Neural antecedents of self-initiated actions in secondary motor cortex

Journal name:
Nature Neuroscience
Volume:
17,
Pages:
1574–1582
Year published:
DOI:
doi:10.1038/nn.3826
Received
Accepted
Published online

Abstract

The neural origins of spontaneous or self-initiated actions are not well understood and their interpretation is controversial. To address these issues, we used a task in which rats decide when to abort waiting for a delayed tone. We recorded neurons in the secondary motor cortex (M2) and interpreted our findings in light of an integration-to-bound decision model. A first population of M2 neurons ramped to a constant threshold at rates proportional to waiting time, strongly resembling integrator output. A second population, which we propose provide input to the integrator, fired in sequences and showed trial-to-trial rate fluctuations correlated with waiting times. An integration model fit to these data also quantitatively predicted the observed inter-neuronal correlations. Together, these results reinforce the generality of the integration-to-bound model of decision-making. These models identify the initial intention to act as the moment of threshold crossing while explaining how antecedent subthreshold neural activity can influence an action without implying a decision.

At a glance

Figures

  1. The waiting task and the behavioral results.
    Figure 1: The waiting task and the behavioral results.

    (a) Schematic of trial events in the waiting task (top). In each trial, after waiting for a certain period at the waiting port, the rat received a tone(s), moved to the reward port and received a water reward, the size of which depended on the number of the tones presented. Inset, probability distributions of the delays to tone 1 (T1, light green) and tone 2 (T2, dark green). Bottom, timeline of the task events and the definition of the behavior parameters. The light green rectangle indicates the presentation of tone 1, the dark green rectangle represents tone 2 and the light blue rectangle indicates reward. Tone 2 is represented by a hatched rectangle to indicate it was not played in the impatient trials. (b) Snapshot of the waiting behavior. The waiting period in each trial is indicated as a gray bar. Light green ticks represent a presentation of tone 1 and dark green ticks represent tone 2. (c) Waiting time histograms of short poke trials (gray), impatient trials (red) and patient trials (blue) of an example rat. The histograms show data pooled across sessions. Inset, cumulative histogram of waiting times in impatient trials from this rat. The arrow indicates the range from 10th to 90th percentile waiting times (ΔWT [0.1–0.9]). (d) Distribution of ΔWT[0.1–0.9] across rats. Filled bars indicate electrophysiology rats. (e) Distribution of CV2 across rats. Filled bars indicate rats used for electrophysiology. (f) A histogram of response time to tone 2 of an example rat (dark blue, n = 1,501 trials). Light blue–shaded area indicates 95% range of response time histograms from shuffled data. The peak response time is indicated by an arrowhead. (g) Distribution of peak response time across rats. Significant peak is shown in dark blue and non-significant peak in black. Filled bars indicate rats used for electrophysiology (all were significant).

  2. Difference in movement times in impatient and patient trials.
    Figure 2: Difference in movement times in impatient and patient trials.

    (a) A scatter plot indicating median movement time in the impatient and the patient trials of different sessions of an example rat. Each gray circle indicates median movement time of the impatient trials and that of the patient trials from one session. The black circle indicates the mean of the median movement times of the impatient trials and that of the patient trials. Error bar represent ±s.e.m. (b) Normalized mean movement times for impatient (red) and patient (blue) trials. For each rat, the movement time is normalized with movement time of the impatient trial. Error bar represents ±s.e.m. Gray circles represent the normalized movement times of the patient trials of individual rats. Filled circles indicate rats used for electrophysiology. Movement time in patient trials was significantly faster than that in impatient trials (Wilcoxon signed-rank test, P < 0.001, n = 37 rats).

  3. Ramp-to-threshold type predictive activity.
    Figure 3: Ramp-to-threshold type predictive activity.

    (af) Example M2 neuron with ramp-to-threshold type activity. (a) Perievent time histograms (PETHs) for an M2 neuron in different waiting time trials, aligned to poke in and smoothed with a Gaussian filter (s.d. = 50 ms). Impatient trials are grouped according to the waiting time, indicated by the color scale in b, and consistent throughout the figure. Dashed lines in PETHs indicate times at which a rat already left the port in some of the trials in that group. Threshold (57 spikes per s) is indicated by the horizontal solid line. (b) PETHs for the neuron presented in a, aligned to poke out. (c) Time to cross a threshold firing as a function of mean waiting time. The analyses with the highest and lowest thresholds with significant correlation (57 spikes per s (triangle) and 16 spikes per s (inverted triangle), respectively) are shown. R = 0.99, P < 0.001, n = 9 for the 57 spikes per s threshold; R = 0.86, P = 0.001, n = 10 for the 16 spikes per s threshold. Dashed lines indicate the regression line. (d) The rate of ramping activity as a function of mean waiting time. The dashed line indicates the regression line. R = −0.93, P < 0.001, n = 9. (e) The firing rate at the poke-out period (50-ms window before the poke out) is plotted against the mean waiting time for each group. Note that firing rate reached almost the same level at the poke out. R = 0.58, P = 0.076, n = 10. (f) Difference between time to cross threshold and the waiting time (prediction time) is plotted against the threshold tested. Black circles represent mean prediction time across groups, error bars represent ±s.e.m. and filled black circles represent significant threshold. (gi) Population data (n = 27 neurons). (g) Distribution of correlation coefficients between the rate of ramping and the waiting time. Ramp-up neurons are shown in pink, ramp-down neurons in blue and neurons with significant correlation in vivid color. (h) Distribution of correlations between the firing rate at the poke-out period and the waiting time. (i) Distribution of the earliest prediction time (Online Methods).

  4. An M2 neuron with transient predictive activity.
    Figure 4: An M2 neuron with transient predictive activity.

    (a) Raster plots (top) represent activity of an M2 neuron, with each row corresponding to a single trial aligned to poke in (white line) and each black tick to a single spike. The impatient trials are shown on the pink background and the patient trials on the blue background. Trials are chronologically ordered from top to bottom in each type of trials. Color ticks represent tone 1 (light green), tone 2 (dark green), poke out (white) and poke in into the reward port (light blue). PETHs at the bottom represent activity in the impatient (red) and patient trials (blue), smoothed with a Gaussian filter (s.d. = 50 ms). (b) The same neuron in the impatient trials. Trials are sorted in ascending order of the waiting time. PETHs (bottom) of trials grouped by waiting time, as indicated by a color scale. (c) Mean waiting time is plotted against mean firing rate at 0–0.4 s from poke in. Trials are grouped according to firing rates (4–spikes per s bin) for the visualization purpose only. Error bars represent ±s.e.m. Circles without error bars represent groups with 1 or 2 trials. Dashed line: regression line. (d) Significance (P value) of the Pearson's correlation coefficient was calculated at each 0.4-s non-overlapping bin and plotted as a function of time (corrected for the multiple comparisons). Significance level (P = 0.05) is indicated by the dashed line. The significance of each time bin is also indicated by the color bar on top. N.S., not significant; N.D., no data.

