Phylogeny and taxonomic synopsis of PoasubgenusPseudopoa (including Eremopoa and Lindbergella) (Poaceae, Poeae, Poinae)

Abstract Eremopoa is a small genus of annual grasses distributed from Egypt to western China. Phylogenetic analyses of plastid and nuclear ribosomal DNA show that Eremopoa species, together with the monotypic genus Lindbergella and a single species of Poa (P.speluncarum), are nested within the genus Poa, in a clade that we accept as Poasubg.Pseudopoa. Here we accept seven species, four subspecies and four varieties in Poasubg.Pseudopoa. Five new combinations are made: Poaattalica, P.diaphora var. alpina, P.diaphora var. songarica, P.nephelochloides and P.persicasubsp.multiradiata; P.millii is proposed as a replacement name for E.capillaris; and Poa sections Lindbergella and Speluncarae are proposed. We provide a diagnosis for Poasubg.Pseudopoa, synonymy for and a key to the taxa. Eight lectotypes are designated: Eragrostisbarbeyi Post, Eremopoanephelochloides Roshev., Glyceriataurica Steud., Nephelochloatripolitana Boiss. & Blanche, Poacilicensis Hance, Poaparadoxa Kar. & Kir., Poapersicavar.alpina Boiss and Poapersicasubsp.cypria Sam. Eremopoamedica is re-identified as a species of Puccinellia.


Introduction
Eremopoa species are annual with some extreme features usually not found in Poa, but, other than abundantly branching panicles, those characteristics are broached in all cases. No one has doubted that Eremopoa was closely related to Poa.
The taxa placed in Eremopoa range from Egypt (Sinai and north coast) across the northern Middle East (Israel, Lebanon, Syria, Iraq, Turkey [Anatolia], Iran), to Afghanistan, Pakistan, northwest India (Himachal Pradesh, Kashmir), western China (Tibet and Xinjiang), north through Transcaucasia into the Caucasus mountains of Russia and across central Asia in Turkmenistan, Uzbekistan, Tajikistan, Kyrgyz Republic and Kazakhstan. Two taxa have been observed elsewhere as waifs: E. persica in western Europe (France, Norway) and E. altaica (Trin.) Roshev. in Canada (see references in Taxonomy section). The geographic region with the most diversity of Eremopoa taxa is clearly Asia Minor; nearly all of the accepted species occur in Turkey.
There have been many differences of opinion on the species and infraspecific ranks to accept in Eremopoa (Table 1). Roshevitz (1934) treated six species in his new genus in the former U.S.S.R (E. altaica, E. bellula (Regel) Roshev., E. oxyglumis (Boiss.) Roshev., E. multiradiata (Trautv.) Roshev., E. persica and E. songarica). Tzvelev (1976) reduced these six species to two species, E. persica and E. altaica, with two and three subspecies, respectively, all of which were accepted as species by Czerepanov (1981Czerepanov ( , 1995. Scholz (1980Scholz ( , 1981 described two new species, E. attalica H. Scholz from Turkey and E. medica H. Scholz from Azerbaijan. The type of E. medica (holotype at W, isotype at B) was determined to be a species of Puccinellia Parl. (Soreng pers. obs. 2015). Mill (1985) treated six species in Turkey, including two new species, E. capillaris R.R. Mill and E. mardinensis R.R. Mill. Rahmanian et al. (2014) accepted four species in Iran, including E. medica and E. persica with three varieties.
Bor's genus Lindbergella (Bor 1968b(Bor , 1969) comprises a single annual species that is morphologically similar to Eremopoa. It differs from Eremopoa only in having firmer lemmas that are 3-veined and obscurely apiculate and panicles with 1-5 branches that are smooth. Lindbergella sintenisii (H. Lindb.) Bor was originally published as Poa sintenisii by Lindberg (1942) and also as P. persica var. cypria by Samuelsson (1950), the type of which is a syntype of P. persica var. alpina Boissier (1884). The species is endemic to Cyprus.
