Postprandial sleep mechanics in Drosophila

Food consumption is thought to induce sleepiness. However, little is known about how postprandial sleep is regulated. Here, we simultaneously measured sleep and food intake of individual flies and found a transient rise in sleep following meals. Depending on the amount consumed, the effect ranged from slightly arousing to strongly sleep inducing. Postprandial sleep was positively correlated with ingested volume, protein, and salt—but not sucrose—revealing meal property-specific regulation. Silencing of leucokinin receptor (Lkr) neurons specifically reduced sleep induced by protein consumption. Thermogenetic stimulation of leucokinin (Lk) neurons decreased whereas Lk downregulation by RNAi increased postprandial sleep, suggestive of an inhibitory connection in the Lk-Lkr circuit. We further identified a subset of non-leucokininergic cells proximal to Lkr neurons that rhythmically increased postprandial sleep when silenced, suggesting that these cells are cyclically gated inhibitory inputs to Lkr neurons. Together, these findings reveal the dynamic nature of postprandial sleep. DOI: http://dx.doi.org/10.7554/eLife.19334.001


Introduction
Despite accumulating evidence of interactions between sleep and metabolism, few studies have documented an increased propensity for sleep that animals might experience after meals. Results on human postprandial behavior have shown an increased sleep propensity following a meal, a lagging response, or no effect at all (Orr et al., 1997;Zammit et al., 1995;Wells et al., 1998;Harnish et al., 1998;Stahl et al., 1983). It has been proposed that postprandial sleep is not invariable; it may be absent in particular individuals or heavily dependent on specific food properties (Stahl et al., 1983). While the behavior likely exists, these intrinsic variabilities have limited our ability to study its molecular basis. A thorough examination of this behavior would be facilitated by use of an animal model although, at present, no clear model exists (Watson, 2014).
In Drosophila there is a well-documented sleep-metabolism interaction in which flies suppress sleep or increase locomotion when starved (Lee and Park, 2004;Thimgan et al., 2010;Keene et al., 2010). However, the acute effects of food consumption on sleep have not been tested, largely because there is no system available to do so. Here we establish a system for simultaneous measurement of discrete feeding events and sleep, allowing us to examine the mechanisms underlying their interaction.

