A large-scale chloroplast phylogeny of the Lamiaceae sheds new light on its subfamilial classification

Lamiaceae, the sixth largest angiosperm family, contains more than 7000 species distributed all over the world. However, although considerable progress has been made in the last two decades, its phylogenetic backbone has never been well resolved. In the present study, a large-scale phylogenetic reconstruction of Lamiaceae using chloroplast sequences was carried out with the most comprehensive sampling of the family to date (288 species in 191 genera, representing approximately 78% of the genera of Lamiaceae). Twelve strongly supported primary clades were inferred, which form the phylogenetic backbone of Lamiaceae. Six of the primary clades correspond to the current recognized subfamilies Ajugoideae, Lamioideae, Nepetoideae, Prostantheroideae, Scutellarioideae, and Symphorematoideae, and one corresponds to a portion of Viticoideae. The other five clades comprise: 1) Acrymia and Cymaria; 2) Hymenopyramis, Petraeovitex, Peronema, and Garrettia; 3) Premna, Gmelina, and Cornutia; 4) Callicarpa; and 5) Tectona. Based on these results, three new subfamilies—Cymarioideae, Peronematoideae, and Premnoideae—are described, and the compositions of other subfamilies are updated based on new findings from the last decade. Furthermore, our analyses revealed five strongly supported, more inclusive clades that contain subfamilies, and we give them phylogenetically defined, unranked names: Cymalamiina, Scutelamiina, Perolamiina, Viticisymphorina, and Calliprostantherina.

Scientific RepoRts | 6:34343 | DOI: 10.1038/srep34343 Based on the combined data sets D270 and D155, all MP, ML and BI analyses yielded very similar topologies, and this was true regardless of whether gaps were treated as simple indels or as missing data. Twelve well-supported primary clades were obtained in all analyses (Figs 1-4; Supplementary Figs S6-S12). A simplified phylogenetic tree shows the phylogenetic backbone of Lamiaceae (Fig. 1), and the 50% majority-rule consensus tree from the BI analysis of the combined D270 data set with simple gap coding shows detailed relationships (Figs 2 and 3). Six of the 12 primary clades in Fig. 1 correspond to subfamilies Ajugoideae, Lamioideae, Nepetoideae, Prostantheroideae, Scutellarioideae, and Symphorematoideae, as recognized by Harley et al. 16 and Olmstead 18 , and one corresponds to subfamily Viticoideae s. str. as recognized by Bramley et al. 47 . The monophyly of Symphorematoideae was confirmed for the first time, with all three genera sampled in one study ( Fig. 2 and 3). Besides these subfamilial clades, the other five primary clades comprise: 1) Acrymia and Cymaria; 2) Hymenopyramis, Petraeovitex, Garrettia, and Peronema; 3) Premna, Gmelina, and Cornutia; 4) Callicarpa; and 5) Tectona. The relationships among these 12 clades were inferred with varying degrees of support. Callicarpa and Prostantheroideae group together in all of the combined-data analyses, with support ranging from low to high, and the Callicarpa-Prostantheroideae clade usually emerges as sister to the remaining Lamiaceae. Symphorematoideae and Viticoideae s. str. are sister groups in all combined-data analyses with high support. Acrymia-Cymaria, Scutellarioideae, and Hymenopyramis-Petraeovitex-Garrettia-Peronema are successive sister groups to Lamioideae (Fig. 2), with each node highly supported in all combined-data analyses (Fig. 1). Overall, the 12 primary clades cluster into four larger clades. Relationships among the four larger clades are poorly resolved, but each usually received moderate to high support in our analyses ( Fig. 1: clade I-IV). Support values for the above-mentioned clades are summarized in Table 2. Phylogenetically defined names (names ending inina in Fig. 1) are hereby proposed for five clades that are moderately to strongly supported in our analyses and do not already have genus or subfamily names.

Discussion
Our phylogenetic reconstruction of Lamiaceae, on the basis of the most comprehensive sampling of Lamiaceae to date, builds on prior studies using chloroplast DNA markers 9,10,[21][22][23][24][25]27,28,[31][32][33][34][35][36][37][38][39][40][41]43,[45][46][47][49][50][51][52][53] . All of our analyses of the concatenated datasets revealed 12 highly supported primary clades (Figs 1-4), which are grouped into four moderately to highly supported larger clades ( Fig. 1: clade I-IV). This set of four clades has not been identified in any previous published analysis but is consistent with the unpublished results of another combined analysis of three cpDNA regions (ycf1 + ycf1-rps15 + trnL-F; B. Drew, pers. comm.). Of the 12 primary clades, five correspond to subfamilies Ajugoideae, Lamioideae, Nepetoideae, Prostantheroideae, and Scutellarioideae as recognized both by Harley et al. 16 and Olmstead 18 . We have no additional findings on these subfamilies relative to previous molecular studies [21][22][23][24][25][26]43,46 , so we will focus our discussion on the other seven primary clades. One of these clades corresponds to Symphorematoideae recognized by Harley et al. 16 10,24,25 was inferred to be sister to the rest of Lamiaceae. Inconsistency among published trees probably reflects taxon sampling or insufficient data, since these conditions could impact the accuracy of phylogenetic analyses 54,55 . Outgroups used by Scheen et al. 21 , Bendiksby et al. 23 , Li et al. 24 , and Chen et al. 25 are distantly related to the Lamiaceae, and Prostantheroideae were not sampled by Schäferhoff et al. 13 . In the present study, outgroups were selected from Mazaceae, Orobanchaceae, Paulowniaceae, and Phrymaceae, which together form the sister group to Lamiaceae in recent studies of Lamiales [12][13][14][15] , and the ingroup was more comprehensively and densely sampled than in previous studies. Thus, there is reason to have greater confidence in our inference of a Callicarpa-Prostantheroideae clade that is sister to the remaining Lamiaceae than the different inferred positions of these taxa in some previous studies.   