Site Selection and Capture

We encircled four vernal pools in Penobscot County, Maine, with drift fence for a related study (Hoffmann, 2017). Pools 1 and 2 were located in Old Town on a parcel managed for forestry by the University of Maine and abutting the Stillwater River (>200 m across). Pools 3 and 4 were located on privately owned residential parcels in the town of Orono. Their forest matrix was penetrated by residential neighborhoods and fields. We constructed the drift fences out of silt fence buried 20 cm into the ground with a pair of #10 aluminum cans every 5 m to act as pitfall traps (Shoop, 1965). We checked the traps every morning from April to September and alternating mornings in October and November (Hoffmann, 2017).

Surgery and Genetic Testing

We implanted 14 unisexual salamanders with radio transmitters in 2013 following the methods of Madison et al. (2010) and implanted another 29 in 2014. We monitored only large emigrating unisexual salamanders because A. laterale (LL) are generally too small to safely implant (Ryan and Calhoun, 2014), and we were outside the geographic range of A. jeffersonianum. We anesthetized salamanders in 3.1 mM tricaine methane sulfonate (MS-222) neutralized to a pH of 7.0 with aqueous NaOH until loss of pain response (toe pinching). We used surgical scissors to remove a 0.5 by 0.3 cm tissue sample (Nöel et al., 2011) from the tip of the tail, which we shipped in 70% ethanol to the University of Guelph to determine genomotype using microsatellite DNA analyses at six loci (AjeD75, AjeD94, AjeD283, AjeD346, AjeD378, and AjeD422; Julian et al., 2003). Our microsatellite DNA methods are described elsewhere (Bogart et al., 2007, 2009). We inserted ATS A2415 transmitters (0.33 g, Advanced Telemetry Systems, Isanti, MN) with the antennas removed and PIT tags (0.09 g, HPT12, 134.2kHz ISO FDXB tag; Biomark, Boise, ID) into coelomic cavities (Ryan and Calhoun, 2014) using 10-mm longitudinal incisions in the left ventrolateral abdominal walls (Ryan and Calhoun, 2014), then closed the wounds with absorbable sutures (Model PDS Plus, RB-1 taper, Size 5-0, Ethicon Inc, Somerville, NJ). Body mass ranged from 7.74–13.38 g, such that the transmitters represented ≤4.26% of body mass. The salamanders recovered overnight and were released under wet leaves outside the drift fence in 2013. After realizing that we might bias movement by placing the animals on one side of the drift fence, we released the salamanders back into the vernal pools in 2014. We extended our 2013 telemetry season from 42 days (battery life of one transmitter) to 92 days by replacing transmitters in six animals (after McDonough and Paton, 2007; Titus et al., 2014). Unfortunately, the skin was weak at the site of the original incision, and we found two animals with an open incision 7 and 12 days after the second reimplant surgery. Therefore, in 2014, we tracked each salamander for the life of only one transmitter.

Telemetry

We relocated 13 salamanders daily in 2013. In 2014, we tracked 27 salamanders and, because of the increase in sample size, we tracked each individual only once every three days. We used a Model R-1000 receiver (Advanced Telemetry Systems) and Model RA-2AK VHF antennae (Telonics, Inc. Mesa, AZ) for direct overhead localization (10-cm accuracy). In 2014 when a transmitter expired, we scanned an ∼20 m radius using a Reader 03 PIT tag reader (West Fork Environmental, Turnwater, Washington, USA) with a custom-built antenna (Blomquist et al., 2008) before dismissing a salamander as lost, extending tracking for 11 salamanders for a mean of 15 days (range: 3–36 days). We recorded locations with a handheld GPS (GPSMAP 62stc and eTrex 10, Garmin International, Inc. Olathe, KS). We considered an animal not to have moved if it was within 3 m on subsequent visits in 2013 and within 0.5 m in 2014 attributable to new information in Ryan and Calhoun (2014), who found that A. laterale was selecting habitat at a small scale.

