Received: 9 February 2017 | Revised: 11 September 2017 | Accepted: 25 September 2017 DOI: 10.1111/mec.14420

ORIGINAL ARTICLE

Living quarters of a living fossil—Uncovering the current distribution pattern of the rediscovered Hula painted ( nigriventer) using environmental DNA

Sharon Renan1 | Sarig Gafny2 | R. G. Bina Perl2,3 | Uri Roll4,5 | Yoram Malka6 | Miguel Vences3 | Eli Geffen1

1Department of Zoology, George S. Wise Faculty of Life Sciences, Tel-Aviv Abstract University, Tel Aviv, One of the greatest challenges of effective conservation measures is the correct 2The School of Marine Sciences, Ruppin identification of sites where rare and elusive organisms reside. The recently redis- Academic Center, Michmoret, Israel 3Division of Evolutionary Biology, covered Hula painted frog (Latonia nigriventer) has not been seen for many decades Zoological Institute, Braunschweig and was therefore categorized extinct. Since its rediscovery in 2011, individuals University of Technology, Braunschweig, Germany from the critically have been found, with great effort, only in 4School of Geography & the Environment, four restricted sites. We applied the environmental DNA (eDNA) approach to search University of Oxford, Oxford, UK for new populations of the Hula painted frog in suitable aquatic habitats. We fur- 5Mitrani Department of Desert Ecology, Ben-Gurion University of the Negev, ther used the eDNA data to classify the landscape factors associated with the spe- Midreshet Ben-Gurion, Israel cies distribution and to predict its suitable habitats. We sampled 52 aquatic sites in 6 Israel Nature and Parks Authority, the during the spring of 2015 and 2016 and amplified the samples with Jerusalem, Israel a species-specific qPCR assay. DNA of the Hula painted frog was detected in 22 of Correspondence the sites, all of which clustered within three main areas. A boosting classification Sharon Renan, Department of Zoology, George S. Wise Faculty of Life Sciences, model showed that soil type, vegetation cover and the current and former habitats Tel-Aviv University, Tel Aviv, Israel. are all key predictors of the frog’s current distribution. Intriguingly, the habitat suit- Email: [email protected] ability models reveal a high affinity of the species to its long-lost habitat of the his- Funding information torical wetlands. Our findings encourage a series of informed searches for new The Israel Nature and Parks Authority; U.S. Fish & Wildlife Services, Grant/Award populations of this threatened frog and provide guidance for future conservation Number: F14AP00886 management programmes. In the era of global conservation crisis of , developing the eDNA approach, a reliable detection method for many and elusive amphibians, is of particular importance.

KEYWORDS amphibians, conservation genetics, genetic monitoring, species distribution modelling

1 | INTRODUCTION years have been focusing on protecting the remaining populations and species across the world. Many of the conservation Amphibians are considered the most threatened class of vertebrates efforts focus on monitoring spatial distributions and determining sta- on earth (Hoffmann et al., 2010; Stuart et al., 2004). The IUCN Red tus and health of amphibian populations. As many of the critically List classifies 41% of amphibian species as threatened with 546 spe- endangered amphibians are highly elusive species, monitoring them cies listed as “critically endangered” (IUCN 2016). Major threats to can be an arduous task. Moreover, when monitoring rare species, amphibians include habitat loss, pollution, diseases and climate especially in aquatic environments, false-negative detections (type II change (Primack & Sher, 2016), and conservations effort in recent errors) are commonly obtained as detection probabilities are often

| Molecular Ecology. 2017;26:6801–6812. wileyonlinelibrary.com/journal/mec © 2017 John Wiley & Sons Ltd 6801 6802 | RENAN ET AL. very low (Gu & Swihart, 2004). In some cases, calling surveys can be The recently rediscovered Hula painted frog (Latonia nigriventer) useful for detecting rare and elusive amphibians (Heyer, Donnelly, has not been detected in the wild for almost 60 years and was McDiarmid, Hayek, & Foster, 1994); however, this tool is not appli- consequently declared extinct by the International Union for the cable when monitoring silent species that produce low-intensity Conservation of Nature (IUCN) in 1996. Having been rediscovered calls. Moreover, when more than one species is present in the in October 2011, the Hula painted frog is now one of the most water, the acoustic monitoring is inefficient in detecting all species, endangered vertebrates on earth and listed by the IUCN Red List especially the rare ones (Pellet & Schmidt, 2005). These challenges as Critically Endangered. It was first described as lead to uncertainty about amphibian occurrence and often result in nigriventer following the discovery of two specimens on the east ineffective conservation action. This lack of systematic and reliable shore of the Hula Lake in northern Israel in 1940 (Mendelssohn & data on extremely rare amphibian species resulted in the designation Steinitz, 1943). In 1951, the Hula Valley wetlands were drained of some of these species as “extinct”; several of which were redis- with the aim of transforming the swamp into agricultural land. The covered at a later stage (Scheffers, Yong, Harris, Giam, & Sodhi, drainage of the Hula Valley, however, had severe ecological conse- 2011). quences, including the of various unique species endemic Environmental DNA (eDNA) is rapidly becoming a ‘go-to’ tool for to the Hula Lake and swamp. During the drainage years, a third detecting species presence in aquatic environments and has been specimen of the Hula painted frog was found (Steinitz, 1955), and utilized in several studies (Bohmann et al., 2014; Goldberg, Strickler, since then, despite frequent surveys in its habitat, the species was & Pilliod, 2015; Taberlet, Coissac, Hajibabaei, & Rieseberg, 2012). In not seen for 56 years. The loss of the Hula painted frog became the last decade, eDNA techniques have been successfully applied to one of the symbols of the ecological catastrophe of the Hula Valley identify the presence of invasive species in the early stages of inva- drainage. sion (Dejean et al., 2012; Ficetola, Miaud, Pompanon, & Taberlet, Following its rediscovery in October 2011, molecular and mor- 2008), for detecting the presence of rare and endangered species phological analysis revealed not only that this species had ‘come (Goldberg, Pilliod, Arkle, & Waits, 2011; Jerde, Mahon, Chadderton, back from the dead’, but that it is also a living fossil belonging to the & Lodge, 2011; Laramie, Pilliod, & Goldberg, 2015; Olson, Briggler, & Latonia known previously only from fossil records (Biton et al., Williams, 2012; Pilliod, Goldberg, Arkle, & Waits, 2013; Sigsgaard, 2013). Intensive efforts have now been made to find more individu- Carl, Møller, & Thomsen, 2015; Spear, Groves, Williams, & Waits, als in order to assess the population condition, learn more about the 2015; Thomsen et al., 2012; Wilcox et al., 2013) and for assessing species’ life history and attempt a genetic study of the population. aquatic and terrestrial biodiversity (Andersen et al., 2012; Lodge Before this study began, only 34 individuals had been found at four et al., 2012). Quantitative PCR (qPCR) has been suggested to be par- restricted sites in the Hula area, all in terrestrial habitats neighbour- ticularly useful for eDNA studies due to its high sensitivity, as it was ing water bodies. Moreover, these individuals are cryptic, burrowing shown to have a higher detection probability than traditional sam- in the moist detritus, and uncovering them is a difficult task. During pling tools (Dejean et al., 2012; Thomsen et al., 2012) and conven- the initial searching for individuals, no tadpoles were found and 18 tional PCR techniques (Spear et al., 2015; Wilcox et al., 2013). This of the discovered individuals were very small juveniles (16–26 mm advantage is important in the study of rare and elusive species as it snout–vent length and 0.5–1.8 g weight; Perl et al., 2017), suggest- increases the probability of finding species at sites of low population ing that the tadpole stage of the Hula painted frog is of short dura- density. It thereby often reduces the probability of false negatives, a tion. As both adults and tadpoles are very difficult to detect, critical factor in conservation practice. This method is of particular monitoring the species and identifying potential new sites have appeal to organisms, such as amphibians, that spend at least part of remained a great challenge. their life cycle in an aquatic medium. In this study, we employed the eDNA approach to detect the Usage of presence data to identify landscape factors that are asso- presence of the Hula painted frog in potentially suitable aquatic ciated with species distribution and to construct habitat suitability habitats across the entire Hula Valley region. We then used this maps is an approach increasingly applied in ecology and conservation presence–absence data to identify the important landscape attri- practice in general (for a review, see Guisan & Zimmermann, 2000), butes associated with the Hula painted frog’s preferred habitat. and in the study of endangered amphibians in particular (Blank & We further constructed a habitat suitability model in order to pre- Blaustein, 2012; Garcıa-Rodrıguez, Chaves, Benavides-Varela, & dict suitable habitats for the species and to identify potential sites Puschendorf, 2012). Nevertheless, the performance of such models is for future colonization and translocation programmes. As the land- poor when sample size is small (Wisz et al., 2008), as is usually the scape has dramatically changed in the recent past and the old case for rare and elusive species. Because the eDNA methodology is swamp is gone, we hypothesized that the species has survived in likely to increase the detection of species presence, using the pres- small refugia left from its original habitat and possible adapted to ence data from eDNA sampling is expected to increase the reliability new man-made habitats as well. The approach and methods we of habitat suitability models when studying a rare species. However, used in this study could be applied in an array of similar cases to to the best of our knowledge, using eDNA data to construct a habitat delineate distribution patterns of elusive organisms and promote suitability model has never previously been attempted. their conservation. RENAN ET AL. | 6803

