Claims 50
From the paper Preconfigured neuronal firing sequences in human brain organoids · claims file · checked 2026-10-21: ready
- methods-burst-count
The analysed segment of each recording holds about 40 population bursts (38 ± 4, mean ± SE).
The recording durations per data source were selected so that each recording contained around 40 bursts (38 ± 4 bursts, mean ± SE)
Methods, Comparing data from different sourcesHow it will be judged
Bursts detected as in the opening comment over HO1 to HO8, MO1 to MO9, MS and Pr1 to Pr8, under the code's recording lengths (reading a). Reproduced when the mean count is within tolerance. Reading (b) fixes the count by construction and is not a test of it.
- fig1c-example-burst-count
In the example human organoid of Figs. 1 and 2, 46 bursts occur in a three-minute interval.
The population firing rate (gray lines) is plotted relative to the burst peak for 46 burst events measured across a three-minute interval for the same organoid
Figure 1C legend; Figure 2C legendHow it will be judged
The example organoid is not named in the paper (the HMM code uses HO1 as its example). Bursts detected as in the opening comment on the first 180 s of each of HO1 to HO4, with a minimum peak distance of 700 ms (Methods) and 800 ms (code variant). Reproduced when one of the four recordings gives a count within tolerance; that recording is the example organoid for claims fig4d-burst-similarity and fig4f-pca-variance.
- ed1c-lognormal-rates
In each of the 8 human organoids the distribution of unit mean firing rates is well fit by a lognormal distribution (R² 0.97 ± 0.04, mean ± STD), while the rates of backbone units alone are not (R² 0.45 ± 0.30).
we observed that the distribution of firing rates follows a heavy-tailed, right-skewed lognormal distribution ( n = 8, R 2 = 0.97 ± 0.04, Extended Data Fig. 1A – C ) ... Backbone neurons alone are not well described by a lognormal distribution. R 2 values are 0.45±0.30 across the 8 human brain organoids
Results, Temporal dynamics of neuronal firing sequences, paragraph 1; Extended Data Fig. 1A to CHow it will be judged
Mean rate per unit is its spike count over the recording length. The fit is open between the Methods and code and the Extended Data Fig. 1C legend. Readings (a) a lognormal density fitted by nonlinear least squares to the probability histogram of rates in 20 bins from 0 to 10 Hz, ordinary R² (Methods, and plot_multi_rec_figs_backbone.m, whose density divides by (2 sigma)^2 where the standard form has 2 sigma^2, shown as a variant); (b) a normal density fitted to the probability histogram of log rates in ceil(n^0.4) bins (legend). Units either all sorted units (code) or those with at least 30 spikes (Methods). Reproduced when, under one reading, both means are within tolerance; partially reproduced when only the all-unit mean is.
- fig1d-burst-duration
Burst durations in the human organoids are on the order of 10² ms.
Bursts exhibited consistent rate profiles within organoids ( Fig. 1C ), with durations spanning timescales on the order of ≈ 10 2 milliseconds ( Fig. 1D , Supplementary Fig. 1 C )
Results, Temporal dynamics of neuronal firing sequences, paragraph 1; Figure 1D; Supplementary Fig. 1CHow it will be judged
Burst duration is end minus start as in the opening comment (10% of peak on the coarse population rate). "On the order of 10² ms" is read as the decade from 100 to 1000 ms, a standard reading. Reproduced when the median burst duration of each of HO1 to HO4 (Fig. 1D) lies in that decade; partially reproduced when three of the four do. The medians of all 31 recordings (Supplementary Fig. 1C) are reported.
- fig1e-peaks-near-burst-peak
Most units reach their peak firing rate close to the peak of the population burst.
A majority of units firing rates peaked in close proximity to the population burst maxima ( Fig. 1E )
Results, Temporal dynamics of neuronal firing sequences, paragraph 1; Figure 1EHow it will be judged
Peak time per unit and burst, the time of its maximum rate between burst start and end relative to the burst peak, in bursts where it fires at least two spikes (Methods). HO1 to HO4. "Close proximity" is not fixed by the paper. Readings for the quantity, (i) each unit's median peak time, (ii) all unit-burst peak times pooled; for the bound, (a) within 50 ms and (b) within 100 ms of the burst peak (the Results speak of lags near 10 ms and sequences of about 10² ms). Reproduced when more than half of the values fall within the bound in each organoid, under every reading.
- fig2-backbone-fraction
Backbone units, which fire at least twice in every burst, are 28% ± 14% (mean ± STD) of the units in the 8 human organoids.
Backbone units, comprising 28% ± 14% (mean ± STD) of total neurons ( n = 8 organoids, Supplementary Fig. 1 )
Results, Temporal dynamics of neuronal firing sequences, paragraph 3; Supplementary Fig. 1AHow it will be judged
Backbone units as in the opening comment. The denominator can be read as (a) all sorted units in the recording or (b) units with at least 30 spikes. Reproduced when the mean and the standard deviation are both within tolerance under one reading; partially reproduced when only the mean is. The tolerance on the mean is the paper's stated spread.
- fig2e-b2b-correlation
Backbone units have higher burst-to-burst correlations of their firing rate than non-rigid units.
