Introduction

There is a lot of ongoing debate regarding the explanation of plant and animal diversification in the Amazon basin and adjacent Guianas. Several historical biogeographic scenarios have been suggested (e.g. Haffer 1997, 2008; Hall and Harvey 2002; Noonan and Wray 2006). This paper focuses on the disturbance vicariance hypothesis (DV), which is described by Bush (1994), Noonan and Gaucher (2005) and Haffer (2008) derived from pollen analyses and patterns of species phylogenies. DV explains incomplete speciation in taxa on the eastern Guiana Shield due to relatively short phases of climate change during Pleistocene. During interglacials, cool-adapted species were retracted to higher elevations and allopatric speciation started, a process which was interrupted (‘disturbed’) as renewed glacials allowed for secondary contact via lowlands. Such a scenario, for instance, is suggested for caesalpinioid trees (Dutech et al. 2003) or bufonid and dendrobatid frogs (Noonan and Gaucher 2005, 2006). According to Bush (1994) and Noonan and Gaucher (2005), cool-adapted Guiana Shield taxa, which have undergone DV, are of Andean origin. This needs further explanation. Since Cenozoic, repeated phases of cool climate forced plant and animal taxa from the eastern Andean versant to occupy altitudinal ranges several hundred meters lower. Accordingly, diversity in the Amazon lowlands of coffee (Rubiaceae) or poison frogs (Dendrobatoidea) is explained, to give two examples recently studied (Antonelli et al. 2009; Santos et al. 2008). However, for a long time, eastward dispersal onto the eastern Guiana Shield was impossible as a result of marine incursions from the Caribbean Sea into western Amazonia (Lake Pebas). With further uplift of the Andes, this incursion vanished around the change from mid to late Miocene, 11–7 mya (e.g. Antonelli et al. 2009) and the Amazon River was born (Hoorn 2006). In the subsequent late Miocene climate, 5.4–9 mya (i.e. the South American Huayquerian), the Amazon has already entrenched to its today’s bed (Figueiredo et al. 2009). The climate was cooler than that of the current postglacial (i.e. Holocene) but not as cool as during glacial periods, allowing for extensive forest cover over Amazonia (Bush 1994). Only during this time span, cool-adapted Andean forest species were able to reach the eastern Guiana Shield (Fig. 1a). With warming during the subsequent Pliocene forest cover persisted, but persistence or dispersal of cool-adapted species would have been impossible (Bush 1994). Cool-adapted species in western Amazonia could easily respond to warming by restriction to higher elevations along the Andean versant. Likewise on the eastern Guiana Shield, cool-adapted species were retracted to the numerous existing hills. Vicariant speciation processes were thus initialized (Fig. 1b). With every Pleistocene glacial (starting only ca. 500,000 years BP), this retraction was ‘disturbed’ as renewed cooling allowed for lowland dispersal, as mentioned above (Fig. 1c–d). New dispersal from western Amazonia or re-dispersal from the eastern Guiana Shield deep into central Amazonia was impossible, as glacial cooling was stronger than that during the late Miocene accompanied by a reduction in precipitation of up to 20% (Bush 1994). As proposed further by Bush (1994), this resulted in forest loss leaving lowland forest fragments in western Amazonia along the Andean versant and on the eastern Guiana Shield plus vicinities only (Fig. 1c).

Fig. 1
figure 1

Schematic illustration of DV with Andes in the West (left) and the eastern Guiana Shield in the East (Guiana Shield species indicated in grey in phylogenetic trees): (a) cooling during the late Miocene allowed for dispersal (arrow in drawing, grey in phylogenetic tree) of cool-adapted forest taxa from the Andean versant eastward up to the eastern Guiana Shield; (b) due to subsequent warming (Pliocene) cool-adapted species were retracted to higher elevations (arrows) resulting in vicariant speciation; (c) speciation is ‘disturbed’ during Pleistocene glacials cool-adapted species as lowland dispersal was possible again (with limitations due to forest loss and the development of a western and an eastern forest fragment); (d) retraction to higher elevations and speciation continued during warmer interglacials. Scenario (d) was followed by (c) for several times. Scheme not to scale

