Claims 57
From the paper Object-centered population coding in CA1 of the hippocampus · claims file · checked 2026-10-20: ready
- cell-counts
The object sessions yielded 1,189 unique CA1 principal cells from 9 rats, the control sessions 219 CA1 principal cells from 3 rats, and up to 620 CA1 cells were recorded simultaneously.
n = 1,189 principal cells recorded from 9 rats) and the corresponding pair of control trials (as in Figure 2B; brown lines; n = 219 principal cells recorded from 3 rats)
Figure 2D legend; Introduction, paragraph 3; Results, Decoding the presence of an object, paragraph 1How it will be judged
Cells are counted after the paper's inclusion rules (Methods, Spike sorting and cell-inclusion criteria, and Histology), CA1 only. Unique cells under both readings of the opening comment. Reproduced when the three counts are within tolerance.
- fig1c-object-tuned-fraction
19.9% of CA1 cells (237 of 1,189) were object-tuned by the template matching procedure, significantly more than the 5.03% (11 of 219) found by the same procedure in control trials without objects.
Using the template matching procedure, 19.9% of CA1 cells were classified as object-tuned (237/1,189 cells) ... This fraction was significantly higher than a chance level determined by running the same template matching procedure on control trials without objects (11/219 cells, 5.03%
Results, Object cells and object-vector cells in the hippocampus, paragraph 3; Figure 1CHow it will be judged
Template matching as in the opening comment, under each template-width reading, on the O and OM trials of every CA1 cell. Chance level as in the Methods (Template matching, Chance level), on the 219 control cells with the object placed at (75, 75) cm in the trial standing for O and at (35, 120) cm in the trial standing for OM. The paper does not say which control trials stand for O and OM; readings (a) CON2 and CON3, the trials in the same place in the session; (b) CON1 and CON2. Rate maps are smoothed as the opening comment's "Smoothing and unvisited bins" says (rate per bin, unvisited bins 0, then smoothed; calculateRateMap.m and the Methods, Firing rate maps). Reproduced when, under one reading, both fractions are within tolerance and a binomial test of the object-tuned count against the chance fraction gives p < 0.05 (two-sided, as all the paper's tests are; Methods, Statistical tests).
- fig1d-field-distances
Firing fields of object-tuned cells lay at a wide range of distances from the object (4.8 to 80.7 cm, mean 37.7 cm, SD 16.0 cm) and covered all directions around it.
The firing fields of the object-tuned cells covered a wide range of distances from the object (range: 4.8–80.7 cm; mean ± SD: 37.7 ± 16.0 cm) and the entire azimuthal range
Results, Object cells and object-vector cells in the hippocampus, paragraph 3; Figure 1DHow it will be judged
The tolerance on the mean is 10% of it, as the SD of 16.0 cm is the spread over cells, not an uncertainty of the mean. Fields detected as in the opening comment. The paper does not say which point of a field gives its distance, in which trial, or which field of a cell with several. Readings (a) the centroid of the field (the point Supplementary Figure 5G plots) in the O trial, nearest field; (b) as (a) averaged over the O and OM trials; (c) the field's peak bin in the O trial. The 1 Hz rule of field detection is not given further (Methods, Firing rate maps); each point reading is run with the field readings (a) peak at least 1 Hz and (b) every bin at least max(3 x MAD, 1 Hz). Reproduced when, under one reading, the mean is within tolerance, the range lies within 4.8 to 80.7 cm widened by 10% of each bound, and every 45 degree sector around the object holds at least one field.
- fig1e-object-cells-and-object-vector-cells
About 4.1% of all recorded CA1 cells (49 of 1,189) were object cells, object-tuned with fields under 15 cm from the object, and 13.3% (158) were object-vector cells, with fields 15 cm or more away; 30 object-tuned cells had no detectable field.
we estimated that 4.1% of all recorded cells were object cells (49/1,189 cells, object-tuned cells with firing fields <15 cm from the object ... and 13.3% were object-vector cells (158/1,189 cells
Results, Object cells and object-vector cells in the hippocampus, paragraph 4; Figure 1E; Methods, Template matching procedureHow it will be judged
Field distance under the readings of fig1d-field-distances. Reproduced when, under one reading, the three values are within tolerance.
- fig1f-orientation-coverage
The nearer an object-tuned cell's firing field is to the object, the larger the range of orientations around the object that its fields cover.
The nearer the cell’s firing field is to the object, the larger is the orientation coverage.
Figure 1F legendHow it will be judged
Orientation coverage is the fraction of 0 to 360 degrees around the object covered by the cell's firing fields (Supplementary Figure 3C legend), taken as the angular span, seen from the object centre, of the field bins; distance as in fig1d-field-distances. The paper reports no test; its stated test for correlations is Pearson's (Methods, Statistical tests). Reproduced when, under one distance reading, the Pearson correlation of coverage with distance over object-tuned cells is negative with p < 0.05.
- fig1g-no-experience-effect
The object-tuning score did not change with the number of days the animal had been exposed to the object (r = -0.080, p = 0.574, 62 cells).
We observed no significant effect of day of exposure to the object on the object-tuning score (Figure 1G)
Results, Object cells and object-vector cells in the hippocampus, paragraph 4; Figure 1GHow it will be judged
A claim of absence, judged on the paper's test, a Pearson correlation at alpha 0.05 of the object-tuning score (opening comment) against the number of days the rat had been exposed to the object, over the cells recorded on more than one day and object-tuned on at least one. The legend says each dot is a cell (n = 62) and does not say which day's score stands for a cell; readings (a) the score on the first day the cell passed as object-tuned; (b) the cell's highest score; (c) every cell-day, in case the dots are cell-days. Reproduced when p >= 0.05; not reproduced when p < 0.05. The r and the number of cells are given either way. Needs cell identities across days.
- fig1h-cross-day-stability
Of 35 object-tuned cells recorded on more than one day, few passed the object-tuning criteria on more than one day, yet object-tuned cells passed again on later days more often than other cells (13/35 against 36/370, p = 3.35e-4).
