Claims 33
From the paper 10.1016/j.neuron.2021.07.029 · claims file · checked 2026-10-22: ready
- fig1d-sessions-blocks
The analysis includes 24, 32, 23 and 33 behavioural sessions and 89, 85, 105 and 166 trial blocks for the four subjects.
n = 24, 32, 23, and 33 sessions per subject, respectively ... n = 89, 85, 105, and 166 trial blocks per subject, respectively
Figure 1D and 1E-F legendsHow it will be judged
Sessions are the full-task sessions of each subject, less sessions 1 and 3 of subject 1 (Methods, Spatial decoding). Trial blocks are counted as in dfs_ripcontent.m part 2, blocks with at least four rewards at the goal (a trial of task phase >= 4, the code's any(ctphase>=4)). Reproduced when all eight counts are within tolerance; partially reproduced when the session counts are and the block counts are not.
- fig1ef-reward-rate
Aligned to the goal change, the outer-arm reward rate starts at chance (0.125) and rises; aligned to the first reward at the new goal, it is well above chance, so all subjects keep visiting a goal once found.
Aligned to the first search trial of each trial block, the reward rate at the outer arms for all subjects began at chance (0.125) and rose (Figure 1E) ... the outer reward rate was well above chance, indicating that all subjects preferentially visited the goal location consistently after finding it (Figure 1F)
Results, Rats learn to perform a flexible spatial memory task, paragraph 5; Figure 1E and 1FHow it will be judged
As dfs_ripcontent.m part 2, per subject over trial blocks with at least four rewards, the final unrewarded trial of the block dropped; a trial scores 1 when rewarded; binomial S.E.M. sqrt(p(1-p)/n), trials 1 to 15. The paper gives no test, so the figure's S.E.M. band is read. Reproduced when, in all four subjects, the band on trial 1 of Figure 1E reaches 0.125 or lies below it, the rate rises over trials 1 to 15, and every point of Figure 1F (trials 1 to 15 after the first reward) has its band above 0.125.
- figs1c-swr-reward-consumption
SWRs occur mainly during reward consumption at the home port, the center ports and the rewarded goal arm, and are rarely detected at unrewarded outer ports.
SWRs occurred primarily during reward consumption (Supplementary Figure 1C) ... As expected, SWRs were rarely detected at non-goal outer arm ports, which did not provide reward
Results, Most SWRs contain interpretable spatial content, paragraph 1; Supplementary Figure 1CHow it will be judged
SWRs detected as in the Methods (consensus ripple envelope, 150 to 250 Hz, over all CA1 cell-layer tetrodes, above 2 SD of the session mean for at least 15 ms, bounds at the mean crossing, immobility below 4 cm/s). Per session, time, SWR count and SWR rate in each trial phase (home port, center port before and after reward, rewarded outer port, unrewarded outer port), as dfs_ripcontent.m part 12. Reproduced when, in all four subjects, the median over sessions of the SWR count at unrewarded outer ports is the lowest of the five phases and its SWR rate is below that of each reward phase.
- figs3abc-decode-accuracy
During movement the decoded position (posterior maximum) is usually on the same or an adjacent maze segment as the real position, and the median deviation between them is less than one 5 cm bin, on every maze segment.
The median deviation between predicted and real position was less than a single spatial bin (5 cm), indicating that the maximum of the posterior and the subject's real position most often corresponded to the same spatial bin, and this was consistent across all segments of the maze
Results, Most SWRs contain interpretable spatial content, paragraph 3; Supplementary Figure 3A-C; Methods, Spatial decoding, paragraph 5How it will be judged
Decoding as the Methods give it (clusterless, encoding on spikes above 100 uV at > 4 cm/s with a 20 uV Gaussian kernel on the four amplitudes and 5 cm bins, 2 ms steps, a continuous and a uniform fragmented movement state, 0.98 probability of staying in a state, acausal smoother, each session decoded alone, a spike's own position not used to decode it). For 30 Hz frames during movement, the frame is same/adjacent when the real and decoded segments match or one is the box; for those frames the deviation in 5 cm bins. Reproduced when, in every subject, the median deviation is 0 bins (less than one bin) on the pooled frames and on each of the nine segments. dfs_decodequant_movement.m part 2 plots the mean deviation per segment; it is shown as a variant.
