Large-Scale 3D Two-Photon Imaging of Molecularly Identified CA1 Interneuron Dynamics in Behaving Mice (DANDI:000625)

Claims

From the paper Large-Scale 3D Two-Photon Imaging of Molecularly Identified CA1 Interneuron Dynamics in Behaving Mice

Reproducing Geiller, Vancura et al. (2020) from dandiset 000625 14

Notebook reproduce_geiller_2020.ipynb

Not reproducedResults, locomotion section; Figure 2C
IvC/NGFC were less modulated by locomotion than other subtypes.
Averaged over sessions IvC/NGFC have a mean correlation of 0.511 against 0.484 for PVBC, SomC, BiC and AAC pooled (one-sided Mann-Whitney p = 0.58), and the same for the absolute correlation. Session 21324 alone shows a lower correlation for IvC/NGFC (0.368 against 0.525, p = 0.005); the other two sessions do not.
How the test differed from the paper’s method
- One mouse with 13 IvC/NGFC, where the paper has 119.
- The comparison group (PVBC, SomC, BiC and AAC pooled) and the test are ours.
Not reproducedResults, locomotion section (Figure S2)
The activity of positively correlated cells followed changes in velocity by up to several seconds, while the activity of negatively correlated cells was more tightly linked in time to changes in velocity.
Positively correlated ROIs lag velocity (median -1.67 s), as stated. The 8 negatively correlated ROIs have a larger median absolute lag (2.06 s against 1.67 s, one-sided p = 0.60), and the same direction holds in each session taken alone. Lags are averaged over sessions even where the sign of r changes; keeping only ROIs with the same sign in every session leaves 1 negatively correlated ROI, too few to test. Figure S2 itself was not available.
How the test differed from the paper’s method
- One mouse; the paper's Figure S2 was not available, so the criterion rests on the Results text.
- Lags depend on imaging and behaviour starting together at time 0, which the files state and which could not be checked independently.
Not reproducedResults, spatial section; Figure 4
17.5 ± 6.4% (mean ± sem over 6 mice, 3 sessions each) of interneurons, blind of subtype, passed the test for significant spatial selectivity.
With the paper's rule, including a peak above 100% dF/F, 1.0% of ROIs are selected on average over sessions, since only 1.1% of ROIs have a tuning curve reaching 1.0 dF/F. Without the peak rule 69.8% are selected, against 8.6% on a surrogate with dF/F rotated against behaviour. Neither reading is within 17.5 ± 6.4%. A peak rule of 50% dF/F gives 13.0%, but that threshold is not the paper's. The paper states no false-positive rate for the rule.
How the test differed from the paper’s method
- One mouse, three sessions, where the paper has six mice with three sessions each.
- Fields may wrap around the end of the belt (not stated in the paper); without wrapping the rule without the peak selects 66.0%.
- The belt is 2.11 m long in the data, divided into 100 bins of 2.11 cm, where the paper has 2 m and 2 cm bins.
- The dF/F in the files may not be scaled as the paper's was; this cannot be checked.
Not reproducedResults, spatial section (Figure S5)
12.7 ± 8.3% of interneurons (mean ± sem) had negative spatial selectivity, slightly fewer than had positive selectivity.
On average 66.4% of ROIs have a negative field (47.0%, 74.5% and 77.7% by session), against 10.8% on the surrogate and 12.7 ± 8.3% in the paper. Under the paper's rule the two kinds of field are coupled, as a strong field raises the rotated shuffles at every bin. On synthetic cells, a cell with only a positive field is also flagged negative, and a cell with only a negative field is flagged positive under the rule without the peak. With the positive rule without the peak, 55.8% of ROIs on average have both kinds, 10.6% only a negative field and 13.9% only a positive one. The coupling explains part of the 66.4%, not all of it; 78.4% of ROIs with a positive field have a negative field, and so do 37.7% of ROIs without one, about 3.5 times the surrogate rate. Counting only ROIs with a negative field and no positive field gives 10.6%, inside the paper's range, but that variant depends on the positive rule without the peak (65.3% with the peak rule), excludes real negative cells flagged positive, and was chosen after the result, so the outcome is kept.