Only 13 of these cells (31.4%) had significant object-tuning scores on more than 1 day (Figure 1H). However, object-tuned cells were more stable than expected by chance, with higher probability of passing the classification on later days compared to other cells (13/35 object-tuned cells vs. 36/370 other cells
Results, Object cells and object-vector cells in the hippocampus, paragraph 5; Figure 1HHow it will be judged
The paper gives 13 in the text and 11 in the Figure 1H legend, with 31.4% for both; the readings are 13 (text) and 11 (legend, which matches 31.4%). Day 0 is the first day a cell was recorded (Figure 1H legend). Reproduced when the counts are within tolerance, the number of object-tuned cells passing on more than one day is within 10% of 13 or of 11, and a two-sided binomial test (Methods, Statistical tests) of that count among the 35 against the fraction of other cells passing on a later day gives p < 0.05. The y axis of Figure 1H runs to 62 cells while its legend says 35; the count found is reported. Needs cell identities across days.
- figs2e-within-trial-stability
Within a single trial, object-tuned cells with larger object-tuning scores were more stable (r = 0.45 between within-trial stability and score).
cells with larger object-tuning scores tended to be more stable than cells with lower scores (Figure S2E) (Pearson correlation between within-trial stability and object-tuning score, r = 0.45
Results, Object cells and object-vector cells in the hippocampus, paragraph 5; Supplementary Figure 2EHow it will be judged
Within-trial stability is the Pearson correlation of rate maps from the first and second halves of the O trial (Supplementary Figure 2E legend), rate maps as in the opening comment. Reproduced when r is within tolerance and p < 0.05.
- figs2c-no-proximodistal-effect
The percentage of object-tuned cells did not depend on the proximodistal location of the recording site (r = 0.42, p = 0.22).
There was no significant linear correlation between the proximodistal site and the percentage of object related cells (r = 0.42, p = 0.2216
Results, Object cells and object-vector cells in the hippocampus, paragraph 3; Supplementary Figure 2CHow it will be judged
A claim of absence, judged on the paper's test. Per implant (rat, with the two probes of rat 27207 apart), the percentage of object-tuned cells against the proximodistal order of the implant site; Pearson correlation. Reproduced when p >= 0.05, not reproduced when p < 0.05; r is given either way. Needs the implant sites, which the paper shows in Supplementary Figure 1.
- figs3a-ca3-object-tuned
Object-related activity was present also in CA3, where 23% of cells (72 of 313) were object-tuned.
In CA3, 23% of the cells were object tuned (72/313 cells; 7 object cells and 58 object vector cells)
Supplementary Figure 3AHow it will be judged
Template matching as for CA1 (opening comment), on the CA3 cells. Reproduced when, under one template-width reading, the fraction is within tolerance.
- fig2d-spatial-change-no-o
Introducing the object changed spatial firing more than the passage of time did, with lower NO-O rate-map correlations than CON1-CON2 correlations in control sessions.
introducing the object led to larger changes in spatial firing patterns (i.e., smaller Pearson correlations of rate maps) than in pairs of control trials with no object
Results, The majority of hippocampal cells change spatial firing patterns, paragraph 1; Figure 2D, top leftHow it will be judged
Spatial correlation as in the opening comment. Reproduced when the NO-O correlations are lower than the CON1-CON2 correlations (lower median) with a two-sided rank-sum p < 0.05.
- fig2d-rate-overlap-no-o-unchanged
The overall rate change between NO and O, measured by rate overlap, did not differ from that between control trials.
the overall rate change, estimated as the mean firing rate in the less active trial divided by mean firing rate in the more active trial (rate overlap), was not different between trial pairs with objects (NO-O) and control trials (CON1-CON2)
Results, The majority of hippocampal cells change spatial firing patterns, paragraph 1; Figure 2D, top rightHow it will be judged
A claim of absence, judged on the paper's test, a two-sided Wilcoxon rank-sum test of NO-O against CON1-CON2 rate overlaps at alpha 0.05. Reproduced when p >= 0.05, not reproduced when p < 0.05; the difference in medians is given either way, under both rate-overlap readings of the opening comment.
- fig2-fraction-changed-no-o
About two thirds of CA1 cells (67.5%, 803 of 1,189) changed their spatial firing, their rate, or both between NO and O.
we estimated that 67.5% of the cells (n = 803/1,189 cells) expressed either spatial change, rate change, or both between NO and O
Results, The majority of hippocampal cells change spatial firing patterns, paragraph 1How it will be judged
The Methods' classification (Spatial correlation and rate overlap). Within-trial distributions from the NO trial, split 500 times at random into two groups of 1 s bins; between- trial distributions from a random half of each trial's 1 s bins, 500 times. A cell changes on a measure when the mean between-trial value is below the 1st percentile of the within-trial distribution and a Wilcoxon signed-rank test of the two distributions gives p < 0.05 (alpha from Methods, Statistical tests; the 500 values are paired by repetition, as the signed-rank test needs pairs and the paper does not say how they are paired). A cell counts when it changes on spatial correlation, rate overlap or both. Reproduced when the fraction is within tolerance.
- fig2d-o-om-change
Moving the object also changed spatial firing and rate more than in control trials, and 69.7% of cells (829 of 1,189) changed between O and OM.
Considerable change was also observed between the O and OM trials ... In these trials, 69.7% of the cells changed their spatial firing pattern, firing rate, or both when the object was moved (n = 829/1,189 cells)
Results, The majority of hippocampal cells change spatial firing patterns, paragraph 1; Figure 2D, second rowHow it will be judged
As fig2d-spatial-change-no-o and fig2-fraction-changed-no-o, for O-OM against CON2-CON3. Reproduced when both rank-sum tests give p < 0.05 with lower O-OM medians, and the fraction is within tolerance; partially reproduced when one of the three holds.
