Acoustic and Morphological Differentiation in the Frog Allobates Femoralis: Relationships with the Upper Madeira River and Other Potential Geological Barriers

THE JOURNAL OF TROPICAL BIOLOGY AND CONSERVATION

BIOTROPICA *(*): ***–*** **** 10.1111/j.1744-7429.2008.00416.x

Acoustic and Morphological Differentiation in the Frog Allobates femoralis: Relationships with the Upper Madeira River and Other Potential Geological Barriers

Pedro Ivo Simoes˜ 1, Albertina P. Lima, William E. Magnusson Coordenac¸ao˜ de Pesquisas em Ecologia, Instituto Nacional de Pesquisas da Amazonia,ˆ Caixa Postal 478, CEP 69011-970, Manaus, Amazonas, Brazil

Walter Hodl¨ Department fur¨ Evolutionsbiologie, Fakultat¨ fur¨ Lebenswissenschaften, Universitat¨ Wien, Althanstrasse 14, 1090, Wien, Austria and

Adolfo Amezquita´ Departamento de Ciencias Biologicas,´ Universidad de Los Andes, CRA 1 # 18a 10, AA 4976, Bogota,´ Colombia

ABSTRACT

We studied patterns of call acoustics and external morphological differentiation in populations of the dart-poison frog Allobates femoralis occurring in forested areas along a 250-km stretch of the upper Madeira River, Brazil. Multivariate analyses of variance using principal components representing shared acoustic and morphological parameters distinguished three groups in relation to call structure and external morphology: (1) populations belonging to a two-note call morphotype; (2) populations with four-note calls inhabiting the left riverbank; and (3) populations with four-note calls inhabiting the right riverbank. Our results report a case of Amazonian anuran diversity hidden by current taxonomy and provide evidence for the upper Madeira River being a boundary between distinct populations of A. femoralis,and suggest a new taxonomic interpretation for these groups. Samples that did not ﬁt into the general differentiation pattern and the existence of a well-deﬁned contact zone between two morphotypes on the left riverbank indicate that mechanisms complementary to river-barrier hypotheses are necessary to explain the phenotypic differentiation between populations. Our study shows that at least one anuran species shows congruence between population differentiation and separation by a large Amazonian river, as documented for birds and mammals. Conservation efforts should not consider the taxon now known as A. femoralis as a homogeneous entity. There is much within-taxon variability, which can be probably explained partly by the existence of cryptic species, partly by geological barriers andpartofwhich currently has no obvious explanation.

Abstract in Portuguese is available at http://www.blackwell-synergy.com/loi/btp.

Key words: bioacoustics; Brazil; Dendrobatidade; external morphology; Madeira River; phenotypic differentiation; population distribution patterns.

MANY AMAZONIAN SPECIES HAVE THEIR DISTRIBUTIONS LIMITED by did not differ between riverbanks, suggesting that the JuruaRiver´ large Amazonian rivers, such as the Solimoes,˜ Negro, and Madeira does not prevent species crossing from one interfluve to another (Wallace 1852, Hellmayr 1910, Capparella 1987, Ayres & Clutton- (Gascon et al. 2000). In contrast to the Jurua,´ the Madeira River is Brock 1992, Patton et al. 2000, Hayes & Sewlal 2004). With a a distributional boundary to several primate and bird taxa (Wallace more comprehensive data base on Amazonian species occurrences, 1852; Hellmayr 1910; Haffer 1974; Capparella 1987; Roosmalen contemporary studies of species distributions and phylogeography et al. 1998, 2000), but studies of its importance as a geographic revealed a myriad of distribution patterns, some of which indi- boundary capable of maintaining population differentiation have cated population or species boundaries that are not coincident not been undertaken for other groups. with large rivers (Haffer 1997a). A considerable number of al- Our study aimed to evaluate the geographic patterns of call ternative hypotheses were proposed to explain the origin of those features and external morphology in the poison-dart frog Allobates distribution patterns and their maintenance through time (Haffer femoralis (Boulenger 1883) on both sides of a 250-km stretch of 1997b, Moritz et al. 2000). Such hypotheses are still controversial the upper Madeira River. Allobates femoralis is a ground-dwelling and largely debated, but it is reasonable to assume that no single diurnal dendrobatid frog, abundant in forested areas of the Madeira process affected all groups of organisms with the same intensity River, which feeds on a wide range of invertebrate taxa (Caldwell (Bush 1994). 1996). Males vocalize from sites slightly elevated from the forest Studies undertaken along the Jurua´ River, a meandering, 100- floor, such as logs and rocks. Advertisement calls are constituted to 400-m wide tributary of the Amazon, did not detect an effect by the repetition of groups of short, frequency-modulated notes, of the river on allozyme, external morphology, or mitochondrial used for territory defense against conspecific males and to attract DNA differentiation in anuran populations (Gascon et al. 1996, females. Courtship and oviposition occurs inside the male’s territory, 1998; Lougheed et al. 1999). Anuran-community composition also which may range from 0.25 to 26.2 m2. Females lay eggs on leaf nests and tadpoles are transported later by the male to a temporary Received 26 June 2007; revision accepted 11 December 2007. puddle. Territorial behavior results in calling sites being at least a few 1Corresponding author; e-mail: [email protected] meters apart from each other, allowing calls to be recorded without

