Impacts of fungal disease on dyadic social interactions in a wild agamid lizard

Emerging infectious fungal diseases are responsible for the extinction of myriad species across a range of phyla. As recently shown by the COVID-19 pandemic, social transmission can be key to disease spread, and in this context, humans are not alone in trying to be alone. In group-living species, individuals have been shown to use social behaviour to avoid infection; diseased individuals can isolate from the group, or healthy animals can avoid diseased conspeci ﬁ cs. However, little is known about social behaviour as a mechanism to avoid fungal infection. In this study, we investigated the extent to which wild urban eastern water dragons, Intellagama lesueurii , a gregarious reptile, modify their social behaviour as a response to infection with a recently emerged infectious fungal disease, caused by the pathogen Nannizziopsis barbatae . Using individual data from a long-term study population inhabiting Roma Street Parkland in Brisbane's Central Business District (QLD, Australia) and focal sampling, we tested whether dragons exhibit self-isolation and social-distancing behaviours in the context of dyadic social approach events. Our results suggested that while the presence of the fungal disease had no effect on individuals' social behaviour, its severity did. Speci ﬁ cally, we found that (1) diseased individuals were no less social than their nondiseased conspeci ﬁ cs, (2) nondiseased individuals did not avoid or spend less time with diseased conspeci ﬁ cs, and (3

Emerging infectious fungal diseases (EIFDs) have caused some of the most significant conservation crises in modern times, driving population declines and threatening numerous wildlife species with extinction (Frick et al., 2010;Langwig et al., 2012). Some of the most detrimental recent EIFDs have included Chytrid fungus, white-nose syndrome (WNS) and snake fungal disease (SFD) (caused by Batrachochytrium dendrobatidis, Pseudogymnoascus destructans and Ophidiomyces ophiodiicola, respectively; Fisher et al., 2012;Hileman et al., 2018;Lorch et al., 2011). Fungal diseases have been responsible for 72% of disease-related extinction events, with the proportion of documented fungal disease records increasing seven-fold in just 15 years (Fisher et al., 2012). EIFDs are thus now considered one of the greatest threats to extinction for host species (Mitchell et al., 2008). Yet, EIFDs and their transmission remain poorly understood (Allender et al., 2015;Fisher et al., 2012;Warnecke et al., 2012).
In animals, early detection and avoidance of infected or diseased individuals are considered the first lines of defence against disease (Curtis, 2014). Various species have been found to change their behaviour in the presence of infectious disease; diseased individuals actively 'self-isolate' to limit pathogen transmission, while healthy individuals 'socially distance' to avoid initial infection (Lopes, 2020). While active self-isolation remains poorly understood outside of studies on eusocial insects (Heinze & Walter, 2010;Rueppell et al., 2010;Stroeymeyt et al., 2018), social distancing is much better understood, as healthy individuals that can recognize disease cues and modify their social behaviour to 'socially distance' from affected conspecifics have been documented across a range of taxa (Albery et al., 2020;Behringer et al., 2006;Croft et al., 2011;Kiesecker et al., 1999;Paciência et al., 2019). For example, healthy lobsters, Panulirus interruptus, were shown to avoid dens harbouring diseased conspecifics, preferring to share dens with other healthy individuals (Behringer et al., 2006). The absence of social distancing in the presence of infectious diseases has, however, also been documented. For instance, house finches, Haemorhous mexicanus, have demonstrated a preference for feeding near conspecifics infected with visible signs of transmissible conjunctivitis (Bouwman & Hawley, 2010).
Previous studies, such as those outlined above, have primarily explored social behavioural modifications in the context of emerging infectious diseases caused by viruses (Behringer et al., 2006) and parasites (Croft et al., 2011), with a distinct scarcity of studies addressing their expression in the context of fungal pathogens and EIFDs. As such, there remains a dearth of knowledge about social behaviour as a mechanism to avoid fungal infection, with the only known examples being tadpoles (i.e. American bullfrog tadpoles, Lithobates catesbeianus, and Candida humicola infection; Kiesecker et al., 1999) and some eusocial insects (i.e. honey bees, Apis mellifera, and Nosema sp. infection; Kralj & Fuchs, 2010).
A newly identified EIFD, yellow fungus disease, caused by Nannizziopsis barbatae, has emerged repeatedly in free-living reptiles across Australia, and poses a significant threat to Australia's herpetofauna biodiversity (Peterson et al., 2020). Reptile-associated Nannizziopsis species demonstrate all aspects of high virulence: they are contagious (Par e et al., 2006;Sigler et al., 2013;Thomas et al., 2002), have no known treatment (Cabañes et al., 2014;Hedley et al., 2010), have spread geographically (Peterson et al., 2020), and have caused lethal infection in phylogenetically and ecologically distinct species (Cogger, 2015;Peterson et al., 2020). In this study, we investigated the extent to which eastern water dragons, Intellagama lesueurii (hereafter referred to as dragons) in an urban free-living population afflicted by N. barbatae, modify their social proximity behaviours as a response to infection with this EIFD. Using longitudinal data on individuals and focal sampling, we tested for the presence of self-isolation and social distancing in the context of dyadic social approach events.
Given the limited previous research, it is not yet known whether N. barbatae is directly transmitted from host to host and/or indirectly transmitted between substrate and host. It may follow a similar transmission pathway to many fungal pathogens which persist saprophytically in the environment (Allender et al., 2015), thus facilitating transmission between substrate and host (Rowley & Alford, 2007). If N. barbatae does follow an indirect transmission pathway and there is limited direct transmission from host to host, then we would not expect to observe any changes in social behaviour to avoid diseased individuals. As such, we tested the predictions that (1) diseased individuals would be no less social than their nondiseased conspecifics, (2) nondiseased individuals would not avoid or spend less time with diseased conspecifics and (3) neither diseased nor nondiseased individuals would avoid more severely diseased conspecifics.

