Drosophila chemotaxis

Abstract

Chemotaxis, the ability to direct movements according to chemical cues in the environment, is important for the survival of most organisms. In our original article, we combined a quantitative behavioral assay with genetic manipulations to dissect the neural substrate for chemotaxis. In this Extra View article, we offer a more chronological narration of the findings leading to our key conclusion that aversion engages specific motor-related circuits and kinematics. We speculate on the separation and crosstalk between aversion and attraction circuits in the brain and the ventral nerve cord, and the implication for valence encoding in the olfactory system.

Olfaction, the sense of smell, is inherently judgmental. It often implies negative intuitions in natural languages, such as “this article smells fishy.” On the other hand, the hedonic experience of food consumption arises to a large extent from our olfaction, and the perfume industry is built upon our liking of fragrances. In mice, olfactory sensory neurons projecting to the dorsal olfactory bulb are necessary for the innate avoidance of the odorants from spoiled food and predator urine,¹ while amygdala, a brain area involved in emotional responses, receives input directly and preferentially from that area.² Drosophila melanogaster also robustly pursue food-related odorants,^3-6 and avoid many volatiles that are either highly concentrated^4,6 or indicative of danger.^7,8 In spite of the evolutionary distance between insects and mammals, hardwired neural circuits also appear to underlie innate olfactory behavior in flies. Specifically, the axon terminals of the projection neurons (PNs, which are secondary neurons in the olfactory circuits) segregate in the lateral horn based on whether they relay information about food or sex.⁹

There are about 50 different odorant receptors in flies and hundreds in mammals. How are such high-dimensional inputs reduced to the approximately binary outputs of attraction or aversion? If stereotypic circuits perform such computations, we should be able to identify the necessary circuit components by genetically targeted silencing of specific neurons and observing the consequences in the chemotactic behavior of the animals. Once we have a genetic handle on these components, we will be able to characterize their computational functions in detail. That was the premise at the initiation of the project.

To observe olfactory chemotaxis, we adopted a star-shaped arena where the spatial locations of flies are recorded while air or odorants are delivered to its quadrants.³ Control flies are robustly repelled by 10% acetic acid.¹⁰ To acutely block synaptic transmission, we screened flies expressing UAS-shibire^ts1 (UAS-shi^ts1) with different GAL4s (GAL4 > shi^ts1 for short) generated in Clandinin Lab.¹¹ We selected over 100 lines for their relative sparseness and their lack of expression in the olfactory receptor neurons (ORNs) or PNs. The 4 candidates with the greatest effects on the Preference Index (PI) were repeated, and are all true positives (Fig. 1A and B, also see Figs. 3B and 6A in the original paper¹⁰). In fact one of the hits, 658 > shi^ts1, not only abolishes aversion, but also induces attraction (Fig. 1A). To our knowledge, this was the first report of such a phenotype, which we will reflect upon later.

Figure 1. (A) 658 > shi^ts1 turns the aversion to 10% acetic acid to attraction, and this effect is suppressed by tsh-GAL80. (B) 756 > shi^ts1 attenuates aversion to 10% acetic acid, and this attenuation is not affected by tsh-GAL80. (C) 441 > shi^ts1 attenuates aversion to a lethal concentration of ethyl butyrate. (D) After pre-incubation at 29 °C for 6 d, 441 > shi^ts1 still abolishes aversion compared with control flies. (A–D) Bars represent mean PIs of multiple independent runs; error bars represent s.e.m.; the definition of PI is the same as in the original paper;¹⁰ the statistical significances reflect t-test with Holm–Bonferroni post hoc correction. (E) Distribution of the differences between the speeds after (0.17s–0.4s) and before [(-0.4s) – (-0.2s)] aversive turns. (F) Distribution of the differences between the speeds after and before attractive turns.

