Evolutionarily conserved neural dynamics across mice, monkeys, and humans

This study demonstrates that despite vast evolutionary distances and behavioral differences, mice, monkeys, and humans share highly conserved neural computations and population dynamics in the motor cortex for movement, suggesting that evolution preserves these fundamental circuit mechanisms even as behavioral repertoires expand.

The Gist

The Big Question: Are Our Brains Built on the Same Blueprint?

Imagine you walk into a hardware store and see three different types of vehicles: a tiny, nimble mouse, a powerful monkey, and a sophisticated human. They all look very different on the outside. They have different body shapes, different hand sizes, and they move in different ways.

Scientists have long wondered: Do these different vehicles use the same engine?

In other words, even though a mouse, a monkey, and a human are separated by millions of years of evolution, do their brains use the exact same "math" or "rules" to figure out how to reach out and grab an object? Or does each species have to reinvent the wheel every time?

This paper says: Yes, the engine is the same.

The Experiment: A Multi-Species "Reach" Test

To test this, the researchers set up a simple game for three very different players:

1. Mice: They had to reach for a lever, grab it, and pull it.
2. Monkeys: They had to reach for various objects (cylinders or cubes), grab them, and pull them.
3. A Human: A participant with a spinal cord injury (who had to use a special wrist movement to grab objects) reached for and moved a cylinder.

While they played these games, the scientists put tiny microphones (electrodes) inside the motor cortex of their brains. This is the part of the brain responsible for planning and executing movement. They listened to the "chatter" of thousands of neurons firing at once.

The Discovery: The "Flow Field" is Identical

Here is the magic part. When the scientists looked at the data, they didn't just look at what the neurons did; they looked at how they did it over time.

Imagine the brain's activity as a river flowing through a landscape.

• The shape of the river (where it goes, how fast it flows, the curves it makes) represents the neural dynamics (the rules of the computation).
• The water itself represents the specific movement (reaching for a small mouse lever vs. a large human cup).

The researchers found that the river flowed through the exact same landscape for the mouse, the monkey, and the human. Even though the mouse was small and the human was large, and even though they were grabbing different things, the "flow field" of their brain activity was nearly identical.

The Analogy: Think of it like a dance.

• A mouse, a monkey, and a human are all dancing to the exact same song (the conserved neural dynamics).
• However, because they have different body sizes, they take different steps. The mouse does a tiny shuffle; the human does a big stride.
• The music (the brain's internal rules) is the same, but the choreography (the specific movement path) changes to fit the dancer's body.

Why This Matters: It's Not Just "Same Species"

To prove this wasn't a fluke, the scientists compared the brain activity in other situations where the "music" should be different:

1. Different Rooms: They compared the motor cortex (movement) to the somatosensory cortex (feeling/touch) in the human. These are neighbors in the brain, but they do different jobs. The "dance music" here was totally different.
2. Different Phases: They compared the monkey's brain when planning a move vs. when doing the move. Again, the music changed.

But when they compared the mouse, monkey, and human all doing the same type of movement (reaching and grabbing), the music was more similar to each other than it was to the other parts of the same brain.

This is a huge deal. It means that evolution didn't just tweak the body; it kept the core "software" of the motor brain exactly the same for over 100 million years.

The "Why": Circuit Constraints

The researchers also built computer simulations (digital brains) to see if this happens by accident or by design. They found that when you build a neural network with specific "architectural constraints" (like how the wires are connected), it naturally produces these same flow patterns.

The Metaphor: Imagine you are trying to pour water into a cup. No matter if you use a tiny thimble or a giant bucket, if the cup has the same shape, the water will flow in the same pattern. The "shape of the cup" is the brain's circuitry. Evolution kept the shape of the cup the same, even as the size of the bucket (the animal) changed.

The Takeaway

1. Shared Heritage: The way our brains compute movement is a shared family heirloom. We didn't invent new math for our brains; we just repurposed the old, efficient math our mouse ancestors had.
2. Geometry vs. Dynamics: The rules of the movement (dynamics) are conserved, but the shape of the path (geometry) changes to fit the specific animal and task.
3. Hope for Medicine: Because the "engine" is the same in mice, monkeys, and humans, we can be much more confident that treatments or brain-computer interfaces developed in animals will work in humans. We are all driving the same model of car, just with different body kits.

In short: Evolution is a frugal engineer. It didn't redesign the brain's operating system for every new species; it just updated the user interface to fit the new body.

Technical Summary

1. Problem Statement

Despite the vast phylogenetic distance and anatomical differences between mammals (e.g., rodents vs. primates), they share homologous brain regions, such as the motor cortex, which control similar behaviors like reaching and grasping. A fundamental question in neuroscience remains unresolved: Do these homologous regions utilize conserved neural computations across species, or have species-specific adaptations fundamentally altered the computational substrate?

While anatomical similarities exist, differences in circuitry (e.g., monosynaptic corticospinal connections in primates vs. polysynaptic in rodents) and behavioral repertoires suggest potential divergence in how neural activity is processed. The authors aim to determine if the underlying "dynamical rules" governing neural population activity are evolutionarily conserved, even as the specific behavioral outputs (geometry of movement) vary.

