Abstract
The retrosplenial cortex (RSC) plays a significant role in spatial learning and memory and is functionally disrupted in the early stages of Alzheimer's disease (AD). In order to investigate neurophysiological correlates of spatial learning and memory in this region we employed in vivo electrophysiology in awake and freely moving male mice, comparing neural activity between wild-type and J20 mice, a transgenic model of AD-associated amyloidopathy. To determine the response of the RSC to environmental novelty local field potentials (LFPs) were recorded while mice explored novel and familiar recording arenas. In familiar environments we detected short, phasic bursts of β (20–30 Hz) oscillations (β bursts), which arose at a low but steady rate. Exposure to a novel environment rapidly initiated a dramatic increase in the rate, size and duration of β bursts. Additionally, θ-α/β cross-frequency coupling was significantly higher during novelty, and spiking of neurons in the RSC was significantly enhanced during β bursts. Finally, excessive β bursting was seen in J20 mice, including increased β bursting during novelty and familiarity, yet a loss of coupling between β bursts and spiking activity. These findings support the concept that β bursting may be responsible for the activation and reactivation of neuronal ensembles underpinning the formation and maintenance of cortical representations, and that disruptions to this activity in J20 mice may underlie cognitive impairments seen in these animals.
SIGNIFICANCE STATEMENT The retrosplenial cortex (RSC) is thought to be involved in the formation, recall and consolidation of contextual memory. The discovery of bursts of β oscillations in this region, which are associated with increased neuronal spiking and strongly upregulated while mice explore novel environments, provides a potential mechanism for the activation of neuronal ensembles, which may underlie the formation of cortical representations of context. Excessive β bursting in the RSC of J20 mice, a mouse model of Alzheimer's disease (AD), alongside the disassociation of β bursting from neuronal spiking, may underlie spatial memory impairments previously shown in these mice. These findings introduce a novel neurophysiological correlate of spatial learning and memory, and a potentially new form of AD-related cortical dysfunction.
Introduction
The retrosplenial cortex (RSC) is considered to play a critical role in spatial learning and memory. Damage to this region results in severe impairments in navigation and landmark processing (for review, see Mitchell et al., 2018). There is a large body of experimental evidence suggesting the RSC is involved in the encoding, retrieval, and consolidation of spatial and contextual memory (for review, see Todd and Bucci, 2015). Optogenetic stimulation of RSC neurons is sufficient to elicit retrieval and consolidation of contextual memories (Cowansage et al., 2014; De Sousa et al., 2019). RSC neurons encode a range of contextual information during navigation, and inactivation of the RSC using glutamate receptor antagonists impairs performance in the Morris water maze and contextual fear memory tasks (Czajkowski et al., 2014; Kwapis et al., 2015), suggesting the RSC is involved in the storage of spatial information. Finally, Iaria et al. (2007) demonstrated that while hippocampal subregions are differentially involved in the encoding and retrieval of spatial information, the entire RSC is active during both processes. Spatial learning and memory impairments have been shown to be one of the earliest aspects of cognitive impairment in Alzheimer's disease (AD). Patients exhibit disturbances in specific spatial memory processes associated with the RSC (Laczó et al., 2009; Vann et al., 2009; Morganti et al., 2013). During the early stages of AD, the retrosplenial gyrus has been shown to exhibit regional hypometabolism (as measured by FDG-PET), and considerable atrophy (Minoshima et al., 1997; Choo et al., 2010). As such, the RSC is a region of great interest in research into the brain's function in health and AD.
Measurable correlates of brain function can have great value in fundamental neuroscience. They can aid the understanding of the complex ways in which the brain processes information and performs its many tasks and indicate how such functionality may be affected in disease. Similarly, these “functional biomarkers” can provide measurable benchmarks against which to test interventions which may affect or restore normal brain function (Walsh et al., 2017). Of the growing number of methodologies available for investigating brain function, in vivo electrophysiology remains a powerful tool with a superior temporal resolution to all others. The coordinated firing of large groups of neurons in the brain gives rise to waves of electrical activity, known as neural oscillations, which can be recorded as intracranial local field potentials (LFPs) or extracranial electroencephalograms (EEGs). It is thought that one of the roles of these oscillations in the brain is to coordinate the spiking activity of neurons, allowing computation and communication between potentially distant brain regions (Canolty et al., 2010). The temporal resolution of electrophysiology combined with the spatial specificity afforded by intracranial recordings make in vivo electrophysiology an invaluable tool for discovering functional correlates of brain function and disease-associated dysfunction.
In order to investigate the function of the RSC in spatial learning and memory, we recorded LFPs and multiunit spiking activity from this region, while mice freely explored either a novel or familiar environment. To probe the effects of AD-associated amyloid pathology on RSC function, we used J20 mice, a widely employed mouse model of amyloidopathy. In this paper, we describe short, phasic bursts of β (20–30 Hz) oscillations, termed “β bursts,” that occur within the RSC, while mice freely explore an environment. β Bursting activity is significantly increased during exposure to a novel environment, and these bursts are correlated with increased neuronal spiking in the RSC. These data demonstrate that β bursting in the RSC is a robust neurophysiological correlate of environmental novelty and may have a role in the storage and retrieval of cortical spatial representations. Finally, we observed excessive β bursting activity and an uncoupling of β bursting from neuronal spiking in the RSC in J20 mice, which may disrupt its function, and underlie spatial learning and memory deficits seen in these mice (Cheng et al., 2007).
Materials and Methods
Ethics
All procedures were conducted in accordance with the United Kingdom Animal (Scientific Procedures) Act 1986 and were approved by the University of Exeter Animal Welfare and Ethical Review Body.
Animals
Eight male J20 mice and five wild-type (WT) littermates were bred in-house at the University of Exeter and housed on a 12/12 h light/dark cycle. This sample size was estimated from a similar previous study performed by one of the authors (Ahnaou et al., 2019), and proved to be sufficient for this study. J20 mice were bred on a C57BL/6 background. Access to food and water was provided ad libitum. All mice underwent surgery at between six and eight months of age. Mice were group housed before surgery, and single housed postsurgery, to prevent damage to the surgical implants.
Surgery
Mice were unilaterally implanted with a 16 channel, single shank silicon probe (NeuroNexus Technologies, A1x16-5 mm-100-177-CM16LP), in the right RSC (AP –2 mm, ML +0.5 mm, DV +1.75 mm, 0° Pitch). Mice were anaesthetized using isoflurane and fixed into a stereotaxic frame. A small craniotomy was drilled over the desired co-ordinate, and at least one hole was drilled in each of the major skull plates, in which miniature screws were placed to act as supports (Antrin Miniature Specialties). The probe was slowly lowered into the desired location, and fixed in place with dental cement (RelyX Unicem, 3M). The ground wire from the probe was connected to a silver wire, attached to a screw overlying the cerebellum. Throughout surgery, body temperature was monitored with a rectal probe and regulated by a feedback-controlled heat mat. Animals were kept hydrated by subcutaneous injections of Hartmann's solution once per hour of surgery (0.01 ml/g body weight).
Behavior
After at least one week of postsurgical recovery, animals underwent a novel/familiar environment task, as shown in Figure 1. Individual mice were tethered to the recording apparatus and placed in one of two high-sided recording arenas: one square, with black and white stripes, and one circular and lacking stripes. Both arenas each had two internal visual cues, placed on opposite sides. The animals were allowed to freely explore their environment for 15 min, after which, they were returned to their home cage. After 15 min in their home cage, the animal was returned to the same recording arena for another 15 min and allowed to freely explore. Following this, the animal was returned to its home cage. This protocol was repeated at the same time of day, for five consecutive days, but on the fifth day, the animal was placed in the other, previously unseen arena. The order of exposure to these arenas was counterbalanced between animals. Each session can therefore be described by the experimental day, and the particular session within that day, with session A being the first, and session B being the second. Using this nomenclature, sessions 1a and 5a were “novel” sessions, while the remaining sessions were “familiar” sessions. In order to reduce the stress associated with the recording process, animals were acclimatized to this process during 10 min test session 3 d before the start of the experiment, in which the animal was tethered and recorded from while in its home cage. An added benefit of this was to familiarize the animals with this experimental procedure, thus ensuring that perceived novelty during the first experimental session was limited to the environment, and not the experience of being tethered to the recording apparatus.
Data collection
Throughout experimental sessions, LFPs were recorded using an Open Ephys Acquisition board (Open Ephys), which was tethered to the probe via a headstage (RHD 16-Channel Recording Headstage, Intan Technologies), and SPI cables (Intan Technologies). LFPs on each channel were sampled at 30 kHz, while the animal's location was monitored using a pair of light-emitting diodes (LED) soldered to the headstage, and a video camera (Logitech HD Pro Webcam C920, Logitech), placed directly above the arena. The location of these LEDs was tracked at 30 frames per second using Bonsai tracking software, so the location and running speed of the animal could be estimated offline. To reduce noise, position information was smoothed using a Savitzky–Golay filter of order three and running speed was calculated on a second-by-second basis.