Full size image

Harlequin frogs (Atelopus) are a species-rich bufonid genus of Andean origin, with more than 100 species occurring in forest or paramo habitats in the Andes (Lötters 1996; La Marca et al. 2005). In this paper we focus on the less than 10 Atelopus (depending on the taxonomy applied; see Lötters et al. 2002) occurring exclusively in forest habitats in the Amazon basin and on the eastern Guiana Shield. In an earlier molecular genetic study, Noonan and Gaucher (2005) showed that the five nominal species of the eastern Guiana Shield harlequin frogs are genetically little differentiated and that they apparently interbreed in nature. Supported by divergence time estimates, these authors advocated that the observed phylogeographical patterns in Atelopus fit DV predictions, i.e. that a single Andean ancestor had invaded the eastern Guiana Shield (likely in late Miocene, as also suggested for other anuran amphibians; Santos et al. 2008) and has started speciation there in the Pleistocene due to the alteration of glacial and interglacial phases (as illustrated in Fig. 1a–d).

To their molecular phylogeny, Noonan and Gaucher (2005) added only four Atelopus species from outside the eastern Guiana Shield. As a result, the validity of their study is pending on additional corroboration. This is especially significant because our knowledge on the current distribution of harlequin frogs in central Amazonia is poorly understood. Lescure and Gasc (1986), with providing data, proposed a continuous distribution of harlequin frogs from the Andes to the eastern Guiana Shield. In contrast, Lötters et al. (2002), in a taxonomic study, were unable to trace Atelopus material in scientific collections from a large part of central Amazonia, casting some doubt on a continuous distribution. Such a hiatus could be well explained by DV predictions, since the recolonisation of central Amazonia, either from the western Amazonian lowlands or from the eastern Guiana Shield plus vicinities, would be impossible during the current postglacial. From a phylogenetic point of view, according to DV predictions and the findings of Noonan and Gaucher (2005), we expect that harlequin frogs from east of this suspected distribution gap in central Amazonia constitute one clade nested within those from the Andes and Amazonian lowlands adjacent to them (Fig. 1d) if more species were included from more of the genus’ entire geographic range than available to Noonan and Gaucher (2005).

Species can respond to climate change in two ways. One is change of geographic range (i.e. increase, decrease down to extinction, shift) and maintenance of the specific climate envelope, termed niche conservancy (e.g. Peterson et al. 1999; Wiens and Graham 2005). In the other the geographical range is retained, necessarily accompanied by climate envelope shift (e.g. Broennimann et al. 2007; Pearman et al. 2007; Rödder et al. 2009). Species climate envelope predictions have never been formulated with regard to DV. According to our understanding of DV, we largely expect climate conservancy in Amazonian and Guianan Atelopus as, under DV, species change their geographic ranges as a response to a changing climate (Fig. 1a–d). Vertical range shift of cool-adapted species along the Andean versant was up to 800 m (Bush 1994). However, maximum altitudes found on the eastern Guiana Shield have been about 300 m above today’s sea level only. As niche shift is facilitated in small populations pushed to their margin of environmental tolerance (Holt and Gomulkiewicz 2004; Holt et al. 2005; Jakob et al. 2010), it may be assumed that within the eastern glacial forest fragment (Fig. 1c) climate envelopes have shifted in those cool-adapted species which have survived warmer periods. As a consequence, when comparing current-day Atelopus populations from the western and eastern Amazonian (including the eastern Guiana Shield) lowlands (Fig. 1c) their climate envelopes under today’s macroclimate are predicted to show some divergence.

The contemporary postglacial was warmest about 8,000–4,500 years BP and temperature has decreased since then. According to DV, harlequin frog species should currently be able to re-expand their distributions into lower areas. When mapping climate envelopes of current-day Atelopus populations from both western and eastern Amazonia under macroclimatic conditions into geographic space, they should range into central Amazonia. However, because of the expected climate envelope shift in eastern Amazonian Atelopus, mapped climate envelopes (which can be understood as species’ potential distributions) are predicted to be rather allopatric than sympatric.

In this paper we combined different methodological approaches to study (i) if extant harlequin frogs display a central Amazonian distribution gap; (ii) if eastern Amazonian Atelopus constitute a single clade nested in a phylogeny comprising an enlarged data set from the Andes and adjacent lowlands; (iii) if climate envelopes of western versus eastern Amazonian populations (i.e., geographically well delimitated by a natural central Amazonian distribution gap) are divergent under today’s macroclimate; (iv) if allopatry is the result rather than sympatry when mapping these climate envelopes into geographic space. We discuss in how far our result reinforce and expand DV predictions.