How the test differed from the paper’s method
- As for fig4-fraction-spatial.
Not reproducedFigure 4D
Position decoded from the small subset of spatially selective interneurons was about as accurate as from all interneurons, and more accurate than from as many randomly chosen interneurons or with position shuffled.
The criterion holds in one session of three. In 21324 the 162 selected ROIs give an error of 13.8 cm, against 13.2 cm from all ROIs, 14.4 cm from random subsets and 48.7 cm with position shuffled. In 21333 the 118 selected ROIs do no better than random subsets (25.2 against 24.8 cm). In 21322 the 6 peak-rule ROIs give 30.3 cm against 15.4 cm from all ROIs. Shuffling position brings the error to about chance (49 to 53 cm, chance 52.7 cm) in every session, and all ROIs decode position with errors of 13 to 21 cm.
How the test differed from the paper’s method
- The selected subset is not small in two sessions (118 and 162 ROIs), because the peak rule selects fewer than 5 ROIs there and the rule without the peak is used.
- A one-vs-one linear SVM (scikit-learn SVC, C = 1); the paper names a linear SVM without the multi-class scheme or penalty.
- Shuffled position made by rotating dF/F against position by at least 60 s, three draws; ten random subsets.
Not reproducedFigure 4H
Spatially selective PVBC had a more stable within-session recruitment (odd-even lap correlation of tuning curves) than spatially selective SomC.
For 22 spatially selective PVBC and 9 SomC, the mean odd-even correlation is 0.678 and 0.736 (unpaired t-test p = 0.42); SomC are higher in each session where both are present.
How the test differed from the paper’s method
- Selection as for fig4d-decoding.
- Each cell's correlation averaged over the sessions in which it was selected.
Partially reproducedResults, first section
About 227 interneurons were imaged per mouse (226.6 ± 27.3, mean ± std over 16 mice), and 121.0 ± 28.4 per mouse were assigned a putative subtype.
The mouse has 231 ROIs, inside 226.6 ± 27.3 (199.3 to 253.9). 74 ROIs carry one of the six subtypes and 83 are not labelled UNIDENTIFIED; both are below 121.0 ± 28.4 (92.6 to 149.4).
How the test differed from the paper’s method
- One mouse, compared with the paper's mean and standard deviation over 16 mice.
- Two readings of "assigned a subtype" were allowed, as the labels NOTHING and UNIDENTIFIED are not explained.
Partially reproducedResults, locomotion section (Figure S2)
Run-start and run-stop responses show that immobility-triggered activation is specific to CCKC.
At run stops the other five subtypes drop by 0.06 to 0.10 dF/F while CCKC stay flat on average (-0.003), a significant difference (one-sided Mann-Whitney p = 0.0009 at separations of 3, 5 and 10 s). CCKC do not rise on average, which the criterion requires. The 6 negatively correlated CCKC rise (0.084) and the other 9 fall (-0.061). By session the mean CCKC response is -0.005, 0.024 and -0.023, while every other subtype falls in every session (-0.021 to -0.173); the CCKC negatively correlated in a session rise in all three (0.019, 0.123 and 0.102).
How the test differed from the paper’s method
- One mouse; event separation of 5 s chosen for "several seconds", with 3 and 10 s shown.
Partially reproducedResults, spatial section (Figure S6A, S6B)
Within-session stability of tuning curves differed between subtypes, PVBC showing the highest and CCKC the lowest odd-even correlations.
CCKC have the lowest mean odd-even correlation (0.500). PVBC (0.598) are second to SomC (0.606). The one-way ANOVA across the five subtypes with at least 5 cells gives p = 0.56, so the statement that subtypes differ does not hold here.
How the test differed from the paper’s method
- One mouse; Figure S6 was not available, so the criterion rests on the Results text.
Partially reproducedResults, spatial section (Figure S6I)
Tuning curve correlations of the same cells after 1 hour and after 24 hours were lower than within-session correlations, and still higher than chance.