Backbone neurons consistently exhibited higher burst-to-burst correlation coefficients compared to non-rigid units ( P < 10 −16 , two-sided linear mixed-effects model, n = 4 organoids and 507 single-units)
Figure 2E legend; Results, Temporal dynamics of neuronal firing sequences, paragraph 3How it will be judged
Burst-to-burst correlation per unit as in the Methods. Burst-only rates; for each pair of bursts, the rate from 250 ms before to 500 ms after each burst peak, MATLAB xcorr with maxlag 10 ms and 'coeff' normalization, maximum over lags; mean over all pairs of bursts in which the unit has at least 2 spikes in that window. Units kept when they have at least 2 spikes in at least 30% of bursts (Methods; the code keeps units with at least 2 spikes in at least 70% of bursts, a variant) and at least 30 spikes (code). HO1 to HO4 (code, Fig 2E comment). Test, correlation on unit type with a random intercept per organoid. Reproduced when backbone units are higher with p < 0.05; the unit count is reported against 507.
- fig2g-variance-latency
Among backbone units, the variance of firing-rate peak times across bursts grows with the unit's position in the sequence, so later-firing units are more variable.
Finally, the variability of firing peak times increased with their average latency ( Fig. 2F ), a relationship consistently observed across organoids ( Fig. 2G , P < 10 −5 , linear mixed-effects model)
Results, Temporal dynamics of neuronal firing sequences, paragraph 3; Figure 2F and 2GHow it will be judged
Per backbone unit of HO1 to HO4, the median and the variance of its peak time over the bursts in which it fires at least two spikes (Methods). The "relative position of the peak (from 0 to 1)" of the Methods is read as (a) the unit's rank by median peak time scaled to 0 to 1, or (b) its median peak time rescaled over the backbone period to 0 to 1. Model, variance on relative position with a random intercept per organoid. Reproduced when the slope is positive with p < 0.05 under both readings.
- ed3f-burst-cluster-variability
When bursts are clustered by their pairwise unit correlation matrices, the firing rates of non-rigid units vary more between clusters than those of backbone units.
Clustering bursts by their pairwise correlation matrix confirmed that backbone patterns were preserved across clusters, whereas non-rigid units varied substantially ( P < 10 −20 , linear mixed-effects model; Extended Data Fig. 3 )
Results, Temporal dynamics of neuronal firing sequences, paragraph 2; Extended Data Fig. 3FHow it will be judged
As in the Supplementary Methods, Burst clustering, and ExtractBurst_ElbowSeeds.py. Per burst, the burst-only rates from 250 ms before to 500 ms after the peak, the correlation matrix of all unit pairs (numpy corrcoef), its lower triangle as a vector; k-means++ on these vectors for k from 2 upward, the best of several seeds per k. The elbow is picked by eye in the paper; readings (a) the knee of the SSE curve by the kneedle rule, (b) the k with the largest silhouette coefficient (the code computes both). Per unit, the CV (STD over mean) of its mean rate across clusters. Recordings, HO1 to HO8; the paper shows HO5 (Or5) as its example and does not name the set, so HO1 to HO4 alone is a second reading. Test, CV on unit type with a random intercept per organoid. Reproduced when non-rigid CV is higher with p < 0.05 under every reading.
- ed4b-gabazine-more-bursts
In murine organoids, blocking GABA-A receptors with gabazine increases the number of population bursts.
Meanwhile, the number of bursts increase after gabazine treatment ( P < 0.05, two-sided linear mixed-effect model)
Results, Temporal dynamics of neuronal firing sequences, paragraph 2; Extended Data Fig. 4BHow it will be judged
Bursts detected as in the opening comment in the baseline and gabazine recordings of MO10 to MO14. Burst count (or rate, if the recordings differ in length) on condition with a random intercept per organoid. Reproduced when gabazine is higher with p < 0.05.
- ed4b-nbqx-cpp-abolish-bursts
Blocking AMPA and NMDA receptors with NBQX and R-CPP abolishes population bursts in murine organoids.
Note that bursting disappears after NBQX and R-CPP treatment reflected as a significant decrease in bursting compared to baseline conditions ( P < 0.001, two-sided linear mixed-effect model)
Results, Temporal dynamics of neuronal firing sequences, paragraph 2; Extended Data Fig. 4BHow it will be judged
As in ed4b-gabazine-more-bursts, baseline against NBQX with R-CPP. Reproduced when the count falls with p < 0.05 and the treated recordings hold no burst or nearly none (at most one per recording, given that the paper says bursting disappears).
- ed4c-gabazine-participation
Gabazine increases the fraction of bursts in which units fire at least two spikes.
The fraction of bursts in which units are active increases significantly after gabazine treatment ( P < 10 −7 , two-sided linear mixed-effect model)
Extended Data Fig. 4CHow it will be judged
Per unit and recording, the fraction of bursts with at least two spikes between burst start and end. Units are matched between conditions only if the dataset identifies them across recordings (the paper records "the same electrode configuration"); otherwise units are pooled per condition. Model, fraction on condition with a random intercept per organoid (and per unit when matched). Reproduced when gabazine is higher with p < 0.05.
- ed4d-gabazine-rank-order
Gabazine makes the order of backbone sequences more consistent across bursts.
There is a significant increase in backbone sequence order similarity after treatment with gabazine ( P < 10 −20 , two-sided linear mixed-effect model)
Extended Data Fig. 4DHow it will be judged
As in the Supplementary Methods, Firing rate peak rank order correlations. Per burst, the backbone units ranked by their peak rate time; Spearman correlation for every pair of bursts; each normalized to the mean over 100 shuffles. The legend says z-scored and the Supplementary Methods say (A - B)/(A + B); both are readings. Backbone units defined per recording with the 80% murine organoid threshold. Model, normalized correlation of burst pairs on condition with a random intercept per organoid. Reproduced when gabazine is higher with p < 0.05 under both readings.
- supp2a-backbone-fraction-declines
The fraction of backbone units declines as human and murine organoids mature.
