Every procedure in this study followed animal ethics rules and were approved by the Ethics Committee of the Plant Protection Institute. The Environment Protection and Nature Conservation Department of the Pest County Bureau of the Hungarian Government granted permission for the study (PE06/KTF/8060-3/2018, PE-06/KTF/8060-1/2018, PE/EA/295-7/2018).

We collected common toad eggs from a Hungarian pond “Bajdázó-tó” (coordinates: N47.903437, E18.978268) surrounded by natural woodlands. We collected ca. 30 freshly spawned eggs from each of six different egg strings and took them into captivity in a room with mean temperature 22°C ± 2.04 and natural outdoors photoperiod (via a window) during the entire investigation. Throughout the study, we used oxygenated aged tap water. The tadpole siblings were kept together until they reached developmental stage 25 (Gosner, 1960), at which point (day 1 of the study) we randomly selected five tadpoles from each sibling group, yielding a total sample size of 30 animals. We released the remaining tadpoles at the egg-collection site.

We kept the 30 tadpoles individually in 2-L white plastic boxes filled with 1 L water. The boxes were placed on four shelves in a random order, and the individuals’ order was changed every week until the experiment ended. We changed the tadpoles’ water twice a week and fed them ad libitum with finely chopped spinach. When the individuals started the climax of metamorphosis and forelimbs appeared (developmental stage 42), we reduced the water to one deciliter, and we angled the boxes to create a dry region within the box. When metamorphosis was complete (at developmental stage 46), we placed a wet paper towel in the bottom of the box to keep the animals moist, as well as an egg carton piece to serve as shelter for the toadlets. Twice a week, we fed the toadlets with ant-sized house crickets (Acheta domesticus) powdered with a mixture of vitamins and minerals. The top of each toadlet’s box was covered with a perforated, transparent lid to allow light in but prevent escape.

We took water-borne CORT samples from each individual on six occasions: one, two, and three weeks after the study began (days 8, 15, and 22), and three, four, and five weeks after the last individual had completed metamorphosis (days 57, 64, and 71). Then each toadlet participated in three behavioral assays to test their foraging performance under ALAN (day 84) and daytime noise (day 90) and their locomotor performance while fleeing from a human (day 93). We did not sample CORT around the expected time of metamorphosis because CORT levels change during that time as part of mediating that process (Jaffe, 1981; Kikuyama et al., 1993; Glennemeier and Denver, 2002), so individual differences in the timing of metamorphosis may have masked repeatability. Furthermore, freshly metamorphosed toadlets are very small and fragile, and we did not want to risk their drowning during water-borne CORT sampling.

After collecting the third CORT sample on each sampling occasion, we took photographs of the tadpoles to estimate their body size, and we measured the body mass of the toadlets (± 0.01 g) after gently patting them dry with a paper towel. We did not directly measure body mass in tadpoles to avoid the stress of drying them. Instead, we followed the methods of Davis et al. (2008), who showed that the dorsal surface area of the tadpoles correlates with their body mass. The tadpoles were photographed from above individually in a small plastic dish with an Olympus TG4 camera fixed on a tripod, under natural light, and including millimeter paper for size reference in the picture. We took several JPG images of each tadpole in macro mode and burst mode, and we chose one in which the tadpole was not moving and its dorsal surface was fully in view. We used the ImageJ software (Schneider et al., 2012) to measure the total surface area of the tadpole’s body (including the tail). We took a similar photograph of each individual when they reached developmental stage 42, upon which we measured their body mass as we did with toadlets, to estimate the relationship between mass and surface area. Using standardized major axis regression, we obtained the following formula: body mass (mg) = 50.1 + (1.8 × surface area in mm²); the relationship was highly significant (P < 0.001). We used this formula to estimate tadpole body mass at CORT sampling.

During the study, 4 out of the 30 animals died for unknown reasons, all belonging to a single sibling group that showed stunted growth. Throughout the paper we present data for the remaining 26 individuals. After the last behavioral assay, the animals were returned to their habitat of origin.

