A fungal pathogen that robustly manipulates the behavior of Drosophila melanogaster in the laboratory

Many microbes induce striking behavioral changes in their animal hosts, but how they achieve these effects is poorly understood, especially at the molecular level. This is due in large part to the lack of a robust system amenable to modern molecular manipulation. We recently discovered a strain of the behavior-manipulating fungal fly pathogen Entomophthora muscae infecting wild adult Drosophila in Northern California, and developed methods to reliably propagate the infection in lab.-reared Drosophila melanogaster. Our lab.- infected flies manifest the moribund behaviors characteristic of E. muscae infections: on their final day of life they climb to a high location, extend their proboscides and become affixed to the substrate, then finally raise their wings to strike a characteristic death pose that clears a path for spores that are forcibly ejected from their abdomen to land on and infect other flies. Using a combination of descriptive, histological, molecular and genomic techniques, we have carefully characterized the progress of infection in lab.-reared flies in both the fungus and host. Enticingly, we reveal that E. muscae invades the fly nervous system early in infection, suggesting a direct means by which the fungus could induce behavioral changes. Given the vast toolkit of molecular and neurobiological tools available for D. melanogaster, we believe this newly established E. muscae system will permit rapid progress in understanding how microbes manipulate animal behavior.


Introduction 31 111
To establish an in vivo infection, wild cadavers were co-housed overnight in a confined space with 112 healthy, lab.-reared CantonS D. melanogaster, and exposed flies were monitored nightly for two weeks to 113 identify E. muscae Berkeley cadavers. We repeated this process daily for several weeks before we were 114 able to passage the infection. We were aware that our standard fly diet contained a small amount of the 115 preservative tegosept (0.09%), but did not anticipate that this would be problematic since infected wild flies 116 still died of infection after being housed on this diet for up to eight days ( Fig S3). However, it was only 117 when we began housing flies on food devoid of the preservative tegosept that we were able to successfully 118 passage the infection.

119
Once we had transferred E. muscae Berkeley to lab.-reared flies, we assessed the impacts of several 120 variables on infection efficacy, ultimately arriving at an optimized propagation protocol ( Fig S4). Briefly, 121 we expose flies to E. muscae by embedding six freshly-killed, infected cadavers headfirst in sucrose agar 122 and confining 50 young (eclosed within the past 24 hours) CantonS adults of mixed sex with these cadavers 123 for 24 hours in a cool, humid environment on an inverted 12:12 light:dark cycle. After 24 hours, 124 confinement is relieved and flies are transferred to a medium free of tegosept. Exposed flies are housed at 125 room temperature with moderate humidity and monitored daily for death by fungus. both morphologically and behaviorally until they begin to exhibit end-of-life behaviors (Fig 2A). Exposed 139 flies bear melanized scars that form following spore entry through the cuticle, which are most apparent 140 when the point of entry is the pale ventral abdomen. However, not all flies that are penetrated by the fungus 141 are successfully infected and killed, as we have observed animals with scarring that survive beyond seven 142 days after exposure, and have found that housing exposed flies on diet with anti-fungal significantly 143 improves survival (Fig S4). At 72 hours after exposure and beyond, infected flies generally have more 144 opaque abdomens than uninfected flies due to abundant fungal growth. Under our conditions, ~80% of 6 CantonS flies are killed four to seven days after exposure to E. muscae Berkeley, with the majority of deaths 146 occurring at 96 and 120 hours (Fig 2B). While by eye infected animals behave normally until the onset of 147 end-of-life behaviors, analysis of infected fly activity revealed that infected flies exhibit a marked decrease 148 in total activity compared to healthy counterparts beginning about 36 hours before time of death, which 149 presently is the best indication of imminent mortality for a given fly (Fig 2C).

150
On the last day of life, E. muscae Berkeley infected flies stop moving 0-5 hours before sunset (Fig 2D).

151
Taking time of last movement as a proxy for time of death, this observation agrees with reports of E. muscae 152 in house flies [9]. Also consistent with previous reports, flies exposed to E. muscae Berkeley and housed 153 under complete darkness die sporadically throughout the day rather than in a gated fashion (Fig S5, [9]).

