1. Author Response:

    Reviewer #1:

    Guo et al. describes interesting experiments recording from various sites along a cortico-cerebellar loop involved in limb control. Using neuropixels recordings in motor cortex, pontine nuclei, cerebellar cortex and nuclei, the authors amass a large physiological dataset during a cued reach-to-grasp task in mice. In addition to these data, the authors 'ping' the system with optogenetic activation of pontocerebellar neurons, asking how activity introduced at this node of the loop propagates through the cerebellum to cortex and influences reaching. From these experiments they conclude the following: the cerebellum transforms activity originating in the pontine nuclei, this activity is not sufficient to initiate reaches, and supports the long standing view that the cerebellum 'fine tunes' movement, since reaches are dysmetric in response to pontine stimulation. Overall these data are novel, of high quality, and will be of interest to a variety of neuroscientists. As detailed below however, I think these data could provide much more insight than they currently do. Thus below I provide some suggestions on improving the manuscript.

    1. Since the loop is the focus of this study, it would be nice if the authors better characterized latencies of responsivity to pontine stimulation through the loop, to address how cortically derived information routed to the cerebellum may loop back to influence cortical function. In the data provided, we know that pontine stimulation modulates Purkinje and deep nuclear firing (but latency to responses are not transparently provided in the main text, if anywhere), while motor cortical responses peak at 120 ms (after stimulus onset?, unclear), and that this responsivity is preferentially observed in neurons engaged early in the reaching movement. Is the idea, then, that cortical activity early in the reach is further modulated by cerebellar processing to (Re) influence that same cortical population? Does this interpretation align with the duration of reaches, the duration of early responsive activity during reach, and the latency of responsivity; or is the idea that independent information from other modalities entering the pontine nuclei modulates early cells? Latency to respond at the different nodes, might aid in thinking through what these data mean for the function of the loop.

    We thank the reviewer for this important suggestion, and we have now added measurements of the latency from the onset of sinusoidal PN stimulation to neural responses in Purkinje cells, DCN neurons, and motor cortex (Supplemental Fig. 7), and observe a progressive recruitment of laser-evoked spiking along this pathway. There is a tradeoff between temporal resolution (which increases with decreasing bin width) and statistical power (which decreases with decreasing bin width), and we have opted to use 10 ms bins in a sliding window, which provides a reasonable compromise between these criteria. Although we potentially detect fewer tagged neurons at shorter latencies than we would with larger bins, this approach enables us to detect the timing of the earliest responses (defined as the earliest time point at which 5% of the neurons eventually recruited are responsive). Note that the sinusoidal stimulation used in these experiments is not ideal for latency measurements, as it takes 6.25 ms for the laser to reach peak power. We have also added a similar analysis for the response latency of PN neurons to pulse train stimulation of motor cortex (Supplemental Fig. 1). Based on these analysis, our estimate of the delay for signals to propagate across the entire loop is 26 ms: PN to motor cortex (21 ms) + motor cortex to PN (5 ms). Given that the movement duration (lift-to-grab) is approximately 110 ms on average, this would allow ~4 full feedback cycles throughout the reach. Thus, these delays are consistent with the possibility that cortical activity during planning or early in the reach is further modulated by cerebellar processing to influence that same cortical population later in the reach. Regarding the earliest motor cortical responses that we observe in PN-tagged units, it's possible that they may result from ponto-cerebellar input driven by other cortical regions. Alternatively, the responses of motor cortical neurons early in the movement may be driven more directly by other cortical areas or the basal ganglia, but these early-responding neurons may also receive strong ponto-cerebellar input due to plasticity during development or learning.

    1. Many of the figures need work to aid interpretation. Axis labels are often missing (eg 2F); color keys are often unlabeled (2F); color gradients often used but significance thresholds are hard to evaluate (using same colors for z scores and control / laser is confusing 6, 8); and within-figure keys would be useful (5D-h). These issues occur throughout the manuscript.

    We have added the axis and color labels in Fig. 2F, and have added additional annotation throughout the main and supplemental figures. For firing rate z-score heatmaps, we have kept the gray color scale for control and laser to facilitate direct comparison between the panels, but have added orange and blue boxes around the heatmaps in Fig. 6, 7, S8, and S9 to emphasize that they reflect different experimental conditions.

    1. Relatedly, but also conceptually, Figure 3B has particular issues, such as identifying where the neuropixel multiunit activity is coming from. I assume that in the gray boxes illustrating the spatio-temporal profile of spiking band activity that the lower part of the box is the ventral direction, upper, dorsal. This is not spelled out. From the two examples it would seem that the spiking band is in different places in the cerebellum, undermining, I think, the objective of the figure. It would be sensible to revisit this entire figure to identify the key takeaways and design figures around those ideas. As it stands, these examples appear anecdotal. Consider moving this to a supplement. Powerband density strength is missing an axis. More importantly, it would be nice to corroborate the interpretation of the MUA with the single unit recordings, since the idea is that many neurons are entraining to the PN activity. Yet, the examples don't seem particularly entrained. Is the activity being picked up on just axonal firing of the PN axons? Fourier analysis of spiking of isolated neurons in cerebellum should be used to corroborate the idea that cerebellar neurons are entraining, rather than the neuropixel picking up entrained PN axons.

    To examine spike entrainment to the 40 Hz PN stimulation for Purkinje cells and DCN neurons, we computed the phase of sinusoidal stimulation coinciding with each individual spike. If a neuron is entrained to the stimulation, the phase distribution for its spikes will differ from the uniform distribution on the circle; this can be assessed for each cell using a Rayleigh test. Furthermore, we can calculate the strength of entrainment and preferred phase by calculating the magnitude and angle of the mean resultant for each cell. If a neuron’s spikes are completely unrelated to the stimulation phase, the mean resultant length will tend to 0 as the number of spikes observed goes to infinity. If, on the other hand, a neuron is completely entrained (with every spike occurring at exactly the same phase), the mean resultant length will be 1. This approach is illustrated schematically in Supplemental Fig. 6A.

