Plastid phylogenomics of the cool-season grass subfamily: clarification of relationships among early-diverging tribes

Whole plastid genomes (plastomes) are being sequenced rapidly from across the green plant tree of life, and phylogenetic analyses of these are increasing resolution and support for relationships that were unresolved in earlier studies. The cool-season grass subfamily, Pooideae, includes important temperate cereals, turf grasses and forage species, yet some aspects of deep phylogeny in the lineage are unresolved. We newly sequenced 25 Pooideae plastomes, and conducted phylogenomic analyses of these and 20 existing plastomes from the subfamily. Most aspects of deep relationship in Pooideae are maximally supported in our analyses, including those among early-diverging tribes.

genetic analyses of these are increasing resolution and support for relationships that have varied among or been unresolved in earlier single-and multi-gene studies. Pooideae, the cool-season grass lineage, is the largest of the 12 grass subfamilies and includes important temperate cereals, turf grasses and forage species. Although numerous studies of the phylogeny of the subfamily have been undertaken, relationships among some 'early-diverging' tribes conflict among studies, and some relationships among subtribes of Poeae have not yet been resolved. To address these issues, we newly sequenced 25 whole plastomes, which showed rearrangements typical of Poaceae. These plastomes represent 9 tribes and 11 subtribes of Pooideae, and were analysed with 20 existing plastomes for the subfamily. Maximum likelihood (ML), maximum parsimony (MP) and Bayesian inference (BI) robustly resolve most deep relationships in the subfamily. Complete plastome data provide increased nodal support compared with proteincoding data alone at nodes that are not maximally supported. Following the divergence of Brachyelytrum, Phaenospermateae, Brylkinieae -Meliceae and Ampelodesmeae-Stipeae are the successive sister groups of the rest of the subfamily. Ampelodesmeae are nested within Stipeae in the plastome trees, consistent with its hybrid origin between a phaenospermatoid and a stipoid grass (the maternal parent). The core Pooideae are strongly supported and include Brachypodieae, a Bromeae -Triticeae clade and Poeae. Within Poeae, a novel sister group relationship between Phalaridinae and Torreyochloinae is found, and the relative branching order of this clade and Aveninae, with respect to an Agrostidinae -Brizinae clade, are discordant between MP and ML/BI trees. Maximum likelihood and Bayesian analyses strongly support Airinae and Holcinae as the successive sister groups of a Dactylidinae -Loliinae clade.

Introduction
Advances in next-generation sequencing technologies (Moore et al. 2006;Cronn et al. 2008;Parks et al. 2009;Wysocki et al. 2014) have resulted in a rapid increase in completed plastid genomes (Jansen and Ruhlman 2012) sampled widely across the green plant tree of life. The use of whole plastomes to infer phylogenies (i.e. phylogenomics) provides opportunities to potentially increase resolution and support for relationships that have varied among or been unresolved and/or poorly supported in earlier single-and multi-gene studies. Plastomes have been used to address diverse phylogenetic questions at deep (Ruhfel et al. 2014) to shallow (Parks et al. 2009) hierarchical levels, and to characterize plastid genome evolution (e.g. patterns of gene loss and organization, GC content, microstructural events, evolutionary rates) (e.g. Barrett and Davis 2012;Jansen and Ruhlman 2012).
Grasses are the fourth largest family of flowering plants in the world, with some 11 000 species and 600 -900 genera. At the family level, numerous phylogenetic studies have been conducted and the deep phylogenetic framework for Poaceae is well established. The family contains three small, deeply diverging subfamilies (Anomochlooideae, Pharoideae, Puelioideae) that are the successive sister groups of a large clade comprising two major lineages, the Bambusoideae, Ehrhartoideae, Pooideae (BEP) and the Panicoideae, Arundinoideae, Chloridoideae, Micrairoideae, Aristidoideae, Danthonioideae (PACMAD) clades (Grass Phylogeny Working Group 2001;Duvall et al. 2007;Sanchez-Ken and Clark 2007;Bouchenak-Khelladi et al. 2008;Saarela and Graham 2010;Grass Phylogeny Working Group II 2012).
The BEP clade includes the bamboo (Bambusoideae), rice (Ehrhartoideae) and cool-season (Pooideae) grass subfamilies. The origin of its crown clade has been dated at 40 -53 to 70.7 -72.6 million years ago, depending on which fossil calibration points are used (Christin et al. 2014). Earlier analyses inferred varying relationships among the subfamilies of the BEP clade (Hsiao et al. 1998;Zhang 2000;Grass Phylogeny Working Group 2001;Duvall et al. 2007) and a few studies found that Pooideae may be the sister group of the PACMAD clade (Soreng and Davis 1998;Duvall et al. 2007), but there is now consensus from plastid multi-gene analyses that Bambusoideae and Pooideae are sister taxa (Bouchenak-Khelladi et al. 2008;Saarela and Graham 2010;Grass Phylogeny Working Group II 2012;Wu and Ge 2012). This is consistent with some symplesiomorphic bambusoid macro-and micromorphological characters in 'early-diverging' lineages within Pooideae that were traditionally included in Bambusoideae (Clark et al. 1995). This topology has been confirmed by a phylogenomic study of 121 orthologous nuclear genes (Zhao et al. 2013) and a species tree inferred from .18 000 maximum likelihood (ML) gene trees (Burleigh et al. 2011), but it was not recovered in a nuclear analysis of 43 putative orthologous cDNA sequences (Peng et al. 2010).
Pooideae, the cool-season grass lineage, is the largest in the BEP clade and the largest of the 12 grass subfamilies, with some 189 genera, 10 nothogenera and 3900 species (Grass Phylogeny Working Group 2001;Clayton et al. 2006 onwards;Soreng et al. 2014). It includes important temperate cereals like wheat (Triticum), barley (Hordeum) and oats (Avena), cool-season turf grasses in the genera Festuca, Poa, Agrostis, Lolium and Schedonorus (Beard 2012), and numerous temperate forage species. Since the establishment of Pooideae in the mid-19th century by Bentham (1861), its circumscription has undergone radical realignment as understanding of the evolutionary history of grasses advanced (reviewed in Soreng and Davis 2000;. Pooideae, in its current circumscription, was first recognized as a natural group in cladistic studies of morphological data (Kellogg and Campbell 1987), and this hypothesis has been corroborated by numerous molecular phylogenetic studies (Soreng et al. 1990; Davis and Soreng 1993Cummings et al. 1994;Nadot et al. 1994;Clark et al. 1995;Catalá n et al. 1997;Hsiao et al. 1998;Soreng and Davis 1998;Mathews et al. 2000;Grass Phylogeny Working Group 2001;Duvall et al. 2007;Bouchenak-Khelladi et al. 2008;Grass Phylogeny Working Group II 2012;Blaner et al. 2014 [matK analyses]).
Multiple phylogenetic analyses based on plastid and nuclear ribosomal data have clarified the major evolutionary lineages within Pooideae (e.g. Davis and Soreng 2007;Dö ring et al. 2007;Bouchenak-Khelladi et al. 2008;Schneider et al. 2009). These are recognized as supertribes, tribes and/or subtribes in two recent classifications that differ only in the ranks chosen for some lineages. The classification by Soreng et al. (2014), modified from Soreng et al. (2003, recognizes 2 supertribes, 14 tribes and 1 subtribe, while that of Rö ser and colleagues (Dö ring et al. 2007;Schneider et al. 2009Schneider et al. , 2011 recognizes 9 tribes, 9 subtribes and  Soreng et al. (2014) and Schneider et al. (2009Schneider et al. ( , 2011. Parallel taxa (taxon names in boldface) are equivalent in circumscription but may differ in rank. Numbers in square brackets are the number of genera in each taxon according to Soreng et al. (2014). Schneider et al. (2009Schneider et al. ( , 2011 do not provide a subtribal classification for their Aveneae/Poeae complex. Soreng et al. (2014) Schneider et al. (2009Schneider et al. ( , 2011 Tribe  (557); Clayton et al. 2006 onwards] species-poor 'earlydiverging' lineages (those resulting from the earliest or deepest splits in the clade), and the core Pooideae, which as defined by , includes the tribes Brachypodieae, Bromeae, Poeae, Triticeae and the genus Littledalea, and includes the majority of species in the subfamily. Most studies identify Brachyelytrum [¼ Brachyelytreae], a small enigmatic genus that has been variously recognized as pooid or bambusoid (reviewed in Saarela et al. 2003), as the sister group of the rest of the subfamily [Catalá n et al. 1997;Grass Phylogeny Working Group 2001;Davis and Soreng 2007 (matK analyses); Duvall et al. 2007;Bouchenak-Khelladi et al. 2008;Davis and Soreng 2010;Grass Phylogeny Working Group II 2012;Blaner et al. 2014], although in a few studies Brachyelytrum falls outside the Pooideae clade [Davis and Soreng 1993;Soreng and Davis 1998;Blaner et al. 2014 (topoisomerase 6 analysis)]. After the divergence of Brachyelytrum, a clade of Lygeum [¼ Lygeeae] and Nardus [¼ Nardeae] is the next successive sister group of the rest of the subfamily in multiple studies [Catalá n et al. 1997;Davis 1998, 2000;Mathews et al. 2000;Grass Phylogeny Working Group 2001;Soreng 2007, 2010;Dö ring et al. 2007 (Nardeae not sampled); Duvall et al. 2007;Bouchenak-Khelladi et al. 2008;Schneider et al. 2011;Grass Phylogeny Working Group II 2012;Blaner et al. 2014], with the exception of two in which Brachyelytrum and Nardus form a moderately supported clade that is sister to rest of the subfamily (Hsiao et al. 1998;. The tribes Meliceae, Brylkinieae, Phaenospermateae, Stipeae and Ampelodesmeae are the next emerging branches in the phylogeny. Brylkinieae has been included only in a few analyses, where it has been placed as the sister group of Meliceae (Schneider et al. 2009(Schneider et al. , 2011Davis and Soreng 2010;Romaschenko et al. 2010Romaschenko et al. , 2012Rodionov et al. 2013;Blaner et al. 2014). The monotypic Ampelodesmos (A. mauritanicus) [¼ Ampelodesmeae] has consistently been nested in Stipeae in plastid and nuclear ribosomal analyses (Hsiao et al. 1998;Soreng 2007, 2010;Grass Phylogeny Working Group II 2012;Romaschenko et al. 2012), although it differs dramatically in spikelet morphology, being multi-flowered rather than single-flowered like Stipeae. Using a low copy nuclear gene, Romaschenko et al. (2014) recently demonstrated that Ampelodesmos originated via hybridization between a phaenospermatoid and an 'early-diverging' stipoid grass (the maternal parent), explaining its placement in Stipeae in plastid trees, and leading Soreng et al. (2014) to accept Ampelodesmeae as a monotypic tribe.
The interrelationships among Stipeae-Ampelodesmeae, Meliceae and Phaenospermateae with respect to the rest of Pooideae have varied substantially among studies based on different taxon samplings, molecular markers and methods of phylogenetic inference. Three studies infer these to be successively diverging lineages with strong support in Bayesian analyses for at least two of the three relevant deep nodes, but in each of these studies the position of at least one of the tribes/lineages differs with respect to the others. The best supported topology is presented in a tree based on two plastid genes, in which Phaenospermateae (not monophyletic in the analysis), Stipeae -Ampelodesmeae and Meliceae are strongly supported as successively diverging lineages (Duvall et al. 2007). In contrast, a tree based on three plastid genes identifies Stipeae -Ampelodesmeae, Phaenospermateae and Meliceae as successively diverging lineages (in other words, the branching order of the first two lineages is reversed), but with weak support for the clade that includes Phaenospermateae, Meliceae and the rest of Pooideae (Grass Phylogeny Working Group II 2012). A nine-region plastid tree supports Phaenospermateae, Brylkinieae -Meliceae and Stipeae -Ampelodesmeae as successively diverging lineages, but with weak support for the clade including Stipeae-Ampelodesmeae and the rest of Pooideae (Romaschenko et al. 2012).
Diarrheneae and Brachypodieae are resolved as the next successively diverging lineages in most analyses,  (Mathews et al. 2000), the two lineages comprised a polytomy with the rest of the core Pooideae (Schneider et al. 2011) and Diarrheneae were part of an unresolved polytomy with Stipeae, Meliceae, Phaenospermateae, Ampelodesmeae and the core Pooideae [Blaner et al. 2014 (based on sequences of the nuclear gene topoisomerase 6)]. In one ITS analysis, Diarrheneae were weakly supported as the sister group of the rest of Pooideae (Hsiao et al. 1998). Among the core Pooideae, Triticeae and Brachypodium are estimated to have diverged 32 -39 Mya, based on comparisons of their nuclear genomes (The International Brachypodium Initiative 2010); this represents the origin of the crown core Pooideae. An independent estimate places the origin of the Triticeae -Poeae split at 26 -33.5 Mya (Sandve and Fjellheim 2010). There is consensus that Triticeae and Bromeae are sister taxa (Catalá n et al. 1997;Hsiao et al. 1998;Davis 1998, 2000;Mathews et al. 2000;Grass Phylogeny Working Group 2001;Soreng 2007, 2010;Dö ring et al. 2007;Duvall et al. 2007;Bouchenak-Khelladi et al. 2008;Schneider et al. 2009Schneider et al. , 2011Grass Phylogeny Working Group II 2012;Blaner et al. 2014) and all studies that have sampled Littledalea have found this genus to be the sister group of the Bromeae -Triticeae clade Soreng 2007, 2010;Dö ring et al. 2007;Schneider et al. 2009Schneider et al. , 2011Blaner et al. 2014), with the exception of one in which cloned topoisomerase 6 sequences of Littledalea tibetica formed a polytomy with the Bromeae -Triticeae clade (Blaner et al. 2014).

