Claims 54
From the paper Causal evidence of a line attractor encoding an affective state. · claims file · checked 2026-10-22: ready
- ed1n-engaging-x1-slower
In freely behaving mice engaging in aggression, rSLDS fits show an integration dimension x1 with a longer time constant than the orthogonal dimension x2, an approximate line attractor.
revealed an approximate line attractor under both conditions, exhibiting ramping and persistent activity aligned and maintained across both performed and observed attack sessions
Extended Data Fig. 1n and 1o; Results, Line attractor for observing aggression, paragraph 1How it will be judged
For each mouse, an rSLDS with 2 latent dimensions fit to the engaging session (number of states by fivefold cross-validation, Methods, or 3 states with the sample code's settings; both readings run). The cvR2 of each fit is reported; no mouse is excluded by the Methods' 70% rule, since the paper reports n = 5 in both conditions while the mean observation cvR2 it gives (67.2%) is below 70%. Time constants from the eigenvalues of each state's dynamics matrix as the sample code computes them; x1 is the dimension with the larger time constant in the state that holds the slow dynamics (the state with the largest time constant). Reproduced when the x1 time constants exceed the x2 time constants across the 5 mice by a two-tailed unpaired Mann-Whitney test at p < 0.05. The line attractor score, log2 of the largest over the second largest time constant (Methods), is reported per mouse.
- ed1p-observing-x1-slower
In the same mice observing aggression, rSLDS fits also show an x1 dimension with a longer time constant than x2.
revealed an approximate line attractor under both conditions
Extended Data Fig. 1p and 1q; Results, Line attractor for observing aggression, paragraph 1How it will be judged
As ed1n-engaging-x1-slower, on the observation sessions.
- ed1m-x1-observe-attack-vs-sniff
In freely behaving mice, average x1 activity differs between sniffing the SW mouse and observing the SW mouse attack a BALB/c intruder.
Average activity in the x 1 dimension during sniffing of the SW mouse, vs observing the SW mouse a BALB\c intruder (n = 4 mice, *p = 0.0286, Two-tailed Mann Whitney U -test
Extended Data Fig. 1mHow it will be judged
The legend is garbled and the panel was not read, so the direction is not fixed from the text. Per mouse, the mean of the x1 projection of the observation session over frames annotated as sniffing the SW mouse and over frames annotated as observing attack. Reproduced when a two-tailed unpaired Mann-Whitney test over mice gives p < 0.05; the direction is reported. The legend's n of 4 (of 5 mice) is not explained; all mice with both annotations are used and the count is reported.
- ed3c-x1-activity-observe-vs-engage
The activity of x1 neurons during observed attack bouts does not differ from their activity during engaged attack.
comparison of the activity during observing attack bouts and engaging in attack (n = 5 mice, p = 0.42, Two-tailed Mann-Whitney U -test
Extended Data Fig. 3cHow it will be judged
A claim of absence, judged on the paper's test. Per mouse, the mean activity of the x1 neurons of each condition during attack bouts (observed or engaged). Reproduced when a two-tailed Mann-Whitney test over mice gives p >= 0.05; not reproduced when p < 0.05. The difference is reported either way. Which neurons are x1 neurons is not fixed beyond the rSLDS weights; the readings are the five neurons with the largest absolute x1 weight (as in Fig. 1i) and the neurons the dataset marks as x1, if it does.
- ed3e-achw-full-session
Over the full interaction, autocorrelation half-widths of x1 neurons are similar when observing (25 +/- 0.8 s) and engaging in aggression (20 +/- 1.7 s).
mean achw during observation: 25 ± 0.8s, mean achw during engagement: 20 ± 1.7s, n=5 mice, p = 0.125
Extended Data Fig. 3eHow it will be judged
Autocorrelation per neuron as in Methods equations 7 and 8 (the sample code's autocorr_matlab). The half-width is not defined beyond its name; the reading is the first lag at which the autocorrelation falls to 0.5 or below, in seconds at 10 Hz. Per mouse, the mean over x1 neurons (readings of x1 neurons as in ed3c). The "full interaction" is not bounded in the paper; readings are the whole session and the span from the first intruder entry to the last removal. Expected values within tolerance, and, as a claim of absence, a two-tailed test over mice with p >= 0.05, run as an unpaired Mann-Whitney test and as a paired signed-rank test (header, Tests; the printed 0.125 is an exact signed-rank value).
- ed3f-achw-after-removal
Aligned to removal of the last intruder, autocorrelation half-widths of x1 neurons are similar when observing (14 +/- 1 s) and engaging (11 +/- 1.6 s).
mean achw during observation: 14 ± 1s, mean achw during engagement: 11 ± 1.6s, n = 5 mice, p = 0.187
Extended Data Fig. 3fHow it will be judged
As ed3e-achw-full-session, on activity from the removal of the last intruder to the end of the session (the paper does not give the window's end; the session end is the only reading the legend allows), with both test readings (the printed 0.187 is the exact signed-rank value 0.1875 at n = 5).