Data analysis
LFPs were down-sampled [spectral analysis: 1 kHz, burst detection and phase-amplitude coupling (PAC): 3 kHz, multiunit activity: 30 kHz] and de-trended, to remove any slow linear drift of the baseline that may occur across the session. The Chronux toolbox (Mitra and Bokil, 2008; http://chronux.org/) was used for the mtspecgramc function, as well as the CircStat toolbox for circular statistics (Berens, 2009). Several built-in MATLAB functions were used as well as some functions from the MATLAB File Exchange including shadedErrorBar (Campbell, 2022) and polyfitZero (Mikofski, 2021). All other scripts used in this study were written in-house and will be made publicly available (see below, Software accessibility). All LFP analyses were performed for a single channel located in the center of the dysgranular RSC (RSCdg). In order to select channels which were at equivalent depths between animals, a combination of post hoc histology and functional outputs were used to determine the exact locations of each channel. One functional output makes use of the fact that the phase of θ oscillations reverses across the pyramidal cell layer of the hippocampus, so for probes which reached or crosses this layer, it is possible to work backwards to find the exact location of each channel to <100 µm.
Power spectra
Multitaper spectral analysis was performed using the mtspecgramc function from the Chronux Toolbox, with a time-bandwidth product of 2 (1 s × 2 Hz), and three tapers, resulting in some smoothing of resulting spectra. The mtspecgramc function generates a power spectrogram by generating multiple power spectra for short segments of time series data, using a moving window; in our case with the window size of 1 s with no overlap. These spectrograms were then logged to the base 10, and multiplied by 10, to correct for the tendency of spectral power to decrease with a 1/f distribution. These individual spectra were averaged, resulting in a single mean power spectrum for the entire session, or for the first minute of each session, as specified in the results. Spectral data from 48 to 52 Hz, which incorporates line frequency noise (50 Hz), were removed, and linearly interpolated. The power of each frequency band was calculated as the mean power in each of the following frequency ranges: δ (1–5 Hz), θ (5–12 Hz), α (12–20 Hz), β (20–30 Hz), and γ (30–100 Hz). Wavelet analysis was performed using the cwt function in MATLAB, with the Morlet wavelet with equal variance and time and frequency. The scale to frequency conversions are set by the sampling rate of 30 kHz.
β Burst detection
The data were bandpass filtered between 20 and 30 Hz, to isolate the β frequency band, using a Butterworth IIR filter with an order of 2. The amplitude and phase of this β signal were calculated as the real and imaginary components of the Hilbert transform, respectively. The amplitude was Z scored, to give the instantaneous standard deviation (σ/SD) of the β signal amplitude from the mean. Epochs of the signal where this z score exceeded 2 SDs from the mean amplitude were detected, as were the corresponding “edges” of these epochs, where the signal magnitude surpassed 1 SD either side of the 2 SD threshold. This was done to capture the time course of these high β amplitude epochs. Events that did not persist longer than a minimum duration of 150 ms (i.e., fewer than three oscillation cycles) were discarded. Furthermore, because of the sensitivity of this method to large, amplitude noise artefacts, any event whose peak amplitude exceeded three scaled median absolute deviations from the median of the events detected in that session were discarded as well. These remaining events were then considered β-bursts. The duration and peak magnitude of each burst was calculated, as well as the distribution and total number of bursts in the session. Running speed during β bursts was estimated as the animal's instantaneous running speed at the time of the β burst.
Phase-amplitude coupling
To calculate phase-amplitude coupling (PAC), and create PAC comodulograms, modulation index (MI) was calculated individually for each pair of phase and amplitude frequencies as described by Tort et al. (2009). This method has been shown to be superior to alternative methods and is less sensitive to changes in amplitude. A full explanation of this method can be found in (Tort et al., 2009), but will be briefly explained here. PAC was calculated between phase frequencies in bins of 0.25 Hz from 2 to 12 Hz, and amplitude frequencies in bins of 2 Hz from 10 to 100 Hz. For each pair, LFPs were filtered in the phase frequency band and the amplitude frequency band, using a Butterworth IIR filter with an order of 2, after which the instantaneous phase and amplitude of each filtered signal was calculated, respectively, using the Hilbert transform. The phases of the “phase signal” were binned in 10° bins, and the average amplitude of the “amplitude signal” was calculated for each phase bin, after which this “amplitude distribution” was normalized so that the sum of all bins is equal to 1. The existence of PAC can be seen in these amplitude distributions as a nonuniform amplitude across the phase bins, and as such, the Kullback–Leibler distance was calculated to quantify the divergence of this amplitude distribution from the uniform distribution (Kullback and Leibler, 1951). In order to convert to Kullback–Leibler distance to MI, with a scalar value between 0 and 1, this value is divided by the natural logarithm of the number of phase bins, which in this case is 18. Although this method is far less sensitive to spurious coupling than other methods, we still normalized the resulting MI. This was done by the generation of 100 surrogates, where the data were time shifted by a random amount between 1 and 59 s, for which the MI was calculated. A Gaussian distribution was then fitted to these surrogate MIs and the actual MI was calculated as a z score from the mean of this distribution. This same mathematical operation was performed for all phase and amplitude frequency pairs to create a comodulograms, and to smooth the resulting comodulograms, the data matrix was linearly interpolated in both dimensions by a factor of 2.
Multiunit activity
Because of the distance between adjacent channels on the recording probe (100 µm) it is highly unlikely that activity of a single neuron would appear on multiple channels. Consequently, each channel was treated as an individual multiunit. Raw LFPs were first common average-referenced, using a mean of the signals from all other 15 channels, then filtered in the range of 500–14,250 Hz, using a Butterworth IIR filter with an order of 4, to isolate the spiking frequency band. Spikes were detected as peaks that crossed a threshold given by the median of the absolute voltage values of the signal, multiplied by 0.6745, as suggested by Quiroga et al. (2004), and had a minimum separation of 0.5 ms. In order to investigate multiunit activity during β bursts, bursts were detected as previously mentioned, and bursts that occurred within a second of each other were discarded, to remove overlapping segments. Peri-event histograms were created by counting the total number of spikes in 50-ms time bins from 0.5 s before burst onset, to 0.5 s after, for all β bursts. Each histogram was then normalized by dividing the count in each bin by the total number of spikes in all bins, averaged across all β bursts, and then across all sessions and Z scored with respect to the baseline epoch (0.5 s preburst). Potential phase-locking of spikes to β oscillations was investigated using circular statistics in a manner similar to Siapas et al. (2005). All spikes in a session were binned depending on the instantaneous phase of the β oscillation at which they occurred and then counted to produce a phase-distribution histogram. The Rayleigh test was then performed to statistically test for nonuniformity in these distributions, which would be indicative of phase-locking to a specific phase of the β oscillation. Rayleigh's Z statistic gives the significance level of this test, and any sessions with a Z statistic equivalent to p < 0.05 were considered to demonstrate significant β phase-locking. The µ and κ parameters were estimated from the von Mises distribution to determine the preferred phase and concentration (strength) of this β phase-locking, respectively.
Software accessibility
All code has been made publicly available at https://github.com/cfle/In-Vivo-Ephys-Code. This code is freely accessible for viewing, or use. If using any of this code in a paper, please, cite this paper as well as the GitHub repository (https://github.com/cfle/In-Vivo-Ephys-Code).
Statistics
All statistical analysis was performed in MATLAB. Thirteen mice in total were used in this study, five wild-type and 8 J20, with each mouse undergoing a total of ten recording sessions (5 d, two sessions per day). Unfortunately, the LFP data from day 3 session 1 (i.e., session 3a) was corrupted for a single wild-type mouse, and therefore data for this mouse from this session was omitted from all figure making and statistics. Therefore, the n numbers for all statistics are [wild-type: n = 5 (except from day 3a where n = 4), J20: n = 8]. All statistical analysis was performed in MATLAB using several different built-in functions. Statistical analysis varied depending on the type of analysis performed, however most of the statistical analysis was performed using mixed ANOVA with varying number of factors. The novel/familiar environment task involved two novel sessions and eight familiar sessions, so to account for this imbalance, data were averaged across all novel and all familiar sessions. For most analyses, mixed ANOVAs had two factors, with genotype as the between-subjects factor, and novelty as the within-subjects factor. Other additional factors included region or age. Significant main effects or interactions from an ANOVA was subsequently followed up with relevant planned comparisons. Statistical tests used for each analysis are noted alongside the results of that analysis, throughout this paper.
Histology and amyloid plaque staining
Upon completion of the experiments, mice were killed using an overdose of sodium pentobarbital (Euthetal), and an isolated stimulator was used to produce electrolytic lesions at the recording sites. Mice were then transcardially perfused with 4% paraformaldehyde (PFA), and their brains were extracted and stored in PFA for 24 h, after which they were transferred to PBS before sectioning. Brains were sliced into 100-µm sagittal sections using a vibratome (Leica), and stained with Cresyl Violet. Digital pictures were taken using QCapture Pro 7 software (Qimaging), and electrode sites were verified by comparing the lesion sites in these photographs to The Allen Mouse Brain Atlas (https://mouse.brain-map.org/static/atlas). Because of the high channel count of these probes, as well as their linear geometry, it was possible to account for small differences in the depth of each probe by selecting channels of similar depths across different probes. This resulted in reduced variability between animals in a range of neurophysiological measures.
