When to care and when to kill: termites shape their collective response based on stage of infection

Termites defend their colonies from disease using an array of social behaviours, including allogrooming, cannibalism, and burial. We tested how groups of eastern subterranean termites (Reticulitermes flavipes) deploy these behaviours when presented with a nestmate at different stages of infection with the entomopathogenic fungus Metarhizium anisopliae. As expected, the termites groomed pathogen-exposed individuals significantly more than mock-treated controls; however, grooming levels were significantly higher after spore germination than before. Cannibalism became prevalent only after exposed termites became visibly ill, and burial was rarely observed. These results demonstrate that termites employ different strategies depending on the stage of infection that they encounter. Grooming intensity is linked not only to pathogen presence, but also to germination status, and, given the temporal correlation between cannibalism and visible signs of illness, the host may play a role in triggering its own sacrifice.


27
Social insects have evolved collective behaviours to protect their colonies from disease. 28 These social immune defences, which include pathogen avoidance, prophylactic secretions, 29 grooming, and corpse disposal, act to protect the colony as a whole, at times at the expense 30 of individual members [1,2]. In this latter case, sick colony members are identified and killed 31 to prevent the spread of disease [1,3,4]. Regulation is therefore essential, both to prevent 32 unnecessary killing and to allow the colony to dynamically adjust its investment in other 33 defences [1]. 34 Of all the social insects, the social Hymenoptera are the most well-studied. In honeybees 35 (Apis spp. Linnaeus) activation of the physiological immune system by an infection results in 36 a changed cuticular hydrocarbon profile [5], which can then trigger the removal of the 37 infected bee by other members of the hive [6]. Likewise, workers respond to volatiles emitted 38 by sick or injured brood by removing them from the hive [7,8], and factors external to the 39 host, such as the odour of a parasite or pathogen inside a brood cell [9], can also play a role. 40 In ants, the situation is similar: invasive garden ant (Lasius neglectus Van Loon,Boomsma & 41 Andrásfalvy) workers groom fungus-exposed pupae to prevent disease, but kill them if 42 alerted to an internal infection by a change in cuticular hydrocarbons [4]. European fire ant 43 (Myrmica rubra (Linnaeus)) workers also behave more aggressively toward fungus-infected 44 adult nestmates once internal proliferation has begun [10]. 45 Comparatively little is known about how termites (Blattodea: infraorder Isoptera) shape their 46 social immune response based on the stage of infection encountered. There is broad 47 host is moribund (an internal infection has been established); and (4) the host is near death. 118 To obtain individuals at each stage of infection, termites were treated with a 1 x 10 8 119 conidia/mL suspension of M. anisopliae or 0.05% Tween 80 as a control, then maintained 120 individually for 2, 12, 15, or 20 hours. These four incubation times were chosen based on the 121 results of a preliminary experiment (Supplementary Material). The reasoning for each is also 122 summarised in Table 1. 123 were 24 replicates of the M. anisopliae treatment (8 per colony for three colonies) and 12 of 126 the control treatment (4 per colony) to control for the effects of handling and isolation. These 127 were split evenly across two experimental replicates. In the first experimental replicate, the 128 conidia used in the experiment were freshly harvested from half of one PDA plate. In the 129 second experimental replicate, conidia were freshly harvested from the other half. 130 Preparation of Petri dish nests 131 Each Petri dish nest consisted of a Petri dish (94 x 16 mm, without vents), two thick Pall 132 cellulose pads (45.5 mm diameter, 0.9 mm thick), two thin Whatman No. 5 filter paper discs 133 (47 mm diameter, 0.2 mm thick), and one standard glass microscope slide (76 x 26 mm). 134 Each thick cellulose pad was placed on top of a thin filter paper disc. The two paper stacks 135 were then placed side-by-side in the Petri dish, with one of the stacks trimmed on one side to 136 fit. Finally, a glass microscope slide was placed on top ( Figure 1). The thickness of the paper 137 stacks (ca. 1.1 mm) was experimentally determined such that when termites dug in the paper 138 under the glass slide, they were able to move freely but had too little space to leave an 139 opaque "ceiling" over their tunnels. The paper was moistened with 3.5 mL tap water. 140

