Recircumscription and taxonomic revision of Siderasis, with comments on the systematics of subtribe Dichorisandrinae (Commelinaceae)

Abstract A new circumscription and a total of six microendemic species, four of them new to science, are herein presented for Siderasis, based on field and herbaria studies, and cultivated material. We provide an identification key to the species and a distribution map, description, comments, conservation assessment, and illustration for each species. Also, we present an emended key to the genera of subtribe Dichorisandrinae, and comments on the morphology and systematics of the subtribe.


Introduction
Siderasis Raf. is currently applied to a small genus of neotropical Commelinaceae, comprising only two microendemic species, restricted to the Atlantic Forest of Southeastern Brazil (Pellegrini 2017). It was originally described by Rafinesque (1836), together with several other small genera, in order to better organize the many species misplaced in Commelina L. and Tradescantia L. Rafinesque (1836) mentioned a possible affinity between Siderasis, Callisia Loefl. and Etheosanthes Raf. (= Belosynapsis Hassk.), and considered Siderasis not at all similar to Tradescantia; but gave no explanation for any of these statements. He also considered T. fuscata Lodd. a synonym of his newly described S. acaulis Raf., which was characterized by its rusty hirsute indumentum covering the entire plant, short stems, flowers emerging from the roots, petals basally connate, dimorphic stamens varying from four to six, and gynoecium 2-3-locular [sic]. After Rafinesque's publication, Siderasis was completely overlooked by all Commelinaceae specialists for the next 120 years (Moore 1956). In the meantime, Hasskarl (1869) described the new genus Pyrrheima Hassk. following his discussions with Schlechtendal during the Botany Meeting in Amsterdam in April of 1865. Hasskarl and Schlechtendal believed that Pyrrheima diverged greatly from Tradescantia and Tinantia Scheidw. due to its non-tubular perianth, six equal and fertile stamens, and lunate anther sacs, and thus merited distinct generic status. Clarke (1881), in his monograph for Commelinaceae, accepted Pyrrheima, including only P. loddigesii Hassk., and reducing P. minus Hassk. to a variety of it. Brückner (1930), noticing that P. loddigesii was an unnecessary replacement name for T. fuscata, made the new combination P. fuscata (Lodd.) G.Brückn., but was unsure if Siderasis and Pyrrheima were indeed congeneric. This was later confirmed by Moore (1956), when he merged the two by transferring P. fuscata to Siderasis.
After further period of neglect, the genus was placed in subtribe Dichorisandrinae (Faden and Hunt 1991), along with its sister-genus Dichorisandra J.C.Mikan, Cochliostema Lem., Geogenanthus Ule, and Plowmanianthus Faden & C.R.Hardy (Faden and Hunt 1991;Hardy 2001;Evans et al. 2003;Hardy and Faden 2004). However, resolution of the relationships within the group remains elusive. While it appears certain that Siderasis is proximally related to Dichorisandra, the subtribe as a whole has been recovered as paraphyletic in most molecular and morphological phylogenies to date. Two separate clades are recovered with one containing Dichorisandra and Siderasis (i.e. subtribe Dichorisandrinae s.s), while the remaining three genera (Geogenanthus and Cochliostema+Plowmanianthus) are recovered as one of the early-diverging clades in tribe Tradescantieae (Hardy 2001;Evans et al. 2003;Wade et al. 2006;Zuiderveen et al. 2011;Hertweck and Pires 2014;Pellegrini 2017;Pellegrini et al., in prep.). Siderasis has hitherto been characterized by the rusty to bright red hirsute indumentum covering the entire plant (except the petals and androecium), its ebracteolate cincinni, filaments three to four times longer than the anthers, anthers with rimose dehiscence (Hardy and Faden 2004), and exarillate seeds (Faden 1998).
Composition of the genus itself also remains unclear with Faden and Hunt (1991) mentioning two Siderasis species, while Faden (1998) mentions two to three species. Barreto (1997), in a survey of the Commelinaceae native to Brazil accepts only S. fuscata and reaffirmed Siderasis as a monospecific genus. Pellegrini (2017) recently described a new species of Siderasis, and provided important information regarding inflorescence and seed morphology in the genus. Clearly, further studies were still necessary to solve the ongoing issues , and with this in mind recent field and herbaria studies have been undertaken to shed further light on this genus. In an attempt to clarify the taxonomy and systematics of neotropical Commelinaceae, and as part of the authors' ongoing studies in subtribe Dichorisandrinae (Hardy and Faden 2004;Aona et al. 2012;Aona et al. 2016;Pellegrini 2017), we recircumscribe and revise Siderasis, with the description of four new species. We also provide detailed comments on the morphology and systematics of subtribe Dichorisandrinae s.l.