- ed3g-decode-observed-attack
A decoder trained on x1 activity during observation separates observed attack bouts from non-attack bouts better than shuffled controls.
A SVM decoder was trained on data from integration dimension activity to separate bouts of observing attack from non attack bouts. Right: Quantification of the decoder accuracy performance (n = 5 mice, p = 0.0079
Extended Data Fig. 3gHow it will be judged
As the Methods describe (Decoding behaviour from integration dimension). Attack bouts separated by less than 5 s merged into trials, classes balanced; a one-dimensional threshold on the normalized x1 signal fit by a linear SVM on training frames and scored on held-out frames; shuffles assign the behaviour labels randomly to bouts, 20 times, and 1,000 times per mouse for the significance test (Methods). The Methods report accuracy "on data from all other trials for each mouse" without giving the split further; the readings are training on one trial and testing on all others, leave-one-trial-out, and fivefold over trials. Reproduced when decoder accuracy exceeds shuffled accuracy across mice by a two-tailed Mann-Whitney test at p < 0.05.
- ed3h-decode-engaged-from-observed
The decoder trained on observing attack decodes engaged attack from x1 activity better than shuffled data, so x1 encodes a similar variable in both conditions.
could be reliably used to decode from held-out data instances of both observation of and engagement in attack, suggesting that this dimension encodes a similar internal state variable under both conditions
Extended Data Fig. 3h; Results, Line attractor for observing aggression, paragraph 1How it will be judged
The decoder of ed3g-decode-observed-attack applied to the engaging session's x1 signal (frames of attack against all others); the paper does not say whether the engaging session is projected with the observation model's weights or its own; both readings run. Reproduced when accuracy exceeds that of the shuffled decoder across mice by a two-tailed Mann-Whitney test at p < 0.05.
- ed4c-x1-neuron-overlap
Around 70% of the neurons that contribute to the line attractor overlap between observing and engaging in aggression.
suggesting that a highly overlapping set of neurons (70%) contributes to line-attractor dynamics during observing or engaging in attack
Extended Data Fig. 4c; Results, Line attractor for observing aggression, paragraph 1How it will be judged
The paper does not say how a neuron is counted as contributing, or the denominator. Readings for contributing are (a) the top five neurons by absolute weight, as in Fig. 1i, and (b) the neurons the dataset marks as x1 neurons, if it does. Readings for the overlap are the shared fraction of (i) the observation set and (ii) the engagement set. Reproduced when one reading gives a mean within tolerance.
- ed4d-weight-dot-product
The x1 weight vector during observation is more aligned with the x1 weight vector during engagement than with the x2 weight vector.
rSLDS weights of the x 1 dimension during observation were compared to model weights of the x 1 and x 2 dimensions during engagement using a dot product of the two weight vectors. (n = 5 mice, ***p = 0.0079
Extended Data Fig. 4dHow it will be judged
Per mouse, the dot product of the observation x1 weight vector with the engagement x1 vector and with the engagement x2 vector, over neurons present in both sessions. The paper does not say whether vectors are normalized or signs aligned; the readings are the raw dot product and the absolute dot product of unit vectors. Reproduced when x1-x1 exceeds x1-x2 across mice by a two-tailed Mann-Whitney test at p < 0.05 under one reading.
- ed4g-x2-more-spontaneous
During the solitary baseline, x2 neurons show more frequent spontaneous calcium transients than x1 neurons (2.3 against 1.6 events).
we did see a slightly but significantly higher frequency of spontaneous calcium transients in x 2 neurons
Extended Data Fig. 4g; Results, Line attractor for observing aggression, paragraph 2How it will be judged
A transient is a crossing above 1.5 standard deviations of the z-scored activity (legend) during the 5 min baseline at the start of the session (Methods, Behaviour annotations). The paper does not say per what time the frequency is, which session's baseline, or which model's x1 and x2 neurons are used; readings are events per baseline and per minute, and the engaging and observing sessions' baselines and models. Reproduced when x2 exceeds x1 across mice by a two-tailed Mann-Whitney test at p < 0.05; the values are compared with the expected ones under each reading.
- ed4hi-amplitude-tau-similar
Baseline transient amplitude (0.58 against 0.71 z) and decay constant (1.7 against 2.3 s) do not differ between x1 and x2 neurons.
We did not detect a difference in amplitude or decay constant (tau) between x 1 and x 2 neurons
Extended Data Fig. 4h and 4i; Results, Line attractor for observing aggression, paragraph 2How it will be judged
A claim of absence, judged on the paper's test, two-tailed over mice for each measure, run as an unpaired Mann-Whitney test and as a paired signed-rank test (header, Tests; the printed 0.188 is the exact signed-rank value 0.1875 at n = 5); reproduced when neither measure gives p < 0.05. Transients as in ed4g-x2-more-spontaneous. The decay constant fit is not described; the reading is an exponential fit to the mean transient from its peak.