  5. Population data of predictive activity.
    Figure 5: Population data of predictive activity.

    (a,b) Other examples of M2 neurons with predictive activity. (a) Shown is an M2 neuron with sustained activation during waiting and whose firing rate was positively correlated with the waiting time. Data are presented as in Figure 4b,d. (b) Shown is an M2 neuron with negative correlation between the firing rate and the waiting time. Data are presented as in Figure 4b,d. The color scale in green indicates a positive correlation and orange a negative correlation. (c) Time course and the sign of the correlation for all the predictive neurons (n = 64 neurons). The significance of the correlation is calculated for 0.4-s overlapping time window in every 0.02-s time step for each neuron and indicated in each row. Neurons are sorted according to a center of mass of log(P value). Only the time bins with significant P value were used to calculate the center of mass. The color code is the same as in b. Bonferroni correction for multiple comparisons was used to select neurons with transient correlation, but the P value here is not corrected for the multiple comparisons, as the main points are the time course and the sign of the predictive activity, not its absolute value. (d) Time course of fraction of predictive neurons. For each neuron, waiting time correlation with firing rate was tested on subsamples of 30 impatient trials, allowing comparisons across time bins. Subsampling was performed 1,000 times and error bars represent 95th percentile ranges. The white dashed line indicates chance level of 0.05.

  6. Action specificity of predictive activity.
    Figure 6: Action specificity of predictive activity.

    (a) A schematic diagram of a waiting task with interleaved blocks of the nose-poke waiting trials and the lever-press waiting trials (Online Methods). (b) An example of nose poke–specific predictive neurons. PETHs indicate activities of the neuron in the nose-poke waiting trials (left) and in the lever-press waiting trials (right). The color of the bar on top indicates the significance of the time bins. Data are presented as in Figure 4b,d. (c) The predictive activities of the two types of waiting trials are represented independently in M2. Each circle represents one neuron, indicating the correlation coefficient between the firing rate and the waiting time in the nose-poke trials on the x axis and the correlation coefficient between the firing rate and the waiting time in the lever-press trials on the y axis. The correlation coefficient was calculated at the most significant time bin for each type of trials.

  7. Action specificity of ramp-to-threshold activity.
    Figure 7: Action specificity of ramp-to-threshold activity.

    (a) An example of nose poke–specific predictive neurons. PETHs indicate activities of the neuron in the nose-poke waiting trials (left) and in the lever-press waiting trials (right). Data are presented as in Figure 3. (b) Poke-out/delay selectivity index and lever-release/delay selectivity index for all of the nose-poke predictive neurons (ramp-to-threshold type). Of six nose-poke predictive neurons, two neurons did not show significant difference between activity at the lever-release period and activity at the delay period (open circle), and were therefore not tested with threshold-type predictive activity. The other four neurons showed significant difference in activity at the lever-release period and delay period (black and red filled circles), and were therefore tested with the threshold-type predictive activity for the lever-release time. One of them showed significant predictive activity in the lever-press trials, but the direction of ramping activity was the opposite (red filled circle). (c) An example of the lever press–specific predictive neurons. Data are presented as in Figure 3. (d) Lever-release/delay selectivity index and poke-out/delay selectivity index for all the lever-press predictive neurons (ramp-to-threshold type). Of 15 lever-press predictive neurons, nine neurons did not show significant difference between activity at the poke-out period and activity at the delay period (open circle). Six neurons showed significant difference in activity at the poke-out period and delay period (black and red filled circle), and were therefore tested with the threshold-type predictive activity for the poke-out time. One of them (the same as the red neuron in b) showed significant, but opposite, predictive activity in the nose-poke trials.

  8. Integrator model.
    Figure 8: Integrator model.

    (a) A schematic diagram of an integrator model. Circles with 'I' indicate input neurons. A circle with '∫' indicates an integrator neuron. A small triangle indicates an excitatory synapse and a small circle indicates an inhibitory synapse. Inset panels show PETHs of example model neurons (top three panels are example input neurons and the bottom left panel is an integrator neuron; data are presented as in Fig. 3a). The bottom right inset panel shows a waiting time histogram of the model (mean ± s.e.m. of 1,000 model sessions of 100 trials). (b) Waiting time correlation for all the input neurons from three example models with different parameters (number of neurons per time: 10 (left), 300 (middle and right); fraction of common noise (β): 0 (left and middle), 0.4 (right)). Color indicates P value of waiting time correlation. Neurons are arranged according to its activation time (x axis) and synaptic weight (y axis, positive weight at the top and negative weight at the bottom). (c) Left, pairwise partial correlation between simultaneously recorded neurons as a function of time difference of the most predictive time bins of each neuron. Pairs are categorized as the same sign (green), opposite sign (blue) or other (gray) according to the sign of waiting time correlation of each neuron. Error bars indicate s.e.m. N = 1,836 pairs. Right, pairwise partial correlation between input neurons in the model as a function of time difference of their activities (number of neurons per time, 300; fraction of common noise (β), 0.4). Mean ± s.e.m. of 100 model sessions. Error bars are too small to be visible. (d) Fraction of predictive neurons as a function of the number of neurons per time step and fraction of common noise in the input neuron activity (β). Mean ± s.e.m. of 1,000 model sessions.

  9. Histograms of response times to tone 2 of all the rats.
    Supplementary Fig. 1: Histograms of response times to tone 2 of all the rats.

    Rats used for electrophysiology are indicated by (E) after a rat name on top. Rats without a significant peak in a response time histogram are indicated by an asterisk after a rat name. The same format as in Figure 1f.

  10. M2 neurons were activated in different phases of the task.
    Supplementary Fig. 2: M2 neurons were activated in different phases of the task.

