Morphological analyses suggest a new taxonomic circumscription for Hymenaea courbaril L. (Leguminosae, Caesalpinioideae)

Abstract Hymenaea is a genus of the Resin-producing Clade of the tribe Detarieae (Leguminosae: Caesalpinioideae) with 14 species. Hymenaea courbaril is the most widespread species of the genus, ranging from southern Mexico to southeastern Brazil. As currently circumscribed, Hymenaea courbaril is a polytypic species with six varieties: var. altissima, var. courbaril, var. longifolia, var. stilbocarpa, var. subsessilis, and var. villosa. These varieties are distinguishable mostly by traits related to leaflet shape and indumentation, and calyx indumentation. We carried out morphometric analyses of 14 quantitative (continuous) leaf characters in order to assess the taxonomy of Hymenaea courbaril under the Unified Species Concept framework. Cluster analysis used the Unweighted Pair Group Method with Arithmetic Mean (UPGMA) based on Bray-Curtis dissimilarity matrices. Principal Component Analyses (PCA) were carried out based on the same morphometric matrix. Two sets of Analyses of Similarity and Non Parametric Multivariate Analysis of Variance were carried out to evaluate statistical support (1) for the major groups recovered using UPGMA and PCA, and (2) for the varieties. All analyses recovered three major groups coincident with (1) var. altissima, (2) var. longifolia, and (3) all other varieties. These results, together with geographical and habitat information, were taken as evidence of three separate metapopulation lineages recognized here as three distinct species. Nomenclatural adjustments, including reclassifying formerly misapplied types, are proposed.


Introduction
Hymenaea L. is a genus of caesalpinioid legumes with 14 species (Lee and Langenheim 1975). Th e genus is distributed throughout tropical America, from Mexico to Paraguay, with one species in coastal East Africa (Mackinder 2005). Hymenaea is included in the Resin-producing Clade of the tribe Detarieae, and is most closely related to the genera Guibourtia Benn. and Peltogyne Vogel (Bruneau et al. 2001(Bruneau et al. , 2008Fougère-Danezan et al. 2007), all presenting similar leaf morphologies, with two asymmetrical pellucid-punctate leafl ets. Hymenaea can be diff erentiated from these related genera by its rather larger and more massive bat-pollinated fl owers with a robust hypanthium, and indehiscent, pulpy and woody pods with very large seeds.
Th e current taxonomy of the genus is largely based on Lee and Langenheim's (1975) revision. In addition to recognizing the 14 currently accepted species, they reduced several species to varietal rank. Th ese polytypic species were viewed by Lee and Langenheim (1975) as showing complex relationships with other species of Hymenaea . Th ey hypothesized, for example, that H. oblongifolia Huber var. oblongifolia is more closely related to H. aurea Lee and Lang. and H. eriogyne Benth., while H. oblongifolia var. davisii (Sandwith) Lee and Lang. is probably more closely related to H. parvifolia Huber, H. rubrifl ora Ducke, and H. reticulata Ducke. According to their concepts, H. oblongifolia should be regarded as a polyphyletic species. Another example of a putative polyphyletic species, according to Lee and Langenheim's (1975) conceptual framework, is H. courbaril , with var. villosa Lee and Andrade-Lima hypothesized as being more closely related to H. martiana Hayne, and var. longifolia (Benth.) Lee and Andrade-Lima to H. velutina Ducke and H. stigonocarpa Mart. ex Hayne.
Hymenaea courbaril is the most widely distributed species of the genus, almost matching the geographic range of Hymenaea in the New World. It also has the greatest economic importance in the genus, due to the high quality of its wood and its resin, the latter being used by native populations as incense, cement, in the manufacture of varnishes, and for medicinal purposes. Its nutritive fruits are sought after by mammals and birds (Rizzini 1971;Langenheim 1967;Lee and Langenheim 1975). Hymenaea courbaril is the most taxonomically complex species, with six varieties: var. altissima (Ducke) Lee and Lang., var. courbaril , var. longifolia , var. stilbocarpa (Hayne) Lee and Lang., var. subsessilis Ducke, and var. villosa . Th ese varieties are diff erentiated by their leafl et sizes, shapes, and indumentation, calyx indumentation, petal shapes, ovary stipe sizes, and pod sizes and shapes. Th ese variations in several diagnostic features make the boundaries of putatively related taxa rather imprecise. Th e widely circumscribed H. courbaril , as defi ned by Lee and Langenheim (1975), is hereafter referred to as the H. courbaril complex.
