CRCNS:pfc-6 · Encore Science

Claims

From the paper 10.1038/nn.2337

Reproducing Peyrache et al. (2009) from CRCNS pfc-6 15

Notebook reproduce_peyrache_2009.ipynb

Reproduced4

ReproducedResults, Transient synchronized replay of awake patterns, paragraph 2; Figure 2b and 2c
For signal components, the average reactivation strength over SWS is greater in the post epoch than in the pre epoch, in each of the three groups of encoding strength.
The average reactivation strength was greater during post epoch SWS than pre epoch SWS for signal components (P o 0.005 all comparisons, n 1⁄4 10, 40 and 273 for the three signal groups observed here, sorted according to their encoding strength, n 1⁄4 811 for nonsignal components; Fig. 2b,c).
Results, Transient synchronized replay of awake patterns, paragraph 2; Figure 2b and 2c
What the paper reports
- post against pre average reactivation strength, each signal group: p < 0.005 (n = 10, 40 and 273 signal components in the three groups; 811 nonsignal components)
Post exceeds pre in all three signal groups under all four readings. With z over the whole rest epoch, paired t p = 0.0015, 0.00014 and 0.00023 for Phi above 1.8, 1.4 to 1.8 and 1 to 1.4 (Wilcoxon 0.0039, 9.2e-05, 8.6e-06); with z over SWS bins, 0.00071, 2.2e-06 and 5.2e-07. Nonsignal components do not differ (p 0.10 to 0.31). The group sizes, 9, 35 and 241 against 10, 40 and 273, are within tolerance only for the first; each is about 88% of the paper's, as the included sessions are 56 of 63. The opposite order of sizes matches none. By rat the effect is found in rats 15, 18 and 20 and not in rat 19 (mean difference -0.003, p = 0.87). Under the 3.5 min SWS variant of fig1b the group sizes are 11, 39 and 265, all within tolerance.
How the test differed from the paper’s method
- The outcome follows the criterion, which tests post against pre; two of the three group sizes listed as expected values fall outside tolerance, because the data set holds fewer sessions that pass the inclusion rule.
ReproducedResults, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 6verifier disagreed at first
Pre and post epoch SWS do not differ in average SWS duration, average firing rates, rates of SPWRs and delta waves, or LFP power in the delta and spindle bands.
Absence of significant differences between PRE and POST SWS in global measures of activity and oscillations (paired t-test, p>0.05).
Results, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 6
What the paper reports
- paired t-test, p > 0.05
Paired t-tests over 56 sessions find no difference in total SWS time (630 against 623 s, p = 0.87), mean SWS episode duration (203 against 222 s, p = 0.41) or session mean SWS firing rate (3.58 against 3.58 Hz, p = 0.93); over 1499 cells, rates do not differ by paired t-test (p = 0.65), while a Wilcoxon test over cells, a variant, gives p = 0.0013 for a 0.8% difference. SPWR rate, delta-wave counts and delta and spindle power cannot be tested, as the data set holds no LFP and no event times; the outcome covers the SWS duration and firing-rate parts only.
How the test differed from the paper’s method
- SWS episodes are the data set's SWS intervals; their merging rule (gaps under 1 min) is the paper's, applied before deposit.
ReproducedResults, Transient synchronized replay of awake patterns, paragraph 3; Supplementary Figures 9 and 10
During sleep, reactivation strengths of different simultaneously recorded signal components tend not to peak together, with a trough at zero lag in their cross-correlograms, while in the awake epoch they coactivate more than by chance.
concomitant activation of different principal component–related patterns was less than expected by chance, as can be inferred from the zero lag trough in their cross-correlograms ... During sleep epochs, pairs of PCs are more unlikely than chance to co-activate
Results, Transient synchronized replay of awake patterns, paragraph 3; Supplementary Figures 9 and 10
Under reading (a), Pearson cross-correlation averaged over 704 pairs of signal components from 56 sessions, the post SWS average has its minimum at zero lag, below chance (-0.0035, p = 0.29 against 0, with z over the epoch; -0.0099, p = 0.00045, with z over SWS) and below its mean at 200 to 500 ms (0.0017 and 0.0016). In the awake epoch the zero-lag value is above chance (0.0329, p = 8.5e-16). Reading (b) is unstable because mean R over SWS is near zero for many components; its sleep averages are not compared with chance (zero-lag s.e.m. 33 and 430). The trough is significant under one z-transform reading only.
How the test differed from the paper’s method
- Cross-correlations are computed within each SWS interval and pooled, so that no pair of bins spans a gap between intervals; lags of -1 to 1 s (a standard choice).
ReproducedResults, Replay of choice point activity increases with learning, paragraph 3; Supplementary Figure 12B and 12C
Rule acquisition does not change trial duration or the running speed along the trajectory.
Average trial length (time from initial movements by rat until arrival at end of arms) before and after learning (10 learning days). The distributions are not significantly different (t-test, p>0.05).
Results, Replay of choice point activity increases with learning, paragraph 3; Supplementary Figure 12B and 12C
What the paper reports
- trial length before against after learning: t-test, p > 0.05 (n = 10 learning days)
In the ten learning sessions, trial length does not differ before and after the learning trial, by session means (5.48 against 8.23 s, paired t p = 0.16, n = 10) or by trials (5.35 against 6.81 s, unpaired t p = 0.28; medians 4.30 s both). Speed along the path does not differ in 9 of 10 bins; the first bin is faster after learning (24.9 against 23.3 units/s, p = 0.034 uncorrected, 0.34 after Bonferroni over the ten bins). The test has little power (a 2.7 s difference over ten sessions is not significant). With the learning points of learning-sessions reading (a) in the same ten sessions, session means give p = 0.19.
How the test differed from the paper’s method
- The learning sessions and trials are those listed in the data set's description, since the paper's definition does not recover them (learning-sessions).
- Speed is shown in ten bins of the fraction of the trial's path length, as the data set holds no maze geometry for the paper's linearization; it is reported, and does not decide the outcome.