Methods

A central Amazonian distribution gap

In order to determine the extant distribution of Atelopus in Amazonia, 87 presence data points from all over Amazonia were employed in this study (Fig. 2). They were taken from published references and obtained through interviews with seven experts (see Appendix). Interviews were open, non-standardized, as described by Atteslander (2008). Only acknowledged experts in anuran taxonomy and with field experience in Amazonia or the eastern Guiana Shield were interviewed. Additional presence data were taken from scientific collections. As an altitudinal limit for pre-Andean/western Amazonia we chose 800 m above sea level, the approximate upper border of the tierra caliente lowlands. Latitude and longitude coordinates for presence data points were obtained from the sources listed in the Appendix. If not provided, they were obtained through the Alexandria Digital Library Gazetteer (Hill and Zheng 1999; http://www.alexandria.ucsb.edu/gazetteer).

Fig. 2
figure 2

Northern South America showing data points of presence (grey and coloured circles) and apparent absence (open circles) of harlequin frogs in Amazonia (see Appendix). Colours refer to presence points of Amazonian taxa processed in the phylogeny. (Color figure online)

Full size image

In addition, 42 data points of apparent absence of harlequin frogs, illustrated in Fig. 2 (see Appendix), were obtained from published references and expert interviews as described above. We only included data points at elevations ≤800 m above sea level and situated in an area defined through a Minimum Convex Polygon (MCP) for all presence data, created with DIVA-GIS 5.4.

We are aware that absence is nearly impossible to prove and should be handled with caution; therefore, we independently analysed presence and absence information. For this, Ripley’s K function, a multi-distance spatial cluster analysis, was used to independently study spatial dependence in both data sets (Fig. 2) by comparison to a random pattern, which follows a Poisson distribution (Ripley 1977; Haase 1995). If the K function of the data differs significantly from that of the random distribution, data points under study are clustered (i.e. aggregated, when above that of the random distribution) or highly dispersed (i.e. when below random expectation). Analysis was performed with the Spatial Statistics (confidence envelope: 99 permutations) tool box of ArcGIS Desktop 9.2 (ESRI; http://www.esri.com).

Nested monophyly of eastern Amazonian Atelopus

Noonan and Gaucher (2005) based their study on fragments of the mitochondrial genes cyt b and ND2. We here chose a fragment of the mitochondrial 16S rRNA gene for two reasons. First, this locus is a widely used marker in amphibian systematics, especially suitable because of strong constancy of priming sites and information content at the species level (e.g. Vences et al. 2005). Second, the use of 16S allowed us to maximize the species sample size in order to study nested monophyly of eastern Amazonian harlequin frogs. As listed in Table 1, sequences of nine Atelopus (three outgroup species) were available from GenBank (http://www.ncbi.nlm.nih.gov; Benson et al. 2004). We supplemented these data by sequencing 16S for 11 additional Atelopus plus four outgroup taxa (Table 1).

Table 1 List of species used for the molecular analysis, their GenBank accession numbers and origin
Full size table

DNA was extracted from toe clips. Tissue samples (stored in 99% ethanol) were digested using proteinase K (final concentration 1 mg/mL), homogenised and subsequently purified following a high-salt extraction protocol (Bruford et al. 1992). Polymerase chain reaction (PCR) primers for the fragment of the 16S rRNA gene were 16SA-L and 16SB-H of Palumbi et al. (1991), used as in Van der Meijden et al. (2007). PCR products were purified via spin columns (Qiagen). Sequencing was performed directly using the corresponding PCR primers. New sequences were combined with existing sequences taken from GenBank in the final dataset containing 27 taxa including bufonid and non-bufonid outgroups (Table 1). Sequences were aligned using ClustalW (Thompson et al. 1994) and subsequently edited by hand. The final alignment contained a total of 570 positions of which 219 were variable and 136 were parsimony-informative.

Phylogeny reconstruction was performed using Maximum Likelihood (ML) and Bayesian Inference (BI) methods. Gaps were treated as unknown characters. The best fitting models of sequence evolution were determined by the AIC criterion as implemented in Modeltest 3.06 (Posada and Crandall 1998). ML tree searches were performed using PhyML, version 2.4.4 (Guindon and Gascuel 2003). Bootstrap branch support values were calculated with 200 replicates. The Bayesian analyses of the combined and separate datasets was conducted with MrBayes 2.0 (Huelsenbeck and Ronquist 2001) with 2 million generations, sampling trees every 10th generation and calculating a consensus tree after omitting the first 200,000 trees as burn-in determined with the Tracer 1.4 (http://beast.bio.ed.ac.uk/Tracer). No well supported topological differences were found between the BI and ML trees; the ML tree was used in the subsequent analysis.

