Source-linked AI summary
Large-scale neural recordings call for new insights to link brain and behavior
Anne E. Urai, Brent Doiron, Andrew M. Leifer, Anne K. Churchland
TL;DR
Large-scale neural recordings now expose population activity at scales that challenge neuron-centered accounts of computation and behavior. The paper reviews recording and behavioral-analysis advances, synthesizes unexpected population-level insights, and examines theoretical frameworks for interpreting them. It concludes that the field has answered key questions while raising new ones requiring further theoretical development.
Problem
Connecting brain-wide neural recordings to computation and behavior remains difficult, especially for neural variability distributed across brain regions and for distinguishing mechanistic models.
Method
The paper reviews technological advances, large-scale neural and behavioral measurements, empirical insights, and theoretical frameworks for interpreting population activity and variability.
Results
Large-scale recordings reveal distributed and sparse representations, population-level computations hidden in single-neuron rates, low-dimensional behaviorally relevant variance, and structured responses from largely unstructured networks.
Takeaways & Limitations
Large-scale recording has changed how neural population activity is analyzed and understood, while motivating new theoretical approaches and deeper attention to behavior, internal states, and circuit variability.
Takeaways & Limitations
Observational large-scale recordings may not distinguish concrete mechanistic models, and standardized spike sorting remains a significant challenge requiring manual curation.
Abstract
from arXiv · showhide
Neuroscientists today can measure activity from more neurons than ever before, and are facing the challenge of connecting these brain-wide neural recordings to computation and behavior. Here, we first describe emerging tools and technologies being used to probe large-scale brain activity and new approaches to characterize behavior in the context of such measurements. We next highlight insights obtained from large-scale neural recordings in diverse model systems, and argue that some of these pose a challenge to traditional theoretical frameworks. Finally, we elaborate on existing modelling frameworks to interpret these data, and argue that interpreting brain-wide neural recordings calls for new theoretical approaches that may depend on the desired level of understanding at stake. These advances in both neural recordings and theory development will pave the way for critical advances in our understanding of the brain.
Introduction
Large-scale recordings expand access to simultaneous neural population activity, enabling questions about distributed representations, interactions with movement and internal signals, and links between neural activity and behavior. The review presents technological progress, four unexpected empirical insights, and theoretical challenges arising from these data.
- Large-scale recordings increase statistical power, reduce reliance on hand-selected neurons and areas, and enable broader surveys of neural responses across brain regions and cell types.
- The review asks how neural representations are distributed, how task-related signals interact with movement and arousal, and how much neural variability is behaviorally relevant.
- The authors identify four insights: representations are distributed and sparse, population dynamics can hide single-neuron computations, behaviorally relevant variance is often low-dimensional, and unstructured networks can yield structured responses.
- These findings challenge theoretical frameworks centered on individual neurons and motivate new approaches for interpreting large-scale neural data.
Insights from large-scale neural recordings
Large-scale and simultaneous recordings reveal distributed, sparse, movement-sensitive, and often low-dimensional population activity, while exposing computations and communication patterns that single-neuron analyses can miss. They also motivate circuit models that incorporate population dynamics, variability, and connectivity.
- Distributed and sparse representations: Choice signals in mice were sparse and widely distributed, with plentiful responses in deep structures; only ~18% of V1 neurons responded to visual gratings in one study.A broader stimulus battery identified receptive fields in up to 70% of V1 neurons, while calcium imaging found many neurons unresponsive to visual stimuli.
- Distributed and sparse representations: Movement-related activity, including task-unrelated fidgets, was stronger and more widely distributed than expected, even during expert cognitive behavior.Unsupervised video analysis captured diverse movements beyond running and pupil diameter.
- Low-dimensional activity: Behaviorally relevant neural variance can often be explained by a small number of dimensions, with a monkey visual-cortex choice decoder using the first principal component performing almost as well as one using the whole dataset.
- Limits of dimensionality: The dimensional structure of V1 responses can balance coding efficiency with robustness to small perturbations in visual images.
- Population dynamics: Population dynamics can reveal movement planning and decision-making computations even when single neurons lack obvious tuning to task variables.This motivates frameworks in which representations are contained in population dynamics rather than static individual-neuron responses.
- Population structure and dynamics: Transient single-neuron responses can coexist with stable, lower-dimensional population coding, while minimally structured networks can generate sequential activity with robustness and flexibility.Sequential firing can emerge through cooperation between recurrent synaptic interactions and external inputs.
- Communication and connectivity: Simultaneous recordings expose shared fluctuations across regions and can connect neural co-fluctuations to behavior, while connectomes make circuit interpretations and model predictions more concrete.In Drosophila, precise PFN–hΔB synaptic-weight offsets supported a circuit model for transforming egocentric into allocentric coordinates.
Box 1: History and future of large-scale neural recordings
Whole-brain single-neuron measurements are feasible in small, transparent animals, while recording technologies have scaled toward larger populations through electrophysiology and optical imaging. Figure 2 situates these measurements against species’ brain sizes and extrapolates future growth.