Across-session correlations are lower than within-session ones in every pair (0.22 against 0.61 for the same-day pair, Wilcoxon p = 4e-07; 0.06 and 0.14 for the two-day pairs). Above chance holds only for the same-day pair with identities shuffled among all cells (p = 0.047), not with shuffles within subtype (p = 0.10), and for neither two-day pair (p = 0.075 and 0.53).
How the test differed from the paper’s method
- Sessions 2.5 h and 54 to 56 h apart stand in for 1 h and 24 h. The order rests on the stated start times, which put 21333 first although its id is the highest; nothing else in the files dates the sessions.
- The ROI index is taken as the same cell across sessions; tuning curves of the same index agree above a shuffled null only for one pair, so a failure here may reflect the ROI mapping.
ReproducedResults, locomotion section; Figure 2C, 2D
The majority of interneurons were positively correlated with running velocity, and about 10% were negatively correlated, across all subtypes (156 negative and 1429 positive in Figure 2D).
Averaged over sessions, 90.5% of the 74 ROIs in the six subtypes are positively correlated with velocity and 0.095 negatively, inside 0.088 to 0.108. Among all 228 imaged ROIs 96.5% are positive and 0.035 negative, outside the tolerance. The claim holds under the six-subtype reading, and only because r is averaged over sessions; per session that reading gives 0.055, 0.143 and 0.164, all outside the range. Per session the all-ROI reading gives 0.065, 0.111 and 0.089, in range in session 21322 only. The sign of r changes between sessions in 17 of the 74 subtyped ROIs (7 of 15 CCKC); 18 are negative in at least one session and 7 on average. The result is at the edge of the range; 6 or 8 of 74 would fall outside it.
How the test differed from the paper’s method
- One mouse, three sessions, where the paper has seven mice.
- Frame rates for sessions 21324 and 21322 fitted from the data; on the stated clock 28% of ROIs in session 21322 are negatively correlated.
- Velocity smoothed with a Gaussian of 0.5 s, a width the paper does not give; 0 and 1 s change the averaged fractions by no more than 0.005 under either reading, and a per-session fraction by at most 0.014.
- Correlations averaged over sessions on the assumption that the ROI index is the same cell, which the data are consistent with but do not show.
- The denominator can be read two ways; the six-subtype reading matches and the all-ROI reading does not.
ReproducedResults, locomotion section; Figure 2B, 2C
Negatively correlated cells were significantly enriched in CCKC, where nearly half of cells were more active during immobility.
6 of 15 CCKC (0.40) are negatively correlated against 1 of 59 cells in the other five subtypes, one-sided Fisher exact p = 0.00017. By session the CCKC fraction is 0.13, 0.50 and 0.43.
How the test differed from the paper’s method
- One mouse, 15 CCKC, where the paper has 44 from seven mice.
- The threshold of 0.35 for "nearly half" is ours.
ReproducedFigure 2C legend
The correlation with velocity differs between the six subtypes (one-way ANOVA, p < 1e-10, with Tukey's post hoc test).
One-way ANOVA across the six subtypes gives p = 0.0015. Tukey's test places the difference with CCKC, lower than PVBC (p = 0.0009) and IvC/NGFC (p = 0.005). Per session the ANOVA is significant in one session of three. The paper's p < 1e-10 rests on seven mice and is not reached here.
How the test differed from the paper’s method
- One mouse; significance threshold 0.05 in place of the paper's reported p-value.
ReproducedFigure 4F
The selectivity index of spatially selective cells did not differ between subtypes (one-way ANOVA, p = 0.35).
One-way ANOVA across the six subtypes of the selectivity index of selected cells gives p = 0.17, so no difference is found. Subtype means range from 0.012 to 0.033 with a standard deviation of 0.020 across cells. The tuning curves of the selected subtyped cells have a median minimum of 0.158 dF/F and a median maximum of 0.255; subtracting each curve's minimum raises the median index from 0.016 to 0.307. The test compares near-flat curves and has little power.
How the test differed from the paper’s method
- Selection without the peak rule in two sessions, as for fig4d-decoding, so the cells are many more than the paper's spatially selective cells.

