Area- and Layer-Specific Organization of Multimodal Timescales in Macaque Motor Cortex

By analyzing intrinsic timescales from both spiking activity and local field potentials across cortical layers in macaques, this study reveals a hierarchical organization where the primary motor cortex (M1) occupies a higher level than the dorsal premotor cortex (PMd), while also demonstrating that laminar temporal patterns differ significantly between signal modalities due to their distinct physiological origins.

The Gist

The Big Picture: Who's the Boss in the Motor Cortex?

Imagine your brain's motor system (the part that tells your body how to move) as a corporate office building. In the visual system (how we see), we know exactly who the bosses are: the lower floors handle simple details like "that's a red line," while the top floors handle complex ideas like "that's a stop sign."

But in the motor cortex (the part that plans and executes movement), the organizational chart has been a mystery. We have two main departments:

1. PMd (Dorsal Premotor Cortex): Think of this as the Project Manager. It's busy planning the "what" and "when" of a movement.
2. M1 (Primary Motor Cortex): Think of this as the Foreman on the Construction Site. It's the one actually shouting orders to the muscles to move.

For a long time, scientists argued about who was higher up in the hierarchy. Is the Project Manager (PMd) the boss giving orders to the Foreman (M1)? Or is the Foreman (M1) actually the one with the bigger picture, just doing the heavy lifting?

Because the motor cortex looks messy under a microscope (it lacks the clear "floors" or layers seen in the visual system), anatomy alone couldn't solve the mystery. So, the researchers decided to listen to the rhythm of the workers instead of just looking at the blueprints.

The Method: Listening to the "Pulse" of the Brain

The researchers recorded two types of signals from monkeys performing a reaching task (like a "Simon Says" game where they had to remember a direction):

1. Spikes (SUA): These are the individual neurons firing. Imagine this as listening to individual employees talking on their phones.
2. LFP (Local Field Potentials): This is the electrical hum of the whole neighborhood. Imagine this as the background noise of the office—the hum of computers, the chatter of the crowd, the vibration of the floor.

They measured "Intrinsic Timescales."

• Short Timescale: A worker who reacts instantly, changes their mind quickly, and forgets things fast. (Fast, reactive).
• Long Timescale: A worker who holds onto information for a long time, thinks deeply, and keeps a steady plan. (Slow, persistent).

The Findings: The Hierarchy Revealed

1. Who is the Boss? (Area Comparison)

The researchers found that M1 (the Foreman) actually has longer timescales than PMd (the Project Manager).

• The Analogy: The neurons in M1 are like slow-cooking stews. They hold onto the "plan" for a long time, integrating information slowly and steadily. The neurons in PMd are like fast-food orders—they react quickly to new cues but don't hold the information as long.
• The Conclusion: In the brain's hierarchy, the area with the longer timescale is usually the "higher" level. This suggests M1 is actually higher up in the hierarchy than PMd. It's not just the muscle-pusher; it's the one maintaining the long-term stability of the movement plan.

2. The Floor Plan: Deep vs. Shallow (Layer Comparison)

Here is where it gets weird. The researchers looked at the "floors" of the building (the layers of the cortex).

• The Theory: Scientists thought the "input floor" (shallow layers) would be fast/reactive, and the "output floor" (deep layers) would be slow/stable.
• The Reality (Spikes): When they listened to individual employees (spikes), there was no difference between the top floor and the bottom floor. Everyone seemed to be working at the same speed.
• The Reality (LFP/Background Noise): When they listened to the office hum (LFP), the deep floors were much slower and more stable than the shallow floors.

Why the difference?

• Spikes are the output of a single neuron. It's like hearing a single person's voice.
• LFPs are the inputs and the dendritic processing (the "listening" part of the neuron). The deep layers have long "antennas" (dendrites) that reach all the way to the top, collecting signals from everywhere. This makes the deep layers act like a big, slow, integrating pool of information, even if the individual neurons firing out of them look fast.

