Routing of Hippocampal Ripples to Subcortical Structures via the Lateral Septum (DANDI:000218)

Claims

From the paper Routing of Hippocampal Ripples to Subcortical Structures via the Lateral Septum

Reproducing Tingley and Buzsaki (2019) from dandiset 000218 21

Notebook reproduce_tingley_2019.ipynb

Not reproducedResults, first paragraph; Figure 1A to C
Transient 120-180 Hz HFOs occur in LS in each of the 6 animals, with a robust increase in the firing probability of LS neurons at the events.
we observed transient high frequency oscillations (HFOs) in the 120–180 Hz frequency range in each animal (Figure 1A, B) and an associated robust increase in the firing probability of LS neurons (Figure 1C)
Results, first paragraph; Figure 1A to C
Median event frequency is 152 to 158 Hz in every animal. LS firing within 10 ms of the events exceeds firing at random times by more than 3 SEM only in DT1 (6.0 SEM); DT2 and DT5 are higher by 1.3 and 1.4 SEM, and DT7, DT8 and DT9 are lower (-3.4, -4.6, -1.4 SEM). In variants, NREM events pass in DT2 (3.3) and DT9 (3.9), and ripple-coupled events in DT9 (4.2). On the LS channel re-referenced within its shank no animal passes (largest 1.5 SEM, DT2). Random times mix states; against each neuron's firing 50 to 100 ms before and after the same events, firing at the events is higher in DT1, DT2, DT5 and DT9 and lower in DT7 and DT8, and exceeds 3 SEM only in DT1 (5.2 SEM; DT2 1.6, DT5 1.7, DT9 0.5). In DT2 and DT5 the differences are large (0.024 and 0.037 spikes) but vary widely across neurons. The test does not show raised LS firing at the events outside DT1, and does not show that the events lack it.
How the test differed from the paper’s method
- Events are detected here on one LS channel per recording chosen by a fixed rule (highest 120-200 Hz SD in a 300 s NREM sample); the paper does not say which channels it used.
- The paper's noise channel is replaced by the EMG estimated from the LFP, which the Methods allow when no noise channel exists.
Not reproducedResults, first paragraph (Figure S1A, B)
On average 16% of LS neurons take part in a given HFO, firing an average of 0.23 spikes per event.
On average, 16% of LS neurons participated in any given event, firing an average of 0.23 spikes per event (Figure S1A, B)
Results, first paragraph (Figure S1A, B)
Pooled over 31,984 events, 2.7% of LS neurons fire in an HFO, with 0.051 spikes per neuron per event; the mean over 76 recordings gives 3.5% and 0.068. The paper reports 16% and 0.23. The fraction is 0.17 (pooled) to 0.22 (recordings) of the paper's, the spike count 0.22 to 0.30. The highest recording reaches 18.2%; the median recording 1.7%. On the re-referenced LS channel the pooled values are 4.5% and 0.060.
How the test differed from the paper’s method
- Spikes are counted between event start and stop; the paper does not give its window.
Not reproducedResults, first paragraph; Figure 1G
LS HFOs are slower in frequency than CA1 ripples (147 +- 5.7 against 159 +- 8 Hz; rank-sum P < 0.01).
These events were significantly ... slower in frequency (147±5.7 vs 159±8 Hz) than CA1 ripples recorded from the same animals (Figure 1F, G; rank-sum test; P < 0.01)
Results, first paragraph; Figure 1G
What the paper reports
- rank-sum test, p < 0.01
LS HFOs average 153.6 +- 8.3 Hz and CA1 ripples 150.8 +- 9.2 Hz; LS events are faster, the reverse of the claim, in four of five animals, and both means are outside tolerance. With NREM events only, LS are slower by 0.8 Hz (150.3 against 151.1 Hz). On the re-referenced LS channel LS events are faster (156.3 against 150.8 Hz).
How the test differed from the paper’s method
- The instantaneous-frequency estimator pulls towards the band centre (the verifier measured 145 Hz for a 140 Hz burst and 167 Hz for 170 Hz); this compresses differences but cannot reverse their order.
Not reproducedResults, first paragraph; Figure 1H
LS HFOs show no significant correlation between event amplitude and peak frequency.
However, they showed no significant correlation between event amplitude and peak frequency (Figure 1H)
Results, first paragraph; Figure 1H
Over 67,186 events, Pearson r = 0.056 (p = 2.4e-47), Spearman rho = 0.123. By animal r is -0.009 to 0.080, significant in four of six; NREM events only give r = 0.028 (p = 7.3e-05). The outcome follows the criterion's test; the correlation is negligible in size (0.3% of the variance) and significant because of the event count.
How the test differed from the paper’s method
- The paper names no test; a Pearson test over all events was fixed in the criterion.
Not reproducedResults, second paragraph; Figure 2A to C
HFO events are seen only at recording sites within LS, not at sites in neighbouring structures.
We only observed HFO events at recording sites that were confined to the LS (Figure 2A–C)
Results, second paragraph; Figure 2A to C
Of 5 DT7 recordings with at least 20 HFOs in the 30 min block, LS channels have the highest HFO-triggered power ratio in 2 (2017-03-18, 5.38 against 5.31 for cingulate cortex; 2017-03-21, 8.25 against 8.05 for medial septum). Medial septum (6.05 against 5.18), corpus callosum (50.1 against 42.0) and fimbria (72.1 against 64.8) channels are higher in the other three. The test rests on one animal, DT7, the only one whose files label electrodes outside LS and the hippocampus. With each channel re-referenced to the median of its shank and the re-referenced LS events as triggers, 4 recordings have at least 20 events and LS is highest in 2; medial septum channels are higher in the other 2 (4.00 against 3.09, 4.21 against 3.65).
How the test differed from the paper’s method
