Abstract
Environmental DNA (eDNA) metabarcoding is revolutionising biodiversity monitoring, but has unrealised potential for ecological hypothesis testing. Here, we validate this potential in a large-scale analysis of vertebrate community data generated by eDNA metabarcoding of 532 UK ponds. We test biotic associations between the threatened great crested newt (Triturus cristatus) and other vertebrates as well as abiotic factors influencing T. cristatus occupancy at the pondscape. Furthermore, we test the status of T. cristatus as an umbrella species for pond conservation by assessing whether vertebrate species richness is greater in ponds with T. cristatus and higher T. cristatus Habitat Suitability Index (HSI) scores. T. cristatus occupancy was positively correlated with amphibian and waterfowl species richness. Specifically, T. cristatus was positively associated with smooth newt (Lissotriton vulgaris), common coot (Fulica atra), and common moorhen (Gallinula chloropus), but negatively associated with common toad (Bufo bufo). T. cristatus occupancy did not significantly decrease as fish species richness increased, but negative associations with common carp (Cyprinus carpio), three-spined stickleback (Gasterosteus aculeatus) and ninespine stickleback (Pungitius pungitius) were identified. T. cristatus occupancy was negatively correlated with mammal species richness, and T. cristatus was negatively associated with grey squirrel (Sciurus carolinensis). T. cristatus occupancy was negatively influenced by larger pond area, presence of inflow, and higher percentage of shading, but positively correlated with HSI score, supporting its application to T. cristatus survey. Vertebrate species richness was significantly higher in T. cristatus ponds and broadly increased as T. cristatus HSI scores increased. We reaffirm reported associations (e.g. T. cristatus preference for smaller ponds) but also provide novel insights, including a negative effect of pond inflow on T. cristatus. Our findings demonstrate the prospects of eDNA metabarcoding for ecological hypothesis testing at landscape scale and dramatic enhancement of freshwater conservation, management, monitoring and research.
1. Introduction
Environmental DNA (eDNA) analysis offers ecologists exceptional power to detect organisms within and across ecosystems. DNA released by organisms into their environment via secretions, excretions, gametes, blood, or decomposition, can be sampled and analysed using different approaches to reveal the distribution of single or multiple species (Rees et al., 2014; Lawson Handley, 2015). eDNA analysis combined with high-throughput sequencing (i.e. eDNA metabarcoding) can yield efficient, comprehensive assessments of entire communities (Deiner et al., 2017), providing a step change in biodiversity monitoring (Hering et al., 2018). eDNA metabarcoding has untapped potential to test ecological hypotheses by enabling biodiversity monitoring at landscape scale with minimal impact to communities under investigation. This potential has already been demonstrated with targeted eDNA analysis by Wilcox et al. (2018), where climate-mediated responses of bull trout (Salvelinus confluentus) to biotic and abiotic factors were revealed using quantitative PCR (qPCR) on crowd-sourced eDNA samples. Although eDNA metabarcoding assessments of alpha and beta diversity along environmental gradients are increasing (e.g. Hänfling et al., 2016; Olds et al., 2016; Kelly et al., 2016; Evans et al., 2017; Li et al., 2018a; Nakagawa et al., 2018), this tool is less commonly used for ecological hypothesis testing, such as the impact of environmental stressors (Li et al., 2018b; Macher et al., 2018).
Aquatic ecosystems are highly suited to eDNA studies as eDNA exists in multiple states with rapid modes of transport and degradation, increasing detectability of contemporary biodiversity (Rees et al., 2014; Barnes & Turner, 2015). Lentic systems provide further opportunities for eDNA research, being discrete water bodies with variable physicochemical properties that do not experience flow dynamics (Harper et al., 2019). Ponds in particular have enormous biodiversity and experimental virtue that has not been maximised in previous eDNA metabarcoding assessments of this habitat (Valentini et al., 2016; Evans et al., 2017; Klymus et al., 2017; Ushio et al., 2017; Bálint et al., 2018). These small and abundant water bodies span broad ecological gradients (De Meester et al., 2005) and comprise pondscapes - a network of ponds and their surrounding terrestrial habitat (Hill et al., 2018). Pondscapes contribute substantially to aquatic and non-aquatic biodiversity across spatial scales, with ponds supporting many rare and protected species in fragmented landscapes (De Meester et al., 2005; Biggs et al., 2016; Hill et al., 2018). Consequently, ponds are model systems for experimental validation and examination of biogeographical patterns (De Meester et al., 2005). Habitat complexity and tools required for different taxa with associated bias (Evans et al., 2017) and cost (Valentini et al., 2016) once hindered exhaustive sampling of pond biodiversity (Hill et al., 2018), but eDNA metabarcoding may overcome these barriers (Harper et al., 2019).
