Control of recollection by slow gamma dominating mid-frequency gamma in hippocampus CA1

Behavior is used to assess memory and cognitive deficits in animals like Fmrl-null mice that model Fragile X Syndrome, but behavior is a proxy for unknown neural events that define cognitive variables like recollection. We identified an electrophysiological signature of recollection in mouse dorsal CA1 hippocampus. During a shocked-place avoidance task, slow gamma (SG: 30-50 Hz) dominates mid-frequency gamma (MG: 70-90 Hz) oscillations 2-3 seconds before successful avoidance, but not failures. Wild-type but not Fmrl-null mice rapidly adapt to relocating the shock; concurrently, SG/MG maxima (SGdominance) decrease in wild-type but not in cognitively inflexible Fmrl-null mice. During SGdominance, putative pyramidal cell ensembles represent distant locations; during place avoidance, these are avoided places. During shock relocation, wild-type ensembles represent distant locations near the currently-correct shock zone but Fmrl-null ensembles represent the formerly-correct zone. These findings indicate that recollection occurs when CA1 slow gamma dominates mid-frequency gamma, and that accurate recollection of inappropriate memories explains Fmrl-null cognitive inflexibility.


Control of recollection by slow gamma dominating mid-frequency gamma in hippocampus CA1
Authors: Dino Dvorak, Basma Radwan, Fraser T. Sparks, Zoe Nicole Talbot, André A. Fenton

Data and code online repository
To obtain data used for the analyses, please visit https://osf.io/j5baw/?view_only=87b1ff00182a437883da06feda992763

Basic LFP properties during active avoidance
Local field potentials analyzed during active avoidance training when mice were still (speed < 2 cm/s) and running (speed ≥ 2 cm/s; Fig S1A,  We detected sharp-wave associated ripple events (Carr et al., 2011;Csicsvari et al., 2000;Jackson et al., 2006;O'Neill et al., 2006) using a published algorithm (Eschenko et al., 2008). The rate of ripple events were generally low (Kay et al., 2016) and did not differ during overall stillness or during the stillness that preceded subsequent active avoidance movements (Fig S1C, F 1,28 = 0.34, p = 0.7).

Mixtures of slow, mid-frequency and fast gamma oscillations in stratum pyramidale of mouse CA1
Concurrent local field potentials (LFPs), reflecting the synchronous synaptic activity at distinct inputdefined anatomical locations within the mouse CA1 region of dorsal hippocampus were recorded using 32-ch linear silicon electrode array (Neuronexus, Ann Arbor, MI) in order to demonstrate the presence of three distinct gamma bands within stratum pyramidale (sp) as shown previously in rat (Fernandez-Ruiz and Herreras, 2013;Schomburg et al., 2014) and mouse Klausberger, 2014, 2016). LFPs were first localized ( Fig S2A) using sharp-wave associated ripples (SWR). The maximum amplitude of the ripple identified stratum pyramidale, the maximum amplitude of the sharp wave identified stratum radiatum, and the sharp wave reversal identified stratum lacunosum moleculare. CSD analysis (Mitzdorf, 1985) was then performed in order to separate individual oscillatory CSD components ( Fig S2B). Theta (5-12 Hz) phase was then used to construct theta-averaged CSD power profiles (Fig S2C) Individual theta cycles are marked by vertical lines. Oscillatory events detected as local maxima in timefrequency 2-D space with peak power > 2.5 S.D. are marked with red crosses.

Mixtures of slow and mid-frequency gamma oscillations during single theta cycles
It was reported that in rat, slow and mid-frequency gamma oscillations tend to occur mutually exclusively during theta cycles (Colgin et al., 2009), but we find instead that slow and mid-frequency gamma oscillations detected as discrete events (Fig S2E) are often mixed in mouse, as reported by others (Lasztoczi and Klausberger, 2016; see Fig S2). It is possible however, that the common, non-exclusive appearance of slow and mid-frequency gamma oscillations in single theta cycles might be specific to mouse.

