History-dependent spiking facilitates efficient encoding of polarization angles in neurons of the central complex

Reviewer #1 (Public review):

Summary:

The authors of this valuable study use linearly polarized UV light rotating at different angular velocities to stimulate photoreceptors in bumblebees and study the response of TL3 neurons to polarized light. Previous work has typically used a single constant rotation velocity of the polarized light, while the authors of this study explore a range of constant rotational velocities spanning from 30deg/s to 1920deg/s. The authors also use linearly polarized UV light rotating at continuously varying velocities following the angular velocity of the head of a flying bumblebee.

Strengths:

The authors investigate the neuronal responses of TL3 neurons to a variety of rotational velocities. This approach has the potential to reveal the neuronal response to dynamically changing stimuli experienced by the animal as it moves around its environment.

The authors make good use of physiology and modeling to validate their hypotheses and findings. If done right, this line of investigation has the potential to provide a very useful methodology for utilizing more complex stimuli in studies of the visual pathway and central complex than traditionally.

Weaknesses:

The attempt of the authors to use more naturalistic stimuli than previous studies is very important, but the stimulus they use, i.e. linearly polarized UV light projected on the whole dorsal rim of the animal's eyes, is very different from the circular pattern of UV light polarization coming through the sky. In particular, as a bumblebee turns under the sky, the light projected on each ommatidium of the dorsal rim area will not smoothly change like the rotating linearly polarized light used in the experiments. The authors need to discuss this and other limitations of their study.

The authors should also commend the light intensity confound common in polarized light setups as discussed by Reinhard Wolf et al, J. Comp. Physiol. 1980 and in the thesis of Peter Weir, California Institute of Technology, 2013. It is unclear whether the authors performed measurements to quantify the intensity pattern and if they took measures to compensate and make the polarized light intensity uniform.

The authors show that the neuronal responses of TL3 neurons depend on the recent history of the polarized light stimulus. They use as evidence, the different neuronal firing rates measured when arriving at the same polarization stimulus by following two different preceding stimulus sequences. It would have been worthwhile to investigate to what extent the difference in neuronal response is due to the history alone and to what extent it is due to spike timing stochasticity inherent in the neurons. According to the raster plots in Figure 2F, there is substantial stochasticity in the timing of the action potential firing events.

The authors appear to base their delay calculations and analysis on the response of one single neuron (Figures 2 and 3) even though they have recorded the responses of several TL3 neurons. There is no reason for the authors not to use all neuron recordings in their calculations and analysis.

Another concern is that while the authors make good use of modeling, like any model, the presented models only partially explain the observed phenomena. However, a discussion about the limitations of their model needs to be provided. Actually, observing the discrepancies between the model's output and the intracellular recordings reveals what the model is missing. That is, careful consideration of the discrepancies would have led the authors to try adding some noise in their model, which would partially resolve the differences observed at the lower rotational speeds (see stars deviating from the fitted line in Figure 2A) and to consider that introducing an asymmetry between the post-stimulus inhibition and excitation time constants could result in a model not deviating as much at the higher rotation velocities during counter-clockwise rotation of the polarized light (see stars deviating from the fitted line in Figure 2A).

In the end, the authors use the observation that during saccades, the average activity in their model-with-history increases to claim that when the animal does not turn, it uses less neuronal activity and energy. This is not a convincing line of reasoning. To make a claim about energy efficiency, the authors must instead compare their model with alternatives and show that the neuronal activity of their model during straight flight is indeed lower than those alternative models. Note that such a comparison would be meaningful only if the alternative models compared against capture physiology equally well in all other respects. However, the evident deviations of the presented model from the physiology measurements and the short duration of the test stimulus used would make any such claims difficult to substantiate.

Finally, for most experiments, the models are stimulated with a single short yaw sequence lasting a few seconds to measure responses. Given the dependence of the model on history, using such a small sample, we cannot see how generalizable the observations are. The authors need to show that the same effect is produced using multiple different trajectories.

Reviewer #2 (Public review):

Summary:

The compass network is a higher-order circuit in insects that integrates sensory cues, like the angle of polarized light, with self-motion information to estimate the animal's angular position in space. This paper by Rother et al. uses share electrode recordings to measure intracellular voltage activity from individual compass neurons while polarization patterns are presented to the bee. They present patterns that rotate with variable speed or simulate the sensory experience created by a flight trajectory. The authors discover that at low rotational speeds, TL neuron responses diverge from the tuning expected from a systematic synaptic delay, suggesting that recent experience (history) impacts TL responses. A population model of 180 TL neurons is then used to argue that having cells that are impacted by spiking history could be advantageous for estimating heading. The model activity showed an anticipation of polarization angle for rapid turns that followed prolonged straight flights or turns in the opposite direction. The model also had reduced spiking activity during translational straight flight.