Study Population
Our study was conducted at Roma Street Parkland (À27 27 0 46 0 S, 153 1 0 11 0 E) in the Central Business District of Brisbane, Queensland's capital city. The parkland spans approximately 16 ha and contains a combination of simulated native vegetation and arranged garden beds, differing in their structure and access to water. The park is nestled within Brisbane's largest transit centre and is surrounded by roads and urban infrastructure .
The study focused on an urban population of dragons inhabiting the parkland. Eastern water dragons are a large semiaquatic agamid lizard, native to the east coast of Australia. They display malebiased sexual dimorphism, with males being larger than females (Thompson, 1993), along with other distinct morphological differences (Baird et al., 2012). This species is long lived, with a life span of up to 15 years and a generation time of 4 years (Thompson, 1993), and exhibits a defined social structure that is similar in complexity to those identified in some long-lived mammals (Sah et al., 2018); they are highly gregarious, show both preferences and avoidances for certain conspecifics (Piza-Roca et al., 2018;Strickland & Frere, 2019) and exhibit sex differences in social strategies Strickland & Frere, 2017).
Roma Street Parkland's population of dragons has an estimated size of over 300 adult individuals at a 1:1 sex ratio . The dragons have been the subject of longitudinal behavioural, morphological and genetic studies since 2010, and are habituated to human presence (Strickland & Frere, 2017). The parkland's urban boundaries prevent immigration and emigration with other geographically close populations (Littleford-Colquhoun et al., 2017). It is estimated that 34.5% of the population currently exhibit clinical signs of N. barbatae infection (Peterson et al., 2020) and are thus considered 'diseased'.

Focal Surveys
To investigate the effects of N. barbatae infection on the social behaviour of dragons, we conducted repeated behavioural focal follows during morning or afternoon surveys. Focal follows were performed by following a transect line through the study area from 0800 to 1500 hours 3 days per week from September 2020 to January 2021. This period represents the breeding season, during which dragons are known to be more social (Strickland & Frere, 2019). We defined a social approach as one individual approaching another to within 2 m ; see Appendix 1 for further details). This spatial proximity distance is used as a proxy for social tolerance given that eastern water dragons will otherwise react aggressively towards one another (Strickland & Frere, 2017). We also excluded any mating attempts or antagonistic interactions given that we sought to capture affiliative behaviour which is defined as nonaggressive social behaviour and can range from social tolerance of conspecifics to group living and cooperation (Ward & Webster, 2016).
For each focal follow, a focal individual was selected upon encounter (randomly with respect to its sex, age or disease status) if it was alone but sufficiently close (between 2 and 10 m) to other individuals (potential partners) to increase the likelihood of observing a social approach. For both the focal dragon and its partner, right or left facial profile images  were captured using a Canon (Tokyo, Japan) EOS 600 digital camera to allow later identification, and fine-scale behavioural observations pertaining to a social approach were recorded.
We recorded the identity of the individual initiating and the individual receiving the social approach. A social approach was considered successful if the initiator approached the receiver to within 2 m without displacing the receiver, or unsuccessful if the receiver immediately moved more than 2 m away following a social approach (Fig. 1). Following a successful approach, the time that two individuals spent within 2 m of one another was recorded as their 'time in proximity'. Time in proximity was recorded until the social approach ended, or for up to 5 min. If the focal individual was not involved in a social approach or moved out of the observer's sight, the focal follow stopped after 10 min. Successive focal follows were not performed within the same immediate area, to avoid observing the same individuals several times (either as focal individuals or as partners) within the same survey day.