Selecting the power threshold for oscillatory events
From the nature of time-frequency representations, when power is reduced, the time-frequency profile loses dominant peaks and instead more low power peaks are detected (Fig S3A). This results in a negative correlation between the number of low power peaks in a given time interval and the total power of all events detected in the interval (Fig S3B). To avoid this bias, we identified an optimal power threshold so that the number of oscillations per unit of time is not influenced by increased counting of low power events. Setting the power threshold to above 2 S.D. from the overall mean band power selects about 35% of all detected events. Setting the power threshold above 3 S.D. selects about 20% of all detected events (Fig S3C). Fig S3D bottom shows the continuous relationship between power threshold applied to detected events in the 30-60 Hz slow and 60-90 Hz mid-frequency gamma range and the ratio of theta cycles classified as having a unique type of oscillation (slow or mid-frequency gamma) and theta cycles classified as having both slow and mid-frequency gamma oscillations. We only included periods when the mouse was running (speed ≥ 2 cm/s) and only included theta cycles with periods between 83 and 250 ms corresponding to frequencies between 4 and 12 Hz. Predictably, low power thresholds lead to high numbers of detected events but a low ratio of single types of event (slow or mid-frequency gamma) present in a given theta cycle (Fig S3D top, left; theta cycles only with slow gamma 'S', midfrequency gamma 'M', mixture of slow and mid-frequency gamma 'S/M' and no detected oscillations 'Ø').
In contrast, high power thresholds lead to a low number of detected events and a high prevalence of single gamma-type theta cycles (Fig S3D top, right).
Because of the continuous relationship between power threshold and the ability to classify a theta cycle as one with just slow or mid-frequency gamma, we used the ratio of slow to mid-frequency gamma events per time period throughout this report.

Identifying a power threshold and frequency bands of interest for slow and medium gamma event rates.
To investigate the apparent asymmetry in the relative declines of slow and mid-frequency gamma events that was observed in the wavelet spectrum prior to avoidance ( Fig 1E in the main text), we first identified the specific frequency bands of interest ( Fig S3E). We computed the rate of oscillatory events in 1-s long intervals that advanced by 250 ms. The event rates around the avoidance onset were averaged across 20 Hz-wide bands between 20 and 110 Hz and for power thresholds z ≥ 1, 2, 2.5 and 3 during the third training session. We then compared these profiles across all bands to look for similarities. The 20-40 Hz band followed similar patterns of activity as the other two slow gamma bands (30-50 Hz and 40-60 Hz), but its average event rates were generally low so we excluded this band from further analysis. The 30-50 Hz and 40-60 Hz slow gamma bands were distinct from bands in the mid-frequency gamma range (60-110 Hz) for power thresholds above z ≥ 1, because the attenuation of slow gamma was reduced prior to avoidance onset, in fact, there were peaks of increased slow gamma rates 2-4 s before the avoidance initiated and there was an earlier increase of the slow gamma event rate at avoidance onset compared to a delayed increase in the rate of mid-frequency gamma events. We selected the 30-50 Hz band to represent slow gamma (SG) oscillations and the 70-90 Hz band to define mid-frequency gamma (MG) oscillations. In all subsequent analyses we only included events that were 2.5 S.D. or greater than the average power in a given frequency band. (B) Proportions of different behavioral events detected during SG dom events during pretraining sessions before ever experiencing shock (filled bars) compared to randomly-selected events (empty bars; Comparisons of SG dom to Random Still: ! " # = 6.5, p = 0.16; Run: ! " # = 8.3, p = 0.08; StillgRun: ! " # = 9.3, p = 0.05; RungStill: ! " # = 0.004, p = 0.99). (C) Average SG dom rates across initial 15 minutes of first and last training session (two-way ANOVA with repeated measures genotype x trial: genotype: F 1,13 = 0.30, p = 0.59; trial: F 1,13 = 6.91, p = 0.02; genotype x trial: F 1,13 = 0.04, p = 0.85). Percentage of (non-isolated) SG and MG events that coincide with SWR events. (B) Relationships between location decoding error (smaller error = more accurate) and running speed (top) and ensemble firing rate (bottom). (C) Bayesian decoding error during slow gamma oscillations that are not accompanied by mid-frequency gamma oscillations or SWRs (blue), mid-frequency gamma oscillations that are not accompanied by slow gamma oscillations or SWRs (yellow) and random events, which are not accompanied by SWRs (gray) in WT and KO mice, corrected for firing-rate bias of the decoding. (D)

Relationship between decoding accuracy, firing rate and speed
Summary of decoding error during isolated slow and mid-frequency gamma oscillations and random events. *p < 0.05 relative to random events.