    This new analysis revealed two key features of the data we had not previously appreciated. First, it revealed PN-stimulation-induced changes in neural activity that were not apparent from the mean firing rate profiles: most Purkinje cells and DCN neurons were significantly entrained to the 40 Hz stimulation. Second, the entrainment strength was higher in the DCN than Purkinje cells (Supplemental Fig. 6B-D), suggesting the corticonuclear pathway amplifies the rhythmic input. This result is strikingly similar to published observations obtained from slice electrophysiology and anesthetized mice (Person & Raman, 2012), which we now discuss in the text. It is also possible that direct excitation from PN collaterals contributes to the DCN entrainment.

    We agree that the original analysis of multiunit activity is difficult to interpret, for two reasons: (1) the signal likely reflects the combined contribution of multiple cell types, including pontine mossy fiber terminals, and (2) the depth profile will differ for different electrode penetrations, due to the geometry of the cerebellar cortex. Furthermore, this analysis is largely redundant, since we have recorded from individual Purkinje cells and added new analyses demonstrating their entrainment to the 40 Hz stimulation (Supplemental Fig. 6). We have now moved this figure to the supplement and added labels to all axes (Supplemental Fig. 3).

    1. The use of the GLM is puzzling. In addressing the question of how cerebellum and motor cortex interact (from the Abstract, "how and why" do these regions interact) it is unclear why these regions are treated separately. I would have expected some kind of joint GLM where DCN activity is used to predict M1 variance (5 co-recordings are reported but nothing to analyze?); or where DCN + M1 activity is used to decode kinematics to see if it is better than one or the other alone. As it stands, we learn that there is more kinematic information in the motor cortex than in DCN. This is not necessarily surprising given previous literature on cerebellar contributions to reaching movements. In principle the idea that 'PN stimulation might perturb reaching kinematics through descending projections to the spinal cord, or by altering activity in motor cortex' is treated as mutually exclusive outcomes, though it is highly unlike to be so.' Analyzing M1+DCN together could address whether DCN activity adds nothing to decoding kinematics that isn't there in M1 or adds something that M1 does not have access to. The main point here is that the physiological datasets could be better leveraged with these fits to derive insight into the interactions of the loop. R2 should be provided in the GLMs (Fig 8) to assess statistically how well they perform relative to one another, not just correlations between the two.

    We have added two additional analyses to address these questions. First, in addition to motor cortex-based and DCN-based decoders for all sessions (Fig.8 and Supp. Fig.12A-D, G-H; all the R2 values are reported in Supp. Fig. 12C-D, G-H) we now also train a decoder using both motor cortical and DCN multiunit activity in sessions with simultaneous recordings (Supp. Fig.12E-F, I-J). When we train only on control trials, the decoder performs about equally well with or without the DCN multi-units for control trials (Supplemental Fig. 12E), but performs slightly worse on laser trials in comparison to using only cortical data (Supplemental Fig. 12F). When we train on both control and laser trials, adding DCN multi-units slightly degrades decoding performance on both control and laser trials in 3 out of 5 sessions (Supplemental Fig. 12I-J). Based on this comparison, it does not appear that DCN contributes kinematic information that is not already present in cortex. However, there are several cautionary notes to consider in interpreting these results. (1) This dataset consist of only 5 sessions, in all of which the recording yield in DCN was not as high as in cortex, so it is possible that dimensions of activity unique to DCN may not have been sampled enough in these experiments. (2) Our task involves only a single reaching target (in comparison to, e.g., center-out reaching tasks with eight targets which are possible in primates) so we cannot assess whether DCN contains directional-specific kinematic information not present in cortex. Thus, in light of these factors, it is difficult to draw strong conclusions from our experiments about differences in kinematic information between motor cortex or DCN. A more rigorous comparison requires carefully controlled experiments with many reaching targets, as in Fortier, Smith, & Kalaska (1993).

    Second, we have added an additional analysis to determine how predictive cortical activity is of DCN activity at the single-trial level, and vice versa. We considered several possible statistical approaches to this issue. Computing pairwise correlations of neurons in the cortex and DCN would be one possible method, but the outcome of this analysis would be difficult to interpret, as the sign and timing of firing rate peaks will vary across neurons. Another approach would be to regress principal component scores in one region - or their derivatives, as in Sauerbrei et al., 2020 - on the scores in another region. However, because cortex and DCN are bidirectionally connected, the choice of which region’s scores should be considered as the dependent variables is ambiguous, and this approach will merely “align” activity in one region (as a projection onto regression coefficients) with activity in the other. Ideally, we would like to find simultaneous linear transformations of both cortical and DCN activity that would maximally “align” them with one another, and to compute the correlations of the aligned neural trajectories. This is precisely what canonical correlation analysis (CCA) does, and CCA has been used increasingly in recent years to align population activity from different brain regions or samples - e.g., Lara et al., Nat. Comm. (2018), Perich et al., Neuron (2020), and Gallego et al., Nat. Neuro (2020). We took this approach with our simultaneous recordings of multiunit activity in the motor cortex and DCN, and found that:

    (a) In each of the 5 sessions, CCA found two pairs of canonical variates that were strongly correlated (Supplemental Fig. 11A, first two columns; Supplemental Fig. 11B, correlations in the range 0.58-0.88 for the first two canonical variates), and two pairs of canonical variates weakly correlated (Supplemental Fig. 11B, correlations <0.27 for the last two canonical variates)

    (b) The first two canonical variates accounted for half or more of the variance in each region (49%-64% in cortex, 51%-70% in DCN; Supplemental Fig. 11C, left column)

    (c) Between a quarter and a half of the variance in each region was accounted for by canonical variates in the other region (25%-50% of variance in DCN explained by cortex, 26%-47% in cortex explained by DCN; Supplemental Fig. 11C, right column)

    From these results we conclude that, within the constraints of our behavioral task, some but not all of the dominant dimensions of cortical and cerebellar activity are strongly correlated. We also performed additional CCA analyses using only laser trials or only control trials, to assess whether PN perturbation strongly affected the similarity in population activity between the two regions, but found limited differences between the results of the two analyses (Supplemental Fig. 11D).