Results
To simultaneously measure sleep and feeding of individual adult flies, we designed the Activity Recording CAFE (ARC), a system that couples machine vision tracking of capillary-based food consumption and animal motion ( Figure 1A-C) (Ja et al., 2007;Donelson et al., 2012;Deshpande et al., 2014). Previous studies have shown that flies inactive for 5 min exhibit multiple hallmarks of sleep (Shaw et al., 2000;Huber et al., 2004;Hendricks et al., 2000). Using this standard, we extracted sleep parameters by defining sleep as periods of inactivity >5 min. Despite different housing and food consistency, we found sleep architecture to be similar to previous studies using high resolution object tracking, a method which has greater resolution of total sleep and sleep duration than the infrared-beam based alternative ( Figure 1C, Figure 1-figure supplement 1A-D) (Donelson et al., 2012). Flies appeared to move less after meals than before-an effect more pronounced with larger meals-suggesting that sleep might also be greater after eating ( Figure 1D). Animals also positioned themselves closer to the food following each meal ( Figure 1E-F), consistent with observations that isolated flies sleep close to food (Hendricks et al., 2000) and the spatial tendency of flies driven to sleep by activation of neurons expressing short neuropeptide F, a satiety signaling hormone (Shang et al., 2013;. To examine the acute effects of feeding on sleep, we quantified the probability of sleep (P sleep ) relative to individual feeding events. The animals experienced a sharp decline in P sleep in the time approaching each meal, as expected since flies cannot sleep and feed simultaneously. Subsequent to the meal, P sleep rose rapidly and exceeded the respective pre-meal P sleep (Figure 2A). To quantify changes in sleep subsequent to each meal, we calculated the change in P sleep of respective time points before and after each event (i.e. t +1 -t À1 , t +2 -t À2 , . . .). This analysis revealed a rise in sleep lasting approximately 20-40 min with a maximum DP sleep amplitude of 0.15 and 0.2 for the control strains, w 1118 and Canton-S, respectively ( Figure 2B). The effect was observed in both males and females ( Figure 2-figure supplement 1A). While the duration of sleep events prior to a meal is inherently limited by the time until an animal wakes to eat, sleep initiation is not limited and was more probable following a meal (Figure 2-figure supplement 1B).
In addition to periods of quiescence, increased arousal threshold is regularly used to demonstrate sleep in flies. To test if arousal threshold is increased after a meal, we adapted methodology from the Drosophila arousal tracking system in which animals are exposed to a series of increasing vibrational stimuli (0.8-3.2 g) once every hour in order to measure the intensity at which inactive animals eLife digest Many of us have experienced feelings of sleepiness after a large meal. However, there is little scientific evidence that this "food coma" effect is real. If it is, it may vary between individuals, or depend on the type of food consumed. This variability makes it difficult to study the causes of post-meal sleepiness.
Murphy et al. have now developed a system that can measure fruit fly sleep and feeding behavior at the same time. Recordings using this system reveal that after a meal, flies sleep more for a short period before returning to a normal state of wakefulness. The sleep period lasts around 20-40 minutes, with flies that ate more generally sleeping more.
Further investigation revealed that salty or protein-rich foods promote sleep, whereas sugary foods do not. By using genetic tools to turn on and off neurons in the fly brain, Murphy et al. identified a number of brain circuits that play a role in controlling post-meal sleepiness. Some of these respond specifically to the consumption of protein. Others are sensitive to the fruit fly's internal clock, reducing post-meal sleepiness only around dusk. Thus, post-meal sleepiness can be regulated in a number of different ways.
Future experiments are now needed to explore the genes and circuits that enable meal size and the protein or salt content of food to drive sleep. In nature, sleep is likely a vulnerable state for animals. Thus, another challenge will be to uncover why post-meal sleep is important. Does sleeping after a meal boost digestion? Or might it help animals to form memories about a food source, making it easier to find similar food in the future?   Schematic of ARC apparatus. Independent computer-controlled cameras record images of capillaries for feeding measurements and of flies for activity and sleep determination. (B) Raw image (one out of 1440 taken over 24 hr) of capillaries containing liquid food (top). Capillaries have an external reference mark and an internal dyed oil band overlaid the aqueous food. Images are thresholded to binary, from which cumulative pixel distances between the reference mark and dye band can be calculated from the image series (bottom). D pixels per frame reveals individual feeding events, selected as events greater than 3.5 standard deviations above noise (red dashed line). (C) Activity, sleep (black shading) and feeding (ml) of an individual fly (activity and feeding in 1 min bins, sleep in 30 min bins). (D) Example motion traces of individuals before (black) and after (red) a meal (yellow circle). Motion traces and averages associated with small (<0.06 ml) or large (!0.06 ml) meals are shown. n = 661 meals from 30 flies, w 1118 ; *p<0.05, Figure 1 continued on next page respond ( Figure 2C) (van Alphen et al., 2013;Faville et al., 2015). Consistent with previous studies, we found that arousal threshold initially increases with the amount of time that a fly is inactive, with the greatest change occurring in the first 5 min ( Figure 2D) (Shaw et al., 2000;Huber et al., 2004;van Alphen et al., 2013;Faville et al., 2015). Since the maximum duration of inactivity is lower following a meal, irrespective of sleep probability ( Figure 2E), we filtered arousal events to compare animals in similar stages of sleep. While animals inactive for more than 5 min are experiencing change in sleep depth, animals inactive for 0-5 min are transitioning into sleep. In this transition state animals showed a significantly greater arousal threshold in the 20, 40, and 60 min post-meal compared to animals during the respective pre-meal times ( Figure 2F). Interestingly, the correlation between arousal threshold and postprandial sleep grew weaker when comparing animals in deeper stages of sleep (Figure 2-figure supplement 2A-B). This suggests that food intake transiently influences sleep induction rather than depth, which might be shaped solely by the amount of time that an animal is inactive.
We further considered the possibility that increases in immobility-derived sleep and arousal threshold could be artifacts of grooming behavior. To test this, we manually scored videos of animals during an ARC recording and annotated grooming events. Falsely calculated sleep, in which animals were grooming rather than immobile, accounted for~7% of sleep surrounding meals and had statistically indistinguishable occurrence before and after a meal ( Figure 2G, Figure 2-figure supplement 3). Re-analysis of sleep surrounding a meal, where sleep was defined as periods of immobility and a lack of grooming exceeding 5 min, lowered total sleep but did not alter the rise in postprandial sleep ( Figure 2H).
We next asked how the quantity of food consumed might influence postprandial sleep. We plotted P sleep or DP sleep for different consumption volumes and found that sleep following meals increased as a function of meal size, while pre-prandial sleep did not ( Figure 3A, Figure 3-figure supplement 1A). Since average postprandial sleep lasted between 20-40 min, and since arousal threshold differences were most apparent in the first 20 min, we re-calculated DP sleep for the 20 min surrounding each meal. We found a significant correlation between DP sleep and consumption volume when meals were analyzed individually (Figure 3-figure supplement 1B) or binned ( Figure 3B). Since this analysis used individual meals rather than individual animals, we tested whether animals that consumed meals more frequently could bias the observed effect. We performed a Monte Carlo simulation by iteratively recapitulating the analysis with an equal number of randomly selected meals from each animal and found that equal sampling accurately portrays the full data set (Figure 3 Since sleep is derived from periods of inactivity, we considered whether increased consumption could affect mobility, which might in turn appear to be sleep. The change in movement rate after a meal (D speed ) showed a much weaker correlation to meal size than DP sleep (Figure 3-figure supplement 2C). Since speed is a strong indicator of mobility, we reasoned that immobility is not likely responsible for the sleep effect.