In the UK, the threatened great crested newt (Triturus cristatus) is an umbrella species for pond conservation. The extensive literature on T. cristatus ecology provides an excellent opportunity to validate ecological patterns revealed by eDNA metabarcoding. Both biotic (e.g. breeding substrate, prey, and predators) and abiotic (e.g. pond area, depth, and temperature) factors are known to influence T. cristatus breeding success (Langton, Beckett & Foster, 2001). The T. cristatus Habitat Suitability Index (HSI [Oldham et al., 2000; ARG-UK, 2010]) accounts for these factors using 10 suitability indices that are scored and combined to calculate a decimal score between 0 and 1 (where 1 = excellent habitat). Larvae are susceptible to fish and waterfowl predation (Edgar & Bird, 2006; Rannap & Briggs, 2006; Skei et al., 2006; Hartel, Nemes & Oellerer, 2010), and adults reportedly avoid ponds containing three-spined stickleback (Gasterosteus aculeatus) (McLee & Scaife, 1992), ninespine stickleback (Pungitius pungitius), crucian carp (Carassius carassius), and common carp (Carassius carpio) (Rannap, Lõhmus & Briggs, 2009a, b). Conversely, T. cristatus and smooth newt (Lissotriton vulgaris) prefer similar habitat and often co-occur (Rannap & Briggs, 2006; Skei et al., 2006; Rannap et al., 2009a; Denoël et al., 2013; Cayuela et al., 2018). T. cristatus individuals thrive in ponds with good water quality as indicated by diverse macroinvertebrate communities (Oldham et al., 2000; Rannap et al., 2009a), and water clarity is important for breeding displays, foraging success, and egg survival (Rannap & Briggs, 2006; Skei et al., 2006). Pond networks encourage T. cristatus occupancy (Joly et al., 2001; Rannap et al., 2009a; Hartel et al., 2010; Denoël et al., 2013), but larger pond area discourages presence (Joly et al., 2001). Ponds with heavy shading (Vuorio, Heikkinen & Tikkanen, 2013) or dense macrophyte cover (Rannap & Briggs, 2006; Skei et al., 2006; Hartel et al., 2010) are unlikely to support viable populations. T. cristatus individuals also depend on terrestrial habitat, preferring open, semi-rural pondscapes (Denoël et al., 2013) containing pasture, extensively grazed and rough grassland, scrub, and coniferous and deciduous woodland (Oldham et al., 2000; Rannap & Briggs, 2006; Rannap et al., 2009a; Gustafson, Malmgren & Mikusiński, 2011; Vuorio et al., 2013).
We assessed vertebrate communities at the pondscape using a dataset generated by eDNA metabarcoding for over 500 ponds with comprehensive environmental metadata. We validated eDNA metabarcoding as a tool for ecological hypothesis testing, and compared its outputs to previous results generated by established methods. Specifically, we tested biotic (community presence-absence data) and abiotic (environmental metadata on ponds and surrounding terrestrial habitat) determinants of T. cristatus at the pondscape - an impractical task by conventional means. Furthermore, we tested the applicability of the HSI to predict eDNA-based T. cristatus occupancy. Finally, we assessed the umbrella species status of T. cristatus by investigating whether T. cristatus presence and the T. cristatus HSI score can predict vertebrate species richness of ponds.
2. Materials and methods
2.1 Samples
We repurposed the taxonomically assigned sequence reads from Harper et al. (2018) that were produced using eDNA metabarcoding of pond water to compare qPCR and eDNA metabarcoding for T. cristatus detection. Samples from 508 ponds included in Natural England’s Great Crested Newt Evidence Enhancement Programme were processed using eDNA metabarcoding alongside 24 privately surveyed ponds. Water samples were collected using established methodology (Biggs et al., 2015), detailed in Supporting Information: Appendix 1. Briefly, 20 × 30 mL water samples were collected from each pond and pooled. Six 15 mL subsamples were taken from the pooled sample and each added to 33.5 mL absolute ethanol and 1.5 mL sodium acetate 3 M (pH 5.2). Subsamples were pooled during DNA extraction to produce one eDNA sample per pond. Targeted qPCR detected T. cristatus in 265 (49.81%) ponds (Harper et al., 2018).