    Reviewer #2:

    Guo et al examine the cortico-cerebellar loop during skilled forelimb movements in mice. The authors use optogenetic stimulation of the pontine nuclei (PN) and recordings in PN, cerebellar cortex, cerebellar nuclei (DCN), and motor cortex to show that PN output is transformed into a variety of activity patterns at different stages of the cortico-cerebellar loop. Stimulation only slightly alters movement-related activity in these structures and degrades movement accuracy. The authors conclude that the cortico-cerebellar loop fine tunes dexterous movement. The study is technically impressive, employing recordings in 4 brain regions, and recordings during optogenetic manipulations and behavior. The experiments are well done and the analyses are appropriate. The comparison across brain regions is comprehensive. The results that PN perturbation alters skilled movement and the perturbed activity could predict perturbed movement are important. The study adds to a long line of work supporting the view that the cortico-cerebellar pathway is required for fine motor control. I have a few comments on the interpretation and analysis which I believe could be addressed with changes to the text and additional analysis.

    1. The authors conclude that the cortico-cerebellar loop "does not drive movement" but "fine tunes" the movement. While I generally agree with this interpretation, I wonder if the authors could flush out the concepts of "driving movement execution" vs. "fine-tuning movement" more clearly. Do authors consider them separate processes? How can they be disentangled? I also feel the data on its own has some limitations that should be considered or discussed. First, the data shows that PN stimulation degrades movement accuracy. However, this does not yet reveal the function of the cerebellar loop in fine motor control. Certain places in the text makes stronger assertions (for example, "cortico-cerebellar loop fine-tunes movement parameters") that I feel the data does not support. It is not clear from the data how the loop tunes movement parameters. Second, Fig. 5F shows that stimulating PN blocked movement initiation in some sessions (this is also mentioned in the Methods). Could the authors consider the possibility that stimulating PN at a higher intensity might block movement? This is related to the distinction between "driving" vs. "fine-tuning" movement. At the very least, the authors should discuss these limitations and possibilities.

    In our view, the claim that a brain area drives reaching means that it is necessary for generating the large changes in muscle activity that set the limb in motion towards the target. The claim that a brain area fine-tunes reaching means that it is necessary for generating smaller changes in muscle activity that subtly adjust the limb trajectory and enable precise and accurate behavior. Previous work has demonstrated that motor cortex drives reaching: if it is transiently silenced, the initiation of reaching is robustly blocked (see Guo et al. 2015, Sauerbrei et al. 2020, and Galinanes et al. 2018). In the present manuscript, we show that perturbation of the PN has a very different effect: mice are usually able to initiate reaching, but they are less skillful (the success rate drops), slower (movement duration increases), and less precise (endpoint standard deviation increases). Our interpretation of these results is that while the total output of cortex drives movement (likely through corticospinal and cortico-reticulospinal routes), the cortico-cerebellar loop makes more subtle adjustments to the ongoing movement; that is, it fine- tunes. We have updated the text (in particular, the Abstract, Introduction par. 1, and Discussion par. 1-2) to clarify the distinction between driving and fine-tuning.

    We agree that several interpretive statements in the previous version (especially concluding sentences at the end of some Results paragraphs) were not clearly connected with the data, and we have removed or modified these statements. We now lay out our interpretation of the data as evidence for a cortico-cerebellar contribution to fine-tuning, rather than driving, in the first two paragraphs of the Discussion, but emphasis that this is an interpretation, rather than a direct description of the data. We have also changed the title to more directly state our experimental observations.

    We now mention the possibility that stronger stimulation or inactivation of PN neurons might have robustly blocked movement, and also mention several experimental variables which might have contributed to animal-to-animal variability in behavioral effects: “It is possible that the variability of behavioral effects ...” (Discussion).

    1. Related to point 1, in Fig. 5F, for stimulation trials in which mice failed to initiate movement, did mice fail to move altogether, or did they move in an abnormal fashion?

    We have added a new video documenting the behavior of the animal with the largest blocking effect from PN stimulation (supplemental video 2). This animal does not struggle through a partial reach, but fails to initiate movement. Small movements of the arm occurred (this also occurred in control trials), but these were not tightly synchronized with the onset of the laser across trials.

    1. In the abstract, the authors state that PN stimulation is "reduced to transient excitation in motor cortex". Also in the results (page 5) and discussion (page 8), "pontine stimulation only led to increases in cortical firing rates". These statements are based on the comparison between Fig 3D, 3F, and 4B. But I think the current presentation is somewhat misleading. First, Fig 3D, 3F, and 4B use different neuron selections that make direct comparison difficult. Fig 3 shows all neuron from Purkinje cell and DCN recordings. Fig 4B shows only PN-tagged motor cortex neurons. Furthermore, based on the methods description, it appears that PN-tagged neurons were defined using one-sided sign-rank test. Since the test is one tailed, does that mean neurons shown in Fig 4B are, by definition, neurons significantly excited by photostimulation? Looking at Fig 4B and 4C closely, there appear to be neurons suppressed by PN stimulation. Could the authors organize the rows in Fig 4 in the same way as Fig 3, where neurons that show suppression are grouped together?