3. What Does This Mean for Movement?

They looked at how these "fast" and "slow" workers helped the monkey move.

• During Planning: The "slow" workers (long timescales) in the PMd were better at holding the plan steady while waiting for the "Go" signal. They didn't forget the direction.
• During Execution: When the monkey actually moved, the "slow" workers in both areas created a wider, more stable signal. It was like a smooth, continuous river of movement, whereas the "fast" workers created a choppy, flickering stream.

The Takeaway

1. M1 is the High-Level Boss: Even though it looks like the "motor output" area, it actually sits higher in the hierarchy than the premotor area because it holds information longer and more stably.
2. Signal Matters: You can't just look at one type of brain signal. If you only listen to individual neurons (spikes), you miss the story. If you listen to the crowd noise (LFP), you see that the deep layers are the true "integrators" of the brain.
3. Stability is Key: To move smoothly, the brain needs a mix of fast reactions and slow, stable holding. The "slow" neurons are the glue that keeps your movement plan from falling apart before you even start moving.

In short: The brain isn't just a ladder of "fast" at the bottom and "slow" at the top. It's a complex building where the "deep" layers act as a slow, stable foundation, and the "Primary Motor Cortex" is the high-level architect keeping the whole plan together.

Technical Summary

1. Problem Statement

The hierarchical organization of the primate motor cortex remains debated, particularly regarding the relative positions of the dorsal premotor cortex (PMd) and the primary motor cortex (M1).

• Anatomical Ambiguity: Unlike sensory systems, the motor cortex has an agranular architecture (lacking a distinct Layer IV) and complex, reciprocal connectivity patterns. This obscures the distinction between feedforward and feedback pathways, making it difficult to determine if PMd projects to M1 (descending) or vice versa (ascending).
• Lack of Functional Markers: While intrinsic temporal properties (timescales) have been established as markers of cortical hierarchy in sensory and prefrontal areas (where higher areas exhibit longer timescales), their application to the motor system is limited.
• Laminar Uncertainty: It is unknown whether temporal organization varies systematically across cortical layers (superficial vs. deep) within motor areas, and whether different neural signal modalities (spikes vs. Local Field Potentials) reveal consistent laminar gradients.

2. Methodology

The study employed a multimodal approach using acute laminar recordings from two adult male macaques performing a Delayed-Match-to-Sample (DMS) reaching task.

• Subjects & Task: Two macaques performed a task requiring them to remember a spatial target based on a color-coded cue, followed by a delayed period before executing a center-out reach.
• Recording:
  • Sites: Simultaneous recordings from PMd and M1 in the left hemisphere (contralateral to the right arm).
  • Probes: Acute laminar probes (linear arrays) and single-tip electrodes.
  • Signals: Single-unit activity (SUA) and Local Field Potentials (LFPs).
• Area Classification: PMd and M1 were functionally distinguished based on dominant beta-band rhythms (low beta <20Hz dominant in M1; high beta >20Hz dominant in PMd).
• Laminar Classification: Layers were divided into Superficial (SUP, ~L2/3) and Deep (DEEP, ~L5/6) based on cortical depth (approx. 1.8mm threshold).
• Metrics for Intrinsic Timescales:
  1. SUA Timescales ($\tau$): Estimated from the autocorrelation of spike trains using an exponential decay fit (Fontanier et al., 2022 method).
  2. LFP Timescales ($\tau$): Estimated from the autocorrelation of band-pass filtered (1–12 Hz) bipolar LFP signals to avoid beta oscillation contamination.
  3. LFP Aperiodic Exponents: Estimated using the FOOOF algorithm (55–95 Hz range) to measure the spectral slope ($1/f$ decay), which inversely correlates with temporal integration.
• Functional Analysis:
  • Decoding: Linear classifiers were used to decode movement direction from population activity.
  • Cross-Temporal Decoding (CTD): Analyzed the stability of representations over time (diagonal width in CTD matrices) to assess how intrinsic timescales relate to the temporal evolution of motor encoding.

