Phylogeny and reclassification of Aconitum subgenus Lycoctonum (Ranunculaceae)

Phylogenetic analyses were performed using multiple nuclear (ITS and ETS) and chloroplast regions (ndhF-trnL, psbA-trnH, psbD-trnT, and trnT-trnL) to test the monophyly of Aconitum subgen. Lycoctonum (Ranunculaceae) and reconstruct the phylogenetic relationships within the subgenus. The subgenus as currently circumscribed is revealed to be polyphyletic. To achieve its monophyly, sect. Galeata and sect. Fletcherum, both being unispecific and each having a unique array of characters (the latter even having the aberrant base chromosome number of x = 6), must be removed from the subgenus. The subgenus Lycoctonum should thus be redefined to include only two sections, the unispecific sect. Alatospermum and the relatively species-rich sect. Lycoctonum. The section Alatospermum, which is both morphologically and karyologically in the primitive condition, is resolved as the first diverging lineage of the subgenus Lycoctonum clade. The monophyly of sect. Lycoctonum is strongly supported, but all the ten series currently recognized within the section are revealed to be para- or poly-phyletic. Five major clades are recovered within the section. We propose to treat them as five series: ser. Crassiflora, ser. Scaposa, ser. Volubilia, ser. Longicassidata, and ser. Lycoctonia. Thus, a formal reclassification of subgen. Lycoctonum is presented, which involves segregating both sect. Galeata and sect. Fletcherum from the subgenus as two independent subgenera within the genus Aconitum, reinstating one series (ser. Crassiflora) and abolishing six series (ser. Laevia, ser. Longibracteolata, ser. Micrantha, ser. Ranunculoidea, ser. Reclinata, and ser. Umbrosa) within sect. Lycoctonum. The series affiliation of some species within the section is adjusted accordingly.

The first comprehensive classification of Aconitum subgen. Lycoctonum was proposed by Lauener and Tamura [2] and Tamura and Lauener [6] based on morphology, in which four sections and 11 series were recognized (Table 1). Later, Tamura [1] basically adopted this classification with only some minor changes in the arrangement order of the sections and series, and in the number of series as well. As shown in Table 1, he divided the subgenus into four sections: sect. Alatospermum Tamura, sect. Galeata Rapaics, sect. Fletcherum Tamura, and sect. Lycoctonum DC. The former three are all unispecific, including each the eastern Himalayan A. novoluridum Munz, the Kashmir Himalayan A. moschatum (Brühl ex Duthie) Stapf, and the eastern Himalayan A. fletcheranum G. Taylor, respectively. The section Lycoctonum, occurring nearly in the same area as subgen. Lycoctonum, comprises the remaining species which were subdivided into nine series: ser. Micrantha Steinb. ex Tamura [6] and Tamura [1]. Later, Kadota [8] ascribed another six species from northern Japan to ser. Umbrosa, five of which were described as new by him. The circumscription of Aconitum subgen. Lycoctonum by Tamura [1] has been questioned by some authors. Kadota [9] transferred A. fletcheranum from this subgenus to his new subgenus, subgen. Tangutica (W. T. Wang) Kadota, and still maintained its independent sectional status, stating that the nectary blade of the petal in this species was not provided with a tubular portion. For the same reason he also considered that the inclusion of A. moschatum in subgen. Lycoctonum was highly doubtful, although he did not pinpoint further its systematic position [9]. Indeed, except for A. fletcheranum and A. moschatum, all the other species within subgen. Lycoctonum have a nectary blade provided with a tubular portion [1,2,6,9]. The inclusion of A. fletcheranum within subgen. Lycoctonum seems also quite abnormal in terms of karyological characters. Most recently Hong et al. [10] reported the chromosome number of this species as 2n = 12. This count represents a new base chromosome number of x = 6 for the genus Aconitum which otherwise has a uniform base chromosome number of x = 8. The karyotype of A. fletcheranum is unique in Aconitum. There are two largest metacentric chromosome pairs in this species, whereas in all the other diploid taxa of the genus with available chromosomal data there is only one such chromosome pair, with the second largest pair being submetacentric [10].
