Hippocampal replay of experience at real-world speeds
- Howard Hughes Medical Institute, University of California, San Francisco, United States
- Departments of Physiology and Psychiatry, University of California, San Francisco, United States
- Kavli Institute for Fundamental Neuroscience, University of California, San Francisco, United States
- Department of Neurobiology, Stanford University School of Medicine, United States
- Department of Neurological Surgery, University of California, San Francisco, United States
- Department of Mathematics and Statistics, Boston University, United States
Figures
Figure 1 with 1 supplement
The model can capture different sequence dynamics on simulated data.
(A) We construct a firing sequence of 19 simulated place cells that exhibits three different movement dynamics. For the first 60 ms, one cell fires repeatedly, representing one stationary location. …
Figure 1—figure supplement 1
The model is robust to change of the probability of persisting in the same dynamic for a wide range of plausible expected durations (25–150 ms).
(A) Each panel shows the probability of each dynamic on simulated data example from Figure 1 with a different diagonal value—which governs the probability of remaining in that dynamic. The …
Figure 2 with 4 supplements
The model can decode hippocampal replay trajectories using either sorted and clusterless spikes from the same SWR event.
(A) Decoding using sorted spikes. The top panel shows 31 cells on a W-track ordered according to linearized position by their place field peaks. The middle panel shows the probability of each …
Figure 2—figure supplement 1
Decoding the same SWR in Figure 2 with 2D position using sorted spikes and clusterless decoding.
(A) The left panel shows the spikes from cells arranged by the linear position of the peak of place field as in Figure 2. The middle panel shows the probability of each dynamic over time from the 2D …
Figure 2—figure supplement 2
More examples of SWRs that have continuous trajectories.
Figure 2—figure supplement 3
Population firing rate on the track is spatially uniform and consistent for each animal.
Multiunit rate of encoding spikes over track positions for each animal. Each gray line represents a recording session for that animal.
Figure 2—video 1
Example of an SWR with continuous content.
Magenta dot represents the animal’s position. Green dot represents the most likely decoded position projected from 1D back to the linearized 2D position. Green line represents the decoded positions …
Figure 3 with 4 supplements
Most SWRs are spatially coherent, but not continuous.
(A-F) Examples of SWRs with non-constant speed trajectories. Figure conventions are the same as in Figure 2. Filtered SWR ripple (150–250 Hz) trace from the tetrode with the maximum amplitude …
-
Figure 3—source data 1
Table of replay statistics for each SWR for Figure 3G.
- https://cdn.elifesciences.org/articles/64505/elife-64505-fig3-data1-v2.csv
Figure 3—figure supplement 1
More examples of SWRs with non-constant speed trajectories.
(A-I) More examples of SWRs with non-constant speed trajectories. Conventions are the same as in Figure 3.
Figure 3—figure supplement 2
Shuffling the position data with replacement decreases the percent of SWRs classified.
Comparison of percentage of SWRs classified (that is, an SWR containing at least one of the five classifications) on real vs. position shuffled data for two recording sessions from different …
Figure 3—figure supplement 3
Shuffling the data by swapping the runs and circularly permuting the position increases the percentage of spatially incoherent SWRs and decreases the spatially coherent SWRs.
The red line represents the percent of spatially coherent or incoherent SWRs in that recording session for actual data. The histogram represents the distribution after 50 shuffles of the run from …
Figure 3—video 1
Example of an SWR with that is not purely continuous.
Conventions the same as Figure 2—video1.
Figure 4 with 2 supplements
Validation of classification using the 95% Highest Posterior Density.
(A–D) Examples of the 95% Highest Posterior Density. In each column: top panel: Probability of dynamic over time. Shading and labels indicate dynamic categories. Middle panel: Posterior distribution …
-
Figure 4—source data 1
Table of replay statistics for each SWR for Figure 4E.
- https://cdn.elifesciences.org/articles/64505/elife-64505-fig4-data1-v2.csv
Figure 4—figure supplement 1
Shuffling the position with replacement increases the average 95% HPD region size.
Red line represents the average 95% HPD region size for all ripples. The histogram represents the distribution after 50 shuffles of the position data. Position data was shuffled by resampling with …
Figure 4—figure supplement 2
Shuffling the position data by swapping the runs and circularly permuting the data increases the average 95% HPD region size for spatially coherent and incoherent classified times.
Data from two recording sessions from different animals. Red line represents the average spatial position spanned by the highest posterior density on real data for all ripples. The histogram …
Figure 5 with 4 supplements
Prevalence of classifications.
(A) UpSet plot (Lex et al., 2014)—which is similar to a Venn diagram with more than three sets—of the most common sets of classifications within each SWR. Each row represents a classification and …
-
Figure 5—source data 1
Table of replay statistics for each SWR for Figure 5.
- https://cdn.elifesciences.org/articles/64505/elife-64505-fig5-data1-v2.csv
Figure 5—figure supplement 1
More examples of stationary-continuous-mixtures.
Conventions are the same as in Figure 3.
Figure 5—figure supplement 2
Control analyses for distribution of dynamics.
Conventions are the same as in Figure 5.
Figure 5—figure supplement 3
Further quantification of spiking and ripple properties by dynamic.
(A) Further quantification of the dynamics with respect to the spiking and ripple properties. (B) Speed and number of tetrodes with spikes for entire SWR.
Figure 5—video 1
Example of a stationary-continuous-mixture.
Conventions the same as Figure 2—video1.
Figure 6 with 2 supplements
The standard decoder MAP estimate speeds are the most similar to the state space decoder.
Event speeds calculated using three common 'Bayesian' decoder approaches (A–C) compared to using the state space model (D). For each panel, the top five rows show the probability density of the …
-
Figure 6—source data 1
Table of replay statistics for each SWR for Figure 6.
- https://cdn.elifesciences.org/articles/64505/elife-64505-fig6-data1-v2.csv
Figure 6—figure supplement 1
Examples of fits from the standard 'Bayesian' decoder with 20 ms bins and state space model.
(A–F) For each panel, the top row is a raster plot of the sorted cells, the second row is the decoded posterior probability of position in 20 ms time bins for the sorted spikes and the corresponding …
Figure 6—figure supplement 2
Examples of fits from the standard 'Bayesian' decoder with 2 ms bins and state space model.
(A–F) For each panel, the top row is a raster plot of the sorted cells, the second row is the decoded posterior probability of position in 2 ms time bins for the sorted spikes and the corresponding …
Author response image 1
Proportion of SWRs for each day vs duration of the dynamic within the SWR.
Each dot represents one day for one animal.
Author response image 2