C 2008 The Author(s) 1 Journal compilation C 2008 by The Association for Tropical Biology and Conservation 2Simoes,˜ Lima, Magnusson, Hodl,¨ and Amezquita´

distressing calling neighbors (Hodl¨ 1987; Roithmair 1992, 1994; Rodr´ıguez & Duellman 1994). The study of interpopulational variation on phenotypic traits, such as external morphology and call characteristics, is useful for evaluating divergence patterns, because populations of a single species presumably developed unique characters in relatively recent times. A more recent differentiation between groups of interest re- duces the probability that groups showing intermediate phenotypes have become extinct, and it is even possible that those populations still inhabit the location where the divergence process took place, providing unique opportunities for the study of the evolutionary mechanisms that triggered divergence (Foster 1999, Moritz et al. 2000). Considering their apparent restriction to forested environ- ments, and consequently reduced interpopulation gene ﬂow across ﬂoodplains (Gascon et al. 1998, Lougheed et al. 1999), populations of A. femoralis are potential models to test the role of the upper Madeira River in differentiation of anuran populations.

METHODS

STUDY AREA.—We sampled forests along the upper Madeira River, from the city of Porto Velho (8◦4334 S, 63◦5524 W) to the mouth of the Abuna˜ River, a western tributary (9◦4041 S, 65◦2558 W), in the State of Rondonia,ˆ Brazil. Along this stretch, the muddy Madeira River is not meandering and flows between nar- FIGURE 1. Location of study area within the State of Rondonia,ˆ Brazil, and row floodplains, limited by pronounced slopes. On the right bank, relative position of sampling sites along the upper Madeira river, from the city the landscape is a continuous flat plateau, originated from Pliopleis- ◦ ◦ of Porto Velho (8 43 34 S, 63 55 24 W) to the mouth of the Abunariver˜ tocenic sediments. On the left bank, the landscape is represented ◦ ◦ (9 40 41 S, 65 25 58 W). ‘N’ sampling sites correspond to areas where no by tabular formations of the same plateau and by fragments of a Allobates femoralis populations were found. White dots represent populations largely eroded plateau associated with pre-Cambrian sediments. The of the two-note call morphotype and black dots represent populations of the drainage system constituting this part of the Madeira River basin four-note morphotype. Relative position of the boundary between geological runs primarily through tectonic valleys. Slope ruptures attributed to domains was adapted from Souza Filho et al. (1999). tectonic faults along the river cause large rapids across the riverbed. Climate is characterized by well-defined wet and dry seasons, which directly influence water levels. On both riverbanks, vegetation is constituted by low rain forest, with open pioneer vegetation on the photypes of A. femoralis were known to occur. Each site was visited river margins (DNPM 1978, Souza Filho et al. 1999). It is not once in the same reproductive season, from November 2004 to known if plant composition and forest structure differ between ge- January 2005. ological domains or between riverbanks in the study area, but large Distance between sites was represented in analyses by the UTM deforested gaps are found on the right riverbank, near populated co-ordinates. As models used are additive, and each UTM co- localities (Porto Velho, Jaci-Parana,´ Mutum-Parana).´ ordinate represents an individual variable, distance was modeled as City-Block distance rather than straight-line Euclidean distance SAMPLING DESIGN.—Initially, we established 19 sampling sites in between the sites. However, as the system was nearly linear, the City- areas of terra firme forest along the upper Madeira River, ten on Block distances are highly correlated with the Euclidean distances. the left bank and nine on the right bank (Fig. 1). Populations of Euclidean distances were not used, as that would have required A. femoralis occur in patches in continuous-forest areas, so each Mantel-type analyses that would have confounded the effects of the sampling site was searched for vocally active populations. We found variables that are not easily represented as distances. no populations of A. femoralis in two of the sampling sites established on the right bank (corresponding to the localities of Abuna˜ ACOUSTIC DATA COLLECTION AND ANALYSES.—We chose recording and Jirau), both of which had sites apparently appropriate for the sites distant from roads and boat routes to prevent anthropogenic species, but no individuals. Therefore, we used samples from 17 sounds from affecting male calling behavior and recording quality. sites. Minimum and maximum distances between effective sam- Recordings were conducted during daytime, at 0600–1800 h. Ad- pling sites were of 1.3 and 168.0 km, respectively. We also searched vertisement calls of 206 A. femoralis males were recorded with a for contact zones along the left riverbank, where at least two mor- Sony WM-D6C tape recorder (2004, Sony Corr, Japan) and AKG Frog Acoustic and Morphological Differentiation 3