Microhabitat Variables

We measured microhabitat variables at paired used and available plots in succession such that meteorological conditions and vegetation phenology were comparable within pairs. We measured variables at two scales: when the salamanders moved ≥3 m (larger scale), we compared 3-m radius used and available plots (Faccio, 2003); and for movements < 3 m (smaller scale), we compared 0.5-m radius used and available plots (2014 only, based on Ryan and Calhoun, 2014).Three available plots evenly surrounded each used plot with one available plot located along a random bearing and the others 120° to either side (Fig. 1). The center of available plots for the smaller movement was always located 6 m from the center of the used plot. Distances between the centers of plots at the larger scale varied by week and were determined by the median distance between sequential locations during the previous two weeks in 2013 (range: 56–6 m), except for during the initial two weeks when we used the first weeks' median distance. We used the distances measured in 2013 to space plots in 2014.

Fig. 1. 

Schematic of used and random points at our larger scale (3-m plots measured when a salamander moved >3 m). The path (thick black line) of a hypothetical unisexual salamander as it moved from a vernal pool (blue oval). Three random points (gray circles) were spaced 120° apart around each used location (red circles). We used the median distance moved by all salamanders in the previous 2 weeks to determine the distance of random plots from used plots (thin gray lines), such that random plots better represented the scale on which salamanders were making decisions than a constant distance.

We measured 22 microhabitat variables within plots based on studies of A. jeffersonianum, A. laterale, and other amphibians (Faccio, 2003; Rittenhouse and Semlitsch, 2007; Ryan and Calhoun, 2014; Groff et al., 2017). We recorded land use as forest, yard/field, or wetland. We visually estimated the percent cover of bare soil, all leaf litter, coniferous leaf litter, water, lawn/hay, moss, rock, vegetation < 1 m tall, vegetation between 1 and 3 m tall, and vegetation > 3 m tall but < 10 cm DBH (diameter at breast height), allowing cover amounts to sum to >100%. We counted the number of stumps. We used a spherical convex densitometer (Model-A, Forestry Suppliers Inc. Jackson, MS) to quantify angular canopy cover because we assumed light interception by tree canopy surrounding the plot rather than just directly above it may be biologically important as a cue for cover (Nuttle, 1997). We measured leaf litter depth to the nearest 0.5 cm, soil moisture (FieldScout TDR200; Spectrum Technologies, Inc. Aurora, IL), and soil temperature in °C (Model 9841; Taylor Precision Products, Oak Brook, IL) near the center of each plot. We measured the diameter and length of coarse woody debris (>10 cm) to calculate the total area covered (cm²) and recorded the maximum decay stage for 3-m plots only (Monti, 1997). Once a salamander had vacated the plot, we brushed away the leaf litter and recorded the number of horizontal and vertical small mammal burrow openings (after Faccio, 2003).