Longitudinal recordings showed that the relative fraction of backbone units declined over development ( Supplementary Fig. 2 )
Results, Temporal dynamics of neuronal firing sequences, paragraph 2; Supplementary Fig. 2AHow it will be judged
No test is reported. Backbone fraction per organoid and age (human organoids at 6, 7 and 8 months per plot_develomental_analysis.m; murine organoids at their recorded weeks). Readings for "declined", (a) the mean fraction over organoids is lower at the oldest age than at the youngest, (b) it is lower at the oldest age in every organoid. Reproduced when this holds for both human and murine organoids under both readings; partially reproduced when it holds for one of them.
- supp2c-b2b-increases-with-age
The burst-to-burst correlation of units increases over development in human and murine organoids, and backbone units are higher than non-rigid units at every age.
Their firing patterns remained stable across hours ( Supplementary Fig. 8 ) and persisted over months, with correlations increasing during development in both human and murine organoids ( Supplementary Fig. 2C , P < 10 −5 , two-way ANOVA)
Results, Temporal dynamics of neuronal firing sequences, paragraph 3; Supplementary Fig. 2CHow it will be judged
Burst-to-burst correlation per unit as in fig2e-b2b-correlation, for each organoid and age. Two-way ANOVA of unit correlations on age and unit type (the second factor is not named; unit type is the reading the panel's layout implies, organoid is a second reading). Reproduced when the age effect has p < 0.05 with correlations higher at older ages in both groups, and backbone means exceed non-rigid means at every age.
- supp8-backbone-stable-4h
Units classed as backbone in one recording are mostly classed as backbone again 4 h later, units classed as non-rigid are mostly classed as non-rigid again, and backbone units keep more consistent firing patterns across the two recordings than non-rigid units.
the majority of units that was detected as a backbone unit at time point 0-hour was also detected as backbone unit at time point 4-hour (top left) for both organoids. Meanwhile, the majority of units that were detected as non-rigid units at time point 0-hour were also detected as non-rigid units at time point 4-hour (bottom right) for both organoids
Supplementary Fig. 8A and 8EHow it will be judged
HO2 and HO3, recordings at 0 h and 4 h. Units must be identified across the two recordings; the paper does not say how, so this is tested only if the dataset links them (for example by joint sorting). Reproduced when more than half of the 0 h backbone units are backbone at 4 h and more than half of the 0 h non-rigid units are non-rigid at 4 h in both organoids, and the mean 0-to-4 h burst-to-burst correlation (bursts paired across recordings, Methods) is higher for backbone than for non-rigid units in both.
- ed5-shuffle-reduces-b2b
Randomizing spike times while keeping unit and population rates reduces burst-to-burst correlations.
Randomized spike trains that preserved mean and single-unit firing rates abolished sequences and significantly reduced burst-to-bust correlations ( Extended Data Fig. 5 , Supplementary Figs. 9 - 10 )
Results, Temporal dynamics of neuronal firing sequences, paragraph 3; Extended Data Fig. 5D; Supplementary Fig. 9AHow it will be judged
Burst-to-burst correlation per unit as in fig2e-b2b-correlation on the real and on the 100 shuffled matrices, HO1 to HO8. No test is named. Readings (a) a linear mixed-effects model of correlation on data type (real or shuffled mean) with random intercepts for unit and organoid, (b) the normalized value (A - B)/(A + B) tested above 0 with a random intercept per organoid, one-sided as in Extended Data Fig. 2F. Reproduced when real exceeds shuffled with p < 0.05 for backbone units under both readings.
- fig3e-pairwise-correlation
Pairwise cross-correlations of firing rates are higher between backbone units than between backbone and non-rigid units or between non-rigid units.
Correlations between backbone units (blue) are significantly higher ( P < 10 −20 , 2-sided linear mixed effects model, n = 4 organoids and 34,056 single unit pairs) than the cross-correlations between pairs of backbone and non-rigid units (gray) and pairs of non-rigid units (yellow)
Figure 3E legend; Results, Backbone units are a highly correlated ensembleHow it will be judged
As in the Methods, Pairwise firing rate correlations. Burst-only rates over the whole recording, MATLAB xcorr with maxlag 350 ms and 'coeff', maximum over lags; units with at least 30 spikes (legend, code). HO1 to HO4 (code comment for Fig. 3E). Model, correlation on pair type with a random intercept per organoid, emmeans contrasts. Reproduced when backbone pairs exceed both other pair types with p < 0.05.
- fig3f-backbone-pairs-in-tail
The distribution of pairwise correlations is skewed and roughly lognormal, and backbone-backbone pairs lie in its upper tail in every organoid.
The histogram of all pairwise cross-correlations follows a skewed, lognormal distribution ( x -axis is log scale). The pairwise connections between pairs of backbone units populate the tail of this distribution for all organoids
Figure 3F legend; Results, Backbone units are a highly correlated ensembleHow it will be judged
Pairwise correlations as in fig3e-pairwise-correlation, HO1 to HO4. The paper marks the mean of the backbone-backbone correlations on each histogram and fixes no bound for the tail. Readings, that mean lies above (a) the 75th or (b) the 90th percentile of all pairs. Reproduced when this holds in all four organoids under both readings. The skew of all pairs and the R² of a normal fit to their logarithm (Supplementary Fig. 12C) are reported, not judged.
- fig4c-similarity-variance
The spread of burst similarity across pairs of bursts is high outside bursts, drops at burst onset and stays low until the bursts end.