18 were covered by gray boxes with different grey level, while new subfamilies and clades proposed in this study were covered by pink boxes and marked in red bold font.  Viticisymphorina. The clade comprising the subfamilies Symphorematoideae and Viticoideae s. str., which we are naming Viticisymphorina (see Phylogenetic Nomenclature, below), was strongly supported in all our analyses, consistent with some previous studies 23,47 . Each subfamily in this clade was confirmed to be monophyletic for the first time (Figs 1-4, Table 2). Though Symphorematoideae is well characterized by many morphological characters, its monophyly had never been tested in previous molecular studies because only one representative had been included 9,10,21,[23][24][25]47 . With all three genera included in the present study, Symphorematoideae was confirmed to be monophyletic (Figs 1-4, Table 2). Viticoideae as circumscribed by Harley et al. 16 have been shown to form two clades that are not sister groups 10,23,25,43,47 , with Vitex (grouped with Paravitex, Petitia, Teijsmanniodendron, Tsoongia, and Viticipremna) being the largest genus in one clade and Premna (grouped with Cornutia and Gmelina) the largest genus in the other. After Premna, Cornutia, and Gmelina were removed from Viticoideae 18 and Paravitex, Tsoongia, and Viticipremna were reduced to synonymy with Vitex 47 , the remaining Viticoideae consist of only four genera: Petitia, Pseudocarpidium Millsp., Teijsmanniodendron, and Vitex. We included representatives of all four of these genera in a single analysis for the first time here and the monophyly of Viticoideae s. str. was strongly supported (Figs 1-4, Table 2). Symphorematoideae and Viticoideae s. str. are morphologically distinct from each other. Symphorematoideae are woody climbers with simple leaves, 3-7-flowered capitate inflorescences with accrescent bracteoles, whereas Viticoideae s. str. are generally shrubs or trees that have simple or palmately compound leaves and lack the distinctive inflorescence structure of Symphorematoideae. However, the two subfamilies are putatively connected by several anatomical structures: 1) Ovaries without a false septum are rare in other Lamiaceae, but can be found both in Symphorematoideae and Viticoideae s. str. 2) In Symphorematoideae, orthotropous and pendulous ovules are borne in the free apex of the locules 16 . Ovules in most Lamiaceae are anatropous or hemianatropous. Junell 5 noted that the distinctive placentation in Symphorematoideae could easily be derived from the type of placentation found in many Viticoideae.
Premnoideae. This clade, comprising Premna, Gmelina and Cornutia, has been partially recovered in previous molecular phylogenetic analyses 10,24,25,43 , in which Premna and Gmelina always grouped together. Cornutia was first included by Bendiksby et al. 23 and revealed to be sister to a clade containing Premna, Gmelina, and Tectona (though the inclusion of Tectona in that clade conflicts with all other studies). In all of our combined analyses, the Premna-Gmelina-Cornutia clade was strongly supported ( Table 2). This clade is part of a larger clade that also includes Ajugoideae, Hymenopyramis-Petraeovitex-Garrettia-Peronema, Scutellarioideae, Acrymia-Cymaria and Lamioideae, consistent with the findings in Chen et al. 25 . Though traditionally placed in Viticoideae 3,16 , previous studies 10,[23][24][25]43,47 as well as the present one have shown that the Premna-Gmelina-Cornutia clade is not sister to the rest of Viticoideae sensu Harley et al. 16 . Furthermore, these three genera cannot be included in any other established subfamily based on our results. In a paper intended to contrast conventional and phylogenetic nomenclature, Cantino et al. 56  Peronematoideae. This clade comprises Petraeovitex, Peronema, Hymenopyramis and Garrettia. The first three of these were inferred to form a well-supported clade 25,47 that is sister to the Scutellarioideae-Acrymia-Cymaria-Lamioideae clade 25 . The same sister position to Scutellarioideae-Acrymia-Cymaria-Lamioideae was found for Garrettia, which was first included by Bendiksby et al. 23 . When Garrettia, Hymenopyramis, and Petraeovitex were included in the same analysis, they formed a moderately supported clade 25 . In our combined analyses, the four genera form a highly supported clade that is sister to a larger clade comprising of Scutellarioideae, Acrymia-Cymaria and Lamioideae (Figs 1-4, Table 2). Olmstead 18 suggested that a new name should be provided to accommodate the Hymenopyramis-Petraeovitex-Peronema clade, while Garrettia was still listed as having uncertain subfamilial placement in his A Synoptical Classification of the Lamiales (Version 2.4). Based on the present phylogeny, the clade comprising Hymenopyramis, Petraeovitex, Garrettia, and Peronema cannot be assigned to any established subfamily, thus we here propose a new subfamilial name: Peronematoideae B. Li, R. G. Olmstead & P. D. Cantino. Formal description of the new subfamily is provided below.