Individual Identification and Disease Status
In this study, the disease status and disease severity of each individual were assessed during its capture, with all individuals having been captured within 60 days of their behavioural observation. As the disease progression is slow and there is currently limited evidence that lesions associated with N. barbatae can heal, this time frame was considered to accurately reflect the disease status of all individuals observed in our focal follows. Visual assessments of disease status were confirmed by qPCR analyses for a subset of individuals (N ¼ 43; see Appendix 1 for further details and Appendix Table A1). Furthermore, as there were no instances where an individual disease status changed during our study period, skin swabs taken days after behavioural data collection were considered to accurately reflect the individual disease status during our behavioural observations. The disease status (clinically diseased or nondiseased) and level of disease severity of a dragon (ranging from 0 signifying 'no lesions' to 5 signifying 'severe lesions'; Fig. 2) was assessed during capture, when evidence for clinical signs of the disease could be identified (i.e. yellowish-brown crusted lesions on the skin (see Figure 1 in Peterson et al., 2020). Investigating disease severity may provide more fine-scaled information than the binary response (i.e. presence/absence), especially if the impacts of the disease were 'dose-dependent' (Ostermann et al., 2013). Additionally, incorporating disease severity allowed us to investigate whether dragons may modify their social behaviour as a result of disease presence itself, or based on the visibility of the lesions.
During focal follows, the observer was not aware of the disease status of most individuals, to avoid any unintentional bias associated with a priori expectations of particular focal individuals. This was facilitated by the fact that lesions were often situated on the ventral side of the animals (except for some moderate and all severe cases), and hence often not visible without capture.
Profile photographs captured during focal follows were later used to identify individuals through an interactive individual identification software (I3S Spot, v4.0.2, https://reijns.com/i3s/), using individuals' unique scale pattern and coloration (their 'fingerprint'). I3S Spot provided an output of the 50 closest matches to an individual's fingerprint, returning the correct identification within the first five matches in 83.3% of searches, and within the first 10 matches in 90.5% of searches ; see Appendix 1 for further details). Age and sex were also recorded, with individuals distinguished as juvenile or adult based on morphological features and sometimes life history (see Appendix 1 for further details), and male or female based on distinct malebiased sexual dimorphisms (Baird et al., 2012). We accounted for age and sex as we know these influence patterns of social behaviour in our studied population (Delm e, 2022).

Effect of disease on social approach occurrence
To examine the effect of the focal individual's disease status and disease severity on its propensity to initiate a social approach, we fitted two generalized linear mixed models (GLMMs) with binomial distributions using the R package lme4 (Bates et al., 2015) in R v. 4.0.4 (R Core Team, 2021). When repeated records of the same individual occurred within the same survey period (i.e. morning or afternoon), only the first record was kept for all analyses. Here, we also excluded any instances where a focal follow began as a result of another individual chasing the focal dragon closer to a conspecific. We fitted disease status as a fixed effect in the first model, and disease severity as a fixed effect in the second model. Both models included age and sex of the focal individual as fixed effects, and the number of days since we began data collection as a scaled linear covariate to account for seasonal effects. We did not include a nonlinear (quadratic) term for seasonal effects in the model as it did not significantly impact the response variable. The identity of the focal individual was included as a random effect to account for repeated records of the same individuals during the study period. Focal follows performed in our study. The outcome of a social approach was considered either successful or unsuccessful. If the receiver of an approach remained within 2 m of the initiator, the approach was characterized as successful. If the receiver immediately moved more than 2 m away following an approach, the approach was considered unsuccessful. Following a successful approach, the time that two individuals spent together was recorded for up to 300 s.
'Date: Time' was included as an additional random effect to account for the potential nonindependence of focal follows within the same survey.

Effect of disease on social approach success
To determine whether the disease status or disease severity of the initiator and the receiver influenced the success of a social approach (Fig. 1), we fitted two more GLMMs with binomial distributions. Here, the response was at the level of the pair of socially interacting individuals (dyad) instead of the focal individual, and we were interested in the directionality of a social approach, essentially 'who approached whom'. As such, the individual that actively sought the social approach was recoded as the 'initiator', while the individual that received the social approach was recoded as the 'receiver'. The identities of the initiator and the receiver were included as random effects along with Date: Time. The disease status of each individual and its interaction were fitted as fixed effects in the first model, while the disease severity of each individual and its interaction were fitted as fixed effects in the second model. Additionally, the age and sex of the initiator and receiver were included as fixed effects in both models, along with linear seasonal effects (as above).