- fig1e-2p-x1-slower
In head-fixed mice observing aggression, rSLDS fits to 2P data show an x1 dimension with longer time constants than x2.
during observation of aggression by the head-fixed mice, rSLDS analysis identified an integration dimension with slow dynamics ( x 1 ) aligned to an approximate line attractor, and an orthogonal dimension with faster dynamics ( x 2 )
Figure 1e; Results, Line attractor for observing aggression, paragraph 4How it will be judged
As ed1n-engaging-x1-slower, on the 2P observation sessions, with the sample code's 2P settings (2 states) as the second reading. Methods add that 2P flow fields were fit with an input term for the presence of the BALB/c intruder; fits with and without that input are both run. The legend's P is read from its star key (four stars, P < 0.0001), the only test in the figure besides panel m.
- fig1f-line-attractor-score
The line attractor score is positive across the 9 head-fixed mice.
How it will be judged
Score = log2 of the largest time constant over the second largest (Methods, Calculation of line attractor score) of each mouse's fit. The paper gives no threshold; the readings are (a) the mean over mice above 0 and (b) every mouse above 0. Reproduced when both hold; partially reproduced when only (a) holds.
- x1-time-constant-range
The experimentally observed network time constants of the integration dimension lie in a range of about 50 to 200 s.
could yield network time constants ( τ n ) in the experimentally observed range (~50–200 s
Results, Line-attractor neurons form ensembles, paragraph 5How it will be judged
A reported range; reproduced when the x1 time constants of the 9 mice lie within 50 to 200 s widened by 10% of each bound (45 to 220 s); partially reproduced when the median lies inside. The paper does not say which fits the range comes from; readings are the 2P observation fits (9 mice) and these together with the miniscope fits. The expected value is the midpoint and is not itself tested.
- methods-subspace-angle
The x1 and x2 dimensions are nearly orthogonal, with a subspace angle of 88.1 +/- 0.87 degrees across 9 mice.
we initially confirmed the orthogonality of these dimensions by computing the subspace angle between them (88.1 ± 0.87°, n = 9 mice)
Methods, Dynamical system models of neural data, paragraph 2How it will be judged
The angle between the two columns of C (the stored factor weights), per mouse, averaged over the 9 2P mice.
- methods-cvr2
rSLDS models of miniscope data explain 84.7% of variance during engagement and 67.2% during observation of aggression (cross-validated R2).
Models fit to miniscope data during engagement of aggression obtained a cvR 2 = 84.7 ± 0.03%, while the same model explains 67.2 ± 0.02% of variance in data obtained from observation of aggression
Methods, Dynamical system models of neural data, paragraph 2How it will be judged
Fivefold cross-validation, training on four portions and testing on the fifth (Methods). The printed spreads (0.03% and 0.02%) are read as fractions (3 and 2 percentage points), the only reading that fits a spread over 5 mice, and are used as tolerances. The paper does not say whether the portions are contiguous blocks or random frames, or how R2 is taken (neural activity predicted from inferred latents); readings are contiguous blocks and random frames. The Methods also require 70% cvR2 before downstream use, which the observation value does not meet; this is reported.
- fig1m-achw-x1-longer
Neurons contributing to x1 have longer autocorrelation half-widths than those contributing to x2 in head-fixed observation.
Neurons contributing to the integration dimension displayed more persistence than those aligned with the faster dimension
Figure 1l and 1m; Results, Line attractor for observing aggression, paragraph 4How it will be judged
Five neurons per dimension per mouse (45 each over 9 mice, legend of Fig. 1l), the five with the largest absolute weight on that dimension (Fig. 1i). ACHW as in ed3e-achw-full-session, on the observation session. Per-mouse means compared by a two-tailed test, unpaired Mann-Whitney and paired signed-rank (header, Tests; 0.0078 at n = 9 is an exact value of both); reproduced when x1 exceeds x2 at p < 0.05.
- ed5e-motion-explains-little
Only a small fraction of neural activity is explained by the head-fixed observer's movements, and the variance explained does not differ between x1 and x2 neurons.
Importantly, only a small fraction of the neural activity could be explained by movements of the observer mouse
Extended Data Fig. 5a-e; Results, Line attractor for observing aggression, paragraph 4How it will be judged
GLM of each neuron on the first ten principal components of 500 motion SVDs at lags up to 10 s, ridge, tenfold cross-validation, cvR2 (Methods). The Methods call the ten features "binary" without saying how they are binarized; readings are the continuous components and components thresholded at their median. Reproduced when the median cvR2 over neurons is below 0.1 (the paper gives no value; "small fraction" read as under 10% of variance, a reading named here and reported with the full distribution) and a two-tailed test of per-mouse x1 against x2 means gives p >= 0.05, run as an unpaired Mann-Whitney test and as a paired signed-rank test (header, Tests; the printed 0.8125 is an exact signed-rank value at n = 7). Needs the face videos or their SVDs.
