Claims 52
From the paper New information triggers prospective codes to adapt for flexible navigation · claims file · checked 2026-10-21: ready
- recordings
Seven mice were recorded over 66 sessions, yielding 3892 hippocampal CA1 units and 2557 medial prefrontal units after curation and cell type classification.
we recorded single unit and local field potential activity in dorsal hippocampal CA1 and medial prefrontal cortex during behavior ( n = 3892 units in hippocampus, n = 2557 units in medial prefrontal cortex
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 1; Supplementary Table 1How it will be judged
Counts from the dandiset. Animals and sessions are design counts and must match exactly (tolerance 0, reason: they are the experiment, not a measurement). Units are counted per region over all sessions, all three cell types (Supplementary Table 1 note: final counts after curation and cell type classification); the per-animal counts of Supplementary Table 1 are reported beside them.
- trial-mix
In the final task 65% of trials are delay only, 25% switch and 10% stay trials.
These trial types consisted of delay only trials (65% of trials)
Methods, Behavioral task and training, paragraph 2; Results, first section, paragraph 1How it will be judged
Fractions of final-task trials (warm-up trials excluded, Methods; code maze_id 4) of each type, pooled over the 66 sessions. Readings for the denominator, all reported: (a) all final-task trials; (b) the per-session fractions averaged over sessions. Reproduced when all three fractions are within tolerance under one reading. Note: the paper's own counts do not give these fractions. Supplementary Table 1 lists 8104 delay, 1840 switch and 717 stay completed trials (0.76, 0.17, 0.07), and the Results counts (7172 delay only, 1840 switch) with the 717 stay trials give 0.74, 0.19, 0.07; both fall outside the tolerances, so the percentages read as the task's programmed trial probabilities (Methods). The realized fractions are reported beside the result.
- task-timing
The delay on delay only trials lasted 10.56 s on average, the shortened delay before the update cue on switch trials 1.40 s, and the delay after the update cue 6.02 s; trials lasted about 21 s from start to choice.
The mice had to maintain the memory of the correct goal arm during the delay period (10.56 ± 0.07 seconds, n = 7172 trials, Supplementary Fig. 1 ). On a subset of trials, a second visual cue appeared when mice reached a specific location after a shortened delay period (1.40 ± 0.03 seconds, n = 1840 trials)
Results, first section, paragraph 1; Supplementary Fig. 1aHow it will be judged
From the trial table of final-task trials. The delay on delay only trials runs from original cue offset to choice made (Supplementary Fig. 1a legend); trial duration from trial start to choice made (Supplementary Fig. 1a legend; code behavior_analyzer, t_choice_made minus start_time). The first delay on switch trials runs from original cue offset to update cue onset. The second delay is open; readings (a) second delay onset (t_delay2) to choice made, (b) update cue onset to choice made. The Results do not say which trials the 6.02 s second delay covers (Methods give a second delay on stay and update trials); readings for it (e) switch trials, (f) switch and stay trials pooled. The trial set is open (n = 7172 delay only and 1840 switch trials in the text, 7313 and 1886 in the legend, 8104 and 1840 completed in Supplementary Table 1); readings (c) all final-task trials, (d) correct trials. Reproduced when each value is within tolerance under one reading; the trial counts under each reading are reported.
- fig1b-above-chance
Animals performed above 50% correct on delay only trials (65 +/- 2%) and on switch trials (74 +/- 2%).
animals learned to remember and follow the cues across all trial types and performed above 50% accuracy (delay only mean ± SEM: 65 ± 2%, switch mean ± SEM: 74 ± 2%, p = 6.197e−10 delay only vs. 50% correct, p = 3.410e−11 switch vs. 50% correct
Results, first section, paragraph 1; Figure 1b; Supplementary Table 2, Figure 1bHow it will be judged
Final-task trials (code maze_id 4). The test is a two-sided one-sample Wilcoxon signed-rank test against 0.5 (Methods, Supplementary Table 2). Its sample is open; readings (a) one proportion correct per session and trial type (the code runs this test on session proportions minus 0.5, and the printed W values fit a sample of tens, not thousands); (b) the 40-trial windows that Supplementary Table 2 names. The mean is open; readings (c) over sessions, (d) over the 7 animals. Reproduced when the test gives p < 0.05 with the mean above 0.5 for both trial types and the means are within tolerance under one reading of the mean.
- fig1b-window-performance
Over rolling windows of trials, proportion correct averaged 0.68 on delay only and 0.78 on switch trials.
Box plots indicate median and quartiles of 40-trial bins. (delay only: 0.68 ± 0.00, n = 6252 windows of 40 trials; switch: 0.78 ± 0.01, n = 1002 windows)
Figure 1b legend; Supplementary Table 2, Figure 1bHow it will be judged
Proportion correct in rolling windows of consecutive trials of one type within a session (Methods "a rolling window of proportion correct"; code behavior_analyzer, pandas rolling with min_periods equal to the window), final-task trials. The window length is 40 trials in the Methods and legend, while the code default used for Figure 1 is 15 trials (trial_window); 15-trial windows give a window count close to the reported 6252 (7172 minus 66 times 14), 40-trial windows cannot. Readings (a) 40 trials, (b) 15 trials; the number of windows of each reading is reported. Reproduced when means and medians are within tolerance under one reading.
- fig1d-heading-switch
On switch trials the animals’ heading (view angle) shifts from the initial toward the new side after the update cue, while on delay only trials it stays toward the initial side.