Environmental metadata (Table S1) were collected for 504 of 532 ponds (Fig. S1) by environmental consultants contracted for Natural England’s Great Crested Newt Evidence Enhancement Programme. Metadata included: maximum depth of ponds; pond circumference; pond width; pond length; pond area; pond density (i.e. number of ponds per km2); terrestrial overhang; shading; macrophyte cover; HSI score (Oldham et al., 2000); HSI band (categorical classification of HSI score [ARG-UK, 2010]); pond permanence; water quality; pond substrate; presence of inflow or outflow; presence of pollution; presence of other amphibians, fish and waterfowl; woodland; rough grass; scrub/hedge; ruderals; other terrestrial habitat (i.e. good quality terrestrial habitat that did not conform to aforementioned habitat types); and overall terrestrial habitat quality.
2.2 DNA reference database construction
A custom, phylogenetically curated reference database of mitochondrial 12S ribosomal RNA (rRNA) sequences for UK fish species was previously constructed for eDNA metabarcoding of lake fish communities (Hänfling et al., 2016). Harper et al. (2018) constructed additional reference databases for UK amphibians, reptiles, birds, and mammals (Supporting Information: Appendix 1). Reference sequences available for species varied across vertebrate groups: amphibians 100.00% (N = 21), reptiles 90.00% (N = 20), mammals 83.93% (N = 112), and birds 55.88% (N = 621). Table S2 lists species without database representation, i.e. no records for any species in a genus. Sanger sequences were obtained from tissue of T. cristatus, L. vulgaris, Alpine newt (Ichthyosaura alpestris), common toad (Bufo bufo), and common frog (Rana temporaria) to supplement the amphibian database (Supporting Information: Appendix 1). The complete reference databases compiled in GenBank format were deposited in a GitHub repository and permanently archived (https://doi.org/10.5281/zenodo.1188710) by Harper et al. (2018).
2.3 Primer validation
Reference databases were combined for in silico validation of published 12S rRNA primers 12S-V5-F (5’-ACTGGGATTAGATACCCC-3’) and 12S-V5-R (5’-TAGAACAGGCTCCTCTAG-3’) (Riaz et al., 2011) using ecoPCR software (Ficetola et al., 2010). Set parameters allowed a 50-250 bp fragment and three mismatches between each primer and reference sequence. Primers were validated in vitro for UK fish by Hänfling et al. (2016) and by Harper et al. (2018) for six UK amphibian species (Fig. S2).
2.4 eDNA metabarcoding
We used the taxonomically assigned sequence reads generated with vertebrate eDNA metabarcoding by Harper et al. (2018). The eDNA metabarcoding workflow is fully described in Harper et al. (2018) and Supporting Information: Appendix 1. eDNA was first amplified with the aforementioned primers, where PCR positive controls (six per PCR plate; n = 114) were cichlid (Rhamphochromis esox) DNA (0.284 ng/µL) and PCR negative controls (six per PCR plate; n = 114) were sterile molecular grade water (Fisher Scientific UK Ltd, UK). PCR products were individually purified using E.Z.N.A® Cycle Pure V-Spin Clean-Up Kits (Omega Bio-tek, GA, USA) following the manufacturer’s protocol. The second PCR bound Multiplex Identification tags to the purified products. PCR products were individually purified using magnetic bead clean-up and quantified with a Quant-IT™ PicoGreen™ dsDNA Assay (Invitrogen, UK). Samples were normalised, pooled, and libraries quantified using a Qubit™ dsDNA HS Assay (Invitrogen, UK). Libraries were sequenced on an Illumina® MiSeq using 2 × 300 bp V3 chemistry (Illumina, Inc, CA, USA) and raw sequence reads processed using metaBEAT (metaBarcoding and Environmental Analysis Tool) v0.97.7 (https://github.com/HullUni-bioinformatics/metaBEAT). After quality filtering, trimming, merging, chimera detection, and clustering, non-redundant query sequences were compared against our reference database using BLAST (Zhang et al., 2000). Putative taxonomic identity was assigned using a lowest common ancestor (LCA) approach based on the top 10% BLAST matches for any query matching with at least 98% identity to a reference sequence across more than 80% of its length. Unassigned sequences were subjected to a separate BLAST against the complete NCBI nucleotide (nt) database at 98% identity to determine the source via LCA as described above. The bioinformatic analysis was archived (https://doi.org/10.5281/zenodo.1188710) by Harper et al. (2018) for reproducibility.