Results
In the present study, we accept six species of Siderasis, with four of them newly described here. All species in the genus are microendemics, restricted to the Atlantic Forest of eastern Brazil. Both Dichorisandra and Siderasis share considerable variation in growth form, inflorescence architecture and androecium arrangement, which may have hindered the emergence of a stable taxonomy. Due to the variation and peculiar morphology of the newly described species, especially the two climbing species, Siderasis is recircumscribed below. The genus can be distinguished from the remaining Dichorisandrinae s.l. based on floral morphology, especially androecium and microstigmatic morphology. An updated identification key for the genera of Dichorisandrinae s.l. is presented, along with comments on the morphology of Siderasis compared to the remaining genera of the subtribe. Description. Herbs or vines, perennial, with a definite base, terrestrial or rupicolous. Roots thin, fibrous, sometimes forming terminal, small, fusiform to oblongoid tubers. Rhizomes present or not, if present short, shallowly to deeply buried in the ground, rarely only covered by leaf litter. Subterraneous stems present or not, when present buried deep in the soil, unbranched, produced directly from the short rhizome; internodes moderately elongate to elongate. Aerial stems with determinate or indeterminate growth, elongated or short to inconspicuous, densely branched or unbranched, when densely branched primary shoot determinate or not, when present secondary shoots determinate; internodes inconspicuous to weakly elongate, or elongate; flagelliform-shoots (ramets) present or not, if present produced after the fertile period, forming a new rosette at the apex, axillary, unbranched, internodes elongate. Leaves spirally-alternate or distichously-alternate, congested at the apex of the stems forming a rosette or evenly distributed along the secondary branches, sessile to subpetiolate or petiolate, sheathing at the base, ptyxis involute; blades membranous to chartaceous or succulent, base symmetric or slightly to completely asymmetric, margins slightly revolute to flat, apex curved or straight. Synflorescence composed of a solitary main florescence or with 1-7 coflorescences. Main florescence (inflorescence) a thyrse, terminal or apparently so, rarely axillary, a many-branched, pedunculate thyrse, with alternate cincinni or reduced to a solitary pedunculate cincinnus; basal bract sessile or amplexicaulous or sheathing; cincinni bracts sessile or amplexicaulous; cincinni pedunculate, 1-many-flowered; bracteoles present or not. Flowers bisexual or staminate, actinomorphic or zygomorphic, chasmogamous, flat, pedicellate or sessile; pedicels erect during pre-anthesis and anthesis, erect or deflexed post-anthesis, generally elongating in fruit; sepals 3, unequal, free, membranous or fleshy, persistent and accrescent in fruit, the uppermost external, broader than the others, sometimes also shorter than the others; petals 3, deliquescent, free, margins entire to irregularly lacerated, glabrous, rarely ciliated with non-moniliform hairs, apex entire to irregularly lacerated, subequal, the lowermost either broader or longer than the others; stamens 6, equal or unequal, straight or curved upwards, filaments free, glabrous, straight or sigmoid, anthers dorsifixed, extrorsely rimose, anther sacs semicircular, divergent, pollen white, connectives expanded, quadrangular to rectangular; ovary sessile, globose to broadly oblongoid to ellipsoid in outline, trigonous with obtuse to round angles in cross-section, densely hirsute or lanate or velutine, 3-locular, locules equal, 3-6-ovulate, ovules hemianatropous, biseriate to partially uniseriate; style terminal, straight or curved upwards; stigma annular-truncate or annular-capitate, marginally papillate leaving the stylar canal evident, papillae unicellular. Capsules loculicidal, thick-walled, 3-valved, globose or subglobose to broadly ellipsoid to broadly oblongoid to oblongoid in outline, trigonous with obtuse to round angles in cross-section, smooth to sparsely reticulate, apiculate due to persistent style base. Seeds (1-)3-6 per locule, arillate, obconic to ellipsoid, dorsiventrally compressed, ventrally slightly flattened or with a mild ridge, testa foveolate or rugose; hilum C-shaped, in a shallow depression; embryotega semidorsal or semilateral, relatively inconspicuous, without a prominent apicule; aril cream-colored to hyaline, slightly to completely translucent, thick or inconspicuous.
Etymology. Siderasis was named in allusion to the peculiar red to bright-red hairs that cover almost the entire plant, but especially the leaves. However, only S. fuscata possesses the aforementioned hairs, and all of the remaining species possess leaf blades covered by hyaline to light brown, rarely rusty hairs.
Habitat, distribution and ecology. Siderasis is endemic to the Atlantic Forest domain in coastal Brazil, occurring in the states of Bahia, Espírito Santo, and Rio de Janeiro (Fig. 2). More specifically, Siderasis is restricted to the Central Corridor of the Atlantic Forest, growing in remnants of semideciduous forests associated with inselbergs, between 90-1350 m above sea level. The genus is composed exclusively by microendemic species distributed in very small and fragmented subpopulations, susceptible to deforestation and illegal collection of specimens for ornamental purposes.
Biogeography. Since most phylogenies for Commelinaceae corroborate the paraphyly of Dichorisandrinae (Evans et al. 2000(Evans et al. , 2003Hardy 2001;Wade et al. 2006;Zuiderveen et al. 2011;Hertweck and Pires 2014;Pellegrini et al., in prep.), we can hypothesize on the independent diversification of these lineages from a biogeographical point of view. The clade composed by Cochliostema, Geogenanthus, and Plowmanianthus is consistently recovered as the second lineage to diverge in tribe Tradescantieae, following the diversion of subtribe Streptoliriinae (Evans et al. 2003;Wade et al. 2006;Zuiderveen et al. 2011;Hertweck and Pires 2014;Pellegrini et al., in prep.). The ancestor of this lineage probably originated in the Amazon Basin, and posteriorly diversified in the Guyana Shield, northern Andes, and Central America reaching Costa Rica (Hardy 2001;Hardy and Faden 2004). On the other hand, the clade composed by Dichorisandra and Siderasis is recovered as the third lineage to diverge in Tradescantieae (Evans et al. 2003;Wade et al. 2006;Zuiderveen et al. 2011;Hertweck and Pires 2014;Pellegrini et al., in prep.). The ancestor of this clade probably originated and diversified in the Atlantic Forest domain, since it is the center of diversity of both genera. Subsequently, the ancestors of various Dichorisandra lineages might have dispersed, more than once, diversifying in the Amazon Basin through gallery forests in the Cerrado domain.
