Pathogen-induced food evasion behavior in Drosophila larvae

Animals in their natural habitat are often exposed to pathogens that can be life threatening. The animal is equipped with several defense strategies to fight the pathogen. The immune system plays a central role together with host learning to avoid, resist and tolerate these dangers (Akira et al., 2006; Medzhitov et al., 2012). One or all of these mechanisms can be seen in various organisms from Caenorhabditis elegans (Melo and Ruvkun, 2012; Meisel and Kim, 2014) to humans (McCusker and Kelley, 2013; Curtis, 2014). However, an additional behavioral strategy has evolved to help limit pathogen growth and fight infection more efficiently. This is manifested as hypothermia, inconsistent sleep pattern, nausea, pain, reduced grooming, depression and loss of appetite/food aversion. These sets of organized behavioral responses are known as sickness syndrome or sickness behavior (SB), generated by the animal rather than being a secondary result of the infection itself (Hart, 1988). In mammals, pro-inflammatory cytokines such as Interleukin 1, Interleukin 6 and interferons, produced in the periphery by activated macrophages during infection, reach the brain directly or indirectly, causing SB (Dantzer, 2001). Although these symptoms seem to be damaging initially, they are generated by the animal's central nervous system (CNS) after evaluating the chance of survival.

Drosophila melanogaster has helped us understand the details of the innate immune system (Lemaitre and Hoffmann, 2007; Kounatidis and Ligoxygakis, 2012; Buchon et al., 2014). Infection behavior has been less well explored. Here, we used D. melanogaster larvae to examine how infection can affect their attraction towards food. Pseudomonas entomophila (Pe) infection can block food intake in D. melanogaster (Liehl et al., 2006). We found that larvae can actively recognize infectious food and generate an alteration in their feeding behavior, limiting damage caused by infection. Like other animals, the larval circuit can also process the information regarding its internal state and generate a behavioral response for survival.

Org-R (Bloom no. 4269) was used as wild-type. Other strains include Hug-Gal4 (Bloom no. 58769) (Melcher and Pankratz, 2005), UAS-rpr;;UAS-hid (Buch et al., 2008), UAS-Kir2.1 (Baines et al., 2001), UAS-lacZ-RNAi (gift from M. Jünger), UAS-hug1A-RNAi (Schoofs et al., 2014a,b) and UAS-CaMPARI (Bloom no. 58761).

Flies were raised on standard fly food at 25^oC with a 12 h:12 h light:dark cycle; 2–4 h egg collections were done on apple juice agar plates with a drop of yeast. First instar larvae (24 h after egg laying, AEL) were used for evasion experiments, unless otherwise stated; larvae 60–68 h AEL were used for CaMPARI experiments and antibody stainings.

Glycerol stocks of all Pe strains – Pe, Pe gacA and Pe pvf (Vodovar et al., 2005; Vallet-Gely et al., 2010) – were freshly streaked onto LB agar plates containing 2% skimmed milk and 100 µg rifampicin ml⁻¹. Plates were incubated at 29^oC for 24–30 h and stored at 4^oC for a maximum of 2 weeks. Glycerol stocks were stored at −80°C. For evasion experiments, single bacterial colonies were picked from LB plates and inoculated in tubes containing LB medium and rifampicin. Tubes were kept at 29^oC and 250 rpm overnight, spun down, washed and resuspended to reach an OD₆₀₀ of 10, 60, 110 or 130, according to the particular experiment. For heat-killed Pe, the suspension was kept at 95^oC for 10 min followed by a cold shock at −20°C for 5 min. Suspensions were mixed with yeast for evasion assays.

A 350 µl sample of bacterial suspension was added to 2 g yeast and mixed. A drop of this paste was put in the center of an apple agar plate and used for evasion assays. First or second instar larvae from collection plates were washed in water until they were free from food residues. Fifty larvae were transferred onto each apple agar plate with yeast paste. Larvae were dropped on the food source. The camera set up was controlled by iSpy software, set to take an image every 30 min during the experiment. After the assay (12 h), the total number of larvae on each plate was counted. The number of larvae that were outside the food source at every time point was also counted. The percentage of larvae outside the food source was calculated (degree of evasion) at each time point and graphs were generated using GraphPad Prism 6.