154
As healthy flies housed for 168 hours in complete darkness maintain circadian rhythm, this suggests that 155 environmental cues and/or a fungal clock are required to coordinate the timing of death, as has been 156 previously suggested [9]. Of note, flies housed in complete darkness are still observed to die in elevated 157 positions. This suggests that summiting behavior relies predominantly on gravitaxis rather than phototaxis.   perturbations (i.e. poking with a paintbrush or jostling their container), they will not take flight. After they 181 have lost the ability (or desire) to fly, moribund flies will begin to exhibit a shaky and slowed gait which is 182 usually coincident with an upward climbing or movement towards a vertical surface. Many flies reach 183 elevated positions before they lose the ability (or desire) to continue moving (even when perturbed by the 184 experimenter), but some succumb to immobility before they leave the ground. When provided a thin, 185 wooden dowel as a summiting substrate, more flies are observed to die in elevated positions, mostly on the 186 dowel itself (Fig 2G). Interestingly, we have noticed that when drips of medium are present on the side of 187 a vial, flies that die on the side of the vial are preferentially found on these drips. It is unclear if this indicates 188 a preference for the medium as a climbing substrate (versus the smooth plastic of a fly vial) or if the flies 189 are attempting to eat until their very last.

190
Once the fly stops walking, it extends its proboscis until it makes contact with the surface on which 191 it is standing (Fig 2E). The extension of the proboscis is shaky and can occur slowly relative to extension 192 in response to a nutritive stimulus, and we have observed in multiple instances that the labella of infected 193 flies do not spread as is typically observed when uninfected flies eat (see Movie S1). Typically, once the proboscis has made contact with the surface, the fly may move its legs in what appears to be an apparent 195 attempt to escape, but the material that emanates from the proboscis is sufficient to keep it anchored in 196 place. After the proboscis has adhered, the fly then begins to raise its wings up and away from the dorsal 197 abdomen (Fig 2F). This process has been observed to take on the order of ~10 minutes, with wing raising 198 occurring in small bursts, reminiscent of the inflation of a balloon (see Movies S2 and S3). Curiously, a 199 persistent minority of infected flies die with their wings lowered down onto their abdomen rather than with 200 wings elevated (Fig 2H). By applying pressure to the thorax of these flies, the wings are observed to 201 "toggle" into the upright position, suggesting that the same muscles are involved in raising and lowering.

202
The fly may continue to twitch its legs and antenna for several minutes after the wings have reached their 203 final position but will shortly cease moving.

204
After death, the fungus inside of the fly continues to differentiate into conidiophores, conidia-205 launching structures, that grow out into the environment through weak points in the fly's cuticle. Over the 206 course of several minutes, each conidiophore forms a single primary conidium (Movie S7) which, upon 207 maturation, is forcibly ejected into the environment. Using time lapse imaging, we observe that conidia 208 begin to launch approximately five hours after sunset and continue doing so for several hours at ambient 209 temperature and humidity ( Fig 3A). We observed that conidia form and launch asynchronously within a 210 given cadaver, and not all conidiophores are guaranteed to launch what appear to be mature conidia. Using 211 high speed videography, we were able to capture the motion of conidial ejection (Fig 3B), and determine 212 that conidia leave the conidiophore at an initial velocity of ~21 miles per hour (~9.4 meters/second). These 213 speeds are comparable to those observed in coprophilous fungi, which are among the fastest observed 214 velocities of organisms relative to their size known in the natural world [25]. In addition, we obtained high 215 speed footage of primary conidia landing (Fig 3C), which shows conclusively that conidia and halo land 216 concurrently, an observation that supports the fungal canon mechanism of spore discharge [26].

233
To compare E. muscae Berkeley with other reported isolates, we collected primary conidia and 234 measured their key morphological traits (e.g. Fig 3D). Our measurements are most similar to primary 235 conidia from E. muscae sensu strictu rather than other members of the E. muscae species complex (Table   236   1

268
We first examined the percentage of reads that aligned to host or fungus in each of our time course 269 samples ( Figure 4A). We observe that E. muscae

280
are separated into controls (healthy animals who were mock exposed), exposed (animals who exposed to 281 E. muscae Berkeley and were alive at the time of sampling) and cadavers (animals who were killed by E.