Partially reproduced6

Partially reproducedResults, mPFC ensemble activity patterns during a rule shift task, paragraph 3; Figure 1b; Online Methods, Signal and nonsignal components
Comparing the awake-epoch correlation matrix eigenvalues with the Marcenko-Pastur bound typically gives 1 to 6 signal components per session; 323 components were signal and 811 nonsignal in total.
we could typically discriminate 1–6 signal components (and occasionally more) in each session (Fig. 1b) ... For the awake epoch, a total of 323 components were classified as signal and 811 were classified as nonsignal.
Results, mPFC ensemble activity patterns during a rule shift task, paragraph 3; Figure 1b; Online Methods, Signal and nonsignal components
Over the 56 included sessions, 285 signal components (expected 291 to 355) and 730 nonsignal (730 to 892); 484 components lie below lambda_min. Per session this is 5.09 signal and 13.04 nonsignal, against 5.13 and 12.87 for the paper's totals over 63 sessions, so the shortfall in the signal total follows the missing sessions. 45 of 56 sessions have 1 to 6 signal components and 11 have 7 to 11. Over all 86 sessions there are 411 and 1050. Leaving out the 12 cells silent in the awake epoch gives 287 and 727. With a 3.5 min SWS threshold, a variant outside the criterion that matches the paper's total of sessions but not its counts per rat, the totals are 315 and 820, both within tolerance, with 51 of 63 sessions at 1 to 6 signal components.
How the test differed from the paper’s method
- The criterion takes the session selection from the matching reading of sessions-and-cells; neither matches, so the paper's stated inclusion rule, reading (b), is used.
- The awake epoch is the data set's single awake interval, binned whole in 100 ms bins.
Partially reproducedResults, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 5verifier disagreed at firstrests on an assumption
Replay is concentrated in SWS; reactivation strength of signal components shows effects of SWS against non-SWS and of post against pre, and no post-pre effect is seen in rest periods not classified as SWS.
No such effects were observed in the rest periods that were not classified as SWS (as shown in Supplementary Fig. 5 online) ... A two-way ANOVA revealed significant effects of both SWS vs non-SWS (p < 0.01) and POST VS. PRE (p < 0.01).
Results, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 5
What the paper reports
- SWS against non-SWS: two-way ANOVA, p < 0.01
- post against pre: two-way ANOVA, p < 0.01
Repeated-measures ANOVA over the per-component means of the 285 signal components (component as subject), state effect p = 0.0019, epoch effect p = 4.9e-08, interaction p = 1.6e-05, so both main effects pass p < 0.01; an ordinary two-way ANOVA, treating the four means as independent, gives state p = 0.014 and epoch p = 0.0005. SWS means pre 0.1005 and post 0.1469, non-SWS 0.0985 and 0.1088. In non-SWS bins post exceeds pre by 0.0103 (paired t p = 0.042, Wilcoxon p = 0.018), against 0.0464 in SWS, so the claim's third part fails under both readings, though within each signal group the non-SWS difference is not significant (p 0.13 to 0.49). With rest epochs limited to the 30 min next to the awake epoch the non-SWS result is the same (non-SWS t p = 0.044; ordinary ANOVA state p = 0.018, epoch p = 0.00047). Two of the three parts hold and the absence outside SWS fails.
How the test differed from the paper’s method
- The data set does not give the bounds of the rest epochs. Assumed, as the claim's criterion directs, from the data set's epoch times, pre from the start of the recording to the start of the awake interval and post from its end to the end of the position record, or to the end of the last post SWS interval if that is later; this makes pre 23 to 58 min and post 22 to 59 min in the included sessions, where the Methods give 20 to 30 min. A 30 min variant is shown.
- Non-SWS bins are z-transformed with the whole rest epoch.
- The ANOVA is a repeated-measures ANOVA with the signal component as subject, since the four means of a component are not independent; the legend names only an ANOVA. The ordinary two-way ANOVA is shown.
Partially reproducedResults, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 7
In pre epoch SWS the relation between reactivation strength and encoding strength is significantly weaker than in post, in slope and in correlation.
During pre SWS, this relationship was significantly weaker with respect to slope (P o 10�20, n 1⁄4 323) and correlation (P o 10�5, n 1⁄4 323; Supplementary Fig. 7).
Results, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 7
What the paper reports
- slope, pre against post: p < 10⁻²⁰ (n = 323 signal components)
- correlation, pre against post: p < 10⁻⁵ (n = 323 signal components)
Pre slope 0.168 against post 0.434, pre r = 0.283 against post 0.620 (z over the epoch; 0.150, 0.398, 0.319 and 0.646 with z over SWS). The correlation is lower with p < 1e-5 under both readings (Fisher z p = 2.5e-07 and 1.9e-07, Steiger p = 6.2e-13 and 2.5e-15). The slope is lower but with p of 2.4e-08 to 3.3e-13 (interaction in a linear model, or bootstrap of the slope difference with a normal approximation), not below 1e-20 under any reading.
How the test differed from the paper’s method
- The bootstrap p-value uses a normal approximation from the bootstrap standard error, since 2000 resamples cannot resolve p below 5e-4; no resample gave a slope difference of zero or less.
Partially reproducedResults, Transient synchronized replay of awake patterns, paragraph 4; Figure 2f
