Molecular systematics, species limits, and diversification of the genus Dendrocolaptes (Aves: Furnariidae): Insights on biotic exchanges between dry and humid forest types in the Neotropics

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. Journal of Zoological Systematics and Evolutionary Research published by Blackwell Verlag GmbH Contributing authors: Antonita Santana (antonitasantana@hotmail.com); Sofia Marques Silva (sofiamarques1@gmail.com); Romina Batista (rominassbatista@gmail.com); Iracilda Sampaio (ira@ufpa.br) 1Programa de Pós-Graduação em Biologia Ambiental, Universidade Federal do Pará, Bragança, Brazil 2Research Center in Biodiversity and Genetic Resources/InBIO Associate Laboratory, Vairão, Portugal 3Coordenação de Zoologia, Museu Paraense Emílio Goeldi, Belém, Brazil 4Instituto Nacional de Pesquisas da Amazônia (INPA) Campus II, Manaus, Brazil 5Gothenburg Global Biodiversity Centre, Gothenburg, Sweden 6Finnish Museum of Natural History, University of Helsinki, Helsinki, Finland


| INTRODUC TI ON
Both dry and humid forest vegetation types account for a significant portion of the Neotropical biomes, but it remains less clear in which way and to what extent diversification processes are coupled between these two distinct habitats (Capurucho, Ashley, Ribas, & Bates, 2018;Garcia-Moreno & Silva, 1997;Silva, 1996, but see Antonelli et al., 2018). For instance, drier climatic conditions typically associated with glacial maxima are thought to promote the expansion of dry biomes, while at the same time causing the fragmentation and reduction in humid forest cover (Hooghiemstra & van der Hammen, 1998;Werneck, 2011). Therefore, historical factors such as extreme variations in climate may have affected dry and humid Neotropical forests synchronously, albeit in very contrasting ways (Silva et al., 2019;Werneck, Nogueira, Colli, & Costa, 2012).
Current evidence points toward the existence of many lineages that have long evolved separately in dry and humid forest types throughout the Neotropics, with only a small number of taxa known to occur indistinctly in both habitats (Capurucho et al., 2018;Fenker et al., 2020;Porzecanski & Cracraft, 2005;Sousa-Neto, Cianciaruso, & Collevatti, 2016). Lineages occurring in both dry and humid forest types are the best targets for evaluating any connections between these habitats because they can provide both spatial resolution and temporal clues on comparative diversification processes within both vegetation types. These investigations are of paramount importance because climate change may not only increase or reduce the area covered by each dry and humid forest types, but also shift their mutual ranges through time, hence providing multiple opportunities for close geographic contact and biotic exchange between them (Capurucho et al., 2018;Oliveras & Mahli, 2016;Sousa-Neto et al., 2016).
Here, we provide the first phylogenies for Dendrocolaptes with dense taxonomic sampling and use them to evaluate species limits within the genus, as well as reconstruct its diversification through time. Specifically, we evaluated whether past changes in the extent and distribution of dry and humid forest types influenced the genus' evolutionary history. If biotic exchanges between Neotropical dry and humid forest types were important events through time, we anticipate that Dendrocolaptes sister lineages will often replace each other across these habitats. In contrast, if biotic exchange was limited, dry and humid forest Dendrocolaptes lineages would have evolved separately within each biome for the most part of their evolutionary history. Finally, we contrast our results with previous taxonomic work on Dendrocolaptes and make recommendations concerning the classification of its taxa as either species or subspecies.
The success of each amplification was confirmed through electrophoresis in 1% agarose gel. Positive PCR amplicons (Table S1) were purified with polyethylene glycol (