    (a) An example neuron which was activated at the poke-in period (from 0.2 s before to 0.2 s after the poke-in). The left panel shows the activity aligned to the poke-in, the middle panel aligned to the poke-out, the right panel aligned to the water delivery. Raster plots (top) represent activity of the neuron around the waiting period, with each row corresponding to a single trial and each black tick to a single spike. The impatient trials in pink background and the patient trials in blue background. Trials are chronologically ordered from top to bottom within each type of trials. Color ticks represent Tone 1 (light green), Tone 2 (dark green), poke-in (white), poke-out (white), poke-in into the reward port (light blue) and water delivery (blue). Perievent time histograms (PETHs) at the bottom represent activity in the impatient trials (red) and the patient trials (blue), smoothed with a Gaussian filter (s.d. = 50 ms). (b) A neuron which was activated at the delay period (from 0.4 s after the poke-in to 0.4 s before the poke-out). (c) A neuron which was activated at the poke-out period (from 0.2 s before to 0.2 s after the poke-out). (d) A neuron which was activated at the water-poke-in period (from 0.2 s before to 0.2 s after the water-poke-in). (e) A neuron which was activated at water delivery (from 0.1 s to 0.5 s after the water-delivery). (f) Fraction of neurons activated (open bar) or suppressed (filled bar) at each period (Mean ± SEM across rats) (Wilcoxon signed-rank test, P < 0.05, comparing the firing rate at the period of interest with that at 0.4 s time window randomly chosen in each trial).

  11. Flow of the analysis for waiting time predictive neurons.
    Supplementary Fig. 3: Flow of the analysis for waiting time predictive neurons.

    (a) Flow of the ramp-to-threshold neuron analysis. Numbers in the parenthesis represent number of neurons in each category. The analysis classified 27 neurons as ramp-to-threshold neurons (0,4,4,6,7,4,1,1 neurons from each rat). (b) Flow of the transient neuron analysis. The analysis classified 64 neurons as transient neurons (3,3,0,6,22,27,2,1 neurons from each rat). The details of each analysis are described in Online Methods.

  12. Schematic drawings of hypothetical activity of ramp-to-threshold type predictive neurons.
    Supplementary Fig. 4: Schematic drawings of hypothetical activity of ramp-to-threshold type predictive neurons.

    A difference in times to cross the threshold in different waiting time trials could result from identical rate but different onset time of ramping activity, which would be expected from a stereotypical movement-related activity27. In contrast, for a neuron reflecting the output of a neural integrator, the difference in the threshold crossing times should arise from the different rates of ramping. (a) Perievent time histograms of a hypothetical “preparatory-type” ramp-to-threshold neuron. The format is the same as in Figure 3. Two horizontal lines indicate a high threshold line and a low threshold line. (b) Perievent time histograms of a hypothetical “movement-type” ramp-to-threshold neuron. The format is the same as in Figure 3. (c) Relationship between the waiting time and the ramp rate for the preparatory-type neuron. The ramp rate was measured between the high and low threshold lines indicated in a. The logarithm of ramp rate is negatively correlated with the logarithm of waiting time. (d) Relationship between the waiting time and the ramp rate for the movement-type neuron. The ramp rate is constant and independent of the waiting time.

  13. Choice of waiting action at block switches between nose-poke waiting blocks and lever-press waiting blocks.
    Supplementary Fig. 5: Choice of waiting action at block switches between nose-poke waiting blocks and lever-press waiting blocks.

    (a) An example session in which a rat performed both nose-poke and lever-press waiting task. The x-axis indicates number of trials and y-axis the waiting time in each trial. The waiting time of the nose-poke trials is indicated in a positive direction in the y-axis and a lever-press waiting time in a negative direction. The lever-press block is indicated with a light yellow background. Gray circle indicates the short poke trials, red the impatient trials, and blue the patient trials. (b) Choice between nose-poke waiting and lever-press waiting around block switches. Fraction of trials in which rats chose to perform nose-poke waiting is plotted as a function of trials (20 trials before switch and 40 trials after switch). The left panel shows the choice of waiting action at the block switch from the nose-poke block to the lever-press block. The right panel shows the choice of waiting action at the block switch from the lever-press block to the nose-poke block. All the recording sessions from 3 rats are combined. The gray shade indicates 95% confidence interval calculated with binomial fitting. The lever-press block is indicated by light yellow background. The rats quickly adapted to the block switch within a few trials.

  14. Transient correlation analysis focused on a period before poke in in the current trial and during intertrial interval (ITI).
    Supplementary Fig. 6: Transient correlation analysis focused on a period before poke in in the current trial and during intertrial interval (ITI).

    Ninety-one neurons are selected with the same transient correlation analysis as in the main result, but focused on different time bins (Eight non-overlapping 1.5s time bins, including 2 bins from a 3s before poke-in in the current trial and 2 bins from the first 3s ITI period of the previous trials (1 – 3 trials back)). An ITI period of 3 trial back is shown in (a), 2 trials back in (b), 1 trial back in (c) and a pre-poke-in period of the current trial in (d). P-value shown here are calculated with 0.4s overlapping time bins with 0.02s time steps for each neuron. The color code is the same as in Figure 5c. Bonferroni correction for multiple comparisons were used for selecting neurons (P < 0.05), but P-value shown here was not corrected. Note that the waiting time predictive activity was already present at 2 – 3 seconds before the poke-in and during ITI period after the previous trial, but much weaker in 2 trials back and 3 trials back.

  15. Difference in activities in the impatient and patient trials of the predictive neurons.
    Supplementary Fig. 7: Difference in activities in the impatient and patient trials of the predictive neurons.