Th e species and varietal limits of H. courbaril , H. stigonocarpa, and H. martiana were investigated by Pestana (2010). Th is author did not employ objective analytical methods and used the same classical taxonomic approach as Lee and Langenheim (1975), thus coming to similar conclusions as the latter authors in keeping H. courbaril as a polytypic species with six varieties.
Delineating precise species boundaries is a key task in plant taxonomy. Th is process has direct impacts on society, as there is a growing demand for credible taxonomic information that allows us to conserve, manage, and understand natural biodiversity (Wheeler et al. 2004). However, questions of species recognition can be aff ected by several theoretical, methodological, and practical issues. De Queiroz (2005Queiroz ( , 2007 defi ned species as separately evolving metapopulation lineages (the Unifi ed Species Concept-USC), and proposed that all other previously considered properties of species should be reinterpreted as contingent rather than critical. Th ese additional contingent properties, such as phenetic distinctiveness, reciprocal monophyly, genetic coalescence, or ecological distinctiveness, are acquired during speciation and should be considered as diff erent lines of evidence relevant to assessing lineage separation. Th e USC, by treating species conceptualization and species delimitation as clearly separate issues, allows the use of properties formerly treated as secondary criteria in species delimitations. More importantly, it allows for currently accepted species limits to be considered as hypotheses to be tested using the presence of any one of those secondary properties as evidence for the existence of a species.
Analyses of morphometric data can be useful in objectively demonstrating species limits, especially when combined with molecular markers (Andrés-Sánchez et al. 2009;Newmaster and Ragupathy 2009). Additionally, the use of morphological information represents the fastest and least expensive manner of assessing taxonomic complexesand has been used to solve problems of species limits in many diff erent plant groups, especially when molecular data was not easily available (Handerson 2006;Estrella et al. 2009;Pedersen 2010;Ceolin and Miotto 2012;Rahman and Rahman 2012;Castello and Galeto 2013;Scrivanti et al. 2013), as was the case of the group studied here.
Th e problem of defi ning species limits in the polytypic H. courbaril complex is revisited here under the USC conceptual framework by exploring morphometric, geographical, and ecological patterns as lines of evidence for the existence of separate metapopulation lineages. Specifi cally, we sought to test Lee and Langenheim's (1975) hypothesis that H. courbaril should be treated as a polytypic species with six varieties.

Materials and methods
A total of 96 specimens of the H. courbaril complex were examined in this study (vouchers listed in Appendix 1). All analyzed materials were sheets from the following herbaria: B, CEN, CEPEC, HRB, HUEFS, IBGE, IPA, LAGU, M, MBM, NY, RB, SP, SPF, U, UB, and UC. We selected specimens that displayed branch ends with fullydeveloped (mature) leaves. Th is criterion avoided considering young leaves from the tips of the branches or leaves at the bases of the branches that are often much larger. Each specimen analyzed was considered an individual, and identifi cations strictly followed Lee and Langenheim (1975), although these names were only used as nomenclatural references in this study.
Th e leaves in the group studied here are bifoliolate; the leafl ets range from oblong to ovate or obovate, with rounded, acute or obtuse apices; the main vein is displaced towards the inner margin resulting in an asymmetrical base, the outer portion being wider than the inner and extending beyond the attachment to the petiolule (Figure 1). Fourteen quantitative (continuous) characters were examined (Table 1; Figure 1). Only leaf traits were measured and quantifi ed, as most herbarium sheets lacked fl owers and/or fruits. In any case, fl ower morphology is much conserved in the species studied and the herbarium material examined usually contained only incomplete or damaged fl owers. Measurements were taken of two fully developed leaves per dried herbarium sheet, using a graduated ruler (precision 1 mm).