Not tested in a notebook 12

11 data missing from the dandiset, 1 other

BlockedFigure 1F· Data missing from the dandiset
The layer distribution of each subtype along the CA1 depth axis is consistent with earlier anatomical studies.
The files hold no depth or 3D coordinates of the ROIs. The plane segmentation has only id and pixel_mask (x, y, weight, no z); the masks lie on a strip of the 231 cropped frames side by side, and the imaging plane's origin_coords and grid_spacing are placeholders. The paper's Z-stacks and coordinates are not public as far as the dandiset and the paper's repository show.
BlockedFigure 2D· Data missing from the dandiset
Negatively correlated cells were enriched in stratum oriens and positively correlated cells in stratum pyramidale.
Needs the depth of each ROI, which the files do not hold.
BlockedResults, machine-learning section (Figure S3)· Data missing from the dandiset
A deep network classified PV-expressing, SOM-expressing, CCK and IvC/NGFC cells from dF/F, velocity and depth with accuracies of 64, 71, 60 and 59%.
The reported accuracies come from a network trained on cells pooled from the paper's mice with depth as a feature. The dandiset holds one mouse (74 identified cells, 13 to 33 per merged class) and no depth. A dF/F-only network could be trained on this mouse, but it would not test the reported accuracies. The code is public (TensorFlow 2.1, Keras 2.3.1); its data files are not.
BlockedFigure 3D· Data missing from the dandiset
During sharp-wave ripples PVBC were generally strongly modulated while most SomC were not.
The ripple experiments used LFP and 40 Hz imaging of about 30 cells in four other mice. The dandiset holds no LFP and no 40 Hz sessions.
BlockedFigure 3E, 3F· Data missing from the dandiset
AAC and BiC showed bimodal modulation during sharp-wave ripples, and all subtypes but CCKC and IvC/NGFC split into a modulated and a non-modulated group.
As for fig3-swr-pvbc-vs-somc.
BlockedFigure 4E· Other
BiC had a higher probability than other subtypes of being spatially modulated again in other sessions.
The dandiset has 3 BiC, below the 5 the criterion requires; the paper had 50.
BlockedFigure 5B, 5C, 5D· Data missing from the dandiset
Interneuron tuning curves decorrelated on exposure to a novel belt, more than with the passage of time, and context could be decoded from population activity.
All three sessions are random foraging; no novel-belt session is included, and the files do not name the belt.
BlockedFigure 5F, 5H· Data missing from the dandiset
PVBC, SomC, BiC and AAC decreased their activity in the first laps on a novel belt; CCKC were not modulated.
As for fig5-context-decorrelation.
BlockedFigure 6D· Data missing from the dandiset
In goal-oriented learning, SomC and BiC had higher reward modulation indices than other subtypes.
No goal-oriented learning session and no lick record are included.
BlockedFigure 6G, 6H· Data missing from the dandiset
After the reward zone moved, SomC and BiC activity increased at the old zone above chance, and this decreased over the session.
As for fig6-reward-somc-bic.
BlockedFigure 7C· Data missing from the dandiset
VIP+/CCK+ and VIP+/M2R+ cells were negatively correlated with velocity, and all ISI subtypes positively.
These cells come from VIP-Cre mice and strategy 3 VGAT-Cre mice; the dandiset's mouse tg19 is a strategy 1 mouse, without VIP, CR or M2R staining.
BlockedFigure 7H, 7I· Data missing from the dandiset
VIP+/CR- cells and type III ISIs were less correlated with SomC and BiC than with PV+ cells.
As for fig7-vip-subtypes-immobility.