2.5 Data analysis
Analyses were performed in R v.3.4.3 (R Core Team, 2017). Data and R scripts have been deposited in a dedicated GitHub repository for this study, which has been permanently archived at: https://doi.org/10.5281/zenodo.2634427. Assignments from different databases were merged, and spurious assignments (i.e. non-UK species, invertebrates and bacteria) removed from the dataset. The family Cichlidae was reassigned to Rhamphochromis esox. The green-winged teal (Anas carolinenisis) was reassigned to Anas (Dabbling ducks) because this species is a rare migrant and available reference sequences were identical across our metabarcode to those for mallard (Anas platyrhynchos) and Eurasian teal (Anas crecca), which are widely distributed across the UK. Scottish wildcat (Felis silvestris) does not occur at the sampling localities (Kent, Lincolnshire and Cheshire) and was therefore reassigned to domestic cat (Felis catus). Wild boar (Sus scrofa) and grey wolf (Canis lupus) were reassigned to domestic pig (Sus scrofa domesticus) and domestic dog (Canis lupus familiaris) given the restricted distribution of S. scrofa and absence of C. lupus in the UK. The genus Strix was reassigned to tawny owl (Strix aluco) as it is the only UK representative of this genus. Where family and genera assignments containing a single UK representative had reads assigned to species, reads from all assignment levels were merged and manually assigned to that species.
Of the 114 PCR negative controls included, 50 produced no reads (Fig. S3). Reads generated for 64 of 114 PCR negative controls ranged from 0 to 49227, and strength of each contaminant varied (mean = 0.021%, range = 0 - 100.0% of the total reads per PCR negative control). To minimise risk of false positives, we evaluated different sequence thresholds. These included the maximum sequence frequency of R. esox DNA in eDNA samples (100%), maximum sequence frequency of any DNA except R. esox in PCR positive controls (83.96%), and taxon-specific thresholds (maximum sequence frequency of each taxon in PCR positive controls). The different thresholds were applied to the eDNA samples and the results from each compared (Fig. S4). The taxon-specific thresholds (Table S3) retained the most biological information, thus these were selected for downstream analysis. Consequently, taxa were only classed as present at sites if their sequence frequency exceeded their threshold. After applying the taxon-specific thresholds, our PCR positive control (R. esox), human (Homo sapiens), and domestic species (Table S4) were removed from the dataset. Higher taxonomic assignments excluding the genus Anas were then removed, thus taxonomic assignments in the final dataset were predominantly of species resolution.
The read count data were converted to a species presence-absence matrix. Hypotheses tested relating to biotic and abiotic determinants of T. cristatus occupancy as well as the umbrella status of T. cristatus are summarised in Table 1. We employed Generalized Linear Mixed-effects Models (GLMMs) using the package lme4 v1.1-12 (Bates et al., 2015) for hypothesis testing. First, we investigated the influence of vertebrate group species richness on T. cristatus occupancy using a binomial GLMM with the logit link function that included species richness of other amphibians, fish, waterfowl, terrestrial birds, and mammals as fixed effects and pond as a random effect (N = 532). We performed a preliminary analysis using the package cooccur v1.3 (Griffith, Veech & Marsh, 2016) to identify species associations between T. cristatus and other vertebrates (N = 532). Identified associations in conjunction with the existing T. cristatus literature informed candidate biotic variables to be modelled against T. cristatus occupancy (n = 504). The existing T. cristatus literature informed candidate abiotic variables to be modelled against T. cristatus occupancy (n = 504).
Selection of a suitable set of explanatory variables and modelling framework is fully described in Supporting Information: Appendix 1. Briefly, candidate biotic and abiotic explanatory variables were assessed for collinearity, relative importance, and non-linearity. We constructed separate binomial GLMMs with the logit link function for biotic and abiotic explanatory variables that included sample as a random effect. We modelled HSI score (fixed effect) and sample (random effect) separately to prevent HSI score masking variation caused by the individual biotic and abiotic variables it encompasses. Using a Poisson GLMM with sample as a random effect, we tested the umbrella species status of T. cristatus by modelling vertebrate species richness against T. cristatus presence-absence and the T. cristatus HSI score.
For each GLMM, we employed an information-theoretic approach using Akaike Information Criterion (AIC) to determine the most parsimonious approximating model to make predictions (Akaike, 1973). Biotic and abiotic models considered respectively were nested thus the best models were chosen using stepwise backward deletion of terms based on Likelihood Ratio Tests (LRTs). The HSI score and vertebrate species richness models were compared to null GLMMs. Final models were tested for overdispersion using a custom function testing overdispersion of the Pearson residuals. Model fit was assessed using the Hosmer and Lemeshow Goodness of Fit Test within the package ResourceSelection v0.2-4 (Lele, Keim & Solymos, 2016), quantile-quantile plots, and partial residual plots (Zuur et al., 2009). Model predictions were obtained using the predictSE function in the package AICcmodavg v2.0-3 (Mazerolle, 2016) and upper and lower 95% CIs were calculated from the standard error of the predictions. Results were plotted using the package ggplot2 v2.1.0 (Wickham, 2016).
