Striatal ensemble activity in an innate naturalistic behavior

Reviewer #1 (Public Review):

In "Striatal ensemble activity in an innate behavior", Minkowicz et al. strive to characterize how the striatum, the primary input nucleus of the basal ganglia, represents grooming. Here, grooming is used as a paradigmatic habitual behavior. The pose dynamics of grooming are stereotyped: mice perform it spontaneously and prior work has shown that it is both represented and controlled by the striatum.

The manuscript presents a valuable contribution to the field by shedding light on how ensembles of neurons encode this innate behavior. Additionally, the use of supervised machine learning allowed the authors to collect and precisely align a large number of grooming repetitions, which enabled most of their downstream analysis.

I found the paper to be well-written and the conclusions are mostly well-supported. However, some of the data analysis was a bit opaque, and some more detail and reanalysis could substantially strengthen the authors' claims.

1. The authors identified grooming bouts using empirically defined thresholds and manual tweaking. Next, the boundaries of grooming were used for trial alignment and linear time warping. This is a completely sensible approach; however, in using only the boundaries of grooming episodes, the dynamics of grooming bouts are ignored. I am particularly concerned that pose dynamics of grooming bouts are most stereotyped at the boundaries (e.g. they always begin and end with specific paw movements). To play devil's advocate, if the striatum encodes pose dynamics and not boundaries and pose dynamics are noisy between the beginning and end of these bouts (either due to the dynamics of the behavior or how it was identified), then a "boundary-like" representation may emerge in the average. I strongly recommend re-running a subset of the analysis after accounting for variability in grooming dynamics. A simple thing to try would be to further cluster grooming bouts using 3D keypoint trajectories. Another would be to warp grooming bouts in a manner that accounts for keypoint trajectories (e.g. DTW or other recent time-warping variants).

2. The authors should consider if the correlation to grooming is due to (at least in part) a correlation with another aspect of movement, e.g. overall velocity, acceleration, height, or angular velocity. This should be straightforward to analyze with the current dataset. To start, I would simply take the velocity and acceleration of the mouse's centroid (head and body could be considered separately). Next, look at the correlation with DLS spiking. If a clear relationship emerges, then check to see how velocity (or another variable) maps onto grooming. It may be that DLS neurons appear to encode the boundaries of grooming when they (at least partially) encode other variables.

3. The ensemble analysis is potentially critical to our understanding of SPNs. Figure 4A suggests that ensembles encode grooming with a probabilistic code - ensembles appear to be engaged for a small number of grooming bouts in the session. First, a basic question is what is the probability a given ensemble is activated during grooming? Second, the more complex question is whether there is an explanation for why one ensemble is engaged for some trials and not others? Related to point 2, I wonder if another aspect of behavior - e.g. vigor, duration, or speed - determines this. I suggest some analysis to at least rule out some simple explanations.

Reviewer #2 (Public Review):

The manuscript by Minkowicz et al., investigates the presence of neuronal ensembles in the striatum that may encode grooming (as a model of a naturalistic behavior). They implemented a semi-automated detection of grooming, and by recording populations of striatal cells they show that individual neurons in the striatum contain activity modulations around the start, end, or during grooming. Then using this activity they identify ensembles of cells in individual sessions/animals at the start, end or during grooming.

The behavioral tracking and recordings are remarkable, the manuscript is clearly written and the finding mostly sound with the proposed conclusions, providing original findings in the field. Nonetheless some points are raised that need further clarification

1. When claiming that the findings show encoding of transitions into or out of grooming (and duration of grooming) one could expect to see specific regressions between the neuronal activity (of individual cells or ensembles) and the parameters mentioned besides the analysis shown in figure 3 and 5.
2. Was the detection of ensembles presented in figure 4 sensible to use less than 5 seconds before/after grooming. I am thinking that 5 seconds are times that could contain behaviors that may have their own ensembles. Why 5 seconds?
3. According to Figure 2-figure supplement 1. The recordings were performed covering the lateral and in some cases the central part of the striatum. Shall it be specified along the text where the specific recordings come from?