Growth form and leaf morphology. Siderasis possesses two clearly differentiated growth patterns: (1) rosette herbs, generally with very short internodes, and spirallyalternate, symmetrical leaves ( Fig. 3A-B); (2) climbing vines, with elongated internodes, and distichously-alternate, asymmetrical leaves ( Fig. 3C-D). The rosette habit has hitherto been the only one recognized in the genus. Faden (1998) mentioned the existence of a climbing species in the genus, but due to the synoptic nature of that publication, no further remarks were made on the subject. The climbing habit is relatively uncommon in the family, but found in the closely related Dichorisandra. However, in Dichorisandra the plants tend to lean on nearby trees and shrubs, later producing pendant branches, or even growing completely intertwined with more robust shrubs ( Fig. 3E-F). In Siderasis, the primary branch grows at the base of a tree (Fig. 3C), posteriorly spirally ascending around the trunk, and finally producing the flowering secondary branches (Fig. 3D). In the remaining genera of Dichorisandrinae, growth form is stable, with almost no variation within each genus. In Cochliostema, the plants tend to be tank-forming rosette herbs, but creeping individuals are also known in C. velutinum Read (Hardy 2001). In Geogenanthus, the plants always possess a dracaenoid habit, with leaves congested at the apex (Hardy 2001). In Plowmanianthus, the plants are always rosette herbs with very short stems (Hardy and Faden 2004).
Considerable variation in leaf morphology occurs in Siderasis, with leaves ranging from: (1) sessile to subpetiolate (Fig. 3B-D); (2) truly petiolate, as in S. fuscata (Fig.  3A, 8C). Truly petiolate leaves are extremely rare in Commelinaceae, being recorded only in a handful of species restricted to the peculiar-looking subtribe Streptoliriinae, mostly comprised of vining plants (Pellegrini and Faden, pers. observ.). Phyllotaxy in Siderasis can range from distichous to spirally-alternate, the arrangement being correlated to symmetry of the leaf blades.
Inflorescence morphology. In all Dichorisandra and two species of Siderasis (i.e. S. spectabilis and S. zorzanellii), the main florescence is a many-branched, pedunculate, terminal or axillary thyrse with alternate cincinni, each cincinnus being multiflowered. In the remaining species of Siderasis (i.e. S. albofasciata, S. almeidae, S. fuscata and S. medusoides), the main florescence is composed of a thyrse reduced to a solitary cincinnus, as described in Pellegrini (2017;Fig. 4A). These reduced thyrsi are arranged into a synflorescence that may contain up to seven coflorescences. The center of the mature Siderasis rosette may contain several terminal or apparently terminal synflorescences. In Dichorisandra and the two climbing species of Siderasis, the main axis of the inflorescence is usually well developed, thus producing a typical looking thyrse (Fig. 4B). Nevertheless, the inflorescences may also be extremely reduced in some species (i.e. D. acaulis group), due to the shortening of the inflorescence's internodes . The cincinni are also very short (i.e. sessile to subsessile), being enclosed by the leaf sheaths and not obvious at first glance . The flowers are peculiarly long-pedicellate, giving the impression that all flowers emerge directly from the apex of the stems (Pellegrini and Almeida 2016; Fig. 4C). Despite the extreme reduction and superficial similarity, this inflorescence pattern differs from the one found in the rosette species of Siderasis, since it still is a many-branched thyrse. In Plowmanianthus the main florescence is also reduced to a solitary cincinnus. Nonetheless, coflorescences only develop after the main florescence has failed to develop or set fruit, and the cincinni from the primary and secondary thyrsi are morphologically distinct (Hardy and Faden 2004). In Geogenanthus the inflorescences are always born at the base of the plant, near the ground. Aside from that, the main florescence is a pedunculate, fascicle-like thyrse, with (1-)2-4-several alternate cincinni (Hardy 2001). Finally, in Cochliostema the main florescence is a many-branched, pedunculate, axillary thyrse, with alternate to verticillate cincinni, each cincinnus being multi-flowered and subtended by showy and cucullate spathaceous bracts (Hardy and Stevenson 2000;Hardy 2001).
Floral symmetry. Two distinct floral patterns can be observed in different species of Siderasis: (1) flowers are always bisexual, actinomorphic, having 6 equal stamens arranged cyclically around the ovary, with straight filaments (Fig. 1A-B); (2) flowers bisexual or staminate, zygomorphic, having 6 unequal stamens curved upwards, with sigmoid filaments (Fig. 1C). Furthermore, in the zygomorphic staminate flowers, the lower antepetalous stamen is longer, and is arranged and curved in the same way as the Pell. & Faden inflorescence type, consisting of a thyrse reduced to a solitary cincinnus B diagram of the basic Dichorisandra inflorescence type (also characteristic of S. spectabilis and S. zorzanellii), consisting of a many-branched thyrse with alternate, many-flowered cincinni C diagram of the basic D. acaulis species group inflorescence type, where the main florescence axis and cincinni axis are greatly reduced, and the pedicels are peculiarly elongated. P = prophyll; pB = peduncle bract on main synflorescence axis; * = aborted or dormant apex of main inflorescence axis (usually not observed); B = cincinnus bract; b = bracteole; f = flower; 1°bud = bud terminating cincinnus; 2°bud = bud in axil of peduncle bract with potential to develop into a secondary thyrse (coflorescence); Modified from Pellegrini (2017). style in bisexual flowers. The first flower morph is very similar to that found in the D. acaulis group (Pellegrini and Almeida 2016; Fig. 1D), while the second is equivalent to that of the D. hexandra and D. incurva groups (Fig. 1E, I). In Dichorisandra, flower symmetry is generally influenced by the positioning of the stamens, rather than by the relationship of stamens and staminodes. Actinomorphic flowers can be found not only in the D. acaulis group (Fig. 1D), but also in a group of still-undescribed species from the Guyana Shield (Faden and Pellegrini, pers. observ.). In all remaining species groups in Dichorisandra, the flowers are clearly zygomorphic, either due to the number of stamens, their size and/or position. In the D. thyrsiflora group, the androecium is generally composed of six fertile stamens, four of them curved towards the center of the flower, and the two lower lateral ones curved towards their opposing sides (Fig. 1F). An exception can be noticed in D. paranaënsis D.Maia et al. (Fig. 1G) and D. nana Aona & M.C.E.Amaral (Fig. 1H). In D. paranaënsis the stamens are curved upwards, varying from five fertile stamens with a staminode (present or not) to six fertile stamens, and introrsely rimose anthers. On the other hand, in D. nana the six fertile stamens are curved upwards, and possess poricidal anthers. In the D. incurva (Fig. 1I), D. penduliflora (Fig.  1J), D. leucophthalmos (Fig. 1K), and D. radicalis groups (Fig. 1L), the androecium is composed of five stamens (generally with an upper staminode; notice the filiform staminode in Fig. 1L), rarely six fertile stamens, curved upwards, and with introrsely rimose anthers. In the D. incurva and D. leucophthalmos groups, the anthers are always yellow (Figs. 1I, K), while in the D. penduliflora and D. radicalis groups, the anthers are white, generally with the anther sacs partially to totally colored in blue, pink or purple (Figs. 1J, L). The remaining genera of Dichorisandrinae possess strongly zygomorphic flowers, especially due to the position and/or number of stamens: (1) 5-6 dimorphic, free and fertile stamens in Geogenanthus (Hardy 2001 ; Fig. 1N); (2) 3 stamens in the upper side of the flower, fused in a hood-like structure, and 3 lower staminodes (the middle one microscopic) in Cochliostema (Hardy 2001;Fig. 1M); (3) and 3 free to partially fused stamens in the upper side of the flower, and 3 lower staminodes (generally all of them microscopic) in Plowmanianthus (Hardy and Faden 2004;Fig. 1O).