568 EB directional microphone (2003, AKG Acoustics GMBH, MORPHOLOGICAL DATA COLLECTION AND ANALYSES.—Males were Austria) positioned 1.0 m from calling males. For each individual collected after recording procedures, euthanized, and weighed after recorded, air temperature at the calling site at the time of record- the removal of the digestive tracts, because gut content represented ing was registered. For each recording session, we selected the three approximately 10 percent of total body mass in our samples (mean call samples that showed the least interference by environmental gut content mass was 0.16 ± 0.08 g for an average body mass of noise, starting from the most central calls. Thus, we avoided select- 1.68 ± 0. 27 g, considering all individuals collected). Our measure ing warm-up calls at the beginning of the calling bout, that vary of body mass therefore reflects long-term differences in morphology greatly in duration and frequency, and calls at the end of the bout, and not short-term differences due to the time an individual encoun- which could show temporal and spectral variations due to fatigue tered a prey item. For each of the 209 individuals used, 19 external (Gerhardt & Huber 2002). morphological variables were measured using a digital caliper and For each of the three call samples, we obtained temporal and a stereoscopic microscope with a graduated ocular lens (precisions spectral variables using sonograms generated by Raven 1.2. software 0.01 and 0.10 mm). All variables were obtained from the left side (Charif et al. 2004). Spectral analysis was done by a fast Fourier of the specimen. Means, standard deviations, and maximum and transform with a frequency resolution of 82 Hz and 2048 points. minimum values of all morphological variables for each sampling Because two acoustic morphotypes were found in the study area, site are presented in Table S2. Voucher specimens were deposited in the number of variables obtained for each group depended on the the Herpetology Collections of the Instituto Nacional de Pesquisas number of notes that constituted its advertisement calls. Means, da Amazonia,ˆ Manaus, Brazil, accession numbers INPA-H16541 standard deviations, and maximum and minimum values of all to INPA-H16823. acoustic variables for each sampling site are given in Table S1. The 19 morphological variables were log10 transformed for We used a principal component analysis (PCA) to reduce di- scale adjustment. The first and second components generated by a mensionality and produce a smaller number of independent acoustic PCA were included in a MANOVA model as dependent variables variables. The first and second components generated by the anal- to test whether the three groups identified in the acoustic analysis ysis were used as dependent variables in a multivariate analysis of presented significant morphological differences. variance (MANOVA) model, to test whether three groups defined Principal component analysis on external morphological vari- by the number of notes of advertisement calls and origin (i.e.,right ables usually produces a first principal component carrying heavy or left bank) were significantly different from each other in terms of size effects while the remaining components contain more informa- acoustic characteristics. Body weight was included as a covariate in tion on body shape. Therefore, because direct measurements usually the MANOVA, as body size can affect spectral features of anuran vary continuously with body size, individual body weight was in- calls (Ryan 1988, Keddy-Hector et al. 1992). Air temperature at the cluded in the MANOVA model as a covariate representing body time of recording was included as another covariate, as variation in size (Reis et al. 1990, Strauss & Bond 1990). All morphological and temperature can influence temporal features of advertisement calls acoustic statistical procedures were done in SYSTAT 8.0 (Wilkinson (Ryan 1988, Keddy-Hector et al. 1992, Gerhardt & Huber 2002). 1990). Sampling-site co-ordinates in UTM units were also included as covariates, to represent distance between sites (Legendre et al. 2002), because morphological and behavioral characteristics could vary RESULTS with geographic distance, independent of physical barriers. Early components produced by PCA usually recover important patterns Two distinct morphotypes of A. femoralis were found in the study of similarity between samples, while later components tend to ac- area. One has reddish posterior ventral coloration and advertisement cumulate variation related to noise (Gauch 1982). For the present calls constituted by two notes (Fig. 2A). This morphotype occurs analysis and for other analyses described below, we used the mini- only on the upper left riverbank, and has a contact zone with the mum number of components necessary to explain at least 50 percent other morphotype on the same bank, where there are no apparent of the total variation in our data sets. barriers to dispersal of individuals at the present time. The second Because two acoustic morphotypes were found in the study morphotype has black and white ventral coloration and advertise- area, the analysis described above was done twice. First, we used ment calls constituted by four notes (Fig. 2B,C). The distribution a set of nine variables obtained only from the first and second of this morphotype is split by the Madeira River (Fig. 1). notes of the advertisement calls of all individuals sampled. In both acoustic morphotypes, the first and second notes showed similar ACOUSTIC ANALYSES.—Principal component analysis using six spec- spectrographic patterns, with first note always narrowly modulated tral variables and two temporal variables present in the first and in frequency and short in duration, followed by a longer, broadly second notes produced a first and a second component explaining modulated second note. Therefore, we assumed first and second together 82.2 percent of the acoustic variation. The first component notes to be homologous in calls of both morphotypes. A second (PC1) explained 58.9 percent of the total acoustic variation, and analysis was carried out using 24 acoustic variables obtained from spectral variables had high loadings on that component. The second the four notes constituting the advertisement calls of one of the component (PC 2) explained 23.3 percent of the acoustic varia- morphotypes that inhabited both riverbanks, to test the acoustic tion, and temporal variables had high loadings on that component divergence between populations of a single acoustic morphotype (Table S3). Both components were used as dependent variables in living on opposite sides of the river. a MANOVA to test if acoustic characters were statistically different 4Simoes,˜ Lima, Magnusson, Hodl,¨ and Amezquita´