Tamura's [1] classification of the largest section (including ca. 40 species) within Aconitum subgen. Lycoctonum, i.e., sect. Lycoctonum, may also be problematic. For example, he placed A. brevicalcaratum (Finet & Gagnep.) Diels and A. chrysotrichum W. T. Wang in ser. Micrantha while A. crassiflorum Hand.-Mazz. in ser. Scaposa, but these three species, all occurring in the southern part (western Sichuan, northwestern Yunnan) of the Hengduan Mountains region in southwestern China, are morphologically most closely similar to each other [11,12]. More importantly, they are all tetraploid (2n = 32) with almost identical karyotypes [10,13]. Yuan and Yang [13] proposed that the three species, together with A. rilongense Kadota, also a tetraploid species from western Sichuan, should be placed in one and the same series.
Utelli et al. [14] used the chloroplast intergenic spacer psbA-trnH and nuclear ITS region to study the relationships of the Aconitum lycoctonum L. species complex from Europe and the Caucasus Mountains. No other molecular phylogenetic study focused on subgen. Lycoctonum has been made, although more or less species of the subgenus (all belonging to sect. Lycoctonum) were chosen as outgroups or placeholders in phylogenetic analyses of subgen. Aconitum [15,16] or of the tribe Delphinieae [4,17]. Significantly, the nine species sampled by Jabbour and Renner [17] and the 18 species sampled by Wang et al. [4] form a well-supported monophyletic clade in the phylograms obtained from combined trnL-F and ITS dataset by using maximum likelihood method. It is also noteworthy that the genus Aconitum, after the segregation of subgen. Gymnaconitum from it, is a monophyletic group, with the genera Delphinium L. and Gymnaconitum as its sister groups [4,17]. This provides us a framework to probe further into the phylogeny of subgen. Lycoctonum.
In the present study, we sampled the majority of the known species within Aconitum subgen. Lycoctonum and used multiple nuclear (ITS and ETS) and chloroplast regions (ndhF-trnL, psbA-trnH, psbD-trnT, and trnT-trnL) to perform phylogenetic analyses on this subgenus. Our aims were to (1) test the monophyly of subgen. Lycoctonum, (2) reconstruct the phylogenetic relationships within the subgenus, and (3) provide a reclassification of the subgenus that is phylogeny driven.

Taxon sampling
We sampled 61 accessions representing 41 species (ca. 87% of the recognized taxa, Table 1) which covered all the sections and series in Aconitum subgen. Lycoctonum (S1 Table). In order to settle the dispute over the phylogenetic relationships of A. fletcheranum and A. moschatum, we further sampled 34 species from subgen. Aconitum, which represent four of its five sections according to Tamura [1] (sect. Austrokoreensia Nakai, not included herein, is unispecific, including only A. austrokoreense Koidz. from southern Korea). Based on the results of Jabbour and Renner [17] and Wang et al. [4], Gymnaconitum gymnandrum (Maxim.) Wei Wang & Z. D. Chen, the single species in Gymnaconitum, and three species of Delphinium were selected as outgroups.

Phylogenetic analyses
Phylogenetic analyses of the nrDNA datasets, the chloroplast datasets, and the combined datasets, were conducted using PAUP v.4.0b10 [24], GARLI (genetic algorithm for rapid likelihood inference) v.2.0 [25] and MrBayes v.3.2.1 [26]. Maximum parsimony (MP) searches were performed using heuristic search methods with tree bisection reconnection (TBR) branch swapping, and equal weighting of all characters. The analyses were repeated 1,000 times with a random order of sequence addition in an attempt to sample multiple islands of most parsimonious trees. Bootstrap tests were carried out to evaluate node support using 1,000 replicates with heuristic search settings identical to those for the original search. We determined the best-fit model of sequence evolution using the program Modeltest v.3.7 [27]. Maximum likelihood (ML) searches were carried out in GARLI v.2.0 using models selected by the Akaike information criterion (AIC) for each dataset. GARLI was run with eight replicates, using the default settings. The topology with the highest likelihood score was chosen as the best tree. For statistical support of branches, non-parametric bootstrap values were computed with 100 replicates, and support values were calculated using PAUP v.4.0b10. Bayesian inferences (BI) were conducted using the different models selected from Modeltest for each partition. Ten million generations were run to estimate parameters relating to sequence evolution and likelihood probabilities using a Markov chain Monte Carlo (MCMC) method. Trees were collected every 1000th generation. Convergence of runs was tested by inspecting whether the standard deviation of split frequencies of the runs was < 0.01 and by using the effective sample sizes (ESS) as calculated with Tracer v.1.4 [28], considering ESS values > 200 as good evidence. After removing 25% of the generations as burn in, a 50% majority rule consensus tree was calculated to generate a posterior probability for each node.