- Tested on the DT7 recordings with channels in other regions, over up to 30 min of all channels, in place of the paper's depth series of Figure 2C.
Not reproducedResults, fourth paragraph; Figure 4D; Figure S3B
Cross-correlograms of LS pairs during HFOs correlate with those during traversal above chance up to about 20-fold temporal compression (P < 0.01), similarly for PRE and POST (P > 0.01 for all compression factors).
We found correlations significantly higher than chance (up to ~20-fold compression; P < 0.01) ... However, the strength of this compression was similar for HFOs recorded prior to, and after, maze traversal (P > 0.01 for all compression factors)
Results, fourth paragraph; Figure 4D; Figure S3B
What the paper reports
- HFO against maze traversal CCG correlations above chance, up to about 20-fold compression: p < 0.01
- PRE against POST, every compression factor: p > 0.01
PRE correlations exceed chance (p < 0.01) at factors 1 and 2 only, and POST at 1, 2, 4 and 20, against up to about 20 in the paper. PRE and POST do not differ at any factor (smallest p = 0.038), as the paper reports. Pairs with a defined correlation fall from 59 (PRE) and 77 (POST) at factor 1 to 5 and 11 at factor 50, so the higher factors rest on few pairs; PRE at factor 4 rests on 24 pairs (p = 0.18).
How the test differed from the paper’s method
- Interpolation and chance by circular shift were fixed in the criterion; the Methods describe the warping but not the chance procedure in detail. One shift per pair is used for all factors.
Not reproducedResults, sixth paragraph; Figure 5Dverifier disagreed at first
LS HFOs are reliably coupled to CA1 ripples, with a median peak-to-peak lag of 6.4 +- 4.2 ms, CA1 leading LS.
Across recordings, we observed a reliable coupling of events across structures (Figure 5D) with a median peak-to-peak temporal lag of 6.4 ± 4.2 milliseconds (CA1 leading LS)
Results, sixth paragraph; Figure 5D
The coupling holds; the pooled CA1-LS correlogram peaks at 0 ms, 166 SD above its flanks. The lag is 0.00 ms under both readings, outside 6.4 +- 4.2 ms, because 35% of LS HFOs within 50 ms of a ripple fall on the same sample (DT8 71%, DT2 62%, DT9 7%). Without the pairs within 1 ms the pooled median is 3.2 ms with LS after CA1 in 55%, but by animal it is -3.2 (DT2), -5.6 (DT5), -6.4 (DT7), 1.6 (DT8) and 8.0 ms (DT9), so LS leads in three animals. Over all pairs, DT9's median of recording medians is 3.8 ms, inside the tolerance (2.2 to 10.6 ms); the other animals give 0.0 ms. On the LS channel re-referenced within its shank, 4.1% of pairs fall on the same sample, the correlogram still peaks at 0 ms (53.4 SD above the flanks), and the lag is 0.00 ms (a) and 0.80 ms (b), LS after CA1 in 52%. The criterion requires both the lag and the correlogram peak and defines no partial outcome, so the claim is not reproduced; the coupling part holds and the lag part does not. Changed from partially reproduced after round 3.
How the test differed from the paper’s method
- Detection as in fig1c-hfo-each-animal; event times are the troughs of the filtered signal.
Not reproducedResults, sixth paragraph; Figure 5E and inset
HPC-LS coupling is stronger in NREM sleep than in waking; the fraction of LS HFOs within +-25 ms of a CA1 ripple is 0.3 in NREM and 0.15 in WAKE (P < 6.6e-6).
HPC-LS coupling was significantly stronger during NREM sleep than during the waking immobile state ... the percentage of LS HFO events that occurred within ±25 milliseconds of a CA1 ripple (NREM = 0.3; WAKE = 0.15; P < 6.6e-6 two-way T-test; Figure 5E inset)
Results, sixth paragraph; Figure 5E and inset
What the paper reports
- CA1 ripple and LS HFO coupling, NREM against WAKE: two-way T-test, p < 0.001, Bonferroni-corrected
- fraction of LS HFOs within ±25 ms of a CA1 ripple, NREM against WAKE: two-way T-test, p < 6.6 × 10⁻⁶
Over 25 recordings with at least 10 HFOs in each state, 0.199 of LS HFOs have a CA1 ripple within 25 ms in NREM and 0.116 in WAKE (t-test p = 0.089); pooled, 0.088 and 0.041. The direction matches, the difference is not significant, and both are below the paper's values. With ripple times shifted by 1 s the fractions are 0.019 and 0.005. On the re-referenced LS channel, over 16 recordings, 0.118 and 0.085 (p = 0.43).
How the test differed from the paper’s method
- Detection as in fig1c-hfo-each-animal.
Not reproducedResults, sixth paragraph; Figure 5F
LS firing around CA1 ripples is higher during NREM than during waking.
This state-dependent coupling could also be observed ... in the SPW-R triggered LS firing rates (two-way T-test with Bonferroni correction; P < 0.001; Figure 5F)
Results, sixth paragraph; Figure 5F
What the paper reports
- two-way T-test, p < 0.001, Bonferroni-corrected (n = 1,313 neurons)
Over 978 LS neurons (paper 1,313), the rate within 25 ms of CA1 ripples is 1.39 Hz in NREM and 1.33 Hz in WAKE (paired t-test p = 0.24). Relative to each neuron's rate in the state, WAKE is higher (1.56 against 1.34, p = 0.00016). By animal, DT5, DT7 and DT9 are higher in NREM and DT2 is lower.
How the test differed from the paper’s method
- A window of +-25 ms and a single test over neurons were fixed in the criterion; the figure tests each bin of the correlogram.
Not reproducedResults, seventh paragraph; Figure 5F inset
CA1-LS phase coherence in the 120-180 Hz band around CA1 ripples does not differ from that of randomly jittered windows, so LS does not inherit the ripple wave by wave.
phase coherence between HFOs and SPW-Rs was not significantly different from a shuffled distribution (Figure 5F inset)
Results, seventh paragraph; Figure 5F inset
