Spontaneous behaviors drive multidimensional, brainwide activity

Spontaneous behaviors drive multidimensional, brainwide activity Stringer Pachitariu Steinmetz Reddy Carandini Harris

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Spontaneous behaviors drive multidimensional, brainwide activity 2019 Science 10.1126/science.aav7893 Stringer2019-spontaneous-behaviors-drive-multidimensional Stringer2019-spontaneous-behaviors-drive-multidimensional.txt, the project's staged copy 187aeebc3fc29e1dc329b4712b8f206d83e60427 not-redistributable neuroscience

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them the structure of spontaneous and stimulus-evoked population activity in mouse visual cortex, and what the apparently noisy component of it encodes Quotation withheld because this record is not cleared for redistribution. is cortical variability random noise, recapitulation of prior sensory experience, or encoding of ongoing behaviour Quotation withheld because this record is not cleared for redistribution. shows spontaneous activity reliably encodes a high-dimensional behavioural signal present brain-wide, and that sensory input adds to it in orthogonal dimensions rather than displacing it Quotation withheld because this record is not cleared for redistribution. existence Quotation withheld because this record is not cleared for redistribution. spontaneous cortical population activity and trial-to-trial response variability Quotation withheld because this record is not cleared for redistribution. calcium-imaging calcium-imaging Quotation withheld because this record is not cleared for redistribution. mechanisms Behavioural and state variables reach primary sensory cortex through neuromodulatory input carrying arousal state and top-down input from higher-order cortices carrying fine-grained motor variables, so sensory and motor information are integrated already at V1 Quotation withheld because this record is not cleared for redistribution. mechanisms The apparent similarity between spontaneous and stimulus-evoked population activity arises because a single common mean-rate dimension fluctuates in both conditions, not because cortex replays past sensory experience Quotation withheld because this record is not cleared for redistribution. reliable variance is defined by double cross-validation: neurons split into two spatially separate sets, timepoints split into train and test, and only variance that covaries across both splits counts Quotation withheld because this record is not cleared for redistribution. the behavioural regressor is not chosen in advance but discovered, by taking a thousand principal components of facial motion rather than naming which movements should matter Quotation withheld because this record is not cleared for redistribution. the hand-picked arousal variables are shown to be redundant once the video is used, which is how the paper argues the old low-dimensional accounts were glimpses of something larger Quotation withheld because this record is not cleared for redistribution. the paper states what its own method cannot see, and distinguishes its phenomenon from the fast one by a timescale its sampling rate excludes Quotation withheld because this record is not cleared for redistribution. the claim of high dimensionality is stress-tested against sample size: reliability of higher components rises with neuron count, so the reported dimensionality is a floor Quotation withheld because this record is not cleared for redistribution. decoding is used as the check on the subspace geometry - if stimulus information really is confined to one subspace, a classifier reading the others should fail Quotation withheld because this record is not cleared for redistribution. the imaging result is repeated in a different preparation and a different recording modality, eight probes across forebrain, before the brain-wide claim is made Quotation withheld because this record is not cleared for redistribution. a spatial null is run on the sorted activity continuum: position on it does not track position in the tissue, so the structure is not topographic Quotation withheld because this record is not cleared for redistribution. trial-to-trial variability is decomposed into a multiplicative part inside the stimulus subspace and an additive part outside it, which is what lets a long-running either/or dispute be answered with both Quotation withheld because this record is not cleared for redistribution. the multiplicative gain is not left as a fitted nuisance parameter but predicted from the behavioural video, tying the statistical model back to an observable Quotation withheld because this record is not cleared for redistribution. the affine model is tested and rejected, so the absence of additive variability within the stimulus subspace is a checked result rather than an assumption Quotation withheld because this record is not cleared for redistribution. The dimensionality estimate is bounded by the number of neurons recorded; larger populations would reveal still more dimensions. what this limits: the dimensionality figure, which is bounded by the recording and not by the brain: THE ESTIMATED RELIABILITY OF HIGHER SVCs INCREASED WITH THE NUMBER OF NEURONS ANALYZED ... RECORDINGS OF LARGER POPULATIONS WOULD IDENTIFY YET MORE DIMENSIONS Quotation withheld because this record is not cleared for redistribution. The imaging frame rate (400 ms per cortical scan) makes fast up/down