- fig2d-om-no-prime-change
Removing the object changed spatial firing more than in control trials (OM-NO' against CON3-CON4), without a significant difference in rate overlap, and many cells kept fields from the preceding trial.
Major changes were evoked also when we removed the object ... Upon removal of the object, many cells retained firing fields from the preceding trial
Results, The majority of hippocampal cells change spatial firing patterns, paragraph 1; Figure 2D, third rowHow it will be judged
Reproduced when the spatial correlations are lower for OM-NO' with rank-sum p < 0.05, and the rate-overlap rank-sum test gives p >= 0.05 (the absence part, judged on the paper's test). Sessions without a NO' trial are left out.
- fig2d-no-no-prime-change
Spatial firing in NO' differed from that in NO more than CON4 differed from CON1.
major changes in spatial firing were also observed between NO and NO0 (Figure 2D, fourth row
Results, The majority of hippocampal cells change spatial firing patterns, paragraph 1; Figure 2D, fourth rowHow it will be judged
Reproduced when NO-NO' spatial correlations are lower than CON1-CON4 correlations with rank-sum p < 0.05. The paper prints the same z and p as for OM-NO'; that is noted, not tested.
- figs3f-ca3-changes
In CA3 too, spatial correlations between trials differed between object and control sessions for every trial pair, while rate overlaps did not.
As for CA1, we found that spatial correlation values between pairs of trials was always significantly different between Object sessions and control sessions, whereas rate overlap was not significantly different between the two types of sessions.
Supplementary Figure 3FHow it will be judged
313 CA3 cells against the 59 CA3 control cells, measures as for CA1. Reproduced when all four spatial-correlation tests give p < 0.05 and all four rate-overlap tests give p >= 0.05 (the absence part judged on the paper's test); partially reproduced when the spatial part holds and one or more rate-overlap tests is significant.
- figs4-object-traces
18.3% of CA1 cells (218 of 1,189) had object-trace fields, firing in NO' where they had fired in OM, well above the 1% chance level.
We found that 18.3% of CA1 cells had trace responses ... (n = 218/1189 cells in total, significantly above chance level of 1%
Supplementary Figure 4B to 4D legend; Methods, Region-of-interest analysis for object tracesHow it will be judged
The Methods' procedure. Cells changing between NO and NO' as in fig2-fraction-changed-no-o; fields in NO' (opening comment); a circular ROI per field moved to the same place in OM; rate overlap between the NO' ROI and the OM ROI against 1,000 shuffles placing the ROI at random in OM with wrap-around; a trace field above the 99th percentile. The ROI's size is open, as the Methods refer to a section that is not in the paper. Readings (a) a circle of the field's area centred on its centroid; (b) the smallest circle centred on the centroid containing the field. A cell counts when any of its fields passes (the Supplementary Figure 4A legend tests each field). Reproduced when, under one reading, both values are within tolerance and a two-sided binomial test (Methods, Statistical tests) against 1% gives p < 0.05.
- figs2d-traces-no-interference
Removing object traces from the OM trial did not significantly change the number of object-tuned cells (256 of 1,189 after removal, against the expected 19.9%).
The number object tuned cells was not significantly different after removing traces (using expected percentage 19.9% from Fig. 1c and observed count of 256/1189
Supplementary Figure 2DHow it will be judged
A claim of absence, judged on the paper's test, a two-sided binomial test of the count of object-tuned cells after removal against 19.9% at alpha 0.05. The paper does not say how traces were removed; readings (a) leave the OM rate-map bins inside the trace ROIs of figs4-object-traces out of the template correlations; (b) set the rate in those bins to zero. Reproduced when p >= 0.05, not reproduced when p < 0.05; the count is given either way.
- ovs-method-extra-cells
The object-vector score method of earlier studies classified 40 cells (3.4%) that the template matching did not, mostly false positives on inspection.
40 out of 1189 cells (3.4%) were classified by this method but not by the template matching procedure.
Methods, Object-vector score and object-vector cell criteria, last paragraphHow it will be judged
The Methods' four criteria, after excluding cells whose fields would fall outside the arena in OM. The vector map's distance and direction bins and its smoothing are not given; the paper directs to the method of its references 11 and 12 (Hoydal et al. 2019; Andersson et al. 2021), and those papers' bins and smoothing are used. Reference 12 (Andersson et al. 2021, Methods, OV score) gives them, read for the answer of 2026-10-21; "distance bins of 2 cm and orientation bins of 10°", rate as spikes over time per bin, "circularly smoothed with a 2D Gaussian kernel with s.d. of 1.5 bins (3 cm, 15°)", 0° East. Neither paper gives the range of distances; readings (r-a) 0 to the largest distance sampled in either trial, the correlation over bins with a value in both maps; (r-b) 0 to the largest distance sampled in both trials. Reproduced when, under one range reading, the count is within tolerance. The manual inspection is not repeated.
- fig3b-decode-presence
A Bayesian decoder using 1 s population spike counts told the NO trial from the O trial above chance (50%) in every session.
The decoding accuracy was above chance (50%) in all experimental sessions (Figure 3B)
Results, Decoding the presence of an object from neural-population activity, paragraph 1; Figure 3BHow it will be judged
The whole-arena decoder (opening comment), on each decoding session. The paper reports no test per session; it gives chance as 50%, and its chance level comes from the label-shuffling procedure (Methods, Shuffling trial labels; 50 shuffles per session, Supplementary Figure 5E). Reproduced when, under both speed readings, every session's accuracy exceeds 0.5 and the mean accuracy of that session's 50 label shuffles.
- fig3c-bin-size
Decoding accuracy grew roughly logarithmically with the testing bin size, rising quickly up to 1 s and saturating beyond.