We took a baseline, stressed, and recovery sample of CORT released by individuals into the surrounding water ( Figure 1 ) using a non-invasive method of water-borne hormone sampling (Gabor et al., 2013b, a; Forsburg et al., 2019). This approach gives a time-integrated CORT measurement that is repeatable within individuals, correlates with plasma levels, and responds to an ACTH challenge (Gabor et al., 2013a, b; Forsburg et al., 2019). Earlier studies with various toad species showed that this approach is appropriate for sampling the CORT profile in both tadpoles and juveniles (Bókony et al., 2021; Monroe et al., 2024). Upon each sampling occasion, we placed each tadpole in a clean plastic insert (a perforated cup, to facilitate removal of tadpoles from the other cup) using a hand-made small net into a 250 ml plastic cup containing 100 ml water. We left the animals undisturbed for an hour to measure baseline CORT release rates. Then we moved them (with the perforated insert) into another plastic cup of 100 ml water, and we agitated them for an hour by gently shaking their cups for 1 minute every 3 minutes to measure stressed corticosterone release rates (Forsburg et al., 2019). We then transferred the tadpoles into a third plastic cup with 100 ml water and left them undisturbed for an hour to measure recovery CORT release rates. After metamorphosis we took CORT samples from the toadlets in a similar way, except that we put them into 50 ml water in a 750 ml plastic box and inserted another, identical box on top to prevent escaping from the water ( Figure 1 ). Each toadlet was moved using a separate mosquito net piece.

Each water-borne hormone sample was filtered using coffee filters (equivalent to grade 4 filter paper) immediately after collection and stored at −20°C. We extracted hormones from the water samples using C18 solid phase extraction (SPE) columns (SepPak Vac3 cc/500 mg; Waters, Inc., Milford, MA, USA) primed with 100% HPLC-grade methanol (4 ml) and distilled water (4 ml). Then we stored the columns at −20°C until analysis. After defrosting, we washed the columns with 2 ml distilled water and eluted them with 4 ml methanol into borosilicate vials, followed by nitrogen gas evaporation. After drying, we re-suspended the residue in a total volume of 500 μl consisting of 25 µl ethanol (95% lab grade) and 475 µl enzyme-immunoassay (EIA) buffer (Cayman Chemicals Inc., Ann Arbor, MI, USA). We measured CORT concentration in duplicates for all samples using Corticosterone EIA kits (№ 501320, Cayman Chemical Company, Inc.; assay has a range of 8.2–5000 pg/ml and a sensitivity (80% B/B₀) of approximately 30 pg/ml). Sample absorbance was read on a spectrophotometer plate reader at 405 nm (BioTek 800XS). Inter-plate coefficient of variation of the concentration standards was 12.8% across 19 plates; intra-assay coefficient of variation (CV) averaged 9.5% (range: 6.7–16.1%). Six plates had CV > 10%; samples with high intra-assay CV were re-run and we used the measurement with the lowest CV for each individual.

We calculated the amount of CORT in each sample by multiplying CORT concentration (pg/ml) by the resuspension volume (0.5 ml). We quantified corticosterone release rates (pg/g/h) as the amount of water-borne corticosterone measured divided by the duration of sample collection and body mass. To quantify the magnitude of stress response, we calculated the relative change of CORT release rate in response to stress (stress-induced change) as: 100 × (stressed − baseline)/baseline. Similarly, we quantified the rate of negative feedback as the relative change from stressed to recovery levels as: 100 × (stressed − recovery)/stressed (Lattin and Kelly, 2020). We also calculated a recently proposed measure of HPA/I flexibility (Zimmer et al., 2021): the square root of the mean squared differences (RMSSD) of corticosterone concentrations between successive sampling occasions. We calculated RMSSD for all CORT release rates (3 × 6 = 18 values per individual), and also separately for tadpoles and for toadlets.

The two assays of foraging behavior were performed in the same order for all individuals. Each assay consisted of an acclimation period, an 8-hours control trial, and an 8-hours disturbance trial, always in this order ( Figure 2 ). For acclimation, we placed the animal-containing boxes in a circle on the floor two days before the control trial, we removed the shelters from the boxes (to prevent the toadlets from hiding) and we left a less damp paper towel in each box (to prevent the crickets from drowning). We expected that the acclimation period would allow the animals to habituate to the slightly altered circumstances of the control trial, and thereby minimize any carry-over effect to the disturbance trial. Because we had no capacity for keeping a control group to quantify the sequential effect, we preferred to expose all individuals to each trial in the same order so that their behavioral data would be directly comparable to each other as well as to the GC measurements.

