Here we test the hypothesis that island species that have evolved in predator-poor environments have reduced locomotor abilities. More than 60% of recent recorded extinctions are from islands 1, and island taxa are often susceptible to invasive predators typically because of the loss of ancestral antipredator behaviors. While locomotor abilities are critical for escaping predation, little is known on how the presence of different types of native predators influences these abilities by maintaining selective pressure. To fill this gap, we documented sprint speed in the Aegean wall lizard (Podarcis erhardii) from Aegean islands (Greece) with varying levels of predation pressure. We show that on islands where mammalian predators were present, lizards sprinted fastest. Lizards sprinted at an intermediate speed where predators other than mammals were present, and lizards sprinted slowest on islands where no predators were present. These results indicate that lizards from the lowest-predation islands are most vulnerable and preventing the introduction of invasive predators should be prioritized for these island systems.

Study system

The Cyclades archipelago is a cluster of land-bridge islands located in the central Aegean Sea (Greece). Since the end of the last glacial maximum approximately 18,000 years ago, rising sea levels have fragmented former mainland coastal regions. As Holocene sea level rise separated the area, populations of plants and animals also became increasingly isolated, thus setting the stage for the evolution of locally adapted island phenotypes.

The study area has warm, long, dry summers and mild, rainy winters as is typical of Mediterranean region climates. Humans have inhabited the region for thousands of years, altering the vegetation through agriculture and animal grazing. Today island habitats consist typically of agricultural fields edged by dry-stone walls and embedded in a matrix of spiny, summer-deciduous, low-growing woody vegetation known as phrygana.

The Aegean wall lizard (Podarcis erhardii, Lacertidae) is a medium-sized lizard, typically ranging in size from 48-78 mm from snout to vent.  P. erhardii is found throughout the southern Balkans, mainland Greece, and many Aegean islands, and is common throughout the Cyclades island group. The species has – with the exception of closed canopy forests – broad habitat preferences and can be found in particularly high densities in areas where appropriate refugia, like dry-stone walls are present. The diet of P. erhardii consists predominately of arthropods, although it occasionally includes vegetation and fruit. Aegean wall lizards are poor overwater dispersers and genetic studies confirm that each population has evolved in response to locally prevailing conditions since isolated by rising sea levels. More than 25 island subspecies are recognized reflecting the broad morphological and genetic variation found in the different Aegean islands and pronounced adaptation to local environments.

Island Characteristics

Study sites (N=22 see Fig. 1, Table 1) were selected to include a range of island sizes and ages (i.e. periods of isolation), as well as predation environments.  Using methods described more fully in Brock et al. and Pafilis et al., we assessed predator presence on an island by combining published information with field surveys for any direct (live or dead individuals) or indirect (burrows, fecal matter, tracks) evidence of predators. Predator species were grouped into six categories: rats (R), predatory mammals (M), raptorial birds (B), vipers (V), sand boas (SB), and other snakes (OS). Each predator category is characterized by distinct hunting strategies, which likely results in different antipredator behaviors. Rats (Rattus rattus), for example, are considered in a category distinct from “mammals” because in the Cyclades the rats are small, opportunistic predators that do not prey on lizards to the same effective capacity as feral cats (Felis catus) and stone martens (Martes foina) which use burst speed to capture prey. Snakes are separated into three categories based on hunting strategies. Vipers (Vipera ammodytes) are sit-and-wait predators that ambush and envenomate prey. Sand boas (Eryx jaculus) are constrictors that will prey on adult and juvenile lizards, but mostly feed on lizard eggs. ‘Other snakes’ includes several diurnal colubrid taxa (e.g. Natrix natrix, Elaphe quatuorlineata, Dolichophis caspius) that actively search for prey.

Lizard population density measurements were taken along one 100 m-long, 5 m-wide transect within each study site.  Following established protocols, lizards seen or heard were counted by an observer (SS) while walking along the transect during peak hours of lizard activity (09:00-11:00, and 15:00-17:00) and under optimal environmental conditions (20-25 ^oC; sunny; windspeeds<2Bf). Previous research has suggested that habitat openness is an important determinant of sprint speed. At each location, we measured habitat openness along a 100 m long transect randomly laid out within the habitat occupied by P. erhardii at that location. Substrate type, vegetation type and height above the ground were recorded at 1 m intervals along the transect. Areas with vegetation <5cm high were categorized as open habitat amenable to rapid running.

Animal Methods and Morphology

Lizards were caught by noosing or baiting with mealworms (Tenebrio sp. larvae) during early summer (May-early July) 2016. A sample of 18-34 lizards were taken from each location. Each lizard was given a unique, temporary number on the back using a non-toxic marker. While in captivity, lizards were held in 60cm x 41.6cm x 33.7cm terraria, and were provided with water ad libitum, and fed mealworms once a day. Morphometric data, including snout to vent length (SVL), front and hind limb length, and hindleg span (hindspan) were measured using digital slide calipers. Lizards were sexed and body mass measurements were collected using a spring-loaded scale, (Pesola 10020). Lizards were returned to the place of capture after all morphological and performance measurements were recorded.