About 40 to 50% of the net reactivation (post minus pre) comes from events with reactivation strength beyond the 99th percentile.
40–50% of the net reactivation (difference between post and pre) came from events with reactivation strengths beyond the 99th percentile.
Results, Transient synchronized replay of awake patterns, paragraph 4; Figure 2f
The measured value is reading (a), z over the epoch, for Phi above 1.8. Reading (a) gives 52.2%, 56.1% and 16.1% for Phi above 1.8, 1.4 to 1.8 and 1 to 1.4 with z over the epoch, and 51.0%, 60.6% and 36.6% with z over SWS bins (one and two of three groups in 36 to 55%). Reading (b) gives two of three in range with z over SWS (46.3%, 45.8%, and an extreme mean for 1 to 1.4) and none with z over the epoch, as components whose net reactivation is near zero give extreme ratios (99 of 241 in the lowest group have a negative net). Per-component medians, not a reading of the criterion, are 38.8% to 51.7%. The extreme reading (b) means (5459%, one ratio above 10^6 %) are not well defined; the count of groups in range rests on reading (a) and on the two highest groups of reading (b) with z over SWS, whose per-component ratios stay within 149%.
Partially reproducedResults, Transient synchronized replay of awake patterns, paragraph 3; Supplementary Figure 11D
Reactivation strength computed after permuting cell identities in each time bin, which keeps global firing rate fluctuations, does not differ between pre and post and is more than an order of magnitude smaller than for the actual patterns.
For shuffled patterns, the reactivation strength was not significantly different between sleep PRE and POST. Furthermore, it was more than one order of magnitude smaller than for actual patterns.
Results, Transient synchronized replay of awake patterns, paragraph 3; Supplementary Figure 11D
Shuffled reactivation is 5.3% to 5.9% of the actual in each signal group with z over the epoch, and 5.1% to 5.7% with z over SWS bins, more than an order of magnitude smaller. It does not differ between pre and post for Phi 1 to 1.4 (paired t p = 0.85 and 0.76) or 1.4 to 1.8 by t-test (p = 0.13 and 0.067), but for Phi above 1.8 post is higher (0.0204 against 0.0162, paired t p = 0.031; 0.034 with z over SWS), and by Wilcoxon the two highest groups differ (p 0.012 to 0.023). The absence holds in two of three groups by the t-test reading and one by the Wilcoxon reading. The differences are under 0.005, against 0.27 for the actual patterns in the highest group.
How the test differed from the paper’s method
- The mean over shuffles is computed in closed form, as the mean over all permutations of the cells within each bin, instead of 1000 random shuffles per bin; on one session it agrees with 1000 shuffles (-0.0082 against -0.0086).
Partially reproducedResults, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 2; Figure 6bverifier disagreed at firstrests on an assumption
Reactivation and SPWR occurrence relative to putative DOWN to UP transitions, with DOWN states defined as decreases of neural activity lasting at least 80 ms, are comparable with the delta wave results.
the same analysis was carried out with reference to putative DOWN to UP state transitions (putative DOWN states were defined as a decrease of neural activity in windows of at least 80 ms; Fig. 6b). The relationships between reactivation and SPWR occurrence with these transitions were comparable with our delta wave results.
Results, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 2; Figure 6b
In post SWS the signal components' event-triggered reactivation rises above baseline 0.5 to 0.3 s before DOWN to UP transitions (three bins at p < 0.001 over 285 components, 15415 transitions; a fourth significant bin is a dip below baseline at 0 s, -0.059), as the paper reports for delta waves (about 400 ms before), and nonsignal components have no significant bin. In pre SWS there is one significant bin, and the rise over -0.5 to -0.3 s is lower (0.063 against 0.107; per component post minus pre 0.043, p = 0.15); the pre average's largest value above baseline (0.145 at +0.2 s) is not lower than the post one (0.132), and that value is at +0.2 s after the transition, pre's one significant bin. The post rise before the transition holds, and the pre average is lower only in the window where post is significant, without a significant difference, so the claim holds in part. Both definitions of DOWN states give the same transitions, since the mean population count per 10 ms SWS bin is 0.21 to 2.22 spikes and a fifth of it is below one spike in every session. The SPWR part of the claim is not tested (no ripple times), and the comparison with delta waves rests on the paper's reported 400 ms, as fig6a-delta-waves is blocked.
How the test differed from the paper’s method
- Assumed that "the pre average is lower" holds only if pre is lower both by the rise above baseline in the window where post is significant (-0.5 to -0.3 s) and by the largest value of the whole average, or if the per-component difference is significant. Pre is lower in the window (0.063 against 0.107) but its largest value is higher (0.145 against 0.132), and the per-component difference is not significant (p = 0.15), so this part is not established. Comparing by the window alone would make the claim reproduced. Both comparisons are shown.
- Event-triggered averages use R over all bins of the rest epoch with Q z-transformed over the whole epoch, since the -4 to 4 s windows reach outside SWS; transitions are taken inside SWS.

