as a contiguous region of at least 4 cm where the firing rate exceeded 20 % of the peak rate. Low threshold for peak rate was 1 Hz. Additional place fields were counted only when the peak position of the field was separated from other fields by more than the width of the field size.

Defining Place Cells.

Place cells were defined as cells with spatial information score above the chance level (Langston et al., 2010). The chance level was determined in each brain region by a random permutation procedure using all cells recorded at that time-point in that region from old WT mice. One hundred permutations were performed for each cell in the sample. For each permutation trial, the entire sequence of spikes fired by the cell was time-shifted along the animal’s path by a random interval between 20 s and 20 s less than the total length of the trial (usually 600 −20 = 580 s), with the end of the trial wrapped to the beginning to allow for circular displacements. This procedure allowed the temporal firing structure to be retained in the shuffled data at the same time as the spatial structure was lost. Spatial information was then calculated for each shuffled map. The distribution of spatial information values across all 100 permutations of all cells in the sample was computed and finally the 95th percentile was determined. Place cells were defined as cells with spatial information scores above the 95th percentile of the distribution from shuffled data for the relevant group.

Gridness score.

The structure of all rate maps was evaluated for all cells by calculating the spatial autocorrelation for each smoothed rate map (Sargolini et al., 2006; Fyhn et al., 2007). Autocorrelograms were based on Pearson’s product moment correlation coefficient with corrections for edge effects and unvisited locations. With λ (x, y) denoting the average rate of a cell at location (x, y), the autocorrelation between the fields with spatial lags of τ_x and τ_y was estimated as:

$$r\left({\tau }_x,{\tau }_y\right)=\frac{n\sum \lambda (x,y)\lambda \left(x-{\tau }_x,y-{\tau }_y\right)-\sum \lambda (x,y)\sum \lambda \left(x-{\tau }_x,y-{\tau }_y\right)}{\sqrt{n\sum \lambda {(x,y)}^2-{\left(\sum \lambda (x,y)\right)}^2}\sqrt{n\sum \lambda {\left(x-{\tau }_x,y-{\tau }_y\right)}^2-{\left(\sum \lambda \left(x-{\tau }_x,y-{\tau }_y\right)\right)}^2}}$$

where the summation is over all n pixels in λ (x, y) for which rate was estimated for both λ (x, y) and λ (x − τ_x, y − τ_y). Autocorrelations were not estimated for lags of τ_x, τ_y where n < 20.

The degree of spatial periodicity (‘gridness score’) was determined for each recorded cell by taking a circular sample of the autocorrelogram, centered on the central peak but with the central peak excluded, and comparing rotated versions of this sample. The Pearson correlation of this circle with its rotation in α degrees was obtained for angles of 60° and 120° on one side and 30°, 90° and 150° on the other. The cell’s gridness score was defined as the minimum difference between any of the elements in the first group and any of the elements in the second. The radius of the excluded central peak was defined as either the first local minimum in a curve showing correlation as a function of average distance from the center, or as the first incidence where the correlation was negative, whichever occurred first.

Defining Grid Cells.

Grid cells were defined as cells in which rotational symmetry-based gridness scores exceeded the 95th percentile of a distribution of grid scores for shuffled recordings from the entire population of cells obtained from old WT mice. Shuffling was performed in the same way as for place cells, i.e. for each permutation trial, the entire sequence of spikes fired by the cell was time-shifted along the animal’s path by a random interval between 20 s and 20 s less than the length of the trial (usually 600 – 20 = 580 s), with the end of the trial wrapped to the beginning. The shuffling procedure was repeated 100 times for each of the cells in the sample. For each permutation, a firing rate map and an autocorrelation map were constructed and a grid score was calculated. Grid scores were computed for all permutations for all cells and the 95th percentile was read out from the overall distribution. Grid cells were defined as cells with grid scores higher than the 95th percentile of the grid scores of the distribution for the shuffled data.

Single-cell Shuffling Analysis.