To investigate how ALAN affects feeding success, we measured the change in the number of crickets ingested, corrected for toadlet body mass. The control trial started at sunrise (0500h), when we took all the crickets from the boxes and measured the toadlets’ mass. The animals spent the day without food to standardize hunger levels. At sunset (2100h), we put twenty crickets into each toadlet’s box and left the animals undisturbed in the dark until the next sunrise. At 0500h, we removed all the remaining crickets from the boxes, counted them, and left the toadlets without food for the day. We began the disturbance trial at 2100h by putting twenty crickets into each box, but this time we left a light on until 0500h ( Figure 2 ). We used an EMOS ZQ5130 (LED-A60-PISW-WW) light bulb (230 V, 50Hz/70mA, −5~+40°C/8W/E27/2700K), which was 193 cm above the ground in the center of the area surrounded by the boxes. Light intensity was 66 LUX directly under the lamp, and 39 LUX in the boxes, as measured with Lux Light Meter Pro application. We counted the remaining crickets at sunrise, and after weighing the toadlets, we placed their boxes back on the rearing shelves.

We performed a similar assay to test how daytime noise affects feeding success. After two days of acclimation, we removed all crickets from the boxes for the night and weighed the toadlets. The next day at 0900h, we started the control trial by placing twenty crickets in each toadlet’s box. We left the animals forage without disturbance until 1700h, when we removed all the remaining crickets from the boxes and counted them. After a night without food, at 0900h we placed twenty crickets in each toadlet’s box again to begin the disturbance trial, during which we continuously played back urban sounds and noises until 1700h, using the SONYC Urban Sound Tagging Database (Cartwright et al., 2019) ( Figure 2 ). The decibel X PRO: Sound Metre application was used to measure the strength of the various noise levels at the boxes, which ranged from 45 to 98 dB.

To control for individual differences in body size, we divided the number of crickets eaten with body mass measured at the start of the control trial of the respective assay (i.e., ALAN and noise test). We are referring to this variable as consumption rate (i.e. number of crickets eaten per each gram of body mass). To express the toadlets’ foraging response to each disturbance with a single variable for each assay, we subtracted the consumption rate during the control trial from the consumption rate during the disturbance trial. This latter variable will be referred to as the change in consumption rate. One toadlet had to be excluded from the analysis of the ALAN test because the crickets hid under the paper towel on bottom of the box.

Locomotor performance is often key in animal fitness. To quantify the toadlets’ performance in a situation mimicking a chase by a human, we conducted a trial that is often used with amphibians (Goater et al., 1993; Sanabria et al., 2013; Zamora-Camacho, 2018; Hudson et al., 2020; Park and Do, 2023). The test was conducted two days following the noise pollution trial. The animals were fed crickets ad libitum before the test, and they had access to shelter as well. We made an oval-shaped arena (110 cm × 70 cm) out of a flexible PVC slab and drew a 5 cm × 5 cm rectangle on the bottom of the arena. To start the trial, we placed a toadlet into the rectangle, and gently prodded on the urostyle with a wooden stick to induce hopping. The trials were recorded using a Canon HD Legria HF R66 type camera (advanced zoom was 57×, optical zoom was 32×, and we utilized optical image stabilizer) into MP4 files. Each trial lasted a maximum duration of 6 minutes or ended if the toadlet stopped hopping when poked twice ( Figure 3 ). After each trial, we cleaned the bottom of the arena with 96% ethanol and waited 2 minutes for the alcohol to evaporate before proceeding to the next animal. The trials were performed between 1100h and 1500h in a random order of individuals.