- figs3df-goal-arm-no-remapping
Decoding error on the goal arm shows no systematic change over the course of a trial block, so repeated reward does not change spatial coding at the ensemble level.
We found no systematic changes in the deviation over the course of the trial block, indicating that the changing reward location did not alter overall spatial coding at the ensemble level
Results, Most SWRs contain interpretable spatial content, paragraph 3; Supplementary Figure 3D-FHow it will be judged
A claim of absence. The legend names no test; dfs_decodequant_movement.m part 6 compares the mean deviation in the last 5 bins (25 cm) of the goal arm on the first against the last rewarded goal visit of each block with a Wilcoxon rank-sum test, and fits a line to the per-visit deviations of each block with at least 4 visits. Reproduced when the rank-sum test gives p >= 0.05 in each subject; the slope distributions (median and its sign) are reported either way.
- figs3g-continuous-fraction
About 95% of SWRs contain spatially continuous representations.
We found that the vast majority of SWRs (~95%) contained spatially continuous representations of the environment
Results, Most SWRs contain interpretable spatial content, paragraph 4; Supplementary Figure 3GHow it will be judged
Per session, spatially continuous SWRs (as in the opening comment, with the 0.3 single-segment share) over all SWRs inside trial intervals, as dfs_ripcontent.m part 3 (fraccohall). Reproduced when the session mean of each subject is within tolerance.
- likelihood-decode-same-segment
A history-independent Bayesian decoder with 20 ms bins and no movement model assigns more than 90% of SWRs to the same maze segment as the state space decoder.
We found that >90% of SWRs decoded to the same maze segment and our main results were unchanged with the use of this simplified decoding approach
Results, Most SWRs contain interpretable spatial content, paragraph 4How it will be judged
The comparison is not described beyond the sentence. Readings, each run, (a) over all SWRs, the segment with the largest summed posterior during the SWR under each decoder; (b) over spatially continuous SWRs only. Reproduced when the fraction is above 0.9 under one reading, pooled over subjects and in each subject. The "main results unchanged" part is tested in robustness-controls.
- fig2f-remote-fraction
A substantial fraction of all detected SWRs contain remote replay, 35.1%, 41.1%, 34.5% and 47.2% for the four subjects.
Across all animals, we found that a substantial fraction of detected SWRs contained remote replay ... 35.1%, 41.1%, 34.5%, and 47.2% for the 4 subjects (Figure 2F)
Results, Most SWRs contain interpretable spatial content, paragraph 6; Figure 2FHow it will be judged
Per session, remote continuous events (segment other than the current one) over all SWRs in the trial intervals, then the mean over sessions, as dfs_ripcontent.m part 3 (fracremoteall, whose printed means are the reported values). Reproduced when all four are within tolerance; partially reproduced when two or three are.
- fig3a-box-majority
The large majority of spatially continuous replay events occur in the box area, before the arm choice, and include local and remote events.
The large majority of spatially continuous replay events occurred in the box area, before arm choice, and included both local and remote representations (Figure 3A)
Results, Replay content is enriched for previous goal locations, paragraph 1; Figure 3AHow it will be judged
Pooled over sessions per subject, the shares of continuous events that are box local, box remote, outer local, outer box and outer remote arm, as dfs_ripcontent.m part 3 (totalnums). Reproduced when, in all four subjects, box events are more than half of all continuous events and box local and box remote are each above zero. "Large majority" is not given a number; the box share is reported.
- fig3b-search-prevgoal-enriched
In search trials where the past, future and previous goal arms are distinct, the previous goal arm is consistently and strongly over-represented in pre-choice remote replay, about 20% of these events, in all subjects.
