Skip to main content

Biological and chemical quantification of tadpole nurseries (phytotelmata)

Cite this dataset

Fouilloux, Chloe et al. (2021). Biological and chemical quantification of tadpole nurseries (phytotelmata) [Dataset]. Dryad.


Many species of Neotropical frogs have evolved to deposit their tadpoles in small water bodies inside plant structures called phytotelmata. These pools are small enough to exclude large predators but have limited nutrients and high desiccation risk. Here, we explore phytotelm use by three common Neotropical species: Osteocephalus oophagus, an arboreal frog that periodically feeds eggs to its tadpoles; Dendrobates tinctorius, a tadpole-transporting poison frog with cannibalistic tadpoles; and Allobates femoralis, a terrestrial tadpole-transporting poison frog with omnivorous tadpoles. We found that D. tinctorius occupies pools across the chemical and vertical gradient, whereas A. femoralis and O. oophagus appear to have narrower deposition options that are restricted primarily by pool height, water capacity, alkalinity, and salinity. Dendrobates tinctorius tadpoles are particularly flexible, and can survive in a wide range of chemical, physical, and biological conditions, whereas O. oophagus seems to prefer small, clear pools and A. femoralis occupies medium-sized pools with abundant leaf litter and low salinity. Together, these results show the possible niche partitioning of phytotelmata among frogs, and provide insight into stressors and resilience of phytotelm breeders.


The study was carried out in the primary lowland terra-firme forest near the Camp Pararé at the CNRS Nouragues Ecological Research Station in the Nature Reserve Les Nouragues, French Guiana (4°02N, 52°41’W) over two field seasons: 1st February to 20th March 2019, and 30th January to 26th February 2020. The study area (approximately 0.2 km2) was chosen specifically because of the high abundance of D. tinctorius (Rojas and Pašukonis 2019). Pools were found with a combination of field methods. We opportunistically searched for pools targeting suitable microhabitats such as fallen trees, trees with buttresses and palm trees. In addition, pools were discovered by using tracking to follow D. tinctorius during previous studies (Pašukonis et al. 2019). We also used experimentally-induced tadpole transport in combination with tracking (Pašukonis et al. 2017) to find additional pools used by A. femoralis. Trees with high arboreal pools were discovered by locating calls produced by the treehole-breeding frogs Trachycephalus resinifictrix and T. hadroceps during night surveys.


Study species

Throughout the course of this work, three species formed the core of our data. D. tinctorius and A. femoralis are both small poison frogs belonging to the superfamily Dendrobatoidea. A. femoralis is a terrestrial frog whose adult males aggressively defend territories during the rainy season (Roithmair 1992, Narins et al. 2003), from which they carry recently hatched tadpoles to a variety of terrestrial pools including phytotelmata close to the ground (Ringler et al. 2009, 2013). Tadpoles of this species are omnivorous (McKeon and Summers 2013), but not cannibalistic (Summers and McKeon 2004). Similarly to A. femoralis, D. tinctorius males care for their offspring by carrying them to pools of water. Males of this species are adept climbers (depositing their tadpoles from the ground to more than 20 m in vertical height; Gaucher 2002, Rojas 2014, 2015), and their tadpoles are aggressive cannibals (Rojas 2014, Rojas and Pašukonis 2019).


Osteocephalus oophagus is a hylid treefrog with bi-parental care and obligately oophagous tadpoles (Jungfer and Weygoldt 1999, Almendáriz et al. 2000). As in our field site, adults have been found to call and breed in bromeliads, tree-holes, and palm axils close to the forest floor (Jungfer and Weygoldt 1999). Tadpoles of this species develop in the same pool in which the eggs are deposited.


Sampled pools

We exclusively considered phytotelmata throughout this study. Pools could be classified into two categories: dead substrates, which included holes in dead branches, fallen trees, and fallen Oenocarpus palm bracts; or live substrates which included live tree trunks, branches, roots, and buttresses. We did not sample bromeliads and non-phytotelm pools as these pools are not used by D. tinctorius. Based on the pools’ height and accessibility to different frog species, we termed the pools as “ground access”, “low arboreal” or “high arboreal” (Fig. 1, 2). Ground-access pools did not require vertical climbing ability to reach and included dead fallen structures as well as pools in live roots or low buttresses. Low arboreal pools were inside vertical structures low on the trunk or on high buttresses. High arboreal pools were high on the trunk or in canopy branches and were accessed for sampling using rope-based canopy access methods.  There was a clear vertical separation between ground-access and low-arboreal pools, which were all under 212 cm in height and between those and high arboreal pools, which were all above seven meters in height. In total, we sampled 84 unique pools across the 2019 and 2020 field seasons.