The decoder’s performance increased approximately logarithmically with the bin size of the testing data vector (Figure 3C)
Results, Decoding the presence of an object from neural-population activity, paragraph 1; Figure 3C; Methods, Performance of decoderHow it will be judged
The whole-arena decoder with testing bins of 10, 50, 100, 500, 1000, 5000 and 10000 ms (Methods). The paper gives no test. Reproduced when the median accuracy over sessions rises from 10 to 1000 ms at each step, and its gain from 1000 to 10000 ms is smaller than its gain from 100 to 1000 ms.
- fig3d-number-of-cells
Decoding accuracy increased with the number of cells used.
the success of decoding increased with the number of cells included (Figure 3D)
Results, Decoding the presence of an object from neural-population activity, paragraph 1; Figure 3D; Methods, Effect of number of cellsHow it will be judged
Random subsets of k cells, 100 per k, whole-arena decoder (Methods). The data are open; readings (a) the Methods', the 620-cell Neuropixels session, k = 1, 5, 10, 20, 30, 50, 100, 200, 300, 400 and 500; (b) the legend's, the four datasets pooled (two tetrode, two Neuropixels), k up to each dataset's size. The paper gives no test; its stated test for correlations is Pearson's (Methods, Statistical tests). Reproduced when, under one reading, mean accuracy is higher at the largest k than at k = 1 and the Pearson correlation of accuracy with k over all subsets is positive with p < 0.05.
- fig4b-distance-decline
Decoding accuracy fell gradually with the animal's distance from the object, over 5 equal-sample distance bins, with significant steps between bins 1 and 2, 2 and 3, and 4 and 5.
we found that decoding performance decreased gradually with increasing distance from the object (Figure 4B)
Results, Decoding the presence of an object from neural-population activity, paragraph 2; Figure 4BHow it will be judged
The distance decoder (opening comment), on the decoding sessions. Reproduced when, under every reading of speed and sessions, the Kruskal-Wallis test across the 5 bins gives p < 0.05 with median accuracy falling from bin 1 to bin 5, and signed-rank tests of bins 1 vs 2, 2 vs 3 and 4 vs 5 each give p < 0.05 with the nearer bin higher. Partially reproduced when the Kruskal-Wallis part holds and one or two of the pairs do not. The population mean of bin 1 minus bin 5 (0.19 in the paper, Results, Distance-specific reorganization, paragraph 2) is reported.
- fig4c-each-session-decreasing
Every session showed decoding accuracy decreasing with distance from the object.
Note that all experimental sessions have decreasing curves.
Figure 4C legendHow it will be judged
The paper's two measures of the decrease (Methods, Measuring strength of distance-specific reorganization) give the readings (a) the least-squares slope of accuracy on distance-bin index is negative; (b) accuracy in bin 1 minus bin 5 is positive. Reproduced when, under one reading, every session meets it; partially reproduced when more than half do; not reproduced otherwise.
- fig4e-angle-flat
Decoding accuracy did not differ between 5 angular sectors around the object, so the distance effect is not specific to compass directions.
The decoding accuracy was the same in different angular bins (Figures 4E and 4F)
Results, Decoding the presence of an object from neural-population activity, paragraph 3; Figure 4E and 4FHow it will be judged
A claim of absence, judged on the paper's test, a Kruskal-Wallis test across the 5 angular bins at alpha 0.05. Distance decoder with 72 degree sectors of allocentric angle from the object in place of distance rings (Figure 4D legend, Methods). The sectors' starting angle is not given; readings (a) sectors starting at 0 degrees (the positive x axis, the code's angle origin in templateMatchingScore.m); (b) sectors starting at 36 degrees. Reproduced when p >= 0.05 under the readings, not reproduced when p < 0.05; H is given either way.
- fig4h-population-vector-correlation
Correlations between NO and O population vectors were lowest near the object and rose stepwise with distance before saturating, with significant steps between bins 1 and 2, 2 and 3, and 3 and 4.
This showed a stepwise linear increase in correlation values as a function of distance from object, with saturation at larger distances (Figure 4H)
Results, Decoding the presence of an object from neural-population activity, paragraph 5; Figure 4G and 4HHow it will be judged
Methods, Population vector correlation. Per session, rate maps of the whole NO and O trials (3 cm bins, 50 x 50, as the Methods give m = n = 50), the Pearson correlation across cells at each spatial bin, and the mean correlation in each of 5 distance bins. The bins are open; readings (a) the Methods' fixed 20 cm rings (0 to 100 cm); (b) 5 bins with equal numbers of spatial bins, which the reported 489 values per bin implies. The Methods do not say whether these maps are speed-filtered or blanked (answered 2026-10-21); readings (pv-a) speed-filtered, as the Methods' general rule for rate maps; (pv-b) not filtered, as the decoder code; each with (bl-a) unvisited bins blanked and left out of that bin's correlation or (bl-b) not blanked, as the decoder's 3 cm maps. Reproduced when, under one reading, the Kruskal-Wallis test gives p < 0.05 with correlation rising from bin 1 to bin 5 and the three signed-rank tests give p < 0.05 with the farther bin higher.
- figs5a-decode-no-from-no-prime
The NO trial could also be decoded from the NO' trial, above chance, consistent with trace fields.
we were also able to decode the NO trial from the NO0 trial using population activity (Figure S5A)
Results, Decoding the presence of an object from neural-population activity, paragraph 1; Supplementary Figure 5AHow it will be judged
As fig3b-decode-presence, with NO' in place of O, on sessions with a NO' trial. Reproduced when the accuracy exceeds 0.5 and the mean of the session's label shuffles in every session; partially reproduced when in more than half.
- figs5b-fixed-distance-bins
With fixed distance bins (0-20 to 80-100 cm) decoding accuracy still fell with distance.
Decoding accuracy as a function of distance from object when using fixed distance bins ... Decoding accuracy decreases as a function of distance from object
Supplementary Figure 5B; Methods, Fixed distance binsHow it will be judged
The distance decoder with bin edges 0, 20, 40, 60, 80 and 100 cm (Methods). Samples beyond 100 cm are left out, as the Methods' bins stop there. Reproduced when the Kruskal-Wallis test gives p < 0.05 with median accuracy highest in bin 1 and lowest in bin 4 or 5.