Grass plastomes
The economically important grasses were among the first organisms to have their plastid genomes sequenced. The plastome of rice (Hiratsuka et al. 1989) was the third species, after tobacco and a liverwort, and the first monocot completed; and the maize plastome (Maier et al. 1995) was the sixth one completed (Jansen et al. 2005). There are now (as of 25 September 2014) some 106 plastid genomes publicly available for grasses, representing 43 genera,78 species and 8 subfamilies: Anomochlooideae (1 species) (Givnish et al. 2010;Morris and Duvall 2010;Jones et al. 2014), Pharoideae (2 species) (Jones et al. 2014), Puelioideae (1 species) (Jones et al. 2014), Bambusoideae (33 species) (Wu et al. 2009;Zhang et al. 2011;Burke et al. 2012Burke et al. , 2014Wu and Ge 2012;Gao and Gao 2014;Ma et al. 2014), Ehrhartoideae (7 species) (Hiratsuka et al. 1989;Shahid Masood et al. 2004;Tang et al. 2004;Wu and Ge 2012;Lin et al. 2014), Panicoideae (7 species) (Maier et al. 1995;Asano et al. 2004;Calsa Jú nior et al. 2004;Saski et al. 2007;Diekmann et al. 2009;Leseberg and Duvall 2009;Young et al. 2011;Besnard et al. 2013), Pooideae (48 species) (Ogihara et al. 2000;Saski et al. 2007;Diekmann et al. 2009;  Several studies have generated whole plastomes for grasses, and phylogenetic analyses of these plastomes have resulted in increased resolution and support for relationships within and among some grass subfamilies compared with trees in earlier single-and multi-gene studies. For example, Jones et al. (2014) found the twogenus subfamily Anomochlooideae to be monophyletic, a result found in some (Clark et al. 1995;Grass Phylogeny Working Group 2001;Duvall et al. 2007; Grass Phylogeny Working Group II 2012) but not all Zhang 2000;Bouchenak-Khelladi et al. 2008;Blaner et al. 2014 (matK analyses)] previous analyses, and they reconstructed species-level relationships in Pharoideae. Several plastome studies support a sister group relationship between Bambusoideae and Pooideae (Zhang et al. 2011;Wu and Ge 2012;Burke et al. 2014;Jones et al. 2014-but see the ML tree in Young et al. 2011) and in Bambusoideae plastomes have substantially improved resolution and support within and among species of the temperate woody bamboo tribe Arundinarieae (Zhang et al. 2011;Ma et al. 2014) compared with earlier multi-locus plastid phylogenies (Triplett and Clark 2010;Zeng et al. 2010). In Triticeae, plastomes have clarified relationships among closely related species of wheat (Triticum) and goatgrass (Aegilops) (Gornicki et al. 2014;Middleton et al. 2014). These examples indicate that whole plastomes hold much promise for resolving relationships among grass clades that have previously been problematic. Here, we report 25 new plastomes of taxa of Pooideae and use these in combination with previously published plastomes to infer phylogenetic relationships among the major lineages in the subfamily.

Taxon sampling and DNA extraction
Silica-dried leaf tissue was obtained from 25 species of pooid grasses (Table 2). Tissue was homogenized manually in liquid nitrogen before extraction. The DNA extraction protocol was followed using the Qiagen DNeasy Plant Mini Kit (Qiagen Inc., Valencia, CA, USA).