The first molecular data on Eremopoa, generated by our lab in 2004/2005, indicated that E. songarica was nested within Poa. That data was first published by Gillespie et al. (2007) using chloroplast DNA sequences from the trnT-trnL-trnF region. Based on this same data, inclusion of Eremopoa in Poa was already applied in the Flora of China account (Zhu et al. 2006, as P. subg. Pseudopoa (K. Koch) Stapf ) and was continued in Gillespie et al. (2008Gillespie et al. ( , 2010, Soreng (2004+) and Soreng et al. (2010Soreng et al. ( , 2015aSoreng et al. ( , 2017a. Although nested within Poa, Eremopoa was positioned on a long branch separate from other Poa clades, justifying its recognition as a distinct subgenus, P. subg. Pseudopoa (Gillespie et al. 2007).
We published our initial DNA results for only one species of Eremopoa (E. songarica) based on trnT-trnL-trnF and, subsequently, nuclear ribosomal (nrDNA) ITS and ETS sequence data (Gillespie et al. 2007, 2008. We subsequently sequenced two additional plastid regions (matK and rpoB-trnC) and added Table 1. Classification history of Eremopoa and other taxa here accepted in Poa subg. Pseudopoa. Species and infraspecific taxa accepted by Roshevitz (1934) and authors of major floras and the region covered by their treatments are given. The last column provides the corresponding names in Poa accepted here.  data for Eremopoa persica (Cabi et al. 2017, as Poa persica). A DNA analysis of ITS sequence data by Hoffmann et al. (2013) showed Lindbergella sintenisii was also nested within Poa near Eremopoa. Since then, we have accumulated nrDNA and plastid sequence data for most of the Eremopoa taxa and L. sintenisii and sampled many more species of Poa from Turkey and around the world. Analysis of our accumulated phylogenetic data on Eremopoa is presented here. All Eremopoa taxa were nested well within Poa, and P. speluncarum J.R. Edm. and L. sintenisii were found to be nested within or sister to the set of Eremopoa species. Here we place these taxa in Poa subg. Pseudopoa and present a taxonomic synopsis of all the species and infraspecies, as well as a key to the taxa we currently accept. Further study is needed before a comprehensive revision of the subgenus can be produced.

Methods
Collections of Eremopoa at E and G (except those not available for loan), several from P and two type specimens from BM and B were loaned to RJS at US. Other material was examined by RJS at B, K, LE, P, US and herbaria in Turkey (ANK, ISTE, NKU). Fieldwork in which 38 specimens of Eremopoa were collected by us was conducted in Kyrgyz Republic (RJS 2006) and Turkey (RJS and associates 1994(RJS and associates , 2013(RJS and associates , 2014(RJS and associates , 2015 LJG & RJS and associates 2011; EC was a co-collector on the 2011 to 2015 expeditions). Additional material was obtained from R. Hand (Lindbergella sintenisii) and M. Assadi and M. Amini-Rad (Iranian Eremopoa). The molecular phylogenetic analysis included 77 samples: 15 Eremopoa, 56 Poa, 1 Lindbergella and 5 outgroup samples (Appendix 1). A diverse set of Poa species was chosen to represent the majority of sections, including all sections in southwest Asia. Outgroup taxa were chosen to include representatives of the two taxa (Phleum L. and Milium L.) and one clade considered most closely related to Poa , Soreng et al. 2015b). Sequences of Lindbergella and the majority of Eremopoa samples, plus many matK and rpoB sequences, are new to this study (Appendix 1). For simplicity, due to the confusing taxonomy and nomenclature, we refer to Eremopoa taxa using names at the species level in the Results, trees and Appendix 1 (see Table 1 for their corresponding names in Poa). The collection TARI 135082 was previously identified as E. medica (Rahmanian et al. 2014), but was re-determined by RJS as P. persica subsp. persica.
DNA was extracted from silica gel dried or herbarium leaf material as described in Gillespie et al. (2008). Three plastid markers (matK, rpoB-trnC and trnT-trnL-trnF [TLF]) and two nuclear ribosomal DNA (nrDNA) markers (internal transcribed spacer [ITS] and external transcribed spacer [ETS]) were sequenced. Amplification and sequencing protocols, including primers used, were described in our previous studies, as follows: ITS and TLF (Gillespie et al. 2008); ETS (Gillespie et al. 2009; matK and rpoB-trnC (Soreng et al. 2015b). Sequences were assembled, edited, aligned and concatenated using Geneious ver. 6.1.5 (http://www.geneious.com). The MAFFT ver. 7.017 plugin (Katoh and Standley 2013) was used for alignment, followed by manual adjustment. All samples are complete for all markers, except for several samples with missing ends. The molecular study was conducted at the Canadian Museum of Nature; sequencing was mostly performed by NA, analyses by LJG.
