Computations underlying Drosophila photo-taxis, odor-taxis, and multi-sensory integration

6 figures, 1 video and 2 tables

Figures

Identifying computations underlying the decision to initiate a turn.

(A) Computation: on the basis of sensory input (light or odor in this work) the larva decides whether or not to end a run and begin a turn. (B) LNP model of the computation: Sensory input is …

https://doi.org/10.7554/eLife.06229.003
Figure 2 with 1 supplement
Unimodal reverse-correlation experiments.

Top row, Berlin wild-type larvae were stimulated with blue (λpeak = 448 nm; max intensity = 74 μW/cm2) light. All other rows, larvae of indicated genotype were stimulated with red light (λpeak = 655 …

https://doi.org/10.7554/eLife.06229.004
Figure 2—figure supplement 1
LNP model parameters are stable for duration of 20 min experiments.

Experiments of Figure 2 analyzed separately using data only from the first 10 min of experiment (teal) or only from the second 10 min of experiment (purple) or from entire 20 min data set (black). …

https://doi.org/10.7554/eLife.06229.005
Figure 3 with 3 supplements
Multi-modal reverse-correlation experiments suggest attractive odor and light signals are combined linearly and early.

Or42a>CsChrimson larvae were presented with independently varying Brownian light intensities. Reverse-correlation analysis was carried out as in Figure 1. (A) TTA. Average change in red (fictive …

https://doi.org/10.7554/eLife.06229.006
Figure 3—figure supplement 1
Graphical explanation of the independent pathways model.
https://doi.org/10.7554/eLife.06229.007
Figure 3—figure supplement 2
Graphical explanation of the early linear combination model.
https://doi.org/10.7554/eLife.06229.008
Figure 3—figure supplement 3
Visual and fictive olfactory stimuli do not cross-talk.

Larvae were presented with same red and blue Brownian light stimuli as in Figure 3. TTA of red and blue stimuli are shown on same axes as in Figure 3A. (A) Reproduced from Figure 3A: Larvae …

https://doi.org/10.7554/eLife.06229.009
Multi-modal step responses support early linear combination of odor and light signals.

Turn rates vs time for Or42a>CsChrimson larvae responding to coordinated increases and decreases of red and blue light. All steps occur at t = 0. Left column: no change in fictive odor, center …

https://doi.org/10.7554/eLife.06229.010
All navigational decisions appear to be based on a single linear combination of odor and light inputs.

(AC) Reverse correlation in rotated coordinate system. μ,ν are linear combinations of the raw input stimuli according to the same scaling as used to combine filtered odor and light signals in Figure…

https://doi.org/10.7554/eLife.06229.011
Probing for attentional shifts during multi-modal noise experiments.

(A) Turn-triggered ensemble (duplicated from Figure 3B), with quadrants highlighted. Color scale the same as in 3B. Quadrants I–IV indicate which stimulus or stimuli likely provoked the larva to …

https://doi.org/10.7554/eLife.06229.012

Videos

Video 1
Calculating the turn-triggered average.

Left panel: annotated video image of individual larva. Thin white line: larva's path (past and future). Gold dots: markers along midline of animal, used to determine posture. Upper left corner: time …

https://doi.org/10.7554/eLife.06229.015

Tables

Table 1

Numbers of experiments, animals, turns, and head sweeps for all figures

https://doi.org/10.7554/eLife.06229.013
Genotype#expts#animalshours#turnsrms turn size#large turns#small turns#accepted head sweeps#rejected head sweeps
Uni-modal reverse-correlation experiments (Figure 2A,B,D,E, Figure 2—figure supplement 1)
 Berlin615052.6659476.42462413241392455
 Canton-S7334117.7882466.03086573855703254
 Or42a>CsChrimson518061.1697168.92531444041222849
 Or42b>CsChrimson624664.4956573.33480608562153350
 Gr21a>CsChrimson522754.2876075.13392536854243336
Uni-modal step experiments (Figure 2C)
 Berlin49534.73674
 Or42a>CsChrimson210736.63905
 Or42b>CsChrimson29921.52599
 Gr21a>CsChrimson211122.12384
Multi-modal reverse-correlation experiments (Figure 3, 5, 6, Figure 3—figure supplement 3)
 Or42a>CsChrimson1260813621,07566.6722513,85012,7958280
  quadrant I10,36362164147
  quadrant II330120861215
  quadrant III16841088596
  quadrant IV572734052322
 GMR-Hid, Or42a>CsChrimson312128.94842
 Canton-S316639.93412
Multi-modal step experiments (Figure 4)
 Or42a>CsChrimson5250507859
  1. #expts: Number of 20 min experiments. For reverse-correlation experiments, each experiment presented a different stimulus sequence with the same statistical properties; for step experiments, the same stimulus pattern was presented in each experiment.

  2. #animals: Approximate number of animals, taken by finding the maximum number of animals tracked in a 30-s window during each experiment.

  3. #hours: total observation time in units of larva-hours. Observing 3 larvae for 20 min each would equal 1 larva-hour.

  4. #turns: total number of turns observed and used in analysis.

  5. rms turn size: root mean square turn size in degrees (defined as angular difference in run heading immediately before and after a turn) for the set of experiments.

  6. #large/small turns: number of turns with angular changes larger/smaller than the rms turn size.

  7. #accepted head sweeps: number of times the first head sweep of a turn was accepted, ending in a new run.

  8. #rejected head sweeps: number of times the first head sweep of a turn was rejected, leading to another head sweep.

Table 2

Kullback-Leibler divergences for Figure 3

https://doi.org/10.7554/eLife.06229.014
KL divergencek-NNmodel data as normally distributedSzegö-PSD method
Figure 3B: DKL((P(xo,xl|turn)||P(xo,xL))0.3510.325
Figure 3B: DKL((P(xo|turn)||P(xo))0.2360.2230.235
Figure 3B: DKL((P(xL|turn)||P(xL))0.1030.1000.104
Figure 3B: DKL((P(u|turn)||P(u))0.3340.3240.333
Figure 3B: DKL((P(v|turn)||P(v))0.0060.00020.003
Figure 3C: DKL(data||model)0.0620.035
Figure 3D: DKL(data||model)0.0300.007
  1. KL divergence: the divergence to be calculated. k-NN: divergence calculated using the k-nearest neighbors algorithm. This value is displayed in Figure 3. model data as normal distributed: the distributions are modeled as Gaussians, whose divergence is calculated analytically. Szegö-PSD method: divergence between 1D distributions calculated by an alternate method.

Download links