Plastome sequencing
Two micrograms of total genomic DNA from Brachyelytrum aristosum, Phaenosperma globosum, Piptochaetium avenaceum, Melica mutica and Diarrhena obovata were used in each library preparation. Libraries were prepared using the TruSeq Nano DNA sample preparation kit (Illumina, San Diego, CA, USA) and sequenced paired-end at Cold Spring Harbor Laboratory, Cold Spring, NY, USA. Libraries were prepared for the remaining taxa using the TruSeq and Nextera library preparation kits (Table 2). Detailed protocols for TruSeq, Nextera and TruSeq Nano are provided in Wysocki et al. (2014), Burke et al. (2014) and Barrett et al. (2013Barrett et al. ( , 2014, respectively. All Illumina-sequenced reads were first quality filtered using DynamicTrim v2.1 from the SolexaQA software package (Cox et al. 2010) with default settings, and then sequences ,25 bp in length (default setting) were removed with LengthSort v2.1 in the same package.

Plastome assembly, annotation and alignment
Plastome assembly was performed entirely de novo. The Velvet software package (Zerbino and Birney 2008) was run iteratively following the methods from Wysocki et al. (2014). Contigs were scaffolded using the anchored conserved region extension method (Wysocki et al. 2014). Any remaining gaps in the plastomes were repaired by locating overlapping regions of 20 bp or higher using contigs or raw reads until the circular map was complete. Fully assembled plastomes were annotated by aligning to a previously published and annotated reference plastome in Geneious Pro (Biomatters Ltd, Auckland, New Zealand) and copying the reference annotations to the assembled plastome when the annotation shared a minimum similarity of 70 %. The banked plastome from Lolium perenne (NC009950) was used as an annotation reference.
Plastomes were arranged with the large single-copy (LSC) region followed by the inverted repeat region B (IRb) and ending with the short single-copy (SSC) region. Inverted repeat region A (IRa) was omitted from the matrix to prevent overrepresentation of the inverted repeat sequence. All newly assembled plastomes were then aligned, along with 20 previously published pooid plastomes and one bamboo outgroup plastome, Bambusa bambos (KJ870988) ( Table 2), using the MAFFT alignment software (Katoh et al. 2005). We used the corrected sequence of Triticum aestivum (Bahieldin et al. 2014), as the earlier plastome (Ogihara et al. 2000) contained sequences from the rice plastome. The alignment was then inspected for structural mutations and adjusted manually to preserve tandem repeat boundaries and to identify inversions. Regions that contained inversion mutations were deleted from the matrix to remove false homology inferences. The alignment file may be obtained from TreeBase (http://purl.org/phylo/treebase/ phylows/study/TB2:S16741).

Indels
Indel mutations were scored in the plastome alignment as in Leseberg and Duvall (2009). We scored indels that  Table 2. Voucher specimen information and GenBank accession numbers for newly sequenced plastomes, place of publication of previously published plastomes and Illumina library preparation methods (TruSeq, Nextera or TruSeq Nano) used to produce reads of the newly sequenced taxa. The tribal/subtribal classification follows Soreng et al. (2014).   TruSeq (i) were ≥2 bp long, excluding mono-nucleotide repeats regardless of length; (ii) could be attributed to slippedstrand mispairing, identified by the presence of a perfect or near-perfect repeated sequence; (iii) were unambiguous and did not overlap with other indels in the alignment; and (iv) were present in two or more individuals (i.e. autapomorphic indels were not scored). These criteria allowed us to focus on indels that could be interpreted as single evolutionary events. Indels were not included in our analyses. Based on the topology of the ML tree, scored indels were interpreted as representing putative synapomorphies or putative homoplasy.

Phylogeny estimation
We conducted maximum parsimony (MP), ML and Bayesian inference (BI) analyses with two different taxon sets and two subsets of the data. The first taxon set included 45 of the 46 taxa in our matrix, excluding a previously published plastome of Schedonorus arundinaceus that has been reported as having some sequencing errors (Hand et al. 2013), and the second included this plastome (46 taxa). For each of these matrices, we conducted analyses of complete plastomes (non-coding and protein coding) and protein-coding sequence (cds). The procedures, analyses and parameters noted below were used for all analyses. To exclude the potential alignment ambiguity from the analyses, nucleotide positions that contained one or more gaps introduced by the alignment were omitted from the matrix. The Akaike Information Criterion (AIC) was used in jModelTest v 2.1.3 (Guindon and Gascuel 2003;Darriba et al. 2012) to compare models of character evolution in each of the gap-free matrices. The General Time Reversible (GTR) model of substitution incorporating invariant sites and a g distribution (GTR + I + G) was among the best-fit models and was used in subsequent analyses. Maximum likelihood analysis was performed using RAxML v 8.0.5 (Stamatakis 2006) with 1000 nonparametric bootstrap replicates. Non-parametric bootstrap values were generated using the Consense function in Phylip (Felsenstein 2005). MrBayes 3.2.2 (Ronquist and Huelsenbeck 2003) was used to perform the BI analyses. A Dirichlet prior was used for base frequencies and the rate matrix, and a uniform prior was used for the shape parameter (a), proportion of invariable sites (I) and topology; these are the default prior settings. Branch lengths were unconstrained and a GTR + I + G model was used with four discrete rate categories. The Metropolis-coupled Markov chain Monte Carlo (MCMCMC) search/sampling analysis was run for 2 × 10 000 000 generations with four chains. Average standard deviation of split frequencies remained ,0.001 after the 50 % burn-in. Heuristic parsimony [1000 random addition sequence (RAS) replicates; TBR branch swapping; best trees kept] and parsimony bootstrap analysis with 1000 pseudoreplicates, 10 RAS each, were performed with PAUP* v4.0b10 (Swofford 2003). All results are presented on ML phylograms.

Testing outgroup selection effects
The set of 45 pooid plastomes was analysed with a randomized set of outgroup taxa to test the effect of outgroup selection on the tree topology. Outgroup taxa were selected from a pool of previously published plastome sequences from Bambusoideae [Acidosasa purpurea (NC006084) and Zea mays (NC001666)]. A representative species from each of the three lineages was chosen randomly using a custom Python script (available on request), and aligned with the pooid plastomes with the MAFFT method. Phylogeny was estimated in an ML framework using RAxML. This was repeated 16 times and the tree with the highest likelihood from each iteration was tested for congruence using the Consense function of the Phylip software package.

Plastome sequencing
Complete plastomes were newly sequenced for 25 pooid grass species. Plastome lengths ranged from 134 287 to 137 897 bp. All lengths of plastomes and their sub-regions are reported in Table 3

Phylogeny
The 45-taxon MP analyses of the complete and protein cds resulted in one and two equally most parsimonious trees, respectively. In the 45-taxon analysis of the complete data, all but four nodes in the ML tree and three nodes in the MP tree are maximally supported (BS ¼ 100 %) and all nodes in the BI tree have posterior probabilities of 1 (Fig. 1). In the 45-taxon analysis of protein cds, all but four nodes in the ML tree and three nodes in the BI tree were maximally supported (Fig. 2). In the MP tree, five nodes received support between 50 and 99 %, and no topology among Helictochloa, Dactylis, Deschampsia and Loliinae received .50 % BS (Fig. 2). Support is lower in the protein cds analyses compared with the complete analyses at nodes that are not maximally supported in both. The topologies of the ML and BI trees are identical in the 45-taxon analyses of both data subsets. The parsimony trees differ in the relative placements of four subtribes (two clades) in one major clade of tribe Poeae, and two to three subtribes in a second major clade of the tribe. Bootstrap support in parsimony trees is generally lower than BS in ML trees at nodes that are not maximally supported in both.
In the 46-taxon analysis of the complete data, all but three nodes in the ML tree and four nodes in the MP tree are maximally supported, and all nodes in the BI tree are maximally supported [see Supporting Information]. In the 46-taxon analysis of protein cds, all but six nodes in the ML tree, seven nodes in the MP tree and four nodes in the BI tree are maximally supported. Support for most nodes is lower in the trees based on protein cds tree compared with the complete plastome data. The remainder of the text focusses on the 45-taxon dataset, unless indicated otherwise.
Ampelodesmeae is nested in Stipeae, and henceforth we refer to this lineage as the Ampelodesmeae -Stipeae clade. Within this clade, Achnatherum and Piptochaetium are successive sisters to an Oryzopsis -Ampelodesmos clade.
Within the core Pooideae, Bromeae and Triticeae are sister taxa, and this clade is sister to Poeae. Within Triticeae, Secale is the sister group of an Aegilops -Triticum clade in which neither Aegilops nor Triticum is monophyletic, and Hordeum is the sister group to this larger   'NR' indicates nodes not resolved or supported above the 50 % bootstrap level in one of the analyses. Crosses indicate clades that were reversed in the MP tree. The MP bootstrap value of the node affected by the reversal of clades reflects that of the reversed topology. The histogram indicates total invariant (stippled) and variable (solid) nucleotide sites in the full plastome analysis (Fig. 1) Fig. 2) supported as the successive sister groups of the rest of the clade, whereas in the MP tree the branching order of these two lineages is reversed and weakly (MPBS ¼ 69 %; Fig. 1) to strongly (MPBS ¼ 91 %; Fig. 2) supported. In Poeae clade 2, Coleanthinae and Poinae are sister taxa, and a second subclade comprises Airinae, Holcinae, Dactylidinae and Loliinae. In ML and BI complete data trees Airinae and Holcinae are strongly supported successive sister groups of the rest of the subclade, whereas in the MP tree Airinae and Holcinae are an unsupported clade (Fig. 1); and relationships at the base of this clade are unresolved in the MP analysis of protein cds (Fig. 2). In ML and BI analyses, Dactylidinae and Loliinae are sister taxa; this relationship is strongly supported in analyses of the complete data but poorly supported in the protein-coding analyses.