We found that on switch trials around the update cue onset, the animals’ heading direction shifted from one direction to the opposite in response to the cue indicating the reward location had changed (Fig. 1d
Results, first section, paragraph 2; Figure 1dHow it will be judged
Average view angle trajectories by position (Methods, Behavioral analysis: the mean per position bin over trials, left and right trials separately; code behavior_analyzer, 50 position bins), across all delay only and switch trials (Figure 1d legend), separately for initial-left and initial-right trials. Reproduced when, for both sides, the mean view angle of switch trials points toward the initial side at the update cue onset position and toward the new side in the last central-arm bin before the choice point, while delay only trials point toward the initial side in that last bin. The paper reports no test.
- fig5a-stay-performance
Animals performed very well on stay trials (proportion correct 0.92 over windows).
accuracy on stay trials showing median and quartiles ( n = 7 animals; stay: 0.92 ± 0.01, n = 117 40-trial windows)
Figure 5a legend; Results, Goal codes for updating behaviors are not due simply to new visual information, paragraph 1How it will be judged
As fig1b-window-performance, for stay trials, with the same two window-length readings. Reproduced when the mean is within tolerance under one reading.
- tuning-tiles-track
Spatial tuning curves of hippocampal and prefrontal neurons tile the virtual track.
we found that neurons in both hippocampus and prefrontal cortex had spatial tuning curves that tiled the virtual environment (Supplementary Fig. 2c , Supplementary Fig. 5a )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 1; Supplementary Figs. 2c and 5aHow it will be judged
Position tuning curves of the encoder model (shared method S2, S3; 80% of delay only trials, Supplementary Fig. 2c legend), all units of included sessions, per region. The paper gives no measure. Readings (a) every one of the 50 position bins is the peak of at least one unit; (b) the distribution of peak positions over units covers the central arm and both goal arms with no run of more than 3 empty bins outside the gaps between track segments. Reproduced for a region when both readings hold.
- decoding-validation
Hippocampal and prefrontal population activity decodes the animal’s current position across the track.
We first confirmed that hippocampal and prefrontal activity could reliably decode the animals’ current location in the environment; decoded positions from neural activity represented the animals’ true position across locations in the environment (Supplementary Figs. 2d and 5b )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 1; Supplementary Figs. 2d and 5bHow it will be judged
Held-out delay only trials decoded with the shared method (S1 to S4); error is the absolute difference between true and peak decoded position (Supplementary Fig. 2d legend). The paper gives no threshold. Readings (a) in most included sessions the median error is below the median error between true positions and positions drawn uniformly over the decoded bins; (b) the session-averaged confusion matrix has its column maximum within 2 bins of the true bin for more than half of the occupied position bins. Reproduced for a region when both readings hold; if they differ the claim is blocked as methods_underspecified.
- fig2e-hpc-goal-increase
After the update cue, hippocampal decoding of both the initial and the new goal arm is higher on switch trials than on delay only trials.
the overall probability of goal locations being decoded, the posterior probability integrated over each goal arm, was elevated on switch trials compared to delay only trials (p = 2.630e-8, initial, switch vs. initial, delay only, p = 4.157e−7, new, switch vs. new, delay only
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 2; Figure 2c to e; Supplementary Table 2, Figure 2eHow it will be judged
CA1 position decoding by the shared method (S1 to S9), correct switch and delay only trials, trial value the mean integrated goal arm probability over 0 to 1.5 s. LME on trial values with goal arm by trial type as predictor, ANOVA, Tukey contrasts. Which trial types enter the model is open; readings (a) switch and delay only only (code plot_goal_coding_stats for Figure 2; the reported denominator df 4400.90 fits about 2 x 2221 rows); (b) switch, delay only and stay together (Supplementary Table 2 note "linear mixed effects models included delay, update, and stay trials"). Reproduced when, under every reading, the ANOVA and both switch against delay only contrasts give p < 0.05 with switch higher; trial counts are reported against 1295 and 926.
- fig2e-hpc-arms-equal
Within trials, hippocampal decoding of the initial and the new arm after the update cue does not differ: the initial minus new difference is not different from zero and does not differ between switch and delay only trials.
when we compared differences between the amount of initial goal location and new goal location representations on a trial-by-trial basis, the differences were not significantly different from zero and were not significantly different between switch and delay only trials ( p = 0.7806 switch vs. delay only
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 3; Figure 2e right; Supplementary Table 2, Figure 2eHow it will be judged
A claim of absence, judged on the paper's tests: an LME (S9) on the per-trial initial minus new difference (0 to 1.5 s) with trial type as predictor, ANOVA at p < 0.05; and two-sided one-sample Wilcoxon signed-rank tests of per-session means of the difference against zero, per trial type (Supplementary Table 2, n = 56 sessions). Reproduced when none of the three tests gives p < 0.05; not reproduced when any does. Effect sizes are reported either way.
- sfig2e-goal-bias
Within single 200 ms bins hippocampal goal decoding is strongly biased to one arm (43% of bins highly biased), yet within a trial both arms are represented (70% of trials have a mean goal bias index between -0.5 and 0.5).
each bin strongly represented one arm or the other with 43% of bins highly biased for one arm or the other (having a goal bias index close to 1 or −1)
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 3; Supplementary Fig. 2eHow it will be judged
Goal bias index per bin = (new minus initial) / (new plus initial) integrated arm probability (Methods); highly biased when above 0.93 or below -0.93 (Methods); per-trial index the mean of its bin indices (code plot_goal_coding_indtrials, ave_bins); correct switch trials, CA1, n = 1295 (Supplementary Fig. 2e legend). The window is open; readings (a) 0 to 1.5 s after the update cue (Methods quantification window), (b) 0 to 2.5 s, all aligned bins at or after the cue (code). Bins on the 0.2 s alignment grid. Reproduced when both fractions are within tolerance under one reading.
