How many taxa can be recognized within the complex Tillandsia capillaris (Bromeliaceae, Tillandsioideae)? Analysis of the available classifications using a multivariate approach

Abstract Tillandsia capillaris Ruiz & Pav., which belongs to the subgenus Diaphoranthema is distributed in Ecuador, Peru, Bolivia, northern and central Argentina, and Chile, and includes forms that are difficult to circumscribe, thus considered to form a complex. The entities of this complex are predominantly small-sized epiphytes, adapted to xeric environments. The most widely used classification defines 5 forms for this complex based on few morphological reproductive traits: Tillandsia capillaris Ruiz & Pav. f. capillaris, Tillandsia capillaris f. incana (Mez) L.B. Sm., Tillandsia capillaris f. cordobensis (Hieron.) L.B. Sm., Tillandsia capillaris f. hieronymi (Mez) L.B. Sm. and Tillandsia capillaris f. virescens (Ruiz & Pav.) L.B. Sm. In this study, 35 floral and vegetative characters were analyzed with a multivariate approach in order to assess and discuss different proposals for classification of the Tillandsia capillaris complex, which presents morphotypes that co-occur in central and northern Argentina. To accomplish this, data of quantitative and categorical morphological characters of flowers and leaves were collected from herbarium specimens and field collections and were analyzed with statistical multivariate techniques. The results suggest that the last classification for the complex seems more comprehensive and three taxa were delimited: Tillandsia capillaris (=Tillandsia capillaris f. incana-hieronymi), Tillandsia virescens s. str. (=Tillandsia capillaris f. cordobensis) and Tillandsia virescens s. l. (=Tillandsia capillaris f. virescens). While Tillandsia capillaris and Tillandsia virescens s. str. co-occur, Tillandsia virescens s. l. is restricted to altitudes above 2000 m in Argentina. Characters previously used for taxa delimitation showed continuous variation and therefore were not useful. New diagnostic characters are proposed and a key is provided for delimiting these three taxa within the complex.


Introduction
Th e subfamily Tillandsioideae comprises 10 genera (Smith and Till 1998, Espejo-Serna 2002, Barfuss et al. 2005) of which Tillandsia is the most diversifi ed. In Argentina, Tillandsia is represented by 53 species belonging to the subgenera Anoplophytum (22 spp.), Diaphoranthema (21 spp.), Phytarrhiza (7 spp.), and Allardtia (3 spp.) (Smith and Downs 1977, Luther and Sieff 1994, Zuloaga et al. 2008. Tillandsia capillaris Ruiz & Pav. belongs to the subgenus Diaphoranthema , which is characterized by small sized species adapted to arid environments, with abundant absorbing trichomes, infl orescences with few inconspicuous fl owers with stamens and styles included in the corolla. Tillandsia capillaris s. l. is distributed from southern Ecuador to central Argentina and Chile, between altitudes from 300 m to 4000 m. (Smith and Downs 1977, Till 1989, Jørgensen and León-Yánez 1999. Plants are epiphytes where the canopy is not a limiting factor (Benzing 1990) and colonize diff erent substrates, from native or exotic trees (Astegiano et al. 2007) to exposed rock, and even power lines, walls and metallic fences. Plants produce chasmogamous and cleistogamous fl owers (Gilmartin and Brown 1985) and then fruits with a large number of seeds (Till 1992).
In the dry forests of central Argentina (called Bosque Serrano , Cabido et al. 2010) T. capillaris s. l. is abundant and represents most of the biomass of epiphytes on trees (Astegiano et al. 2007). It has been argued that it causes damage to trees when the abundance is high, due to a decrease in the surface of the host shoot buds (Benzing 1990;Caldiz and Fernández 1995;Soria 2007). Current studies attribute medicinal properties to this taxon (Barboza et al. 2006) and it has been also considered as a bioindicator of air quality (Wannaz et al. 2006).