Cymarioideae. This small clade comprising Acrymia and Cymaria received high support in all our analyses of combined dataset D270, as well as in BI, ML, and MP analyses of D155 with gaps coded. It was inferred to be sister to subfamily Lamioideae with strong support in all analyses (Figs 1-4, Table 2). Our findings corroborate previous molecular phylogenetic analyses, where Cymaria 21 or Acrymia-Cymaria 23,25 was revealed to be closely related to Lamioideae. Now that the phylogenetic relationships seem to be well established, a taxonomic decision is needed whether to expand Lamioideae to include these two genera or name a new subfamily to accommodate them. The two approaches are equally consistent with the molecular phylogenetic results; i.e., both Lamioideae s. str. and a broader Lamioideae, expanded to include Acrymia and Cymaria, are well supported in our analyses and previous studies 23, 25 . Bendiksby et al. 23 and Chen et al. 25 have argued that an expanded Lamioideae would be more morphologically heterogeneous and difficult to diagnose, and they therefore recommended excluding Acrymia and Cymaria from Lamioideae. Because the Acrymia-Cymaria clade was only moderately supported in their analyses, Chen et al. 25 noted that if future evidence were to strongly corroborate the existence of this clade, a new subfamily could be named to accommodate them. This hypothesis is confirmed with strong confidence in our analyses (

Taxonomy and Nomenclature
An Updated Subfamilial Classification of Lamiaceae. The most recent and widely adopted classification of Lamiaceae was proposed by Harley et al. 16 in the first global, genus-level treatment of the entire family in more than a century since Briquet 2 . Harley et al.'s landmark work includes 226 genera assigned to seven subfamilies (Ajugoideae, Lamioideae, Nepetoideae, Prostantheroideae, Scutellarioideae, Symphorematoideae, and Viticoideae), and ten genera listed as incertae sedis (Acrymia, Callicarpa, Cymaria, Garrettia, Holocheila, Hymenopyramis, Ombrocharis, Peronema, Petraeovitex, and Tectona). A decade later, numerous new findings have improved the classification incrementally. The results reported here provide the basis for a revised subfamilial classification. We take this opportunity to describe three new subfamilies and to update the subfamilial classification of the family incorporating new findings since Harley et al. 16 . For each subfamily, we provide a brief summary of its historical classification and presently understood phylogenetic position, generic and species diversity, morphology, synapomorphies, and distribution. Of the ten genera treated as incertae sedis by Harley et al. 16 , there are only two that we do not assign to a subfamily-Callicarpa and Tectona; these genera are inserted among the subfamilies in positions consistent with the phylogeny in Fig. 1. Prostantheroideae Luerssen. This endemic Australian subfamily includes 17 genera and ca. 300 species belonging to two major clades, Chloantheae (12 genera) and Westringieae (5 genera). Prior to the cladistic study of Cantino 4 , these tribes were usually placed in Verbenaceae (or Chloanthaceae 57,58 ) and Lamiaceae, respectively (e.g., Briquet 2,3 ). However, based on gynoecial anatomy, Junell 5 transferred "Chloanthoideae" (i.e., Chloantheae) to Lamiaceae and suggested that it shares a common origin with "Prostantheroideae" (i.e., Westringieae). Cantino et al. 59 first placed the two groups together in a subfamily ("Chloanthoideae"). Monophyly of each tribe (excluding Spartothamnella Briq. and Tectona from Chloantheae, contrary to Munir 60 and Cantino et al. 59 , respectively) and of the combined Prostantheroideae was confirmed by molecular phylogenetic analysis 20 , which also produced the first evidence that Callicarpa is sister to Prostantheroideae. Phylogenetic studies of Prostantheroideae 20,44-46 (as well as from T. Wilson 63 , and indications that additional realignments to several genera will be forthcoming 44,64 (also T. Wilson, pers. comm.). A probable synapomorphy is a dry schizocarp that splits into four one-seeded mericarps. This feature also characterizes several other clades within Lamiaceae, but our results indicate that it evolved independently in each. Because this fruit type is found in all members of Westringieae and in Brachysola Rye, which is sister to the rest of Chloantheae 4,20,45 , it is the most parsimonious assignment to the most recent common ancestor of Prostantheroideae.
Callicarpa Linnaeus. Callicarpa contains about 140 species occurring in both temperate and tropical regions 16 . The plants are small trees or shrubs with actinomorphic, 4-5 (− 7)-parted flowers and drupaceous fruits. Despite being one of the largest genera in Lamiaceae, its phylogenetic position had not previously been confirmed. In previous molecular studies, Callicarpa was included with only one or few representatives, and has been inferred to be sister to the rest of the family 9,21,23 , to group with subfamily Prostantheroideae 15,20,43 , or to be variably isolated in different positions 10 . Bramley 49 sampled more representatives and indicated that Callicarpa is monophyletic, but she could not infer its phylogenetic position because of poor sampling from the whole family. In the present study, Callicarpa was sampled much more extensively (18 spp.), taking into consideration its morphological and geographic breadth and its infrageneric classification. In all our analyses, the monophyly of Callicarpa was well supported (Figs 1-4; Table 2). A sister relationship between Callicarpa and subfamily Prostantheroideae was moderately to highly supported in analyses of the combined dataset D270 (Figs 1-3, Table 2), as well as in BI and ML analyses of the combined dataset D155 (Fig. 4).

Symphorematoideae Briquet.