- sfig3c-goals-over-retrospective
On switch trials after the update cue, both goal arms are decoded more than retrospective non-local positions at the start of the track.
We found that both initial and new goal arms were represented significantly more than retrospective non-local locations after the update cue on switch trials (retrospective vs. initial: p = 0.04, retrospective vs. new: p = 0.0011, Supplementary Fig. 3c )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 4; Supplementary Fig. 3cHow it will be judged
As fig2e-hpc-goal-increase with a third zone, the retrospective zone, the first 0.15 of the track (Results); the code integrates linearized positions 5 to 50 (quantify_aligned_data with prospective_reps). LME with zone as predictor on correct switch trials, Tukey contrasts. Reproduced when both goal arms exceed the retrospective zone with p < 0.05.
- sfig3c-retrospective-unchanged
Retrospective non-local decoding after the update cue does not differ between switch and delay only trials.
Comparing switch and delay only trials, we found that non-local retrospective codes did not differ significantly (retrospective switch vs delay only: p = 0.0513, Supplementary Fig. 3c )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 4; Supplementary Fig. 3c iiHow it will be judged
A claim of absence, judged on the paper's test: the LME of fig2e-hpc-goal-increase with zone (retrospective, initial, new) by trial type, Tukey contrast of the retrospective zone between switch and delay only. Reproduced when p >= 0.05, not reproduced when p < 0.05; the effect is reported either way.
- sfig3ab-original-vs-update-cue
Goal arm decoding also rises during the original cue, but then favours the back half of the goal arms, while after the update cue it favours the front half.
during the original cue we found higher decoding of the back half of the goal arm (initial: p = 4.842e−9, new: p = 3.919e−14), while during the update cue decoding of the front half of the goal arms was higher (initial: p = 0.0100, new: p = 1.020e−7
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 6; Supplementary Fig. 3a, bHow it will be judged
Correct switch trials, CA1. Decoding aligned to the original cue (trial start, code time_label start_time with no time before it) and to the update cue, each over 0 to 1.5 s. Each goal arm split into front (toward the choice point) and back halves by its bins (code quantify_aligned_data, half front and rearhalf; the Figure code passes front for both panels, so it does not settle the back half, and the split by bins is used). LME with cue by arm half, Tukey contrasts between cues. Reproduced when the four contrasts give p < 0.05 in the stated directions.
- sfig2f-all-trials-encoder
With all trial types in the encoding model, the increase of hippocampal goal arm decoding after the update cue on switch trials remains similar.
we also ran the analysis using all trials in the encoding model and found that while there was more accurate representation of the animals’ current location during the update cue, the non-local coding results remained similar (Supplementary Fig. 2f )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 5; Supplementary Fig. 2fHow it will be judged
The test of fig2e-hpc-goal-increase repeated with encoders built from 80% of all trial types (delay only, switch, stay; Methods "we used all trial types ... to build our encoding model as a control"). Which trials are decoded is open. Supplementary Fig. 2f lists 1322 switch and 4538 delay only trials; 4538 is close to the 926 held-out correct delay only trials of Figure 2e divided by 0.2 (about 4630), so the decoded trials do not look restricted to a 20% test set. Readings (a) encoder built from all trials of all types and every trial decoded with it; (b) encoder from 80% of all trial types, the held-out 20% decoded. Readings for outcome (c) correct trials as in Figure 2e (1322 is near the 1295 correct switch trials), (d) correct and incorrect. Trial counts under each reading are reported against 1322 and 4538. Reproduced when both arms remain higher on switch than on delay only trials with Tukey p < 0.05 under every reading.
- sfig5f-pfc-wider-spread
Decoded position distributions are spatially wider in prefrontal cortex than in CA1.
Indeed, we found the spatial spread of decoded positions was significantly wider in PFC than in CA1 (CA1 vs PFC: p = 0.0001, Supplementary Fig. 5f )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 7; Supplementary Fig. 5fHow it will be judged
Sessions with at least 20 cells in each region and 50 trials; the region with more cells randomly subsampled to the same number; delay only trials, correct and incorrect (Methods; Supplementary Fig. 5f legend). Spread per decoding bin is the distance in front of and behind the posterior peak at which the posterior drops below the peak minus one standard deviation of that posterior (Methods); the code takes it in feature bins, using bins where the animal is past 120.35 (after the original cue), and averages per trial (plot_region_error_distribution). The unit is open; readings (a) feature bins (code), (b) linearized track units. Expected values are compared under reading (a). Region compared by LME with animal and session random effects (Methods; the Methods also list region and trial as random effects, which cannot be read as written, so the S9 model is used). Reproduced when PFC spread exceeds CA1 with p < 0.05 and the values are within tolerance.
- sfig5g-ca1-more-goal-coding
Before the update cue, goal arms are decoded more strongly from CA1 than from prefrontal cortex.
Goal locations were more strongly represented in CA1 than in the prefrontal cortex before any changes due to the update cue (CA1 vs PFC initial: p = 4.641e−14; new: p < 4.462e−61, Supplementary Fig. 5g )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 7; Supplementary Fig. 5gHow it will be judged
Correct delay only trials, sessions with both regions meeting S1, cells subsampled to equal numbers (Methods; legend), at positions before the update cue location; the code uses -2 to 0 s around the pseudo update time (time_window (-2, 0)). The CA1 initial arm value is printed 0.010, beside 0.11 for the new arm, and is not used as an expected value. LME with region by arm, Tukey contrasts. Reproduced when CA1 exceeds PFC for both arms with p < 0.05.
- fig2g-pfc-no-goal-increase
Prefrontal position decoding of the goal arms after the update cue does not differ between switch and delay only trials, and the two arms are represented equally.