phases undetectable in this dataset. what this limits: any statement about up/down dynamics from this dataset: UP/DOWN PHASES COULD NOT EVEN HAVE BEEN DETECTED IN OUR RECORDINGS, WHICH SCANNED THE CORTEX EVERY 400 ms Quotation withheld because this record is not cleared for redistribution. The unexplained dimensions of reliable variance are not identified; they may be motor activity invisible on the face or internal cognitive variables. what this limits: the behavioural account of the variance: THE REMAINING DIMENSIONS AND VARIANCE MIGHT IN PART REFLECT MOTOR ACTIVITY NOT VISIBLE ON THE FACE ... OR THEY MIGHT REFLECT INTERNAL COGNITIVE VARIABLES SUCH AS MOTIVATIONAL DRIVES Quotation withheld because this record is not cleared for redistribution. the recording rate makes the fast up/down phenomenon undetectable in the imaging data duplicate of: limitations[1] · what this limits: the same scope as limitations[1] - the scan rate excludes the phenomenon rather than measuring it as absent Quotation withheld because this record is not cleared for redistribution. most of the reliable variance is still unexplained, and the paper offers two quite different candidates without deciding what this limits: the same scope as limitations[2] - most of the reliable variance is unexplained, with two very different candidates offered and neither tested Quotation withheld because this record is not cleared for redistribution. the dimensionality reported is bounded by the number of neurons recorded, not by the brain what this limits: the same scope as limitations[0] - more neurons would give more dimensions Quotation withheld because this record is not cleared for redistribution. the recapitulation account is not excluded, only shown to be unsupported by the similarity statistic on its own what this limits: the recapitulation literature: OUR RESULTS DO NOT EXCLUDE THAT GENUINE RECAPITULATION COULD OCCUR IN OTHER BEHAVIORAL CIRCUMSTANCES, but EVEN A SINGLE DIMENSION OF COMMON RATE FLUCTUATION IS SUFFICIENT FOR SOME PREVIOUSLY-APPLIED STATISTICAL METHODS TO REPORT it Quotation withheld because this record is not cleared for redistribution. a train-test split applied simultaneously along two axes - cells and time - to define which variance counts as signal rather than relying on a noise model Quotation withheld because this record is not cleared for redistribution. motion energy and its principal components, used to turn a video of a face into a thousand-dimensional behavioural regressor without deciding in advance which movements matter Quotation withheld because this record is not cleared for redistribution. reduced rank regression, used to ask how few behavioural dimensions suffice rather than how much total variance behaviour explains Quotation withheld because this record is not cleared for redistribution. comparing recordings made in complete darkness against recordings made with a gray screen eliminates: that the spontaneous population activity is driven by visual input from the gray screen rather than being genuinely spontaneous · outcome: held Quotation withheld because this record is not cleared for redistribution. cross-validated stimulus decoding from the stimulus subspace compared against the spontaneous-only and behaviour-only subspaces eliminates: that the spontaneous and behavioural dimensions carry the stimulus information, rather than being orthogonal to it · outcome: held Quotation withheld because this record is not cleared for redistribution. cortical variability - as random noise, as recapitulation of prior experience, or as encoding of behaviour; three incompatible readings of one measured quantity, named together by the authors Quotation withheld because this record is not cleared for redistribution. variability - independent across cells (harmless to coding) versus correlated (harmful or not, depending on its geometry); the distinction that makes the orthogonality result matter Quotation withheld because this record is not cleared for redistribution. correlated variability - mimicking stimulus differences, and so corrupting the code, versus lying orthogonal to them, and so carrying something else Quotation withheld because this record is not cleared for redistribution. slow multi-second alternation between two sub-populations versus fast up/down phases modulating all neurons together; the same raster read at two timescales Quotation withheld because this record is not cleared for redistribution. similarity between spontaneous and evoked activity - as evidence of replayed experience versus as a mean-rate artefact; the paper argues the statistic cannot distinguish them Quotation withheld because this record is not cleared for redistribution. arousal as three hand-picked variables versus behaviour as a thousand videographic dimensions; the first turns out to be a projection of the second Quotation withheld because this record is not cleared for redistribution. reliable variance - defined operationally as what survives a split across both cells and time, not as what exceeds a noise threshold Quotation withheld because this record is not cleared for redistribution. method: two-photon calcium imaging of over 10,000 neurons with simultaneous facial videography, cross-validated by a purpose-built shared-variance method, and replicated with eight simultaneous Neuropixels probes a second reading DIFFERING from the recorded value. Recorded: calcium-imaging — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. phenomenon: spontaneous cortical activity and trial-to-trial variability of sensory responses a second reading DIFFERING from the recorded value. Recorded: spontaneous cortical population activity and trial-to-trial response variability — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. extends → Previous reports of low-dimensional locomotion and arousal correlates in visual cortex (refs 8, 10-16, 33) Quotation withheld because this record is not cleared for redistribution. contradicts → Studies reporting similarity between spontaneous activity and sensory responses as evidence for recapitulation of sensory experience (refs 2-7) Quotation withheld because this record is not cleared for redistribution. extends → Conflicting prior findings on whether cortical variability is additive or multiplicative (refs 39-42) Quotation withheld because this record is not cleared for redistribution. critiques → studies inferring recapitulation of sensory experience from statistical similarity between spontaneous and evoked activity Quotation withheld because this record is not cleared for redistribution. extends → reports of low-dimensional locomotion and arousal correlates in visual cortex, which the paper recasts as glimpses of a much larger set Quotation withheld because this record is not cleared for redistribution. supports → the resolution of the additive-versus-multiplicative dispute about cortical variability, by assigning each to a different subspace Quotation withheld because this record is not cleared for redistribution. contradicts → the view that trial-to-trial variability in sensory cortex is primarily unstructured noise Quotation withheld because this record is not cleared for redistribution. builds-on → the anatomical fact that primary sensory cortex receives both neuromodulatory state input and higher-order cortical input Quotation withheld because this record is not cleared for redistribution. spontaneous activity —distinct-from→ recapitulation spontaneous activity recapitulation strength of the relation: 0.75 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the similarity 'need not imply recapitulation'; distinct-from is loose. Quotation withheld because this record is not cleared for redistribution. reliable variance —operationalised-as→ Shared Variance Component Analysis reliable variance Shared Variance Component Analysis strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. EVIDENCE CORRECTED (RE-GROUNDED) · stimulus-only subspace —distinct-from→ behavior-only subspace stimulus-only subspace behavior-only subspace EVIDENCE CORRECTED: RE-GROUNDED. The old span named neither subspace. This clause names both and states their independence, which is the distinctness the edge records. — strength of the relation: 0.95 Quotation withheld because this record is not cleared for redistribution. facial motion —causes→ multiplicative gain facial motion multiplicative gain strength of the relation: 0.85 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the gain 'could be predicted from' facial motion energy; causes is loose. Quotation withheld because this record is not cleared for redistribution. Shared Variance Components —operationalised-as→ reliable variance Shared Variance Components reliable variance strength of the relation: 0.9 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'the first SVC'; the endpoint is a variant, and operationalised-as is loose. Quotation withheld because this record is not cleared for redistribution. thalamus —part-of→ forebrain thalamus forebrain strength of the relation: 0.85 Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span lists recorded regions; part-of 'forebrain' is not stated. Quotation withheld because this record is not cleared for redistribution. Wilcoxon signed-rank test Wilcoxon signed-rank test Quotation withheld because this record is not cleared for redistribution. Wilcoxon rank-sum test Wilcoxon rank-sum test Quotation withheld because this record is not cleared for redistribution. correlation correlation Quotation withheld because this record is not cleared for redistribution. correlation correlation Quotation withheld because this record is not cleared for redistribution. principal component analysis principal component analysis Quotation withheld because this record is not cleared for redistribution. regression regression Quotation withheld because this record is not cleared for redistribution. cross-validation cross-validation Quotation withheld because this record is not cleared for redistribution. linear classification linear classification Quotation withheld because this record is not cleared for redistribution. standard deviation standard deviation Quotation withheld because this record is not cleared for redistribution. Shared Variance Component Analysis Shared Variance Component Analysis Quotation withheld because this record is not cleared for redistribution. reduced rank regression reduced rank regression Quotation withheld because this record is not cleared for redistribution. power law power law Quotation withheld because this record is not cleared for redistribution. two-sided Wilcoxon sign-rank test two-sided Wilcoxon sign-rank test Quotation withheld because this record is not cleared for redistribution. Wilcoxon rank-sum test Wilcoxon rank-sum test Quotation withheld because this record is not cleared for redistribution. cross-validated cross-validated Quotation withheld because