Effects of historic and projected climate change on the range and impacts of an emerging wildlife disease

The global trend of increasing environmental temperatures is often predicted to result in more severe disease epidemics. However, unambiguous evidence that temperature is a driver of epidemics is largely lacking, because it is demanding to demonstrate its role among the complex interactions between hosts, pathogens, and their shared environment. Here, we apply a three‐pronged approach to understand the effects of temperature on ranavirus epidemics in UK common frogs, combining in vitro, in vivo, and field studies. Each approach suggests that higher temperatures drive increasing severity of epidemics. In wild populations, ranavirosis incidents were more frequent and more severe at higher temperatures, and their frequency increased through a period of historic warming in the 1990s. Laboratory experiments using cell culture and whole animal models showed that higher temperature increased ranavirus propagation, disease incidence, and mortality rate. These results, combined with climate projections, predict severe ranavirosis outbreaks will occur over wider areas and an extended season, possibly affecting larval recruitment. Since ranaviruses affect a variety of ectothermic hosts (amphibians, reptiles, and fish), wider ecological damage could occur. Our three complementary lines of evidence present a clear case for direct environmental modulation of these epidemics and suggest management options to protect species from disease.

There is strong evidence supporting the rapid international spread of a global panzootic lineage of B. dendrobatidis during the 20th century (Farrer et al., 2011;O'Hanlon et al., 2018), but the proposed conflict between the two hypotheses now appears to have been reconciled in a framework that incorporates both as key drivers of emergence and outcomes, explaining observations of decline in regions where the impacts have been greatest (Cohen, Civitello, Venesky, McMahon, & Rohr, 2018;Cohen et al., 2017;Raffel et al., 2013;Rohr et al., 2008). Thus, it seems likely that the previous mindset of treating environmental change and pathogen range expansion as conflicting has hampered understanding of the patterns of emergence and the focusing of mitigation efforts.
Establishing a role for climate in disease emergence can be very challenging. Increasing environmental temperature is a key component of climate change, which is cited as a driver of infectious disease emergence and severity, but evidence for this is scarce and it is often difficult to discriminate between the effect of temperature and other aspects of climate (Harvell et al., 2002). The direct and indirect influences of temperature on host-pathogen interactions (Altizer et al., 2006;Clare et al., 2016;Garner, Rowcliffe, & Fisher, 2011) and its nonlinear effects on incidence and severity (Bosch, Carrascal, Durán, Walker, & Fisher, 2007;Raffel et al., 2013;Walker et al., 2010) represent considerable challenges to a better understanding of disease emergence (Rohr et al., 2011). Most research effort in this area has focused on human diseases (Aguirre & Tabor, 2008), and vector-borne diseases (e.g., malaria, dengue, chikungunya) in particular (Harvell et al., 2002;McMichael, Woodruff, & Hales, 2006).
In the current study, we investigate the effect of temperature on the interaction between ranaviruses and their amphibian hosts, a host-pathogen system that offers the possibility of direct experimental manipulation, and a well characterized recent history of pathogen invasion into the United Kingdom (Price, Garner, Cunningham, Langton, & Nichols, 2016). Ranaviruses are large double-stranded DNA viruses (family Iridoviridae) that can be highly pathogenic to ectothermic vertebrates (Gray, Miller, & Hoverman, 2009;Price et al., 2014;Rosa et al., 2017). Ranavirus infections of amphibians are notifiable to the World Organization for Animal Health due to their potential to cause severe disease outbreaks as well as the risks of international spread through trade (Schloegel, Daszak, Cunningham, Speare, & Hill, 2010;Schloegel et al., 2009). Ranavirus growth and virulence can be affected by temperature (Ariel et al., 2009;Bayley, Hill, & Feist, 2013;Brand et al., 2016;Rojas, Richards, Jancovich, & Davidson, 2005) and environmental temperature is considered to be one possible explanation for observations of seasonality in outbreaks (Brunner, Storfer, Gray, & Hoverman, 2015). Indeed, incidents of ranavirosis in frogs in the USA were recently shown to be uncoupled from a pulse in transmission or the density of susceptible hosts, and instead were coincident with temperature increases and developmental changes in frog larvae (Hall, Goldberg, Brunner, & Crespi, 2018).
Ranaviruses are distributed globally but outbreaks of disease are extremely patchy-a pattern which is not yet understood. Some disease outbreaks have been shown to result from human translocations of ranavirus Picco & Collins, 2008;Price et al., 2016), while other studies have found infections to be widespread at national scales without evidence for disease, which may reflect an historic association (Warne, LaBumbard, LaGrange, Vredenburg, & Catenazzi, 2016;Whitfield et al., 2013). The seasonal patterns and the observations of a temperature effect in laboratory studies raise the possibility that environmental conditions could drive invasion success and routes in cases where ranaviruses are undergoing range expansion as well as climate change being a driver of disease emergence in regions where the associations between viruses and hosts are historic and widespread.
In the United Kingdom, recurrent amphibian mass-mortality incidents caused by ranavirus have resulted in severe population declines of the common frog (Rana temporaria; Teacher, Cunningham, & Garner, 2010). Genetic evidence supports multiple pathogen introductions into the United Kingdom while spatiotemporal models suggest that ranavirus spread rapidly, facilitated by translocations of unspecified infectious materials by people (Hyatt et al., 2000;Price et al., 2016). Disease outbreaks are strongly seasonal, peaking in the summer months and appearing to mostly affect adult animals, contrasting with other regions where larvae or metamorphic animals are the worst-affected age-classes (Brunner et al., 2015). However, the detectability of the main UK host, the common frog, is also strongly seasonal and there has been no previous attempts to explicitly control for host population density, host activity, or observer effort in examining the periodicity of outbreaks (Cunningham, 2001;Cunningham et al., 1996;Teacher et al., 2010).
In this study, we investigated the role of temperature as a driver of disease outbreaks in UK common frogs infected with ranaviruses in the frog virus 3 (FV3) lineage through a combination of epidemiological modeling of a long-term study of disease in wild populations (Cunningham, 2001;Price et al., 2016), in vitro experiments involving manipulation of the host-pathogen environment, and similar in vivo experiments using natural hosts. Our aims were to examine the role of temperature in shaping host-pathogen interactions to address whether it (a) has been a factor explaining the pattern of invasion which can be used to predict future changes in epidemiology under projections of climate change, and (b) can explain the observed seasonal patterns in disease occurrence and the contrast in affected host age-class compared to other temperate regions experiencing amphibian mortality incidents due to the same type of ranavirus.

