Inter-species comparison of the orientation algorithm directing larval chemotaxis in the genus Drosophila

Animals differ in their appearances and behaviors. While many genetic studies have addressed the origins of phenotypic differences between fly species, we are still lacking a quantitative assessment of the variability in the way different fly species behave. We tackled this question in one of the most robust behaviors displayed by Drosophila: chemotaxis. At the larval stage, Drosophila melanogaster navigate odor gradients by combining four sensorimotor routines in a multilayered algorithm: a modulation of the overall locomotor speed and turn rate; a bias in turning during down-gradient motion; a bias in turning toward the gradient; the local curl of trajectories toward the gradient (“weathervaning”). Using high-resolution tracking and behavioral quantification, we characterized the olfactory behavior of eight closely related species of the Drosophila group in response to 19 ecologically-relevant odors. Significant changes are observed in the receptive field of each species, which is consistent with the rapid evolution of the peripheral olfactory system. Our results reveal substantial inter-species variability in the algorithms directing larval chemotaxis. While the basic sensorimotor routines are shared, their parametric arrangements can vary dramatically across species. The present analysis sets the stage for deciphering the evolutionary relationships between the structure and function of neural circuits directing orientation behaviors in Drosophila.


154
To compare larval chemotaxis across species, we tracked the locomotor behavior of larvae in 155 controlled odor gradients. A set of 19 odorant molecules ( Figure 2E) was selected based on the 156 biosynthesis of the molecules in native or introduced plant species reported as breeding sites in was taken to avoid potential activation of other chemosensory receptors, such as gustatory behavior directed by airborne olfactory cues. A total of 152 species-odor combinations were tested ( Figure 2E). As a control, we characterized the behavior of each species to the 178 unstimulated (no-odor) condition.

179
The throughput of the screen was increased by testing the behavior of 15 larvae 180 simultaneously in a group assay featuring 2 odor sources ( Figure 2B). Under these 181 conditions, the group of larvae tended to split equally between the two sources, which limited In analogy to the similarity tree in the behavioral space ( Figure 3B), we constructed a 239 similarity tree in the physicochemical space of the odors ( Figure 3C). For this, we applied a  Figure 3C depicts the closeness between the odors primarily based on their molecular 248 weight. According to this representation, the smallest molecules (e.g., acetic acid and methyl 249 acetate) are furthest away from the largest molecules (e.g., octanoic acid and citral) in the 250 similarity tree.

251
As a trend, we observed that odors eliciting strong attraction were predominantly smaller 252 molecules with lower molecular weights, while odors eliciting weak attraction were 253 predominantly larger molecules with higher molecular weights. A notable exception to this 298 biarmipes that showed considerable inter-odor variability ( Supplementary Fig. 4B).

299
In addition to an overall change in speed, larvae could adjust their run speed relative to their  increase in the run speed during movements towards the HOP source (e.g., D. yakuba), other 304 species did not modulate the run speed at all (e.g., D. erecta). As will become clear in the rest of the analysis (Figure 6), a lack of modulation of the run speed in response to the local 306 gradient did not imply that a species was unable to detect changes in odor concentration.
corresponds to the ability of a larva to turn with a higher frequency during down-gradient 309 motion and to turn less frequently during up-gradient motion (Figure 5Bi). While the baseline 310 turn rate was unique to each species, resulting in the separation of species into turners and 311 non-turners ( Figure 4B), the sensory-dependent modulation of the turn rate was a feature 312 common to all species upon HOP stimulation ( Figure 5Bii). This result could be extended to 313 the majority of odors tested in the study (Figure 5Biii), but a counter-example is highlighted in Figure 7. Thus, when-to-turn is a primary sensorimotor routine that is conserved across all 315 species.

316
The ability of a larva to turn toward the odor gradient (where-to-turn-to) is a fundamental 317 sensorimotor routine underlying accurate chemotaxis (Figure 6Ai). When exposed to their HOP odor, each species displayed a higher proportion of turns oriented towards the odor odor-specific or species-specific: it is a sensorimotor routine conditioning strong chemotaxis 322 in all species.

323
Finally, we examined the contribution of weathervaning to orient in odor gradients ( Figure   324 6Bi). This navigational strategy is achieved through smooth reorientation resulting in the 325 continuous correction of the animal's heading to improve its alignment with the local odor 326 gradient. In Figure 6Bii, weathervaning is quantified as a modulation of the instantaneous 327 reorientation rate by the bearing with respect to the odor gradient. The S-shape of the 328 function implies that larvae tended to steer their runs toward the left side when the odor 329 gradient pointed to the left, and vice versa. We found that all species implemented result could be generalized to a majority of the tested odors ( Figure 6Biii). Nonetheless, we 332 noted that the strength of the weathervaning was species-dependent. D. erecta displayed 333 strong weathervaning for most odors whereas the bias implemented by D. biarmipes tended 334 to be weak.
Since a given odor did not produce the same level of attraction in each species ( Figure 2E),

336
we further examined chemotaxis elicited by the set of HOP odors, which produced the most 337 attractive behaviors in each species. Therefore, the comparative analysis based on the HOP 338 behavior provided information about the characteristic sensorimotor strategies implemented 339 by each species for effective chemotaxis. In the rest of the analysis, we extended the HOP 340 comparison to special case studies that enabled an assessment of the conservation of differences in the responses elicited by the same odor across species (Figures 8 and 9).

343
While the HOP odor of each species elicited a unique modulation of the sensorimotor 344 routines, this modulation varied within a species. Different odors could elicit different 345 parametric arrangements of the sensorimotor routines, resulting in the same overall attraction 346 level. This point is illustrated in Figure 7A by comparing the olfactory responses of D. erecta.