We visualized the expression pattern of these GAL4s using a membrane-tagged GFP, and realized that all 4 candidates label neurons relatively extensively, despite our intentional enrichment of sparse lines. As a first step of narrowing down the causal neurons, we tested the role of the brain vs. the ventral nerve cord (VNC) by suppressing GAL4 expression in the VNC with tsh-GAL80 (GAL4 NOT tsh > shi^ts1 for short). tsh-GAL80 significantly rescues the loss of aversion in 3 out of the 4 cases (Figs. 1A-B, also see Figs. 5C and 6A in the original paper¹⁰), indicating that the expression of these GAL4s in the VNC contributes to the loss of aversion.

Conceptually flies can distribute motor control between the brain and the VNC in 2 ways. One possibility is that descending projections from the brain instruct basic elements of locomotion. For example, the signal from a specific descending neuron may always direct the animal to turn right, no matter what sensory task triggers the turn. Alternatively, descending signals can be task-specific. For example, 2 different navigational tasks may recruit 2 distinct VNC circuits to produce different sequences of motor events, although both of them might involve right turns. The former seems to be the case in zebra fish,¹² yet our data favor the latter in flies, as suggested by a comparison between the 658 > shi^ts1 and 658 NOT tsh > shi^ts1 flies (Fig. 1A). If the 658+tsh+ neurons were necessary for general navigation, compared with 658 NOT tsh > shi^ts1, the additional inactivation of these neurons in 658 > shi^ts1 would lead to a passive loss of aversion rather than the gain of attraction that we observed. To further test this unexpected revelation, we examined the attractive behavior in the other 2 lines, 441 > shi^ts1 and 918 > shi^ts1, whose aversion phenotypes also have a VNC origin. Despite the great reduction of aversion by these manipulations, the effects on attraction were much subtler, and the enrichment of experiment flies in the vinegar quadrant was identical to the controls as judged by PI.¹⁰ Thus, the seemingly symmetric tasks of aversion and attraction require different neurons in the VNC.

The net effect of 10% acetic acid is aversion, yet the “valence reversion” between 658 NOT tsh > shi^ts1 flies and 658 > shi^ts1 flies implies that the olfactory cue sends both attractive and aversive signals to the VNC, where the former is normally suppressed by the latter. This is another counterintuitive finding. If a unitary decision were computed in the brain before being forwarded to the VNC, it would be more economic because of the reduced amount of information passing through the cervical connectives. Although we have to caution that the broad inactivation by 658 > shi^ts1 might confound the interpretation, since the attraction signal might be an artifact originating from some substantial disturbance of the nervous system.

Line 658 is informative, but it expresses so broadly that identifying the causal neurons becomes very difficult, if possible at all. Thus, we tried to refine the expression pattern of a sparser line, 441, with GAL80s to suppress GAL4 in different neuronal populations. In particular, we noted that tsh-GAL80 does not fully rescue aversion in 441 > shi^ts flies, which could be due to either the incomplete suppression of 441 by tsh-GAL80 in VNC neurons, or due to 441 expression in the brain. Since 441 prominently labels the ellipsoid body (EB) in the brain, and the role of EB has been implicated in other aversion assays,¹³ we sought to test its contribution. We found that EB-GAL80s partially rescue the loss of aversion in 441 > shi^ts1 flies as well.¹⁰ Also taking into account that EB-GAL4s, despite using the same enhancers as the EB-GAL80s, do not affect aversion when driving shi^ts1,¹⁰ we proposed that at least 2 redundant circuits mediate aversive chemotaxis, and they are fortuitously disrupted by 441 expressions in the EB and in the VNC, respectively. Although our screen has yet to reach saturation, the redundancy is also consistent with the lack of hits targeting single populations of neurons.

In theory, the intact capacity for attraction in 441 > shi^ts1 flies can be adapted for aversion, as long as the sign of the underlying algorithm is reversed. Such an adaptation was never observed. The 441 > shi^ts1 flies still cannot avoid ethyl butyrate when the odorant is at a lethal concentration (Fig. 1C), demonstrating that an emergency does not suffice to induce adaptation. Raising 441 > shi^ts1 flies at a restrictive temperature for a week before testing did not ameliorate the condition either (Fig. 1D). Interestingly, compared with the loss of aversion in acute inactivation, the long-term blocking of 441+ neurons made flies attracted to acetic acid. It implies that the nervous system failed to compensate for chronic dysfunction in aversion; it also serves as a second example of “valence reversion,” where the inactivation is not too broad and the interpretation less subject to the caveat mentioned for 658 > shi^ts1. Taken together, the 441 > shi^ts1 data highlighted the redundancy and rigidity of the neural circuits underlying robust aversive chemotaxis.