| MATERIAL S AND ME THODS
2.1 | Temperature as a predictor of frog mortality and incident severity 2.1.1 | Temperature dependence of ranavirus incidence We used data from the Frog Mortality Project (FMP), a flagship citizen science project which collected information on amphibian mortality incidents for over 25 years in order to study disease occurrence in wild populations (Price et al., 2016). The project provided a structured reporting system and guidance to volunteer reporters on how to report amphibian mortality incidents, including a range of variables describing the environment and its management. The FMP dataset has been reliably filtered for incidents of ranavirosis previously (North, Hodgson, Price, & Griffiths, 2015;Price et al., 2016;Teacher et al., 2010) based on detailed postmortem examinations of dead and sick amphibians reported from multiple sites over multiple years (Cunningham, 2001;Cunningham et al., 1996;. In this study, the same criteria as Cunningham (2001) and Price et al. (2016) (the presence of indicative signs of disease ["ulceration," "red spots on the body," and "limb necrosis/loss of digits"; see Appendix S1 and Figure S1] and a minimum of five dead animals) were used to create a binary variable describing the ranavirus status of each incident. To date, no diagnosis other than ranavirosis has been made from pathological examinations of common frogs from such mortality incidents in the United Kingodm (Cunningham, 2001;Cunningham et al., 1996;Price et al., 2017). The detectability of amphibians varies seasonally, while reporting to the FMP was subject to possible temporal and spatial variation in effort (Price et al., 2016).
We controlled for these potential biases indirectly through the inclusion of mortality incidents caused by factors other than ranavirosis as previously (Price et al., 2016). Data on the timing of the onset of mortality (available at the resolution of month only) and the incident location were used to download the monthly average of the daily maximum temperature (T MAX30 ) for the month of onset of disease for each incident from the Met Office (2017b) (further details in Appendix S2).
Factors affecting ranavirus incidence were investigated using a standard logistic regression model incorporating variables describing the environment (temperature), pond (volume, shading, and the presence of marginal and floating vegetation [e.g., see Raffel, Michel, Sites, & Rohr, 2010]), other aquatic vertebrate species present in addition to common frogs (toads, newts, fish), and geographic region (government office region) as predictors of ranavirus status (Model 1) fitted with the R function glm2 (Marschner, 2011;. To explore the relationship between temperature and the probability of a ranavirus-positive observation further, it was modeled as a sigmoid (logistic) transition between an upper and lower mean frequency (Model 2). The four parameters of the curve (upper and lower limits, the location, and the slope of the transition) were fitted using the mle2 function in the R package, bbmle (Bolker & R Core Team, 2017;. Starting values for the mle2 algorithm were obtained for slope and location (intercept), using the glm2 function with a bespoke link function for binomial data (Data S2). A confidence interval around the fitted line was generated using the delta method to compute variance of a function with the emdbook package (Bolker, 2008(Bolker, , 2016. Model 2 was compared using Akaike's information criterion (AIC) to a standard logistic curve (a simplified version of Model 1 with only temperature terms retained as predictors).