Not reproduced5

Not reproducedResults, mPFC ensemble activity patterns during a rule shift task, paragraph 2
The paper recorded 1,692 mPFC cells from four rats in 63 sessions (rat 15, 16 sessions; rat 18, 11; rat 19, 12; rat 20, 24), and analyzed only sessions with at least ten cells and at least 4 min of SWS in each rest epoch.
We recorded 1,692 cells in the mPFC (Supplementary Fig. 1) from four rats during a total of 63 recording sessions (rat 15, 16; rat 18, 11; rat 19, 12; rat 20, 24). Only sessions with a minimum of ten cells and at least 4 min of SWS in each rest epoch were analyzed.
Results, mPFC ensemble activity patterns during a rule shift task, paragraph 2
Measured values are those of reading (b), the inclusion rule, which the Figure 2b legend supports. Reading (a), all sessions, gives 86 sessions and 2155 cells; reading (b) gives 56 sessions and 1499 cells, against 63 (57 to 69) and 1692 (1523 to 1861). Under (b) rats 18 and 20 match the paper's per-rat counts (11 and 24), rat 19 has 11 against 12 and rat 15 has 10 against 16. The data set holds only 14 sessions of rat 15 in all, so at least two of the paper's rat-15 sessions are not in it. With a 3.5 min SWS threshold the totals match (63 sessions, 1682 cells) but the per-rat counts do not (11, 12, 11, 29).
How the test differed from the paper’s method
- SWS time per rest epoch is the summed length of the data set's SWS intervals; the paper's own SWS detection, from the LFP, cannot be repeated.
Not reproducedOnline Methods, Behavioral task, last paragraph
All four rats learned the right-arm and lit-arm rules; two learned the left-arm rule and one the dark-arm rule.
All four rats learned the right and light arm rules (according to above criteria), whereas only two learned the left arm rule and one learned the dark arm rule.
Online Methods, Behavioral task, last paragraph
Both readings (within a session, and across consecutive sessions with the same rule) give right arm 3 rats (15, 19, 20), lit arm 4, left arm 3 (15, 18, 20) and dark arm 1 (20), against 4, 4, 2 and 1. Rat 18 never reaches criterion on the right-arm rule in the recorded trials (9 and 13 trials in sessions 181011 and 181012), though its rule was then changed. Counting instead the rules after which the rule was changed, a variant outside the criterion, gives 4, 3, 3 and 1. The recorded trials may not include every trial run. Over the right-arm trials rat 18 has 22 trials, a longest correct run of 9 and at most 10 correct in 12, one trial short of either form of the criterion; on the left-arm rule rats 15 and 18 pass clearly (runs of 14 and 13).
How the test differed from the paper’s method
- Sessions are ordered by the month and day in their ids, with rat 20's December sessions put before its January to March ones; the data set gives no year.
Not reproducedFigure 7b legend; Online Methods, Behavioral task, paragraph 3
Ten sessions, from four rats, contained rule learning, with a learning point defined as the first trial of a block of at least three consecutive rewarded trials after which performance to the end of the session stayed above 80%.
during sessions with learning (n 1⁄4 10, four rats)
Figure 7b legend; Online Methods, Behavioral task, paragraph 3
The measured value is reading (a)(i). No reading gives 9 to 11 sessions over four rats. (a)(i) 34 sessions, (a)(ii) 16, (b)(i) 22 and (b)(ii) 14, each over four rats. Under reading (b) a block of correct trials that runs to the end of the session leaves no trials after it and counts as passing. The data set's description lists ten learning sessions from four rats with their learning trials, the paper's count. The definition under reading (a) gives the listed trial in 6 of the 10 (150628, 150630, 181012, 190214, 190228, 201229), and reading (b) in 4 (150628, 150707, 181012, 190228). Neither recovers 181020 (trial 4 against 10), 201222 (37 or none against 23) or 201227 (31 or 40 against 23).
How the test differed from the paper’s method
- The learning point is computed over all trials of the session, as the definition says, even where the rule changed within the session.
- All 86 sessions are searched, not only those passing the inclusion rule; the table shows which of the listed sessions are included.
Not reproducedResults, Transient synchronized replay of awake patterns, paragraph 2; Figure 2b; Supplementary Figure 7
Across signal components, post epoch reactivation strength correlates positively with encoding strength (r squared 0.61).
Reactivation strength correlated positively with encoding strength (r2 1⁄4 0.61, P o 10�30, Pearson correlation test, n 1⁄4 323; Fig. 2b and Supplementary Fig. 7 online).
Results, Transient synchronized replay of awake patterns, paragraph 2; Figure 2b; Supplementary Figure 7
What the paper reports
- Pearson correlation test, r2 = 0.61, p < 10⁻³⁰ (n = 323 signal components)
Over 285 signal components, r = 0.620 (r squared 0.384, p = 1.2e-31) with z over the epoch and r = 0.646 (r squared 0.417, p = 4.5e-35) with z over SWS bins; the expected r squared is 0.549 to 0.671. The correlation is positive and its p-value is below 1e-30, but r squared is too low. The r itself is close to the 0.61 the paper prints as r squared. By rat r squared is 0.20, 0.43, 0.03 and 0.50 (rats 15, 18, 19, 20). Reading (b), inter-SPWR bins, cannot be run.
Not reproducedResults, Transient synchronized replay of awake patterns, paragraph 3; Figure 2d and 2e
Post epoch SWS reactivation strength distributions are heavy tailed, more skewed than pre epoch ones, most markedly for components of higher encoding strength.
The difference in tail weight was reflected in the significant difference between the post and pre epochs in the skewness of the signal reactivation–strength histograms (P o 0.05, t test, n 1⁄4 10–40), most markedly for patterns with higher encoding strengths (Fig. 2e).
Results, Transient synchronized replay of awake patterns, paragraph 3; Figure 2d and 2e
What the paper reports
- t test, p < 0.05 (n = 10 to 40 signal components)
Post minus pre skewness of R over SWS bins is positive and significant only for Phi 1.4 to 1.8 (3.44, p = 0.0092, n = 35, z over the epoch; 3.25, p = 0.0023 with z over SWS). For Phi above 1.8 it is negative (-1.19, p = 0.62, n = 9; -1.52, p = 0.54), and for 1 to 1.4 near zero (p = 0.79 and 0.40). Under neither grouping reading is every tested group positive with p < 0.05, and the difference does not rise with Phi. Skewness values are large (5 to 11), set by rare extreme bins.

