HAL Id: hal-02373798
https://hal.archives-ouvertes.fr/hal-02373798
Submitted on 21 Nov 2019
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
ARE THERE DISCRETE GAMMA SUB-BANDS IN HIPPOCAMPAL NETWORKS DURING SPATIAL
LEARNING ?
Vincent Douchamps, Demian Battaglia, Romain Goutagny
To cite this version:
Vincent Douchamps, Demian Battaglia, Romain Goutagny. ARE THERE DISCRETE GAMMA SUB-BANDS IN HIPPOCAMPAL NETWORKS DURING SPATIAL LEARNING ?. Society for Neuroscience annual meeting, Oct 2019, Chicago, United States. �hal-02373798�
Vincent Douchamps 1 , Demian Battaglia 2 , Romain Goutagny 1
1
Laboratoire de Neurosciences Cognitives et Adaptatives, CNRS UMR 7364, Université de Strasbourg;
2Institut de Neurosciences des Systèmes, INSERM UMR 1106, Aix-Marseille Université
Are there discrete gamma sub-bands
in hippocampus during spatial learning?
Institut de
Neurosciences des Systèmes
Fonds Paul Mandel pour les neurosciences
Fédération Neuropôle Université de Strasbourg
Power across learning Introduction
Theta (θ) and gamma (γ) oscillations are believed to
organise hippocampal activity, via their cross-freq coupling
Classic view: Slow γ (~30-60 Hz), possibly related to memory retrieval; and Medium γ (~60-100 Hz), possibly related to
sensory infomation and memory encoding
However, recent evidence suggests a wider repertoire of coupling patterns when considering individual θ cycles
Schomburg et al., 2014
AIM OF THE STUDY
Explore theta-gamma oscillatory dynamics in CA1 and DG across learning in an ecologic spatial navigation task and without a priori on frequencies and phases that should be observed.
The Arm-to-Arm task
Subjects: water-restricted mice (n=4) Maze: 8-arm radial arm maze; distal visual cues available
Task: finding the rewarded arm starting from another (semi-random) arm.
Fixed reward arm across the 10-day training, 4 daily trials
Reward: 0.05 ml of water at last arm end
Day 1 - Trial 1 Day 4 - Trial 4
0 50 100
Time to reach target (s)
Learning day
1 2 3 4 5 6 7 8 9 10 F(9,63) = 7.956
p < 0.0001
target
-90 opposite
0 0.2 0.4 0.6
Ratio of visits in arm
+90
* *
*
*
F(7,49) = 12.38 p < 0.0001
Identification of θ cycles: preservation of θ
asymmetry (1-25 Hz filtering); phase estimation with piece-wise linear interpolation
γ spectral content of each θ cycle: coincident seg- ment of the spectrogram of γ-composite signal
(complex Morlet wavelets; 1-200 Hz; 1 Hz steps)
Mean θ-γ motifs from repre- sentative channels and trials do correspond to the “classical
model”
However, the possibility of multi- ple γ bouts per θ cycle is not
“classical”, as well as the pres-
ence of both medium and slow γ episodes in both dendritic
layers
The large variability of individ- ual cycles and the broad count distributions suggest that the landscape of possible θ-γ cou- plings is better described as a structured continuum
270
90
0 180
90
0 180
90
0 180
At all stages of learning and all anatomical layers, cycles arehighly variable, but phase, freq and power are not uniformly distributed
The “swarms” of cycles morph along
learning (e.g. migration of hi power SLM cycles toward higher frequencies)
Yet, despite complexity, non-trivial
coding of navigation speed or location?
We trained classifiers (Random forests, via RUSBoost method) to predict speed and maze location as a function of individual cycle properties
Decoding is possible, revealing that different
inputs convey different information at different times (but not simply “recall” or “encoding”!)
Raw LFP (hippocampal fissure)
Unsupervised EEMD decomposition
Unsupervised empirical ensemble mode decomposition (EEMD for cycle-by-cycle analy- sis [Wu & Huang, 2009] ; CEEMD for power spectrum [Torres et al., 2011] )
Power spectra of EEMD-derived IMFs is stable along training Need to correct for theta harmonics!
SLM spectra
n.s.
n.s.
n.s.
θ cycle-by-cycle analyses
Peak-to-peak segmentation in θ cycles (ref. in LFP from hippocampal fissure)
θ-γ coupling: means vs counts
20 40 60 80 100 120 140 160 180 200
0
-π 0 π
Frequency (Hz) LFP power
max
min
SP
20 40 60 80 100 120 140 160 180 200
0
Frequency (Hz)
20 40 60 80 100 120 140 160 180 200
0
Frequency (Hz)
Theta phase (°)
Hippocampal fissure
CSD amplitude
min max
SR
SLM
Arbitrary units
Mean Count
Frequency (Hz)
Mean Count
Mean Count
0 50 100 150
-π 0 π
Theta phase (°)
Hippocampal fissure
15 85 160
Frequency (Hz)
-π 0 π
Theta phase (°)
Hippocampal fissure
-π 0 π
Theta phase (°)
Hippocampal fissure
15 85 160
Frequency (Hz)
15 85 160
Frequency (Hz)
INDIVIDUAL CYCLES MEAN θ CYCLES Mean vs Count
Daunting variability and deviations from “prototypical” case (usually, only ~2% of cycles are similar to mean template)
40 80 120
SLM
Frequency (Hz)
SP
Cycle counts (a.u.)
SR
CYCLE COUNTS (all trials and channels)
LO and HI power together
LO power only
HI power only
A structured continuum of cycles...
180 -90
0
90
40 60
80 100
120
40 60
80 100
120
40 60
80 100
40 60
80 100
120
40 0 60
80 100
120
40 60
80 100
120 40
60 80 100
120
40 60
80 100
120
40 60
80 100
40 60
80 100
120
40 60
80 100
120
40 60
80 100
120
SO
Early Late
SP
Early Late
SR
Early Late
SLM
Early Late
DG mol
Early Late
DG gran
Early Late
CENTER:
low freq cycles
PERIPHERY:
hi freq cycles One dot = one θ/γ event
Cycle peak power
5% quantile 95% quantile
Phase
30 Hz 120 Hz
Frequency
e.g., in SLM,
cycles occurring during high-speed epochs
tend to cluster in patches in this
dimensionally-
reduced projection of all SLM cycles
... which conveys decodable information
Trial group
Early Late Early
Trial groupLate SPEED PREDICTION TARGET ARM
PREDICTION
Correlation between predicted and real speed
0.5
0.3
0.62 0.70 0.78
Area under ROC curve (AUC)
SLM SR
SR confusion matrix (% of classifications)
Other
Target arm Reward
Other Target arm Reward
0% 100%
50%
60%
82%
REPRESENTATIONAL SIMILARITY
Trial group
Early
Late
SLM predicts speed
Trial group
Early
Late
SLM predicts target arm
Trial group
Early Late
low high
Cross-classification performance
predicts