Effect of disease on time spent in proximity following a successful social approach
To examine the effect of the initiating and receiving individual's disease status and disease severity on their time spent in proximity, we ran two mixed-effects Cox proportional hazards regression models (COXMEs) using the R package coxme (Therneau, 2015) in R v. 4.0.4 (R Core Team, 2021). These models are typically used in survival analyses to examine how time to an event (here, the end of a social interaction) is predicted by covariates, with the inclusion of (a) The disease status and level of disease severity used to categorize dragons was assessed during capture, when evidence for clinical signs of the disease could be identified.
(a) An individual disease rating of 0 (no obvious lesions). Clinical signs include a normal appearance, with no skin lesions observed (e.g. only injury observed or unsure of skin condition). (b) An individual disease rating of 1 (mild). Clinical signs include one to three focal skin lesions 5 mm diameter. (c) An individual disease rating of 2 (mild/moderate). Clinical signs include four to five focal skin lesions 5 mm diameter or one lesion 5e10 mm in diameter. (d) An individual disease rating of 3 (moderate). Clinical signs include up to three lesions 5e10 mm diameter or one to two skin lesions 10e20 mm in diameter. (e) An individual disease rating of 4 (moderate/severe). Clinical signs include at least four skin lesions 10e20 mm in diameter with roughly 5e10% of the skin surface affected. (f) An individual disease rating of 5 (severe). Clinical signs include more than 10% of total skin surface affected or 5e10% affected and in poor condition. random effects to account for multilevel nested data (Austin, 2017). In the first model, we fitted the initiator and receiver's disease status as a fixed effect to examine 'time in proximity', the time dyads spent within 2 m of each other following a successful approach. In the second model, we fitted disease severity as a fixed effect. The age and sex of the initiator and receiver were included as fixed effects in both models, along with linear seasonal effects (as above). The individual identities of the initiator and receiver were included as random effects in both models. Any instances where an interaction was interrupted by a chase were excluded from this analysis, along with any instances where humans (e.g. park visitors) interrupted the focal follow. Across all analyses that comprised our study, statistical inference was based on the full fitted models.

Ethical Note
Our study was approved by the animal ethics committee of the University of the Sunshine Coast (ANS1858 and ANA20161). Our behavioural observations did not involve any physical contact with the animals.

RESULTS
A total of 57 surveys were completed across 31 days. We recorded an average of 10.73 behavioural focal follows per survey, and an average of 20.42 behavioural focal follows per day. We documented an average of 5.36 follows per focal individual. The entire data set comprised 123 unique individuals (63 females and 60 males) which were involved in 394 social interactions. Of these unique individuals, 45 were categorized as diseased, while 78 were categorized as nondiseased. The majority of individuals ranged from mild to moderately diseased (1e3), with only two individuals recorded as severely diseased (5). Our data set reflected the known population demography with a 1:1 sex ratio . In our sample, 36.6% of individuals were diseased, which was close to the 34.5% of the population estimated to be infected in 2018 (Peterson et al., 2020).

Social Approach Occurrence
First, we examined the effects of the focal individuals' disease status (0, 1) and disease severity (0e5) on their propensity to initiate a social approach. A total of 633 observations were recorded, with 212 of these involving a social approach that was initiated by the focal individual. Estimates from the model fitting disease status (Appendix 2, Table A2) revealed a small but not significant increase (3.4%) in the probability of initiating a social approach, these probabilities being 35.1% for diseased focal individuals (95% confidence interval, CI ¼ 0.239e0.483) and 31.7% for nondiseased focal individuals (95% CI ¼ 0.218e0.436). Second, estimates from the model fitting disease severity (Table 1) also showed a small positive but not statistically significant effect of disease severity on the probability of initiating a social approach.

Social Approach Success
When we explored the effect of the initiating and receiving individuals' disease status or disease severity on the success of a social approach, the model fitting disease status revealed a small effect of the initiator or receiver's disease status and their combinations on the percentage probability of a successful social approach (Appendix 2, Table A3). This probability was 89.7% for both diseased (95% CI ¼ 0.699e0.970), 83.9% for initiator only diseased (95% CI ¼ 0.613e0.945), 96.0% for receiver only diseased (95% CI ¼ 0.834e0.991) and 87.9% for both nondiseased (95% CI ¼ 0.553e0.977). In contrast, the model fitting disease severity (Table 2) showed that the probability of a successful social approach decreased significantly as the initiator's disease severity increased (Fig. 3). However, we did not find an effect of the interaction between both partners' disease severity.