    We now display the PN stimulation-aligned firing rates in the same format for Purkinje cells (Fig. 3B), DCN neurons (Fig. 3D), and motor cortical cells (Fig. 4A, lower), with all neurons in a single panel, sorted by response magnitude, for each area. The dominant response pattern in the cortical population is a transient firing rate increase, and this is more readily apparent with the new panel in Fig. 4A (lower). We also use a two-tailed test (which has slightly less statistical power, but allows us to test for both firing rate increases and decreases) for the identification of PN-tagged cortical neurons, and display neurons with stimulation-locked increases (n = 94) and decreases (n = 13) separately (Fig. 4B). In Fig. 4B-C, we still sort the neurons by their reach- related responses, as this reveals a difference in lift-aligned patterns between tagged and non- tagged neurons, which would be masked if we ordered according to stimulation-aligned responses. In Fig. 4D-E, we pool neurons with PN-stimulation-aligned increases and decreases into a “PN-tagged” group, as the small number of stimulation-aligned decreasing neurons (n = 13) does not allow adequate statistical power for a 3x3 contingency table test or for within-group averaging of lift-aligned firing rates.

    1. Fig 7 shows that PN stimulation has only subtle effects on movement-related activity in motor cortex. However, only a small portion (1/8) of the motor cortex neurons show modulation to PN stimulation. Fig 7 shows all neurons. Would the results look similar for PN-tagged neurons?

    We have added a new analysis to address this question, shown in Supplemental Fig. 10. The laser - control difference in lift-aligned activity are indeed larger for PN-tagged neurons; however, the largest peak in this difference occurs before lift, when the laser has been turned on, but the animal hasn’t started to move (Supplemental Fig. 10C).

    1. Page 3 "Our observation that the activity of some motor cortex-recipient PN neurons is aligned both to the cue and movement suggests that these neurons might integrate signals of multiple modalities." Presumably, motor cortex neurons also have cue and movement-related activity and PN simply inherits this activity from the motor cortex.

    As described in our response to the first reviewer’s seventh comment, we cannot conclude that the cue-related responses in the PN are inherited entirely from motor cortex. Briefly, (1) it has been difficult for us to reliably disassociate cue and movement responses for individual motor cortical cells (for instance, the GLM approach we took with PN neurons resulted in very poor model fits when applied to cortical cells), though our previous work has suggested that at the population level, the dominant signal in motor cortex is aligned to movement onset. To reliably disentangle cue and movement responses in cortex, we would need to train mice to wait for a relatively long and variable delay period before reaching. (2) The PN receive convergent input from many cortical areas, and there is likely a convergence of multiple inputs onto the motor- cortex-tagged PN units (c.f. the convergence of inputs from visual and somatosensory cortex onto individual PN neurons in rats reported in Potter, Ruegg, & Wiesendanger,1978). Hence it is possible (if not likely) that the multi-modal activity we observe in PN neurons results from the integration of inputs from different cortical areas, rather than being entirely inherited from motor cortex.

    1. Do Purkinje cells follow the 40 Hz PN stimulation like in the multi-unit recordings. The PSTHs in Fig 3 are too smoothed out to see this.

    As described in the response to reviewer 1.3 above, we have added a new analysis to the manuscript to address this question (Supplemental Fig. 6). Most Purkinje cells and DCN neurons are entrained to the 40 Hz stimulation, and the entrainment is much stronger in the DCN, consistent with previous work (Person & Raman, 2012).

    1. For the correlation analysis in Fig 6C top and 7C top, is the correlation computed from z-scored firing rates rather than on raw firing rates? This is not clear from the text. If computed on raw firing rates, one would expect the correlation to be above 0 even before photostimulation, since different neurons exhibit different baseline firing rates that presumably will be the same across control and stim trials.

    The correlations were indeed computed on z-scores, rather than raw firing rates, for this reason. We have clarified this in the Methods section. This analysis was designed to capture correlations in movement-related modulation between control and laser trials, and we z-scored the firing rates to avoid the confound that would have been introduced by baseline differences.

    Reviewer #3:

    It is generally thought that the cerebellum is primarily involved in the short-timescale control of movements, while motor cortex is involved in motor planning. The present paper follows classic studies in primates and a recent study in mouse that investigated the role of cortico-cerebellar loops in motor control. To date, studies in both species applied perturbations to the cerebellum to then study changes in cortical activity. For example, it has been long known that cooling deep cerebellar nucleus produces changes in the responses of motor cortex neurons in primate (e.g., Meyer-Lohmann et al., 1975). Further, Gao and colleagues' recent paper (Nature 2018) used optogenetics to perturb responses in the deep cerebellar nucleus before licking movements. The authors of this 2018 nature paper conclude that persistent neural dynamics are maintained during voluntary movements by connectivity in within this cortico-cerebellar loop.

    The experiments are well performed, and the results are logically organized and presented. However, a main concern is that the authors have not well justified that these experiments prove a conceptual advance. The conclusions appear to be largely consistent with those of prior work, both regarding changes in the responses of motor cortex neurons, and resultant (subtle) changes in behavior (i.e., altered arm kinematics). The impact of the paper would be improved if the authors adapted a more precise style of reporting the novelty of their results throughout.

    Major concerns:

    1. The experiments are well performed, and the results are logically organized and presented. However, a main concern is that the authors have not well justified that these experiments prove a conceptual advance. As noted above, prior studies have probed the role of cortico-cerebellar loops by applying perturbations to cerebellar activity (cerebellar cortex and/or deep cerebellar nuclei) and quantifying changes in cortical activity prior to and during movement. The main novelty of the present study is that the authors perturbed the loop at a different locus, namely in the pontine nuclei (PN) which send projections to the cerebellum rather than directly to the cerebellum. The rationale for why this specific perturbation provides a conceptual advance to the field was not adequately motivated.