All multivariate analyses were carried out using Past software (Hammer et al. 2001). Cluster analyses used the Unweighted Pair Group Method with Arithmetic Mean (UP-  Table 1. GMA) based on Bray-Curtis dissimilarity matrices. Principal Component Analyses (PCA) were carried out based on the same morphometric matrix. Two sets of Analyses of Similarity (ANOSIM; Warwick, Clarke and Suharsono 1990) and Non Parametric Multivariate Analysis of Variance (NPMANOVA; Anderson 2001) were carried out to evaluate statistical support for: (1) the major groups recovered in the UPGMA and PCA; and (2) the varieties as defi ned by Lee and Langenheim (1975). Hymenaea courbaril var. villosa was not included in the second analysis because it is known from only two specimens (Lee and Langenheim 1975) and only one was located during this study. Both ANOSIM and NPMANOVA used Bonferroni corrections, 10,000 permutations, and Bray-Curtis distances. As H. courbaril var. villosa was represented by only a single specimen, we carried out a second set of UPGMA, PCA, and similarity analyses for the major groups without including this variety in order to test its infl uence on the results.
Distribution maps of the specimens studied were prepared using DIVA-GIS software (Hijmanns et al. 2005), based on the geographic coordinates recorded on the herbarium sheet labels. For material lacking original coordinates, a central coordinate for the municipality was used as provided by the Species Link website (available at http://specieslink.org.br).
Species limits were tested following the USC framework ( de Queiroz 2005de Queiroz , 2007. Species ranks were ascribed to groups that showed morphological and habitat distinctiveness and geographical consistency. We considered as morphologically distinct groups those that were recovered in UPGMA and PCA and that exhibited statistical signifi cance in both ANOSIM and NPMANOVA tests. Habitat distinctiveness was assessed from the vegetation type where the taxon occurs, following the UNESCO (1973) classifi cation. Species diagnoses were prepared based on vegetative and reproductive characters. Table 1. List of the quantitative leaf characters in specimens of the Hymenaea courbaril complex. Letters in the second column refer to measurements depicted in Figure 1.

Leaf characters (cm)
Measurements in Figure

Results and discussion
Both UPGMA and PCA recovered three major groups ( Figure 2): Group 1 included all specimens of var. altissima ; Group 2 all specimens of varieties courbaril , stilbocarpa , subsessilis , and the single specimen of villosa ; and Group 3 all specimens of var. longifolia . Within Group 2, individuals of the diff erent varieties did not cluster together and appeared intermixed in UPGMA, or formed highly overlapping groups in two fi rst axes of PCA. Th e fi rst PCA axis accumulated 88.3% of the total variance, with the two fi rst axes summing 93% of the observed variation. Leafl et length was the trait that explained most of the variation found in fi rst axis, and the three major groups were sorted mostly by leafl et size. Th ese results indicate that Group 3 includes specimens with largest leafl ets, and Group 1 the smallest leafl ets (Figure 3). Th e results of the UPGMA and PCA were consistent with both ANOSIM and NPMANOVA. Comparisons of the varieties of H. courbaril showed signifi cant differences (p < 0.05) between the varieties altissima and longifolia and all other varieties, but no signifi cant diff erences between the varieties courbaril , silbocarpa , and subsessilis that clustered in Group 2 (Table 2). Th e same analyses comparing the three groups recovered in UPGMA and PCA found signifi cant morphological diff erences between them (Table 3).
Th e complex morphological variations found in Group 2 appear to refl ect its wide geographic range and large genetic variability (Ramos et al. 2009). Phylogeographic studies in part of the geographic range of H. courbaril var. stilbocarpa clearly demonstrated the wide genetic base and geographic structure of this genetic variation (Ramos et al. 2009). Th e morphological distinctions among the varieties clustered in Group 2 are made by rather continuous characters, such as the color of the calyx lobes (ochraceous to golden or rusty brown), leafl et shape (broadly to narrowly falcate), and the shapes of the outer sides of the leafl et base (straight to nearly straight or rounded). Th e var. subsessilis presents a short ovary stipe (c. 2 mm long) that grades to a medium-sized stipe in var. stilbocarpa (3-4 mm) and then large in var. courbaril (4-6 mm).
