Morphological Variation in the Vriesea procera Complex (Bromeliaceae, Tillandsioideae) in the Brazilian Atlantic Rainforest, with Recognition of New Taxa

Abstract The family Bromeliaceae is essentially Neotropical, with high endemism and diversity in the Atlantic Rainforest Domain. Species circumscription is a major problem in the family systematics, especially in the most diverse genera. Species of the Vriesea procera complex, which occur in forests and restinga (coastal vegetation) along the South American Atlantic coast from Venezuela to southern Brazil, share the same basic vegetative and reproductive morphological patterns. However, they vary widely in the number and position of inflorescence branches as well as in the dimensions, position, and shape of the leaves and flowers in different populations. Here we aimed 1) to evaluate the morphological variation in the V. procera complex, through morphometric analyses of natural populations along the Brazilian Atlantic Rainforest; and 2) to determine the taxonomic relationships among these species, establishing the validity and the limits of variation of the taxa through taxonomic treatment. Fourteen natural populations, 271 individuals, and 36 morphometric variables were analyzed. Kruskal-Wallis tests and discriminant analyses were conducted to test statistical differences between previously established groups. Of an original three species and three varieties, our data allowed us to recognize six species, including three new taxa (Vriesea aureoramosa, V. magna, V. rubroviridis) and one new name and status (V. flexuosa). The resulting taxa are distinguished by the inflorescence and leaf sizes and especially by floral characteristics such as the length of bracts, sepals, petals, stamens, and pistils, besides the petal apex posture and relative position of the stamen during anthesis. Recognition of cryptic species under the names V. procera and V. neoglutinosa is an important step toward a better understanding of the biodiversity of the Brazilian Atlantic Rainforest.

Bromeliaceae Juss. is one of the most morphologically and ecologically diverse Neotropical angiosperm families (Givnish et al. 2011). Studies on diversity have shown that increases in the number of bromeliad species may be related to greater environmental heterogeneity and the consequent structural diversity of the forest (Nadkarni and Matelson 1989). Tillandsioideae ranges more widely across the Neotropics than the other seven subfamilies (Smith and Downs 1977;Barfuss et al. 2016). It is also the richest and morphologically diverse, with 22 newly recircumscribed genera (Barfuss et al. 2016;Leme et al. 2017). Among them, Vriesea Lindl. has about 200 species, occurring in different phytophysiognomies of Atlantic Rainforest and Cerrado domains, its main center of diversity (Martinelli et al. 2008;Costa et al. 2014; Gomes-da-Silva and Souza-Chies 2017) and richness (Stehmann et al. 2009). The genus distribution extends to northwestern Argentina, Bolivia, Peru, Venezuela, and the Greater Antilles (Barfuss et al. 2016).
The genus is especially diverse in floral morphology (petal shape, corolla type, and position of the stamens in relation to the corolla fauce) and in the color of inflorescence bracts ). These features characterize the two traditional sections, V. sect. Vriesea and V. sect. XiphionÉ.Morren (Smith and Downs 1977;Costa et al. 2014), whose monophyly was not corroborated in any phylogeny performed on the genus to date (Gomes-da-Silva et al. 2012;Costa et al. 2015; Barfuss et al. 2016; Gomes-da-Silva and Souza-Chies 2017). On the other hand, different groups of morphologically related species were recovered and are recognized as natural (Costa et al. 2015; Gomes-da-Silva and Souza-Chies 2017). However, some of them have species with difficult delimitations, and conflicting nomenclatural and taxonomic histories .
Recently, some of these species groups or complexes have been studied in order to enable a taxonomic revision of Vriesea ; Gomes-da-Silva and Costa 2011; Versieux 2011; Moura and Costa 2014; Neves et al. 2018).