The subfamily has three genera: Congea (ca. 7 species), Sphenodesme Jack (ca. 14 species) and Symphorema Roxb. (3 species). All genera are endemic to continental Asia (India to Indochina and southern and eastern China) and parts of Malesia (Peninsular Malaysia, Sumatra, Borneo, Java, Tanimbar Islands, and the Philippines). Congea tomentosa Roxb. and to a lesser extent C. griffithiana Munir are cultivated as ornamental climbers. There are several morphological traits that unite the three genera. All are climbers with inflorescences of 3-7-flowered capitate cymes. These are usually surrounded by conspicuous bracteoles, often coloured and accrescent. Flowers of Sphenodesme are 5 or 6-merous; the flowers of Symphorema are 6 to 16 (-18)-merous. The corolla of Congea is 2-lipped, but the corollas of Sphenodesme and Symphorema are actinomorphic. The ovaries are incompletely 2-locular, and the ovules are orthotropous and pendulous. Although unique to this subfamily, the ovary type was interpreted by Junell 5 as being derived from that found in genera of "Viticoideae" (in which Junell included not only Viticoideae s. str. but also Callicarpa, Tectona, Premnoideae, Peronematoideae, Cymarioideae, and Ajugoideae in our classification). The fruit in all three genera is indehiscent, weakly drupaceous or dry, and is 1 (− 2) -seeded by abortion. Vitex, Petitia, and Teijsmanniodendron. Because several smaller viticoid genera were not included in these early analyses, the circumscription of the subfamily was not altered by Harley et al. 16 . Bramley et al. 47 further elucidated the relationships among the viticoid genera, focusing particularly on Southeast Asian taxa traditionally allied to Vitex. Based on their results, Viticipremna, Tsoongia, and Paravitex were included in Vitex, reducing the number of viticoid genera to seven. The generic status of Petitia and Pseudocarpidium was unchanged, because of poor support for the position of the former, lack of any data for the latter, and poor sampling among Neotropical taxa. In the present study, we find sufficient evidence to include Petitia, but not Pseudocarpidium, in Vitex. Similarly, the generic status of Teijsmanniodendron remains problematic. Neither the analyses of Bramley et al. 47 nor our analyses provide convincing support to include Teijsmanniodendron in Vitex. Identification of species in these two genera is often confused; the sole morphological character that can be used to delimit them in most cases is a swelling present at the base and apex of the petiole in Teijsmanniodendron. Traditionally, Teijsmanniodendron species were also recognised by their capsule-like rather than drupaceous fruit that is 1-seeded (by abortion) 66 , but a reduction in the number of mature seeds can also occur in Vitex species.

Viticoideae
Viticoideae are distributed predominantly in the Tropics (Vitex throughout; Teijsmanniodendron in Malesia; Pseudocarpidium in the Caribbean), although there are a few temperate species of Vitex. Madagascar may be home to a number of currently unrecognized species of Vitex 67 . Analysis of a greater number of viticoid taxa could result in further changes to generic boundaries in this subfamily. Possible synapomorphies for the subfamily are phytochemical (see Pedersen 65 ). There has been no comparative study of morphological or anatomical characters including all of the viticoid genera aside from Junell's 5 work on gynoecial structure. Although Junell noted that Vitex, Petitia, Pseudocarpidium, and the other genera now recognized as Vitex have a very similar ovary structure, he found some differences in Teijsmanniodendron. Further comparative studies including subfamily Viticoideae, particularly focusing on gynoecial structure, may elucidate unifying characters.

Nepetoideae (Dumortier) Luerssen.
Nepetoideae is the largest subfamily of Lamiaceae, containing almost half of the genera and species. It now contains 118 genera (compared to 105 recognized by Harley et al. 16 ) and ca. 3400 species, which are widely distributed across tropical and temperate regions of the northern and southern hemispheres but with few native species in Australia and New Zealand. Probable synapomorphies for Nepetoideae include hexacolpate and three-celled pollen, investing embryos, myxocarpy, gynobasic style, and the presence of rosmarinic acid 4,16,[68][69][70][71] . Three tribes are now recognized within Nepetoideae 16 : Elsholtzieae, Mentheae, and Ocimeae. The monophyly of each of these tribes is well supported by molecular phylogenetic studies 34,37,40,43,72 , but there are conflicting findings about relationships among the three tribes.
There have been several genus-level changes since the treatment of the subfamily by Harley et al. 16 16 ) and showed that Ombrocharis (unassigned to subfamily by Harley et al. 16 ) is sister to Perillula Maxim. within tribe Elsholtzieae. Drew and Sytsma 37 found Heterolamium C. Y. Wu to be nested within Meehania Britton, but Deng et al. 74 found that the specimen of Heterolamium studied by Drew and Sytsma 41 was misidentified and was in fact a member of Meehania. Thus, the systematic position of Heterolamium within Nepetoideae is still uncertain.
Tectona Linnaeus f. Tectona is a genus of large trees comprising three species distributed from India to southeast Asia. The large drupaceous fruits contain a hard four-celled endocarp and are enclosed in an enlarged persistent calyx. Tectona has been included in several molecular studies 9,10,21,23,47 , but its phylogenetic position has never been determined definitively. It has been inferred to be sister to a clade comprising Hymenopyramis-Petraeovitex-Peronema and Premna-Gmelina 47 , to a large clade containing Ajugoideae, Lamioideae, Scutellarioideae, Peronema-Petraeovitex, and Premna-Gmelina-Cornutia (B. Drew, pers. comm.), to another larger clade comprising Ajugoideae, Lamioideae, Premna-Gmelina, Prostantheroideae, Scutellarioideae, and Vitex-Petitia 10 , or to group with Gmelina 21,23 . In our more comprehensive analyses, Tectona is inferred to be sister to a large clade comprised of Lamioideae, Acrymia-Cymaria, Scutellarioideae, Hymenopyramis-Petraeovitex-Garrettia-Peronema, Ajugoideae, and Premna-Gmelina-Cornutia, with moderate to strong support in BI and ML analyses of both datasets D270 and D155, with or without gaps coded (Figs 1-4). The distinct morphology of Tectona including an actinomorphic 5-7-lobed calyx and corolla, greatly enlarged and inflated persistent calyx, and 4-celled endocarp with small central cavity between the cells 16 , contributed to the difficulty of placing it in previous classifications. Our results suggest that Tectona is an early diverging lineage from the major clade IV (Figs 1-4).