Maximum parsimony (MP) analyses were performed in PAUP* 4.0b10 (Swofford 2002) using the heuristic search command with default settings, including tree bisection-reconnection (TBR) swapping, saving all multiple shortest trees (Multrees) with a maximum number set to 100,000. Branch support was assessed using MP bootstrap analyses performed in PAUP* with heuristic search strategy, 10,000 bootstrap replicates, each with ten random addition sequence replicates, saving ten trees per replicate.
Bayesian Markov chain Monte Carlo analyses were conducted in MrBayes (Ronquist et al. 2011). Optimal models of molecular evolution for individual markers were first determined using the Akaike information criterion (AIC; Akaike 1974) conducted through likelihood searches in jModeltest with default settings (Darriba et al. 2012). Models were set at GTR + Γ for ITS, ETS and rpoB-trnC partitions and GTR + I + Γ for matK and TLF partitions based on the AIC scores and the models allowed in Mr-Bayes. Two independent runs of four chained searches were performed for either two or three million generations (analyses were stopped when split frequency of 0.005 was reached or closely approached), sampling every 500 generations, with default parameters. A 25% burn-in was implemented prior to summarising a 50% majority rule consensus tree and calculating Bayesian posterior probabilities (pp).
MP heuristic searches and bootstrap analyses were performed initially on the separate marker alignments. Strict consensus trees were examined for conflicting topologies with incongruence identified by branch conflicts with ≥75% bootstrap support (BS). No supported incongruence was found between ITS and ETS trees, nor amongst the three plastid trees. Further MP and Bayesian analyses were performed on the separate concatenated nrDNA (77 samples, 1251 aligned characters) and plastid (77 samples, 4465 characters) alignments. Since supported incongruence was detected between the nrDNA and plastid strict consensus trees, species and clades determined to be incongruent were removed prior to performing analyses on the concatenated combined nrDNA and plastid alignment (68 samples, 5599 aligned characters). Trees were viewed in FigTree v1.4.0 (Rambaut 2006+). Clade designations follow Soreng et al. (2010) with modifications as in Cabi et al. (2017) and Soreng et al. (2017b), wherein well-supported major clades are assigned letters.

Results
Plastid and nrDNA Bayesian trees are given in Fig. 1 with summary statistics in Suppl. material 1. There are 100 new sequences reported in GenBank and these are given in Appendix 1. MP trees (bootstrap values shown below branches in Fig. 1) were very similar to the Bayesian trees with a few minor unsupported differences. Major clades (shown by letter and colour in Fig. 1) are identical in both nrDNA and plastid trees, with two exceptions: Poa arctica R. Br. and P. sect. Secundae members (P. curtifolia Scribn., P. secunda J. Presl and P. stenantha Trin.), each belonging to different major clades in the two trees. The position of three major clades differs significantly between the nrDNA and plastid trees: J clade (sect. Jubatae: P. jubata A. Kern.), S clade (sects. Stenopoa and Abbreviatae) and V clade (sect. Pandemos: P. trivialis L.). Poa major clades have been described elsewhere (Gillespie et al. 2007, 2008, 2009, 2017b, Cabi et al. 2017); here we focus on the position of Eremopoa.
In the combined nrDNA and plastid tree (Fig. 2), the E clade is strongly supported as sister (pp = 1, BS = 100%) to a clade comprising Poa supersects. Homalopoa (H clade) and Poa (P clade) and the N clade (P. sect. Nanopoa plus unassigned species). In the nrDNA analysis, the E clade is strongly supported as sister to clades P+H (not differentiated), N, and X (represented here by P. arctica) (Fig. 1). In the plastid analysis, the E clade is sister to a larger clade comprising clades H, N, and P, plus J, S and V (Fig. 1).

Discussion
Our molecular analyses of plastid and nuclear ribosomal DNA strongly support the position of Eremopoa and Lindbergella within the genus Poa. Eremopoa and Lindbergella were united in a clade along with Poa speluncarum with strong support in the plastid and combined trees (weak support in the nuclear tree). We call this set the E clade   , Cabi et al. 2017) and accept it as Poa subg Pseudopoa. In its recent usage, this subgenus was initially considered to include only Eremopoa (Zhu et al. 2006, Gillespie et al. 2007); here it is expanded to include Lindbergella and P. speluncarum.
