PREreview of Neural representation of action symbols in primate frontal cortex

This manuscript addresses a fundamental and long-standing question in cognitive neuroscience: how symbolic representations are implemented at the neuronal level. The authors trained two monkeys to do a drawing-like task characterized by symbol invariance, discreteness, and recombinational use. They then recorded single-cell electrophysiological activity from eight brain areas and found that, strikingly, only PMv exhibited all three properties when encoding motor primitives. Overall, the manuscript provides novel and compelling evidence for the unique role of PMv in encoding action symbols, and the behavior paradigm offers a promising tool for future investigations in how symbols are organized and linked in the brain. Nevertheless, some of the conclusions are not fully supported by the current analyses.

Major points:

·      The first major point concerns how PCA was implemented in Fig.4k and Ext.Fig.10c. The conclusion from these analyses (along with Fig.4i&j) is that PMv encodes motor primitive consistently across changes in location or size, but not vice versa. However, the manuscript provides very little information on how PCA was implemented there, and from what is described, it appears that PCA was performed in a way that maximized the neural activity variance related to motor primitive, which may have inadvertently down-weighted the encoding of location and size. As an alternative possibility, PMv could encode primitive, location, and size alike, but only the encoding of motor primitive got through the PCA pre-processing. To rule out the possibility and verify the conclusion, the authors should either provide additional methodological details on their PCA procedure or test their result with a PCA approach that treats primitive, location, and size in a balanced manner.

·      Secondly, the manuscript has insufficient evidence for the claim of “recombination”. A crucial distinction between “recombining/constructing actions into sequences” and “performing a series of actions”: the former requires some relational structure between action elements, whereas the latter can arise simply from the passage of time. In principle, the monkeys could accomplish the character task without true recombination by repeatedly reassessing the remaining shape on the screen and applying a single motor primitive at a time until the whole shape is filled out. To support the claim that monkeys recombine primitives into sequences, the authors should present additional evidence showing how one primitive is linked to another with a certain structure. For example, one observation from the example plots in Fig.2 and Fig.6 is that one monkey consistently used two strokes to fill in a circle within a complex character. More detailed analysis of such patterns would strengthen the argument for recombination as sequences.

Minor points:

·      One minor point is about the rationale behind different selections on motor primitives in Fig.4, 6, and Ext.Fig.10. Is that from a limitation of what images got tested in each individual task setting? Or was any motor primitive excluded due to sample size, consistency, etc.? The authors should provide the above information along with the analyses in the methods section.

·      The last minor point is about the example figures of neural activity in Fig.4,5&6. There are plots labeled as “from both subjects”, and there are two plots in Fig.3 and Fig.6 labeled as “from subject 2”, but the rest are unclear. The author should clarify if the rest are from subject 2 or not. And it would be ideal if authors could show the plots for the other subject (e.g. in some extended figures).

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.