TA B L E 1 (Continued)
ABI PRISM BigDye Terminator Cycle sequencing protocol (Applied Biosystems ® ). Sequencing products were run in an ABI PRISM 3130 Genetic Analyzer (Applied Biosystems ® ).
Sequence chromatograms were edited using the software Geneious 9.1.2 (http://www.genei ous.com; Kearse et al., 2012) and aligned in BioEdit 7.2.6.1, using the ClustalW algorithm (Hall, 1999;Thompson, Higgins, & Gibson, 1994). Heterozygous sites for the nuclear genes were confirmed by the presence of double peaks in both complementary strands of DNA and coded according to IUPAC codes. All DNA sequences generated were deposited on GenBank (Table 1; see also Alignments S1-S6).

| Phylogenetic analyses and divergence time estimates
Two distinct phylogenies were estimated using Bayesian inference  Table S2). Three independent runs were made for a total of 10 7 generations with parameters sampled every 1,000 generations and a 25% burn-in. The multilocus concatenated phylogeny was used as a guide tree to test species boundaries based on coalescent methods (see below). To this end, statistically wellsupported (PP ≥ 0.95) reciprocally monophyletic groups recovered in this analysis that were also geographically structured and morphologically diagnosable according to previous analyses (Marantz, 1997;Marantz & Patten, 2010) were assumed as hypothesized species.
We used the algorithm *BEAST implemented in BEAST 1.8.0 (Drummond, Suchard, Xie, & Rambautt, 2012) to generate a species tree (ST) and estimate divergence times between the clades supported by previous analyses. This algorithm was also run in CIPRES (Miller et al., 2010; www.phylo.org/porta l/). We assumed an uncorrelated lognormal molecular clock model for each locus and applied a calibration derived from the Cytb mutation rate estimated as 0.0105 substitutions/ million years (SD = 0.0034) (Weir & Schluter, 2008). Mutation rates for the other genes were esti-

| Species delimitation
To investigate interspecific boundaries among lineages of the genus Dendrocolaptes, we used the Bayesian Phylogenetics and Phylogeography (BP&P) software 3.3 (Yang, 2015). BP&P uses a multilocus coalescence-based Bayesian modeling approach, inferring phylogenetic relationships and describing the likelihood of independence between lineages (Yang, 2015). We used the ST topology as our guide tree in BP&P, which included 15 reciprocally monophyletic lineages in total. After an initial test evaluating different parameters, we implemented Yang's (2015) approach

| Historical biogeography reconstruction
To reconstruct the geographical ancestral history of the genus
Within the genus, two main highly supported clades were present: one grouping all taxa formerly attributed to the polytypic D. certhia (hereafter referred to as the "certhia complex"; PP = 1), including D. c. concolor, D. c. juruanus, D. c. medius, D. c. radiolatus, D. c. retentus, D. c. ridgwayi, and D. sanctithomae The mitochondrial tree recovered a similar topology to that of the concatenated multilocus tree, except for the basal relationships within the picumnus complex, which involved nodes with little statistical support (PP ≤ 0.7; Figure S1).

| Divergence time estimates and species limits
The divergence chronogram reconstructed in *BEAST was consistent for the most part with the concatenated phylogeny obtained with Bayesian methods, with minor discrepancies in the tree topology, but significant differences in the statistical support for most nodes (Figures 1 and 2). The only difference in topology pertained to the relationships within the D. c. medius/ retentus/ ridgwayi clade, which was consistently recovered as a highly supported clade by both analyses, but whereby the position of the sister group to the entire clade alternated between D. c. ridgwayi (according to the concatenated estimate; Figure 1) and D. c. retentus (according to the multilocus *BEAST estimate; Figure 2). With respect to nodal support values, while most basal nodes recovered by the coalescent tree were well-supported statistically, just like in the concatenated tree, nodes closer or associated with the tips received generally lower values than those in the concatenated tree (Figures 1 and 2).  D. c. certhia, D. c. concolor, D. c. juruanus, D. c. medius, D. c. radiolatus, D. c. retentus, D. c. ridgwayi, D. p. costaricensis, D. p. pallescens, D. p. picumnus, D. p. transfasciatus, D. p. validus, D. hoffmannsi, D. platyrostris, and D. sanctithomae (Figure 2 and Table 2).