Within the E clade, three taxa of southwest Turkey and Cyprus, E. attalica, P. speluncarum and Lindbergella sintenisii, are phylogenetically isolated from all the other species of Eremopoa sampled (the core Eremopoa clade). All three taxa formed a strongly supported clade in the plastid tree, while in the nuclear tree only the first two species form a clade and L. sintenisii is sister to this clade plus the core Eremopoa clade. The position of L. sintenisii is moderately supported as incongruent between the nuclear and plastid trees suggesting that the genus may be of hybrid origin; however, further studies are needed to confirm incongruence over lack of support.
All Eremopoa taxa sampled, excluding E. attalica, form a strongly supported clade in all trees, called here the core Eremopoa clade. This clade includes two strongly supported subclades in the combined nuclear-plastid tree, corresponding to E. persica s.l. and E. altaica s.l. In the first subclade, E. multiradiata is nested amongst E. persica samples, as is the sample originally determined as E. medica (TARI 35082). The E. multiradiata sample (Soreng 9240) comes from the type locality of E. mardinensis in SW Turkey and is a good match for that species, but we believe that E. mardinensis should be treated as a synonym of E. multiradiata. The E. altaica s.l. subclade in the combined tree includes a strongly supported and divergent clade of three E. songarica samples and a clade of E. oxyglumis plus one sample of E. songarica (identification needs confirmation). The position of E. songarica (tetraploid) with E. oxyglumis (diploid and hexaploid) is strongly supported in the combined and nuclear trees, but is weakly supported with E. persica (diploid) in the plastid tree. This, together with ploidy level, is suggestive of a possible hybrid origin for E. songarica, but this hypothesis needs to be further explored.
As noted in the introduction and Table 1, there has been no consensus on the taxonomy of Eremopoa species. Bor (1970, p. 49) wrote "As far as the genus Eremopoa Roshev. is concerned I am prepared to accept two species only: Eremopoa persica (Trin.) Roshev. and E. bellula (Regel) Roshev." He considered E. songarica, multiradiata and oxyglumis "only worthy of varietal rank" as the single taxon, E. persica var. songarica. Tzvelev (1976), Cope (1982) and Mill (1985) dismissed the E. bellula form as indistinct, yet it was maintained as a species by Bor (1970) and Rahmanian et al. (2014). As such, the array of taxa has been treated as a series of species, subspecies or varieties. The taxonomy proposed by Tzvelev (1976) seems the most useful for treating E. persica s.l. and E. altaica s.l.; each is treated as a separate species with subspecies. His classification, supported by molecular data, is adopted here with some minor modifications.
Here, we present a synopsis of P. subg. Pseudopoa based on our current understanding. Further herbarium and molecular study is needed before a more comprehensive revision of the subgenus can be produced. We treat all Eremopoa species, Lindbergella sintenisii and P. speluncarum in P. subg. Pseudopoa. We merge all Eremopoa taxa and L. sintenisii into Poa and treat the Eremopoa taxa as five species. Poa diaphora Trin. is the correct name for E. altaica within Poa. Two subspecies, subsp. diaphora and oxyglumis (Boiss.) Soreng  Emended diagnosis. Like species of other Poa subgenera, but annual (P. speluncarum a weak stooling perennial) and differing from other annual species of Poa by combination of sheath margins fused only near the base (basal sheaths fused to 16%, top sheath 4-12% [to 50% in P. speluncarum]), panicle branches scabrous along angles (P. sintenisii smooth), arranged in whorl-like groups of 5 to 27 per node (sometimes fewer in P. diaphora and P. sintenisii), sometimes the lower whorls of branches naked or with only a few sterile spikelets, flowers bisexual, glumes short (lower glume 2/7-2/3 (-3/4) the first lemma in length), 1-veined (3-veined in P. sintenisii), apex sharply pointed, sometimes apiculate, rachilla internodes exposed, scaberulous, callus glabrous (or with a short crown of hairs in P. sintenisii), lemmas membranous to subchartaceous (P. sintenisii chartaceous), 3-5 veined, the intermediate veins faint when present, laterally compressed, but the keel not pronounced, glabrous or keel and marginal veins short sericeous (also sericeous between the veins in P. sintenisii), but keel scabrous distal to the hairs. Distribution. Southwest Asia from Israel, Lebanon, Cyprus and Turkey eastwards through Transcaucasia, Iran, central Asia to western China and northwest India. Sporadic elsewhere, possibly adventive on Egypt's North African coast but native east of the Red Sea, adventive in Europe and Canada.