Leafl et indument distinguishes var. villosa from the remaining varieties of Group 2 (Lee and Andrade- Lima 1974Lima , 1975. Th e ovary is also described as slightly pilose on one side, a condition not observed in other varieties of H. courbaril . Th e exclusion of var. villosa from the similarity analyses did not alter the results obtained with its inclusion (Table 3). Th e var. villosa is known from only two specimens from the eastern coastal areas of the states of Paraíba and Pernambuco (northeastern Brazil) where the range of var. stilbocarpa overlaps with that of H. martiana . Th is species ( H. martiana ) has tomentose leafl ets and an ovary with a tuft of trichomes near its base. Th e rarity of H. courbaril var. villosa , as well as the transitional nature of the leafl ets and ovary indumentation in relation to H. martiana and the other varieties of the H. courbaril complex clustered in Group 2, all suggest that var. villosa could represent a hybrid between H. martiana and H. courbaril .
Th e habitats of the taxa of Group 2 include mostly tropical ombrophilous alluvial (gallery) forests in areas subject to seasonally dry climates from Mexico and the  Lee and Langenheim (1975) Figure 1 and Table 1 for measurement details and Bray-Curtis distances).
Varieties are represented by the same symbols used for the map C Scatter diagram showing the fi rst two axes of the PCA using the same data matrix as the UPGMA analysis.
Ellipses represent the varieties as recognized by Lee and Langenheim (1975), and they are represented by the same symbols used for the map.  Caribbean islands to central Brazil, but not the Amazonian region (var. courbaril and stilbocarpa ), tropical ombrophilous lowland forests in central and eastern Amazon basin (var. subsessilis ), or tropical ombrophilous submontane forests in the northeastern Brazilian state of Paraíba (var. villosa ) (Figure 2).
Th e var. altissima (Group 1) constitutes a morphologically, geographically and ecologically consistent taxon. It is distinguished from the other varieties of H. courbaril complex by having smaller, falcate and acuminate leafl ets, not exceeding 6.5 cm long on the fl owering branches (Table 4). Th e leafl ets of this taxon have a thinner texture than the remaining varieties, allowing its tertiary venation to appear as raised and reticulate veins. Th e fl owers of var. altissima are also the smallest within the H. courbaril complex, measuring less than 15 mm long Langenheim 1975, Pestana 2010). Th is variety is distributed in tropical ombrophilous lowland and submontane forests of the Atlantic Forest phytogeographical domain along the eastern coast of Brazil (Figure 2). It was cited for the southeastern Brazilian states of São Paulo and Rio de Janeiro (Mattos 1968, Lee and Langenheim 1975, Rizzini 1978, Pestana 2010), but we found specimens occurring northwards into Espírito Santo and Bahia states. It was originally described as H. altissima Ducke (Ducke 1935). Lee and Langenheim (1974: 448) considered that the foliar characters and relative fl ower size "do not appear suffi ciently signifi cant to warrant specifi c status for this taxon" and considered it as a variety of H. courbaril .
Var. longifolia (Group 3) comprises the morphs with largest leafl ets. Th is variety also diff ers from the remaining groups by having oblong leafl ets with an obtuse apex, clawed petals, and fruits compressed and enlarged toward the apex (Table 4). It occupies a very particular habitat in submontane tropical drought-deciduous thorny forests in the Caatinga and Cerrado phytogeographical domains that extend from western Bahia and Piauí states to the Araripe mountain range in the border area between the states of Ceará and Pernambuco (Figure 2). It was fi rst proposed as a variety of Hymenaea splendida Vogel. Langenheim (1974, 1975) considered H. splendida var. splendida to be synonymous with H. courbaril var. stilbocarpa and segregated H. splendida var. longifolia as another variety of H. courbaril .