| Biogeographic reconstruction
Of the six models evaluated, the best fit to our dataset was Numbers of asterisks at the tips (*) correspond to the number of times the respective taxon was recovered as an independent lineage (probability ≥ 0.98) out of a total of three demographic models combining different priors on ancestral population sizes and divergence times (see text and Table 2 for details)

TA B L E 2
Results of different divergence scenarios tested for reciprocally monophyletic lineages in Dendrocolaptes revealed by a Bayesian concatenated analysis using BPP. θ and τ are priors for population size parameters and divergences time at the root of the species tree, respectively. Results at p > .95 indicate that the taxon in question has coalesced significantly with respect to all other lineages for the scenario tested Scenarios 1 2 3 Priors θ = 2 2,000 τ = 1 10 θ = 1 10 τ = 1 10 θ = 2 2,000 τ = 2 2,000

| Phylogenetic relationships and plumage evolution
Our estimated phylogenies provide an unprecedented resolution into the diversification of the genus Dendrocolaptes in the Neotropics, even considering the sampling limitations of our study (see below). First, we have confirmed with the largest number of specimens and taxa sequenced so far, previous anatomical and molecular results pointing to the monophyly of Dendrocolaptes (Derryberry et al., 2011;Raikow, 1994). In contrast, a study based exclusively on plumage characters failed to recover the genus monophyly, mainly because the "streaked" Dendrocolaptes group shares this particular plumage character with other woodcreeper genera such as Xiphocolaptes and Hylexetastes, which has been interpreted as a source of homoplasy (Marantz, 1997). Second, our study provides the most robust evidence to date supporting two reciprocally monophyletic groups in Dendrocolaptes, herein named the certhia and picumnus complexes, confirming a previous family-wide molecular phylogeny (Derryberry et al., 2011). From a phenotypic perspective, previous taxonomists had already recognized the certhia and picumnus complexes, respectively, as the "barred" and "streaked" Dendrocolaptes groups, although membership of D. hoffmannsi to each of them remained controversial due to its intermediate anatom- ical and plumage characteristics (Raikow, 1994;Willis, 1982). Herein, our data confirm with high support that D. hoffmannsi belongs to the "streaked" picumnus complex, as also supported by a previous molecular study (Derryberry et al., 2011), as well as vocal characters and behavior (Marantz, 1997;Marantz & Patten, 2010;Willis, 1982).
Phylogenetic relationships within each the certhia and picumnus complexes were also generally well resolved in our phylogenies, contrasting strongly with a previous study based on plumage characters that did not recover well-resolved phylogenetic relationships within these groups (Marantz, 1997 Figure 1). Indeed, the more uniform plumage characteristics shared between the largely sympatric D. c. concolor and D. hoffmannsi could be due to either convergence or parallel evolution, given their distant phylogenetic affinities (Marantz, 1997;   (Batista et al., 2013;Marantz, 1997).