Unique plastome features
The plastome from Brachyelytrum aristosum shares a 196 bp insertion with Bambusa bambos in the rps16-trnQ intergenic spacer. The two insertions exhibit 85 % identity. This insertion is not present in any other genera of Pooideae but is represented in all subfamilies of grasses except for Danthonioideae and Micrairoideae. Phaenosperma globosum and both species of Melica share a 198 bp insertion in the same region that shows no significant sequence similarity to the insertion shared by Bambusa bambos and Brachyelytrum aristosum. The insertions found in Melica and P. globosum are not found in any other species of Pooideae included here, but a BLAST query reveals similar insertions in one species of Puelioideae (Puelia olyriformis), an 'early-diverging' member of the Panicoideae (Thysanolaena latifolia) and many members of the temperate bamboo lineage Arundinarieae. Two undocumented putative insertions of non-plastid homology were found in two previously published plastomes (Fig. 3). The full plastome for Triticum urartu contains a 1060 bp insertion in the region of the inverted repeat that ordinarily contains the coding sequence for the rpl23 gene. A BLAST query of this sequence against GenBank shows the region to have homology to various nuclear genes within multiple species of Triticum. Triticum monococcum contains a 1077 bp insertion in the rpl23-ndhB intergenic spacer of the inverted repeat region. A BLAST query of this sequence against GenBank shows the region to have homology to the Triticum timopheevii mitochondrial genome and many other grass mitochondrial regions. The top hit was located in the rrn26-1 region of the T. timopheevii mitochondrial genome, covered nearly 100 % of plastome insertion and exhibited over 99 % identity.

Indels
A total of 177 indels were scored that we interpreted as single evolutionary events [see Supporting Information]. Eighty-four of these could be attributed to slipped-strand mispairing. All of the scored indels could be straightforwardly interpreted as being putatively synapomorphic or putatively homoplastic. Given the current level of taxon sampling, 34 indels were putatively homoplastic with respect to the topology of the ML tree, while the remaining 143 represented putative synapomorphies for clades of various composition, ranging from multi-tribal Figure 3. Two putative insertions of nuclear and mitochondrial homology located in IR regions of the Triticum urartu (NC_021762) and T. monococcum (NC_021762) plastomes, respectively, are shown here and compared with the typical Hordeum jubatum plastome (KM974741). The insertion in the T. urartu plastome exhibits 90 % identity to chromosome 3B of the T. aestivum nuclear genome (HG670306) and the insertion in the T. monococcum plastome exhibits 99 % identity to the T. timopheevii mitochondrial genome (AP013106). Gene orientation is indicated and thin lines denote regions in which gaps were introduced to preserve the alignment. Gene position is relative and lengths are not to scale.
AoB PLANTS www.aobplants.oxfordjournals.org clades to those of a single species (in cases where more than one individual per species was sampled) ( Table 4). One indel was putatively synapomorphic for the large clade comprising Diarrheneae plus the core Pooideae; 2 indels each were putatively synapomorphic for a threetribe and a four-tribe lineage; 30 indels were putatively synapomorphic for 6 two-tribe lineages, with 20 of these supporting the Bromeae -Triticeae clade; 37 indels were putatively synapomorphic for 8 tribes, with 13 of these supporting Meliceae and 7 supporting Triticeae and 9 indels were putatively synapomorphic for three species. Within Triticeae, 7 indels represented putative synapomorphies for clades of subsets of Aegilops and Triticum species, 4 were putatively synapomorphic for an Aegilops -Triticum clade, 10 were putatively synapomorphic for an Aegilops-Triticum -Secale clade and 6 were putatively synapomorphic for Hordeum. Within Loliinae 4 indels were putatively synapomorphies for a clade of all sampled taxa except Festuca ovina, and 14 were putative synapomorphies for a Lolium-Schedonorus clade. Seven and ten indels were putative synapomorphies for Poeae clades 1 and 2, respectively. A total of 111 indels are located in intergenic spacer regions, 33 in introns and 14 in protein-coding regions (genes).

Discussion
Phylogenomic analyses of 45 whole plastomes, including 25 newly sequenced here, resulted in a highly resolved and strongly supported phylogeny of Pooideae, with the caveat that whole plastomes for three tribes (Nardeae, Lygeae, Brylkineae) and Littledalea are not yet sampled. The few topological differences between the MP, ML and BI trees are in tribe Poeae and at the base of the core Pooideae. Each of the discordant inferred clades is subtended by a very short branch. As most branches in the Poeae clade define genera and are fairly long, the topological differences observed may be a reflection of long-branch attraction (LBA) in the MP analysis due to the proximity of very short and very long branches in the tree. This is a common problem in phylogenetic inference and has been observed in other whole plastome phylogenomic studies (Soltis and Soltis 2004;Stefanović et al. 2004;Givnish et al. 2010;Barrett et al. 2014). Strategies to overcome LBA include using inference methods that are less prone to long branch effects and, therefore, may be more accurate (such as ML or BI inference, which we have done here); excluding third codon positions, which may be saturated or randomized; representing clades in analyses with only short-branched taxa by excluding taxa with long branches; adding taxa to break up large branches and adding data (Bergsten 2005). An alternative strategy to adding more data could involve data filtration. For example, quality measures of 'tree-likeness' for data partitions could be used to identify the 'data core.' Aside from LBA, factors that might lead to differences in  topology and/or support among data partitions include model mis-specification, intrinsic biases among the data (e.g. large differences in GC content among clades; Ruhfel et al. 2014) and heterotachy. For most tribes and subtribes we have sampled only one or two exemplar taxa, so removing taxa from the analyses would be a poor approach. In contrast, with rapidly decreasing costs for sequencing whole plastomes, there is ample opportunity to increase taxon sampling within and among tribes, particularly those that are species-rich, such as Poeae-the clade in which the incongruences among analyses were found and in which there are numerous unresolved phylogenetic questions. Given our current strategy of sampling whole plastomes, adding more data from this linkage group is impossible, by definition. Adding data from the mitochondrial or nuclear genomes would be beneficial for deciphering the true evolutionary history of taxa (i.e. the species tree), as genes from these genomes may have evolutionary histories that are different than those of maternally inherited plastid genomes, such as the nuclear gene history found by Triplett et al. (2014), but this would not help with the long-branch problem in the plastid data. Coalescent-based analyses of multiple unlinked nuclear genes could help accurately reconstruct species relationships, especially for parts of the plastome phylogeny that may be at odds with the genealogies of some or many nuclear genes. Because probability-based methods of phylogenetic inference are less prone to longbranch effects than parsimony, our discussion below is based primarily on the topologies of our nearly identical ML and BI trees, which also are generally better supported than the MP trees.
Previous phylogenomic studies of relationships among orders and families of plants have examined only proteincoding genes (e.g. Barrett et al. 2013Barrett et al. , 2014Davis et al. 2013;Kim and Kim 2013;Martin et al. 2013;Ruhfel et al. 2014), while those-including the current study-that have included non-coding data have generally focussed on more recent or 'shallow' phylogenetic relationships (Zhang et al. 2011;Burke et al. 2012;Hand et al. 2013;Gornicki et al. 2014;Huang et al. 2014;Ma et al. 2014;Middleton et al. 2014). Even though spacers/introns are 'non-coding' some plastid intergenic regions are expected to be conserved due to containing enzymebinding sites or important secondary structures (Peredo et al. 2012) and they may too be under purifying selection similar to coding loci, and thus useful in larger-scale phylogenetic reconstructions. Nevertheless, aligning noncoding regions (or a subset of them) across divergent taxa can be difficult and may introduce error into phylogenetic analyses when alignments are not accurate, and we have attempted to minimize the possibility of this type of error by excluding from analysis all positions in the alignment that include a gap in one or more taxon. The improvement in support in our analyses of complete plastomes versus protein cds indicates that the non-coding complement of the genome contributes important characters for resolving these mostly deep nodes in pooid grasses, similar to the results of other phylogenomic studies of grasses (Burke et al. 2012;Ma et al. 2014) and other plants (Eserman et al. 2014) in which analyses of complete plastomes and/or non-coding data and protein cds were explicitly compared. Inclusion of non-coding data in plastome phylogenomic analyses of such groups as Zingiberales (Barrett et al. 2014) may improve support levels at deep nodes that are weakly supported based on protein cds alone.

