We used mitochondrial and nuclear DNA sequences to examine patterns of differentiation and evolution in the Musophagidae, an avian family endemic to sub-Saharan Africa; attention was focused on the subfamily Musophaginae, the turacos, or louries. Phylogeographic analysis of 410 individual ND2 sequences from throughout the ranges of the currently recognized species revealed multiple instances of unexpectedly large genetic divergences and cryptic taxa. Within both montane and lowland species, including Tauraco hartlaubi and T. schalowi, Menelikornis leucotis, Musophaga macrorhyncha, and Gallirex johnstoni, fixed private haplotypes were found in disjunct portions of the ranges, suggesting negligible recent gene flow and evolutionary independence of populations. Two taxa originally described as subspecies (T. schalowi loitanus and T. s. marungensis), but not recognized for over 50 years, were found to be 100% diagnosable based on the mitochondrial sequences. The data also revealed the existence of two polyphyletic traditional species, Tauraco livingstonii and T. schuettii, as well as the polyphyly or paraphyly of all traditional superspecies complexes involving members of the genus Tauraco. Overall, our analyses of genetic and morphological variation revealed substantial and unexpected geographic diversity within the Musophagidae. We recognize 33 species-level taxa that represent the appropriate units for phylogenetic and biogeographic analyses (phylogenetic species).
We used complete mitochondrial ND2 sequences and nuclear DNA sequences of an Aconitase intron and of the RAG-1 exon to infer the phylogenetic relationships among those species. The results include all the phylogenetic species and, for the first time, nuclear data. We present a new classification of the Musophagidae based on our phylogeographic and phylogenetic results. We allocate the 33 species to seven previously recognized genera, an average of 4.7 species per genus.
INTRODUCTION
The turacos and go-away-birds (Musophagidae) are a charismatic family of colorful birds now endemic to sub-Saharan Africa. A fossil from Wyoming placed as a sister to the crown clade suggests ancestral taxa were once more widely distributed (Field and Hsiang, 2018). The extant species are distributed throughout forest, bush, and savannah habitats (Chapin, 1963; Forshaw and Cooper, 2002), and occur from sea level to over 3000 m in elevation. Usually treated as three subfamilies, both their higher-level relationships and species limits have long been controversial. Sibley and Ahlquist (1990) reviewed the history of opinion on affinities of the musophagids with other family-level taxa; they had often been associated with cuckoos, based on their foot arrangement, but at various times had also been thought to be allied with the hoatzin (Opisthocomus) or with gallinaceous birds. Their relationships remain unresolved. Nuclear DNA sequences have suggested a sister-group relationship between turacos and a large clade of water birds (Hackett et al., 2008), bustards (Jarvis et al., 2014), or cuckoos plus bustards (Prum et al., 2015); Prum et al. (2015) suggested they have evolved separately from other avian lineages for 50–60 million years.
The taxonomies of two of the three subfamilies have generally been stable. One of these, the Corythaeolinae, is monotypic and comprised of the great blue turaco, Corythaeola cristata. A second subfamily, Criniferinae, comprises the mostly gray or brown go-away-birds; five generally recognized species possess largely allopatric ranges, although in some cases their ranges overlap with segregation by habitat (Snow, 1978).
The largest subfamily, the Musophaginae, consists of green or violet birds possessing unique, copper-based pigments (turacin and turacoverdin) in their plumage. The nomenclature, number, and composition of species in this group have long been problematic (table 1). Peters (1940) recognized 15 species (plus 18 additional subspecies) in four genera, but seven years later von Boetticher (1947) listed 17 species in 13 genera and subgenera. The most comprehensive analysis in recent times was that of Moreau (1958a, 1958b, 1958c) who recognized only 12 species in two genera. Subsequently, Snow (1978) recognized 16 species, Brosset and Fry (1988) recognized 14 species and, most recently, Turner (1997) and Dickinson and Remsen (2013) both recognized 17 species. However, these numbers obscure additional complexity: von Boetticher's (1947) 17 species are not identical to those of Dickinson and Remsen (2013).
Geographic variation in plumage and morphology of all the described taxa in this family was thoroughly described by Moreau (1958b). In addition to discrete differences, he also found cases of disagreement among characters and apparent clinal variation. He chose not to recognize some forms and combined others into species and superspecies based on a heuristic application of the biological species concept. Almost all the variation in recent species-level classifications of the birds have been based on differing interpretations of his characters in nonphylogenetic frameworks. The general problem has been that many of these birds are quite similar in appearance and it has been difficult to place species limits on the parapatric, geographically replacing forms that possess subtle differences or conflicting patterns of phenotypic traits. A second complication arising in these birds is the historic association, among avian taxonomists, of particular anatomical characters with rank (Mayr, 1982); for example, nostril shape was often considered a generic character, while bill color was usually treated as a species- or subspecies-level trait; crest length might be a subspecific or clinal trait (e.g., Moreau, 1959). Thus, for many taxonomists, the green turacos of the Congo Basin possessing black bills had to be related, even if aspects of their plumage suggested otherwise. These various issues have resulted in idiosyncratic and conflicting classifications that need to be resolved; species limits are a serious concern because studies of evolutionary diversification and historical biogeography require monophyletic species taxa as the basic units of analysis (Barrowclough et al., 2016).
We investigated geographic variation, species limits, and phylogeny of the Musophagidae using mitochondrial and nuclear DNA sequences to resolve turaco systematics. Phylogeography has frequently been useful for identifying cryptic evolutionary units and determining species limits in difficult groups of birds, including those in Africa. For example, African chats (Cercomela: Outlaw et al., 2010), bush-shrikes (Laniarius: Voelker et al., 2010a), and cisticoline warblers (Scepomycter: Bowie et al., 2009) all represent difficult taxa in which molecular studies of geographic variation have been highly informative. Although the phylogenetic relationships among turacos have recently been investigated using both morphology (Veron, 1999) and molecules (Veron and Winney, 2000; Njabo and Sorenson, 2009), these studies were all restricted to exemplars of the generally recognized taxa, accepting traditional classifications based on the biological species concept and either postulated hybrid zones or an inferred ability to hybridize. The monophyly of those biological species was assumed, but not assessed.
For the most part, turacos have large, often parapatric ranges. Consequently, they provide important data concerning large-scale patterns of African biogeography and associated historical processes (Dowsett-Lemaire and Dowsett, 2001). Indeed, these birds have figured in several attempts to assay general patterns of distribution and endemism (e.g., Crowe and Crowe, 1982) with subsequent linkage to Quaternary climate change (Livingstone, 1975). They also have figured in arguments concerning the efficacy of Pleistocene refuge models of speciation (Mayr and O'Hara, 1986), and the role of ecological isolation in species' distributions (e.g., Lack, 1971). An improved knowledge of turaco diversification should benefit such investigations.
MATERIALS AND METHODS
Nomenclature
The results we report here for the Musophaginae require a species-level classification incompatible with all prior treatments; some names we recognize have not been used in more than 75 years. Consequently, it is difficult to discuss parts of our results using the most recent classification of turacos (Dickinson and Remsen, 2013). In some cases, we use specific epithets in the following text, tables, and figures that may not be widely recognized. In table 1 we provide a brief synopsis of recent species-level classifications of this subfamily, relevant synonymies, and the species level taxonomy supported by our data.