Recently, an alternative method for shuffling spatial information and gridness score was introduced using single-cell data for defining spatially-tuned cells and grid cells (Diehl et al., 2017). We reanalyzed our data using this method (Fig. S9). Our result showed that, in CA1, the spatially tuned neurons did not differ in number between WT and APP-KI mice in both young and old mice (P>0.05, Wilcoxon rank sum test, Fig. S9A1). However, in the MEC, spatially tuned cells, as well as grid cells, were reduced in both young and old APP-KI mice (P<0.05 or better, Wilcoxon rank sum test, Fig. S9 A₁). The analysis also confirmed the notion that the deterioration of spatially tuned cells in young APP-KI mice appeared only in MEC (Fig. S9A₄).

Remapping Analysis.

Remapping analysis was performed as described previously (Miao et al., 2015). For experiments on the linear track, instantaneous firing rates on individual laps were estimated using a Gaussian kernel on the spike data for temporal smoothing:

$$r(t)=\sum _{i=1}^{N}g\left(\frac{t_i-t}{h}\right)$$

where g is a 1D Gaussian kernel, h is bandwidth, N is the total number of spikes, and t_i is the time of the i^th spike. The bandwidth was set at 100 ms. Due to minimal coverage per bin, temporal smoothing gives more robust rate estimates compared to those based on spatial bins during fast running on the track. Spike rate at each track position (1 cm bin) on each lap was estimated using a linear interpolation method applied to the temporally-smoothed spike rates. Because the same cells often had different firing fields on left and right runs, firing fields of the same cell in each run direction were analyzed as two distinct data sets. A segment of 5 cm was excluded from each end of the linear track for analysis.

Spatial correlation was obtained by calculating the Pearson correlation coefficient for mean firing rates across 1 cm wide bins from the track on a pair of sessions. For population vector analysis, population vectors were defined for each 3.6-cm bin of rate maps (25 bins in total) from all cells in the experimental group. To maintain independence of neighboring spatial bins, we estimated mean firing rates in spatial bins without smoothing. For lap-by-lap population vector analyses, a total of 10 laps (defined as pairs of forward and backward runs) were taken from each recording session. Population vector correlations were determined for each spatial bin of each lap.

Time course analysis.

The age-dependent progression of impairment in behavioral context discrimination, impaired spatial tuning, impaired remapping found in our results were plotted for CA1 and MEC. In these plots, Impairment Index (II) was defined as following: Each functional property (spatial tuning score for spatial tuning, PVec correlation for remapping and Discrimination Index for context discrimination) obtained from APP-KI mice was first normalized by the that obtained from age-matched WT mice. This ratio was then subtracted from 1 (II = 1 − APP-KI / WT) so that II become 0 if there is no impairment and II become 1 if the functional property is completely impaired. We used the normalization by age-matched WT, rather than just normalizing by old mice data, in order to take age effects into consideration. For Aβ plaque where no data are available for correcting age effect, percent areas with Aβ plaque in young APP-KI mice were plotted so that the values obtained from the same area in old mice become 1 (II_Aβ = Aβ plaque area / Aβ plaque area_old).

LFP ANALYSIS

Animals with single-ended LFP recordings were used for the analysis.

Time-frequency analysis of power.

Time-resolved power across frequencies was computed using a wavelet transform method. Signals were convolved by a family of complex Morlet’s wavelets w(t, f), one for each frequency, as a function of time:

$$w(t,f)=A\text{ exp}\left(-t^2/{\sigma }_t^2\right)\text{exp}(2i\pi \text{ft})$$

in which the family of wavelets was characterized by a constant ratio f/σ_f, which was set to 7, and with σ_f = ½πσ_t. The coefficient A was set at:

$${\left({\sigma }_t\sqrt{\sigma }\right)}^{-1/2}$$

in order to normalize the wavelets such that their total energy was equal to 1.

Normalization of LFP power.

To control for impedance differences between tetrodes, LFP power was normalized, for each tetrode, to the total power in 1–100 Hz band.

Bandpass filtering.