Although meal size has an apparent impact on postprandial sleep, the standard diet contains numerous physiological and nutritive components. To determine the potential contribution of individual components to postprandial sleep, we designed a paradigm that allowed us to test a range of feeding volumes or of ingested nutrients. To test volumetric effects, animals were fed a low protein concentration food (0.25% tryptone) to elicit feeding without impacting baseline sleep, since sucrose and sweet tasting substances have been shown to modulate sleep (Keene et al., 2010;Linford et al., 2012). We found a highly significant correlation between meal volume and 20 min  ). The inset graph shows the average position over 20 min pre-meal (À20) and post-meal (+20) with the axis scaled to the parent graph. ***p<0.001, Wilcoxon matched-pairs sign rank test. All bars represent mean ± s.e.m. DOI: 10.7554/eLife.19334.003 The following figure supplement is available for figure 1: Data are shown as averages of 1 min bins. DP sleep is defined as the difference between postprandial P sleep and the corresponding time-matched pre-meal P sleep (i.e. t 1 -t À1 , t 2 -t À2 , . . .). For clarity, a mirror image of pre-meal P sleep is replotted in the postprandial period (dashed line). Inset graphs show average P sleep (± s.e.m.) for the 20 min before (À20) and after (+20) meals, with the axis scaled to the parent graph. n = 757 meals from 50 flies, Canton-S; 661 meals from 30 flies, w 1118 ; ***p<0.001, Wilcoxon matched-pairs sign rank test. (B) DP sleep calculated from data in A. (C) Diagram of stimulus delivery system showing shaft-less vibration motors attached to the back of ARC chamber. Increasing vibrations are delivered to the chamber via a microcontroller using pulse width modulation. (D) Arousal threshold shows an initial increase with the time an animal is inactive. 180 flies, 11,479 arousal events, Canton-S; 5 min bins, circles represent mean ± s.e.m., a secondorder polynomial trendline is shown. (E) Superimposition of time inactive (blue) over P sleep (black) relative to meals during periodic vibrational stimuli. The inset graph shows P sleep and time inactive in the 20 min before and after each meal. n = 2245 meals from 180 flies, Canton-S; ***p<0.001, Wilcoxon matched-pairs sign rank test. (F) Stimulus response from 0-20, 20-40, and 40-60 min pre-and post-meal (red) superimposed onto P sleep (black). Arousal events are filtered to 5 min intervals for prior time inactive to control for sleep depth (0-5 mins shown, minimum 1 s inactivity). Circles represent mean ± s.e.m.; n = 2245 meals from 180 flies, Canton-S; *p<0.05, **p<0.01, Mann Whitney test. (G) Percent of calculated sleep that is actual immobility versus grooming. The inset graph shows the percent grooming in the 20 min before and after each meal. n = 55 meals from seven flies, Canton-S; p=0.69, Wilcoxon matched-pairs sign rank test. (H) Comparison of P sleep before and after meals calculated using immobility criteria "I" versus immobility criteria paired with grooming criteria "I, -G". Inset shows average P sleep (± s.e.m.) in the 20 min before and after each meal. ***p<0.001, Wilcoxon matched-pairs sign rank test. DOI: 10.7554/eLife.19334.005 The following figure supplements are available for figure 2: To determine if individual nutrients could modulate postprandial sleep, we fed animals a concentration series of protein (tryptone), salt (NaCl), or sucrose. All diets contained a base of 0.25% protein to induce feeding while minimizing effects on baseline sleep that might arise by differences in sucrose consumption. Within each series we compared nutrient consumption within a narrow volume range (0.02-0.04 ml, representing 33-38% of recorded meals) to limit the contribution of ingested volume on sleep. We found that DP sleep was significantly correlated to both protein and salt ingestion ( Figure Figure 4B). To compare the kinetics of DP sleep for each effective driver, we quantified maximum DP sleep amplitude and the total DP sleep during its decay. For any given increase in total DP sleep during the decay, meal volume induced a greater rise in amplitude than protein or salt (Figure 4-figure supplement 2B). We hypothesize that multiple mechanisms operating on different time scales might underlie postprandial sleep regulation by the various drivers.
We next sought to identify a neuronal mechanism by which feeding drives postprandial sleep. A previous study showed that Lk was involved in meal size regulation, suggesting that this system acts rapidly during feeding to signal a behavioral shift (Al-Anzi et al., 2010). To test if Lk signaling regulates postprandial sleep, a subset of Lkr neurons labeled by the line, R65C07 (hereafter referred to as Lkr GAL4 ), were silenced by overexpressing Kir 2.1 , an inward rectifying K + channel (Johns et al., 1999). This manipulation eliminated any rise in postprandial sleep and instead caused slight arousal ( Figure 5A). There was no significant change in postprandial activity (D motion for 20 min after meal:    The following figure supplement is available for figure 5: revealing several cell populations stemming from the lateral horn which innervate the dorsal fanshaped body of the central complex-a region of the brain which functions in sleep homeostasis, visuomotor function, and hunger-related behavior ( Figure 5B) (Donlea et al., 2011(Donlea et al., , 2014Seelig and Jayaraman, 2013;Krashes et al., 2009). This circuitry suggests that the Lk system may act directly on the sleep and motor controllers of the fly brain to regulate postprandial sleep. We hypothesized that Lkr circuitry could receive convergent signals from volume, protein, and salt to drive postprandial sleep. Alternatively, it could specifically relay the signals of a single driver to some point of convergence. We first tested whether the subset of Lkr neurons labeled by Lkr GAL4mediated the sleep response to meal volume by feeding a low nutrient diet to Lkr GAL4 -silenced animals. Silenced animals and their controls showed similar DP sleep in response to ingested volume, indicating that the Lkr GAL4 neurons are not responsible for the volumetric modulation of postprandial sleep ( Figure 5C). To test the role of Lkr neurons in the postprandial sleep response mediated by nutrients, we supplemented a sucrose diet with either salt or tryptone. Lkr GAL4 -silenced animals and their controls showed statistically indistinguishable DP sleep responses to salt manipulation ( Figure 5D). In contrast, while protein ingestion increased postprandial sleep in controls, Lkr GAL4silenced animals showed a strong waking response, or a negative shift in postprandial sleep ( Figure 5D). Interestingly, this waking response exceeded that of control animals on sucrose alone, indicating that protein can also drive wakefulness in the absence of Lkr GAL4 neuronal activity. We did not observe changes in meal size, total sleep, or total consumption, indicating that these behaviors are independent of Lkr influence on postprandial sleep (Figure 5-figure supplement 1A-C). To test whether this phenotype was caused by changes in development, we conditionally silenced the cells by co-expressing a temperature sensitive GAL4 suppressant protein, GAL80 ts , throughout development ( Figure 5E). Inactivation of GAL80 ts to silence Lkr GAL4 -labeled cells in adulthood was sufficient to induce a postprandial sleep defect ( Figure 5F).
Subsets of Lk neurons have been found in the lateral horn, with the diffuse puncta in close proximity to the cell bodies of the Lkr neurons, suggesting a neuromodulatory connection between these cells (Al-Anzi et al., 2010;de Haro et al., 2010;Cavey et al., 2016). Recent findings also show that incubating Lkr neurons with Lk inhibits their response to the cholinergic agonist carbachol, suggesting that Lk inhibits Lkr neuronal activity (Cavey et al., 2016). We first sought to confirm that cells labeled by the Lk GAL4 driver express Lk. Indeed, immunostaining of Lk overlapped with Lk GAL4labeled cells in the lateral horn, as well as in two cells in the suboesophageal ganglion (SOG) ( Figure 6A). Using thermogenetic activation of these cells with the transient receptor potential channel Trp A , we found that postprandial sleep was significantly reduced upon activation ( Figure 6B), furthering the notion that these cells inhibit Lkr neurons. To see if Lk was necessary to drive this inhibitory effect, we used two independent RNAi lines to knockdown Lk in cells labeled by Lk GAL4 . We found that this knockdown increased postprandial sleep, consistent with the idea that expression of Lk in Lk GAL4 -labeled cells is necessary for the inhibitory effect on Lkr neuronal activity ( Figure 6C).
The broad arborizations of the Lkr neurons suggest that these cells might receive inputs from a number of spatially or type distinguished cells in the brain. R20G03 (20G03), a line expressing GAL4 driven by an enhancer fragment 316 base pairs upstream of Lk, labels cell bodies in the lateral horn and SOG, as well as a subset of mushroom body output neurons (MBONs) ( Figure 7A), all of which coincide with Lkr neurons labeled by the Lkr GAL4 driver ( Figure 7B). The cholinergic MBONs of 20G03 have been shown to play a role in short-term appetitive memory following food intake (Aso et al., 2014). These features, paired with their neuroanatomical distribution, suggested that these cells might play a role in postprandial sleep. To test this, we again employed electrical silencing with Kir 2.1 and found that constitutive and adult-restricted silencing of these neurons increased postprandial sleep, suggesting that these cells might directly inhibit the activity of proximal Lkr neurons ( Figure 7C-D). Interestingly, immunostaining of Lk did not co-localize with any of the 20G03 cells, nor did downregulation of Lk in these cells affect postprandial sleep (Figure 7-figure supplement 1A-B), suggesting that their inhibitory action stems from an alternative neuropeptide or neurotransmitter. By analyzing behavior throughout a 24 hr period, we found that the increase in postprandial sleep driven by 20G03 silencing was most pronounced at ZT 12 ( Figure 7E-F). Closer examination within this period showed that, rather than raising sleep evenly within ZT 10-14, the sleep after but not before each meal was increased ( Figure 7G), implying that these cells activate in response to food intake.