Time in Proximity
Last, we analysed the effect of the initiating and receiving individuals' disease status and disease severity on the amount of time dyads spent together following a successful social approach. Estimates from the first model fitting disease status (Appendix 2, Table A4) did not reveal any significant differences in time spent together following a successful social approach between any type of dyad. Survival curves representing all four combinations of disease status within dyads (both diseased, initiator only diseased, Variance and SD are given for random effects, estimate and SE for fixed effects. ID: Identity. I: initiator, the individual that actively sought out an interaction by approaching a conspecific within 2 m. R: receiver, the individual that was approached. The identity of the initiator and receiver were both included as random effects. Date: Time represents the survey during which the focal follow was recorded. Date alone represented a continuous linear covariate (e.g. days since September) and was scaled to have a mean of 0 and SD of 1. Severity Numbered: the degree of clinical fungal infection exhibited by the initiator (I) or receiver (R), ranging from 0 or 'no lesions present' to 5 or 'severely diseased'. Age was categorized as juvenile or adult and sex as male or female. *P <0.05; P***<0.001. receiver only diseased, both nondiseased; Fig. 4) revealed a small but not significant decrease in the time that individuals spent in proximity when both individuals were diseased. Instances where both the initiator and receiver were diseased (Fig. 4, in red) had a survival curve that decreased faster and remained lower than curves for all other disease status categories. Here, 50% of successful social approaches that involved two diseased dragons had ended after a median of 72 s, the shortest median time across all combinations of disease status within dyads (Appendix 2, Table A5).
Similarly, a second model fitting disease severity (Table 3) revealed a small but not significant effect of the initiator or receiver's disease severity on the time spent together between any type of dyad. Thus, it is unlikely that nondiseased individuals spent less time with diseased conspecifics, or that the disease severity of either individual influenced the length of a social interaction. Interestingly, both models suggested that dyads spent less time together following a successful social approach if the initiator was male compared to female (Appendix 2, Table A4, Table 3). The first model also suggested that dyads spent less time together following a successful social approach if the receiver was a juvenile compared to an adult (Appendix 2, Table A4).

DISCUSSION
While it has been suggested that individuals that can plastically modify their social behaviour to avoid infected conspecifics are likely to avoid a disease-related death (Bouwman & Hawley, 2010), empirical studies that test for such disease-triggered social responses in wild animals are scarce (but see Stroeymeyt et al., 2018). Moreover, we still do not understand whether and how EIFDs impact the social behaviour of animals in nature. Our study addresses these gaps and worryingly suggests that dragons, plagued by an EIFD, show limited social behavioural disease avoidance mechanisms. In particular, we found that (1) diseased individuals were no less social than their nondiseased conspecifics, (2) nondiseased individuals did not avoid or spend less time with diseased conspecifics, and (3)    caused by N. barbatae, instead of their presence or absence, suggested that individuals avoided more severely diseased conspecifics regardless of their own disease presence. Below, we discuss our findings in the context of potential plastic responses to visible clinical signs of disease, with a particular focus on active selfisolation and social distancing (Stockmaier et al., 2021).

Self-isolation
While self-isolation could help individuals avoid infecting others when sick, self-isolation in the form of direct avoidance of conspecifics has rarely been documented outside of eusocial insects (Stroeymeyt et al., 2018), with some of the most intriguing examples recorded in honey bees (Kralj & Fuchs, 2006;Rueppell et al., 2010), black garden ants, Lasius niger (Stroeymeyt et al., 2018), and Temnothorax ants (Heinze & Walter, 2010). We are aware of only two studies outside eusocial insects that reported active selfisolation from diseased individuals: one anecdotal case of a tuberculosis-infected badger, Meles meles, leaving its group to die alone (Cheeseman & Mallinson, 1981), and one study (Lopes et al., 2016) that found experimentally immune-challenged mice, Mus musculus, socially disconnected from their groups as they reduced their movement. In contrast to these two studies, and consistent with most other studies outside eusocial insects, we found no evidence of active self-isolation of diseased individuals in our studied population, where diseased dragons sought social interactions as often as their nondiseased conspecifics. Similar patterns have also been identified in the highly social banded mongoose, Mungos mungo (Fairbanks et al., 2015): like diseased dragons, tuberculosisinfected individuals were just as social as their healthy conspecifics. These patterns may, however, differ during the nonbreeding season when dragons are less social, and this may influence the overall effect sizes presented here. Nevertheless, the lack of self-isolation of diseased dragons during the breeding season may have significant implications for the transmission of this disease in this population. However, the sparsity of studies that have investigated active self-isolation of diseased animals outside of eusocial insects means that we poorly understand its prevalence across species, along with any proximate or ultimate mechanisms that may contribute to its expression (Stockmaier et al., 2021). Nevertheless, self-isolation could still have direct impacts upon disease dynamics (Stroeymeyt et al., 2018).