The complex related to Vriesea procera (Mart. ex Schult. & Schult. f.) Wittm. is part of the V. sect. Vriesea, and is composed of taxa that have the same basic pattern in rosette and inflorescence morphology, and encompasses three species and four varieties ( Table 1). Tillandsia procera (its basionym) was collected by Carl Friedrich Philipp von Martius in the forests of the Itaipé River, in the south of the state of Bahia, in a Brazilian Atlantic Rainforest area, in the early 19th century. Later, the monographers of Bromeliaceae, especially Mez (1896) and Smith and Downs (1977), included in the circumscription of the species records from a wide geographic area including the north of South America, greatly extending its morphological and geographic limits. Moreover, the description of new related taxa (including varieties and species) also contributed to the misunderstanding about their limits. This way, the taxa of the complex vary enormously in: 1) dimensions, posture, and shape of the leaves, and 2) number and position of the inflorescence branches, apex posture and shape of the petals, and the position of the stamens and pistil in relation to the corolla fauce during anthesis ; Moura et al. 2007). These variations, together with the imprecise delimitations in the original descriptions, led to conflicts in determinations, as reflected in herbarium collections, especially around the names V. procera and V. neoglutinosa Mez Costa et al. 2014; Moura and Costa 2014). Throughout the area of occurrence of this species complex, especially in lowland forests and restingas, distinguishing between these taxa is difficult (Smith and Downs 1977;Reitz 1983;Araújo 2000;Moura et al. 2007).
Considering the huge morphological variation throughout the wide geographic distribution of the complex, we performed morphometric analyses in natural populations along the Atlantic Rainforest Domain due to the occurrence of large populations, and the type locality of Vriesea procera at south Bahia, as a first attempt to the understanding of its morphological variation.
Here we aimed 1) to evaluate the morphological variation in the V. procera complex, through morphometric analyses of natural populations along the Atlantic Rainforest domain; and 2) to determine the taxonomic relationships between these species, establishing the validity and the limits of variation of the taxa through taxonomic treatment. Recognition of cryptic species presently under the names V. procera and V. neoglutinosa is an important step toward a better understanding of the biodiversity of the Brazilian Atlantic Rainforest.

Materials and Methods
Morphometric Analysis-Populations and Localities-For morphometric study, vegetative and reproductive structures of natural populations were analyzed. To select populations, specific literature and collections of the herbaria R, RB, GUA, HB, SP, and MBM (see acronyms in Thiers 2017) were consulted. We analyzed 14 natural populations morphologically related to V. procera (Table 2), along the Atlantic coast from southern Bahia (BA) to northeast Paraná (PR) (Fig. 1). The individuals were randomly sampled; the size of the samples varied according to their availability and accessibility. Voucher specimens were deposited in the Herbarium of the National Museum of Rio de Janeiro (R). During the field work, the locations indicated in the herbarium collections for the states of Paraíba, Pernambuco, Alagoas, Sergipe, Minas Gerais, Santa Catarina, and Rio Grande do Sul were visited. However, flowering individuals were observed in none of the populations, hampering their inclusion in morphometric analyses.
Data Analyses-The analyses were perfomed using STATISTICA 8.0 (StatSoft Inc. 2007) and PAST v. 2.04 (Hammer 2010). The normality and homoscedasticity of the data variances were tested using the Shapiro-Wilk (Shapiro and Wilk 1965) and Levene (Levene 1960) tests, respectively. The Kruskal-Wallis test (KW) was also performed to determine statistical differences.
A discriminant analysis (DA) were performed to test for statistical differences between groups that were previously established, to assess the adherence among individuals from each group, and to indicate the characteristics that contributed most to the discrimination (Hair et al. 2009). The groups were composed of individuals from the 14 natural populations. Data available from Dryad (Uribbe et al. 2020).  The variables were considered significant for DA based on the distance of each variable from the center of the axis: d $ √2 / n, where n is the number of variables (Legendre and Legendre 1998), corresponding to d 5 0.39 in this case.
Taxonomic Treatment-For the taxonomic analysis, the main reference works on Bromeliaceae (Mez 1894(Mez , 1896(Mez , 1934(Mez -1935Harms 1930;Smith and Downs 1977;Luther and Sieff 1994, 1997a, 1997bLuther 2001;Luther and Rabinowitz 2010;Luther 2012) were consulted for information on protologues and taxonomic history. The collections of the herbaria CEPEC, HB, HST, IPA, MBML, R, RB, SP, TEPB, UFPB, and VIES were consulted and images from the K herbarium were analyzed through the website http://apps.kew.org/herbcat/navigator.do (abbreviations according to Thiers 2017). The descriptions are based on analyses of herbarium material, collections, field observations, and the results of the morphometric analyses. The general terminology used in the morphological description follows Stearn (1973), Smith and Downs (1977), Radford (1986), and Scharf and Gouda (2008). The material examined is listed in a sequence from north to south, followed by the chronological order and collector number.