This new subfamily contains three genera: Premna (50-200 spp. in tropical to subtropical Asia, Africa, Australia, and the Pacific Islands), Gmelina (31 spp. in tropical and subtropical Asia to Australia and western Pacific Islands), and Cornutia (12 spp. in tropical America). The three genera were traditionally placed in subfamily Viticoideae of Verbenaceae 3 , and transferred to Lamiaceae together with the subfamily 5,16,59 , and then excluded from Viticoideae by Olmstead 18 based on molecular evidence that Viticoideae is non-monophyletic if they are included 23,47 . A possible synapomorphy for Premnoideae is a drupaceous fruit with one four-seeded pyrene. However since a similar fruit structure is also found in Tectona and some species of Vitex, it may instead be a synapomorphy at a deeper level in the phylogeny with subsequent reversals. With the number of species estimated from 50 75 to 200 76 , Premna now ranks among the most taxonomically difficult and complicated genera of Lamiaceae. Though some regional revisions of the genus have been done in recent decades [77][78][79][80][81][82] , there is no treatment of the genus throughout its range. A global taxonomic revision of Gmelina was published by de Kok 83  Shrubs, trees, and lianas. Leaves opposite, petiolate, simple or ternately, biternately, or pinnately compound. Inflorescence cymose, axillary, and/or terminal, highly variable in form. Calyx actinomorphic, 4-5-lobed, and usually accrescent (not accrescent in Peronema) in fruit. Corolla white to yellow, 4-5-lobed, and nearly actinomorphic to zygomorphic. Stamens 4 or posterior pair reduced to staminodes (Peronema), equal or didynamous, included or exserted, thecae parallel to divaricate, usually separate (confluent in Garrettia) at dehiscence; pollen tricolpate, tectate-perforate. Ovary unlobed, stigma 2-lobed, equal or unequal; disc absent or poorly developed. Fruit dry, globose or turbinate, glabrous or pubescent to villous, indehiscent or breaking into two or four mericarps, abscission-scar as long as the mericarp.
This new subfamily comprises four small, mostly tropical Asian genera that were treated as incertae sedis by Harley et al. 16 : Garrettia (1 sp., southwest China, Thailand, and Indonesia), Hymenopyramis (7 spp., India, China, and Indo-China), Peronema (1 sp., Thailand to Malaysia and western Indonesia), and Petraeovitex (8 spp., Burma, Thailand, Malaysia, Indonesia, Philippines, New Guinea, and Melanesia). Previously, Hymenopyramis, Peronema, and Petraeovitex have been placed in Caryopteridoideae 3 or transferred to Viticoideae 5 , or Hymenopyramis was retained in Viticoideae but Peronema and Petraeovitex were transferred to Teucrioideae 59 . Garrettia was always placed in Caryopteridoideae 92,93 before being transferred to Ajugoideae 59 . The four genera of Peronematoideae differ greatly in morphology and have never been linked in any previous classification, but Chen et al. 25 has found some traits in common, including woody stems (small or climbing shrubs, lianas or large trees), white to yellowish corolla, unlobed ovary, nectar disc poorly developed or absent, and dry fruit. Each of these traits is probably either synapomorphic at a more inclusive level within Lamiaceae or plesiomorphic in the family as a whole 25 .

Scutellarioideae (Dumortier) Caruel.
A taxon centered on Scutellaria was recognized as a distinct element within Lamiaceae in early classifications (e.g., Bentham 1 ; Briquet 2 ), often comprising only Scutellaria and the segregate genera, Perilomia Kunth and Salazaria Torr., now included within Scutellaria 16,94 . Early phylogenetic studies based on morphology 4,8 and DNA sequences 10 expanded this clade to include Renschia Vatke, Tinnea Kotschy ex Hook. f., and Holmskioldia Retz., the latter formerly assigned to Verbenaceae. The rediscovery of the extremely rare Wenchengia C. Y. Wu & S. Chow permitted Li et al. 24 to confirm its placement in Scutellarioideae by Harley et al. 16 . A characteristic two lobed, untoothed calyx is shared by a clade of Scutellaria, Renschia, and Tinnea. Holmskioldia is sister to this clade and has an expanded saucer-shaped calyx with five, often indistinct, lobes, which form the dominant part of the floral display, unlike the other genera. Wenchengia is sister to the rest of the clade and has a two-lobed, but five-toothed calyx. Probable synapomorphies for Scutellarioideae include pericarps with tuberculate or elongate processes 24 , high densities of xylem fibers in the calyces 95 , and possibly racemose inflorescences (but they are cymose in Holmskioldia and most species of Tinnea, suggesting that independent origin of racemes within Scutellarioideae may be equally parsimonious). Scutellaria includes approximately 360 species, is cosmopolitan in distribution, occurs in a wide range of habitats, and includes annual and perennial herbs and shrubs. A global taxonomic revision of Scutellaria established infrageneric classification and Shrubs and subshrubs. Leaves simple, opposite, petiolate, elliptic or ovate to rhombic, crenate to crenulate or repand. Cymes axillary, lax, long-pedunculate, with secund, monochasial branches, sometimes grading into a terminal paniculiform thyrse. Calyx campanulate, accrescent, broadly campanulate to urceolate or subglobose in fruit, 5-lobed, lobes equal to subequal, triangular. Corolla white to yellowish, 2-lipped, posterior lip entire to deeply 2-lobed, anterior lip with median lobe largest. Stamens 4, didynamous (anterior pair longer), included or exserted, thecae divaricate, confluent at dehiscence; pollen tricolpate, tectate-perforate, suprareticulate, columellae simple to sparsely branched. Ovary shallowly 4-lobed; style sub-terminal, stigma lobes subequal to unequal; disc absent. Nutlets obovoid, reticulately ridged, pubescent, abscission-scar lateral, 0.4-0.6× the length of the nutlet.
