HIGHER INFECTION PREVALENCE IN AMPHIBIANS INHABITING HUMAN-MADE COMPARED TO NATURAL WETLANDS

The fungal pathogen Batrachochytrium dendrobatidis (Bd) causes the emerging infectious disease, chytridiomycosis, in amphibians (Longcore et al. 1999). Chytridiomycosis is a leading factor in global population declines and species extinctions of amphibians in recent decades (Scheele et al. 2019). Batrachochytrium dendrobatidis infections can vary spatially within and among species due to differences in habitat quality, temperature, and host community composition (Kriger and Hero 2007; Searle et al. 2011). Exploring amphibian communities and their diseases across varying levels of human activity is important for developing methods to control infectious disease and manage productive wetlands.

Anthropogenic practices can influence Bd infection dynamics through direct effects on the pathogen, often decreasing disease risk for hosts (Kriger and Hero 2007; Stockwell et al. 2015). For example, elevated levels of salinity (Stockwell et al. 2012, 2015), and pH at or below 4 and at or above 8 reduce growth and suggest reduced transmission (Piotrowski et al. 2004). Similarly, pesticides from agricultural fields can inhibit in vitro Bd growth and abundance, and increase survival rates of infected individuals (Reeves et al. 2017). Water temperature, influenced by runoff or canopy shading, can alter Bd growth, which grows best between 17 C and 25 C (Piotrowski et al. 2004; Becker et al. 2012; Heard et al. 2014). Human-caused habitat loss in the tropics was correlated with decreased Bd infection prevalence and intensity, possibly due to increased temperature from reduced canopy cover (Becker and Zamudio 2011). Conversely, increased human population densities and traffic can increase Bd introductions, thereby spreading or maintaining disease risk (Morgan et al. 2007; Murray et al. 2011). The anthropogenic practices that can impact Bd infection dynamics are also known to influence host population structure and susceptibility.

Environmental factors that negatively affect Bd proliferation can also reduce amphibian survival and fitness (Relyea 2004). Human-made wetlands are established for water retention, water regulation, and aesthetics (Brand and Snodgrass 2010), and are more prone to anthropogenic disturbances such as runoff pollution or fragmentation. Natural wetlands are associated with vegetation or forest buffers from anthropogenic disturbances that are experienced in human-made wetlands. Exposure to pesticides and heavy metals in human-made wetlands can have immediate and long-term effects on growth, survival, reproductive fitness, and the population stability of amphibians (Sanzo and Hecnar 2006; Harless et al. 2011). Wood frog (Lithobates sylvaticus) tadpoles exposed to increased concentrations of chloride showed decreased survivorship, weight, and activity, while also experiencing faster time to metamorphosis and increased rates of developmental abnormalities (Sanzo and Hecnar 2006). Additionally, contaminant exposure can cause immunosuppression in frogs, increasing their susceptibility to pathogen infection (Belden and Kiesecker 2005; Peterson et al. 2013; McMahon et al. 2017; Rollins-Smith 2017). Human-induced changes to habitat connectivity and patch size can indirectly affect host-pathogen interactions through altered species density and diversity (Cumming 2002; Molyneaux et al. 2002; Power and Mitchell 2004; Johnson et al. 2008). Habitat connectivity can support frog metacommunities, in part by providing environmental refugia from pathogens (Heard et al. 2015). Investigating how anthropogenic disturbances affect both wildlife and their pathogens will clarify of how humans can impact ecological communities.

Human-made wetlands, both those established unintentionally or for water retention, are typically more prone to anthropogenic influence (e.g., increased runoff pollution) than are wetlands established naturally (Borne 2014; Chumchal and Drenner 2015). Such anthropogenic impacts can negatively affect local amphibian species and influence disease dynamics. It is not known if human-made and natural wetlands present the same risk of Bd infection for amphibian hosts.

We compared Bd infection prevalence between human-made versus natural wetlands to better understand the role of anthropogenic influence on Bd infection dynamics. We recorded frog density, community composition, water parameters, and prevalence of Bd infection at nine sites. We hypothesized that prevalence of Bd infection in human-made sites would be higher than in natural sites because human-related environmental impacts would negatively affect frogs' ability to resist infection. We also hypothesized that frog community composition would differ between natural sites because anthropogenic disturbances in human-made sites would be detrimental to frog population growth and stability. Examining the habitat characteristics that impact Bd infections can shed light on the risk of disease outbreaks in various locations and the environmental factors keeping infections at endemic levels.