2 | MATERIALS AND METHODS 2.3 | Study site

The Hula Valley in northern Israel covers an area of 177 square kilo- 2.1 | The eDNA approach metres (25 km long by 6–8 km wide) at an elevation of about 70 m In spite of the promise of the eDNA technique, there are several above sea level (Figure 1). This northern basin of the Dead Sea Rift inherent key challenges in order to implement it successfully. These Valley (Afro-Syrian Rift Series) was created during the formation of include: (i) the samples have many nontarget sources of DNA; (ii) most the rift in the Pleistocene. The Hula Valley is bordered by the steep samples have low quantity and quality of target DNA; and (iii) closely slopes of the Golan Heights on the east, the Upper Galilee moun- related species may co-occur at the same site with the study target tains on the west and the basalt ‘plug’ on the south. The Hula Valley species. All these issues can increase the chances for both false-posi- climate is Mediterranean, characterized by relatively hot, dry sum- tive and false-negative detection errors (Darling & Mahon, 2011). To mers (mean temperature SD, 32.9°C 2.7) and cool, wet winters address these challenges, in this study species-specific primers and a (15.7°C 3.7). Annual rainfall ranges from 400 mm in the south to specific probe were developed, a pilot study testing for the best col- 800 mm in the north of the valley. Water coming from underground lection–extraction protocol was conducted (following Renan et al., springs, mainly from the Hermon Mountains precipitation, is the 2012; Deiner, Walser, M€achler, & Altermatt, 2015), and all genetic main source for the flowing water throughout the valley (Hambright analyses were conducted in a ‘clean’ laboratory that was designed only & Zohary, 1998). for the eDNA samples, using conservative protocols such as those used in ancient or noninvasive DNA studies (Beja-Pereira, Oliveira, 2.4 | Testing different collection–extraction Alves, Schwartz, & Luikart, 2009; Cooper & Poinar, 2000). protocols for the eDNA methodology

To test the amplification success of the Latonia assay on eDNA sam- 2.2 | Designing the species-specific assay ples and the amplification success of different collection–extraction A PrimeTimeâ qPCR species-specific assay (primers and probe) was protocols, we collected water samples from one water body in the designed for the Hula painted frog by sequencing the same sections Hula Valley that the Hula painted frog and all the other three nontar- of the mitochondrial DNA (mtDNA) in this species and in the other get species are known to inhabit. Water samples were taken using three amphibian species that may co-occur in the Hula Valley: the two different collection protocols: (i) three tubes of 15 ml of water Levantine frog Pelophylax bedriagae (previously Rana levantina), the were kept frozen at 20°C and centrifuged before extraction (follow- Middle Eastern tree frog Hyla savignyi and the green toad Bufotes ing Thomsen et al., 2012); and (ii) 1 L of water was filtered using dis- viridis (previously Bufo viridis). For our sequencing protocol, we used posable funnels with 47-mm-diameter cellulose nitrate filter paper published primers for three mitochondrial genes: 12S rRNA, 16S with a 0.45-lm pore size (Thermo Scientific), and the filters were rRNA and cytochrome-b, which are known to be highly polymorphic placed into a sterile 2-ml tube and kept frozen at 20°C until extrac- among amphibian species (Biton et al., 2013; Goebel, Donnelly, & tion (following Pilliod et al., 2013). Three replicates of each collection Atz, 1999; Zangari, Cimmaruta, & Nascetti, 2006). Although all method were extracted using the DNeasy Blood & Tissue kit (QIA- sequences were previously published for all the four amphibian spe- GEN, Cat No. 69504) protocol with several modifications (see details cies, we sequenced those regions again using tissue samples, in in the “DNA extraction, PCR and sequencing” section), and another order to have a better comparable set of sequences for our assay three replicates of the filtering method were also extracted using the design. We targeted mtDNA sequences, as it is a more suitable mar- Power water DNA Isolation kit (MOBio). This resulted in three replica- ker for low quality and quantity DNA samples (Waits & Paetkau, tions of eDNA samples for each of the three different collection–ex- 2005). By comparing the mtDNA sequences of all four species in traction combinations (i.e., Centrifuge DNeasy, Filtering DNeasy, those three regions, we selected a target amplicon of 110-base pair Filtering MOBio). All samples from all three combinations were tested sequence in the 12S rRNA that enabled assay design of maximum using the Latonia assay with four amplification repeats for each sam- base pair differences in both the primer-binding and the probe-bind- ple. In each reaction, we used DNA extracted from a tissue sample as ing regions between the Hula painted frog and the three other non- a positive control and double-distilled water as a negative control. target species (following Wilcox et al., 2013). This Latonia assay All eDNA samples yielded successful amplifications, confirming included F-primer: 50- GAA CTA CGA GCC TCA GCT TAA A-30,R- the ability of the eDNA methodology to detect the presence of the primer: 30-GGC AAG AAG TGG TGA GGT TA-50 and probe: 50-CAA Hula painted frog from the water. One amplicon of each collection– ACC CAC CTA GAG GAG CCT GTT-30. PrimerQuestâ Tool was used extraction combination was sequenced to ensure that only the target for the primer design and the probe was manually designed using species was amplified in each reaction. The protocol of sample filter- oligoanalyser tools to check homo- and heterodimerization (Inte- ing and extraction with the DNeasy kit had a slightly higher amplifi- grated DNA Technologies, Inc. via Syntezza Bioscience Ltd). The pri- cation success (i.e., more of the samples repeats were successfully mers and probe were screened via BLAST to check specificity, and amplified) than the other two collection–extraction protocols, indi- were tested on tissue samples from the three nontarget species to cating on higher DNA concentration or less inhibitors in the extract, rule out possible cross-amplification. and was therefore used for all subsequent experiments. 6804 | RENAN ET AL.