Not tested in a notebook 14

14 data missing from the dataset

Blocked14

BlockedResults, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 8· Data missing from the dataset
The replay effect persists when cell pairs from the same tetrode are left out, and pairs from the same tetrode alone show no replay.
they persisted when cell pairs discriminated from the same tetrode were ignored. Moreover, cell pairs from the same tetrodes, when considered alone, showed no replay effect
Results, Transient synchronized replay of awake patterns, paragraph 2; Supplementary Figure 8
Needs the tetrode of each cell. The spike file gives only a cell number per session, and neither the data description nor any other file of the data set maps cells to tetrodes.
BlockedResults, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5a; Supplementary Figure 7B· Data missing from the dataset
In post epoch SWS, average reactivation strength of signal components is greater in bins coinciding with SPWRs than in other bins, and the effect is stronger for higher encoding strength.
The average reactivation strength in post epoch SWS was considerably greater for bins coinciding with SPWRs than for non-SPWR bins (all comparisons P o 0.005, n 1⁄4 8, 37 and 225, including only sessions with reliable discrimination of ripple signals; Fig. 5a). The effect was stronger for components with higher encoding strengths (Pearson’s correlation test, P o 10�20
Results, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5a; Supplementary Figure 7B
What the paper reports
- SPWR against non-SPWR bins, each signal group: p < 0.005 (n = 8, 37 and 225 signal components)
- correlation of the effect with encoding strength: Pearson's correlation test, p < 10⁻²⁰
Needs SPWR times or the hippocampal LFP, neither of which is in the data set (its files are spikes, cell types, behaviour, position, and awake and SWS intervals) or published elsewhere that the paper or the data set names.
BlockedResults, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5b· Data missing from the dataset
Around SPWRs the reactivation strength of signal components rises by about 70% above baseline in post, but not pre, epoch SWS, returns to baseline within 1 s, and shows no such rise for nonsignal components.
the reactivation strength for signal components during post, but not pre, epoch SWS increased by B70% at the time of the sharp waves with respect to baseline (P o 10�10, n 1⁄4 270). Reactivation in mPFC declined to baseline values within 1 s before and after the peak of the SPWR events.
Results, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5b
What the paper reports
- p < 10⁻¹⁰ (n = 270 signal components)
Needs SPWR times; see fig5a-spwr-bins.
BlockedResults, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5c· Data missing from the dataset
At higher time resolution, reactivation peaks about 40 ms after SPWRs.
A similar analysis at higher time resolution (Fig. 5c) showed that reactivation peaked B40 ms after SPWR occurrences
Results, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5c
Needs SPWR times; see fig5a-spwr-bins.
BlockedResults, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5d· Data missing from the dataset
Overall mPFC population activity rises transiently with SPWRs and stays elevated after them, an asymmetric profile unlike reactivation, with no difference between pre and post.
on average, prefrontal population activity transiently increased with the SPWRs and maintained sustained activity thereafter20,21 (with no difference between pre and post epochs)
Results, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5d
Needs SPWR times; see fig5a-spwr-bins.
BlockedResults, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5e· Data missing from the dataset
Autocorrelograms of post SWS reactivation strength and of SPWR occurrences decay with similar time constants, 150 and 160 ms in exponential fits.
autocorrelograms of both reactivation strength and SPWR occurrences (Fig. 5e) decayed with very similar time constants (150 and 160 ms for exponential fits, respectively)
Results, Preferential mPFC replay during hippocampal SPWRs, paragraph 2; Figure 5e
Needs SPWR times for the SPWR autocorrelogram and the time constant it is compared with; see fig5a-spwr-bins.
BlockedResults, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 1· Data missing from the dataset
Episodes of strong replay are concentrated in periods of elevated prefrontal LFP delta (2 to 4 Hz) and spindle (10 to 20 Hz) power.
episodes of strong replay were significantly concentrated into periods of elevated prefrontal LFP oscillatory activity in the delta (2–4 Hz) and spindle (10–20 Hz) ranges (P o 10�5 for all, t test; Supplementary Fig. 5)
Results, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 1
What the paper reports
- t test, p < 10⁻⁵
Needs the prefrontal LFP for delta and spindle power, which is not in the data set.
BlockedResults, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 1; Figure 6a· Data missing from the dataset
In post epoch SWS, reactivation strength of signal components rises significantly about 400 ms before the peak of cortical delta waves; the modulation is smaller in pre SWS and absent for nonsignal components.
During post epoch SWS, the reactivation strength for signal components showed a significant (P o 0.001, t test) increase B400 ms before the peak of the delta wave (Fig. 6a). This was experience dependent and possibly memory related, as the modulation was smaller for pre epoch SWS and null for nonsignal components.