An acausal (zero phase shift), frequency domain, finite impulse response bandpass filter was applied to the signals. For theta filtering, 5 and 6 Hz were chosen for the stopband and passband, respectively, for the low cut-off frequencies; 10 and 11 Hz were chosen for passband and stopband high cut-off frequencies. For filtering of gamma oscillations, 19 and 20 Hz were chosen for the stopband and passband, respectively, for the low cut-off frequencies; 90 and 91 Hz were chosen for passband and stopband high cut-off frequencies. For filtering of slow gamma oscillations, 19 and 20 Hz were chosen for the stopband and passband, respectively, for the low cut-off frequencies; 40 and 41 Hz were chosen for passband and stopband high cut-off frequencies. For filtering of fast gamma oscillations, 39 and 40 Hz were chosen for the stopband and passband, respectively, for the low cut-off frequencies; 90 and 91 Hz were chosen for passband and stopband high cut-off frequencies.

Selection of theta cycles.

Theta cycles were selected by bandpass filtering the signal from 6–10 Hz and selecting local minima in the filtered signal (i.e., θ(t − 1) > θ(t) and θ(t + 1) > θ(t)). Segments of the recording were collected and defined as a theta cycle if the time between detected points fell within a range criterion that corresponded to the period of an ~8 Hz theta cycle (i.e. 125 ± 25 ms). Local minima of detected theta cycles were required to be separated by at least 100 ms.

Detection of oscillation episodes.

To extract periods of gamma oscillatory activity in the LFP, we first computed time-varying power within the frequency bands for each recording. Power at each time point was averaged across the frequency range to obtain time-varying estimates of oscillatory power. Time points were collected when the power exceeded 2 SD of the time-averaged power. Time windows, 160 ms in length, were cut around the identified time points. In each 160 ms segment, the maxima of gamma oscillatory amplitude were determined from the gamma bandpass filtered versions of the recordings. Duplicated gamma oscillatory periods, a common consequence of extracting overlapping time windows, were avoided by discarding identical maxima values within a given gamma oscillatory subtype and further requiring that maxima of a given subtype be separated by at least 100 ms. Individual gamma oscillatory windows were finally constructed from the original, non-bandpass filtered recordings as 400 ms long windows centered around the gamma oscillatory amplitude maxima.

Relationship of gamma to theta phase.

LFP recordings were bandpass-filtered in the theta range (6–10 Hz), and theta phases for each time point were estimated using the Hilbert transform function from the Signal Processing Toolbox in MATLAB. Theta phases at the time points associated with gamma maxima (determined as described above) as well as theta phases for spikes were collected. Theta phases for each gamma oscillatory event were sorted into 30° bins, allowing the phase distribution of each event to be determined. For a given recording, the distributions of gamma oscillations were normalized by dividing the bins by the total number of gamma oscillatory episodes within a given recording. In this analysis, and in all analyses involving oscillation phase, the oscillation peak was defined as 0°.

Time-frequency representation of power across individual theta cycles.

Time-varying power in 2-Hz-wide frequency bands, from 2 Hz to 140 Hz, was obtained for individual theta cycles using the wavelet transform method described above. Time frequency representations for multiple theta cycles recorded from the same site and session within the same animal were then averaged using the theta phase, and then normalized using the total average power in the individual 2-Hz bin across the phase (Schomburg et al., 2014).

Strength of theta-gamma coupling.

The theta phase at the time of gamma oscillation maxima occurrence was calculated by bandpass-filtering the LFP in the theta range, performing a Hilbert transform on the filtered signal, and then locating theta phase of individual gamma oscillation maxima. Resultant vectors were calculated from the phase distributions of gamma maxima. Lengths of resultant vectors were used as an index for the strength of theta-gamma coupling.

Phase-locking of spikes to gamma oscillations and theta oscillations.

Neurons recorded in animals with single-ended LFP recordings were used for the analysis. The oscillatory phase at the time of spike occurrence was estimated. This was performed by first bandpass-filtering the LFP, performing a Hilbert transform on the filtered signal, and then extracting the phase component at the spike times. Cells were considered to be phase-locked if their phase distribution differed significantly from a uniform distribution (P < 0.05, Rayleigh test). Resultant vectors were calculated from the distributions of phase component. Lengths of resultant vectors were used as an index for the strength of phase-locking.

