Adaptation of Drosophila larva foraging in response to changes in food resources

  1. Marina E Wosniack
  2. Dylan Festa
  3. Nan Hu
  4. Julijana Gjorgjieva  Is a corresponding author
  5. Jimena Berni  Is a corresponding author
  1. Computation in Neural Circuits Group, Max Planck Institute for Brain Research, Germany
  2. School of Life Sciences, Technical University of Munich, Germany
  3. Department of Zoology, University of Cambridge, United Kingdom
  4. Brighton and Sussex Medical School,, University of Sussex, United Kingdom

Abstract

All animals face the challenge of finding nutritious resources in a changing environment. To maximize lifetime fitness, the exploratory behavior has to be flexible, but which behavioral elements adapt and what triggers those changes remain elusive. Using experiments and modeling, we characterized extensively how Drosophila larvae foraging adapts to different food quality and distribution and how the foraging genetic background influences this adaptation. Our work shows that different food properties modulated specific motor programs. Food quality controls the traveled distance by modulating crawling speed and frequency of pauses and turns. Food distribution, and in particular the food–no food interface, controls turning behavior, stimulating turns toward the food when reaching the patch border and increasing the proportion of time spent within patches of food. Finally, the polymorphism in the foraging gene (rover–sitter) of the larvae adjusts the magnitude of the behavioral response to different food conditions. This study defines several levels of control of foraging and provides the basis for the systematic identification of the neuronal circuits and mechanisms controlling each behavioral response.

Editor's evaluation

This paper contributes to the growing body of literature that investigates foraging in complex sensory landscapes. It is therefore of interest to both neuroscientists and ecologists. Using behavioral analysis and computational modeling, the authors characterize different behavioral components of the foraging strategy adopted by the Drosophila larva as a function of food quality and food distribution. Altogether, this works sets the stage for investigating the genetic and neural-circuit bases underlying the control of foraging behavior.

https://doi.org/10.7554/eLife.75826.sa0

Introduction

Most moving organisms need to explore their surroundings to increase their chances of finding nutritious resources. This is a challenging task in natural environments, where food quality varies both in time (e.g., seasonal effects) and space (e.g., patchy distribution). Therefore, the exploratory behavior of animals has to be flexible and adapt to environmental challenges. From the perspective of evolutionary ecology, foraging strategies have evolved to maximize lifetime fitness under distinct constraints (Stephens and Krebs, 1987) including the concentration of food inside patches (Charnov, 1976). Accordingly, several hypotheses and models have been developed to predict the optimal foraging strategy that an animal will adopt (Stephens and Charnov, 1982; Viswanathan et al., 2011). These models postulate that animals will use different strategies depending on the distribution of the resources. In environments where resources are abundant, animals will search and exploit them performing short movements in random directions, in patterns well approximated by Brownian random walks. When resources are sparse, and foragers have incomplete knowledge about their location, a more diffusive strategy is needed, with an alternation between short- and long-range movements, which can be modeled as a Lévy random walk. Analysis of animal movements in the wild has demonstrated that environmental context can induce the switch between Lévy to Brownian movement patterns (Humphries et al., 2010), but the effective mechanisms behind the implementation of such behavior (e.g., cognitive capacity, memory) often remain elusive (Budaev et al., 2019). Understanding the motor mechanisms that regulate the execution of different movement strategies and the transitions between them could provide insight into how the nervous system can drive the search for resources in complex and ever-changing environments. Drosophila larva is an excellent model to study this question, because the movement of single animals can be tracked for long periods of time in a controlled environment.

Larvae of the fruit-fly are constantly foraging and feeding to fulfill their nutritional needs for the following non-feeding pupal stage. They explore the substrate by executing sequences of crawls, pauses, and turns (Berni, 2015; Berni et al., 2012) and can efficiently explore an environment even without brain input (Sims et al., 2019). Larvae approach (or avoid) sources of odor by triggering oriented turns during chemotaxis (Gomez-Marin et al., 2011) and can also navigate through gradients of light intensity (Kane et al., 2013; Humberg and Sprecher, 2018), temperature (Luo et al., 2010; Lahiri et al., 2011), and mechanosensory cues (Jovanic et al., 2019). Their natural habitat is decaying vegetable matter distributed in patches (Ringo, 2018), and due to food decay and intraspecific competition larvae are constantly deciding what patch to visit and how long to stay before exploring for new higher quality food patches. This constant exploration comes at a high energetic cost since crawling behavior is very demanding (Berrigan and Lighton, 1993; Berrigan and Pepin, 1995).

The foraging behavior of Drosophila both in the larval and adult stages is influenced by the foraging (for) gene (Sokolowski, 2001; Sokolowski et al., 1997). Larvae with the rover allele crawl significantly longer paths on a yeast paste than larvae with the sitter allele, and a proportion of 70% rovers and 30% sitters is observed in natural populations (Sokolowski, 2001). Due to the higher dispersal of rover larvae, their pupae are usually found in the ground while those from sitter are usually found on the fruit (Sokolowski et al., 1986). However, it is not known if the behavioral differences between rover and sitter larvae can be observed in food substrates of different compositions, nor how rovers and sitters behave in a patchy environment of regions with and without food (even though it has been hypothesized that rover larvae are more likely than sitter to leave a patch of food to search for a new one, Sokolowski, 2001).

Previous studies on larval foraging focused on the behavior in homogeneous substrates, where larvae engage in a highly exploratory movement pattern if no food is available (Berni et al., 2012; Godoy-Herrera et al., 1984; Sims et al., 2019). However, the natural habitat of larvae is very patchy and it is not clear how they select feeding vs. exploring when the environment has food patches separated by areas without food. Previous studies have shown that larvae are more willing to leave a patch if the protein concentration is low but tend to stay in the patch if its nutritional content is adequate (Ringo, 2018). Nevertheless, these studies lack an individualized tracking of the path executed by larvae during patchy exploration.

Here, we investigate the mechanisms of foraging that adapt to changes in food distribution. To address this challenge, we investigate how (1) the quality of the food and (2) its distribution, homogenous vs. constrained in small patches, influence larval foraging. We test the effect of the rover and sitter genetic dimorphism in the different food distributions and disentangle the role of olfaction in remaining in food patches using anosmic animals. By combining a detailed analysis of individual larval trajectories from behavioral experiments and computational modeling, we characterize the elements of the navigation routine and show how they adapt to a changing environment. Our results show a modular adaptation to different food characteristics. Food quality modulates crawling speed, turning frequency, and fraction of pauses controlling the distance traveled and therefore the area explored. The patchy distribution of food triggers oriented turns toward the food at the patch interface, increasing the time larvae exploit the food inside the patch. The foraging polymorphism of rovers and sitters adjusts the degree of the behavioral response to different food conditions. The detailed description of the larval behavior and the model presented here provide the basis for the systematic identification of the neuronal circuits and mechanisms controlling each behavioral response modulated by different food resources.

Results

Food quality controls the distance traveled modulating the speed and the frequency of pauses

To study the effect of different food substrates in foraging larvae, we devised a behavioral assay where larvae explore different substrates with minimal external stimuli (Figure 1A). The three different substrates (agar, sucrose, yeast) had the same agar density but distinct nutritional quality (with yeast being the richest due to its high content of protein) (Materials and methods). Wildtype larvae from different polymorphisms – rovers and sitters – were separately recorded because of previously reported differences in foraging behavior (Sokolowski et al., 1997). We recorded the free exploratory behavior of groups of 10 third-instar larvae in large arenas (240 × 240 mm2) for 50 min and then tracked each individual trajectory (Risse et al., 2013; Sims et al., 2019). Three independent replicates were analyzed. To identify salient turning points in the trajectory and to obtain the distribution of turning angles of each larva, we used the Ramer–Douglas–Peucker algorithm (Materials and methods). Larvae explored the three different substrates (Figure 1B and Figure 1—figure supplement 1A) by executing sequences of crawls, turns (marked as circles in the trajectories), and pauses. Interestingly, we observed that a preferential orientation – clockwise (CW) or counter-clockwise (CCW) – is present in many trajectories, and the paths described often have circular shapes (Figure 1B).

Figure 1 with 1 supplement see all
Drosophila larva exploratory behavior in homogeneous substrates.

(A) Experimental setup: 10 larvae of the same phenotype (rover or sitter) were placed on the top of an agar-coated arena and recorded for 50 min, experiments were repeated three times with independent samples. Three types of substrates were used: agar-only (blue), sucrose (green), and yeast (orange). (B) Sample trajectories of rover larvae in the different substrates (top: agar, bottom left: sucrose, bottom right: yeast) with turning points identified by the RDP algorithm. Corresponding turning angle distributions are shown as an inset. (C) Average crawling speeds of rovers (N = 30, 30, 29) and sitters (N = 29, 30, 30) in the different substrates: agar (A, blue), sucrose (S, green), and yeast (Y, orange). The speed was calculated during bouts of crawls. Horizontal line indicates median, the box is drawn between the 25th and 75th percentiles, whiskers extend above and below the box to the most extreme data points within 1.5 times the interquartile range, points (gray for rovers, white for sitters) indicate all data points. (D) Average number of turns per minute registered in each trajectory. (E) Fraction of time in which larvae did not move (pauses). (F) Total distance traveled in 50 min. (G) Handedness score. The horizontal dashed line corresponds to a score of 0.5, that is, an equal number of counter-clockwise (CCW) and clockwise (CW) turns. Mann–Whitney–Wilcoxon test with Bonferroni correction was performed since the data were not normally distributed. ns: 0.05 < p < 1, *0.01 < p < 0.05, **0.001 < p < 0.01, ****p < 0.0001. The number of larvae tested is detailed in Table 1. Statistical power and Cohen’s size effect of non-significant comparisons are included in Table 4.

We found that the presence of food in the substrate had a strong effect on the larval crawling speed. Rover (sitter) larvae crawl on average at a speed of 0.84, 0.68, and 0.37 mm/s (0.96, 0.68, and 0.31 mm/s) in the agar, sucrose, and yeast, respectively (Figure 1C). In addition to changing speed, larvae suppressed turning in the food substrates, with rover (sitter) larvae executing an average of 2.65, 2.44, and 2.00 (2.79, 2.13, and 1.71) turns per minute in the agar, sucrose, and yeast, respectively (Figure 1D). Larvae also paused more often in the yeast substrate (Figure 1E and Video 1). Most pausing larvae were completely still, except for internal movements in their gut, suggesting they were digesting (Video 1). As a consequence, the total distance traveled showed a clear dependence with food quality, with yeast, the most nutritious food, generating the shorter path and consequently often a smaller explored area (Figure 1B, F and S1A).

Video 1
Video of a larva pausing.

A sitter larva was allowed to feed on a fine layer of yeast supplemented with 0.1% Bromophenol blue sodium salt (B5525, Sigma). During pauses the larvae are immobile and only the movement of their gut content can be seen.

We next quantified the individual orientation preference of each larva based on its turning angle distributions. We defined the handedness score H of a larva as the number of CCW turns divided by the total number of turns in the trajectory, that is, CCW and CW combined. Larvae with H > 0.5 (H < 0.5) have a bias to turn CCW (CW). Surprisingly, in both rover and sitter populations we found larvae with a very strong handedness, meaning that larvae have individual biases when turning in homogeneous environments that do not provide orientation cues in the form of sensory stimuli (Figure 1G).

Finally, we contrasted the differences in exploratory behavior of rovers and sitters in the different homogeneous substrates (Figure 1—figure supplement 1B–G). In particular, we were interested in evaluating if sitter larvae crawled significantly less than rovers in the first 5 min of the recording in the food substrates, as previously observed in experiments using yeast substrates (Sokolowski, 1980). We did not find significant differences between the crawled distances of rovers and sitters in the substrates that we tested. Thus, when the resources are distributed homogenously, the genetic foraging dimorphism could not be detected.

In summary, we have provided a detailed characterization of larval foraging behavior in homogenous substrates with different types of food. We found that larval crawling speed and probabilities to turn and to pause are behavioral elements that are adapted according to the quality of food.

A phenomenological model of crawling describes larval exploratory behavior in patchy substrates

In ecological conditions, the fruit on which Drosophila eggs are laid and on which the larvae forage decays over time. To maximize their survival chances, and reduce competition, larvae therefore move toward food patches that are more nutritious and less crowded (Del Pino et al., 2015). Here, we designed a phenomenological model to simulate larval exploratory trajectories in different substrates based on our collected data (Figure 1 and Methods). The model predicted the fraction of time larvae spent inside patches of food, as a measure of food exploitation, if larvae only used the information about the substrate while foraging. Each type of larva (rover, sitter) had a distribution of crawling speeds v and probabilities to crawl Pcrawl, to turn Pturn, and to pause Ppause in a given time step for each type of homogeneous substrate: agar, sucrose, and yeast (Figure 2A). To capture the variability in the turning behavior, each simulated larva had its own set of parameters for the turning angle distribution based on a single recorded larva. The simulated trajectories preserved the CW or CCW orientation inherited from the turning angle distributions characterized in the experiments (Figure 2B).

Figure 2 with 1 supplement see all
Model of larva crawling in different substrates.

(A) Simulated larva crawls at time steps tk and tk+2, turns at tk+1, and makes a pause at tk+3. Crawling speed and turning angle are sampled from normal and von Mises probability distributions, respectively. At each time step, there is a constant probability to turn Pturn or to pause Ppause. (B) Sample model trajectories and turning angle distributions of sitter larvae simulated in different homogeneous substrates: agar (left), sucrose (middle), and yeast (right). (C) Simulations with patchy environments: food (sucrose or yeast) is distributed inside two circular regions, with agar in the remaining substrate. Crawling speeds, turning and pause probabilities are sampled from different distributions when the simulated larva is inside (green) or outside (blue) the patch. (D) Sample model trajectories and turning angle distributions of sitter larvae in simulated patchy substrates: sucrose (left) and yeast (right) patches. (E) Average fraction of time each simulated larva (rover (r), sitter (s), N = 30) spent inside patches (sucrose and yeast) in the simulations. Horizontal line indicates median, the box is drawn between the 25th and 75th percentiles, whiskers extend above and below the box to the most extreme data points that are within a distance to the box equal to 1.5 times the interquartile range and points indicate all data points. Mann–Whitney–Wilcoxon paired test two-sided. ns: 0.05 < p < 1, ****p < 0.0001.