Androecium and gynoecium morphology. The anthers in Siderasis are dorsifixed, with extrorsely rimose dehiscence, two times wider than long, three to four times shorter than the filaments, with semicircular, divergent anthers sacs, and expanded connectives ( Fig. 1A-C). In Dichorisandra the anthers are basifixed, with poricidal or introrsely rimose (but functionally poricidal) dehiscence, three to four times longer than wide, and three to four times longer than the filaments, rarely equal to the filaments, with elongate, parallel anther sacs, and inconspicuous connectives (Aona 2008;Figs. 1D-L). In Cochliostema, Geogenanthus and Plowmanianthus the anthers vary from dorsifixed to basifixed, with extrorsely rimose dehiscence, as wide as long to two times shorter than the filaments, with semicircular to spirally-coiled, appressed anther sacs, and inconspicuous connectives (Hardy and Stevenson 2000;Hardy 2001;Hardy and Faden 2004;Figs. 1M-O).
The gynoecium is fairly homogeneous in Dichorisandrinae s.l., with all genera having sessile, 3-locular ovaries, with all locules fertile, ovules hemianatropous, biseriate to partially uniseriate, style terminal, straight or bent at the apex, stigma annular-truncate to annular-capitate, peripherally ciliate with moniliform hairs (i.e. Cochliostema and Plowmanianthus) or not (i.e. Dichorisandra, Geogenanthus and Siderasis). In Siderasis, the stigmatic papillae are unicellular, and restricted to the margins of the stigma, leaving the stylar canal evident (Owens and Kimmins 1981). On the other hand, in Dichorisandra, the stigmatic papillae are multicellular, and evenly distributed on the stigma, completely concealing the stylar canal (Owens and Kimmins 1981).
Fruit and seed morphology. The capsules of Dichorisandra and Siderasis can be differentiated from capsules of other Commelinaceae by their thick and tough walls. In Commelinaceae the fruits are commonly (1-)2-3-locular, thin walled, septicidal capsules (Faden 1998). Dichorisandra and Siderasis possess 3-locular, 3-valvar capsules, and arillate seeds. The aril in Dichorisandra is generally opaque (rarely hyaline), usually thick (rarely inconspicuous), and colored from white to grayish or bright orange (rarely colorless) ( Fig. 5A-B). Whereas the aril in Siderasis can be hyaline to slightly hyaline, inconspicuous or thick, and cream-colored to colorless ( Fig. 5C-D).
The seeds in both genera are very similar in gross morphology, varying in shape from obconic to ellipsoid to quadrangular; in ornamentation from foveolate to scrobiculate to rugose, with a semilateral to semidorsal embryotega, and with a C-shaped hilum. In Cochliostema, Geogenanthus and Plowmanianthus the capsules are thinwalled, 3-locular, 3-valvar, and with exarillate seeds. In Cochliostema the capsules are narrowly cylindrical, and the seeds vary from subcylindrical to narrowly oblongoid, with a smooth testa that becomes sticky when hydrated, semidorsal embryotega, and a linear hilum with curved edges. In Geogenanthus and Plowmanianthus the capsules are fusiform to ellipsoid, the seeds range from reniform to ellipsoid, with rugose to foveolate, farinose testa, lateral embryotega, and a C-shaped hilum (Hardy 2001;Hardy and Faden 2004).
In Dichorisandra and Siderasis capsule and seed morphology differences may have great taxonomic potential. In Dichorisandra, many of the aforementioned species groups display characteristic capsule and seed morphology, as exemplified in the D. acaulis group by . In the D. thyrsiflora group, capsule morphology can easily differentiate most known species, based on shape, coloration, texture and pubescence (Pellegrini, pers. observ.). In Siderasis, capsule morphology shows a similar potential, with the fruits of S. zorzanellii being completely deviant in shape, texture and pubescence from the remaining species. Unfortunately, since the fruits of S. spectabilis are still unknown, it is impossible to know if this change in capsule morphology is correlated to the change in habit from rosette to vining herbs. Siderasis fuscata possesses unique seed morphology, being the only known species with an inconspicuous and hyaline aril, testa light gray to gray, and foveolate. Field expeditions focused on collecting fruit and seed samples of all species of Siderasis could be of great taxonomic value. It is possible that most, if not all, presently accepted species could be differentiated based exclusively on fruit and seed morphology.