3. Results
3.1 eDNA metabarcoding
A total of 532 eDNA samples and 228 PCR controls were processed across two sequencing runs. The runs generated raw sequence read counts of 36,236,862 and 32,900,914 respectively. After quality filtering, trimming, and merging of paired-end reads, 26,294,906 and 26,451,564 sequences remained. Following removal of chimeras and redundancy via clustering, the libraries contained 14,141,237 and 14,081,939 sequences (average read counts of 36,826 and 36,671 per sample respectively), of which 13,126,148 and 13,113,143 sequences were taxonomically assigned. The final dataset (assignments corrected, thresholds applied, and assignments removed) contained 53 vertebrate species (Table S5), including six amphibians, 14 fish, 17 birds, and 16 mammals (Fig. 1).
3.2 Pondscape biodiversity
All native amphibians were found as well as the non-native marsh frog (Pelophylax ridibundus). T. cristatus (n = 148), L. vulgaris (n = 151) and R. temporaria (n = 122) were widespread, but B. bufo (n = 42), palmate newt (Lissotriton helveticus [n = 5]) and P. ridibundus were uncommon (n = 1). The threatened European eel (Anguilla anguilla [n = 15]), European bullhead (Cottus gobio [n = 14]), and C. carassius (n = 2) were detected alongside native fishes, such as pike (Esox Lucius [n = 17]) and roach (Rutilus rutilus [n = 71]), but also introduced species, including C. carpio (n = 40), ruffe (Gymnocephalus cernua [n = 1]), and rainbow trout (Oncorhynchus mykiss [n = 3]). Some identified waterfowl were ubiquitous, such as common moorhen (Gallinula chloropus [n = 211]), whereas others were less common, e.g. grey heron (Ardea cinerea [n = 1]) and Eurasian oystercatcher (Haematopus ostralegus [n = 1]). Terrestrial fauna were often detected in fewer than five ponds (Figs. 1c, d). Buzzard (Buteo buteo [n = 4]), Eurasian jay (Garrulus glandarius [n = 7]), dunnock (Prunella modularis [n = 4]), and starling (Sturnus vulgaris [n = 4]) were the most frequently detected terrestrial birds. Introduced mammals (Mathews et al., 2018), such as grey squirrel (Sciurus carolinensis [n = 57]) and Reeve’s muntjac (Muntiacus reevesi [n = 3]), outweighed native mammals. Nonetheless, we detected several mammals with Biodiversity Actions Plans and/or of conservation concern (Mathews et al., 2018), including otter (Lutra lutra [n = 1]), water vole (Arvicola amphibious [n = 16]), European polecat (Mustela putorius [n = 1]), brown hare (Lepus europaeus [n = 1]) and water shrew (Neomys fodiens [n = 8]). Notably, the invasive American mink (Neovison vison) was absent despite widespread UK distribution (Mathews et al., 2018). All species and their detection frequencies are listed in Table S5.
3.3 Biotic determinants of T. cristatus occupancy
T. cristatus occupancy was positively influenced by amphibian and waterfowl species richness, yet negatively influenced by mammal species richness. T. cristatus occupancy was reduced as fish and terrestrial bird species richness increased, but these trends were not significant (Fig. 2; GLMM: overdispersion θ = 0.994, χ2525 = 521.778, P = 0.532; fit χ28 = 48.537, P < 0.001, R2 = 10.91%). T. cristatus had significant (P < 0.05) positive associations with three species (Fig. S5), including L. vulgaris, common coot (Fulica atra), and G. chloropus. However, T. cristatus had significant (P < 0.05) negative associations with six species (Fig. S5), including B. bufo, C. carpio, G. aculeatus, P. pungitius, common pheasant (Phasianus colchicus), and S. carolinensis. Only presence-absence of L. vulgaris, B. bufo, C. carpio, G. aculeatus, and G. chloropus were retained by model selection as explanatory variables for the biotic GLMM of T. cristatus occupancy (Figs. 3a-e; GLMM: overdispersion θ = 1.001, χ2495 = 495.297, P = 0.488; fit χ28 = 101.820, P < 0.001, R2 = 27.80%). Waterfowl presence-absence was also retained in the biotic GLMM (Fig. 3f). Results of analyses are summarised and compared to previously reported determinants in Table 1.