Results, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 1; Figure 6a
What the paper reports
- t test, p < 0.001
Needs delta-wave times from the prefrontal LFP, which is not in the data set.
BlockedResults, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 2; Figure 6c· Data missing from the dataset
Reactivation strength is higher in the about 1 s before spindle troughs than after them, echoing a higher SPWR probability before spindles, while population activity peaks symmetrically at spindles.
Reactivation event-triggered averages centered on spindles troughs were asymmetric, with an increase in reactivation in the B1 s preceding spindles compared with the period thereafter (P o 0.001, t test; Fig. 6c).
Results, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 2; Figure 6c
What the paper reports
- t test, p < 0.001
Needs spindle times from the prefrontal LFP, which is not in the data set.
BlockedResults, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 2; Figure 6d· Data missing from the dataset
Relative to each of delta waves, DOWN to UP transitions and spindles, the time course of reactivation strength and that of SPWR occurrence are strongly correlated.
The respective cross-correlograms and event-time averages of reactivation relative to these three cortical events were strongly correlated (Fig. 6d).
Results, Relation of slow oscillations with SPWRs and mPFC replay, paragraph 2; Figure 6d
What the paper reports
- the three panels of Figure 6d; the text does not say which value belongs to which event: r = 0.79, r = 0.53 and r = 0.96, p < 0.001
Needs SPWR times and the delta-wave and spindle times; see fig5a-spwr-bins.
BlockedResults, Replay of choice point activity increases with learning, paragraph 1; Figure 7b; Supplementary Table 1· Data missing from the dataset
In learning sessions, significantly more principal components from post epoch SPWR activity are stronger in the awake epoch after the learning point than before (15 of 30 against 1 of 30).
For a significant number of components (P o 0.05) extracted from SPWR bins (Fig. 7a,b, Supplementary Table 1 and Supplementary Fig. 12 online), coactivations became stronger as the rat started the first run of correct trials marking rule acquisition.
Results, Replay of choice point activity increases with learning, paragraph 1; Figure 7b; Supplementary Table 1
What the paper reports
- multinomial confidence intervals (Methods); two-way ANOVA followed by t test (Figure 7b legend), p < 0.05 (n = 30 principal components from 10 learning sessions, four rats)
The templates are computed from SWS bins at SPWRs, and the data set holds no SPWR times. The learning sessions and trials it needs are in the data set's description.
BlockedResults, Replay of choice point activity increases with learning, paragraphs 1 and 3; Figure 7b; Supplementary Table 1; Online Methods, Detection of sleep-related patterns, paragraph 1· Data missing from the dataset
No such difference is found for components from pre epoch SPWRs or from inter-SPWR periods, between the halves of sessions without learning (31 of 111 lower, 22 higher), or across switches between unrewarded strategies (7 of 24 lower, 8 higher).
This difference was not significant (P 4 0.05) for principal components computed from the inter-SPWR intervals in post or pre epochs ... For nonlearning days, a total of 111 principal components were considered; 31 were lower (28%) and 22 were higher (20%) when the halves of the session were compared. For nonrewarded strategy switches, 24 principal components were considered; seven were lower (29%) and eight were higher (33%).
Results, Replay of choice point activity increases with learning, paragraphs 1 and 3; Figure 7b; Supplementary Table 1; Online Methods, Detection of sleep-related patterns, paragraph 1
What the paper reports
- p > 0.05
Every control uses templates from SPWR or inter-SPWR bins, and the data set holds no SPWR times.
BlockedResults, Replay of choice point activity increases with learning, paragraph 2; Figure 7c· Data missing from the dataset
Post SPWR components appear mainly when the rat is on the central platform, and the effect of learning on their spatial distribution is significant only in the part of the maze from the central platform toward the end of the target arm.
the principal components computed from post SPWR appeared primarily when the rat was on the central platform of the Y maze ... A significant effect (P o 0.05) of learning on the spatial distribution of principal components from post SPWR appeared only in the part of the maze going from the central platform toward the end of the target arm (Fig. 7c).
Results, Replay of choice point activity increases with learning, paragraph 2; Figure 7c
What the paper reports
- t test, p < 0.05 (n = 10)
Depends on fig7b-post-spwr-after-learning, blocked for want of SPWR times.
BlockedResults, Replay of choice point activity increases with learning, paragraph 2; Figure 7d· Data missing from the dataset
Factor analysis of the spatial distributions of post SPWR patterns gives two main factors, concentrated on the central platform and on the target arm.
a factor analysis of these spatial distributions revealed that the two most important factors were concentrated on the platform and on the target arm, respectively (Fig. 7d and Online Methods).
Results, Replay of choice point activity increases with learning, paragraph 2; Figure 7d
Depends on fig7c-choice-point, blocked for want of SPWR times.