'Early-diverging' Pooideae
Among the tribes of Pooideae major uncertainties of relationship have revolved around the relative branching orders of Phaenospermateae, Brylkinieae -Meliceae and Ampelodesmeae -Stipeae following the respective divergences of Brachyelytrum and Lygeae -Nardeae. Plastomes are not yet available for Lygeum and Nardus. As expected, Brachyelytrum represents one of the lineages resulting from the deepest split in the subfamily, and we find strong support for Phaenospermateae, Brylkinieae-Meliceae and Ampelodesmeae -Stipeae to be the next successive sister groups to the rest of Pooideae. The branch subtending all Pooideae except Brachyelytrum is quite long, indicating a long period of time prior to the next divergence and/or a rapid rate of plastid evolution along this branch. Conversely, the next two deep branches at the base of the plastome trees identifying Phaenospermateae, Meliceae and Ampelodesmeae-Stipeae as successively diverging taxa are extremely short compared with most other deep branches in the tree subtending lineages recognized as tribes, possibly reflecting rapid radiation of these lineages. Given these short branches, it is not surprising that most previous studies were not able to robustly resolve the relationships among these taxa, particularly those based on one or a few gene regions (Catalá n et al. 1997;Mathews et al. 2000;Dö ring et al. 2007;Schneider et al. 2009;Davis and Soreng 2010;Blaner et al. 2014). Our results are congruent with the largest study, in terms of gene regions sampled, that sampled these lineages (Romaschenko et al. 2012), although not all nodes in that study were strongly supported as they are in our plastome trees. In contrast, our results are discordant with two-and three-gene studies that recovered strong support in BI analyses for the opposite order of divergence for Ampelodesmeae -Stipeae and Meliceae (Duvall et al. 2007;Bouchenak-Khelladi et al. 2008). The reasons for this are not clear, but may be due to differing AoB PLANTS www.aobplants.oxfordjournals.org levels of taxon sampling in and near this part of the tree, outgroup choice (although tests of different grass outgroups here had no effect on the ingroup topology) or varying phylogenetic signal in different partitions of the data [as demonstrated for monocots in Davis et al. (2013), for example]. Considering the latter possibility, support was much lower for the current topology in the MP analyses of protein-coding data compared with the complete dataset, while support in the ML and BI analyses was strong in both. Diarrheneae represent the next successive sister group to the rest of the subfamily (the core Pooideae) in most of our plastome trees, congruent with the results of other studies [Clark et al. 1995;Catalá n et al. 1997;Soreng 2007, 2010;Dö ring et al. 2007;Duvall et al. 2007 (Bayesian analysis)]. An alternative topology recovered only in the ML analysis of protein cds, in which Brachypodium and Diarrhena are a clade, was poorly supported. This topology was also found in Minaya et al. (2013), with strong support.