Discussion
The ARC is the first platform capable of simultaneously measuring sleep and resolving discrete feeding events in individual flies. Using this tool, we found that flies exhibit a rise in sleep following a meal. Interestingly, consumption of a meal did not necessarily affect the depth of sleep when controlling for time inactive. There are many instances in which genes or neural processes have been found to influence particular features of sleep while leaving others unaffected (Harbison et al., 2009;Shi et al., 2014;Shang et al., 2011). Similarly, our work suggests that sleep probability and sleep depth can also be dissociated. By further examination of the shift in sleep probability, we found that it increased as a function of meal size. This may be paralleled in humans where cranial EEG power increases with meal size and during certain stages of sleep (Reyner et al., 2012).
In examining individual meal components, we identified volume, protein, and salt as effectors of postprandial sleep. While human studies have been unable to resolve a link between meal volume and measures of wakefulness (Landströ m et al., 2001), artificial distension of the gut has been shown to increase sleep in rats (Lorenz et al., 1998). Although there are no studies on the effect of protein on postprandial sleep, protein consumption modulates sleep in Drosophila and positively correlates with napping in humans (Catterson et al., 2010;Grandner et al., 2010). Similarly, although there are no studies on the influence of salt on postprandial sleep, salt intake increases mammalian oxytocin signaling, which has been implicated in sleep-wake behavior (Krause et al., 2011;Lancel et al., 2003). Sucrose has been shown to play a critical role in long-term sleep homeostasis and architecture in flies (Keene et al., 2010;Linford et al., 2012), although our findings indicate that it does not modulate postprandial sleep. Although we demonstrate the effectiveness of these dietary components in isolation, we note that meals typically incorporate multiple elements. Thus, further work is necessary to identify the mechanisms by which each constituent is sensed and We demonstrate that the Lk system plays a role in postprandial sleep ( Figure 7H). A subset of Lkr neurons were necessary to initiate postprandial sleep in the presence of protein. While we expected that animals defective in protein sensing would experience postprandial sleep similar to that of animals fed only sucrose, we found instead that they had a waking response. This suggests that ingested protein promotes both sleep and wakefulness, and that the wakefulness is normally counterbalanced by Lkr neuronal activity. While this study does not specifically identify the waking output for protein, it was previously shown that Translin acts through Lk-expressing cells to drive wakefulness in response to sucrose starvation (Murakami et al., 2016). While this does not explain rapid wake promotion in response to protein ingestion, it does demonstrate that Lk-expressing cells are capable of promoting wakefulness.
Another recent finding identified Lkr neurons as circadian output neurons, which modulate activity rhythmically. The same study found that incubation of Lkr cells with Lk shunts their calcium response to the cholinergic agonist carbachol (Cavey et al., 2016), suggesting an inhibitory connection between Lk and downstream Lkr neurons. By stimulating these same Lk neurons, we observed a reduction of postprandial sleep matching the effect of Lkr silencing, providing additional evidence that the lateral horn harbors an inhibitory node between Lk and Lkr neurons. We also found that knockdown of Lk in these cells increased postprandial sleep, supporting this notion. However, it has been suggested that Lk may have additional receptors (Terhzaz et al., 1999), and there is still a lack of definitive evidence for the inhibitory mechanism between Lk and Lkr neuronal activity in the lateral horn. Additional genetic tools, including lines with restricted expression patterns, will be needed to fully uncover the genetic interactions and neuronal connectivity of this system.
We also found a subset of cells labeled by 20G03 that modulate postprandial sleep during the period near 'dusk' or 'lights-off'. Interestingly, the subset harbors acetylcholine MBONs-cells necessary for the short-term associative memory between sucrose consumption after starvation and an odor (Aso et al., 2014). This suggests that these cells activate following a feeding event, in agreement with our results. This is further supported by the anatomical distribution of cells in the SOG, a region of the fly brain that regulates feeding behavior. The distribution also shows overlap with Lkr neurons and regions of the neuronal clock network, suggesting a diversity of regulatory inputs and outputs. Further studies might employ GFP reconstitution across synaptic partners (GRASP) (Gordon and Scott, 2009) or neuronal co-labeling to demonstrate these connections. The apparent diversity of neuronal modulators of postprandial sleep opens the door to future work on determining the various genes and circuits involved.
The tachykinin family to which Lk belongs has been implicated in the regulation of satiety and sleep, however any physiological connections between these behavioral outputs has not been explored (Massi et al., 1988;Zielinski et al., 2015). The orexin system is regulated by a diversity of signaling molecules (Scammell and Winrow, 2011) and has already shown promise as a target for sleep therapeutics. Such genetic diversity and therapeutic potential may also be true of systems governing postprandial sleep. Accordingly, the ARC provides a platform for future studies aimed at uncovering additional genes, circuitry, and physiological roles of postprandial sleep. Furthermore, the ability to measure both sleep and feeding in individual animals may shed new light on the fundamental relationship between these behaviors in other paradigms.