Amylo-Glo staining was performed on formalin fixed slices according to the recommended protocol. 70% ethanol solution was applied to the slices for 5 min, after which the slices were rinsed in distilled water for 2 min. 100× Amylo-Glo stock solution was diluted 1:100 using 0.9% saline solution, and slices were then incubated in this 1× solution for 10 min. After this they were rinsed in 0.9% saline solution for 5 min, and then distilled water for ∼15 s, before cover slipping with Dako Fluorescence Mounting Medium (Dako). Imaging of plaques was performed on a confocal microscope (ThorLabs), or a Nikon Eclipse E800 Fluorescence Microscope (Nikon).
Parvalbumin staining
Mice were terminally anaesthetized with an intraperitoneal injection of sodium pentobarbital (Euthetal) before being transcardially perfused (5 ml/min) with PBS followed by 4% PFA in PBS. Following the perfusion the brains were extracted and stored in 4% PFA for 22 h at 4°C, then cryoprotected in 30% Sucrose in PBS-Azide (PBS, 0.02% sodium azide) for at least 3 d. Using a freezing sledge microtome (Leica SM2010R with Physitemp BFS-5MP temperature controller) 30-µm coronal sections were taken from frozen brains (−20°C) and stored in cryoprotectant solution (25% glycerol, 30% ethylene glycol, 25% 0.2 m phosphate buffer, 20% ddH2O) at −20°C.
For parvalbumin staining, all steps were conducted at room temperature unless stated otherwise; 30-µm free-floating sections stored in cryoprotectant solution were washed in PBS (3 × 10 min), before being incubated in PBS with 0.09% hydrogen peroxide for 20 min to quench endogenous peroxidase. Next, sections were washed in PBS (3 × 10 min), then blocked and permeabilised in PBS-Tx (PBS, 0.2% Triton X-100) with 3% normal goat serum (NGS; Vector Laboratories, S-1000). The sections were then incubated in 1:5000 rabbit anti-PV primary antibody (Swant, PV27) in PBS-Tx with 3% NGS at 4°C overnight. The following day, the sections were washed in PBS (3 × 10 min), then incubated for 2 h in 1:600 goat biotinylated anti-rabbit secondary antibody (Vector Laboratories, BA-1000) in PBS-Tx with 1% NGS. Sections were washed in PBS (2 × 10 min), then incubated in an Avidin-Biotin complex solution (Vector Laboratories, PK-4000) for 1 h. After a further three washes in PBS, the sections were incubated with 0.04% 3,3'-diaminobenzidine-tetrahydrochloride (DAB; Hello Bio, HB0687) with 0.04% hydrogen peroxide and 0.05% ammonium nickel(II) sulfate (Merck Life Sciences, A1827) in PBS for ∼10 min. After a final two washes in PBS, sections were mounted on Superfrost Plus slides (Fisherbrand) and left to dry overnight. The next day, sections were serially dehydrated in graded ethanol baths and then cleared in Histo-Clear II (Scientific Laboratory Supplies, NAT1334) for 20 min. Slides were then sealed and coverslipped using Histo-Mount mounting medium (Scientific Laboratory Supplies, NAT1310).
Semi-automated cell counting was performed using Fiji (Schindelin et al., 2012). Briefly, images at 10× magnification were locally thresholded using the Sauvola method, to account for potential differences in brightness across the image. The image was then despeckled and eroded to remove noise, and cells were automatically counted using the Analyze Particles function. These results were manually checked to remove clearly spurious detections, and then the number of cells in each subregion of the RSC were counted. Analysis was performed fully blinded with regards to genotype.
Results
To investigate neurophysiological correlates of spatial learning and memory in the RSC, LFPs were recorded from across the entire dorsoventral axis of the RSC, while animals underwent a novel/familiar environment task. The RSC is made up of two distinct subdivisions: dysgranular (RSCdg) and granular (RSCg); however, for this study, we focused on recordings from the RSCdg (Fig. 1C, channel shown in red).
Experimental design. A, Diagrams of the recording arenas used for this study. Both are roughly equal sized, one is square, with black and white stripes along the walls and floor (left) and the other is cylindrical with plain brown floor and walls. B, Experimental procedure for the novel/familiar environment task. A mouse is placed in one of the recording arenas for two 15-min sessions, referred to as sessions A and B, with a 15-min break in their home cage between the two sessions. This is repeated in the same arena for four consecutive days, after which the arena is switched for the fifth and final day. C, Single shank, 16 channel silicon probe electrodes were implanted in the retrosplenial cortex (green), so that they spanned the dysgranular (upper green section) and granular (lower green section) subregions. In order to verify the location of the electrodes, electrolytic lesions were made before perfusion, and slices were histologically prepared using Cresyl violet stain. An example is shown (right).
Spectral analysis
LFPs from RSCdg show a clear peak in θ frequency band (5–12 Hz) throughout recording sessions (Fig. 2A), as well as smaller peaks at higher frequencies. In order to investigate any changes in oscillatory activity in RSCdg during environmental novelty, power spectral analysis was performed on the entire 15 min of each session. These power spectra were averaged across novel and familiar sessions for wild-type and J20 mice. γ Power was significantly higher overall during novel sessions (γ: main effect novelty, F(1,11) = 21.6, p = 7e-4, mixed ANOVA). γ Power was significantly higher during novel sessions in wild-type (Nov: 10.6 ± 0.1 dB; Fam: 10.2 ± 0.1 dB, p = 0.01) and J20 (Nov: 11.2 ± 0.2 dB; Fam: 10.8 ± 0.2 dB, p = 0.004) mice. There were significant interactions between the effects of genotype and novelty on δ, α, and β power (δ: interaction, F(1,11) = 9.4, p = 0.01, mixed ANOVA; α: interaction, F(1,11) = 6, p = 0.03, mixed ANOVA; β: interaction, F(1,11) = 5.2, p = 0.04, mixed ANOVA). δ Power was significantly higher during familiar sessions in wild-type (Nov: 22.1 ± 0.7 dB; Fam: 22.5 ± 0.7 dB, p = 0.02) but not J20 mice. β Power was significantly higher during novel sessions in both wild-type (Nov: 14.2 ± 0.2 dB; Fam: 13.7 ± 0.2 dB, p = 0.02) and J20 (Nov: 16.7 ± 0.3 dB; Fam: 15.8 ± 0.3 dB, p = 2e-5) mice. Moreover, β power was significantly higher in J20 mice than in wild-type mice, for both novel (Nov, WT: 14.2 ± 0.2 dB; J20: 16.7 ± 0.3 dB, p = 0.001) and familiar sessions (Fam, WT: 13.7 ± 0.2 dB; J20: 15.8 ± 0.3 dB, p = 0.002). Upon closer inspection of power spectrograms (Fig. 2A), it was clear that spectral activity changed within novel sessions. Power in the α, β, and low γ range appeared to be higher in the first minute of the session and diminish over time. As exemplified in (Fig. 2C), transient epochs of high power in the 12–40 Hz range are seen throughout the early stages of the session. By performing the same power spectral analysis as before on only the first minute of each session, clear differences appeared between novel and familiar sessions. θ, α, β, and γ power were significantly higher overall during novel sessions (θ: main effect novelty, F(1,11) = 14.7, p = 0.003, mixed ANOVA; α: main effect novelty, F(1,11) = 24.3, p = 4e-4, mixed ANOVA; β: main effect novelty, F(1,11) = 47.5, p = 3e-5, mixed ANOVA; γ: main effect novelty, F(1,11) = 19.9, p = 0.001, mixed ANOVA). There was a significant interaction between the effects of genotype and novelty on δ power (interaction, F(1,11) = 8.3, p = 0.01, mixed ANOVA). δ Power was significantly higher during novel sessions in J20 mice (Nov: 22.8 ± 0.2 dB; Fam: 21.8 ± 0.2 dB, p = 0.006), but not wild-type mice. Moreover, α and β power were significantly higher overall in J20 mice (α: main effect genotype, F(1,11) = 7.2, p = 0.02, mixed ANOVA; β: main effect genotype, F(1,11) = 21.9, p = 7e-4, mixed ANOVA). In order to demonstrate that the effect of novelty on spectral power was consistent across both novel sessions (day 1a and day 5a), we statistically compared spectral power between these two sessions (Fig. 3). We found that there were no significant differences in spectral power between day 1a and day 5a for either wild-type or J20 mice, except for a small significant increase in α power in day 5a compared with day 1a in wild-type mice, when the entire session was analyzed (day 1a: 16.2 ± 0.3 dB; day 5a: 16.7 ± 0.3 dB, p = 0.04). This result demonstrates that the effects of novelty on spectral power are not specific to a single session but occur equally during both exposures.