The CA1-LS 120-180 Hz phase-locking value at ripple times is 0.405 against 0.202 +- 0.007 for jittered windows (31 SD above), higher in 70 of 71 recordings, and 90 of 125 lags within 50 ms exceed the bound. The jittered value is itself high. With a median of 1722 ripples per recording random phases would give about 0.021, and the jittered value per recording is 0.030 to 0.602 (median 0.134), so the two channels keep a phase relation outside ripples. On the LS channel re-referenced within its shank, the value at ripples is 0.164 against 0.074 +- 0.004 jittered (23.9 SD above, higher in 60 of 71 recordings), so the excess remains.
How the test differed from the paper’s method
- The coherence measure (phase-locking value of the phase difference across ripples) was fixed in the criterion; the paper does not define it.
- The LS channel is the detection channel; the paper does not say which channel it used.
Partially reproducedResults, first paragraph; Figure 1C legend; Figure 5B legend; STAR Methods, Experimental model and subject details
LS was recorded in 6 rats, with 1989 LS neurons; 37 recordings had CA1 and LS and 25 had CA3 and LS recorded simultaneously; LS HFO rates come from 89 recordings and CA1 ripple rates from 53.
During initial recordings in LS (N = 6 rats) ... The dataset used for Figure 6 consists of 62 recordings with both CA1 and LS simultaneously recorded (N = 37) or CA3 and LS simultaneously recorded (N = 25)
Results, first paragraph; Figure 1C legend; Figure 5B legend; STAR Methods, Experimental model and subject details
The animal count matches (6). LS neurons number 2448 labelled `ls`, and 2046 with more than 50 spikes in high-power epochs, within 10% of 1989. Files with LS and `hpc` units number 37, as in the paper; files with LS electrodes 94 (paper 89). Files with LS and CA3 units number 10 against 25, and the CA1 recording count is 40 (units), 72 (electrodes) or 43 (electrodes and states) against 53. Three of five counts beyond the animals are within tolerance.
How the test differed from the paper’s method
- The paper's recordings cannot be matched to files one by one; counts are compared under the readings the criterion lists.
Partially reproducedResults, first paragraph; Figure 1D, E; STAR Methods, Phase locking analysis
The vast majority of LS neurons (1840 of 1989, 92%) are significantly phase-locked to the 120-200 Hz band (Rayleigh P < 0.05), with a strong preference for the trough of each cycle.
the vast majority of LS neurons (1840 of 1989; 92%) were significantly phase-locked to the 120–200 Hz frequency band, with a strong unimodal phase preference for the trough of each cycle (Figure 1D, E)
Results, first paragraph; Figure 1D, E; STAR Methods, Phase locking analysis
1151 of 2046 tested LS neurons (56.3%) are significantly locked (47.0% of all 2448), against 92.5%. The trough preference holds, with preferred phases centred on 189 degrees (resultant 0.755). By animal the fraction is 5.7% (DT7) to 70.4% (DT5). Median resultant 0.124 against 0.048 with the phase shifted. With the phase taken on an LS channel on another shank than the unit's, 1048 of 2003 tested neurons (52.3%) are locked, centred on 189.7 degrees (resultant 0.743), and 44 have a resultant above 0.6, so the trough preference does not come from the units' own spikes. Locking is stronger for units on the detection shank (76.4%) than on other shanks (51.4%).
How the test differed from the paper’s method
- The phase is taken on the recording's detection channel, not on a channel near each unit; the paper does not say which channel it used.
Partially reproducedResults, fifth paragraph (Figure S2A, B)
Firing rates increase during post-behavior NREM sleep in CA1, CA3 and LS.
we observed a systematic increase in firing rates during post-behavior NREM sleep in CA1, as well as CA3 and LS populations
Results, fifth paragraph (Figure S2A, B)
LS units fire more in POST NREM than in PRE NREM (median ratio 1.06, Wilcoxon p = 8.4e-07, 1326 units). The `hpc` units (ratio 1.00, p = 0.52, 2552 units) and the `ca3` units (1.03, p = 0.22, 466 units) do not.
How the test differed from the paper’s method
- The paper's supplementary figure was not available; the test was fixed in the criterion.
Partially reproducedResults, sixth paragraph; Figure 5A, B
LS HFOs occur primarily during NREM sleep, and their rates are lower than those of hippocampal SPW-Rs.
we primarily observed HFO events during NREM sleep (Figure 5A) ... the overall rates of LS HFOs were significantly lower than hippocampal SPW-Rs (Figure 5B)
Results, sixth paragraph; Figure 5A, B
LS HFO rates in NREM are far below CA1 ripple rates (medians 0.006 and 0.461 per s, lower in 34 of 39 recordings, Wilcoxon p = 3.1e-06). The detected LS events are not more frequent in NREM than in WAKE (NREM higher in 18 of 49 recordings, lower in 29, p = 0.44), while CA1 ripples are (about 9 times). On the re-referenced LS channel the same holds (NREM higher in 17, lower in 31, p = 0.73; below CA1 in 35 of 39, p = 6.7e-06).
How the test differed from the paper’s method
- Detection as in fig1c-hfo-each-animal; 4 recordings without NREM or WAKE time are left out.
Partially reproducedResults, eighth paragraph; Figure 7D, E
The number of CA1 spikes in a SPW-R significantly predicts the number of spikes of 38% of LS neurons.
For many LS neurons (38%), this measure significantly predicted the number of action potentials elicited in the postsynaptic LS neuron (Figure 7D, E)
Results, eighth paragraph; Figure 7D, E
The number of CA1 spikes predicts 79.4% of 880 LS neurons under reading (a) (chance 4.7%) and 88.4% under (b) (chance 40.5%); above chance and far above 38%. Counting from ripple start to stop makes the LS and CA1 counts both grow with duration; with a fixed +-25 ms count window, 24.7% under (a) (chance 5.6%). Reading (b) is miscalibrated by construction (chance about 40%) and is not used.