We sampled nine sites in Tippecanoe County, Indiana, USA from May 2017 to October 2017 (Fig. 1 and Table 1). Only Mulvey Pond was sampled three times (May, June, and July), whereas all other sites except one were sampled five or six times (Ross Reserve was sampled seven times). All sites fall within the Central Till Plain Natural Region of the Wabash Drainage System. Site elevations ranged from 177 m to 234 m (mean=200 m). Average monthly rainfall for this area during sampling was 12.7 cm, with an average monthly temperature of 19 C (range=14–22 C), which overlaps with optimum growth temperature for Bd (17–25 C; Voyles et al. 2012; US Climate Data 2020). The months during which our sampling occurred also overlapped with the breeding phenology of the known frog species inhabiting these sites, and provide a temporal range to examine changes in frog community and infection parameters. All sampling events were conducted at night between 2100 hours and 0300 hours. Sites were selected if they were permanent wetlands and supported at least one frog species. We identified site origin (human-made or natural) by consulting with landowners and managers. Human-made wetlands were established by complete or partial development with human involvement and consisted of dugout drainage ditches, water retention ponds, and other means to regulate river flow (Fig. 2A, B; Brand and Snodgrass 2010). Wetlands were classified as natural if there was no evidence of human involvement in their creation or water flow control, and included ponds, marshes, and prairie pothole wetlands (Fig. 2C, D). Characteristics of water quality were collected during each visit along a randomly chosen transect to document microhabitat characteristics of sampled frogs, and included water and air temperature, salinity, total dissolved solids (TDS), conductivity, and pH with a Multi-Parameter PCSTestr^™ (Oakton Instruments, Vernon Hills, Illinois, USA) at 5–8 cm below the water surface. We disinfected all equipment between sites to minimize transmission of Bd and other pathogens with 10% bleach, or by rinsing off organic matter and allowing equipment to completely dry for 24 hr between sites (Johnson et al. 2003).

In addition to on-site data, we characterized surrounding land-use type using a buffer analysis within a 500 m radius of each site in R version 3.4.1 (packages: raster, rgdal; Hijmans and van Etten 2012; Bivand et al. 2014; R Core Team 2018). This allowed us to calculate the proportion of surrounding land use based on the 2011 National Land Cover Database (Homer et al. 2015). Based on the National Land Cover Database, we grouped land use into an agriculture category that contained cultivated-pasture classes, an urban category that contained developed classes, and a forest category that contained forest classes (Fig. 1). Nearest road proximity was measured using aerial photos with Google Earth Pro 7.3 (Google Inc., Mountain View, California, USA) as the shortest distance (m) between a paved road and the shoreline of the wetland.

To measure frog communities, three to four visual encounter surveys (VES) were conducted at every site during each survey period, depending on wetland size. Each VES was 30 m in length and evenly distributed across each wetland. We captured up to 20 individuals per species with gloved hands or large landing net at each site. Gloves were changed and discarded between each captured individual during VES and swabbing, and nets were cleaned of organic debris. Upon capture, frogs were immediately placed in separate plastic bags to record individual data (species, age, sex, and body length [snout-vent length]), and to conduct skin swabbing for Bd infections.

To collect samples for Bd testing, one MW 113 rayon-tipped culture swab (Medical Wire and Equipment Co. Ltd., Corsham, England) was passed along three areas of each frog for a total of 40 swipes (10 on the ventrum, 10 on each inner thigh, one under each toe) while wearing latex-free nitrile gloves (Hyatt et al. 2007). The swab was immediately placed in a 1.5 mL micro-centrifuge tube and allowed to air dry away from continued swabbing for approximately 15 min to reduce potential inhibitory effects of water in the sample. Swabs were then stored in a –20 C freezer until DNA extraction.

Extraction and quantification of Bd from swabs was performed following standard protocols for a qPCR Taqman assay (Boyle et al. 2004; Hyatt et al. 2007) using gBlocks (Integrated DNA Technologies, Coralville, Iowa, USA) for Bd internal transcribed spacer genes as standards, which included four serial dilutions in duplicate on each plate. Each plate was run on a CFX Connect Real-Time System machine (Bio-Rad Laboratories, Hercules, California, USA). To detect inhibition within the reactions, TaqMan Exogenous Internal Positive Controls (Applied Biosystem, Foster City, California, USA) were included in each sample (Hyatt et al. 2007). When no amplification occurred and if the qPCR reaction was not inhibited, the sample was considered negative for the presence of Bd.

All field sampling was conducted under permission of the Purdue Animal Care and Use Committee (no. 1607001450) and the Indiana Department of Natural Resources Scientific Purposes license (no. 17-020).