We used Boris video analysis software (Friard and Gamba, 2016) to clip the images from the video recordings for measuring the hopping distances as follows. We selected the frames when the animal landed after each hop and saved them to PNG files. Then, by overlaying the photos ( Figure 3 ), we calculated the distance (in cm) between the consecutive landing spots using the ROI management tools (Dobretsov et al., 2017) of the ImageJ program. For each toadlet, we calculated the total number, total distance, and average distance of all hops, and the average distance of the first four hops (Goater et al., 1993). To control for individual differences in body size, we divided each of these variables with the last body mass measurement (taken at the end of the noise trial).

All analyses were run in R4.3.2 (R Core Team, 2023). To analyze how CORT release rates varied between the three consecutive GC profile samples (“sample type”) within occasion and between occasions (“age”), we used a linear mixed-effects (LME) model (function “lme” of package “nlme”). The dependent variable was log₁₀-transformed CORT release rate (18 values per individual), the explanatory variables were sample type and age and their interaction, and the random factors were individual identity nested in sibling group. We allowed for heterogeneous variances among the combinations of sample type and age by the compound symmetry autocorrelation structure (function “varIdent” of package “nlme”). Using a likelihood ratio test, we compared this model to a similar model without the sibling group variable to test if this random effect was significant. Because it was not (see Results), we omitted it. From the resulting model, we calculated pairwise comparisons (function “emmeans” from package “emmeans”) and corrected their p-values for multiple testing by the false-discovery rate (FDR) method (Pike, 2011).

To test the repeatability of the CORT profile, we estimated the intra-class correlation coefficient (ICC) using function “rpt” from package “rptR” (Nakagawa and Schielzeth, 2010). First, we quantified ICC for all CORT release rates, adjusted for the interaction between sample type and age. Then we calculated ICC for various aspects of the CORT profile as follows, adjusting to age (the number of sampling occasion, treated as a categorical variable). Within each sampling occasion, let us designate the three consecutive samples as B, D, and F, short for before, during, and following agitation. For tadpoles, we tested the repeatability of the following variables across the first 3 sampling occasions: baseline CORT release rate (B levels measured in the first hour; log-transformed), stressed CORT release rate (D levels measured in the second hour; log-transformed), recovery CORT release rate (F levels measured in the third hour; log-transformed), the magnitude of stress response quantified as 100 × (D − B)/B and transformed as square-root(x+65) following Bókony et al. (2021), and the rate of negative feedback quantified as 100 × (D − F)/D and transformed as (x+239)³ following Bókony et al. (2021). Because the shape of the CORT profile changed from the second to the third tadpole sampling occasion (see Results), we also tested the above repeatabilities across the first two sampling occasions only. For toadlets, we tested the repeatability of CORT release rates in the 3 samples (B, D, and F) similarly as in tadpoles. However, due to the unexpected shape of the CORT profile of toadlets (see Results) we could not quantify the magnitude of the stress response in toadlets, and we estimated the repeatability of their rate of negative feedback calculated as 100 × (B − average(D,F))/B (this variable had several outliers so it was rank-transformed). To test repeatability for the aspects of the CORT profile over all 6 sampling occasions, as stressed CORT release rates we used the second samples for tadpoles (D) and the first samples for toadlets (B), whereas as negative feedback we used the respective calculation for tadpoles and toadlets based on the shape of their CORT profiles, as explained above. We also tested the correlation between tadpole RMSSD and toadlet RMSSD using Pearson correlation. Note that we could not use the reaction-norm approach to quantify CORT profile repeatability because the reaction-norm shape was non-linear and changed with ontogeny; furthermore, our sample size would not have permitted good power for estimating repeatability of reaction-norm slopes (Martin et al., 2011; van de Pol, 2012).

To test the effects of ALAN and noise on foraging success, we used an LME model in which the dependent variable was consumption rate (4 values per individual), the explanatory variables were trial type (disturbance or control) and time of day (night for ALAN, day for noise) and their interaction, and the random factors were individual identity nested in sibling group. Because the random effect of sibling group was non-significant (p > 0.999) in a likelihood ratio test, we omitted it. Then, to test if any aspect of the CORT profile explained the changes in consumption, we added each CORT variable (i.e. those that were tested for repeatability) into the model, one at a time, in a three-way interaction with the other two explanatory variables. To test if the toadlets’ responses to ALAN and noise were correlated, we performed a Spearman rank-correlation test between the two foraging assays using the change in consumption rate (from control to disturbance) for each assay. To test if measures of fleeing performance were correlated with changes in consumption rate in the two foraging assays, and with CORT variables, we performed Spearman rank-correlation tests, and we corrected the p-values for multiple testing by the FDR method.