shorter notes in relation to four-note call populations from the left riverbank. It was expected that the two-note call morphotype on the left side of the river would have sufficiently different acoustic features to form a vocally distinct group. Thus, a second PCA was carried out, using only samples of the four-note call morphotype. Besides providing a larger number of acoustic variables (15 spectral and 9 temporal), the four-note call morphotype was present on both riverbanks, so the river effect on acoustic traits could be tested for a single morphotype. The first and second components explained 51.2 and 18.0 percent of the acoustic variation, respectively. Again, spectral variables had high loadings on PC 1 and temporal variables had high loadings on PC 2 (Table S4). MANOVA showed that the covariates body mass (Pillai trace = 0.293; P < 0.001, df = 140) and air temperature (Pillai trace = 0.227; P < 0.001, df = 140) significantly affected acoustic features of the four notes. Nonetheless, acoustic characteristics differed statistically between riverbanks, independent of covariate effects (Pillai trace = 0.099; P = 0.001, df = 140). Dis- tances between sampling sites did not influence call features (Pillai trace = 0.028; P = 0.410, df = 140). Distribution of sampling sites along the first and second PCA axes showed that samples from opposite riverbanks were separated mainly along PC1, which described FIGURE 2. Spectrographic patterns of advertisement calls of individuals of the the spectral variation of calls (Fig. 3), reflecting lower frequencies of two-note morphotype (A) and four-note morphotype recorded on the left (B) calls recorded on the right bank (Table S1). A PC3 was related to the and right (C) riverbanks. Calls were digitized from tape recordings using Raven variation in silent interval duration, but variation in this acoustic 1.2 software, at 22,050 Hz and 16 bits. trait occurred between individual sampling sites, and not between groups of sites defined by riverbanks. Variation explained by other components was never greater than expected for components constituted by random variables, among the three groups defined apriori: (1) two-note advertisement and inclusion increases the chance of type II errors. However, anal- call morphotype; (2) four-note advertisement call morphotype from yses using more components, not reported here, produced similar the left riverbank; and (3) four-note advertisement call morphotype qualitative results to analyses with two components. from the right riverbank. MANOVA indicated that body mass (Pillai trace = 0.225; MORPHOLOGICAL ANALYSES.—A principal component analysis of 19 P < 0.001, df = 192) and air temperature (Pillai trace = 0.043; transformed morphological variables generated a first component P = 0.015, df = 192) significantly affected the acoustic features explaining 43.9 percent of the external morphological variation. All of the two notes. However, the three groups defined aprioriwere variables used had positive loadings on that component. The second significantly different from each other in relation to acoustic char- principal component explained only 6.8 percent of the morpholog- acteristics, independent of covariate effects (Pillai trace = 0.354; ical variation. P < 0.001, df = 192). Geographic distance between sampling sites First and second principal components were used as depen- did not influence call features (Pillai trace = 0.035; P = 0.144, dent variables in a MANOVA to test for morphological differences df = 192). between the three groups as defined in the acoustic analysis. Body We plotted the variable means per site on PC1 and PC2, mass (Pillai trace = 0.507; P < 0.001, df = 199) and geographic which represented spectral and temporal characters, respectively. distance between sites (Pillai trace = 0.086; P = 0.002, df = 199) Populations from the right and left bank formed distinct groups had significant effects on morphological traits, but there were sta- along PC1 (Fig. 3) as a result of the lower frequencies of calls and tistically significant differences among groups (Pillai trace = 0.208; notes of the populations inhabiting the right bank (Table S1). Sam- P < 0.001, df = 199), after accounting for the effects of covariates. pling site No. 17 (St.Antonio-L), on the left bank, was closer in Groups were generally separated along the principal compo- multidimensional space to sampling sites on the right bank, show- nents, especially along PC1. As this component relates mostly to ing low values for spectral characteristics. This site was located ca body size, separation of populations on this axis is probably due to 30 km from another sampling site on the left bank, where in- larger individuals inhabiting the right riverbank (Table S2). Con- dividuals showed acoustic characteristics typical of that riverbank sidering only the sampling sites on the left riverbank, samples of (Fig. 3). Two-note morphotype populations (sites from no. 1–5) the two-note call morphotype tended to form a group distinct from tended to be different in temporal characteristics of calls, presenting the two-note morphotype along PC 2, which accounted mostly Frog Acoustic and Morphological Differentiation 5