Non-local representations in medial prefrontal cortex were not significantly different across trial types and both goal locations were equally represented until the animals approached the goal arm ( p = 0.3496, initial, switch vs. initial, delay only, p = 0.3148, new, switch vs. new, delay only, p = 0.9111, switch vs. delay only, Fig. 2g
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 7; Figure 2f, g; Supplementary Table 2, Figure 2gHow it will be judged
A claim of absence, judged on the paper's tests, applied to PFC position decoding as in fig2e-hpc-goal-increase and fig2e-hpc-arms-equal (shared method, correct trials, 0 to 1.5 s). Reproduced when the ANOVA on arm by trial type, the two Tukey contrasts, the ANOVA on the within-trial difference and the two per-session Wilcoxon tests all give p >= 0.05; not reproduced when any does not. Effects are reported either way.
- sfig5e-pfc-windows
Prefrontal goal arm decoding on switch and delay only trials does not differ in shorter windows after the update cue either.
There was a small increase in prefrontal decoding of goal locations on switch trials versus delay only trials, however these trial types did not differ significantly over several time windows examined (0–0.5 s, 0–1 s, 0–1.5 s after the update cue onset, Supplementary Fig. 5e )
Results, Enhanced non-local codes of both goal locations in hippocampus, paragraph 7; Supplementary Fig. 5eHow it will be judged
A claim of absence, judged on the paper's test (as fig2g-pfc-no-goal-increase) in each window the paper names; the text lists 0 to 0.5, 0 to 1 and 0 to 1.5 s and the legend 0 to 0.5, 0 to 1 and 0.5 to 1 s, so all four windows are run. Reproduced when no switch against delay only contrast gives p < 0.05 in any window; not reproduced otherwise.
- fig3c-theta-phase-segregation
After the update cue on switch trials, hippocampal non-local goal codes are strongest in the first half of the theta cycle and local codes in the second half.
we found that codes for prospective goal locations (non-local) were represented significantly more on the opposite theta phase as current location representations (local) (p = 0.0275, −π to 0, initial vs. central arm, p = 0.0106, −π to 0 new vs. central arm, p = 9.284e−9, 0 to π, initial vs. central arm, p = 6.672e−14, 0 to π, new vs. central arm, Fig. 3a–c )
Results, Theta phase segregation of non-local and local position codes, paragraph 2; Figure 3b, c; Supplementary Table 2, Figure 3cHow it will be judged
CA1 decoding in 25 ms bins (Methods; code decoder_bin_size 0.025), correct switch trials. Theta from the CA1 pyramidal-layer channel, 4 to 12 Hz FIR equiripple band-pass, phase from peaks and troughs, 0 set at the phase of maximal CA1 pyramidal cell firing in 30 degree bins per session (Methods; code theta_phase/get_phase_reference.py). The phase at each decoding bin centre; integrated probability per zone averaged per trial in each half cycle (-pi to 0, 0 to pi) over 0 to 1.5 s after the cue (Methods; Figure 3b legend; code calc_theta_phase_data). The central zone is open; readings (a) the whole central arm, (b) a central window the size of a goal arm centred on the update cue location (code home_boundaries). LME with location by half, Tukey contrasts. Reproduced when, under every reading, both goal arms exceed central in the first half and central exceeds both in the second half, all four with p < 0.05.
- fig3b-theta-pre-post
At the theta timescale, goal arm decoding increases and central decoding decreases from before to after the update cue on switch trials.
When comparing before and after the update cue onset, we found that there was a significant increase in the amount of goal location (non-local) decoding, but the local vs. non-local phase preferences were maintained ( p = 6.439e−14, pre- vs. post-update initial arm, p = 5.329e-14 pre- vs. post-update new goal, Fig. 3b–d )
Results, Theta phase segregation of non-local and local position codes, paragraph 2; Figure 3b; Supplementary Table 2, Figure 3bHow it will be judged
As fig3c-theta-phase-segregation with pre (-1.5 to 0 s) and post (0 to 1.5 s) windows (Figure 3b legend). Which phases enter the values is open; readings (a) the first half cycle (Supplementary Table 2 labels the Figure 3b data "first half of theta cycle"), (b) the whole cycle. LME with time by location, Tukey contrasts. Reproduced when, under every reading, both goal arms increase and central decreases with p < 0.05.
- sfig6-theta-switch-over-delay
At the theta timescale, goal arm decoding after the update cue is larger on switch than on delay only trials.
non-local codes at the timescale of individual theta cycles were also significantly larger on switch trials compared delay-only trials ( p = 1.450e−27, initial arm, switch vs. delay only, p = 1.367e−13 new arm, switch vs. delay only Supplementary Fig. 6 )
Results, Theta phase segregation of non-local and local position codes, paragraph 2; Supplementary Fig. 6; Supplementary Table 2, Figure 3c, dHow it will be judged
As fig3c-theta-phase-segregation, one LME per goal arm with trial type (switch, stay, delay only, as Supplementary Table 2 lists) by phase half, correct trials, 0 to 1.5 s. Reproduced when, for both arms, the ANOVA and the switch against delay only contrasts in both halves give p < 0.05 with switch higher.
- fig3d-new-arm-theta-modulated
After the update cue, decoding of the new goal arm differs between theta half cycles while decoding of the initial arm does not.
The new goal location was more strongly theta modulated compared to the old goal location ( p = 0.0008, new arm − π to 0 vs 0 to π , p = 0.6496, initial arm − π to 0 vs 0 to π
Results, Theta phase segregation of non-local and local position codes, paragraph 2; Figure 3d; Supplementary Table 2, Figure 3dHow it will be judged
As fig3c-theta-phase-segregation, one LME per arm with phase half as predictor (Supplementary Table 2), correct switch trials, 0 to 1.5 s. Reproduced when the new arm gives p < 0.05 and the initial arm, a claim of absence judged on the same test, gives p >= 0.05. The text and the table print different initial arm p values (0.6496, 0.4203); both are above 0.05.