The delimitation of the T. capillaris complex and its taxonomic history
Th e T. capillaris complex constitutes a group of related taxa with a gradual morphological variation. Th e nomenclatural history itself refl ects the complex nature of T. capillaris and allies. Th e available classifi cations (Smith and Downs 1977;Till 1989) do not allow an unequivocal recognition of the entities and suggest the existence of gradients between them. Th ese authors considered these plants diffi cult to identify; Smith (1935, p. 210) mentioned "this very variable species has a number of forms whose extremes are easily diff erentiated, but which show all degrees of intergradations in any large collection" . Otherwise, Till (1989;p. 33) referring to the complex said "still remains to be clarifi ed by additional studies, if the abundance of the two species have genetic underpinnings, or hybridization processes fade the boundaries between the diff erent forms in both species; names exist in abundance" .
In the past two centuries several species that are currently included in the T. capillaris complex were described. Ruiz and Pavon (1802) described T. capillaris Ruiz & Pav. and T. virescens Ruiz & Pav. using  Th e currently accepted classifi cation in Argentina (Zuloaga et al. 2008) is based on the works of Smith (1935Smith ( , 1969Smith ( , 1970 and Smith and Downs (1977), who conducted a review of the genus, defi ning a single species T. capillaris with 5 forms: T. capillaris Ruiz & Pav. f Th e lastest revision of the subgenus Diaphoranthema in South America was done by Till (1989), who accepted two species, T. capillaris (reducing to the synonymy T. capillaris f. incana , and T. capillaris f. hieronymi ) with a distribution area ranging from southern Ecuador to central Argentina, reaching altitudes of 3500 m; and T. virescens (including T. capillaris f. virescens and T. capillaris f . cordobensis ) with a similar distribution area (the main diff erence is that it also occurs in Chile), but thriving up to 4300 m. Th is proposal is also based on a few traits: mainly on diff erences in the connation of the sepals, the indument of the fl oral bracts, and the architecture of the sepal veins. Although this classifi cation (Till 1989) can be considered more comprehensive, the classifi cation of Smith and Downs (1977) prevails in the literature.
In this contribution we analyzed the morphological variation of T. capillaris taking as the starting point the fi ve forms defi ned by Smith and Downs (1977) and cited for Argentina, using 35 fl oral and vegetative characters with a multivariate approach. Th ese infraspecifi c taxa are present in many vegetation types in northwestern and central Argentina, where the southern limit of the species is found.

Materials and methods
Herbarium specimens from CORD and LIL (Holmgren and Holmgren 2001), that were annotated by Lyman B. Smith during his visit to Argentina in 1968 and included in his monograph on Bromeliaceae (Smith and Downs 1977), were used together with new additional specimens from fi eld collections, previously identifi ed with Smith and Downs (1977) keys. Th e herbarium material inquired by Till during his visit to Argentina in 1990 were also included. Supplementary specimens from B, GOET, MA, P, W, WU were screened but not incorporated in the analyses. A total of 100 specimens were analyzed (20 of T. capillaris f. capillaris , 26 of T. capillaris f. hieronymi , 19 of T. capillaris f. incana , 12 of T. capillaris f. virescens and 23 of T. capillaris f. cordobensis ) from northern and central Argentina, therefore the results are valid for Argentina but not for the whole range of the complex (see supplementary material 1). Each specimen was treated as a taxonomical operational unit (OTU), and 35 fl oral and vegetative morphological characters were registered, including 12 continuous variables, 7 discontinuous (or discrete) variables, 11 binary variables and 5 multistate variables (Table 1). Th e morphological features selected include those traits used as key characters in species descriptions by Mez (1896), Castellanos (1945b), Smith and Downs (1977) and Till (1989). All characters were measured in the longest fertile shoot, foliar characters in the most developed leaf, and the character number of leaves per linear cm of shoot in the middle portion of the shoot. Table 1. Qualitative and quantitative characters used for the morphometric study of the complex Tillandsia capillaris .