This new subfamily consists of two small, tropical Asian genera: Acrymia (1 sp., Peninsular Malaysia) and Cymaria (2-3 spp., Hainan, Indo-China and Malesia)(species numbers and ranges from Harley et al. 16 ). The two genera have been included in subfamily Ajugoideae 2,59,96,97 or treated as incertae sedis 16 . A probable synapomorphy of Cymarioideae is its inflorescence structure: the cymes are axillary, lax, and long-pedunculate, with secund, monochasial branches 16 . This form of inflorescence is rare in the family but also occurs in Garrettia, where it apparently evolved independently.
When describing Cymarioideae, we realized that the type genus of the new subfamily, Cymaria, is currently without a type species. Bentham 98 simultaneously named two species, C. dichotoma Benth. and C. elongata Benth., at the same time he described the genus, but he did not designate either as the type. Cymaria was recognized by several subsequent authors, and the two species were always listed in parallel without any type designation 1,2,97,99 . We take this opportunity to designate C. dichotoma as the type species for Cymaria, because it is more widely distributed and better represented in herbaria.

Lamioideae Harley.
The largely Old World subfamily Lamioideae is second in size only to Nepetoideae with over 60 genera and ca. 1200 species. Tribe Stachydeae is cosmopolitan in distribution, and Synandreae is endemic to North America. The other eight tribes are largely Eurasian, but four of them include some African species, and Pogostemon also occurs in Australia 16,21,23 . All Lamioideae have a gynobasic style, a synapomorphy that arose independently in Nepetoideae and Scutellarioideae 9,10 . Another possible synapomorphy is the presence of seed oils with an allenic component present 4 , but this character has been studied in too few species to be fully evaluated. Recent phylogenetic studies based on cpDNA sequence data have identified ten clades that have been ranked as tribes, with several genera unassigned to tribe 21,23 , but the monophyly of some of these tribes is not supported by nuclear DNA (PPR) data 26 . Further studies of relationships within some tribes (Synandreae-Scheen et al. 29 53 ) have led to changes in the composition of some genera, acceptance of genera not recognized by Harley et al. 16 , and elimination of some genera that were recognized by Harley et al. 16 . In addition, Chen et al. 25 showed that the formerly unplaced genus Holocheila belongs in Lamioideae. Besides Holocheila, four genera have been added to Lamioideae since 2004: Rydingia Scheen & V. A. Albert 103 , Betonica L. 21 , Acanthoprasium (Benth.) Spenn. 23 104 . As a result of these changes, Lamioideae is now considered to have 62 genera (versus 63 recognized by Harley et al. 16 ).

Phylogenetic Nomenclature
Our analyses revealed five strongly supported but previously unnamed clades that contain subfamilies (labeled in Fig. 1). These clades warrant naming to facilitate communication about them, but there is no standard rank between family and subfamily. We therefore have given them unranked names, which are defined below, following the rules and recommendations of the draft PhyloCode 105 . For readers unfamiliar with phylogenetic nomenclature, the preface of the draft PhyloCode provides a good introduction (https://www.ohio.edu/phylocode/preface.html). The names of a variety of plant clades have been defined following the draft PhyloCode, including a set of major tracheophyte and angiosperm clades 106 , but there have been few previous applications of phylogenetic nomenclature to Lamiaceae. Cantino et al. 56 provided phylogenetic definitions for some clade names within Lamiaceae to illustrate differences between phylogenetic and traditional nomenclature, but the PhyloCode did not yet exist, and the names and definitions in that paper were not intended to have any formal nomenclatural status. Salmaki et al. 31  Primary reference phylogeny. Figure 2; see also Fig. 1 (this paper), Bendiksby et al. 23 (Fig. 1), Chen et al. 25 (Fig. 4), Roy and Lind qvist 26 (Fig. 1a), and Chen et al. 43 (Fig. 3, where Cymaria represents Cymarioideae). Composition. Cymarioideae and Lamioideae.
Synapomorphies. Confluent anther thecae may be a synapomorphy. It is shared by Cymarioideae and one of the two basal subclades of Lamioideae (i.e., Pogostemoneae), but the anthers in the other basal subclade (i.e., the rest of Lamioideae) vary from distinct (e.g., Gomphostemmateae, Synandreae, Galeopsis L.) to confluent (e.g., Colquhounia Wall. and some Stachydeae). The closest outgroups to Cymalamiina (i.e., Scutellarioideae and Peronematoideae) have distinct thecae (except Garrettia in Peronematoideae), as do the more distant outgroups Premnoideae and Tectona; this character varies in Ajugoideae, another more distant outgroup. Given the distribution of the character states and its variability within many tribes of Lamioideae, it is not clear whether confluent thecae is a synapomorphy of Cymalamiina with a series of reversals within Lamioideae or, alternatively, that it evolved independently in Cymarioideae, Pogostemoneae, and in scattered other members of Lamioideae. Primary reference phylogeny. Figure 2; see also Fig. 1 (this paper), Bendiksby et al. 23 (Fig. 1), Chen et al. 25 (Fig. 4), and Chen et al. 43 (Fig. 3).