| Effect of increased temperature on the severity of disease incidents
In order to analyze the effect of temperature on the severity of mortality incidents, the number of dead frogs and the number of surviving frogs (estimated by reporters) at each reported incident were combined as estimates of the proportion of the population that died (each pond/frog population was included in the analysis only once) and used as the response term in a generalized linear model using the quasibinomial family to account for overdispersion in the data (Crawley, 2013). Temperature (T MAX30 ), the predicted ranavirus status, their interaction, and additional variables which described the pond environment (the presence of other species and physical characteristics of ponds) were used as predictors of severity. Since the total number of dead animals at each incident was used to generate the proportion of the population that died when filtering the dataset for ranavirosis-consistent incidents, the criterion requiring mortality incidents to include five or more dead frogs was dropped but the signs of disease data were used more stringently instead (observations of two of the three indicative signs of disease used above were now required).

| Temperature preceding ranavirus outbreaks with precise timestamps
We previously screened a UK amphibian and reptile tissue archive for ranavirus and returned a database of mortality incidents for which the presence of ranavirus was established using molecular diagnostic methods . After filtering for incidents with a precise georeference (postcodes or grid references) and timestamp (date found, submitted, or examined if a postmortem examination was conducted on receipt, which were each assumed to approximate closely to the day of death due to the rapid decomposition of amphibian carcasses), this dataset contained a total of 197 incidents, for which ranavirus had been detected from 31. All incidents were overlaid on the UK grid of 5 × 5 km squares and the maximum daily temperatures for the date matching the timestamp and the 50 days prior to the death(s) were extracted from plain text data files in the Met Office UK Climate Projections 2009 dataset (Met Office, 2017a), which were downloaded using the R package Rcurl (Price, 2018;Temple Lang & CRAN Team, 2018).
Ranavirus status was used as the response variable in a series of logistic regression analyses (generalized linear models using the binomial family and the logit link function) with the average maximum daily temperature in the week preceding the mortality incident (T MAX7 ) or the number of consecutive days in the previous seven where temperature exceeded 16°C (DAYS T > 16 ) as predictors. These models were also run with region (Government Office Region) or latitude as an additional predictor to further control for any effect of spatial variation in temperature. As above, seasonal variation in the detectability of amphibians was controlled for by inclusion of mortality incidents caused by factors other than ranavirosis.