How the test differed from the paper’s method
- The paper does not state its model; a linear regression with cross-validated MSE and shuffles was fixed in the criterion. Under reading (b), the paper's t-test wording, the chance rate is about 40%.
- The `hpc` units are taken as CA1.
Partially reproducedResults, eighth paragraph; Figure 7E
Brain state (WAKE or NREM) in which a SPW-R occurs significantly predicts the spike count of 26% of LS neurons.
the rate of spiking for some LS neurons (26%) was significantly predicted by brain state (Figure 7E)
Results, eighth paragraph; Figure 7E
State predicts 34.3% of 813 LS neurons under (a) (chance 5.4%) and 67.4% under (b) (chance 41.7%), above chance and above 26 +- 2.6%. With a fixed window, 34.9% under (a) (chance 6.0%).
How the test differed from the paper’s method
- As for fig7e-ca1-spikes.
Partially reproducedResults, eighth paragraph; Figure 7E
The duration of hippocampal SPW-Rs significantly predicts the spike count of about 20% of LS neurons.
The duration of hippocampal events could also be used to significantly predict firing rates for ~20% of the LS neurons (Figure 7E)
Results, eighth paragraph; Figure 7E
Duration predicts 86.0% of 880 LS neurons under (a) (chance 5.9%) and 93.2% under (b), far above about 20%. Counting spikes from ripple start to stop makes the count grow with duration; with a fixed +-25 ms window it is 17.8% under (a) (chance 7.2%), just below the tolerance of 18 to 22%, shown as a variant.
How the test differed from the paper’s method
- As for fig7e-ca1-spikes; the count window (ripple start to stop) was fixed in the criterion.
ReproducedFigure 1C legend
Across 6 animals, 29,955 LS HFO events were analysed.
Average power spectrum across all events (N = 6 animals, 29,955 events)
Figure 1C legend
Under reading (b), recordings with LS units, 31,984 events are detected, within 10% of 29,955. Reading (a), all 94 recordings with LS electrodes, gives 67,186, of which 38,008 come from DT7. More events fall in WAKE (27,676) than in NREM (20,019). The count matching does not show that the same events were found, and is probably a coincidence, as DT7 contributes most events and the count depends on the channel; on the re-referenced LS channel the recordings with LS units give 66,458 events.
How the test differed from the paper’s method
- Detection channels and noise exclusion differ from the paper's (see fig1c-hfo-each-animal).
ReproducedResults, first paragraph; Figure 1F
LS HFOs are shorter than CA1 ripples from the same animals (27 +- 9 against 31 +- 14 ms; rank-sum P < 0.01).
These events were significantly shorter in duration (27±9 vs 31±14 milliseconds) ... than CA1 ripples recorded from the same animals (Figure 1F, G; rank-sum test; P < 0.01)
Results, first paragraph; Figure 1F
What the paper reports
- rank-sum test, p < 0.01
LS HFOs last 33.9 +- 25.0 ms and CA1 ripples 38.3 +- 23.7 ms (medians 23.2 and 32.0), inside the tolerances of the paper's SDs, and LS HFOs are shorter (rank-sum p underflows to 0). The means are 6.9 (LS) and 7.3 ms (CA1) above the paper's. DT8 shows no difference (27.5 against 28.0 ms, p = 0.83). With NREM events only, the means are 38.0 and 38.4 ms, and on the re-referenced LS channel 38.2 and 38.3 ms. With 56,871 and 230,161 events the rank-sum p is small for any difference.
How the test differed from the paper’s method
- The tolerances are the paper's stated SDs, as the protocol asks, which makes them wide.
- The `hpc` channel is taken as CA1.
ReproducedResults, third paragraph; Figure 4B
The co-activation of LS neuron pairs during HFOs (excess correlation) is not modulated by waking experience on a familiar maze (PRE against POST, rank-sum P = 0.36).
It also suggested that the degree of co-activation observed is not modulated by waking experience in a familiar environment (Figure 4B; rank-sum test; P = 0.36; PRE vs POST)
Results, third paragraph; Figure 4B
What the paper reports
- rank-sum test, p = 0.36 (n = 7,172 pairs of LS neurons, 53 sessions)
PRE and POST excess correlations do not differ (medians 0.0062 and 0.0059, rank-sum p = 0.81; within animals p = 0.49 to 0.88). Only 106 pairs in 10 recordings pass the inclusion rule, against 7,172 in 53 sessions, so the test has much less power than the paper's.
How the test differed from the paper’s method
- The CCG normalization, the smoothing at the window edges and the reading of the 5-spike inclusion rule were fixed in the criterion; the Methods do not give them.
- HFO spikes are those inside this attempt's detected events, which hold few LS spikes.
ReproducedResults, third paragraph; Figure 4B; Figure S3A
Excess correlations during HFOs (PRE and POST) are higher than during maze traversal (rank-sum P < 0.001), and those of PRE and POST HFOs exceed chance from spikes jittered by +-200 ms (P < 0.01).
the excess correlations for HFOs during the PRE and POST epochs were much greater than expected by chance ... during LS HFOs there is a higher degree of co-activation than observed during behavior or expected given LS firing rates (rank-sum test; P < 0.001 for PRE vs BEHAV and BEHAV vs POST)
Results, third paragraph; Figure 4B; Figure S3A
What the paper reports
- PRE and POST HFO excess correlations against spikes jittered by ±200 ms: rank-sum test, p < 0.01
- PRE against BEHAV and BEHAV against POST: rank-sum test, p < 0.001
HFO excess correlations exceed BEHAV (median 0.0003; PRE p = 9.1e-24, POST p = 3.8e-25) and their jittered values (p = 3.6e-31 and 7.8e-35); BEHAV is slightly above its jitter (p = 0.0022). 106 pairs in 10 recordings.
How the test differed from the paper’s method
- As for fig4b-pre-post-unchanged.