Incongruence tests
To evaluate the congruence of datasets from different gene markers we employed the incongruence length difference (ILD) test [29] implemented in PAUP v.4.0b10 [24]. We used simple taxon addition, TBR branch swapping, and heuristic searches with 999 repartitions of the data. The ILD test was carried out with pairwise partition for each gene dataset as well as with the combined dataset. P-values below 0.05 were considered as evidence of significant incongruence [29]. As the ILD test suggested a significant difference between the cpDNA and nrDNA data, we visually compared the cpDNA and nrDNA trees and located five samples that were incongruently placed with strong support. These include Aconitum apetalum (Huth) B. Fedtsch. ex Steinb., two accessions of A. barbatum Pers. var. barbatum (ZY69 and GQ150), A. fletcheranum, and A. gigas var. hondoense Nakai ex Tamura & Lauener. Both Wilcoxon signed-ranks (WSR) test [30,31] and approximately unbiased (AU) test [32] were further employed to assess the level of contribution of these samples to the conflict between the cpDNA and nrDNA data. Aconitum moschatum was of special interest in the phylogenetic position and thus also subjected to the WSR and AU tests, although its sister relationship to subgen. Aconitum in the nrDNA tree did not receive strong support. These six samples were first pruned from the original nrDNA and cpDNA datasets and then re-added individually. For each sample, the relationship inferred from one dataset was used as a constraint topology to test against the alternative one inferred from the other data-set. For WSR tests, PAUP v.4.0b10 [24] was employed to optimize the constraint topologies using MP approach. Before AU tests, GARLI v.2.0 [25] was used to optimize the constraint topologies, and then to calculate the site-log-likelihood values for both the best and the optimized constraint trees. The AU tests were implemented in CONSEL v.0.2 [33] using the default settings. P-values below 0.05 were considered to indicate significant differences.

Phylogenetic analyses of chloroplast sequence data
There was no significant incongruence among the four chloroplast regions and the p-values resulting from the ILD test of pairwise sequences are shown in Table 3. Therefore, we combined all markers into a single dataset. For the combined cpDNA dataset, information of each aligned DNA data, tree statistics for the MP analysis and the best-fit model of each region are given in Table 4. The analyses of three approaches (MP, ML, and BI) revealed largely congruent tree topologies and the ML tree is shown in Fig 1 with support values. The monophyly of the genus Aconitum was strongly supported (PP/MP/ML = 1.00/96%/99%). However, the monophyly of subgen. Lycoctonum as currently circumscribed was not supported since its two members, A. moschatum (the single species of sect. Galeata) and A. fletcheranum (the single species of sect. Fletcherum), did not cluster together with other members in the same clade. Aconitum moschatum appeared to be sister to all the other species of Aconitum (PP/MP/ML = 1.00/91%/96%), whereas A. fletcheranum formed a clade together with species of subgen. Aconitum (PP/MP/ ML = 1.00/82%/79%) and held a sister position to this subgenus, although with weak support (PP/MP = 0.77/52%). Except for A. moschatum and A. fletcheranum, all the remaining taxa of subgen. Lycoctonum formed a well-supported clade (PP/MP/ML = 1.00/99%/97%). Aconitum novoluridum (the single species of sect. Altospermum) was the first diverging lineage of the clade. The section Lycoctonum was resolved as a monophyletic group, although with somewhat

Phylogenetic analyses of nuclear ribosomal DNA sequence data
The p-value resulting from the ILD test between nrDNA ETS dataset and nrDNA ITS dataset showed no significant incongruence (Table 3). We thus combined them in the phylogenetic analyses. For the combined nrDNA dataset, information of each aligned DNA data, tree statistics for the MP analysis and the best-fit model of each region are given in Table 4.