| Effects of historic warming and seasonality
To check whether prior warming (over the time course of the dataset; 1991-2010) had altered the rate of ranavirosis incidents and to assess seasonality in the data, we first decomposed the annual signal and trend across years in the time series of temperatures and rates of ranavirosis incidents. The mean of the average daily maximum temperature during the month of onset of mortality incidents from all reports in the FMP dataset (1991-2010) was calculated for each month with reports, as well as the numbers of reports that were consistent or otherwise with ranavirosis. Generalized additive mixed models (GAMM; R function gamm; package mgcv [Wood, 2003[Wood, , 2004[Wood, , 2017 were used to fit smooth splines to both the within-year (seasonal; cyclic cubic regression spline) and across-year (cubic regression spline) patterns, and autocorrelation structures (of order one) were used to model residual correlation within years. The number of ranavirosis incidents as a proportion of total reports was then modeled as a function of the seasonal trend (smoothed with a cubic regression spline as above), the trend in temperature across years (predicted using the GAMM above and smoothed with a cubic regression spline), and time using a generalized additive model (GAM) and the binomial family with logit link function (function gam; package mgcv).  2002). Model predictions and residuals were extracted and visualized using functions in the R packages, visreg and ggplot2 (Breheny & Burchett, 2013;Wickham, 2016). The predictive power of the model incorporating the across-year temperature trend was then assessed by comparison to a model containing the temporal trend in the rate of ranavirosis incidents by dividing the dataset into two training and test sets (taking half and three-quarters of the data for training, respectively). GAMs incorporating the smoothed seasonal trend and either a smoothed (across-year) time trend or the smoothed temperature trend were fitted using the training datasets and compared in terms of their ability to predict patterns in the test datasets.

| Virus growth in vitro
To investigate the effect of temperature on viral growth, two UK isolates of FV3 (RUK11, isolated from the kidney of a diseased common frog that died with systemic hemorrhages in Middlesex in 1992, and RUK13, isolated from the skin of a diseased common frog that died with skin ulceration in Suffolk in 1995 [Cunningham, 2001]) were grown at a range of temperatures in two cell lines (epithelioma papulosum cyprini [EPC, derived from the fathead minnow fish, Pimephales promelas (Winton et al., 2010); ECACC 93120820] and the iguana heart reptile line [IgH2; ECACC 90030804]). Cells were grown on 96-well plates until more than 90% confluent and then inoculated with virus in a 10-fold dilution series ranging from an estimated multiplicity of infection of approximately 2 × 10 −6 to 2 × 10 3 (five wells per dilution with an additional one well per dilution receiving a sham exposure of cell culture media only as a negative control). Titers of viral isolate stocks were equalized by reference to qPCR scores [following Leung et al. (2017)]. Plates were incubated at six temperatures (10, 14, 18, 22, 26, and 30°C) and monitored daily for cytopathic effect (plaques in the cell layer). After 6 days, plates were scored for viral growth by counting the number of replicates at each dilution where cytopathic effect was evident and calculating the tissue culture 50% infective dose (TCID 50 ) using the method of Reed and Muench (1938). The effects of temperature and cell line on viral growth (the mean titers of the two isolates) were analyzed using a linear model in R.