this record is not cleared for redistribution. Ongoing population activity in mouse visual cortex reliably encoded a latent signal of at least 100 linear dimensions. Quotation withheld because this record is not cleared for redistribution. The single largest dimension of ongoing activity tracks arousal and drives roughly half the neurons up while driving the other half down, so it is not a global rate increase. Quotation withheld because this record is not cleared for redistribution. A high-dimensional description of facial movement predicts about twice as much neural variance as the three standard arousal measures (running, pupil diameter and whisking) combined. Quotation withheld because this record is not cleared for redistribution. The same multidimensional behaviour-related activity structure is not confined to visual cortex but is present in neurons sampled across the forebrain. Quotation withheld because this record is not cleared for redistribution. The stimulus and behaviour subspaces overlap primarily because both can change the population mean rate; the finer patterns of increase and decrease were essentially orthogonal. Quotation withheld because this record is not cleared for redistribution. Stimulus identity can be decoded accurately from the stimulus-only subspace but not at all from the behaviour-only or spontaneous-only subspaces, which perform no better than random dimensions. Quotation withheld because this record is not cleared for redistribution. Trial-to-trial response variability decomposes into multiplicative gain modulation inside the stimulus subspace plus additive ongoing activity in dimensions orthogonal to it. Quotation withheld because this record is not cleared for redistribution. Thalamic neurons are the most strongly behaviour-modulated population recorded, with 35.7% of their total variance predictable from facial video. Quotation withheld because this record is not cleared for redistribution. 97% of the variance in the first shared component was reliable, leaving 3% for independent noise Quotation withheld because this record is not cleared for redistribution. reliable spontaneous variance decayed across dimensions as a power law of exponent 1.14, steeper than for stimulus responses but still high-dimensional Quotation withheld because this record is not cleared for redistribution. the three explicit arousal variables together explained about half the variance of the first neural principal component Quotation withheld because this record is not cleared for redistribution. about 57% of neurons correlated positively with the first PC and the rest negatively, so two sub-populations alternate Quotation withheld because this record is not cleared for redistribution. multidimensional facial behaviour predicted roughly twice the variance the three arousal variables did Quotation withheld because this record is not cleared for redistribution. prediction kept improving up to at least 16 behavioural dimensions, suggesting visual cortex encodes at least that many motor dimensions Quotation withheld because this record is not cleared for redistribution. those dimensions captured 21% of total population variance against 11% for the three-variable model Quotation withheld because this record is not cleared for redistribution. adding running, pupil and whisking to the video improved prediction by less than 1%, even though running is not visible in the face video Quotation withheld because this record is not cleared for redistribution. every recorded brain area contained neurons across the whole activity continuum, so the structure is distributed rather than localised Quotation withheld because this record is not cleared for redistribution. thalamic neurons were the best predicted by behaviour, at 35.7% of total variance Quotation withheld because this record is not cleared for redistribution. behavioural prediction was best at zero lag and decayed over seconds, full width at half maximum 2.5 s Quotation withheld because this record is not cleared for redistribution. stimulus and behaviour subspaces overlapped in essentially one dimension, which is the population mean rate Quotation withheld because this record is not cleared for redistribution. a classifier decoded stimuli at 10% error from the stimulus subspace and at chance from the behaviour or spontaneous subspaces Quotation withheld because this record is not cleared for redistribution. the behaviour subspace was active before the stimulus and continued unchanged through it, while the stimulus subspace was silent then active Quotation withheld because this record is not cleared for redistribution. a purely multiplicative trial-varying gain accounted for 89% of the variance in the stimulus subspace Quotation withheld because this record is not cleared for redistribution. that gain was predictable from facial motion at r = 0.61 cross-validated Quotation withheld because this record is not cleared for redistribution. adding an additive term to the multiplicative model did not significantly improve it Quotation withheld because this record is not cleared for redistribution. stimulus and behaviour representations were mixed in the same neurons rather than carried by separate populations Quotation withheld because this record is not cleared for redistribution. recordings in darkness and with a gray screen did not differ Quotation withheld because this record is not cleared for redistribution. position on the sorted