How the insect central complex could coordinate multimodal navigation

  1. Xuelong Sun
  2. Shigang Yue
  3. Michael Mangan

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This paper builds on a previously published computational model of the insect central complex developed to generate a biologically plausible neural circuit for producing visually guided navigation behavior to show how the same model can be used to produce navigation behavior in response to multimodal sensory information. In particular, the authors show that olfactory navigation as well as wind-guided navigation can be seamlessly integrated with visual behaviors. The work is significant, valuable and of broad interest to circuit and computational neuroscientists.

    (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. Reviewer #1 agreed to share their name with the authors.)

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

The central complex of the insect midbrain is thought to coordinate insect guidance strategies. Computational models can account for specific behaviours, but their applicability across sensory and task domains remains untested. Here, we assess the capacity of our previous model (Sun et al. 2020) of visual navigation to generalise to olfactory navigation and its coordination with other guidance in flies and ants. We show that fundamental to this capacity is the use of a biologically plausible neural copy-and-shift mechanism that ensures sensory information is presented in a format compatible with the insect steering circuit regardless of its source. Moreover, the same mechanism is shown to allow the transfer cues from unstable/egocentric to stable/geocentric frames of reference, providing a first account of the mechanism by which foraging insects robustly recover from environmental disturbances. We propose that these circuits can be flexibly repurposed by different insect navigators to address their unique ecological needs.

Article activity feed

  1. Version published to 10.7554/elife.73077 on eLife
  2. eLife

    Author Response:

    Reviewer #1:

    The paper by Sun et al uses a previously published computational model of the insect central complex and expands the applicability of this model. While the original model was developed to generate a biologically plausible neural circuit for producing visually guided navigation behavior (integrating three distinct navigation strategies), the new paper shows that the same model can be used to produce navigation behavior in response to multimodal sensory information. In particular, the authors show that olfactory navigation as well as wind-guided navigation can be seamlessly integrated with visual behaviors.

    The authors link the computational model to postulate neural mechanisms that are inspired by known features of the insect central complex. Using the model, behavioral observations, in particular from ants, can be readily reproduced, including tasks in which the animals had to switch between guidance cues, e.g. from visually driven path integration to odor based location of a nest entrance, or were blown off course by wind.

    The manuscript clearly requires that the first paper by the same group is read first, as many core concepts of the computational model are introduced in that paper. When viewed as such an extension (as intended by the 'Research Advances' article type), the paper adds valuable insights and stimulates thought and hypothesis development regarding concepts of multimodal integration. In light of the increasing amount of data on insect brain connectomics, hypotheses based on biologically inspired computational models are highly useful.

    While the authors refer to their computational model as 'biologically realistic', several features (e.g. a ring attractor circuit in the fan shaped body) are speculative and, to date, unconfirmed in any insect, despite the existence of the fruit fly connectome. This does not mean that the model is conceptually wrong, especially as it allows to faithfully reproduce complex behavioral data and shifting of activity 'bumps' across the width of the central complex (as is key for the model) is likely one of the principal functions of the fan-shaped body circuit. Yet the exact nature of the neural implementation might have to be adjusted once relevant data on ants becomes available. The model, as it stands, should hence be seen as 'biologically plausible" rather than 'realistic'.

    That said, the addition of the new aspects of the model shows how flexible the proposed circuit is for coordinated navigational control in insects, and, interestingly, highlights analogous concepts found in the basal ganglia of mammals - a thought-provoking parallel that is in line with ideas of deep homology between these distant brain regions.

    Thanks for your comments. We have carefully check thorough the paper to update our statements: we use 'bio-plausible' when there is no direct biological evidence supporting this computation.

    Reviewer #2:

    In their previous manuscript, Sun et al. combined existing and hypothesized circuit motifs within the insect central complex (CX) to propose an integrated model for how the region might enable visual route following and homing. In their original framework, circuit motifs within the fan-shaped body allowed for appropriate context-switching. They now show how roughly the same motifs could also allow the model to (optimally) incorporate other sensory inputs, such as odor concentration gradients and wind direction cues, thereby enabling an insect to use the CX for additional behaviors in a context-dependent manner. The model's performance is evaluated in comparison to the behavior of larval and adult insect behavior (flies and ants, for example). This study represents a useful extension of the model's scope, but it would benefit from some additional computational exploration and explanation. As it stands, the figures and figure legends are not self-contained enough to be clearly understandable to the average reader. This new piece would also benefit from a greater focus on alternative models and alternative neural pathways that also subserve at least some of the additional navigational behaviors. The existence of direct olfactory-motor pathways is mentioned in Discussion for example, but deserves to be explored in Results as well. Otherwise, the significance of the authors' model reproducing Drosophila larval chemotaxis is not clear: note that larvae do not have CX circuits of the sort that the model proposes.