Notes. A subgenus of seven species with several infraspecies, distributed mainly in semi-arid midlands to uplands (usually 300 m plus) to alpine, with winter spring / summer drought precipitation pattern, often along trails and roads, cultivated fields and pastures, around puddles, shallow springs, swales and vernal pools, snow beds, in pine/ oak forests to open grasslands and deserts, also in shallow caves, in shallow sandy or stony soils or screes of igneous or metamorphic rocks of igneous or sedimentary origin, including pumice, lava, serpentine, shale, sandstone, limestone and marble.
Key to Poa subgen. Pseudopoa taxa and other annual species of Poa in the coincident geographic region Plants annual (infrequently perennial or perenniating); anthers mostly 0.2-1 mm (to 1.7 mm in the weak stemmed, stooling perennial P. speluncarum, to 2.8 mm in the annual species Poa persica). Lemmas all glabrous or rarely with a few hairs near the base of the keel or marginal veins; spikelets (4-)5-10 ( 7)

Notes.
We provisionally retain this species in sect. Pseudopoa, despite its divergent phylogenetic placement. The species is morphologically similar to other members of the section. As noted by Mill (1985), it is most like Poa nephelochloides Roshev., but the anthers are smaller. Some populations of P. millii approach P. attalica and are problematical to separate (see under P. millii). Further molecular study is needed to determine if the three species are closely related and if a new section is warranted. Notes. Separating the four forms of Poa diaphora s.l. treated here is often difficult. Here we choose to recognise two subspecies as divided in the molecular plastid analysis. Subspecies diaphora and oxyglumis are most easily separated by the minute anthers (0.2-0.6 mm) combined with glabrous or nearly glabrous lemmas in the former and slightly longer anthers (0.6-1.1 mm) combined with hairy lemma keels and marginal veins in the latter. The other forms, diaphora s.s., songarica and alpina are essentially intergrading and are here treated as varieties in subsp. diaphora.

Notes.
Morphologically Poa millii is intermediate between P. persica subsp. persica and P. attalica. However, we are not sure which of these it is actually related to or if it is a hybrid between them. The type approaches P. persica in having anthers 1.2-1.3 mm long and P. attalica in having abundant branching and sometimes having some sterile branches amongst the lower branch whorls. Much of the material of P. millii from further west than the type location from the Taurus Mts. has smaller anthers and is problematical to separate from P. attalica. Notes. Due to its sterile whorls of branches, this species seems very close to Poa millii and P. attalica, but may be a derivative of P. persica since it has longer anthers than the previous taxa. Roshevits cited two gatherings of Köie: "Kechwar, 700 m (3 May 1937;no. 475). Chah-Bazan, 500 m" (Kechvar is about 60 km north of Dizful). The specimen at C has the same date and collection number as Roshevits cited and was annotated by Roshevits as this taxon; we select it as the lectotype. The anthers are ca. 1.1-1.2 mm as measured from the C photo and other characters seem to match P. attalica. The anther length is given as 1.5 mm in Roshevits' diagnosis. The specimen clearly has the hyaline lemma apices of P. persica s.l. (in contrast to P. diaphora). However, these features are also present in the type of E. capillaris (=P. millii). Poa attalica has shorter anthers, ca. 0.8 to 1 mm, on the type (anthers not described by Scholz 1980or Mill 1985. Poa nephelochloides and P. attalica may represent the same species, diagnosed as different from P. persica by sterile branches and from Nephelochloa orientalis Boiss. by glabrous lemmas (P. nephelochloides has pubescent lemmas). However, Poa nephelochloides and P. attalica are geographically isolated by over 1500 km and have different anther lengths. Notes. The presence of hairs on the lemmas in material treated as "multiradiata" is confused in the literature. Mill (1985) indicates that E. multiradiata and E. persica s.s. have lemma keels hairy in the lower ⅓-½. We concur with Tzvelev (1976), who keyed E. persica subsp. persica as lemmas short pilose along the base of keel and marginal