    Because Tone 2 terminated the waiting of rats in patient trials, the rats’ willingness to wait was presumably longer than the actual waiting time. This gives rise to two additional predictions that we were able to test. First, transient waiting-time predictive neurons that fired at a higher rate in long waiting impatient trials should fire even more vigorously in patient trials, while neurons which fired less in long waiting trials should fire even less in the patient trials. Second, in patient trials, the ramp-to-threshold neurons should not reach the threshold firing rate for leaving before Tone 2. (a) A scatter plot showing the relationship between the correlation between firing rates and waiting times in the impatient trials and impatient/patient selectivity index. We calculated an impatient/patient selectivity index using the time window(s) with significant predictive activity in the correlation analysis. Because we wished to compare activities in impatient and patient trials without a contribution of difference in waiting time distribution, we selected subset of impatient and patient trials to match the waiting time distributions from two trial types, as described for the analysis of the movement time. The impatient/patient selectivity index was defined as the difference in a mean firing rate in the impatient and patient trials normalized by the sum of the mean firing rates. The value ranges from –1 to 1; 1 indicates a neuron selective for the impatient trials. Each circle represents a time bin with significant correlation between firing rate and waiting time in the impatient trials. A neuron which has multiple time bins with significant correlation is represented more than once in the plot. Filled circles indicate time bins with significant impatient/patient selectivity index (Wilcoxon signed-rank test, P < 0.05). Open circles indicate time bins with non-significant selectivity index. (b) Distribution of the impatient/patient selectivity index for all time bins with significant positive correlation between firing rates and waiting times in impatient trials. Arrow indicates the median selectivity index. The distribution is significantly shifted toward negative (Wilcoxon signed-rank test, P < 0.01) (c) The same as in b but for time bins with negative correlation. The distribution is significantly shifted toward positive (Wilcoxon signed-rank test, P < 0.0001). Analysis of patient trials for the ramp-to-threshold type neurons (d,e). (d) Activity of an example ramp-to-threshold neuron (the same neuron as in Figure 3) in impatient and patient trials aligned to poke-out. A gray shaded area indicates the poke-out time window (from –500 ms to –250 ms from poke-out) used for the population analysis in e. A green dashed line indicates a range of possible Tone 2 onset time in the patient trials. Activity was higher in the impatient trials than in patient trials in the analysis window (P < 0.05, Wilcoxon signed-rank test). (e) Distribution of impatient/patient selectivity index for ramp-to-threshold neurons. Patient trials with more than 250 ms reaction time were excluded to prevent contamination of Tone 2 response in the analysis time window. Patient trials with less than 60 ms response time were also excluded, because probably those are trials in which Tone 2 presentation happened to coincide with the time the rat was about to leave, estimated from the response time distribution pooled across rats (data not shown). After exclusion of those trials, the procedure for matching the waiting time distribution of impatient and patient trials was performed, as described before. Ramp-up and ramp-down neurons were pooled together. The sign of impatient/patient selectivity index for the ramp-down neurons was flipped. A red histogram indicates neurons with significant correlation between ramp rates and waiting times. A black histogram indicates neurons without significant correlation. While a distribution of the selectivity index of neurons with significant ramp rate correlation was shifted toward positive (Wilcoxon signed-rank test, P < 0.05), that of neurons without significance was not (P > 0.5).

  16. Effect of waiting times and ongoing movements on M2 neural activity.
    Supplementary Fig. 8: Effect of waiting times and ongoing movements on M2 neural activity.

    We used a stepwise regression (MATLAB, stepwisefit), because the number of independent variables is relatively large compared with the data size. The significance of each coefficient was estimated by running the same analysis after shuffling a variable of interest across trials (1000 times). We first tracked the position (X, Y) and orientation of the rat’s body, using custom-written software with Python using the OpenCV library. For the firing rate correlation analysis, the average firing rate of each 400 ms time bin was regressed using stepwise regression using for independent variables waiting time and 6 movement variables of the same time bin: X-, Y-, angular-position, X-, Y- and angular-velocity. P-values were corrected for multiple comparisons (Bonferroni correction). (a) An example neuron that showed a significant correlation between waiting time and firing rate with the stepwise regression analysis. Perievent time histograms (PETHs) around start of waiting shown in the same way as in Figure 4 (left). A scatter plot shows the relationship between X position (x-axis), waiting time (y-axis) and firing rate (color axis) (right). Note that the color gradient (i.e. firing rate gradient) is obvious in the waiting time axis but not in the X position axis. None of the other movement variables explained the firing rate (data not shown). (b) An example neuron that showed a spurious correlation between waiting time and firing rate in the original analysis, which disappears if movement variables are included in the regression analysis. PETHs around start of waiting (left). A scatter plot shows the relationship between Y position (x-axis), waiting time (y-axis) and firing rate (color axis) (right). Note that the firing rate gradient is obvious in the Y position axis but not as much in the waiting time axis. (c) Fractions of neurons with firing rate significantly correlated with waiting times, as described in the main text, applied to the neurons from video recorded sessions (n = 164 neurons) is shown in gray bar. Fractions of neurons with firing rate significantly correlated with each of the variables shown in the bottom labels using a stepwise regression analysis are shown in black and white bars. (d) Fractions of neurons with significant correlation with each variable shown in the label among waiting time predictive neurons in the original analysis. Eighty percent (20 of 25) of neurons remained significantly correlated with waiting times with the stepwise regression. (e) Fractions of neurons with rate of ramping activity significantly correlated with waiting times among ramp-to-threshold type predictive neurons from video recorded sessions, as described in the main text, are shown in gray. Fractions of neurons with ramp-rate significantly correlated with each of the variables described in the label are shown in black and white bars. Seventy-five percent (6 of 8) of ramp-rate correlated neurons remained significant with the stepwise regression. For the ramp-rate correlation analysis, we regressed the rate of ramping activity with the same variables described above. Motor variables were calculated at a single time bin from 0 to 400 ms relative to the start of waiting, because this is the period common to all the trials. We also performed the analysis using the entire waiting time as a single time bin (different bin size for different waiting time trials). This yielded similar results (data not shown).

  17. Trial history analysis using a stepwise multiple linear regression.
    Supplementary Fig. 9: Trial history analysis using a stepwise multiple linear regression.

    (a) A fraction of neurons with activity significantly correlated with waiting times (current impatient trials) using a standard simple linear regressions analysis (analysis described in the main text) is shown in gray (left most bar). Fraction of neurons significantly correlated with each of the trial history variables shown in the bottom labels using a stepwise multiple linear regression are shown in black (waiting time of current impatient trials) and white bars (other trial history variables). We used a stepwise regression (MATLAB, stepwisefit), because the number of independent variables is relatively large compared with the data size (number of trials). The significance of each coefficient was estimated by running the same analysis after shuffling a variable of interest across trials (1000 times). Note that using the stepwise multiple linear regression analysis including trial history variables, 15.7 % of neurons are significantly correlated with waiting times of the current trial (a black bar), comparable to 18.0% with the analysis shown in Figures 5 and 6 (a gray bar). (b) Time course and the sign of correlation for all the waiting time predictive neurons with stepwise multiple linear regressions (n = 56 neurons). The format is the same as is in Figure 5c. The time course of the predictive activity was also comparable to the original analysis. We also observed neurons that carried information about trial history, consistent with previous studies21. We did not further analyze those correlations because it was beyond the scope of our current study.