- figs5c-half-trial-control
Decoding the first from the second half of the NO trial showed no decrease of accuracy with distance.
we observed no decrease in decoding accuracy across the same distance bins when attempting to decode the first half of the NO trial from the second half of the NO trial (Figure S5C)
Results, Decoding the presence of an object from neural-population activity, paragraph 4; Supplementary Figure 5CHow it will be judged
A claim of absence, judged on the paper's test. The NO trial is split at its midpoint and each half split again into training and testing quarters (Methods, Decoding 1st half vs. 2nd half); distances from the O trial's object position (the Methods give no other). Reproduced when the Kruskal-Wallis test across the 5 bins gives p >= 0.05, not reproduced when p < 0.05.
- figs5d-control-sessions
In control sessions without an object (CON1 against CON2), decoding accuracy did not fall with distance.
attempting to perform the decoding in control experiments without the object (Control trial 1 vs. Control trial 2). We observed no decrease in accuracy as a function of distance
Results, Decoding the presence of an object from neural-population activity, paragraph 4; Supplementary Figure 5DHow it will be judged
A claim of absence, judged on the paper's test. Distance decoder on CON1 against CON2. The object position used for distances in control sessions is not given for this analysis; readings (a) the arena centre (75, 75) cm, the pseudo-object position the Methods use for the control chance level; (b) (35, 120) cm, the other one. The control sessions used are not stated; those with at least 25 simultaneously recorded cells are used (the rule for population analyses). Reproduced when p >= 0.05 under the readings, not reproduced when p < 0.05.
- figs5e-shuffled-labels
With shuffled trial labels the decoder performed at 50% in every distance bin.
On the shuffled data, the decoder performed with a mean accuracy of 50% in all distance bins (Figure S5E)
Results, Decoding the presence of an object from neural-population activity, paragraph 4; Supplementary Figure 5EHow it will be judged
Labels of the testing samples shuffled 50 times per session (Methods, Shuffling trial labels); session values are the means over shuffles. Tolerances are the SDs the legend gives. Reproduced when every bin's mean is within 0.49 to 0.51 and no one-sample t test against 0.5 gives p < 0.05.
- figs5f-random-partition
Partitioning training and testing samples at random gave the same decline of accuracy with distance.
The results from this decoder showed the same decline in decoding accuracy as a function of distance from the object (Figure S5F)
Results, Decoding the presence of an object from neural-population activity, paragraph 4; Supplementary Figure 5FHow it will be judged
The distance decoder with training and testing timepoints drawn by a random permutation (Methods, Random partitioning). The timepoints are the 1 s spike-count vectors (answered 2026-10-21); the random partitioning is "analogous to the shuffling procedure above", where the permuted timepoints are "N×1 spike count vectors" (Methods, Shuffling trial labels). Training rate maps are built from the position and spike samples within the training windows. Reproduced when the Kruskal-Wallis test gives p < 0.05 with accuracy highest in bin 1 and lowest in bin 5.
- figs5ghi-fields-and-rates
Firing fields were spread across the arena, not concentrated at the object, and firing rates were similar between NO and O and highest at middle distances (bin 3).
across all cells, firing fields were widely distributed across the environment and not concentrated at the object location (Figures S5G and S5H), whereas firing rates were similar between trials, being largest at middle distances from the object (peaking in distance bin 3; Figure S5I)
Results, Decoding the presence of an object from neural-population activity, paragraph 4; Supplementary Figure 5G to 5IHow it will be judged
The paper gives no test. Field centroids in NO and O relative to the object (Supplementary Figure 5G); per session, the mean firing rate of each cell in each of the 5 equal-sample distance bins of the whole trial, averaged over cells, then over sessions. Reproduced when (i) the proportion of fields per unit area within 20 cm of the object is not higher in O than in NO by a two-sided rank-sum test over sessions at p < 0.05 (the paper's test for unpaired data); (ii) the mean rate peaks in bin 2, 3 or 4 in both trials, and in bin 3 in at least one; (iii) no bin shows a NO against O difference by signed-rank test at p < 0.05. Partially reproduced when two of the three hold.
- figs6-behaviour-decoders-flat
Decoders of trial type based on speed, head direction or occupancy performed only slightly above 50% and their accuracy did not depend on distance from the object.
Decoders for speed, head direction, and occupancy performed only slightly above 50% correct, with accuracies independent of object distance (Figures S6A, S6C, and S6E)
Results, Behavioral analyses, paragraph 1; Supplementary Figure 6A, 6C and 6EHow it will be judged
A claim of absence, judged on the paper's tests. Behavioural decoders as in the Methods (Behavioral analyses I), with gamma likelihoods for speed (Lowess-smoothed, 25 Hz, all speeds), two-Gaussian mixtures for head direction, and occupancy ratios, 50 x 50 spatial bins, same training and testing split and distance bins. The Lowess span is not given; it cannot change the absence much, and a span of 0.5 s is used as a standard choice. Reproduced when all three Kruskal-Wallis tests give p >= 0.05; not reproduced when any test gives p < 0.05. The paper gives no number for "slightly above 50%", so the median accuracy of each decoder is reported, not judged.
- figs6hij-behaviour-matched
After matching the distributions of speed, head direction or occupancy between NO and O within each distance bin, decoding accuracy on neural data still fell with distance.
We still observed a graded decrease in decoding accuracy for the neural activity decoder after matching distributions for speed (Figure S6H), head direction (Figure S6I) and occupancy (Figure S6J)
Results, Behavioral analyses, paragraph 2; Supplementary Figure 6H to 6JHow it will be judged
Matching as in the Methods (Behavioral analyses II), speed bins of 5 cm/s to 100 cm/s, head direction bins of pi/10, occupancy bins of 30 cm, random removal of the excess samples from the trial with more. The samples matched are the 1 s testing samples (the decoder's samples), with the behaviour averaged over the second; the Methods do not say, so a second reading matches 25 Hz position samples and keeps the 1 s bins whose samples all survive. Reproduced when, under one reading, each of the three Kruskal-Wallis tests gives p < 0.05 with bin 1 highest.