Findings about the data

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

Metadata errorThe imaging rate is stated as 5.7981 Hz for all three sessions. For sessions 21324 and 21322 that makes imaging 680 s long against 900 s and 879 s of behaviour, and the population activity does not follow running (r = 0.16 and 0.09). Rates of 4.395 and 4.385 Hz give r = 0.72 and 0.87, decode running at 0.94 and 0.89 balanced accuracy, and make all three sessions about 898 s long. Session 21333's stated rate fits.
Metadata errorThe unit of the dF/F series is given as lumens; the values are dF/F fractions.
Data missingThe Fluorescence series is the same HDF5 object as the DfOverF series, so the raw fluorescence the description lists is not in the files.
Metadata errorThe description of the subtype column reads "Recording location in Middle Earth", and the subtype table has 232 rows for 231 ROIs (id 231 has no ROI).
Description missingThe labels NOTHING and UNIDENTIFIED, the label "AAC (RIGHT CELL)", and the reward and water intervals have no description.
Identifier missingNothing states that a ROI index refers to the same cell in every session. The data are consistent with it (velocity correlations of the same index agree across sessions above a shuffled null) but tuning curves agree above the null for only one pair of sessions.
Data missingThe files hold no depth or position of the ROIs in the imaged volume, which several of the paper's analyses use, and no lick record.
Metadata errorAll three sessions are described as "session 1 of 4", and the experiment description names four random-foraging sessions; only three are in the dandiset. The session ids (21322, 21324, 21333) and the start times (21333 two days before the other two) give different orders, and nothing else in the files dates the sessions.
Description missingThe pixel masks lie on a strip of the 231 cropped frames side by side (x from 0 to 10625), the imaging plane's origin_coords (0, 0) and grid_spacing (50, 50 um) are placeholders, and there are no reference images, so a ROI has no position in the field of view or in depth.
Selection not documentedThe dandiset holds one of the paper's 16 mice and only its random-foraging sessions, without saying which part of the paper it covers.
Metadata errorThe emission wavelength is given as 920 nm, equal to the excitation, and the subject's age is P60D in sessions two days apart, where the paper gives 3 to 5 months.
Description missingThe description lists raw images of individual cropped ROIs, but each file holds a single cropped series (33 x 46 or 25 x 46 pixels) without saying which ROI it shows.
OtherWith the paper's place-field rule, which requires a peak above 100% dF/F, 1.0% of ROIs qualify, as only 1.1% of tuning curves reach 1.0 dF/F. The dF/F in the files may be scaled or smoothed differently from the paper's.

Notebooks

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

Reproduction Molecularly identified CA1 interneurons during random foraging (Geiller, Vancura et al. 2020)

Tests the findings of Geiller, Vancura et al. (2020) that can be tested on dandiset 000625, which holds three random-foraging sessions of one mouse with 231 interneurons, 74 of them assigned one of the paper's six subtypes. It checks the imaging clock against the behaviour, then measures each interneuron's correlation with running speed, run-stop responses, spatial selectivity, position decoding and the stability of tuning curves within and across sessions. The sharp-wave ripple, novel-context, goal-oriented learning and VIP experiments are not in the dandiset. All files

Reproducing Geiller, Vancura et al. (2020) from dandiset 000625
reproduce_geiller_2020.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-10-14Agreed with 26 of 26 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The first executed notebook with the conclusion cells and the summary removed, the helper module, the claims file and the paper's full text (PMC7736348). The verifier checked the archive's version and asset metadata through the DANDI REST API, and ran synthetic checks of place_fields, run_events, sparsity_index, velocity_correlation, decode_position and lap_tuning.
Round 2
2026-10-14Agreed with 26 of 26 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The full executed notebook after the first revision, with the conclusion cells and the summary, the helper module, the claims file, the README, the attempt record and the paper's full text (PMC7736348).