Diets
Liquid diets were prepared by boiling ingredients in 100 ml ddH 2 O followed by filtration (0.2 mm cellulose acetate syringe filter, VWR, Radnor, PA). Listed percentages of Bacto Tryptone, Bacto yeast extract, sucrose, or NaCl were calculated as weight/volume. All reagents were obtained from Fisher Scientific (Waltham, MA) or VWR. Feeding behavior was recorded on a standard liquid diet of 2.5% sucrose + 2.5% yeast extract unless otherwise noted.

ARC behavioral chamber
Animals were housed in a plastic chamber (acrylonitrile butadiene styrene) containing small halfcylindrical cells (7 mm wide, 4.5 mm depth, 1.15 mm spacing) fit with a 2 mm thick clear acrylic panel for visualization. Strips of infrared led lights were placed~16 cm behind each chamber on a heat sink to provide backlighting for feeding and motion tracking. Each animal was given access to a capillary filled with liquid food. Above the chambers was a 2 mm gap with a small overhang allowed for a sliding gate (5 mm Â 2 mm) to seal individual fly cells. The gate contained evenly spaced 2.5 mm vertically oriented holes above the center of each fly cell. This allowed for the insertion of tightly fitting 200 ml pipette tip sleeves that were cut to hold the food-containing capillaries (5 ml glass capillaries, VWR). Sleeves were made by cutting the narrow end of 200 ml pipette tips to fit capillaries tightly. The dye used for liquid meniscus tracking was composed of 75% mineral oil, 25% dodecane, and 1% Copper (II) 1,4,8,11,15,18,22,St. Louis,MO). Dye solution was vortexed for 1 min and centrifuged briefly. Supernatant dye was collected and re-dispensed into 1.5 ml tubes and vortexed for 1 min. Dye solution was loaded into each capillary to make a 1 mm plug before loading liquid diet. Food was loaded until dye bands were approximately 0.5 mm below the capillary reference mark. To improve visualization of the dye and reference mark, thin white construction paper was gently attached to the back of capillaries using double-sided tape.