High-power LED of 405 nm [M405L2-UV (405 nm) Mounted LED, 1000 mA, 410 mW; Thorlabs, Newton, NJ, USA] was driven with an LED controller (LEDDB1, Thorlabs) and positioned 6.5 cm above the solution containing the larvae. For the hugin experiments with Pe, the well of a 96-well PCR plate was filled with 50 µl PBS and dead or live Pe (OD₆₀₀=60), larvae were placed in the well for 5 min and UV light was applied for 30 s. For hugin experiments under starvation conditions, the well of a 96-well PCR plate was filled with 50 µl tap water, larvae that had been on fly food or starved for 1 or 2 h were placed in the well for 2 min and UV light was applied for 30 s. Brains were dissected and processed as previously described (Hückesfeld et al., 2016).

Larvae were immediately washed with ice-cold PBS and kept on ice until dissection. Dissected larval brains in ice-cold PBS were fixed in paraformaldehyde (4%) in PBS and washed in 1% PBT containing 5% goat serum. Rabbit anti-hug-PK2 and rabbit anti-hug-gamma primary antibodies (each 1:500, Pankratz laboratory) were incubated overnight at 4°C. After washing with 1% PBT secondary antibody, Alexa 568 (LifeTechnologies, lot no. 1301874), was applied and incubated overnight at 4°C. On the third day, samples were washed with 1% PBT and dehydrated using an ethanol dilution series (30%, 50%, 75%, 95%, 3×100% each for 10 min). After dehydration, samples were treated for 3×5 min in xylenes (Sigma, lot no. STBG1981V) to clear brain tissue. Samples were then mounted in DPX (Sigma, lot no. BCBR1545V) on a cover slide.

For quantification of hugin antibody signals, samples were scanned using the same settings for all scans. Regions of interest were drawn in ImageJ (https://imagej.nih.gov/ij/index.html) and the mean intensity was measured. HuginPC somata were identified based on soma position and morphology. Background normalization was obtained with a 100×100 pixel oval from the subesophageal zone below HuginPC somas.

Single young third instar larvae were placed in the middle of a Petri dish (5 cm diameter) filled with 2% water agar and videotaped with a Logitech C920 HD Pro webcam for 15 min with iSpy software. Videos were then analyzed using FIJI (ImageJ) with a modified version of MTrack2 plugin. The path length that larvae crawled within the 15 min was converted from pixels to cm and plotted on a graph for each genotype.

Evasion box plots represent the cumulative percentage from 6 to 9 h for each genotype. Evasion percentages were compared using Mann–Whitney rank-sum test. For experiments using quantification of CaMPARI red/green ratios or antibody fluorescence intensity, all values were compared using Mann–Whitney rank-sum test with exact P-value test using GraphPad Prism 6 software.

We started out by asking whether Drosophila larvae can identify the presence of a pathogen in their yeast food source. We tested the behavior of wild-type first instar larvae by exposing them to yeast paste containing infectious Pe. Pe is known to be lethal to both larvae and adult flies when ingested in high doses (Vodovar et al., 2005). As an additional control, yeast paste mixed with heat-killed Pe was also provided. Heat-killed Pe are no longer infectious but are still a source of bacterial components that could be identified by the larvae. When we analyzed the plates at the end of the assay, we found larvae on the control plates in and around the food source, while 80% of larvae on the live Pe plates were found outside the infectious food source (Fig. 1A). To understand how this behavior developed, the percentage evasion was plotted over different time points (Fig. 1B). The shape of the graph revealed that the behavior is manifested gradually over 12 h. Evasion behavior starts only after 2–3 h of exposure, unlike sensory avoidance behavior such as bitter aversion in Drosophila larvae (Hückesfeld et al., 2016), which is a much faster response. Hence, we call this behavior larval evasion as opposed to avoidance. Thus, by 6 h of exposure, 50% of the larvae had moved out of the food source mixed with virulent Pe (Fig. 1C). When larvae were exposed to the less virulent strain Erwinia carotovora carotovora (Ecc15), which is non-lethal but induces the immune deficiency (IMD) pathway (Buchon et al., 2009), we did not observe such evasion behavior. Only the strong virulent Pe strain was capable of inducing evasion. This evasion behavior was not restricted to first instar larvae but was also apparent in second instar larvae, albeit at a higher OD₆₀₀ (130 versus 10) of Pe (Fig. S1).