Using our model based on crawling behavior in homogeneous substrates, we next tested how changes in the food distribution influence the exploratory trajectories of rovers and sitters. We modeled heterogeneous environments with two circular patches of food substrate with agar substrate in the rest of the arena (Figure 2C, see Materials and methods). The two patches had a fixed radius (25 mm) that corresponds to the surface area of a grape (Xie et al., 2018). We simulated larval exploration of rovers and sitters in patches of two food substrates – sucrose and yeast (Figure 2D). The initial position was picked at random in each simulation, but always inside one of the two food patches to match the experiments.

We next quantified the fraction of simulation time that rovers and sitters spent inside patches. For each larva, this was averaged over 30 simulation runs (Figure 2E and Figure 2—figure supplement 1A, B). Inside sucrose patches, the percentage of time spent inside patches was small for both rovers and sitters (9.2% and 9.8%, respectively). These values were only slightly larger than those in a simulated environment with patches made of agar (7.47% for rovers and 6.99% for sitters – Figure 2—figure supplement 1A, C, D) – that is, the same speed and probabilities to turn and pause inside and outside patches. This result is unsurprising since in our homogeneous substrate experiments with rover and sitter larvae both had similar behavior in the agar and sucrose arenas. In simulations with yeast patches, the percentage of time spent inside patches was higher for both rovers (22.6%) and sitters (26.9%). This increase can be linked to the slower speeds and more frequent pauses in the homogeneous yeast substrate executed by the larvae. In spite of non-significant differences in the crawling of rovers and sitters in the homogeneous yeast substrate (Figure 1—figure supplement 1B–G), in our model simulated sitter larvae remained longer inside the yeast patches due to their lower (though not significantly different average crawling speed in the homogeneous yeast substrate experiments).

Thus far, our model predicts that, in patchy environments, larvae spend a relatively small proportion of time inside patches (approximately 1% for sucrose and 3% for yeast) while exploring takes up most of their time with a significant energy cost. However, our model does not integrate other possible mechanisms that a larva might employ to remain inside a food patch besides decreasing its crawling speed and increasing the fraction of pause events. We therefore compared the model predictions on foraging efficiency in patchy environments with behavioral experiments.

Increased proportion of time in patches relies on turns toward the patch center at the food–no food interface

We next recorded the larval behavior in arenas with patchy substrates. We used the same size and distribution of food patches as in our simulations (Figure 3A). Food was distributed inside, with agar outside patches (Figure 3A and Materials and methods). We tested sucrose and yeast at the same concentration as in the homogenous substrate. We also performed experiments using apple juice as a nutrient, motivated by the fact that it is ecologically relevant and that, unlike sucrose, the fructose contained in apple juice is volatile, which makes it detectable by smell and not only by taste. Groups of five larvae of the same type (rovers or sitters) were placed inside each patch (total of two) at the beginning of the recordings (total of ten larvae of the same type per replicate, repeated in three independent experiments).

Figure 3 with 2 supplements see all
Larval exploratory behavior in patchy substrates.

(A) Experimental setup: Five larvae of the same phenotype were placed on top of each food patch (two patches, total: 10 larvae per experiment). Three types of food patches were tested: sucrose (green), yeast (orange), and apple juice solution (magenta). Agar was uniformly spread in the arena outside the food patches. (B) Sample trajectories of sitter larvae in the three patch substrates with inward (outward) turns marked in black (white) circles. Distribution of turning directions is shown on the bottom of each trajectory. (C) Larval crawling speeds of rovers and sitters measured inside (colored bars) and outside (blue bars) food patches: sucrose (S, green), yeast (Y, orange), and apple juice (AJ, magenta). Horizontal line indicates median, the box is drawn between the 25th and 75th percentiles, whiskers extend above and below the box to the most extreme data points that are within a distance to the box equal to 1.5 times the interquartile range and points indicate all data points. (D) Average number of turns executed per minute. (E) Fraction of time in which larvae did not move (pauses). (F) Handedness score. (G) Total distance crawled by rover (r; darker colors) and sitter (s; lighter colors) larvae in the first 5 min of the recording. (H) Fraction of time spent inside patches of rovers (r) and sitters (s). (I) Left: Identification of turning angle as inwards (θ2<θ1, black) or outwards (θ2>θ1, white). Right: Circular regions with fixed distances relative to the patch center. The yellow line represents the patch border. (J) Relative fraction of inward turns calculated as a function of the distance from the patch center. The distance bin that includes the patch radius is highlighted in yellow. Left: Sucrose, middle: apple juice, right: yeast patches. Top: Rovers, bottom: sitters. Mann–Whitney–Wilcoxon test two-sided was performed since the data are not normally distributed. ns: 0.05 < p < 1, *0.01 < p < 0.05, **0.001 < p < 0.01, ***0.0001 < p < 0.001, ****p < 0.0001. The total number of larvae tested is detailed in Table 1. Statistical power and Cohen’s size effect of non-significant comparisons are included in Table 4.

We tracked the trajectories with the same methods used in the homogeneous environment (Figure 3B and Figure 3—figure supplement 1A). Then, we performed the analysis separately for the two different regions: inside and outside the patches, and quantified features of the larval exploratory behavior. Inside yeast and apple juice patches, larvae crawled significantly slower than outside them (Figure 3C). In yeast patches, both rovers and sitters executed fewer turns inside than outside (Figure 3D). All larvae made significantly more pauses inside the food patches than outside (Figure 3E). We also observed that the handedness score of the larvae is less broad than in the homogeneous substrates (Figure 3F), which may be caused by reorientations that are triggered to prevent the larva from exiting the food patch. As expected from the phenotype, sitter larvae crawled a shorter distance in the first 5 min of the recording in the yeast but also the sucrose substrates (Figure 3G). In general, sitter larvae had slower crawling speeds and executed fewer turns in the patchy environments than rovers (Figure 3—figure supplement 1A–C). We also noticed that sitters paused more inside patches than rovers (Figure 3—figure supplement 1D). Outside yeast and apple juice patches, the crawling speed increased but did not return to levels similar to the agar-only condition, suggesting that the behavior of larvae that exit the patch is influenced by the recent food experience or that larvae might still be sensing the food (Figure 3—figure supplement 1E). In line with this, in yeast the number of turns outside the patch was higher than inside the patch.

Our model predicted that fraction of time spent inside patches should vary according to the substrate: larvae should remain longer inside yeast patches than inside sucrose patches (Figure 2E). In particular, simulated sitter larvae stayed longer than simulated rovers inside yeast patches. In the experiments, the same trend was observed: for both rovers and sitters the fraction of time spent inside patches was higher in the yeast compared to both sucrose and apple juice patches (Figure 3H). Sitter larvae stayed significantly longer inside yeast patches than rovers (Figure 3H). Nevertheless, the percentage of time the larvae spent inside patches in the experiments was very different from our model predictions. Rover (sitter) larvae remained on average 72.6% (72.3%) of the experiment inside sucrose, 85.7% (90.0%) inside yeast, and 75.6% (81.3%) inside apple juice patches. Those values were much higher in the experiments than what we predicted with our simulations, and suggest that larvae might employ other mechanisms in addition to slower crawling and more frequent pauses to remain inside the food.

To gain more insight into the strategies used by larvae to increase the time spent inside the food patches, we studied the distribution of turns in the food–no food interface. First, we labeled each turn as inwards or outwards depending on whether they were oriented toward or away from the patch center (Tao et al., 2020; Figure 3I, left). We observed that inward turns occur more often than outward turns at the border of the patch for the three substrates (Figure 3B, inward turns are shown in black). To control for possible mechanosensory effects due to the border edges, we prepared new arenas with patches that contained no nutrients, either using the same agar that composed the rest of the arena, or using ultrasound gel (Methods). Larvae in the agar–agar or the agar–gel border did not show any changes in their preference to turn toward the patch center, confirming that the behavioral change observed in response to food is specific (Figure 3—figure supplement 2).

We then studied the fraction of turns toward the patch center as a function of the distance to the patch center (Figure 3I, right). For the three types of substrates, the bias to turn inwards was clearly manifested when the larvae experienced the patch border (patch radius: 25 mm, distance bin: 20–30 mm) (Figure 3J). The bias persisted when the larva exited the patch (distance bins: 30–40, 40–50, 50–60 mm). We did not consider further distance bins in our analysis because most larvae did not reach those locations in our experiments.

Therefore, our model predictions do not seem to be well supported by experiments with patchy substrates. In particular, we conclude that when larvae reach the food–no food interface their turning behavior changes. This is accomplished by turning toward the patch center while maintaining the handedness (Figure 3J and Figure 3—figure supplement 1F) and represents an important mechanism to remain inside the food.

Anosmic larvae also select turns toward the patch center when reaching the food–no food border, but not on the yeast

It is well known that Drosophila larvae can efficiently navigate toward or away an odor source using chemotaxis (Louis et al., 2008; Gomez-Marin et al., 2011; Schulze et al., 2015). Chemosensory information from gustatory and olfactory receptors is combined to allow larvae to locate food sources in the environment (Vosshall and Stocker, 2007). We next wondered how much of the tendency to turn toward the patch center once outside the patch could be attributed to processing olfactory cues.

Thus, we repeated the patchy experiments with mutant anosmic larvae, where Orco, the obligatory co-receptor for all olfactory neurons, apart the CO2 sensing ones, is mutated (Vosshall and Stocker, 2007) and tested if they show the same distant-dependent bias when exploring the patchy substrate.

Anosmic larvae extensively explored the patchy substrate (Figure 4A). In general, they exhibited a small difference in crawling speeds when comparing their behavior inside vs. outside of food patches (Figure 4B). Curiously, this difference in speeds was non-significant inside vs. outside yeast patches. We also found that the fraction of pauses of anosmic larvae in yeast patches was smaller than that of rovers and sitters (Figures 3G and 4D). This suggests that yeast patches are not attractive to anosmic larvae, in agreement with the lower fraction of time spent inside yeast patches relative to sucrose and apple juice patches (Figure 4F).

Figure 4 with 1 supplement see all
Exploratory behavior of anosmic larvae in patchy environments.

(A) Sample trajectories of anosmic larvae in the three patch substrates with inward (outward) turns marked in red (gray) circles. Distribution of turning directions is shown on the bottom of each trajectory. (B) Crawling speeds of anosmic larvae measured inside (colorful bars) and outside (blue bars) food patches: sucrose (S, green), yeast (Y, orange), and apple juice (AJ, magenta). Horizontal line indicates median, the box is drawn between the 25th and 75th percentiles, whiskers extend above and below the box to the most extreme data points that are within a distance to the box equal to 1.5 times the interquartile range and points indicate all data points. (C) Average number of turns per minute inside and outside patches. (D) Fraction of pauses inside and outside patches. (E) Handedness score of anosmic larvae inside and outside patches. (F) Fraction of time spent inside patches for different types of food. (G) Relative fraction of inward turns calculated as a function of the distance from the patch center, top: sucrose, middle: yeast, bottom: apple juice. The distance bin that includes the patch radius is highlighted in yellow. Mann–Whitney–Wilcoxon test two-sided was performed since the data are not normally distributed. ns: 0.05 < p < 1, **0.001 < p < 0.01, ***0.0001 < p < 0.001, ****p < 0.00001. The number of larvae tested is detailed in Table 1. Statistical power and Cohen’s size effect of non-significant comparisons are included in Table 4.

Next, we investigated if anosmic larvae can bias their turns at the patch border interface without navigating odorant cues. Turns in the trajectory were labeled as inwards or outwards (as in Figure 3I) and the fraction of turns toward the patch center was analyzed as a function of the distance away from the patch center.

In sucrose and apple juice substrates, anosmic larvae consistently increased the fraction of inward turns near the patch border (20–30 mm; Figure 4G). This was not the case in the yeast patches, where no bias was detected at the patch border.

In sum, we found that anosmic larvae, apart from on yeast, trigger turns toward the patch center at the food–no food interface, suggesting that olfaction is not the only mechanism responsible for the turning bias that increases the fraction of time larvae spend inside patches.

Taste very likely influences the probability that larvae remain in the patches. To control for the diffusion of nutrients (sucrose and apple juice) at the edge of a patch, we evaluated the maximum distance at which an increased fraction of turns toward the center was significantly different when compared to the yeast non-responsive anosmic control. At a distance greater than 0.5 cm from the edge, anosmic larvae on sucrose, apple juice, and yeast were indistinguishable, suggesting that diffusion has a limited impact on behavior (Figure 4—figure supplement 1A).

Finally, to control for possible effects of diffusion over time, we compared the fraction of turns toward the center in the first and second half of the experiment. For most distance and nutrients, the two distributions were not significantly different (Figure 4—figure supplement 1B).

To remain inside of the food patch larvae combine turning bias with other strategies

To understand the impact of the turning bias on the percentage of time that larvae spend inside patches, we included a distance-dependent probability of turning toward the patch center in our model (Figure 5A). After drawing a turning angle from the probability distribution, the turn was implemented toward the patch center with probability Pbias that depends on the distance between the current position and the center of the closest patch (Figure 5B). For each simulated substrate, larva type, and relative distance, Pbias corresponds to the fraction of turns toward the patch center quantified in our experiments (Figures 3J and 4G).