Strengths:

One strength of this paper is that it focuses on a question that is underexplored in the field: How does the compass network handle the processing delay caused by multi-synaptic relay from the DRA to the sensory input neurons (TL) to the compass network why the insect is turning rapidly and thus sampling distinct polarization angles in rapid succession? Another strength is the fact that they were able to present neurons with both simulated naturalistic polarization patterns that could occur during flight and synthetic stimuli with a range of rotational velocities. This provides an important data set where these responses can be compared. Another strength is the exploration of how adding a history term to a model of a population of TL neurons can lead to the population coding of polarization angle to vary in how delayed it is from changes to the sensory stimuli. They find that angular coding is more anticipatory (shorter delay) following prolonged periods of fixating a single angle, such as what occurs during translation movement, or following turns in the opposite direction of the current turn.

Weaknesses:

A challenge for this experimental approach is the relatively low power for data sets in some of the experimental conditions. Low throughput is expected for this experimental approach, as intracellular recordings are a challenging and time-consuming method. A weakness of the manuscript in its current form is that the data from all cells that were able to be recorded is not always presented or quantified. For example, only a single neuron example is used to show the impact of history on preferred polarization and how this tuning varied with rotation velocity. This is also true for the claim that TL3 neurons exhibit post-inhibitory excitation and post-excitatory inhibition. Another concern is regarding the use of the term "spiking-history" as potentially confusing to readers who might assume this process is cell intrinsic. The authors presented data shows evidence of an effect of stimulus history on the responses of the neurons. However as the authors describe in the discussion this current data set does not distinguish between an effect that occurs in the recorded neurons (e.g. an effect of intrinsic excitability) vs adaptation elsewhere in the circuit or DRA photoreceptors. A final challenge for this approach, shared with other studies that measure neural responses from an insect fixed in place, is that it assumes that these TL neurons are purely sensory and that their response properties (or those upstream of them) do not change when the bee performs a motor action or maneuver. This caveat should be considered when interpreting these data, however these data still represent novel information and important progress in exploring this question.

Reviewer #3 (Public review):

This manuscript reports the temporal history dependence of central complex TL/ring neuron spiking activity to polarized light patterns. Using sharp recording in tethered bumblebees with synthetic and natural visual stimulation, the authors nicely measured activities to rotating polarized UV light, and made the interesting finding that spiking activity depends on not just current stimulus but also recent activity.

(1) History dependence has been reported before in ring neurons in Drosophila (Sun et al., Nature Neuroscience, 2017; Shiozaki et al., Nature Neuroscience, 2017). While there are differences in the nature of the visual stimulation used, the basic phenomenology of temporal history dependence bears some resemblance. Where are the differences in the physiological properties of ring/TL neurons between different insect species in relevance to history dependence? What are the structural similarities and differences in the circuits that may help to explain history dependence? Just to name a few. To gain further insight into this question, the manuscript may benefit from putting the findings here into context.

(2) Figure 3b serves as a critical evidence for history-dependence. However, it is unclear from this data if this is history dependence, or other physiological processes such as OFF response to sensory stimulation, or sensory adaptation. One way to test this is to examine whether such an effect can be detected after a delay period. For example, history dependence in fly ring neurons is mediated by delay period activity present for several seconds. This can be easily tested here as well.

(3) The properties of the history dependence can be better characterized to help understand its nature. What are the statistical characteristics of post-stimulus inhibition to preferred AoP and post-stimulus excitation to anti-preferred AoP? What are the temporal dynamics of such an effect, e.g., how long does it take to return to baseline? Are the differences in these properties recorded across the TL neuron population? Is it possible to categorize these TL neurons based on these properties and morphology? These properties are important to under the physiological basis of such effect. The authors only presented two traces in Figure 3b, beautiful example traces, but without any further population data and statistical analysis.

(4) A major point of the manuscript is energy efficiency via reduction of firing rate. However, the only evidence comes from simulation, and it seems to be a weak effect of 0.5 APs/s.

(5) Another major point of the manuscript is "increases sensitivity for course deviations during straight flight". However, this again is supported by simulation only. To validate these claims, empirical support of behavioral experiments is highly desired. Otherwise, it is recommended to minimize emphasizing such behavioral predictions.

(6) A substantial portion of the text emphasizes the importance of natural stimulation. While natural stimulation is indeed a desirable experimental approach, it is unclear if natural stimulation is exploited to its full in this manuscript. History dependence can be explored with synthetic stimulation.

(7) A phenomenological model was used to account for the history effect, by assuming a linear integration process and a linear history effect. However, such an assumption is not adequately backed up by rigorous statistical analysis of experiment data or at least proper conceptual discussion.

(8) Population responses, as in Figure 4, are based on strong assumptions of neuronal properties without clear experimental support, thus seeming to be quite a stretch.

(9) There are interesting observations in simulation results from Figure 5; it would be nice to experimentally test at least some of these ideas.

(10) "anticipate future head directions" seems to be quite a stretch to me without mechanistic explanations.

(11) The visual stimulation design used can be improved and expanded. The synthetic stimulation used in Figure 1c follows a stereotyped order, according to angular velocities. As the focus of the manuscript is to probe the history effect and to test again the findings made with this stimulation, randomized stimulation should ideally be examined.

(12) State dependence was observed in ring neurons in Drosophila (Sun et al., Nature Neuroscience, 2017) which might be related to ongoing neural activity and history dependence. While I realize that the animal is tethered, I was wondering if there was any signature of neural activity state dependence observed in this study.