Reproductive biology. Little is known regarding the floral biology of subtribe Dichorisandrinae, although this subtribe possesses the greatest range in inflorescence architecture and floral patterns in the family. In Siderasis the anthers are always extrorsely rimose, but apart from the floral specialization (i.e. zygomorphic, bisexual or staminate flowers, and unequal and sigmoid stamens) in the two climbing species and the petals with margins ciliated with non-moniliform hairs, a character unique in the family, in S. spectabilis, the flowers are relatively unspecialized. Dichorisandra possesses a wide variation in flower morphology and androecium arrangement. Its flowers can range from actinomorphic to zygomorphic, the stamens can vary from (5-)6, sometimes with the upper stamen reduced to a staminode in some species. The filaments can be either straight, slightly sigmoid or slightly twisted depending on their position in the flower, while the anthers can be introrsely rimose and functionally poricidal or truly poricidal (Aona 2008;Fig. 1D-L). On the other hand, in Cochliostema, Geogenanthus and Plowmanianthus, the flowers are highly specialized, being zygomorphic (in all genera), scented (in Cochliostema), with a high frequency of cleistogamous flowers (in Plowmanianthus), petals and stigma fringed with moniliform hairs (fringed petals in all genera, stigma fringed exclusively in Cochliostema and Plowmanianthus), filaments bearded with moniliform hairs (in all genera), functionally poricidal androecium (in Cochliostema, due to the hood-like structure enclosing the anthers), and curved to spirally-coiled anthers (in all genera) (Hardy and Stevenson 2000;Hardy 2001;Hardy and Faden 2004;Fig. 1M-O). Only three species of Dichorisandra have had their reproductive biology investigated, presenting typical buzz-pollination, performed by bumblebees (Apidae) and/or sweatbees (Halictidae) (D. thyrsiflora, Boaventura and Matthes 1987;D. hexandra and D. incurva, Sigrist and Sazima 2015). Information regarding flower visitation in Cochliostema, Geogenanthus, and Plowmanianthus is completely lacking from the available literature. During our field studies and while observing the Siderasis specimens grown at the greenhouse of Jardim Botânico do Rio de Janeiro, the first author has observed flowers of S. albofasciata, S. almeidae, and S. fuscata being visited by stingless honey bees (Apidae, tribe Meliponini). Siderasis medusoides was not seen in the field, but high-resolution photographs sent by one of the collectors clearly show several small ants walking around the flowers and cincinni (Fig. 9C). The bees might either represent pollen robbers or potential pollinators, but the presence of the ants is hard to explain, since nectaries are unknown for Commelinaceae (Faden 1992(Faden , 1998. Further studies on the reproductive biology of Siderasis are clearly needed. Aside from the peculiar floral diversity, Dichorisandrinae s.l. has two genera (out of five) and the majority of species in the family with arillate seeds (Pellegrini 2017). Nonetheless, no study has ever focused on vector-mediated (i.e. zoo-choric) seed dispersal in the family. In Dichorisandra, the seeds in the D. hexandra group are most certainly dispersed by birds (Faden 1992), due to the plants vining habit (Fig. 3E), which help in displaying the seeds, covered by an orange to bright orange, thick and opaque aril (Fig.  5B). The seeds in the D. thyrsiflora group are covered by a thick and opaque, white to cream-colored aril (Fig. 5A), and are generally easy to see in the field, due to the plants high stature (Pellegrini pers. observ.). Nonetheless, these species lack the characteristic colors that are generally associated with bird pollination/dispersal (i.e. pink, red, orange and yellow; Fleming and Estrada 1993), always present in the D. hexandra group. The species in the D. acaulis group possess seeds also covered by a thick and opaque, white aril, lacking the visual attraction associated with bird dispersal, and also lack an elevated display, since they are always shorter than 1 m long . These seeds might be dispersed by ants, or by small terrestrial vertebrates (e.g. small rodents), instead of being dispersed by birds, as hypothesized for other species of Dichorisandra. The seeds from the rosette species of Siderasis have similar morphological and ecological features to the species from the D. acaulis group. These species also have small stature and seeds with hyaline and inconspicuous, or cream-colored, slightly translucent, thick arils ( Fig. 5C-D), being most probably dispersed by animals similar to the ones dispersing the seeds of the species in the D. acaulis group.
From a phylogenetic point of view, it seems that vector-mediated seed dispersal has evolved several times in the family: (1) arillate seeds are recorded for Dichorisandra and Siderasis in Dichorisandrinae, Amischotolype Hassk., Coleotrype C.B.Clarke and Porandra Hong in Coleotrypinae (Pellegrini 2017), and Spatholirion Ridl. in Streptoliriinae (Thitimetharoch 2004); (2) appendaged seeds are recorded for at least two separate lineages in tribe Commelineae (i.e. some species of Commelina L. and Murdannia Royle; ; (3) truly fleshy fruits are known only from Palisota Rchb. ex Endl. (Faden 1998);(4) in Tradescantia zanonia (L.) Sw. the fleshy sepals cover the indehiscent fruit at post-anthesis, producing a sweet and atro-vinaceous berry-like fruit, dispersed by birds (Pellegrini, obs. pers.); (5) in Pollia Thunb. the fruits are dry, crustaceous and indehiscent, and due to their vibrant colors (metallic blue to shiny black) mimic real berries (Faden 1978); (6) in some Commelina (i.e. the species originally placed under Phaeosphaerion Hassk. and Commelinopsis Pichon) the fruits are morphologically similar to those of Pollia, being also crustaceous, but either dehiscent or indehiscent (Faden and Hunt 1987); (7) and sticky capsules covered by a mixture of hook and minute glandular hairs, in Rhopalephora Hassk. (Pellegrini et al., in prep.). Nonetheless, further investigations are needed to better understand the ecology and evolution of vector-mediated seed dispersal in Commelinaceae.
Distribution and habitat. Siderasis albofasciata is known exclusively from the municipalities of Santa Teresa and Fundão, state of Espírito Santo (Fig. 2). It occurs in the understory of evergreen forests, in shady areas with shallow and rocky soil, with great leaf-litter accumulation.