Figure 5 with 2 supplements see all
Interplay between food quality and patches distribution.

(A) Schematic showing inward turn (clockwise, CW) being selected by the simulated larva. By selecting inward turns, the trajectory approaches the patch center. (B) Spatial-dependent probability of turning toward the patch center. Each region is a concentric circle with a fixed probability of drawing inward turns (see Figure 3I, right). The yellow line shows the patch border. (C) Sample simulated trajectories for a sitter larva with biased inward turns: sucrose patch (green), yeast patch (orange), and apple juice patch (magenta). (D) Fraction of time spent inside patches of rovers (r; darker colors) and sitters (s; lighter colors) in the different substrates: sucrose (S, green), yeast (Y, orange), and apple juice (AJ, magenta). Each point is 30 simulation runs of one larva (total: 30 larvae simulated per substrate). Horizontal line indicates median, the box is drawn between the 25th and 75th percentiles, whiskers extend above and below the box to the most extreme data points that are within a distance to the box equal to 1.5 times the interquartile range and points indicate all data points. (E) Sample trajectories of sitter larvae in environments with varying number of randomly located patches, with a fixed total area of yeast substrate being distributed (Np = 1, 2, 8, 32 from left to right). (F) Average fraction of time spent inside patches of distinct substrates (S: sucrose, green; Y: yeast, orange, and A: apple juice, magenta) for rovers (left) and sitters (right) as a function of the number of patches. Each point is the average of 30 larvae (30 simulation runs each). Bars show the standard deviation. (G) Same as (F) but for the average fraction of visited patches. Mann–Whitney–Wilcoxon paired test was performed since the data are not normally distributed. ns: 0.05 < p < 1, ***0.0001 < p < 0.001.

We observed that the simulated trajectories with this distance-dependent turning bias resemble the experimental ones much more (Figure 5C), with larvae often returning to a patch when leaving its border. Indeed, larvae spent three times longer inside a patch in the new simulations compared to the model without biased orientations (Figure 5D and Figure 5—figure supplement 1A): now rover (sitter) larvae remain on average 31.1% (28.9%) of the simulation inside sucrose patches and 63.8 (68.4%) of the simulation inside the yeast patches. Simulated anosmic larvae also showed a gain in the ratio of time inside patches (Figure 5—figure supplement 1B, C). Therefore, biased orientations at the patch border are an important mechanism employed by larvae to return to a food source when they detect a change in the substrate quality. This can be achieved without olfactory orientation cues, since anosmic animals can also perform biased turns (Figure 4G). However, the ratio of time that simulated larvae remain inside patches was still smaller than that measured in the experiments (Figures 3H and 4F). We reason that other mechanisms, such as working or short-term memory (Louis et al., 2008; Schleyer et al., 2015), or other sensory modalities at the vicinity of the border of the patch (see discussion) can contribute to increasing the time inside the food.

Our model reveals the interplay between food quality and patches fragmentation

We next used our model to investigate how a further fragmentation of the food patches affects the ability of larvae to stay in patches where they can feed. To test we fixed the total area of food S and varied the number of patches choosing the center coordinates for each patch randomly (Figure 5E). We tested seven levels of fragmentation from 1 to 64 patches and to compensate for different patch radii, we adjusted the distance-dependent probability to turn inwards of each larva (Figures 3J and 4G, see Materials and methods). We modeled the three types of food tested thus far, for rover, sitter, and anosmic larvae. In total, this would represent 1575 hr of experiment, highlighting the advantage of the model.

First, we quantified the average fraction of the time spent inside patches relative to the whole simulation for the different food substrates as a function of the number of patches (Figure 5F and Figure 5—figure supplement 2A, B). As expected, both rovers and sitters spent less time inside a patch as the number of patches increases (and thus the patches radius decreases) (Figure 5F). Larvae spent longer inside patches in more nutritious environments, for example yeast, irrespective of the number of available patches. Interestingly, despite the small differences we previously quantified, our results showed that sitter larvae consistently spent more time inside yeast patches than rovers for each number of patches (Figure 5—figure supplement 2D). This was not observed in the sucrose or apple juice patches. Anosmic animals also spent less time inside patches when the number of patches increases, but the dependence on the quality of food was much less pronounced (Figure 5—figure supplement 2B).

Next, we investigated the effect of different food substrates on the number of patches larvae explore to understand how fractioning environment would affect exploitation, which is key for survival. We quantified the fraction of new patches a larva visits during the simulation (discounting the source patch, since all the simulations start with the larva inside one patch) (Figure 5G). Rovers and sitters explored more patches in the less nutritious substrate (sucrose), with a slightly higher fraction of visited patches for rovers in the sucrose and yeast patches (Figure 5G; Figure 5—figure supplement 1E). Anosmic larvae showed a weaker effect of the substrate on the fraction of patches visited (Figure 5—figure supplement 1C).

Our model predicts a trade-off between the quality of the food and the fraction of patches visited: when exploring a substrate with low-quality (high-quality) food, the larvae are more (less) likely to leave and more (fewer) patches are visited.

Larvae experience a trade-off between food consumption and exploration

To confirm that larvae adapt their behavior as modeled in response to different quality and fragmentation of food, we compared the behavior of larvae in two and eight patches. We conducted new experiments in arenas with eight patches of sucrose and yeast with rover and sitter larvae. Three sets of random positions of patches were used for each replicate (Figure 6B). Each larva (total of eight) was placed inside a different patch and left to crawl for 50 min. The data were compared to the experiments with two patches (Figure 3).

Behavioral response to changes in food quality and fragmentation.

(A) Sample simulated trajectories for sitter and rover larvae exploring in eight patches: sucrose patch (green) and yeast patch (orange). (B) Sample experimental trajectories of rover and sitter larvae in an arena with eight patches of food. Three random distributions (exp1; exp2; exp3) were used for each type of food: sucrose patches (green) and yeast patches (orange). (C) Fraction of time spent inside patches of rovers (darker colors) and sitters (lighter colors) on sucrose (green) and yeast (orange). Horizontal line indicates median, the box is drawn between the 25th and 75th percentiles, whiskers extend above and below the box to the most extreme data points that are within a distance to the box equal to 1.5 times the interquartile range and points indicate all data points. (D) Average fraction of time spent inside patches of distinct substrates (sucrose, green; yeast, orange) for rovers and sitters as a function of the number of patches. Data represent mean ± standard deviation. (E) Same as (D) but for the average fraction of visited patches. analysis of variance (ANOVA) test was performed for (C) and (D) (normally distributed) and Mann–Whitney–Wilcoxon paired for (E) (non-normally distributed). ns: 0.05 < p < 1, *0.01 < p < 0.05, **0.001 < p < 0.01, ****0.00001 < p < 0.0001. The number of larvae tested is detailed in Table 1. Statistical power and Cohen’s size effect of non-significant comparisons are included in Table 4.

A first comparison of the trajectories of simulated and experimental larvae exploring in an environment with eight patches shows great similarity (Figure 6A, B). As predicted by the model, both rovers and sitter spent half of the time inside patches when the area of food was divided in eight compared to two patches (Figures 5F and 6C, D). Furthermore, the larvae stayed longer on the yeast patches compared to the sucrose ones (Figure 6C, D), supporting the prediction of the model that larvae will spend less time in less nutritious patches irrespective of the number of available patches.

We then analyzed the effect of food quality on the proportion of patches visited by the larvae. There were no significant differences comparing the larvae in yeast and sucrose apart for rover in yeast for two patches. In this case, the model had predicted a difference between yeast and sucrose that is not present experimentally, probably because the larvae spend more time on the patches than what the model predicted via other mechanisms. However, it is clear that larvae spent more time looking for new patches (outside patches, Figure 6C, D) when the quality of food was lower (in sucrose) compared to higher quality (yeast), but they did not reach more patches in our experimental timeline. It is possible that having left a source of poor food, the larvae were more interested in exploring in search of food of better quality.

Finally, we were particularly interested in testing the prediction that larvae would reach a steady state in the proportion of patches visited as the food would become more fragmented. This was supported by the experiments with two and eight patches despite our suspicion that.

Overall, the experiments show how larvae tune the elements of the navigation routine to generate a foraging behavior that adapts to the quality and spatial distribution of food resources.

Discussion

Foraging behavior is a complex process influenced by many internal factors (locomotion style, sensory perception, cognitive capacity, age) and external variables (spatiotemporal distribution of resources, presence of predators, social interactions with co-specifics). Here, we focused on the detailed characterization of foraging in a single model organism, the fruit fly Drosophila larva, using extensive experiments and modeling. This allowed us to study the role of both internal and external factors on foraging: (1) genetics (rovers, sitters, and later orco null anosmic animals), (2) food quality (agar, yeast, sucrose, and apple juice), and (3) food spatial distribution (homogeneous and heterogeneous environments).

We systematically investigated larval exploratory behavior first in experimental arenas with homogeneously distributed food. Larval crawling speed, turning frequency and fraction of pausing events adapted according to the quality of the food substrate (Figure 1C–E). The quality of the food had a strong impact on the distance traveled by the larvae. In yeast, larvae moved less and their speed and turn frequency were decreased. They also made more pauses, with the majority remaining stationary, except for internal gut movements (Video 1), which suggested that they were digesting the yeast. The pauses were rarely observed in sucrose, which is metabolized more quickly than yeast, even when mixed with agar (Figure 1E).

We observed that larval trajectories often had a circular shape, revealing an individual preference for a given turning direction in the absence of direction cues, which we quantified as the larval handedness (Figure 1B, F). The population variability in the handedness has been quantified in adult flies (Buchanan et al., 2015), but to our knowledge not until now at the larval stage. In adult walking flies, individual preferences of turning left or right in maze tests have been shown to persist across days (Buchanan et al., 2015) and recently have been linked to anatomical differences in the synaptic distribution of bottleneck neurons downstream of the central complex (Skutt-Kakaria et al., 2019). It is therefore possible that, as found in adults, larval individual differences in neuronal connections could define handedness. It would be interesting to understand the evolutionary advantage of handedness, if there is one, and to relate it to the ‘hard-wired’ circuitry controlling Lévy search behavior (Sims et al., 2019).

It is expected that animals change their foraging behavior depending on the quality and spatial distribution of food, with more localized exploitation of resources where they are abundant and a more exploratory behavior when resources become scarce (Humphries et al., 2010). We tested this in a phenomenological model of larval foraging behavior in patchy substrates (Humphries et al., 2010; Figure 2). We reasoned that crawling speed, turning frequency and fraction of pauses are the behavioral elements that adapt when the larva crosses the food–no food interface at the patch boundary. To quantify the food exploitation, we measured the fraction of the time each larva spent inside the patches. We found that decreasing the speed and turning frequency and increasing the fraction of pauses is not sufficient to explain why larvae remain inside the food for longer periods.

In experiments with patchy substrates, we found that larvae spend a longer time inside food patches than predicted with our model (Figure 3H). The lack of agreement between the experiments and our model was not surprising, since the latter does not include additional mechanisms that could guide the larva back to the patch when it leaves it, such as chemotaxis (Louis et al., 2008; Gomez-Marin et al., 2011). Since in chemotaxis larvae redirect their turns toward a source of odor, we classified each turn in their recorded trajectory as toward or away from the patch center. We observed that the fraction of inward turns is very high around the patch border (Figure 3J). To test whether larvae could redirect their turns toward the food when exiting it using olfactory cues, we repeated the experiments with anosmic mutants. Surprisingly, in sucrose and apple juice substrate anosmic larvae bias their turns toward the patch center when in the neighborhood of the patch border (Figure 4G). Therefore, this reorientation at the border does not seem to rely solely on olfaction. When exiting the food patch, larvae sense the lack of taste and it is possible that the turn bias changes as a result of temporal integration of the recent sensory-motor experience allowing them to return to the patch, as observed when navigating in an olfactory or light intensity gradient. Also, the patches of sucrose and apple juice were in direct contact with the surrounding agar arena. This has the advantage of generating a smooth transition in the substrate (Figure 3—figure supplement 1E, F), but it also allows diffusion at the interface which the larvae can sense as they crawl away from the food (Lebrun and Junter, 1993). In anosmic larvae, the fraction of turns toward the center for sucrose and apple juice patches was only higher compared to the one for the yeast patch (where there was no food effect) within the first half centimeter outside the patch, suggesting that the impact of diffusion could be significant only in that region (Figure 4—figure supplement 1).

An experiment using the gustatory sweet sensor Gr43a mutant on sucrose, which is not volatile and does not produce smell, could help discerning the contribution of taste at the border of the patch (Fujishiro et al., 1984; Marella et al., 2006; Miyamoto et al., 2013; Wang et al., 2004; Mishra et al., 2013). For yeast, the lack of smell completely changed the response of the larvae, which did not show differences inside and outside the patch for most foraging parameters (Figure 4B, C, E, G). In this instance, taste was not sufficient to retain larvae inside the yeast patch (compare Figure 3H with Figure 4F) even though several gustatory receptors have been shown to be activated by yeast metabolites (Wisotsky et al., 2011; Ganguly et al., 2017; Croset et al., 2016).

Another sensory modality that could have influenced the larval behavior at the food–no food interface, is mechanosensation. We excluded the possible role of the border of the patches performing experiments in patches without food (Figure 3—figure supplement 2). However, when larvae are crawling, they leave a print of their denticle attachment on the agar, that could inform them about their previous location and help returning to the food. Overall, the differences in behavior of larvae exposed to different foods, revealed the complexity of the sensory-motor processing involved in foraging.