Mediterranean (a) (b) Sea Sea of Galilee

Dead Sea Israel

33.17 Egypt Jordan FIGURE 1 Locations of the eDNA sampling sites projected onto the landscape classification maps of the Hula Agricultural field Valley. Red points denote sites where Batha eDNA of the Hula painted frog was Forest 33.10 Lake detected and yellow points are sites Orchard without detection. The black star denotes Reservour the location of initial discovery by Riverine Mendelssohn and Steinitz (1943). (a) Settlement Classification map of the habitats during Swamp the 1940s (when the Hula painted frog 33.03 was first discovered); (b) current 04 classification map of the habitats [Colour km figure can be viewed at 35.57 35.60 35.63 35.57 35.60 35.63 wileyonlinelibrary.com]

digestion stage. Samples were amplified on the StepOnePlusTM Real- 2.5 | Field sampling time PCR System (Applied Biosystems). We ran 20 ll qPCR reactions As the number of field replicates was shown to be positively related with 10 ll qPCR BIO Probe Mix Hi-Rox (Integrated DNA Technolo- to eDNA detection (Willoughby, Wijayawardena, Sundaram, Swihart, gies, Inc), 2 ll PrimeTimeâ qPCR assay (0.25 lM Probe and 0.5 lM

& DeWoody, 2016), four replicates of 1 L water samples were col- of each primer, Integrated DNA Technologies, Inc), 6 ll of ddH2O lected from 52 aquatic sites during spring 2015 and 2016 (2 9 1L and 2 lM of sample extract. The qPCR cycling protocol began with each year except for five sites that were sampled in 2015 but were 2 min at 95°C followed by 40 cycles of 95°C for 5 s and 60°C for found dry in spring 2016 and were replaced by four additional 25 s. All samples were run in duplicates with both positive and nega- potential sites) covering over ca. 177 km2 in the Hula Valley. To tive controls. Six independent qPCR replications were performed for avoid cross-contamination between sites, each water sample was each of the four field sample replicates (24 PCR for each sampling taken from the shore using new disposable gloves without entering site) to increase detection probability (Willoughby et al., 2016). A with waders into the water. Sampling was conducted during the sample was considered as positive if at least two of its six amplifica- breeding period in order to increase detection probability (Spear tion repeats came up positive. In addition, one PCR product of every et al., 2015). Sampling sites included drain ditches, natural and artifi- positive sample was sequenced to validate the amplified fragment as cial ponds and natural streams (Table S1). All samples were filtered Hula painted frog DNA. Sequencing was performed at G.S. Wise within the same day in a field laboratory, using manifold that allows Faculty of Life Sciences DNA Sequencing Unit on an ABI 3500xl to simultaneously pump water through 12 disposable funnels. Each Genetic analyser (Applied Biosystems). All extractions and PCR set- filter paper was folded using disposable tweezers and placed into a up were conducted in separated laboratory benches dedicated to 2.0-ml tube (in the case of a very cloudy sample, two filter papers eDNA. per sample were used). To avoid cross-contamination, each filter was handled using new disposable gloves. All 2.0-ml tubes were stored 2.7 | Classification modelling of eDNA presence/ at 20°C and were extracted within 1–2 weeks after collection. absence data

We used a supervised machine learning procedure to classify 2.6 | DNA extraction, PCR and sequencing between sites in which Hula painted frog eDNA was present or All samples were extracted using the DNeasy Blood & Tissue kit absent Olden, Lawler, and Poff (2008). Our modelling approach (QIAGEN, Cat No. 69504) protocol with the following modifications: incorporated 14 environmental variables for each sampling site that (i) the ATL buffer and proteinase K were added directly to the 2.0- are potentially relevant for amphibians. These variables were either ml tubes with the filter paper. (ii) We added 90 ll to the 180 ll ATL recorded in the field during sampling (i.e., depth, latitude, longitude, buffer in the first digestion stage as much of the lysis buffer absorbs vegetation cover, wideness, muddy and site type) or obtained from in the filter. (iii) We incubated the samples for at least 2 hr in the remotely sensed data and publicly available sources (i.e., current RENAN ET AL. | 6805 habitats, habitats in the 1940s, altitude, temperature, NDVI (normal- the sum of the squared improvements in error rate over all the inter- ized difference vegetation index) in April, NDVI in August and soil nal nodes for which it was selected as the splitting variable. These types). All field variables were collected by one person in order to improvements are then averaged over all classifications, and calcu- maintain uniformity of sampling and were estimated qualitatively in lated as a percentage of the best predictor, to give the relative the field (Table S2). We used a high-resolution aerial photograph to importance value per predictor (Hastie et al., 2001). We also pro- classify the current habitats of the region where the samples were duced cross-validation partial dependence plots for the predictors found (Figure 1b). This region had undergone severe habitat modifi- using the “INTERPRETR” package in R (Ballings & Van den Poel, 2016). cations over the past 75 years, since the initial discovery of the Hula These plots show the change in classification rate along the variable painted frog (Mendelssohn & Steinitz, 1943), and we therefore gradient of classes. attempted also to classify the habitats of this region in the 1940s (Figure 1a). This was done since past habitats may have had an 2.8 | Habitat suitability models for the Hula painted important effect on where this relict species resides today. Past frog habitats were classified using a topographical map of the region from 1942 (Houle; Levant 1:50,000 file N1-36-XII-2a, edition of Septem- In order to construct a habitat suitability model for the Hula painted ber 1942), augmented by aerial photographs of this region from frog over the entire region of the Hula Valley, we used our presence 1945 (680 p.s. 23, 1:15,000 from 30/1/1945), as well as surface locations together with pseudo-absence points, representing a ran- photographs of the various habitats of this region from the 1940s dom spatial pattern in the landscape. We placed 500 random points (obtained from http://salkkl.kkl.org.il/form/photos/ArchivePrePage). in our study extent (Figure 1) and ran these and our presence-only We used the same nine landscape category types to classify the points, using a MaxEnt (maximum entropy) model with a fivefold entire landscape for both current and past habitats. These categories cross-validation. For this modelling procedure, we used only the pre- are found in the legend of Figure 1. When plotting the actual sam- dictors for which we had full coverage in the entire landscape (i.e., pling locations on both the current and past habitat classifications, not those collected in the field while sampling the eDNA; Table S1). these locations were not located in all of the nine different habitat We projected the model predictions to a region of the entire Hula types found in the landscape explored. Our sampling locations were Valley (10 9 24 km, 240 km2), at a spatial resolution of 20 m2.We located in five landscape categories on the past habitat classification also plotted the predicted presence locations of the model, using a and seven on the current habitat classification (Table S2). Spatial cross-validated threshold—separating predicted presence and classification and mapping were done in ARCGIS version 10 (ESRI absence locations. The threshold employed was calculated via a sen-