- fig4a-lstm-choice-estimate
An LSTM predicting the final choice from behaviour is near chance at the start of the trial and highly accurate by the end, with session-averaged log likelihoods of -0.98, -0.63, -0.43 and -0.17 in the original cue, delay, update and second delay phases.
The choice estimate varied throughout the course of a trial, starting at chance-level accuracy at the start of the trial on average, and ending with a highly accurate, or −0.17 ± 0.02 log likelihood on average, prediction of the animals’ final choice
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 1; Figure 4aHow it will be judged
LSTM as shared method S10, one model set per session, 66 sessions. Log likelihood at each time point is y log2(p) + (1 - y) log2(1 - p) of the final choice y (Methods; code _log2_likelihood). Phases by forward position as fractions of the track, with boundaries at the original, delay, update and second delay cue start locations 5, 120.325, 145.125 and 215.125 of a track from 5 to 285 (code choice_visualizer plot_choice_commitment), on held-out delay only trials; mean over time points per phase and session, then over sessions. Reproduced when the four session means are within tolerance; partially reproduced when the order original < delay < update < second delay holds but some value falls outside. The LSTM is stochastic; one run with the code's seeds is reported.
- fig4f-pfc-choice-flips
After the update cue, prefrontal decoding of the initial choice is lower and of the new choice higher on switch than on delay only trials.
We observed that the choice estimates of the initial and new choice began to diverge preceding the update and then rapidly flipped after the reward location changed ( p = 5.192e−49 initial switch vs. initial, delay only, p = 2.934e−7 new, switch vs. new, delay only Fig. 4b–f
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 2; Figure 4b to f; Supplementary Table 2, Figure 4fHow it will be judged
PFC choice decoding by shared method S1, S3 to S9 and S11, correct trials, 0 to 1.5 s. LME with choice by trial type (readings for the trial types in the model as in fig2e-hpc-goal-increase), Tukey contrasts. Reproduced when, under every reading, initial choice is lower and new choice higher on switch trials, each with p < 0.05.
- fig4f-pfc-difference
The within-trial initial minus new choice difference in prefrontal cortex is negative on switch trials and positive on delay only trials, and differs between them.
Furthermore, the difference between initial and new choice decoding was significantly different between switch and delay trial types (p = 8.610e-33 initial – new for switch vs. delay only Fig. 4f )
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 2; Figure 4f rightHow it will be judged
As fig4f-pfc-choice-flips, on the per-trial difference, LME with trial type, ANOVA. Reproduced when p < 0.05, the switch mean is below zero, the delay only mean above zero, and both are within tolerance.
- fig4d-pfc-precedes-movement
On average, the switch of prefrontal choice decoding from the initial to the new choice precedes the change in rotational velocity.
On average, the change in neural activity choice decoding preceded the movement change (Fig. 4d )
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 2; Figure 4dHow it will be judged
Correct switch trials, PFC choice decoding aligned to the update cue on the 0.2 s grid (S5), rotational velocity mirrored like the decoding. Two times defined in the Figure 4d legend; the neural time, the first time after the cue at which the trial-averaged new choice decoding exceeds the initial; the motor time, the first time after the cue at which the slope of the trial-averaged rotational velocity changes to negative. Reproduced when the neural time is earlier than the motor time; not reproduced when it is later; equal times are reported as not decided at this grid. The paper reports no test.
- sfig7d-pfc-before-motor-single-trials
Aligned to each trial's motor response, prefrontal initial choice decoding has already decreased and new choice decoding increased before the response, and new overtakes initial at about the time of the response.
Examining when animals began changing head direction (rotational velocity began decelerating away from the initial choice) on a single trial basis, we found initial choice codes decreased and new choice codes increased before the motor response (Supplementary Fig. 7d )
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 2; Supplementary Fig. 7dHow it will be judged
Correct switch trials. The motor response is the first time after the update cue at which the derivative of rotational velocity switches from positive to negative (Supplementary Fig. 7d legend). Variant (code plot_motor_timing), the first bin after the cue at which the mirrored rotational velocity is negative. Decoding re-aligned to each trial's motor response and averaged. Reproduced when the mean initial decoding in the bin before the motor response is below its value at update cue onset and the mean new decoding above it, and the mean new decoding first exceeds the initial within one 0.2 s bin of the motor response. The paper reports no test.
- sfig7e-pfc-pre-update
On correct trials, prefrontal initial choice decoding 1.5 to 0.5 s before the update cue is lower on switch than on delay only trials, a difference attributed to selecting correct trials.
We also found differences in pre-update cue decoding of the initial choices in prefrontal cortex between switch and delay only trials (p = 0.0031 switch vs. delay-only 1.5–0.5 s preceding update cue, Supplementary Fig. 7e )
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 2; Supplementary Fig. 7eHow it will be judged
As fig4f-pfc-choice-flips over -1.5 to -0.5 s (Results; code time_window (-1.5, -0.5)), correct trials, and again with all outcomes. Reproduced when the correct-trial initial contrast gives p < 0.05 with switch lower, and the all-trial initial contrast gives p < 0.05 with switch higher.
- fig4h-hpc-choice-switches
In hippocampus, initial choice decoding is lower and new choice decoding higher after the update cue on switch than on delay only trials.