3. Key Contributions

• First Characterization of Laminar Spike Timescales: This is the first study to directly estimate intrinsic spiking timescales across cortical layers in any cortical area.
• Multimodal Convergence: Demonstrates that different signal modalities (spikes and LFPs) converge on the same inter-areal hierarchy but diverge in their laminar organization.
• Functional Link: Establishes a direct link between intrinsic timescales and the temporal stability of motor planning and execution representations.

4. Key Results

A. Inter-Area Hierarchy (PMd vs. M1)

• SUA: M1 exhibited significantly longer intrinsic timescales ($\tau$) than PMd.
• LFP Aperiodic Exponents: M1 exhibited significantly smaller spectral exponents (flatter slopes) than PMd. Smaller exponents indicate stronger low-frequency dominance and longer temporal integration.
• LFP Autocorrelation: No significant difference in LFP $\tau$ between areas was found.
• Conclusion: Convergent evidence places M1 higher in the motor hierarchy than PMd, consistent with recent anatomical tracing studies suggesting an ascending influence from PMd to M1.

B. Laminar Organization (Superficial vs. Deep)

• SUA (Spikes): No significant laminar differences were found in intrinsic timescales within either PMd or M1. Spiking activity did not show a systematic gradient across depth.
• LFP (Population):
  • Aperiodic Exponents: Significantly smaller (flatter) in deep layers compared to superficial layers in both PMd and M1.
  • Autocorrelation Timescales: Significantly longer in deep layers compared to superficial layers in M1 (and a trend in PMd).
• Dissociation: There is a clear dissociation: while spikes show no laminar gradient, LFPs (reflecting synaptic/dendritic integration) show consistent depth-dependent differences, with deep layers exhibiting longer integration timescales.

C. Functional Relevance of Timescales

• Motor Planning (PMd): Neurons with longer $\tau$ maintained more stable representations of the memorized movement direction during the delay period. This was evidenced by a higher proportion of significant pixels in cross-temporal decoding matrices compared to short-$\tau$ neurons.
• Motor Execution (PMd & M1): During movement execution, longer $\tau$ neurons exhibited slower temporal dynamics. This was quantified by wider diagonals in cross-temporal decoding maps, indicating that the neural code for movement direction persisted longer over time in long-$\tau$ neurons compared to the rapid, transient updates of short-$\tau$ neurons.

5. Significance and Discussion

• Resolving Motor Hierarchy: The study provides functional evidence supporting a hierarchical model where M1 is "higher" than PMd, resolving anatomical ambiguities through temporal dynamics.
• Physiological Origins of Divergence: The lack of laminar gradient in spikes versus the presence in LFPs suggests distinct physiological origins:
  • Spikes reflect the output of individual neurons, which may be homogenized across layers in terms of integration time.
  • LFPs reflect synaptic inputs and dendritic processing. Deep layers (L5/6) contain pyramidal neurons with extensive apical dendrites spanning the column, potentially integrating inputs from multiple layers and resulting in longer population-level timescales.
• Computational Role: The findings suggest a division of labor within motor populations:
  • Long-$\tau$ neurons: Support stable, persistent coding (maintenance of motor plans) and sustained execution dynamics.
  • Short-$\tau$ neurons: Support rapid switching and transient updates, allowing for flexible, moment-to-moment adjustments.
• Methodological Impact: The study highlights the necessity of multimodal approaches. Relying solely on spikes might miss laminar temporal organization that is evident in population-level signals (LFPs).

In conclusion, the paper establishes that intrinsic timescales are a robust marker of motor cortical hierarchy, with M1 positioned above PMd. Furthermore, it reveals that while inter-areal hierarchy is consistent across modalities, laminar temporal organization is modality-specific, reflecting the distinct computational roles of neuronal output (spikes) versus synaptic integration (LFPs).