    Thanks for the considered feedback. 1) to be more self-contained, we added a new Figure 1 that introduces the previous models’ key features. We also updated the legends and captions throughout to make them clearer; 2) to justify our simulation of Drosophila olfactory navigation in larvae that do not possess of the CX, we added a specific discussion to the text (Results and Discussion), and added a panel showing the different brain structures for adult and larvae in Figure 2 (revised version). We hope that every important thing is clear now.

    Reviewer #3:

    Sun et al. propose an excellent study on multi-sensory fusion in ant when the animal is confronted to both wind and odour source or when conflict exists between chemotaxis and path integration. The paper is very well written and the figures are very clearly designed. The list of references is complete. On my opinion, this paper might be considered as a companion paper of a previous paper published in eLife (Sun et al 2020) featuring a strong impact on the plausible strategy that can be used by ant to integrate various cues (odour, wind, proprioception) but a weaker impact (because already published in the first paper) on the neuronal model of the various structures involved in the central complex of the ant's brain: protocerebrum bridge (PB), fan-shape body (FB) and ellipsoid body (EB). An interesting copy and shift function already described in (Sun et al 2020, eLife) seems to be well suited to generate the appropriate motor commands of the heading in response to a difference between the current heading and the measured heading (sensory feedback control). To summarize, this copy and shift function tends to minimize the heading error by making ant turn left or right. It is worth noting that the simulated responses of the ant have been compared to the real data published in previous papers by others.

    I have the following main concerns about this work:

    • First, the steering function as regard of the shift and copy mechanism should be recalled and carefully explained (figure 2B of the previous paper published in eLife)

    We have added a new Figure 1 that introduces the previous model (including the crucial copy-and-shift mechanism) to the reader.

    • About figure 1C and the on-off response of the ant: authors argue that their model replicates faithfully the ant's response: however to be absolutely convinced by this statement, authors must take into account in their simulation the following parameters used by Alavarez-Salvada et al. : ground speed, angular velocity, curvature and turn probability. If the simulated and real responses are similar, we should observe an ON response consisting of upwind orientation coupled with faster and straighter trajectories, and an OFF response consisting of slower and more curved trajectories. It is not clearly the case in the current version of the paper due to a lack of thorough analysis between the parameters listed previously. I can not see more curved trajectories in the OFF response.

    Thanks for these comments. We have included more detailed analysis of the paths of the model in the supplementary material of the revised paper (Figure 2-Figure supplement 3 and Figure 3-Figure supplement 1) in preference for a clear behavioural example in the main figures (Figure 2C and Figure 3C). Specifically, the supplementary analysis includes angular velocity (as requested), upwind speed (a function of path directness rather than animal velocity), and the perceived odour concentrations. These were chosen to match the data presented by Álvarez-Salvado et al. (2018) and allow direct comparison with their results.

    Indeed, the model generates higher 'upwind speed' and smaller 'angular velocity' (more straight) during ON-response phases than that of OFF-response as reported in real animals.

    Regarding plotting ground speed, in the current model the speed of motion is constant which we have clarified in the Methods section.

    Regarding 'curvature' and 'turn probability', given our simulation settings we do not think these two parameters are informative for the following reasons. The ground speed is constant within each simulation, and thus the 'curvature' which is calculated by dividing angular velocity by ground speed provides identical information as the 'angular velocity'. Also, the model alters the heading angle at each step so there is no notion of turning vs. not turning against which to assign a probability value.

    Given the above, we have softened the claims in the Results section from 'closely matches the behavioural data' to 'similar to the behavioural data' which we believe more closely aligns with the reviewer’s perspective.