Social Distancing
Social distancing, how healthy individuals avoid coming into contact with sick conspecifics to reduce their own disease risk, is much better understood than active self-isolation, with many species to date having been documented to exercise social distancing (i.e. Behringer et al., 2006;Bulmer et al., 2019;Croft et al., 2011;Langwig et al., 2012;Paciência et al., 2019;Stephenson et al., 2018). Although we did not find that nondiseased dragons avoided socially interacting with diseased conspecifics altogether, we did find that they avoided more severely diseased conspecifics. Our findings suggest that it may not be the presence of disease that influences social-distancing behaviours in dragons, but the severity of the clinical signs of infection. Future work based on long-term data and investigating social relationships before and after an individual becomes diseased could provide further insight into our findings. The pathways involved in both disease presence recognition and in 'severity-sensitive' avoidance behaviour are poorly understood generally (Fairbanks et al., 2015). Most research, for instance, has focused on the presence or absence of disease rather than its severity; male house finches preferentially fed near diseased conspecifics (Bouwman & Hawley, 2010), while banded mongooses did not avoid resting on top of or grooming one another, even when conspecifics exhibited clear signs of tuberculosis (Fairbanks et al., 2015). The role of disease severity has, however, been shown to be important in guppies, Poecilia reticulata, where healthy individuals began to avoid infected conspecifics when these were in the later stages of infection (Stephenson et al., 2018).
Most likely, learning to socially avoid diseased conspecifics would require a suite of cues to (1) distinguish sick from healthy individuals (i.e. sickness behaviours, visual, olfactory or chemosensory cues) and (2) trigger a 'disgust-eliciting' response, as individuals learn about the potential fitness costs posed by the pathogen in question (e.g. disease and death; Curtis, 2014;Schaller, 2011;Weinstein et al., 2018). Moreover, this disgust-eliciting response would need to be effective for it to be maintained in the long term. Below, we discuss what may be some of the reasons why dragons have not learned to modify their social behaviour to avoid the risk of disease transmission altogether, despite N. barbatae having emerged in wild populations 8 years ago (Peterson et al., 2020).
Clinical symptoms of N. barbatae can be visually observed, typically beginning as a cutaneous infection which often progresses to fatal systemic disease. Gross presentation of diseased dragons includes severe, proliferative yellow to tan/brown crusted lesions which can cover 10e60% of the skin surface of the body (see Figure 1 in Peterson et al., 2020). Lesions are also often accompanied by ulcerated, bleeding skin and loss of digits (Par e & Sigler, 2016) and, in the late stage of the disease, diseased dragons can be lethargic and emaciated (Peterson et al., 2020). Similar to Stephenson et al. (2018), we found that severely diseased individuals with conspicuous clinical symptoms were less socially tolerated compared to less diseased or nondiseased individuals. While this finding could initially appear to be driven by the limited number of severely diseased individuals in this data set (two females), we found this was not the case. The negative relationship between disease severity and social tolerance remained even when these two severely diseased females were removed (Appendix 2, Table A6). Variance and SD are given for random effects, estimate and SE for fixed effects. I: initiator, the individual that actively sought out an interaction by approaching a conspecific within 2 m. R: receiver, the individual that was approached. Severity Numbered: the degree of clinical fungal infection exhibited by the initiator (I) or receiver (R), ranging from 0 or 'no lesions present' to 5 or 'severely diseased'. Date: Time represents the survey during which the focal follow was recorded. Date alone represented a continuous linear covariate (e.g. days since September) and was scaled to have a mean of 0 and SD of 1. Age was categorized as juvenile or adult and sex as male or female. *P <0.05.
Given diseased individuals vary in the transmission risk they pose, it would make sense for natural selection to favour risk-or 'severity-sensitive' avoidance behaviour that could optimally balance the costs and benefits of sociality. However, a limited number of studies have investigated how animals detect fungal disease (Kiesecker et al., 1999), let alone how they may detect the severity of a fungal infection, and we are yet to understand how dragons may detect N. barbatae.