| In vivo assessment of effect of temperature on virulence
To investigate whether an effect of temperature on in vitro viral growth was reflected in altered virulence in vivo, 60 overwintered common frog metamorphs (R. temporaria) were randomly allocated to one of six treatments (10 animals per treatment): three exposure treatments (sham, RUK11, and RUK13) crossed with two temperatures (20°C ["low"] and 27°C ["high"]). Temperature was maintained by placing five individually housed frogs selected at random from each of the three exposure treatments into one of four climate-controlled chambers constructed from polystyrene boxes-two held at 20°C and two held at 27°C ( Figure S2; see Appendix S3 for a comprehensive description of the set up). RUK13 was used at a titer of 1.58 × 10 5 TCID 50 /ml and RUK11 at a titer of 1.58 × 10 7 TCID 50 /ml (see Appendix S3 for details of preparation of inocula and exposure methods).
Individuals reaching endpoints (either gross signs of hemorrhaging or ulceration; Figure S1) and all surviving individuals at the end of the experiment (day eight postexposure) were euthanized following a schedule 1 method for amphibians. The number of hours postexposure that animals were found dead or at endpoints was recorded for survival analysis. Survival data were analyzed by fitting a mixed effects Cox proportional hazards regression model in R using the coxme package (Therneau, 2018) with exposure and temperature as fixed effects and the climate-controlled chamber as a random effect to account for pseudoreplication due to placement of individual experimental units inside the four climate-controlled chambers.
A second in vivo experiment was performed in the same host species, using one ranavirus isolate (RUK13) and a single temperature (20°C), but two exposure doses ("low" and "high"; detailed methods in Appendix S1). The effects of temperature and dose on disease progression and survival were compared.

| Effect of projected climate change on timing of ranavirus outbreaks
Baseline temperature data were generated by calculating the mean daily maximum temperature for each calendar month for the period

| Effect of temperature on disease occurrence and severity in the wild
The finalized FMP dataset used in this study contained 4,385 unique records, of which 1,497 were classed as ranavirosis consistent. The logistic model (Model 1) revealed a highly significant, nonlinear effect of temperature on the proportion of ranavirosis-consistent incidents observed (Table S1). The minimal adequate model also retained newts, fish, and shading: the presence of either type of animal in ponds increased the proportion of ranavirosis-consistent incidents observed while shading reduced this proportion (Table S1).
To explore the relationship with temperature further, a model with a transition between an upper and lower frequency was fitted (Model 2), which improved the fit to the data compared to a simplified version of Model 1 comprising only the terms describing the nonlinear relationship with temperature (AIC scores were 5565 and 5576, respectively). Model 2 shows a step change: below approximately 16°C, 25.1% of incidents were ranavirosis-consistent, rising to 38.5% after the temperature threshold was crossed (Figure 1a). The difference between incidents that were ranavirosis-consistent and the remainder ("non-ranavirus") is also apparent in the distribution of temperature records: the non-ranavirus category being strongly bimodal with peaks at both low and high temperatures, whereas most of the ranavirosis-consistent incidents were reported at higher temperatures with few outliers at much lower temperatures ( Figure S3).
Temperature was again a highly significant predictor of ranavirus status when the records with precise timestamps and confirmed ranavirus-positive status were analyzed. This more precise information about timing enabled a fine-scale examination of the effect of temperature in the days preceding incidents. The average temperature in the 7 days preceding incidents (T MAX7 ) was a significant predictor of ranavirus status (odds ratio = 1.20, 95% confidence interval = 1.09-1.31; Figure 1b). The temperature threshold where the proportion of ranavirosis incidents increased sharply in the analysis of the full FMP dataset (Model 2) was approximately 16°C. A second model-with the number of consecutive days where the daily maximum temperature in the week preceding incidents exceeded 16°C (DAYS T>16 ) as a predictor-also indicated that warmer temperatures were a good predictor of ranavirus status with each additional warm day raising the odds that incidents were caused by ranavirus (odds ratio = 1.33, 95% confidence interval = 1.16-1.51; Figure 1c). The 16°C threshold model had a slightly lower AIC score than the model using the average temperature as a predictor (155 compared to 158).
The FMP database contains data on the severity of outbreaks (the estimated proportion of the frog population that died) for the years 1991-2000. After removing records with missing values, we produced a dataset for investigating severity that contained 2,667 records, of which 427 incidents were classified as ranavirosis-consistent. The effects of temperature and covariates previously identified as having an influence on the occurrence or severity of ranavirosis in UK common frogs (North et al., 2015) were explored using a logistic model. The ranavirus status and temperature (T MAX30 at month of mortality onset) plus their interaction, as well as the log-transformed pond volume, shading around ponds, the amount of both the marginal and floating vegetation ("none/little" or "lots"), the presence of toads, the presence of newts, the presence of fish, and the region were all included as predictors in the model. After model simplification, the minimal adequate model retained temperature, ranavirus status, and their interaction but there were also significant effects of the presence of toads, the presence of fish, shading, pond volume (nonlinear), marginal vegetation, and region (Table S2). Temperature had a significant, positive effect on incident severity, each 1°C increase in temperature leading to a 3.1% increase in the proportion of the population that died (p = 7.56 × 10 −12 ). There was a significant interaction with ranavirus status (p = 7 × 10 −4 ); at low temperatures, the severity of ranavirosis-consistent incidents was lower than other types of incident but at higher temperatures ranavirosis-consistent incidents were more severe (Figure 1d; Figure S4). Toads reduced the severity of mortality incidents while the presence of fish increased severity ( Figure S5) as found previously (North et al., 2015).