2011), and the modelling was conducted in the R programming lan- sitivity–specificity sum maximization approach (Liu, Berry, Dawson, & guage (R Core Team 2016). Pearson, 2005). The model was constructed using MaxEnt (Phillips,

We used a gentle adaptive stochastic boosting classification Anderson, & Schapire, 2006) and ran using the “DISMO” package in R model (Friedman, Hastie, & Tibshirani, 2000) to distinguish between (Hijmans, Phillips, Leathwick, & Elith, 2016). sites with presence or absence of the Hula painted frog DNA. This While not covering the entire landscape, the identity of our method can combine the outputs of many weak classifiers in order sampled sites where Hula painted frog eDNA was absent may pro- to achieve, through iterations, a powerful classification with low vide further information on suitable/unsuitable habitats for the error rates (Hastie, Tibshirani, & Friedman, 2001). At each step, the Hula painted frog. The conservative approach, of using the pres- observation weights are modified such that misclassified observa- ence-only data, came from the concern that in some of our “ab- tions increase in weight and correctly classified observation decrease sence” sites, there may be a population that has been undetected. in weight. Therefore, as the procedure progresses through iterations, However, if absence data are available, in most cases, this informa- observations that are more difficult to classify correctly receive ever- tion can produce a better habitat suitability model. We therefore increasing influence. Put together, the boosting algorithm sequen- constructed a second habitat suitability model using only the eDNA tially applies weak classifiers (whose error rate is only slightly better sampling locations with presence–absence data as inputs to our than random guessing) to repeatedly modified versions of the data model, which was constructed in a similar manner as the model until the predictions from all the weak classifiers are combined above. through a weighted majority vote to produce the final prediction (Hastie et al., 2001).

Our modelling was conducted using the “ADA” package in R (Culp, 3 | RESULTS Johnson, & Michailidis, 2016) and incorporated an exponential loss function with 5,000 iterations. In order to obtain mean absolute pre- One PCR product of each of the positive samples was sent for diction error rates, we employed a fivefold cross-validation proce- sequencing (n = 26 in 2015 and n = 13 in 2016; Table S3). All of dure. Furthermore, we computed the relative importance of the these sequences consistently matched the Hula painted frog DNA different predictors. At each internal node, the variable giving the (>99% identity in BLAST), indicating that our assay is highly species- maximal improvement in error for that node is selected, and its specific and does not amplify any of the three nontarget species improvement in error rate is noted. The importance of a predictor is (Pelophylax bedriagae, Hyla savignyi, Bufotes viridis). Out of the 52 6806 | RENAN ET AL. sampling sites, positive amplifications of the Hula painted frog DNA 3.2 | Habitat suitability models for the Hula painted were obtained at 22 sites (Figure 1). We found 16 positive sites in frog 2015 and 11 positive sites in 2016. Five of the 2016 sites were also found positive in 2015 and six sites were new (Table S3). No amplifi- The results of the habitat suitability model, based on presence and cation was obtained in any of the negative controls. pseudo-absence points, are presented in Figure 3a. This map clearly Out of the 22 positive sites, 10 sites had positive amplifications shows the importance of the former swamp and lake (Figure 1a) for for more than one replicate (two positive replicates, n = 6; three the prediction of currently suitable habitat for the Hula painted frog. positive replicates, n = 3 and four positive replicates, n = 1), while In addition, the model highlights the southern area of the Hula Val- for the other 12 sites, positive amplifications were obtained for only ley and the Agamon Ha-Hula nature park as highly suitable habitats one of the site replicates. Nevertheless, it is important to note that and the Hula nature reserve and its surroundings as the most suit- for all of those sites with only one positive replicate, positive amplifi- able habitat for the species. This pattern remains when observing cations were obtained in two or more of the qPCR reactions (out of the map depicting the predicted presence using a threshold based the six amplification repeats). on maximizing the sum of specificity and selectivity (0.0118, Fig- Four of the 22 positive sites were our positive control sites, with ure 3b). Overall, the model performed well, with a value of 0.886 for previously known presence of the Hula painted frog, supporting the the mean cross-validated AUC (i.e., area under the curve—which is reliability of the eDNA method. Most of the positive sites were the probability that a classifier will rank a randomly chosen positive located in the Hula nature reserve and its surroundings (n = 16) and instance higher than a randomly chosen negative one; Fawcett, the few additional positive sites were found in the Agamon Ha-Hula 2006). The variable contributions for this model are given in Table 1. nature park area (n = 4) and in the Ein Te’o nature reserve (n = 2). Altitude, past habitat and soil type had the highest contribution to No positive site was found along the eastern shore of the former the model. Hula Lake, where the three-first specimens were found in 1940 and The presence–absence model predicted similar suitable habitats in 1955 (Figure 1). for the Hula painted frog in the landscape we studied. Importantly, the most suitable habitats were placed in the region occupied by the former Hula Lake and swamp (Figure S1). The model performed well, 3.1 | Classification model for eDNA presence/ but not as good as the presence-only model, with a mean cross-vali- absence dated AUC value of 0.677. The overall maximum probabilities of the Our boosting classification model of sites where Hula painted frog model were much higher in the presence–absence model compared eDNA was present or absent resulted in a perfect classification (from with the presence-only model. the fivefold cross-validation procedure) with a mean absolute error rate of zero. This means that the predictors we used enabled us to correctly predict the presence of this frog in a given sampled location. 4 | DISCUSSION The relative importance of each predictor to the model is presented in Figure 2a. Both current and past habitats, as well as soil types and The Hula painted frog that seemingly disappeared and came back spring NDVI values, were found to have an important influence on the after almost 60 years effectively demonstrates our limitations in Hula painted frog’s presence. Finally, the dominance of latitude and monitoring elusive species using only traditional sampling tools. longitude in the classification suggests a spatial clustering effect for Intensive fieldwork conducted in parallel to the present study, and in the locations where Hula painted frog DNA was present. which both adults and tadpole individuals were found, determined The partial dependence plots (Figure 2) display the relationship that the Hula painted frog is nocturnal and well camouflaged, the between different values/categories of the predictors and the mean tadpoles are rare and hard to detect due to the short larval period

(logit(Ppresence)/2). When examining these plots for top classifiers, and small tadpole size (with maximal total lengths of 26 mm), and several interesting patterns arose. Latitude showed a high probability the calls produced by the species are of a very low intensity, barely of presence in the southern part of the Hula Valley, with a sharp audible to the human ear from a few metres’ distance (Perl et al., decrease above 33.1° (Figure 2b), and longitude showed a high prob- 2017). These findings may explain why the species was not detected ability of presence in the western part of the Hula Valley, with a during the numerous amphibian surveys that have been conducted sharp decrease above 35.62° (Figure 2c). April NDVI values showed in the Hula area over the last 60 years, all of which applied the com- an increase in the probability of the Hula painted frog presence from monly used survey methods for amphibians (i.e., direct observations, the value of 5,000 (Figure 2d). The categories from the current habi- sampling with hand-held nets and bioacoustic monitoring). Our study tat classification showing most association with the frog’s presence demonstrated that in contrast to traditional sampling tools, the were agricultural fields, orchards and reservoirs (Figure 2e), and eDNA approach is highly effective in detecting the presence of the when looking at past habitats, the most associated category was the Hula painted frog, as previous studies have indeed shown in other former area of the swamp (Figure 2f). Two soil types, organic and rare and elusive species (Goldberg et al., 2011; Jerde et al., 2011; colluvial–alluvial soils, were also linked to a high probability of the Laramie et al., 2015; Olson et al., 2012; Pierson et al., 2016; Pilliod Hula painted frog (Figure 2g). et al., 2013; Sigsgaard et al., 2015; Spear et al., 2015; Thomsen RENAN ET AL. | 6807