Samples
We obtained preserved tissue samples of turacos from the American Museum of Natural History, the Field Museum of Natural History, the Peabody Museum of Natural History at Yale University, the Burke Museum of the University of Washington, the Zoological Museum of the University of Copenhagen, and the British Museum (Natural History) at Tring. However, a relatively small number of geographically dispersed samples were available for most of the generally recognized species-level taxa and, in the case of ruspolii, none were available. Consequently, we relied on toe pads of traditional museum specimens for the vast majority of our geographic sampling for DNA sequencing. In addition, we examined traditional museum skins in the collections of the American Museum of Natural History and the Field Museum of Natural History to confirm plumage and soft-part differences among the recognized forms that had been previously described and illustrated by Moreau (1958b) and Forshaw and Cooper (2002).
Laboratory Methods
We extracted DNA from the tissue and toe pad samples using standard procedures previously described (e.g., Barrowclough et al., 2011). We sequenced approximately one half of the mitochondrial ND2 gene from the specimens available as toe pad samples. These samples were, in some cases, over 100 years old; consequently, a set of turaco specific PCR primers was designed to amplify 100 to 200 base pair (bp) fragments for those specimens for which longer fragments could not readily be amplified. ND2 extraction and sequencing from large numbers of traditional skin preparations previously has been shown to provide useful data for avian phylogeography (e.g., Reddy, 2008; Perktaş et al., 2011). In addition, we amplified the entire ND2 gene for single exemplars of each of the generally recognized species of musophagids, as well as of subspecific taxa that we found to be genetically divergent in our analyses. Each of these exemplars was also sequenced for the entire 15th intron of the nuclear aconitase-1 gene, using methods previously described (Barrowclough et al., 2011). Finally, we amplified and sequenced a large portion of the nuclear RAG-1 gene for those taxa of turacos for which preserved tissue samples were available, again using previously described procedures (Groth and Barrowclough, 1999). All the PCR products were Sanger sequenced and the chromatograms recorded on an ABI 3730xl DNA analyzer.
Analyses
The DNA sequences were assembled, aligned, and analyzed using Sequencher software (version 5.1). We used the program PAUP* version 4.0b10 (Swofford, 2001) to infer minimum-length networks for both the entire set of partial ND2 sequences and for the reduced set of unique haplotypes found among those sequences. For each of the three exemplar data sets of differentiated taxa (complete ND2, ACO1-I15, and RAG-1), we checked for unusual nucleotides, excessive proportions of ambiguity codes, heterogeneity of base composition at all three coding positions, signatures of contamination (chimerism) and, for the two protein-coding genes, unexpected stop codons and indels not a multiple of three bp in length. We used PAUP* to infer most parsimonious trees for the data sets using 25 TBR heuristic searches (random stepwise addition of taxa, gaps treated as missing, and ambiguities treated as uninformative). We also performed a bootstrap analysis of the parsimony procedure with 100 replicates.
We used the program MODELTEST version 3.06 (Posada and Crandall, 1998) to find initial maximum likelihood models for the three exemplar data sets using the AIC criterion. We then followed the protocol suggested by Sullivan et al. (2005) and performed initial maximum likelihood heuristic TBR searches, using PAUP*, with fixed parameters from the model specified by MODELTEST, starting at the maximum parsimony tree. Upon completion of the likelihood search, we reestimated the likelihood model parameters on the resulting tree; if the parameters had changed, we fixed the new parameters and started a subsequent TBR search. We repeated this procedure until the resulting likelihood tree was consistent with its initial parameters. Finally, we performed likelihood bootstrap analyses using the fixed final parameter sets with 100 TBR replicates, starting from trees obtained using neighbor-joining.
We examined base composition variation among taxa for all three loci. Base composition heterogeneity in vertebrate nuclear genomes is frequently characterized by variation in C plus G content. For the nuclear, noncoding aconitase intron, we computed the overall C plus G fraction for each of the members of the Musophaginae, the Criniferinae, Corythaeola, and the four outgroups (see below). For the coding RAG-1 nuclear gene, we computed the overall C plus G fraction for third position synonymous sites and for first plus second position amino-acid replacement sites. In the vertebrate mitochondrial genome, transitions greatly outnumber transversions; consequently, the A plus G fraction is nearly independent of the C plus T fraction for closely related taxa. Therefore, we computed the A and C fractions at third codon positions for the mitochondrial ND2 locus (e.g., Groth et al., 2015).
We used the program GARLI version 2.01 (Zwickl, 2006) to obtain an overall estimate of turaco phylogeny across loci. Each of the three genes was treated as a separate partition in this likelihood analysis, with its own model parameters optimized during the search, using the autostopping criterion. A bootstrap analysis was performed with 100 replicates and five stepwise random addition searches within each replicate.
In all our phylogenetic analyses, we used a bustard, crane, cuckoo, and stork as outgroups based on recent opinions concerning the phylogenetic relatives of turacos. For our ND2 analyses, we used four sequences from GenBank for that purpose: Otis tarda: NC014046; Antigone canadensis: FJ769855; Coccyzus americanus: EU327609; and Ciconia ciconia: NC002197. For ACO1, we sequenced the same four species ourselves. For RAG-1, we used two outgroup sequences from GenBank (A. canadensis: AF143732 and C. americanus: DQ482640) and sequenced the other two.
Within each species or traditional species complex, we grouped individuals into populations composed of samples taken from collecting localities within the same small country or small region for larger political entities. For those populations with sample sizes of three or more, we estimated Nei's (1987) nucleotide diversity (π) and computed Holsinger and Mason-Gamer's (1996) Gst statistic. The latter is an estimator of the ratio between the among-population and total coalescent times (Slatkin, 1991).
We used the program ARLEQUIN version 3.5.2.2 (Excoffier et al., 2005) to estimate hierarchical components of genetic variance between previously recognized subspecies for those cases in which we observed substantial divergence in the ND2 phylogeographic analysis. In these estimates, transitions and transversions were equally weighted; the levels of analysis were: among subspecies; among populations within subspecies; and within populations. The estimate of Fst among subspecies was taken as the hierarchical component of genetic variance among those taxa.
RESULTS
Phylogeography
We obtained an aligned set of 410 partial sequences of the ND2 gene from representatives of all generally recognized species, and most subspecies, of musophagine turacos. The partial ND2 data set was 607 bp long and corresponded to base pairs 5405 through 6011 of the published complete mitochondrial genome of Gallus (Desjardins and Morais, 1990). No indels or ambiguous bases were required in the alignment. These sequences have been deposited in GenBank; GenBank accession number, museum voucher information, as well as available data on the geographic origin of each specimen are provided in appendix 1. Unfortunately, many of these specimens were very old and the correspondence between their label data and currently recognized place names was sometimes difficult to assess; some specimens could be allocated only to present-day countries, others were assignable to counties, provinces, regions, states, or more specific localities, based on label data, atlases, and gazetteers.
The number of individuals sampled from each currently recognized species (Dickinson and Remsen, 2013) and the number of haplotypes found within each are shown in table 2. Overall, the 410 sequences corresponded to 116 unique haplotypes. Nucleotide diversities, averaged over populations represented by three or more individuals, generally ranged between 10-4 and 10-3, but we observed no variation among the three sampled specimens of ruspolii (table 2).
A minimum-spanning network for the 116 haplotypes is shown in figure 1. No haplotypes were shared between any pairs of currently recognized species. However, there were frequent instances of geographically allopatric or parapatric taxa, currently considered subspecies, that shared no haplotypes and, in many cases, were reciprocally monophyletic. In two cases, livingstonii and schuettii, the currently recognized species were not monophyletic.