We found that initial choice representations in hippocampus were higher than the alternative choice preceding the update, and then decreased after the update cue was presented and the new choice representation increased (Fig. 4g, h
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 4; Figure 4g, h; Supplementary Table 2, Figure 4hHow it will be judged
CA1 choice decoding as fig4f-pfc-choice-flips, over 0 to 1.5 s after the cue (Methods quantification window). Variant (code, plot_goal_coding_stats with time_window (0, 2) for this panel), 0 to 2 s, shown but not deciding, since the Methods come first where they disagree with the code. Reproduced when initial is lower and new higher on switch trials with p < 0.05.
- fig4h-hpc-choices-equal
After the update cue on switch trials, hippocampus represents the initial and new choices equally; the within-trial difference is not different from zero.
both choices were represented equally and the difference on a trial-by-trial basis was not significantly different from zero ( p = 0.5625, initial – new vs. zero, Fig. 4h )
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 4; Figure 4h rightHow it will be judged
A claim of absence, judged on the paper's test, a two-sided one-sample Wilcoxon signed-rank test of the initial minus new difference against zero (Methods). The sample is not fixed by the paper; readings (a) per-trial differences (the legend gives n = 1205 trials), (b) per-session means (as Supplementary Table 2 does for Figures 2e and 2g); window 0 to 1.5 s as fig4h-hpc-choice-switches. Reproduced when no reading gives p < 0.05; not reproduced when all do; otherwise blocked as methods_underspecified with every result shown.
- fig4g-hpc-slower-than-pfc
The switch from initial to new choice decoding evolves more slowly in hippocampus than in prefrontal cortex.
However, while the decoded activity followed a similar pattern of switching from the initial to the new choice as in prefrontal cortex, this change evolved over a longer period of time in hippocampus.
Results, Rapid switch from old to new choice estimates in prefrontal cortex, paragraph 4; Figure 4d, gHow it will be judged
Correct switch trials, trial-averaged choice decoding per region on the 0.2 s grid. The crossing time is the first time after the update cue at which mean new choice decoding exceeds mean initial (Figure 4d legend arrow). Reproduced when the CA1 crossing is later than the PFC crossing, or CA1 does not cross within 2.5 s while PFC does. The paper reports no test.
- fig5d-stay-hpc-like-delay
Hippocampal goal arm decoding after the second cue on stay trials does not differ from delay only trials.
we found that the amount of non-local representations on stay trials was not significantly different from delay-only trials (p = 0.9927, initial, stay vs. delay only; p = 0.6267, new, stay vs. delay only)
Results, Goal codes for updating behaviors are not due simply to new visual information, paragraph 2; Figure 5c, d; Supplementary Table 2, Figure 5dHow it will be judged
A claim of absence, judged on the paper's test, the LME of fig2e-hpc-goal-increase with three trial types (switch, stay, delay only), correct trials, 0 to 1.5 s, Tukey contrasts. Reproduced when both stay against delay only contrasts give p >= 0.05; not reproduced when either gives p < 0.05.
- fig5d-stay-hpc-below-switch
Hippocampal goal arm decoding after the second cue is lower on stay trials than on switch trials.
Non-local codes were in fact significantly lower on stay trials compared to switch trials ( p = 2.358e-8 initial, stay vs. switch, p = 0.0274 new, stay vs. switch, Fig. 5c, d )
Results, Goal codes for updating behaviors are not due simply to new visual information, paragraph 2; Figure 5c, dHow it will be judged
As fig5d-stay-hpc-like-delay. Reproduced when both stay against switch contrasts give p < 0.05 with stay lower.
- sfig10-stay-transient-new-arm
In the first 0.5 s after the stay cue, hippocampal decoding of the new arm is transiently higher on stay than on delay only trials, with no difference for the initial arm.
Examining that period (0–0.5 s after the update cue), we found significantly more decoding of the new arm on stay trials than on delay only trials (initial, stay vs delay only: p = 0.9806; new, stay vs delay only: p = 0.0375, Supplementary Fig. 10a, b )
Results, Goal codes for updating behaviors are not due simply to new visual information, paragraph 2; Supplementary Fig. 10How it will be judged
As fig5d-stay-hpc-like-delay over 0 to 0.5 s after the cue. Reproduced when the new arm contrast gives p < 0.05 with stay higher; the initial arm part is a claim of absence judged on the same test (p >= 0.05). The text and the table print different p values; both are below 0.05 for the new arm.
- fig5f-stay-pfc-new-choice
Prefrontal new choice decoding after the second cue on stay trials is similar to delay only trials and lower than on switch trials.
We found that new choice coding on stay trials was also at similar levels to delay only trials, and significantly lower than switch trials ( p = 0.7914, stay vs. delay only, p = 0.0072, stay vs. switch, Fig. 5e, f )
Results, Goal codes for updating behaviors are not due simply to new visual information, paragraph 3; Figure 5e, f; Supplementary Table 2, Figure 5fHow it will be judged
PFC choice decoding as fig4f-pfc-choice-flips with three trial types, correct trials, 0 to 1.5 s. The stay against delay only part is a claim of absence judged on its Tukey contrast (p >= 0.05); the stay against switch part holds when its contrast gives p < 0.05 with stay lower. Reproduced when both hold.
- fig5f-stay-pfc-initial-choice
Prefrontal initial choice decoding after the second cue on stay trials stays higher than on switch trials but is lower than on delay only trials.
The decoded initial choice estimate remained relatively constant on stay trials, in contrast to switch trials, though also significantly lower than on delay trials ( p = 0.0032, stay vs. delay only, p = 1.908e-10 stay vs. switch)
Results, Goal codes for updating behaviors are not due simply to new visual information, paragraph 3; Figure 5fHow it will be judged
As fig5f-stay-pfc-new-choice. Reproduced when stay is lower than delay only and higher than switch, both with p < 0.05.