Core Pooideae
Brachypodieae. Accurate placement of Brachypodium in the context of the phylogeny of Pooideae is important, as B. distachyon has become a model species for the genomics of cereal crops, grass biofuels and forage grasses (The International Brachypodium Initiative 2010; Mochida and Shinozaki 2013). Relationships among the four sampled core tribes of Pooideae inferred by whole plastomes are consistent with most previous plastid trees, in which Brachypodieae is the sister group of a clade comprising a Bromeae -Triticeae lineage and Poeae (Catalá n et al. 1997;Duvall et al. 2007;Schneider et al. 2009;Davis and Soreng 2010;Grass Phylogeny Working Group II 2012;Blaner et al. 2014). This topology is also present in the nuclear topoisomerase 6 phylogeny of Blaner et al. (2014). Two other nuclear-based phylogenies, however, infer alternative highly supported topologies that are discordant with each other and with the plastome phylogeny. The phytochrome B tree of Mathews et al. (2000) identifies Diarrhena as the sister group to the core Pooideae (excluding Brachypodium), with Brachypodium sister to this clade, whereas the b-amylase tree of Minaya et al. (2013) identifies a strongly supported (Brachypodieae + Diarrheneae), [Stipeae, (Lygeae + Nardeae)] clade that is sister to the rest of the core Poeae. The combined nuclear (ITS, b-amylase) and plastid tree in Minaya et al. (2013), however, has the same topology as our plastome tree with respect to the relationships among the core Pooideae, with all the relevant nodes strongly supported. This level of support is surprising given the strongly discordant b-amylase tree, which was incongruent with the ITS tree in a partition homogeneity test, as were the nuclear and plastid data (Minaya et al. 2013). Unfortunately, the ITS and plastid trees were not reported in that study and we are therefore unable to compare them with the b-amylase tree.
Additional nuclear genes have been sequenced for Brachypodium and representatives of Triticeae and Meliceae in the context of understanding the evolutionary history of Brachypodium species, but these studies did not have sufficient sampling to address the affinities of Brachypodium with respect to other Pooideae (Wolny et al. 2011;Catalá n et al. 2012). Similarly, gene trees of nuclear loci coding for economically important traits, such as grain endosperm texture (Charles et al. 2009) have been studied, but these are based on data from the few sequenced grass genomes (rice, sorghum, wheat, Brachypodium), a level of taxon sampling that is insufficient for reconstructing finer-scale aspects of phylogeny. Although there is consensus on the position of Brachypodium in plastid-based phylogenies as the sister group of the rest of the core Pooideae, additional nuclearbased phylogenies with dense sampling are needed to distinguish gene trees from the species tree to characterize the precise affinities of Brachypodium and other pooid grasses, given the discordances among existing nuclear gene trees.
In the current study, the single plastome representative of Brachypodieae, the annual species Brachypodium distachyon, lies on a very long branch and is subtended by a very short one. Although the topologies are the same in five of the six MP, ML and BI trees, the branch length may be distorting the analyses. Available plastid data indicate that B. distachyon has the highest substitution rate in the genus. For example, within the crown Brachypodium clade in an ndhF phylogeny B. distachyon lies on a branch two and a half to eight times the length of the terminal branches of other taxa in the genus, which are perennials (Catalá n et al. 1997). Rate variation between annual and perennial plant taxa has been widely documented (e.g. Yue et al. 2010;Gaut et al. 2011) and was recently demonstrated in grasses in plastome phylogenies of woody vs. herbaceous bamboos (Burke et al. 2012. Until recently B. distachyon was considered to be the only annual species in its genus, but the taxon has now been split into three annual species on the basis of morphological, cytogenetic and molecular data (Catalá n et al. 2012). The plastome included here was sequenced from the Bd21 (2n ¼ 10) genotype of Brachypodium (Bortiri et al. 2008) and corresponds to B. distachyon s.s. in the revised classification. Although an earlier phylogenetic analysis identified B. distachyon (s.l.) as the sister group of the rest of the genus (Catalá n and Olmstead 2000), a more detailed study identifies two perennial species and one of the newly described annuals (B. stacei) as representing the first (but poorly supported) splits in the genus, with B. distachyon s.s. and the other newly described annual, B. hybridum, sister to a clade of the remaining perennial taxa (Catalá n et al. 2012). Sampling plastomes from one of the core perennial species and one or more of the putative 'early-diverging' taxa would contribute to an understanding of plastome variation in the lineage and might break up the long branch in the crown Brachypodium clade, which may improve support in analyses of protein cds, and may possibly affect support levels elsewhere in the tree.
Poeae clade 1. As in previous plastid studies, the tribe Poeae is divided into two strongly supported clades, Poeae clades 1 and 2, that are further supported by 7 and ten unambiguous indels, respectively. The plastome phylogenies of Poeae clade 1 are maximally supported at all nodes in the ML and BI trees and at all but one node in the MP trees, and the topologies of the ML/BI and MP trees differ only in the relative placements of Aveninae and Phalaridinae -Torreyochloinae as the sister group of the rest of the clade. Our sampling represents all six subtribes that are part of Poeae clade 1 in Soreng et al. (2014). Brizinae is divided into two informal groups in the classification: Brizinae s.s., which includes the Old World genera Airopsis and Briza s.s., and Brizinae s.l. 'Calotheca clade', which includes the New World genera Chascolytrum s.l. (see Essi et al. 2011 for details on the recent expansion of Chascolytrum) and Relchela. This division is based on the results of phylogenetic work that shows Brizinae to be non-monophyletic. In an early study, Soreng and Davis (2000) found Brizinae to be paraphyletic and suggested that the subtribe represents parallel evolution of a 'brizoid' lemma (i.e. lemmas that are as long as broad) in Eurasia and South America. This paraphyly was supported by numerous subsequent phylogenetic studies: species of Chascolytrum s.l. and subtribe Agrostidinae form a clade, and Briza s.s. and Airopsis represent a distinct lineage that, in most analyses, is the sister group of the former clade Dö ring et al. 2007;Quintanar et al. 2007;Grass Phylogeny Working Group II 2012). Chascolytrum s.l. is anomalous morphologically in the Agrostidinae clade as it has multiple florets per spikelet, compared with a single floret per spikelet, which is the diagnostic character for the subtribe. The distinction between these Old World and New World Brizinae lineages is clearly shown in Essi et al. (2008), although that study unfortunately did not include any closely related non-Brizinae taxa and thus the two Brizinae lineages appear to be reciprocally monophyletic sister taxa, which may not be accurate. Although we have not sampled the Brizinae s.l. 'Calotheca clade' here, our plastome trees similarly identify Brizinae s.s. as the sister group of Agrostidinae, with improved support compared with most earlier studies. Characterizing relationships between the Brizinae s.l. 'Calotheca clade' (i.e. Chascolytrum s.l.) and Agrostidinae taxa will require further work. The former group has been resolved as monophyletic in some, but not all studies, and its affinities with Agrostidinae taxa have been variously resolved Quintanar et al. 2007;. The monotypic Relchela has only been sampled in one study, where in an ITS tree it falls in a clade of Agrostidinae taxa (Refulio-Rodriguez 2007). We have sampled only two (Agrostis, Ammophila) of the 16 genera recognized in Agrostidinae, and these form a strongly supported clade. Other studies have similarly identified Agrostidinae as a monophyletic group (often including Chascolytrum s.l., as noted above) (Soreng and Davis 2000;Davis and Soreng 2007;Dö ring et al. 2007;Quintanar et al. 2007;Minaya et al. 2013) but further research is needed to clarify generic concepts and relationships within the subtribe .
The strongly supported sister group relationship in our plastome trees between Anthoxanthinae and Brizinae -Agrostidinae has been found previously only in two plastid trees, in which this topology was weakly supported (Bouchenak-Khelladi et al. 2008;). This topology is incongruent with the placement of Anthoxanthinae in a strongly supported clade with Aveninae in other plastid, nuclear and combined trees Dö ring et al. 2007;Quintanar et al. 2007;Schneider et al. 2009;Minaya et al. 2013) or sister to Phalaris (Soreng and Davis 2000). The reasons for these varying topologies are unclear, but may be related to density of taxon sampling, variable signal in different data partitions and/or rates of evolution in Poeae clade 1.
The strongly supported sister group relationship between Phalaridinae and Torreyochloinae in our plastome trees has not been identified previously. This may be because both lineages have been sampled together in only a few analyses, at least when considering the current circumscription of Phalaridinae. This subtribe has traditionally been circumscribed as including two or three genera, the closely related Anthoxanthum and Hierochloe (often included in Anthoxanthum) and Phalaris, which have a similar and unique spikelet structure in Poeae (Clayton and Renvoize 1986;). However, this circumscription has not been supported by most molecular studies-including the current one-which have identified AoB PLANTS www.aobplants.oxfordjournals.org Phalaris and Anthoxanthum s.l. as separate lineages (Dö ring et al. 2007;Quintanar et al. 2007;Bouchenak-Khelladi et al. 2008;Gillespie et al. 2008;Minaya et al. 2013), hence their current recognition in monotypic subtribes. Within Anthoxanthinae, a molecular study has demonstrated the need for recognition of one or three genera . The shared spikelet characteristics of Anthoxanthinae and Phalaridinae may be plesiomorphies or may have evolved in parallel. Some studies represented Phalaridinae solely by Anthoxanthum (e.g. Soreng and Davis 2000), which in retrospect is uninformative in shedding light on the origins of Phalaridinae as now understood (since Anthoxanthum is part of a separate lineage). Torreyochloinae includes two genera, Amphibromus and Torreyochloa, a circumscription based on plastid and nuclear ribosomal phylogenies (Soreng and Davis 2000;Davis and Soreng 2007;. One set of studies that sampled all three genera of Phalaridinae and Torreyochloinae was based on restriction site and morphological characters and found Phalaris, Amphibromus and Torreyochloa to be part of a clade of Aveneae taxa, but none of these three taxa was closely related Davis 1998, 2000). In -so far the only study based on DNA sequence data to sample both subtribes and all three genera-all deep branches in plastid and nuclear ribosomal trees were weakly supported and/or unresolved. Although not incongruent with our current results, the unsupported trees in  do not provide support for the lineage.
Numerous studies have sampled either Phalaridinae or Torreyochloinae. In those that sampled only Phalaris, its placement was either unresolved (Dö ring et al. 2007) or variously inferred to be the sister group of the rest of Poeae clade 1 (Quintanar et al. 2007;Bouchenak-Khelladi et al. 2008Gillespie et al. 2008;Schaefer et al. 2011), Agrostidinae (Minaya et al. 2013) or Aveninae (Grass Phylogeny Working Group II 2012). Studies that sampled only Torreyochloinae identified it as the sister group of the rest of Poeae clade 1 (Soreng and Davis 2000;Davis and Soreng 2007; or Agrostidinae plus Brizinae s.s. (Davis and Soreng 2010). The studies that placed Phalaridinae or Torreyochloinae as sister to the rest of Poeae clade 1 are consistent with our ML and BI trees, in which the Phalaridinae -Torreyochloinae clade is strongly supported (at least based on complete plastome data) as the sister group of the rest of the clade. However, those previous studies and our ML and BI trees contrast with the topology of our MP trees, in which the Aveninae clade is weakly (complete plastome data) to strongly (protein cds) supported as the sister group of the rest of the clade. This latter set of relationships has not been found in other studies and likely represents systematic error in the MP tree here given the extremely short internal branches and long terminal branches at the base of the clade.
The novel relationship identified between Phalaridinae and Torreyochloinae requires consideration of possible non-molecular synapomorphies for this lineage. In terms of their gross morphologies, Phalaridinae and Torreyochloinae are distinctive, differing in their inflorescence shape, the number and fertility of florets per spikelet, glume length and the presence or absence of an awn. Phalaridinae is characterized by inflorescences of false spikes, spikelets laterally compressed with a single fertile floret and two proximal sterile lemmas that are shorter than the fertile lemma, glumes exceeding the florets, lemmas awnless and calluses glabrous (Voshell et al. 2011). Torreyochloinae is a morphologically heterogeneous subtribe. Torreyochloa is characterized by terminal paniculate inflorescences, spikelets laterally compressed to terete with 2 -8 florets, glumes rounded to slightly keeled, unawned and shorter than the lowest lemma, lemmas 5 -7-nerved (these prominent and scaberulous) and unawned and calluses glabrous (Watson and Dallwitz 1992 onwards;. It is one of several genera in Poeae clade 1 (also including genera of Brizinae s.l.) with spikelet structure characteristic of the traditional Poeae (i.e. simple spikelets with short glumes, several florets, 3 -5-nerved lemmas vs. long glumes, 1 -several florets, (3)5 -11-nerved lemmas and/or geniculate dorsal awns in traditional Aveneae; Clayton and Renvoize 1986), which are mostly part of Poeae clade 1 ). Its sister taxon, Amphibromus, is characterized by terminal paniculate inflorescences, spikelets laterally compressed with 2 -10( -12) fertile florets, glumes rounded to slightly keeled, unawned and shorter than or subequal to the lowest lemma, lemmas 2 -4-toothed with teeth extending into short bristles, lemmas dorsally awned from about the middle and calluses hairy (Watson and Dallwitz 1992 onwards;Weiller et al. 2009). The spikelet structure of Torreyochloa may be plesiomorphic in Poeae, or may be a result of convergent evolution in Poeae clades 1 and 2 (Soreng et al. 1990;Soreng and Davis 2000), but a possible hybrid origin for this taxon should not be ruled out without supporting evidence. We are not able to identify any putative synapomorphies for the Phalaridinae -Torreyochloinae clade.
Poeae clade 2. Poeae clade 2 is strongly supported in our analyses, and comprises two major subclades based on the current sampling. One strongly supported subclade includes Dactylidinae, Holcinae, Loliinae and Airinae, and a sister relationship between Dactylidinae and Loliinae is supported in most analyses. In the ML and BI analyses, Airinae and Holcinae are identified as the successive sister groups of the rest of the subclade, whereas in one MP tree (complete data) these taxa are recovered as clade but with no support (BP ,50), and in the other (protein cds) their relationships are unresolved. As in Poeae clade 1, these discordant topologies are likely a function of the very short branches at the base of the subclade. Recovery of this four-tribe subclade is consistent with previous studies that identified a larger clade comprising these subtribes as well as Ammochloinae, Cynosurinae, Parapholiinae and Sesleriinae, based on one to three plastid regions Davis and Soreng 2007;Dö ring et al. 2007;Quintanar et al. 2007;Bouchenak-Khelladi et al. 2008;Schneider et al. 2009Schneider et al. , 2012Schaefer et al. 2011;Grass Phylogeny Working Group II 2012) and combined plastid (four regions) and nuclear (ITS and b-amylase) data (Minaya et al. 2013).
To properly interpret the results of earlier studies in the context of the current subtribal classification of Poeae, the recently revised nomenclature of some previously sampled taxa in light of knowledge of their evolutionary histories must be considered. Some species treated as Avenula (A. albinervis, A. compressa, A. gervaisii, A. hookeri, A. pratensis, A. sulcata) and Helictotrichon (H. bromoides, H. schellianum) are now recognized in the genus Helictochloa (Romero-Zarco 2011), which is included in Airinae, and Deschampsia flexuosa is now recognized in Avenella (Chiapella 2007), also included in Airinae. Only Avenula pubescens remains in Avenula s.s. (Romero-Zarco 2011), and it combines characters of Helictotrichon and Helictochloa, but its plastid is phylogenetically isolated from any of the above, possibly aligning with Poinae s.l. Soreng and Davis (2000), as yet unaware of the nrDNA problem, commented on the intermediate nature of Avenula pubescens between Avenula subgen. Pratavenastrum (now Helictochloa) and Helictotrichon. Helictochloa and Deschampsia are apparently part of a set of taxa (also including Avenella, Sesleria, Scolochloa and others) possibly derived from reticulation between plastid-based Poeae clade 1 and Poeae clade 2, or involved in that event or events. They and others align with traditional Poeae in plastid analysis (i.e. Poeae clade 2) and are intermediate between those and taxa with traditional Aveneae-type plastids (i.e. Poeae clade 1) in nrDNA trees (in nrDNA trees the Aveneae taxa arise from within Poeae, and the placements of various genera are quite incongruent) (e.g. Quintanar et al. 2007). This is a complex area that needs further study.
In earlier plastid trees, relationships among the lineages in the subclade were unresolved and/or weakly supported, with the exception of a sister group relationship inferred between Cynosurinae and Parapholiinae Soreng and Gillespie 2007;Schneider et al. 2012;Minaya et al. 2013), and are not in conflict with the set of relationships inferred here. The combined nuclear and plastid tree in Minaya et al. (2013) is much better resolved, identifying the following strongly supported topology for this subclade: {Airinae (Helictochloa bromoides, as Avenula bromoides) [Dactylidinae (Cynosurinae, Parapholiinae)]}, and identifying a second clade comprising Holcinae, Airinae (Avenella flexuosa, as Deschampsia flexuosa, and Corynephorus) and Sesleriinae. Sessleriinae falls within Poeae clade 2 in plastid analyses, but is nested within Poeae clade 1 taxa in nrDNA analyses. Sesleria has a strange morphology, including a bract below the inflorescence and oddly shaped spikelets, that are suggestive of some disruption of the developmental patterns, possibly resulting from its reticulate origin (R. J. Soreng, pers. obs). The placement of Dactylidinae in the combined nuclear and plastid tree of Minaya et al. (2013) contrasts with its strongly supported placements as sister to Loliinae in our ML and BI trees based on the complete plastome data. We are not able to compare our plastome tree with their plastid tree, as it was not shown.
In earlier studies, two subtribes in the subclade were not monophyletic with respect to their current circumscriptions: Airinae, with Aira, Avenella, Corynephorus and Periballia comprising a lineage separate from Helictochloa Dö ring et al. 2007;Quintanar et al. 2007;Schaefer et al. 2011;Minaya et al. 2013); and Holcinae, with a Holcus -Vahlodea clade and Deschampsia s.s. representing separate lineages Dö ring et al. 2007;Quintanar et al. 2007;). We do not yet have sufficient plastome sampling to address the monophyly of these subtribes, having sampled only a single exemplar from each. A first strategy for future work should be to obtain plastomes from the currently unsampled subtribes Ammochloinae, Cynosurinae, Parapholiinae and Sesleriinae, and each of the putative lineages representing the non-monophyletic Airinae and Holcinae (at least two of these are currently unsampled). Further sampling in Helictochloa (some 30 species) and Deschampsia s.s. should aim to maximize the phylogenetic diversity in these genera as characterized in recent studies (Chiapella 2007;Winterfeld et al. 2014).
The second major subclade in Poeae clade 2 comprises representatives of Poinae and Coleanthinae. Like the other subclade, this major lineage has been identified in plastid trees in numerous studies Gillespie et al. 2007;Quintanar et al. 2007;Bouchenak-Khelladi et al. 2008;Schaefer et al. 2011;Grass Phylogeny Working Group II 2012;Schneider et al. 2012). Several subtribes that were recognized recently , including Phleinae, Cinninae, Alopecurinae and Beckmanniinae, are now included in a more broadly defined subtribe Poinae on the basis of phylogenetic data, although numerous aspects of deep relationship in this large subtribe remain unresolved for plastid data (Gillespie et al. , 2008(Gillespie et al. , 2010. The maximally supported pattern of relationships in our plastome trees, with Coleanthinae identified as the sister group of a Phleum -Poa clade (i.e. Poinae), agrees with the earlier plastid trees of Gillespie et al. (2007Gillespie et al. ( , 2008. The affinities of the monotypic subtribe Miliinae, not sampled here, remain somewhat unclear and are discordant in plastid and nuclear ribosomal trees. In plastid trees the tribe lies on a very long branch, is weakly allied with Phleum, and the Phleum -Miliinae lineage is identified as the sister group of Poinae, but with poor support (Gillespie et al. 2008). In contrast, in nuclear ribosomal trees a weakly supported Phleum -Miliinae lineage falls outside the Poinae clade, with unclear affinities to a clade of Poinae taxa excluding Poa, a Poa clade and Coleanthinae (Gillespie et al. 2008(Gillespie et al. , 2010. In combined plastid and nuclear analyses a weakly supported Phleum -Miliinae lineage is the sister group of a Poa clade (Gillespie et al. 2010), but this may not be an accurate reflection of evolutionary history given the discordance between the plastid and nuclear ribosomal data partitions. In a nuclear b-amylase tree, Milium is sister to a clade that includes Poinae plus a paraphyletic Coleanthinae, whereas in a combined nuclear (ITS, b-amylase) and plastid tree Milium is sister to Poinae, and the Milium-Poinae clade is sister to Coleanthinae (Minaya et al. 2013). Although plastome data for Miliinae may help clarify its maternal affinities with respect to Phleum and the rest of the Poinae clade, additional nuclear data will be necessary to reconstruct its possible hybrid origins.
Within Loliinae the plastome data indicate that Lolium is nested within a paraphyletic Schedonorus, that Festuca altissima (also known as Schedonorus altissimus) is more closely related to Lolium -Schedonorus (these being part of a 'broad-leaved' clade of Festuca s.l.) than F. ovina ('fine-leaved' clade), and that recognition of Schedonorus and Lolium as genera renders Festuca paraphyletic. All of these findings agree with the results of previous studies (Torrecilla and Catalá n 2002;Catalá n et al. 2004;Inda et al. 2008Inda et al. , 2014Hand et al. 2010Hand et al. , 2013. In Soreng et al. (2014), Schedonorus is treated as a synonym of Lolium, based on the phylogeny in Catalá n et al. (2009) and consistent with the plastome phylogeny here. The Lolium -Schedonorus clade lies on a fairly long branch, and is further supported by 14 unambiguous indels. Schedonorus arundinaceus is a hexaploid species complex comprising three morphologically and physiologically distinct forms recognized as Continental, Mediterranean and rhizomatous (Hand et al. 2010). Hand et al. (2010) sampled each of these forms and related species in the clade to reconstruct the evolutionary history of this agriculturally important pasture grass, and found each of the forms to have different origins. In their plastid tree, the rhizomatous and Continental forms were part of a clade that is the sister group of the clade including Lolium and S. pratensis, with the Mediterranean form placed elsewhere. Schedonorus arundinaceus was similarly not monophyletic in their nuclear ribosomal or other nuclear trees, and some aspects of their interrelationships differed with respect to the plastid tree.
Two complete plastomes are now available for Schedonorus arundinaceus, one of which was sequenced here. The two accessions of this species analysed in our 46-taxon trees are not resolved as a clade in either of the MP analyses [see Supporting Information], as they are in the ML and BI trees; this is likely a function of the extremely long branch of the accession sequenced by Cahoon et al. (2010), rather than the possibility that the accessions may represent different forms. The previously published plastome and our new one for the species both represent the Contintental form-the most common form in North America. Cahoon et al. (2010) made this identification explicit, as they sampled 'KY31' [Kentucky 31], a widespread cultivar of the Continental form. Our plastome is from a field-collected specimen from British Columbia, which we identify as the Continental form based on its lack of rhizomes, the main diagnostic character for differentiating these morphotypes (Hand et al. 2010). This identification is consistent with BLAST comparisons of matK variation among our sample and those of the Continental and rhizomatous forms sequenced by Hand et al. (2010). The matK sequences of these samples differ by one substitution that varies within and among the two forms, compared with the more divergent matK sequences in the Mediterranean form, as illustrated in the matK tree in Hand et al. (2010). Although the BLAST search does not unambiguously identify our sample as either the Contintental or rhizomatous form, it does rule out the possibility of it being the Mediterranean form, at least based on current knowledge of variation in matK in that form. Hand et al. (2013) included the Schedonorus arundinaceus plastome generated by Cahoon et al. (2010) in an analysis with four Festuca, Lolium and Schedonorus plastomes they generated, and observed considerable divergence among S. arundinaceus and the other Loliinae taxa, as we do here. They suggested that this divergence may be due to sequencing errors in the S. arundinaceus genome, and that additional plastomes should be generated from the species to determine whether the observed variation is real or artefactual. The previously sequenced plastome for S. arundinaceus shows 2.7 % difference from the new plastome from this taxon sequenced here. The top hits from a BLAST query of the previously sequenced regions rpoC2 and ccsA, which contain numerous substitutions in the alignment, show the top hits to be the respective regions from species within the PACMAD clade (data not shown), suggestive of some errors in this genome. Variation in our new plastome from S. arundinaceus is more in line with the other three species in the Lolium -Schedonorus clade, further indicating problems with the earlier plastome. Inclusion of the long-branch S. arundinaceus genome differentially affects ML BS for the core Pooideae in the complete and protein-coding analyses compared with the 45-taxon analyses, for reasons that are unclear.
Bromeae -Triticeae. As in previous phylogenies, Bromeae and Triticeae are a strongly supported clade in our analyses, and the clade is further supported by 20 unambiguous indels. We have not sampled Littledalea, which is the sister group of Bromeae -Triticeae. The pattern of relationships among the three sampled Triticeae genera agree with plastid (Petersen and Seberg 1997;Mason-Gamer et al. 2002), nuclear (Mason-Gamer 2005 and combined plastid/nuclear phylogenies (Escobar et al. 2011) of the subtribe. An earlier version of the T. aestivum plastome (Ogihara et al. 2000) was recently found to contain sequence from the rice plastid genome (Bahieldin et al. 2014). This contaminated genome was included in an early version of our plastome matrix, and in our preliminary MP and ML analyses the terminal branch of T. aestivum was considerably longer than those of the other Aegilops and Triticum taxa in the trees, indicative of this error (data not shown). In the current matrix containing the corrected T. aestivum plastome sequence, the terminal branch for this taxon is very similar in length to those of all other taxa in the clade. Plastomes for the Aegilops and other Triticum species (except T. aestivum) were sequenced by Middleton et al. (2014). Their phylogeny depicting relationships between wheat, rye and barley was based on a 37 kb subset (,30 %) of the whole plastome. We included the full plastomes for these species in our analyses, and find the relationships among Triticeae taxa to be maximally supported and identical to those reported by Middleton et al. (2014).