All the three analyses revealed approximately congruent tree topologies and the ML tree is shown in Fig 2 with support values. The monophyletic status of the genus Aconitum was confirmed with strong support values (PP/MP/ML = 1.00/100%/99%). However, as the case with cpDNA dataset, the monophyly of subgen. Lycoctonum as currently circumscribed was not supported. Its member A. moschatum was resolved as the sister to the subgenus Aconitum clade (PP/MP/ML = 0.99/91%/73%), although with relatively weak support (PP/MP/ ML = 0.71/68%/63%). The remaining species of subgen. Lycoctonum formed a highly supported clade (PP/MP/ML = 0.96/94%/90%). Within this clade, A. fletcheranum and A. novoluridum were the first two species successively diverging. The section Lycoctonum was again resolved as a monophyletic group (PP/MP/ML = 1.00/66%/80%). Within this section, five clades with moderate to strong support were resolved. Comparison of the series classification by Tamura [1] on this section (Table 1) with our ML tree indicates that all the series recognized by him were para-or polyphyletic, although species of ser. Volubilia and of ser. Lycoctonia were almost nested together in a well-supported clade of their own respectively. The series Umbrosa was also not retrieved. It is noteworthy that the placements of A. apetalum, two accessions of A. barbatum var. barbatum, and A. gigas var. hondoense in the ML tree resulting from nrDNA dataset were different from their placements in the ML tree resulting from cpDNA dataset.

Phylogenetic analyses of combined cpDNA and nrDNA data
The ILD test indicated strong incongruence between nuclear markers and chloroplast markers with a p-value of 0.001 (Table 3). The results of the WSR and AU tests are provided in Table 5. For Aconitum gigas var. hondoense and two accessions of A. barbatum var. barbatum (ZY69 and GQ150), all the tests suggested significant differences between the relationships inferred respectively from cpDNA and nrDNA datasets, indicating that they contributed greatly to the conflict. These samples were thus included in phylogenetic analyses as two entries, once as a cpDNA-only entry and once as an nrDNA-only entry. For A. apetalum, A. fletcheranum and A. moschatum, at least the tests of the nrDNA dataset showed no significant difference between the relationships inferred from the cpDNA and nrDNA datasets, suggesting that they should not be the main causes of the conflict between the two datasets. We thus combined their nrDNA and cpDNA sequences for phylogenetic analyses. However, it is to be noted that, even after excluding the six samples mentioned above, the p-value of the ILD test between the cpDNA and nrDNA datasets was still less than 0.001 (Table 4). A visual comparison showed that both the original (containing all the six samples) and pruned (containing only each of the six samples) datasets suggested the same positions for the six samples (results not shown here), indicating that their placements were not influenced by the other five samples.
All the three analyses revealed approximately congruent tree topologies and the ML tree is shown in Fig 3 with support values. The genus Aconitum was confirmed again to be monophyletic with strong support values (PP/MP/ML = 1.00/100%/100%). The sister relationship of A. moschatum with the remaining species of Aconitum was supported by both BI and ML analyses (PP/ML = 1.00/91%). Aconitum fletcheranum was sister to subgen. Lycoctonum (PP/ MP = 0.78/82%). The rest of species in subgen. Lycoctonum formed a well-supported clade (PP/MP/ML = 1.00/100%/100%). Aconitum novoluridum was the first split lineage. The monophyly of sect. Lycoctonum was well supported (PP/MP/ML = 1.00/93%/94%). Within this section, five major highly supported clades (Clades A-E) were resolved. These clades conformed largely to those revealed by the nrDNA dataset in taxon composition if regardless of those accessions with incongruent placements as revealed by nrDNA and cpDNA markers. All the series recognized by Tamura [1] were para-or poly-phyletic when superimposed on the ML tree (Fig 3), although Clade C consisted of taxa mostly from ser. Volubilia and Clade E was comprised of those mostly from ser. Lycoctonia. The series Umbrosa was also not recovered here, with its type species, A. umbrosum, deeply embedded in Clade D.