T. cristatus individuals were more likely to occupy ponds with more amphibian species (Fig. 2a). T. cristatus was detected in 51.66% of ponds (n = 151) containing L. vulgaris, but in only 11.91% of ponds (n = 42) with B. bufo (Fig. 1a). Probability of T. cristatus occupancy was lower in ponds with more fish species, and T. cristatus was absent from ponds with more than four fish species (Fig. 2b). T. cristatus was only found in 15.00% (n = 40), 14.55% (n = 55) and 6.67% (n = 15) of ponds inhabited by C. carpio, G. aculeatus and P. pungitius respectively (Fig. 1b). In contrast, T. cristatus individuals were more likely to occur in ponds with more waterfowl species (Fig. 2c). T. cristatus occupied 41.67% (n = 48) and 36.02% (n = 211) of ponds with F. atra and G. chloropus respectively (Fig. 1c). T. cristatus occupancy was negatively influenced by higher terrestrial bird species richness, but not significantly so (Fig. 2d). However, T. cristatus was negatively associated with P. colchicus, being found in only 12.00% (n = 25) ponds with P. colchicus records (Fig. 1c). T. cristatus were less likely to occupy ponds with more mammal species (Figs. 2e). Specifically, T. cristatus was negatively associated with S. carolinensis and found in only 15.79% (n = 57) of ponds with S. carolinensis records (Fig. 1d).
3.4 Abiotic determinants of T. cristatus occupancy
Three explanatory variables were retained in the abiotic GLMM explaining T. cristatus occupancy (GLMM: overdispersion θ = 1.004, χ2499 = 500.995, P = 0.467; fit χ28 = 14.409, P = 0.072, R2 = 9.73%). The probability of T. cristatus occupancy decreased in ponds with inflow present, larger area, and higher percentage of shading (Table 1, Figs. 3g-i).
3.5 T. cristatus HSI and umbrella status
HSI score positively correlated with T. cristatus occupancy (GLMM: overdispersion θ = 1.012, χ2501 = 506.893, P = 0.418; fit χ28 = 7.270, P = 0.508, R2 = 5.45%), where T. cristatus individuals were more likely to occupy ponds with a higher HSI score (Table 1, Fig. 3j). Vertebrate species richness was positively associated with T. cristatus occupancy (GLMM: overdispersion θ = 0.977, χ2500 = 488.687, P = 0.633; fit χ28 = −158.03, P = 1.000, R2 = 13.63%), with more species detected in ponds occupied by T. cristatus (Fig. 4a; χ21 = 40.985, P < 0.001). However, vertebrate species richness did not significantly increase with the T. cristatus HSI score (Fig. 4b; χ21 = 1.207, P = 0.272).
4. Discussion
We have validated eDNA metabarcoding for ecological hypothesis testing using the community data generated by this tool in combination with environmental metadata for ponds. We tested biotic and abiotic determinants of T. cristatus occupancy, whether the HSI can be applied to T. cristatus eDNA survey, and whether T. cristatus is truly an umbrella species for pond conservation. T. cristatus occupancy was higher in ponds containing L. vulgaris and G. chloropus, and ponds without B. bufo, C. carpio, and G. aculeatus. T. cristatus individuals were also more likely to occupy ponds where waterfowl occurred. Ponds inhabited by T. cristatus were typically small, absent of inflow, and not excessively shaded. The T. cristatus HSI was appropriate for predicting T. cristatus occupancy, but not vertebrate species richness. Nonetheless, more vertebrates were present in ponds occupied by T. cristatus thus presence of this amphibian may indicate good quality habitat for other vertebrates. Our findings demonstrate the power of eDNA metabarcoding to enhance freshwater monitoring and research by providing biodiversity data en masse at low cost.
4.1 Pondscape biodiversity
eDNA metabarcoding detected six amphibian, 14 fish, 17 bird, and 16 mammal species across 532 UK ponds. This diverse species inventory emphasises the importance of ponds as habitat for aquatic taxa, but also as stepping stones for semi-aquatic and terrestrial taxa (De Meester et al., 2005; Hill et al., 2018) through provision of drinking, foraging, dispersive, and reproductive opportunities (Biggs et al., 2016; Klymus et al., 2017). Some species detections may be the result of eDNA transport from water bodies in the surrounding area (Hänfling et al., 2016) to ponds via inflow. However, this signifies the capacity of ponds to provide natural samples of freshwater and terrestrial biodiversity in the wider catchment (Deiner et al., 2017).
4.2 Biotic determinants of T. cristatus occupancy
T. cristatus were more likely to occupy ponds with higher amphibian species richness - particularly ponds containing L. vulgaris and absent of B. bufo. T. cristatus and L. vulgaris have similar habitat requirements and tend to breed in the same ponds (Skei et al., 2006; Rannap et al., 2009a; Denoël et al., 2013; Cayuela et al., 2018), with >60% overlap reported (Rannap & Briggs, 2006). However, L. vulgaris can inhabit a broader range of habitat (Rannap & Briggs, 2006; Skei et al., 2006) than T. cristatus, which depends on larger, deeper ponds with abundant macrophytes and no fish located in open, semi-rural landscapes (Denoël et al., 2013). B. bufo can inhabit fish-containing ponds (Manenti & Pennati, 2016) and T. cristatus may predate B. bufo eggs and larvae (Langton et al., 2001). This may explain the negative association between B. bufo and T. cristatus as opposed to the positively associated T. cristatus and L. vulgaris.