We report the results of the LME models with type-2 analysis-of-deviance tables (function “Anova” from package “car”). In all analyses, we used 95% confidence levels. However, when graphically presenting the results, we provide 84% confidence intervals (CI) to facilitate visual comparisons between groups, because non-overlap between two 84% CIs is equivalent to a significant difference between them (i.e. the 95% CI of their difference excludes zero).

Individual corticosterone release rates varied significantly with age and sampling type (sample type: χ² = 13.32, df = 2, p = 0.001; age: χ² = 1227.72, df = 5, p < 0.001, sample type × age: χ² = 112.7, df = 10, p < 0.001). At the same time, the random effect of sibling group was practically zero (likelihood-ratio test, p = 0.999). The pairwise comparisons revealed a significant stress response at all three tadpole sampling occasions ( Figure 4 ), as well as significant recovery following agitation at the first two (but not at the third) tadpole sampling occasions ( Figure 4 ). Also, the stressed CORT release rate of agitated tadpoles decreased from the second to the third sampling occasion ( Figure 4 ). In toadlets, however, CORT release rates at all sampling occasions were higher in the first sample (before agitation) than in the second and third samples (i.e. during and following agitation; Figure 4 ). Between occasions, CORT release rate of the first sample decreased from the second to the third toadlet sampling occasion ( Figure 4 ), and the CORT release rate of the third sample increased from the first to the third toadlet sampling occasion ( Figure 4 ).

All investigated aspects of the CORT profile had low repeatability that did not differ significantly from zero both within tadpoles and within toadlets, as well as when we included all six sampling occasions ( Table 1 ). There were two exceptions to this general trend. First, the repeatability of CORT release rates across all sampling occasions and all sample types was significantly higher than zero, although its value was still very low (ICC = 0.05; Table 1 ). Second, early tadpole stress response had relatively high repeatability (ICC = 0.30; Figure 5A ) compared to all other CORT variables, although estimation uncertainty was high and thus the p-value fell just above the traditional significance threshold ( Table 1 ). RMSSD over the tadpole stage did not correlate significantly with RMSSD over the toadlet stage (r = 0.28, p = 0.163, n = 26).

Consumption rate was significantly greater in the disturbance trials than in the control trials (main effect of trial type: χ² = 41.59, df = 1, p < 0.001; Figure 6 ) and smaller during the day than during the night (main effect of time of day: χ² = 61.2, df = 1, p < 0.001; Figure 6 ), but the interaction was not significant (trial type × time of day: χ² = 2.46, df = 1, p = 0.117). Thus, toadlets responded to both ALAN and daytime noise by a similar increase in consumption rate ( Figure 6 ). None of the CORT variables affected this response, as the three-way interaction between trial type, time of day, and each CORT variable was always non-significant (p > 0.117). The change in consumption rate from control to disturbance trials was correlated between the ALAN test and the noise test (r_s = 0.42, p = 0.034, n = 25; Figure 5B ).

All variables measured in the fleeing performance test were significantly correlated with each other (r_s > 0.50, p ≤ 0.010, n = 26), except for mean distance with hop count (r_s = 0.28, p = 0.170, n = 26) and with total time of hops (r_s =0.26, p = 0.200, n = 26). However, none of these variables correlated significantly with the change of consumption rate in response to ALAN (r ≤ 0.22, p > 0.696, n = 25) or noise (r ≤ 0.19, p > 0.962, n = 26). Hop performance variables were uncorrelated with all CORT variables, with two exceptions. Longer average distances over the first four hops correlated significantly with lower magnitude of stress response at the second tadpole sampling occasion (r_s = −0.42, p = 0.034, n = 26; Figure 5C ) and with higher HPI flexibility measured as RMSSD over the tadpole stage (r_s = 0.42, p = 0.033, n = 26; Figure 5D ). However, neither of these relationships were significant after FDR correction for multiple testing (p > 0.806).