Results
We visited 25 different localities. We constructed a spreadsheet of 575 individuals analyzed, together with the herbarium materials and field collections.
Morphological Analysis-Twenty-five areas were visited, totaling 32 days of fieldwork, which provided better observations of the differences among populations and individuals. We analyzed 239 individuals, with a minimum of six and a maximum of 20 per population sampled. For all variables, the Shapiro-Wilk and Levene tests showed absence of normality and homoscedasticity, respectively (Supplemental Tables S1, S2; Uribbe et al. 2020).
Kruskal-Wallis H Test-The results of the KW H test for the 36 variables analyzed and the most important variables selected are shown in Table S3 (Uribbe et al. 2020) and Figs. 2 and 3, respectively. The Ilha Grande population (RJ5) was distinguished from the others by the widest sheath ( Fig. 2A), and the Itaguaçú (ES3), Canoas (RJ1), andÁgua Viva (ES2) populations by the narrowest sheaths. The RJ5 population had the longest sheath while the RJ1 population had the shortest one (Table S3; Uribbe et al. 2020). Characters of the leaf blade also separated the populations. Caraíva (BA1) and RJ5 had the widest blades and RJ1 and ES3 the narrowest ones (Fig. 2B). RJ5 had the longest blades, while Canoas (RJ1) had the shortest ones (Table S3; Uribbe et al. 2020).
Regarding the inflorescence, Ilha Grande (RJ5) was distinguished from the other populations by the largest total length (Fig. 2C), the longest peduncle, and the largest peduncle diameter, while RJ1 (Canoas) had the shortest peduncle (Table S3; Uribbe et al. 2020 Table 1  largest number of branches ( Fig. 2D). Considering the length of the inflorescence median branch, RJ5 had the longest ones with the largest number of flowers, and BA1, the longest internodes (Table S3; Uribbe et al. 2020). In relation to the inflorescence distal branch, RJ5 also had the longest branch with the largest number of flowers (Table S3; Uribbe et al. 2020). The Ilha do Cardoso (SP1) and Caraíva (BA1) populations had the longest distal-branch internodes. Populations SP1 and PR had the longest distal-branch peduncle (Table S3; Uribbe et al. 2020).
Discriminant Analysis-In the DA performed with all natural populations studied, the first two axes explained 64.6% of the variation between groups (46.71% and 17.9% respectively) ( Table 4). Analysis of all populations combined showed that the null hypothesis must be rejected, and the type 1 error was accepted, with a separation into four distinct groups ( Fig. 4). On Axis 1, the population at Canoas (RJ1) was clearly separate, with the most important variables being the number of branches, sepal length, and sepal width. Also on Axis 1, and segregated by the same variables that separated RJ1, was a second group, formed by the Ilha do Cardoso (SP1), Picinguaba (SP2), and Ilha do Mel (PR) populations. The third group was formed by the Itaguaçú population (ES3), which differed in the number of branches, length of the proximal branch, and sepal width. The fourth group was formed by the remaining populations.
Additionally, aiming to obtain a better resolution, an additional DA was performed only sampling individuals of that fourth group, which encompassed the populations BA1, BA2, ES1, ES2, RJ2, RJ3, RJ4, RJ5, RJ6, and RJ6. In this analysis the first two axes explained 51.3% of the total variance (26.6% and 24.7%, respectively) ( Table 5). Again we rejected the null hypothesis and accepted the type 1 error, with separation of three groups (Fig. 5). The first was formed by RJ6, which was separated on Axis 2 due to the number of branches, length of the median branch, and pistil length. The second group was formed by RJ5, which was discriminated on Axis 2, even with some overlapping, through the total length of the inflorescence, peduncle length of the distal branch, and petal length. The third group was formed by the remaining populations.
Summary of the Results of the Different Analyses-The morphometric analyses, added to the analyses of herbaria collections, including the nomenclatural types, allowed us to distinguish six taxa, including a new status and three new species.