Findings about the data

What attempts found about the dataset itself, for its owner and the archive.

Data missingThe data set holds no LFP, no SPWR, delta-wave or spindle times, and no tetrode of each cell. The paper's analyses of hippocampal sharp-wave ripples, cortical delta waves and spindles (Figures 5 to 7) and its same-tetrode control cannot be repeated, 14 of 29 claims.
Data missingThe data set holds 14 sessions of rat 15, where the paper reports 16 analysed. With the paper's inclusion rule (at least ten cells and 4 min of SWS in each rest epoch) 56 sessions and 1499 cells remain, against 63 and 1,692 in the paper; rats 18 and 20 match the paper's counts, rat 19 has one fewer and rat 15 six fewer.
Selection not documentedThe data set holds 86 sessions, including 23 of rat 20 that fail the 4 min SWS rule, and does not say which sessions the paper analysed.
Description missingThe bounds of the pre and post rest epochs are not given; only the awake interval and the SWS intervals are. Position units and the maze's coordinates are not stated.
OtherTwo of the ten learning sessions the description lists (201227 and 201229) have less than 4 min of SWS in one rest epoch by the data set's SWS intervals (2.2 min pre, and 3.3 min post), so they would fail the paper's inclusion rule, though the paper's Figure 7 uses ten learning sessions.
OtherSessions without SWS in a rest epoch store a single [0, 0] interval (four sessions) or an empty file (150704 SwsPRE), not a documented empty value.
OtherCell types are all 0 (unknown) in 200107 and all 1 (interneuron) in 200209, and mostly unknown in 200131, 200224 and 200301, which is unlikely for prefrontal recordings.
OtherIn five sessions spikes continue 12 to 176 min after the position record ends (200106 by 176 min), and in 200213 the position record ends before the awake interval does.
OtherIn the recorded trials rat 18 never reaches the rule-change criterion on the right-arm rule, yet its rule was changed, and three rats, not two, reach criterion on the left-arm rule.
Methods not availableFor the authors. Is the match epoch z-transformed over the whole rest epoch or over its SWS bins alone? The Methods say "For all epochs, spike trains ... were z transformed", and analyses are restricted to SWS. Both were run here and no outcome changes, but the values differ (for example r squared 0.384 against 0.417 for Figure 2b).
Methods not availableFor the authors. How were the pre and post rest epochs bounded in the deposited sessions, and how were DOWN states detected for Figure 6b (beyond the 80 ms minimum) and the pre epoch's smaller modulation judged? Here DOWN states are 80 ms without spikes, and the pre modulation is compared over the window where post is significant; compared by the largest value, pre is not smaller.
Methods not availableFor the authors. The tests behind Figure 2b, Figure 2e and Supplementary Figure 11D are not named, and the r squared of 0.61 for Figure 2b is near the r found here (0.62, r squared 0.38). Which tests were used, and is 0.61 an r or an r squared?

Notebooks

Every registered entry that uses this dataset. An entry can hold several notebooks, and can use more than one dataset.

Reproduction Replay of rule-learning related patterns in rat prefrontal cortex during sleep (Peyrache et al. 2009)

Tests the findings of Peyrache, Khamassi, Benchenane, Wiener and Battaglia (2009) on CRCNS pfc-6, which holds the prefrontal spike trains, behaviour, position and the awake and SWS intervals of 86 sessions from four rats. It counts sessions, rules and signal components, and tests reactivation of awake patterns in post against pre SWS, its relation to encoding strength, its tail, the shuffle control, cross-correlations between patterns, reactivation around DOWN to UP transitions, and behaviour around learning. The LFP, the SPWR, delta and spindle times and the tetrode of each cell are not in the data set. All files