In contrast, we saw consistent and strong enrichment for replay of the previous goal arm ... such replay accounted for roughly 20% of all pre-choice remote replay events during these search trials (Figure 3B, previous goal)
Results, Replay content is enriched for previous goal locations, paragraph 2; Figure 3BHow it will be judged
As the Methods (Quantification of replay content) and dfs_ripcontent.m part 4. Box remote events, sessions after the first trial block, search trials with past, future and previous goal all distinct, sessions with at least 5 such trials; per session the count of events of each category over all remote events. The null is the open detail; readings, each run, (a) Methods, per session 5000 draws of a random arm 1 to 8 for each event, the null values of all included sessions pooled; (b) dfs_ripcontent.m, 100 draws per session pooled; (c) 5000 draws over the subject's events pooled across sessions. Per subject, a two-sided Wilcoxon rank-sum test of session fractions against the null. "Roughly 20%" does not say which summary; the session median (the boxplot's) and the session mean are both read. Reproduced when the previous goal fraction is above the null at p < 0.05 in all four subjects under every null reading and the median or the mean is within tolerance; partially reproduced when it is enriched in all four but neither is within tolerance.
- fig3b-search-past-depleted
In those search trials the immediately past arm is replayed far less often than chance.
we found far fewer replays of the immediate past arm than would be expected by chance (Figure 3B, past)
Results, Replay content is enriched for previous goal locations, paragraph 2; Figure 3BHow it will be judged
Selection, measure, null readings and test as fig3b-search-prevgoal-enriched. Reproduced when the past fraction is below the null at p < 0.05 in all four subjects under every reading; partially reproduced in two or three.
- fig3b-search-future-inconsistent
In those search trials the upcoming (future) arm shows a small but significant enrichment in two of the four subjects only.
while we did see a small but significant enrichment for the upcoming arm in two out of four subjects, this enrichment was not consistent across animals (Figure 3B, future)
Results, Replay content is enriched for previous goal locations, paragraph 2; Figure 3BHow it will be judged
Selection, measure, null readings and test as fig3b-search-prevgoal-enriched. Reproduced when the future fraction is significantly above the null (p < 0.05) in fewer than four subjects and its session median is below that of the previous goal in every subject. The count of subjects with significant enrichment is reported; the paper's count is two.
- fig3b-search-other
Each of the other arms is replayed at about chance per arm, yet the other arms together make up the majority of pre-choice remote replay.
Finally, replay of other arms occurred at roughly chance levels per arm (Figure 3B, other), but the total fraction of "other" events comprised the majority of replay events overall (Figure 3B, right)
Results, Replay content is enriched for previous goal locations, paragraph 2; Figure 3B rightHow it will be judged
Per session, other-arm events over all remote events divided by 5 (per arm) and not divided (total), as the Methods and dfs_ripcontent.m part 4. "Roughly chance" has no test in the text; the per-arm session median is compared with 0.125 using the protocol's 10% tolerance (0.1125 to 0.1375) and the rank-sum p against the null is reported. Reproduced when, in all four subjects, the per-arm median is within that range and the total-other median is above 0.5.
- fig3c-repeat-prevgoal
In repeat trials where past, future and current goal are the same arm and the previous goal is distinct, all animals still strongly over-replay the previous goal, while the future (past and goal) arm shows only a small and variable effect; other arms together are again the majority.
During the large subset of repeat trials in which the past, future, and goal arm refer to the same location, distinct from the previous goal arm, all animals still showed a strong and highly significant tendency to replay the previous goal location (Figure 3C, previous goal). In contrast, replay of the future arm still showed only a small and variable effect
Results, Replay content is enriched for previous goal locations, paragraph 3; Figure 3CHow it will be judged
As fig3b-search-prevgoal-enriched, for repeat trials (task phase > 1) after the first block where past, future and current goal are one arm and the previous goal differs; other arms per arm are divided by 6. Reproduced when, under every null reading, the previous goal fraction is above the null at p < 0.01 ("highly significant", the figure's ** level) in all four subjects, the future/past/goal fraction is significantly above the null in fewer than four subjects or has a median below that of the previous goal in every subject, and the total-other median exceeds 0.5 in all four.
- fig3d-glm-search
In a Poisson GLM over all search trials, being the previous goal roughly doubles an arm's replay rate, the strongest effect; the past arm is consistently replayed less; the future arm has a small increase significant in three of four subjects.