142
Forty-five workers were added to each Petri dish, 1 soldier was added to 20 dishes per 143 colony (10 in the first experimental replicate, 10 in the second), and every dish received 3 144 representatives of the reproductive caste, with the exception of 7 plates from colony 5, which 145 only received 2 in the second experimental replicate due to difficulty retrieving nymphs from 146 the colony. R. flavipes caste ratios vary [33], but workers are by far the dominant caste [19]. 147 The numbers of reproductives and soldiers were taken into account in the statistical analysis, 148 but neither had a significant effect. 149 In total, each dish contained 48 or 49 termites, not including the focal individual. Becker [30] 150 recommends using a minimum of approximately 50 termites to maintain R. flavipes in the lab. 151 In pilot experiments, we confirmed that groups this size could survive for three or more 152 weeks in a Petri dish setup, and that they displayed typical social and hygienic behaviours, 153 including cannibalism and burial. Smaller groups had lower survivorship and sometimes 154 displayed abnormal behaviour, such as leaving corpses uneaten and unburied. 155 The dishes were sealed with parafilm to prevent desiccation and left in a dark room at 27°C, 156 70% humidity for two weeks. At least half an hour prior to a behavioural experiment, a cotton 157 swab was used to sweep debris off the glass slide. This was necessary to ensure a clear 158 view into the nest. 159 Marking focal termites 160 Focal termites were marked with Nile blue (AppliChem GmbH, Darmstadt, Germany), a 161 moderately toxic fat-soluble stain that has previously been used to mark termites in 162 behavioural studies [16,19,34]. As an internal stain, it cannot be removed and does not 163 interfere with grooming. 164 Our protocol is a faster version of Evans' fast marking technique [35]. As termites will 165 swallow any liquid that they are immersed in, we dispensed with his desiccation step and 166 lastic etri dish without vents mm all cellulose pad 5.5 mm . mm thick lass slide mm hatman o. 5 filter paper mm . mm thick immersed all termites that needed to be stained in 0.025% Nile blue. This is the minimum 167 concentration needed to reliably stain R. flavipes. 168 Large (≥ 4 mm) workers were poured into 2 mL microcentrifuge tubes, one per colony, using 169 a small funnel. Only workers that appeared healthy and active were used. Sufficient 0.025% 170 Nile blue was added to cover them, and they were flicked to mix for 1 minute, then tipped out 171 onto a dry cellulose pad. Initially, all appeared unstained. The termites were transferred to 172 one of three labelled round plastic containers (ca. 52mm inner diameter), one per colony, 173 each lined with a clean cellulose pad moistened with 1 mL tap water and closed with a tight-174 fitting lid, and then left overnight in a dark room at 27°C, 70% humidity. Only termites that 175 were successfully stained and appeared healthy and active were used in the subsequent 176 experiment. Because the intensity of the colour varied widely, and because of the known 177 toxicity of the stain, termites of different shades were randomly distributed amongst 178 treatments and controls. 179