Given the identification of motor-related areas specific for aversion, we wondered whether the circuit specificity is reflected in behavioral kinematics. Our paradigm allowed us to quantify the locomotion patterns of free-walking flies, and we demonstrated for the first time the modulation of turn initiation and direction in both aversive and attractive chemotaxis.¹⁰ For most of the parameters examined, aversion and attraction appear to be qualitatively symmetric, and the only aversion-specific features are the increase of speed after turns and the concomitant decrease of angular speed, which we managed to mimic by artificially activating 441+ neurons with optogenetics.¹⁰ It is worth noting that, despite the magnitude and statistical significance of the post-turn speed increase in aversion, the variation is large. In aversion, the differences between the speeds after and before each turn distribute broadly from negative to positive values (Fig. 1E); this distribution also overlaps considerably with that of attraction (Fig. 1F). Changes in angular speed show similar overlaps between aversive and attractive turns (data not shown). In studies driven by quantitative behavioral analysis, trajectories of flies are typically segmented and assigned to different states based on the classification of locomotion parameters,¹⁴ yet such an approach, if not guided by genetic perturbations, would most likely miss the subtle distinction between aversive and attractive turns. Recently, methods to quantify the gaits of flies have been established,^15,16 and it would be interesting to examine the exact motor defects caused by our manipulations.

Our unexpected discoveries placed the project in the context of the output end of the sensory-motor transformation, yet, as this article approaches its end, we would like to revisit the question raised in the beginning about the computing of positive and negative values in olfaction. Conceptually, the valence can be encoded in different manners. One possibility is that only 1 population of neurons encodes both negative and positive values on a single scale; alternatively, 2 antagonizing populations simultaneously read aversive and attractive signals from the olfactory circuit, and the net valence is decided by specific rules weighing the contributions of these 2 circuits. Our data favor the latter.

How can 1 input pattern be read in 2 ways for opposite values? We postulate that the separation takes place in the temporal domain. Specifically, the “aversion circuit” reads the transient rising of PN firing in response to a new odorant, and the “attraction circuit” reads the decayed and steady firing during the continued exposure to the odorant. This hypothesis takes into account the following facts. (1) PN responses preferentially peak at the rising phase of ORN activity.¹⁷ (2) An odorant is more likely to be aversive as its concentration increases.^4,6 (3) The peak firing of PNs is less subject to normalization than the steady firing,¹⁸ so the former carries more information about the concentration whereas the latter is more about the identity. This hypothesis is also consistent with the “valence reversion” we observed. Because PN firing should peak when a fly enters the odorant quadrant from an air quadrant, if the animal misses that time window to complete an aversive turn, the PN firing will reach steady-state and the aversiveness will no longer be sensed.

Since most genetically encoded manipulators globally change the activity of neurons, our hypothesis about the causal role of a specific period of neural activity is difficult to test. One possible strategy would be to express light-gated inactivators with 441, and silence these neurons only when the flies are entering the odorant field.¹⁹ But such a strategy would require the knowledge of the timing between the animal’s encounter of the odorant and the responses in its PNs and downstream circuits. The currently available genetic toolbox is sufficient for probing the layout of neural circuits, but a deeper understanding of the nervous system still awaits a paradigm of manipulation in unrestrained animals that matches the temporal resolution of functional imaging or electrophysiology.

Gao XJ, Potter CJ, Gohl DM, Silies M, Katsov AY, Clandinin TR, Luo L. Specific kinematics and motor-related neurons for aversive chemotaxis in Drosophila. Curr Biol. 2013;23:1163–72. doi: 10.1016/j.cub.2013.05.008.

10.4161/fly.26685