During search trials, previous goal arm status had the strongest impact on replay, such that if an arm was the previous goal, there was an approximately two-fold increase in replay rate from baseline (Figure 3D). The past arm showed a significant and consistent reduction in the rate of replay. The future category showed a small increase in replay rate that was significant in three of the four subjects but not in the fourth
Results, Replay content is enriched for previous goal locations, paragraph 5; Figure 3DHow it will be judged
As the Methods (Generalized linear models to measure the effect of arm category) and dfs_ripcontent.m part 5. Per subject, eight rows per search trial (task phase <= 1), response the count of box remote continuous events of that arm, binary predictors future, past and previous goal, Poisson log link, 99% Wald intervals, exponentiated. The trial set is the open detail; (a) Methods, trials after the first block of each session (main); (b) dfs_ripcontent.m, all search trials with the previous goal predictor zero in the first block, whose trial counts appear to be those of the legend (711 here against 534 search trials after the first block in Figure 5A), run as a variant. Reproduced under (a) when, in all four subjects, the previous goal fold change has its interval above 1 and is the largest of the three, the past fold change has its interval below 1, and the future interval is above 1 in at most three subjects; the previous goal values are compared with 2 within tolerance, and partially reproduced when only that fails. The pooled fit of the code is shown too. The legend gives 711, 848, 725 and 1065 trials per subject.
- fig3e-glm-repeat
In a Poisson GLM over all repeat trials the previous goal arm is again the strongest and most consistent driver of replay and the past arm is reliably replayed less.
Applying the GLM approach to repeat trials yielded similar results (Figure 3E), confirming that for both search and repeat phase, the previous goal arm was the strongest and most consistent driver of replay content, while the replay of the past arm was reliably reduced
Results, Replay content is enriched for previous goal locations, paragraph 5; Figure 3EHow it will be judged
As fig3d-glm-search with a fourth predictor, current goal, over repeat trials (task phase > 1), with the same two trial-set readings. In dfs_ripcontent.m part 5 the current goal on the final unrewarded trial of a block is already the new goal; this is kept. Reproduced when, in all four subjects, the previous goal interval is above 1 and its fold change is the largest of the four predictors, and the past interval is below 1. The legend gives 1347, 1661, 1154 and 1730 trials per subject.
- fig4ab-current-goal-late
Replay of the current goal is consistently enhanced only late in the repeat phase (five or more rewards at the goal), while the effects of the other three categories do not differ much between early and late repeat trials.
Indeed, we found a consistent enhancement of current goal replay only in the late portion of the repeat phase. In contrast, the three other categories did not differ substantially between early and late conditions
Results, Replay of current goal location increases with experience, paragraph 2; Figure 4A-BHow it will be judged
The repeat GLM of fig3e-glm-repeat fit separately to early and late repeat trials. Early is task phase > 1 and < 5, as dfs_ripcontent.m part 5 (the Figure 4A n equal the early correct plus early error counts of Supplementary Figure 7 for subjects 1 to 3, and differ by one trial for subject 4, 793 against 515 + 277). Late is open; readings, each run, (a) task phase >= 5, the complement of early; (b) the code's task phase > 4, under which errors after four rewards (4.3, 4.5, 4.8) enter both sets. The trial-set readings of fig3d-glm-search also apply. Reproduced when, under each reading, the current goal interval is above 1 in all four subjects for late trials and in fewer than four for early trials, and for future, past and previous goal the early and late 99% intervals overlap in at least three of four subjects. The legends give 594, 801, 573 and 793 early and 801, 895, 600 and 967 late trials per subject.
- fig4c-current-goal-reward-correlation
On correct repeat trials, the number of current goal replays correlates positively with the number of rewards already received at that goal in every subject, with low R2.
Quantified on a trial-by-trial basis, we found significant positive correlations between the number of replays of the current goal and the number of rewards already received at that location, although there was substantial variability across trials, as indicated by the low R2 values (R2=0.033, 0.015, 0.012, and 0.025; p=1.54e-6, 0.0003, 0.0004, and 2.87e-11; n=923, 1007, 847, and 1340 trials per subject, respectively)
Results, Replay of current goal location increases with experience, paragraph 2; Figure 4CHow it will be judged
As dfs_ripcontent.m part 6. Correct repeat trials after the first block (task phase integer > 1 or n + 0.9, with the old goal kept on the n + 0.9 trial), per subject pooled over sessions (n = 923 is the Figure 5B correct count, so the first-block variant of fig5b-future-correct-error is run too); Pearson correlation (corrcoef) of the count of box remote events of the current goal with reward experience. Reward experience is open; (a) the code's ceil(task phase), the reward count including the current trial's (n + 1 on the final trial); (b) rewards received before the trial (n - 1 for a rewarded trial, n on the final trial). Reproduced when, under one reading, r > 0 with p < 0.05 in all four subjects and the R2 values are within tolerance; partially reproduced when the correlations are significant and positive but some R2 is outside tolerance. The Figure 4C means per reward count (groups of at least 10 trials) are shown.
