Disrupted social memory ensembles in the ventral hippocampus underlie social amnesia in autism-associated Shank3 mutant mice

The ability to remember conspecifics is critical for adaptive cognitive functioning and social communication, and impairments of this ability are hallmarks of autism spectrum disorders (ASDs). Although hippocampal ventral CA1 (vCA1) neurons are known to store social memories, how their activities are coordinated remains unclear. Here we show that vCA1 social memory neurons, characterized by enhanced activity in response to memorized individuals, were preferentially reactivated during sharp-wave ripples (SPW-Rs). Spike sequences of these social replays reflected the temporal orders of neuronal activities within theta cycles during social experiences. In ASD model Shank3 knockout mice, the proportion of social memory neurons was reduced, and neuronal ensemble spike sequences during SPW-Rs were disrupted, which correlated with impaired discriminatory social behavior. These results suggest that SPW-R-mediated sequential reactivation of neuronal ensembles is a canonical mechanism for coordinating hippocampus-dependent social memories and its disruption underlie the pathophysiology of social memory defects associated with ASD.


Social discrimination test (SDT)
Mice were individually habituated to the investigator by handling for several minutes on each of two consecutive days (Day-1 and Day-2). Habituation to the social arena and recording cable was performed for 5 min on each of two consecutive days (Day-2 and Day-3). The social arena was a white acrylic box (area, 38×38 cm 2 , height, 30 cm); two custom-made social chambers with quadrant bottom shape (7.5cm radius) printed with a 3D printer (Original Prusa i3 MK3S; Prusa Research Prague, Czechia) using white PETG filaments were placed at the opposite corners of the arena. A social chamber consists of Disrupted social memory ensembles in autism bars with ~1-cm gaps and a cap, which can confine a stimulus mouse inside while allowing animals to interact with each other through the gaps. Social chambers On Day-4, a stimulus mouse (C3H or BALB/c; A) was placed into the home cage of the subject mouse for familiarization for 2 hours. Five minutes later, the familiarized mouse (A) and a novel mouse (BALB/c or C3H; B) were placed in the left or right chamber, and the subject mouse was allowed to explore inside the arena for 5 min (A-B trial). This procedure was repeated by alternating the position of the two stimulus mice (B-A trial) and then in their absence (E-E trial). Trials were repeated with 5-min intertrial intervals. Trials were video recorded at 25 frames per s from the top view of the arena using an area scan camera (acA1300-60gmNIR; Basler, Ahrensburg, Germany) equipped with a Computer 4-8-mm C-Mount lens. Video recording system was controlled by Bonsai software 2.6.2 [62]. Immediately after the E-E trial (typically ~1 min), the subject mouse was moved back to the home cage and allowed to rest for 2 h. Five subjects repeated experimental sessions with at least a 24-h interval using a different group of stimulus mice. Novel object recognition (NOR) test was performed following the same time-course of the SDT but using inanimate objects. All objects were in different shapes, made of distinct materials (wood, glass, metal, or plastic), and about 10 cm tall and 6×6-cm footprint.

Histology
After completion of the experiments, mice were deeply anesthetized and transcardially perfused with phosphate-buffered saline and then 4% paraformaldehyde. Thereafter, brains were dissected out and post-fixed in 4% paraformaldehyde at 4 °C overnight, and fixed samples were sectioned into 100-μm coronal slices using a vibratome (Leica, VT1000S; Leica Microsystems, Wetzlar, Germany). For verifying the electrode tracks, images were acquired using a microscope (BZ-X710; Keyence, Tokyo, Japan).

Social activity of hippocampal units
Discrimination score for each unit was computed as( ! − " )/( ! + " ), where fA and fB were the firing rates of a unit when the subject was located in the social zone around the stimulator mouse A or B, respectively. Significance of the calculated score was determined by a permutation test of spike counts in each video frame (repeated 10,000 times). The P value represents the proportion of shuffled values larger than the actual observed value. Neurons with a significant (P < 0.05 by permutation test) discrimination score were classified either as mouse-A neuron (score < 0) or mouse-B neuron (score > 0), respectively. The egocentric directional tuning function for each unit was the ratio between the peaks of spike count and total time spent in each direction in bins of 5° and smoothed with a Gaussian kernel of 2 bins.

LFP spectrum analysis
Root mean square (RMS) of the LFP signal across channels having pyramidal cells was calculated at each time point. The signal was then binned into 10-s epochs and the instantaneous power spectrum density (PSD) was estimated using Welch's method with a Hamming window (size = 1.33 s, overlap = 0.67 s). The obtained PSD was used to display spectrograms across offline recordings, to compare the spectral structure of the LFP between genotypes, and to assess the behavioral state during offline recordings as described below. To display spectrograms of SPW-Rs, continuous wavelet transform (CWT) was applied to unfiltered LFP signals using complex Morlet wavelets with a parameter of 7.
Peak ripple frequency was identified as the frequency bin higher than 100 Hz in which the CWT has the largest value on the ripple peak timings.

Sleep analysis
To assess the behavioral state during offline recordings, signals from a 3-axis accelerometer on the headstage were utilized to calculate the derivatives of roll and pitch, which reflect angular velocities of the subject's head. The L2 norm of these derivatives was used as a proxy for the overall movement of the subject. The signal was then binned into 10-s epochs and sleep epochs were defined as the period of sustained immobility (0.1 arbitrary unit/s). The delta (2-5 Hz) power calculated by Welch's method was referenced to confirm that the movement signal recapitulated the physiological sleeping state.

Ripple event detection
The LFP signal was bandpass filtered (150-250 Hz) on each channel having pyramidal neurons, and the root mean square across channels was calculated at each time point. The power was then Gaussian smoothed (4 ms standard deviation [SD]). Epochs with a peak power exceeding the mean by at least 5 SD for at least 15 ms were detected. The sample points at which the power reduced below 1 standard deviation were determined as the points of onset and offset of the epochs. Ripple pairs with peaks closer than 50 ms were merged into single events.