Synapomorphies.
A four-lobed ovary appears to be a synapomorphy for Scutelamiina. It is shallowly four lobed in Cymarioideae and the more basal members of Scutellarioideae (though unlobed in Holmskioldia) and deeply lobed in Lamioideae and Scutellaria. A developmentally related feature, a schizocarpic fruit with four mericarps ("nutlets"), may be another synapomorphy of Scutelamiina. However, within its sister group (Peronematoideae), the 4-seeded capsule of Garrettia and Peronema breaks with pressure into four mericarps, possibly representing a stage in the evolution of the schizocarp of Scutelamiina. Another likely synapomorphy is suprareticulate (sometimes called bireticulate) pollen, which is found in all genera of Scutellarioideae and Cymarioideae and most genera of Lamioideae 90,107-109 . However the presence of suprareticulate pollen in Garrettia (Peronematoideae) 90,110 and suprarugulose pollen in Peronema 110 raises the possibility that this feature may be a synapomorphy for Perolamiina rather than Scutelamiina, though Petraeovitex and Hymenopyramis (the more distal genera within Peronematoideae) have psilate pollen 110 . Furthermore, suprareticulate sculpturing could be synapomorphic at an even more inclusive level because it occurs in some genera of Premnoideae (Gmelina, but not Cornutia or Premna, based on very few species 90,110 ).
Comments. Cantino et al. 56 applied the name Lamiina to this clade and provided a phylogenetic definition, but the names used in that paper were intended only to illustrate differences between phylogenetic and traditional rank-based nomenclature. They were not intended to have nomenclatural precedence under either system, and the PhyloCode did not yet exist. Primary reference phylogeny. Figure 3; see also Fig. 1 (this paper), Bendiksby et al. 23 (Fig. 1, where Congea represents Symphorematoideae, Vitex and Petitia represent Viticoideae), and Bramley et al. 47 (Fig. 1, where Sphenodesme represent Symphorematoideae).
Comments. We intend this name to be applicable only if Symphorematoideae and Viticoideae are sister groups. Although our analyses strongly support the monophyly of this grouping, it is not supported in some other analyses 24,25,43 . For this reason, the definition includes external specifiers to make the name inapplicable under many alternative phylogenies. Primary reference phylogeny. Figure 3; see also Fig. 1 (this paper), Olmstead et al. 20 (Fig. 1), Refulio-Rodriguez and Olmstead 15 (Fig. 1A, where Prostanthera represents Prostantheroideae), and Chen et al. 43 (Fig. 3, where Prostanthera and Westringia represent Prostantheroideae).

Composition. Callicarpa and Prostantheroideae.
Apomorphies. Branched trichomes and actinomorphic corollas, two characters that are infrequent in Lamiaceae, are shared by Callicarpa and Chloantheae (one of the two basal subclades of Prostantheroideae). Branched trichomes also occur in scattered species of the other subclade (Westringieae), increasing the likelihood that this feature characterizes Calliprostantherina. However, it could be apomorphic at a deeper level in the phylogeny since Symphorematoideae and Tectona also have branched hairs (see Fig. 1 for the relationship of the latter taxa to Calliprostantherina). Actinomorphic corollas also occur in Tectona and some genera of Symphorematoideae; furthermore, all species of Westringieae have zygomorphic flowers, weakening the hypothesis that actinomorphy is a synapomorphy for Calliprostantherina. There are similarities in pollen surface sculpturing between some species of Callicarpa and some genera of Prostantheroideae 111 , but there is considerable variation in both groups, and the polarity of the character is unknown.
Comments. Because the molecular support for this clade is only moderate and the potential morphological synapomorphies discussed above are not convincing, the definition is designed to become inapplicable under many alternative phylogenies (e.g., Bendiksby et al. 23 : Fig. 1).  Note:-One of the sheets at S [sheet number S04-2601] is designated as lectotype because E. L. Ekman was based there, and it has attached to it a handwritten note "it is a pleasure to dedicate this fine new species to Prof. Urban, the admirable botanist, the never tiring worker, and the best friend a man ever had".

Materials and Methods
Choice of markers, taxon sampling and molecular data. Five chloroplast DNA markers-matK, ndhF, rbcL, rps16, and trnL-F-were employed in this study because (1) they have been widely used in phylogenetic reconstructions of Lamiaceae at generic, tribal or subfamilial level, and (2) many species of Lamiaceae have already been sequenced for these markers in previous molecular studies 9,10,101 . No comparable source of data exists for any nuclear DNA region for a broad sample of Lamiaceae.