First, we tested the influence of biotic and abiotic variables on infection status of frogs. We used a generalized linear mixed-effects model (GLMM) with binomial error distribution to identify factors that best predicted the response variable of infection status (infected or not infected) of each individual sampled for Bd. We fit a GLMM model using the glmer function in the lme4 package in R (Bates et al. 2015). All explanatory variables were tested for multicollinearity using variance inflation values (VIF), and a cutoff value of five was used to begin removing collinear variables (James et al. 2013). Sex and age had VIF values greater than five. We then removed sex and recalculated VIF values; all remaining values were below 5. Included as fixed effects were site origin, age, sampling month, species, snout-vent-length, and road proximity. We included sampling site nested within sampling date as a random effect to control for the fact that infection status of individuals from the same site was not independent, and sites sampled on the same date might have infection status levels that were more similar to one another than from other dates (Harrison et al. 2018). Starting with this global model, a set of GLMMs was created, followed by model selection (Burnham and Anderson 2002). Model assessment was ranked by Akaike information criterion (AIC), and multiple comparisons of differences within single significant factors were determined using the drop1 function (Bates et al. 2015). A threshold ΔAIC of two was used to distinguish differences among models (Burnham and Anderson 2002).

Because all habitat characteristic parameters were not measured at every pathogen sampling event (all parameters measured at 23/34 events) due to lack of or malfunctioning equipment, we used data from only those sampling events that included all water parameter measurements in a separate binomial GLMM to examine effects on Bd infection status (all other variables collected on same day habitat parameters were still included). We constructed a GLMM with binomial error distribution similar to above. All explanatory variables were scaled (mean=0, SD=1) to bring the size and variability of the different data to a common scale. We tested for multicollinearity of explanatory variables using VIF, and a cutoff value of five was used to remove collinear variables (James et al. 2013). Conductivity, salinity, and TDS had VIF values greater than five. We then removed conductivity and salinity and all remaining values were below five. Included as fixed effects were air temperature, water temperature, and pH. We included sampling site nested within date collected as a random effect. We constructed a set of GLMMs starting with a global model that included all variables, followed by model selection similar to above (Burnham and Anderson 2002).

Next, we examined how frog density was influenced by multiple environmental variables we collected. We used stepwise model selection of constructed linear regression models to determine the most impactful explanatory variables for frog density (mean frogs/m²) and interpreted the best fit model determined by AIC with correction for small sample size. A threshold ΔAIC of two was used to distinguish differences among models (Burnham and Anderson 2002). Explanatory variables analyzed were TDS, salinity, conductivity, pH, surrounding land use, site origin, and distance to nearest road. All explanatory variables were scaled (mean=0, SD=1) to bring the size and variability of the different data to a common scale. We found pH, conductivity, salinity, and TDS were collinear, so we removed all but pH, which had the lowest VIF value. We also found all land-use proportion estimates were collinear and removed all except proportion urban land use. We recalculated VIF values and all remaining values were below five.

To determine if there were differences in water quality in wetlands that were natural versus human-made, we conducted Welch's two sample t-tests for average water parameters (pH, salinity, conductivity, TDS, and water temperature) and amphibian density. Data were log-transformed to meet assumptions of parametric statistics.

Differences in species composition between human-made and natural sites were examined via nonmetric multidimensional scaling (NMDS) with the Bray-Curtis dissimilarity measurement (Venables and Ripley 2002) using the R package vegan (Oksanen et al. 2011). After performing the NMDS, a permutational multivariate analysis of variance to test for statistical differences in species composition between site types (human-made versus natural). To directly compare alpha diversity, we tested Shannon-Wiener diversity between site types (human-made versus natural) via one-way analysis of variance. All statistical analyses were performed in R version 3.5.1 (R Core Team 2018).

Mean Bd infection prevalence was 33% (118/361) across all samples, with five of seven species having at least one infected individual (Table 2 and Supplementary Material Table S1). The best supported GLMM model predicting infection included site origin, species, and month (Table 3). The mean prevalence of Bd infection was higher at human-made sites (38% [87/230]) than at natural sites (24% [31/131]; Fig. 3). Species and month were also significantly related to Bd infection prevalence. Multiple comparison analyses of species' probability to test positive showed American bullfrogs (Lithobates catesbeianus) was most likely to be infected, whereas gray treefrogs (Hyla versicolor), pickerel frogs (Lithobates palustris), and spring peepers (Pseudacris crucifer) were the least likely to be infected (Table 2). Infection prevalence among months showed a bimodal distribution with lowest percent infection in July (Fig. 4 and Table 2). The most robust model predicting infection based on habitat characteristics contained only water temperature (Table 4), in which there was a decreasing probability of infection with increasing temperature.