Feeding behavioral capture and analysis
For ARC serial image capture, we used a Lifecam 1080p HD webcam (Microsoft, Redmond, WA) with the infrared filter replaced by an infrared pass filter. The camera was mounted with a custom bracket at equal height to the base of the capillaries. PhenoCapture software was used to capture one image per min using the time-lapse image capture tool (www.phenocapture.net). Each capillary had an exterior reference mark and was pre-loaded with an oil-based dye band, marking the meniscus of the liquid food. ImageJ was used for image processing (Schneider et al., 2012). Images were cropped to include all capillaries and dye bands. The images were background subtracted using the built-in subtract background function with a rolling ball radius of five and disabled smoothing. The images were then thresholded manually until the dye bands and reference marks were faithfully displayed without background noise. The images were split to contain image stacks of individual capillaries and tracked using a custom ImageJ plugin. The relative distance between the food meniscus and the reference mark was calculated at each frame using the custom plugin. We used a meal selection algorithm that estimates noise by an iterative filtration of distal values 4.0 s.d. above the D pixel mean (see Supplementary file 1). Meals were then selected as values which were 3.5 s.d. above the estimated noise, based on the one frame/min sampling rate. In a setup tracking 30 capillaries, this method provides 10 nL resolution although greater resolution can be achieved by increasing camera resolution and sampling rate. Though infrequent, temporally adjacent feeding bouts were combined and time stamped with the latter meal time. Thus, feeding bouts followed by at least 1 min of non-feeding were considered meals. Each feeding event was corrected for effects of evaporation and luminance fluctuation on observed meal size by subtracting the average noise value from all non-feeding frames. Meal output contains meal associated time and corrected volume.