Variance in similarity showed the opposite trend: it was high during non-bursting periods, rapidly declining at burst onset, and remaining low until bursts ended ( Fig. 4C )
Results, Population firing rate timing across burst epochs, paragraph 1; Figure 4CHow it will be judged
Burst similarity over all units as in fig4d-burst-similarity, for every pair of bursts at each frame from the earliest burst start to the latest burst end relative to the peak (Methods, Burst similarity score); at each frame, the standard deviation over burst pairs (Fig. 4C legend). Example organoid from fig1c-example-burst-count. No test is reported, and the paper does not say which frames count as inside a burst; readings, inside is (a) from the median burst start to the median burst end relative to the peak, (b) the backbone window. Reproduced when the mean standard deviation over frames inside is lower than over the frames outside, under both readings. HO1 to HO8 are reported.
- fig4d-burst-similarity
Across pairs of bursts, the population firing vectors of backbone units are more similar (cosine similarity) throughout the backbone window than those of non-rigid units or of shuffled data.
Backbone units consistently display significantly higher similarity throughout the backbone window compared with non-rigid or shuffled controls ( P < 10 −20 , paired two-sided t-test)
Figure 4D legend; Results, Population firing rate timing across burst epochs, paragraph 1How it will be judged
As in the Methods, Burst similarity score, and comp_burst_sim in the code. At each 1 ms frame relative to the burst peak, the cosine similarity of the unit rate vectors of every pair of bursts. The code z-scores each unit's rate over the recording first (the Methods are silent, the code is followed) and uses full rates for real data and burst-only rates for the shuffles. Backbone window, from the earliest to the latest median peak time of backbone units. At each frame of the window, a paired two-sided t-test over burst pairs (lower triangle) of backbone against non-rigid and backbone against shuffled, at Bonferroni alpha 0.05 over frames (code). Example organoid from fig1c-example-burst-count. Reproduced when backbone is higher and significant at every frame of the window for both comparisons; partially reproduced when that holds for at least 90% of frames. HO1 to HO8 are reported.
- fig4e-similarity-rises-at-backbone-onset
In every human organoid the burst similarity rises from the onset of the backbone period and stays high during it.
Across organoids, burst similarity increased at backbone onset and plateaued during their activation ( Fig. 4E , Supplementary Fig. 13D , n = 8 organoids)
Results, Population firing rate timing across burst epochs, paragraph 1; Figure 4E; Supplementary Fig. 13DHow it will be judged
Burst similarity over all units as in fig4d-burst-similarity, for HO1 to HO8, within each backbone period. No test is reported. Reproduced when, in all 8 organoids, the mean similarity over the backbone period exceeds its value at the first frame of the period; partially reproduced when that holds in 6 or 7.
- fig4f-pca-variance
In the example organoid, the first two principal components of backbone unit rates explain 73% of their variance against 25% for non-rigid units, and across organoids backbone units need fewer components than all units while non-rigid units need more.
When populations were divided, backbone units captured far more variance (73%) in the first two PCs compared to non-rigid units (25%) ( Fig. 4F middle/right, Fig. 4G , Supplementary Figs. 14 - 15 )
Results, Population firing rate timing across burst epochs, paragraph 2; Figure 4F and 4G; Supplementary Fig. 14How it will be judged
PCA as in compute_manifolds of the code, on burst-only rates, each unit mean-centred and scaled to unit standard deviation (the Methods' z-score), by SVD, for all, backbone and non-rigid units. The example organoid is the one of fig1c-example-burst-count; if none matches there, any of HO1 to HO8 that matches counts, as the number's subject is not named. Reproduced when both percentages are within tolerance and, in every organoid of HO1 to HO8, the cumulative variance of the first two PCs is higher for backbone than for all units and lower for non-rigid than for all units; partially reproduced when only the across-organoid pattern holds.
- supp18-hmm-likelihood-real-above-shuffled
Hidden Markov models fitted to the organoid spike trains give a higher held-out log likelihood on real data than on randomized data with the same unit and population rates.
Log likelihood is always greater for real than for randomized data, with no overlap between the distributions for real and shuffled data
Results, Revealing temporal structure with a hidden Markov model, paragraph 1; Supplementary Fig. 18How it will be judged
Poisson-observation HMMs (SSM), 30 ms bins, K from 10 to 30, fitted by EM (code tolerance 1e-3, at most 5000 iterations), for HO1 to HO8 (Supplementary Methods). Validation by 5-fold cross-validation. The Supplementary Methods fit each fold on a random 80% of the data and score the other 20%; the code (cv_scores) holds out a random 10% of the time bin by unit entries instead, shown as a variant. Score, held-out log likelihood per entry on real data against the randomized (Okun) surrogate in the held-out region. Reproduced when real exceeds surrogate in every fold, K and organoid.
- fig5c-states-converge-at-peak
The diversity of HMM states is high before and after bursts and reaches a minimum near the burst peak, where states follow a stereotyped path.
Notably, entropy consistently approaches a local minimum near burst peaks, indicating a convergence toward specific states with higher transition probabilities during the initiation of sequential activation patterns
Results, Revealing temporal structure with a hidden Markov model, paragraph 2; Figure 5C; Supplementary Fig. 19BHow it will be judged
HMMs as in supp18-hmm-likelihood-real-above-shuffled fitted to the full data; most likely state per bin (the code uses SSM most_likely_states); per bin from 500 ms before to 500 ms after each burst peak, the Shannon entropy of the state distribution across bursts (Supplementary Fig. 19B), per model, then averaged over K. "Near burst peaks" is not fixed; readings, the minimum lies (a) within 90 ms (three bins) of the peak, (b) within the backbone window. Reproduced when, in at least 6 of the 8 organoids, the minimum is near the peak and lower than the entropy at both ends of the window, under both readings.