| Species limits in Dendrocolaptes
A previous phylogenetic study that sampled all Dendrocolaptes taxa known at the time produced plumage-based trees that failed to recover well-resolved relationships among taxa grouped in both the certhia and picumnus complexes (Marantz, 1997). While those trees did recover the certhia complex as monophyletic (with overall little internal resolution), the picumnus group was recovered as polyphyletic, with only a few taxa in it grouped as each other closest relatives depending on the tree building criterion chosen (Marantz, 1997). A subsequent study that sampled the same taxa through morphometric characters also obtained relatively little resolution with respect to classifying different groups according to accepted species and subspecies boundaries (Marantz & Patten, 2010 Marantz, Aleixo, Bevier, & Patten, 2020c;Peters, 1951;Piacentini et al., 2015;Pinto, 1938Pinto, , 1978Willis, 1982). This is perhaps best exemplified by Marantz et al. (2020b) statement concerning the most frequent treatment of the picumnus complex as including three distinct species (D. hoffmannsi, D. picumnus, and D. platyrostris), that "…work [is] needed to justify their continued separation." In contrast, molecular studies sampling multiple Dendrocolaptes taxa have recovered well-resolved and for the most part statistically well-supported phylogenies (Batista et al., 2013;Derryberry et al., 2011;Silva et al., 2019), which allow for a more explicit evaluation of interspecific limits under an evolutionary perspective.
This contrasts markedly with the current number of only five to six species recognized in the genus according to most modern taxonomic sources (Gill et al., 2020;del Hoyo et al., 2020;Remsen et al., 2020), but see Piacentini et al., (2015). With respect to D. c. ridgwayi, Batista lineage that is morphologically closer to the weakly barred D. c. concolor, although it belongs in a clade that includes both weekly barred (D. c. retentus: not described before Marantz, 1997 andMarantz &Patten, 2010) and barred (D. c. medius) taxa. As discussed above, the boldness of barring on plumage is not correlated with the phylogeny in Dendrocolaptes and other woodcreeper genera, and this was the only criterion used by Marantz (1997) to regard D. c. ridgwayi as a hybrid swarm.
In conclusion, the lineage delimitation scheme provided by BP&P herein objectively recognizes genetically and morphologically diagnosable Dendrocolaptes taxa that can be regarded as consistent basal evolutionary units for classification purposes (Figure 2). The use of BP&P, and the multispecies coalescent model upon which it is based (MSC), has been criticized on the grounds that it captures mainly population-level structure rather than species divergences (Jackson, Carstens, Morales, & O'Meara, 2017;Sukumaran & Knowles, 2017).
On a more recent study, Leaché et al., (2019) performed simulations to address these criticisms and showed that the Bayesian model selection implemented in BP&P may indeed over-split and recognize too many species, particularly in subdivided populations with ongoing gene flow. However, this trend is particularly true when hundreds or thousands of loci are analyzed, which contrasts strongly with the total of three independent loci sampled in our study (Leaché et al., 2019).
These authors also pointed out that the MSC model (and any other species delimitation model currently available) is not intended to split lineages based on "speciation genes" or those genes accounting for reproductive isolation, and conclude that "Even if the genomic data or the BP&P program cannot distinguish populations and species, the genetic distinctness of the populations signifies the presence of reproductive barriers or isolation mechanisms." The 15 significantly coalesced lineages identified by BP&P in Dendrocolaptes fell within a range of population divergence parameters consistent with at least advanced degrees of reproductive isolation (Figure 2), which were also matched in most cases by existing (albeit subtle) phenotypic diagnoses (Marantz, 1997;Marantz & Patten, 2010). These patterns are totally consistent with a novel perspective of species delimitation whereby the burden of proof is placed on splitting rather than lumping taxa, as long as empirical evidence suggests the existence of distinct and reciprocally monophyletic sister populations, which are then interpreted as exhibiting some level of reproductive isolation in opposition to free interbreeding if occurring in sympatry (Gill, 2014).
Although some sampling limitations of our study preclude the immediate translation of all significantly coalesced lineages revealed by BP&P into independent species-level taxa (see below), our results nevertheless represent a major step forward with respect to the historical taxonomic practice in Dendrocolaptes, which has essentially consisted in assessing species limits based on qualitative interpretations of comparative levels of plumage and vocal differentiation across taxa (Cory & Hellmayr, 1925;Marantz, 1997;Peters, 1951;Pinto, 1938Pinto, , 1978; but see Batista et al., 2013).