Interestingly, while we found that disease severity decreased the probability of a successful approach, it did not influence the time associating dyads spent together. Similar patterns were also reported in the Tasmanian devil, Sarcophilus harrisii: Cameron et al. (2020) documented that while nondiseased Tasmanian devils were generally found to avoid conspecifics displaying clinical signs of devil facial tumour disease, long-term associating dyads continued to persist even if one of them became symptomatic. Together these patterns underline that the potential cost of avoiding or spending less time with long-term associates may not outweigh the cost associated with increased risk of disease transmission. This is because engaging in avoidance behaviour can result in lost social benefits (Stephenson et al., 2018). Thus, a costebenefit trade-off may mean that the time dyads spend together may not be impacted by the disease severity of one or both individuals. In addition, more research is needed to understand whether and how such fine-scale changes in social behaviour would influence population level social networks and thus disease transmission overall.
While several studies have demonstrated that a range of species use chemosensory detection methods in their avoidance of diseased conspecifics (Arakawa et al., 2010;Behringer et al., 2006;Kavaliers et al., 2004), and others have suggested visual detection methods (Bouwman & Hawley, 2010), none of these have included reptiles. As such, we argue that despite the clear visual cues available to dragons, it may be the long period of clinical infection (dragons have displayed clinical symptoms for up to 7 years postdiagnosis) accompanied by a lack of any mass mortality events that may be limiting dragons' ability to associate the visual clinical signs of N. barbatae with a pathogen fitness cost (e.g. disease and death), and as such the need to trigger a 'disgust-eliciting' response (Curtis, 2011;Schaller, 2011;Weinstein et al., 2018). In contrast, in species that have adopted social distancing to avoid diseased conspecifics, the onset of clinical symptoms of disease or sickness behaviours are often shortly followed by death (Kurze et al., 2020). For instance, in captive little brown bats, Myotis lucifugus, death occurred in as little as 3 months after experimental infection with P. destructans (Warnecke et al., 2012). It has been suggested that little brown bats change their clustering behaviour in the presence of P. destructans to limit disease transmission (Langwig et al., 2012). Additionally, black garden ants, a eusocial species known for their advanced behavioural responses to fungal diseases, can take just 24 h to die once fungal spores have penetrated the body (Stroeymeyt et al., 2018). In these instances, visual evidence of clinical infection has also been documented alongside mass mortality events (Meteyer et al., 2009), which may represent a key opportunity for animals to associate clinical visual signs of disease or sickness behaviour with widespread death.
Finally, the biology of fungal pathogens may also contribute to a limited capacity for social avoidance behaviours. Unlike many other viral and bacterial pathogens, fungal pathogens persist saprophytically in the environment (Allender et al., 2015), which makes them difficult to detect and facilitates transmission between substrate and host, as well as between hosts (Rowley & Alford, 2007). EIFDs therefore pose real challenges to wildlife, as this additional transmission pathway may mean that socially distancing would not be a sufficient mechanism to mitigate the risk of pathogen exposure or disease transmission (Fairbanks et al., 2015;Loehle, 1995). However, many parasites are similarly difficult to detect, and avoidance by their hosts often relies on indirect cues that are driven by long-term associations (Weinstein et al., 2018). Many species avoid faeces and carcasses, regardless of infection status (Curtis, 2014). Moreover, infection can alter the chemical composition of sweat, breath and faeces, enabling more targeted avoidance of infected individuals (Shirasu & Touhara, 2011). Thus, parasite avoidance is akin to disgust, an innate aversion to cues associated with parasites (Curtis, 2014). Such disgust may drive the formation of a three-dimensional landscape (Weinstein et al., 2018), with 'mountains' of infection risk and 'valleys' of safety. Some might argue that such 'landscapes of disgust' may end up evolving and trigger the development of social avoidance behaviours.

