Neural circuit mechanisms for steering control in walking Drosophila

Reviewer #1 (Public review):

Summary:

The paper addresses the knowledge gap between the representation of goal direction in the central complex and how motor systems stabilize movement toward that goal. The authors focused on two descending neurons, DNa01 and 02, and showed that they play different roles in steering the fly toward a goal. They also explored the connectome data to propose a model to explain how these DNs could mediate response to lateralized sensory inputs. They finally used lateralized optogenetic activation/inactivation experiments to test the roles of these neurons in mediating turnings in freely walking flies.

Strengths:

The experiments are well-designed and controlled. The experiment in Figure 4 is elegant, and the authors put a lot of effort into ensuring that ATP puffs do not accidentally activate the DNs. They also have explained complex experiments well. I only have minor comments for the authors.

Weaknesses:

(1) I do not fully understand how the authors extracted the correlation functions from the population data in Figure 1. Since the ipsilateral DNs are anti-correlated with the contralateral ones, I expected that the average will drop to zero when they are pooled together (e.g., 1E-G). Of course, this will not be the case if all the data in Figure 1 are collected from the same brain hemisphere. It would be helpful if the authors could explain this.

(2) What constitutes the goal directions in Figures 1-3 and 8, as the authors could not use EPG activity as a proxy for goal directions? If these experiments were done in the dark, without landmarks, one would expect the fly's heading to drift randomly at times, and they would not engage the DNa01/02 for turning. Do the walking trajectories in these experiments qualify as menotactic bouts?

(3) In Figure 2B, the authors mentioned that DNa02 overpredicts and 01 underpredicts rapid turning and provided single examples. It would be nice to see more population-level quantification to support this claim.

Reviewer #2 (Public review):

The data is largely electrophysiological recordings coupled with behavioral measurements (technically impressive) and some gain-of-function experiments in freely walking flies. Loss-of-function was tested but had minimal effect, which is not surprising in a system with partially redundant control mechanisms. The data is also consistent with/complementary to subsequent manuscripts (Yang 2023, Feng 2024, and Ros 2024) showing additional descending neurons with contributions to steering in walking and flying.

The experiments are well executed, the results interesting, and the description clear. Some hypotheses based on connectome anatomy are tested: the insights on the pre-synaptic side - how sensory and central complex heading circuits converge onto these DNs are stronger than the suggestions about biomechanical mechanisms for how turning happens on the motor side.

Of particular interest is the idea that different sensory cues can converge on a common motor program. The turn-toward or turn-away mechanism is initiated by valence rather than whether the stimulus was odor or temperature or memory of heading. The idea that animals choose a direction based on external sensory information and then maintain that direction as a heading through a more internal, goal-based memory mechanism, is interesting but it is hard to separate conclusively.

The "see-saw", where left-right symmetry is broken to allow a turn, presumably by excitation on one side and inhibition of the other leg motor modules, is interesting but not well explained here. How hyperpolarization affects motor outputs is not clear.

The statement near Figure 5B that "DNa02 activity was higher on the side ipsilateral to the attractive stimulus, but contralateral to the aversive stimulus" is really important - and only possible to see because of the dual recordings.

Reviewer #3 (Public review):

Summary:

Rayshubskiy et al. performed whole-cell recordings from descending neurons (DNs) of fruit flies to characterize their role in steering. Two DNs implicated in "walking control" and "steering control" by previous studies (Namiki et al., 2018, Cande et al., 2018, Chen et al., 2018) were chosen by the authors for further characterization. In-vivo whole-cell recordings from DNa01 and DNa02 showed that their activity predicts spontaneous ipsilateral turning events. The recordings also showed that while DNa02 predicts transient turns DNa01 predicts slow sustained turns. However, optogenetic activation or inactivation showed relatively subtle phenotypes for both neurons (consistent with data in other recent preprints, Yang et al 2023 and Feng et al 2024). The authors also further characterized DNa02 with respect to its inputs and showed a functional connection with olfactory and thermosensory inputs as well as with the head-direction system. DNa01 is not characterized to this extent.

Strengths:

(1) In-vivo recordings and especially dual recordings are extremely challenging in Drosophila and provide a much higher resolution DN characterization than other recent studies that have relied on behavior or calcium imaging. Especially impressive are the simultaneous recordings from bilateral DNs (Figure 3). These bilateral recordings show clearly that DNa02 cells not only fire more during ipsilateral turning events but that they get inhibited during contralateral turns. In line with this observation, the difference between left and right DNa02 neuronal activity is a much better predictor of turning events compared to individual DNa02 activity.