  18. Anatomical location of recording sites.
    Supplementary Fig. 10: Anatomical location of recording sites.

    (a) Nissl stained coronal section of a rat frontal cortex. The arrow indicates a site of an electrolytic lesion. (b) Fluorescent image of an adjacent section. The arrow indicates a track of DiI coated tetrode. (c) Corresponding section from the rat brain atlas52. (d) Recording sites of all the neurons analyzed for the ramp-up predictive activity (correlation between time to cross a threshold firing rate and waiting time). Neurons with significant correlation between times to cross a threshold firing rate and waiting times are shown with color. Open pink circle, ramp-up neurons without significant correlation between ramp rates and waiting times. Filled pink circle, ramp-up neurons with significant correlation between ramp rates and waiting times. No obvious clustering of predictive neurons was observed. (e) The same as in d but for ramp-down neurons. Open blue circle, ramp-down neurons without significant correlation between ramp rates and waiting times. Filled blue circle, ramp-down neurons with significant correlation between ramp rates and waiting times. No obvious clustering of predictive neurons was observed. (f) Recording site of all the neurons analyzed for the predictive activity (correlation between firing rate and waiting time at a certain time window). Neurons whose activity was positively correlated with waiting time are shown in green, while neurons with negative correlations are shown in orange. No obvious clustering of predictive neurons was observed.

Videos

  1. Behavior from an example recording session.
    Video 1: Behavior from an example recording session.
    For this rat, the nose-poke waiting port was located at the left side, lever-press waiting port at the right side and reward port at the center. Entering the nose-poke waiting port was indicated by a filled white circle on top of the waiting port, which is an open circle outside the waiting period. Presentation of Tone 1 is indicated by light green bars at both sides and Tone 2 by dark green bars. Delivery of a water reward was indicated by a blue circle appearing on top of the reward port. Although the onset of the presentation of tones and water is accurate, the duration of those are not accurately represented in the movie, because they are brief in the actual experiment and hard to see in the movie.

References

  1. Uchida, N., Kepecs, A. & Mainen, Z.F. Seeing at a glance, smelling in a whiff: rapid forms of perceptual decision making. Nat. Rev. Neurosci. 7, 485491 (2006).
  2. Wang, X.J. Probabilistic decision making by slow reverberation in cortical circuits. Neuron 36, 955968 (2002).
  3. Mazurek, M.E., Roitman, J.D., Jochen, D. & Shadlen, M.N. A role for neural integrators in perceptual decision making. Cereb. Cortex 13, 12571269 (2003).
  4. Gold, J.I. & Shadlen, M.N. The neural basis of decision making. Annu. Rev. Neurosci. 30, 535574 (2007).
  5. Hanes, D.P. & Schall, J.D. Neural control of voluntary movement initiation. Science 274, 427 (1996).
  6. Ratcliff, R., Cherian, A. & Segraves, M. A comparison of Macaque behavior and superior colliculus neuronal activity to predictions from models of two-choice decisions. J. Neurophysiol. 90, 13921407 (2003).
  7. Roitman, J.D. & Shadlen, M.N. Response of neurons in the lateral intraparietal area during a combined visual discrimination reaction time task. J. Neurosci. 22, 94759489 (2002).
  8. Berns, G.S., Laibson, D. & Loewenstein, G. Intertemporal choice—toward an integrative framework. Trends Cogn. Sci. 11, 482488 (2007).
  9. McGuire, J.T. & Kable, J.W. Rational temporal predictions can underlie apparent failures to delay gratification. Psychol. Rev. 120, 395410 (2013).
  10. Mischel, W., Ebbesen, E.B. & Zeiss, A.R. Cognitive and attentional mechanisms in delay of gratification. J. Pers. Soc. Psychol. 21, 204218 (1972).
  11. Cardinal, R.N., Winstanley, C.A., Robbins, T.W. & Everitt, B.J. Limbic corticostriatal systems and delayed reinforcement. Ann. NY Acad. Sci. 1021, 3350 (2004).
  12. Roesch, M.R., Calu, D.J., Burke, K.A. & Schoenbaum, G. Should I stay or should I go? Transformation of time-discounted rewards in orbitofrontal cortex and associated brain circuits. Ann. NY Acad. Sci. 1104, 2134 (2007).
  13. Deecke, L., Scheid, P. & Kornhuber, H.H. Distribution of readiness potential, pre-motion positivity and motor potential of the human cerebral cortex preceding voluntary finger movements. Exp. Brain Res. 7, 158168 (1969).
  14. Libet, B., Gleason, C.A., Wright, E.W. & Pearl, D.K. Time of conscious intention to act in relation to onset of cerebral activity (readiness-potential). Brain 106, 623642 (1983).
  15. Okano, K. & Tanji, J. Neuronal activities in the primate motor fields of the agranular frontal cortex preceding visually triggered and self-paced movement. Exp. Brain Res. 66, 155166 (1987).
  16. Romo, R. & Schultz, W. Role of primate basal ganglia and frontal cortex in the internal generation of movements. III. Neuronal activity in the supplementary motor area. Exp. Brain Res. 91, 396407 (1992).
  17. Fried, I., Mukamel, R. & Kreiman, G. Internally generated preactivation of single neurons in human medial frontal cortex predicts volition. Neuron 69, 548562 (2011).
  18. Lebedev, M.A., O'Doherty, J.E. & Nicolelis, M.A.L. Decoding of temporal intervals from cortical ensemble activity. J. Neurophysiol. 99, 166186 (2008).
  19. Erlich, J.C., Bialek, M. & Brody, C.D. A cortical substrate for memory-guided orienting in the rat. Neuron 72, 330343 (2011).
  20. Guo, Z.V. et al. Flow of cortical activity underlying a tactile decision in mice. Neuron 81, 179194 (2014).
  21. Sul, J.H., Jo, S., Lee, D. & Jung, M.W. Role of rodent secondary motor cortex in value-based action selection. Nat. Neurosci. 14, 12021208 (2011).
  22. Reep, R.L., Corwin, J. & Hashimoto, A. Efferent connections of the rostral portion of medial agranular cortex in rats. Brain Res. Bull. 19, 203221 (1987).
  23. Reep, R.L. & Goodwin, G. Topographic organization in the corticocortical connections of medial agranular cortex in rats. J. Comp. Neurol. 294, 262280 (1990).
  24. Jaramillo, S. & Zador, A.M. The auditory cortex mediates the perceptual effects of acoustic temporal expectation. Nat. Neurosci. 14, 246251 (2011).
  25. Lauwereyns, J., Watanabe, K., Coe, B. & Hikosaka, O. A neural correlate of response bias in monkey caudate nucleus. Nature 418, 413417 (2002).
  26. Zohary, E., Shadlen, M.N. & Newsome, W.T. Correlated neuronal discharge rate and its implications for psychophysical performance. Nature 370, 140143 (1994).
  27. Maimon, G. & Assad, J.A. A cognitive signal for the proactive timing of action in macaque LIP. Nat. Neurosci. 9, 948955 (2006).
  28. Harvey, C.D., Coen, P. & Tank, D.W. Choice-specific sequences in parietal cortex during a virtual-navigation decision task. Nature 484, 6268 (2012).
  29. Fujisawa, S., Amarasingham, A., Harrison, M.T. & Buzsáki, G. Behavior-dependent short-term assembly dynamics in the medial prefrontal cortex. Nat. Neurosci. 11, 823833 (2008).
  30. Lau, B. & Glimcher, P.W. Action and outcome encoding in the primate caudate nucleus. J. Neurosci. 27, 1450214514 (2007).
  31. Stuphorn, V., Brown, J.W. & Schall, J.D. Role of supplementary eye field in saccade initiation: executive, not direct, control. J. Neurophysiol. 103, 801816 (2010).
  32. Milosavljevic, M., Malmaud, J. & Huth, A. The drift diffusion model can account for the accuracy and reaction time of value-based choices under high and low time pressure. Judgm. Decis. Mak. 5, 437449 (2010).
  33. Ratcliff, R. A theory of memory retrieval. Psychol. Rev. 85, 59108 (1978).
  34. Machens, C.K., Romo, R. & Brody, C.D. Flexible control of mutual inhibition: a neural model of two-interval discrimination. Science 307, 11211124 (2005).
  35. Renart, A. et al. The asynchronous state in cortical circuits. Science 327, 587590 (2010).
  36. Soltani, A. & Wang, X.-J. Synaptic computation underlying probabilistic inference. Nat. Neurosci. 13, 112119 (2010).
  37. Gremel, C.M. & Costa, R.M. Premotor cortex is critical for goal-directed actions. Front. Comput. Neurosci. 7, 110 (2013).
  38. Gremel, C.M. & Costa, R.M. Orbitofrontal and striatal circuits dynamically encode the shift between goal-directed and habitual actions. Nat. Commun. 4, 2264 (2013).
  39. Narayanan, N.S. & Laubach, M. Top-down control of motor cortex ensembles by dorsomedial prefrontal cortex. Neuron 52, 921931 (2006).
  40. Botvinick, M.M., Cohen, J.D. & Carter, C.S. Conflict monitoring and anterior cingulate cortex: an update. Trends Cogn. Sci. 8, 539546 (2004).
  41. Brecht, M. et al. Organization of rat vibrissa motor cortex and adjacent areas according to cytoarchitectonics, microstimulation, and intracellular stimulation of identified cells. J. Comp. Neurol. 479, 360373 (2004).
  42. Neafsay, E.J. et al. The organization of the rat motor cortex: a microstimulation mapping study. Brain Res. 396, 7796 (1986).
  43. Stuesse, S.L. & Newman, D.B. Projections from the medial agranular cortex to brain stem visuomotor centers in rats. Exp. Brain Res. 80, 532544 (1990).
  44. Felsen, G. & Mainen, Z.F. Neural substrates of sensory-guided locomotor decisions in the rat superior colliculus. Neuron 60, 137148 (2008).
  45. Lo, C.-C. & Wang, X.-J. Cortico-basal ganglia circuit mechanism for a decision threshold in reaction time tasks. Nat. Neurosci. 9, 956963 (2006).
  46. Zingg, B. et al. Neural networks of the mouse neocortex. Cell 156, 10961111 (2014).
  47. Kepecs, A., Uchida, N., Zariwala, H.a & Mainen, Z.F. Neural correlates, computation and behavioural impact of decision confidence. Nature 455, 227231 (2008).
  48. Lak, A. et al. Orbitofrontal cortex is required for optimal waiting based on decision confidence. Neuron (in the press).
  49. Schall, J.D. Neural basis of deciding, choosing and acting. Nat. Rev. Neurosci. 2, 3342 (2001).
  50. Dehaene, S., Changeux, J.-P., Naccache, L., Sackur, J. & Sergent, C. Conscious, preconscious, and subliminal processing: a testable taxonomy. Trends Cogn. Sci. 10, 204211 (2006).
  51. Holt, G.R., Softky, W. & Koch, C. Comparison of discharge variability in vitro and in vivo in cat visual cortex neurons. J. Neurophysiol. 75, 18061814 (1996).
  52. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates (Academic Press, 2005).