- fig5d-backbone-more-states
Backbone units are active across a larger pool of HMM states than non-rigid units.
Backbone ensembles spanned a larger pool of HMM states than non-rigid units ( Fig. 5D , Supplementary Fig. 21 )
Results, Revealing temporal structure with a hidden Markov model, paragraph 3; Figure 5D; Supplementary Fig. 21How it will be judged
Consistency of a unit in a state, the fraction of the state's realizations (bins where it is most probable) in which the unit fires at least once (Supplementary Methods). The size of the "pool" is not defined; readings (a) the number of states with consistency above 0.5, (b) the mean consistency over states. HO1 to HO8, K from 10 to 30. No test is reported; a linear mixed-effects model on unit type with random intercepts for organoid and K. Reproduced when backbone is higher with p < 0.05 under both readings.
- fig5f-svm-states-beat-rate
A linear classifier on HMM state consistency separates backbone from non-rigid units with 83.9% ± 12.0% accuracy, above the 63.2% ± 10.4% reached with firing rate alone.
Linear support vector machine classifiers trained on hidden state features distinguished backbone from non-rigid units with 83.9% ± 12.0% accuracy, significantly exceeding the 63.2% ± 10.4% achieved when using firing rate alone ( Fig. 5F )
Results, Revealing temporal structure with a hidden Markov model, paragraph 3; Figure 5FHow it will be judged
Features per unit, its K consistencies; classifier, standardization then a linear SVM (the code uses sklearn SGDClassifier with its default hinge loss), leave-one-out accuracy over units; baseline, the same classifier on the unit's mean rate (Fig5.py). K from 10 to 30, so 21 models per organoid; means and SDs pooled over models (states) and over organoids (rate), as Fig5.py prints them. The tolerances are the paper's stated spreads. The test is open; readings (a) a paired two-sided t-test over the 8 organoids of mean state accuracy against rate accuracy (legend, n = 8), (b) per organoid a one-sample t-test of the 21 state accuracies against the rate accuracy (Fig5.py), required in all 8 ("in all cases"). Classifier seeds are a standard choice. Reproduced when both means are within tolerance and states beat rate with p < 0.05 under both readings.
- fig6f-slice-b2b-correlation
In neonatal mouse cortical slices, backbone units have higher burst-to-burst correlations than non-rigid units.
Backbone units consistently exhibited higher correlations than non-rigid units (P < 10 −20 , two-sided linear mixed-effects model, n = 6 mice and 959 single units)
Figure 6F legend; Results, Endogenous sequences in early developing neonatal cortex, paragraph 2How it will be judged
As fig2e-b2b-correlation, on the six slices, with the backbone threshold at 90% (Methods) and at 70% (Fig. 6B legend), and a random intercept per slice or per animal. Reproduced when backbone units are higher with p < 0.05 under every reading.
- fig6g-slice-pairwise-correlation
In neonatal slices, backbone-backbone pairs are more strongly correlated than mixed or non-rigid pairs.
Backbone–backbone correlations were significantly stronger than mixed or non-rigid pairs (P < 10 −20 , two-sided linear mixed-effects model, n = 6 mice and 125607 single unit pairs)
Figure 6G legendHow it will be judged
As fig3e-pairwise-correlation, on the six slices, with the slice readings of fig6f-slice-b2b-correlation. Reproduced when backbone pairs exceed both other types with p < 0.05 under every reading.
- fig7a-backbone-higher-rate
Backbone units fire at higher rates than non-rigid units in every model type (human and murine organoids, neonatal slices and primary cultures).
Backbone neurons had significantly higher firing rates than non-rigid counterparts ( Fig. 7A , P < 10 −6 , linear mixed-effects model, Supplementary Fig. 23 )
Results, Comparing sequences across neurodevelopmental models, paragraph 1; Figure 7AHow it will be judged
Mean rate per unit (spike count over recording length), units with at least 30 spikes, log10-transformed (Fig7.ipynb plots log10 rate; Methods log-transform skewed data). Model, log rate on unit type, model type and their interaction with a random intercept per recording; emmeans contrasts of backbone against non-rigid within each model type. Thresholds per model type as in the opening comment, with both slice readings. Reproduced when backbone is higher with p < 0.05 in all four model types; partially reproduced in two or three. Unit counts are reported (the paper gives 1048 for both human organoids and primary cultures).
- fig7a-primary-higher-rate
Primary cultures fire at higher rates than organoids and neonatal slices, in both unit groups.
However, primary cultures exhibited overall higher firing rates across both populations ( P < 10 −10 for organoids and neonatal slices)
Results, Comparing sequences across neurodevelopmental models, paragraph 1; Figure 7AHow it will be judged
Model of fig7a-backbone-higher-rate; emmeans contrasts of primary cultures against each of the other three model types, within backbone and within non-rigid units. Reproduced when all six contrasts favour primary cultures with p < 0.05; partially reproduced when at least four do.
- fig7b-normalized-b2b
After normalization to shuffled data, backbone units remain more correlated from burst to burst than non-rigid units in organoids and neonatal slices, and backbone units of organoids and slices are more correlated than those of primary cultures.
backbone units in organoids and neonatal slices showed significantly stronger normalized burst-to-burst single-unit firing correlations compared to primary cultures ( Fig. 7B , Supplementary Fig. 9 ). After normalization relative to randomized data, backbone units remained significantly more correlated than non-rigid units in organoids and slices ( P < 10 −10 and P = 7×10 −4 , linear mixed-effects model)
Results, Comparing sequences across neurodevelopmental models, paragraph 1; Figure 7BHow it will be judged
Burst-to-burst correlation per unit as in fig2e-b2b-correlation, normalized as (A - B)/(A + B) to the 100 shuffles (Methods; btb_corr_m in the code's single_unit table). Model as in fig7a-backbone-higher-rate on the normalized value (not log-transformed). Reproduced when backbone exceeds non-rigid with p < 0.05 in human organoids, murine organoids and slices, and backbone units of each of these three exceed those of primary cultures with p < 0.05; partially reproduced when only the within-type contrasts hold. The Results and the legend print different p-values for the same panel; both are recorded. Supplementary Fig. 9B describes the normalization as subtracting the shuffle mean (A - B); that is shown as a variant, as for fig7e-pca-backbone-variance.