Canoas (RJ1) was distinguished from the others mainly by the sepal length and width and number of branches (DA and KW), as well as the width and length of the sheath, width and length of the leaf blade, length of the peduncle of the inflorescence, length and width of the floral bracts, length and width of petals, and lengths of stamens and pistil (KW test). This population represents V. procera var. tenuis. The statistical analyses, along with the taxonomic treatment, led us to elevate this taxon to species rank.
The populations SP1, SP2, and PR together formed the second group, which was characterized by sepal length and width, and number of branches (DA and KW test), median peduncle internode length, median branch internode length, distal branch, rachis diameter, petal, sepal, stamen, and pistil length (KW test). This group represents a new species.
For the ES3 population, the variables number of branches, proximal branch length, and sepal width (DA), along with petal length and width, and stamen and pistil length (KW test) contributed to characterize this population as a second new species.
The RJ5 population was determined by number of branches, median branch and pistil length (DA), together with sheath width and length, leaf blade width and length, total length of the inflorescence, and peduncle of the inflorescence length (KW test), is interpreted as the third new species.
The RJ6 population represents V. neoglutinosa. This population stands out from the others by total length of the inflorescence, peduncle length of the distal branch, and petal length (DA), along with length of the median peduncle internode length, sepal width, and pistil length (KW test).
The remaining populations constitute the V. procera species (sensu stricto).
Distribution and Habitat-Plant endemic to Espírito Santo State, occurring as an epiphyte on inselbergs at altitudes between 300 and 650 m a. s. l. Vriesea aureoramosa has a restricted distribution and is presently known from two inselbergs in the state. The first of these, Morro da Torre de TV in the municipality of Itaguaçú, is the type locality. The other is the Pedra Lisa inselberg, district of Burarama, in the municipality of Cachoeiro do Itapemirim. The species blooms from January to March.
Individuals of V. aureoramosa were abundant and frequent in the above-mentioned localities. The ES3 population (Table 2) represents the taxon in the morphometric analyses.
Etymology-The epithet "aureoramosa" refers to the yellow color of the branches.
Distribution and Habitat-Plant endemic to the Atlantic Rainforest in the states of Bahia, Espírito Santo, Rio de Janeiro, and São Paulo. Vriesea flexuosa is found in low-montane and montane forests, and blooms from January to March.
The population of Canoas (RJ1) ( Table 2) represents the taxon in the morphometric analyses.
Etymology-The epithet "flexuosa" refers to the delicate flexuous branches.
Herb epiphytic or terrestrial. Rosette forming a tank, infundibuliform. Leaves recurved; sheath 12.8-31.7 3 6.3-12.8 cm, ovate, pale brown or brown at base and green near the blade, lepidote on both surfaces; blade 17.8-30 3 5.5-8 cm, narrowly elliptic, apex obtuse with acuminate projection, green, sparsely lepidote on the abaxial surface and dense on the adaxial one. Inflorescence 137-247 cm long, suberect, with 7-22 branches, flowers lax; peduncle 71-145 3 0.8-1.5 cm, red, glabrous; lower peduncle bracts leaf-like and the upper ones 5-23 cm long, ovate, acuminate apex, red with greenish apex, longer or shorter than the internode, imbricate or not, sparsely lepidote on the abaxial surface and dense on the adaxial one; branches 13.5-52.4 3 0.2-0.3 cm, suberect to pendulous, bearing sterile bracts on the peduncle, red, sparsely lepidote, peduncle with sterile bracts; rachis 66-102 cm long, straight, red, glabrous; primary bracts 2.7-14 cm long, shorter than the branch peduncles, ovate to narrowly elliptic, attenuate apex, red, sparsely lepidote on the abaxial surface and dense on the adaxial one; floral bracts 2.7-3.9 3 1.5-2.6 cm, shorter than the sepals, ovate, striate, carinate, involute, yellow, with a transparent,  Distribution and Habitat-Plant endemic to Rio de Janeiro State. The species was found in the restinga of Ilha Grande, where it is abundant and frequent, blooming from January to March, and fruiting all year round. The taxon is represented in the morphometric analyses as the RJ5 population (Table 2).
Etymology-The epithet "magna" refers to the large dimensions of the plant.