The network shown (fig. 1) is one of many, and the various resolutions of all the networks resulted in 1936 alternate minimum length trees in the PAUP* analysis. The only major differences among these were that, in one third of the trees, M. verreauxii was paraphyletic with respect to M. macrorhyncha in an alternate local rooting and, in approximately 30% of the trees, either T. schalowi (sensu latu) or T. schuettii was paraphyletic with respect to the other taxon. None of the alternate trees, however, resulted in polyphyly of any of the taxa identified in figure 1, and there was no haplotype sharing anywhere in the network with the exception of two individuals of porphyreolophus possessing the predominant haplotype found in chlorochlamys.
TABLE 2.
Genetic variation in turaco populations at the ND2 locus.a
The geographic distributions of haplotypes are shown in figures 2–6 for each of the 17 species-level taxa of musophagines recognized by Dickinson and Remsen (2013). These show the geographic pattern and extent of our sampling within those taxa, the haplotype network for each species, and suggest regions where the genetic units might contact between the sampled populations. The magnitude of genetic divergence among populations is provided by our estimates of Gst (table 2); in many cases, these estimates were large and reflect clades of haplotypes restricted to divergent, currently subspecific, taxa. Consequently, we also provide estimates of Gst within such subspecies. For example, 88% of the total genetic variation was distributed among populations of the traditional T. persa across the west coast of Africa, but substantially less among populations within each of our restricted persa and buffoni (table 2). In other instances, large values of Gst occurred within much smaller regions (e.g., T. hartlaubi and M. leucotis plus M. donaldsoni).
Species Delimitation
Based on the pattern of geographically parapatric and allopatric clades of haplotypes found in our phylogeographic analysis, in addition to abrupt geographic transitions in plumage and soft-part morphology, as documented by Moreau (1958b), we concluded that there are 27 phylogenetic species-level taxa within the subfamily Musophaginae; these are the taxa appropriate for studies of phylogenetic diversification and historical biogeographic analysis. Our assessment incorporates phylogeography, morphology, and geography, as indicated in table 3. The associated hierarchical estimates of genetic variance, based on the ARLEQUIN results, were very high (>47%, table 3), with the sole exception of that between G. porphyreolophus and G. chlorochlamys (20%), in which a shared haplotype was found in one population.
Phylogeny
We obtained complete sequences of the mitochondrial ND2 gene from 33 taxa of musophagids. The sequences have been deposited in GenBank (KU160188–KU160218); specimen voucher data is provided in appendix 1. All sequences were 1041 bp in length, including the outgroups; there were no ambiguous bases, and they could be aligned without indels.
We obtained complete sequences of intron 15 of the nuclear aconitase-1 gene from the same 33 taxa of turacos sequenced for the ND2 gene, as well as for the same four outgroups. These sequences have been deposited in GenBank (KT372802–KT372836, MF766008–MF766009); specimen voucher data is provided in appendix 2. The sequences varied in length from 544 bp to 560 bp among the musophagids, and from 549 bp to 562 bp among the outgroups. We obtained an overall alignment of 574 bp; this required ten indels within the ingroup, four of which represented synapomorphies. There were 16 indels inferred among the outgroups. The ACO1 gene resides on the Z chromosome in birds and consequently is diploid in males and haploid in females; we observed a range of heterozygosities within individuals of 0.0 to 0.009 among the turacos.
FIGURE 1.
Minimum-spanning network for 116 ND2 haplotypes found among 410 individuals of musophagine turacos. Area of each pie diagram is proportional to number of individuals sampled with that haplotype; black dots indicate positions of unobserved (inferred) ancestral haplotypes. Approximate position of branch from sister taxa of Musophagidae indicated by arrow. Alternate colors and names correspond to species-level taxa recognized here.
FIGURE 2.
Geographic distribution of ND2 haplotypes in the Gallirex and Proturacus complexes. Approximate geographic ranges of taxa are indicated by colored shading (Gallirex in dark green; Proturacus in light green). Areas of pie diagrams on map are proportional to sample sizes at each locality. Colors of pie diagrams are keyed to haplotype networks shown in associated capsules, where area of haplotype circle is again proportional to sample size. Approximate position of division between G. porphyreolophus and G. chlorochlamys phenotypes (Zambesi River), based on Moreau (1958b), is indicated by dashed line.
We obtained new nuclear RAG-1 sequences from 23 taxa of musophagids and two outgroups for which preserved tissues were available. The sequences have been deposited in GenBank (KT424072–KT424096); specimen voucher data are provided in appendix 3. The sequenced fragment is identical to that described by Groth and Barrowclough (1999); it corresponds to base pairs 84 through 2967 of the Gallus gene (GenBank: M58530; Carlson et al. 1991). We also used the RAG-1 sequence of one additional species of turaco (P. erythrolophus: DQ482643) previously deposited in GenBank. The four outgroups represent the same taxa used for the other genes. A single, 3 bp indel in one outgroup (Antigone canadensis) was required to align the sequences. The overall alignment of the gene fragment was 2872 bp in length; heterozygosity ranged from 0.0 to 0.006 for this autosomal gene. No length heterozygotes were encountered in either of the nuclear loci.
FIGURE 3.
Geographic distribution of ND2 haplotypes in turacos of the genus Musophaga. Approximate geographic range of the macrorhyncha complex is indicated in dark green, that of rossae in light green, and of violacea in violet. Areas of pie-diagrams on map are proportional to sample sizes at each locality. Colors of pie diagrams are keyed to haplotype networks shown in capsule; area of haplotype circle is proportional to sample size.
We used PAUP* to infer maximum likelihood trees, along with bootstrap confidence indices, for each of the three genes using a GTR plus G model of sequence evolution, based on the MODELTEST AIC results. The ND2 and ACO1 trees both included 33 taxa of musophagids; their bootstrap consensus trees are shown in figure 7 (outgroups not shown). The RAG-1 analysis included the reduced set of 24 ingroup taxa for which fresh tissues were available (fig. 8), plus the four outgroups (not shown). Each of the three genes resulted in monophyletic clades corresponding to the three traditional subfamily-level taxa. In fact, the only major discordance among the three trees was the position of Corythaeola in the two nuclear trees versus the mitochondrial tree. For the ND2 and ACO1 results, three minor discrepancies also occurred within species groups (figure 7); in two of those cases, high bootstrap support from ND2 conflicted with weak bootstrap support from ACO1. For example, P. leucolophus was sister to P. erythrolophus in the ND2 tree, but sister to P. bannermani in the ACO1 tree. The four synapomorphic indels at the ACO1 locus, not used in our phylogenetic inference, nevertheless were each consistent with nodes that also had appreciable bootstrap support. The RAG-1 results (fig. 8) were generally consistent with the other two trees; some minor discrepancies within the green turaco complex were present, but lacked bootstrap support.
FIGURE 4.
Geographic distribution of ND2 haplotypes in several currently recognized species (Dickinson and Remsen, 2013) of green turacos (Tauraco) across forested portions of Africa; approximate ranges of taxa are indicated by alternate shades of green. Areas of pie diagrams are proportional to sample sizes at each locality. Colors of pie-diagrams are keyed to haplotype network shown in capsule for each species complex.
We performed a combined analysis of the three genes using Garli. Each gene was treated as a separate partition, using GTR plus G models, with bootstrap replicates. The resulting consensus (fig. 9) placed Corythaeola with the go-away-birds (90% bootstrap) and showed more hierarchical structure within the Musophaginae than did any of the three individual gene trees.