Shading and marginal vegetation decreased the severity of incidents.
Notwithstanding, and irrespective of which covariate was considered, the effect of increasing temperature increased the severity of disease. This is perhaps best illustrated by the effects of pond shading, where increasing the amount of shading (and, presumably, decreasing the maximum temperatures that frogs would have been exposed to) was associated with reduced severity of ranavirosis and a reduction in the disparity in the severity of incidents between ranavirosis-consistent and non-ranavirus incidents (Figure 1e).  Table S5). Model validation indicated conformity to assumptions. The temperature model, when trained on subsets of the data, predicted trends in test datasets effectively in contrast to models incorporating only the across-year trend in ranavirosis rates, which showed no predictive power ( Figure S6). Temperature showed a strong seasonal pattern which was reflected in the pattern of ranavirosis incidents where seasonality was also marked (GAM: edf = 7.65, p = 2.45 × 10 −13 ; Figure 2c). Overall, after accounting for the effect of temperature, there was a small but significant decrease in the proportion of ranavirosis incidents between 1991 F I G U R E 1 Warm temperatures increased the frequency and severity of incidents of ranavirosis involving wild common frog populations of the United Kingdom. (a) The effect of temperature on ranavirosis-incident rate (the proportion of citizen science reports of frog mortality that were classified as ranavirosis consistent). The line represents the fitted maximum likelihood model of a logistic transition between a lower and upper frequency. The shaded area around the line represents the 95% confidence interval, calculated using the delta method. Points represent the observed data, grouped in 30 windows each containing 146-148 individual records. (b, c) Temperature in the week preceding frog mortality incidents screened using molecular methods predicted ranavirus status ("Positive" or "Negative"). (b) T MAX7 by ranavirus status. (c) DAYS T>16 by ranavirus status. (d) The severity of frog mortality incidents (estimated proportion of population that died) was consistently greater at higher temperatures, particularly in the case of ranavirosis-consistent incidents (orange). The plot shows fitted lines (and 95% confidence intervals) from a generalized linear model (quasibinomial regression) of severity as a function of ranavirus status and temperature (T MAX30 for the month of onset of mortality incidents, including quadratic terms). Points are a summary of raw data after binning individual data points (mode of 20 data points per bin) and calculating the mean temperature and overall proportion of dead frogs within each bin. (e) Large amounts of shading around ponds reduced the severity of ranavirosis-consistent mortality incidents (orange) compared to other incidents (green). In panels (b)

