Behavioral dissection of hunger states in Drosophila

Reviewer #1 (Public Review):

This study aims to identify the existence of hedonic feeding and to distinguish it from homeostatic feeding, in Drosophila. The authors use direct observation of feeding events, a novel automated feeding event detector, inventive behavioral assays, and genetics to separate out the ways that Drosophila interacts with food. Using two choice assays, the authors find an increased duration of interactions with high-concentration sugars under conditions of expected satiety, which is considered to be hedonic feeding.

Strengths:

The technical advances in the measurement of animal interactions with food will help advance the understanding of feeding behavior and motivational states.

The correlation of specific types of food interactions across satiation state, sex, and circadian time will help drive forward the understanding of the scope of an animal's goals with feeding, and likely their relation between species and eating disorders.

The assessment of mushroom body circuitry in a type of food interaction is helpful for understanding the coding of feeding control in the brain.

Limitations:

All feeding data presented in the manuscript are from the interactions of individual flies with a source of liquid food, where interaction is defined as 'physical contact of a specific duration.' It would be helpful to approach the measurement of feeding from multiple angles to form the notion of hedonic feeding since the debate around hedonic feeding in Drosophila has been ongoing for some time and remains controversial. One possibility would be to measure food intake volumetrically in addition to food interaction patterns and durations (e.g. via the modified CAFE assay used by Ja).

Some of the statistical analyses were presented in a way that may make understanding the data unnecessarily difficult for readers. Examples include:

1. In Table I the authors present food interaction classifications based on direct observation. These are helpful. However, the classification system is updated or incompletely used as the manuscript progresses, most importantly changing from four categories with seven total subcategories to three categories and no subcategories. In subsequent data analyses, only one or two of these categories are assessed. It would be helpful, especially when moving from direct observation to automated categorization, to quantify the exact correspondences between all of the prior and new classifications, as well as elaborate on the types of data that are being excluded.

2. The authors switch between a variety of biological and physiological conditions with varying assays, which makes following the train of reasoning nearly impossible to follow. For example, the authors introduce us to circadian aspects of feeding behavior to introduce the concept of 'meal' and 'non-meal' periods of the day. It is then not clear in which of the subsequent experiments this paradigm is used to measure food interactions. Is it the majority of the subsequent figure panels? However, the authors also use starved flies for some assays, which would be incompatible with circadian-locked meals. The somewhat random and incompletely reported use of males and females, which the authors show behave differently, also makes the results more difficult to parse. Finally, the authors are comparing within-fly for the 'control environment' and between flies for their 'hedonic environment' (Figure 3A and subsequent panels), which I believe is not a good thing to do.

3. Statistical analyses are not always used consistently. For example, in Figures 3B and C, post hoc test results are shown for sucrose vs. yeast interactions, but no such statistics are given for 3E and 3F, preventing readers from assessing if the assay design is measuring what the authors tell us it is measuring.

Reviewer #2 (Public Review):

Weaver et al. used video analysis of flies that were feeding in their previously developed FLIC assay to begin to dissect the mechanisms of feeding. FLIC or Fly Liquid Interaction Counter records electrical signals that are generated when a fly touches a liquid food substrate with its legs or proboscis or both. Using video data of the liquid food interactions in the FLIC assay allowed the authors to precisely identify what a fly is doing in the feeding chamber and what the relationship is between the flies' behavior and the electrical signal recorded in the assay. This analysis produced the first detailed behavioral profile of feeding flies and allowed the authors to categorize different types of feeding in the FLIC assay, from tasting food (using their legs) to fast and long feeding bouts (using their proboscis).

After establishing what FLIC signals correspond to the different types of feeding, they used these signals to examine the food choices of starved and sated flies when presented with a sugar-rich (2% sucrose) or protein-rich (2% yeast + 1% sucrose) liquid food source. To represent hedonic feeding, they also presented flies with a choice between super sweet (20% sucrose) food or protein-rich (2% yeast + 1% sucrose) liquid food. Although fully fed flies show no difference in the number of times they visit either food choice, the flies spend more time feeding during their visits on 20% sucrose food than they do on regular sugar and on the yeast food source, suggesting that 20% sucrose is a more pleasurable food source. To make sure this was not due to the higher caloric content of 20% sucrose, they also offered flies food with the same sweetness as 20% sucrose (2% sucrose + 18% arabinose) but without caloric content and food with the same caloric content but the sweetness of 2% sucrose (2% sucrose + 18% sorbitol). This experiment showed that sweetness was the driver for the longer feeding bouts, confirming that sweeter food is apparently perceived as more pleasurable. They also looked at the effect of starving flies on the hedonic drive and found that starvation increases the time spent feeding on pleasurable food, consistent with findings in mammals that homeostatic feeding affects the hedonic drive.