Phenology. It blooms from November to February. This species was collected in fruit in April, when mature and immature capsules were seen.
Affinities. Siderasis albofasciata is similar to S. fuscata due to its leaves being of a different color along the midvein of the adaxial side, abaxially vinaceous, and inflorescences covered with rusty hairs. However, S. albofasciata can be readily differentiated by its sessile to subpetiolate leaves covered by hyaline to light brown indumentum (vs. petiolate leaves with bright red to red indumentum, in S. fuscata), a well-defined white stripe along the midvein on the adaxial side of the blade (vs. sometimes blotched silver to metallic light green), main axis of the synflorescence elongate (vs. inconspicuous), bracteoles present (vs. bracteoles absent), cincinni (3-)5-8-flowered [vs. 1-3(-4)-flowered], anthers purple, filaments and style apically purple (vs. androecium and gynoecium completely white), testa brown and rugose (vs. grey to light grey and foveolate), and aril cream-colored, thick and slightly hyaline (vs. aril colorless and inconspicuous). It is also similar to S. almeidae and S. medusoides due to the leaf blades adaxially hispid, abaxially lanate, and presence of bracteoles in the cincinni. Siderasis albofasciata can be easily differentiated from all the accepted species in the genus by the peculiar coloration pattern in its androecium and gynoecium.

Specimens seen (paratypes). BRAZIL. Bahia
Distribution and habitat. Siderasis almeidae is confined to the municipalities of Itamarajú and Prado, Bahia (Fig. 2). It occurs in the "mata higrófila" vegetation with emerging rocky formations, in shady and moist areas. In the type locality, the subpopulations were found growing in great accumulations of leaf litter, among dense clusters of Marantaceae. The area is greatly disturbed, and within private property.
Phenology. It was found in bloom from September to April, beginning to fruit in September, but mature fruits are unknown.
Conservation status. Siderasis almeidae has considerably narrow EOO (ca. 180.390 km 2 ) and AOO (ca. 2800 km 2 ). Most of the known collections were made in the type locality, in a small forest patch inside a private cattle farm. None of the known subpopulations is protected by a conservation unit, and the southern region of Bahia has few undisturbed areas of Atlantic Forest, being subjected to ongoing deforestation, cattle breeding, and several crops. The subpopulations of S. almeidae are small to medium-sized (with ca. 20 individuals), but mainly composed of clonal individuals. Thus, following the IUCN (2001) criteria, we suggest S. almeidae to be considered Critically Endangered [CR, A2abcd+B2ab(i, ii, iii, iv, v)+C1].

Distribution and habitat.
Siderasis fuscata is endemic to the municipalities of Rio de Janeiro (with several localities inside Floresta da Tijuca) and Niterói (with just one locality, Alto Mourão), in the Rio de Janeiro state (Fig. 2). It occurs in the vegetation on hillsides (mata de encosta) near the littoral, in shady areas with shallow and rocky soil.
Phenology. It blooms from August to May and fruits from January to May, although fructification seems to be an uncommon event since few fruiting specimens were seen or collected.
Common name. "violeta-silvestre", "orelha-de-urso", "pelo-de-urso", "trapoerabapeluda", "brown spiderwort", "bear ears". The subpopulation from Niterói is disjunct from the others in Rio de Janeiro, due to the urban area of both cities. It possesses a considerably small EOO (ca. 7000 km²), with the population being severely fragmented. Despite all the extant subpopulations being inside conservation units (i.e. Parque Nacional da Tijuca and Parque Estadual Serra da Tiririca), they are considerably small, composed mainly of clonal individuals, with no more than 30 mature individuals. Only a small number of fertile individuals can be found during the flowering season in each population, and very few fruits are produced. All these areas are extremely susceptible to real-estate development, deforestation, and have many invasive species, with areas like Parque Estadual Serra da Tiririca being especially affected by human-related forest fires. The subpopulations from Pedra da Gávea and Corcovado are probably extinct, or nearly so, since no recent collection in either areas is known by the authors. A total of 250 mature individuals is estimated for the overall population, based on our field observations. Added to the above factors, S. fuscata is appreciated as an ornamental plant all over the world due to its exotic foliage and beautiful flowers, so the few known extant subpopulations are also a target of illegal collection for exotic plant growers from all over the world. Thus, following the IUCN criteria (2001), we suggest S. fuscata be considered Critically Endangered [CR,A2abcde+B1ab(i,ii,iii,iv,v) Affinities. Siderasis fuscata is similar to S. albofasciata in their variegated leaf blades, and similar to S. almeidae and S. medusoides in their white anthers. Nevertheless, it can be readily distinguished from all species of Siderasis by its petiolate leaves, red to bright red indumentum covering almost the entire plant (vs. sessile to subsessile leaves, light brown to hyaline indumentum), cincinni without bracteoles (vs. bracteoles present), acuminate flower buds and sepals (vs. obtuse to rounded), androecium and gynoecium completely white (vs. androecium and gynoecium partially bluish, lilac or purple), ovary and capsules hirsute (vs. velutine, hispid or lanate), seeds with light grey to grey and foveolate testa (vs. medium to dark brown and rugose or scrobiculate testa), and hyaline and inconspicuous aril (vs. aril cream-colored, slightly translucent and thick). Diagnosis. Similar to S. almeidae due to its sessile to subpetiolate, entirely green leaves, present bracteoles, sessile flowers, purple filaments and style combined with white anthers, and oblongoid to broadly oblongoid capsules. Siderasis medusoides is distinct due to its membranous leaves, elongate and tangled cincinni, small flowers, and purple to dark blue and elliptic to narrowly obovate or spatulate petals.
Specimens seen ( Etymology. The epithet alludes to the extremely elongated cincinni, common in mature individuals of this species, due to their resemblance to the snakes that composed the hair of Medusa, one of the three Gorgon sisters from Greek mythology. Distribution and habitat. Siderasis medusoides is known from the municipalities of Marilândia and Santa Leopoldina, in the state of Espírito Santo (Fig. 2). It grows in lowland Atlantic Forest, in shady and moist areas with great leaf litter accumulation, 90-550 m above the sea level.