(2) Another technical feat in this work is driving local excitation in the head-direction neuronal ensemble (PEN-1 neurons), while simultaneously imaging its activity and performing whole-cell recordings from DNa02 (Figure 4). This impressive approach provided a way to causally relate changes in the head-direction system to DNa02 activity. Indeed, DNa02 activity could predict the rate at which an artificially triggered bump in the PEN-1 ring attractor returns to its previous stable point.

(3) The authors also support the above observations with connectomics analysis and provide circuit motifs that can explain how the head direction system (as well as external olfactory/thermal stimuli) communicated with DNa02. All these results unequivocally put DNa02 as an essential DN in steering control, both during exploratory navigation as well as stimulus-directed turns.

Weaknesses:

(1) I understand that the first version of this preprint was already on biorxiv in 2020, and some of the "weaknesses" I list are likely a reflection of the fact that I'm tasked to review this manuscript in late 2024 (more than 4 years later). But given this is a 2024 updated version it suffers from laying out the results in contemporary terms. For instance, the manuscript lacks any reference to the DNp09 circuit implicated in object-directed turning and upstream to DNa02 even though the authors cite one of the papers where this was analyzed (Braun et al, 2024). More importantly, these studies (both Braun et al 2024 and Sapkal et al 2024) along with recent work from the authors' lab (Yang et al 2023) and other labs (Feng et al 2024) provide a view that the entire suite of leg kinematics changes required for turning are orchestrated by populations of heterogeneous interconnected DNs. Moreover, these studies also show that this DN-DN network has some degree of hierarchy with some DNs being upstream to other DNs. In this contemporary view of steering control, DNa02 (like DNg13 from Yang et al 2023) is a downstream DN that is recruited by hierarchically upstream DNs like DNa03, DNp09, etc. In this view, DNa02 is likely to be involved in most turning events, but by itself unable to drive all the motor outputs required for the said events. This reasoning could be used while discussing the lack of major phenotypes with DNa02 activation or inactivation observed in the current study, which is in stark contrast to strong phenotypes observed in the case of hierarchically upstream DNs like DNp09 or DNa03. In the section, "Contributions of single descending neuron types to steering behavior": the authors start off by asking if individual DNs can make measurable contributions to steering behavior. Once more, any citations to DNp09 or DNa03 - two DNs that are clearly shown to drive strong turning-on activation (Bidaye et al, 2020, Feng et al 2024) - are lacking. Besides misleading the reader, such statements also digress the results away from contemporary knowledge in the field. I appreciate that the brief discussion in the section titled "Ensemble codes for steering" tries to cover these recent updates. However, I think this would serve a better purpose in the introduction and help guide the results.

(2) The second major weakness is the lack of any immunohistochemistry (IHC) images quantifying the expression of the genetic tools used in these studies. Even though the main split-Gal4 tools for DNa01 and DNa02 were previously reported by Namiki et al, 2018, it is important to document the expression with the effectors used in this work and explicitly mention the expression in any ectopic neurons. Similarly, for any experiments where drivers were combined together (double recordings, functional connectivity) or modified for stochastic expression (Figure 8), IHC images are absolutely necessary. Without this evidence, it is difficult to trust many of the results (especially in the case of behavioral experiments in Figure 8). For example, the DNa01 genetic driver used by the authors is also expressed in some neurons in the nerve cord (as shown on the Flylight webpage of Janelia Research Campus). One wonders if all or part of the results described in Figure 8 are due to DNa01 manipulation or manipulation of the nerve cord neurons. The same applies for optic lobe neurons in the DNa02 driver.

(3) The paper starts off with a comparative analysis of the roles of DNa01 and DNa02 during steering. Unfortunately, after this initial analysis, DNa01 is largely ignored for further characterization (e.g. with respect to inputs, connectomics, etc.), only to return in the final figure for behavioral characterization where DNa01 seems to have a stronger silencing phenotype compared to DNa02. I couldn't find an explanation for this imbalance in the characterization of DNa01 versus DNa02. Is this due to technical reasons? Or was it an informed decision due to some results? In addition to being a biased characterization, this also results in the manuscript lacking a coherent thread, which in turn makes it a bit inaccessible to the non-specialist.