(a) Depth Muddy Vegetation Wideness Ditch/Pool/Stream Temperature Altitude August NDVI 1940s habitats Longitude Soil type April NDVI Current habitats Latitude

0 20 40 60 80 100 Relative scores

(b) (c) (d) 5 −2 −2 0 −4 −4 −5 −6 −6 −10 −8 Parital dependence Parital dependence Parital dependence −8 −15 33.04 33.08 33.12 33.16 35.58 35.60 35.62 35.64 4,000 5,000 6,000 7,000 8,000 Latitude Longitude April NDVI

(e) (f) (g) −3 −2 −3.5 −4 −3 −4.0 −5 −4 −5 −4.5 −6 −6 −5.0 −7 −7

Parital dependence Parital dependence −5.5 −8 Parital dependence −8

Batha Lake Batha Swamp Forest Riverine Swamp Riverine OrchardsReservour Agriculture Agriculture Organic soils Basaltic soil Lacustrine gley Current habitats 1940s habitats Grumusolic gley Brown rendzina Colluvial-alluvial soil Hydromorphic grumusol Alluvial brown grumusol Soil type

FIGURE 2 Relative variable importance scores for modelling the Hula painted frog positive sites using stochastic boosting. The mean classification importance for all predictors in a gentle adaptive stochastic boosting classification model is computed and shown as percentage of the best predictor (which receives 100% relative importance). Also displayed are the partial dependence plots for the top six predictors: latitude, longitude, April NDVI, current habitat classifications, past habitat classifications and soil types et al., 2012; Wilcox et al., 2013). The data even suggest that for this future study seeking to detect and monitor populations of the Hula extremely elusive species, eDNA might be the only efficient tool for painted frog. a reliable detection of the species’ presence. This study established All four sites with a previously known record of the Hula painted an eDNA monitoring protocol, including sample collection, eDNA frog were found positive using the eDNA methodology. This sup- extraction procedures and a species-specific qPCR assay, for any ports the reliability of the method for our system. In addition to 6808 | RENAN ET AL.

(a) (b) N N

33.17

33.10 FIGURE 3 Spatial predictions from a MaxEnt habitat suitability model for the Hula painted frog, using eDNA presence information and 500 randomly placed Probability .016 pseudo-absences. (a) The predicted presence probabilities for the Hula painted frog in the landscape; (b) the predicted 0 presence values based on a cross-validated 33.03 04 threshold (maximizing specificity and km selectivity). The dashed line represents the former lake and swamp area [Colour figure 35.57 35.60 35.63 35.57 35.60 35.63 can be viewed at wileyonlinelibrary.com]

interconnected in a complex network of canals and streams and that TABLE 1 Estimates of relative variable contributions to the water flow is mainly from north to south, and in part also from east MaxEnt model. Variable per cent contribution is determined in each iteration of the training algorithm by adding or subtracting the and west, to the centre of the valley. Consequently, sites testing increase or decrease in regularized gain for each corresponding positive for eDNA might not always indicate a Hula painted frog variable. Permutation importance is calculated per variable by population at a specific site, but rather a Hula painted frog popula- randomly permutating its values for the presence and absence data tion is present in the adjacent area, from a specific point upstream (on the training set) and then evaluating the drop in the model AUC (for review, see Goldberg et al., 2015). Nevertheless, all positive sites value due to this shuffling of cases. These values for all variables are then normalized to percentages constitute good potential candidates for the presence of a local pop- ulation, and should be regarded as of high conservation interest until Per cent Permutation proven otherwise. Variable contribution importance Half of the sites where DNA of the Hula painted frog was Altitude 50.1 75.0 detected were positive in only one of four site replicates. This high- 1940s habitats 20.3 1.4 lights the importance of several field replicates, especially when Soils 11.6 0.5 searching for rare species with very small population sizes (Wil- Current habitats 7.7 1.8 loughby et al., 2016). As the target species of this study is one of Temperature 5.8 19.6 the most endangered vertebrates known today, the decision to April NDVI 3.1 1.2 include also those sites with only one positive replicate in our data- August NDVI 1.4 0.5 set derived from the conception that from a conservation point of view a false negative (i.e., misdetection of an actual population) has these four known sites, 18 new sites were positive for the frog’s a more severe impact than a false positive. Additional support for DNA, most of them clustered in three regions: the Hula nature considering sites that were positive for Hula painted frog in only reserve and its surroundings, the Agamon Ha-Hula nature park and one of four replicates was the fact that two of the four positive con- the Ein Te’o nature reserve. These findings reveal the current main trol sites (where live Hula painted frog were collect before) also areas of the Hula painted frog populations in the Hula Valley and came up positive in only one of four replicates. Nevertheless, to thereby facilitate the search efforts of future studies on this endemic ensure population presence, any positive detection by the eDNA species. The fact that all three positive regions are protected areas approach should be verified by physical collection of individuals. highlights the important role of nature reserves and nature parks in We observed a small reduction in positive sites of the Hula safeguarding wildlife populations. Given these findings, it is impor- painted frog from spring 2015 to spring 2016 (from 16 to 11 sites, tant to note that almost all the water bodies of the Hula Valley are respectively). Tadpole surveys conducted during the same time RENAN ET AL. | 6809 period in the Hula Valley revealed a similar pattern of fewer tadpole in designating pseudo-absences). However, the higher overall maxi- observations of all amphibian species in spring 2016 compared to mum probabilities of the presence–absence model indicate on the spring 2015 (unpublished data). As eDNA is expected to increase greater predicted power of using both presence and absence data during the breeding period due to more frequent visits of adults to when available (Lobo & Tognelli, 2011; Phillips & Dudik, 2008). the water, as well as to potential male aggression and injuries and the release of gametes into the water (Spear et al., 2015), the noted reduction in eDNA-positive sites of the Hula painted frog in spring 5 | CONCLUSIONS 2016 might be a result of a less suitable year for amphibian repro- duction in the Hula area due to low rainfall. Our study shows the powerful utility of the eDNA approach in In this study, we did not attempt inferring population densities detecting highly elusive aquatic species. By taking only 4 L of water from eDNA concentration. This was due to the huge diversity and from each of the potential aquatic sites in the Hula region, we were significant differences among sampling sites. Some of these were able to reveal 22 sites of Hula painted frog occurrence, clustered in small ponds while others were very large water pools. Some sites three main areas of the valley. Employing the eDNA data to create had a relatively strong water flow while others had nearly stagnated classification and habitat suitability models further revealed the main water. A few sites had dense vegetation cover compared to other habitat characteristics associated with the presence of the Hula sites that were bare and exposed to direct solar radiation. All those painted frog and enabled construction of a predictive map of suit- differences can strongly affect the DNA concentration (Klymus, Rich- able habitats for this species. These maps are of great importance ter, Chapman, & Paukert, 2015; Pilliod, Goldberg, Arkle, & Waits, for the development of conservation management programmes and 2014; Strickler, Fremier, & Goldberg, 2015), making the comparison for identifying potential sites for future colonization or translocation of DNA concentration among sites meaningless in terms of popula- programmes of this critically endangered species. In the aim of con- tion density. Therefore, in this study we used the qPCR methodol- tinuing the ecological, biological and genetic study of the Hula ogy solely because it provides a higher sensitivity and higher Painted Frog, future study will focus on searching for live individuals detection ability compared with conventional PCR (Spear et al., in both the new eDNA-positive sites and the suitable areas inferred 2015; Wilcox et al., 2013). by model. eDNA sampling, which is likely to increase significantly the We are facing a global conservation crisis of amphibians that are detection probability of a species, should be the preferred approach declining more rapidly than any other vertebrate class (Beebee & for establishing habitat suitability models of aquatic rare species. In Griffiths, 2005). As the major threat to amphibians is habitat degra- this study, we were able to correctly classify between sites with dation and loss (Hoffmann et al., 2010), the most crucial need for presence/absence eDNA samples, indicating that these sites may be amphibian conservation is reliable knowledge on species distribu- inherently different in their attributes. The classification model high- tions, in order to guide efficient habitat conservation practices. How- lights the importance of several landscape factors for the Hula ever, as many of these critically endangered amphibians are highly painted frog distribution. This frog species occurs in a clustered dis- elusive and their detection probability very low, such knowledge is tribution, situated within the former area of the Hula swamp (Fig- difficult to acquire. In this study, we demonstrated the utility of ure 2). Soil type also had a strong effect on the frog’s presence eDNA in revealing the distribution patterns of one of the most criti- (similarly to its effect on other amphibian species, Dayton, Jung, & cally endangered vertebrates on earth. This cost-effective and rapid Droege, 2004; Blank & Blaustein, 2012), with organic and colluvial– approach has great potential in elucidating habitat requirements of alluvial soils, indicative to the presence of the past wetland, showing many other elusive and critically endangered amphibians or other the highest presence probability (Figure 2g). The high probability of aquatic organisms. In doing so, it can facilitate conservation efforts the frog’s presence at high spring NDVI values (Figure 2d) might be and aid in the selection of sites for protection. due to dense vegetation requirements for safer movement during the breeding period. Our two habitat suitability models depicted similar patterns of 6 | ACKNOWLEDEGMENTS suitable habitat for the Hula painted frog. Interestingly, the main pre- dictors of the current suitable habitat were the former lake and The Israel Nature and Parks Authority and the U.S. Fish & Wildlife swamp area and the soil type. This suggests that although more than Services (USFWS, grant no. F14AP00886) funded this study. SR was 60 years have passed, the distribution of this relict species mirrors supported by a postdoctoral scholarship from the Steinhardt the long-lost habitat rather than the current one. Our results may also Museum of Natural History and National Research Center. U.R. was imply on a certain resilience of the species to the massive habitat supported by the Kreitman Postdoctoral Fellowship at the Ben-Gur- modifications that have taken place—it still persists in lower densities ion University of the Negev. The study was conducted under permit in artificial habitats and small refugia left from its original habitat. number 2015/40926 of the Israel Nature and Parks Authority. We When comparing the two models, the presence-only model had a are grateful to the numerous volunteers and colleagues who pro- higher AUC, most likely due to the larger number of modelled loca- vided assistance during the laboratory and fieldwork and especially tions (as we were not constrained by the number of actual absences to Orly Cohen, Michal Hadas-Sasson, Naomi Gordon, Noa Trosknov, 6810 | RENAN ET AL.

