Natural behavior relaxes zoning divisions in the brain

Abstract

Technological advances allow neurophysiologists to explore the brain during natural behaviors, revealing new functional principles and challenging old ones. In this issue of Neuron, Li¹ and colleagues show that the traditional parcellation of the marmoset frontal cortex does not apply to naturalistic conversations.

What would our social life look like if every frown or smile had to be repeated tens of times for the receiver to understand it? Foundational neuroscience experiments, spanning decades of exploration, relied on paradigms that needed tens or hundreds of repetitions of the same behavior or stimulus to extract meaningful information from single neurons. A major leap forward came with technological innovations that allowed neurophysiologists to monitor the activity of populations of neurons^2,3 and the subsequent demand for new analytical methods. Together, these developments revealed brain states emerging from the joint activity of neurons assembled in functional networks. Neural networks, often distributed across multiple brain areas, replaced single neurons as the functional unit of brain activity.⁴ The collateral advantage of capturing the activity of entire networks of neurons was the replacement of reductionist and trial-based experimental designs with natural behaviors that are spontaneous, less predictable, and that carry the risk of never repetitively occurring. These challenges notwithstanding, the neurophysiology of natural behaviors offers a significant advantage for discovering neural dynamics and computational principles that evolved to serve the behaviors under scrutiny.⁵

In this issue, Li and colleagues¹ embraced this challenge. They abandoned the strictly scripted, trial-based experimental design and exploited instead the quasi-predictable sequence of natural vocal exchanges between marmosets. They simulated natural conversations (antiphonal calling) by delivering through speakers pre-recorded vocal calls that engage the listener monkey into responding to these calls with delays seen during natural behavior. They also recorded the spontaneous solicitation calls of the subject. Instead of timing the presentations of calls independently of the receiver’s behavioral state, the pre-recorded calls were broadcast either as solicitation calls or as responses to calls of the subject. The resulting sequence of calls was temporally variable, even overlapping at times. They analyzed the activity of neurons in multiple areas of the frontal cortex both at the single-cell level, using traditional methods (e.g., peri-stimulus time histograms), and at the population level, using a method that considered the history of calls and was flexible enough to accommodate the temporal variability inherent to unconstrained behaviors. Accordingly, they applied a generalized linear model (GLM) that included behavioral state, calls heard, calls produced, and interactions between them as regressors. Previous experiments showed how critical it is to include behavioral state in these analyses, as the activity of neurons in the frontal cortex of marmosets predicted whether they would engage in a conversation with a conspecific seconds before they heard the invitation to respond.⁶

The traditional, cell-by-cell approach confirmed that the majority of neurons in the prefrontal cortex and the premotor cortex were responsive to facets of the ongoing communication. However, the GLM captured additional facets of dependence between brain activity and the receiving-emitting cycle of vocal communication. Through dimensionality reduction, the authors identified functional clusters related to the perceptual and motor elements of the communication but also to internal states and social context.

Remarkably, the functional clusters did not map onto the well-established functional parcellation of the primate frontal cortex.⁷ The premotor and prefrontal cortical areas of a freely moving, unrestrained receiver and emitter of calls responded differently during the vocal exchange than what would be expected based on a trial-based design with behavioral restrictions. Moreover, the sliding-window GLMs suggested that both areas contributed to a continuously unfolding progression of neural states distributed over multiple subdivisions of the frontal cortex. Although not monitored by the authors, these states could be informed by body posture, head and eye movements, social decision-making, autonomic and energetic regulation, and perhaps even less obvious elements of the unfolding mental processes.

The zoning violation perpetrated by these functional clusters is not as surprising as it may appear at first glance, given that removing behavioral constraints may add new variables processed by neurons in both areas. Areas of the prefrontal cortex are highly inter-connected with the premotor cortex, and both areas are required to contribute to the elaboration of spontaneous social behaviors. The current functional parcellation may have been established by tasks and behaviors that by design, or by necessity, elicited lower dimensional representations. By contrast, high-dimensional representations evidenced by multidimensional selectivity at the individual cell level⁸ may reveal relative rather than absolute cortical specializations. Indeed, the fluid transitions between absolute and relative specialization might be a better criterion of frontal homology across mammals than localization of function or hierarchical organization.⁹ Highly social animals, like marmosets, need behavioral flexibility arising from high-dimensional representations to handle parallel streams of relatively unscripted behaviors. The quiet revolution toward natural behavior may dissolve the absolute functional parcellation and the strict division of labor among cortical areas.