(4) There seems to be a discrepancy with regard to what is emphasized in the main text and what is shown in Figures S3/S4 in relation to the role of these DNs in backward walking. There are only two sentences in the main text where these figures are cited.
a) "DNa01 and DNa02 firing rate increases were not consistently followed by large changes in forward velocity (Figs. 1G and S3)."
b) "We found that rotational velocity was consistently related to the difference in right-left firing rates (Fig. 3B). This relationship was essentially linear through its entire dynamic range, and was consistent across paired recordings (Fig. 3C). It was also consistent during backward walking, as well as forward walking (Fig. S4)."
These main text sentences imply the role of the difference between left and right DNa02 in turning. However, the actual plots in the Figures S3 and S4 and their respective legends seem to imply a role in "backward walking". For instance, see this sentence from the legend of Figure S3 "When (ΔvoltageDNa02>>ΔvoltageDNa01), the fly is typically moving backward. When (firing rateDNa02>>firing rateDNa01), the fly is also often moving backward, but forward movement is still more common overall, and so the net effect is that forward velocity is small but still positive when (firing rateDNa02>>firing rateDNa01). Note that when we condition our analysis on behavior rather than neural activity, we do see that backward walking is associated with a large firing rate differential (Fig. S4)." This sort of discrepancy in what is emphasized in the text, versus what is emphasized in the figures, ends up confusing the reader. More importantly, I do not agree with any of these conclusions regarding the implication of backward walking. Both Figures S3 and S4 are riddled with caveats, misinterpretations, and small sample sizes. As a result, I actually support the authors' decision to not infer too much from these figures in the "main text". In fact, I would recommend going one step further and removing/modifying these figures to focus on the role of "rotational velocity". Please find my concerns about these two figures below:
a) In Figures S3 and S4, every heat map has a different scale for the same parameter: forward velocity. S3A is -10 to +10mm/s. S3B is -6 to +6 S4B (left) is -12 to +12 and S4B (right) is -4 to +4. Since the authors are trying to depict results based on the color-coding this is highly problematic.
b) Figure S3A legend "When (ΔvoltageDNa02>>ΔvoltageDNa01), the fly is typically moving backward." There are also several instances when ΔvoltageDNa02= ΔvoltageDNa01 and both are low (lower left quadrant) when the fly is typically moving backwards. So in my opinion, this figure in fact suggests DNa02 has no role in backward velocity control.
c) Based on the example traces in S4A, every time the fly walks backwards it is also turning. Based on this it is important to show absolute rotational velocity in Figure S4C. It could be that the fly is turning around the backward peak which would change the interpretation from Figure S4C. Also, it is important to note that the backward velocities in S4A are unprecedentedly high. No previous reports show flies walking backwards at such high velocities (for example see Chen et al 2018, Nat Comm. for backward walking velocities on a similar setup).
d) In my opinion, Figure S4D showing that right-left DNa02 correlates with rotational velocity, regardless of whether the fly is in a forward or backward walking state, is the only important and conclusive result in Figures S3/S4. These figures should be rearranged to only emphasize this panel.

(5) Figure 3 shows a really nice analysis of the bilateral DNa02 recordings data. While Figure S5 shows that authors have a similar dataset for DNa01, a similar level analysis (Figures 3D, E) is not done for DNa01 data. Is there a reason why this is not done?

(6) In Figure 4 since the authors have trials where bump-jump led to turning in the opposite direction to the DNa02 being recorded, I wonder if the authors could quantify hyperpolarization in DNa02 as is predicted from connectomics data in Figure 7.

(7) Figure 6 suggests that DNa02 contains information about latent steering drives. This is really interesting. However, in order to unequivocally claim this, a higher-resolution postural analysis might be needed. Especially given that DNa02 activation does not reliably evoke ipsilateral turning, these "latent" steering events could actually contain significant postural changes driven by DNa02 (making them "not latent"). Without this information, at least the authors need to explicitly mention this caveat.

(8) Figure 7 would really benefit from connectome data with synapse numbers (or weighted arrows) and a corresponding analysis of DNa01.

(9) In Figure 8E, the most obvious neuronal silencing phenotype is decreased sideways velocity in the case of DNa01 optogenetic silencing. In Figure S2, the inverse filter for sideways velocity for DNa01 had a higher amplitude than the rotational velocity filter. Taken together, does this point at some role for DNa01 in sideways velocity specifically?

(10) In Figure 8G, the effect on inner hind leg stance prolongation is very weak, and given the huge sample size, hard to interpret. Also, it is not clear how this fits with the role of DNa01 in slow sustained turning based on recordings.

Author response:

We thank the reviewers for their feedback. We are currently revising the manuscript to address their questions and concerns. Here we briefly summarize our planned revisions.

Reviewer 1 requested clarification on three points. We will clarify all these points with text edits. One point is brief enough to be addressed here: in cases when we pooled data from the left and right hemispheres, the reviewer wants to know how this was done. Simply put, we defined the “ipsi” side of the body as the side where the recorded DN resided, and we defined “contra” as the other side.