To begin dissecting circuits underlying hedonic drive, the authors used CaMPARI expression in all neurons. CaMPARI is a green fluorescent reporter that turns red in the presence of Ca2+ (a measure of neuronal activity) and UV exposure. Fully fed flies in the super sweet food choice condition showed more red fluorescence in the mushroom bodies. Inhibiting a subset of these neurons acutely shows that horizontal lobes are required for the increased duration of feeding bouts on super sweet food. These lobes are innervated by a cluster of DA neurons and inhibiting them also blocks the increased super sweet feeding times.

The data in the paper largely support the conclusions. The application of this tool to distinguish between homeostatic and hedonic feeding is innovative and very compelling. As proof of principle of the strength of their paradigm, the authors identify a distinct brain circuit involved in hedonic feeding. The methods established in the paper make a deeper understanding of feeding mechanisms possible at both a genetic and brain circuit level.

Some of the data presentation is dense and could be improved to make this paper easier for readers to understand.

1. The dissection of feeding into distinct behavioral elements and its correlation with electrical FLIC signals that allow interpreting feeding types is a fundamental new method to dissect feeding in flies. However, the categories of micro-behaviors in Table 1 are not intuitive.

2. The details for the behavioral data analysis are not clear and should be made more obvious. For example, how many males and females were used in each experiment? Were any of the females mated or were they all virgins? If all virgins, why not use mated females? Mating status may have an effect on the feeding drive. If mated and virgin females were used, are there any differences between them? Similarly, for diurnal feeding experiments, it is not immediately clear from the graphs how many animals were used and how the frequencies were obtained (Fig. 1F, presumably averages for each category per fly but that is inconsistent with the legend in the supplement for this figure). Why does the transition heat map not include all micro-behaviors (Fig. 1E, no LQ data which are significant in diurnal feeding)?

3. The CaMPARI images do not look great, particularly in the pan-neuronal condition (Fig. 5A). It would be useful to include the movie of the stack. Did any other brain regions show activity differences, such as SEZ or PI? These regions are known to be involved in feeding so it seems surprising they show no effect.

Author Response:

Reviewer #1 (Public Review):

1. All feeding data presented in the manuscript are from the interactions of individual flies with a source of liquid food, where interaction is defined as 'physical contact of a specific duration.' It would be helpful to approach the measurement of feeding from multiple angles to form the notion of hedonic feeding since the debate around hedonic feeding in Drosophila has been ongoing for some time and remains controversial. One possibility would be to measure food intake volumetrically in addition to food interaction patterns and durations (e.g. via the modified CAFE assay used by Ja).

We acknowledge that our FLIC assays address only one dimension of feeding behavior, physical interaction with liquid food. However, there is clear evidence that interactions are strongly predictive of consumption, and it would be technically difficult to measure feeding durations at the resolution of milliseconds using a Café assay. Nevertheless, we appreciate the spirit of this comment and agree that expanding our inference to other measures of feeding, as well as feeding environments, is an important next step. To this end, we will include measures of feeding on more traditional solid food, using the ConEx assay, and find that flies in the hedonic environment consume twice as much sucrose volume compared to flies in the control environment. These will be added as supplemental data (Figure 1 – Figure Supplement 1A), and the text will be updated to reflect our findings.

1. Some of the statistical analyses were presented in a way that may make understanding the data unnecessarily difficult for readers. Examples include:

a) In Table I the authors present food interaction classifications based on direct observation. These are helpful. However, the classification system is updated or incompletely used as the manuscript progresses, most importantly changing from four categories with seven total subcategories to three categories and no subcategories. In subsequent data analyses, only one or two of these categories are assessed. It would be helpful, especially when moving from direct observation to automated categorization, to quantify the exact correspondences between all of the prior and new classifications, as well as elaborate on the types of data that are being excluded.

We appreciate the feedback on our usage of the behavioral classification system and will make several adjustments to improve it. We will rename some of the behaviors to make them more intuitive (see Reviewer #2, comment #1), and update the main text and Table 1 to reflect these changes. We will update the text and figures to be more transparent about when we group subcategories into main categories for quantification and when we quantify all subcategories separately. Because these videos required manual scoring by an experimenter, after our initial characterizations we opted to score only main categories (which contain subcategories). We agree that it would be useful to quantify correspondence between subcategories and the automated FLIC signal. However, we believe this task is better suited for more advanced and automated video tracking software, and, incidentally, more sophisticated analysis of FLIC data, which has a very high-dimensional character that has yet to be properly exploited. At the moment, therefore, we are not confident in the ability to understand the data at the desired resolution.

b) The authors switch between a variety of biological and physiological conditions with varying assays, which makes following the train of reasoning nearly impossible to follow. For example, the authors introduce us to circadian aspects of feeding behavior to introduce the concept of 'meal' and 'non-meal' periods of the day. It is then not clear in which of the subsequent experiments this paradigm is used to measure food interactions. Is it the majority of the subsequent figure panels? However, the authors also use starved flies for some assays, which would be incompatible with circadian-locked meals. The somewhat random and incompletely reported use of males and females, which the authors show behave differently, also makes the results more difficult to parse. Finally, the authors are comparing within-fly for the 'control environment' and between flies for their 'hedonic environment' (Figure 3A and subsequent panels), which I believe is not a good thing to do.