In addition to morphological distinctions, ecological and geographical information can be used for interpreting the three groups recovered by UPGMA and PCA as separately evolving lineages that would be considered diff erent species under the USC framework (see Taxonomy section). With respect to the variation in Group 2, studies of wider samplings, including especially the collections made after Lee and Langenheim's (1975) work, shows that the diagnostic features used for defi ning the varieties included in Group 2 have more complex variations. Indumented leafl ets, for example, the diagnostic character for var. villosa , can be found to varying degrees in var. stilbocarpa . A short stipe was used as a diagnostic feature of var. subsessilis, but this trait is variable and grades into var. courbaril . Finally, the distinctions between vars. courbaril and stilbocarpa rely on the types and colors of the calyx lobe indumentation, traits that tend to change during the duration of the fl ower. Th us, a more parsimonious way to treat the taxonomy of Group 2 is to consider it a variable species, without recognizing varieties. Description. Large trees, to 30 m tall. Petiole 1.1-1.7 (-2) cm long; petiolule 4-5 mm long; leafl ets 6-9 (-10.2) × 2.8-4 (-5.1) cm, elliptic, rarely ovate, straight or slightly falcate due to central vein with an angle c. 10°, apex acute or obtuse, rarely acuminate, base acute along the inner margin and rounded along the outer margin, distance from the inner margin to central vein 7-10 mm in the basal region. Flower buds 2.1-2.5 cm long; fl ower 2.2-3.1 cm long; hypanthium 7-15 mm long; petals 1.1-2.0 cm long. Fruit cylindrical, of uniform width, apex mostly rounded and apiculate.
Hymenaea courbaril is defi ned here more narrowly than the circumscription adopted by Lee and Langenheim (1975), as we are proposing the exclusion of the varieties altissima and longifolia and their recognition as distinct species. In this narrower sense, H. courbaril is characterized by leafl ets with slightly convex inner margins and midrib slightly arched, resulting in an elliptic or ovate outline, not or only slightly falcate, apex mostly acute or obtuse (Figure 3). It presents fl owers larger than the other species of the H. courbaril complex, and cylindrical fruits mostly with more than fi ve seeds (Table 4). It has a wide geographical range, mostly in diff erent nuclei of Seasonally Dry Forests in Mexico, Central America, Caribbean, northern South America, Bolivia, and Brazil (from the Amazon region to Paraná State in the south).
Nomenclatural problems with several names associated with H. courbaril were not amended by Lee and Langenheim (1975) when they undertook their taxonomic revision of the genus. No formal type for H. stilbocarpa was cited by Lee and Langenheim (1975), who stated that no specimen was cited in the original description of Hayne (1830). However, Hayne (1830) based H. stilbocarpa on material collected by Martius, citing "Wäscht in Brasilien in Wäldern der Provinzen S. Paulo, Minas Geraes und Bahia (Martius)". We found a specimen in the M herbarium collected by Martius in the Brazilian state of São Paulo (Santana farm) with an attached label indicating that this plant was distributed over the "Prov. Rio de Jº., S. Paulo, Minas Geraes, Bahia", which probably served as the original material for Hayne's description of H. stilbocarpa . Th us, we are lectotypifying this species with Martius' specimen held in M under the barcode number M-0215314.
Hymenaea confertifolia Hayne was based on material collected by Sellow and Olfers in Brazil ("Wächst in Brasilien (Olfers u. Sellow)"; Hayne 1830: table 9). Lee and Langenheim (1975: 88) stated that this name was based on Sellow 1025 and that the "holotype" in  Hymenaea courbaril var. obtusifolia Ducke was published based on a tree cultivated in the Pará Botanical Garden (now Museu Paraense Emílio Goeldi, Belém, state of Pará, Brazil). Lee and Langenheim (1975) misinterpreted this as evidence that the holotype was the specimen in the herbarium of this museum (MG). However, Ducke (1925: 47) explicitly stated that "specimina fl orifera and fructus in herb. Jard. Bot. Rio n. 16.906", and thus the RB material should be considered as the holotype of this name. Ducke (1925: 265) did not cite any specimen when describing H. courbaril var. subsessilis Ducke. Lee and Langenheim (1975) did not explicitly designate a type for this variety, but chose a "representative specimen" collected "in the same general area believed to be the type locality". We are accepting it as an inferential typifi cation, and the status of this material should be a neotype since no other specimen was refereed in the protologue. Description. Large trees, up to 38 m tall. Petiole 1-1.3 (-1.7) cm long; petiolule 3-4 (-5) mm long; leafl ets 4-5.7 (-6.4) × 2-2.5 cm, oval, narrow-elliptic or elliptic, strongly falcate, due to central vein with an angle c. 35°, apex acuminate, rarely acute, base acute along the inner margin and truncate along the outer margin, distance from the inner margin to central vein 4-5 mm in the basal region. Flower buds 1.5-2.5 cm long; fl ower 2-2.7 cm long; hypanthium 7-12 mm long; petals 1.2-1.6 cm long. Fruit cylindrical, of uniform width, apex mostly rounded and apiculate.