Our ND2 sequences placed Corythaeola as sister to the turacos, whereas the two nuclear loci placed it as sister to the go-away-birds; a prior mitochondrial study had placed Corythaeola as sister to the rest of the family (Veron and Winney, 2000). Because base composition heterogeneity can interfere with phylogenetic reconstruction, we examined base composition in the three loci used here. For both RAG-1 (fig. 10) and ACO1 (not shown), there was little variation in base composition within the Musophagidae; however, for ND2 (fig. 11) variation was substantial within the ingroup, and especially within the go-away-birds, as it was among the outgroups. In particular, Corythaeola possessed an A nucleotide fraction within the range of that of the turacos, but substantially greater than that of the go-away-birds. For C nucleotides, Corythaeola was closer to Gallirex, sister to the rest of the musophagines, than it was to mean of the widely dispersed criniferines. These three subfamilies are from 14% to 17% divergent for ND2 and it is possible that base composition heterogeneity may have attracted Corythaeola to the musophagine portion of the evolutionary network for this mitochondrial gene and interfered with recovering actual evolutionary relationships.
FIGURE 5.
Geographic distribution of ND2 haplotypes in the Menelikornis complex in the Ethiopian highlands; approximate ranges of three taxa in the complex are indicated by alternate colors. Areas of pie diagrams on map are proportional to sample sizes at each locality. Colors of pie diagrams are keyed to haplotype network shown in capsule.
DISCUSSION
Phylogeography and Species Limits in the Musophaginae
Based on our analyses of the mtDNA sequences and our assessment of prior descriptions of the external morphology of the birds, we treat allopatric or parapatric forms that are diagnosable as phylogenetic species (Barrowclough et al., 2016). These are the proper units for studies of evolutionary divergence and historical biogeography; they document diversity hidden in current avian species lists (Collar, 2018). In addition, they play a critical role in setting priorities for conservation planning (Peterson and Navarro-Sigüenza, 1999; Goldstein et al., 2000). In the absence of detailed behavioral data, it is not evident whether each of these corresponds to a traditional biological species. Nevertheless, our results reinforce the suggestion of Barrowclough et al. (2016) that there is substantial unrecognized phylogenetic divergence in birds. Below we discuss each taxon given species rank in the most recent, widely used classification, that of Dickinson and Remsen (2013).
FIGURE 6.
Geographic distribution of ND2 haplotypes of Tauraco hartlaubi across East Africa. Areas of pie diagrams on map are proportional to sample sizes at each locality. Colors of pie diagrams are keyed to haplotype network shown in capsule. Darker tan on map indicates 1500 m contour.
Gallirex johnstoni: Short et al. (1990) suggest that “Rwenzori” is the proper spelling for the vernacular of this turaco. The two currently recognized subspecies, G. j. johnstoni and G. j. kivuensis, have ranges that are allopatric (fig. 2). They possess discrete differences in plumage and soft-part coloration (Moreau, 1958b; Forshaw and Cooper, 2002), and their mitochondrial haplotype networks resolve as two differentiated (table 3), reciprocally monophyletic clades (fig. 1). These represent two phylogenetic species. Their divergence is most likely the result of a history of isolation within high-elevation, montane forest fragments that are separated from each other by unsuitable, lower-elevation habitat. Moreau (1958b) thought a third taxon, G. j. bredoi, confined to Mt. Kabobo in the Democratic Republic of Congo, possessed “good” characters, but he later (Moreau, 1958c) changed his opinion, based on the examination of additional specimens. We were not able to examine any specimens during this research.
Gallirex porphyreolophus: The purple-crested turaco occurs from southern Kenya to the northeastern portion of the Republic of South Africa. The two generally recognized subspecies, G. p. porphyreolophus and G. p. chlorochlamys, have discrete well-marked differences in both plumage pattern and soft parts, as described by Moreau (1958b) and illustrated by Forshaw and Cooper (2002). Moreau (1958b), based on his examination of the large series at the British Museum, suggested the taxa were isolated by the Zambesi River; all (22) specimens at the AMNH are consistent with that interpretation (appendix 4). The suggestion of intergradation between these two (Moreau, 1958b) appears to be based on a few specimens, with reduced brownish pink on their breast and mantle, taken from the upper Zambesi and its tributaries, toward the western edge of the taxon's range in present day Zambia and Zimbabwe (Smithers, 1951). We have not seen those specimens; they might be relevant under the biological species concept, but not under a phylogenetic concept in which historical isolation trumps limited hybridization.
FIGURE 7.
Bootstrap consensus maximum likelihood trees for complete ND2 and ACO1-I15 DNA sequences of species-level taxa of turacos; bootstrap values indicated for nodes with greater than 50% support. Phylogenetic positions of four synapomorphic indels among the ACO1 sequences are indicated by vertical bars. Four conflicts between the two genes are indicated.
FIGURE 8.
Maximum likelihood phylogram for 24 species-level taxa of turacos based on sequences of the RAG-1 exon. Bootstrap values indicated for nodes with greater than 50% support.
We found several private haplotypes restricted to one or the other taxon, and a single haplotype present at high frequency in chlorochlamys (83%) and at low frequency (28%) in porphyreolophus (fig. 2). The occurrence of that common haplotype in porphyreolophus was not proximal to the edge of the range of chlorochlamys, near the Zambesi, where one might expect it on the basis of hybridization, but rather in the center of the range in the South African Mpumalanga Province (formerly Transvaal), consistent with a hypothesis of incomplete lineage sorting. The two Mpumalanga specimens that possessed the common, chlorochlamys ND2 haplotype (AMNH624120 and AMNH624123) both possess the diagnostic porphyreolophus character traits of brownish-pink breast and a reduced bare spot in front of the eye (appendix 4). This complex clearly merits future investigation, but, for the present, we recognize two phylogenetic species, based on the combination of mtDNA and morphology, the apparent result of geographic isolation in the relatively recent (e.g., incomplete lineage sorting) past.
Menelikornis leucotis/M. ruspolii: Our data indicate that the two currently recognized subspecies of white-cheeked turaco, along with Prince Ruspoli's turaco, are a closely related geographical assemblage (fig. 1) in northeast Africa (fig. 5). The three are diagnosable based on either our ND2 sequences or the morphological characters described by Moreau (1958b) and depicted in Forshaw and Cooper (2002). M. donaldsoni was as genetically divergent from leucotis as was ruspolii. They represent three allopatric, species-level taxa. Nucleotide divergence within the leucotis complex, Gst, was 0.5, and the hierarchical Fst was 0.67; these values were almost entirely due to the inclusion of the donaldsoni samples with those of leucotis (table 2, table 3). We did not find any genetic variation in either donaldsoni or ruspolii.
FIGURE 9.
Phylogenetic relationships among species-level taxa of turacos based on partitioned maximum likelihood analysis of ND2, ACO1-I15, and RAG-1 DNA sequences; bootstrap values indicated for nodes with greater than 50% support. Polyphyletic relationships of taxa within two currently recognized biological species (e.g., Dickinson and Remsen, 2013) are highlighted with arrows and alternate colors.
FIGURE 10.
Base composition variation among 24 taxa of the Musophagidae and 4 outgroups at the nuclear RAG-1 locus.
Musophaga rossae: Lady Ross's turaco, a monotypic species, has an extensive distribution through much of the southern and eastern Congo River Basin and its fringes (fig. 3). Nevertheless, it showed less genetic divergence over that range than the statistical error associated with our population samples (table 2). A sample from an isolated population from northern Cameroon shared haplotypes with a population sample from eastern Haut-Zaire, more than 1500 km away; this suggests that recently a more extensive distribution must have existed across the northern edge of the Congo Basin.
Musophaga violacea: The monotypic violet turaco showed substantial nucleotide diversity in our sample from Ghana (table 2). Although the species has a wide geographic distribution from Senegal to Cameroon (fig. 3), five of our six samples were smaller than three individuals per population; consequently, we did not calculate Gst across the range. We note, however, that the Senegal and Guinea-Bissau samples were fixed for a haplotype, two substitutions divergent from any others, that was not found east of those locations; this suggests there may be some additional structure within western Africa. We were not able to examine any specimens of this species from an isolated population in northeastern Central African Republic/southern Chad; a note by Moreau (1958b) suggests that population may be morphologically divergent.