| Historical diversification in dry and humid forests of the Neotropics
Our ancestral area estimates favored lowland Amazonia as the center of diversification for the genus Dendrocolaptes (Figure 3). While the earliest splits in the genus continued to take place in Cis-Andean South America, an early dispersal event across the Andes occurred during the Plio-Pleistocene (3.1-1.5 mya), leading to the establishment of D. sanctithomae in Central America, where it eventually expanded into a wide range of distinct forest types up to ca. 1,800 m (i.e., cloud forest, gallery, mangrove, pine-oak, and semi-deciduous), albeit still favoring mature humid forest in the lowlands (Marantz, Aleixo, Bevier, Patten, & Kirwan, 2020d). Our analysis indicates that the remaining lineages of the certhia complex diversified almost entirely in the Amazonian lowland humid forests (Figure 3), although D. c. medius has reached the northern part of the also humid Atlantic Forest in northeastern Brazil (Marantz et al., 2020c). We did not successfully amplify DNA out of D. c. medius specimens from the Atlantic Forest since only relatively old skins were available of this now rare and highly endangered population . However, no apparent morphological differentiation has been detected between Amazonian and Atlantic Forest populations of D. c. medius (Marantz, 1997;Marantz & Patten, 2010), which could also indicate a lack of complete genetic isolation between them, analogous to what has been reported for at least two other suboscine lineages sharing similar disjunct distributions between these forest biomes (Rocha et al., 2015;Thom & Aleixo, 2015).
Albeit also centered in Amazonia, the diversification of the picumnus complex included three independent dispersal events outside of that region that also led to differentiation into different forest montane habitats (Marantz et al., 2020b). The third and last episode occurred at ca. 0.3 mya, with D. p. pallescens differentiating in the southern "dry diagonal" Chaco biome and neighboring foothills of the Andes (Figure 3). Therefore, unlike inferred for other groups (Porzecanski & Cracraft, 2005;Werneck, 2011), Dendrocolaptes lineages replacing each other across the "dry diagonal" biomes of South America do not support a sister relationship linking the Chaco and Cerrado to the expense of the Caatinga biome, or a closer Chaco-Caatinga historical connection (Hayes, 2001;Lopes, Chaves, de Aquino, Silveira, & dos Santos, 2018). Instead, our results indicate that Dendrocolaptes lineages distributed in the "dry diagonal" are not monophyletic, but derived from two temporally independent Amazonian invasions. The diversification of the picumnus complex in particular supports that the Middle and Late Pleistocene were periods of intense biotic exchange both between lowland and montane biotas and between humid (particularly Amazonia) and drier forest types, such as those distributed along the South American dry diagonal. These results also support a keystone role for Amazonia as a major source of diversification in the Neotropics, mirroring previous findings (Antonelli et al., 2018).