β (20–30 Hz) power is significantly higher during novelty in the RSCdg in wild-type and J20 mice. A, Example power spectrogram for an entire novel session in a wild-type mouse. B, Average power spectra for the entire 15 min of all novel (N) and familiar (F) sessions, for wild-type and J20 mice. Frequency bands are marked with dashed lines (δ: 1–5 Hz, θ: 5–12 Hz, α: 12–20 Hz, β: 20–30 Hz, γ: 30–100 Hz). β Power was significantly higher during novel sessions in both wild-type (p = 0.02) and J20 (p = 2e-5) mice. Moreover, β power was significantly higher in J20 mice than in wild-type mice, during novel (p = 0.001) and familiar sessions (p = 0.002). C, Example power spectrogram shown in A, expanded to show the first 60 s of the session (before the white line). Short epochs of increased power in the 20- to 40-Hz range can be seen. D, Average power spectra for the first minute of all novel and familiar sessions, for wild-type and J20 mice. β Power was significantly higher overall during novel sessions (p = 3e-5) and was significantly higher overall in J20 mice (p = 7e-4; data shown as mean ± SEM, WT: n = 5, J20: n = 8).
Power spectral changes because of novelty are consistent across both exposures. A, Average power spectra for the entire 15 min of novel sessions day 1a and day 5a, for wild-type mice. Frequency bands are marked with dashed lines (δ: 1–5 Hz, θ: 5–12 Hz, α: 12–20 Hz, β: 20–30 Hz, γ: 30–100 Hz). There were no significant differences between day 1a and day 5a for power in any frequency band, except for a small increase in α power in day 5a compared with day 1a (p = 0.04). B, Average power spectra for the entire 15 min of novel sessions day 1a and day 5a, for J20 mice. There were no significant differences between day 1a and day 5a for power in any frequency band. C, Average power spectra for the first minute of novel sessions day 1a and day 5a, for wild-type mice. There were no significant differences between day 1a and day 5a for power in any frequency band. D, Average power spectra for the first minute of novel sessions day 1a and day 5a, for J20 mice. There were no significant differences between day 1a and day 5a for power in any frequency band (data shown as mean ± SEM, WT: n = 5, J20: n = 8).
Across these time series, increased β power occurred in brief, discrete epochs, as shown in the expanded power spectrogram in (Fig. 4A). This can also be seen clearly in β-filtered LFPs, where these periods of high β amplitude intersperse an otherwise very low-amplitude oscillation. In order to understand the timescale and frequency domains of these events, a continuous wavelet transform was performed using the Morse analytic wavelet, to investigate them further. As exemplified in (Fig. 4C), these individual events were short in duration, and peaked in the 20- to 30-Hz, β band.
Retrosplenial LFPs are marked by short, phasic increases in β power, referred to as β bursts. A, Example power spectrogram showing transient increases in β power. B, LFPs of data shown in A, both unfiltered (top), and filtered in the β band (bottom), with the envelope amplitude in blue for clarity. The β-filtered LFP shows clear epochs of high β amplitude, which intersperse a low-amplitude continuous β oscillation. C, Expanded trace of the dashed area shown in B (bottom), and a continuous wavelet spectrogram of this time series (bottom). Because of the high temporal resolution of wavelet-based methods, these periods of high β amplitude can be seen to be brief, only lasting around 100–200 ms.
β Bursting activity
In order to investigate this phasic β activity in more depth, an algorithm was written to detect these epochs of high β oscillatory amplitude, as described in the methods and illustrated in (Fig. 5A). With these transient epochs of high β power now classified as discrete “β bursts,” it is possible to compare this β activity between sessions. As shown in Figure 5B, there were significantly more β bursts detected overall during novel sessions (main effect novelty, F(1,11) = 20.9, p = 8e-4, mixed ANOVA). Furthermore, there were significantly more β bursts detected overall in J20 mice (main effect genotype, F(1,11) = 16.8, p = 0.002, mixed ANOVA). Furthermore, it is possible to investigate the distribution of β bursts within sessions. As shown in Figure 5C, right, during familiar sessions the rate of β busting was reasonably steady, as indicated by the linear relationship between time and burst number shown in the cumulative frequency plot, for both wild-type and J20 mice. During novel sessions, however, there was a high rate of β bursting during the first 1–3 min of the session, which gradually decreased over time to a steady rate (Fig. 5C, left).
β Bursting activity in the RSCdg is significantly higher during novelty. A, Diagram illustrating how β bursts were detected. B, Graph showing the average number of β bursts detected in RSCdg in each session, for wild-type (black) and J20 (green) mice. Novel sessions day 1a and day 5a are highlighted in blue for clarity. Significantly more β bursts were detected during novel sessions than during familiar sessions (p = 8e-4). Moreover, significantly more β bursts were detected overall in J20 mice (p = 0.002). C, Cumulative frequency graphs of β bursts detected in novel (left) and familiar sessions (right), for wild-type and J20 mice. While β bursting occurred monotonically during familiar sessions, during the first minute of a novel session, β bursting was substantially increased. D, Graphs showing β burst rate during novel (left) and familiar sessions (right), for wild-type and J20 mice. Burst rate was quantified for the initial minute of each session, and final 10 min. β Burst rate was significantly higher overall during the initial minute of novel sessions than during the final 10 min of novel sessions for both wild-type (p = 0.01) and J20 (p = 0.004) mice (data shown as mean ± SEM, WT: n = 5, J20: n = 8, *p < 0.05).
β Bursting rate during the initial part of the session (first minute) and the final part of the session (last 10 min), was calculated for each session and averaged across novel and familiar sessions (Fig. 5D). The rate of β bursting was significantly higher overall during novel sessions (main effect novelty, F(1,11) = 18.6, p = 0.001, mixed ANOVA), and also significantly higher overall during the initial part of recording sessions (main effect time, F(1,11) = 24.5, p = 4e-4, mixed ANOVA). During novel sessions, initial burst rate was significantly higher than final burst rate for wild-type (Nov initial: 13 ± 1.6 bursts per minute; final: 0.9 ± 0.1 bursts per minute, p = 0.01) and J20 (Nov initial: 14.1 ± 3.6 bursts per minute; final: 2.0 ± 0.4 bursts per minute, p = 0.004) mice. Furthermore, there was no significant difference between wild-type and J20 mice for initial burst rate or final burst rate (Nov initial, WT: 13 ± 1.6 bursts per minute; J20: 14.1 ± 3.7 bursts per minute, p = 0.8; Nov final, WT: 0.9 ± 0.1 bursts per minute; J20: 2 ± 0.4 bursts per minute, p = 0.08). During familiar sessions, initial burst rate was significantly higher than final burst rate for J20 mice (Fam initial: 5.1 ± 0.5 bursts per minute; final: 2.1 ± 0.2 bursts per minute, p = 1e-4), but not wild-type mice (Fam initial: 2.7 ± 0.5 bursts per minute; final: 1.4 ± 0.1 bursts per minute, p = 0.07). Furthermore, initial burst rate and final burst rate were significantly higher in J20 mice than in wild-type mice (Fam initial, WT: 2.7 ± 0.5 bursts per minute; J20: 5.1 ± 0.5 bursts per minute, p = 0.006; Fam final, WT: 1.4 ± 0.1 bursts per minute; J20: 2.1 ± 0.2 bursts per minute, p = 0.03).
β Burst characteristics
In order to attempt to understand the nature of retrosplenial β bursts, and the mechanisms which underlie them, several β burst characteristics were investigated. For each β burst, the duration and magnitude were calculated, as shown in Figure 6A. β Burst magnitude was significantly higher overall during novel sessions (main effect novelty, F(1,11) = 43.6, p = 4e-5, mixed ANOVA). As shown in Figure 6B, β bursts were significantly larger in magnitude during novel sessions in wild-type (Nov: 93.3 ± 3 µV; Fam: 78.5 ± 2.7 µV, p = 0.004) and J20 (Nov: 121 ± 4.4 µV; Fam: 102 ± 3.3 µV, p = 8e-5) mice. Moreover, β bursts were significantly larger in magnitude overall in J20 mice (main effect genotype, F(1,11) = 14.3, p = 0.003, mixed ANOVA). β Burst duration was also significantly higher overall during novel sessions (main effect novelty, F(1,11) = 28.1, p = 3e-4, mixed ANOVA). As shown in Figure 6C, β bursts were significantly longer in duration during novel sessions in wild-type (Nov: 190 ± 2.5 ms; Fam: 177 ± 1 ms, p = 0.003) and J20 (Nov: 189 ± 2 ms; Fam: 180 ± 0.9 ms, p = 0.004) mice.
β Burst characteristics in the RSCdg. A, Diagram illustrating how the magnitude and duration of β bursts were calculated. B, Graph showing the average β burst magnitude in RSCdg in each session, for wild-type and J20 mice. β Bursts were overall significantly larger in magnitude during novel sessions (p = 4e-5). Moreover, β bursts were also significantly larger overall in J20 mice (p = 8e-5). C, Graph showing the average duration of β bursts in RSCdg in each session, for wild-type and J20 mice. β Bursts were overall significantly longer in duration during novel sessions (p = 3e-4); however, there was no significant overall difference between β burst duration in wild-type and J20 mice. D, Average power spectra for β burst, and preburst epochs. β Bursts were associated with a large, significant increase in β power during β bursts (p = 7e-16; data shown as mean ± SEM, WT: n = 5, J20: n = 8).