Download references

Author information

Affiliations

  1. Champalimaud Neuroscience Programme, Champalimaud Centre for the Unknown, Lisbon, Portugal.

    • Masayoshi Murakami,
    • M Inês Vicente,
    • Gil M Costa &
    • Zachary F Mainen

Contributions

M.M. and Z.F.M. designed the experiments, analyses and models and wrote the manuscript. M.M. conducted the experiments with assistance from M.I.V. and G.M.C. M.M. analyzed the data and implemented the model.

Competing financial interests

The authors declare no competing financial interests.

Corresponding author

Correspondence to:

Author details

Supplementary information

Supplementary Figures

  1. Supplementary Figure 1: Histograms of response times to tone 2 of all the rats. (150 KB)

    Rats used for electrophysiology are indicated by (E) after a rat name on top. Rats without a significant peak in a response time histogram are indicated by an asterisk after a rat name. The same format as in Figure 1f.

  2. Supplementary Figure 2: M2 neurons were activated in different phases of the task. (233 KB)

    (a) An example neuron which was activated at the poke-in period (from 0.2 s before to 0.2 s after the poke-in). The left panel shows the activity aligned to the poke-in, the middle panel aligned to the poke-out, the right panel aligned to the water delivery. Raster plots (top) represent activity of the neuron around the waiting period, with each row corresponding to a single trial and each black tick to a single spike. The impatient trials in pink background and the patient trials in blue background. Trials are chronologically ordered from top to bottom within each type of trials. Color ticks represent Tone 1 (light green), Tone 2 (dark green), poke-in (white), poke-out (white), poke-in into the reward port (light blue) and water delivery (blue). Perievent time histograms (PETHs) at the bottom represent activity in the impatient trials (red) and the patient trials (blue), smoothed with a Gaussian filter (s.d. = 50 ms). (b) A neuron which was activated at the delay period (from 0.4 s after the poke-in to 0.4 s before the poke-out). (c) A neuron which was activated at the poke-out period (from 0.2 s before to 0.2 s after the poke-out). (d) A neuron which was activated at the water-poke-in period (from 0.2 s before to 0.2 s after the water-poke-in). (e) A neuron which was activated at water delivery (from 0.1 s to 0.5 s after the water-delivery). (f) Fraction of neurons activated (open bar) or suppressed (filled bar) at each period (Mean ± SEM across rats) (Wilcoxon signed-rank test, P < 0.05, comparing the firing rate at the period of interest with that at 0.4 s time window randomly chosen in each trial).