Sprint Trials

Sprint speed was measured using a 230 cm long and 40 cm wide wooden racetrack. Every lizard ran 3 sprint trials to ensure that at least one good measure of sprint speed was obtained. Before each sprint trial, lizards were allowed to thermoregulate for at least an hour in a temperature gradient. Temperatures were taken by cloacal thermometer immediately after each trial. Trials were spaced at least 1 hour apart to allow the lizards time to recover. Using a video camera (GoPro; Hero Black 4; 1280 x 720 px), video of each trial was recorded at 240 frames per second; the camera was positioned 1.5 m above the racetrack, so that a clear dorsal view of the running lizard was visible for at least one full meter of the racetrack. Using the video analysis tool SAVRA (code: https://github.com/bkazez/savra) custom-built for Donihue, a measurement of the distance travelled between every 5 frame sequence was calculated, and fit with a quintic spline using the SPAPI function in MatLab (MathWorks Inc. 20) as described in Donihue in order to calculate velocity of the lizard in m/sec. The fastest of the three trials for each lizard was selected for use in analyses. Gravid females were excluded from the analysis, as were trials where individual lizards did not run normally. Stamina and sprint trials were run more than 12 hours apart to allow the lizards time to recover between measurements.

Stamina Trials

In addition to sprint trials, a subset of 20 lizards was randomly selected from each island sample to participate in stamina trials. Gravid females were further excluded from the analysis. Lizards were allowed to thermoregulate for an hour before each trial and temperatures were recorded by cloacal thermometer immediately before the start of each trial. Lizards were placed in a circular track, 2.71 m in circumference, with the bottom covered with sand. Researchers worked in pairs, the animal handler encouraged the lizard to run for as long as possible and the observer used a stopwatch to record the length of each trial to the nearest second. Lizards were encouraged to run along the track by light tapping on the base of the tail. A trial ended when an animal did not start running after ten tail taps and was unable right itself when flipped on its back. Three stamina trials were performed on each lizard, with at least 90 minutes between trials to allow for recovery.

Statistical Analysis

Performance (sprint speed and stamina), and most morphological values (SVL, hindlimb length, and hindspan length) and island characterics (island age, island area, and lizard density) were Log₁₀-transformed to ensure normality of residuals. Visual inspection and heterogeneity tests confirmed that error variation was approximately Gaussian and uniformly distributed in all analyses.

Pearson correlations were used to explore relationships between locomotor performance and morphological characteristics, as well as island characteristics (island age, island area, % open habitat, lizard density, and number of predator categories).

We used a linear mixed model approach to analyze how sprint speed and stamina were influenced by different ecological drivers. Different predation variables were included in different models. All models included sex, lizard body size (snout-vent length), and lizard body temperature to control for these factors that are known to affect lizard running performance. Snout-vent length is typically used as a measure of body size, which is linked to stride length and sprint speed, and body temperature has also been broadly linked to sprint speed. Previous research in this study system suggested island age was important to other antipredator defenses, so it was included in the linear mixed model analysis for further examination.

            We compared 8 a priori models to identify which predator types were most important in shaping sprint speed and stamina. Models looked at the presence or absence of any predators, total predation, and individual predator types. The binary “Zero Predation” model separated islands with no predators present from islands with any type of predator. The “Sum Predation” model summed all the predator types found on each island, assuming each predator type added equal predation pressure to the system as described in Brock et al. Alternatively, another model estimated an effect of each predator type individually and simultaneously (R+SB+V+B+OS+M). This model considers the fact that predator types are not the same and might each affect sprint speed or stamina in unique ways. Based on a priori knowledge, the other five models emphasized mammals (M) and predatory birds (B) as the predators most likely to affect locomotor performance. In order to catch prey, mammals stalk and chase prey, often capturing an animal after a burst of speed. Predatory birds utilize a similar approach: after locating prey from above they swoop down to capture an animal. Models of birds, mammals, and birds and mammals combined were calculated with and without separating the zero predation islands. Including zero predation has the effect of calculating whether other predators beyond mammals and/or birds had an impact on sprint speed. In order to, determine whether time of isolation influenced sprint speed and stamina, island age (Log₁₀-transformed) was added as another fixed effect to the 8 models stated above. Models were then compared using Akaike Information Criteria, using AICc values and the associated Akaike weights.

Finally, we examined a potential trade-off between stamina and sprint speed, after controlling for other potentially confounding factors. In a general linear model, stamina for each individual was predicted by the combined effects of maximum sprint speed, individual sex, snout vent length, and trial temperature.

The three main files used in the analyses were the Dataset 1,2, and 3 files. These files were used for sprint speed analysis, stamina analysis, and island averages respectively.

However, there is a lot of additional data from video analysis. Every sprint video is available and is labeled as the island name and the number of the video (eg. Amorgos1.mp4) the can be found in the zip file for each location (i.e. Amorgos.zip). Video files were analyzed using the program SAVRA (see the link in the methods below), values from this program were then saved into an Excel file that was labeled with the island name (eg. Amorgos5.xlsx). These files were then used in Matlab to calculate velocity (code is in files: get_quintic_spline_and_max_va.m and run_for_all_the_lizards.m). Using this code we obtained velocity (sprint speed) values, that were then saved into Microsoft word documents (eg. amorgosvalues.docx). These values were then matched to the appropriate video, lizard, and trial. This information is available in the file (SemegenetAl Sprint Speed Trial Numbers.xlsx) some trials have more that one video listed because the quality of the video was poor or the lizard failed to run the full distance on camera. These values were then input into the Dataset 1 sprint speed file. Some trials were thrown out because values were significantly off. This data was then input into the main files used for analyses. Correlation tables are in the file labeled “SemegenetAl Pearson Correlations.” Results and syntax of mixed models are in the files labled “Sprint Mixed Model Outputs Log10 SemegenetAl” and (Stamina Mixed Model Outputs Log10 SemegenetAl”. A detailed methodology is included in the Methods section above.