Divergence in climate envelopes and allopatry

Climate envelopes for western and eastern Amazonian Atelopus were modelled, subsequently mapped into geographic space and compared. For our approach we used the presence data points listed in the Appendix (30 for all western and 54 for all eastern Amazonian Atelopus; Fig. 2). We created models based on seven macroscale bioclimatic parameters (Table 2) describing the availability of thermal energy and water, widely used in climate envelope models (e.g. Carnaval and Moritz 2008; Rödder and Lötters 2009). Using DIVA-GIS 5.4 (Hijmans et al. 2001), bioclimatic parameters were extracted from the WorldClim 1.4 interpolation model with grid cell resolution 2.5 min for the period 1950–2000 (Hijmans et al. 2005) at (i) the species records as well as (ii) at 1,000 random points within both the MCP of the western and eastern Atelopus presence. For comparison, we computed boxplots with XLSTAT 2009 (Addinsoft). Subsequently, climate envelope models were generated and mapped with MaxEnt 3.2.19 (Phillips et al. 2006) based on the principle of maximum entropy (Jaynes 1957). This approach yields more reliable results than comparable methods (e.g. Elith et al. 2006; Heikkinen et al. 2006; Wisz et al. 2008), especially when data points for species number relatively few (e.g. Hernandez et al. 2006). Using default settings, 25% of the data points were randomly reserved for model testing (duplicate presence records in one grid cell were automatically removed). Prediction accuracy was evaluated through threshold-independent receiver operating characteristic (ROC) curves and the calculation of the area under the curve (AUC) method (e.g. Hanley and McNeil 1982). We acknowledge that there is currently some discussion about the suitability of AUC (Lobo et al. 2008). However, for our application AUC is the best possible choice. Elith and Graham (2009) pointed out that none of the frequently applied statistics in AUC are misleading and that appropriate statistics relevant to the application of the model need to be selected. The logistic MaxEnt output was chosen which is continuous and linear scaled (0–1, with 0.1 being the minimum Maxent value at the training records already suggesting suitability to the species under study; Phillips et al. 2006).

Table 2 AUC values per model, climate envelope overlap in terms of I and D values and assessment of their similarity and equivalency via randomization tests (see text)
Full size table

We quantitatively compared climate envelopes of western and eastern Amazonian Atelopus with Schoener’s index (D) and Hellinger distance (I) as modified by Warren et al. (2008). Both indices allow for testing climate envelope similarity between two probability distributions of (e.g. climate envelope) distributions over geographic space, whereby D and I values range from 0 to 1 (i.e. models have no to entire overlap). We evaluated the significance of D and I values with null models regarding climate envelope similarity and equivalency representing two extremes within the spectrum of niche conservatism (Warren et al. 2008). Tests were performed separately for each bioclimatic parameter in the manner of Rödder and Lötters (2009). Moreover, for climate envelope equivalency, we applied a randomization test as proposed by Warren et al. (2008) which relies on the metrics D and I. For western and eastern Amazonian harlequin frog occurrences 100 pseudoreplicate datasets were created by randomly partitioning the combined number of western and eastern occurrences into sets of the same size of the original of western and eastern datasets. Climate envelope models were built from each pseudoreplicate in order to generate null distributions. The overlap between models computed with the original data sets were compared to the percentiles of these null distributions in a one-tailed test to evaluate the hypothesis that climate envelope models for western and eastern records were not significantly different. This test allows for an assessment of climate envelope maintenance (i.e. niche conservancy) in a strict sense, i.e. the effective equivalency of the climate envelope in the western and eastern geographic ranges. It is expected to be only met if western and eastern harlequin frogs tolerate exactly the same set of climatic conditions and have the same set of environmental conditions available to them.

In order to assess climate envelope similarity, we again used a randomization test of Warren et al. (2008). It compares the actual similarity of climate envelopes in terms of D and I values to the distribution of similarities obtained by comparing them to a climate envelope model created through randomly choosing cells from among the cells in the study area. The same procedure was performed in both directions (western to eastern records vice versa) 100 times to construct an expected distribution of D and I values between a climate envelope model generated using actual occurrences and another one generated from random background data points extracted within a MCP enclosing one set of records. These null distributions served as a two-tailed test to assess the null hypothesis that measured climate envelope overlap between western and eastern Amazonian Atelopus is explained by regional similarities or differences in available habitat (‘background effects’). This hypothesis is rejected if the actual similarity falls outside the 95% confidence limits of the null distribution suggesting active habitat choice. Significantly higher values suggest that climate envelopes are more similar than expected by chance and lower values indicate greater differences. Computations of D, I, climate envelope similarity and equivalency were performed with a Perl script developed by Warren et al. (2008).