- figs6k-uniform-prior
Replacing the occupancy prior by a uniform prior did not change the decline of decoding accuracy with distance.
removal of this prior had no effect on our results (Figure S6K)
Results, Behavioral analyses, paragraph 2; Supplementary Figure 6KHow it will be judged
Distance decoder with a uniform prior over the spatial bins of each distance bin (Methods, Removing effects of prior). Reproduced when the Kruskal-Wallis test gives p < 0.05 with bin 1 highest.
- figs6l-sampling-thresholds
Using only spatial bins visited for at least 100, 500, 1000 or 2000 ms in both trials, the decline of accuracy with distance remained.
Regardless of the threshold value, the result was the same as before, with accuracy declining along with object distance (Figure S6L
Results, Behavioral analyses, paragraph 3; Supplementary Figure 6LHow it will be judged
Distance decoder with rate-map bins of 10 cm (the Methods' "10 cm2" read as 10 cm sides, as 3 cm2 is used for the 3 cm bins), spatial bins below the threshold in either trial left out of both. The Methods do not say whether the sampling time is that of the whole trial or of the training half, nor whether testing samples in a left-out bin are kept (answered 2026-10-21; the code holds no sampling analysis); the four combinations are readings. Reproduced when, under one reading, all four Kruskal-Wallis tests give p < 0.05 with bin 1 highest.
- tables1-without-bin-1
The effect of distance on decoding accuracy held when the bin nearest the object was left out.
all the main statistics hold even after excluding distance bin 1
Supplementary Figure 6L legend, last sentence; Supplementary Table 1, first rowHow it will be judged
The distance decoder results of fig4b-distance-decline, bins 2 to 5 only. The table's p-value has lost its exponent in the text; it was read from PDF page 49. Reproduced when the Kruskal-Wallis test gives p < 0.05 with bin 2 highest; H is compared with the expected value.
- fig5b-words-more-informative
Population words (patterns of spiking and silence in 10 ms bins) carried much more information about the trial than population spike counts.
Words were much more informative than counts (Figure 5B)
Results, Objects are encoded by distinct patterns of joint activity, paragraph 1; Figure 5BHow it will be judged
Methods, Definition of words and counts, and Mutual information. Spike counts in 10 ms bins; tetrode sessions with all cells, Neuropixels sessions subsampled to 30 cells; the all-zero word and zero count left out; mutual information with trial, from frequency distributions, minus the mean over 50 label shuffles. Distance bins as in Figure 4B (Figure 5B legend), each 10 ms bin taking the distance of the nearest position sample, 5 equal-sample bins per trial. Open readings (a) one random subset of 30 Neuropixels cells; (b) the mean over 20 subsets (the number the Methods use for subsampling the decoder). Which samples are compared is open; readings (c) per distance bin, 16 against 16 sessions; (d) all bins pooled. Reproduced when, under one combination, every rank-sum test gives p < 0.05 with words higher.
- fig5bc-words-information-falls-with-distance
The information from words decreased with distance from the object, as decoding did.
Importantly, the amount of information gained from words decreased as a function of distance from the object (Figures 5B and 5C)
Results, Objects are encoded by distinct patterns of joint activity, paragraph 1; Figures 5B and 5CHow it will be judged
Mutual information of words as in fig5b-words-more-informative, under its subset readings. Reproduced when, under one reading, the Kruskal-Wallis test across the 5 bins gives p < 0.05 with bin 1 highest.
- fig5b-counts-uninformative
The information from counts did not depend on distance and was not significantly above zero.
In contrast, the amount of information from counts did not depend on object distance ... and was not significantly above zero
Results, Objects are encoded by distinct patterns of joint activity, paragraph 1; Figure 5BHow it will be judged
A claim of absence, judged on the paper's tests. Bias-subtracted mutual information of counts as in fig5b-words-more-informative. Reproduced when the Kruskal-Wallis test gives p >= 0.05 and no per-bin two-sided one-sample t test against zero gives p < 0.05; not reproduced otherwise. Values given either way.
- fig5e-kl-divergence
The KL divergence between the word distributions of NO and O fell with distance from the object, from bin 1 to bin 5, while shuffled data showed no decrease.
divergence between the NO and O distributions became smaller with increasing object distance (Figure 5E, blue curve) ... No decrease in the KL divergence was present in shuffled data
Results, Objects are encoded by distinct patterns of joint activity, paragraph 3; Figure 5EHow it will be judged
Methods, Kullback-Leibler divergence, words and distance bins as in fig5b-words-more-informative, D(NO||O) over all words, probabilities as counts over all timepoints. Words seen in NO and never in O make the sum infinite, and the paper does not say how that was handled; the one finite reading its text allows is to sum over words seen in both trials, recorded as a deviation. Shuffled data assign trial labels to words at random (Figure 5E legend). Reproduced when the Kruskal-Wallis test and the bin 1 against bin 5 signed-rank test give p < 0.05 with bin 1 higher, and the shuffled divergences show no Kruskal-Wallis effect at 0.05.
- fig5-entropy
The entropy of words was higher than that of counts (1.65 against 0.84 bits).
the entropy was higher for words than counts (entropy of words: 1.65 bits ± 0.06; entropy of counts: 0.84 bits ± 0.02; mean ± SEM)
Results, Objects are encoded by distinct patterns of joint activity, paragraph 4How it will be judged
Entropy as in the Methods, words and counts as in fig5b-words-more-informative. The SEM is over an unstated unit, so the tolerance is 10%. The unit and whether the all-zero word is in are open; readings (a) per session and trial over the whole trial, all-zero word left out (as in the mutual information); (b) as (a) with it included; (c) per session, trial and distance bin. Reproduced when, under one reading, both means are within tolerance.