- fig2d-x1-integration
Photostimulation of five x1 neurons with four pulses at 20 s ISI produces integration in the x1 dimension, with ISI activity rising with each pulse and not returning to baseline.
optogenetic reactivation of cohorts of five individual x 1 neurons yielded robust integration of activity in the entire x 1 dimension-weighted population, as evidenced by progressively increasing peak activity during the 20 s ISI after each consecutive pulse
Figure 2c and 2d; Results, Holographic activation shows integration, paragraph 2How it will be judged
The x1 projection of the x1-activation session (sample code). Per mouse, the mean over baseline and over each ISI under readings Wa and Wb; Kruskal-Wallis with Dunn's correction under readings Ca and Cb (header). Reproduced when, under one reading, at least three Dunn comparisons give p < 0.05 and the mean over mice rises from the first to the third ISI. Partially reproduced when the rise holds but fewer comparisons are significant.
- fig2gh-x1-moves-along-attractor
x1 photostimulation moves the population activity vector progressively along the line attractor; the state-space distance from the initial point grows with each pulse, both at the end of stimulation and at the end of each ISI.
This revealed that the x 1 perturbations resulted in progressive, stable on-manifold movement along the attractor with each consecutive stimulation, as measured by the increase in both metrics
Figure 2e-h; Results, Holographic activation shows integration, paragraph 3How it will be judged
State = (x1 projection, x2 projection) of the x1-activation session. Euclidean distance from t_initial to t_stim-end (onset + 2 s) and to t_post-stim (the frame before the next onset; for the last pulse, onset + 21 s). t_initial is not fixed beyond "during the baseline period"; readings are the mean baseline state and the last baseline frame. What the distances are compared with is not given; readings are (a) each pulse against the distances of baseline frames from t_initial and (b) each pulse against the first. Kruskal-Wallis with Dunn's correction. Reproduced when, under one reading, the mean distance rises from pulse 1 to pulse 3 for both metrics and three of four comparisons give p < 0.05 for each metric, as the legends report.
- ed7f-activated-x1-natural-levels
Photoactivated x1 neurons reach peak activity comparable to their peak during observation of aggression.
Activated x 1 neurons exhibited activity levels comparable to their response during observation of aggression
Extended Data Fig. 7d-f; Results, Holographic activation shows integration, paragraph 2How it will be judged
A claim of absence, judged on the paper's test. Per mouse, the mean over the five targeted x1 neurons of their peak z-score during observation and during the activation session; reproduced when a two-tailed Mann-Whitney test gives p >= 0.05. The difference is reported.
- ed8a-x1-integration-8s
x1 activation with an 8 s ISI also produces integration in the x1 dimension.
Similar results were obtained using an 8 s ISI
Extended Data Fig. 8a; Results, Holographic activation shows integration, paragraph 2How it will be judged
As fig2d-x1-integration, on the 8 s ISI x1-activation session (onsets at frames 140, 240, 340, 440). Reproduced when, under one reading, three of four comparisons give p < 0.05 and the mean rises from the first to the third ISI.
- ed8ef-x1-scales-with-power
x1 integration scales with laser power, being larger with 5 mW than with 1 mW per neuron.
This activity also scaled with different laser powers
Extended Data Fig. 8e and 8f; Results, Holographic activation shows integration, paragraph 2How it will be judged
Per mouse and period (baseline and each ISI, readings Wa and Wb), the mean x1 projection at 1 mW and at 5 mW; per period, a two-tailed Mann-Whitney test. Reproduced when three of the four ISIs give 5 mW higher at p < 0.05 under one reading (the legend's four values have three below 0.05; the fourth is read as the baseline or one ISI).
- ed8c-model-integrates
Given the same pulsed input, the rSLDS model fit to observation data also integrates along x1, as in the data.
Providing the same (digital optogenetic) input to the fit rSLDS model also resulted in integration by the model along the x 1 dimension, similar to that observed in the data
Extended Data Fig. 8c; Results, Holographic activation shows integration, paragraph 2How it will be judged
The paper does not say how the input enters the model. Readings are (a) the input weight learned for the BALB/c-presence input of the observation fit (Methods, flow fields fit with input terms) driven by a 2 s pulse at each stimulation onset; (b) a pulse added to x1 at each onset, of the size of the first evoked x1 response. The model is simulated forward from the baseline state with its discrete states. Reproduced when the simulated x1 at the end of each ISI rises from the first to the third ISI in at least 6 of the 8 mice under one reading; the cross-validated R2 against the data (about 85%, Methods) is reported.
- ed8gi-x1-stim-no-x2
x1 stimulation does not evoke appreciable activity in the x2 dimension.