- fig5a-future-replay-rate
Replay of the future arm occurs on average less than once per trial, and more often in search than in repeat trials for three of the four subjects.
Future replay per trial tends to occur less than once per trial, and more often during search trials than repeat trials for three of the four subjects ... Search vs repeat rate p = 0.001, 5.002e-11, 0.264, and 4.111e-5 for each subject, respectively, using Wilcoxon rank-sum test
Results, Subsequent arm choice is not biased by replay, paragraph 2; Figure 5AHow it will be judged
As dfs_ripcontent.m part 7. Per trial after the first block, the count of box remote continuous events of the arm visited on that trial; per subject the mean over search and over repeat trials, and a Wilcoxon rank-sum test of the pooled trial counts. Reproduced when all eight means are below 1 and the search mean is above the repeat mean at p < 0.05 in three subjects (any three); partially reproduced in two or four.
- fig5b-future-correct-error
On repeat trials, future arm replay does not differ between correct and error trials in two subjects and is higher on error trials in the other two, the opposite of what planning predicts.
However, we found no difference in future replay rate between correct and incorrect repeat trials for two out of the four subjects (Figure 5B). For the other two subjects, future arm replay was actually higher, on average, for error as compared to correct trials
Results, Subsequent arm choice is not biased by replay, paragraph 2; Figure 5BHow it will be judged
As dfs_ripcontent.m part 6. Repeat trials after the first block, correct and error as in the opening comment; per trial the count of box remote events of the visited arm; per subject a Wilcoxon rank-sum test of correct against error trials pooled over sessions. The trial set is open; (a) after the first block, as the Methods (Quantification of replay content) and part 6 (~isnan of the previous goal), the main reading; (b) all repeat trials, first block included, run as a variant, because the legend's correct plus error counts (1347, 1661, 1154, 1730) equal the Figure 3E counts, which part 5 takes with the first block included. Reproduced when no subject has significantly (p < 0.05) more future replay on correct than on error trials, and the number of subjects with significantly more on error trials is two. Partially reproduced when no subject is higher on correct trials but the count is not two. The p entry above records the figure's lowest marked level (**).
- fig5c-current-goal-correct-error
Current goal replay does not differ consistently between correct and error repeat trials for three subjects, and the fourth has more current goal replay on error trials.
Again, we found no consistent significant difference between the amount of current goal replay on correct compared to error trials for three subjects (Figure 5C), and the fourth subject showed a significant increase
Results, Subsequent arm choice is not biased by replay, paragraph 4; Figure 5CHow it will be judged
As fig5b-future-correct-error, counting box remote events of the current goal arm (the old goal on the final unrewarded trial, as dfs_ripcontent.m part 6). Reproduced when no subject has significantly more current goal replay on correct trials and exactly one has significantly more on error trials; partially reproduced when none is higher on correct trials but the count of subjects higher on error trials is not one.
- figs7-early-late-correct-error
The lack of a correct over error difference in future and current goal replay holds both early and late in the repeat phase.
Importantly, these results were consistent for trials that occurred both early and late in the repeat phase of the trial block, indicating that familiarity with the goal location did not modulate the effect (Supplementary Figure 7A-D)
Results, Subsequent arm choice is not biased by replay, paragraph 4; Supplementary Figure 7A-DHow it will be judged
As fig5b and fig5c, split into early (task phase < 5) and late (task phase >= 5) repeat trials, as dfs_ripcontent.m part 6 (its counts, 268 early correct and 326 early error trials for subject 1, are in the legend). Reproduced when, in both halves, for future and for current goal replay, no subject has significantly (rank-sum p < 0.05) more replay on correct than on error trials.
- figs7e-remote-rate-error
Error trials have more remote replay than correct trials regardless of content, while SWR amplitude, event duration and spike rate of remote events do not differ.