| In vitro assessment of viral growth rates and in vivo tests of virulence
We examined in vitro viral growth using two UK isolates of FV3 The effect of temperature on the response of common frogs to viral exposure was assessed in order to validate results from cell culture models in vivo. Temperature was a highly significant predictor of survival; 20 of 60 animals died or were euthanized on reaching humane endpoints, of which 14 were from high-temperature treatments and 6 were from low-temperature treatments (Figure 4). Overall, there was a 5.33 times higher risk of death in the high-temperature treatments (p = 0.005; Table 1). Titers of viral inoculates were not equalized between isolates. All individuals exposed to RUK11 (at a high dose) and maintained at high temperature died or reached endpoint by the eighth day postexposure compared to 6 of 10 individuals maintained at low temperature (Figure 4). Of the animals exposed to RUK13 (at a relatively low dose compared to RUK11), three individuals died or reached endpoint in the high-temperature treatment compared with none at the low temperature. There was also a significant effect of exposure treatment: the expected hazard of animals exposed to RUK11 was 41.6 times higher than animals receiving a sham exposure (p = 0.0004). These results are largely in line with a study examining survival of common frog tadpoles exposed to a North American isolate of FV3, which showed that mortality was increased at 20°C compared to 15°C (Bayley et al., 2013).
The second in vivo experiment examining the effect of dose on disease outcome and progression in juvenile common frogs (Appendix S1) complements the findings of the in vivo temperature experiment. The dose experiment suggests that a viral load threshold exists which must be crossed before gross signs of disease develop. We found the outcome and presentation of disease, as well as the viral quantity in tissues at death to be largely independent of dose; all animals exposed to either low or high viral doses died, presented with the same set of signs ( Figures S1 and S7), and had similar quantities of virus in their tissues at death ( Figure S8a). However, the onset and progression of disease were delayed at low dose (the development of disease and death both occurred significantly later; Figure S7) reflecting the lower viral loads of individuals (measured by swabbing animals after infection; Figure S8b) in this treatment. Also, viral loads of dead individuals were greater than those of live individuals (both when repeated-measures from the same individuals were compared [ Figure S8c] and when individuals that were euthanized part-way through the experiment were compared to those that died [ Figure S8d]). These results all suggest that elevated viral loads lead to the onset of disease and that the viral capacity to cross a threshold concentration is a more important determinant of whether disease develops than the initial dose. It seems likely that higher temperature and higher initial dose each serve as ways to reach this putative threshold for disease sooner-either through more rapid viral growth or a greater initial intensity of infection, respectively-and explain the delays and/or reductions in observations of severe outcomes in the other respective treatments (low temperature or low dose).

| Impact of future climate on timing of outbreaks
The UK climate is expected to warm considerably over the remainder of the century (Chen & Tung, 2018;Jenkins et al., 2009) investigated. These are months that we expect to have experienced limited incidence and severity of ranavirosis previously, but temperatures are projected to change to such a degree that this limitation will be removed for large areas of the Untied Kindgom ( Figure 5). Survival TA B L E 1 Effect of temperature and ranavirus exposure treatments on the survival of common frogs. Coefficients and hazard ratios from a mixed-effects Cox proportional hazards model of survival in response to exposure ("Sham" as the reference level) and temperature ("Low" as the reference level) treatments  . The association between temperature and the rate of ranavirosis incidents across years suggests that temperature is driving the rate of disease incidents in the long term and may also drive the seasonality, since both a correlated seasonal factor and a correlated nonseasonal factor acting across years would otherwise be required to explain the observed patterns. Furthermore, the use of in vitro and in vivo studies in combination with modeling of field datasets serves as a "triangulation" process (Plowright, Sokolow, Gorman, Daszak, & Foley, 2008) and strongly suggests a causal link between temperature and disease occurrence in this system.