Analysis

We plotted locations in ArcGIS 10.3 (ERSI, Redlands, CA) and used the “adehabitatLT” package (Calenge, 2006) in Program R (R Core Team, 2016) to determine step length, cumulative distance, and maximum straight line distance from the vernal pool for each salamander. Unless otherwise noted, summary statistics are reported as mean ± SD. We used a Kruskal–Wallace test to determine whether maximum straight-line distances varied by pool and Spearnman's rank to determine whether distance was correlated with salamander mass (Jehle and Arntzen, 2000). We found the farthest distances from the pool for each unisexual salamander and compared the mean of these distances and 95% life zones for each pool to those calculated from published data for the parent species. Faccio (2003) tracked six A. jeffersonianum and followed Semlitsch (1998) in using a 95% confidence interval to determine the radius of a life zone that would include 95% of a study population's distances, suggesting managers keep habitat within this buffer intact to conserve populations. In contrast, we argue the use of confidence intervals to establish 95% life zones is incorrect and that quantiles are more appropriate. Confidence intervals are intended to give precision of estimation of the population mean (μ) such that the maximum of this interval is a possible population mean (P[lower CI ≤ μ ≤ upper CI] = 95%), not the area that includes 95% of the salamanders. We think Semlitsch may have intended to use quantiles, as we simply use the standard deviation rather than the standard error, such that the distance is not inflated by low sample size. Ryan and Calhoun (2014) sorted the distances traveled by A. laterale in ascending order and determined which distance included 95%. We recalculated life zones using the 95% quantile for these published data, the radio-isotope tracked A. jeffersonianum of Williams (1973), and our own observations using t-scores. Neither of these A. jeffersonianum populations has been genomotyped. However, Williams' animals were outside the geographic range of unisexual salamanders (Petranka, 1998; Charney, 2011) and Faccio's salamanders were likely mostly JJ based on sex ratios of the breeding population (S. Faccio, personal communication). Also, half of Faccio's telemetered salamanders were males, and, therefore, almost certainly JJ. Ryan and Calhoun's (2014) A. laterale population was known to contain no unisexual salamanders.

We used conditional logistic regression models to examine variables related to microhabitat selection by comparing plots used by individuals to their own available plots at a given time (i.e., study design IV; Erickson et al., 2001). We used the themes of land use, shelter, ground cover, microclimate, and vegetation to develop 22 specific a priori models (Table 1) about how salamanders may be using the landscape (i.e., if salamanders seek rotten wood when choosing where to settle rather than just any cover object, then a model containing just stumps and coarse woody debris would rank better than one also containing rocks) and 3 composite models (literature models in Table 1) based on the top models to predict plots selected by A. laterale from Ryan and Calhoun (2014). These models were not exclusive, such that variables in our shelter models may also appear in our ground cover models. We Z-standardized continuous variables, checked for collinearity (Pearson r ≥ 0.7) and ranked all models seperately for the 3-m plot and 0.5-m plot scales. We weighed each plot by the proportion of days the salamander spent there, such that the experimental units were animals rather than relocations (Aebischer et al., 1993; Thomas and Taylor, 2006). We used the “survival” package (Therneau, 2015) in program R to conduct conditional logistic regression to examine selection of microhabitat features. We used Akaike Information Criterion adjusted for small sample size (AICc; Burnham and Anderson, 2002) to rank models separately for each scale with package “AICcmodavg” (Mazerolle, 2016). We used the “support.Ces” package (Aizaki, 2012) to determine McFadden's R² (ρ). Rather than make inference based solely on the fit of individual models, we found the model averaged estimates of the odds ratios of the variables from models with a cumulative model weight of ≤0.9 and considered these variables important if the confidence intervals of their odds ratios did not include 1.

Table 1. 

Conditional logistic regression models used to compare each 3-m and 0.5-m plot used by unisexual salamanders to three paired random plots in central Maine. The variables CWD (coarse woody debris) area, CWD decay stage, and forest were not included in models for 0.5-m plots because of lack of field measurements and lack of convergence caused by low variation.

Distances

We tracked unisexual salamanders from 5 to 94 days (mean = 51 days), during which they moved an average straight-line distance of 172 m (range = 6–403 m) from the breeding pool (Table 1, Appendix 1). In 2013, the mean cumulative distance was 191 ± 76 m (range = 6–410 m). The mean cumulative distance in 2014 was 209 ± 140 m (range = 47–463 m). Maximum straight-line distance from the pool was not related to the body mass of the salamander (r = 0.090, P = 0.581) but varied by pool (Kruskal-Wallis χ² = 18.45, df = 3, P = 0.004). Salamanders in Old Town remained closer to the pools (mean = 112 ± 44 m and 36 ± 19 m) than did salamanders in Orono (mean = 244 ± 76 and 214 ± 113 m). Distance to the pool generally reached an asymptote within a week, as salamanders made large initial migrations with few short subsequent movements (Fig. 2).