T. cristatus occupancy marginally decreased with higher fish species richness, and T. cristatus was negatively associated with C. carpio, G. aculeatus, and P. pungitius. These fishes are common in and typical of ponds. All T. cristatus life stages may be predated by fishes (Langton et al., 2001) and negative effects of fish presence-absence on T. cristatus occupancy, distribution, and abundance are repeatedly reported (Joly et al., 2001; Rannap & Briggs, 2006; Skei et al., 2006; Denoël & Ficetola, 2008; Rannap et al., 2009a, b; Hartel et al., 2010; Denoël et al., 2013). G. aculeatus predates T. cristatus eggs and larvae (McLee & Scaife, 1992; Jarvis, 2010), and has non-consumptive effects on T. cristatus embryos (Jarvis, 2010). T. cristatus larvae were also found to alter their behaviour when exposed to predatory G. aculeatus but not non-predatory C. carassius (Jarvis, 2012), another fish characteristic of ponds.
In our study, we detected T. cristatus in 50% of ponds inhabited by C. carassius, but <20% of ponds containing large and/or predatory fishes, e.g. C. carpio, G. aculeatus, E. lucius. Although fewer ponds contained C. carassius than C. carpio, G. aculeatus or E. lucius, previous research also indicates large and/or predatory fish are more detrimental to T. cristatus occurrence (Skei et al., 2006; Hartel et al., 2010; Chan, 2011). C. carassius does not hinder T. cristatus oviposition, larval behaviour, or recruitment success (Chan, 2011; Jarvis, 2012), or pond invertebrate and macrophyte diversity (Stefanoudis et al., 2017). In contrast, C. carpio foraging reduces invertebrate density and macrophyte cover (Maceda-Veiga, López & Green, 2017), which lowers T. cristatus reproductive and foraging success and heightens predator exposure (Rannap & Briggs, 2006; Gustafson et al., 2006; Chan, 2011). C. carassius and C. carpio are both included among fish species assumed to negatively impact T. cristatus and whose presence-absence is assessed for the T. cristatus HSI (ARG-UK, 2010). However, it is evident that C. carassius does not directly predate T. cristatus or indirectly alter its behaviour, reproductive success, or habitat. Therefore, we advocate a systematic re-evaluation of problematic fish species for T. cristatus conservation.
T. cristatus was positively associated with waterfowl species richness, namely presence of F. atra and G. chloropus. These waterfowl species share macrophytes and macroinvertebrates as resources with amphibians, feeding on both directly (Perrow et al., 1997; Paillisson & Marion, 2001; Wallau et al., 2010). F. atra and G. chloropus crop emergent macrophytes to search for invertebrate prey (Paillisson & Marion, 2001; Wallau et al., 2010), which may indirectly benefit T. cristatus foraging. Although Fulica spp. can also pull up submerged vegetation and damage vegetation banks (Lauridsen, Jeppesen & Andersen, 1993), diet is macrophyte-dominated in late summer and autumn (Perrow et al., 1997) and unlikely to impact T. cristatus breeding in spring (Langton et al., 2001). The positive association identified here between T. cristatus and these waterfowl most likely reflects a shared preference for macrophyte-rich ponds.
T. cristatus were less likely to occupy ponds with higher mammal species richness. Our preliminary cooccur analysis indicated T. cristatus had negative associations with P. colchicus and S. carolinensis. The terrestrial associations identified are most likely indirect and a reflection of land-use rather than direct as a result of predation or competition, but further investigation would be worthwhile.
4.3 Abiotic determinants of T. cristatus occupancy
T. cristatus was less likely to inhabit large ponds with inflow present and a greater percentage of shading. Although our results indicate T. cristatus prefers smaller ponds, pond area does not always influence occupancy (Maletzky, Kyek & Goldschmid, 2007; Denoël & Ficetola, 2008; Gustafson et al., 2011) and was deemed a poor predictor of reproductive success (Vuorio et al., 2013). T. cristatus has been found to utilise small and large ponds (Rannap & Briggs, 2006; Skei et al., 2006); however, very small ponds (<124 m2) may be unable to support all life stages, and larger ponds may contain fish and experience eutrophication due to agricultural or polluted run-off (Rannap & Briggs, 2006). Inflow to ponds may exacerbate these problems by facilitating entry of agricultural or polluted run-off and connections to streams and rivers containing large, predatory fish (Freshwater Habitats Trust, 2015). Our results corroborate existing research where viable T. cristatus populations were unlikely in ponds that were shaded (Vuorio et al., 2013) or had dense macrophyte cover (Rannap & Briggs, 2006; Skei et al., 2006; Hartel et al., 2010).