  3. Supplementary Figure 3: Flow of the analysis for waiting time predictive neurons. (160 KB)

    (a) Flow of the ramp-to-threshold neuron analysis. Numbers in the parenthesis represent number of neurons in each category. The analysis classified 27 neurons as ramp-to-threshold neurons (0,4,4,6,7,4,1,1 neurons from each rat). (b) Flow of the transient neuron analysis. The analysis classified 64 neurons as transient neurons (3,3,0,6,22,27,2,1 neurons from each rat). The details of each analysis are described in Online Methods.

  4. Supplementary Figure 4: Schematic drawings of hypothetical activity of ramp-to-threshold type predictive neurons. (96 KB)

    A difference in times to cross the threshold in different waiting time trials could result from identical rate but different onset time of ramping activity, which would be expected from a stereotypical movement-related activity27. In contrast, for a neuron reflecting the output of a neural integrator, the difference in the threshold crossing times should arise from the different rates of ramping. (a) Perievent time histograms of a hypothetical “preparatory-type” ramp-to-threshold neuron. The format is the same as in Figure 3. Two horizontal lines indicate a high threshold line and a low threshold line. (b) Perievent time histograms of a hypothetical “movement-type” ramp-to-threshold neuron. The format is the same as in Figure 3. (c) Relationship between the waiting time and the ramp rate for the preparatory-type neuron. The ramp rate was measured between the high and low threshold lines indicated in a. The logarithm of ramp rate is negatively correlated with the logarithm of waiting time. (d) Relationship between the waiting time and the ramp rate for the movement-type neuron. The ramp rate is constant and independent of the waiting time.

  5. Supplementary Figure 5: Choice of waiting action at block switches between nose-poke waiting blocks and lever-press waiting blocks. (109 KB)

    (a) An example session in which a rat performed both nose-poke and lever-press waiting task. The x-axis indicates number of trials and y-axis the waiting time in each trial. The waiting time of the nose-poke trials is indicated in a positive direction in the y-axis and a lever-press waiting time in a negative direction. The lever-press block is indicated with a light yellow background. Gray circle indicates the short poke trials, red the impatient trials, and blue the patient trials. (b) Choice between nose-poke waiting and lever-press waiting around block switches. Fraction of trials in which rats chose to perform nose-poke waiting is plotted as a function of trials (20 trials before switch and 40 trials after switch). The left panel shows the choice of waiting action at the block switch from the nose-poke block to the lever-press block. The right panel shows the choice of waiting action at the block switch from the lever-press block to the nose-poke block. All the recording sessions from 3 rats are combined. The gray shade indicates 95% confidence interval calculated with binomial fitting. The lever-press block is indicated by light yellow background. The rats quickly adapted to the block switch within a few trials.

  6. Supplementary Figure 6: Transient correlation analysis focused on a period before poke in in the current trial and during intertrial interval (ITI). (66 KB)

    Ninety-one neurons are selected with the same transient correlation analysis as in the main result, but focused on different time bins (Eight non-overlapping 1.5s time bins, including 2 bins from a 3s before poke-in in the current trial and 2 bins from the first 3s ITI period of the previous trials (1 – 3 trials back)). An ITI period of 3 trial back is shown in (a), 2 trials back in (b), 1 trial back in (c) and a pre-poke-in period of the current trial in (d). P-value shown here are calculated with 0.4s overlapping time bins with 0.02s time steps for each neuron. The color code is the same as in Figure 5c. Bonferroni correction for multiple comparisons were used for selecting neurons (P < 0.05), but P-value shown here was not corrected. Note that the waiting time predictive activity was already present at 2 – 3 seconds before the poke-in and during ITI period after the previous trial, but much weaker in 2 trials back and 3 trials back.

  7. Supplementary Figure 7: Difference in activities in the impatient and patient trials of the predictive neurons. (89 KB)

    Because Tone 2 terminated the waiting of rats in patient trials, the rats’ willingness to wait was presumably longer than the actual waiting time. This gives rise to two additional predictions that we were able to test. First, transient waiting-time predictive neurons that fired at a higher rate in long waiting impatient trials should fire even more vigorously in patient trials, while neurons which fired less in long waiting trials should fire even less in the patient trials. Second, in patient trials, the ramp-to-threshold neurons should not reach the threshold firing rate for leaving before Tone 2. (a) A scatter plot showing the relationship between the correlation between firing rates and waiting times in the impatient trials and impatient/patient selectivity index. We calculated an impatient/patient selectivity index using the time window(s) with significant predictive activity in the correlation analysis. Because we wished to compare activities in impatient and patient trials without a contribution of difference in waiting time distribution, we selected subset of impatient and patient trials to match the waiting time distributions from two trial types, as described for the analysis of the movement time. The impatient/patient selectivity index was defined as the difference in a mean firing rate in the impatient and patient trials normalized by the sum of the mean firing rates. The value ranges from –1 to 1; 1 indicates a neuron selective for the impatient trials. Each circle represents a time bin with significant correlation between firing rate and waiting time in the impatient trials. A neuron which has multiple time bins with significant correlation is represented more than once in the plot. Filled circles indicate time bins with significant impatient/patient selectivity index (Wilcoxon signed-rank test, P < 0.05). Open circles indicate time bins with non-significant selectivity index. (b) Distribution of the impatient/patient selectivity index for all time bins with significant positive correlation between firing rates and waiting times in impatient trials. Arrow indicates the median selectivity index. The distribution is significantly shifted toward negative (Wilcoxon signed-rank test, P < 0.01) (c) The same as in b but for time bins with negative correlation. The distribution is significantly shifted toward positive (Wilcoxon signed-rank test, P < 0.0001). Analysis of patient trials for the ramp-to-threshold type neurons (d,e). (d) Activity of an example ramp-to-threshold neuron (the same neuron as in Figure 3) in impatient and patient trials aligned to poke-out. A gray shaded area indicates the poke-out time window (from –500 ms to –250 ms from poke-out) used for the population analysis in e. A green dashed line indicates a range of possible Tone 2 onset time in the patient trials. Activity was higher in the impatient trials than in patient trials in the analysis window (P < 0.05, Wilcoxon signed-rank test). (e) Distribution of impatient/patient selectivity index for ramp-to-threshold neurons. Patient trials with more than 250 ms reaction time were excluded to prevent contamination of Tone 2 response in the analysis time window. Patient trials with less than 60 ms response time were also excluded, because probably those are trials in which Tone 2 presentation happened to coincide with the time the rat was about to leave, estimated from the response time distribution pooled across rats (data not shown). After exclusion of those trials, the procedure for matching the waiting time distribution of impatient and patient trials was performed, as described before. Ramp-up and ramp-down neurons were pooled together. The sign of impatient/patient selectivity index for the ramp-down neurons was flipped. A red histogram indicates neurons with significant correlation between ramp rates and waiting times. A black histogram indicates neurons without significant correlation. While a distribution of the selectivity index of neurons with significant ramp rate correlation was shifted toward positive (Wilcoxon signed-rank test, P < 0.05), that of neurons without significance was not (P > 0.5).