Discussion
Phylogenetic position and taxonomic status of Aconitum sect. Fletcherum Aconitum sect. Fletcherum is unispecific, including only A. fletcheranum distributed in the eastern Himalayan region (Bhutan, southeastern Xizang in China, and Assam in northern India) [1-3, 9, 34-36]. When Tamura [34] established the section, he stated that the species was very peculiar in the short and scapose stem, the single flower terminal to the stem which was contrary to the indeterminate raceme in all other species of subgen. Lycoctonum, and in the navicular upper sepal and the 6-8 carpels as well. Lauener and Tamura [2], Wang [3], and Tamura [1] all accepted this section and its affiliation with subgen. Lycoctonum, but Kadota [9] transferred A. fletcheranum from subgen. Lycoctonum to his subgen. Tangutica, stressing that the nectary blade of the petal in this species, contrary to the other species (except for A. moschatum) within subgen. Lycoctonum but the same as members within subgen. Tangutica, was not provided with a tubular portion.
Most notably, the chromosome number of 2n = 12 in Aconitum fletcheranum, which represents a new base chromosome number of x = 6 for the genus Aconitum [10], is remarkably aberrant in the genus which otherwise has uniformly a base chromosome number of x = 8 [1,13,[37][38][39][40][41][42][43][44][45]. Moreover, the karyotypic constitution of this species is also unique in the genus Aconitum. There are two largest metacentric chromosome pairs in this species, whereas in all the other diploid taxa of Aconitum there is only one such chromosome pair, with the second largest pair being submetacentric [10]. Hong et al. [10] considered that the base number of x = 6 in Aconitum may have originated from x = 8 (descending dysploidy) through asymmetric reciprocal translocations of some telocentric chromosomes. All these further suggest the abnormality of A. fletcheranum within subgen. Lycoctonum. In our analyses, Aconitum fletcheranum is revealed to form a sister relationship to subgen. Aconitum in the cpDNA tree (PP/MP/ML = 1.00/82%/79% in Fig 1) and to all the remaining species of subgen. Lycoctonum in the nrDNA tree (PP/MP/ML = 0.96/94%/90% in Fig 2), respectively. There are several causes which may be invoked to account for this incongruence. All the topology tests show no significant difference between these two phylogenetic hypotheses, suggesting that stochastic errors, which are caused mainly by finite sequences used in studies, could not be ruled out confidently [46,47]. Hybridization, probably combined with some other biological factors (reviewed in [48][49][50][51][52]), seems to be a more reasonable explanation for the incongruence of A. fletcheranum in view of its conflicting placements between the cpDNA and nrDNA trees, its aberrant chromosome number of 2n = 12 and unique karyotypic constitution. The signs of hybridization are nevertheless not evident in the nrDNA sequences of A. fletcheranum. We have failed to amplify the ETS sequence of A. fletcheranum, but we have successfully sequenced the ITS sequences of five individuals of this species. The chromatograms all contain well-formed, distinctive single peaks (no overlapping peaks) with very little background "noise", and the sequences are different by only two nucleotides. It is regrettable that only one population of this species has been available for the present study. Sampling of more populations and further analyses using single-or low-copy nuclear genes are expected to reveal a more comprehensive and convincing pattern for this incongruence. Systematic errors, such as long-branch attraction, which result primarily from model misspecification [46], seem to be less likely a cause for the incongruence of A. fletcheranum. This is indicated by the results that, for each dataset, all the analyses (MP, ML, and BI) suggest the same placement for A. fletcheranum and that the branch length is generally even. Moreover, the positions of A. fletcheranum are stable in the cpDNA and nrDNA trees, at least not influenced by A. apetalum, the two accessions of A. barbatum var. barbatum (ZY69 and GQ150), A. gigas var. hondoense, and A. moschatum.