Unique plastome features
Within Pooideae the presence of the rps16-trnQ insertion solely in Brachyelytrum suggests that the loss of this insertion may be synapomorphic for the remaining genera. The insertion in the rps16-trnQ region of Phaenosperma and Melica is of note as, after Brachyelytrum, these two genera are the earliest to diverge in our analyses. A greater level of sampling within Phaenospermateae and Meliceae as well as sampling from the early diverging Nardeae and Lygeeae are needed to clarify the evolutionary history of these indels.
One insertion of mitochondrial origin was identified in the plastome of Triticum monococcum. This is currently the fourth documented case of plastome regions exhibiting mitochondrial homology. This type of gene transfer was first documented by Goremykin et al. (2009) in Daucus carota. The second was located by Straub et al. (2013) in Asclepias and the third was located by Wysocki et al. (2015) in two species within the Parianinae lineage of the Bambusoideae (Eremitis sp., Pariana radiciflora). While the first three documented insertions were extensively tested for erroneous assembly, the T. monococcum plastome was not sequenced by our team and cannot currently be verified.
One insertion of nuclear homology was identified in the plastome of Triticum urartu. The presence of this insertion, as well as the mitochondrial insertion, in the inverted repeat region suggests that it may have been retained due to the conserved nature of this region. This plastome was also not sequenced by our team so the presence of this insertion cannot be verified and may be an artefact of mis-assembly.