- figs7cd-number-of-words
The number of distinct words did not differ between trial types or across distance bins.
In contrast to the identity of words, the number of words was independent of trial type and distance from the object (Figures S7C and S7D)
Results, Objects are encoded by distinct patterns of joint activity, paragraph 3; Supplementary Figure 7C and 7DHow it will be judged
A claim of absence, judged on the paper's tests (from the Supplementary Figure 7 legend, an image on PDF page 46, not in the text). Distinct words per session, trial and distance bin as in fig5b-words-more-informative. The values of the rank-sum test are not given (answered 2026-10-21); the legend says "n = 16 experiments" and Supplementary Figure 7C and 7D plot only values per session and distance bin. Readings (w-a) each session's distinct words over the whole trial, 16 against 16; (w-b) the 80 session and distance bin values of each trial. Reproduced when, under one reading, all three tests give p >= 0.05, not reproduced when any gives p < 0.05.
- figs7e-downsampled-words
After downsampling words to as many states as counts, words still carried more information than counts.
after downsampling the number of states to be equal for words and counts, words were still much more informative (Figure S7E)
Results, Objects are encoded by distinct patterns of joint activity, paragraph 4; Supplementary Figure 7EHow it will be judged
Methods, Downsampling words. Per session and distance bin, keep the c most frequent words, c the number of distinct counts, and renormalise. Whether the bias subtraction is applied is not said; readings (a) with, (b) without. Reproduced when, under one reading, a signed-rank test over the 80 pairs gives p < 0.05 with words higher.
- fig6ab-single-cell-distance-tuning
Single cells decoded with almost flat accuracy across distance; their distance tuning (bin 1 minus bin 5 accuracy) averaged 0.02, 98.7% lay between -0.2 and 0.2, 12.2% (145) had significant positive and 3.8% (45) significant negative tuning.
CA1 cells had a slight positive bias in distance tuning (mean: 0.02; for reference, the same mean calculated from populations is 0.19; Figure 4C) and nearly all cells were contained between the bounds ... and 0.2 (98.7% of cells) ... some cells had significant positive distance tuning (12.2%, n = 145/1,189 cells) ... and some cells had significant negative distance tuning (3.8%, n = 45/1,189 cells)
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 2; Figure 6A and 6BHow it will be judged
The distance decoder run with one cell at a time; distance tuning is accuracy in bin 1 minus bin 5 (Equation 1). Significance from 100 circular shifts of the spike train by 20 s to the trial length less 20 s, above the 99th or below the 1st percentile (Methods). The tolerance on the within-bounds fraction is 10% and stops at 1. The mean of 0.02 is printed to one significant figure, so the tolerance on it is 0.005 (half the last digit) in place of 10%. Which cells are open; readings (a) all 1,189 unique cells, each in its session; (b) the cells of the decoding sessions (Figure 6A legend). Reproduced when, under one reading, all four values are within tolerance; partially reproduced when the mean and within-bounds fraction are and a significant fraction is not.
- figs5l-without-positively-tuned
After removing the cells with significant positive distance tuning, decoding accuracy still fell gradually with distance.
after removing cells with significant positive distance tuning (12.2%), we still observed a gradual drop in decoding accuracy ... obtained with population data as a function of object distance (Figure S5L)
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 2; Supplementary Figure 5LHow it will be judged
Distance decoder on the decoding sessions without the cells found in fig6ab-single-cell-distance-tuning. Reproduced when the Kruskal-Wallis test gives p < 0.05 with bin 1 highest.
- figs5m-without-object-tuned
After removing all object-tuned cells, decoding accuracy still fell gradually with distance.
After removing all object-tuned cells (19.9%) from the data and re-running the decoder, we observed the same gradual fall in decoding accuracy as a function of distance (Figure S5M)
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 5; Supplementary Figure 5MHow it will be judged
Distance decoder on the decoding sessions without the object-tuned cells of fig1c-object-tuned-fraction (each template-width reading). Reproduced when, under one reading, the Kruskal-Wallis test gives p < 0.05 with bin 1 highest.
- fig6c-subsets-decline
In the 620-cell session, random subsets of every size except a single cell showed decoding accuracy falling with distance.
All curves showed reduced decoding accuracy as a function of distance, except the curve constructed using only a single cell (Figure 6C)
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 3; Figure 6CHow it will be judged
Distance decoder on 20 random subsets per size of the 620 cells (Methods, Subsampling for decoder). The sizes are open, as three lists are given; readings (a) Results, 1, 5, 10, 30, 50, 100, 200, 300, 400, 500; (b) legend, 1, 5, 20, 30, 50 to 500; (c) Methods, 1, 5, 10, 30, 100 to 500. The test is open, as the printed W values are not integers; readings (d) signed-rank of bin 1 against bin 5 over the 20 subsets; (e) Kruskal-Wallis across the 5 bins over the 20 subsets. Reproduced when, under one reading of the test, the single-cell curve gives p >= 0.05 and every other size gives p < 0.05 with bin 1 higher.
- fig6d-reorganization-grows-with-cells
The difference between bins 1 and 5 grew, and the slope of accuracy over distance became more negative, as more cells were used, both reaching their extremes at 500 cells.
The difference between bins #1 and #5 grew systematically with the number of cells (Figure 6D; left y axis, blue)
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 3; Figure 6DHow it will be judged
The 20 subsets per size of fig6c-subsets-decline; per subset, accuracy in bin 1 minus bin 5 and the least-squares slope over bin indices 1 to 5 (Methods, Measuring strength). Reproduced when, under one size reading, both Kruskal-Wallis tests across sizes give p < 0.05 and the largest mean difference and most negative mean slope are at 500 cells, as the Results say.
- figs8-smaller-datasets
The growth of distance reorganization with the number of cells held in the tetrode sessions and in the 135-cell Neuropixels session.
