TL;DR
This paper reveals that despite individual differences in brain structure, animals of the same species exhibit remarkably similar patterns of neural activity (latent dynamics) when performing the same behavior. These shared dynamics are not limited to motor cortex but extend to older brain regions and even during covert planning, suggesting they are a fundamental, evolutionarily conserved property of behavior.
What Problem Does This Paper Address?
We observe that animals within a species perform remarkably similar behaviors, like reaching for an object. This is fascinating because each individual animal has a unique brain, with variations in neuron numbers, connections, and other cellular details. The big question is: how can such consistent, species-typical behaviors emerge from such idiosyncratic neural circuitry? One possibility is that each unique brain generates unique neural activity patterns to achieve the same outcome. This paper challenges that idea, proposing instead that there are underlying, shared "latent dynamics" – specific patterns of neural population activity – that are conserved across individuals and are the true basis of species-specific behaviors.
Methods
The core of this research involves analyzing neural recordings from monkeys and mice performing specific motor tasks.
- Neural Recording: High-density electrode arrays were used to record neural activity from motor cortex and the dorsal striatum.
- Latent Dynamics Estimation: Principal Component Analysis (PCA) was employed to reduce the high-dimensional neural activity data into a lower-dimensional "neural manifold," representing the dominant patterns of neural co-variation (latent dynamics).
- Alignment of Dynamics: Canonical Correlation Analysis (CCA) was used to "align" the latent dynamics from different animals. This method finds linear transformations that maximize the correlation between the neural activity patterns of two individuals, effectively comparing their underlying dynamics.
- Decoding: Long Short-Term Memory (LSTM) networks were trained to predict hand trajectories based on the aligned neural dynamics. This tested whether the shared dynamics contained behaviorally relevant information.
- Recurrent Neural Networks (RNNs): To control for the possibility that behavioral similarity alone could explain the observed neural similarity, the researchers trained RNNs to perform the same task. They then compared the latent dynamics of "standard" RNNs with "constrained" RNNs designed to have distinct dynamics despite similar behavioral output.
Datasets
The study utilized neural recordings from:
- Monkeys: Three macaque monkeys performing a center-out reaching task.
- Mice: Four mice performing a grasping and pulling joystick task.
These datasets were collected by the researchers and, in the case of monkey data, some are publicly available. No new benchmarks were introduced; rather, existing datasets were used to investigate the phenomenon of preserved neural dynamics.
Key Insights and Ideas
- Shared Neural Blueprints: The most striking finding is that animals of the same species, despite individual brain differences, share remarkably similar neural "blueprints" (latent dynamics) when performing the same behavior. This suggests a conserved neural strategy for common tasks.
- Beyond the Cortex: These preserved dynamics aren't confined to the motor cortex, which is directly involved in movement execution. They extend to the dorsal striatum, an evolutionarily older brain structure, implying a deeper, more fundamental neural basis for behavior.
- Planning and Action are Linked: The same preserved dynamics are observed not only during overt movement but also during covert movement planning, suggesting that the neural mechanisms for preparing and executing actions are tightly linked and share common underlying dynamics.
- Behavioral Stereotypy Matters: The degree of similarity in neural dynamics across individuals is directly related to how stereotyped their behavior is. More consistent behavior leads to more consistent neural dynamics.
- Evolutionary Constraints: The findings suggest that these preserved dynamics are not accidental but are likely shaped by evolutionary constraints on brain development, providing a "generative model" for behavior.
Key Results
- High Correlation: Aligned latent dynamics across different monkeys performing the same reaching task showed correlations nearly as high as those within the same monkey performing the task.
- Cross-Species Decoding: Decoders trained on the aligned latent dynamics of one monkey could predict the hand position of another monkey with high accuracy, comparable to decoders trained and tested on the same animal.
- Striatum Preservation: Similar preservation of latent dynamics was observed in the mouse dorsal striatum, further supporting the idea that these dynamics are widespread.
- Planning Dynamics: Latent dynamics during covert movement planning in monkeys were also highly correlated across individuals and predictive of behavior.
- Behavioral Link: The similarity in neural dynamics across individuals was strongly correlated with the similarity of their behavioral output.
- RNNs Confirm: Trained RNNs demonstrated that while similar behavior can be achieved, distinct neural dynamics can arise if not constrained, reinforcing that behavioral similarity alone doesn't guarantee neural similarity.
Limitations and Future Work
- Causality: While the study shows a strong correlation between preserved latent dynamics and behavior, it doesn't definitively prove causality. Future work could explore causal links through interventions.
- Generalizability: The study focused on motor tasks. Investigating whether these preserved dynamics extend to more complex cognitive or social behaviors would be valuable.
- Mechanisms: The paper posits evolutionary constraints as the driver of these dynamics. Further research could delve into the specific developmental and genetic mechanisms that enforce these constraints.
- Cross-Species Differences: While the study highlights similarities, understanding the specific differences in latent dynamics that contribute to species-specific behavioral nuances remains an open question.
Summary updated