Statistical analyses
Non-parametric Kruskal-Wallis tests (KW) were run for all the variables among the taxa considered. Box-plots were made for continuous variables. A two-steps analysis was carried out to detect the most informative characters. First, a Principal Component Analysis (PCA) was run using all characters (Woods et al. 2005, Denham et al. 2006, Blanco-Dios 2007, Nicolalde-Morejón 2005, obtaining a correlation matrix with the Pearson coeffi cient (Sokal and Rohlf 1995), and selecting afterwards characters with coeffi cient >0.20 as input for a second analysis. With the new matrix (which contained less variables), a Principal Coordinates Analysis (PCoA) was run, using the Gower coeffi cient (Gower 1971;Hernández 1997;Correa et al. 2007). Dispersion graphs were done for PCA and PCoA with INFOSTAT software (Di Rienzo et al. 2009).

Results
Results of character comparisons showed signifi cant diff erences ( Fig. 1; except for "number of branches" -KW test; H =4.68; P= 0.24) among the putative taxa but with unclear trends (Fig. 1). For example, T. capillaris f. virescens showed signifi cantly lower values compared to the other taxa for several traits (length of: fertile shoot, leaf blade, scape, fruit, seed and of embryo; Figs 1a, c, d, g, h, i respectively). T. capillaris f. cordobensis diff ered from all the other forms by the longer size of leaves, of fl oral bracts, of sepals, the lower number of leaves per linear cm of shoot and the highest number of nerves joined together at the apex of bracts (Figs 1c, e, f, j, k, respectively). T. capillaris f. hieronymi showed signifi cant diff erences with an intermediate size of the fertile shoot, the scapes, and lower number of nerves joined together at the apex of bracts (Figs 1a,d,k). T. capillaris f. capillaris and T. capillaris f. incana did not show significant diff erences and these forms are overlapped with the other forms considering this set of characters (Figs 1a-l).
In the PCA, the fi rst three components explained 50.5 % of variability (25.9, 16.5, and 8.1 % respectively) (results not shown). Analyzing the variables individually, only 19 variables were selected to explain the variance among taxa (see material and methods), considering the ones which showed values up to 0.20 (Table 2). A second PCA using these 19 characters showed that the principal two axes provide a clear ordination of the OTUs into separate groups (Figure 2). Th e two principal axes together account for 64.4% of the variability. Th e variance of the fi rst component included quantitative variables (lengths of leaf blade, bract, and sepals, number of leaves per linear cm of shoot, and fusion degree of the adaxial sepals), and qualitative variables (arrangement of the leaf, sepal dimension, type of leaf blade, fl oral bract shape, sepal shape, indument of the fl oral bract and leaf sheath exposure). Th e variance of the second component was supported by quantitative variables concerning the vegetative and the infl orescence size (lengths of fertile shoot, stem, scape, fruit, seed, embryo, and width of bract; Table 2). Figure 2 shows the grouping tendency among the OTUs for this set of 19 variables.
PCoA showed that the two principal axes provide a clear ordination of the OTUs into three separate groups (Fig. 3). Th e two principal axes together account for 54.6%    Table 2   Scatterplots of the fi rst two axis based on 19 characters selected in the PCA and using the Gower distance (sqrt(1-S)). References: Characters used (see Table I Many of the 19 most infl uential characters are useful to separate T. capillaris f . incana-hieronymi (= T. capillaris sensu Till) (Fig. 4a, b) , T. capillaris f. cordobensis (= T. virescens s. str . sensu Till) (Fig. 4c, d) and T. capillaris f. virescens (= T. virescens s. l. sensu Till) (Fig. 4e, f) . For example, characters such as: triangular and elongated fl oral bract; sepals long, acute and equaling the bract; elongated and curved leaf blades; and low number of leaves per linear cm of shoot are useful to delimit T. capillaris f. cordobensis. Th e second group formed by T. capillaris f. capillaris , T. capillaris f. incana and T. capillaris f. hieronymi can be circumscribed by: ovate-lanceolate sepals, exceeding in length the fl oral bract; round and wide fl oral bract; and straight and half-curved leaf blade. Finally, T. capillaris f. virescens (= T. virescens s. l. ) has the smaller sizes of the fertile shoot, scape, leaf blade, fruit, seed, and embryo. Th is last form showed statistically similarities in some of the characters (length of the fertile shoot, scape and sepals; Figs 1a, d, f, respectively) with T. capillaris f. hieronymi. Nevertheless, the characters indument of the bract, shape of the sepals and fusion degree of the adaxial sepals allowed to separate the forms in two diff erent groups (Fig. 2).
Key for the recognition of the morphotypes proposed for the T. capillaris complex in Argentina