Importantly, x 1 stimulation did not evoke appreciable activity in x 2 dimension neurons
Extended Data Fig. 8g-i; Results, Holographic activation shows integration, paragraph 2How it will be judged
A claim of absence, judged on the paper's test. The x2 projection of the x1-activation session, per mouse means of baseline and ISIs (Wa, Wb), Kruskal-Wallis with Dunn's correction (Ca, Cb). Reproduced when no comparison gives p < 0.05 under the reading used for fig2d-x1-integration; the size of each difference is reported.
- fig2k-x2-no-integration
Photostimulation of five x2 neurons does not produce integration; x2 activity is transient and decays to baseline during each ISI.
Importantly, activation of x 2 neurons did not lead to integration
Figure 2i-k; Results, Holographic activation shows integration, paragraph 4How it will be judged
A claim of absence, judged on the paper's test. As fig2d-x1-integration, on the x2 projection of the x2-activation session; reproduced when no Dunn comparison gives p < 0.05.
- ed8b-x2-no-integration-8s
With an 8 s ISI, x2 activation also produces no integration in x2.
we observed stimulus-locked transient activity in the x 2 dimension followed by a decay back to the baseline during the ISI period, across stimulation paradigms
Extended Data Fig. 8b; Results, Holographic activation shows integration, paragraph 4How it will be judged
As fig2k-x2-no-integration, on the 8 s ISI x2-activation session.
- ed8jl-x2-stim-little-x1
x2 activation has little effect on x1 activity; in the paper's test two of four comparisons are not significant (P = 0.276) and two are (P = 0.0072 and 0.03).
with little to no effect on x 1 neurons
Extended Data Fig. 8j-l; Results, Holographic activation shows integration, paragraph 4How it will be judged
The x1 projection of the x2-activation session, tested as fig2d-x1-integration. The paper's own test finds two significant comparisons, so the claim is judged against that pattern. Reproduced when, under one reading, at most two comparisons give p < 0.05 and the mean ISI x1 response is smaller than under x1 activation (fig2d-x1-integration) in every ISI. The legend reports n = 8, though Fig. 2k has 7 mice with x2 activation; the number found is reported.
- fig2no-x2-off-manifold
x2 activation moves the population vector transiently off the attractor during each pulse, after which it relaxes back near its initial location.
we observed that x 2 neuron activation caused transient off-manifold movements of the PAV orthogonal to the attractor axis during each pulse ... After each stimulus, the PAV relaxed back into the attractor, near to the initial location that it occupied before the stimulus
Figure 2l-o; Results, Holographic activation shows integration, paragraph 4How it will be judged
As fig2gh-x1-moves-along-attractor, on the x2-activation session. Reproduced when, under one reading, at least three of four stimulation-end comparisons give p < 0.05 and at most one ISI-end comparison does (the legends' pattern), and the mean ISI-end distance is smaller than the mean stimulation-end distance.
- ed9d-attractor-finite
Integration of pulses saturates after the third pulse; a model with a finite attractor fits the x1 response better than an infinite one.
we found that integration of optogenetic stimulation pulses saturated in the x 1 dimension after the third pulse, suggesting that the line attractor occupies a finite portion of the neural state space
Extended Data Fig. 9a-d; Results, Holographic activation shows integration, paragraph 3How it will be judged
The two models are described only by their predictions (further pulses add no ramp once the system reaches a point, against each pulse adding the same step). Readings are (a) a step model fit to the x1 trace whose cumulative level is capped at a fitted maximum, against the same model uncapped; (b) the mean ISI levels compared directly, saturation being ISI4 not above ISI3 while ISI3 is above ISI2. Reproduced when under (a) the capped model has lower mean square error in the whole trace and after the third pulse by Mann-Whitney across mice at p < 0.05, or under (b) the pattern holds across mice.
- ed9i-points-along-line-attractive
x2 stimulation after x1 has moved the system along the attractor perturbs it off the line, after which it relaxes to the position reached after x1 stimulation and not to baseline.
This x 2 perturbation also resulted in transient off-manifold movements of the PAV orthogonal to the line attractor, followed by relaxation to the position occupied after the previous x 1 stimulation
Extended Data Fig. 9e-i; Results, Holographic activation shows integration, paragraph 4How it will be judged
In the combined session, the distance from baseline of the state after each stimulation (the legend's "activity distance from baseline after each photostimulation"). Reproduced when the state after x2 stimulation remains farther from baseline than the baseline state (Dunn p < 0.05 for the post-x1 point) and does not differ from the state before x2 stimulation (p >= 0.05). The time points are not given; readings are the frame before the next onset and the mean over each ISI (Wa).
