Sampling motion trajectories during hippocampal theta sequences

  1. Balazs B Ujfalussy  Is a corresponding author
  2. Gergő Orbán
  1. Laboratory of Biological Computation, Institute of Experimental Medicine, Hungary
  2. Laboratory of Neuronal Signalling, Institute of Experimental Medicine, Budapest, Hungary
  3. Computational Systems Neuroscience Lab, Wigner Research Center for Physics, Budapest, Hungary
7 figures, 4 tables and 1 additional file

Figures

Theta sequences, uncertainty and variability.

(a) Schematic showing the way hippocampal place cell activity represents possible trajectories in subsequent theta cycles during navigation. (b-f) Schemes for representing a target probability …

Neural variability increases within theta cycle.

(a) Example spiking activity of 250 cells (top) and raw (black) and theta filtered (green) local field potential (bottom) for 6 consecutive theta cycles (vertical lines). (b) Place fields of 6 …

Figure 3 with 3 supplements
Theta sequences in simulated data.

(a) Motion trajectory of the simulated rat (gray, 10 s) together with its own inferred and predicted most likely (mean) trajectory segments (colored arrows) in five locations separated by 2 s …

Figure 3—figure supplement 1
Inference and movement in the generative model.

(a) Graphical model of the processes underlying the generation of the simulated animal’s trajectory. Arrows represent the individual steps in the generative process, orange arrows highlight sensory …

Figure 3—figure supplement 2
Comparison of the motion profile of the simulated animal and one of the analysed experimental sessions.

(a-h) Motion profile of the simulated animal. (a) Histogram of the running speed. Orange vertical line indicates the mean of the distribution. (b) Auto-correlation of the running speed. Orange line …

Figure 3—figure supplement 3
Place cell firing in the synthetic and in the experimental data.

(a-e) Place cell activity in the simulated data using mean encoding. (a) Histogram of the average firing rate of the 200 simulated neurons. Numbers indicate the average firing rate across cells and …

Figure 4 with 3 supplements
Product scheme: population gain decreases with uncertainty.

(a) Decision-tree for identifying the encoding scheme. (b) Schematic of encoding a high-uncertainty (left) and a low-uncertainty (right) target distribution using the product scheme with 4 neurons. …

Figure 4—figure supplement 1
Population gain as a hallmark of the product representation.

(a) The 110 tuning curves ψ(x) from the example session R1D2 used in this analysis as basis function h(x)=logψ(x) (Equation 36), with ψ(x)=ψ(x)/ψ(x)dx. (b) Representing distributions of increasing uncertainty in the product …

Figure 4—figure supplement 2
Theta sequence bias and variability in all recording sessions.

(a) Location and direction-aligned motion trajectories and 0.5 confidence interval ellipses for 1Δt1. (b) Bias and spread of motion trajectories as a function of time in an example session. (c) Left: …

Figure 4—figure supplement 3
No evidence for encoding parameters of distributions via product or DDC schemes in linear track data.

(a) Normalized firing rate of all putative excitatory neurons recorded in a single session (Achilles, up) ordered by the location of the peak activity on rightward runs from a previously published …

Figure 5 with 3 supplements
DDC scheme: diversity increases with uncertainty.

(a) Schematic of encoding a narrow (left) and a wide (right) distribution with spike-based DDC using four neurons. Intuitively, the standard deviation (SD) is represented by the diversity of the …

Figure 5—figure supplement 1
Reducing the bias of the decoded SD in the DDC scheme.

(a) Illustration of the problem of decoding bias. In the DDC scheme, the firing rate of the neurons is proportional to the overlap between the target distribution and the basis functions of the …

Figure 5—figure supplement 2
No evidence for DDC code when decoding spikes using the estimated basis functions instead of the empirical tuning curves.

No evidence for DDC code when decoding spikes using the estimated basis functions instead of the empirical tuning curves. (a-b) Estimated standard deviation of the DDC-decoded distribution from …

Figure 5—figure supplement 3
Summary figure showing the decoded SD and EV-index for all encoding schemes.

(a) Similar to Figure 5c, d but also including data for the product scheme. Estimated standard deviation of the DDC-decoded distribution from spikes at early, mid and late theta phases using the …

Figure 6 with 2 supplements
Sampling scheme: excess variability increases with uncertainty.

(a) To discriminate sampling from mean encoding, we defined the EV-index which measures the magnitude of cycle-to-cycle variability (CCV) relative to the trajectory encoding error (TEE). (b) …

Figure 6—figure supplement 1
EV-index calculated from all theta cycles.