    The authors do clearly review prior literature showing that perturbation of cortico-cerebellar projections impacts the rest of the loop and behavior, they also well explain the application of their exciting new tool to specifically target PN neurons with their optogenetic stimulation. Yet, the authors do not motivate why it is important to specifically perturb the pontine nuclei (PN) to gain new insights into the role of "cortico-cerebellar loops" nor do they provide any reason to expect a difference in changes in loop dynamics for perturbations applied versus to the DCN. Indeed, the conclusions appear to be largely consistent with those of prior work, both regarding changes in the responses of motor cortex neurons, and resultant (subtle) changes in behavior (i.e., altered arm kinematics). Generally, these results are similar to those previously reported in primate DCN cooling experiments characterizing changes in hand movement in in a voluntary tracking task (e.g., Brooks et al., 1973; Conrad and Brooks 1974).

    We agree that the rationale and conceptual advance require clarification. Previous work has established that silencing motor cortex blocks reaching (Guo et al. 2015, Sauerbrei et al. 2020, Galinanes et al. 2018), but the perturbations used in these studies were not selective to specific output channels (e.g., corticospinal, corticoreticulospinal, or corticocerebellar), and simultaneously influenced many projection targets of motor cortex. Other work from the Brooks, Prut, Person, and Svoboda groups has shown that altering cerebellar output impairs movement planning or execution, but their methodology did not test the effects of disrupting specific cerebellar inputs (e.g., from cortex). Thus, we would argue that previous studies have not provided direct evidence of the behavioral and neural effects of disrupting cortico-cerebellar signals. The central goal of the present manuscript is to test how selective impairment of cortico-cerebellar communication - not the simultaneous impairment of corticospinal, corticoreticulospinal, and cortico-cerebellar communication, and not a nonselective disruption of cerebellar output - disrupts behavior and neural dynamics across the cortico-cerebellar loop. Our conceptual advance, then, is to show that impairment of cortico-cerebellar communication does not typically block movement execution (as simultaneous perturbation of all motor cortical outputs does), but disrupts the fine kinematic details, similar to a direct manipulation downstream in the cerebellum. We have updated the text, particularly the Abstract, Introduction par. 1, and Discussion par. 1-2, to clarify this rationale and conclusion.

    1. The description of the connectivity of the loop illustrated in Figure 1 is straightforward. Motor cortex recipient PN neurons project to PN neurons, which then project directly to the cerebellar cortex and deep cerebellar nuclei, etc. Thus, the effect of any perturbation to PN neurons should be realized rapidly within neurons in the cerebellar cortex and deep cerebellar nuclei if they are part of this direct loop. However, onset latencies for the effect of the perturbations are not documented for these experiments (Figs 3&6 in the test/reaching conditions, and associated text). Similarly, latencies are not reported for the onset of changes in motor cortex neuron responses to PN perturbations in either condition (Figs 4&7 in the test/reaching conditions, and associated text). The only reference I could find to latencies specified the that required to reach the peak firing rate - not latency of the change. Specifically: "these were stereotypical, mostly consisting of transient excitation (Fig. 4B, left; median time of firing rate peak 120 ms)" - 120ms seems very long for the loop in Fig 1. It would be useful to know the latency between optogenetic stimulation in PN and changes in PN firing rate. And then the question is at what latency are the neurons in subsequent nodes altered? Quantification of latencies of the effects that are observes in the different nodes of the cortico-cerebellar loops would strengthen the authors' conclusion that they are actually studying the direct loop in Figure 1 which would then make the study's conclusions more compelling.

    We agree that it is important to characterize the latencies of neural responses to PN stimulation, and now provide these numbers for Purkinje cells, DCN neurons, and motor cortical neurons in the text and Supplemental Fig. 7. On stimulation of the PN, activity propagates first to Purkinje cells, then the DCN, and finally to motor cortex. We also quantify the latency of PN responses to motor cortical stimulation in Supplemental Fig. 1. (For a discussion of the rationale and limitations of our method, see also our response above to reviewer 1’s first comment.) Unfortunately, we have not been able to measure the delay from stimulation onset to the earliest spikes induced by ChR2 currents in PN neurons, as this would require simultaneous insertion of a stimulation fiber and recording probe to a deep target in the PN. Furthermore, we note that the earliest measurable response in Purkinje cells occurs 10 ms after stimulation onset, and this is likely an overestimate of the minimum latency, as it takes 6.25 ms for the laser to reach peak power under sinusoidal stimulation.

    1. Overall, there was often a sharp incongruity between the complexity of many of the findings described in results and accompanying figures and the short summary conclusion provided for the Results. Here is one of many examples (bottom of page 5), where the authors conclude "These results demonstrate that the cortico-cerebellar loop does not drive reaching, but fine-tunes the behavior to enable precise and accurate movement." Yet, what the results above describe is considerable heterogeneity and variability across animals and cases. These conclusion should be more aligned with/ justified by the author's description of their actual results.

    Throughout the Results section, we have now tied the interpretations more closely to the data. For example, in the instance the reviewer mentions, we now state: “These results demonstrate that PN stimulation impairs reaching performance, typically by disrupting precision, accuracy, duration or success rate of the movement.” In the first two paragraphs of the Discussion, we lay out our interpretation of the data as evidence that the cortico-cerebellar loop contributes to fine- tuning the movement, rather than driving it, but emphasize that this is an interpretation rather than a description of experimental results. Furthermore, we now address possible factors that could underlie the diversity of behavioral effects in the fourth paragraph of the Discussion (“It is possible that the variability of behavioral effects ...”).

    1. A related issue is the disconnection between description and summary, in the description of Figure 6- 8. The emphasis on correlation, yet the authors' main point here seems to be that there are changes in the activity in cortex and DCN induced by the PN stimulation during movement explain the changes in hand trajectory. For example, Figure 6D and its implications are not effectively described in the text.

