Trade-offs between aquatic and terrestrial locomotion and functional parallelism in Desmognathus salamanders

Experimental subjects

Ten adult semi-aquatic (D. amphileucus) and 10 adult aquatic (D. aureatus) morphs were collected from a stream near DeSoto Falls Recreational Area, Lumpkin Co., Georgia. These represent distinct species in the southern clade. Twelve adult semi-aquatic (D. sp.) and 12 adult aquatic (D. marmoratus) morphs were collected from Little Laurel creek along Whitetop Rd., Smyth Co. Virginia. These represent the ecomorphs in the northern clade. The northern semi-aquatic taxa cannot be distinguished visually and either D. mavrokoilius or D. kanawha could be present at the Virginia locality (Pyron & Beamer, 2022). Collections were made in 2010, before the taxonomic revision of Pyron & Beamer (2022). Based on the locality and genetic data of Pyron et al. (2022), the Virginia locality is in or near a contact zone between D. mavrokoilius and D. kanawha, and it is likely that they interbreed where they co-occur (Jones & Weisrock, 2018; Pyron et al., 2022). As a pragmatic matter for this analysis, they are lumped collectively as the northern semi-aquatic ecomorph, Desmognathus sp. Aquatic ecomorphs were collected by looking under rocks underwater in stream riffles and semi-aquatic ecomorphs were collected by looking under rocks along the edge of the stream or above water in the stream.

All 44 individual salamanders were evaluated for fast-start performance after 8–12 weeks in captivity. The sexes of the individuals were not known. Based on previous studies (Azizi & Landberg, 2002; Fitzpatrick, Benard & Fordyce, 2003), we intended to test 10–20 individuals per species. In practice, we let the less common ecomorph dictate the sample size for each collection (the aquatic morph in both cases). Individuals with tail damage were not collected. Individuals smaller than 50 mm snout-vent length were assumed to be juveniles and not collected (Martof, 1962). The average and standard deviation of the snout-vent lengths for each group, measured at the time of capture, were as follows: 66.5 ± 0.69 mm for southern aquatic, 74.9 ± 1.52 for southern semi-aquatic, 72.2 ± 1.23 for northern aquatic, and 60.6 ± 0.89 for northern semi-aquatic.

Salamanders were maintained at 15 °C and 70% r.h. in Sterilite 6-quart (5.7 L) plastic shoeboxes (34.5 × 20.3 × 12.7 cm). Semi-aquatic individuals were housed on a damp paper towel substrate, while aquatic individuals were housed in two L of de-chlorinated and well-aerated water. Bedding was changed weekly and water changed every three to four days. All salamanders were provided with a terra cotta refuge and were fed either crickets or California blackworms ad libitum. Food was withheld for two days prior to performance assessment to minimize effects of recent feeding on locomotion (Yan et al., 2015; Zhao et al., 2020). Salamanders were monitored daily for signs infection, loss of appetite, or impaired mobility. There were no adverse events.

Experiments were carried out in accordance with the University of Tennessee Institutional Animal Care and Use Committee protocol # 1870-0510, a scientific collection permit from the Virginia Department of Game and Inland Fisheries (# 37891) and the Georgia Department of Natural Resources. All animals were alive and healthy at the end of the experiment and were retained for additional research on mating behavior.

Escape performance trials

Salamander escape performance was assessed as fast-start velocity in response to a threat stimulus (Domenici & Blake, 1997; Walker et al., 2005; Fitzpatrick, 2008). A few studies have demonstrated that swimming and running velocity are good predictors of survival in amphibians and reptiles (Watkins, 1996; Walker et al., 2005; Husak, 2006). The Desmognathus species studied here are more active at night, but can be observed moving about at any time of day (Martof, 1962; Huheey & Stupka, 1967; Valentine, 1974; Dodd, 2004). They are not chemically defended and adults are preyed upon by large aquatic predators such as trout (Salvelinus fontinalis, Oncorhyncus mykiss), and semi-aquatic/terrestrial predators such as snakes (e.g., Nerodia and Thamnophis), river otters (Lutra canadensis), raccoons (Procyon lotor), and other medium sized mammals and birds (Martof, 1962; Petranka, 1998; Sánchez-Hernández, 2020; Dempsey, Roden & Bidwell, 2021). Thus, the velocity of the initial burst of swimming or running in response to attack is likely an important component of ecological performance. Moreover, sprinting speed can be correlated with other components of performance requiring physical vigor, including foraging and territory defense (e.g., Peterson & Husak, 2006).

Escape response was recorded three times for each individual salamander in both a terrestrial and an aquatic arena. The terrestrial arena consisted of a glass container with a gravel substrate and water level with the substrate. The aquatic arena consisted of a glass container with enough water to fully submerge a salamander. Salamanders were given at least 30 min to acclimate to room temperature because the camera setup was in a different room from where the salamanders were housed. The temperature in the camera room was 21–22 C and preliminary tests indicated that 200g of water equilibrated within 30 min. Previous work demonstrated that salamanders typically equilibrate to a 6–7 C change in temperature within 15 min (e.g., Hutchison, 1961; Lunghi et al., 2016).