The ingroup sample included representatives of all seven subfamilies and all ten genera incertae sedis recognized by Harley et al. 16 and all 14 tribes recognized by Olmstead 18 . Nomenclature of Lamiaceae and Viticoideae s. str. followed Olmstead 18 and Bramley et al. 47 , respectively. Initially, we downloaded data for all taxa of Lamiaceae with sequence information for any of the five gene regions deposited in Genbank as of August 2015. In the five subfamilies whose monophyly is well supported (viz., Ajugoideae, Lamioideae, Nepetoideae, Prostantheroideae and Scutellarioideae), sampling was designed to cover their genus-level diversity. Generally, genera with at least two sequenced regions were selected, and each selected genus was represented by one or two species. Particular emphasis was placed on sampling Symphorematoideae, Viticoideae s. str., all genera incertae sedis, and three genera formerly assigned to Viticoideae-Cornutia, Gmelina, and Premna. In three large genera-Callicarpa, Premna, and Vitex, sampling was designed to cover their morphological and geographic breadth. In total, 288 species representing 191 genera were included, representing approximately 78% of the genera of Lamiaceae. Five outgroup species were selected representing the closest relatives to Lamiaceae in Lamiales [12][13][14][15] . They are Lindenbergia philippensis (Cham. & Schltdl.) Benth. and Pedicularis groenlandica Retz. from Orobanchaceae, Paulownia tomentosa (Thunb.) Steud. from Paulowniaceae, Mazus reptans N. E. Br. from Mazaceae and Phryma leptostachya L. from Phrymaceae. Information on sampled taxa and Genbank accession numbers is assembled in Supplementary Table S1.
The five separate molecular data sets matK, ndhF, rbcL, rps16 and trnL-F contained 202, 160, 170, 181, and 259 sequences with 54, 83, 59, 57, and 88 newly reported sequences, respectively. The dataset combining the five markers included 270 taxa (D270), with 39.65 % missing data. According to investigations by Wiens 113 and Wiens and Moen 114 , the proportion of missing data should not affect the accuracy of the phylogenetic analysis; however, just to make sure, a reduced dataset was assembled including 155 taxa (D155) with at least three of the five regions or 50 % of the total aligned sequence length available for each terminal taxon. The total amount of missing data in D155 was 23.51 %. For most species in the combined datasets, data were available for all five regions, but there were some genera of Ajugoideae, Lamioideae, Nepetoideae, Prostantheroideae, and Scutellarioideae in which different species were used for different gene regions. When data were pooled in this way, generic names, rather than species names, were used to represent the combined sequences in the phylogenetic trees.
DNA extraction, amplification, and sequencing. The 2x CTAB method of Doyle and Doyle 115 was used to extract total genomic DNA of the samples with silica dried leaf tissue, and DNEasy ® Plant Mini Kit (QIAGEN ® , Valencia, California, USA) was used for herbarium materials according to the manufacturer's specifications. The DNA extracts were dissolved in TE buffer and preserved at − 20 °C for further use.
Primer pairs used in Polymerase chain reaction (PCR) amplification of the five regions are listed in Table S2 with their sequences and references. The PCR reaction system and amplification protocol were identical for all five fragments. PCR reactions used 2.5 μ L sample DNA, 0.5 μ L Dream Taq DNA polymerase, 5 μ L 10 × DreamTaq Green Buffer, 5 μ L dNTP Mix (2 mM each), 1 μ L of bovine serum albumin (BSA, 20 mg/mL), 1 μ L of each primer in a final reaction volume of 50 μ L. The PCR program was as follows: an initial template denaturation at 94 °C for 5 min, 35 cycles of 30 second denaturation at 94 °C, 1 minute primer annealing at 52 °C, 1.5 min extension at 72 °C, with a final extension of 8 min at 72 °C. Sequencing was done by the Invitrogen sequencing service (Invitrogen, commercial sequencing facility, Guangzhou, China) using the same primers for PCR amplifications.

Sequence alignment and phylogenetic analyses.
Sequencher v.4.5 116 was used to evaluate chromatograms for base confirmation and to edit contiguous sequences. All DNA sequences were initially aligned using Clustal X v.2.0. 117 and adjusted manually in BioEdit Sequence Alignment Editor v.7.0.0 118 .
The separate data sets were first analyzed using Maximum parsimony (MP) and Maximum likelihood (ML) methods, with gaps treated as simple indels determined by the program Gapcoder 119 and added to the matrix as binary presence/absence characters. The combined data sets D270 and D155 were analyzed using MP, ML and Bayesian inference (BI) methods with gaps treated either as missing data or as simple indels.
MP analyses were conducted using PAUP* v.4.0b10 120 with all characters unordered and equally weighted. Heuristic search was conducted using 1000 random addition sequence replicates, tree-bisection-reconnection (TBR) branch swapping, MulTrees in effect, and steepest descent off. Bootstrap support values (BS) were estimated using a heuristic search strategy with 500 bootstrap replicates and 1000 random sequences additions.
ML analyses were performed on the web server RAxML Black Box 121 . Before each submission, the "Maximum likelihood search" and "Estimate proportion ofinvariable sites" options were selected, with a total of 1000 bootstrap replicates performed.
BI analysis was executed using MrBayes version 3.2.2 122 on the CIPRES Science Gateway 123 with the default parameters. The best substitution types (Nst) and rate distribution models (rates) were determined by the Akaike information criterion (AIC) using Model Test v.3.7 124 with the hierarchical likelihood ratio tests. Four Markov chain Monte Carlo (MCMC) chains were run, each beginning with a random tree and sampling one tree every 1000 generations for 30 000 000 generations. Mixing, convergence and a suitable burn-in were assessed with the statistics provided by the program and with Tracer v. 1.6 125 . Post burn-in samples from the four runs were merged using LogCombiner v1.7.5. (available at http://beast.bio.ed.ac.uk/, LogCombiner) prior to the calculation of a 50 % majority-rule consensus tree.