301
of genes overexpressed (red) or under-expressed (blue) in exposed animals compared to controls.

303
We next surveyed gene expression in E. muscae Berkeley across our exposed samples. As different 304 exposed individuals vary in their rate of infection by E. muscae Berkeley, we reasoned that it would be most 305 informative to order our samples based on E. muscae Berkeley titer, which we approximated using the 306 proportion of reads that aligned to E. muscae Berkeley of total reads aligned to either the E. muscae Berkeley 307 or D. melanogaster references ( Figure 4B). The bulk of transcripts are not expressed until three days after 308 exposure, which could simply be a consequence of the fungus being low abundance until this time point.

309
Interestingly, there are three groupings of genes (Groups i-iii) that demonstrate patterns that cannot be

316
Next, we examined host gene expression patterns across all of our samples, again ordering samples 317 based on the proportion of E. muscae Berkeley aligned reads among all total and clustering genes by 318 expression pattern ( Figure 4C). Host gene expression segregates into six major groupings (Table S1). Group

331
Following our initial overview of host transcription, we next looked at genes that were consistently 332 different between control and exposed samples from 24-72 hours ( Figure 4D). We excluded all cadaver starving is consistent with these enrichments and also with the observation that basic cell metabolism 342 (macromolecule synthesis) is substantially decreased at 72 hours ( Fig S7).

343
The same analysis shows that genes that are over-expressed in exposed animals are enriched for

407
At 48 hours after exposure, fungal cells are consistently observed in the brain and/or ventral nerve 408 cord (VNC; 4 out of 5 samples). In the one case where fungus had not invaded the nervous system, hyphal bodies were apparent immediately adjacent to the brain, abutting the blood brain barrier.

493
Fungus is observable within the host brain and VNC, though overall fungal titer is still quite low. As the 494 abdomen is the most likely point of initial entry for the fungus (it is the biggest target for the fungus to hit),

504
While the trajectory of infection is consistent, it is important to recognize that just because two 505 animals have been exposed for the same duration of time that these two animals will not progress through infection identically. In our RNAseq data, we noticed that the host gene expression in exposed animals at 507 24 and 48 hours tended to be more variable than those for 72 hours. We imagine that this was due to chance, 508 that we picked animals that were at similar points in infection at 72 hours, whereas we picked animals that 509 were more variable at other time points. This may at least partly explain why we observed so much 510 differential expression at the 72 hour time point in exposed versus controls whereas less was observed at

526
Our data show that there is a robust initial response to E. muscae Berkeley exposure. Many of the 527 immune genes that are induced with E. muscae Berkeley have also been observed to be induced by exposure 528 to other, more generalist fungal pathogens (e.g. Beauveria bassiana, Metarhizium anisopliae [33,34]).

529
These data clearly indicate that the host detects an invader early on in infection. Furthermore, there is a 530 detectable immune response through the length of infection (Fig S8), though at this point we cannot say if 531 this response is a slow disengagement of the initial response or stimulated de novo by the growing fungus. not know if every instance of a spore hitting a fly leads to a productive fungal infection. There is some evidence to the contrary: we have consistently observed that some highly-exposed flies die prematurely.

541
These animals are generally smaller than others in the vial and are often covered in spores. This could 542 indicate that getting hit by too many spores (an unlikely outcome in the natural world) leads to an 543 overwhelmed fly (e.g. overactive immune system or accelerated fungal growth) that dies before being 544 manipulated. These flies do not sporulate, though it is possible that they do produce resting spores. On the 545 other hand, we have observed that survival of exposed flies is substantially increased when flies are exposed 546 to small quantities of anti-fungal. This indicates that there are ways of either halting or slowing an infection, 547 though whether the fly's immune system is generally capable of doing this is unknown.

548
Though five of our six cadavers sampled for transcriptomics have similar levels of immune gene 549 transcripts compared to E. muscae Berkeley-exposed animals sampled at 72 hours, the sixth cadaver

565
Berkeley could alter host behavior before jumping to this conclusion.

566
We can imagine four general mechanisms by which E. muscae is able to achieve behavioral 567 manipulation. The fungus could invade the nervous system in order to localize adjacent to and impinge on 568 the activity of particular neurons through chemical or physical means. However, we are skeptical that this 569 is the case as our observations do not support specific localization of the fungus in the CNS.