Not tested in a notebook 10

2 metadata insufficient, 3 other, 2 data missing from the dataset, 3 methods underspecified

BlockedResults, second paragraph; Figure 2D to F; Figure S1C, D· Metadata insufficient
HFO power and the proportion of phase-locked neurons are highest in intermediate LS and taper towards its dorsal and ventral borders.
Both the power of HFOs and the proportion of phase-locked neurons were highest in the intermediate LS and tapered off as recordings approached either the dorsal or ventral border of LS
Results, second paragraph; Figure 2D to F; Figure S1C, D
The claim needs each recording's depth relative to the dorsal and ventral borders of LS, which the paper estimated from implant depths, drive turns and histology. The files give only the drive depths in the session names and a region label per channel; the borders, the histology and the relative depths are not in the dandiset, and the names do not say which of their two depths is the LS drive. For the record, the notebook shows the HFO band SD and the locked fraction against each depth within each animal (rank correlations -0.61 to 0.41 and -0.84 to 0.36); the band SD is largest at the shallowest or deepest recording in 9 of 11 series.
BlockedResults, second paragraph; Figure 3C· Metadata insufficient
LS HFOs are travelling waves, posterior to anterior at 1.01 m/s (4 animals), medial to lateral at 0.53 m/s (1 animal) and dorsal to ventral at 2.19 m/s (4 animals).
we estimated that the travel speed was 1.01 meters per second in the posterior-to-anterior direction (N = 4 animals), 0.53 meters per second in the medial-to-lateral direction (N = 1 animal), and 2.19 meters per second in the dorsal-to-ventral direction (N = 4 animals)
Results, second paragraph; Figure 3C
The speeds need the positions of the electrodes along the anterior-posterior, medial- lateral and dorsal-ventral axes. The electrode x, y and z are NaN in every file, and the probe layouts are named only in the device descriptions, without their orientation in the brain. The electrode groups are described only as "GroupN electrodes.", and the devices give the layout (shanks by sites) without site pitch or orientation.
BlockedResults, third paragraph· Other
Only 7 of 53 sessions have significantly different excess correlations in PRE and POST (rank-sum P < 0.01), three with PRE below POST and four with PRE above.
Only 7 out of 53 sessions had significantly different excess-correlation values for PRE and POST epochs (rank-sum test; P < 0.01), three where PRE was less than POST and four where PRE was greater than POST
Results, third paragraph
Too few recordings under the claim's own rule; decided after the pair selection. The shortfall comes from this attempt's HFO detection, whose events hold few LS spikes, not from the dandiset, which holds the spikes, epochs and trials. 7 recordings have at least 5 pairs, against 53 sessions in the paper; at the paper's rate of 7 in 53 about one significant recording would be expected, so the count cannot test the claim. For the record, none of the 7 has p < 0.01 (smallest p = 0.050).
BlockedResults, fourth paragraph; Figure 4C· Other
The temporal ordering of LS spike pairs during HFOs before and after the maze correlates with the ordering during traversal to a similar degree (P = 0.85).
the spiking structure during HFOs prior to and after maze performance showed correlations with the spike ordering during maze traversal that were similar (Figure 4C; N = 53 recordings; Two-way T-test; P = 0.85)
Results, fourth paragraph; Figure 4C
What the paper reports
- Two-way T-test, p = 0.85 (n = 53 recordings)
Too few recordings under the claim's own rule; decided after the pair selection, for the reason given under fig4-sessions-few-change. 4 recordings have at least 10 pairs, against 53 in the paper, too few for a t-test over recordings. For the record, the mean correlations are 0.042 (PRE-BEHAV) and 0.084 (BEHAV-POST), t-test p = 0.81.
BlockedResults, seventh paragraph; Figure 6· Data missing from the dataset
Optogenetic activation of LS GABAergic neurons with a Gaussian light waveform induces local oscillations of about 140 Hz.
Using a Gaussian waveform of blue light (450 nm), mimicking the envelope of SPW-R-induced hippocampal excitation, local oscillations of ~140 Hz could be reliably induced (Figure 6C–E; S2)
Results, seventh paragraph; Figure 6