The best-supported model predicting frog density included only road proximity (Table 5), with denser frog communities at further distances to the nearest road (F_(1,7)=11.66, P=0.011, r²=0.571; Fig. 5). Mean amphibian density did not differ between site origins (t_(3,5)=–0.57, P=0.601), nor was site origin included in the best-supported model predicting frog density. Welch's t-tests revealed no significant differences between site origin for salinity (t_(6.5)=0.95, P=0.37), conductivity (t_(6.5)=0.97, P=0.37), pH (t_(6.4)=1.01, P=0.35), or TDS (t_(10.2)=1.26, P=0.24; Fig. 6), or water temperature (t_(4.2)=–0.22, P=0.83).

Based on the NMDS plot of amphibian assemblages (Fig. 7), community composition did not differ between habitat types. The overlap in datapoints and lack of separation suggest the compositions are quite similar between site types (F_(1,8)=0.223, R=0.031, P=0.943). Further, we found no difference in Shannon-Wiener diversity between human-made and natural sites (F_(1,7)=0.004, P=0.951).

We found that Bd infection prevalence in frogs was higher in human-made sites than in natural sites (Fig. 3), consistent with our hypothesis. This suggests that frogs occupying habitats with greater potential of human disturbance are more likely to acquire or maintain infection. Frog density and community composition did not differ between wetlands with different site origins, indicating that human-made wetlands can support frog populations that are similar to natural wetlands. Therefore, human-made wetlands can play a role in sustaining frog communities, but could pose a greater risk for Bd infection than natural wetlands.

The mechanism by which human-made wetlands have increased Bd infection compared to natural wetlands is unclear, but factors associated with human activity can impact wildlife disease dynamics (Bradley and Altizer 2006). In our study, human-made sites had, on average, higher levels of salinity, conductivity, TDS, and pH, although differences were nonsignificant. In contrast, Bd growth and host infection load has been shown to decrease at salinity concentrations over 1 part per thousand (1 g/L) in field and laboratory settings (Stockwell et al. 2012, 2015). Because we did not record salinity above 0.3 g/L (mean=0.15 g/L, SD=0.09), the salinity concentrations (and collinear with conductivity) we observed at our sites likely had little to no direct impact on Bd even if they were having sublethal effects on the amphibian hosts. It is also possible that other immunosuppressive abiotic factors that we did not measure, such as nitrate concentration, dissolved oxygen, pesticide, or heavy metal concentrations, are higher in human-made wetlands and are increasing host susceptibility to Bd infections (Christin et al. 2003; Parris and Baud 2004; Calderon et al. 2019). Examining immunocompromising stressors that are common in human-made wetlands could reduce negative effects associated with pathogen infection.

Differences in infection prevalence among months revealed patterns that are useful for monitoring Bd outbreaks and amphibian population dynamics (Fig. 4). The prevalence of Bd infection typically declines in many regions of North America in the summer due to high temperatures that limit Bd growth or cause Bd mortality, similar to our findings (Piotrowski et al. 2001; Longcore et al. 2007). This pattern was further supported by the negative relationship that we found between water temperature and infection probability, as temperatures increased above the optimal Bd growth range of 17–25 C (Piotrowski et al. 2004). Higher infection prevalence in spring could be due to increased frog densities when breeding, and frogs spending more time in Bd-inoculated water, providing more opportunities for transmission (Carey et al. 2006). Alternatively, stressful conditions caused by springtime reproduction, development, or contaminant runoff with winter snowmelt, can limit frog resistance or tolerance to Bd (Rollins-Smith 2001; Sanzo and Hecnar 2006; Karraker et al. 2008; Denver 2009). High infection prevalence in the fall suggests that environmental conditions become more suitable for Bd, likely driven by lower temperatures compared to summer months (Berger et al. 2004). Alternatively, amphibians experience increased susceptibility due to a depressed immune system following temperature decreases or cold exposure (Maniero and Carey 1997; Greenspan et al. 2017), as well as fall acclimation, potentially creating a period of increased susceptibility to pathogens (Raffel et al. 2006). Studying seasonal differences in infection prevalence is important to identify the time of year when species are most at risk of disease-related effects, and when to quantify infection in amphibian communities.