Asaf Moran and Yael Balon. We would like to thank Amit Dolev, Blank, L., & Blaustein, L. (2012). Using ecological niche modeling to pre- Noam Leader and Yehoshua Shkedy from the Israel Nature and dict the distributions of two endangered amphibian species in aquatic breeding sites. Hydrobiologia, 693, 157–167. https://doi.org/10.1007/ Parks Authority for their advice and support. Finally, we would like s10750-012-1101-5 to thank Shalom Tramzi (Tel-Hai College) for access to the old maps Bohmann, K., Evans, A., Gilbert, M. T. P., Carvalho, G. R., Creer, S., collection of the Hula Valley and to Gili Greenbaum for helpful com- Knapp, M., ... De Bruyn, M. (2014). Environmental DNA for wildlife ments on the manuscript. The comments of S. Creer and four anony- biology and biodiversity monitoring. Trends in Ecology & Evolution, 29, 358–367. https://doi.org/10.1016/j.tree.2014.04.003 mous reviewers greatly improved the manuscript. Cooper, A., & Poinar, H. N. (2000). Ancient DNA: Do it right or not at all. Science, 289, 1139–1139. https://doi.org/10.1126/science.289.5482. 1139b DATA ACCESSIBILITY Culp, M., Johnson, K., & Michailidis, G. (2016). ada: The R Package Ada for Stochastic Boosting. R package version 2.0-5. Retrieved from The primers used for sequencing the 12S region of the mitochondrial https://CRAN.R-project.org/package=ada. DNA in the Hula painted frog and in the three nontarget amphibian Darling, J. A., & Mahon, A. (2011). From molecules to management: species are previously published in NCBI, accession: KC867711.1 Adopting DNA-based methods for monitoring biological invasions in (www.ncbi.nlm.nih.gov). aquatic environments. Environmental Research, 111,1–11. Detailed information on the environmental variables used for Dayton, G. H., Jung, R. E., & Droege, S. (2004). Large-scale habitat associa- tions of four desert anurans in Big Bend National Park, Texas. Journal classification is available in Table S2. of Herpetology, 38, 619–627. https://doi.org/10.1670/125-04N Spatial modelling code in R follows that given in the vignette of Deiner, K., Walser, J. C., M€achler, E., & Altermatt, F. (2015). Choice of the R “DISMO” package (Hijmans et al., 2016; https://cran.r-project. capture and extraction methods affect detection of freshwater biodi- – org/web/packages/dismo/vignettes/sdm.pdf). versity from environmental DNA. Biological Conservation, 183,53 63. https://doi.org/10.1016/j.biocon.2014.11.018 Dejean, T., Valentini, A., Miquel, C., Taberlet, P., Bellemain, E., & Miaud, C. (2012). Improved detection of an alien invasive species through AUTHOR CONTRIBUTIONS environmental DNA barcoding: The example of the American bullfrog S.R., S.G. and E.G. developed the study design. S.R., S.G., E.G., B.P. Lithobates catesbeianus. Journal of Applied Ecology, 49, 953–959. https://doi.org/10.1111/j.1365-2664.2012.02171.x and Y.M. collected the eDNA samples. S.R. conducted the laboratory ESRI (2011). ArcGIS Desktop: Release 10. Redlands, CA: Environmental work and analysed the genetic data. E.G. and U.R. constructed and Systems Research Institute. analysed the classification model and the habitat suitability model. Fawcett, T. (2006). An introduction to ROC analysis. Pattern Recognition M.V. contributed to the research design. S.R. wrote the first version Letters, 27, 861–874. https://doi.org/10.1016/j.patrec.2005.10.010 Ficetola, G. F., Miaud, C., Pompanon, F., & Taberlet, P. (2008). Species of the manuscript and all authors contributed their comments to the detection using environmental DNA from water samples. Biology Let- manuscript. ters, 4, 423–425. https://doi.org/10.1098/rsbl.2008.0118 Friedman, J., Hastie, T., & Tibshirani, R. (2000). Additive logistic regres- sion: A statistical view of boosting (with discussion and a rejoinder ORCID by the authors). The Annals of Statistics, 28, 337–407. https://doi. org/10.1214/aos/1016218223 Sharon Renan http://orcid.org/0000-0002-7316-7907 Garcıa-Rodrıguez, A., Chaves, G., Benavides-Varela, C., & Puschendorf, R. Uri Roll http://orcid.org/0000-0002-5418-1164 (2012). Where are the survivors? Tracking relictual populations of Eli Geffen http://orcid.org/0000-0002-3028-0045 endangered in Costa Rica. Diversity and Distributions, 18, 204– 212. https://doi.org/10.1111/j.1472-4642.2011.00862.x Goebel, A. M., Donnelly, J. M., & Atz, M. E. (1999). PCR primers and amplification methods for 12S ribosomal DNA, the control region, REFERENCES cytochrome oxidase I, and cytochrome b in bufonids and other frogs, and an overview of PCR primers which have amplified DNA in Andersen, K., Bird, K. L., Rasmussen, M., Haile, J., Breuning-Madsen, H. amphibians successfully. Molecular Phylogenetics and Evolution, 11, E. N. R. I. K., Kjaer, K. H., ... Willerslev, E. (2012). Meta-barcoding 163–199. https://doi.org/10.1006/mpev.1998.0538 of ‘dirt’ DNA from soil reflects vertebrate biodiversity. Molecular Goldberg, C. S., Pilliod, D. S., Arkle, R. S., & Waits, L. P. (2011). Molecular Ecology, 21, 1966–1979. https://doi.org/10.1111/j.1365-294X.2011. detection of vertebrates in stream water: A demonstration using 05261.x Rocky Mountain tailed frogs and Idaho giant salamanders. PLoS ONE, Ballings, M., Van den Poel, D. (2016). interpretR: Binary Classifier and 6, e22746. https://doi.org/10.1371/journal.pone.0022746 Regression Model Interpretation Functions. R package version 0.2.4. Goldberg, C. S., Strickler, K. M., & Pilliod, D. S. (2015). Moving environ- Retrieved from https://CRAN.R-project.org/package=interpretR mental DNA methods from concept to practice for monitoring aqua- Beebee, T. J. C., & Griffiths, R. A. (2005). The amphibian decline crisis: A tic macroorganisms. Biological Conservation, 183,1–3. https://doi.org/ watershed for conservation biology? Biological Conservation, 125, 10.1016/j.biocon.2014.11.040 271–285. https://doi.org/10.1016/j.biocon.2005.04.009 Gu, W., & Swihart, R. K. (2004). Absent or undetected? Effects of non- Beja-Pereira, A., Oliveira, R., Alves, P. C., Schwartz, M. K., & Luikart, G. detection of species occurrence on wildlife–habitat models. Biological (2009). Advancing ecological understandings through technological Conservation, 116, 195–203. https://doi.org/10.1016/S0006-3207 transformations in noninvasive genetics. Molecular Ecology Resources, (03)00190-3 9, 1279–1301. https://doi.org/10.1111/j.1755-0998.2009.02699.x Guisan, A., & Zimmermann, N. E. (2000). Predictive habitat distribution Biton, R., Geffen, E., Vences, M., Cohen, O., Bailon, S., Rabinovich, R., ... models in ecology. Ecological Modelling, 135, 147–186. https://doi. Gafny, S. (2013). The rediscovered Hula painted frog is a living fossil. org/10.1016/S0304-3800(00)00354-9 Nature Communications, 4,1–6. RENAN ET AL. | 6811