Reviewer 2 requested clarification on two minor points. We will clarify these points with text edits and with an additional analysis.

Reviewer 3 had a number of substantive concerns. Briefly:

(1) The reviewer asks us to improve its discussion of some relevant literature. We will provide updated information on the DN steering network, and in particular, we will cite Bidaye et al. 2020 and Sapkal et al. 2024. We apologize for the oversight.

(2) The reviewer asks us for immunofluorescent images documenting the expression patterns of our effector transgenes. With regard to GtACR1::eYPF expression, we will include these images in our resubmission. With regard to ReachR expression, we expressed this reagent stochastically under hs-FLP control, and so different brains had different expression patterns; however, we carefully documented the number of DNa02 cells that expressed ReachR in each brain. With regard to GFP expression, these expression patterns are available online from the FlyLight documentation associated with Namiki et al. eLife 2018 (https://splitgal4.janelia.org/precomputed/Descending%20Neurons%202018.html). The UAS-GFP transgene used by Namiki et al. 2018 (pJFRC200-10XUASIVS-myr::smGFP-HA in attP18) is different from the UAS-GFP transgene we used (10XUAS-IVS-mCD8::GFP(su(Hw)attP8), and so there may be minor differences in expression pattern. However, it should be noted that we only used GFP expression to target somata for patch clamp recording, and DNa01 and DNa02 somata have a distinctive location and a distinctive size; when we performed these recordings, we only targeted a soma in this location, and we verified that there were no “distractor” somata in this vicinity with similar size and appearance. The same applies to patch clamp recordings targeted via Halo7 expression (SiR110-HaloTag fluorescence). In paired recordings from both DNa02 and DN01, we verified the identity of each cell as described in Fig. S1.

(3) The reviewer asks why we focused on DNa02 in the latter part of the manuscript, rather than DNa01. We made this decision because DNa02 is more highly predictive of steering behavior, as compared to DNa01 (Fig. 1H). Also, an impulse of DNa02 activity is followed by a relatively large turning maneuver, on average, whereas an impulse of DNa01 activity is followed by a relatively small turning maneuver (Fig. 1E-F). Moreover, DNa02 has many more synaptic inputs in the brain (Fig. 7A), and it has many more direct synaptic connections onto motor neurons (Fig. 1B).

(4) The reviewer highlights difficulties in interpreting DN activity during backward movement (Figs. S3/S4). We included this material in the spirit of completeness, but we agree with the reviewer that it is difficult to interpret. In our revision, we will omit Fig. S3C and Fig. S4A-B, and we will revise these legends to improve clarity.

(5) The reviewer asks why do a systematic analysis of paired DNa01 recordings, as we did for DNa02. It is difficult to get paired right/left recordings from two DNs of the same type in the same fly, while the fly is walking vigorously, and we were only able to get two such paired recordings from DNa01. We did not feel this was a sufficiently large sample size to support a systematic analysis. We chose not to invest more time in getting more paired DNa01 recordings because we thought that DNa02 was more important, for the reasons noted above.

(6) The reviewer asks for an analysis of trials where bump-jump led to turning in the opposite direction to the DNa02 being recorded. We will provide this analysis in the revision.

(7) The reviewer points out that “latent” steering drives might not be latent, as they might produce small postural changes we are not capturing. This is a fair point, and we will note this in our revision.

(8) The reviewer asks for a systematic analysis of DNa01 inputs in Figure 7, similar to our analysis of DNa02 inputs. Here we would prefer to focus on DNa02, for three reasons. First, we think DNa02 is likely more important, for the reasons noted above. Second, there has been some uncertainty as to the identity of DNa01 in connectome data; indeed, in the hemibrain data set, the cell recently identified as DNa01 was annotated as VES006 (Schlegel et al. Nature 634: 139-152). Third, the cell now identified as DNa01 does not receive direct input from either the central complex or the mushroom body, and for this reason, we felt that the inputs to DNa01 might be less interesting to a general audience.

(9) The reviewer wonders whether DNa01 is more involved in sideways movement, rather than rotational movement. Our data do not support this conclusion: rather, our data show that DNa01 is only weakly correlated with sideways movement. Thus, the forward filter (Fig. 1F) shows that an impulse of DNa01 activity is (on average) followed by a relatively small amount of sideways movement. Conversely, the reverse filter (in Fig. S2I) shows that an impulse of sideways movement is (on average) preceded by a relatively large amount of DNa01 activity.

(10) The reviewer points out that the phenotype associated with optogenetic suppression in Fig. 8G is weak. We will highlight this point and discuss potential reasons for this weak phenotype in the revision.