FIGURE 3. Distribution of population means along the original principal components scores generated from acoustic variables of the two notes present in calls of both morphotypes (A) and variables obtained from the four notes of a single morphotype (B). Spectral divergence between populations of the right (R) and left (L) riverbanks were evident in both analyses. On the left riverbank, populations of the two-note call morphotype (in gray) signiﬁcantly differed in call characteristics from populations of the four-note call morphotype. See text for statistical comparisons between groups.

1987, Hayes & Sewlal 2004). However, the only studies evaluating river boundaries on anurans in Amazonian lowlands were restricted to the JuruaRiver,inBrazil(Gascon´ et al. 1996, 1998, 2000; Lougheed et al. 1999). Those studies found no evidence of differentiation between populations or communities of anurans on opposite sides of the river, but some authors argue that shifts in the river course and sediment deposit dynamics, typical of Amazonian meandering rivers, such as the Jurua,´ can cause terrain to move from one riverbank to another, passively transporting individuals between riverbanks from time to time and homogenizing populations via interbreeding (Ayres & Clutton-Brock 1992, Moritz et al. 2000). Our results clearly suggest that the upper Madeira River limits distinct populations of the poison-dart frog A. femoralis.The two-note call morphotype is restricted to the left riverbank and there is significant differentiation in acoustic and morphological traits between populations of the four-note morphotype inhabiting FIGURE 4. Distribution of population means along the original scores of the opposite riverbanks. In the study area, the Madeira River has an first and second principal components using external morphological variables. average width of 1.0 km and certainly represents a considerable PC1 accounts for great part of the morphological variation related to size. barrier to forest-specialist anurans. The river runs mostly along an Individuals taken from the right bank populations (R) were larger, on average, incisive tectonic valley, predominantly crystalline and geologically than individuals from the left riverbank (L). ancient (DNPM 1978), so it is improbable that changes of river- course orientation occurred in geologically recent times. A system of successive tectonic faults (altitude decreases from west to east) for head and limb measurements, except for sampling site no. 3, also results in a system of rapids along this section of the river, caus- Mutum-L (Fig. 4). ing water velocity to be particularly fast. Passive transportation of individuals from one riverbank to another by means of sedimentary island dynamics is unlikely due to slow rates of sediment deposit. DISCUSSION Although populations of the four-note call on opposite banks of the Madeira River had statistically significant differences, those Many studies suggested that large Amazonian rivers represent distri- were greatly influenced by size-related characteristics of morphology bution boundaries for species and distinct populations of primates and calls. Body size affects calls due to physical relationships between (Wallace 1852; Ayres & Clutton-Brock 1992; Peres et al. 1996; frequency wavelength and the dimensions of primary and secondary Roosmalen et al. 1998, 2000) and birds (Hellmayr 1910, Capparella oscillators (i.e., larynx, vocal cords, nostrils and vocal sacs; Gerhardt 6Simoes,˜ Lima, Magnusson, Hodl,¨ and Amezquita´