veins and subsp. multiradiata as lemmas glabrous or with a few solitary hairs. Mill (1985) distinguished his new species Eremopoa mardinensis from E. multiradiata based on its glabrous lemmas, 8-12-flowered spikelets and florets strongly divergent from the rachilla. However, subsp. multiradiata also has glabrous lemmas (as noted above) and divergent florets (when spikelets are in flower) and its (4)5-9(10)-flowered spikelets overlap in number; therefore, we treat E. mardinensis as a synonym of E. multiradiata. The type material of Eragrostis barbeyi is from the same place as E. mardinensis and is clearly the same form (spikelets many-flowered); Nephelochloa tripolitana, with ca. 12-14-flowered spikelets, also appears to belong to this form. If E. mardinensis were accepted as a species, the basionym names Eragrostis barbeyi or Nephelochloa tripolitana would have priority.  Edmondson (1985) as an annual species of Poa sect. Ochlopoa Asch. & Graebn (≡ Poa sect. Micrantherae Stapf. Type: Poa annua). Our investigation found it to be a feeble, stooling perennial with sparsely scabrous panicle branches, uppermost sheaths closed up to half their length, spikelets sparsely scaberulous, mostly 1-flowered, the distal-most ones frequently 2(-3) flowered, anthers 1.1-1.7 mm, caryopsis 1.7-1.8 mm long, hilum 0.3 mm long and grain adherent to the palea. DNA data have clearly placed it in the Poa clade that includes Eremopoa species (E clade), either as sister to P. attalica (nuclear data) or as sister to P. attalica + P. sintenisii (plastid data). The species is odd in subgenus Pseudopoa for its perennial habit (albeit weak) and more closed sheaths, and in Poa generally by its mostly uniflorous spikelets. It is a very rare species that lives in the backs of shallow, moist, cool caves in the Taurus Mts., along with other cave endemics.  Boissier [1884]). Tzvelev (1976, pg. 480) noted that the holotype collection of E. bellula appeared to be a mix of altaica (diaphora) and songarica forms ("p.p. max" = E. altaica subsp. songarica, somewhat intermediate between this subsp. and subsp. altaica, and "p.p. minor" = E. altaica subsp. altaica); he considered E. bellula to be a synonym of E. altaica subsp. songarica. Further study is needed to clarifiy the placement of Eremopoa bellula and determine if it is synonymous with P. diaphora var. alpina. Notes. Tzvelev (1976, pg. 480) included E. glareosa as a synonym under E. altaica subsp. songarica, but noted that it is somewhat intermediate between this taxon and E. altaica subsp. altaica. As the protologue indicates the plants are 10-28 cm tall, with 3 to 4 florests per spikelet, spikelets 4-7 mm long and anthers 2.5 mm long, this is more likely to be Poa persica, perhaps subsp. multiradiata, since no pubescence is indicated. Note. There is no location in the species protologue beyond the article title "Beitrage zu einer Flora des Orients". Tzvelev (1976) indicated this name and the next, Festuca polygama, probably apply to Eremopoa persica and that the types of these were in Berlin (B). Clayton et al. 2002+ (GrassBase) reflect the same information. RJS was unable to locate type material of either of these two names at B, P or via internet searches. Notes. Tzvelev (1976) indicates "Caucasus?", but there is no location in the species protologue beyond the article title "Beitrage zu einer Flora des Orients". Notes. The type collection of Eremopoa medica is clearly a perennial species of Puccinellia (possibly P. gigantea (Grossh.) Grossh.) with lemmas rounded on the back, a distinct short crown of callus hairs and papillae common on vegetative structures (pedicels and leaves). Material cited as E. medica in Rahmanian et al. (2014, fig. 5) appears to us to be Poa persica subsp. persica; their description and illustration indicate an annual habit, pubescent lemmas and panicles with 10 or more branches per whorl. The single specimen (TARI 35082) cited was included in our molecular analysis and formed a clade with other P. persica accessions in all trees. Notes. Kew GrassBase (Clayton et al. 2002+) indicates it is equal to E. persica. The specimen K00078950 (ex P) (image!), Voyage V. Jacquemont aux Indes orient. no. 1902, has this name on the label. The specimen is certainly P. diaphora, not P. persica.