Hambright, K. D., & Zohary, T. (1998). Lakes Hula and Agmon: Destruc- Phillips, S. J., Anderson, R. P., & Schapire, R. E. (2006). Maximum entropy tion and creation of wetland ecosystems in northern Israel. Wetlands modeling of species geographic distributions. Ecological Modelling, Ecology and Management, 6,83–89. https://doi.org/10.1023/A: 190, 231–259. https://doi.org/10.1016/j.ecolmodel.2005.03.026 1008441015990 Phillips, S. J., & Dudik, M. (2008). Modeling of species distributions with Hastie, T., Tibshirani, R., & Friedman, J. (2001). The elements of statistical Maxent: New extensions and a comprehensive evaluation. Ecography, learning, 2nd ed. NY: Springer. https://doi.org/10.1007/978-0-387- 31, 161–175. https://doi.org/10.1111/j.0906-7590.2008.5203.x 21606-5 Pierson, T. W., McKee, A. M., Spear, S. F., Maerz, J. C., Camp, C. D., & Heyer, W. R., Donnelly, M. A., McDiarmid, R. W., Hayek, L.-A. C., & Fos- Glenn, T. C. (2016). Detection of an enigmatic plethodontid salaman- ter, M. S. (Eds.) (1994). Measuring and monitoring biological diversity: der using environmental DNA. Copeia, 104,78–82. https://doi.org/ Standard methods for amphibians. Washington, D.C: Smithsonian Insti- 10.1643/CH-14-202 tution Press. Pilliod, D. S., Goldberg, C. S., Arkle, R. S., & Waits, L. P. (2013). Estimating Hijmans, R. J., Phillips, S. J., Leathwick, J., & Elith, J. (2016). dismo: Spe- occupancy and abundance of stream amphibians using environmental cies Distribution Modeling R package version 1.1-1. Retrieved from DNA from filtered water samples. Canadian Journal of Fisheries and https://CRAN.R-project.org/package=dismo Aquatic Sciences, 70, 1123–1130. https://doi.org/10.1139/cjfas- Hoffmann, M., Hilton-Taylor, C., Angulo, A., Bohm,€ M., Brooks, T. M., 2013-0047 Butchart, S. H., ... Darwall, W. R. (2010). The impact of conservation Pilliod, D. S., Goldberg, C. S., Arkle, R. S., & Waits, L. P. (2014). Factors on the status of the world’s vertebrates. Science (New York, N.Y.), influencing detection of eDNA from a stream-dwelling amphibian. 330, 1503–1509. https://doi.org/10.1126/science.1194442 Molecular Ecology Resources, 14, 109–116. https://doi.org/10.1111/ IUCN (2016). IUCN Summary of number of species in each IUCN 1755-0998.12159 Red List Category by taxonomic class. Retrieved from http://www. Primack, R. B., & Sher, A. A. (2016). An introduction to conservation biol- iucnredlist.org/). ogy, 1st ed. Massachusetts, USA: Sinaure Associates Inc. Jerde, C. L., Mahon, A. R., Chadderton, W. L., & Lodge, D. M. (2011). R Core Team (2016). R: A language and environment for statistical comput- “Sight-unseen” detection of rare aquatic species using environmental ing. Vienna, Austria: R Foundation for Statistical Computing. DNA. Conservation Letters, 4, 150–157. https://doi.org/10.1111/j. Retrieved from https://www.R-project.org/. 1755-263X.2010.00158.x Renan, S., Speyer, E., Shahar, N., Gueta, T., Templeton, A. R., & Bar-david, Klymus, K. E., Richter, C. A., Chapman, D. C., & Paukert, C. (2015). Quan- S. (2012). A factorial design experiment as a pilot study for noninva- tification of eDNA shedding rates from invasive bighead sive genetic sampling. Molecular Ecology Resources, 12, 1040–1047. carp Hypophthalmichthys nobilis and silver carp Hypophthalmichthys https://doi.org/10.1111/j.1755-0998.2012.03170.x molitrix. Biological Conservation, 183,77–84. https://doi.org/10.1016/ Scheffers, B. R., Yong, D. L., Harris, J. B. C., Giam, X., & Sodhi, N. S. j.biocon.2014.11.020 (2011). The world’s rediscovered species: Back from the brink? PLoS Laramie, M. B., Pilliod, D. S., & Goldberg, C. S. (2015). Characterizing the ONE, 6, e22531. https://doi.org/10.1371/journal.pone.0022531 distribution of an endangered salmonid using environmental DNA Sigsgaard, E. E., Carl, H., Møller, P. R., & Thomsen, P. F. (2015). Monitor- analysis. Biological Conservation, 183,29–37. https://doi.org/10. ing the near-extinct European weather loach in Denmark based on 1016/j.biocon.2014.11.025 environmental DNA from water samples. Biological Conservation, 183, Liu, C. R., Berry, P. M., Dawson, T. P., & Pearson, R. G. (2005). Selecting 46–52. https://doi.org/10.1016/j.biocon.2014.11.023 thresholds of occurrence in the prediction of species distributions. Spear, S. F., Groves, J. D., Williams, L. A., & Waits, L. P. (2015). Using Ecography, 28, 385–393. https://doi.org/10.1111/j.0906-7590.2005. environmental DNA methods to improve detectability in a hellbender 03957.x (Cryptobranchus alleganiensis) monitoring program. Biological Conserva- Lobo, J. M., & Tognelli, M. F. (2011). Exploring the effects of quantity tion, 183,38–45. https://doi.org/10.1016/j.biocon.2014.11.016 and location of pseudo-absences and sampling biases on the perfor- Steinitz, H. (1955). Occurrence of Discoglossus nigriventer in Israel. Bulletin mance of distribution models with limited point occurrence data. of the Research Council Israel B, 5, 192–193. Journal for Nature Conservation, 19,1–7. https://doi.org/10.1016/j. Strickler, K. M., Fremier, A. K., & Goldberg, C. S. (2015). Quantifying jnc.2010.03.002 effects of UV-B, temperature, and pH on eDNA degradation in aqua- Lodge, D. M., Turner, C. R., Jerde, C. L., Barnes, M. A., Chadderton, L., tic microcosms. Biological Conservation, 183,85–92. https://doi.org/ Egan, S. P., ... Pfrender, M. E. (2012). Conservation in a cup of 10.1016/j.biocon.2014.11.038 water: Estimating biodiversity and population abundance from envi- Stuart, S. N., Chanson, J. S., Cox, N. A., Young, B. E., Rodrigues, A. S. L., ronmental DNA. Molecular Ecology, 21, 2555–2558. https://doi.org/ Fischman, D. L., & Waller, R. W. (2004). Status and trends of amphib- 10.1111/j.1365-294X.2012.05600.x ian declines and worldwide. Science, 306, 1783–1786. Mendelssohn, H., & Steinitz, H. (1943). A new frog from Palestine. https://doi.org/10.1126/science.1103538 Copeia, 4, 231–233. https://doi.org/10.2307/1438135 Taberlet, P., Coissac, E., Hajibabaei, M., & Rieseberg, L. H. (2012). Envi- Olden, J. D., Lawler, J. J., & Poff, N. L. (2008). Machine learning methods ronmental DNA. Molecular Ecology, 21, 1789–1793. https://doi.org/ without tears: A primer for ecologists. The Quarterly review of biology, 10.1111/j.1365-294X.2012.05542.x 83, 171–193. https://doi.org/10.1086/587826 Thomsen, P. F., Kielgast, J., Iversen, L. L., Wiuf, C., Rasmussen, M., Gil- Olson, Z. H., Briggler, J. T., & Williams, R. N. (2012). An eDNA approach bert, M. T., ... Willerslev, E. (2012). Monitoring endangered freshwa- to detect eastern hellbenders (Cryptobranchus a. alleganiensis) using ter biodiversity using environmental DNA. Molecular Ecology, 21, samples of water. Wildlife Research, 39, 629–636. 2565–2573. https://doi.org/10.1111/j.1365-294X.2011.05418.x Pellet, J., & Schmidt, B. R. (2005). Monitoring distribution using call sur- Waits, L. P., & Paetkau, D. (2005). Noninvasive genetic sampling tools for veys: Estimating site occupancy, detection probabilities and inferring wildlife biologists: A review of applications and recommendations for absence. Biological Conservation, 123,27–35. https://doi.org/10. accurate data collection. Journal of Wildlife Management, 69, 1419– 1016/j.biocon.2004.10.005 1433. https://doi.org/10.2193/0022-541X(2005)69[1419:NGSTFW] Perl, R. G. B., Gafny, S., Malka, Y., Renan, S., Woodhams, D., Rollins- 2.0.CO;2 Smith, L., ... Vences, M. (2017). Natural history and conservation of Wilcox, T. M., McKelvey, K. S., Young, M. K., Jane, S. F., Lowe, W. H., the rediscovered Hula painted frog, Latonia nigriventer. Contributions Whiteley, A. R., & Schwartz, M. K. (2013). Robust detection of rare to Zoology, 86,11–37. species using environmental DNA: The importance of primer 6812 | RENAN ET AL.