    • About the simulated behaviour shown in figure 2C: this is a very interesting and critical point because here a conflict is produced between two sensory modalities: path integration and chemotaxis. Authors must clarify how is this conflict processed/managed by the ring attractor? Is it due to changes in the dynamics of the measurements (odor and path integration) or due to changes in the ring attractor itself? By the way, I strongly encourage authors to provide temporal simulation of the ring attractor state: plot the input and ring's output signals at different time steps to see clearly a shift in the bump output.

    Thank you, we have considered this comment in detail and agree that this could be clearer. For clarity, the resultant paths are driven by the ring attractor and integrating the dynamic values of PI and olfactory based on their dynamic certainties. This capability is inherent in ring attractor networks and something that we want to highlight. To make this point clearer, we have added a new video (Figure 3-video 1) to demonstrate the dynamics of the process.

    • About the angular resolution of the PB, FB and EB: how many different angular directions are coded? How many neurons are simulated in each structure? In MM section (equation 10 page 13) it is indicated that the angular resolution (shifting accuracy) has been improved by a factor 10 (from 45{degree sign} to 4.5{degree sign}) to achieve better performance. This point must be indicated and discussed in the main text because it is related to the accuracy of the heading measurement and thus to the behaviour of the simulated ant. How can the ant improve its heading accuracy despite a coarse resolution of central complex in the heading measurement?

    Apologies for the confusion here. The accuracy of the heading direction system is not course due to the population encoding across the 8 neural bins of the PB. Much in the same way that many colours can be encoded using just red, green and blue values combined appropriately.

    When we said that we improved the shifting 'accuracy' we were referring to resolution by which the 'activity bump' could be shifted across the population. In our previous model, 'shifts' of 45deg (i.e., one column each step) was sufficient for accurate visual guidance, but insufficient for accurate olfactory navigation (from experiments). Thus, we improved the resolution by which the activity bump could be shifted to 4.5deg. To clarify this point we have changed the term 'accuracy' for 'resolution' in the main text.

    An interesting research question for future work could be an investigation of the mechanism of shifting and their impact on performance in various guidance tasks.

  3. eLife

    Evaluation Summary:

    This paper builds on a previously published computational model of the insect central complex developed to generate a biologically plausible neural circuit for producing visually guided navigation behavior to show how the same model can be used to produce navigation behavior in response to multimodal sensory information. In particular, the authors show that olfactory navigation as well as wind-guided navigation can be seamlessly integrated with visual behaviors. The work is significant, valuable and of broad interest to circuit and computational neuroscientists.

    (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. Reviewer #1 agreed to share their name with the authors.)

  4. eLife

    Reviewer #1 (Public Review):

    The paper by Sun et al uses a previously published computational model of the insect central complex and expands the applicability of this model. While the original model was developed to generate a biologically plausible neural circuit for producing visually guided navigation behavior (integrating three distinct navigation strategies), the new paper shows that the same model can be used to produce navigation behavior in response to multimodal sensory information. In particular, the authors show that olfactory navigation as well as wind-guided navigation can be seamlessly integrated with visual behaviors.

    The authors link the computational model to postulate neural mechanisms that are inspired by known features of the insect central complex. Using the model, behavioral observations, in particular from ants, can be readily reproduced, including tasks in which the animals had to switch between guidance cues, e.g. from visually driven path integration to odor based location of a nest entrance, or were blown off course by wind.

    The manuscript clearly requires that the first paper by the same group is read first, as many core concepts of the computational model are introduced in that paper. When viewed as such an extension (as intended by the 'Research Advances' article type), the paper adds valuable insights and stimulates thought and hypothesis development regarding concepts of multimodal integration. In light of the increasing amount of data on insect brain connectomics, hypotheses based on biologically inspired computational models are highly useful.

    While the authors refer to their computational model as 'biologically realistic', several features (e.g. a ring attractor circuit in the fan shaped body) are speculative and, to date, unconfirmed in any insect, despite the existence of the fruit fly connectome. This does not mean that the model is conceptually wrong, especially as it allows to faithfully reproduce complex behavioral data and shifting of activity 'bumps' across the width of the central complex (as is key for the model) is likely one of the principal functions of the fan-shaped body circuit. Yet the exact nature of the neural implementation might have to be adjusted once relevant data on ants becomes available. The model, as it stands, should hence be seen as 'biologically plausible" rather than 'realistic'.