activity continuum bore little relation to position in the tissue Quotation withheld because this record is not cleared for redistribution. Shared Variance Component Analysis (SVCA) Shared Variance Component Analysis (SVCA) Quotation withheld because this record is not cleared for redistribution. Shared Variance Components (SVCs) Shared Variance Components (SVCs) Quotation withheld because this record is not cleared for redistribution. motion energy motion energy Quotation withheld because this record is not cleared for redistribution. Shared Variance Component Analysis Shared Variance Component Analysis Quotation withheld because this record is not cleared for redistribution. up and down phases up and down phases Quotation withheld because this record is not cleared for redistribution. facial motion facial motion Quotation withheld because this record is not cleared for redistribution. visual stimulation visual stimulation Quotation withheld because this record is not cleared for redistribution. dimensions of the video signal dimensions of the video signal Quotation withheld because this record is not cleared for redistribution. population activity population activity Quotation withheld because this record is not cleared for redistribution. running speed running speed Quotation withheld because this record is not cleared for redistribution. facial motion facial motion Quotation withheld because this record is not cleared for redistribution. reliable variance reliable variance Quotation withheld because this record is not cleared for redistribution. time lag time lag Quotation withheld because this record is not cleared for redistribution. multiplicative gain multiplicative gain Quotation withheld because this record is not cleared for redistribution. pupil area pupil area Quotation withheld because this record is not cleared for redistribution. whisking whisking Quotation withheld because this record is not cleared for redistribution. 11,262 ± 2,282 — mean and s.d. of the number of excitatory neurons simultaneously imaged, over nine sessions in six mice Quotation withheld because this record is not cleared for redistribution. 97% — fraction of the variance along the first shared variance component that was reliable rather than independent noise Quotation withheld because this record is not cleared for redistribution. 1.14 — exponent of the power law by which reliable spontaneous variance decays across dimensions Quotation withheld because this record is not cleared for redistribution. 21%± 1% SE — share of total population variance captured by the 16 predictive facial-behaviour dimensions (31% of the reliable variance) Quotation withheld because this record is not cleared for redistribution. 10.1 ± 4.0 % — cross-validated error rate for decoding which of 32 natural images was shown, from the stimulus subspace Quotation withheld because this record is not cleared for redistribution. 89% ± 0.1% SE — variance in the stimulus subspace explained by a purely multiplicative trial-gain model Quotation withheld because this record is not cleared for redistribution. r = 0.61 ± 0.02 SE — cross-validated correlation between the per-trial multiplicative gain and its prediction from facial motion energy Quotation withheld because this record is not cleared for redistribution. less than 1% — extra explained variance gained by adding running, pupil and whisking to the facial video predictors Quotation withheld because this record is not cleared for redistribution. 11,262 ± 2,282 neurons — excitatory neurons imaged per session, over nine sessions in six mice Quotation withheld because this record is not cleared for redistribution. 97% — reliable variance fraction of the first shared variance component Quotation withheld because this record is not cleared for redistribution. exponent 1.14 — power law decay of reliable spontaneous variance across dimensions Quotation withheld because this record is not cleared for redistribution. 16 dimensions — behavioural dimensions at which neural prediction saturated Quotation withheld because this record is not cleared for redistribution. 21% ± 1% SE against 11% ± 1% SE — total variance explained by videographic behaviour versus the three arousal variables Quotation withheld because this record is not cleared for redistribution. 35.7% — thalamic variance predicted from behaviour, the highest of any region Quotation withheld because this record is not cleared for redistribution. 2.5±0.4 s — full width at half maximum of the behavioural prediction across time lags Quotation withheld because this record is not cleared for redistribution. 10.1 ± 4.0 % — stimulus decoding error from the stimulus subspace, against 53% and 57% from the others Quotation withheld because this record is not cleared for redistribution. 89% ± 0.1% SE — variance in the stimulus subspace captured by a multiplicative gain model Quotation withheld because this record is not cleared for redistribution. r = 0.61 ± 0.02 SE — cross-validated prediction of the trial gain from facial motion Quotation withheld because this record is not cleared for redistribution. 2,296, 2,668 and 1,462 units — units held stable for about an hour in the three electrophysiology mice Quotation withheld because this record is not cleared for redistribution. 57% ± 6.7% SE — share of neurons positively correlated with the first principal component Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1126/science.aav7893.