2.1 Study site and field sampling
The only know site for M. admirabilis is located in municipality of Arvorezinha, State of Rio Grande do Sul, Brazil (52°18’’W, 28°51’S), in the Southern portion of the Atlantic Forest (Di-Bernardo et al. 2006). The climate is humid subtropical, and seasons (autumn, winter, spring, and summer) are differentiated by temperature (Zepner et al. 2021). We conducted our study along ~ 400 m on the Forqueta river’s bank, where most individuals of this microendemic species (range size of 1.6 km2, IUCN, 2023) concentrate to breed on temporary pools formed on flat rock outcrops. We surveyed the site at least twice a year, from September 2019 to December 2022 (exclusively between September and December), totaling nine field campaigns (Table 1). We captured each toad using new disposable latex gloves and swabbed the skin using MedicalWire MW113 swabs, following standard swabbing protocols (Hyatt et al. 2007). To calculate the body condition of each toad, we measured the snout-vent length (SVL) of each individual using a digital caliper with a precision of 0.01 mm and the body mass using a field scale with precision of 0.1 g. After sampling, we immediately released all toads at the exact point of capture.
We applied a photo-identification standardized procedure (Bardier et al. 2019; Caorsi et al. 2012) as a mark-recapture method. The ventral region of each toad was photographed (Figure S1), and we used the Wild-ID open-source software (Bolger et al. 2012) to identify recaptures. The software compares all images for pairwise similarity and returns the 20 top-ranked potential matches for each focal image; all recaptures were visually confirmed afterwards. This mark-recapture method has been successfully applied to M. admirabilis (Fonte et al. 2022) and other species of Melanophryniscus (Caorsi et al. 2012; Bardier et al. 2019) due to their black, brown, or green background with red, yellow, white, green, or orange spots on their belly (Figure S1).
2.2 Pathogen detection and quantification
To assess a sensitivity performance parameter for Bd diagnostic in our QuantStudio™ 6 Real-Time PCR equipment, we performed a series of replicate standard curves, totaling 20 replicates per standard concentration to calculate the limit of detection “LoD”. LoD is defined as the lowest amount of target DNA sequence that can be detected with 95 % probability. We ran a dilution series of a known amount of total Bd DNA, ranging from 1.83 x 105 GE/µL to 10-3 GE/µL. Each plate included 4 technical replicates of each standard concentration, resulting in a total of 32 standards samples per plate, as well as 4 negative controls consisted of DNA-free water. Bd DNA was extracted from a culture (isolate CLFT 159, global panzootic lineage) using PrepMan ULTRA® (Life Technologies).
For the qPCR assay, we utilized a final volume of 25 µL, containing 5 μL of template DNA, 12.5 μL of TaqMan Fast Master Mix (Applied Biosystem), 3.75 μL of ddH2O, 1.25 μL of forward primer (ITS-1 Chytr CCTTGATATAATACAGTGTGCCATATGTC, 18 μM), 1.25 μL of reverse primer (5.8S Chytr AGCCAAGAGATCCGTTGTCAAA, 18 μM), and 1.25 μL of probe (Chytr MGB2 GCAGTCGAACAAAAT, 5 μM). We used thermal cycling at 50 °C for 2 minutes and 95 ºC for 20 seconds, followed by 50 cycles at 95 ºC for 1 second and 60 °C for 20 seconds. The outcome of our analysis indicated that the lowest standard concentration with detection rate of 95 % or greater detection was 0.1 copies per reaction (1.83 x 10-1 GE/μL) (Table S1, Figure S2).
To determine the presence and infection loads of Bd in each swab sample, we extracted DNA from skin swabs using PrepMan ULTRA® (Life Technologies). To quantify Bd infection loads, we used a Taqman® qPCR Assay (Life Technologies) with standards ranging from 10-1 to 103 genomic equivalents of zoospores, hereafter referred as GE (Boyle et al. 2004; Lambertini et al. 2013). We considered samples to be Bd-positive (Bd+) when the infection loads were ≥ 0.1 GE.
2.3 Abiotic data
We obtained the mean temperature for the 15 days leading up to the sampling from two automated weather station located approximately 23 and 37 km from the study site (Instituto Nacional de Meteorologia do Brasil). We used 15 d prior to sampling based on the Bd life cycle (i.e. this period is enough to allow at least 1 generation of Bd; Berger et al. 2005). To assess temperature anomaly metrics, we used the TerraClimate dataset (Abatzoglou et al. 2018). We calculated the deviation from the historical temperatures by subtracting the historical monthly mean temperature over the past 50 years from the mean temperature of the target periods. These temperature deviations (ºC) were calculated for one, two, and three months preceding our sampling, incorporating one-month lagged deviations. The resulting deviation values could be negative for colder-than-average months, and positive for warmer-than-average months. Additionally, we recorded the accumulated rainfall for each month prior to sampling using data from a neighboring hydrometeorological station located at a similar altitude, 5.5 km from the study site (Agência Nacional de Águas). To assess rainfall anomaly metrics, we calculated the deviation from historical rainfall by subtracting the historical monthly mean of accumulated rainfall over the past 50 years from the mean accumulated rainfall of the target periods (Abatzoglou et al. 2018). Like our temperature anomaly metrics, we extracted rainfall deviations (mm) for one, two, and three months prior to the sampled month. This analysis provided us with negative values for dryer-than-average months and positive values for wetter-than-average months.
2.4 Data analyses and modelling
For statistical analysis, we employed two model selection approach. Firstly, to screen for important climatic anomaly metrics explaining Bd prevalence and infections loads, we employed a model selection approach based on Generalized Linear Models (GLM), thus reducing potential multicollinearity bias in downstream pruned models. Based on the Akaike Information Criterion (AIC) (Mazerolle, 2006), the three-month temperature deviation, and two-month rainfall deviation were the best predictor explaining Bd prevalence. Additionally, two-month temperature deviation was the best predictor explaining Bd infection load, while three-month rainfall deviation was the best predictor variable. A detailed description of these models can be found in the supplementary information (Table S2).
Secondly, to test for the potential effect of monthly climatic fluctuations and climatic anomalies explaining Bd prevalence and infection loads, we employed a model selection approach based on GLMs. For Bd prevalence, we fit a GLM with binomial distribution and logit link function. We also fit a Gaussian GLM with log link function, with Bd infection loads (log10-transformed GE; only Bd+ samples) as the response variable. The explanatory variables in the global models were: year, month, mean temperature of the 15 days prior to sampling date, accumulated monthly rainfall from the month prior to sampling date, and climatic anomaly metrics depicting temperature deviation and rainfall deviation. For each GLM, we included year, month (Bd prevalence) and season (warm or cold, Bd infection load model) and as fixed effects. A detailed description of these models can be found in the supplementary information (Table S3). We ran models with all possible combination of explanatory variables, and we ranked the most parsimonious models by employing a backward stepwise procedure based on the Akaike Information Criterion (AIC) (Mazerolle, 2006; Table S3).
Finally, to investigate any significant differences in body condition between Bd infected and uninfected toads (explanatory categorical variable), we performed a Student’s t-test, including data of body condition from adult males sampled until August 2021 (n = 155). We performed a linear regression analysis to test for the potential effect of Bd infection loads on body condition of Bd+ males (n = 33). The body condition metric utilized the Scaled Mass Index (SMI) approach, based on standardized major axis regression between mass and snout−vent length (Peig and Green, 2009). We excluded females from this analysis because the presence of eggs could bias our results. All statistical analyses were performed using R version 2023.3.1 (R Core Team 2022).