FIGURE 11.
Base composition variation among 33 taxa of the Musophagidae and 4 outgroups at the mitochondrial ND2 gene.
Musophaga macrorhyncha: Although it has a range extending from Sierra Leone in West Africa to northern Angola on the west coast of central Africa, there is a discontinuity in the range of the yellow-billed turaco across the Dahomey Gap (Moreau, 1958b; Brosset and Fry, 1988). This gap also separates the ranges of the two described taxa, M. macrorhyncha to the west and M. verreauxii to the east. Our data suggest the gap also marks the division between two distinct, reciprocally monophyletic clades of haplotypes (fig. 1, fig 3) that were responsible for most (88%) of the overall genetic variance (table 3). These two taxa are well marked morphologically, with, among other traits, red versus blue crests (Moreau, 1958b; Forshaw and Cooper, 2002). They represent two phylogenetic species with the Dahomey Gap as a biogeographic barrier.
Proturacus bannermani: Bannerman's turaco is known from high montane forest (above 1700 m) from a single range in western Cameroon (fig. 2); it is monotypic. We found two haplotypes among four sequenced specimens (table 2).
Proturacus erythrolophus: The red-crested turaco has a restricted geographical range along the coast of Angola; there are no described subspecies. Our eight samples all derive from the central portion of that range (fig. 2) and showed moderate variation (table 2).
Proturacus leucolophus: The white-crested turaco has a range across north-central Africa from Nigeria to Kenya. We observed moderate nucleotide diversity among five population samples. Genetic variance among these samples was significant (0.18). This was the result of three private haplotypes, one restricted to Cameroon and two to the Democratic Republic of the Congo; however, both those locality samples possessed the widespread, common haplotype.
Tauraco corythaix: The Knysna turaco has a distribution along the South African coast from Cape Province to KwaZulu-Natal and Swaziland, with an isolated population (T. c. phoebus) in the highlands of Limpopo and Mpumalanga Provinces (Moreau, 1958b). We identified four haplotypes in the four populations we sampled (fig. 4). Nucleotide diversity was low because the variation was largely distributed among, rather than within, populations (Gst = 0.83, table 2). In fact, our single sample of phoebus possessed a haplotype not found elsewhere and our sample of four individuals from the Cape Province also was fixed for a private haplotype. This species shows little variation in plumage; phoebus is not a well-marked subspecies (Moreau, 1958b), but the distribution of fixed private haplotypes suggests there may be genetic differentiation within this species. This bird requires further investigation with denser geographic sampling.
Tauraco fischeri: Fischer's turaco has a restricted range on the coast of East Africa, with a separate subspecies on Zanzibar (fig. 4). We found one common haplotype, plus three singletons, in samples from five localities. Nucleotide diversity was low and divergence among populations did not exceed sampling error (table 2). We sequenced one of the two extant specimens from the Zanzibar population (T. f. zanzibaricus); it had a unique haplotype one substitution away from the most common haplotype found on the mainland. We agree with Pakenham (1938) and Moreau (1958b) that these birds possess a markedly bluer back than do the Kenya and Tanzania birds. Therefore, it is plausible that the Zanzibar birds represent a separate taxon; however, with a single sequence only one step away from the common, mainland haplotype, we are unable to determine whether the island population is fixed for a novel haplotype or simply possesses a local polymorphism. The situation requires further molecular and morphological investigation.
Tauraco hartlaubi: Hartlaub's turaco is found in the highlands and on isolated mountain ranges and volcanoes of East Africa in Kenya, northern Tanzania, and eastern Uganda. We found several common haplotypes distributed widely among our central Kenyan population samples, but the geographic isolates elsewhere in the range were largely fixed for private haplotypes, rendering several populations, for example, those of Mt. Elgon, Kilimanjaro, and the Usambara Range, 100% diagnosable (fig. 6). This geographic pattern was associated with 81% of the genetic variance distributed among populations (table 2). It is probable that there are several species-level taxa within this complex. However, the toe pad samples were not adequate for substantive nuclear DNA sequencing and morphological variation among the isolated populations was minimal. Consequently, we have chosen not to describe new taxa solely based on a fragment of one mitochondrial gene. The hartlaubi complex deserves substantial additional field and lab work.
Tauraco livingstonii: We identified two clades of Livingstone's turaco, not each other's closest relatives (figs. 1, 9), which corresponded to described subspecies from northern and southern portions of the species range (fig.4). The clades were 14 substitutions apart, and 96% of the genetic variance was distributed between the subspecies (table 3). Unfortunately, our population samples were not uniformly distributed over space, and we had no samples from the far southern portion of the range. Nevertheless, our results would seem to place the transition between the haplotype clades in southern Tanzania or northern Mozambique. This division does not closely correspond to published descriptions of the ranges of the morphologically based (blue-green vs. green plumage on back) subspecies, livingstonii and reichenowi (e.g., Moreau, 1958b; Forshaw and Cooper, 2002; Dickinson and Remsen, 2013).
Dickinson and Remsen (2013) restrict their T. l. livingstonii to the highlands of southern Malawi, west through adjacent Mozambique, to eastern Zimbabwe. Forshaw and Cooper (2002) give it a larger range, extending north to southwestern Tanzania. Both treatments correspond to interior versus more coastal distributions, as does Map V of Clancey (1971) and the discussion of Moreau (1958b). Turner (1997) similarly restricted livingstonii, but also limited his coastal reichenowi subspecies by recognizing a third taxon, T. l. cabanisi, for the southern coastal populations. Further sampling is clearly warranted here, but our results strongly support the existence of two unrelated species with northern and southern, rather than eastern and western ranges. The type locality of livingstonii is southern Malawi (Nyasaland) and that of reichenowi is in central east Tanzania (Tanganyika Territory); thus, we assign our southern species the name livingstonii, and our northern taxon the name reichenowi. The extent and geographic distribution of morphological variation in this bird is complex (Moreau, 1958b), and now clearly requires further evaluation.
Tauraco persa: The green turaco has an extensive range across West Africa from Senegal to Cameroon and south to Angola. We found two clades of haplotypes that correspond to regions west (T. buffoni) and east (T. persa) of Ivory Coast (fig. 4). Most of the genetic variation in the complex was distributed between those two regions (table 3). These represent two well-differentiated, morphologically (Moreau, 1958b; Forshaw and Cooper, 2002) as well as genetically, phylogenetic species. The division between them appears to be west of the well-known Dahomey Gap; this is consistent with the subspecific distributions reported by Moreau (1958b). T. p. zenkeri is an occasionally recognized taxon with a distribution to the east of our available samples; it requires further investigation.