Differences in biotic conditions or human activities that we did not measure could explain the higher infection prevalence that we found in human-made sites. For example, zooplankton, which can consume Bd zoo-spores, potentially reducing infection risk in amphibians (Buck et al. 2011; Searle et al. 2013; Schmeller et al. 2014), might be negatively affected by undetected aspects of human-made sites. Further, wetland structure and vegetative diversity can influence frog diversity (Wassens et al. 2010), which can itself impact disease dynamics (Searle et al. 2011). Therefore, differences in diversity or structure of aquatic vegetation might have impacted disease dynamics, and should be considered in future studies. Additionally, human-made sites might also have higher rates of anthropogenic activity such that increased traffic at sites with easy public access results in more human-caused introductions of Bd (Morgan et al. 2007). More human activity and impacts can also lead to increased stress and corticosterone levels that depress immune functions (Dhabhar 2002; Padgett and Glaser 2003; Rollins-Smith et al. 2011). Erecting buffers, creating habitat with refugia harmful to pathogens, increasing habitat connectivity, or implementing shoe and equipment cleaning for the public and researchers can help to reduce the human-caused spread of pathogens and stress to wildlife (Phillot et al. 2010; Heard et al. 2018).

Infection prevalence was highest in American bullfrogs and green frogs (Table 2), two species that are suspected reservoirs for Bd with typically high infection loads (Daszak et al. 2004; Longcore et al. 2007; Klemish et al. 2012). Their year-round occupancy in wetlands suggests they might harbor infection through summer and winter, playing an important role in the increased infection prevalence we observed in spring and fall. Percent infection in gray treefrogs was high (19% [10/52]) relative to other studies, some of which find 0% infection (Longcore et al. 2007). Terrestrial species such as gray treefrogs lose infection outside of the breeding season (Longcore et al. 2007), suggesting that spring breeding events increases their risk of Bd infection. Measuring Bd infection in gray treefrogs and other predominately terrestrial species outside of the breeding season can better illustrate patterns of infection in these species.

We found that both pH and road proximity significantly predicted frog density. Although water quality parameters did not differ between site types (Fig. 6), pH (which was collinear with TDS, salinity, and conductivity) was positively correlated with frog density in our final model (Table 5). Low water pH is associated with lower larval amphibian survival, delayed metamorphosis, and increased adult mortality, and thus adults prefer more neutral environments (Sadinski and Dunson 1992; Horne and Dunson 1995; Vatnick et al. 1999). Future studies should explore how these and other habitat quality characteristics (e.g., pesticide types and concentrations), impact disease dynamics and amphibian communities of larvae and postmetamorphic individuals. Our results show distances to the nearest paved road are positively correlated with frog density across all sites (Fig. 5). This could be an important driver of frog community composition. Closer proximity to anthropogenic features, such as roads or residential landscaping, can result in amphibians being less common due to poor migration success from habitat fragmentation between wetlands and upland forests, and impacts of surface water pollution (Lehtinen et al. 1999; Knutson et al. 2000; Gagne and Fahrig 2007; Dodson 2008). Amphibians are at particularly high risk of road mortality in the region we sampled (Tippecanoe County) where a 2–yr-long study found 93% (9,809/10,515) vertebrate road mortalities were amphibians (Glista et al. 2008). Thus, human disturbances pose great risk to frog communities, regardless of whether the sites are natural or human-made. The similarity of these risks across sites could be why we did not observe differences in density or community composition between site types.

Our findings suggest that human-made and natural wetlands can act as comparably suitable habitats for frogs, but the conditions in human-made wetlands might be making frogs more susceptible to pathogen infection. The combined effect of anthropogenic disturbances and frog diseases can ultimately influence frog population numbers, and subsequently cause breakdowns of local food webs and the community structure of wetland habitats (Petchey et al. 1999). The creation of wetlands by humans, whether intended or not, can act as viable habitat for species being displaced by human encroachment. Therefore, ditches, water retention ponds, and other human-made wetlands should receive greater attention by conservation managers to increase and maintain the quality of these wetlands for the preservation of native species while also considering the effects of these activities on infectious disease.

We thank the Purdue University Department of Biological Sciences for funding and resources. We thank the Department of Forestry and Natural Resources at Purdue University and the Ross Biological Reserve for access to field sites. M.C.B. was funded through the Summer Stays program at Purdue University. We thank A. Catenazzi, T. DeBlieux, J. Hoverman, A. Shepack, and H. Zumbado-Ulate for training and processing of swab samples. We thank R. Lim, L. Rosario, and J. Toth for help with sample collection and processing.

Supplementary material for this article is online at http://dx.doi.org/10.7589/2019-09-220.