& Huber 2002). Therefore, even within the same species, bigger potential vicariant events associated with past tectonic phenomena. individuals tend to produce calls and notes with lower frequencies. Thus, at this time, it is not possible to determine whether there was In at least one case on the left riverbank (sample site no. 17, Sto. an effective vicariant barrier at the contact zone between A. femoralis Antonio-L),ˆ we found individuals as large as the ones found on the morphotypes or their divergence occurred in sympatry. right riverbank, resulting in pronounced acoustic and morphologi- Tectonic events have been previously suggested as possible cal variation between localities separated by only a few kilometers, factors generating genetic population divergence in A. femoralis. on the same side of the river. Such exceptions to geographic patterns Lougheed et al. (1999) suggested that factors other than the riverbed are sometimes hypothesized to result from rare migration events and could have generated patterns of genetic divergence in populations weak selection pressures after a phenotypic divergence pattern was along the Jurua´ River. The tectonic arch of Iquitos (a rocky crest developed under vicariance (Sokal & Oden 1978). Events like that formed during the genesis of the Andes), was found to cross the could occur downstream from the city of Porto Velho, where the Jurua´ River at a right angle and coincides with sampling sites show- Madeira River reaches the Amazonian plains and strong sedimen- ing high genetic divergence. The contact zone between A. femoralis tation produces extremely dynamic fluvial islands (DNPM 1978). morphotypes on the left bank of the Madeira River is not coinci- Thus, unfixed differences between populations of the four-note dent with any of the proposed locations of tectonic arches (Ras¨ anen¨ morphotype inhabiting opposite riverbanks could be a result of mi- 1990). Also, if a tectonic arch was responsible for the differentiation, gration of individuals across the downstream section, where there is evidence of a contact zone between different groups of populations greater potential for gene flow between riverbanks. should be present on the right riverbank, where only relatively uni- Although the upper Madeira River represents a clear bound- form populations of the four-note morphotype occur. ary between the two- and four-note call morphotypes in a NW– Haffer (1997a) described the occurrence of at least 20 contact SE direction, there is no obvious barrier between morphs on the zones between 40 bird taxa in Amazonia. Those zones occur in left riverbank, in a NE–SW direction. The contact zone between continuous rain forest areas and generally cross large Amazonian them coincides with the limits between two geomorphological do- rivers at right angles, providing a similar pattern to what is observed mains, evidenced on the surface as a flattened plateau downstream— for A. femoralis. He considered them as suture zones between taxa Planalto Rebaixado da Amazoniaˆ Ocidental—and by an eroded, older that diverged in forest refugia during the Pleistocene, supporting his plateau upstream—Planalto Dissecado Sul da Amazoniaˆ (DNPM refuge theory as an explanation for the existence of such lines, and 1978, Souza Filho et al. 1999). This boundary crosses the left bank discarding ecological factors as determinants of those zones. As we of the Madeira River almost at a right angle and causes a slope lack detailed information on microhabitat structure and vegetation rupture, forming a group of rapids, locally known as Cachoeira do composition in our study area, we cannot reject at this point that Jirau (Fig. 5). At the present time, the geomorphological contact an ecological mechanism of adaptation to a discrete environmental zone does not show any physical barriers that could prevent indi- gradient across geological domains (sensu Endler 1982) is the driving vidual A. femoralis from crossing. The recent tectonic history of the force that originated or maintains the A. femoralis contact zone. upper Madeira River is poorly known and little can be said about Our results provide scope for a taxonomic reinterpretation of two- and four-note call morphotypes based on their phenotypic differences, as it is highly probable that each morphotype represents a well-delimited evolutionary lineage. Ongoing phylogenetic and population genetic approaches will help to clarify their evolutionary relationships and evaluate current gene flow. Information provided by those analyses will also be used to test alternative diversification hypotheses (river-barrier, forest refugia, tectonic arches, among others) that can explain the current distribution pattern of phenotypic differentiation in A. femoralis along the Madeira River. Establishing the taxonomic limits between species is a key step for planning conservation strategies based on beta diversity. Cryptic anuran species are thought to be frequent among widespread taxa in the Amazon region, and lack of detailed data on variation within species is still a great impediment to the recognition of diversity at this level (Azevedo-Ramos & Gallati 2002). Mapping population variability within-species is also an important conservation tool to maximize potential adaptation and persistence of species in face of contemporary environmental changes. Our study reports that pop- FIGURE 5. The contact zone between Allobates femoralis morphotypes on the ulations included in what is thought to be a single anuran species left riverbank meets the riverbed almost at a right angle and coincides with the (Allobates femoralis) present significant phenotypic differences, part boundary between two geomorphological domains, represented by plateaus of of which is concordant with their division by a large Amazonian different origins. river. These findings have implications for both the planning of Frog Acoustic and Morphological Differentiation 7