Discussion
Th e criteria used in previous classifi cations Downs 1977, Till 1989) are not satisfying to resolve the complexity of this group. Nevertheless, when the number of qualitative and quantitative characters is increased, a pattern emerged through a multivariate test allowing the separation of the putative taxa into three recognizable groups. Th e application of morphometric studies in the Bromeliaceae currently prevails in the literature to resolve diff erent complex groups (Wendt et al. 2000, Costa et al. 2009, Pinzón et al. 2011) stressing the potential of this statistical tool to evaluate the limits between closely related taxa (Sokal and Rohlf 1995). Among the characters analyzed, most were informative (approximately 63% of the quantitative and 44% of the qualitative characters). Th e quantitative characters are signifi cant to separate groups, while the categorical characters were useful when the taxa had similar plant size (e.g. T. capillaris f. virescens and T. capillaris f. hieronymi ). Within the non-informative qualitative characters, those referred to the peltate trichomes (TyPH, WHPL and PiLf ) were cited in previous classifi cations (Hieronymus 1885). We suggest to avoid the use of these characters for taxonomic purposes, unless a detailed morphometric study is done. Once, environmental variation for the number of trichomes and the size of the wing area in Tillandsia has been suggested (Stefano et al. 2008).
Diagnostic characters used by Smith and Downs (1977) were useless to diff erentiate the forms T. capillaris f. capillaris , T. capillaris f. hieronymi and T. capillaris f. incana ocurring in Argentina, because these taxa showed a gradual variation in many characters (e.g., the number of nerves in the fl oral bract, the length and position of the scape, the length and diameter of the leaves) . For example, T. capillaris f. incana described by Smith and Downs (1977) with short, wide and appressed leaf blades did not show statistically signifi cant diff erences in any of these characters when it was compared with the other forms. Tillandsia capillaris f. hieronymi was previously circumscribed by the 3 nerves joined together at the apex of the fl oral bract (Smith and Downs 1977). Nevertheless, this was a variable character, varying from 3-5 nerves in the three forms of the " capillaris" complex defi ned here. Also, T. capillaris f. capillaris did not show statistical diff erences in the scape position (Smith and Downs 1977). Th is variable seems to be related with the development of the stem (Castellanos 1945b), and it was registered for all the forms terminal and axillary infl orescences.
On the contrary, other diagnostic characters established by Smith and Downs (1977) to determine T. capillaris f. cordobensis and T. capillaris f. virescens showed statistically signifi cant diff erences. Tillandsia capillaris f. cordobensis can be circumscribed by both the indument of the fl oral bracts (Fig. 4c, d) and wide spreading leaves (the leaf sheaths are prominent and visible so that the leaf blades are detached). In addition, some complementary characters allow an easier delimitation of this taxon, as the exposure of the leaf sheaths or the leaf density per cm of stem. Tillandsia capillaris f. virescens can be recognized by the lacking of a scape (Smith and Downs 1977) (Fig. 4e, f). It is interesting to note that the reduction of the reproductive organs in this taxon could be a feature related to environment constraints (altitude) where this form lives (Gilmartin and Brown 1985). Till (1989) used the fusion degree of the adaxial and abaxial sepals as the main character to delimit the complex into two groups: one with adaxial sepals partially fused (20-60%) (Fig 4b) and abaxial sepals fused by their bases (10-35%) ( T. capillaris = T. capillaris f. incana-hieronymi ); a second group with adaxial sepals more fused (50-90%) (Fig.  4d, f), and abaxial sepals less fused (5-20%) ( T. virescens s. str . and s. l. =T. capillaris f. cordobensis and T. capillaris f. virescens ). Data for the material studied here showed values of 40-70% fusion for the adaxial sepals in the fi rst group ( T. capillaris = T. capillaris f. incana-hieronymi ), and 60-90% in the second group ( T. virescens s. str . and s. l.=T. capillaris f. cordobensis and T. capillaris f. virescens ). Although the trend described by Till (1989) for the abaxial sepals was detected, statistical diff erences supported the separation of the complex into three groups. Till (1984) considered T. virescens as one species, and T. cordobensis as a taxonomic synonym of T. virescens (Till 1984, p. 135-136), and defi ned fi ve aggregates for " T. virescens" . After revising the herbarium material that he studied, we interpreted that T. virescens s. str . (= T. capillaris f. cordobensis ) comprise "group 1: T. cordobensis " and "group 2: T. cordobensis "var." tucumanensis nom. nud.". Otherwise, T. virescens s. l. (= T. capillaris f. virescens ) include "group 4: T. propinqua "var."" (Till 1984). Till (1989) also used the architecture of the veins of the sepals as a character, but we did not consider it here because to exam such feature we would cause severe damage to her-barium specimens. Other characters briefl y mentioned by Till (1989; shape of the bract and size of the sepals) were measured here and were signifi cant to separate the taxa.
Summarizing, our results partially support the classifi cation of Till (1989) that considered the fi rst group as T. capillaris since the OTUs for T. capillaris f. capillaris , T. capillaris f. hieronymi and T. capillaris f. incana tend to form a single ensemble. On the other hand, concerning the diff erences between T. capillaris f. cordobensis and T. capillaris f. virescens , the classifi cation by Smith and Downs (1977) is still appropiatted. However, we are also evaluating the taxonomical thesis of Till (1984) and considered his classifi cation (" T. virescens and aggregates") in future taxonomical work for the complex, since we saw that the characters he used (the cohesion of the sepals and the indument of the fl oral bract) were useful to defi ned these groups. We expect to propose conclusive nomenclatural changes, however, only after gathering additional data. Taxonomic resolution of complex groups, ideally, should be done combining morphological data from the whole range of distribution.