We also tested two mutant Pe strains. gacA Pe is a complete avirulent strain against Drosophila as it lacks the main GacS/GacA two-component system which is critical for bacterial virulence (Vodovar et al., 2005). The pvf Pe strain, in contrast, has a reduced pathogenicity as a result of four missing genes (pvfA, pvfB, pvfC and pvfD) together termed Pseudomonas virulence factors (pvf). This mutation makes the strain less persistent in the gut (Lemaitre, 2015). Larval evasion behavior exposed to mutant Pe was comparable to that on the control plates; there was no significant evasion behavior when the virulence was compromised (Fig. 1D,E). Larvae in some cases even preferred yeast containing gacA Pe over the control. This preference could be due to the presence of the bacterial wall which is avirulent but still high in protein content. We additionally checked the level of anti-microbial peptides (AMPs) induced in larvae after infection (Fig. S2). Neither gacA Pe nor pvf Pe could induce AMP production in larvae while infectious wild-type Pe could. This experiment suggests that larvae can distinguish a pathogenic from a non-pathogenic food source and actively evade the infectious source upon exposure.

As exposure to an infectious food source forces the larvae to wander away from food, we wanted to test whether physiological factors such as the internal metabolic state of the organism would have an effect on evasion behavior. To address this question, wild-type Org-R larvae were starved for 6 h on PBS-soaked filter paper. Larvae fed on yeast were taken as the control condition. The evasion assay was performed to compare the fed versus starved larval response. Interestingly, starvation led to an overall decrease in evasion (Fig. 2). When compared with fed larvae, only 50% of the starved larvae showed evasion. More than half of the starved larvae preferred to stay in the infectious food throughout the assay. This suggests that evasion behavior is also dependent on the internal nutritive state of the larvae. Hunger is perceived as a stronger signal than infection in a starved larva, overriding the evasion response. Thus, larvae prefer to feed on infectious food rather than continue starvation.

Activation of hugin neurons leads to an increase in wandering-like behavior and a decrease in feeding (Schoofs et al., 2014a,b). Moreover, the hugin circuit is involved in the sensory pathway for the bitter aversion response (Hückesfeld et al., 2016). Thus, hugin could also play a role in generating the bacteria-induced evasion behavior. To test this hypothesis, we first decided to manipulate hugin activity by ablating the hugin neurons. This was achieved by expressing two pro-apoptotic genes reaper (rpr) and head involution defective (hid) using Hug-Gal4. Hugin neuronal ablation resulted in a lower evasion percentage (Fig. 3A,B). In the absence of hugin, larvae stayed longer in the infected food compared with the control genotypes. In spite of the food being toxic, larvae did not move out as quickly as the controls. This suggests that hugin neurons are necessary to generate a timely evasion behavior. To confirm these data, we utilized a different mode of manipulation. Hugin neurons were inactivated by expressing Kir2.1 (Baines et al., 2001), which hyperpolarizes neurons. Consistent with the earlier observation, inactivation of hugin neurons also resulted in a 50% drop in evasion percentage compared with the control genotypes (Fig. 3C,D).

To confirm the role of hugin neuropeptide in evasion behavior, the level of hugin neuropeptide was reduced using RNAi (Schoofs et al., 2014a,b). Lowering the level of hugin neuropeptide alone resulted in lower evasion percentage (Fig. 3E,F). Analyzing locomotion of the larvae also showed that the above-mentioned manipulations of hugin neurons had no effect on locomotion per se (Fig. S3). These experiments confirm the role of hugin in generating the larval evasion response. Thus, we have shown that the decrease or absence of hugin activity leads to lower sensitivity towards both an infectious food source (Fig. 3) and bitter food substrates (Hückesfeld et al., 2016).