- fig5f-stay-pfc-difference
The prefrontal within-trial initial minus new choice difference on stay trials differs from both switch and delay only trials.
Furthermore, the difference between initial and new choice decoding was significantly different between trial types ( p < 4.462e−61 initial-new for switch vs. stay, p = 0.0033 initial-new for stay vs. delay only)
Results, Goal codes for updating behaviors are not due simply to new visual information, paragraph 3; Figure 5f rightHow it will be judged
As fig5f-stay-pfc-new-choice on the per-trial difference. Reproduced when both contrasts give p < 0.05, with stay between delay only (0.08) and switch (-0.02).
- fig6b-hpc-new-lower-incorrect
On incorrect switch trials, hippocampal new goal arm decoding after the update cue is less elevated than on correct trials, after accounting for speed and rotation.
We found that indeed on incorrect trials in hippocampus, the new goal choice coding was not as elevated as correct trials (p = 0.0025, new correct vs. incorrect, Fig. 6a, b )
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 1; Figure 6a, b; Supplementary Table 2, Figure 6bHow it will be judged
CA1 position decoding, switch trials of both outcomes, 0 to 1.5 s. Residual goal codes (Methods, GLM analysis): a GLM with rotational and translational velocity as predictors and the integrated goal probability as response, one model per goal representation, 200 ms bins over 2.5 s before and after the cue, Pearson residuals; the code uses a Poisson family and statsmodels GLM (aggregator get_residuals). The GLM is fit on the bins of all trial types and animals together (Methods "across all trial types and animals"); the code fits it per session on the switch trials of the comparison (by_session=True), which is shown as a variant and does not decide, since the Methods come first where they disagree with the code. Trial value the mean residual in the window; LME with outcome by arm, Tukey contrasts. Reproduced when the new arm contrast gives p < 0.05 with correct higher.
- fig6b-hpc-initial-no-difference
Hippocampal initial goal arm decoding after the update cue does not differ between correct and incorrect switch trials.
There was also no significant difference in initial goal choice coding on correct versus incorrect trials after the update cue (p = 0.2967 initial correct vs. incorrect, Fig. 6b )
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 1; Figure 6bHow it will be judged
A claim of absence, judged on the paper's test, the Tukey contrast of fig6b-hpc-new-lower-incorrect for the initial arm, with its Methods GLM (the per-session code GLM shown as a variant). Reproduced when p >= 0.05; not reproduced when p < 0.05.
- fig6-hpc-pre-cue-initial-higher-incorrect
Before the update cue, hippocampal initial goal arm decoding is higher on incorrect than on correct switch trials, beyond motor differences.
Interestingly, preceding the update cue there was a noticeable increase in the initial goal arm representation on incorrect trials that could not be explained by motor differences alone ( p = 0.002, initial correct vs. incorrect)
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 1; Supplementary Table 2, Figure 6b pre updateHow it will be judged
As fig6b-hpc-new-lower-incorrect over -1.5 to 0 s (Methods baseline window). Reproduced when the initial arm contrast gives p < 0.05 with incorrect higher (Methods GLM; code GLM as a variant).
- fig6d-pfc-new-lower-incorrect
On incorrect switch trials prefrontal new choice decoding is smaller after the update cue and does not overtake the initial choice.
This lack of switching choice representations was driven by significantly smaller decoding of the new choice in prefrontal cortex on incorrect trials ( p = 0.002, new correct vs. incorrect)
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 2; Figure 6c, d; Supplementary Table 2, Figure 6dHow it will be judged
PFC choice decoding with the residuals of fig6b-hpc-new-lower-incorrect (Methods GLM; code GLM as a variant), switch trials of both outcomes, 0 to 1.5 s. Reproduced when the new choice contrast gives p < 0.05 with correct higher.
- fig6d-pfc-initial-no-difference
Prefrontal initial choice decoding does not differ between correct and incorrect switch trials after the update cue, nor, after accounting for motor differences, before it.
In contrast, there was no significant difference in initial choice coding on correct and incorrect trials when accounting for motor differences ( p = 0.5874, initial correct vs. incorrect, Fig. 6d )
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 2; Figure 6dHow it will be judged
A claim of absence, judged on the paper's tests, the Tukey contrasts of fig6d-pfc-new-lower-incorrect for the initial choice over 0 to 1.5 s and over -1.5 to 0 s, Methods GLM (code GLM as a variant). Reproduced when both give p >= 0.05; not reproduced when either gives p < 0.05.
- fig6d-pfc-within-trial
Within trials, prefrontal new choice decoding exceeds initial on correct switch trials but not on incorrect ones.
on incorrect trials the old and new choices were decoded at similar amounts even after new information was presented, in contrast to the significantly stronger new choice coding on correct trials (p = 0.0001, initial-new for correct vs. incorrect, p = 0.3531 initial-new vs. zero, incorrect trials, p = 0.0113 initial-new vs. zero, correct trials)
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 2; Figure 6d rightHow it will be judged
As fig6d-pfc-new-lower-incorrect on the per-trial difference. One-sample two-sided Wilcoxon signed-rank tests against zero; their sample is open, readings per-trial values and per-session means. Reproduced when the correct against incorrect test gives p < 0.05 and the correct-trial difference is below zero with p < 0.05 while the incorrect-trial difference gives p >= 0.05, under every reading.
- sfig12-delay-hpc-no-outcome-difference
On delay only trials, hippocampal goal arm decoding in the update period does not differ between correct and incorrect trials.