In order to understand the frequency profile of β bursts, and to verify that these oscillations conformed to the β frequency band (20–30 Hz), power spectral analysis was performed on individual β bursts. As a control, these burst spectra were compared with power spectra of epochs of equal length directly before each burst. These power spectra were averaged across all bursts and “prebursts,” for wild-type and J20 mice (Fig. 6D). Overall, β bursts were associated with a large significant increase in β power (main effect burst, F(1,11) = 4811, p = 7e-16, mixed ANOVA), and smaller significant increases in α and γ power (α: main effect burst, F(1,11) = 169, p = 5e-8, mixed ANOVA; γ: main effect burst, F(1,11) = 46, p = 3e-5, mixed ANOVA). α, β, And γ power were significantly higher during β bursts in both wild-type (α: WT, preburst: 16.5 ± 0.4 dB; burst: 17.6 ± 0.3 dB, p = 3e-6; β: WT, preburst: 12.8 ± 0.3 dB; burst: 19.9 ± 0.3 dB, p = 9e-14; γ: WT, preburst: 8.3 ± 0.2 dB; burst: 8.6 ± 0.1 dB, p = 0.002) and J20 (α: J20, preburst: 17.7 ± 0.4 dB; burst: 18.6 ± 0.4 dB, p = 7e-7; β: J20, preburst: 15.2 ± 0.3 dB; burst: 22.3 ± 0.3 dB, p = 8e-15; γ: J20, preburst: 9.4 ± 0.2 dB; burst: 9.8 ± 0.2 dB, p = 1e-4) mice. Additionally, these findings confirm that these β oscillations are not the merely the result of a harmonic of θ oscillations.
Running speed analysis
In order to investigate behavioral responses to novelty in this novel/familiar environment paradigm, we calculated each animal's running speed from the tracking data. Many previous studies have shown that environmental novelty is associated with increased exploration in rodents, as reflected by increased locomotor activity and therefore higher average running speeds (Dellu et al., 1996; Stone et al., 1999; Kabbaj et al., 2000), which would suggest that exploration should decrease in familiar environments, and that an absence of this decrease may indicate that this familiar environment is being incorrectly perceived as novel. In order to test this, running speed was calculated from tracking data and averaged across the entire 15 min of both novel sessions (day 1a and day 5a), and the familiar sessions immediately following them (day 1b and day 5b; Fig. 7A, left). At the time of the second session in each environment, animals will have only spent a total of 15 min in that environment, so uncertainty about whether the environment is novel or familiar would be most likely to manifest in these sessions. Conversely, by day 4b animals will have spent a total of 2 h in the first arena, so they should be able to recognize it with relative ease. Average running speed was significantly lower during familiar sessions in wild-type (Nov: 6.8 ± 0.3; Fam: 5.6 ± 0.4, p = 0.04) but not J20 (Nov: 8.0 ± 0.5; Fam: 7.8 ± 0.7, p = 0.6) mice. Moreover, there was no significant overall difference between average running speed in wild-type and J20 mice (main effect genotype, F(1,12) = 3.1, p = 0.11, mixed ANOVA). As we have shown, neurophysiological responses to novelty are greatest during the first minute of recording sessions, so as before average running speed was also calculated for the first minute of both novel sessions (day 1a and day 5a), and the familiar sessions immediately following them (day 1b and day 5b; Fig. 7A, right). There was a significant interaction between the effects of genotype and novelty on average running speed (interaction, F(1,12) = 6.8, p = 0.02, mixed ANOVA). Average running speed was significantly lower during the first minute of familiar sessions in wild-type (Nov: 8.4 ± 0.5; Fam: 4.8 ± 0.6, p = 8.3e-4) but not J20 (Nov: 9.6 ± 1.0; Fam: 8.8 ± 1.0, p = 0.29) mice. Finally, average running speed during the first minute of familiar sessions was significantly higher in J20 mice than in wild-type mice (WT: 4.8 ± 0.6; J20: 8.8 ± 1.0, p = 0.04). These results suggest that J20 mice may have been experiencing some difficulty in discriminating whether their environment is novel or familiar
β Bursting in the RSCdg appears to be unrelated to the running speed of the animal. A, Graphs showing the average running speed of wild-type and J20 mice during novel sessions (day 1a and day 5a) and the first familiar session in each arena (day 1b and day 5b), averaged across the whole session (left), and the first minute of each session (right). Across the whole session, average running speed was lower during familiar sessions than novel sessions in wild-type (p = 0.036), but not J20 (p = 0.64), mice. Similarly, when only the first minute was considered, average running speed was lower during familiar sessions than novel sessions in wild-type (p = 8.3e-4), but not J20 (p = 0.29), mice. B, Graphs showing the relationship between the rate of β bursting and the animal's running speed, across a range of running speed bins (left), and pooled data with the individual slopes and Fisher Z-transformed correlation coefficients for each animal. The rate of β bursting was uncorrelated with running speed in wild-type mice; however, in J20 mice, there was a strong positive correlation (p = 0.0079). C, Graphs showing the relationships between the average magnitude of β bursts and the animals running speed (left), and pooled data with the individual slopes and Fisher Z-transformed correlation coefficients for each animal. β Burst magnitude was positively correlated with running speed for both wild-type (p = 1.6e-5) and J20 (p = 8e-5) mice. D, Graphs showing the relationships between the average duration of β bursts and the animals running speed (left), and pooled data with the individual slopes and Fisher Z-transformed correlation coefficients for each animal. β Burst duration was uncorrelated with running speed in wild-type mice; however, in J20 mice, there was a weak positive correlation (p = 0.018; data shown as mean ± SEM, WT: n = 5, J20: n = 8).
As we have shown, β bursts are more prevalent during novel sessions, and both larger in magnitude and longer in duration, however it is unclear whether β bursting varies with respect to the animal's movement speed. Hippocampal θ and γ power have been shown to be positively correlated with running speed, so an increase in running speed may result in increased β power during novel sessions (Chen et al., 2011; Ahmed and Mehta, 2012). In order to directly investigate the relationships between β bursting and running speed, we grouped β bursts based on the animals running speed at the time of the burst. First, to determine whether the prevalence of β bursts increases with increased running speed, the average number of β bursts detected in each running speed bin was divided by the average amount of time spent in each running speed bin, equivalent to the average β burst rate at each running speed (Fig. 7B, left). In wild-type mice, β bursts occurred at the same rate regardless of running speed (R = −0.31, p = 0.39), while in J20 mice β bursts increased in prevalence with increasing running speed (R = 0.78, p = 0.0079). Second, to investigate potential relationships between β burst characteristics and running speed, we calculated the average magnitude and duration of β bursts across a range of running speed bins (Fig. 7C,D, respectively). β Burst magnitude showed a clear linear relationship with running speed for both wild-type (R = 0.96, p = 1.6e-5) and J20 (R = 0.93, p = 8e-5) mice. Furthermore, the pooled data in (Fig. 7C, right) shows that while average β burst magnitude appeared to be higher overall in J20 mice, there was no significant difference in the slope or correlation coefficient of this relationship between the two genotypes (slope, WT: 0.72 ± 0.24 µV/cm/s; J20: 0.57 ± 0.16 µV/cm/s; t(11) = 0.53, p = 0.6; unpaired t test; R – WT: 0.44 ± 0.16; J20: 0.41 ± 0.12; t(11) = 0.2, p = 0.9; unpaired t test). Conversely, as shown in Figure 7D, left, while there was a weak positive relationship between running speed and β burst duration in J20 mice (R = 0.72, p = 0.02), this was absent in wild-type mice (R = 0.03, p = 0.9). Furthermore, the pooled data in (Fig. 7D, right) shows that there was no significant difference in the slope or correlation coefficient of this relationship between the two genotypes (slope, WT: 0.03 ± 0.28 µV/cm/s; J20: 0.48 ± 0.2 µV/cm/s; t(11) = −1.34, p = 0.2; unpaired t test; R, WT: 0.14 ± 0.08; J20: 0.15 ± 0.05; t(11) = −0.08, p = 0.9; unpaired t test). These data demonstrate increased β bursting in J20 mice may, at least in part arise from a combination of a trend toward increased running speeds in these animals, and a positive correlation between β bursting and running speed that is absent in wild-type mice. Conversely, while β burst magnitude is positively correlated with running speed in both wild-type and J20 mice, the slope of this relationship is equivalent in both animals, suggesting that the trend toward increased running speeds in J20 mice is unlikely to underlie higher average β burst magnitudes in these animals.