(a) Cumulative distribution of CCV (dashed) and TEE (solid) for early, mid, and late theta phase (colors) in the mean scheme using simulated data. Note the logarithmic x axis. (b) Mean CCV and TEE

Figure 6—figure supplement 2
Spatial representation and EV-index is similar across task phases.

(a) Violin plots showing that the distribution of the running speeds are highly overlapping in home (goal directed navigation) and away (random foraging) trials in all recorded sessions. Symbols …

Figure 7 with 2 supplements
Signature of efficient sampling: generative cycling.

(a) Examples of sampled trajectories with positive (left) and negative (right) correlation between the direction of subsequent trajectory endpoints (squares) relative to the current position …

Figure 7—figure supplement 1
Cycling index is unbiased when the encoded position is compared to the internally estimated rather than the true position.

Similar to Figure 7d and f, using the simulated animal’s internal position and motion direction estimate to calculate the relative direction of the encoded trajectory endpoints in each theta cycle. …

Figure 7—figure supplement 2
Model-free replication of the main findings of the paper.

(a) Ten segments of the recorded motion trajectory of a real animal (black, target) and examples of potential motion trajectories with initial motion direction and velocity matching those of the …

Tables

Table 1
Summary of the symbols used in the model.
SymbolMeaning
nindex of time step in the generative model measured as the number of theta cycles
xnposition at theta cycle n (two-dimensional)
ynsensory input (two-dimensional)
unmotor command (two-dimensional)
x¯nplanned position (2×T-dimensional)
x¯nplanned position (two-dimensional)
y1:npast sensory input until theta cycle n
μnmean of the filtering posterior
Σncovariance of the filtering posterior
φtheta phase
xnx(nnp):(n+nf)trajectory of the animal around theta cycle n
μn(φ)posterior mean trajectory at theta cycle n
Σn(φ)posterior variance of trajectory at theta cycle n
x~ntrajectory sampled from P(xn|y1:n,u1:n)
ϕi(x)encoding basis function of cell i - firing rate as a function of the encoded position
ψi(x)empirical tuning curve of cell i - firing rate as a function of the real position
λifiring rate of cell i
sn(φ)spikes recorded in theta cycle n encoding trajectory xn
x^n(φ)trajectory decoded from the observed spikes assuming direct encoding (Equation 18)
μ^n(φ)estimated trajectory mean assuming DDC encoding (Equation 19)
Σ^n(φ)estimated trajectory variance assuming DDC encoding (Equation 19)
Table 2
Parameters controlling the auto-correlation of the sampled trajectories.
ParameterStrong +Weak +IndependentWeak -Strong -
γϑ(slope)-8-5055
ϑ0(threshold)π/8π/4π/2π/4π/2
Table 3
p-values associated with Figure 6.

p-values for panels f,g and k were calculated using a one sample t-test. p-values for panel l were estimated by bootstrapping.

Panel: f
regularjitter: 0510203040
8.9e-101e-064.4e-071.5e-071e-055.7e-071.8e-07
panel: g
regularjitter: 0510203040
0.00014e-153.9e-151e-133.4e-111.5e-112.4e-06
panel: k
rat1 day1rat1 day2rat2 day1rat2 day2rat3 day1rat3 day2rat4 day1rat4 day2
5e-052.5e-182.5e-050.00012.1e-102.8e-071.4e-082.4e-06
panel: l
rat1 day1rat1 day2rat2 day1rat2 day2rat3 day1rat3 day2rat4 day1rat4 day2
<0.001<0.001<0.001<0.001<0.001<0.001<0.001<0.001
Table 4
p-Values associated with Figure 6—figure supplement 1.

p-Values for panels e,f and j were calculated using a one sample t-test. p-Values for panel k were estimated by bootstrapping.

Panel: e
regularjitter: 0510203040
1.8e-051.4e-059e-054.8e-050.00160.00250.0006
panel: f
regularjitter: 0510203040
0.014e-057e-050.00070.00060.0140.16
panel: j
rat1 day1rat1 day2rat2 day1rat2 day2rat3 day1rat3 day2rat4 day1rat4 day2
0.00184.5e-096e-084.9e-073.7e-194e-113.3e-252.4e-07
panel: k
rat1 day1rat1 day2rat2 day1rat2 day2rat3 day1rat3 day2rat4 day1rat4 day2
<0.001<0.001<0.001<0.001<0.001<0.001<0.001<0.001

Additional files

Download links