2. Methodology

A. Multi-Species Dataset Compilation

The authors compiled intracortical neural recordings from three species performing skilled forelimb tasks:

• Mice ($N=4$): Trained on a lever-pull task with varying positions and loads. Recorded via Neuropixels probes.
• Monkeys ($N=4$): Trained on object-pull tasks (cylindrical and cubic objects) with varying grip types and loads. Recorded via chronically implanted microelectrode arrays (Utah and floating arrays).
• Human ($N=1$): A participant with C4-level spinal cord injury performing an object transport task using a modified tenodesis grasp. Recorded via NeuroPort microelectrode arrays in both motor and somatosensory cortices.

B. Data Processing

• Dimensionality Reduction: Neural population activity was projected into a 20-dimensional latent space using Principal Component Analysis (PCA) to capture the majority of variance.
• Trajectory Extraction: Neural trajectories were extracted for specific movement phases (reaching, grasping, pulling/manipulation).

C. Analytical Frameworks

1. Dynamical Similarity Analysis (DSA):
   • Used to quantify the similarity of the underlying dynamical systems (the "rules" of evolution) rather than just the trajectory shapes.
   • Mechanism: DSA linearizes the nonlinear neural dynamics via time-delay embedding and fits a state-transition matrix ($K$) using Dynamic Mode Decomposition (DMD).
   • Metric: A "dynamical distance" is calculated by finding the optimal linear transformation ($C$) that aligns the state-transition matrices of two datasets. A low distance indicates conserved dynamical rules.
2. Canonical Correlation Analysis (CCA):
   • Used to quantify geometric similarity (the shape of the trajectory in state space).
   • Trajectories were time-interpolated to a fixed length to remove speed/temporal scaling effects, isolating geometric structure.
3. Recurrent Neural Network (RNN) Simulations:
   • To test causality, the authors trained RNNs (using the MotorNet toolbox) to control a biomechanical primate arm simulation.
   • They manipulated architectural variables (e.g., GRU vs. Vanilla RNN, learning rates, sensory feedback, muscle non-linearities) to observe how circuit constraints affect dynamics and geometry.

3. Key Contributions

• First Cross-Species Dynamical Comparison: This is the first study to directly compare the dynamical rules of motor cortical population activity across mice, monkeys, and humans.
• Decoupling Dynamics and Geometry: The study establishes a theoretical and empirical distinction between the dynamics (the flow field/rules) and the geometry (the specific path taken) of neural trajectories.
• Circuit Constraint Validation: Through RNN simulations, the authors demonstrate that conserved dynamics arise from shared circuit architectural constraints, not just behavioral similarity.

4. Key Results

A. Conservation of Neural Dynamics Across Species

• Finding: The dynamical distance between motor cortex recordings of mice, monkeys, and humans was statistically indistinguishable from the dynamical distance observed between different individuals of the same species.
• Implication: Despite millions of years of evolution and significant anatomical differences (e.g., body mass, limb length, corticospinal tract organization), the motor cortex employs highly conserved computational rules to generate movement.

B. Dynamics Change with Computational Demands (Controls)

To validate that the observed conservation was specific to the computational task and not an artifact, the authors showed that dynamics diverge when computations change:

• Motor vs. Somatosensory Cortex: In the human participant, the dynamical distance between the motor cortex and the simultaneously recorded somatosensory cortex was high (comparable to the "shuffle" control), indicating distinct computational rules.
• Preparation vs. Execution: In monkeys, the dynamical distance between motor preparation and execution phases was significantly higher than the distance between species during execution.
• Conclusion: Neural dynamics are conserved for a specific computation (movement execution) but change when the computational goal changes, regardless of species.

C. Behavioral Variability is Encoded in Geometry

• Finding: While dynamics were conserved, the geometric similarity (trajectory shape) varied significantly.
  • Geometric similarity strongly correlated with behavioral similarity (kinematics).
  • Human trajectories showed almost no geometric alignment with mice or monkeys, despite sharing the same underlying dynamics.
• Interpretation: The motor cortex uses a conserved dynamical system (flow field) but generates species-specific behaviors by tracing different paths (geometries) through that field.

D. RNN Simulation Insights

• Architecture Matters: Small changes in RNN architecture (e.g., switching from GRU to Vanilla RNN) or learning parameters drastically altered both dynamics and geometry.
• Conservation Requires Shared Constraints: Only networks with similar architectural constraints produced both similar dynamics and similar geometries.
• Decoupling: Networks could produce similar behaviors with different geometries, but producing similar behaviors with different dynamics was difficult. This reinforces that the observed cross-species dynamical conservation in biological data likely stems from shared evolutionary circuit constraints.

5. Significance and Implications

• Evolutionary Insight: The study suggests that evolution preserves the core "dynamical engine" of the motor cortex while allowing the "output geometry" to adapt to specific ecological niches and biomechanical constraints.
• Translational Neuroscience: The finding that neural computations are conserved across species validates the use of animal models (mice, monkeys) for understanding human motor control and developing neural interfaces.
• Brain-Computer Interfaces (BCI): The results support the development of "multi-species neural foundation models." If the underlying dynamics are conserved, algorithms trained on animal data may generalize more effectively to humans, accelerating the clinical translation of neuroprosthetics and neuromodulation therapies.
• Theoretical Framework: It provides a rigorous framework for distinguishing between the rules of neural computation (dynamics) and the expression of behavior (geometry), offering a new lens for studying neural coding across the animal kingdom.