  8. Supplementary Figure 8: Effect of waiting times and ongoing movements on M2 neural activity. (109 KB)

    We used a stepwise regression (MATLAB, stepwisefit), because the number of independent variables is relatively large compared with the data size. The significance of each coefficient was estimated by running the same analysis after shuffling a variable of interest across trials (1000 times). We first tracked the position (X, Y) and orientation of the rat’s body, using custom-written software with Python using the OpenCV library. For the firing rate correlation analysis, the average firing rate of each 400 ms time bin was regressed using stepwise regression using for independent variables waiting time and 6 movement variables of the same time bin: X-, Y-, angular-position, X-, Y- and angular-velocity. P-values were corrected for multiple comparisons (Bonferroni correction). (a) An example neuron that showed a significant correlation between waiting time and firing rate with the stepwise regression analysis. Perievent time histograms (PETHs) around start of waiting shown in the same way as in Figure 4 (left). A scatter plot shows the relationship between X position (x-axis), waiting time (y-axis) and firing rate (color axis) (right). Note that the color gradient (i.e. firing rate gradient) is obvious in the waiting time axis but not in the X position axis. None of the other movement variables explained the firing rate (data not shown). (b) An example neuron that showed a spurious correlation between waiting time and firing rate in the original analysis, which disappears if movement variables are included in the regression analysis. PETHs around start of waiting (left). A scatter plot shows the relationship between Y position (x-axis), waiting time (y-axis) and firing rate (color axis) (right). Note that the firing rate gradient is obvious in the Y position axis but not as much in the waiting time axis. (c) Fractions of neurons with firing rate significantly correlated with waiting times, as described in the main text, applied to the neurons from video recorded sessions (n = 164 neurons) is shown in gray bar. Fractions of neurons with firing rate significantly correlated with each of the variables shown in the bottom labels using a stepwise regression analysis are shown in black and white bars. (d) Fractions of neurons with significant correlation with each variable shown in the label among waiting time predictive neurons in the original analysis. Eighty percent (20 of 25) of neurons remained significantly correlated with waiting times with the stepwise regression. (e) Fractions of neurons with rate of ramping activity significantly correlated with waiting times among ramp-to-threshold type predictive neurons from video recorded sessions, as described in the main text, are shown in gray. Fractions of neurons with ramp-rate significantly correlated with each of the variables described in the label are shown in black and white bars. Seventy-five percent (6 of 8) of ramp-rate correlated neurons remained significant with the stepwise regression. For the ramp-rate correlation analysis, we regressed the rate of ramping activity with the same variables described above. Motor variables were calculated at a single time bin from 0 to 400 ms relative to the start of waiting, because this is the period common to all the trials. We also performed the analysis using the entire waiting time as a single time bin (different bin size for different waiting time trials). This yielded similar results (data not shown).

  9. Supplementary Figure 9: Trial history analysis using a stepwise multiple linear regression. (82 KB)

    (a) A fraction of neurons with activity significantly correlated with waiting times (current impatient trials) using a standard simple linear regressions analysis (analysis described in the main text) is shown in gray (left most bar). Fraction of neurons significantly correlated with each of the trial history variables shown in the bottom labels using a stepwise multiple linear regression are shown in black (waiting time of current impatient trials) and white bars (other trial history variables). We used a stepwise regression (MATLAB, stepwisefit), because the number of independent variables is relatively large compared with the data size (number of trials). The significance of each coefficient was estimated by running the same analysis after shuffling a variable of interest across trials (1000 times). Note that using the stepwise multiple linear regression analysis including trial history variables, 15.7 % of neurons are significantly correlated with waiting times of the current trial (a black bar), comparable to 18.0% with the analysis shown in Figures 5 and 6 (a gray bar). (b) Time course and the sign of correlation for all the waiting time predictive neurons with stepwise multiple linear regressions (n = 56 neurons). The format is the same as is in Figure 5c. The time course of the predictive activity was also comparable to the original analysis. We also observed neurons that carried information about trial history, consistent with previous studies21. We did not further analyze those correlations because it was beyond the scope of our current study.

  10. Supplementary Figure 10: Anatomical location of recording sites. (129 KB)

    (a) Nissl stained coronal section of a rat frontal cortex. The arrow indicates a site of an electrolytic lesion. (b) Fluorescent image of an adjacent section. The arrow indicates a track of DiI coated tetrode. (c) Corresponding section from the rat brain atlas52. (d) Recording sites of all the neurons analyzed for the ramp-up predictive activity (correlation between time to cross a threshold firing rate and waiting time). Neurons with significant correlation between times to cross a threshold firing rate and waiting times are shown with color. Open pink circle, ramp-up neurons without significant correlation between ramp rates and waiting times. Filled pink circle, ramp-up neurons with significant correlation between ramp rates and waiting times. No obvious clustering of predictive neurons was observed. (e) The same as in d but for ramp-down neurons. Open blue circle, ramp-down neurons without significant correlation between ramp rates and waiting times. Filled blue circle, ramp-down neurons with significant correlation between ramp rates and waiting times. No obvious clustering of predictive neurons was observed. (f) Recording site of all the neurons analyzed for the predictive activity (correlation between firing rate and waiting time at a certain time window). Neurons whose activity was positively correlated with waiting time are shown in green, while neurons with negative correlations are shown in orange. No obvious clustering of predictive neurons was observed.

Video

  1. Video 1: Behavior from an example recording session. (7.01 MB, Download)
    For this rat, the nose-poke waiting port was located at the left side, lever-press waiting port at the right side and reward port at the center. Entering the nose-poke waiting port was indicated by a filled white circle on top of the waiting port, which is an open circle outside the waiting period. Presentation of Tone 1 is indicated by light green bars at both sides and Tone 2 by dark green bars. Delivery of a water reward was indicated by a blue circle appearing on top of the reward port. Although the onset of the presentation of tones and water is accurate, the duration of those are not accurately represented in the movie, because they are brief in the actual experiment and hard to see in the movie.

PDF files

  1. Supplementary Text and Figures (2,655 KB)

    Supplementary Figures 1–10

  2. Supplementary Methods Checklist (394 KB)

Additional data