Preparation of conidial suspensions 180
Conidia were harvested after a minimum of one month of growth. A sterile cotton swab 181 moistened with sterile 0.05% Tween 80 was used to wipe the conidia off the plate and 182 suspend them in sterile 0.05% Tween 80. The suspension was inverted and vortexed to mix, 183 then filtered through a piece of sterile cheese cloth that had been folded to reduce the 184 effective pore size. The filtered conidia were washed by centrifuging for 10 minutes at 5000 g 185 in a centrifuge cooled to 4°C, discarding the supernatant, and resuspending the pellet in 186 sterile 0.05% Tween 80. This step was performed a total of three times. 187 A BLAUBRAND® Thoma counting chamber (depth 0.1 mm; BRAND, Wertheim, Germany) 188 was used to estimate the concentration of the conidial suspension. Conidial suspensions 189 were adjusted to 1 x 10 8 conidia/mL with 0.05% Tween 80, aliquoted for ease of use, and 190 used within 48 hours. Suspensions were stored at 4°C when not in use. 191 To ensure that the conidia were viable, PDA plates were streaked with conidia from an 192 aliquot of the same 1 x 10 8 conidia/mL suspension used to inoculate the termites. The plates 193 were parafilmed and placed upside-down in the same room as the termites (27°C, 70% 194 humidity). After 21 hours, at least 300 conidia were evaluated for germination at 200 to 400x 195 magnification on one of the plates to calculate the germination rate. A conidium was 196 considered germinated if the length of the germ tube was at least half the diameter of the 197 conidium. For confirmation, at least 100 conidia were evaluated in the same manner on the 198 second plate. The germination rate was ca. 94% in the first experimental replicate and ca. 199 98% in the second. A germination rate lower than 90% would have indicated a problem with 200 the conidial suspension. 201 Inoculation with conidia or 0.05% Tween 80 202 For the M. anisopliae treatment, previously-marked (blue) termites were placed in a round-203 bottomed 2 mL microcentrifuge tube, then covered with the 1 x 10 8 conidia/mL suspension to 204 a volume of 42 µL per termite. The tube was flicked to mix for 10 seconds, then poured out 205 onto a dry cellulose pad. Termites that remained inside were tapped out, or, if needed, 206 carefully removed with soft forceps. When the termites had recovered enough to walk, they 207 were transferred one-by-one into separate Petri dishes, each containing a cellulose pad 208 moistened with 1 mL tap water. The dishes were sealed with parafilm to prevent desiccation. 209 Control termites were immersed in sterile 0.05% Tween 80 (42 µL per termite) instead of the 210 conidial suspension and handled in the same way. This inoculation method is a variation on 211 that used by Yanagawa and Shimizu [15]. 212 The M. anisopliae-treated and control termites were incubated for 2, 12, 15, or 20 hours at 213 27°C, 70% humidity before use in the behavioural experiment. Scan sampling [36] was used to observe the interactions between the focal termite and its 224 nestmates within each Petri dish nest. Scans typically took less than 1 minute. They were 225 performed every 5 minutes for a total of 3 hours using a magnifying glass (up to 3x 226 magnification) to better distinguish between similar behaviours and a Samsung S7 227 smartphone as a digital voice recorder. interactions with nestmates cannot be seen.  (Table S1).

318
Lower and upper hinges correspond to first and third quartiles, the upper whisker extends to the largest value if it is no greater than 1.5 319 times the inter-quartile rage from the hinge, and the lower whisker extends to the smallest value if it is no smaller than 1.5 times the 320 inter-quartile range from the hinge.

321
Only workers were observed grooming the focal termites, and grooming was visibly more  Table S2). 2h/M.a+ was not significantly 326 different from any of the controls, and 15h/M.a+ and 20h/M.a+ were not significantly different 327 from each other. Non-focal termites in treatments with more intense grooming were 328 frequently observed to engage in vibratory displays (jittering), a known pathogen alarm 329 response [11,48]; however, our sampling method, which focused on direct interactions with 330 the focal individual, precluded analysis of this behaviour. 331  (Table S2).

335
Cannibalism 336 The probability of remaining unharmed during the observation period was significantly 337 different from the controls, which experienced no cannibalism, in the 15h/M.a+ (z=3.958 338 p=0.00167) and 20h/M.a+ (z=4.809 p<0.001) treatments, but not in 2h/M.a+ or 12h/M.a+ 339 (Table S3) Figure 5). In all but two cases, the 344 first cannibalism-related state recorded was biting. In those two exceptions, both in 345 15h/M.a+, it was dismemberment. The previous scan recorded intense grooming of the focal 346 individual by 5 to 6 groomers: it is possible that that state was misidentified, or that biting 347 began between scans. Cannibalism was performed primarily by workers, but on two 348 occasions, a brachypterous neotenic was observed to also partake. This is too few for meaningful statistical analysis. In each case, the focal termite appeared to 358 be alive but moribund and largely immobile at the beginning of the burial process. This 359 immobility was caused by nestmates in one 20h/M.a+ replicate: the legs were first bitten off, 360 and the maimed termite was left for approximately half an hour before burial began. 361 There was no sudden switch from grooming or cannibalism to burial. In one 20h/M.a+ 362 replicate, the focal termite was initially groomed, then bitten, then had a piece of paper 363 placed on top of it (burial), then groomed again for half an hour, during which time the paper 364 was removed, then bitten again. Burial did not resume, and the termite was eventually 365 dismembered. 366