- fig7b-primary-no-b2b-difference
In primary cultures, normalized burst-to-burst correlations do not differ between backbone and non-rigid units.
whereas no significant difference was observed in primary cultures ( P = 0.06)
Results, Comparing sequences across neurodevelopmental models, paragraph 1; Figure 7BHow it will be judged
A claim of absence, judged on the paper's test, the contrast of backbone against non-rigid in primary cultures in the model of fig7b-normalized-b2b at alpha 0.05. Reproduced when p >= 0.05, not reproduced when p < 0.05; the effect size is given either way.
- fig7c-normalized-pairwise
After normalization to shuffled data, backbone-backbone pairs are more correlated than non-rigid pairs in human and murine organoids.
Backbone pairs showed significantly higher correlations in organoids (human: P = 9 × 10 −7 ; murine: P = 1 × 10 −8 ), but not in slices ( P = 0.10) or primaries ( P = 0.16)
Figure 7C legendHow it will be judged
Pairwise correlations as in fig3e-pairwise-correlation, normalized as (A - B)/(A + B) to the 100 shuffles; per recording, the median over pairs of each type (med_corr in the code's unit_pair_avg table, which Fig7.ipynb plots). Model, the per-recording median on pair type (backbone-backbone, non-rigid-non-rigid), model type and their interaction with a random intercept per recording; emmeans contrast within each model type. Reproduced when backbone pairs are higher with p < 0.05 in both human and murine organoids. Pair-level values with a random intercept per recording are a variant. Supplementary Fig. 12B describes the normalization as subtracting the shuffle mean (A - B); that is shown as a variant.
- fig7c-slices-primaries-no-pair-difference
Normalized pairwise correlations of backbone pairs do not differ from those of non-rigid pairs in slices or in primary cultures.
but not in slices ( P = 0.10) or primaries ( P = 0.16)
Figure 7C legendHow it will be judged
A claim of absence, judged on the contrasts of the model of fig7c-normalized-pairwise at alpha 0.05, with both slice readings. Reproduced when both are non-significant; partially reproduced when one is; not reproduced when neither is. Effect sizes are given.
- fig7c-backbone-pairs-above-random
Backbone-backbone pairs in organoids and slices are more correlated than in randomized data, while in primary cultures they are not.
Backbone-to-backbone pairs in organoids and slices exhibited significantly higher correlations compared to randomized controls ( P < 10 −10 , linear mixed-effects model), an effect not observed in primary cultures ( P > 0.99)
Results, Comparing sequences across neurodevelopmental models, paragraph 2How it will be judged
Normalized backbone-backbone pair correlations, (A - B)/(A + B), tested above 0 per model type with a linear mixed-effects model and a random intercept per recording, one-sided as in Extended Data Fig. 2F (the Results do not name the sidedness; two-sided is a variant). Reproduced when the value is above 0 with p < 0.05 in organoids and in slices and not significant in primary cultures (the absence judged on the same test at alpha 0.05).
- fig7d-lag-times
Lags of maximum cross-correlation between backbone units are longer in human and murine organoids than in primary cultures, and lags in human organoids differ from those in murine organoids and in slices.
Correlation Lag-times between backbone units were larger for organoids compared to primary cultures (human organoid vs primary: P = 1 × 10 −8 ; murine organoids vs primary: P = 0.011), with human organoid differing from murine organoids ( P = 0.0096) and slices ( P = 0.0002)
Results, Comparing sequences across neurodevelopmental models, paragraph 2; Figure 7DHow it will be judged
Per backbone pair, the lag of the maximum pairwise correlation (fig3e-pairwise-correlation, maxlag 350 ms), absolute and log10-transformed (lag_vals_log in the code's lag_data table). Zero lags have no logarithm and the code's handling is not shown; readings (a) drop them, (b) set them to 1 ms. Model, log lag on model type with a random intercept per recording; emmeans contrasts. The text gives the direction only for the contrasts against primary cultures (larger in organoids); for human organoids against murine organoids and against slices it says only that they differ, so either direction counts and the direction found is reported. Reproduced when the two contrasts against primary cultures are larger in organoids and all four have p < 0.05, under both readings; partially reproduced when the two contrasts against primary cultures hold.
- fig7d-murine-organoids-slices-same-lag
Backbone lag times do not differ between murine organoids and slices.
Murine organoids vs slices ( P = 0.47) and primary cultures were not significant ( P = 0.53)
Figure 7D legendHow it will be judged
A claim of absence on the model of fig7d-lag-times at alpha 0.05. The legend's second contrast reads as murine organoids against primary cultures, which the same legend reports as P = 0.011, so it is read as slices against primary cultures. Reproduced when murine organoids against slices is non-significant; the slices against primary cultures contrast is reported and judged the same way.
- fig7e-pca-backbone-variance
In human and murine organoids, backbone units explain a larger normalized fraction of variance in the first three principal components than non-rigid units; slices and primary cultures show no such difference.