One of the strengths of our phenomenological model is that it incorporates a modular organization of foraging that could reflect how the crawl and turn modules are controlled. First, we modeled a stochastic search where no information regarding food is available outside of the current location, because food is absent or because the larvae cannot sense it. This corresponds to an autonomous search behavior implemented by circuits located in the ventral nerve cord without input from the brain (Berni et al., 2012; Sims et al., 2019). Second, we incorporated a goal-directed navigation that allows larvae to return to the food. Our phenomenological model includes a distance-dependent probability to turn inwards that mimics the effect of chemotaxis (when present), as much as any other possible mechanism that contributes to the turning probability. As a consequence, we observed that simulated larvae, even when the resources are fractioned in eight patches, could stay inside the food patch for longer periods, in line with experimental observations (Figures 5 and 6). The model could be improved by setting the turning properties outside the patch to match as closely as possible experimental observations. To this end, we could consider studies of larvae crawling in different attractive gradients, where the changes in turning probability and angle, including weathervaning, have been investigated in relation to precise spatiotemporal information of odorants (Louis et al., 2008; Gomez-Marin et al., 2011; Davies et al., 2015). It would also be helpful to have information about other attractive gradients, like taste, to know if a common set of mechanisms is used regardless of the sensory modality. Using this information, our model could be used to investigate how crawling speed and turning properties are controlled via descending pathways from the brain (Tastekin et al., 2018; Jovanic et al., 2019). Finally, in the presence of nutrients, our model adjusts movements to stay on the food patch. The concerted decrease in turning rate and crawling speed and increase in the number of pauses, suggests that a neuromodulatory depression of movement (Marder, 2012; Lin et al., 2019) could be relevant in this phase. It would be interesting to investigate more generally how neuromodulators influence the decision to remain or explore new food resources in relation to the resources available and the larval motivational state.

Overall, we found both in our experiments and modeling that larvae spend less time exploiting patches of less nutritious food (e.g., sucrose). What could be the effect of this when several patches are available in the substrate? Our model results predict that larvae would spend more time exploring and more patches would be visited when food quality is lower (Figure 5G). In natural environments, this would enhance the chances that larvae will eventually find a better food source in the surroundings. Our experiments show a slightly different picture, where larvae indeed explore for a longer period when on less nutritious food but the number of patches they find is not increased compared to when they are on a more nutritious food (Figure 6C, D). It is possible that having left a poor food source, the larvae are more likely to continue looking for a more nutritious one, in the short term, instead of visiting and exploiting a new poor patch. Therefore, the internal state of the animal is probably playing an important role in the decision of choosing a new patch of food to exploit (Ringo, 2018; Branch and Shen, 2017).

The differences we found in the foraging behavior of rovers and sitters are not as drastic as previously reported, where the length of the path of rovers was roughly twice that of sitters when crawling in a yeast paste for 5 min (Sokolowski, 2001). In the homogeneous agar, sucrose, and yeast substrates, we did not observe significant differences in the path length of rovers and sitters (Figure 1—figure supplement 1). This was expected for the no-food condition (agar substrate; Kaun et al., 2007; Yang et al., 2000), but not in the presence of yeast (Sokolowski, 2001). This could be attributed to differences in the food preparation protocol: we applied a thin layer of yeast on top of the agar surface instead of thick yeast suspension as in Sokolowski, 1980 to allow recording from underneath the food (Risse et al., 2013). Also, our experiments were conducted in the dark, which might influence behavior (Sokolowski, 1980).

Interestingly, when the food is constrained inside patches, as done in the classical work studying the foraging polymorphism, we observed significantly shorter crawling paths of sitters in sucrose and yeast patches (Figure 3G). Sitters’ crawling speed was also slower and they perfomed fewer turns per minute and more pauses (Figure 3—figure supplement 1). It is possible that the presence of a patch border plays a significant role for the foraging polymorphism phenotypic expression.

In summary, we have identified a set of behavioral elements – the crawling speed, frequency and biasing of turns, and fraction of pauses – that adapt when larvae explore environments with a patchy distribution of food sources. This adaptation leads to an efficient substrate exploration, as larvae either increase the time inside nutritious food patches or continue exploring the substrate depending on the local quality of food.

Materials and methods

Animals

Rover and sitter flies were a gift of Marla Sokolowski (University of Toronto) and Orco[2] from Bloomington stock center (stock 23130). Flies were allowed to lay eggs for 1 day in standard corn meal food, which consists of 420 g of cornmeal; 450 g of dextrose; 90 g of yeast; 42 g of agar; 140 ml of 10% Nipagin in 95% EtOH; 22 ml of propionic acid, and 6.4 l of water. Larvae that were 72 hr old were collected for the experiment.

Larva tracking

Request a detailed protocol

We recorded movies of larval exploratory behavior in arenas with minimal external stimuli – the recordings were made in the dark with a constant temperature of 25°C. Each trial lasted 50 min and the larvae were simultaneously tracked in a 240 × 240 mm2 arena with a 2-mm thick layer of 0.4% agar-based coating (see the protocol of substrate preparation below).

At each trial, 10 young third-instar larvae (72–80 hr since egg laying) of approximately the same size were washed to remove traces of food and allowed to crawl freely for 5 min on a clean 0.4% agar coated plate before being transferred to the arena (Table 1). We used a Frustrated Total Internal Reflection (FTIR)-based imaging method to record the larval exploratory behavior (Risse et al., 2013). Movies (duration 50 min) were recorded with a Basler acA2040-180km CMOS camera at 2048 × 2048 px2 resolution, using Pylon and StreamPix software, mounted with a 16-mm KOWA IJM3sHC.SW VIS-NIR Lens and 825-nm high-performance longpass filter (Schneider, IF-093). We recorded the movies at 2 frames per second to obtain forward movement displacements and actual pause turns that are recorded accurately rather than to include ‘flickering’ movements associated with peristaltic movements.

Table 1
Number of larvae per recording.
SubstrateNumber of trialsAverage number of larvae per trialTotal larvae
Agar homogeneous3 rovers
3 sitters
10 rovers
10 sitters
30 rovers
29 sitters
Sucrose homogeneous3 rovers
3 sitters
10 rovers
10 sitters
30 rovers
30 sitters
Yeast homogeneous3 rovers
3 sitters
10 rovers
10 sitters
29 rovers
30 sitters
Agar 2 patches3 sitters9 sitters28 sitters
Gel 2 patches3 sitters11 sitters33 sitters
Sucrose 2 patches2 rovers
2 sitters
3 anosmic
8 rovers
8 sitters
10 anosmic
15 rovers
15 sitters
28 anosmic
Sucrose 8 patches3 rovers
3 sitters
8 rovers
8 sitters
24 rovers
19 sitters
Yeast 2 patches3 rovers
3 sitters
3 anosmic
10 rovers
10 sitters
10 anosmic
30 rovers
27 sitters
21 anosmic
Yeast 8 patches3 rovers
3 sitters
8 rovers
8 sitters
25 rovers
21 sitters
Apple juice 2 patches3 rovers
3 sitters
3 anosmic
10 rovers
10 sitters
10 anosmic
30 rovers
30 sitters
29 anosmic

Substrate preparation

Request a detailed protocol

The following food substrates were prepared for our experiments, and stored refrigerated for up to 1 day:

  1. Agar substrate: 0.8 g of agar was melted in 200 ml of distilled water;

  2. Sucrose substrate: 0.8 g of agar with 3.42 g of sucrose was dissolved in 200 ml of distilled water;

  3. Apple juice substrate: 0.8 g of agar with 0.342 g of sucrose and 5 ml apple juice (Del Monte Quality Pure Apple Juice from Concentrate) was dissolved in 195 ml of distilled water;

  4. Yeast substrate: 0.8 g of agar was melted in 200 ml of distilled water with a layer of 5 ml of 20% yeast in water on top.

In the case of agar and sucrose homogeneous substrates, the solution was homogeneously spread on top of the acrylic arena and we waited for it to reach room temperature before transferring the larvae to the arena. Yeast homogeneous arenas were obtained by spreading 5 ml of 20% yeast in water with a soft metallic disk. For sucrose or apple juice patchy arenas, first, the agar solution was homogeneously spread in the acrylic arena. When the solution cooled down, two holes in the agar were made at fixed positions (Figure 3A) using circular-shaped Petri dishes with a 25-mm radius. We carefully removed the agar inside the holes and transferred the food solutions to the holes with the same thickness as the agar around them. Control two agar patches were filled with 04% agar alone. For each one of the two yeast patches, 100 µl of yeast solution was placed on a 25-mm-radius metal disc and printed on the agar. For gel 2 patches control we stamped a drop of 150–200 mg of ultrasound gel for TENS machine (Boots ingredients: purified water, glycerin, propylen glycol, hydroxyethylcellulose, sodium citrate, citric acid, domiphen bromide). The viscosity of the gel is not identical to the one of yeast, but it informs us about the transition from viscous and smooth (gel-yeast) to agar. For eight patches, we chose three distributions randomly generated in the modeling experiment (Figure 6A, lower panel). Using a 12.5-mm-radius disc we printed the patches with 25 µl of yeast solution. For sucrose 8 holes were made using a cylinder and then filled with food solution. One larva was placed in each patch, meaning that each larva was exposed to a different distribution of the resources. The experiments were repeated three times.

Descriptive statistics of larval trajectory

Request a detailed protocol

The data (x,y coordinates of individual larvae) were extracted from the behavioral movies using the FIM track free software (Risse et al., 2017). We used a Kalman filter to the (x,y) coordinates of each larva (code will be available at github after the paper is accepted). The position of each larva in video frame j is represented as the vector:

R(tj)=(x(tj),y(tj)),j=1,2,,N

where xtj and ytj are the centroid coordinates, tj=jΔt(j=1,,N) , Δt=0.5 s, and N=6000 is the number of frames continuously recorded during the experiment. We defined the following quantities that were used in our analysis.

Velocity:

V(tj)=R(tj)R(tj1)Δt

Heading:

H(tj)=V(tj)s(tj)

Scalar speed:

s(tj)=V(tj)

Instantaneous turn rate:

|ΔθΔt|(tj)=cos1(H(tj1)HT(tj))Δt,(0Δθπ)

Next, the Ramer–Douglas–Peucker algorithm (https://pypi.org/project/rdp/) was used to simplify the larval trajectories and therefore identify the locations where larvae executed turns. After visual inspection of the simplified trajectories, we fixed the distance dimension ε, that represents the maximum distance between the original points and the simplified curve. ε=2.5 mm to the analysis with agar, sucrose, and apple juice and ε=1.25 mm to the yeast analysis.

With turning points identified in the trajectory, the turning angles were obtained in the range [π,π] using the atan2 function in python. As a convention, clockwise turns were in the range -π,0 and counter-clockwise turns in the range 0,π . The handedness index of each larva was obtained as:

H=NCCWNCCW+NCW

where NCCW is the number of counter-clockwise turns in the trajectory and NCW the number of clockwise turns. Thus, if H>0.5 (H<0.5) the larva has a bias to execute more counter-clockwise (clockwise) turns.

From the turning points identified by the RDP algorithm, we built a vector that registers 1 in the time points where turns were registered and 0 otherwise. The length of this vector is the number of frames in the recording. Next, we applied a rolling window of 120 frames (1 min) to this vector and summed the elements within the window. Then, we averaged the number of turns registered within each 1-min window to obtain the average number of turns per minute.

Patch radius and center coordinates

Request a detailed protocol

We used imageJ to determine the center and radius of each patch in the experiments. A frame of the recording was adjusted for contrast and brightness until the borders of the patch became visible. Circular regions of interest were drawn for each patch and the center coordinates and radius were obtained.

Classification of turns as toward the patch center

Request a detailed protocol

Let Stk be the trajectory simplified by the RDP algorithm, where each point is a turning point of the original trajectory. To classify the k th turn in the trajectory as inwards or outwards, we define the following vectors:

V1=S(tk)S(tk1)
V2=Stk+1-Stk
U=P-Stk

where Stk-1 and Stk+1 are the previous and the following turning locations and P is the center of the patch that is closest to Stk . The following angles are then computed (Figure 3I, left):

θ1=cos1(V1UV1U)
θ2=cos1(V2UV2U)

and the turn at Stk is classified as inwards (outwards) if θ2<θ1 (θ2>θ1) (Tao et al., 2020).

Model

Homogeneous substrate

Request a detailed protocol

The simulated crawling substrate has rigid boundaries and the same dimensions as the behavioral arenas used in the experiments (240 × 240 mm2). At each time step tk the simulated larva can be at one of three different states (Figure 2A):

  1. with probability Pcrawl crawling with speed v(tk)>0 sampled from a normal distribution;

  2. with probability Pturn turning an angle θ(tk) sampled from a von Mises distribution;

  3. with probability Ppause paused (v(tk)=0).

The parameter values and distributions were obtained from our experimental data of larval crawling in homogeneous substrates and are unique for each type of larva and substrate (Table 2). Crawl, turn or pause events were registered with a constant probability per time step (Pcrawl = 1 − (Pturn + Ppause)) and the simulation duration was the same as our behavioral recordings (50 min). To capture the variability in the turning behavior, each larva was simulated with its own set of parameters for the turning angle distribution according to one recorded larva (with an average of 30 sitter and 30 rover larvae recorded at each type of substrate). The RDP algorithm was then used to identify salient turning points in the simulated trajectory (Figure 2B).

Table 2
Parameters of model in homogeneous and patchy substrates obtained in homogeneous substrate experiments.
SubstrateLarvaMean v (mm/s)Std v (mm/s)Pturn/sPpause/s
AgarRover0.840.130.0440.0083
AgarSitter0.960.130.0460.0063
SucroseRover0.680.0920.0410.012
SucroseSitter0.680.0850.0350.021
YeastRover0.370.130.0330.25
YeastSitter0.310.110.0280.25

Patchy substrate

Request a detailed protocol
Without biased turns toward the food
Request a detailed protocol

We modeled patchy environments initially with two circular patches (radius 25 mm) of food substrate (sucrose or yeast) with agar substrate in the rest of the arena (Figure 2C). Crawling speed and probabilities to turn or pause were drawn based on the current position of the simulated larva. The parameters were sampled from the corresponding food experiment when the larva was inside a patch, and sampled from the agar experiment when the larva was outside the patch. The turning angle distribution of each simulated larva corresponded to one from the recordings in the agar substrate. The same turning angle probability distribution was used whether the larva is inside or outside the patch. The initial position was picked at random in each simulation, but always inside one of the two food patches to match the experiments.