Fig. 2. 

Distances from the breeding pool for emigrating unisexual salamanders quickly reached asymptotes in central Maine. Each color represents an individual salamander.

Ninety-five percent life zones for unisexual salamanders in our study also varied by pool (195 m, 74 m, 383 m, 415 m for Pools 1, 2, 3, and 4, respectively) and extended 362 m with all pools combined (Fig. 3). The 95% quantile for distance traveled by A. jeffersonianum in Indiana, was 478 m (mean = 252 ±136 m, N = 86, based on Williams, 1973), whereas the zone for A. jeffersonianum/unisexual salamanders in Vermont, was 143 m (mean = 92 ± 25, N = 6, based on Faccio, 2003). The life zone for A. laterale was 149 m (mean = 64.9 ± 50.1, N = 43, based on Ryan and Calhoun, 2014). These life zones are slightly smaller than those calculated by the authors (e.g., Faccio reported a 157-m zone based on confidence intervals, Ryan and Calhoun reported a 152-m life zone based on sorting distances by size, and Williams did not calculate a zone).

Fig. 3. 

Maximum distances (mean ± SD) from breeding pools for unisexual salamanders (this study) and their sperm-hosts (based on Williams, 1973; Faccio, 2003; and Ryan and Calhoun, 2014). Dashed lines represent the 95% quantile.

Macrohabitat

Most unisexual salamanders (36 of 40) remained within the forest matrix for the entire study, although 12 of these emigrated to post-breeding home ranges that were within about 20 m of forest-lawn or forest-hay field edges (Fig. 4). Seven salamanders occupied swamps dominated by Alder (Alnus incana) and Highbush Blueberry (Vaccinium corymbosum). The remaining four salamanders crossed lawns during emigration and spent the majority of the season underneath buildings (two salamanders under separate sheds with wooden floors and two salamanders under the same garage on a concrete slab).

Fig. 4. 

Paths of emigration unisexual salamanders (black lines) radiating from breeding pools (black polygons) in central Maine. Pool 1 (A) and Pool 2 (B) are in Old Town, and Pool 3 (C) and Pool 4 (D) are in Orono. Lawns and hay fields (white), forest (medium gray), roads and buildings (dark gray), and water (hatched) are shown.

Microhabitat Selection

At the larger movement scale (≥3 m movements), only the “All tunnels” model had substantial support (ΔAICc = 0, Table 2). Seven other models related to shelter, vegetation, and microclimate, including important models for A. laterale based in Ryan and Calhoun (2014) had some support (ΔAICc < 7), though McFadden's adjusted pseudo R² was low for the five models that did not include “Mammal burrows” as a covariate. We model averaged the β estimates and found the 95% confidence intervals for the odds ratio of each variable that appeared in larger movement scale models with a cumulative 0.9 model weight. Selection of these top models appears to be driven by three important covariates (Table 3), with unisexual salamanders more likely to use plots with more horizontal burrows, lower substrate temperatures, and less vegetation <1 m tall than available plots.

Table 2. 

Top-ranked unisexual salamander paired logistic regression models for used and random locations in central Maine. Only models with ΔAICc < 7 are shown. K is the number of parameters, adjusted ρ² is McFadden's adjusted pseudo R², and w is the model weight.

Table 3. 

Important model averaged parameter estimates (β), odds ratios, and descriptive statistics for unisexual salamander paired logistic regression models for plots of used and random locations in central Maine. Only covariates from models included in 90% of the cumulative model weight and with 95% confidence intervals or odds ratios that did not include one are shown.

At the smaller movement scale (<3 m), both the “All tunnels” model and “Mammal burrows” model had substantial support, and the global shelter and soil moisture models had some support, although all models had low McFadden's adjusted pseudo R². At the small movement scale, only horizontal burrows had an odds ratio with confidence intervals that did not include 1.