The two rats used for optogenetics are not in the dandiset, whose 6 subjects are the recording rats, and no file holds an optogenetic series (notebook, fig6-optogenetic-hfos cell). Archive searches for "Tingley" and "lateral septum" find only 000213, 000218 and 000233; 000213 and 000233 are named for other papers.
BlockedResults, eighth paragraph; Figure 7D, E; STAR Methods, Relative anatomical depth within CA1· Data missing from the dataset
The mean depth within the CA1 pyramidal layer of the neurons active in a SPW-R predicts the spike count of 23% of LS neurons.
For a set of LS neurons (23%), the recording depth of the active CA1 population correlated significantly with the number of action potentials elicited (Figure 7D, E)
Results, eighth paragraph; Figure 7D, E; STAR Methods, Relative anatomical depth within CA1
The depth of a CA1 neuron is estimated from its waveform amplitude on the three channels where it is largest. The units carry no waveforms and no channel, only a shank; the units table holds clu_id, group_id (the shank) and location, or only ids in nine files. The notebook's fig7e-ca1-depth cell lists the wide-band data. 20 kHz wide-band series are in 27 of the 98 files (all 12 DT1 files, all 13 DT2 files and two DT9 files, which take 1.15 TB). Twelve of them (10 DT2, 2 DT9) hold hippocampal units, 1.00 TB of wide-band data, from which waveforms could be re-extracted; streaming them within the 30 min run limit would need about 556 MB/s, so for these recordings the block is the run time (not_runnable). The other 1579 of 2778 hippocampal units, including all of DT5 and DT7, have no wide-band data, so their waveforms are missing, which is the main reason given.
BlockedResults, ninth paragraph; Figure 7D, E· Methods underspecified
The sequential content of SPW-Rs, by three replay scores, predicts the firing of only 7.9, 6.6 and 7.1% of LS neurons.
The firing rates of only a minority of LS neurons—7.9, 6.6, and 7.1% respectively—could be significantly predicted by the sequential structure of hippocampal events captured by these methods (Figure 7D, E)
Results, ninth paragraph; Figure 7D, E
The three replay scores need place-field decoding of the CA1 units on the tracks (bin size, smoothing, trial selection, cell inclusion, and the shuffles for the scores) and the model that relates the scores to LS firing. Neither the paper nor any published code fixes them; the paper's repository holds no code for this figure. The files do hold positions and trials; 74 files in DT2, DT5, DT7, DT8 and DT9 have a position series, covering all trials except in DT7 (93%).
BlockedResults, tenth paragraph; Figure 7E; Figure S5· Methods underspecified
Template sequences extracted from SPW-Rs without supervision predict the firing of only 8.1% of LS neurons.
we found that the percentage of LS neurons whose action potential firing could be significantly predicted with these templates was small (Figure 7E; 8.1%; P < 0.05)
Results, tenth paragraph; Figure 7E; Figure S5
The seqNMF decomposition is described only by its reference and the 40 templates; its lag, regularization, binning and the test against LS firing are not given, and no code for this paper is published.
BlockedResults, eleventh paragraph; Figure 7E, bottom row· Methods underspecified
All spike-content-independent ripple features predict LS firing above chance, while all spike-content-based features do not.
all spike content-independent ripple features were significantly different from chance (left seven columns), while all spike content-based ripple features were not significantly different from chance (Figure 7E; right four columns; bottom row)
Results, eleventh paragraph; Figure 7E, bottom row
Depends on fig7e-replay, fig7e-seqnmf (methods underspecified) and fig7e-ca1-depth (data missing). The three content-independent features tested here are all above chance (see their outcomes).
BlockedResults, twelfth paragraph; Figure 8C, D· Other
In a model HPC-LS circuit, forward, reverse and scrambled compressed sequences evoke nearly identical responses of the LS neuron.
By providing the model with temporally compressed forward, reverse, or scrambled replay events (Figure 8C), we observed that the response of the postsynaptic model LS neuron was nearly identical for all three event types (Figure 8D)
Results, twelfth paragraph; Figure 8C, D
A simulation of a model circuit, not a finding about the recorded data, so the dandiset cannot test it. The model's code is in https://github.com/DavidTingley/papers (notebook dynamicWeightingModel-ripple.ipynb) and was not run here.