- Simultaneous whole-brain measurements of single-neuron activity have been acquired in C. elegans, larval zebrafish, and hydra.
- Figure 2 plots simultaneously recorded neuron counts for electrophysiology and optical imaging, with exponential fits and present/future extrapolations compared against approximate species brain sizes.
Box 2. Tools and technologies to observe brain and behavior
New recording and behavioral-quantification technologies greatly expand the scale, coverage, and ecological relevance of brain–behavior measurements. Each modality offers distinct strengths, while spike sorting and calcium kinetics remain important constraints.
- Neuropixels probes with up to ~10,000 recording sites simultaneously sample large neural populations spanning multiple brain areas.
- Spike sorting remains a major electrophysiology challenge because densely spaced recordings still require substantial manual curation and lack consensus standards.Simultaneous juxtacellular and extracellular recordings provide ground-truth benchmarks for evaluating sorting algorithms.
- Optical imaging provides high spatial resolution and coverage, cell-type or projection-target labeling, spatial activity patterns, and long-term monitoring.In mice, imaging has reached one million simultaneously recorded cells, about a tenth of the cortex.
- Calcium imaging measures spike timing and rate only coarsely because calcium kinetics and indicator dynamics are slower than neural firing.
- Whole-brain imaging in transparent animals can capture most neurons at cellular resolution during natural movement such as swimming or crawling.C. elegans recordings have also served as a testbed for nonlinear dynamical-systems models of whole-brain activity.
- Video tracking and processing enable data-driven parsing of spontaneous behavior, allowing neural activity to be interpreted during both controlled tasks and ethological behaviors.
Theoretical frameworks: more is different
Large-scale recordings expose gaps in mechanistic theories of brain-wide dynamics while motivating models that connect variability across spatial scales and behavior. The review therefore favors complementary models of varying complexity rather than a single unified brain simulation.
- Theoretical frameworks: more is different: Descriptive recordings cannot by themselves distinguish mechanistic models, because diverse variables can produce similar neural signatures.For example, ramping activity may resemble evidence accumulation while also arising from other processes.
- Theoretical frameworks: more is different: Trial-to-trial variability provides a window into interactions spanning synapses, membranes, local circuits, and whole-brain networks.At local scales, balanced excitatory-inhibitory networks can account for correlated and low-dimensional shared variability.
- Theoretical frameworks: more is different: The review proposes a hybrid framework in which high-level and fine-grained models coexist to explain different features of large-scale data.This avoids requiring one model to reproduce every observed brain computation.
- Theoretical frameworks: more is different: Communication subspaces show that only a small subset of V1 population variability predicts V2 variability, with largely non-overlapping feedback dimensions.This suggests selective routing of activity and reduced downstream co-fluctuations.
- Theoretical frameworks: more is different: Internal states and spontaneous behaviors are important predictors of neural variability, but their integration with circuit computations remains largely descriptive.The review argues that future theories must accommodate these varied signals alongside core neural computations.
- Theoretical frameworks: more is different: Large-scale recordings reveal richer brain-wide dynamics but also expose gaps in existing tools for interpreting neural activity.Whole-brain recordings in small animals show both predictive promise and substantial unexplained activity.
Conclusions and outlook
The authors expect continued recording and computational advances to improve behavioral prediction and clarify distributed neural computation. They also emphasize that neuron recordings alone will not fully support causal inference or complete understanding of the brain.
- Conclusions and outlook: Large-scale neural recordings have grown by orders of magnitude and have answered important questions while raising new theoretical ones.The authors frame this expansion as a source of both empirical progress and unresolved challenges.
- Conclusions and outlook: Behavioral prediction and decoding may extend to larger-brained species, with potential translational impact for brain-computer interfaces.This expectation builds on successes in small invertebrate models.
- Conclusions and outlook: Future progress will require richer cross-region recordings, behavior-centered analysis, improved dimensionality reduction, and direct cross-species comparisons.These developments are expected to clarify variability, distributed computations, circuit complexity, and transfer of insights across species.
- Conclusions and outlook: Recording many neurons, even with quantified behavior and a connectome, may remain insufficient for causal inference and behavioral prediction.The authors identify a full directed network description with synaptic weights, along with inaccessible biological signals, as remaining gaps.
Citation diversity statement
The authors explicitly considered citation diversity because women and minority scholars are under-cited relative to their representation. They quantified reference gender balance and compared it with expected proportions from five leading neuroscience journals.
- Citation diversity statement: Citation practices can under-cite papers from women and other minority scholars relative to their representation in the field.The statement motivates a proactive approach to reference selection.
- Citation diversity statement: The authors sought references reflecting field diversity and quantified citation gender balance from first names of first and last authors.Expected proportions were estimated from five top neuroscience journals since 1997.
- Citation diversity statement: Expected reference proportions were 6.7% woman/woman, 9.4% man/woman, 25.5% woman/man, and 58.4% man/man.These proportions provide the comparison baseline for the authors’ reference list.