specificity. PLoS ONE, 8, e59520. https://doi.org/10.1371/journal. SUPPORTING INFORMATION pone.0059520 Willoughby, J. R., Wijayawardena, B. K., Sundaram, M., Swihart, R. K., & Additional Supporting Information may be found online in the sup- DeWoody, J. A. (2016). The importance of including imperfect detec- porting information tab for this article. tion models in eDNA experimental design. Molecular Ecology Resources, 16, 837–844. https://doi.org/10.1111/1755-0998.12531 Wisz, M. S., Hijmans, R. J., Li, J., Peterson, A. T., Graham, C. H., & Guisan, A. (2008). Effects of sample size on the performance of species distri- How to cite this article: Renan S, Gafny S, Perl RGB, et al. bution models. Diversity and Distributions, 14, 763–773. https://doi. Living quarters of a living fossil—Uncovering the current org/10.1111/j.1472-4642.2008.00482.x distribution pattern of the rediscovered Hula painted frog Zangari, F., Cimmaruta, R., & Nascetti, G. (2006). Genetic relationships of the western Mediterranean painted frogs based on allozymes and (Latonia nigriventer) using environmental DNA. Mol Ecol. mitochondrial markers: Evolutionary and taxonomic inferences 2017;26:6801–6812. https://doi.org/10.1111/mec.14420 (Amphibia, Anura, Discoglossidae). Biological Journal of the Linnean Society, 87, 515–536.