- ed9mn-random-neurons-no-effect
Activating randomly selected neurons that are not weighted by either dimension does not produce activity along x1 or x2.
activation of randomly selected neurons that were not weighted by either dimension did not produce activity along either the x 1 or x 2 dimension
Extended Data Fig. 9j-n; Results, Holographic activation shows integration, paragraph 4How it will be judged
A claim of absence, judged on the paper's test. As fig2d-x1-integration on the x1 and on the x2 projection of the random-activation session; reproduced when no comparison gives p < 0.05 in either.
- ed5hj-no-behaviour-change
Activating x1 or x2 ensembles does not change the head-fixed mouse's movements.
Activation of either ensemble did not result in overt changes in the behaviour of the head-fixed mouse
Extended Data Fig. 5f-j; Results, Holographic activation shows integration, paragraph 4How it will be judged
A claim of absence, judged on the paper's test. Per mouse, the mean of each of the top five motion SVDs before and during stimulation; reproduced when no two-tailed Mann-Whitney test gives p < 0.05. Needs the face videos or their SVDs.
- ed6k-no-spatial-clustering
x1 and x2 neurons are not spatially clustered; distances within and between the two groups do not differ.
did not observe spatial clustering of x 1 or x 2 neurons
Extended Data Fig. 6i-k; Results, Holographic activation shows integration, paragraph 1How it will be judged
A claim of absence, judged on the paper's test. Per mouse, the mean pairwise distance within x1, within x2 and between x1 and x2 neurons, from cell centroids; reproduced when no Dunn comparison gives p < 0.05. Which neurons count as x1 and x2 is read as in ed3c-x1-activity-observe-vs-engage.
- ed7bc-no-spatial-decoding
A linear SVM cannot separate x1 from x2 neurons by position better than shuffled labels.
Accuracy of SVM decoder trained on data versus shuffled control (n=10 mice, p=0.156
Extended Data Fig. 7a-cHow it will be judged
A claim of absence, judged on the paper's test. Linear SVM on cell positions per FOV, shuffles by random assignment of identity, 20 repeats, mean accuracy (Methods, Spatial cluster decoder). The cross-validation is not given; readings are leave-one-out and fivefold. Reproduced when the two-tailed Mann-Whitney test of accuracy gives p >= 0.05. The n of 10 mice is more than the 9 of Fig. 1; the number found is reported.
- ed6g-offtarget-15um
Photostimulation through the GRIN lens is spatially confined; the response of a reference cell is close to zero when the spiral is 15 um from its soma.
Note that at 15 µm the average response in the reference cells is close to zero
Extended Data Fig. 6d-hHow it will be judged
Per reference cell (17 cells), the mean response at each offset from the soma. Reproduced when the mean at 15 um is below 10% of the mean at 0 um (the paper gives no value; "close to zero" read as under a tenth of the on-soma response, a reading named here). Needs the calibration recordings.
- ed10c-followers-integrate
Non-targeted x1 neurons outside 50 um of the targeted cells integrate during group x1 activation.
even after excluding the targeted x 1 neurons themselves as well as potentially off-target neurons located within a 50 µm radius of the targeted cell ... we observed integration in the remaining x 1 neurons
Extended Data Fig. 10a-c; Results, Line-attractor neurons form ensembles, paragraph 1How it will be judged
In the x1-activation session, x1 neurons that are not targeted and lie more than 50 um from every targeted cell; their weighted average activity (weights from the x1 factor), tested as fig2d-x1-integration. Reproduced when at least one ISI exceeds baseline at p < 0.001 under one reading and the mean rises from the first to the third ISI.
- ed10g-follower-decoding
Integrated activity of targeted x1 neurons is decoded better from non-targeted x1 neurons (about 80% variance explained) than from non-targeted x2 or random neurons (about 40%).
Decoding from non-targeted x 1 neurons can explain significantly more variance (80% versus 40%) than non-targeted x 2 or randomly selected neurons (n = 8 mice, *p = 0.01, ***p = 0.0003
Extended Data Fig. 10d-g; Results, Line-attractor neurons form ensembles, paragraph 1How it will be judged
A support vector machine decoder (legend) predicting the mean activity of the targeted x1 neurons from non-targeted neurons. Not given and named as open, with readings, are the regression form (linear support vector regression), the number of neurons in each set (equal to the smaller set, random subsets of the same size for the random set), and the validation (fivefold over contiguous blocks). Reproduced when x1 exceeds both x2 and random by Dunn's test at p < 0.05 and the values are within tolerance; partially when the order holds without the values.
- ed10hi-responses-mostly-positive
Responses of non-targeted neurons to single x1 neuron stimulation are mostly positive.
These experiments revealed a slowly decaying elevation of activity during the ISI period in non-targeted x 1 neurons after each pulse of activation (Fig. 3b,d ) that was mostly positive
Extended Data Fig. 10h and 10i; Results, Line-attractor neurons form ensembles, paragraph 2How it will be judged
A neuron's response is positive or negative when its mean response after photostimulation is 1.5 standard deviations above or below its baseline (legend of Extended Data Fig. 10h). The post-stimulation window is not given; readings Wa and Wb. Reproduced when positive responders outnumber negative responders, pooled over the unitary x1 stimulation sessions (the legend reports fractions of non-targeted neurons, not a count of mice); the count per mouse is reported.