Our analyses of the combined data place Aconitum fletcheranum as a sister to all the remaining species of subgen. Lycoctonum (PP/MP = 0.78/82% in Fig 3), the same position as suggested by the nrDNA data. Although further studies are still needed to disentangle its evolutionary history, A. fletcheranum should deserve a subgeneric status in the genus Aconitum given that it always occupies a sister position to subgen. Aconitum (Fig 1) or to all the remaining taxa in subgen. Lycoctonum (Figs 2 and 3) and that it has a unique array of morphological and cytological characters.
Our molecular work lends no support to the establishment of Aconitum subgen. Tangutica. In all our analyses, A. tanguticum (Maxim.) Stapf, the type species of this subgenus, is always embedded in subgen. Aconitum (S1-S3 Figs), conforming to the results of previous molecular works on the genus Aconitum [15] or on the tribe Delphinieae [17]. Karyological data also indicate the membership of A. tanguticum within subgen. Aconitum. This species has the same chromosome number (2n = 16) and chromosome morphology as other taxa within subgen. Aconitum [44].

Phylogenetic position and taxonomic status of Aconitum sect. Galeata
Aconitum sect. Galeata is also unispecific, including only A. moschatum endemic in the alpine zone of Kashmir [1,2,9,34,53]. Tamura [1,34] and Lauener and Tamura [2] accepted this section and its affiliation with subgen. Lycoctonum, although they stated that A. moschatum was very particular within subgen. Lycoctonum by having the navicular or depressed galeate helmet broader than long, lurid purple flowers, and the very obtuse spur of the petal, not opposite to the labium but a continuation of the stalk. Kadota [9], however, pointed out that A. moschatum should not be placed in subgen. Lycoctonum, stressing that its nectary blade of the petal, contrary to the other species (except for A. fletcheranum) within the subgenus, was not equipped with a tubular portion. He left the problem aside for a further study. The chromosome number of A. moschatum was reported to be 2n = 16 [54] but regrettably the chromosome morphology has yet been unknown.
In our ML tree resulting from cpDNA dataset, Aconitum moschatum is revealed as the sister to all other members of the genus Aconitum with strong support (PP/MP/ML = 1.00/91%/96% in Fig 1). In the ML tree resulting from nrDNA dataset, however, it is resolved as the sister to subgen. Aconitum, but with weak support (PP/MP/ML = 0.71/68%/63% in Fig 2). All the topology tests, except for the WSR test of the cpDNA data, show no significant difference between the two phylogenetic hypotheses (Table 5). This may suggest that the incongruence of A. moschatum between cpDNA and nrDNA is not a "hard" one [55]. Similar to the case with A. fletcheranum, systematic errors cannot be invoked to account for this incongruence due to the stable positions suggested by all the analyses (the MP, ML, and BI analyses, and the analyses based on complete or pruned datasets) and the relatively short branch length as well. We thus combine the nrDNA and cpDNA data of A. moschatum for phylogenetic analyses. The analyses of combined data show that this species is the sister to all other species of Aconitum (PP/ML = 1.00/91% in Fig 3), the same relationship as suggested by cpDNA dataset.
In summary, our phylogenetic analyses indicate that Aconitum moschatum is not closely related to any of the major groups of Aconitum revealed here. Taking into account its unique array of morphological characters, we consider it justifiable to remove sect. Galeata from subgen. Lycoctonum and treat it as an independent subgenus of its own in the genus Aconitum.