Differences were significant for organoids (human: P = 2 × 10 −7 ; murine: P = 0.036), but not slices ( P = 0.29) or primaries ( P = 0.21)
Results, Comparing sequences across neurodevelopmental models, paragraph 3; Figure 7E; Supplementary Fig. 15How it will be judged
PCA as in fig4f-pca-variance. Per recording and unit set, the variance of the first three PCs divided by that of the first X PCs, X the smallest number of PCs among all, backbone and non-rigid units; normalized as (A - B)/(A + B) to 100 shuffles (Methods; m_value in the code's pca table; Supplementary Fig. 15 says the shuffle mean was subtracted, a variant). Model, value on unit type, model type and their interaction with a random intercept per recording. The legend plots all units, backbone and non-rigid and says only "Differences were significant", and the Methods compare "the three selections", so which contrast is meant is open; readings (a) backbone against non-rigid (unit types backbone and non-rigid), (b) backbone against all units (all three selections in the model). Reproduced when backbone exceeds the comparison group with p < 0.05 in human and murine organoids and the contrasts in slices and primary cultures are non-significant at alpha 0.05 (absence on the same test), under both readings; partially reproduced when only the organoid contrasts hold. The Results also say organoid backbone units exceed those of other model types; that contrast is reported.
- fig7f-hmm-dimensionality
HMM state spaces are higher-dimensional in human and murine organoids than in primary cultures.
With θ = 0.75, human organoids were separable from slice data ( P = 0.04, linear mixed-effects model with Poisson observations), while primary cultures had significantly lower dimensionality than both human ( P < 0.0001) and murine organoids ( P = 0.003) ( Fig. 7F )
Results, Comparing sequences across neurodevelopmental models, paragraph 4; Figure 7F; Supplementary Fig. 25How it will be judged
HMMs as in supp18-hmm-likelihood-real-above-shuffled for all 31 recordings, K from 10 to 30. Per model, PCA of the K by units matrix of Poisson rates (exp of the fitted log lambdas, Supplements.py); dimensionality, the smallest number of components whose cumulative explained variance exceeds theta = 0.75 (counted up to 10). Test as in lmem.R, a Poisson GLMM (glmer, log link) of dimensionality on model type with a random intercept per recording, emmeans pairwise contrasts (Tukey adjustment, the emmeans default). Reproduced when primary cultures are lower than human and murine organoids with p < 0.05. The human organoid against slice contrast is reported; the legend's "but not compared with slices" next to P = 0.036 is not judged.
- supp24c-nonrigid-more-variable
Non-rigid units fire more irregularly (higher CV2) than backbone units in every model type.
The CV2 scores are significantly larger for non-rigid units compared to backbone units, reflecting a more Poisson-like firing of non-rigid units (Human organoids P ≤ 10-20, Murine organoids P ≤ 10-20, Murine slices P = 7x10-13, Murine primary cultures P ≤ 10-20
Results, Comparing sequences across neurodevelopmental models, paragraph 4; Supplementary Fig. 24CHow it will be judged
CV2 per unit as in Holt et al. (mean over consecutive interval pairs of 2|I2 - I1|/(I2 + I1)), units with at least 30 spikes. The panel does not say whether raw or shuffle z-scored CV2 (Methods) is tested; readings (a) raw, (b) z-scored to the 100 shuffles. Model, CV2 on unit type, model type and their interaction with a random intercept per recording. Reproduced when non-rigid exceeds backbone with p < 0.05 in all four model types under both readings. (The paper prints "P ≤ 10-20"; p_relation "<" stands for it.)
- supp24d-backbone-non-poisson
Given the HMM hidden state, non-rigid units fire mostly like Poisson units, while backbone units fire less variably than Poisson in a sizable fraction of states.
Across models, non-rigid units were predominantly Poisson, while backbone units were non-Poisson ( Supplementary Fig. 24 )
Results, Comparing sequences across neurodevelopmental models, paragraph 4; Supplementary Fig. 24D; Supplementary Methods, Non-Poisson UnitsHow it will be judged
As the Supplementary Methods, Non-Poisson Units, and Supplements.py (poisson_test, unit_consistency, mean_consistency). HMMs as in fig7f-hmm-dimensionality (all 31 recordings, K from 10 to 30). For each unit, model and state, the spike counts of the unit in the realizations of the state (bins where it is the most likely state; the code keeps only bins inside burst windows); a one-sided chi-square test of variance below the Poisson variance (the mean count) at the 1% level, statistic (n - 1) s² / mean against the lower tail of chi-square with n - 1 degrees of freedom (code). A unit that never fires in a state counts as Poisson there (code). Per unit, the fraction of states passing, weighted by their numbers of realizations, over all its fitted HMMs. No test of the group difference is reported. "Sizable" and "almost all states" are not fixed; reproduced when, in every model type, the median fraction of backbone units exceeds that of non-rigid units and the median fraction of non-rigid units is at most 0.1 (read as at least 90% of states Poisson). The medians and a linear mixed-effects model of the fraction on unit type with a random intercept per recording are reported.
- supp10b-rank-order-by-model
The consistency of backbone sequence order across bursts differs between model types, and is higher in human organoids than in primary cultures.
Main effect of sample type P = 0.004 ... Human organoids vs Murine primary cultures P = 0.0004
Supplementary Fig. 10BHow it will be judged
Normalized Spearman rank-order correlations of backbone sequences for all burst pairs as in ed4d-gabazine-rank-order, both normalization readings. Model, value on model type with a random intercept per recording. Reproduced when the main effect and the human organoid against primary culture contrast (higher in organoids) have p < 0.05 under both readings. The other five contrasts (P = 0.066, 0.061, 0.99, 0.38, 0.65) are reported.