With biased turns toward the food
Request a detailed protocol

Except for the choice of turning angles, the model was the same as the one described above. The biased choice of turns toward the food followed the implementation in Tao et al., 2020. After drawing a turning angle from the von Mises probability distribution, the turn direction was chosen such that the larva points toward the patch center with probability Pbias that depends on the distance between the current position relative to the center of the closest patch (Figure 5B). When the simulated larva was further than 60 mm away from the closest patch center, no bias was applied in the turning direction since the data were very sparse in this region (most larvae never crawled such long distances away from the patch of food in the experiments). Each turn was defined by a set of three points {p1,p2,p3} where p1 is where the turn initiates, p2 is the end location of a left turn, and p3 the end location of a right turn. Three movement vectors that characterize the turn options (to the left or to the right) were defined as:

v1=p2p1
v2=p3p1
u=-p1

We next calculated the angle θ the larval trajectory makes with the inward vector u when turning to the left (p2) or to the right (p3). The inward turn is the turn that results in the smallest θ (as shown in Figure 5A).

With more patches
Request a detailed protocol

We fixed the total surface area of food to be distributed in N patches as S=2πR2 , where R=25mm is the radius of the patches from the previous simulations and experiments. Then, the radius of each N th patch is given by R`=S/Nπ. The simulated larvae started within a random food patch, and were tracked for 50 min. The simulation parameters were kept the same as in the two patches model, except that the distances in the distance-dependent probability to turn inwards were adjusted for smaller patch radius, by multiplying the distance values by R`/R.

Model parameters (Tables 2–4)

Request a detailed protocol
Table 3
Parameters of corrected model in patchy substrates obtained in patchy substrate experiments.
Patchy substrateLarvaMean v inside (outside) (mm/s)Std v inside (outside) (mm/s)Pturn/s inside (outside)Ppause/s inside (outside)
Agar + sucroseRover0.60 (0.65)0.27 (0.25)0.037 (0.044)0.039 (0.0014)
Agar + sucroseSitter0.52 (0.57)0.28 (0.26)0.030 (0.030)0.088 (0.040)
Agar + yeastRover0.37 (0.44)0.19 (0.20)0.039 (0.068)0.17 (0.023)
Agar + yeastSitter0.26 (0.36)0.14 (0.17)0.025 (0.048)0.32 (0.053)
Agar + apple juiceRover0.44 (0.53)0.24 (0.26)0.026 (0.017)0.096 (0.076)
Agar + apple juiceSitter0.39 (0.48)0.21 (0.22)0.021 (0.031)0.13 (0.065)
Table 4
Statistical power and Cohen’s effect size of non-significant comparisons.
Figure 1
D – Avg. number of turns per minPower (1 − β)Cohen’s size effect (d)
RoverAgarSucrose0.240.34
RoverSucroseYeast0.480.52
SitterSucroseYeast0.470.51
G – Handedness
RoverAgarSucrose0.13−0.22
RoverAgarYeast0.050.01
RoverSucroseYeast0.190.31
SitterAgarSucrose0.140.25
SitterAgarYeast0.080.16
SitterSucroseYeast0.08−0.15
Figure 3
C – Crawling speedPower (1 − β)Cohen’s size effect (d)
Sucrose
RoverInOut0.14−0.34
SitterInOut0.20−0.44
D – Avg. number of turns per min
Sucrose
RoverInOut0.29−0.56
SitterInOut0.05−0.02
F – Handedness
Rover
SucroseInOut0.11−0.30
YeastInOut0.10−0.24
Apple juiceInOut0.150.31
Sitter
SucroseInOut0.06−0.15
YeastInOut0.06−0.14
Apple juiceInOut0.16−0.39
G – Crawl dist. 5 min
Apple juiceRoverSitter0.200.34
H – Fraction of time spent inside patch
SucroseRoverSitter0.080.20
Apple juiceRoverSitter0.32−0.40
Figure S3
Apple juice
Crawling speed
Power (1 − β)Cohen’s size effect (d)
InRoverSitter0.610.60
OutRoverSitter0.320.40
Avg. number of turns per min
InRoverSitter0.530.55
OutRoverSitter0.090.18
Fraction of pauses
InRoverSitter0.32−0.40
OutRoverSitter0.070.10
Figure 4
Anosmic
B – Crawling speed
Power (1 − β)Cohen’s size effect (d)
SucroseInOut0.51−0.56
YeastInOut0.23−0.39
C – Avg. number of turns per min
SucroseInOut0.10−0.19
YeastInOut0.32−0.50
E – Handedness
SucroseInOut0.100.24
YeastInOut0.07−0.16
Apple juiceInOut0.450.58
F – Fraction of time spent inside patch
SucroseYeast0.260.40
SucroseApple juice0.44−0.50
Figure 6
C – Fraction of time spent inside patch – eight patchesPower (1 − β)Cohen’s size effect (d)
SucroseRoverSitter0.14−0.31
YeastRoverSitter0.310.45
E – Fraction of visited patches
Rover
Sucrose2 patches8 patches0.050
Sitter
Sucrose2 patches8 patches0.070.15
Yeast2 patches8 patches0.56−0.62

Data availability

All data generated or analyzed during this study are included in the manuscript and supporting files 1 and 2. Source data files have been provided formal experimental data: Figures 1, 3, 4 and 6. Code available on GitHub: https://github.com/comp-neural-circuits/adaptation-of-drosophila-larva-foraging (copy archived at swh:1:rev:d501cd1f0df3df2fda0f286d58e98ee0f72b4898).

References

  1. Book
    1. Branch A
    2. Shen P
    (2017) Central and peripheral regulation of appetite and food intake in Drosophila
    In: Harris RBS, editors. Appetite and Food Intake. Boca Raton: CRC Press. pp. 17–38.
    https://doi.org/10.1201/9781315120171
    1. Stephens DW
    2. Charnov EL
    (1982)
    Behavioral ecology and sociobiology optimal foraging: some simple stochastic models
    Behavioral Ecology and Sociobiology 10:251–263.

Decision letter

  1. Matthieu Louis
    Reviewing Editor; University of California, Santa Barbara, United States
  2. K VijayRaghavan
    Senior Editor; National Centre for Biological Sciences, Tata Institute of Fundamental Research, India
  3. Mason Klein
    Reviewer; University of Miami, United States

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Adaptation of Drosophila larva foraging in response to changes in food resources" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and K VijayRaghavan as the Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Mason Klein (Reviewer #4).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1. Defining a potential contribution of taste to foraging. Is it known that yeast is 'unnoticeable' by taste? Did the authors try or consider testing the Gr43a mutant on these substrates? When the larvae are pausing, are they trying to eat the substrate?

2. Defining the role of other modalities: On patchy substrates, is the border completely smooth or could the larvae also sense the border as a rough edge? Are there effects from mechanosensation due to the different preparation of the patch edge (printing for yeast versus cutting for juice and sucrose)? Does diffusion of patch edge play a role for apple juice and sucrose?

3. It would be important to highlight the contribution of the model. Could the model be used to guide new experiments or to suggest new ideas? Would the integration of chemotaxis in the model improve the reproduction of specific behavioral features observed experimentally? Related to this point, one aspect where the work could be strengthened is by highlighting or speculating about where the differences in patch residence times between the model and data might arise.

4. Technical concerns related to the methodology:

– Define whether the simulation's initial conditions matter.

– Specify whether the model samples a steady-state/ ergodic distribution of trajectories or if time-dependent properties should be accounted for.

– Compared to other models, no patch depletion is taken into account. Wouldn't this affect the leaving rates in small patches especially and more so than larger patches?

5. Statistics: For an animal that tends to have a very high variance in its behavior, the number of larvae used in each experiment seems to be pretty low. This limitation should be acknowledged in the conclusions drawn in the paper. More specifically, low sample sizes can lead to false negatives. To address this concern, the authors should perform a statistical power analysis to increase the confidence in the "not statistically significant" results.

The reviewers agreed that points 3 and 4 could be mainly addressed through computational work and/or editing. If any of the other concerns can be adequately addressed without additional experimental work, we will generally support this option so that it reduces the efforts entailed by the revision.

Reviewer #2 (Recommendations for the authors):

I have a few suggestions that I believe could strengthen the manuscript:

– Please comment on the simulation's initial conditions and if they matter? From the trajectories shown in Figures 2 and 5, it seems like there could be some finite duration effect

– Are you sampling a steady-state/ergodic distribution of trajectories or are there time-dependent properties?

– Could you show how well single-trial simulation data matches single-trial experimental data with respect to trajectory features?

It would be helpful for the reader if the model predictions from Figure 2 could be articulated more explicitly in the section discussing Figure 2. It is mentioned later, but it would be useful to summarize these results before the next section of experiments.

– What are the navigational metrics in apple juice on homogeneous patches?

– Does diffusion of patch edge play a role for apple juice and sucrose? Or are there effects from mechanosensation due to the different preparation of the patch edge (printing for yeats versus cutting for juice and sucrose)?

– Please explain the exact nature of the defect in anosmic larvae. The results appear somewhat confusing otherwise. Is it known that yeast is 'unnoticeable' by taste?

– Compared to other models, no patch depletion is taken into account. Wouldn't this affect the leaving rates in small patches especially and more so than larger patches? Relatedly, the paper could benefit if the phenomenological results of the model could be connected to optimal foraging theory, as the authors cite the marginal value theorem. However, this might be beyond the scope of this work.

Reviewer #3 (Recommendations for the authors):

I like the paper overall, the main results are interesting and well supported. I have some issues with the logic of the simulations and how they inform the experiments. My only "major" change would be to rearrange the paper a bit and put the simulations after what is now Figure 3, maybe even after all of the experimental data figures.

General/broader comments:

(1) Using 30 or fewer larvae for each experiment type feels quite low in my experience. I'm not suggesting performing a bunch of new experiments (I almost never do in reviews), but I think it's important to be careful with claims about any comparison being "the same" (specific instances pointed out below) -- finding "n.s." does not mean two things are the same, especially when a small number of data points is used in the comparison.

(2) The introduction feels like it's missing at least a whole paragraph that should probably be explaining to a non-fly, non-foraging researcher why this study is important and useful. I study fly larvae too, and it's certainly exciting and interesting to me, but I think some more effort to draw in other researchers is warranted.

(3) The simulations. I like the method of using results from actual experiments to make the probability distributions that the simulation draws from in its random-walk-style tracks. I'm sure the simulations are done properly and make sense. What dampens my enthusiasm for them in this specific paper is that they don't seem to be informing the experiments or suggesting new ideas to the extent that they could. Numerous statements that start with "as predicted by the model" make it sound like the simulations are providing some kind of insight, but often that doesn't seem true -- like the model predicting the animals spend more time in yeast patches vs. other food, of course they do, because the model makes them crawl slower and pause more in yeast. It's a cool result, but it comes from your real experiments. It's just not very clear what benefit the simulations have here. I don't mean they aren't useful or you should remove them, I mean it should be much more clear what they are for:

– they are used as a check to confirm what is very strongly implied by your own experiments. If you include the modulations from crawling in isotropic substrates, and you include chemotaxis, and then the simulations match your experiments pretty closely, then you have successfully identified the important behavioral features, right? –

I like studying larva behavior and I like building simulations of larva behavior too, but something about the flow and logic of how they are deployed here feels off. I'll try to be more specific in the detailed comments.

(4) When the larvae are pausing, are they trying to eat the substrate? It seems like with third instar larvae and a pretty low % gel, they could if they wanted to.

(5) You talk about distance traveled throughout the paper, but I don't think mention displacement. Wouldn't that be important for describing motion too? Like the animals diffusing away from their start point vs. traveling a lot of distance, these aren't quite the same thing when exploring an environment for food. In particular, when you mention reduced turn and increased pause rate -- don't these do opposite things? Turning less frequently means the animals leave a space faster and the higher pause rate keeps them there longer right? They are often mentioned together in the paper, with the net effect being that larvae spend longer times in such regions, and I trust that's true, but that's because the pausing rate (and duration of pauses) carries more weight in this case?

(6) Does what the larvae do in between turns (or pauses) matter? The simulations draw them as straight lines, but don't larvae drift towards preferable odors? (I could be remembering that wrong, but I think I saw that in a Louis lab paper at some point).

(7) Is there a reason the simulations are never referred to as Monte Carlo simulations, or as (modified) random walks? Those are two pretty important reference points, especially for a general audience I think?

More detailed comments:

(1) line 26: what does "permanence time" mean? The time they spend in the patches?

(2) line 35: "all living organisms need to explore their surroundings…" That doesn't sound true. Not all living organisms even move. And some animals in the ocean just sit around and let food come to them.

(3) line 53: "gradients of light" --> "gradients of light intensity"? (light has a lot of properties).

(4) line 53: the paper used as a thermotaxis reference isn't really a thermotaxis paper, if you want one from the same group I would choose Luo, Gershow, et al. 2010, J. Neurosci. (For the record, I am an author on the one you already cite, not on the one I'm suggesting you add).

(5) line 94-95. This sentence is hard for me to understand.

(6) line 95-98. This is stated several times (basis for neuronal circuits), but is it explained why this is true? (in the Discussion section?)

(7) line 103-104. This section title maybe needs a comma and "pauses frequency" should say something else?