Phenology. It blooms from December to March and fruits between January and March.
Conservation status. Siderasis medusoides possesses narrow EOO (ca. 11037 km 2 ) and AOO (ca. 2000 km 2 ), and based solely on distribution data should be treated as Endangered (EN). Nonetheless, it is known from only five collections in three different localities. They were made within urban areas, and these localities have suffered greatly with direct anthropic influence and deforestation in recent years. We have made several attempts to recollect S. medusoides, but they were all unsuccessful. Thus, we suggest that S. medusoides be considered Critically Endangered [CR, A2abcd+B2ab(i, ii, iii, iv)+C1+C2b+D2].
Affinities. Siderasis medusoides is similar to S. almeidae and S. albofasciata, due to their sessile to subpetiolate leaves, inflorescence with elongate main axis, bracteolate cincinni, sessile flowers, and purple filaments and style combined with white anthers. Nevertheless, it can be easily differentiated from S. almeidae by its inconspicuous subterraneous and aerial stems (vs. subterraneous stems absent and aerial stems elongate, in S. almeidae), membranous leaves appressed against the soil (vs. succulent and ascending), membranous and internally light green sepals (vs. fleshy and internally lilac to purple), narrowly obovate to spatulate petals (vs. rhomboid to broadly obtrullate), and hispid ovary (vs. lanate). It can be differentiated from S. albofasciata by lacking flagelliform-shoots (vs. flagelliform-shoots produced after the fertile period, in S. albofasciata), concolorous and membranous leaves (vs. adaxially variegated, abaxially vinaceous, succulent leaves), petals entirely purple to bluish purple (vs. petals with white basal third), and white anthers (vs. anthers purple to bluish purple). Siderasis medusoides is peculiar due to its membranous leaves appressed to the soil, tangled and elongate cincinni, small flowers, and narrow petals. Diagnosis. Very distinctive due to its vining habit, distichously-alternate leaves, blades asymmetric at base, main florescence a many-branched thyrse, with alternate cincinni, flowers zygomorphic, bisexual or staminate, stamens unequal, curved upwards, sigmoid filaments, and capsules globose and shallowly foveolate. It can be differentiated from S. zorzanellii by its membranous and velutine leaves, inflorescences always terminal in the secondary branches, petals dark mauve to vinaceous, rarely light pink or white, with margins ciliate with non-moniliform hairs.
Distribution and habitat. Siderasis spectabilis is confined to the type locality, in the native vegetation of the Horto Santos Lima (currently the headquarters of the Desengano State Park), in Santa Maria Madalena, state of Rio de Janeiro (Fig. 2). Nothing is known about this species habitat, since the original labels give no information on the area and all field expeditions to recollect this plant have been unsuccessful.
Phenology. Since all known collections were done on the same day, S. spectabilis is only known to bloom during January. Fruits and seeds are unknown for this species.
Conservation status. Due to the complete lack of information on the distribution, ecology and lack of any collections aside from the type specimens, according to the criteria proposed by IUCN (2001), S. spectabilis should be considered Data Deficient (DD), until new collections and data become available.
Affinities. Siderasis spectabilis is morphologically closely related to S. zorzanellii, but S. spectabilis can be easily differentiated due to its inflorescences being always ter-  minal in the secondary branches (vs. axillary in the primary branches or terminal in the secondary branches, in S. zorzanellii), and petals dark mauve to vinaceous, rarely light pink or white, and margins ciliate with non-moniliform hairs (vs. white and glabrous margins). All studied specimens were in excellent condition, and color of most organs could be easily described. Regarding color pattern in the androecium and gynoecium, S. spectabilis is similar to S. albofasciata. These are the only two species in the genus to present the upper third of filaments and style, and the anthers in the same color as the petals, contrasting greatly with the white base of filaments and style, and the white ovary of other species. Nevertheless, both species can be easily differentiated using vegetative or reproductive characters. One specimen (L.E. Mello-Filho 1171) is peculiar in being the only specimen with light-colored flowers. In the label, it is described by the collector as possessing white flowers. Nonetheless, while analyzing the duplicates available at R, RB, SPF and US, we noticed that a few flowers possessed pale pink pigment (particularly noticeable in the petals and stamens). We believe that these specimens might represent albino or semialbino individuals, and thus merit no taxonomic status, especially since they were collected at the same place and date as the remaining darkflowered specimens. Aona (2008), in her unpublished Ph.D. thesis, lists one of the paratypes of S. spectabilis under Dichorisandra incurva Mart. This is justified by her due to the specimens climbing habit, decumbent apical branches, distichously-alternate and sessile leaves, inflorescence composed of a pedunculate, many-branched thyrse, with alternate cincinni, and "white" [sic] flowers. Nevertheless, S. spectabilis can be easily differentiated by its erect inflorescences (vs. pendant to curved downwards, hence the name, in D. incurva), flower buds broadly ellipsoid to broadly obovoid, with truncate apex (vs. ellipsoid, with acute apex), sepals fleshy (vs. membranous), petals dark mauve to vinaceous, rarely light pink or white, with margins ciliate with non-moniliform hairs (vs. white with glabrous margins), stamens 6, anthers dorsifixed, 3 to 4 times shorter than the filaments, dehiscent by extrorse slits, and anther sacs divergent, semicircular, and expanded connectives (vs. stamens 6 or 5 + the upper one modified into a staminode, anthers basifixed, 3 to 4 times longer than the filaments, dehiscent by introrse slits, but functionally poricidal, anthers sacs parallel, elongate, and inconspicuous connectives). All these floral characters can be easily observed with the dissection of mature flower buds in herbarium specimens. The floral morphology of D. incurva is illustrated in Fig. 1I Diagnosis. Similar to S. spectabilis due to its vining habit, distichously-alternate leaves, blades asymmetric at base, main florescence a many-branched thyrse, with alternate cincinni, flowers bisexual or staminate, zygomorphic, stamens unequal, curved upwards and sigmoid filaments. It can be differentiated from by its chartaceous and sparsely velutine leaves, inflorescences axillary in the primary branches or terminal in the secondary branches, and petals white with glabrous margins.
