TL;DR
This study reveals that the dorsal premotor cortex (PMd) in monkeys doesn't just represent the chosen reach direction, but rather a distribution of potential actions, especially when faced with uncertainty. This representation persists even during movement execution, suggesting PMd's role extends beyond simple decision-making to maintaining a flexible plan.
What Problem Does This Paper Address?
When we make a movement, especially reaching for an object, we often rely on imperfect sensory information. This means we're not always certain about the exact target location. While previous research has shown that areas like PMd can represent multiple potential actions when given explicit choices, it was unclear how PMd handles situations where the target is ambiguous and requires estimation rather than selection. This paper investigates how the brain, specifically PMd and the primary motor cortex (M1), represents reach direction when the target information is noisy and uncertain, and how this representation evolves during movement planning and execution.
Methods
The researchers trained two monkeys (Rhesus and Cynomolgus macaques) to perform a reaching task using a robotic manipulandum. The task involved two main conditions:
- Center-out block (zero uncertainty): Monkeys reached to one of eight clearly defined targets.
- Uncertainty blocks: Targets were sampled from a probabilistic distribution (von Mises), and the visual cue was a set of noisy lines, with varying degrees of uncertainty (low and high variance).
They recorded neural activity from both PMd and M1 using microelectrode arrays. To analyze the neural data, they:
- Characterized directional tuning: Determined each neuron's preferred direction (PD) for reaching.
- Quantified uncertainty effects: Compared neural activity between low and high uncertainty conditions, focusing on neurons with PDs aligned with (Same Direction - SD), opposite to (Opposite Direction - OD), or orthogonal to (Orthogonal Direction - ORTH) the reach direction.
- Spatiotemporal activity maps: Visualized how neural activity changed over time and across different directional tuning groups.
- Decoding analysis: Used neural activity to predict the monkey's chosen reach direction.
- Behavioral analysis: Modeled the monkeys' choices using a Bayesian framework to estimate their weighting of visual cues versus prior knowledge and their overall behavioral uncertainty.
Datasets
The study utilized data collected from two monkeys performing the described reaching task. No new benchmark datasets were introduced; rather, the experimental design itself created novel conditions of varying target uncertainty.
Key Insights and Ideas
- PMd maintains a "distribution of actions": Unlike M1, which seems to commit to a single reach direction, PMd neurons show activity patterns that reflect a range of potential movements, particularly when uncertainty is high.
- Uncertainty amplifies "unlikely" representations: When monkeys were more uncertain about the target, PMd neurons with preferred directions away from the most likely reach direction (OD and ORTH neurons) showed increased activity. This suggests PMd is actively representing less probable actions.
- Persistence of uncertainty effects: The influence of uncertainty on PMd activity wasn't confined to the planning phase; it persisted throughout movement execution, a finding not previously observed in target selection tasks.
- Behavioral uncertainty drives neural changes: The degree of uncertainty-related neural activity in PMd was strongly correlated with the monkeys' own estimated behavioral uncertainty, not just the objective properties of the visual cue or prior.
- M1 is less affected by uncertainty: Primary motor cortex (M1) activity primarily reflected the executed reach direction and was largely insensitive to the level of uncertainty, suggesting it receives a more "denoised" signal.
Key Results
- PMd activity increases with uncertainty: Neurons with preferred directions orthogonal or opposite to the reach direction showed significantly higher firing rates during high uncertainty trials compared to low uncertainty trials. This effect was most pronounced during the delay and movement periods.
- Correlation with behavioral uncertainty: The difference in OD and ORTH activity between high and low uncertainty conditions was positively correlated with the difference in behavioral uncertainty across sessions.
- Decoding accuracy: PMd decoding of reach direction was less accurate in high uncertainty trials compared to low uncertainty trials, and this difference persisted throughout the trial. M1 decoding was generally more accurate and less affected by uncertainty, especially at the time of movement.
- No significant M1 uncertainty effects: Unlike PMd, M1 neurons did not show significant modulation of their activity based on uncertainty levels.
Limitations and Future Work
- Theoretical model comparison: The authors note that their experimental design (static cues, instructed delay) limits a comprehensive test of existing theoretical models of uncertainty.
- Persistence of representation: While the persistence of uncertainty effects in PMd is a key finding, the authors suggest future work should explore why this representation is maintained for target estimation but not necessarily for target selection.
- Functional role of persistent representation: Future experiments could investigate if this heightened representation of alternative actions in PMd aids in rapid error correction or facilitates subsequent motor learning.
- Mechanism of "denoising": The paper proposes that the connections between PMd and M1 might "denoise" the PMd signal to produce a single, unambiguous motor command. Further research could explore this interaction and potential MAP (maximum a posteriori) readout mechanisms.
Summary updated