We also examined features of remote events on correct and error trials, independent of their content, and found that while characteristics such as SWR amplitude, event duration, and spike rate did not differ, the rate of remote replay was consistently higher on error trials
Results, Subsequent arm choice is not biased by replay, paragraph 4; Supplementary Figure 7EHow it will be judged
As dfs_ripcontent.m part 14. Repeat trials (task phase > 1), error trials the n + 0.3, 0.5 and 0.8 trials; per trial the number of box remote continuous events, and the mean amplitude (SD threshold at which the event would still be detected), duration and spike rate of those events. The legend says number per trial; the code also shows events per second in the box, a variant. Per subject a Wilcoxon rank-sum test. Reproduced when the number per trial is significantly higher on error trials in all four subjects; the amplitude, duration and spike rate comparisons are judged as absence on the same test (p >= 0.05) and partially reproduce the claim when one of them differs in some subject.
- fig5d-glm-predict-chance
Binomial GLMs predicting correct or error repeat trials from the counts of future arm, current goal, or all four category replays perform at or very near chance (50% misclassification), far worse than the animals.
In all cases the models performed at or very close to chance levels, far worse than the animals (Figure 5D)
Results, Subsequent arm choice is not biased by replay, paragraph 6; Figure 5DHow it will be judged
As the Methods (Generalized linear models to predict correct and incorrect repeat phase trials). Per subject, repeat trials pooled, correct trials subsampled to the number of error trials, five folds, a binomial GLM (logit) on counts of future replay, of current goal replay, or of future, past, current goal and previous goal replay; misclassification is the mean over folds of the test-set error. Open details, each run, (a) number of subsamples, 1000 (Methods, main) or 100 (dfs_ripcontent.m part 6, variant); (b) how a test trial is classified, a draw from the fitted probability (the code's random(mdl, ...)) or a threshold at 0.5. Reproduced when every subject and model mean lies within tolerance under every reading; when readings differ, the claim is methods_underspecified with all results shown. The legend gives 848, 1308, 614 and 780 trials per subset per subject, twice the error trial counts of Figure 5B; the trial-set readings of fig5b-future-correct-error apply.
- pg-replay-visits-correlation
Across trial blocks, the rate of previous goal replay shows no consistent correlation with the number of visits to the previous goal during that block.
There was no consistent correlation between the rate of previous goal replay with the number of subsequent visits to the previous goal location during the subsequent trial block (for each subject: R2=0.008, 0.019, 0.021, and 0.023; p=0.49, 0.32, 0.18, and 0.08, respectively)
Results, Subsequent arm choice is not biased by replay, paragraph 8How it will be judged
A claim of absence, on the paper's test as dfs_ripcontent.m part 16. Per trial block from the second on (trailing blocks with no trial of task phase above 3 dropped), previous goal replay rate is the count of box remote events of the previous goal over all the block's trials divided by its number of trials, and visits are the block's visits to the previous goal arm, search and repeat together; Pearson correlation across blocks per subject. Reproduced when p >= 0.05 in every subject; not reproduced when p < 0.05 with the same sign in all four ("consistent", as the paper uses it); otherwise partially reproduced. R2 and p are reported.
- fig5e-pg-replay-no-avoidance
Pre-choice replay of the previous goal is no more or less likely to precede a visit to the previous goal than a visit to any other arm.
However, we found that previous goal replay was no more or less likely to precede visits to the previous goal as it was to precede visits to any other arm (Figure 5E)
Results, Subsequent arm choice is not biased by replay, paragraph 8; Figure 5EHow it will be judged
A claim of absence, per subject a Wilcoxon rank-sum test over sessions at p < 0.05. The measure is open; readings, each run, (a) dfs_ripcontent.m part 8, per session the fraction of trials visiting the previous goal that have at least one box remote event of the previous goal, against the same fraction among trials visiting any other arm; (b) the legend's "fraction of all trials", the count of trials with previous goal replay and a previous goal visit, and of those with previous goal replay and another visit, each over all trials with a defined previous goal. Trials after the first block, search and repeat. Reproduced when p >= 0.05 in every subject under reading (a), the code's; (b) is shown and its outcome recorded. The legend gives 23, 32, 23 and 33 sessions per subject and no p.