(8) line 113. 50 minutes seems like a long time for these experiments. If I'm back-of-the-enveloping something moving a 1 mm/s, turning every 20 s… wouldn't they leave the arena a lot faster than that? Does the arena have walls? Do their pauses last a really long time on average?

(9) Figure 1.

In 1B. Could the pauses be marked too? Maybe with open/empty circles instead of a solid/filled one? The pausing seems so important in this paper, it seems weird to leave it out.

In 1A, could there be labels for the camera and the red thing (the IR filter?).

In 1B (this figure and others), I think you mean 3pi/2 for the lower quadrant in the turn angle distributions, not 2pi/3?

Also, I don't think you need the same scale bar three times for the three panels.

Could it maybe be more clear that 1F is kind of a summary/consequence of C/D/E? It's kind of the main summary statistic here.

Some of the panels have four * symbols, but the p-value for that is not given in the caption (this figure and others).

1D: (here and elsewhere) when the turning rate is calculated, I get that turns in just the number of turns, but what's in the denominator? The actual time elapsed, or only the time during forward crawling (i.e., the total time when they are could start a turn, excluding when they are already turning or paused).

Maybe the panels and overall figure could be a bit larger? It's a little hard to read at this size.

(10) line 156. would it make sense to do a significance test for your strong handedness? The very long experiments help you here because you measure a lot of turns for individuals, but presumably, a 50/50 binomial distribution of turns would yield some "strong handedness" animals too right?

(11) line 164-167. Are you sure about that? Your data in S1F might turn out to be significant with more data taken? Similarly for the yeast substrate in Figure S1B (mentioned in line 223). I would hesitate with claims about non-significance when comparing a pretty small number of larvae.

(12) Figure 2

Why would only 30 larvae be simulated? It's a simulation, you can simulate millions of them if you wanted?

(13) Line 226ff. This is where the simulation logic seems weird to me. Maybe the order of the sections in the paper throws it off. This section laid out a simulation method, then ran simulations with patches of food substrates. But at this point you know the simulation is missing chemotaxis, and you know that real animals can smell things and move towards/away from them. I'm not sure what the point is of setting this up and then "testing the predictions of the model" with your (very cool) food patch experiments. Similarly in line 281ff, saying "the model predicted" that larvae spend more time in certain substates, doesn't feel like the right statement, because the model is only telling you what you directly put in yourself. This part is written as if the model is helping you figure out what's going on, but that doesn't seem true in this part. I think it might be better to put the simulation stuff after you have looked at the experimental patch data, then build up its capabilities if you want, or just include chemotaxis right away. The in between thing with only using the isotropic data kind of feels like a waste of time in this paper, and really interrupts the flow of some really interesting experimental results.

(14) line 232-233. Also, an oddly phrased section heading.

(15) line 237. What is the reason for apple juice showing up here suddenly? The other substances seem more fundamental -- what chemicals are in apple juice that makes it attractive? Why weren't there isotropic experiments done with an apple juice substrate?

(16) Figure 3.

3B, could this have a legend that says what the white and black circles are? (I know it's in the caption, but it would help to put it in both places).

3I. Could this be drawn more clearly with labels? (the part defining the angles). To understand what you are doing here I had to read the methods section, draw my own picture, then go look up the Tao reference (which has a better picture). Also, a general issue/concern with this definition of "towards the center". Wouldn't there be many circumstances where either a right or a left turn (of the same size) would point the animal more towards the center? Is the rate that the larvae turn different as they crawl away from a food source? Would anything change if you defined "towards the center" as the larva picking the direction that points them more toward the center than the other direction? It seems like some crawling directions would have more crucial turning decisions in them, like crawling perpendicular to the vector from food patch center (matters the most) vs. crawling parallel (directly away from the food, doesn't matter which way you turn). Pooling the turn decisions into those subsets (like you do with distance away) might make the effect more pronounced.

(17) line 300. I think you mean J instead of B?

(18) line 304. Why is this surprising? Larvae can smell, you cited a bunch of papers earlier that show this in detail.

(19) line 329 + Figure 4 -- again, a claim about non-significance with a small data set is maybe not be warranted.

(20) line 384. "Therefore…" Again, this implies the simulation result is telling you something, but you already knew this before simulating, based on the real anosmic data.

(21) line 390-392, this seems like a really interesting idea -- could this be expanded/discussed further in the discussion part of the paper?

(22) line 426ff. I think it would help to more clearly state which simulation results are obvious based on what you put in the simulation, and which gives you something unexpected.

(23) line 480-482. Here too, larvae spending less time in less nutritious patches, isn't that a direct result of putting your empirical result into the simulation? They crawl faster and pause less, doesn't this have to be true?

(24) line 636: is that hours after egg laying or after eclosion?

(25) line 641ff: I don't quite follow -- you are choosing a lower frame rate in order to prioritize spatial resolution because the camera doesn't run faster when recording at 2048x2048?

(26) line 690: when is instantaneous turn rate used? You are generally finding the total for long trajectories right?

(27) line 695: What is the distance dimension parameter? Could you briefly explain what it means?

(28) line 699: typo with the brackets vs. parentheses.

https://doi.org/10.7554/eLife.75826.sa1

Author response

Essential revisions:

1. Defining a potential contribution of taste to foraging. Is it known that yeast is 'unnoticeable' by taste? Did the authors try or consider testing the Gr43a mutant on these substrates? When the larvae are pausing, are they trying to eat the substrate?

Foods are complex in nature: they have different texture, odour and taste. The goal of our paper was not to investigate how different sensory modalities affect the foraging strategy, but rather to understand how different foods, with their combined properties, modulate foraging. Taste is however a key sensory input that contributes to the decision of an animal to feed, so it is indeed interesting to discuss its role in the behavioural adaptation we observed. We have added a new section where we discuss how taste and smell contribute to the perception of food including yeast.

Line 634: “An experiment using the gustatory sweet sensor Gr43a mutant on sucrose, which is not volatile and does not produce smell, could help discerning the contribution of taste at the border of the patch (Fujishiro et al. 1984; Marella et al., 2006; Miyamoto et al. 2013; Wang et al.,2004; Mishra et al.,2013). For yeast, the lack of smell completely changed the response of the larvae, which did not show differences inside and outside the patch for most foraging parameters (Figure 4B, C, E, G). In this instance, taste was not sufficient to retain larvae inside the yeast patch (compare Figure 3H with Figure 4F) even though several gustatory receptors have been shown to be activated by yeast metabolites (Wisotsky et al., 2011, Ganguly et al.,2017, Croset et al., 2016).”

To address the question of what the larvae do during the pauses, we recorded high magnification movies of larvae that paused extensively in yeast. We found that the majority of these larvae were actually completely motionless and only showed movements of food in their gut. We have added a description and video 1 in results, line 165: “Most pausing larvae were completely still, except for internal movements in their gut, suggesting they were digesting (Video 1).”

In discussion, line 579: “They also made more pauses, with the majority remaining stationary, except for internal gut movements (Video 1), which suggested that they were digesting the yeast. The pauses were rarely observed in sucrose, which is metabolized more quickly than yeast, even when mixed with agar (Figure 1E).”

2. Defining the role of other modalities: On patchy substrates, is the border completely smooth or could the larvae also sense the border as a rough edge? Are there effects from mechanosensation due to the different preparation of the patch edge (printing for yeast versus cutting for juice and sucrose)? Does diffusion of patch edge play a role for apple juice and sucrose?

The point raised by the reviewers regarding the edge of the food is quite relevant. To test if the edge is sufficient to generate changes in behaviour comparable to what we observed with food, we performed new experiments and included two controls. First, we built the arena exactly as we did in the two patch experiments, but we filled the circular patches with agar only. In this situation there is a smooth edge between the “out” and “in” area. Second, we imitated the texture of the yeast by spreading a drop of ultrasound gel (see Methods, page 35 line 796). Although the viscosity is not exactly the same, this control experiment helps us understand how the larvae behave when crawling across an edge between a smooth and viscus substrate (the gel) and a flat and harder one (the agar).

The results have been added as Figure 3-supplement 2 and they show that: Line 339:

“To control for possible mechanosensory effects due to the border edges, we prepared new arenas with patches that contained no nutrients, either using the same agar that composed the rest of the arena, or using ultrasound gel (Methods). Larvae in the agaragar or the agar-gel border did not show any changes in their preference to turn inwards, towards the patch center, confirming that the behavioral change observed in response to food is specific (Figure 3—figure supplement 2).”

Diffusion is certainly happening at patch edges for sucrose and apple juice. To account for changes in behavior that depend on diffusion, we have also added an analysis of turn probability in anosmic larvae:

Line 408: “Taste very likely influences the probability that larvae remain in the patches. To control for the effect of diffusion of nutrients (sucrose and apple juice) at the edge of a patch, we evaluated the maximum distance at which an increased fraction of turns toward the center was significantly different when compared to the yeast non-responsive anosmic control. At a distance greater than 0.5 cm from the edge, anosmic larvae on sucrose, apple juice, and yeast were indistinguishable, suggesting that diffusion has a limited impact on behavior (Figure 4, figure supplement 1)”

And a paragraph in discussion, line 629: “In anosmic larvae, the fraction of turns towards the center for sucrose and apple juice patches was only higher compared to the one for the yeast patch (where there was no food effect) within the first half centimeter outside the patch, suggesting that the impact of diffusion could be significant only in that region (Figure 4—figure supplement 1). ”

In addition to this, to control for the effects of diffusion over time, we added a new analysis and a new supplementary figure.

Line 415: “Finally, to control for possible effects of diffusion over time, we compared the fraction of turns towards the center in the first and second half of the experiment. For most distance and nutrients, the two distributions were not significantly different (Figure 4—figure supplement 1B).

3. It would be important to highlight the contribution of the model. Could the model be used to guide new experiments or to suggest new ideas? Would the integration of chemotaxis in the model improve the reproduction of specific behavioral features observed experimentally? Related to this point, one aspect where the work could be strengthened is by highlighting or speculating about where the differences in patch residence times between the model and data might arise.

The model plays a crucial role in this study. Its first advantage is to extract essential features that explain how larvae move across different environments. Second, the model enabled us to simulate a much wider range of conditions, parametrized by food fragmentation level, than possible experimentally (Figure 5E-G). For example, although we originally developed the model based on the 2-patch experiments, the model predicted larval behavior on 8-path experiments without further adjustments which was then confirmed in real larvae. Therefore, the model was already used to guide new experiments.

See for e.g. line 538: “As predicted by the model, both rovers and sitter spent half of the time inside patches when the area of food was divided in eight compared to two patches (Figure 5F and 6C,D).”

Our model is phenomenological rather than mechanistic. Although it does not explicitly include chemotaxis, by incorporating “biased turns” where the probability of turning towards a patch depends on the distance from its center, the model already implicitly includes the effect of chemotaxis. To capture further aspects of the data (including patch residence times) will likely require a more detailed mechanistic model where different sensory modalities are included, as opposed to our phenomenological model based on behavioural statistics. This is very interesting but beyond our scope (and goals). Nonetheless, our phenomenological model can be used as a term of comparison and benchmark for future bottom-up mechanistic models that could explicitly represent sensory information, internal variables (e.g. short-term memory traces, satiety level) and neural circuit components.

We have rephrased several sections to highlight the predictions of the model: Line 257: “We therefore compared the model predictions on foraging efficiency in patchy environments with behavioral experiments.”

Line 473: “We next used our model to investigate how a further fragmentation of the food patches affects the ability of larvae to stay in patches where they can feed.” […] “In total, this would represent 1575 hours of experiment, highlighting the advantage of the model.”

Line 651: “One of the strengths of our phenomenological model is that it incorporates a modular organization of foraging that could reflect how the crawl and turn modules are controlled. First, we modelled a stochastic search where no information regarding food is available outside of the current location, because food is absent or because the larvae cannot sense it. This corresponds to an autonomous search behavior implemented by circuits located in the ventral nerve cord without input from the brain (Berni et. al 2012; Sims et al. 2019). Second, we incorporated a goal-directed navigation that allows larvae to return to the food. Our phenomenological model includes a distance-dependent probability to turn inwards that mimics the effect of chemotaxis (when present), as much as any other possible mechanism that contributes to the turning probability. As a consequence, we observed that simulated larvae, even when the resources are fractioned in eight patches, could stay inside the food patch for longer periods, in line with experimental observations (Figure 5 and Figure 6). The model could be improved by setting the turning properties outside the patch to match as closely as possible experimental observations. To this end, we could consider studies of larvae crawling in different attractive gradients, where the changes in turning probability and angle, including weathervaning, have been investigated in relation to precise spatio-temporal information of odorants (Louis et al., 2008; Gomez-Marin et al., 2011; Davies et al.,2015). It would also be helpful to have information about other attractive gradients, like taste, to know if a common set of mechanisms is used regardless of the sensory modality. Using this information, our model could be used to investigate how crawling speed and turning properties are controlled via descending pathways from the brain (Tastekin et al. 2018; Jovanic et al. 2019). Finally, in the presence of nutrients, our model adjusts movements to stay on the food patch. The concerted decrease in turning rate and crawling speed and the increase in the number of pauses suggests that a neuromodulatory depression of movement (Marder, 2012) could be relevant in this phase. It would be interesting to investigate more generally how neuromodulators influence the decision to remain or explore new food resources in relation to the resources available and the larval motivational state.”

4. Technical concerns related to the methodology:

– Define whether the simulation's initial conditions matter.

The initial conditions and the time duration of the simulation were chosen in line with experiments, because our main goal is to make a quantitative comparison between the two. A different choice of initial conditions, for example, starting outside of a food patch, could lead to different statistics since the area without food is larger than the patches with food, but this would require new experiments to make the comparison to the model. We added these details in the text:

Line 900: “The initial position was picked at random in each simulation, but always inside one of the two food patches to match the experiments.”

Line 925: “The simulated larvae started within a random food patch, and were tracked for 50 minutes.”