- ed10lm-distance-not-explaining
Putative follower x1 neurons lie on average 139 +/- 35 um from the targeted cell, 11% within 50 um, and their response is not positively correlated with distance.
Notably, the strength of functional connectivity was not positively correlated with the distance from the targeted photostimulated cell
Extended Data Fig. 10j-n; Results, Line-attractor neurons form ensembles, paragraph 2How it will be judged
A putative follower is a non-targeted x1 neuron that responded to stimulation of a single targeted x1 neuron, by the 1.5 SD rule of the legend of Extended Data Fig. 10h (ed10hi-responses-mostly-positive). The Methods (Terminology) define putative followers as lying outside 50 um, which cannot give the legend's 11% within 50 um; here responders at any distance are counted, the only reading that allows that number. Expected values within tolerance, and the correlation of response with distance (the legend's regression line) not positive at p < 0.05. The paper gives no test; the reading is a Pearson test, named here.
- ed10p-exclusion-no-change
Excluding neurons within 50 um of the targeted cell does not change the follower response to single x1 neuron stimulation.
was still observed even after excluding neurons in a 50 µm zone surrounding the targeted neuron
Extended Data Fig. 10o and 10p; Results, Line-attractor neurons form ensembles, paragraph 2How it will be judged
A claim of absence, judged on the paper's test. Per mouse and ISI, the mean response of non-targeted x1 neurons with and without the exclusion; reproduced when no Dunn comparison gives p < 0.05.
- ed10r-scales-with-number
The response of non-targeted x1 neurons scales with the number of targeted x1 neurons, larger with five than with one.
revealed that the response strength of the non-targeted x 1 neurons scaled with the number of targeted x 1 neurons
Extended Data Fig. 10q and 10r; Results, Line-attractor neurons form ensembles, paragraph 2How it will be judged
Per mouse and ISI (Wa, Wb), the mean x1 projection over non-targeted x1 neurons in the one-neuron and the five-neuron sessions, compared per ISI by Dunn's test. Reproduced when five exceeds one in all four ISIs at p < 0.05 under one reading; partially when in two or three.
- fig3d-single-x1-drives-x1
Stimulating single x1 neurons raises the activity of non-targeted x1 neurons, with a slowly decaying elevation during the ISI.
These experiments revealed a slowly decaying elevation of activity during the ISI period in non-targeted x 1 neurons after each pulse of activation
Figure 3a, 3b and 3d; Results, Line-attractor neurons form ensembles, paragraph 2How it will be judged
The average z-scored activity of non-targeted x1 neurons in the unitary x1 sessions, per mouse (averaged over its targeted neurons), tested as fig2d-x1-integration. Reproduced when, under one reading, three of four comparisons give p < 0.05, as the legend reports.
- fig3e-single-x1-not-x2
Stimulating single x1 neurons does not raise activity in non-targeted x2 neurons.
we observed little activity in non-targeted x 2 neurons after activation of unitary x 1 or x 2 neurons
Figure 3c and 3e; Results, Line-attractor neurons form ensembles, paragraph 3How it will be judged
A claim of absence, judged on the paper's test. As fig3d-single-x1-drives-x1 on non-targeted x2 neurons; reproduced when no comparison gives p < 0.05.
- fig3i-single-x2-not-x1
Stimulating single x2 neurons gives a trend toward rising activity in non-targeted x1 neurons that the text calls not statistically significant.
While there was a trend to a gradual increase in activity in non-targeted x 1 neurons after repeated activation of unitary x 2 neurons (Fig. 3f–h ), that increase was not statistically significant
Figure 3f-i; Results, Line-attractor neurons form ensembles, paragraph 3How it will be judged
A claim of absence, judged on the paper's test, as fig3d-single-x1-drives-x1 on non-targeted x1 neurons in the unitary x2 sessions. Reproduced when no comparison gives p < 0.05. The legend itself reports one comparison at P = 0.0316 against the text's "not statistically significant"; the attempt reports which comparisons reach 0.05.
- fig3j-single-x2-not-x2
Stimulating single x2 neurons does not raise activity in non-targeted x2 neurons.
Quantification of activity in non-targeted x 2 neurons after targeting of single x 2 neurons (NS)
Figure 3g and 3j; Results, Line-attractor neurons form ensembles, paragraph 3How it will be judged
A claim of absence, judged on the paper's test. As fig3d-single-x1-drives-x1 on non-targeted x2 neurons in the unitary x2 sessions; reproduced when no comparison gives p < 0.05.