- fig6a-pg-replay-persists
Replay of a goal arm stays enriched long after it stops being rewarded, peaking several trials after the switch, extended and significant in all subjects; two subjects also show elevated replay before the switch.
The enrichment for replay of a goal location peaked several trials after it ceased to provide reward and continued for many subsequent trials. This enrichment was extended and significant for all subjects (Figure 6A). We also saw elevated replay before the transition for two out of the four subjects
Results, Previous goal replay persists through multiple trial blocks, paragraph 2; Figure 6AHow it will be judged
As the Methods (Trial-wise replay rate) and dfs_ripcontent.m part 9. Per session, count of box remote events per arm per trial, smoothed with a Gaussian kernel of five trials and SD one trial; for each block (blocks 1 to 8) the goal arm's trace from 40 trials before to 40 after the last rewarded trial, NaN beyond the session; the same window for every arm never a goal in the session. Per trial a Wilcoxon rank-sum test of goal against non-goal traces, Bonferroni corrected. The number of tests is 80 in the legend and 81 (0.05/81) in the code; both are run. Reproduced when, in all four subjects, the mean goal curve peaks after trial 0, some trials after 0 are significant with goal above non-goal, and two subjects (any two) have significant trials with goal above non-goal before trial 0. The number of significant trials after 0 is reported per subject.
- fig6b-visits-decay
The behavioural bias to the previously rewarded arm decays very quickly after the switch, returning to baseline within a few trials.
In contrast, behavioral bias to the previously rewarded arm decays extremely quickly after the switch, returning to baseline levels within a few trials (Figure 6B)
Results, Previous goal replay persists through multiple trial blocks, paragraph 2; Figure 6BHow it will be judged
As fig6a-pg-replay-persists for visits (1 if the arm was visited on that trial), unsmoothed. "A few trials" has no number; readings, (a) no significant trial after trial +3, (b) after trial +5. Reproduced when, in all four subjects, the last trial after 0 with a significant goal over non-goal visit difference is no later than the bound under one reading and the replay enrichment of fig6a extends beyond it.
- fig6c-pg-replay-two-blocks
Enhanced replay of a previous goal persists for at least two complete trial blocks after it served as the goal, in all four animals.
This analysis revealed that the enhanced replay of a previous goal location persisted for at least two complete trial blocks for all four animals (Figure 6C)
Results, Previous goal replay persists through multiple trial blocks, paragraph 3; Figure 6CHow it will be judged
As dfs_ripcontent.m part 9. Per session, blocks 1 to 8, each block split into search (task phase <= 1) and repeat (> 1) sub-blocks; a repeat sub-block counts only when the block reached a fourth reward; per sub-block the unsmoothed count of box remote events per arm over its number of trials. For each goal arm, the rates in the search and repeat sub-blocks of the three following blocks, against the same sub-blocks for arms never a goal in the session; per sub-block a Wilcoxon rank-sum test at 0.05/6. Reproduced when, in all four subjects, the first four sub-blocks (s1, r1, s2, r2) are each significant with goal above non-goal.
- fig7-recency
During search, never-rewarded arms not visited for 5 trials are replayed more than those visited one or two trials ago, and replay rate correlates with how long ago the arm was visited (1 to 5 trials) in every subject.
We found consistently higher replay rates for arms that had not been visited within 5 trials compared to those visited within one or two trials (Figure 7). Although highly variable, the replay rate of arms visited within 1-5 trials was strongly correlated with visit recency (for each subject: R2=0.011, 0.028, 0.012, and 0.018; p=1.04e-4, 4.04e-13, 7.94e-5, and 2.76e-8, respectively)
Results, Pre-choice replay is biased toward non-recent past, paragraph 1; Figure 7How it will be judged
As dfs_ripcontent.m part 8. Search trials (task phase <= 1) of all blocks; for each trial and each arm last visited k trials ago (lockout trials not counted), the count of box remote events of that arm. Open details, each run, (a) "never rewarded", the code's arms that were not a previous goal up to that trial, or the Results' arms never a goal in the session; (b) the "5 trials ago" bar, exactly 5 (the code's plotted and tested bar) or 5 or more (the Results' "not been visited within 5 trials"). Per subject, Wilcoxon rank-sum of the 5 bar against the 1 and the 2 bars, pooled trials, and Pearson correlation of count with k over k = 1 to 5. Reproduced when, under every reading, the 5 bar is significantly above the 1 and 2 bars (p < 0.05) and replay rises with k at p < 0.05 in all four subjects, with R2 within tolerance under one reading; partially reproduced when only R2 fails.