Positional tracking
Behavioral tracking was carried out with a custom, freeware, cross-platform analysis framework (Jav-aGrinders Library, available for free download at http://iEthology.com/). This library utilizes and extends machine vision functions provided by the OpenCV project for high-resolution analysis of spatial and behavioral data. With an integrated interface for standard USB video device class (UVC) cameras, the software implements multi-stage object detection and analysis for 8-bit grayscale/32bit color frame sequences in real-time. Following the subtraction of a reference frame, the tracker locates the frame coordinate with the highest remaining pixel value. Provided this value exceeds the object's threshold, the detection algorithm proceeds outwards until it identifies a starting point along the object's thresholded border. By following this edge until it returns to the starting point, the detection algorithm characterizes the object's outline as a polygonal shape. The object's center location is estimated by the polygon's centroid, directional attributes (e.g., long axis) which are obtained via a Singular Value Decomposition of the polygon's outline. The projected size is represented by the shape's area. The size of a fly tracked in the ARC measured approximately 15-20 pixels in length. The present study used a Microsoft Lifecam Studio camera with the infrared filter replaced by an infrared pass filter. Although much higher rates are possible within this system, a reduced sampling rate of 1 Hz was chosen as it proved sufficient for identifying the extended periods of inactivity indicative of sleep. The maximum frame rate capability depends on a number of factors, including the video's resolution, the object's size, the performances of processor and graphics card, and the specifications of the hardware drivers used to interface with the camera. Distance traveled was obtained from Euclidean measures between object centers in consecutive frames. Motion values less than 50% of the fly body length (FBL) were discarded. Dropped positions were reset to the last known location for motion and sleep calculation purposes. This system has been previously tested for Drosophila tracking sufficiency and demonstrated enhanced motion/sleep characterization compared to alternative methods (Donelson et al., 2012). An animal was considered sleeping during any period of immobility exceeding 5 min, based on previous studies showing that this period of inactivity is highly correlated with hallmarks of sleep (Shaw et al., 2000;Huber et al., 2004). Speed of each motion event was calculated as the distance traveled during that event divided by its uninterrupted duration. This is the first time, to our knowledge, that sleep has been measured in vertically oriented chambers, and on a liquid diet. However, we found sleep patterning in the ARC to be in line with the more commonly used Drosophila Activity Monitor (TriKinetics, Waltham, MA). Notably, object detection provides a higher resolution for sleep bout duration (Figure 1-figure supplement 1D).

ARC experimental design
Each fly well was loaded with 300 ml of 1% (w/v) Bacto agar to allow ad libitum access to water. Flies were loaded into chambers using a standard mouth aspirator to avoid behavioral perturbation from CO 2 anesthetization. After loading each fly, a sleeve and capillary were quickly inserted to prevent escape. Flies were habituated in the recording chamber for 20-24 hr with access to high nutrient food (5% sucrose + 5% yeast extract) to obviate the need for more than one food change every 24 hr. All experiments had daily food changes from Zeitgeber time (ZT) 0-0.5 hr. Infrequently, food consumption exceeded the volume administered requiring an additional food change at ZT 8-12.

Behavioral data analysis
Motion and feeding data were analyzed using custom python-based software and custom or built-in Matlab functions (The Mathworks, Natick, Massachusetts, USA). Instructions and 3D print files for ARC setup and data analysis software are available upon request.
Probability of sleep is calculated as the population mean of sleep probability/unit time. 1-min bins were used for all P sleep plots. DP sleep represents a comparison of pre-meal to post-meal sleep by calculating the difference in P sleep between post-meal time bins and their respective pre-meal time bins (i.e. (t 0-1 min post-meal) -(t 0-1 min pre-meal), (t 1-2 min post-meal) -(t 1-2 min pre-meal). . .). This was either calculated in 1-min bins for time-course analysis, or in 20-min time bins following the meal. The 20-min bin was selected since this length of time covered the mean duration of postprandial sleep for both Canton-S and w 1118 .