367
Our results demonstrate that R. flavipes colonies employ different collective immune defence 368 strategies at different stages of infection with M. anisopliae. Before conidia germinate, the 369 social immune response is dominated by grooming; however, contrary to our first hypothesis, levels of grooming rise significantly after conidial germination, and it becomes visibly more 371 intense. Contrary to our second hypothesis, the onset of cannibalistic behaviour coincides 372 with the stage of infection in which the termite becomes moribund, with a more rapid switch 373 to cannibalism at later stages. All cannibalised individuals were eaten alive. This is consistent 374 with observations by Rosengaus and Traniello [3], who observed that termites were usually 375 cannibalised when near death, but contradicts Strack [24] who observed more "agonism" 376 toward healthy individuals that had been thickly dusted with conidia. Burial was rarely 377 observed, reinforcing the view that termites preferentially eliminate sick individuals through 378 cannibalism [3,23]. 379 The unexpectedly low levels of grooming observed before conidial germination may be 380 explained by their weak attachment to the cuticle: since most conidia can be removed within 381 hours by relatively few individuals [26], there may be no reason to divert resources away 382 from other colony functions or endanger additional members of the colony. The effectiveness 383 of allogrooming, even at the observed low intensity, can also be seen in survivorship studies 384 [12]. Increased levels of grooming after germination, then, could be linked to increased 385 physical difficulty removing fungal material, especially after germ tube penetration. 386 This explanation is unsatisfying, because the longer a pathogen persists on or in members of 387 a colony, the more we would expect it to affect colony fitness. Conidia-exposed, non-388 moribund individuals are mobile and can transfer conidia to many colony members [13,14], 389 all of which would need to be groomed by workers that could otherwise be performing other 390 tasks. Should the infection progress to the next stage, the risk to the colony would increase 391 significantly. This should favour early "clearance" of the infection from the colony via 392 aggregation and intense grooming of conidia-exposed individuals, but that is not what we 393

observed. 394
A second possibility is that the fungus-associated molecules that stimulate grooming (e.g. the 395 fungal "odour" [49]) are partly masked, or present in lower quantities, before germination. 396 Based on response threshold models of division of labour in social insects [50], even partial 397 masking would result in a weaker collective grooming response with fewer participating 398 workers. This need not be a specific adaptation to evade termite social immunity, nor would 399 we expect it in a generalist entomopathogen. Masking of immunogenic components of the 400 fungal cell wall before (but not after) germination has previously been reported in an 401 opportunistic human pathogen, Aspergillus fumigatus Fresenius [51]. Should this prove to be 402 the case in M. anisopliae, it could be harnessed to develop strains with higher epizootic 403

potential. 404
In contrast to grooming, in which fungal factors appear to be the primary trigger, the strong 405 temporal correlation between moribundity and cannibalistic behaviour suggests that the host 406 plays a central role in its own sacrifice. Focal termites appeared healthy at 12 hours and 407 moribund (a reliable sign of internal infection [31]) at 15, and cannibalism was only prevalent 408 in the 15h/M.a+ and 20h/M.a+ treatments. Even in the 12h/M.a+ treatment, which was not 409 significantly different from the control, there was an uptick in cannibalism in the last half hour 410 of the observation period, i.e. at approximately 14.5 hours post-exposure. With the caveat 411 that this is a correlation, and that moribundity could coincide with some fungus-derived 412 stimulus reaching the necessary threshold for cannibalism, the hypothesis that sick 413 individuals might flag themselves for destruction is supported by research in the social 414 Hymenoptera. Ant pupae "advertise" the presence of an internal infection through modified 415 cuticular hydrocarbon profile [4], and aggressive behaviour was observed toward adults at 416 the same stage of infection [10]; however, more work will be required to determine whether 417 the social Blattodea and the social Hymenoptera have independently evolved separate 418 mechanisms to identify fatally ill colony members, or if they have separately co-opted 419 evolutionarily conserved sickness cues for social immune defence. 420

421
We have demonstrated that termites can deploy different collective immune defences when 422 confronted with a worker at different stages of infection with an entomopathogenic fungus. 423 Whereas grooming is favoured earlier in the infectious process, moribund individuals are 424 readily sacrificed to protect the colony. Cannibalism appears to be triggered by some factor 425 associated with moribundity: what this might be remains unclear. Paradoxically, the termites 426 did not display a robust social immune response at the earliest stages, when conidia had not 427 yet germinated, although grooming was somewhat elevated. This may indicate that the 428 ungerminated fungus is less visible to the "social immune system" of the colony but this 429 hypothesis remains to be tested. 430 This study adds to the body of knowledge surrounding termite social immunity and sheds 431 light on how colonies resist fungal disease and regulate destructive immune behaviours. By 432 dividing the infection into stages [29] and studying how the social immune response differs 433 over time, we can better understand how termites, and insects in general, defend their 434 colonies from disease. 435