Coherence.

Coherence was computed using a multitaper method from the Chronux open source MATLAB toolbox (http://chronux.org/). We used 3 for time-bandwidth product and 5 for the number of tapers.

MEC:CA1 power ratio.

All 200 ms time windows of LFP with fast gamma episodes in CA1 were extracted and averaged. Corresponding LFP time windows from the MEC were then extracted and averaged. Power of fast gamma oscillations were calculated from these averaged traces, and then MEC:CA1 power ratio was calculated.

HISTOLOGY AND RECONSTRUCTION OF RECORDING POSITIONS

Electrode positions were confirmed by anaesthetizing the drive-implanted mice using isoflurane and performing small electrolytic lesions through passing current (10 μA for 20 s) through the electrodes. Immediately after this, the mice received an overdose of isoflurane and were perfused intracardially with saline followed by 4% freshly depolymerized paraformaldehyde in phosphate buffer (PFA). The brains were extracted and stored in the same fixative overnight. After overnight cryoprotection in phosphate buffer saline with 30% sucrose at 4°C, tissue samples were embedded in O.C.T. mounting medium and sagittal sections (40 μm) were cut and stained with cresyl violet. All tetrodes were identified, and the tip of each electrode was found by comparison with adjacent sections. Only data from tetrodes in the hippocampal CA1 or the layer 2/3 of dorsal MEC was collected for analysis. The electrical lesions often made tip holes in the brain section larger so that they span across cell layers. To ensure that the LFP was recorded from cell layers (either in pyramidal cell layer in CA1 or layers 2/3 in the MEC) we used LFP data only from tetrodes that had spiking activity of principal neurons for LFP analyses.

For immunostaining, sections were rinsed 3 times for 10 min in 1 x PBS (pH 7.6) at room temperature, preincubated for 1 hr in 10% normal goat serum in PBST (1 x PBS with 0.5% Triton X-100). Between incubation steps, sections were rinsed in PBST. Sections were incubated with antibodies against Aβ, raised in mouse (MCSA1, Medimabs, 1:1000) for 24 hr in antibody-blocking buffer at 4°C. After three 15 min washes in PBST at room temperature, sections were incubated in a goat-anti mouse antibody conjugated with Alexa Fluor 568nm (Invitrogen, 1:250) for 2 hr at room temperature. After rinsing in PBS, sections were mounted onto glass slides with 40 ,60 -diamidino-2-phenylindole (DAPI)-containing Vectashield mounting medium (Vector Laboratories), and a coverslip was applied. Digital photomicrographs were acquired with an Olympus BX53 fluorescent microscope equipped with a digital camera.

For amyloid beta plaque counting analysis, 500μm by 500 μm block images representing the recorded areas in CA1 or MEC were taken from the original section. All blocks were taken from sagittal sections. For CA1, four blocks from each animal were taken spanning 2.2 – 3.2mm lateral to midline. For the medial entorhinal cortex, four blocks from each animal were taken spanning 2.8 – 3.8mm lateral to midline. Blocks were taken from dorsal aspect of each region and they were chosen to reflect our recording positions. The areas of pixels with fluorescence values higher than two standard deviations of background were quantified for all blocks using Image J software, and then divided by the whole area of the image for percent Aβ plaque. Data from eight blocks from two mice were used for each area/genotype/age. All analyses were performed in a blind manner.