| Taxonomic implications
Our analyses recovered with strong support the monophyly of most Dendrocolaptes species recognized by current taxonomy (Gill et al., 2020;Marantz & Patten, 2010;Remsen et al., 2020), except for the polytypic D. picumnus, which appeared consistently as a paraphyletic species with respect to D. hoffmannsi in both phylogeny estimates we obtained (Figures 1 and 2). However, statistical support for this relationship was weak in both trees, and a monophyletic D. picumnus cannot be ruled out given that the basal-most relationships in the picumnus complex are characterized by the presence of one ( Figure 1) or two ( Figure 2) short nodes with low statistical support (Figures 1 and 2). Interestingly, the apparent closer relationship of D. hoffmannsi to most taxa in the polytypic D. picumnus to the exclusion of D. p. transfasciatus (Figures 1 and 2) mirrors previous results pointing to a greater morphological similarity between the latter taxon and D. platyrostris, mainly due to the distinct blackish crown shared by these largely parapatric taxa (Marantz, 1997;Willis, 1982). Furthermore, the average pairwise mitochondrial sequence divergence level between D. p. transfasciatus and the remaining taxa of D. picumnus sampled herein is higher (2.65%) than that between D. hoffmannsi and the latter taxa (1.95%: Table S3), which are currently regarded as separate species (Gill et al., 2020;Marantz & Patten, 2010;Remsen et al., 2020). In sum, both significant levels of morphological and genetic differentiation, coupled with a significant degree of coalescence with respect to other closely re- Dendrocolaptes (Cory & Hellmayr, 1925). Recently, D. p. transfasciatus (which is endemic to Brazil) has been regarded as "vulnerable" at the national level (Leal, Silva, & Marques, 2018).
The lack of four trans-Andean (D. p. costaricensis, D. p. multistrigatus, D. p. puncticollis, and D. p (Marantz, 1997). However, it has been proposed that D. s. punctipectus intergrades with trans-Andean D. sanctithomae taxa in northern Colombia (Marantz, 1997), which could also not be addressed by us due to sampling limitations.
Similarly, D. s. sheffleri from the northernmost part of D. sanctithomae range is also distinct in terms of plumage patterns and coloration (Marantz, 1997), and could well represent a distinct lineage.
The remaining D. sanctithomae subspecies not sampled by us (D. s. hesperius) is the closest morphologically to the nominate subspecies, which argues in principle against a distant phylogenetic relationship. Considering the degree of uncertainty and the sampling limitations across D. sanctithomae taxa that affected not only our study, but also those analyzing their morphological (Marantz, 1997) and vocal (Boesman, 2016) differentiation, we recommend not treating D. s. punctipectus as a full species (thus, following Marantz, 1997;Marantz & Patten, 2010;Gill et al., 2020, andRemsen et al., 2020) until evidence becomes available concerning its phylogenetic position and degree of evolutionary independence with respect to other D. sanctithomae lineages.
In contrast, ample evidence is available by now that treating all Cis-Andean taxa of the certhia complex as one species (Gill et al., 2020;Remsen et al., 2020) implies in tremendously underestimating the true species diversity in this group (Batista et al., 2013;Silva et al., 2019;this study (Marantz, 1997;Willis, 1992; AA pers. obs.), which mirrors the apparent lack of vocal variation among members of the picumnus complex, particularly between D. hoffmannsi and the polytypic D. picumnus (Marantz, 1997;Willis, 1982;AA, pers. obs.). Other woodcreeper genera in which little vocal variation is present, despite pronounced genetic differentiation and significant degree of coalescence consistent with the existence of cryptic species, are Dendrexetastes (Ferreira, Aleixo, & Silva, 2016) and Hylexetastes (Azuaje-Rodriguez et al., 2020). These significantly coalesced lineages are totally inconsistent with the hypothesis of free interbreeding among them, indicating that conservatism in vocal characters is not synonymous with lack of strong genetic isolation and speciation. Indeed, as shown recently for other Furnariidae, not even the mere existence of gene flow can be equated with lack of speciation, as suboscine lineages may evolve full reproductive isolation in a much longer time period than many Northern Hemisphere bird lineages (Pulido-Santacruz, Aleixo, & Weir, 2018, Pulido-Santacruz, Aleixo, & Weir, 2020Weir & Price, 2019). In fact, the recognition of D. certhia as a single species is inconsistent even when considering comparative levels of genetic differentiation within the genus Dendrocolaptes. For instance, as discussed above, D. picumnus and D. hoffmannsi, which differ by 1.95% in their mitochondrial DNA (Table S3), are regarded as independent "biological" species by the same sources that lump D. c. certhia and the remaining taxa members of the same purported polytypic species, and which differ in their mitochondrial DNA by an average of 2.4% (Table S3; Gill et al., 2020;Remsen et al., 2020).
Comparative levels of mitochondrial sequence divergence alone cannot be safely used as yardsticks for establishing species limits, but in this particular case they illustrate the subjectivity and lack of consistency in the application of a particular implementation of the "Biological Species Concept" (Johnson, Remsen Jr, & Cicero, 1999). For all these reasons, we recommend the treatment of the following significantly coalesced taxa according to our multilocus analyses (Figures 1 and 2) and currently regarded as subspecies of D. certhia by some sources (Gill et al., 2020;Marantz et al., 2020c;Remsen et al., 2020), as full species, mirroring the treatment by Piacentini et al. (2015): D. certhia, D. concolor, D. juruanus, D. medius, D. radiolatus, D. retentus, and D. ridgwayi.

ACK N OWLED G EM ENTS
We thank the curators and curatorial assistants of the Academy of   Table S1. Primer sequences used in this study. Table S2. Best fit models and partitions selected by PartitionFinder.