In our study, most environmental metadata available were qualitative, preventing detailed analyses on pond properties and terrestrial habitat in relation to T. cristatus occupancy. Better understanding of T. cristatus occupancy in relation to species interactions and habitat quality could be achieved with quantitative data on pond properties (e.g. water chemistry), terrestrial habitat (e.g. type, density, distance to ponds), and aquatic and terrestrial habitat usage by different vertebrate species. Furthermore, given the metapopulation dynamics of T. cristatus, future research should investigate spatial drivers (e.g. land cover, pond density, climate variables, roads, rivers, elevation) of T. cristatus occupancy using innovative modelling approaches, such as individual-based models (Messager & Olden, 2018). However, acquiring this data to perform these models is a phenomenal task for large numbers of ponds across a vast landscape (Denoël & Ficetola, 2008).
4.4 T. cristatus HSI and umbrella status
We found the HSI can predict eDNA-based T. cristatus occupancy at the UK pondscape. This contradicts conventional studies which deemed the index inappropriate for predicting T. cristatus occupancy or survival probabilities (Unglaub et al., 2015). We detected more vertebrates in ponds containing T. cristatus, which may support its status as an umbrella species for pond biodiversity and conservation (Gustafson et al., 2006). We also observed a non-significant increase in vertebrate species richness with increasing T. cristatus HSI score. An adapted HSI, designed to predict species richness, could help select areas for management and enhancement of aquatic and terrestrial biodiversity. Until then, presence of T. cristatus and its HSI may confer protection to broader biodiversity by identifying optimal habitat for pond creation and restoration to encourage populations of this threatened amphibian. The HSI is not without issue due to qualitative data used for score calculation and subjective estimation of indices (Oldham et al., 2000). For future application of this index in T. cristatus eDNA survey, we recommend metabarcoding to quantify some qualitatively assessed indices (e.g. water quality via macroinvertebrate diversity, fish and waterfowl presence) alongside T. cristatus detection. Provided rigorous spatial and temporal sampling are undertaken, eDNA metabarcoding can also generate site occupancy data to estimate relative species abundance (Valentini et al., 2016; Hänfling et al., 2016).
4.5 Prospects of eDNA metabarcoding for freshwater conservation, management, and research
We have demonstrated the effectiveness of eDNA metabarcoding for landscape-scale biodiversity monitoring and ecological hypothesis testing. We combined metabarcoding with environmental metadata to revisit hypotheses relating to biotic and abiotic determinants of a threatened amphibian at the UK pondscape. Our findings will guide T. cristatus conservation in the face of increasing land-use and habitat fragmentation - a poignant issue as protective legislation for this species in the UK is changing. Whilst conservation of threatened species and their habitat should be a priority, the bigger picture should not be ignored. eDNA metabarcoding could enhance our understanding of freshwater networks, particularly pondscapes, to enable more effective monitoring, protection, and management of aquatic and terrestrial biodiversity. We are only now beginning to realise and explore these opportunities.
Data Availability
The taxonomically assigned sequence reads used in this study were generated by Harper et al. (2018). The raw sequence reads were archived on the NCBI Sequence Read Archive (Bioproject: PRJNA417951; SRA accessions: SRR6285413 - SRR6285678). The bioinformatics analysis was deposited in a GitHub repository and permanently archived (https://doi.org/10.5281/zenodo.1188710). R scripts and corresponding data for this study have been deposited in a separate GitHub repository which has been permanently archived (https://doi.org/10.5281/zenodo.2634427).
Author Contributions
B.H, L.R.H, L.L.H and N.B conceived and designed the study. H.C.R and N.B contributed samples for processing. L.R.H performed laboratory work and analysed the data. I.P.A and
E.L offered advice on and supervised sequencing. C.H assisted with bioinformatics analysis. P.B and S.P contributed datasets for analysis. L.R.H wrote the manuscript, which all authors revised.
Acknowledgements
This work was funded by the University of Hull. We would like to thank Jennifer Hodgetts (Fera Science Ltd) for assisting with sample collection, and Jianlong Li (University of Hull) for primer design and advice on laboratory protocols. Tissue samples for primer validation and Sanger sequencing were provided by Andrew Buxton and Richard Griffiths (DICE, University of Kent) under licence from Natural England, and Barbara Mabel and Elizabeth Kilbride (University of Glasgow).