After acclimating, a salamander was placed in an arena and allowed to acclimate until it ceased moving. A gentle pinch at the base of the tail with forceps was used to stimulate an escape response. I had no way of measuring the strength of the stimulus. Animals were always approached from above and behind, and presumably they could see the forceps approaching. If they moved before being touched, the forceps were withdrawn and the stimulus was attempted again when the animal stopped moving. On the rare occasions when this occurred three times in a row, the trial was stopped and the animal was moved to the end of the list of animals to be tested that day. Variation in the stimulus and/or motivation may affect escape responses (Barry & Harrison, 1957), but I assume this variation was unbiased with regard to the comparisons of interest for this study.

Escape responses were recorded at a rate of 500 frames per second using a high speed digital camera (NAC Memrecam Ci digital system). Trials were repeated only after at least a 30 min break to reduce stress on the animals and to prevent habituation to the stimulus. Azizi & Landberg (2002) found no evidence of fatigue across five escapes during 15 min trials with salamanders, so I assumed a 30 min break between single escapes was more than enough to avoid fatigue accumulation. If fatigue affected performance, I assumed the effects were unbiased with regard to the comparisons of interest for this study. The order of the trials was randomized to eliminate confounding treatment (aquatic vs. terrestrial) with ecomorph, species, or individual experience.

Kinematic analysis

Fast start performance data were collected using ImageJ software. Each video was analyzed independently by two technicians without prior knowledge of the species or ecomorphological hypothesis. Head, trunk, and tail measurements were taken once from each video when the animal was stationary, prior to the stimulus. Four timepoints were recorded from each trial. The initial position (position 0) was defined as the frame just prior to the first perceived movement. The subsequent three positions (positions 1–3) were the frames in which the trunk experienced maximum curvature during each of the first three lateral undulations (Fig. 2). The frame of maximum trunk curvature was determined by finding the frame with the minimum distance between shoulder and pelvis. Salamanders generate forward motion by generating a traveling wave down the trunk and tail when swimming or a standing wave in the trunk when walking (Frolich & Biewener, 1992). By recording the salamander’s position at each maximum curvature of the trunk, we can determine the velocity generated by each wave. Here I focus on the net velocity generated by the first three waves, estimated as the linear distance between positions 0 and 3 (cm) divided by the time in seconds.

I used the mean net velocity across trials and observers as a single fast start performance metric for each individual in each medium. My goal was not to estimate maximal performance, but rather to compare performance between ecomorphs. Mean velocity across trials is highly correlated with the maximum of many trials, and unlike the sample maximum, the mean is unbiased and yields more reliable correlations with explanatory variables (Adolph & Pickering, 2008; Head, Hardin & Adolph, 2011).

For the purposes of this study, I assumed that velocity (distance from the attacker over time, regardless of direction) is a good indicator of ecological performance. Other components of escape behavior, such as changes in direction or speed, can be important for survival (e.g., Brodie III, 1992). By focusing on the velocity of the initial burst, I intended to isolate basic physical ability from behavioral tactics. While direction was not quantified, most salamanders in this study tended to turn toward the center of the arena during aquatic trials, but moved directly forward during terrestrial trials.

Statistical analysis

To test for an overall signal of a trade-off, I estimated the Pearson correlation between terrestrial running velocity and aquatic swimming velocity expressed as body lengths per second. Further, to test whether the strength of the trade-off differs between northern and southern clades, I estimated separate correlation coefficients for each regional sample (Northern and Southern clades) and compared their 95% confidence intervals estimated via the normal approximation with the stats package in R (R Core Team, 2024).

To test whether the semi-aquatic and aquatic ecomorphs differed in performance, I used a mixed-effect linear model (Fox & Weisberg, 2019) with velocity as the dependent variable, species and trial habitat (aquatic or terrestrial) as factors, body length (measured from the snout to the groin), head length, and tail length as covariates, and individual as a random effect (because each individual was tested for both aquatic and terrestrial performance). I included all interaction terms, particularly a species by environment interaction term to test whether a performance difference between ecomorphs depended on the environmental context (aquatic vs. terrestrial). I then extracted the marginal mean velocity for each species x habitat combination and four planned contrasts to compare the species (different ecomorphs) within each clade under both aquatic and terrestrial conditions (Lenth, 2025).

Neither visual inspection of bivariate scatter plots nor the Shapiro–Wilk test of normality on the residuals indicated that any variable transformations were warranted (W = 0.9856, p-value = 0.4418). I used the variance inflation factor, VIF (Fox & Weisberg, 2019) to check for multicollinearity and found that head and tail length were highly correlated with body length. VIFs were acceptable and the model fit improved after eliminating head and tail length.