amphibian conservation strategies in the Amazon and the manage- FOSTER, S. A. 1999. The geography of behaviour: An evolutionary perspective. ment of conservation units along the Madeira River. Trends Ecol. Evol. 14: 190–195. GASCON, C., S. C. LOUGHEED, AND J. P. BOGART. 1996. Genetic and morphological variation in Vanzolinius discodactylus: A test of the river hypothesis ACKNOWLEDGMENTS of speciation. Biotropica 28: 376–387. GASCON, C., S. C. LOUGHEED, AND J. P. BOGART. 1998. Patterns of genetic WethankK.Vulinec,J.Nicolas´ Urbina-Cardona, and one anony- population differentiation in four species of Amazonian frogs: A test of the Riverine Barrier Hypothesis. Biotropica 30: 104–119. mous reviewer for many helpful comments that improved this paper. GASCON, C., J. R. MALCOLM,J.L.PATTON,M.N.F.SILVA,J.P.BOGART,S.C. We thank R. Lima, E. Vasconcelos, A. Coelho, A. da Silva, F. Bar- LOUGHEED,C.A.PERES,S.NECKEL, AND P. T. B OAG. 2000. Riverine bosa, M. de Oliveira, C. de Miranda, and J. Rocha for helping barriers and the geographic distribution of Amazonian species. Proc. us in the field. Collecting permits were provided by RAN-IBAMA Natl. Acad. Sci. USA 97: 13672–13677. (license no. 131/04-RAN, p. 02010.003199/04-76). We thank L. GAUCH, H. G. 1982. Noise reduction by eigenvector ordinations. Ecology 63: 1643–1649. K. Erdtmann and S. Victoria Flechas for helping us with the sound GERHARDT, H. C., AND F. H UBER. 2002. Acoustic communication in insects analyses. Furnas Centrais Eletricas´ S. A. provided funding and logis- and anurans: Common problems and diverse solutions. University of tics for this project. Travel expenses of WH and E. Vasconcelos, and Chicago Press, Chicago, Illinois. flight tickets for PIS were provided by the Austrian Science Foun- HAFFER, J. 1974. Avian speciation in Tropical South America: With a system- dation (project FWF 15345-B03, proj. leader: WH). PIS received atic survey of the toucans (Ramphastidae) and jacamars (Galbulidae). Publications of the Nuttall Ornithological Club 14: 1–390. a fellowship from Brazilian CAPES during this study. HAFFER, J. 1997a. Contact zones between birds of southern Amazonia. Ornithol. Monogr. 48: 281–305. HAFFER, J. 1997b. Alternative models of vertebrate speciation in Amazonia: An overview. Biodivers. Conserv. 6: 451–476. SUPPLEMENTARY MATERIAL HAYES,F.E.,AND J. N. SEWLAL. 2004. The Amazon River as a dispersal barrier to passerine birds: effects of river width, habitat and taxonomy. J. Biogeogr. The following supplementary material for this article is available 31: 1809–1818. online at: www.blackwell-synergy.com/loi/btp HELLMAYR, C. E. 1910. The birds of the Rio Madeira. Novitates Zoologicae 17: 257–428. Table S1. Means, standard deviations and amplitude of acoustic HODL¨ , W.1987. Dendrobates femoralis (Dendrobatidae): A handly fellow for frog variables obtained from advertisement calls of Allobates femoralis. bioacoustics. In Proceedings of the Fourth Ordinary General Meeting of Table S2. Means, standard deviations and amplitude of body the Societas Europaea Herpetologica. Nijmegen, 1987, pp. 201–204. weight, snout-to-vent length and 19 other morphological variables ob- KEDDY-HECTOR, A. C., W. WILCZYNSKI, AND M. J. RYAN. 1992. Call patterns tained from Allobates femoralis males. and basilar papilla tuning in cricket frogs. II. Intrapopulation variation and allometry. Brain Behav. Evol. 39: 238–246. Table S3. Results of PCA using three temporal and six spectral LEGENDRE,P.,M.R.T.DALE,M.J.FORTIN,J.GUREVITCH,M.HOHN, AND D. acoustic variables from the first and second notes present in the adver- MYERS. 2002. The consequences of spatial structure for the design and tisement calls of both Allobates femoralis morphotypes. analysis of ecological field surveys. Ecography 25: 601–615. Table S4. Results of PCA using nine temporal and 18 spectral LOUGHEED, S. C., C. GASCON,D.A.JONES,J.P.BOGART, AND P. T. B OAG. variables obtained from the 4-note calls of the morphotype present on 1999. Ridges and rivers: A test of competing hypothesis of Amazonian diversification using a dart-poison frog (Epipedobates femoralis). Proc. R. both riverbanks. Soc. Lond. Ser. B 266: 1829–1835. MORITZ, C., J. L. PATTON,C.J.SCHNEIDER, AND T. B. S MITH. 2000. Diversi- fication of rainforest faunas: An integrated molecular approach. Annu. Rev. Ecol. Syst. 31: 533–563. LITERATURE CITED PATTON, J. L., M. N. F. SILVA, AND J. R. MALCOLM. 2000. Mammals of the Rio Jurua´ and the evolutionary and ecological diversification of Amazonia. AYRES,J.M.,AND T. H. C LUTTON-BROCK. 1992. River boundaries and species Bull. Am. Mus. Nat. Hist. 44: 1–305. range size in Amazonian primates. Am. Nat. 140: 531–537. PERES,C.A.,J.L.PATTON, AND M. N. F. SILVA. 1996. Riverine barriers and AZEVEDO-RAMOS, C., AND U. GALATTI. 2002. Patterns of amphibian diversity gene flow in Amazonian Saddle-Back Tamarins. Folia Primatologica 67: in Brazilian Amazonia: Conservation implications. Biol. Conserv. 103: 113–124. 103–111. RAS¨ ANEN¨ , M. E., J. S. SALO,H.JUNGNER, AND L. R. PITTMAN. 1990. Evolu- BUSH, M. B. 1994. Amazonian speciation: A necessarily complex model. J. tion of the Western Amazon lowland relief: Impact of Andean foreland Biogeogr. 21: 5–17. dynamics. Terra Nova 2: 320–332. CALDWELL, J. P. 1996. The evolution of myrmecophagy and its correlates in REIS,S.F.,L.M.PESSOAˆ , AND R. E. STRAUSS. 1990. Application of size-free poison frogs (Family Dendrobatidae). J. Zool. 240: 75–101. canonical discriminant analysis to studies of geographic differentiation. CAPPARELLA, A. P.1987. Effects of riverine barriers on genetic differentiation of Rev. Bras. Genetica´ 13: 509–520. Amazonian forest undergrowth birds. PhD Dissertation. Louisiana State RODRÍGUEZ,L.O.,AND W. E. D UELLMAN. 1994. Guide of the frogs of the Iquitos University. Baton Rouge, Louisiana. Region, Amazonian Peru. University of Kansas Publications [Natural CHARIF, R. A., C. W. CLARK, AND K. M. FRISTRUP. 2004. Raven 1.2 User’s History Museum -Special Publications, 22], Lawrence, Kansas. Manual. Cornell Laboratory of Ornithology, Ithaca, New York. ROITHMAIR, M. E. 1992. Territoriality and male mating success in the Dart- DNPM - DEPARTAMENTO NACIONAL DA PRODUÇAO˜ MINERAL –BRASIL. 1978. Poison Frog, Epipedobates femoralis (Dendrobatidae, Anura). Ethology Projeto RADAMBRASIL—Levantamento dos recursos naturais. DNPM, 92: 331–343. Rio de Janeiro, Brazil. ROITHMAIR, M. E. 1994. Field studies on reproductive behaviour in two dart- ENDLER, J. A. 1982. Problems in distinguishing historical from ecological factors poison frog species (Epipedobates femoralis, Epipedobates trivittatus)in in biogeography. Am. Zool. 22: 441–452. Amazonian Peru. Herpetol. J. 4: 77–85. 8Simoes,˜ Lima, Magnusson, Hodl,¨ and Amezquita´