    That said, the addition of the new aspects of the model shows how flexible the proposed circuit is for coordinated navigational control in insects, and, interestingly, highlights analogous concepts found in the basal ganglia of mammals - a thought-provoking parallel that is in line with ideas of deep homology between these distant brain regions.

  5. eLife

    Reviewer #2 (Public Review):

    In their previous manuscript, Sun et al. combined existing and hypothesized circuit motifs within the insect central complex (CX) to propose an integrated model for how the region might enable visual route following and homing. In their original framework, circuit motifs within the fan-shaped body allowed for appropriate context-switching. They now show how roughly the same motifs could also allow the model to (optimally) incorporate other sensory inputs, such as odor concentration gradients and wind direction cues, thereby enabling an insect to use the CX for additional behaviors in a context-dependent manner. The model's performance is evaluated in comparison to the behavior of larval and adult insect behavior (flies and ants, for example). This study represents a useful extension of the model's scope, but it would benefit from some additional computational exploration and explanation. As it stands, the figures and figure legends are not self-contained enough to be clearly understandable to the average reader. This new piece would also benefit from a greater focus on alternative models and alternative neural pathways that also subserve at least some of the additional navigational behaviors. The existence of direct olfactory-motor pathways is mentioned in Discussion for example, but deserves to be explored in Results as well. Otherwise, the significance of the authors' model reproducing Drosophila larval chemotaxis is not clear: note that larvae do not have CX circuits of the sort that the model proposes.

  6. eLife

    Reviewer #3 (Public Review):

    Sun et al. propose an excellent study on multi-sensory fusion in ant when the animal is confronted to both wind and odour source or when conflict exists between chemotaxis and path integration. The paper is very well written and the figures are very clearly designed. The list of references is complete. On my opinion, this paper might be considered as a companion paper of a previous paper published in eLife (Sun et al 2020) featuring a strong impact on the plausible strategy that can be used by ant to integrate various cues (odour, wind, proprioception) but a weaker impact (because already published in the first paper) on the neuronal model of the various structures involved in the central complex of the ant's brain: protocerebrum bridge (PB), fan-shape body (FB) and ellipsoid body (EB). An interesting copy and shift function already described in (Sun et al 2020, eLife) seems to be well suited to generate the appropriate motor commands of the heading in response to a difference between the current heading and the measured heading (sensory feedback control). To summarize, this copy and shift function tends to minimize the heading error by making ant turn left or right. It is worth noting that the simulated responses of the ant have been compared to the real data published in previous papers by others.

    I have the following main concerns about this work:

    - First, the steering function as regard of the shift and copy mechanism should be recalled and carefully explained (figure 2B of the previous paper published in eLife)

    - About figure 1C and the on-off response of the ant: authors argue that their model replicates faithfully the ant's response: however to be absolutely convinced by this statement, authors must take into account in their simulation the following parameters used by Alavarez-Salvada et al. : ground speed, angular velocity, curvature and turn probability. If the simulated and real responses are similar, we should observe an ON response consisting of upwind orientation coupled with faster and straighter trajectories, and an OFF response consisting of slower and more curved trajectories. It is not clearly the case in the current version of the paper due to a lack of thorough analysis between the parameters listed previously. I can not see more curved trajectories in the OFF response.

    - About the simulated behaviour shown in figure 2C: this is a very interesting and critical point because here a conflict is produced between two sensory modalities: path integration and chemotaxis. Authors must clarify how is this conflict processed/managed by the ring attractor? Is it due to changes in the dynamics of the measurements (odor and path integration) or due to changes in the ring attractor itself? By the way, I strongly encourage authors to provide temporal simulation of the ring attractor state: plot the input and ring's output signals at different time steps to see clearly a shift in the bump output.

    - About the angular resolution of the PB, FB and EB: how many different angular directions are coded? How many neurons are simulated in each structure? In MM section (equation 10 page 13) it is indicated that the angular resolution (shifting accuracy) has been improved by a factor 10 (from 45{degree sign} to 4.5{degree sign}) to achieve better performance. This point must be indicated and discussed in the main text because it is related to the accuracy of the heading measurement and thus to the behaviour of the simulated ant. How can the ant improve its heading accuracy despite a coarse resolution of central complex in the heading measurement?

  7. Version published to 10.1101/2021.08.18.456777v1 on bioRxiv