On delay only trials we found no differences in hippocampal decoding of position for either initial or new goal arms ( p -values ≥ 0.942)
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 3; Supplementary Fig. 12aHow it will be judged
A claim of absence, judged on the paper's tests, the residual analysis of fig6b-hpc-new-lower-incorrect applied to delay only trials at the pseudo update time (S5; Results "the same period as we examined for switch trials"), Methods GLM (code GLM as a variant). Reproduced when all three p >= 0.05; not reproduced when any p < 0.05.
- sfig12-delay-pfc-outcome-difference
On correct delay only trials prefrontal cortex decodes the initial (rewarded) choice more strongly, and on incorrect trials the new (unrewarded) choice.
On correct trials the initial goal arm (the one that will be rewarded on delay only trials) was decoded more strongly than the other goal arm throughout the delay period ( p = 5.934e−11, initial correct vs. incorrect, Supplementary Fig. 12b )
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 3; Supplementary Fig. 12bHow it will be judged
As sfig12-delay-hpc-no-outcome-difference for PFC choice decoding. Reproduced when the initial contrast gives p < 0.05 with correct higher and the new contrast p < 0.05 with incorrect higher (Methods GLM; code GLM as a variant).
- fig6e-predict-outcome
The post-update difference between initial and new goal codes predicts whether a switch trial is correct, above shuffled labels, with 61.43% accuracy from prefrontal choice decoding and 55.83% from hippocampal position decoding.
We found that in both hippocampus and prefrontal cortex, the population representation of the initial and new goals after the update cue was presented predicted whether the animals would choose the correct or incorrect side (hippocampus: p < 4.462e−61, prefrontal cortex: p < 4.462e−61, permutation test, Fig. 6e )
Results, Goal codes predict ability to accurately update decisions in response to new information, paragraph 3; Figure 6eHow it will be judged
Per region, switch trials of both outcomes; one feature, the initial minus new integrated probability averaged over 0 to 1.5 s after the cue (Methods; code calc_prediction_data on raw values, not residuals). Classes balanced by resampling the smaller class with replacement to the size of the larger (code random_state 123). RBF SVM; C and gamma by randomized search, 10 iterations, leave-one-out (Methods); the code draws C from loguniform(1, 1000) and gamma from loguniform(1e-4, 1). Accuracy by leave-one-out. Shuffled control, the same procedure on shuffled labels 1000 times (Methods; code iterations=1000). Reproduced when, for both regions, accuracy exceeds the 95th percentile of shuffled accuracies and lies within tolerance; partially reproduced when only the permutation test holds.
- fig7c-hpc-commitment-new-arm
The more committed the animal was to the initial arm at the update cue (view angle toward it), the larger the post-cue increase in hippocampal decoding of the new arm.
the view angle preceding new information was positively correlated with the increase in non-local goal representation of the new arm (new goal codes vs. view angle: p = 1.330e-5, r s = 0.12370, Fig. 7b, c
Results, Increased goal representations are correlated with choice commitment, paragraph 1; Figure 7b, c; Supplementary Table 2, Figure 7cHow it will be judged
Correct switch trials, CA1 (code plot_group_comparisons_full). View angle at update cue onset (Methods), taken at the last aligned sample before time 0 (code calc_choice_commitment_data), signed so that positive points toward the initial arm (Results; the code negates it on trials with a right original cue). Change in decoding per trial is the mean over 0 to 1.5 s minus the mean over -1.5 to 0 s (Methods; code normalize=True). Two-sided Spearman correlation across trials. Quartile means of Figure 7c are reported beside it. Reproduced when rho > 0 with p < 0.05 and rho is within tolerance; partially reproduced when only the sign and p hold.
- fig7c-hpc-commitment-initial-arm
The change in hippocampal initial arm decoding after the update cue is not correlated with view angle at the cue.
It is important to note that the correlation between view angle and prospective codes was only observed for the new arm and not the initial arm
Results, Increased goal representations are correlated with choice commitment, paragraph 1; Figure 7cHow it will be judged
A claim of absence, judged on the paper's test, the Spearman correlation of fig7c-hpc-commitment-new-arm for the initial arm. Reproduced when p >= 0.05; not reproduced when p < 0.05.
- fig7e-pfc-commitment-initial
The more committed the animal was to the initial side at the update cue, the larger the decrease in prefrontal initial choice decoding.
In prefrontal cortex, when the animals were more committed to the initial side, the update cue induced a larger decrease in the initial choice representation compared to trials where animals were less committed (initial goal codes vs. view angle: p = 0.0002, r s = −0.1108
Results, Increased goal representations are correlated with choice commitment, paragraph 1; Figure 7d, e; Supplementary Table 2, Figure 7eHow it will be judged
As fig7c-hpc-commitment-new-arm for PFC choice decoding. Reproduced when rho < 0 with p < 0.05 and rho within tolerance; partially reproduced when only the sign and p hold.
- fig7e-pfc-commitment-new
The post-cue increase in prefrontal new choice decoding is weakly correlated with view angle at the cue, smallest when the animal was already heading to the new arm.
After the update cue, prefrontal decoding of the new arm increased on all trial types, but the increase was the smallest when animals were already more committed to the new arm.
Results, Increased goal representations are correlated with choice commitment, paragraph 1; Figure 7e; Supplementary Table 2, Figure 7eHow it will be judged
As fig7e-pfc-commitment-initial for the new choice. The sign of rho differs between the text (positive, which fits the quartile means of the Figure 7e legend and the sentence quoted) and Supplementary Table 2 (negative). Reproduced when rho > 0 with p < 0.05; a result with rho < 0 and p < 0.05 matches only the table and is partially reproduced; not reproduced when p >= 0.05.
Pipeline
- Paper
- done (2 of 2)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none