ROOSMALEN,M.G.M.,T.ROOSMALEN,R.A.MITTERMEIER, AND G. Some biological implication and four applications of evolutionary and A. B. FONSECA. 1998. A new and distinctive species of mar- ecological interest. Biol. J. Linn. Soc. 10: 229–249. moset (Callitrichidae, Primates) from the lower river Aripuana,˜ State SOUZA FILHO,P.W.M.,M.L.E.S.QUADROS,J.E.SCANDOLARA,E.P. of Amazonas, central Brazilian Amazonia. Goeldiana Zoologia 22: SILVA FILHO, AND M. R. REIS. 1999. Compartimentaçao˜ morfoestru- 1–27. tural e neotectonicaˆ do sistema fluvial Guapore-Mamor´ e-Alto´ Madeira, ROOSMALEN,M.G.M.,T.ROOSMALEN,R.A.MITTERMEIER, AND G. A. B. Rondoniaˆ – Brasil. Rev. Bras. Geocienciasˆ 29: 469–476. FONSECA. 2000. A taxonomic review of the Titi Monkeys, genus Cal- STRAUSS,R.E.,AND C. E. BOND. 1990. Taxonomic methods: morphology. In licebus, Tomas, 1903, with the description of two new species, Cal- P.Moyle and C. Schreck (Eds.). Methods for fish biology, pp. 109–140. licebus bernhardi and Callicebus stephennashi, from Brazilian Amazonia. American Fisheries Society, Special Publication. Neotrop. Primates 10: 1–52. WALLACE, A. R. 1852. On the monkeys of the Amazon. Proc. Zool. Soc. Lond. RYAN, M. J. 1988. Energy, calling, and selection. Am. Zool. 28: 885– 20: 107–110. 898. WILKINSON, L. 1990. SYSTAT: The system for statistics. SYSTAT inc., Evanston, SOKAL,R.R.,AND N. L. ODEM. 1978. Spatial autocorrelation in biology 2. Illinois.