Phase-amplitude coupling
Phase-amplitude coupling (PAC) involves coupling between the amplitude of an oscillation and the phase of a lower frequency oscillation (Canolty et al., 2006). This interaction is generally thought to allow slow, large amplitude oscillations to coordinate faster, small amplitude local oscillations. θ-γ Coupling is the most well studied form of PAC but PAC has been previously demonstrated for a range of other oscillation frequencies (Canolty et al., 2006; Tort et al., 2009; Daume et al., 2017). We investigated PAC across a range of frequencies in this study to determine whether retrosplenial PAC was associated with contextual novelty. PAC efficacy was calculated for a range of phase and amplitude frequencies, and the effect of novelty and genotype determined. As shown in Figure 8A, there were two large peaks in these comodulograms: one between θ phase and γ amplitude, and another between θ phase and 12- to 30-Hz amplitude. This second peak did not conform to a single frequency band, and as such was treated as a composite of α and β frequency. The strength of PAC was quantified for θ-α/β and θ-γ coupling for each session (Fig. 8B). There was a significant interaction between the effects of genotype and novelty on θ-α/β coupling (interaction, F(1,11) = 8.9, p = 0.01, mixed ANOVA). θ-α/β Coupling was significantly higher during novel sessions for wild-type (Nov: 2.9 ± 0.1; Fam: 1.6 ± 0.1, p = 4e-4) but not J20 (Nov: 2.3 ± 0.2; Fam: 2 ± 0.2, p = 0.15) mice. There were no significant effects of novelty or genotype on θ-γ coupling (main effect novelty, F(1,11) = 0.2, p = 0.7, mixed ANOVA; main effect genotype, F(1,11) = 0.7, p = 0.4, mixed ANOVA). It is important to note that to focus on the most physiologically and behaviorally relevant part of the session, this analysis was performed for the first minute of each session. When the same analysis was performed on the last minute of each session, there was no effect of genotype or novelty on coupling on either θ-α/β coupling (main effect genotype, F(1,11) = 0.4, p = 0.56, mixed ANOVA; main effect novelty, F(1,11) = 4.6, p = 0.054, mixed ANOVA; Fig. 9B, left) or θ-γ coupling (main effect genotype, F(1,11) = 3.7, p = 0.08, mixed ANOVA; main effect novelty, F(1,11) = 0.2, p = 0.69, mixed ANOVA; Fig. 9B, right).
θ-α/β PAC is increased in the RSCdg during the first minute of novel sessions. A, Average comodulograms showing the strength of cross-frequency PAC in RSCdg during the first minute of novel and familiar sessions, for wild-type and J20 mice. Note the presence of two peaks in the θ-α/β and θ-γ ranges (the boundaries of which are denoted by the dotted lines). B, Average MI in the θ-α/β (left) and θ-γ (right) ranges, for each session, for wild-type (black) and J20 (green) mice. Novel sessions day 1a and day 5a are highlighted in blue for clarity. θ-α/β Coupling was significantly higher during novel sessions for wild-type (p = 0.01), but not J20 (p = 0.15), mice. There was no significant effect of genotype or novelty on θ-γ coupling (p = 0.4, p = 0.7, respectively; data shown as mean ± SEM, WT: n = 5, J20: n = 8).
θ-α/β PAC in the RSCdg is unaffected by novelty during the final minute of novel sessions. A, Average comodulograms showing the strength of cross-frequency PAC in RSCdg during the final minute of novel and familiar sessions, for wild-type and J20 mice. Note the presence of two peaks in the θ-α/β and θ-γ ranges (the boundaries of which are denoted by the dotted lines). B, Average MI in the θ-α/β (left) and θ-γ ranges (right), for each session, for wild-type (black) and J20 (green) mice. Novel sessions day 1a and day 5a are highlighted in blue for clarity. There was no significant effect of genotype or novelty on θ-α/β coupling (p = 0.56, p = 0.054, respectively) or θ-γ coupling (p = 0.08, p = 0.69, respectively; data shown as mean ± SEM, WT: n = 5, J20: n = 8).
Multiunit activity
In order to determine whether β bursting was associated with a change in neuronal firing, multiunit activity was investigated. Because of the linear geometry of the silicon probes, and the 100 µm distance between channels, it was not possible to reliably identify single unit activity, as activity from a single neuron was unlikely to appear on multiple channels, limiting spatiotemporal clustering methods such as those enabled by tetrodes or higher density silicon probes. Therefore, spikes appearing on a single channel could be from one or more nearby neurons. This, however, does mean that it is possible to treat each individual probe channel as a single multiunit, to facilitate investigation of the relationship between neuronal spiking activity and β bursting. As shown in Figure 10A, left, individual spike waveforms can be readily discerned, and there was no significant difference in the mean amplitude of these waveforms between wild-type (black) and J20 (green) mice (WT: −90.3 ± 6.4 µV; J20: −82 ± 7.7 µV; t(11) = −0.8, p = 0.5; unpaired t test). Furthermore, as shown in Figure 10A, right, there was no significant difference between average firing rate in wild-type and J20 mice (WT: 46.7 ± 10.1 Hz; J20: 43 ± 10.5 Hz; t(11) = 0.24, p = 0.8; unpaired t test). Given that β bursting in wild-type mice appears to be associated with increased neuronal spiking, it was of interest to investigate whether there was also a broad increase in the rate of neuronal spiking during this same time period. As shown in Figure 10B, during both novel (left) and familiar sessions (right), the rate of neuronal spiking was reasonably steady, as indicated by the linear relationship between time and spike number shown in the cumulative frequency plot, for both wild-type and J20 mice. The rate of neuronal spiking during the initial part of the session (first minute) and the final part of the session (last 10 min), was calculated for each session and averaged across novel (Fig. 10C, left) and familiar sessions (Fig. 10C, right). The rate of neuronal spiking was significantly higher overall during familiar sessions (main effect novelty, F(1,11) = 8.3, p = 0.02, mixed ANOVA), and also significantly higher overall during the final part of recording sessions (main effect time, F(1,11) = 21.4, p = 7e-4, mixed ANOVA). During novel sessions, final spike rate was significantly higher than initial spike rate for J20 mice (Nov initial: 29.1 ± 7.0; final: 41.9 ± 10.0, p = 0.002), but not wild-type mice (Nov initial: 37.7 ± 9.5 Hz; final: 46.3 ± 11 Hz, p = 0.06). Furthermore, there was no significant difference between wild-type and J20 mice for initial spike rate or final spike rate (Nov initial, WT: 37.7 ± 9.5 Hz; J20: 29.1 ± 7.0 Hz, p = 0.5; Nov final, WT: 46.3 ± 11 Hz; J20: 41.9 ± 9.9 Hz, p = 0.8). During familiar sessions, final spike rate was significantly higher than initial spike rate for both wild-type (Fam initial: 41.6 ± 9.3 Hz; final: 48.3 ± 10.2 Hz, p = 0.03) and J20 (Fam initial: 37.7 ± 9.8 Hz; final: 45.8 ± 11.2 Hz, p = 0.003) mice. Furthermore, there was no significant difference between wild-type and J20 mice for initial spike rate or final spike rate (Fam initial, WT: 41.6 ± 9.3 Hz; J20: 37.7 ± 9.8 Hz, p = 0.8; Fam final, WT: 48.3 ± 10.2 Hz; J20: 45.8 ± 11.2 Hz, p = 0.9). These data show that while the rate of β bursts is higher during novelty, the rate of neuronal spiking in the RSC is higher during familiarity.
Spiking activity in RSCdg is coupled to β bursting in wild-type mice, but disrupted in J20 mice. A, Average spike waveforms for multiunit activity in wild-type (black) and J20 (green) mice (left) and graph of average firing rate for detected multiunits across all sessions (right). There was no significant difference between the mean amplitude of spike waveforms in wild-type and J20 mice (p = 0.5). Moreover, there was no significant difference between average firing rate in wild-type and J20 (p = 0.8) mice. B, Cumulative frequency graphs of neuronal spikes detected in novel (left) and familiar sessions (right), for wild-type and J20 mice. Neuronal spiking occurs as a relatively steady rate throughout both novel and familiar recording sessions. C, Graphs showing the average spike rate during novel (left) and familiar sessions (right), for wild-type and J20 mice. Spike rate was quantified for the initial minute of each session, and final 10 min. The rate of neuronal spiking was significantly higher overall during familiar sessions (p = 0.02), and significantly higher overall during the final part of recording sessions (p = 7e-4). D, Graphs showing β amplitude (top) and multiunit activity spike rate (bottom) over time for β bursts, time locked to the onset of the burst (dotted line), and averaged across all detected bursts, for wild-type mice (left) and J20 (right) mice. β Bursting was associated with a monophasic increase in β amplitude that returns to baseline after around 250 ms. Spiking data are shown as Z score from baseline (preburst epoch) and averaged across all β bursts with nonoverlapping time segments. Dotted vertical line denotes the burst onset, while the solid horizontal line is shown to indicate the baseline of zero. β Bursts were associated with a significant increase in spike rate in wild-type (p = 0.04), but not J20 (p = 0.4), mice (data shown as mean ± SEM, WT: n = 5, J20: n = 8, *p < 0.05).
The average β amplitude during β bursts is shown in Figure 10D, top, averaged across all bursts with nonoverlapping time segments. β Bursts in both genotypes are associated with a brief, monophasic increase in β amplitude that lasts no more than 200 ms on average. Finally, Figure 10D, bottom, shows peri-event time histograms for spike rate during β bursts, as a Z score from the preburst epoch (left of the dotted line). In order to investigate potential statistically significant changes in spike rate during β bursts, we calculated the average Z scored spike rate between 0 and 250 ms after burst onset. β Bursting in the RSCdg of wild-type mice was associated with a significant increase in spike rate during β bursts (mean Z scored spike rate: 0.9 ± 0.3; t(4) = 2.9, p = 0.04; one-sample t test; Fig. 10D, bottom left). Conversely there was no significant increase in spike rate during β bursts in J20 mice (mean Z scored spike rate: 0.48 ± 0.53; t(7) = 0.9, p = 0.4; one-sample t test; Fig. 10D, bottom right). These data suggest that β bursts are coupled to neuronal spiking in RSCdg in wild-type mice, and that this relationship is lost in J20 mice.