Specimens seen ( Etymology. The epithet honors the collector of the type specimens, João Paulo Fernandes Zorzanelli, Brazilian botanist and dear friend of the authors. JPFZ is an active and prominent collector in the state of Espírito Santo, with collections currently focused on Serra do Valentim, the type locality of S. zorzanellii. Distribution and habitat. Siderasis zorzanellii is confined to the municipality of Iúna, Espírito Santo (Fig. 2). It occurs in the "Floresta Ombrófila Densa Montana" vegetation, at 1200-1350 m above the sea level, generally near disturbed sites, being less frequent in well-preserved areas. This could be related to its climbing habit and the need of more sunlight exposure then the rosette species of the genus. This pattern is common in other liana and vine groups, such as Bignoniaceae, Malpighiaceae, and Sapindaceae (Acevedo-Rodríguez, pers. comm.), especially evident in big families such as Asteraceae, where the primarily climbing genus Mikania Willd. is almost exclusively found at the edge of forests, along trails, and in disturbed areas (Oliveira 2015).
Phenology. It was found in bloom from December to March and in fruit in March. Conservation status. Siderasis zorzanellii is very narrowly distributed, with an EOO of ca. 7.779 km 2 and an AOO of ca. 300 km 2 . The subpopulations are small, with no more than 10 mature individuals each. Unlike for the rosette species in the genus, it is still uncertain if the two climbing species reproduce vegetatively through cloning. Flowering seems to be frequent, although fruits have been collected only once. Thus, following the recommendations from IUCN (2001), S. zorzanellii should be considered Critically Endangered [CR, A2abde+B1ab(iii, iv, v)+ B2ab(iii, iv, v)+C2a(ii)+D1+D2].
Affinities. Siderasis zorzanellii is morphologically similar to S. spectabilis. Nevertheless, both species can be differentiated based on consistency of the leaf blades (chartaceous in S. zorzanellii vs. membranous in S. spectabilis), density of their pubescence (sparsely minutely velutine vs. minutely velutine), position of the inflorescences (terminal in the secondary branches or axillary in the older nodes of the primary branches vs. exclusively terminal in the primary branches), floral morphology (flowers 0.7-0.9 cm diameter, petals white, margins glabrous vs. flowers 1-1.3 cm diameter, petals dark mauve to vinaceous, rarely light pink or white, margins ciliate with non-moniliform hairs), and by their disjunct distribution (southern montane Espírito Santo state vs. northern montane Rio de Janeiro state).

Final remarks
The present work adds four new species to Siderasis, along with the addition of new morphological characters that help clarify the circumscription of the group. Siderasis Raf. emend. M.Pell. & Faden may be uniquely characterized as comprising small perennial rosette herbs or robust perennial vines, with shoots determinate or indeterminate, leaves spirally-or distichously-alternate. The inflorescences are terminal or axillary, either a many-branched thyrse with alternate cincinni or reduced to a solitary cincinnus, cincinni always several-flowered. The flowers are chasmogamous, bisexual or male, actinomorphic or zygomorphic, and petals with glabrous margins or ciliated with non-moniliform hairs. The androecium is composed of 6 fertile stamens, filaments straight or sigmoid, anthers dorsifixed and extrorsely rimose, anther sacs semicircular, divergent, connectives expanded and quadrangular. In the gynoecium, the stigma is annular-truncate or annular-capitate, marginally papillate with unicellular papillae restricted to the margin of the stigmatic regions. Also, similar to Dichorisandra, the capsules are thick-walled, and the seeds are arillate, biseriate to partially uniseriate, with semidorsal or semilateral embryotega, and a C-shaped hilum. All species accepted by us are easily diagnosed by a unique and constant combination of morphological character states. Furthermore, each species can be easily separated based on their geographical distribution, since they are microendemics, with non-overlapping distribution areas (Fig. 2).
As indicated by several systematic studies in Commelinaceae (Evans et al. 2000(Evans et al. , 2003Hardy 2001;Wade et al. 2006;Zuiderveen et al. 2011;Hertweck and Pires 2014) and by the morphological evidence presented here and by Pellegrini (2017), the need to recircumscribe subtribe Dichorisandrinae is pressing. Aside from the cytological character of x=19 large chromosomes described by Jones and Jopling (1972) and hypothesized by Faden and Hunt (1991), no macro or micromorphological synapomorphies were found so far for subtribe Dichorisandrinae in its current circumscription. On the other hand, if subtribe Dichorisandrinae is recircumscribed to exclusively contain Dichorisandra and Siderasis, Dichorisandrinae s.s. can be easily morphologically characterized by its thick-walled capsules, the biseriate to partially uniseriate arillate seeds, semidorsal to semilateral embryotega, and C-shaped hilum. The lineage composed by Geogenanthus(Cochliostema+Plowmanianthus) needs to be formally recognized as a subtribe, and can be easily circumscribed by its petals with marginally fringed with moniliform hairs, and anthers sacs curved to spirally-coiled and appressed to each other. Phylogenetic studies using both nuclear and chloroplast sequences seem promising in elucidating phylogenetic incongruences in Commelinaceae (e.g. Burns et al. 2011). However, most phylogenetic in the family so far completely disregard morphological data, with the exception of Evans et al. (2000Evans et al. ( , 2003. Studies focusing on the systematics and recircumscription of Dichorisandrinae are currently being conducted, combining morphological and molecular data (Pellegrini et al., in prep.), and should shed some light on the evolution of the reproductive biology in the family.