Tauraco schalowi: Schalow's turaco, as traditionally recognized, has a range from Angola, east across Zambia, through the southeastern portion of the Democratic Republic of Congo, to the Rift Valley lakes. There are two additional isolated populations that have been named, but not generally recognized: one in the Crater and Mbulu highlands of central Tanzania (T. s. chalcolophus) and the other in the Loita Hills of southwestern Kenya (T. s. loitanus). Although the birds appear quite similar in plumage across this range (Moreau, 1958b), we found a complex pattern of differentiated haplotypes (fig. 4) and a large fraction (75%) of genetic variation distributed among regions (table 2). First, our sample of chalcolophus from Tanzania was fixed for a haplotype, found nowhere else, that placed that population as sister to the rest of the complex (fig. 1). Our Angola sample, from the western portion of the range (schalowi), possessed a network of haplotypes that was restricted to that country (fig. 3). Derived from within that network was a clade of eight haplotypes present only in the eastern portion of the bird's range. One of those was fixed in a small sample from the allopatric population in the Loita Hills (loitanus); the remaining seven (marungensis) were found in the eastern contiguous portion of the range. Consequently, there were two probably parapatric and two completely allopatric taxa in the complex, each of which was 100% diagnosable on the basis of mtDNA sequences. Forshaw and Cooper (2002) thought chalcolophus was separable based on plumage; the other three represent nearly cryptic taxa (Chapin, 1939; Moreau, 1958b), two of which are currently paraphyletic. The precise geographic boundary between T. schalowi and T. marungensis, probably somewhere in eastern Angola or western Zambia, is not clear (Peters, 1940; Moreau, 1958b), but if the distribution given by Snow (1978) is correct, the two might actually be geographically disjunct. The distribution of genetic variance among these taxa (table 3) indicates there are four phylogenetic taxa.
Tauraco schuettii: The black-billed turaco occupies much of the central and eastern Congo Basin, and extends into Uganda, South Sudan, and western Kenya. We found two reciprocally monophyletic clades of haplotypes that were not sister taxa (figs. 1, 9) and explained approximately 95% of the overall genetic variation (table 3). They correspond to the traditional subspecies T. s. schuettii in the west and T. s. emini in the east (fig. 4). The forms are well marked, with violet (schuettii) versus green (emini) plumage on the back, wings, and tail (e.g., Forshaw and Cooper, 2002), and represent nonsister taxa. The suggestion in the literature of intermediates (Moreau, 1958b) or possible hybridization (Brosset and Fry, 1988) “in a relatively narrow band” (Chapin, 1939) in the northeastern Congo Basin (e.g., Schouteden, 1950) would represent secondary contact of nonsister species. It seems clear that these two taxa have been recognized as conspecifics in the past based on their shared black bill; their striking plumage differences warranted them only subspecific rank.
Phylogeny
Evolutionary relationships among turacos, in the recent, quantitative phylogenetics era, have been investigated using morphological traits (Veron, 1999), karyotypes (Van Tuinen and Valentine, 1986), protein electrophoresis (Brush and Witt, 1983), and DNA sequences (Veron and Winney, 2000; Njabo and Sorenson, 2009). None of those prior studies included all the species-level taxa; additionally, the two previous DNA studies were based solely on mitochondrial sequences.
Our phylogenetic results, based on the ND2 mtDNA sequences, provide greater resolution, but are in general agreement with the results from our two nuclear loci (figs. 7, 8). In terms of generally recognized higher taxa, we obtained, for each of the three genes, monophyletic clades of go-away-birds (Criniferinae) and of green turacos (Musophaginae), plus a monotypic great blue turaco (Corythaeolinae). Veron and Winney (2000) reported similar results based on partial sequences of the mitochondrial cytochrome b locus for 22 taxa, as did Veron (1999) based on a cladistic analysis of 34 morphological characters for 23 species. Criniferinae and Musophaginae are also separated by two chromosomal inversions (Van Tuinen and Valentine, 1986).
The Aconitase data strongly (98% bootstrap) support Corythaeola as sister to the Criniferinae, while the ND2 data place it as sister to the Musophaginae with modest support (70%); the partitioned, three gene results are congruent (90%) with those of ACO1. Veron's (1999) cladistic analyses and Veron and Winney's (2000) neighbor-joining analyses placed Corythaeola as sister to the rest of the family; however, Njabo and Sorenson's (2009) Bayesian likelihood reanalysis of those cyt-b data placed Corythaeola as sister to the Musophaginae, albeit with very weak (0.56) support. The divergences among these three subfamily level taxa are old (perhaps 23 MYA according to Prum et al., 2015, or as much as 30–40 MYA in the dating by Njabo and Sorenson, 2009), and difficult for rapidly evolving mtDNA sequences to resolve, especially given the substantial base-composition heterogeneity we identified in that gene. The intermediate evolutionary rate of the aconitase intron, with its reduced base-composition heterogeneity, provided clearer signal (e.g., consistency index, bootstrap value).
Within the go-away-birds, our results, as well as all prior phylogenetic work, have indicated sister relationships between Crinifer piscator and C. zonurus and between Corythaixoides personatus and C. concolor. The relationships of Criniferoides leucogaster are not resolved: our combined analysis placed it as sister to Crinifer with moderate (78%) bootstrap support; Njabo and Sorenson (2009) placed it as sister to Crinifer and Corythaixoides with a weak Bayesian posterior of 0.83. Thus, the precise branching pattern within the Criniferinae remains uncertain.
Within the green turaco (Musophaginae) clade, the major differences between our phylogenetic results and much of the more recent work reflect our addition of Prince Ruspoli's turaco to the study, our addition of taxa previously treated as subspecies, and our much-improved sampling of individuals and populations. For the ND2 sequences, we found five major clades of these birds with bootstrap support of 98% to 100%; each of the five clades were many substitutions apart in the ND2 haplotype network of turacos (fig. 1).
A branch arising from the first node within the green turacos led to a clade comprised of members of the genus Gallirex (sensu Dickinson and Remsen, 2013). All recent authors have recognized a close relationship among these birds, although G. johnstoni and G. kivuensis were often placed in the genus Ruwenzorornis (table 1). Snow (1978) treated Gallirex and Ruwenzorornis as a superspecies, as did subsequent accounts such as those of Brosset and Fry (1988) and Forshaw and Cooper (2002). That the Gallirex (including Ruwenzorornis) complex is sister to the remaining green turacos was reported by Veron and Winney (2000) and confirmed by Njabo and Sorenson (2009).
The ND2 and aconitase genes both identified a second turaco clade comprised of leucotis, donaldsoni, and ruspolii. The prior mtDNA studies of Veron and Winney (2000) and Njabo and Sorenson (2009) included only leucotis (sensu stricto); the latter's Bayesian analysis placed leucotis in a position equivalent to our result for the clade of three taxa. None of the previous DNA studies included ruspolii; we placed it in the leucotis clade with high (100%) mitochondrial and combined bootstraps. Brosset and Fry (1988) thought hartlaubi was related to this assemblage and allied it with leucotis and ruspolii in a superspecies; our data strongly reject that hypothesis.
The third clade was comprised of two species, rossae and violacea, traditionally placed in the genus Musophaga, plus the macrorhyncha complex. The previous sequencing studies both identified this clade. Brosset and Fry (1988) thought there was a close relationship between the first two species, traditional Musophaga, and our first clade, the Gallirex complex; they based this treatment on plumage pattern and color, and on the cytological results of Van Tuinen and Valentine (1986). However, this latter justification was not supported by those karyotypic results; it represented a misinterpretation of symplesiomorphy as evidence for close relationship.
Our fourth clade, consisting of bannermani, erythrolophus, and leucolophus, present in all three genetic loci, was also identified by Njabo and Sorenson (2009). Traditional classifications (e.g., Brosset and Fry, 1988) have treated bannermani and erythrolophus as members of a superspecies; the autapomorphic plumage traits of leucolophus apparently masked its close relationship with the other two species. For example, Veron (1999) did not recover this clade in his cladistic analysis of 34 plumage and other morphological characters.