Hymenaea altissima shows the smallest leafl ets and fl owers in the species group related to H. courbaril (Table 4). Additionally, its leafl ets are strongly falcate due to the strongly curved and displaced midvein, with an abruptly acuminate apex. Th e base of the leafl et is acute along the inner margin and truncate along the outer margin (Figure 3). Th e fruits are mostly cylindrical, as in H. courbaril, but usually shorter than those of this species (4-5 vs. 8-20 cm long) because they have only one or two seeds, while in H. courbaril they usually have six or more seeds. Th is species is restricted to the coastal rain forests of eastern Brazil, from southern Bahia State to São Paulo and Paraná states.
Th is species was described by Ducke (1935) as H. altissima . Lee and Langenheim (1974) treated it as a variety of H. courbaril . Later, these latter authors highlighted the differences between this taxon and the other varieties as having smaller and falcate leafl ets and smaller fl owers (Lee and Langenheim 1975: 86), but kept it as a variety of H. courbaril , a position also adopted by Pestana (2010). Our results indicated that the leaf traits of H. altissima are clearly distinct of those of H. courbaril and H. longifolia . Th ese results, together with the distinctive fl ower and fruit traits and the coherent distribution and habitat data, all give support to the original view of Ducke (1935) in considering it a diff erent species. Lee and Langenheim (1975: 84) stated that the lectotype of H. altissima was designated by Egler (1963) as the specimen " A. Ducke RB 30306 ". However, Egler (1963: 51) simply indicated that the type should be "Type: RB 23.306". We could not track any specimens of Hymenaea in the RB herbarium with the numbers RB 30306 or RB 23306. We encountered, however, a specimen annotated by Ducke as H. altissima that was collected in the state of "Rio de Janeiro, Avelar" by G. M. Nunes in 1925 and could be linked to the protologue of H. altissima (Ducke 1935). Th is specimen is numbered RB 20306, and we are assuming that both Egler (1963) and Lee and Langenheim (1975) erroneously noted the RB number when referring to the lectotype of H. altissima . Description. Medium-sized trees c. 5-12 (-18) m tall. Petiole 2.6-3 cm long; petiolule 7-8 mm long; leafl ets 10.1-12.5 (-15.4) × 4.5-6.5 (-6.9) cm, elliptic, narrowelliptic, or oblong, not falcate, due to central vein with an angle c. 24°, apex obtuse or acute, base acute or rounded along the inner margin and rounded or slightly rounded along the outer margin, distance from the inner margin to central vein 10-11 mm in the basal region. Flower buds 2.2-3.3 cm long; fl ower 2.6-3.8 cm long; hypanthium 9-14 mm long; petals 1.3-2.1 cm long. Fruit oblongoid, compressed, wider at distal region, apex slightly truncate and apiculate. Hymenaea longifolia , as circumscribed here, is characterized by large leafl ets, both longer and wider than those of the related species H. courbaril and H. altissima , with longer petioles and petiolules (Table 4). Th e leafl ets are elliptic or oblong with obtuse (rarely acute) apices and rounded bases along the outer margin (Figure 3). Th is species occurs in Seasonally Dry Forests within the Caatinga and Cerrado phytogeographical domains in northeastern Brazil, from Ceará State to the northern portion of Bahia State. It was fi rst described by Bentham (1870) as H. splendida var. longifolia, and considered related to H. courbaril and H. stilbocarpa because of the glabrous leafl ets, but diff ering from them by having larger leafl ets. Lee and Langenheim (1975) treated all these taxa under a more widely circumscribed H. courbaril , a position not supported by the results presented here. Lee and Langenheim (1975: 86) misinterpreted the specimen Blanchet 3135 (R) as the holotype of H. splendida var. longifolia . However, when describing this variety, Bentham (1870: 236) cited two syntypes, Blanchet 3135 ("ad Villa da Barra prov. Bahia") and Gardner 1938 ("Serra da Araripé, prov. Ceara"). Th us, the material cited by Lee and Langenheim (1975) as the holotype should be considered as a lectotype.