Results and discussion

A central Amazonian distribution gap

Figure 2 suggests that indeed Amazonian harlequin frogs display a distribution gap in central Amazonia. Ripley’s K function for presence data points revealed that they are above the function for randomly distributed points (Fig. 3), i.e. that the presence data are significantly clustered. Clustered presence data points endorse that a distribution gap exists, excluding the possibility that this pattern is caused by different sampling efforts in different areas, however. With respect to data of apparent absence, we acknowledge that it has to be regarded with care. Interpreting them under Ripley’s K function, they fall within the confidence intervals of a random distribution (Fig. 3). This lets us tentatively conclude that it is unlikely that limited sampling efforts can be made responsible for the distribution gap identifiable in Fig. 2.

Fig. 3
figure 3

Ripley’s K functions showing that presence data points (left) are significantly inhomogeneous (i.e. clustered) while apparent absence data points are homogeneously distributed (compare Fig. 2). Bold black line: expected K function with lower and upper confidence envelopes (dashed), bold grey line: observed K function

Full size image

The existence of a natural distribution gap is expectable under DV (Fig. 1c) and therefore reinforces our hypothesis of Amazonian harlequin frog historical biogeography. However, it needs to be noted that this explanation for the observed geographic pattern is a single possibility out of many possible causes. A gap alone leaves also space for other explanations than DV.

Nested monophyly of eastern Amazonian Atelopus

Figure 4 illustrates a ML phylogram for 20 harlequin frogs and outgroups. All Amazonian Atelopus comprise a well supported monophyletic lineage, which is sister to all other members in the genus (i.e. a combination of Andean and trans-Andean species; Table 1). The species included in this group are the only Atelopus showing bright pink, violet or red coloured hand and foot soles, perhaps representing a synapomorphy (e.g. Lötters 1996; Lötters et al. 2002). Moreover, the only harlequin frogs known to possess a middle ear are included in this group (absent in most members of the genus; Lötters 1996). However, not all species used in our phylogeny have been studied for ear ossicle conditions, so that phylogenetic information can only be expected here (Fig. 4). Within this Amazonian clade, two sub-clades are evident, supported by high bootstrap and Bayesian posterior probability values. One includes the species from central to southern Peru and Bolivia, i.e. an A. tricolor-clade (see Fig. 4). The other is comprised of all studied species from the upper portion of the Amazon River plus the eastern Guiana Shield and the portion of the Amazon basin adjacent to it. Our data strongly support the eastern Guiana Shield Atelopus forming a monophyletic subset of this clade. Similar to the results of Noonan and Gaucher (2005), Guianan Atelopus are little differentiated, as reflected by the weak support of groupings among them. Our findings fully support Noonan and Gaucher (2005) who suggested that DV predictions are well applicable to harlequin frogs.

Fig. 4
figure 4

ML phylogram of different Atelopus species from all over the genus’ range (Table 1) based on the mitochondrial 16S rRNA gene showing that Amazonian Atelopus constitute a monophyletic unit with those from the eastern Guiana Shield nested within them. Numbers above branches indicate Maximum Likelihood bootstrap support/Bayesian posterior probabilities values. Species names are accompanied by GenBank accession numbers. This tree was rooted with Eleutherodactylus cf. johnstonei (not shown). It is also indicated in the Atelopus species if presence (*) or absence (**) of a middle ear is known

Full size image

Atelopus species from the Venezuelan Andes and the Caribbean coastal range, i.e. proximate to the Guiana Shield, show osteological and external morphological characters suggesting a closer relationship to Colombian Andean taxa (McDiarmid 1971). However, we lack other characters, such as those from molecular phylogenetics studies, to validate or dispose this view.

Divergence in climate envelopes and allopatry

Prediction accuracy of MaxEnt climate envelope models was high as suggested by ‘excellent’ AUC values (western Amazonian Atelopus: test 0.955, training 0.980; eastern Amazonian Atelopus: test 0.979, training 0.985) following the AUC classification accuracy of Swets (1988).