Findings about the data

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

Data missingThe files hold no detected LS HFOs or hippocampal ripples, nor the detection channels and parameters, which the Methods say are stored with each event structure. Every analysis of events has to detect them anew, and the detection channels cannot be matched to the paper's.
Description missingNine files (DT5 2016-10-07, DT7 2017-03-14, -16, -17, -19, 2017-04-07, -10, DT9 2017-04-28, 2017-05-26) store units without a location column, so their 268 units cannot be assigned to LS or hippocampus.
Description missingHippocampal electrodes and units are labelled `hpc` or `ca3`; `hpc` does not say whether a site is in CA1, which the paper's analyses distinguish. Four DT7 files label all electrodes `unknown`.
Data missingThe paper's CA3 recordings used a CA1 reference electrode to detect ripples. The 10 files with `ca3` electrodes hold 64 `ca3` and 32 `ls` channels and no CA1 channel, and the paper reports 25 CA3 and LS recordings against 10 here.
Description missingElectrode x, y and z are NaN in every file. The probe geometry and its orientation in the brain, needed for the travelling-wave analysis of Figure 3, are not given, and the relative depth of each recording within LS (Figure 2) is not given.
Data missingThe units carry no waveforms, no channel and no quality metric (the Methods compute isolation distance), so the CA1 depth of each neuron (Figure 7) cannot be estimated and single units cannot be told from multi-unit clusters; 11 units fire above 50 Hz on average. Wide-band data, from which waveforms could be re-extracted, are in 27 of the 98 files (DT1, DT2 and two DT9 files); 1579 of 2778 hippocampal units have none.
Data missingThe two rats of the optogenetic experiments (Figure 6) are not in the dandiset.
File format error25 units in 10 files have spike times out of order (1 to 2782 inversions each), against the NWB rule that spike times are sorted.
Methods not availableThe paper's Key Resources point to github.com/DavidTingley/papers for scripts that recreate each main figure, but the repository holds only the network model of Figure 8 and the code of the 2018 paper; the analyses of Figures 1 to 7 are not published.
Paper inconsistentThe HFO band is given as 120-180 Hz in the Summary and the legends of Figures 1, 2 and 5 and as 120-200 Hz in the Results and Methods. The Methods call the 62-recording dataset that of Figure 6, which shows optogenetics, where Figure 7 is meant, and cite Figure 3A and 3B (travelling waves) for the jitter control and the state of phase-locking epochs, which appear to be Figure S3A and Figure 5.
OtherIn this attempt's detection, 35% of LS HFOs within 50 ms of a CA1 ripple fall on the same 0.8 ms sample as the ripple (71% in DT8, 62% in DT2, 7% in DT9), and the CA1-LS 120-200 Hz phase difference is consistent even at random times (median phase-locking value 0.134 against about 0.021 for random phases). Re-referencing the LS channel to another LS channel on its shank lowers the same-sample share to 4.1%. This points to a signal common to the LS and hippocampal channels, such as the reference, which the files do not document; some files list a device of "reference sites a few millimeters dorsal" without saying which channels use it.

Notebooks

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

Reproduction Lateral septum HFOs and hippocampal ripples (Tingley and Buzsaki 2019)

Tests the findings of Tingley and Buzsaki (2019) on dandiset 000218, which holds LFP, sorted units, sleep states and trials from simultaneous recordings of the lateral septum and hippocampus in 6 rats. LS high-frequency oscillations and CA1 ripples are detected from the LFP, as the files hold no detected events, and the notebook tests their properties, their coupling and state dependence, the phase locking and co-activation of LS neurons, and which ripple features predict LS firing. The optogenetic experiments are not in the dandiset. All files