- supp26-state-traversal
HMM state traversal rates during the backbone period are higher in primary cultures than in human organoids and slices, while human and murine organoids do not differ.
In particular, human and murine organoid cannot be distinguished (P = 0.93), whereas the higher state traversal rate of murine primary cultures is significantly different from that of human organoids (P = .021) and murine acute slices (P < 0.0001) but not significant relative to murine organoids (P = 0.064)
Results, Comparing sequences across neurodevelopmental models, paragraph 4; Supplementary Fig. 26How it will be judged
As state_traversal_df in hmmsupport.py. Per model and burst, the number of distinct most likely states in the bins of the backbone window after the peak offset, averaged over bursts and divided by the window length. Model as in lmem.R, rate on model type plus K (as a factor) with a random intercept per recording, emmeans pairwise contrasts. Reproduced when primary cultures exceed human organoids and slices with p < 0.05 and the human against murine organoid contrast is non-significant (absence at alpha 0.05); the primary against murine organoid contrast is reported.
- supp27-d2-near-critical
A subset of every preparation type shows near-critical temporal dynamics (d2 between 0 and 0.1), and shuffling spike times raises d2 well above that.
Spike-time shuffling consistently abolished temporal criticality across preparations, producing significantly larger d 2 values ( P < 10 −10 compared to intact data, two-sided Welch t-test)
Results, Firing pattern stability, paragraph 1; Supplementary Fig. 27A; Supplementary Methods, d2 calculationHow it will be judged
As the Supplementary Methods, d2 calculation. Spike counts over all units in 30 ms bins, z-scored; an order-20 AR model by Yule-Walker; d2 from the AR coefficients with the code of Sooter et al. (figshare article 25927081, version 1, as r_download.sh fetches). Shuffled control, each neuron's spike times shuffled independently keeping its rate (not the Okun swap). Preparations are kept when the AR model's simulated avalanche size and duration distributions and power spectrum lie within one order of magnitude of the data's; the comparison is not specified further, so counts that differ by one are reported as near matches. The counts are exact numbers of recordings and are given no tolerance; reproduced when shuffled d2 exceeds intact d2 by a two-sided Welch t-test with p < 0.05 and the kept intact preparations have d2 at most 0.1; partially reproduced when the test holds but the counts differ. The code's table lists 7 primary cultures (Pr5 absent) of the 8.
- ed2d-alam-el-din-rates
Backbone units have higher mean rates than non-rigid units in the Sharf organoid slices (HO1 to HO4) but not in the whole Alam El Din organoids (HO5 to HO8).
(Sharf et al. P < 10 −20 , Alam El Din et al . P = 0.40, two-sided linear mixed-effect model, n = 4 organoids each)
Extended Data Fig. 2DHow it will be judged
Log mean rate per unit (units with at least 30 spikes) on unit type with a random intercept per organoid, fitted separately to HO1 to HO4 and to HO5 to HO8. Reproduced when the Sharf fit has backbone higher with p < 0.05 and the Alam El Din fit has p >= 0.05 (absence judged on the same test); the effect sizes are given.
- ed2e-alam-el-din-b2b
Rate-normalized burst-to-burst correlations are higher for backbone than non-rigid units in both sets of human organoid recordings.
Note the significant difference between backbone and non-rigid units present for both sets of recordings (Sharf et al . P = 1×10 −10 , Alam El Din et al . P = 0.0092, two-sided linear mixed-effect model, n = 4 organoids each)
Extended Data Fig. 2EHow it will be judged
Normalized burst-to-burst correlation as in fig7b-normalized-b2b ("after average rate normalization"), on unit type with a random intercept per organoid, fitted separately to HO1 to HO4 and HO5 to HO8. Reproduced when backbone is higher with p < 0.05 in both.
- ed2f-backbone-pairs-above-zero
In both sets of human organoid recordings, only backbone pairs have rate-normalized pairwise correlations significantly above zero.
Note that only the normalized correlation for backbone pairs are significantly larger than 0 for both sets of recordings (Sharf et al . P = 0.0001, 0.145 and 1 respectively, Alam El Din et al . P = 7×10 −11 , 1 and 1 respectively, one-sided linear mixed-effect model, n = 4 organoids each)
Extended Data Fig. 2FHow it will be judged
Normalized pairwise correlations as in fig7c-normalized-pairwise, per pair, for each pair type; one-sided test of the intercept above 0 in a linear mixed-effects model with a random intercept per organoid, separately for HO1 to HO4 and HO5 to HO8. Reproduced when backbone pairs are above 0 with p < 0.05 in both sets and the other two pair types are not (absence at alpha 0.05); partially reproduced when the backbone result holds but another type is also significant.
- supp5-murine-organoid-forebrain-identity
Single-cell RNA sequencing shows that the murine ESC organoids are of predominantly forebrain identity and express AMPA, NMDA and GABA receptor subunit genes.
Expression patterns of brain-region specific marker genes reveal predominant forebrain identity of the murine brain organoids used in this study
Supplementary Fig. 5B and 5CHow it will be judged
The PIPseq data of 30-day organoids from three lines (4031 cells after filtering), as in the Supplementary Methods (Seurat 5.1.0, SCTransform, 40 PCs, Leiden clustering at resolution 0.5, canonical markers). Not testable on the dandiset, which holds electrophysiology only. The markers are those of Supplementary Fig. 5B, which the text does not list, so the set of markers and what counts as predominant are open; the claim is recorded and not tested here.
Pipeline
- Paper
- done (4 of 4)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none