Several unique pools were sometimes found and sampled in the same tree. For all pools, we recorded the pool type, location (latitude/longitude), height from the ground to the pool edge, largest width and length parallel to the water surface, and the pool depth (maximum possible water depth of the phytotelmata) from the solid sediment bottom to the maximum water level line. Based on these measurements, we estimated the maximum water-holding capacity of each pool using the volume formula of a semi-ellipsoid as in Rojas (2014). Other sampling methods differed between the two field seasons.


2019 field season sampling

In 2019, we quantified physical measures (height, pool dimensions, leaf-litter volume), biotic measures (amphibian and invertebrate counts and diversity), as well as chemical measurements (see Appendix 2 for description of all variables measured). For pools accessible from the ground and smaller arboreal pools, we attempted to sample all tadpoles and Odonata larvae (predators of tadpoles; Caldwell 1993, Finke 1999, Summers and McKeon 2004) in each pool. Initially, we carefully observed the undisturbed pool and attempted to catch all tadpoles and Odonata larvae using a variety of tools. We then syphoned the entire volume of the water and sediment from the pool, emptied the leaf litter and searched for tadpoles and Odonata larvae. The volumes of water, sediment, and leaf litter were measured. For deep arboreal pools, we repeatedly netted and observed the pool until no more tadpoles were caught during five minutes of continuous netting. We carefully scraped the inner walls of the pools and caught as many Odonata larvae as possible. The leaf-litter volume could not be accurately measured for some deep arboreal pools, but they typically were protected from falling leaves and had little leaf litter in them.


We used visually apparent morphological traits to identify tadpoles, except for Allobates femoralis, A. granti, and Ameerega hahneli, which we could not reliably differentiate in the field. Because Allobates femoralis was more common in our study area than A. granti and Am. hahneli and we never observed A. granti and Am. hahneli directly at the pools we classified all A. femoralis-like tadpoles as such. Is it important to note that some A. granti and Am. hahneli tadpoles may have been misclassified as A. femoralis. However, this does not affect the interpretation of our results as all three species are cryptic terrestrial poison frogs similar in appearance, ecology and behavior. We also opportunistically recorded all species of adult frogs heard or seen at each pool throughout the sampling period. This was used as an amphibian diversity index between 0 and 8 species observed at each pool. Tadpoles of only three out of seven recorded species, namely D. tinctorius, O. oophagus and A. femoralis, were detected in pools with sufficient frequency for further analysis (N = 34 (2019), N = 7 and N = 10 pools, respectively).


Sampled invertebrates were counted, photographed, and classified only to a group level (usually order or class) apparent in the field. To estimate the predation pressure on tadpoles we used the total count and average size of all Odonata larvae detected in the procedure described above. To estimate density and diversity of aquatic invertebrates, we carefully searched and counted invertebrates in a sample of up to 10 liters of water and up to one liter of sediment in proportion to the total estimated pool volume. For each liter of the water volume sampled, we sampled ~ 100 mL of sediment from the bottom of the pool. When the water volume was less than one liter or the amount of sediment was less than 100 mL, we sampled the entire pool and recorded the exact volumes. In the final analysis, we used the invertebrate density (count divided by the volume sampled) and the diversity index corresponding to our classification (between 0 and 12). The following 12 categories were used to quantify invertebrate diversity: Odonata Zygoptera larvae, Odonata Anisoptera larvae, surface Coleoptera adults, diving Coleoptera adults, Coleoptera Scirtidae larvae, Trichoptera larvae, Diptera Culicidae larvae, Diptera Chironomidae larvae, Diptera Tipulidae larvae, other Diptera larvae, small red Annelida, other unidentified larvae. All water, sediment, tadpoles and invertebrates were released back into the pool after sampling.