Reproducing Peyrache et al. (2009) from CRCNS pfc-6
reproduce_peyrache_2009.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-11-05
Agreed with 26 of 29 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The handover of the first run, the executed notebook with the conclusions and summary removed, the helper module and the claims file, with the paper under paper/.
- supp5-sws-only: the verifier’s outcome was “Not reproduced”. The state main effect has p = 0.0138, above the required 0.01, and in non-SWS bins post exceeds pre significantly (paired t p = 0.042, Wilcoxon p = 0.018); the 30 min variant gives the same result. The ordinary two-way ANOVA on four non-independent means per component could be replaced by a repeated-measures form, which could bring the state effect below 0.01, but the non-SWS result still fails the claim.
- supp6-pre-post-sws-alike: the verifier’s outcome was “Partially reproduced”. No difference by paired t-test in total SWS (p = 0.87), episode duration (0.41), session-mean rate (0.93) or cell rate (0.65). SPWR rate, delta waves and LFP power are blocked (data_missing).
- fig6b-down-up: the verifier’s outcome was “Partially reproduced”. Post signal components have significant bins at -0.5 to -0.3 s, nonsignal components have none, and pre is lower in that window (0.063 against 0.107). Pre's peak is higher than post's (0.145 against 0.132), and the per-component post-minus-pre difference is not significant (p = 0.15). With 10 ms bins the mean count per bin is below 5, so reading (b) is the same as (a), and the notebook does not say so. It asked for the lag of the one significant pre bin and the full pre profile.
Round 2
2026-11-05Agreed with 29 of 29 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The full executed notebook of the first revision, the helper module, the claims file and the attempt record, with the paper under paper/.

Method differences

How the tests differed from the paper’s method or data, and the claims each applies to.

SWS time per rest epoch is the summed length of the data set's SWS intervals; the paper's own SWS detection, from the LFP, cannot be repeated.sessions-and-cells
Sessions are ordered by the month and day in their ids, with rat 20's December sessions put before its January to March ones; the data set gives no year.rules-learned
The learning point is computed over all trials of the session, as the definition says, even where the rule changed within the session.learning-sessions
All 86 sessions are searched, not only those passing the inclusion rule; the table shows which of the listed sessions are included.learning-sessions
The criterion takes the session selection from the matching reading of sessions-and-cells; neither matches, so the paper's stated inclusion rule, reading (b), is used.fig1b-signal-components
The awake epoch is the data set's single awake interval, binned whole in 100 ms bins.fig1b-signal-components
The outcome follows the criterion, which tests post against pre; two of the three group sizes listed as expected values fall outside tolerance, because the data set holds fewer sessions that pass the inclusion rule.fig2b-post-greater-than-pre
The data set does not give the bounds of the rest epochs. Assumed, as the claim's criterion directs, from the data set's epoch times, pre from the start of the recording to the start of the awake interval and post from its end to the end of the position record, or to the end of the last post SWS interval if that is later; this makes pre 23 to 58 min and post 22 to 59 min in the included sessions, where the Methods give 20 to 30 min. A 30 min variant is shown.supp5-sws-only
Non-SWS bins are z-transformed with the whole rest epoch.supp5-sws-only
The ANOVA is a repeated-measures ANOVA with the signal component as subject, since the four means of a component are not independent; the legend names only an ANOVA. The ordinary two-way ANOVA is shown.supp5-sws-only
SWS episodes are the data set's SWS intervals; their merging rule (gaps under 1 min) is the paper's, applied before deposit.supp6-pre-post-sws-alike
The bootstrap p-value uses a normal approximation from the bootstrap standard error, since 2000 resamples cannot resolve p below 5e-4; no resample gave a slope difference of zero or less.fig2b-pre-weaker
Cross-correlations are computed within each SWS interval and pooled, so that no pair of bins spans a gap between intervals; lags of -1 to 1 s (a standard choice).supp10-no-coactivation
The mean over shuffles is computed in closed form, as the mean over all permutations of the cells within each bin, instead of 1000 random shuffles per bin; on one session it agrees with 1000 shuffles (-0.0082 against -0.0086).supp11-shuffle-control
Assumed that "the pre average is lower" holds only if pre is lower both by the rise above baseline in the window where post is significant (-0.5 to -0.3 s) and by the largest value of the whole average, or if the per-component difference is significant. Pre is lower in the window (0.063 against 0.107) but its largest value is higher (0.145 against 0.132), and the per-component difference is not significant (p = 0.15), so this part is not established. Comparing by the window alone would make the claim reproduced. Both comparisons are shown.fig6b-down-up
Event-triggered averages use R over all bins of the rest epoch with Q z-transformed over the whole epoch, since the -4 to 4 s windows reach outside SWS; transitions are taken inside SWS.fig6b-down-up
The learning sessions and trials are those listed in the data set's description, since the paper's definition does not recover them (learning-sessions).supp12-behavior-unchanged
Speed is shown in ten bins of the fraction of the trial's path length, as the data set holds no maze geometry for the paper's linearization; it is reported, and does not decide the outcome.supp12-behavior-unchanged

Provenance

Attempt record: crcns-pfc-6-peyrache-2009-01
Paper: 10.1038/nn.2337
Date: 2026-11-05
Made by: An agent, run by CatalystNeuro
Dataset: CRCNS:pfc-6
Version: unversioned (doi:10.6080/K0KH0KH5)
Run on: GitHub Actions
Harness: Claude Code
Model: claude-opus-5-5
Protocol: pilot-13

Transcripts of the agent runs

Each run’s prompt, messages and tool calls, with long output cut and the text of the paper left out, and its final report.

← CRCNS:ofc-4 CRCNS:ret-1 →