    The main conclusion of figures 6 and 7 is that PN stimulation during movement alters movement-aligned cortical and DCN activity, but this modulation is typically subtle; that is, activity on control and laser trials is highly correlated for most neurons and time points. This is in contrast with more dramatic effects observed for perturbations delivered to other nodes in the loop; for instance, thalamic perturbations can robustly prevent the generation of the cortical pattern that drives movement (Sauerbrei et al. 2020). Supplemental Fig. 8D-E and Supplemental Fig. 9D-E suggest that these subtle stimulation-induced changes during movement are largely consistent with the changes that would be expected based on neural responses to laser alone, outside engagement with the task. Finally, the decoding analysis in Fig. 8 allows us to interpret these subtle neural changes: they do not appear to be random, but are consistent with the effects of stimulation on the hand. That is, the difference in hand velocity between laser and control trials decoded from neural activity is correlated with the observed hand velocity difference. We have added a video (supplemental video 3) to better visualize this result in all three spatial dimensions simultaneously, and have edited the text in the Results section to clarify these findings.

    1. Finally, the authors conclude that changes in the activity in cortex and DCN induced by the PN stimulation during movement explain the subtle deviations in hand trajectory and conclude that the cortico-cerebellar loop is responsible for fine-tuning movement parameters (bottom pf page 5 and top of page 8). However, i) the statement that this pathway fine-tunes motion is not justified by the analysis, and ii) the novelty is not made clear relative to prior work that has investigated cortico-cerebellar loop (beyond the experimental difference in perturbation site).

    Regarding (i), we agree that the fine-tuning is an interpretation rather than a direct reflection of the data presented in the paragraph, and have altered the statement accordingly: “Overall, these results show that the subtle changes in the activity in cortex and DCN induced by the PN stimulation during movement are consistent with the changes in hand trajectory for individual mice.” We now explain our interpretation of the data as supporting a fine-tuning role in the Discussion, rather than the Results. Regarding (ii), we have now clarified in the Abstract, Introduction, and Discussion that perturbation of the PN enables us to test the effects of a selective disruption of cortico-cerebellar communication, in contrast with direct manipulations of motor cortex or cerebellum (see also our response to comment 3.1 above).

    Overall, the text that follows in the discussion presented the findings in a far more clear and compelling way than much of the text in the Abstract, Introduction and Results "perturbing cortico-cerebellar communication did not block movement execution: animals were typically able to generate the basic motor pattern during optogenetic stimulation of the PN, and neural activity in cortex and cerebellum largely recapitulated the firing patterns observed during normal movement. Instead, PN perturbation altered arm kinematics, decreasing the precision and accuracy of the reach, and perturbation-induced shifts in neural activity explained these behavioral effects." The paper would be improved if the authors adapted this more precise style of reporting throughout.

    We have edited the main text throughout to improve clarity and precision.

    Read the original source
    Was this evaluation helpful?
  2. Reviewer #3 (Public Review):

    It is generally thought that the cerebellum is primarily involved in the short-timescale control of movements, while motor cortex is involved in motor planning. The present paper follows classic studies in primates and a recent study in mouse that investigated the role of cortico-cerebellar loops in motor control. To date, studies in both species applied perturbations to the cerebellum to then study changes in cortical activity. For example, it has been long known that cooling deep cerebellar nucleus produces changes in the responses of motor cortex neurons in primate (e.g., Meyer-Lohmann et al., 1975). Further, Gao and colleagues' recent paper (Nature 2018) used optogenetics to perturb responses in the deep cerebellar nucleus before licking movements. The authors of this 2018 nature paper conclude that persistent neural dynamics are maintained during voluntary movements by connectivity in within this cortico-cerebellar loop.

    The experiments are well performed, and the results are logically organized and presented. However, a main concern is that the authors have not well justified that these experiments prove a conceptual advance. The conclusions appear to be largely consistent with those of prior work, both regarding changes in the responses of motor cortex neurons, and resultant (subtle) changes in behavior (i.e., altered arm kinematics). The impact of the paper would be improved if the authors adapted a more precise style of reporting the novelty of their results throughout.

    Major concerns:

    1. The experiments are well performed, and the results are logically organized and presented. However, a main concern is that the authors have not well justified that these experiments prove a conceptual advance. As noted above, prior studies have probed the role of cortico-cerebellar loops by applying perturbations to cerebellar activity (cerebellar cortex and/or deep cerebellar nuclei) and quantifying changes in cortical activity prior to and during movement. The main novelty of the present study is that the authors perturbed the loop at a different locus, namely in the pontine nuclei (PN) which send projections to the cerebellum rather than directly to the cerebellum. The rationale for why this specific perturbation provides a conceptual advance to the field was not adequately motivated.

    The authors do clearly review prior literature showing that perturbation of cortico-cerebellar projections impacts the rest of the loop and behavior, they also well explain the application of their exciting new tool to specifically target PN neurons with their optogenetic stimulation. Yet, the authors do not motivate why it is important to specifically perturb the pontine nuclei (PN) to gain new insights into the role of "cortico-cerebellar loops" nor do they provide any reason to expect a difference in changes in loop dynamics for perturbations applied versus to the DCN. Indeed, the conclusions appear to be largely consistent with those of prior work, both regarding changes in the responses of motor cortex neurons, and resultant (subtle) changes in behavior (i.e., altered arm kinematics). Generally, these results are similar to those previously reported in primate DCN cooling experiments characterizing changes in hand movement in in a voluntary tracking task (e.g., Brooks et al., 1973; Conrad and Brooks 1974).