- fig8-outer-prevgoal
Replay at the rewarded outer arm port also favours the previous goal; beyond the expected strong over-representation of the current arm, the previous goal arm is replayed more than other arms, consistently in a GLM over all rewarded trials.
However, we also observed greater replay of the previous goal compared to other arms (Figure 8A). We used a Poisson GLM to quantify these effects during all rewarded trials and found that in addition to overrepresentation of the current arm (the goal arm), we also saw consistent, strong enrichment for replay of the previous goal arm (Figure 8B)
Results, Enhanced replay of previous goal at outer arms, paragraph 2; Figure 8A-BHow it will be judged
Figure 8A as dfs_ripcontent.m part 10, outer events on rewarded trials (task phase >= 1) after the first block where past, current and goal arm are one arm distinct from the previous goal, sessions with at least 5 such trials, fractions over all continuous outer events (local included); other arms per arm divided by 6. Figure 8B as part 5, all rewarded trials (integer task phase >= 1), counts of outer continuous events per arm, predictors past, current goal and previous goal, Poisson GLM with 99% intervals; the trial-set readings of fig3d-glm-search apply. Reproduced when, in all four subjects, the session mean previous goal fraction exceeds the per-arm other fraction, and the previous goal and current arm intervals are above 1. The legends give 333, 958, 1161 and 3316 events (8A) and 948, 1026, 868 and 1372 trials (8B) per subject.
- figs8-outbound-final-events
Among outbound remote events, and among the final remote event before each choice, the previous goal is still the most consistently enriched arm, and future and current goal replay still do not differ consistently between correct and error trials.
For neither of these subsets of events did we see any difference in the relationships between content and behavior ... previous goal replay remained more strongly and consistently enriched compared to future arm or current goal replay and we still did not observe consistent differences between correct and error trials (Supplementary Figure 8)
Results, Subsequent arm choice is not biased by replay, paragraph 5; Supplementary Figure 8How it will be judged
Outbound events as the Methods and dfa_ripcontent_ripspeed.m, the slope of a line fit through the posterior-maximum bin of each 2 ms step after correcting for linearisation, positive for box to arm; final events as dfs_ripcontent.m part 15, the last box remote event before the arm choice. For each subset, the search and repeat GLMs of fig3d and fig3e, and the correct against error rank-sum tests of fig5b and fig5c. Reproduced when, for both subsets, the previous goal interval is above 1 in all four subjects for search and repeat, no subject has significantly more future or current goal replay on correct than on error trials, and the future and current goal effects are not significant (p < 0.05) in the same direction in all four subjects.
- robustness-controls
The main results are unchanged with MUA-based event detection, with a more permissive definition of remote events, after excluding stationary events identified by a three-state decoder, and with a simple 20 ms Bayesian decoder.
Additionally, we verified that our conclusions were unchanged by using a multiunit activity-based event detection strategy ... instead of a SWR-based one ... Use of this alternative definition did not alter any of our conclusions
Results, Most SWRs contain interpretable spatial content, paragraphs 1, 4 and 5; Methods, Spatial decoding, paragraphs 4 and 7How it will be judged
The paper does not say which results were compared. Read as the patterns of Figures 3D and 3E (previous goal interval above 1 in all subjects, past interval below 1 in all subjects) and Figure 5B (no subject with more future replay on correct trials). Each control as the Methods give it and its script (dfs_muacontent.m, MUA 1 ms bins smoothed with a 15 ms Gaussian, events above 3 SD of the immobility mean; dfs_ripcontent_contentthresh.m, a local-majority event is remote when at least 30% of its density is on one remote segment; dfs_ripcontent_3state.m, the three-state decoder with stationary events excluded; dfs_likcontent.m, 20 ms non-overlapping bins without a movement model). Reproduced when the three patterns hold under all four controls; partially reproduced when they hold under some. The share of remote events classed stationary (paper, < 30%) is reported.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none