The individuals of the species were abundant and frequent in all the localities visited. The taxon is represented in the morphometric analyses as the RJ6 population (Table 2). Table 5. Canonical coefficients derived from discriminant analysis, partial analysis of the populations of the Vriesea procera complex, with BA1, BA2, ES1, ES2, RJ2, RJ3, RJ4, RJ5, and RJ6. Variables that most influenced the separation of groups are in bold. *Values that are outside the contribution circle of the variables (√2 / n variables). Codes and names of variables as in Table 3. Etymology-Vriesea neoglutinosa is a new name given by Mez (1894) when he transferred Tillandsia glutinosa Mart. to Vriesea. The new name was necessary because the epithet "glutinosa" already in use for in Vriesea glutinosa Lindl., a species from Trinidad (Smith and Downs 1977).
Notes-Vriesea neoglutinosa was described from a specimen collected by Martius in Botafogo, city of Rio de Janeiro, probably on an inselberg or in a restinga area. It was found with only preanthesis flower buds, a stage when all the morphologically related species are very similar, resulting in an unclear circumscription. Our description is based on individuals observed in the field, whose characteristics were similar to those observed in individuals from the type locality. This species is very close to V. procera and is often confused with it. Its main differences are the position of the leaves (suberect vs. recurved, Fig. 7E, A, respectively), posture of the petals (strongly recurved vs. slightly recurved), and position of the stamens during the anthesis (more exserted in V. neoglutinosa) (Figs. 6A, 7B, F). Differentiation of the two species is especially difficult in herbarium material, since the main differences are lost. Therefore, it is essential to provide detailed characterization on the specimen labels.    Table 2.
Vriesea procera may have a terrestrial habit but is more commonly found as an epiphyte.
Etymology-The epithet "procera" from the Latin adjective "procerus" meaning very tall (Stearn 1973), refers to the length of the inflorescence.
Notes-Vriesea procera was described from a fruiting specimen, making its circumscription complex and unclear. Vriesa procera is a very polymorphic species, usually with branched inflorescences of varying sizes. However, individuals with simple inflorescences were frequently observed in the populations visited. The specimens of V. graciliscapa and V. procera var. rubra and V. procera var. debilis show a pattern of variation matching that of V. procera, so all three taxa should be better regarded as belonging to a single biological entity and should be included in synonymy.
The morphologically closest species is V. neoglutinosa, which is often confused with this taxon. Its main differences are the position of the leaves (recurved vs. suberect, Figs  Vriesea rubroviridis is distinguished from V. flexuosa F.P.Uribbe & A.F.Costa by the erect to suberect branches (vs. patent or pendulous), peduncle and branches green and sometimes reddish (vs. green), primary and floral bracts reddish with green apex (vs. green), sepals and petals longer (2.2-2.8 cm and 2.7-3.4 cm, respectively) and yellowish green (vs. pale yellow colored and sometimes slightly greenish).
Distribution and Habitat-Plant endemic to the Atlantic Rainforest in the states of São Paulo, Paraná, Santa Catarina, and Rio Grande do Sul. Found in restingas and lower montane and montane forests, the species blooms from January to March and fruits year-round. Individuals of V. rubroviridis were abundant and frequent in all the localities visited. The SP2, SP1, and PR populations (Table 2) represent the taxon in the morphometric analyses. The population from Torres, state of Rio Grande do Sul, was not included in the morphometric analyses due the presence of only few flowering individuals found during the field work.
Etymology-The epithet "rubroviridis" refers to the green to pale-red blurred peduncle and rachis, and the base of the peduncle bracts, primary bracts, and floral bracts.
Notes-Vriesea rubroviridis is distinguished from the remaining taxa of the complex by its smaller flowers, almost erect petals, and stamens clearly included (Figs. 6A, D, 7J). It has a pale-red blurred peduncle and bracts (vs. green in V. flexuosa), branches suberect (vs. patent to pendulous in V. flexuosa) (Figs. 7I, C, respectively), and yellow sepals and petals (vs. sepals and pale yellow colored and sometimes slightly greenish petals in V. flexuosa) (Fig. 7J, D, respectively). Vriesea catharinensis F. Müller was for a long time assigned as a synonym of V. procera (Mez 1934(Mez -1935Smith 1955;Smith and Downs 1977). However, no original material was found and no illustration is available. In the protologue, Müller (1893) described a plant with several stolons, which is the only information on its morphology. As this structure is absent in V. procera, we inferred that V. catharinensis is likely a plant related to V. vagans (L.B.Sm.) L.B.Sm., a very common species in the Atlantic Rainforest remnants in southern Brazil.