We apologize for our difficulties conveying our inference, which was also noted by Reviewer #2. We will work hard to improve this component in the revision. With respect to the confusion about circadian feeding, we introduced circadian meal-times to complement starvation as a second (perhaps more natural) way to measure behaviors associated with hunger. Importantly, we do not use circadian meal-times beyond Figure 1; all subsequent FLIC experiments were conducted during non-meal times of day for 6 hours, which avoids confounding our data with circadian-locked meals even when we use starved flies. We will clarify this point in the revision.

The reviewer also points out that we make both within-fly and between-fly comparisons, which is a point that we note. Perhaps some concern arises, again, from the challenges that we faced in properly delineating our inferences about different types of feeding measures (and motivations). Inference about homeostatic feeding was made using within-fly measures, comparing events on sucrose vs. those on yeast. Inference about hedonic feeding was made using between fly measures (average durations of different flies on 2% vs. 20% sucrose). Treatment comparisons to control always used measures of the same type, such that inference was not made using between-fly measures for treatment and within-fly for control (i.e., all of our figure panels were either within-fly or between fly). We will clarify this in the revision.

Importantly, our approach to all experiments avoided confounding by used randomized design at multiple levels (e.g., randomizing control and hedonic environments to FLIC DFMs, alternating food choice sidedness in the DFMs), by ensuring that flies in both environments are sibling flies that came from the same vial environment before being tested, and by performing each experiment multiple times.

c) Statistical analyses are not always used consistently. For example, in Figures 3B and C, post hoc test results are shown for sucrose vs. yeast interactions, but no such statistics are given for 3E and 3F, preventing readers from assessing if the assay design is measuring what the authors tell us it is measuring.

We report p-values for two-way ANOVA interaction terms for all appropriate experiments. If (and only if) the interaction term is significant, we conduct post-hoc tests for more detailed statistical analysis and report the p-values. The reviewer points out that we do not perform post-hoc tests in figures 3E and 3F. These figures had a non-significant interaction term, and thus, we did not feel a post-hoc test was warranted.

Reviewer #2 (Public Review):

1. The dissection of feeding into distinct behavioral elements and its correlation with electrical FLIC signals that allow interpreting feeding types is a fundamental new method to dissect feeding in flies. However, the categories of micro-behaviors in Table 1 are not intuitive.

We agree and will update the Table, figures, and main text. Please see also our response to Reviewer #1, comment #1.

1. The details for the behavioral data analysis are not clear and should be made more obvious. For example, how many males and females were used in each experiment? Were any of the females mated or were they all virgins? If all virgins, why not use mated females? Mating status may have an effect on the feeding drive. If mated and virgin females were used, are there any differences between them? Similarly, for diurnal feeding experiments, it is not immediately clear from the graphs how many animals were used and how the frequencies were obtained (Fig. 1F, presumably averages for each category per fly but that is inconsistent with the legend in the supplement for this figure). Why does the transition heat map not include all micro-behaviors (Fig. 1E, no LQ data which are significant in diurnal feeding)?

We will clarify the number of flies and events for each behavioral experiment in Figure 1, and we will update the figure legend appropriately. We note that these behavioral datasets are non-overlapping, and each time we mention the number of events scored in the text, that number includes only “new” videos. Female and male flies for all experiments were mated, and we will clarify this in the main text and methods.

For the diurnal experiment in Figure 1F, we scored over 700 events from new (non-overlapping) video compilations and updated the number of flies and event number in the figure legend. The diurnal data we present in the supplement for this figure is a separate experiment conducted on 38 flies, intended only to demonstrate the circadian nature of fly feeding.

For the transition heat map, analysis of this sort seems to require a large amount of data to have sufficient power to return a transition matrix. LQ events are relatively low in frequency, so we opted to combine them with L events for this analysis. We have updated the figure and figure legend to reflect this.

1. The CaMPARI images do not look great, particularly in the pan-neuronal condition (Fig. 5A). It would be useful to include the movie of the stack. Did any other brain regions show activity differences, such as SEZ or PI? These regions are known to be involved in feeding so it seems surprising they show no effect.

We find that CaMPARI imaging is subject to high levels of noise and background, especially when using a broad driver as the reviewer has pointed out. This is why we opted to follow-up our pan-neuronal CaMPARI experiment using a more specific mushroom body driver and to test our correlational findings of increased MB activity in hedonic environments with genetic approaches in the remainder of Figure 5. We will include movies of the confocal stacks for both CaMPARI experiments, as requested.