The final, largest, clade of green turacos was also identified in the other two molecular studies. However, because of our broad sampling, the clade here includes more species-level taxa than in those prior results. Additionally, the phylogeographic results indicate that some taxa formerly treated as conspecific (e.g., Dickinson and Remsen, 2013) are not even monophyletic: for example, T. schuettii and T. emini, and T. livingstonii and T. reichenowi. In addition, each of the superspecies assemblages recognized by Snow (1978) within these birds was either para- or polyphyletic. One genetically well-differentiated taxon, T. chalcolophus, was not even recognized as a valid subspecies by Turner (1997) or Dickinson and Remsen (2013), and many taxa that were 100% diagnosable based on the sequences and morphology were not recognized. As is apparent from the phylogeographic network (fig. 1), the green turaco clade represents a rapid, probably recent series of speciation events.
Classification
As pointed out above, our phylogeographic results require the recognition of species-level taxa of turacos not afforded that rank in recent classifications of the family (Brosset and Fry, 1988; Turner, 1997; Dickinson and Remsen, 2013). In addition, our results indicate that the genus Tauraco is not monophyletic in the first two of those classifications, and that several superspecies complexes recognized by Moreau (1958a), Snow (1978), Brosset and Fry (1988), Short et al. (1990), Sibley and Monroe (1990) and Turner (1997) are not monophyletic (table 4). The application of the biological species concept, and the subsequent allocation of such species to superspecies, has consistently obscured evolutionary relationships in this family of birds, even by authors (e.g., Short et al., 1990) who have themselves pointed out the danger of overreaching superspecific taxa (Amadon and Short, 1992). Although the failure to recognize nearly cryptic species and the recognition of polyphyletic species is understandable given the similarity of plumages in the green turacos, the recognition of polyphyletic superspecies represents an active error of using parapatry or of weighting some characters more highly than others in the absence of any phylogenetic analysis. Fry (1988) recognized this as a potential issue in this group. As a consequence of all these problems, we propose a classification that corrects the errors and provides a list of the taxa that would be essential for any study of evolution or biogeography of these birds or for a larger investigation of patterns of historical diversification on the African continent.
We recognize three subfamilies of musophagids: Corythaeolinae, Criniferinae, and Musophaginae. Recent classifications have varied in recognizing between two (e.g., Sibley and Monroe, 1990) and four (e.g., Verheyen, 1956) subfamilies; however, most have used three (Turner, 1997). Bock (1994) pointed out that Tauracidae and Tauracinae (Verheyen, 1956), originally Turacidae (Rafinesque, 1815), have priority over Musophagidae (Lesson, 1828), but Musophagidae has been used consistently as a family-group name for over 150 years.
There has been a long history of tension between lumping and splitting at the generic level. However, as Mayr (1943) pointed out, in a Linnaean classification the scientific name of a species consists of generic and specific designations; these are intended to represent alternate aspects of relationship: the specific epithet emphasizes differences and individuates the species, whereas the generic allocation is a collective and should carry information about similarity among species. The balance between the two in a classification is a matter of convenience and opinion. As the ratio of genera to species in a classification approaches one, the generic name becomes redundant—all the information content is in the species epithet; as the ratio of genera to species gets small, hierarchical content of the classification is lost.
Although von Boetticher (1947) used four genera for the five species of go-away-birds, most recent classifications have used three (e.g., Dickinson and Remsen, 2013). Nevertheless, with an average of only 1.67 species each, such genera are almost redundant in a Latin binomial. A more efficient classification would use a single genus; Crinifer has priority. We also note that in our maximum likelihood tree based on complete ND2 sequences (not shown), the divergences among the five go-away-birds are of the same magnitude as those among species within our genera Gallirex and Musophaga, and much less than among our proposed genera in the Musophaginae.
Within the green turacos, there were five major clades (e.g., fig. 1) identified by ND2 that were either concordant or consistent with the nuclear loci (fig. 7, fig. 8). Because species traditionally placed in the genus Tauraco were found in four of these, one must either recognize a very large genus Tauraco that includes at least 23 species, including Musophaga (as did Veron and Winney, 2000), or allocate traditional members of Tauraco to other genera. The former, a classification with two genera comprised of 4 and 23 species, seems unbalanced. Alternatively, von Boetticher (1947) recognized 13 genera and subgenera for his 17 species, a ratio of 1.3 species per generic-level taxon. We recognize five genera with an average of approximately five species per genus (and a range of three to 13). Names are already available for each of these.
TABLE 4.
Phylogenetic status of musophagine superspecies.
In several cases in which we have elevated taxa to the species rank, English common names were not available in the literature. Where appropriate, we have added geographic modifiers, such as eastern and western or northern and southern, to the current vernaculars. In other cases, we have suggested the use of modifiers based on relevant geographical or political names.
FAMILY MUSOPHAGIDAE Lesson, 1828
SUBFAMILY CORYTHAEOLINAE Verheyen, 1956 – blue turacos Genus CORYTHAEOLA Heine, 1860
Corythaeola cristata (Vieillot, 1816) great blue turaco
SUBFAMILY CRINIFERINAE Verheyen, 1956 – go-away-birds Genus CRINIFER Jarocki, 1821
Crinifer leucogaster (Rüppell, 1842a) white-bellied go-away-bird
Crinifer piscator (Boddaert, 1783) western grey plantain-eater
Crinifer zonurus (Rüppell, 1835a) eastern grey plantain-eater
Crinifer concolor (Smith, 1833) grey go-away-bird
Crinifer personatus (Rüppell, 1842b) bare-faced go-away-bird
SUBFAMILY MUSOPHAGINAE (Lesson, 1828) – turacos Genus GALLIREX Lesson, 1844
Gallirex porphyreolophus (Vigors, 1831) southern purple-crested turaco
Gallirex chlorochlamys Shelley, 1881 northern purple-crested turaco
Gallirex johnstoni Sharpe, 1901 Rwenzori turaco
Gallirex kivuensis (Neumann, 1908a) Kivu turaco Genus MENELIKORNIS von Boetticher, 1947
Menelikornis leucotis (Rüppell, 1835b) white-cheeked turaco
Menelikornis donaldsoni (Sharpe, 1895) Donaldson's turaco
Menelikornis ruspolii (Salvadori, 1896) Prince Ruspoli's turaco Genus MUSOPHAGA Isert, 1789
Musophaga rossae Gould, 1852 Lady Ross'S turaco
Musophaga violacea Isert, 1789 violet turaco
Musophaga macrorhyncha (Fraser, 1839) western yellow-billed turaco
Musophaga verreauxii Schlegel, 1854 eastern yellow-billed turaco Genus PROTURACUS Bates, 1923
Proturacus bannermani Bates, 1923 Bannerman's turaco
Proturacus leucolophus (von Heuglin, 1855) white-crested turaco
Proturacus erythrolophus (Vieillot, 1819a) red-crested turaco Genus TAURACO Kluk, 1779
Tauraco persa (Linnaeus, 1758) eastern green turaco
Tauraco buffoni (Vieillot, 1819b) western green turaco
Tauraco emini (Reichenow, 1893) eastern black-billed turaco
Tauraco hartlaubi (Fischer and Reichenow, 1884) Hartlaub's turaco
Tauraco fischeri (Reichenow, 1878) Fischer's turaco
Tauraco reichenowi (Fischer, 1880) Reichenow's turaco
Tauraco corythaix (Wagler, 1827) Knysna turaco
Tauraco livingstonii (Gray, 1864) Livingstone's turaco
Tauraco schuettii (Cabanis, 1879) western black-billed turaco
Tauraco chalcolophus (Neumann, 1895) Ngorongoro turaco
Tauraco schalowi (Reichenow, 1891) Schalow's turaco
Tauraco loitanus (Neumann, 1908b) Loita turaco
Tauraco marungensis (Reichenow, 1902) Zambia turaco
Biogeography
Nearly a century ago, Chapin (1923) published a map summarizing general avian distribution patterns across Africa; he recognized six ecological provinces and 17 districts, largely based on vegetation, which he thought reflected overall faunal diversity. Moreau (1966) provided a more detailed update of such patterns in his monograph on the African avifauna, but it was not until the compendia of Hall and Moreau (1970) and Snow (1978) that detailed distribution maps of most species-level taxa of African birds were assembled. Crowe and Crowe (1982) analyzed those data in a statistical assessment of the efficacy of vegetation as a surrogate for defining avifaunal zones and boundaries, and provided a hierarchical classification of such zones. Dowsett-Lemaire and Dowsett (2001) and de Klerk et al. (2002) reexamined those patterns in greater detail. More recently, Linder et al. (2012) summarized African biogeographic patterns across plants and vertebrates. Of course, most of those analyses were based on assessments of species limits prevalent at the time, before detailed molecular studies were available. Our phylogeographic results suggest there is substantially more diversity and genetic structure at varying geographic scales than prior authors had suspected. This represents an additional example of a pattern that has been suggested to be general across birds (Barrowclough et al., 2016).