– Specify whether the model samples a steady-state/ ergodic distribution of trajectories or if time-dependent properties should be accounted for.

– Compared to other models, no patch depletion is taken into account. Wouldn't this affect the leaving rates in small patches especially and more so than larger patches?

Even in the eight-patches experiments, the amount of food in a single patch cannot be depleted in the 50 minutes of recording. For this reason, effects of depletion or food degradation have not been included in the model.

5. Statistics: For an animal that tends to have a very high variance in its behavior, the number of larvae used in each experiment seems to be pretty low. This limitation should be acknowledged in the conclusions drawn in the paper. More specifically, low sample sizes can lead to false negatives. To address this concern, the authors should perform a statistical power analysis to increase the confidence in the "not statistically significant" results.

The low number of larvae is compensated by the 50 minutes recording time. We have added table 4 where we have calculated the power and Cohen’s size effect of all non-significant comparisons in our behavioural experiments.

The reviewers agreed that points 3 and 4 could be mainly addressed through computational work and/or editing. If any of the other concerns can be adequately addressed without additional experimental work, we will generally support this option so that it reduces the efforts entailed by the revision.

Reviewer #2 (Recommendations for the authors):

I have a few suggestions that I believe could strengthen the manuscript:

– Please comment on the simulation's initial conditions and if they matter? From the trajectories shown in Figures 2 and 5, it seems like there could be some finite duration effect

– Are you sampling a steady-state/ergodic distribution of trajectories or are there time-dependent properties?

– Could you show how well single-trial simulation data matches single-trial experimental data with respect to trajectory features?

In simulations and in experiments, larvae started from a food patch, since larval eggs are laid on fruit, and then were tracked for the following 50 minutes. Rather than measuring the ergodic properties of the simulated walk, in our simulations (by investigating the effects of initial conditions and duration), we aimed to quantitatively compare the model to experimental data.

We now added this detail in the text:

Line 193: “In ecological conditions, the fruit on which Drosophila eggs are laid and on which the larvae forage decays over time.”

Line 900: “The initial position was picked at random in each simulation, but always inside one of the two food patches to match the experiments.”

Line 925: “The simulated larvae started within a random food patch, and were tracked for 50 minutes.”

Regarding single trials, Figure 6A and 6B show that simulated and measured paths resemble each other at a qualitative level. Due to the variability between animals, quantitative comparisons at the single-trial level require models based on single larvae, and not on average measures across many larvae as in our current model. The current model takes parameters such as pause duration and step length from the homogeneous experiments, whereas the bias in turning towards patch centre comes from the two patches experiment. Fitting a model on a single larva would therefore require a new experiment where the same individual is tracked in different arenas. We agree this is very interesting but leave it for future work.

It would be helpful for the reader if the model predictions from Figure 2 could be articulated more explicitly in the section discussing Figure 2. It is mentioned later, but it would be useful to summarize these results before the next section of experiments.

They have been clarified, thanks for the suggestion.

Line 252: “Thus far, our model predicts that, in patchy environments, larvae spend a relatively small proportion of time inside patches (approximately 1% for sucrose and 3% for yeast) while exploring takes up most of their time with a significant energy cost.”

– What are the navigational metrics in apple juice on homogeneous patches?

We did not perform experiments with apple juice in a homogeneous arena. We included apple juice only when we performed experiments with two and eight food patches. We now explain:

Line 268: “We also performed experiments using apple juice as a nutrient, motivated by the fact that it is ecologically relevant and that, unlike sucrose, the fructose contained in apple juice is volatile, which makes it detectable by smell and not only by taste.”

– Does diffusion of patch edge play a role for apple juice and sucrose? Or are there effects from mechanosensation due to the different preparation of the patch edge (printing for yeats versus cutting for juice and sucrose)?

To investigate the role of diffusion, we re-analysed the data by separating it into the first and second half of the experiment (Figure 4—figure supplement 1). To investigate the role of mechanosensation, we performed new control experiments to test the impact of the edge (Figure 3—figure supplement 1E and F) and calculated the fraction of turns towards the center in anosmic larvae. See above for details.

– Please explain the exact nature of the defect in anosmic larvae. The results appear somewhat confusing otherwise. Is it known that yeast is 'unnoticeable' by taste?

We have clarified what the defect in anosmic larvae is, by explaining the Orco receptor. Line 368:

“Thus, we repeated the patchy experiments with mutant anosmic larvae, where Orco, the obligatory co-receptor for all olfactory neurons, apart the CO2 sensing ones, is mutated (Vosshall and Stocker, 2007) and tested if they show the same distant-dependent bias when exploring the patchy substrate.”

Metabolites from yeast are known to be sensed by taste but not enough information is available about the exact nature of sensing. We have added information in the discussion (see above).

– Compared to other models, no patch depletion is taken into account. Wouldn't this affect the leaving rates in small patches especially and more so than larger patches? Relatedly, the paper could benefit if the phenomenological results of the model could be connected to optimal foraging theory, as the authors cite the marginal value theorem. However, this might be beyond the scope of this work.

Larvae eat relatively little compared to the amount of food available to them in the patches, even in the 8 patches arena, so we have not considered patch depletion in the model.

Adding a measure of optimality would require new analyses and model extensions, therefore we consider applications to optimal foraging theory beyond the scope of this

Reviewer #3 (Recommendations for the authors):

I like the paper overall, the main results are interesting and well supported. I have some issues with the logic of the simulations and how they inform the experiments. My only "major" change would be to rearrange the paper a bit and put the simulations after what is now Figure 3, maybe even after all of the experimental data figures.

General/broader comments:

(1) Using 30 or fewer larvae for each experiment type feels quite low in my experience. I'm not suggesting performing a bunch of new experiments (I almost never do in reviews), but I think it's important to be careful with claims about any comparison being "the same" (specific instances pointed out below) -- finding "n.s." does not mean two things are the same, especially when a small number of data points is used in the comparison.

As noted above, the low number of larvae is compensated by the 50 minutes recording time. We note that this is much longer than what most other experiments tracking crawling larvae have achieved so far. We have added Table 4 where we calculated the power and Cohen’s size effect of all non-significant comparisons in our behavioural experiments. As expected, the power tends to be low, but it gives a better idea of the confidence of each comparison.

(2) The introduction feels like it's missing at least a whole paragraph that should probably be explaining to a non-fly, non-foraging researcher why this study is important and useful. I study fly larvae too, and it's certainly exciting and interesting to me, but I think some more effort to draw in other researchers is warranted.

We have edited the text to make it more clear why our results apply beyond larval crawling (see response to reviewer 2 above)

(3) The simulations. I like the method of using results from actual experiments to make the probability distributions that the simulation draws from in its random-walk-style tracks. I'm sure the simulations are done properly and make sense. What dampens my enthusiasm for them in this specific paper is that they don't seem to be informing the experiments or suggesting new ideas to the extent that they could. Numerous statements that start with "as predicted by the model" make it sound like the simulations are providing some kind of insight, but often that doesn't seem true -- like the model predicting the animals spend more time in yeast patches vs. other food, of course they do, because the model makes them crawl slower and pause more in yeast. It's a cool result, but it comes from your real experiments. It's just not very clear what benefit the simulations have here. I don't mean they aren't useful or you should remove them, I mean it should be much more clear what they are for:

– they are used as a check to confirm what is very strongly implied by your own experiments. If you include the modulations from crawling in isotropic substrates, and you include chemotaxis, and then the simulations match your experiments pretty closely, then you have successfully identified the important behavioral features, right? –

I like studying larva behavior and I like building simulations of larva behavior too, but something about the flow and logic of how they are deployed here feels off. I'll try to be more specific in the detailed comments.

Since our model is purely phenomenological and directly based on the experimental measures, it cannot offer explicit insights into the mechanisms that determine the animal’s behaviour. However, the model still produces relevant, quantitative results that have been used to guide our experiments, and that can, in turn, inspire future work. As we answer in the required revisions, we used the model’s predictions from the two-patch experiments to predict the behavior for the eight-patch experiments without further adjustments. The model also highlights the consistency of behavioural parameters across conditions: the values measured in isotropic substrates could still be used in the patch experiments where chemotaxis effects exist.

We also added a new section in the discussion to emphasize the benefits of using the model (line 651, already referenced in the answer to the other reviewers).

(4) When the larvae are pausing, are they trying to eat the substrate? It seems like with third instar larvae and a pretty low % gel, they could if they wanted to.

Pausing larvae are generally still, except for gut movements. We included a supplementary video as an example of this behavior.

Line 165: “Most pausing larvae were completely still, except for movements of food in their gut suggesting they were digesting (Video 1).”

(5) You talk about distance traveled throughout the paper, but I don't think mention displacement. Wouldn't that be important for describing motion too? Like the animals diffusing away from their start point vs. traveling a lot of distance, these aren't quite the same thing when exploring an environment for food.

In this paper we did not consider displacement because it is not very informative in the patches experiments where the patch size restricts the distance larvae explore. We also noticed that when they leave the patch, larvae generally stay in proximity (maximum 6 cm from centre of patch) before returning (Figure 3).

The current model samples step sizes from a normal distribution based on the data. Accounting for large displacements would require different statistics, for example a heavytailed distribution, as in Lévy random walks. It would be interesting to extend the model in this direction, however, since larvae moved slowly and with frequent changes of directions, we did not consider it necessary in the current work.

In particular, when you mention reduced turn and increased pause rate -- don't these do opposite things? Turning less frequently means the animals leave a space faster and the higher pause rate keeps them there longer right? They are often mentioned together in the paper, with the net effect being that larvae spend longer times in such regions, and I trust that's true, but that's because the pausing rate (and duration of pauses) carries more weight in this case?

What unifies all these differences in behaviour is the fact that larvae move less. So indeed they reduce turn frequency but that doesn’t mean that they go farther. They reduce the turn frequency and crawling speed at the same time. As a result, they do not leave the patch of food on which they were placed. We have added a comment in discussion:

Line 675: “The concerted decrease in turning rate and crawling speed and increase in the number of pauses, suggests that a neuromodulatory depression of movement (Marder, 2012) could be relevant in this phase.”

(6) Does what the larvae do in between turns (or pauses) matter? The simulations draw them as straight lines, but don't larvae drift towards preferable odors? (I could be remembering that wrong, but I think I saw that in a Louis lab paper at some point).

Yes, the reviewer is correct: weathervaning has been described in larvae during chemotaxis (Gomez-Marin and Louis, 2014). It is an orientation mechanism that occurs as larvae crawl in a gradient. In our case, it could have influenced the behaviour at the border of the patch but we have not included it in the model since we have no information about the gradient the larvae are exposed to. We have mentioned it in discussion line 665 and see above.

(7) Is there a reason the simulations are never referred to as Monte Carlo simulations, or as (modified) random walks? Those are two pretty important reference points, especially for a general audience I think?

We preferred not to use the term “Monte Carlo”, because it might generate confusion on how we conducted our simulations. In particular, the results of our simulations are restricted to initial conditions set on food patches, and for a time of 50 minutes, in line with the experiments. Typical Monte Carlo approaches instead operate in an ergodic / steady-state regime (i.e. completely random initial conditions and time that tends to infinity).

Regarding random walks, they are now mentioned in the introduction.

Line 46: “In environments where resources are abundant, animals will search and exploit them performing short movements in random directions, in patterns well approximated by Brownian random walks. When resources are sparse, and foragers have incomplete knowledge about their location, a more diffusive strategy is needed, with an alternation between short-range and long-range movements, which can be modelled as a Lévy random walk. Analysis of animal movements in the wild has demonstrated that environmental context can induce the switch between Lévy to Brownian movement patterns (Humphries et al., 2010), but the effective mechanisms behind the implementation of such behavior (e.g., cognitive capacity, memory) often remain elusive (Budaev et al., 2019).”

However, we did not refer directly to our model as a “random walk”, since that would then require us to formalize and explain the aspects in which our approach differs from classical random walk processes (e.g. dependence on location, biases in heading).

More detailed comments:

(1) line 26: what does "permanence time" mean? The time they spend in the patches?

This has been corrected.

(2) line 35: "all living organisms need to explore their surroundings…" That doesn't sound true. Not all living organisms even move. And some animals in the ocean just sit around and let food come to them.

Replaced by” most moving organisms”

(3) line 53: "gradients of light" --> "gradients of light intensity"? (light has a lot of properties).

True, this has been corrected.

(4) line 53: the paper used as a thermotaxis reference isn't really a thermotaxis paper, if you want one from the same group I would choose Luo, Gershow, et al. 2010, J. Neurosci. (For the record, I am an author on the one you already cite, not on the one I'm suggesting you add).

Thank you, reference added.

(5) line 94-95. This sentence is hard for me to understand.

Rephrased by replacing “weight” by “degree” of the behavioural response.

(6) line 95-98. This is stated several times (basis for neuronal circuits), but is it explained why this is true? (in the Discussion section?)

We have added a section to explain this:

Line 671: “Using this information, our model could be used to investigate how crawling speed and turning properties are controlled via descending pathways from the brain (Tastekin et al. 2018; Jovanic et al. 2019).” But see the whole new paragraph.

(7) line 103-104. This section title maybe needs a comma and "pauses frequency" should say something else?

We have slightly changed the title so it now reads:

Line 119: “Food quality controls the distance travelled modulating the speed and the frequency of pauses”. We believe it is now clearer.

(8) line 113. 50 minutes seems like a long time for these experiments. If I'm back-of-the-enveloping something moving a 1 mm/s, turning every 20 s… wouldn't they leave the arena a lot faster than that? Does the arena have walls? Do their pauses last a really long time on average?

When feeding, larvae are not interested in leaving the arena. When searching in homogenous agar, they do not like the dry edges of the arena and stay inside.

(9) Figure 1.