As we have shown, in wild-type mice β bursting is associated with increased neuronal spiking in the RSC, so to better understand the relationship between β oscillations and neuronal spiking, we performed additional analyses. First, we investigated potential correlations between β burst characteristics and their effect on neuronal spiking. There was no linear relationship between the change in spiking rate during β bursts and the magnitude of the burst for either wild-type (R = −0.082, p = 0.0065) or J20 mice (R = −0.0039, p = 0.84). Conversely, there was a weak negative correlation between the change in spiking rate during β bursts and the duration of the burst for both wild-type (R = −0.3, p = 2.2e-24) and J20 (R = −0.22, p = 4.1e-30) mice. Second, we investigated potential phase-locking of neuronal spiking to a specific phase of the β oscillation, which would support the idea that β oscillations in the RSC are generated locally and not merely volume conducted from another region of the brain (França et al., 2021). For each recording session, histograms showing the probability of neuronal spiking at different phases of the β oscillation were generated, and Rayleigh's test was performed on these individual phase histograms to investigate potential nonuniformity of these distributions. As shown in Figure 11C, neuronal spiking in both wild-type (top) and J20 (bottom) mice did not occur uniformly throughout the β cycle, with increased spiking probability during the rising phase of the β oscillation. The distribution of these significance levels is shown in Figure 11D, top, as the natural logarithm of Rayleigh's Z statistic, which itself is equivalent to the negative natural logarithm of p. The vertical red line shows the equivalent of p = 0.05, so all sessions to the right of this line demonstrate significant β phase locking, quantified in the pie charts below. β Phase-locking of neuronal firing was seen in ∼90% of sessions for both genotypes. Finally, the histograms in Figure 11E show the distribution of preferred phase (µ, top) and concentration (κ, bottom) parameters for all significantly β Phase-locked sessions. RSC spikes tended to occur at around 0°, during the rising phase of the β oscillation, which was consistent across genotypes, while increased concentration (κ) values in J20 mice suggest a stronger degree of modulation in these animals. These results demonstrate that while there is no discernible relationship between the magnitude of β bursts and neuronal spiking, the change in the rate of neuronal spiking during β bursts appears to be higher during shorter β bursts. Furthermore, the presence of significant β phase-locking of neuronal spiking suggests that β oscillations are generated locally in the RSC.
The effect of β bursting on neuronal spiking is generally unaffected by the characteristics of the β burst. A, Individual peri-event histograms of the data shown in Figure 10D, showing multiunit activity spike rate during β bursts, for all wild-type (top) and J20 (bottom) mice. Data are shown as Z score from baseline (preburst epoch) and averaged across all β bursts with nonoverlapping time segments. Dotted vertical line denotes the burst onset, while the solid horizontal line is shown to indicate the baseline of zero. Increased neuronal spiking during β bursts was consistent across wild-type mice. B, Scatter plots showing the change in spike rate during each β burst against the magnitude (left) and duration (right) of each β burst, for all detected β bursts in wild-type (left) and J20 (right) mice. There was no linear relationship between the change in spiking rate during β bursts and the magnitude of the burst for either wild-type (R = −0.082, p = 0.0065) or J20 (R = −0.0039, p = 0.84) mice. Conversely, there was a weak negative correlation between the change in spiking rate during β bursts and the duration of the burst for both wild-type (R = −0.3, p = 2.2e-24) and J20 (R = −0.22, p = 4.1e-30) mice. C, Probability distributions for neuronal spiking at different phases of the β oscillation, averaged across all sessions for wild-type (top) and J20 (bottom) mice. D, Distribution of log-transformed Rayleigh's Z statistics for each session (top), which denotes the significance of β phase-locking and is equivalent to the negative natural logarithm of p. Vertical red line illustrates an approximate significance level of p = 0.05; therefore, all sessions to the right of this line therefore demonstrate significant β phase locking, quantified in the pie charts below for wild-type (left) and J20 (right) mice. E, Histograms showing the estimated preferred phase (µ, top) and concentration (κ, bottom) parameters for all significantly β phase-locked sessions. RSC spikes tended to occur at around 0°, during the rising phase of the β oscillation, which was consistent across genotypes. Furthermore, increased concentration (κ) values in J20 mice suggest a stronger degree of modulation in these animals (data shown as mean ± SEM, WT: n = 5, J20: n = 8).
Immunohistochemistry
In order to confirm the presence of age-related amyloid pathology in our J20 mice, we stained brains using Amylo-Glo (Biosensis; Schmued et al., 2012), a fluorescent marker which binds to amyloid plaques. We stained coronal slices from a subset of our seven- to eight-month-old experimental cohort, as well as an additional cohort of 12- to 13-month-old, J20 mice and wild-type littermates. Amylo-Glo staining revealed a high density of amyloid plaques in both the RSC and hippocampus of 12- to 13-month-old J20 mice (Fig. 12B), and a complete absence of amyloid plaques in age-matched wild-type littermates (Fig. 12A). Amyloid plaques seemed to vary greatly in size and density and were found in greatest number in the molecular layers of the dentate gyrus. Seven- to eight-month-old J20 mice had far fewer amyloid plaques, which were generally limited to the hippocampus. Quantification of amyloid plaques supported our findings of an age-related increase in amyloid plaque burden (main effect age, F(1,4) = 13.6, p = 0.02, mixed ANOVA; Fig. 12E), and an overall higher number of plaques in the hippocampus, compared with the RSC (main effect region, F(1,4) = 8.4, p = 0.04, mixed ANOVA; Fig. 12E); however, this regional difference was only significant for the younger age group (young, RSC: 0.7 ± 0.3; HC: 1.3 ± 0.2, p = 0.02).
J20 mice undergo age-related deposition of amyloid plaques throughout the hippocampus and RSC. A, Example photomicrograph of the hippocampus (HC) and retrosplenial cortex (RSC) of a 12- to 13-month-old wild-type mouse, stained with Amylo-Glo, and excited with 405-nm violet light. This image was taken under 10× magnification using a Nikon Eclipse N800 epifluorescence microscope (Nikon). Brains from wild-type mice are completely lacking in amyloid plaque pathology. B, Example photomicrograph of the hippocampus and RSC of a 12- to 13-month-old J20 mouse, taken under the same conditions as A. Abundant amyloid plaque pathology can be seen throughout the hippocampus and RSC. The two regions highlighted are shown in greater detail in C, D, taken under 20× magnification using a confocal microscope (ThorLabs). C, Example amyloid plaque found in the RSCdg. D, Example amyloid plaque found in the dentate gyrus of the hippocampus. Amyloid plaques in J20 mice were found in greatest number in the molecular layers of the dentate gyrus. E, Graphs showing the average number of amyloid plaques found in the RSC and hippocampus of 7- to 8-month-old (left) and 12- to 13-month-old (right) J20 mice and wild-type littermates, with individual animals shown as well. Plaque load was higher overall in 12- to 13-month-old J20 mice (p = 0.02), and higher overall in the hippocampus (p = 0.04). In seven- to eight-month-old J20 mice, plaques could be found in the hippocampus, and to a lesser extent in the RSC (p = 0.02), but overall, amyloid plaque load was low at this age point (data shown as mean ± SEM, 7- to 8-month-old WT: n = 1, 7- to 8-month-old J20: n = 3, 12- to 13-month-old WT: n = 5, 12- to 13-month-old J20: n = 3).
The generation of fast oscillations such as γ oscillations is thought to involve the activity of fast-spiking parvalbumin-positive interneurons (Csicsvari et al., 2003; Buzsáki and Wang, 2012), and parvalbumin-positive interneuron dysfunction is thought to underlie some of the neuronal network disturbances in AD (for review, see Xu et al., 2020). In order to investigate whether the altered neuronal network activity in the RSC in J20 mice may be explained, at least in part by changes in the number of parvalbumin-positive interneurons in this region, we performed parvalbumin staining on brains from a group of 12-month-old wild-type and J20 mice (Fig. 13). Overall, the number of parvalbumin-positive interneurons was significantly higher in the RSCdg than the RSCg (main effect region, F(1,7) = 56.5, p = 1.3e-4, mixed ANOVA; Fig. 13E). The number of parvalbumin-positive interneurons was significantly higher in the RSCdg than in the RSCg for both wild-type (RSCdg: 32.3 ± 4.6; RSCg: 13.8 ± 2.8, p = 7.6e-4) and J20 (RSCdg: 33.4 ± 4.4; RSCg: 19 ± 2.9, p = 0.002) mice. Conversely, there was no significant difference in the overall number of parvalbumin-positive interneurons in the RSC in wild-type and J20 mice (main effect genotype, F(1,7) = 0.4, p = 0.5, mixed ANOVA; Fig. 13E).