Phylogenetic position and taxonomic status of Aconitum sect. Alatospermum Aconitum sect. Alatospermum is also unispecific, including only A. novoluridum distributed in the eastern Himalayan region (Bhutan, southeastern Xizang in China, Sikkim in northern India, and Nepal) [1-3, 34, 35]. When establishing the section under subgen. Lycoctonum based on this species, Tamura [34] pointed out that the species was clearly distinguishable from other species of the subgenus by the depressed hemi-elliptic helmet gradually descending into the long beak and much broader than long, the hammer-shaped petals with the nectary blade at a right angle to the stalk and longer than the stalk, the longitudinally winged seeds along the three ridges, and the lurid, reddish, brownish red or purple flowers. This section has since been recognized by Lauener and Tamura [2], Tamura [1], and Kadota [35] but rejected by Wang [3], who placed A. novoluridum in sect. Lycoctonum ("Paraconitum").
Both the sectional status and subgeneric affiliation of Aconitum novoluridum are strongly supported by our molecular analyses. In our ML trees resulting from cpDNA (Fig 1), nrDNA (Fig 2) or the combined cpDNA and nrDNA dataset (Fig 3), this species is always resolved as the first split lineage in subgen. Lycoctonum and forms a sister relationship to sect.

Lycoctonum.
Tamura [1] regarded Aconitum sect. Alatospermum, which has the seeds without transverse squamae and the hemi-elliptic upper sepal, as a primitive group in subgen. Lycoctonum, listing it as the first section in this subgenus (Tamura and Lauener [6] had previously listed sect. Galeata ("Galeatum") as the first section). The primitive condition of A. novoluridum in seed morphology is confirmed by Kong et al. [56]. According to Hong et al. [10], the karyotype of A. novoluridum is the most symmetric in the subgenus and thus very probably represents a primitive condition. The primitive condition of A. novoluridum in subgen. Lycoctonum is also supported by our molecular work. This species is always the first split lineage in the subgenus in our ML trees resulting from different datasets.
Our molecular work, therefore, strongly favors the treatment of Aconitum novoluridum as an independent section within subgen. Lycoctonum, i.e. sect. Alatospermum, lending no support to its placement in sect. Lycoctonum. Evidence from morphology, karyology, and molecular phylogeny all suggests the primitive condition of sect. Alatospermum within subgen. Lycoctonum.

Phylogenetic relationships within Aconitum sect. Lycoctonum and its reclassification
Our molecular work indicates that Aconitum sect. Lycoctonum is a monophyletic group within subgen. Lycoctonum (PP/MP/ML = 0.71/72%/60% in Fig 1; PP/MP/ML = 1.00/66%/80% in  Fig 3), agreeing with the results of past molecular analyses involving this section [4,17]. All the nine series recognized by Tamura [1] and ser. Umbrosa proposed additionally by Kadota [7] in this section, however, are revealed to be polyor para-phyletic in our analyses of the combined cpDNA and nrDNA dataset (Fig 3). Five major clades are nevertheless recovered within the section, and they can be conveniently treated at series rank: ser. Crassiflora, ser. Scaposa, ser. Volubilia, ser. Longicassidata, and ser. Lycoctonia.
Aconitum ser. Scaposa includes only A. scaposum (Clade A: PP/MP/ML = 1.00/100%/100% in Fig 3). This species, fairly widespread in central and southwestern China [3] and recently reported, under the name A. chloranthum Hand.-Mazz., to occur also in Bhutan [35], shows great variation in the relative development of basal and cauline leaves, the leaf shape, pedicel pubescence, and flower color. Several varieties were once described in this species, and some of them were even recognized as independent species [2,3,57]. Yang [11] and Luo and Yang [12] treated A. scaposum as a polymorphic species including all the varieties and species. The four accessions that we have chosen in our molecular analyses represent two types of plants in respect of the relative development of basal and cauline leaves: two accessions (LJP206 from Sichuan, GQ92 from Gansu) have cauline leaves nearly aggregated in the middle part of the stem, while the other two accessions (GQ247 from Hubei, LJP78 from Chongqing) have cauline leaves nearly equally distantly arranged along the stem. All these accessions are nested together with each other in the same clade.