347
While D. erecta larvae were strongly attracted to both ethyl acetate and E-2-hexenal (a high 348 preference index marked as "+++"), the baseline run speed was higher for E-2-hexenal 349 compared to ethyl acetate ( Figure 7B). Consistent with the results of the HOP odor (2-350 heptanone) of D. erecta, ethyl acetate and E-2-hexenal did not elicit orientation-dependent 351 run speed modulation ( Figure 7B). While ethyl acetate did not elicit orientation-dependent turn 352 rate modulation (when-to-turn), E-2-hexenal did ( Figure 7C). Finally, both odors induced a

360
This behavioral comparison between two strongly attractive odors suggests that losing the 361 when-to-turn routine (control of timing of stop-turn) does not necessarily abolish chemotaxis.

362
In Figures 5 and 6, we established that each HOP odor elicited a species-specific 363 sensorimotor strategy that modulated the baseline run speed and turn rate. Despite these 364 differences, three sensorimotor routines that control turning -when-to-turn, where-to-turn-to 365 and weathervaning -were universally implemented by all tested species when stimulated by 366 the HOP odors. In Figure 8, we asked whether this conclusion held true for other odors that

391
While many odors elicited attraction levels that were relatively conserved across species (e.g.,

392
E-2-hexenal, Figure 8), some odors elicited heterogeneous behavioral responses. This inter-393 species variability is illustrated in Figure 9 by the odor pair prenyl acetate and methyl benzoate.

394
We found a remarkable qualitative change in the attraction level elicited by prenyl acetate in D.

395
sechellia (strong attraction) and D. yakuba (virtually no response) ( Figure 9A). By contrast, methyl benzoate was strongly attractive in D. yakuba, but weakly attractive in D. sechellia. The opposed preferences for prenyl acetate and methyl benzoate did not, however, indicate a switch 398 in the underlying sensorimotor controls.
In response to prenyl acetate, D. sechellia modulated neither the run speed nor the when-toattraction to prenyl acetate emerged from a marked use of the where-to-turn-to and                  belongs to the yakuba species complex). We observe a similar split in the two Asian species 576 since D. suzukii is a turner while D. biarmipes is a non-turner. Therefore, being a turner or a 577 non-turner is not a behavioral property that is shared among species that are closely-related 578 phylogenetically.

579
Next, we explored in more detail the modulatory effects of HOP stimulation on the run speed 580 and the turn rate independently. Stimulation by the HOP odor produced either an increase, a 581 decrease or no change in run speed compared to the no-odor stimulation condition ( Figure 4C).

582
As outlined in Figure 10A, three species demonstrated an acceleration: D. melanogaster, D.

615
Turn rate is another locomotor response that is strongly modulated by the animal's bearing in an 616 odor gradient. Larvae tend to turn more during movement away than toward the gradient. This 617 modulation forms the basis of the when-to-turn sensorimotor control (Figure 5Bii). Remarkably,

618
this sensorimotor routine appears to be independent of the net effect that the odor has on the 619 overall turn rate. For some species, such as D. sechellia, HOP stimulation produces an increase 620 in turn rate, whereas it yields either no change or an overall decrease in turn rate in other 621 species such as D. simulans. Even the species classified as non-turners ( Figure 4B) modulate 622 their turn rate as a function of the bearing. In contrast with the species-specific modulation of the 623 run speed, the when-to-turn modulation of the turn rate appears to be universal ( Figure 10B).

624
The conservation of this sensorimotor routine across species of the genus Drosophila is

659
A striking correlation is found between the inability of larvae to chemotax toward specific odors, 660 butyl butyrate, citral and octanoic acid ( Figure 2E) and the absence of turning bias -where-to-essential to ensure that the navigation algorithm produces strong odor attraction. Specieswith the inability of D. yakuba to perform robust when-to-turn (Figure 5Biii) and where-to-turn-to 668 modulation in response to this odor (Figure 6Aiii). We speculate that peripheral changes in the

725
After having examined how the same odor can produce strong attraction through different few odors that produce active aversion, as judged from the larger distance to the source that produces an overall increase in run speed indicates that this odor is perceived by D. yakuba.      qualitatively similar, the algorithm implemented by D. melanogaster is not a "one size fits all" 763 solution adopted by all its closely related species. At the most basic level, different species 764 display a different basal turn rate and locomotor speed ( Figure 4C). Moreover, substantial 765 differences are observed in the way that a given species achieves chemotaxis in response to distinct odors (Figure 7). The degree of algorithmic variability observed across strains of the 767 same species remains to be determined in future work.

768
Not all the basic sensorimotor routines available to the larva ( Figure 1B

935
The when-to-turn metric was also quantified by calculating the proportion of turns that occurred

972
We z-score normalized the descriptors before performing the PCA.

973
Hierarchical clustering: All dendrograms were built by performing a hierarchical clustering 974 analysis on the behavioral data. Using the pdist Matlab function the similarity of dissimilarity 975 between every pair of objects in our data set was measured by calculating the Euclidean 976 distance between objects. Using the linkage Matlab function the objects were then grouped into 92% were found for all conditions. The statistical-power analysis was conducted by using the crawling at the level of organism, segment, and somatic body wall musculature." Journal of proteins OBP57d and OBP57e affect taste perception and host-plant preference in Drosophila   one species across all odors. The level of attraction is quantified with the preference index (PI).

1229
The green squares indicate the odors forming the high-olfactory-preference (HOP) set for each              area that corresponds to 1.96 SEM (95% confidence interval), and a colored interval that 1404 corresponds to 1 SD. In the graph of the speed and reorientation rate, the mean is surrounded by

1411
The color-coding does not convey any information about the statistical significancy of differences 1412 between the no-odor and odor conditions.