As it is known that bitter receptors also play a role in bacterial detection (Tizzano et al., 2010; Maurer et al., 2015; Soldano et al., 2016), we tested whether the HuginPC neurons – bitter taste interneurons in the larval brain (Hückesfeld et al., 2016) – play an active part in the processing of virulent bacterial signals. We made use of the calcium-modulated photoactivatable ratiometric integrator (CaMPARI; Fosque et al., 2015) to monitor calcium activity of hugin neurons. We put larvae for 5 min in a suspension with either dead or live Pe (both with an OD₆₀₀ of 60) and photo-converted the hugin neurons with UV light for 30 s (Fig. 4A). Analysis of the HuginPC neurons showed that they responded with higher calcium activity to live Pe than to dead Pe and PBS controls (Fig. 4B,C). To test whether hugin neuropeptide content in HuginPC neurons was altered after 5 min incubation with dead and live Pe, we performed immunohistological staining of hugin neuropeptide. While we could observe no significant difference when analyzing the HuginPC somas (Fig. 4D), we were able to detect a significant increase in neuropeptide content with live Pe at the protocerebral HuginPC release sites (Fig. 4E). These results indicate that HuginPC bitter taste interneurons are activated by virulent Pe. Inhibition of the hugin neurons thus leads to reduced evasion behavior, thus HuginPC neurons potentially play an important role in processing the bacterial signal in the Drosophila larval brain.

To address whether activity of hugin neurons and the production of hugin neuropeptide are also altered when larvae show starvation-dependent evasion behavior, we expressed CaMPARI in the Hugin neurons. HuginPC calcium activity was analyzed in larvae taken directly out of fly food, or starved for 1 and 2 h on PBS (Fig. 5A). We observed stronger photo-conversion from green to red in fed larvae compared with starved larvae (Fig. 5B). Quantification indicated that with prolonged starvation, HuginPC neurons show less calcium activity (Fig. 5C). To analyze hugin neuropeptide, we performed antibody stainings with hugin antibody after the larvae had been on yeast paste for 5 h or starved for 5 h on PBS. We could detect significantly higher immunolabeling in the somas when larvae were starved versus fed (Fig. 5D). We observed the opposite when analyzing the release site of HuginPC neurons in the protocerebrum (Fig. 5E). Thus, HuginPC neuronal activity and release of the hugin neuropeptide were decreased when larvae were starved. These observations point towards hugin providing a general ‘stop feeding’ command to the larva, where the absence of hugin would make the larvae less sensitive to aversive food sources. Using this simple yet powerful assay, one could also screen for additional potential candidates involved in signaling to the brain during different nutritional states and infection.

In this study, we have reported a behavioral response in Drosophila larvae exposed to pathogenic bacteria. Given that feeding is their strongest innate behavior, it is fascinating to see larvae generating an altered preference when a change in the internal environment was processed by the CNS. Evasion behavior is in many ways reminiscent of the sickness behavior that can be seen in higher animals. Decision making is so finely tuned in Drosophila larvae that a less pathogenic strain of the same bacteria (i.e. gacA Pe and pvf Pe) could not generate a similar evasion phenotype. Moving away from the source of infection provides time to recover from the infection and to fight the infection efficiently. We showed that hugin neurons, by releasing the hugin neuropeptide, play a critical role in processing these signals and inducing evasion behavior. At this point we do not know how the pathogen signal is conveyed from the periphery to the brain, i.e. whether it is detected at the peripheral gustatory organs via a direct neural connection or via the endocrine system in the gut or a combination of the two (Schoofs et al., 2014a,b). As yet unidentified presynaptic peripheral neurons might act on the hugin neurons in order to tune evasion behavior. The evasion assay we have established could be one way to address how information from the periphery reaches the brain.

We thank Bruno Lemaitre for providing the Pe strains. We also thank Ingo Zinke, Andreas Schoofs and Anton Miroschnikow for comments and discussions on the manuscript.