    1. The description of the connectivity of the loop illustrated in Figure 1 is straightforward. Motor cortex recipient PN neurons project to PN neurons, which then project directly to the cerebellar cortex and deep cerebellar nuclei, etc. Thus, the effect of any perturbation to PN neurons should be realized rapidly within neurons in the cerebellar cortex and deep cerebellar nuclei if they are part of this direct loop. However, onset latencies for the effect of the perturbations are not documented for these experiments (Figs 3&6 in the test/reaching conditions, and associated text). Similarly, latencies are not reported for the onset of changes in motor cortex neuron responses to PN perturbations in either condition (Figs 4&7 in the test/reaching conditions, and associated text). The only reference I could find to latencies specified the that required to reach the peak firing rate - not latency of the change. Specifically: "these were stereotypical, mostly consisting of transient excitation (Fig. 4B, left; median time of firing rate peak 120 ms)" - 120ms seems very long for the loop in Fig 1. It would be useful to know the latency between optogenetic stimulation in PN and changes in PN firing rate. And then the question is at what latency are the neurons in subsequent nodes altered? Quantification of latencies of the effects that are observes in the different nodes of the cortico-cerebellar loops would strengthen the authors' conclusion that they are actually studying the direct loop in Figure 1 which would then make the study's conclusions more compelling.

    2. Overall, there was often a sharp incongruity between the complexity of many of the findings described in results and accompanying figures and the short summary conclusion provided for the Results. Here is one of many examples (bottom of page 5), where the authors conclude "These results demonstrate that the cortico-cerebellar loop does not drive reaching, but fine-tunes the behavior to enable precise and accurate movement." Yet, what the results above describe is considerable heterogeneity and variability across animals and cases. These conclusion should be more aligned with/ justified by the author's description of their actual results.

    3. A related issue is the disconnection between description and summary, in the description of Figure 6- 8. The emphasis on correlation, yet the authors' main point here seems to be that there are changes in the activity in cortex and DCN induced by the PN stimulation during movement explain the changes in hand trajectory. For example, Figure 6D and its implications are not effectively described in the text.

    4. Finally, the authors conclude that changes in the activity in cortex and DCN induced by the PN stimulation during movement explain the subtle deviations in hand trajectory and conclude that the cortico-cerebellar loop is responsible for fine-tuning movement parameters (bottom pf page 5 and top of page 8). However, i) the statement that this pathway fine-tunes motion is not justified by the analysis, and ii) the novelty is not made clear relative to prior work that has investigated cortico-cerebellar loop (beyond the experimental difference in perturbation site).

    Overall, the text that follows in the discussion presented the findings in a far more clear and compelling way than much of the text in the Abstract, Introduction and Results "perturbing cortico-cerebellar communication did not block movement execution: animals were typically able to generate the basic motor pattern during optogenetic stimulation of the PN, and neural activity in cortex and cerebellum largely recapitulated the firing patterns observed during normal movement. Instead, PN perturbation altered arm kinematics, decreasing the precision and accuracy of the reach, and perturbation-induced shifts in neural activity explained these behavioral effects." The paper would be improved if the authors adapted this more precise style of reporting throughout.

    Read the original source
    Was this evaluation helpful?
  3. Reviewer #2 (Public Review):

    Guo et al examine the cortico-cerebellar loop during skilled forelimb movements in mice. The authors use optogenetic stimulation of the pontine nuclei (PN) and recordings in PN, cerebellar cortex, cerebellar nuclei (DCN), and motor cortex to show that PN output is transformed into a variety of activity patterns at different stages of the cortico-cerebellar loop. Stimulation only slightly alters movement-related activity in these structures and degrades movement accuracy. The authors conclude that the cortico-cerebellar loop fine tunes dexterous movement. The study is technically impressive, employing recordings in 4 brain regions, and recordings during optogenetic manipulations and behavior. The experiments are well done and the analyses are appropriate. The comparison across brain regions is comprehensive. The results that PN perturbation alters skilled movement and the perturbed activity could predict perturbed movement are important. The study adds to a long line of work supporting the view that the cortico-cerebellar pathway is required for fine motor control. I have a few comments on the interpretation and analysis which I believe could be addressed with changes to the text and additional analysis.

    1. The authors conclude that the cortico-cerebellar loop "does not drive movement" but "fine tunes" the movement. While I generally agree with this interpretation, I wonder if the authors could flush out the concepts of "driving movement execution" vs. "fine-tuning movement" more clearly. Do authors consider them separate processes? How can they be disentangled? I also feel the data on its own has some limitations that should be considered or discussed. First, the data shows that PN stimulation degrades movement accuracy. However, this does not yet reveal the function of the cerebellar loop in fine motor control. Certain places in the text makes stronger assertions (for example, "cortico-cerebellar loop fine-tunes movement parameters") that I feel the data does not support. It is not clear from the data how the loop tunes movement parameters. Second, Fig. 5F shows that stimulating PN blocked movement initiation in some sessions (this is also mentioned in the Methods). Could the authors consider the possibility that stimulating PN at a higher intensity might block movement? This is related to the distinction between "driving" vs. "fine-tuning" movement. At the very least, the authors should discuss these limitations and possibilities.

    2. Related to point 1, in Fig. 5F, for stimulation trials in which mice failed to initiate movement, did mice fail to move altogether, or did they move in an abnormal fashion?

    3. In the abstract, the authors state that PN stimulation is "reduced to transient excitation in motor cortex". Also in the results (page 5) and discussion (page 8), "pontine stimulation only led to increases in cortical firing rates". These statements are based on the comparison between Fig 3D, 3F, and 4B. But I think the current presentation is somewhat misleading. First, Fig 3D, 3F, and 4B use different neuron selections that make direct comparison difficult. Fig 3 shows all neuron from Purkinje cell and DCN recordings. Fig 4B shows only PN-tagged motor cortex neurons. Furthermore, based on the methods description, it appears that PN-tagged neurons were defined using one-sided sign-rank test. Since the test is one tailed, does that mean neurons shown in Fig 4B are, by definition, neurons significantly excited by photostimulation? Looking at Fig 4B and 4C closely, there appear to be neurons suppressed by PN stimulation. Could the authors organize the rows in Fig 4 in the same way as Fig 3, where neurons that show suppression are grouped together?