The importance of montane regions of Africa as centers of diversity is widely recognized (Stuart et al., 1993; Burgess et al., 2007; Fjeldså and Bowie, 2008). Dowsett (1986) summarized the organization of the montane avifaunal regions of Africa and these are generally concordant with our clades (fig. 1) of montane turacos. However, with the exception of Bannerman's turaco, we observed significant taxonomic and geographic structure within his montane groups (e.g., Gallirex johnstoni, Tauraco hartlaubi); that is, phylogeography revealed finer geographic structure, presumably due to more recent historical events, than Dowsett (1986) identified using biological species as units of history. Multiple recent phylogeographic studies are consistent with this pattern of cryptic diversity within assemblages of African montane birds (e.g., Bowie et al., 2004, 2006, 2009). Voelker et al. (2010b) suggested that these montane patterns were driven by Pliocene forest dynamics.
There have been few avian phylogeographic surveys at the larger spatial scale of the extensive African lowland forest; this is perhaps due to the difficulty of obtaining fresh DNA samples from multiple political entities. Nevertheless, our results are generally consistent with those of several recent studies in uncovering significant geographic diversity within lowland avian species and subspecies (e.g., Marks, 2010; Fuchs et al., 2016; Huntley and Voelker, 2016). For example, in West Africa, in two lowland forest-associated traditional complexes, those of M. macrorhyncha and T. persa, we found strong genetic differentiation across or near the Dahomey Gap. The concordance of avian divergence across the Dahomey Gap/lower Niger River is a well-known pattern (e.g., Fuchs and Bowie, 2015) and corresponds to a gap between probable forest refuges (e.g., Diamond and Hamilton, 1980; Mayr and O'Hara, 1986; Maley, 2001). On the other hand, in the more northerly distributed, savannah-associated M. violacea, divergence across the region was not observed. Similarly, in P. leucolophus, distributed in dry forest and savannah north of the Congo Basin, we observed no pronounced geographic structure, but in the forest-dwelling T. schalowi complex, we observed substantial geographic structure. Thus, our results indicate the possible existence, both within montane and lowland forest avifaunas, of largely unappreciated diversity in many taxa. Parallel surveys and attendant discoveries have already begun on the mammalian fauna (e.g., Moodley and Bruford, 2007; Anco et al., 2017).
Perhaps the most interesting of our geographic observations were those of taxa with marked genetic breaks in locations not associated with major phenotypic divergence or current habitat discontinuities. Tauraco schalowi/T. marungensis and T. schuettii/T. emini appear to be parapatric, or possibly even allopatric, somewhere in the eastern or central Congo River Basin; this may be the result of past habitat fragmentation in the eastern Congo during the last (schalowi/marungensis) or an earlier (schuettii/emini) glacial cycle (e.g., Maley, 2001). Similarly, Gallirex porphyreolophus/G. chlorochlamys and T. livingstonii/T. reichenowi (the latter not a species pair) are differentiated in forested habitat in coastal eastern Africa; Fuchs et al. (2017) found mixed haplotype clades in some drongos in this same region. Future, denser sampling will be required to better characterize those zones.
Further Research
Not surprisingly, given the geographical heterogeneity of specimen collecting in sub-Saharan Africa, a great many problems remain in our understanding of turaco systematics. Perhaps the real import of our research is the discovery of how much remains to be investigated concerning geographic variation and species limits in turacos. First, we have yet to obtain DNA sequences for two potentially important populations, originally described as subspecies, that may represent differentiated taxa. These are the population of T. persa at the east-central portion of its range (T. p. zenkeri), and the population of Gallirex johnstoni on Mt. Kabobo in the eastern Democratic Republic of Congo (G. j. bredoi). In addition, the taxonomic status of the population of T. fischeri on Zanzibar is enigmatic and requires further attention. Second, our geographic sampling of green turacos has left large portions of the ranges of several of the species unsampled (e.g., Tauraco schalowi, P. leucolophus, and M. rossae). The T. hartlaubi complex probably harbors several additional species-level taxa and requires additional, especially nuclear, sequencing. In other cases, further sampling might allow for the quantitative characterization of possible zones of contact between sister taxa across western Africa (e.g., T. persa and T. buffoni) and zones of contact between sister taxa, such as T. corythaix and T. livingstonii in southern Africa and Menelikornis leucotis and M. donaldsoni in Ethiopia (Erard and Prévost, 1971). Distantly related pairs, such as T. livingstonii and T. reichenowi in eastern Africa, as well as Tauraco schalowi and T. marungensis, and T. schuettii and T. emini, both in the Congo Basin, all require attention to establish range limits. The Gallirex porphyreolophus plus G. chlorochlamys complex in Zambia, Zimbabwe, and Mozambique particularly requires attention. Third, thorough geographic surveys of the great blue turaco (Corythaeola) and of the go-away-birds (Crinifer) are necessary.
Finally, all our phylogeographic analysis has been based on mitochondrial DNA sequences. Although mtDNA is expected to provide a more sensitive indicator of recent geographical isolation than is nuclear DNA (Zink and Barrowclough, 2008), apparent instances of mitonuclear discordance are known in birds (Toews and Brelsford, 2012). A multilocus nuclear DNA survey of turaco phylogeography would be welcome.
ACKNOWLEDGMENTS
We thank Jonas Lai and Bill Mauck for assistance with lab work associated with this research. Curators and collection managers at the British Museum (Natural History) at Tring (Mark Adams and Robert Prys-Jones), the Burke Museum of the University of Washington (Sharon Birks and John Klicka), the Field Museum of Natural History (John Bates, Dave Willard, and Ben Marks), the University of Kansas Biodiversity Institute (Rob Moyle and Mark Robbins), the Yale Peabody Museum (Rick Prum and Kristof Zyskowski), and the Zoological Museum of the University of Copenhagen (Jon Fjeldså and Jan Kristensen) generously provided data and tissue and toe pad samples for our DNA sequencing; Paul Sweet and Tom Trombone assisted with those efforts. Mary LeCroy offered critical assistance with tracing the nomenclatural history of the Musophagidae; she and Tom Trombone helped identify obscure African localities. Gary Voelker, Rauri Bowie, and several anonymous reviewers offered useful comments on preliminary versions of the manuscript. Lab expenses for this research were provided by the Sackler Institute for Comparative Genomics, the Ambrose Monell Laboratory, and the Cullman Program for Molecular Systematics. Utku Perktaş was a Frank M. Chapman postdoctoral fellow at the American Museum of Natural History when this research was begun; Helen Hays generously assisted with his support and logistics in New York.