We measured water conductivity, salinity and total dissolved solids (TDS), dissolved oxygen and temperature with electronic sensors (EZDO 7200 and pHenomenal OX4110H). Water chemistry (KH (also known as alkalinity), hardness and NO3) was recorded using aquarium water testing strips (JBL EasyTest). All measures were taken from the undisturbed surface water of the pool.


2020 field season sampling

The 2020 dataset focused solely on D. tinctorius tadpole counts and pH measurements of weekly resampled ground access phytotelmata (N = 26) over the time period of a month (February 2020). Rainfall data were provided by the Nouragues Ecological Research Station  from an above-canopy weather station in the study area. High arboreal pools (N = 8 (2020)) were only measured once. pH was recorded using a pH meter (AMTAST Waterproof pH Meter). The pH meter was calibrated once per day, prior to pool sampling, using both acidic (pH = 4) and neutral (pH = 7) calibration solutions. The pH of ground access pools was taken directly by submerging the pH probe into the pool, and the measurement was recorded once read-out stabilized. For arboreal pools, a sample of water was collected using a syringe, which was then sealed at both ends. Once on the ground, one end of the syringe was opened, and the pH was measured by submerging the pH probe into the syringe. Syringes were never reused. Between pool sampling, the pH probe was wiped with a clean cloth and rinsed with aquifer water.


Statistical Analyses

The presence of D. tinctorius in pools was analyzed using 2019 field data. As a result of the high collinearity between variables in the 2019 dataset (see Supp Fig 1), we used a principal components regression to analyze phytotelm ecology data. We first checked data for a non-random structure following Björklund (2019); then, we established that the correlation matrices were significantly different from random (? = 10.22, p = 0; ϕ= 0.238, p < 0.001) to ensure they were suitable for a PCA. Based on each PCs difference from random matrices, we selected the first three principal components as predictors of probability for D. tinctorius tadpole presence as a binomial response in the principal component regression (PC1-3 explained about 53% of the variability of the data (where PC1 = 0.24 ± 0.48, PC2 = 0.17 ± 0.40, PC3 = 0.11 ± 0.33 (variance explained ± SE )). We evaluated the fit of negative binomial GLM models based on second-order AIC ranks (AICc) using the package AICcmodavg (Mazerolle 2020) which are specialized for smaller sample sizes (Akaike 1974; see Supp. Table 1). Models within two AIC scores of each other were further evaluated by assessing the significance of interactions between model terms.


To better understand which variables contributed significantly to each principal component, we calculated which variables had index loadings larger than random data. Following the methods outlined by Björklund (2019) and Vieira (2012), we randomized the data and calculated new correlation matrices which we permuted 1,000 times. We then compared the index of loadings (ILij= uij2 × λi2, see Vieira (2012) for details) between each observed PC and the randomly generated data to see which variables contributed significantly to each principal component.


The 2020 dataset consisted of weekly resampled pools throughout February 2020. Models took repeated measures of pool ID into account as a random effect. Both the presence of D. tinctorius tadpoles (count; negative binomial family) and pH (Gaussian family) from resampled pools were modeled using a mixed effects generalized linear model in the package “glmmTMB” (Magnusson et al. 2020). Predictor structure for both pH and D. tinctorius models were built based on biologically relevant variables (pool substrate, time, D. tinctorius tadpole count (for pH model), water capacity, surface area:depth ratio). Using these variables, models were first fit with relevant interactions (see Supp. Table 2, 3), which were then removed if they did not contribute significantly to the model using single term deletions (using base R function, drop1; Zuur et al. 2009).  Residuals were diagnosed using the package “DHARMa” (Hartig 2020). Final models were then checked for overdispersion and zero-inflation (using DHARMa); none of the final models required correction. All code was done in R (Core 2015).

Usage notes

Please refer to ReadMe file attached in the file uploads for variable descriptions.


Investissement d’Avenir funds of the ANR , Award: AnaEE France ANR-11-INBS-0001; Labex CEBA ANR-10-LABX-25-01

Academy of Finland, Award: Project No.21000042021

European Research Council, Award: Marie Sklodowska-Curie grant agreement No 835530

National Science Foundation, Award: IOS-1845651: Stanford University

Investissement d’Avenir funds of the ANR, Award: AnaEE France ANR-11-INBS-0001; Labex CEBA ANR-10-LABX-25-01