The finding was upheld in data from animals with smaller numbers of cells (135 cells in a Neuropixels recording; 25–39 cells in tetrode recordings)
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 3; Supplementary Figure 8A to 8DHow it will be judged
As fig6d-reorganization-grows-with-cells, on the tetrode decoding sessions pooled (sizes up to each session's count) and on the 135-cell session. Subset sizes for the tetrode sessions are not given beyond n = max (29 to 35 cells); sizes from reading (a) of fig6c-subsets-decline up to each session's count, then the session's full count, are used. Reproduced when all four Kruskal-Wallis tests give p < 0.05 with the difference rising and the slope falling with size.
- fig6ef-least-tuned-cells
Even the 20% of cells with the lowest distance tuning showed distance reorganization, with a bin 1 minus bin 5 difference of 0.121 and a slope of -0.0267.
even the 20% of cells with lowest tuning was sufficient to demonstrate the distance modulation, in the form of positive distance tuning as well as a negative slope (Figures 6E and 6F; rightmost point on blue curve: 0.121; on orange curve
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 4; Figure 6E and 6FHow it will be judged
Cells of the 620-cell session sorted by single-cell distance tuning (fig6ab-single-cell-distance-tuning), the lowest 100, 80, 60, 40 and 20% kept, distance decoder run on each. Reproduced when both 20% values are within tolerance; partially reproduced when the difference is positive and the slope negative but outside tolerance.
- fig7-single-cell-properties
Single-cell properties barely predicted distance tuning; correlations with object-tuning score (r = 0.12), spatial information (r = -0.13), spatial correlation (r = -0.08) and rate overlap (r = -0.18) were significant but weak.
The object-tuning score had a significant positive correlation with distance tuning, but very weakly so (r = 0.12 ... Spatial information had a significant negative correlation, but also this was very weak ... spatial correlation and rate overlap were also negatively but weakly correlated with distance tuning
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 5; Figure 7A to 7DHow it will be judged
Pearson correlations over cells of distance tuning with the object-tuning score (opening comment), spatial information in bits per spike in the O trial (Methods), and NO-O spatial correlation and rate overlap. The mean rate lambda is not defined beyond "the mean firing rate in the trial" (answered 2026-10-21); readings (si-a) the trial's speed-filtered spikes over speed-filtered time; (si-b) the occupancy-weighted mean of the smoothed map, sum of Pi li with Pi the occupancy over the trial's duration. The r values are printed to two digits, so a tolerance of 0.05 is used in place of 10%, being half a unit in the first digit at the scale the claim calls weak. Reproduced when, under one cell reading of fig6ab-single-cell-distance-tuning and one lambda reading, all four r are within 0.05 of the printed values with the same sign and p < 0.05, and each |r| is below 0.3.
- fig7-slope-variant
Using the slope of accuracy against distance in place of distance tuning gave similar weak correlations with the same four properties.
We obtained similar results when, instead of distance tuning, we used the slope of the regression line fitted to decoding accuracy vs. distance as a measure of reorganization
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 5How it will be judged
As fig7-single-cell-properties with the single-cell least-squares slope. Reproduced when all four r are within 0.05 of the printed values with the same sign. The printed sign for spatial correlation (negative for both measures, while the others flip) is noted.
- figs5jk-walls-no-pivot
The arena walls did not act as a pivot as the object did; position decoding error and population vector correlations were comparable near (under 15 cm) and far from the walls.
The internal walls of the recording environment did not play a similar role as the free-standing object, since position decoding accuracy and population vector correlations were comparable near and far away from the walls (Figures S5J and S5K)
Results, Decoding the presence of an object from neural-population activity, paragraph 5; Supplementary Figure 5J and 5KHow it will be judged
A claim of absence. The paper reports no test; its test for paired data is the Wilcoxon signed-rank test (Methods, Statistical tests), applied here over the decoding sessions, near against far, at alpha 0.05. Near is under 15 cm from a wall, far more than 15 cm (Supplementary Figure 5J legend). (J) Position decoding error in the NO trial, the Euclidean distance between the animal's position and the decoded position. The position decoder is not described in the Methods; the reading allowed is the paper's Bayesian decoder (Methods, Bayesian decoding algorithm) with the posterior over the 50 x 50 spatial bins of the NO trial, 1 s samples, the same training and testing halves, and the decoded position the bin of maximum posterior. Open, could change the outcome; reading (b) takes the posterior mean. (K) The population vector correlation (Methods, Population vector correlation) between rate maps of the first and second halves of the NO trial, mean over the near and over the far spatial bins. Reproduced when neither signed-rank test gives p < 0.05 under any reading; not reproduced when one does. Differences are given either way.
- figs8eh-least-tuned-smaller-datasets
In the tetrode sessions and the 135-cell Neuropixels session too, keeping only the cells with the lowest distance tuning weakened but did not at first remove the distance reorganization; the slope reached zero when 60% (tetrode) or 50% (135-cell session) of the least-tuned cells were kept.
As before, the finding also held in smaller-sized tetrode and Neuropixels data (Figures S8E–S8H)
Results, Distance-specific reorganization of the hippocampal map is widely distributed, paragraph 4; Supplementary Figure 8E to 8HHow it will be judged
As fig6ef-least-tuned-cells (100, 80, 60, 40 and 20% of cells with the lowest single-cell distance tuning kept), on the tetrode decoding sessions pooled (Supplementary Figure 8E, 14 sessions) and on the 135-cell session. The tolerance is one step of the kept fraction (the paper's grid is in steps of 20%), not 10%. Reproduced when, for both datasets, the 100% curve has a positive bin 1 minus bin 5 difference and a negative slope, both shrink as fewer cells are kept, and the largest kept fraction with a mean slope of zero or above is within tolerance. The Results say the 20% finding "held", while the legends say the slope reached zero at 0.6 and 0.5; the 20% values are reported.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none