In 1B. Could the pauses be marked too? Maybe with open/empty circles instead of a solid/filled one? The pausing seems so important in this paper, it seems weird to leave it out.

We tried this, but the plot of trajectories becomes really crowded and it becomes very hard to make sense of the larval movement. On a more practical note, the first author of the paper is on maternity leave (why we added a second author to perform the analysis of the new experiments) and it is very difficult for her to explore other aspects of figure design now. We are sorry but we won’t be able to add them at this stage.

In 1A, could there be labels for the camera and the red thing (the IR filter?).

Thank you, done.

In 1B (this figure and others), I think you mean 3pi/2 for the lower quadrant in the turn angle distributions, not 2pi/3?

Yes, thank you, this has been corrected.

Also, I don't think you need the same scale bar three times for the three panels.

We wanted to emphasize that the scale was the same because the tracks look very different in length. We have only kept 3 scales in that figure.

Could it maybe be more clear that 1F is kind of a summary/consequence of C/D/E? It's kind of the main summary statistic here.

In line 166 we now write: “as a consequence”.

Some of the panels have four * symbols, but the p-value for that is not given in the caption (this figure and others).

This has now been corrected in all Figures

1D: (here and elsewhere) when the turning rate is calculated, I get that turns in just the number of turns, but what's in the denominator? The actual time elapsed, or only the time during forward crawling (i.e., the total time when they are could start a turn, excluding when they are already turning or paused).

We provide this information in the figure legend: “Average number of turns per minute registered in each trajectory”.

Maybe the panels and overall figure could be a bit larger? It's a little hard to read at this size.

We agree that it is very important for the reader to clearly see all aspects of the figures. We considered splitting each figure into multiple figures but decided against that because each figure relays its own message. We hope that since the journal publishes only online, the readers will be able to zoom into the different regions of each figure for details that are not easily seen in the printed version. We are making sure that the final figures have the maximum possible width to maximize readability.

(10) line 156. would it make sense to do a significance test for your strong handedness? The very long experiments help you here because you measure a lot of turns for individuals, but presumably, a 50/50 binomial distribution of turns would yield some "strong handedness" animals too right?

Yes, but we do not feel it is necessary here. A statistical test just confirms what we already see, and it is not important for the main message of the paper, which is focused on exploration strategies in different substrates.

(11) line 164-167. Are you sure about that? Your data in S1F might turn out to be significant with more data taken? Similarly for the yeast substrate in Figure S1B (mentioned in line 223). I would hesitate with claims about non-significance when comparing a pretty small number of larvae.

We have changed it accordingly:

Line 184: “Thus, when the resources are distributed homogenously, the genetic foraging dimorphism could not be detected.”

(12) Figure 2

Why would only 30 larvae be simulated? It's a simulation, you can simulate millions of them if you wanted?

The idea was to generate a representative sample comparable to the actual animals using the information provided by the animals. As explained in the paper: “each simulated larva had its own set of parameters for the turning angle distribution based on a single recorded larva.”

(13) Line 226ff. This is where the simulation logic seems weird to me. Maybe the order of the sections in the paper throws it off. This section laid out a simulation method, then ran simulations with patches of food substrates. But at this point you know the simulation is missing chemotaxis, and you know that real animals can smell things and move towards/away from them. I'm not sure what the point is of setting this up and then "testing the predictions of the model" with your (very cool) food patch experiments. Similarly in line 281ff, saying "the model predicted" that larvae spend more time in certain substates, doesn't feel like the right statement, because the model is only telling you what you directly put in yourself. This part is written as if the model is helping you figure out what's going on, but that doesn't seem true in this part. I think it might be better to put the simulation stuff after you have looked at the experimental patch data, then build up its capabilities if you want, or just include chemotaxis right away. The in between thing with only using the isotropic data kind of feels like a waste of time in this paper, and really interrupts the flow of some really interesting experimental results.

We have rewritten the section. The experiments were designed and performed in line with the classical approach used in modelling of Levy flights (here foraging) where only the diffusion is considered when evaluating the foraging in different environments. It is clarified in line 198: “The model predicted the fraction of time larvae spent inside patches of food, as a measure of food exploitation, if larvae only used the information about the substrate while foraging“. The diffusion models can be really good to quantify encounters at a large scale (as previous work on Levy flights has shown) but clearly they are insufficient to understand exploitation.

(14) line 232-233. Also, an oddly phrased section heading.

This has been re-organised.

(15) line 237. What is the reason for apple juice showing up here suddenly? The other substances seem more fundamental -- what chemicals are in apple juice that makes it attractive? Why weren't there isotropic experiments done with an apple juice substrate?

We now explain that we added apple juice because we wanted a substance that was ecologically relevant and attractive to the larvae. This is the case because it contains fructose and smells.

Line 268: “We also performed experiments using apple juice as nutrient, motivated by the fact that it is ecologically relevant and that, unlike sucrose, the fructose contained in apple juice is volatile, which makes it detectable by smell and not only by taste.”

(16) Figure 3.

3B, could this have a legend that says what the white and black circles are? (I know it's in the caption, but it would help to put it in both places).

Done

3I. Could this be drawn more clearly with labels? (the part defining the angles). To understand what you are doing here I had to read the methods section, draw my own picture, then go look up the Tao reference (which has a better picture). Also, a general issue/concern with this definition of "towards the center". Wouldn't there be many circumstances where either a right or a left turn (of the same size) would point the animal more towards the center? Is the rate that the larvae turn different as they crawl away from a food source? Would anything change if you defined "towards the center" as the larva picking the direction that points them more toward the center than the other direction? It seems like some crawling directions would have more crucial turning decisions in them, like crawling perpendicular to the vector from food patch center (matters the most) vs. crawling parallel (directly away from the food, doesn't matter which way you turn). Pooling the turn decisions into those subsets (like you do with distance away) might make the effect more pronounced.

Crawling is definitely more complex than just turning in one direction, but we were trying to find the simplest characterization to use in the model. The rate and angle of turns are likely to be highly dependent on gradients. We have a new section in discussion (see above). And we have re-drawn the figure.

(17) line 300. I think you mean J instead of B?

It is B but the sentence should say “shown in black.”

(18) line 304. Why is this surprising? Larvae can smell, you cited a bunch of papers earlier that show this in detail.

You are right, we deleted the word “Surprisingly”.

(19) line 329 + Figure 4 -- again, a claim about non-significance with a small data set is maybe not be warranted.

We have added the power analysis for all non-significant results. We can’t claim anything other than what we have quantified. We feel the data of anosmic larvae are quite convincing because the behaviour changes so obviously.

(20) line 384. "Therefore…" Again, this implies the simulation result is telling you something, but you already knew this before simulating, based on the real anosmic data.

We have deleted the entire sentence.

(21) line 390-392, this seems like a really interesting idea -- could this be expanded/discussed further in the discussion part of the paper?

Thanks, we have expanded the idea in the discussion (see above).

(22) line 426ff. I think it would help to more clearly state which simulation results are obvious based on what you put in the simulation, and which gives you something unexpected.

We agree that it is a good idea to guide the reader in that sense. We have added a few comments:

Line 484: “As expected…”

Line 488: “Interestingly, despite the small differences we previously quantified, our results showed that sitter larvae consistently spent more time inside yeast patches than rovers for each number of patches (Figure 5—figure supplement 1D).”

Line 496: “to understand how fractioning environment would affect exploitation, which is key for survival.”

(23) line 480-482. Here too, larvae spending less time in less nutritious patches, isn't that a direct result of putting your empirical result into the simulation? They crawl faster and pause less, doesn't this have to be true?

The result of the simulation is indeed not surprising, but we can consider it as a sanity check for the model and for the choice of fitting parameters.

(24) line 636: is that hours after egg laying or after eclosion?

“Since egg laying” has been added.

(25) line 641ff: I don't quite follow -- you are choosing a lower frame rate in order to prioritize spatial resolution because the camera doesn't run faster when recording at 2048x2048?

It wasn’t clear. Line 767 “We recorded the movies at 2 frames per second to obtain forward movement displacements and actual pause-turns that are recorded accurately rather than to include ‘flickering’ movements associated with peristaltic movements.”

(26) line 690: when is instantaneous turn rate used? You are generally finding the total for long trajectories right?

We no longer use this phrase.

(27) line 695: What is the distance dimension parameter? Could you briefly explain what it means?

In a Douglas-Peucker algorithm “ε, represents the maximum distance between the original points and the simplified curve” added in line 829.

(28) line 699: typo with the brackets vs. parentheses.

Corrected, thanks!

https://doi.org/10.7554/eLife.75826.sa2

Article and author information

Author details

  1. Marina E Wosniack

    Computation in Neural Circuits Group, Max Planck Institute for Brain Research, Frankfurt, Germany
    Contribution
    Conceptualization, Software, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2175-9713
  2. Dylan Festa

    School of Life Sciences, Technical University of Munich, Munich, Germany
    Contribution
    Formal analysis, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3803-1542
  3. Nan Hu

    Department of Zoology, University of Cambridge, Cambridge, United Kingdom
    Contribution
    Formal analysis, Investigation
    Competing interests
    No competing interests declared
  4. Julijana Gjorgjieva

    1. Computation in Neural Circuits Group, Max Planck Institute for Brain Research, Frankfurt, Germany
    2. School of Life Sciences, Technical University of Munich, Munich, Germany
    Contribution
    Conceptualization, Supervision, Funding acquisition, Validation, Methodology, Project administration, Writing – review and editing
    For correspondence
    gjorgjieva@tum.de
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-7118-4079
  5. Jimena Berni

    1. Department of Zoology, University of Cambridge, Cambridge, United Kingdom
    2. Brighton and Sussex Medical School,, University of Sussex, Brighton, United Kingdom
    Contribution
    Conceptualization, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Project administration, Writing – review and editing
    For correspondence
    j.berni@sussex.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5068-1372

Funding

Royal Society (105568/Z/14/Z)

  • Jimena Berni

Wellcome Trust (105568/Z/14/Z)

  • Jimena Berni

Max-Planck-Gesellschaft

  • Marina E Wosniack
  • Julijana Gjorgjieva

Alexander von Humboldt-Stiftung

  • Marina E Wosniack

The funders had no role in study design, data collection, and interpretation, or the decision to submit the work for publication. For the purpose of Open Access, the authors have applied a CC BY public copyright license to any Author Accepted Manuscript version arising from this submission.

Acknowledgements

MW was supported by a Capes-Humboldt postdoctoral fellowship. JG was funded by the Max Planck Society. JB was funded by a Sir Henry Dale fellowship from the Wellcome Trust and Royal Society 105568/Z/14/Z. The authors thank Dr. Bertram Gerber and Michael Schleyer for helpful discussions during early stages of this work and also Andre Maia Chagas for technical assistance, Omar D Perez for statistical recommendations, Nick Humphries, Marla Sokolowski, Carlotta Martelli, and Alex Kacelnik for constructive comments on the manuscript.

Senior Editor

  1. K VijayRaghavan, National Centre for Biological Sciences, Tata Institute of Fundamental Research, India

Reviewing Editor

  1. Matthieu Louis, University of California, Santa Barbara, United States

Reviewer

  1. Mason Klein, University of Miami, United States

Publication history

  1. Preprint posted: June 21, 2021 (view preprint)
  2. Received: November 24, 2021
  3. Accepted: November 29, 2022
  4. Accepted Manuscript published: December 2, 2022 (version 1)
  5. Version of Record published: January 6, 2023 (version 2)

Copyright

© 2022, Wosniack et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 518
    Page views
  • 131
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Marina E Wosniack
  2. Dylan Festa
  3. Nan Hu
  4. Julijana Gjorgjieva
  5. Jimena Berni
(2022)
Adaptation of Drosophila larva foraging in response to changes in food resources
eLife 11:e75826.
https://doi.org/10.7554/eLife.75826

Further reading

    1. Neuroscience
    2. Physics of Living Systems
    Sabrina A Jones, Jacob H Barfield ... Woodrow L Shew
    Research Article

    Naturally occurring body movements and collective neural activity both exhibit complex dynamics, often with scale-free, fractal spatiotemporal structure. Scale-free dynamics of both brain and behavior are important because each is associated with functional benefits to the organism. Despite their similarities, scale-free brain activity and scale-free behavior have been studied separately, without a unified explanation. Here we show that scale-free dynamics of mouse behavior and neurons in visual cortex are strongly related. Surprisingly, the scale-free neural activity is limited to specific subsets of neurons, and these scale-free subsets exhibit stochastic winner-take-all competition with other neural subsets. This observation is inconsistent with prevailing theories of scale-free dynamics in neural systems, which stem from the criticality hypothesis. We develop a computational model which incorporates known cell-type-specific circuit structure, explaining our findings with a new type of critical dynamics. Our results establish neural underpinnings of scale-free behavior and clear behavioral relevance of scale-free neural activity.

    1. Neuroscience
    Barna Zajzon, David Dahmen ... Renato Duarte
    Research Article

    Information from the sensory periphery is conveyed to the cortex via structured projection pathways that spatially segregate stimulus features, providing a robust and efficient encoding strategy. Beyond sensory encoding, this prominent anatomical feature extends throughout the neocortex. However, the extent to which it influences cortical processing is unclear. In this study, we combine cortical circuit modeling with network theory to demonstrate that the sharpness of topographic projections acts as a bifurcation parameter, controlling the macroscopic dynamics and representational precision across a modular network. By shifting the balance of excitation and inhibition, topographic modularity gradually increases task performance and improves the signal-to-noise ratio across the system. We demonstrate that in biologically constrained networks, such a denoising behavior is contingent on recurrent inhibition. We show that this is a robust and generic structural feature that enables a broad range of behaviorally-relevant operating regimes, and provide an in-depth theoretical analysis unravelling the dynamical principles underlying the mechanism.