Parvalbumin-positive interneurons in the RSC of wild-type and J20 mice. A, Example photomicrograph of the RSC of a 12-month-old wild-type mouse, stained with a parvalbumin-specific antibody. This image was taken under 4× magnification using a Nikon Eclipse N800 epifluorescence microscope (Nikon). The area highlighted is shown under 10× magnification in B. In wild-type mice, parvalbumin-positive interneurons can be seen in low numbers throughout the RSC. C, Example photomicrograph of the RSC of a 12-month-old J20 mouse, taken under the same conditions as A. The area highlighted is shown under 10× magnification in D. As in wild-type mice, parvalbumin-positive interneurons can be seen in low numbers throughout the RSC of J20 mice. E, Graph showing the average number of parvalbumin (PV)-positive interneurons in the RSCdg and RSCg of 12-month-old J20 mice and wild-type littermates, with individual animals shown as well. While numbers of PV-positive interneurons were higher on average in the RSCdg (p = 1.4e-4), there was no significant difference between numbers of PV-positive interneurons in the RSC in wild-type and J20 mice (p = 0.5; data shown as mean ± SEM, WT: n = 4, J20: n = 5).
Discussion
In this study, we attempted to identify neurophysiological correlates of environmental novelty in the mouse RSC, and investigate how these may be affected by amyloid pathology. We observed transient, high amplitude β frequency oscillations, termed β bursts, which occurred more frequently and with larger amplitude during novelty and were positively correlated with neuronal spiking. Several neurophysiological changes were seen in the RSC in J20 mice, many of them novel, which are indicative of aberrant neuronal network activity. These results together indicate that β bursting activity is a neurophysiological correlate of environmental novelty in the RSC, which is disrupted in J20 mice and may underlie the apparent contextual memory impairments in these animals.
Numerous studies have noted changes in β activity in a range of brain regions, during a variety of behaviors (for review, see Spitzer and Haegens, 2017). Berke et al. (2008) reported a large increase in β power in the hippocampus when mice explore a novel environment, that persists for around a minute after exposure. The authors concluded that these oscillations may be a “dynamic state that facilitates the formation of unique contextual representations.” Coherent 20- to 40-Hz oscillatory activity between the hippocampus and lateral entorhinal cortex has been shown to increase during odor discrimination, with the development of odor-specific neural representations (Igarashi et al., 2014). Work by França et al. (2014) demonstrated that β power was also transiently enhanced in the hippocampus during exploration of novel objects, but not familiar objects. Furthermore, they found that administration of an amnestic agent, namely haloperidol, resulted in a similar increased β activity on re-exposure to previously encountered objects, suggesting they had been “forgotten.” Subsequently, França et al. (2020) investigated novelty-associated β bursting in the hippocampus, prefrontal cortex and parietal cortex during environmental and object novelty. Novelty-associated increases in β power were seen in the prefrontal cortex during environmental novelty, and authors demonstrated significant PAC of δ and θ to β oscillations, which increased during novelty. In the RSC we see strong coupling between θ phase and α/β amplitude, which is significantly higher during novelty in wild-type mice, alongside weaker coupling between θ phase and γ amplitude, which is unaffected by novelty. While θ-γ coupling is well established (for review, see Canolty and Knight, 2010), far less is known about θ-α/β coupling. Previous studies have demonstrated θ-β PAC in humans, both in the hippocampus during a working memory task (Axmacher et al., 2010), and in the inferior temporal cortex during object novelty (Daume et al., 2017). θ-γ Coupling is the dominant form of PAC in the hippocampus and is thought to support memory processes (Colgin et al., 2009; Axmacher et al., 2010; Newman et al., 2013; Lega et al., 2016), so our findings suggest that θ-α/β coupling may be the cortical equivalent of θ-γ coupling, and support memory process during contextual novelty. If this is true, then the absence of novelty-associated increases in θ-α/β coupling in J20 mice may therefore have detrimental effects on memory processes. Interestingly, previous studies into β oscillations during novelty tend to view β activity as continuous oscillations, rather than discrete events (Berke et al., 2008; França et al., 2014, 2020). This is despite Berke et al. (2008) noting that β appears as pulses, and a brief mention of burst detection and characterization by França et al. (2014). As demonstrated in this study, novelty-associated β oscillations in the RSC conform well to a model of discrete, rhythmic bursts, where their rate, magnitude and duration can vary depending on environmental novelty. Because of the use of averaging across trials or long temporal segments, the phasic nature of transient oscillatory events can be easily lost. Furthermore, in the somatosensory cortex, β synchronicity appears as short events in both mice and humans; the features of which, such as duration and frequency range, are highly conserved across tasks and species (Shin et al., 2017). Sherman et al. (2016) hypothesized that transient β oscillations may arise as a result of synchronous bursts of excitatory activity at the proximal and distal dendrites of pyramidal neurons. Furthermore, computational modeling suggests that continuous 10 Hz stimulation at these sites is sufficient to produce β bursts, which could explain the strong PAC between θ and β oscillations in this paper, especially during novelty.
Many groups have previously shown that information may be rapidly represented and stored in the RSC (Cowansage et al., 2014; Czajkowski et al., 2014; Koike et al., 2017; Vedder et al., 2017). β Oscillations have also been shown to carry a variety of different forms of contextual information in a range of brain regions, and phasic increases in β power during working memory maintenance may represent reactivation of encoded information (Spitzer and Haegens, 2017). Supporting this is a study in which the authors employed transcranial magnetic stimulation to activate a currently unattended memory, as shown by an increase in content-specific β activity (Rose et al., 2016). The theory put forth by Spitzer and Haegens (2017) is that β oscillations can activate and reactivate neuronal ensembles to create and recall cortical representations. This theory is consistent with the data shown in this study: high β bursting activity during perceived novelty activates neurons in the RSC, which may encode content about the novel environment, and subsequent β bursting may continuously reactivate these ensembles, further consolidating or altering this representation. Recent breakthroughs in real-time burst detection and neurofeedback have made it possible to artificially induce β bursts in awake behaving animals, creating the possibility of testing this hypothesis directly (Karvat et al., 2020).
Several neurophysiological changes were seen in the RSC in J20 mice. Increases in α and β power are suggestive of cortical hyperexcitability, which has been previously shown in this strain (Palop et al., 2007; Palop and Mucke, 2009). We noted increases in the rate of β bursting and burst magnitude, and an absence of coupling between β bursting activity and neuronal spiking in J20 mice, potentially impairing the ability to form neuronal ensembles that represent information in the RSC. Interestingly, excessive β bursting in the basal ganglia and cortex is associated with the severity of motor impairments in Parkinson's disease (for review, see Brittain et al., 2014). Amyloid plaque formation in J20 mice increases dramatically with age and is most severe in the hippocampus and RSC (Whitesell et al., 2019), however at the age point used in this study, amyloid plaques were sparse in the RSC. This does not preclude the presence other potentially toxic species of amyloid β, such as oligomers, which have been detected in the hippocampus (Wright et al., 2013) and cortex (Castanho et al., 2020) of J20 mice before substantial plaque load, but suggests that functional deficits in these animals occur independent of plaque formation. At four months of age, J20 mice have demonstrated impaired spatial learning and memory in the Morris Water Maze (Cheng et al., 2007) and the radial arm maze (Wright et al., 2013). Functional disturbances and cognitive impairments may arise from interneuron dysfunction, which has been demonstrated in J20 mice and has been shown to lead to cortical network hypersynchrony and spontaneous epileptiform discharges in animals four to seven months of age (Verret et al., 2012). Reduced excitability of parvalbumin-positive interneurons in the hippocampus of 30 d old J20 mice has been associated with impaired θ-γ PAC (Mondragón-Rodríguez et al., 2018), and while parvalbumin-positive interneuron function was not necessary for the generation of ripples in the hippocampus, they were necessary for the coupling of neuronal spiking to ripples (Xia et al., 2017). These findings suggest that interneuron dysfunction in our J20 mice may underlie the altered PAC and the loss of coupling of neuronal spiking to β bursting shown in this paper. Reduced interneuron function in J20 mice could result in disinhibition and therefore underlie increases in β burst magnitude and α/β power. While the total number of parvalbumin-positive interneurons in the RSC did not differ between wild-type and J20 mice, it is possible that interneuron function may be altered. Similarly, in a different mouse model of AD, Booth et al. (2016) found altered γ oscillations in the medial entorhinal cortex, but to the total number of parvalbumin-positive interneurons in this region. Finally, while β power is higher and β bursts are more prevalent and larger in amplitude in J20 mice, their relationships with novelty are preserved, which may allow for the preservation of some contextual memory function. These findings demonstrate a novel form of AD-related cortical dysfunction, which may underlie or exacerbate cognitive dysfunction seen in these mice, and in people with AD (Cheng et al., 2007; Verret et al., 2012; Wright et al., 2013).
In conclusion, phasic bursts of β oscillations may be a functional means of activating neural ensembles to form and subsequently reactivate cortical representations. Dysregulated β bursting and an uncoupling of β bursting from spiking are suggestive of network dysfunction in J20 mice which may underlie cognitive impairments in these mice.
Footnotes
C.W. was supported by a University of Exeter and Janssen Pharmaceutica studentship. T.R. was supported by the ARUK Major Project Grant ARUK-PG2017B-7 awarded to J.T.B. and A.D.R.
The authors declare no competing financial interests.
- Correspondence should be addressed to Callum Walsh at cw685{at}exeter.ac.uk or Jonathan T. Brown at j.t.brown{at}exeter.ac.uk