Comparing box plots (Fig. 5), the available climate space as well as climate envelopes of western and eastern Amazonian Atelopus are similar as ranges of all bioclimatic parameters in our modelling approach largely overlap. Two of the temperature parameters, ‘annual mean temperature’ and ‘maximum temperature of the warmest month’, are rather alike (i.e. not significantly different; Table 2), while one temperature and two of the precipitation parameters, ‘mean monthly temperature range’, ‘precipitation of the wettest month’ and ‘precipitation of the driest month’, show significant divergence (Table 2). These observations are highly coincident with the D and I values, which characterize the climate envelope overlap (Table 2). The niche identity tests revealed that the climate envelopes of eastern and western harlequin frogs were identical in terms of annual means of temperature and precipitation. The null hypothesis that climate envelopes are equivalent in the western and eastern ranges was rejected for all other parameters. The climate envelope similarity test revealed that overlap in the ‘annual mean temperature’ and the ‘maximum temperature of the warmest month’ can be most likely traced back to active habitat choice. These findings corroborate our expectation that climate envelopes of western and eastern Amazonian harlequin frogs show some divergence. However, background effects (i.e. wide availability of suitable climate conditions) may at least partly explain the overlap observed for the other parameters. Whereas eastern Amazonian Atelopus actively chose their habitats according to some climate components which are only limitedly available to them, these same climate components may be widely available within the range of western Atelopus, where other components may be actually limiting. Such patterns are reasonable since different parameters may be widely available or limiting in eastern or western ranges influencing habitat choice. Hence, our findings suggest once cool-adapted Atelopus ancestors, under warm conditions, were forced to change climate envelopes.

Fig. 5
figure 5

Box plots of seven bioclimatic parameters in climate envelope models of western (W) and eastern Amazonian Atelopus (E) and available climate space within MCPs (W BAC; E BAC). Values given in the upper row refer to temperature in °C and those in the lower row refer to precipitation in mm. Broad horizontal bars indicate the first and third quartiles as well as the median. Short horizontal bars indicate minimum/maximum values while dots do represent extremes outside 95% confidence intervals. Mean values are indicated by crosses

Full size image

Because ‘excellent’ AUC values suggest a high prediction accuracy (see above), we mapped climate envelope of western and eastern Amazonian Atelopus into geographic space on the full presence data point sets (i.e. this time no data points were set aside for testing). Doing so, it is possible to take advantage of all available information and to provide best estimated prediction maps (see Phillips et al. 2006). Results are shown in Fig. 6. Fitting well with the comparison of the climate envelops of the two units studied (Fig. 5; Table 2), their geographic distributions are largely allopatric with overlap corresponding to lower suitability (i.e. lower MaxEnt values). Areas of higher suitability of climate envelopes (i.e. warmer colours in Fig. 6) of western and eastern Amazonian Atelopus show little overlap.

Fig. 6
figure 6

Potential distributions of western (a) and eastern (b) Amazonian Atelopus. Colours from green via yellow to red refer to MaxEnt values of probability with warmer colours standing for areas with better predicted conditions (range 0–1, logistic MaxEnt output). Illustrations were performed with DIVA-GIS 5.4. (Color figure online)

Full size image

Conclusion

We provide molecular phylogenetic evidence that all Amazonian Atelopus constitute a monophyletic group and find support that a natural distribution gap in central Amazonia for these amphibians exists. Harlequin frogs from east of this gap are a monophyletic subset, suggesting that they have derived from a single ancestral stock which subsequently has started vicariant speciation. Our findings corroborate the results of Noonan and Gaucher (2005). These authors advocated that DV predictions are met in Amazonian and in particular eastern Guiana Shield Atelopus. We here demonstrate that DV predictions are also met when genetic sampling is expanded by inclusion of more species from the entire genus’ distribution.

The justified spatial breakup into western and eastern Amazonian groups afforded us for the first time to derive DV predictions regarding climate envelope change in taxa of Andean origin. These predictions were met, as we were able to show that climate envelopes of both groups were similar regarding some parameters but that other parameters significantly differed. These different parameters result in allopatric potential distributions of western and eastern Amazonian Atelopus. Geographic range shift does not strictly result in climate envelope change, as commonly species tend to change their distributions with changing climate being bound to physiological constrains hampering climate envelope shifts regarding some parameters (e.g. Parmesan 2006). Because of the limited elevational range in the eastern Guiana Shield, cool-adapted taxa facing extinction risk were forced with a strong selective pressure to change their climate envelopes. We suggest that this is a prediction which is generally applicable to Andean species under DV.