| D ISCUSS I ON
Our results are consistent with an historic effect of climate on the rate and timing of ranavirosis incidents and suggest that the invasiveness of this introduced pathogen may have been influenced by the suitability of local climate. We have previously shown that ranavirus was introduced to the United Kingdom and spread rapidly in England (Price et al., 2016). The small overall decrease in the rate of disease incidents observed over the 20 years of data analyzed, which was observed after controlling for the effect of temperature Although it is challenging to detect disease and mortality in larval amphibians, all evidence points to adult common frogs as the major life history stage and species affected by ranavirosis in the wild in the United Kingdom (Cunningham, 2001;Duffus, 2009;Price et al., 2017). This observation is intriguing since larval forms are usually more affected by FV3 elsewhere in the world and common frog larvae have been shown to be highly susceptible to wild-type ranaviruses in the laboratory (Duffus, Nichols, & Garner, 2013Gray et al., 2009  October outbreaks will alter the life history stages at risk. If common frog tadpoles become affected, the abundance of susceptible hosts will be increased with concomitant impacts on the ranavirus basic reproductive number (R 0 ) (Altizer et al., 2006); that is, the dynamics of outbreaks will be fundamentally changed, making predictions of their impacts more challenging.
Additional research on the effect of climatic changes on the breeding phenology of UK amphibians is needed, but it has been suggested that the common frog differs from other UK species in having shown no response in the timing of breeding to increasing absolute temperatures (Beebee, 1995). Therefore, any compensatory changes in common frog behavior to future climate change may be negligible.  Our laboratory findings that virulence is reduced at lower temperatures, that frogs might be better able to manage infections at lower temperatures, together with field records showing a mitigating effect of shading, pond volume, and vegetation (which might also be due to lowered temperatures of frogs) on incidence and severity of ranavirosis, point to possible steps for mitigation. Thermoregulatory behavior leading to an increase in body temperature above normal range ("behavioral fever") is known to be important for disease mitigation in a range of ectotherms (Boltaña et al., 2013;Elliot, Blanford, & Thomas, 2002;Monagas & Gatten, 1983). Whether thermoregulatory behavior serves as an amphibian strategy for managing infections remains to be elucidated (Sauer et al., 2018) but shading has been proposed as an environmental feature that might have negatively impacted the ability of newts to clear chytrid infections because it limited the availability of warm-water patches (Raffel et al., 2010). Staying cool can be a more difficult challenge than getting warm for many ectotherms (Kearney, Shine, & Porter, 2009), but the provision of suitable opportunities for behavioral regulation of body temperature in the form of shading, log piles, and larger ponds might help to manage the severity of future outbreaks and warrants further study.
Ranavirosis has had a major impact on common frog populations in southeast England (Teacher et al., 2010) and the current study suggests that these impacts may become greater and more widespread (in the United Kingdom and elsewhere) if future climate change projections are realized. Our results-as well as the predictions that follow from them-are strengthened through the use of a model system that allows us to investigate possible drivers of field epidemiology using laboratory experiments, both at the cellular and whole animal levels. Together, our results present a clear case of the environment modulating an important host-pathogen interaction.
Few previous studies have convincingly shown how climate change affects disease emergence in wild animal populations, but we have been able to demonstrate a historic impact of warming in the wild and then tease apart relationships between the environment, host, and pathogen in the laboratory. These results underline the importance of accounting for the effects of local environmental drivers to predict the dynamics of an invading pathogen (Cohen et al., 2018(Cohen et al., , 2017Raffel et al., 2013).

ACK N OWLED G EM ENTS
We thank Rob Knell for help and advice about climate-chamber construction. This work was funded by NERC grants NE/M000338/1, NE/M000591/1, and NE/M00080X/1. All in vivo experimental procedures and husbandry methods were approved by the ZSL Ethics Committee before any work was undertaken and procedures were performed under UK Home Office licenses P8897246A and 80/2214. All data and code required to reproduce the analyses in this article are included in the Supporting information files (Data S1-S8). We thank Froglife (Registered Charity No. 1093372 in England and Wales) for careful administration and promotion of the Frog Mortality Project. We also thank reviewers for helpful comments which improved the manuscript.

CO N FLI C T O F I NTE R E S T
The authors declare that there are no conflicts of interest.

AUTH O R CO NTR I B UTI O N S
SJP and AAC designed the in vitro experiments; CO and WL performed the laboratory work; and SJP with help from CO performed the analysis. SJP with help from TWJG and RAN designed the in vivo experiments; SJP, WL, and CS performed the laboratory work; and SJP analyzed the results. AAC oversaw collection of the epidemiological datasets; SJP, RP, TWJG, RAN, and FB planned the analyses which were conducted by SJP and RAN. SJP and RP planned the climate change projections, which were performed by SJP. SJP and FB wrote the first draft of the manuscript, which was edited by all authors.