Reproducing Tingley and Buzsaki (2019) from dandiset 000218
reproduce_tingley_2019.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-10-17
Agreed with 30 of 31 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The executed notebook of the first run with the conclusion cells and the summary removed, the helper module and the claims file. The attempt record was not among its materials.
- fig5d-lag: the verifier’s outcome was “Not reproduced”. The median lag is 0.00 ms under both readings, against 6.4 +- 4.2; 35% of pairs fall on the same sample. The CCG peak criterion holds (166 SD above the flanks, at 0 ms). Problem 1, that the LS detection channel carries the hippocampal or a common-reference 120-180 Hz signal (same-sample pairs, 0 ms lag, jittered PLV of 0.20), and no local re-referencing was tried.
Round 2
2026-10-17
Agreed with 30 of 31 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The full executed notebook of the second run, with its conclusions, figures and summary, the helper module, the claims file and the attempt record.
- fig5d-lag: the verifier’s outcome was “Not reproduced”. Not reproduced, as in round 1. The criterion defines no partial outcome. The part of the claim that can be measured, a lag of 6.4 ms with CA1 leading, fails on both channels, 0.00 ms on the detection channel and 0.00 or 0.80 ms re-referenced. A correlogram peak at 0 ms does not show CA1 leading. Also, cell 61 and the notes say "DT9 gives 3.8 ms" without those pairs, but the 3.8 ms comes from all pairs, as the median of DT9's per-recording medians, and it is inside the tolerance (2.2 to 10.6 ms), which neither text says.
Round 3
2026-10-17Agreed with 1 of 1 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The proposed change of outcome for fig5d-lag only, with the full executed notebook of the final revision (cells 60, 61 and 68), the claims file and the attempt record. No paper text was downloaded this round. Report in transcripts/000218/verifier-3-report.md.

Method differences

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

The paper's recordings cannot be matched to files one by one; counts are compared under the readings the criterion lists.dataset-composition
Events are detected here on one LS channel per recording chosen by a fixed rule (highest 120-200 Hz SD in a 300 s NREM sample); the paper does not say which channels it used.fig1c-hfo-each-animal
The paper's noise channel is replaced by the EMG estimated from the LFP, which the Methods allow when no noise channel exists.fig1c-hfo-each-animal
Detection channels and noise exclusion differ from the paper's (see fig1c-hfo-each-animal).fig1c-event-count
Spikes are counted between event start and stop; the paper does not give its window.figs1-participation
The phase is taken on the recording's detection channel, not on a channel near each unit; the paper does not say which channel it used.fig1d-phase-locking
The tolerances are the paper's stated SDs, as the protocol asks, which makes them wide.fig1f-shorter-than-ripples
The `hpc` channel is taken as CA1.fig1f-shorter-than-ripples
The instantaneous-frequency estimator pulls towards the band centre (the verifier measured 145 Hz for a 140 Hz burst and 167 Hz for 170 Hz); this compresses differences but cannot reverse their order.fig1g-slower-than-ripples
The paper names no test; a Pearson test over all events was fixed in the criterion.fig1h-no-amplitude-frequency-correlation
Tested on the DT7 recordings with channels in other regions, over up to 30 min of all channels, in place of the paper's depth series of Figure 2C.fig2-confined-to-ls
The CCG normalization, the smoothing at the window edges and the reading of the 5-spike inclusion rule were fixed in the criterion; the Methods do not give them.fig4b-pre-post-unchanged
HFO spikes are those inside this attempt's detected events, which hold few LS spikes.fig4b-pre-post-unchanged
As for fig4b-pre-post-unchanged.fig4b-hfo-above-behav
Interpolation and chance by circular shift were fixed in the criterion; the Methods describe the warping but not the chance procedure in detail. One shift per pair is used for all factors.fig4d-compression
The paper's supplementary figure was not available; the test was fixed in the criterion.figs2-post-rate-increase
Detection as in fig1c-hfo-each-animal; 4 recordings without NREM or WAKE time are left out.fig5b-nrem-and-rates
Detection as in fig1c-hfo-each-animal; event times are the troughs of the filtered signal.fig5d-lag
Detection as in fig1c-hfo-each-animal.fig5e-state-coupling
A window of +-25 ms and a single test over neurons were fixed in the criterion; the figure tests each bin of the correlogram.fig5f-ripple-triggered-firing
The coherence measure (phase-locking value of the phase difference across ripples) was fixed in the criterion; the paper does not define it.fig5f-no-phase-coherence
The LS channel is the detection channel; the paper does not say which channel it used.fig5f-no-phase-coherence
The paper does not state its model; a linear regression with cross-validated MSE and shuffles was fixed in the criterion. Under reading (b), the paper's t-test wording, the chance rate is about 40%.fig7e-ca1-spikes
The `hpc` units are taken as CA1.fig7e-ca1-spikes
As for fig7e-ca1-spikes.fig7e-state
As for fig7e-ca1-spikes; the count window (ripple start to stop) was fixed in the criterion.fig7e-duration

Provenance

Attempt record: 000218-tingley-2019-01
Paper: Routing of Hippocampal Ripples to Subcortical Structures via the Lateral Septum
Date: 2026-10-16
Made by: An agent, run by CatalystNeuro
Dataset: DANDI:000218
Version: 0.250624.0439 (doi:10.48324/dandi.000218/0.250624.0439)
Run on: GitHub Actions
Harness: Claude Code (GitHub Actions)
Model: claude-opus-5-5
Protocol: pilot-7

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.

← DANDI:000213 DANDI:000226 →