570
A second possibility is that the fungus invades the nervous system in order to gain access to either 571 a particular group or groups of neurons or all neurons generally, but does not localize within the CNS in a 572 stereotyped manner. Rather it is sufficient that it has crossed the blood brain barrier, which insulates the 573 nervous system from the activities in the hemolymph and allows for the selective transport of compounds to and from the hemolymph, allowing the fungus to modulate the activity of neurons by secreting 575 compounds that diffuse throughout the CNS. The secreted compounds could be specific, only altering the 576 activity of a subset of susceptible neurons, or could be more general, changing activity over many or all 577 neurons.

578
A third possibility is that the fungus does not need to invade the nervous system in order to change 579 the host's behavior. The fungus could be secreting a compound into a hemolymph that is capable of crossing 580 the blood brain barrier and altering neuronal activity. Alternatively, the fungus could be secreting a 581 compound into the hemolymph that changes the host's internal state (either directly or by leading the host 582 to respond in a way that causes the internal state to change) which leads the animal to respond by executing 583 one or more of the end-of-life behaviors.

584
Lastly, it's possible that the fungus does not secrete compounds to induce these behaviors, but by 585 destroying fly tissues elicits the series of observed behaviors. While we believe this last scenario to be 586 highly unlikely, it cannot yet be ruled out.

587
For these last two proposed mechanisms, the fungus would not need to invade the CNS in order to 588 affect behavior. In these cases, the fungus could be invading the CNS as a means of escaping immune

598
Interestingly, another Entomophthoralean fungus, Strongwellsea magna, is also known to invade the 599 nervous system of its lesser house fly host (Fannia canicularis) during infection [36]. In this case, the 600 author proposed that this did not have consequences for behavior.

601
Surprisingly, transcriptomic analysis of dissected brains from exposed females at 24, 48 and 72 602 hours with confirmed E. muscae Berkeley infections failed to show differential gene expression compared 603 to uninfected controls but did show an increase in E. muscae Berkeley titer (taking E. muscae Berkeley 604 reads as a proxy) (Fig S9). Though these samples were not collected at the point of behavioral manipulation 605 by the fungus, it is surprising that there are no major transcriptional changes within the brain at these time 606 points, and suggests that behavioral modification may be largely independent of transcriptional changes in 607 the brain. was pushed down to confine the flies within a few centimeters to improve the likelihood that they would 706 encounter flying spores. Leaving the exposed flies with the spent cadavers was initially problematic as we 707 were working without access to anesthesia or a microscope and had to identify new cadavers by naked eye.
Additionally, the raw banana began to ferment and break down, leading to excess moisture which was 709 prematurely killing some of our exposed flies. To avoid these issues, the exposed flies were transferred to    Unexposed flies (i.e. controls) were always processed before proceeding to exposed flies. CO2 pad was 777 wiped down with 70% ethanol between vial types to present cross-contamination. DAMs were loaded from 778 bottom to top row, filling a row and securing each tube with rubber bands before proceeding to the next.

866
Brains were individually dissected and sampled from first three control and then three exposed females.

867
Each animal was dissected in sterile 1x PBS in its own disposable dissection chamber (35mm petri dish 868 lined with 2-3% agar) and dissecting forceps were treated with 3.5% hydrogen peroxide then rinsed with 869 sterile water between samples to prevent nucleic acid carryover. The body of each animal was saved and electrophoresis to confirm that all exposed animals had come into contact with E. muscae Berkeley and that 877 control animals were uninfected.
Two mock and two exposure vials were started daily for seven days each with 50 CantonS WF flies 0-1 912 days old with either 0 (mock) or 6 (exposure) cadavers embedded in AS. Flies were incubated for the first 913 24 hours at 18C confined to 2 cm with cadavers, then moved to 21C where the confinement was relieved.

914
Flies were transferred to GB+ at 48 hours where they continued to be housed at 21C. Vials were sampled

1045
ANOVA grouping control vs. exposed samples. Genes with p-value < 0.001 are shown in color. Right:

1046
Panther GO-term analysis (complete biological process) of genes overexpressed in exposed animals (red)