Conclusions
Our phylogenomic analysis of whole plastomes resolve relationships at the base of Pooideae that have varied in earlier studies, and provide new insights into several aspects of relationship among tribes of Poeae, including a strongly supported novel relationship between Torreyochloinae and Phalaridinae. Plastomes representing Lygeae, Nardeae, Brylkinieae and Littledalea are needed to complete tribal-level plastome sampling, and several subtribes of Poeae are as yet unsampled. Our results demonstrate that inclusion of non-coding data in whole plastome analyses provides important characters for recovering robust support at deep nodes, compared with protein-coding data alone. Given rapid advances in next-generation sequencing, achieving a densely sampled plastome-based phylogeny of Pooideae is a realistic goal that we are working towards.

Conflict of Interest Statement
None declared.

Supporting Information
The following additional information is available in the online version of this article - Table S1. Summary of the Pooideae classification of Soreng et al. (2014). Table S2. List of completed and publicly available whole plastomes in Poaceae (as of 25 September 2014). Figure S1. ML phylogram of complete plastomes of 45 pooid grasses and one outgroup taxon, including the highly divergent, previously banked sequence of Schedonorus arundinaceus (taxon in bold; GenBank accession NC_011713). Bootstrap values are indicated only when at least one is ,100 (ML bootstrap value precedes MP bootstrap value). ML and BI topologies were identical. Posterior probabilities at all nodes ¼ 1.00. 'NR' indicates a node not resolved or supported above the 50 % bootstrap level in the MP analysis. In the MP analysis, the boldfaced S. arundinaceus is sister to a clade comprising the other S. arundinaceus sample, Festuca pratensis, Lolium multiflorum and L. perenne; the latter clade receives 71 % BS. Clades marked with a cross are reversed in the MP tree. The MP bootstrap value of the node affected by the reversal of these clades reflects that of the reversed topology. Taxa marked with diamonds are a maximally supported clade in the MP tree. Figure S2. ML phylogram of the protein cds of 45 pooid grasses and one outgroup taxon, including the highly divergent, previously banked sequence of Schedonorus arundinaceus (taxon in bold). Support values are indicated only when at least one is less than the maximum possible value (ML bootstrap value precedes MP bootstrap value, which precedes posterior probability value). ML and MP topologies were identical. 'NR' indicates a node not resolved in the strict consensus of most parsimonious trees or not supported above the 50 % bootstrap level. In the MP analysis, the boldfaced S. arundinaceus is sister to a clade comprising the other S. arundinaceus sample, Festuca pratensis, Lolium multiflorum and L. perenne; the latter clade receives 97 % BS. A star indicates clades that were reversed in the BI tree. The posterior probability of the node affected by the reversal of these clades reflects that of the reversed topology.
Dataset S1. Details of scored indels in the 46-taxon matrix.