Discussion
The distinction between V. neoglutinosa and V. procera has long been doubtful, and consequently erroneous determinations and citations are common in the literature (Smith and URIBBE ET AL.: MORPHOMETRICS OF VRIESEA PROCERA COMPLEX 65 202065 ] Downs 1977Reitz 1983;Araújo 2000;Cogliatti-Carvalho et al. 2001;Costa and Dias 2001;Rocha et al. 2003;Nunes-Freitas et al. 2006;Moura et al. 2007). The present morphometric analyses performed with natural populations allowed us to recognize six species: V. procera, V. neoglutinosa, V. magna, V. aureoramosa, V. flexuosa, and V. rubroviridis. The new species emerged from the morphological variation in taxa that have existed under the names V. procera and V. neoglutinosa. The resulting taxa are distinguished by their inflorescences and leaf sizes, and especially by floral characters such as the length of bracts, sepals, petals, stamens, and pistils, besides the petal apex posture and the relative position of the stamens during anthesis. Recognition of cryptic species under the names V. procera and V. neoglutinosa is an important step toward a better understanding of the biodiversity of the Brazilian Atlantic Rainforest.
The morphometric analyses provided clear geographical distributions for most species of the complex. The results indicated a separation of the populations into two large groups along the Atlantic Rainforest: the first one from the latitude S 25°towards the south, between the states of São Paulo to Rio Grande do Sul (V. rubroviridis and V. flexuosa), and the second one from the latitude S 22°towards the north between the states of Rio de Janeiro and Bahia (V. procera, V. flexuosa, V. neoglutinosa, V. magna and V. aureoramosa). Populations of other Vriesea species showed a north-south genetic discontinuity in the Atlantic Rainforest ( V. carinata, Zanella 2013;and V. gigantea, Palma-Silva et al. 2009), as do other groups of plants and animals (Harris et al. 2005;Pellegrino et al. 2005;Grazziotin et al. 2006;Cabanne et al. 2007;Carnaval et al. 2009;Fitzpatrick et al. 2009;Pinheiro et al. 2011;Ribeiro et al. 2011). This phytogeographic division is explained by the hypothesis of Pleistocene refuges, where species survived isolated in favorable environments along the coast. In the Quaternary, climate oscillations during the Pleistocene caused changes in sea levels and expansions and contractions in forest areas, influencing the vegetation type and distribution of the species in the coastal region of Brazil. After the Last Glacial Maximum, with the increases in temperature and precipitation, the Atlantic Rainforest expanded toward the south. The northernmost areas appear to have remained stable during the climate oscillations; the present-day northern populations are older and the southern populations more recent (Haffer 1969;Vanzolini and Williams 1981;Rull 2006;Carnaval et al. 2009).
Phylogenetic and biogeographical hypotheses on Tillandsioideae showed the monophyly of the Brazilian lineage of Vriesea (Barfuss et al. 2016; Gomes-da-Silva and Souza-Chies 2017; Kessous et al. 2019), encompassing about 90% of the species of the genus. The Amazonian, Caribbean, and Central American species were poorly sampled and their relationships with the Brazilian lineage remain unknown. The occurrence of widely distributed species (e.g. V. procera, V. simplex (Vell.) Beer, V. wawranea Antoine) suggests the importance of the inclusion of extra-Brazilian accessions in phylogenies with molecular data in order to infer if these species were dispersed from the Atlantic Rainforest and Cerrado domains, or if they are the result of morphological convergence. Besides, the phylogenies proposed so far for Vriesea have failed to delineate natural infrageneric groups (Costa et al. 2015; Gomes-da-Silva and Souza-Chies 2017). On the other hand, some small groups of morphologically similar species were recovered, especially among species with simple inflorescences (e.g. Gomes-da-Silva and Souza-Chies 2017). The compound inflorescence species complexes like those related to V. procera, V. friburgensis Mez, and V. morrenii Wawra, to mention just a few examples, are poorly sampled in the phylogenies, and lack studies that address their processes of diversification and speciation. Therefore, the species morphologically circumscribed in the present study are interesting hypotheses to be tested in the future from a phylogenetic, phylogeographic, and reproductive perspective.