    4. Fig 7 shows that PN stimulation has only subtle effects on movement-related activity in motor cortex. However, only a small portion (1/8) of the motor cortex neurons show modulation to PN stimulation. Fig 7 shows all neurons. Would the results look similar for PN-tagged neurons?

    5. Page 3 "Our observation that the activity of some motor cortex-recipient PN neurons is aligned both to the cue and movement suggests that these neurons might integrate signals of multiple modalities." Presumably, motor cortex neurons also have cue and movement-related activity and PN simply inherits this activity from the motor cortex.

    6. Do Purkinje cells follow the 40 Hz PN stimulation like in the multi-unit recordings. The PSTHs in Fig 3 are too smoothed out to see this.

    7. For the correlation analysis in Fig 6C top and 7C top, is the correlation computed from z-scored firing rates rather than on raw firing rates? This is not clear from the text. If computed on raw firing rates, one would expect the correlation to be above 0 even before photostimulation, since different neurons exhibit different baseline firing rates that presumably will be the same across control and stim trials.

    Read the original source
    Was this evaluation helpful?
  4. Reviewer #1 (Public Review):

    Guo et al. describes interesting experiments recording from various sites along a cortico-cerebellar loop involved in limb control. Using neuropixels recordings in motor cortex, pontine nuclei, cerebellar cortex and nuclei, the authors amass a large physiological dataset during a cued reach-to-grasp task in mice. In addition to these data, the authors 'ping' the system with optogenetic activation of pontocerebellar neurons, asking how activity introduced at this node of the loop propagates through the cerebellum to cortex and influences reaching. From these experiments they conclude the following: the cerebellum transforms activity originating in the pontine nuclei, this activity is not sufficient to initiate reaches, and supports the long standing view that the cerebellum 'fine tunes' movement, since reaches are dysmetric in response to pontine stimulation. Overall these data are novel, of high quality, and will be of interest to a variety of neuroscientists. As detailed below however, I think these data could provide much more insight than they currently do. Thus below I provide some suggestions on improving the manuscript.

    1. Since the loop is the focus of this study, it would be nice if the authors better characterized latencies of responsivity to pontine stimulation through the loop, to address how cortically derived information routed to the cerebellum may loop back to influence cortical function. In the data provided, we know that pontine stimulation modulates Purkinje and deep nuclear firing (but latency to responses are not transparently provided in the main text, if anywhere), while motor cortical responses peak at 120 ms (after stimulus onset?, unclear), and that this responsivity is preferentially observed in neurons engaged early in the reaching movement. Is the idea, then, that cortical activity early in the reach is further modulated by cerebellar processing to (Re) influence that same cortical population? Does this interpretation align with the duration of reaches, the duration of early responsive activity during reach, and the latency of responsivity; or is the idea that independent information from other modalities entering the pontine nuclei modulates early cells? Latency to respond at the different nodes, might aid in thinking through what these data mean for the function of the loop.

    2. Many of the figures need work to aid interpretation. Axis labels are often missing (eg 2F); color keys are often unlabeled (2F); color gradients often used but significance thresholds are hard to evaluate (using same colors for z scores and control / laser is confusing 6, 8); and within-figure keys would be useful (5D-h). These issues occur throughout the manuscript.

    3. Relatedly, but also conceptually, Figure 3B has particular issues, such as identifying where the neuropixel multiunit activity is coming from. I assume that in the gray boxes illustrating the spatio-temporal profile of spiking band activity that the lower part of the box is the ventral direction, upper, dorsal. This is not spelled out. From the two examples it would seem that the spiking band is in different places in the cerebellum, undermining, I think, the objective of the figure. It would be sensible to revisit this entire figure to identify the key takeaways and design figures around those ideas. As it stands, these examples appear anecdotal. Consider moving this to a supplement. Powerband density strength is missing an axis. More importantly, it would be nice to corroborate the interpretation of the MUA with the single unit recordings, since the idea is that many neurons are entraining to the PN activity. Yet, the examples don't seem particularly entrained. Is the activity being picked up on just axonal firing of the PN axons? Fourier analysis of spiking of isolated neurons in cerebellum should be used to corroborate the idea that cerebellar neurons are entraining, rather than the neuropixel picking up entrained PN axons.

    4. The use of the GLM is puzzling. In addressing the question of how cerebellum and motor cortex interact (from the Abstract, "how and why" do these regions interact) it is unclear why these regions are treated separately. I would have expected some kind of joint GLM where DCN activity is used to predict M1 variance (5 co-recordings are reported but nothing to analyze?); or where DCN + M1 activity is used to decode kinematics to see if it is better than one or the other alone. As it stands, we learn that there is more kinematic information in the motor cortex than in DCN. This is not necessarily surprising given previous literature on cerebellar contributions to reaching movements. In principle the idea that 'PN stimulation might perturb reaching kinematics through descending projections to the spinal cord, or by altering activity in motor cortex' is treated as mutually exclusive outcomes, though it is highly unlike to be so.' Analyzing M1+DCN together could address whether DCN activity adds nothing to decoding kinematics that isn't there in M1 or adds something that M1 does not have access to. The main point here is that the physiological datasets could be better leveraged with these fits to derive insight into the interactions of the loop. R2 should be provided in the GLMs (Fig 8) to assess statistically how well they perform relative to one another, not just correlations between the two.

    Read the original source
    Was this evaluation helpful?
  5. Evaluation Summary:

    The present paper investigated the role of cortico-cerebellar loops in motor control with high density physiological recordings and by using optogenetics to perturb responses of precerebellar neurons in the pontine nuclei during reaching. The study adds to a long line of work supporting the view that the cortico-cerebellar pathway is required for fine motor control. The experiments are well performed, but a number of revisions in analysis and presentation are required.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. The reviewers remained anonymous to the authors.)

    Read the original source
    Was this evaluation helpful?