CONTEXT DISCRIMINATION TASK

We used a context discrimination task for mice used in McHugh et al (2007). Four male and four female mice from each group (WT and APP-KI) between 3–5 months old for young group and 12–13 months old for old group were used in this experiment. All experiments were conducted and analyzed by scientists blind to the genotypes of the animals. The mice were caged in single housing at least one week prior to the experiment to become acclimatized. The mice were housed in plastic tubs and had ad libitum access to food and water and lived on a 12:12 hour dark/light cycle. All procedures occurred during the dark cycle. During fear conditioning, the mice received three days of conditioning in Context A before the discrimination phase of the task began, allowing for greater generalization in both genotypes of mice. Context A chamber consisted of black Plexiglas side walls while the front door, back wall, and ceiling were made of clear Plexiglas. The floor of each chamber consisted of 33 stainless steel rods, separated by 6 mm, which were wired to a shock generator and scrambler. A stainless steel pan coated with isobenzaldehyde in 100% alcohol (0.25% concentration) was placed under the grid floor in each box to provide a distinct odor. Each chamber was cleaned thoroughly with an odorless 5% sodium hydroxide solution before the animals were placed in the chambers. The overhead fluorescent room lights remained on. Context B chamber consisted of Plexiglas side walls sloped inward at a 60 degrees angle from the floor (28 × 21 × 21 cm). As in context A, the floor of each chamber consisted of 33 stainless steel rods, separated by 6 mm, which were wired to a shock generator and scrambler. This context was cleaned and scented with a 1% acetic acid solution. The room was lit with a 30-W red overhead light and a 30-W red light located in the corner of the room opposite the chambers.

Fear Conditioning in Context A.

Each day, the animals were transported to a room adjacent to the experimental room. They were left undisturbed for at least 20 min. On days 1–3 the mice were carried to the A-context conditioning room in their home tub and placed into the conditioning chambers. After 192sec, they received a single foot shock (2 sec; 0.65mA) and were removed from the chambers 1min following foot shock termination.

Familiarization.

Across the subsequent two consecutive days (day 1 and 2 of familiarization), mice were placed into the A-context and B-context conditioning chambers in separate tests (counterbalanced order). Each session consisted of an 8-minute exposure to the chamber without the delivery of foot shock.

Discrimination training.

On days 1 through 5 of discrimination training, mice were exposed to both A-context and B-context conditioning chambers daily. The order of exposure on each day followed a BAABA design such that on days 2, 3 and 5 all animals were exposed to Chamber A first and Chamber B second. Across the entire discrimination phase, all animals received a single foots hock during each Chamber A exposure and never received foot shock during Chamber B exposures. The dependent measure employed was freezing behavior, defined as behavioral immobility except for movement necessary for respiration (Fanselow et al., J Biol Sci 1980). An observer, blind to the genotypes of the mice, scored each mouse as either freezing or not freezing every 8 s for the duration of each 192 s in each context on each day during spatial discrimination training. These scores were then converted into a percentage of observations spent freezing. Discrimination Index were calculated for discrimination training using these freezing percentage scores according to the following formula: (Chamber A − Chamber B)/(Chamber A + Chamber B).

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical testing assumed a non-parametric distribution and the Wilcoxon rank sum test was used. For comparing the distribution of place cells and grid cells, Kolmogorov-Smirnov (KS) tests were applied. All statistical methods used are summarized in Supplementary Tables S2. All data are represented as mean ± SEM.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-α-Aβ	Medimabs	Cat#: MCSA1
Goat anti-Mouse IgG Alexa Fluor 568 conjugated	Invitrogen	Cat#: A-11004
Experimental Models: Organisms/Strains
C57BL/6J	Jackson Laboratory	000664
APP Knock-in	RIKEN Bio Resource Center Saito et al., 2014	RBRC06344
Software and Algorithms
Custom-built MATLAB analysis programs	MATLAB	https://www.mathworks.com/
Image J	NIH	https://imaeej.nih.eov/
Other
Digital Lynx SX	Neuralynx	Digital Lynx SX-M
.0007 inch diameter Platinum 10% Iridium	California Fine Wire	100167
Customized 64-channel drive	Igarashi Lab	N/A
Chambers for Contextual Fear Conditioning	Med Associates	https://www.med-associates.com/

Highlights.

• Remapping of place and grid cells was investigated in APP knock-in mice

• Place cells were mildly deteriorated but showed severely disrupted remapping

• Grid cells were severely impaired, leaving with few neurons with ability to remap

• Disruption of grid cells, but not place cells, emerged in younger APP knock-in mice