Method differences

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

One mouse, compared with the paper's mean and standard deviation over 16 mice.fig1-cells-per-mouse
Two readings of "assigned a subtype" were allowed, as the labels NOTHING and UNIDENTIFIED are not explained.fig1-cells-per-mouse
One mouse, three sessions, where the paper has seven mice.fig2c-most-positive
Frame rates for sessions 21324 and 21322 fitted from the data; on the stated clock 28% of ROIs in session 21322 are negatively correlated.fig2c-most-positive
Velocity smoothed with a Gaussian of 0.5 s, a width the paper does not give; 0 and 1 s change the averaged fractions by no more than 0.005 under either reading, and a per-session fraction by at most 0.014.fig2c-most-positive
Correlations averaged over sessions on the assumption that the ROI index is the same cell, which the data are consistent with but do not show.fig2c-most-positive
The denominator can be read two ways; the six-subtype reading matches and the all-ROI reading does not.fig2c-most-positive
One mouse, 15 CCKC, where the paper has 44 from seven mice.fig2c-cck-negative
The threshold of 0.35 for "nearly half" is ours.fig2c-cck-negative
One mouse; significance threshold 0.05 in place of the paper's reported p-value.fig2c-subtypes-differ
One mouse with 13 IvC/NGFC, where the paper has 119.fig2c-ngfc-less-modulated
The comparison group (PVBC, SomC, BiC and AAC pooled) and the test are ours.fig2c-ngfc-less-modulated
One mouse; the paper's Figure S2 was not available, so the criterion rests on the Results text.s2-lags
Lags depend on imaging and behaviour starting together at time 0, which the files state and which could not be checked independently.s2-lags
One mouse; event separation of 5 s chosen for "several seconds", with 3 and 10 s shown.s2-cck-run-stop
One mouse, three sessions, where the paper has six mice with three sessions each.fig4-fraction-spatial
Fields may wrap around the end of the belt (not stated in the paper); without wrapping the rule without the peak selects 66.0%.fig4-fraction-spatial
The belt is 2.11 m long in the data, divided into 100 bins of 2.11 cm, where the paper has 2 m and 2 cm bins.fig4-fraction-spatial
The dF/F in the files may not be scaled as the paper's was; this cannot be checked.fig4-fraction-spatial
As for fig4-fraction-spatial.fig4-fraction-negative
The selected subset is not small in two sessions (118 and 162 ROIs), because the peak rule selects fewer than 5 ROIs there and the rule without the peak is used.fig4d-decoding
A one-vs-one linear SVM (scikit-learn SVC, C = 1); the paper names a linear SVM without the multi-class scheme or penalty.fig4d-decoding
Shuffled position made by rotating dF/F against position by at least 60 s, three draws; ten random subsets.fig4d-decoding
Selection without the peak rule in two sessions, as for fig4d-decoding, so the cells are many more than the paper's spatially selective cells.fig4f-selectivity-no-difference
Selection as for fig4d-decoding.fig4h-pvbc-more-stable
Each cell's correlation averaged over the sessions in which it was selected.fig4h-pvbc-more-stable
One mouse; Figure S6 was not available, so the criterion rests on the Results text.s6-within-session-order
Sessions 2.5 h and 54 to 56 h apart stand in for 1 h and 24 h. The order rests on the stated start times, which put 21333 first although its id is the highest; nothing else in the files dates the sessions.s6-later-lower-above-chance
The ROI index is taken as the same cell across sessions; tuning curves of the same index agree above a shuffled null only for one pair, so a failure here may reflect the ROI mapping.s6-later-lower-above-chance

Provenance

Attempt record: 000625-geiller-2020-01
Paper: Large-Scale 3D Two-Photon Imaging of Molecularly Identified CA1 Interneuron Dynamics in Behaving Mice
Date: 2026-10-14
Made by: An agent, run by CatalystNeuro
Dandiset version: 0.231114.0019
Harness: Claude Code (GitHub Actions)
Model: claude-opus-5-5
Protocol: pilot-4

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.

← Dandiset 000582 Dandiset 000691 →