Arousal threshold analysis
To test arousal threshold, we adapted methodology previously used for quantifying sleep and sleep depth (van Alphen et al., 2013;Faville et al., 2015). Animals were exposed hourly to a series of vibrations of increasing intensity ranging from 0.8 to 3.2 g, in steps of 0.6 g. Stimuli trains were composed of 200 ms vibration with 800 ms inter-vibration interval and 15 s inter-stimuli train interval. Stimulation intensity and timing were controlled using pulse-width modulation via an Arduino UNO and shaft-less vibrating motors (Precision Microdrives, model 312-110) ( Figure 2C). Arousal to a given stimulus was assigned when an animal (1) was inactive at the time of the stimulus, (2) satisfied a given inactivity criteria at the time of the stimulus, and (3) moved within the inter-stimuli train period (15 s) of that stimulus.

Grooming analysis
To determine the impact of grooming behavior on calculated sleep, we simultaneously recorded video of flies during a normal ARC experiment using equipment and videocapture methods previously described (King et al., 2016). Grooming start and stop times were manually noted, where the observer was blind with respect to feeding times. Times were then computer annotated as a binary (grooming = 1, not grooming = 0) for all time points within the experiment. Individual housing precluded the need for annotation of courtship behaviors.

Meal size, volume, and nutrient grading
For meal volume gradings, animals were fasted for 20 hr and then placed on a 0.25% tryptone diet. Feeding events were recorded for 24 hr. This paradigm provided a high range of feeding volumes while maintaining a small but sufficient amount of nutrients to elicit feeding. Meal volumes and associated sleep events were grouped into 0.01 ml bins according to volume consumed. Bins greater than the 97th percentile were excluded to maintain sufficient sample size for each bin. Alternative experiments for testing volumetric effects used small amounts of sucrose (1%) or both sucrose and tryptone (1% sucrose + 0.25% tryptone) and gave similar results (data not shown). For sucrose grading, the same paradigm was used with a low tryptone diet (0.25%) supplemented with varying concentrations of sucrose (0, 1, 5, 15, and 25%). Feeding events were then filtered to include volumes within the range 0.02-0.04 ml which included 37% of all recorded meals (1131/3038 meals). This range was chosen to maintain a small and narrow volumetric input, and because all dietary groups maintained meal volumes within this range. Meals were grouped into 1.333 mg bins of sucrose consumed during the meal (meal volume Â nutrient concentration). Salt grading was performed similarly but with varying concentrations of salt (0, 0.25, 0.3, 0.4, 0.5, 0.6, 0.75, and 1%) and 0.05 mg bins, and included 33% of all recorded meals (478/1446 meals). Protein grading was performed by providing diets of varying concentrations of tryptone (0.25, 1, 2.5, 3, 4, and 5%) and analyzed similarly to sucrose and salt experiments using 0.333 mg bins, and included 38% of all recorded meals (714/1872 meals).

Lkr GAL4 dietary manipulations
For Lkr GAL4 silencing experiments, flies were fed standard liquid diets: 2.5% sucrose + 2.5% yeast extract, respectively. To test volume response deficiency, animals were fed a low nutrient (1% sucrose + 0.25% tryptone) food which elicited a large range and frequency of meals. To test protein response deficiency, animals were fed 2.5% sucrose supplemented with the standard diet equivalent of protein (1.67% tryptone). The standard diet contained a negligible amount of salt (with respect to inducing a postprandial sleep response). To test a salt response deficiency, animals were fed 2.5% sucrose supplemented with 1% NaCl. Average meal size between groups did not differ-hence, meal size filtering was not applied.

Statistics
Statistical analysis was performed using Matlab statistical toolbox or GraphPad Prism version 5.04 (GraphPad Software, La Jolla, CA). All reported values are mean ± s.e.m. Shapiro-Wilk test was used to determine data normality. Data were analyzed using either Wilcoxon matched-pairs sign rank test for comparisons of non-parametric P sleep of animals before and after meals, Kruskal-Wallis test followed by Dunn's multiple comparisons for non-parametric comparisons of DP sleep , or the Mann-Whitney test when comparing two non-parametric groups, non-paired observations. Comparisons of meal size to given features were analyzed using Spearman rank correlation, due to non-parametric distribution, while binned data were analyzed using Pearson correlation.