The T. capillaris complex in Central Argentina
All the forms analyzed are distributed in the central and northern Argentina, in the southern distributional range of the complex. Th e taxa T. capillaris (=f. incana-hieronymi ) and T. virescens s. str . (=f. cordobensis ), co-occur in almost the same sites and altitude levels. Tillandsia virescens s. str . was mentioned by Smith and Downs (1977) for altitudes above 900 m. We found populations at lower altitudes (300 m), co-occurring with T. capillaris . On the other hand, T. virescens s. l. (=f. virescens ), with saxicolous habit, was found restricted to higher altitudes in the central Argentina (above 2000 m; Achala batholith), or in the western foothill of the Andes (between 2000-3500 m). It is interesting to note that in some regions and at higher altitudes (2000 m, for example in the central Argentina), the three forms can co-occur, but at lower altitudes (as in the woodland mountains of Bosque Serrano at, 400-1100 m) only two of these forms can be found ( T. capillaris and T. virescens s. str .).

Conclusion
Th e main goal of this contribution was to analyze the available classifi cations of the T. capillaris complex using a relatively large sample of material with a multivariate perspective. Th is methodological approach allowed us to defi ne three taxa in Argentina (Figs 3 and 4) with clear morphological limits, and to inquire into the confl icts between the available classifi cations. Th e next step is to compare these results using new material from other populations within the geographical range of the complex, specifi cally from Bolivia and Peru. We are not proposing new nomenclature combinations until the whole distribution area are investigated.