Conclusion
Human activities have been causing an increased emergence of viral, bacterial, parasitic and fungal pathogens (Sehgal, 2010;Soul e, 1986), along with major changes to the geographical range and incidence of vector-borne infectious diseases (Daszak et al., 2000). Our study contributes to important research on disease-triggered social behavioural modifications. We corroborated evidence that suggests a lack of active self-isolation behaviours in animals outside eusocial insects and, consistent with findings across a diverse range of taxa, observed limited social distancing in the face of emerging infectious disease. It is crucial that future research aims to understand whether animal populations can mitigate the costs of EIFDs, as the proportion of EIFD records has increased seven-fold in the last 15 years (Fisher et al., 2012) and is predicted to increase in the future (Mitchell et al., 2008).

Data Availability
The data and the associated R scripts can be found at https://doi. org/10.5061/dryad.905qfttpv.

Declaration of Interest
The authors have no conflicts of interests.
work was funded under Associate Professor Celine Frere ARC Future Fellowship FT200100192.
Diagnostic screening for presence of N. barbatae-specific DNA using qPCR In this study, disease status was attributed to an individual animal visually, based on the presence of characteristic skin lesions associated with N. barbatae infection at the time of capture. The presence of infection was confirmed in a previous study validating Nannizziopsis disease diagnosis using a combination of histology and molecular-based typing (Peterson et al., 2020;Powell et al., 2023). Observers were trained to recognize such characteristic skin lesions in the field; however, a sampling regime was also initiated to swab infected skin lesions for testing using a newly developed Nannizziopsis-specific qPCR assay. A total of 43 animals included in this study were swabbed and tested for the presence of Nannizziopsis fungi to confirm the visual diagnostic-based disease status assignment. We found that 97% (33/34) of individuals considered diseased during this study through visual assessment were qPCR positive while all the individuals considered nondiseased (N ¼ 9) were qPCR negative. The results of this screening are presented in Table A1.

Individual identification and disease status
Dragons have unique tympanic scale patterns which were used as a 'fingerprint' to identify them using the method detailed in Gardiner et al. (2014). As part of the long-term study of individuals' behaviours in this population, each individual had been assigned a unique name against this fingerprint, and this information was stored in an established picture library of individuals along with its sex and age (juvenile or adult). As in Gardiner et al. (2014), we compared facial profile images captured during focal follows to this library to identify individuals.

Distinguishing adults from juveniles
The following morphological features were used to distinguish between adults and juveniles. (1) snout to vent length (adult female: SVL > 171; adult male: SVL > 205); (2) length and shape of spikes from the head crest; (3) coloration pattern between the eye and ear; (4) size of the eye; and (5) occurrence of scars and missing spikes from the head crest. In contrast to juveniles, both male and female adults have long and sharp spikes on their head crest, and the coloration pattern (yellow/black) between the eye and ear is more contrasted (C.H. Fr ere & C. Delm e, personal observation). Furthermore, adult males have noticeably larger jaws than juveniles, and the size of the eye seems more prominent in juveniles (C.H. Fr ere & C. Delm e, personal observation). Adults also tend to exhibit more scars and missing spikes from the crest.
In addition to differences in morphology, an individual seen in the population for more than 3 years was reliably defined as an adult, as this represented the average number of years that both adult males and females had been seen in the population. Life history data were particularly helpful, for instance, in ageing small adult females. Disease status is at time of sampling for testing. One sample returned a negative result although it was clearly diseased, highlighting a false-negative bias attributed to sampling method.  Variance and SD are given for random effects, estimate and SE for fixed effects. ID: Identity. DS: Disease status. I: initiator, the individual that actively sought out an interaction by approaching a conspecific within 2 m. R: receiver, the individual that was approached. The identity of the initiator and receiver were both included as random effects. Date: Time represents the survey during which the focal follow was recorded. Date alone represented a continuous linear covariate (e.g. days since September) and was scaled to have a mean of 0 and SD of 1. DS I and DS R indicated whether the individuals in a focal follow exhibited clinical signs of fungal infection (Yes or No). Age was categorized as juvenile or adult and sex as male or female. *P <0.05; **P <0.01; P***<0.001. Variance and SD are given for random effects, estimate and SE for fixed effects. DS: Disease Status. I: initiator, the individual that actively sought out an interaction by approaching a conspecific within 2 m. R: receiver, the individual that was approached. DS I and DS R indicated whether the dragons in an observation exhibited clinical signs of fungal infection (Yes or No). For instance, DS I No R Yes represents successful social approaches where the initiator was not diseased and the receiver was diseased. For this analysis, the effects of the coefficients estimated for disease status were contrasted to pairs of diseased individuals. Changing the baseline group did not reveal further significant differences between groups. Date: Time represents the survey during which the focal follow was recorded. Date alone represented a continuous linear covariate (e.g. days since September) and was scaled to have a mean of 0 and SD of 1. Age was categorized as juvenile or adult and sex as male or female. *P <0.05.  Variance and SD are given for random effects, estimate and SE for fixed effects. ID: Identity. I: initiator, the individual that actively sought out an interaction by approaching a conspecific within 2 m. R: receiver, the individual that was approached. The identity of the initiator and receiver were both included as random effects. Date: Time represents the survey during which the focal follow was recorded. Date alone represented a continuous linear covariate (e.g. days since September) and was scaled to have a mean of 0 and SD of 1. Severity Numbered: the degree of clinical fungal infection exhibited by the initiator (I) or receiver (R), ranging from 0 or 'no lesions present' to 5 or 'severely diseased'. More severely diseased individuals (i.e. categorized as '4' or '5') were excluded. Age was categorized as juvenile or adult and sex as male or female. *P <0.05; **P <0.01; P***<0.001.