- fig4a-dense-x1-connectivity
About 36% of x1-to-x1 pairs have influence scores above the highest x1-to-x2 pair (a z-score of about 0.6), a relatively dense functional ensemble.
we considered the percentage of x 1 → x 1 pairs that had influence scores higher than the highest x 1 → x 2 pair, which had a z score of around 0.6 ... The fraction of x 1 → x 1 pairs above this threshold was around 36%
Figure 4a; Results, Line-attractor neurons form ensembles, paragraph 4How it will be judged
Influence score = the average evoked z-scored activity of each non-targeted photoactivated neuron after photostimulation of a single targeted cell (Results). The averaging window is not given; readings Wa and Wb pooled over the four ISIs. The legend counts 99 pairs from 7 mice; the pair counts found are reported. Reproduced when both values are within tolerance under one reading.
- fig4ce-excitatory-network
In purely excitatory spiking networks, only relatively sparse integration subnetworks (around 8 to 12%) with slow synaptic time constants (around 20 s) give network time constants of 50 to 200 s; fast glutamatergic networks fail over the same densities.
we found that only artificial networks based on relatively sparse connectivity (around 8–12%) and slow synaptic time constants (around 20 s) could yield network time constants ( τ n ) in the experimentally observed range (~50–200 s; Fig. 4c,e (red shading)). By contrast, networks with fast glutamatergic connectivity failed to do so over the same range of connection densities
Figure 4c-f; Results, Line-attractor neurons form ensembles, paragraph 5How it will be judged
A simulation as the Methods give it (Mechanistic modelling), 1,000 neurons, a 200-neuron subnetwork, weights U(0, 1/N) and U(0, 1/Np), tau_n = tau_s / (1 - lambda_max). Reproduced when, at tau_s = 20 s, tau_n falls in 50 to 200 s (widened 10%, 45 to 220 s) for some density between 8% and 12% (widened, 7.2% to 13.2%), and at tau_s of 0.6 s or less no stable density in the simulated range of 0 to 20% (legend of Fig. 4c) gives tau_n of 45 s or more. Uses no recorded data.
- fig4hj-inhibition-slow-dense
With fast feedback inhibition, a network with slow transmission (20 s) and the measured 36% density integrates pulsed input like the data, while a fast glutamatergic network (100 ms) does not.
networks simulated with a long τ s (20s) and dense σ (36%) could integrate digital optogenetic stimulation in a manner like that observed experimentally ... By contrast, purely glutamatergic networks ( τ s = 100 ms) were unable to integrate at the observed timescales given the measured connectivity density
Figure 4g-l; Results, Line-attractor neurons form ensembles, paragraph 6How it will be judged
The Methods network with inhibition (tau_I 50 ms, g = 1, g_inh 5 or 10, and 1 by the Methods or 1.25 by the legend of Fig. 4h, both run), four smoothed input pulses at 20 s ISI to a random 25% of neurons, Euler steps of 1 ms, noise N(0,1)/5. The pulse width and the calcium kernel are not given (readings 2 s, as in vivo, and an exponential kernel of 1 s decay as a standard choice that cannot change whether activity integrates). Reproduced when the subnetwork's mean rate at the end of each ISI rises from the first to the third ISI at tau_s 20 s and does not at tau_s 0.1 s, for one of the g_inh values. Uses no recorded data.
- fig5g-time-constant-vs-connectivity
Across mice, the x1 time constant during observation correlates strongly with the functional connectivity among x1 neurons measured as the AUC of non-targeted x1 neurons after the third pulse (R2 = 0.87).
The strength of this correlation was higher after the third stimulus ( r 2 = 0.87) compared with after the first stimulus ( r 2 = 0.59)
Figure 5g; Results, Attractor stability ties to connectivity, paragraph 2How it will be judged
Per mouse, the x1 time constant of its observation fit (the "intrinsic leak rate" over the whole session, Methods; readings are the largest time constant over states and the x1 time constant of the state occupied after BALB/c removal) and the AUC of the average z-scored activity of non-targeted x1 neurons in its unitary x1 sessions after the third pulse (readings Wa and Wb for the window). Pearson R2 and its p-value over 8 mice. Reproduced when R2 is within tolerance and p < 0.001 under one reading; partially when p < 0.05 with a positive slope.
- ed12-first-stimulus-weaker
The correlation is weaker after the first stimulus (R2 = 0.59) than after the third.
The strength of this correlation was higher after the third stimulus ( r 2 = 0.87) compared with after the first stimulus ( r 2 = 0.59)
Extended Data Fig. 12c and 12d; Results, Attractor stability ties to connectivity, paragraph 2How it will be judged
As fig5g-time-constant-vs-connectivity, after the first pulse. Reproduced when R2 is within tolerance and below the third-pulse R2 under the same reading. The Extended Data Fig. 12b legend gives r2 0.87 for x2 neurons after the third stimulus, which conflicts with the Results and with Fig. 12e-h; it is not used.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none