Aconitum ser. Crassiflora includes six taxa: A. apetalum, A. brevicalcaratum var. brevicalcaratum, A. brevicalcaratum var. parviflorum, A. crassiflorum, A. chrysotrichum, and A. rilongense (Clade B: PP/MP/ML = 1.00/52%/84% in Fig 3). The close affinity of the latter five taxa is also indicated by gross morphology, geographical distribution, and in particular, karyology. They are very similar to each other in general appearance [11,12,58] and all concentrated in the southern part of the Hengduan Mountains region in southwestern China [10]. They are all tetraploid (2n = 32) and share similar chromosome size and karyotype constitution [10]. The close affinity between A. scaposum (a diploid species with 2n = 16) and A. crassiflorum as previously regarded by Tamura [1] is not supported by our molecular analyses. In fact, no diploid species are clustered with these five tetraploid taxa. All the five tetraploid taxa in this series might have originated through only one polyploidization event, but we have been unable to ascertain their type of polyploidy (autopolyploidy vs. allopolyploidy) and parental origin from morphological, karyological and molecular data currently available.
Aconitum apetalum is distributed in central Asia (Xinjiang in China, Kazakhstan, and Tajikistan). It is readily distinguishable by its long, many-flowered raceme, very small flowers, and petals with a short, capitate spur, and the upper sepal narrowly cylindrical with a prominent peak [1,6]. Tamura [1] placed it in ser. Micrantha together with A. brevicalcaratum, A. chrysotrichum, and A. sajanense Kumin mainly due to their short spur of petals, but in our molecular analyses A. brevicalcaratum and A. chrysotrichum are shown to be members in ser. Crassiflora, while A. sajanense, a diploid species with 2n = 16 [59], is shown to be a member in ser. Lycoctonia (Fig 3). The results of our molecular analyses of A. apetalum are somewhat similar to those of A. fletcheranum. As a hexaploid species (2n = 48), A. apetalum is nested with those tetraploid taxa (2n = 32) within ser. Crassiflora in the cpDNA tree (PP/MP/ML = 1.00/ 95%/95% in Fig 1) and sister to A. scaposum (2n = 16) in the nrDNA tree (PP/MP/ML = 0.98/ 87%/93% in Fig 2). This seems to strongly support a hybrid origin followed by subsequent polyploidization for A. apetalum. However, the chromatograms of ITS and ETS of A. apetalum, with only distinctive single peaks, show no obvious evidence of hybridization. The WSR and AU tests of the nrDNA data do not significantly reject the hypothesis inferred from the cpDNA data, further indicating the probable existence of a stochastic error. We thus also combine the cpDNA and nrDNA sequences of A. apetalum for phylogenetic analyses. The analyses place A. apetalum sister to all the tetraploid taxa from the Hengduan Mountains region, a relationship similar to that suggested by the cpDNA data. We therefore tentatively refer A. apetalum to ser. Crassiflora. It is worth mentioning that our molecular results do not support the reduction of A. monticola Steinb., also a central Asian species, to the synonymy of A. apetalum [60]. Both of them are nested in different clades (Fig 3).
Aconitum ser. Volubilia defined by Tamura and Lauener [6] and Tamura [1] is largely supported by our molecular work (Figs 1-3). Nine species traditionally classified in this series are clustered in this clade together with A. angustius (Clade C: PP/MP/ML = 1.00/92%/94% in Fig  3), a species mainly distributed in southeastern and central China and previously placed, under the name A. sinomontanum Nakai var. angustius W. T. Wang, in ser. Lycoctonia [6]. Aconitum angustius has long been considered to be most closely related to A. sinomontanum [57] or even treated as a variety of it [3,61,62]. This is not supported by our molecular results. The two taxa are nested in different clades (Fig 3). In fact, Gao et al. [63] and Hong et al. [10] has previously revealed that A. angustius is distinct from A. sinomontanum in ploidy level (tetraploid with 2n = 32 vs. diploid with 2n = 16) and that the former is morphologically more closely similar to A. finetianum Hand.-Mazz., a diploid species mainly distributed in southeastern China and placed (as a synonym of the Japanese species A. pterocaule Koidz.) in ser. Volubilia by Tamura and Lauener [6]. Moreover, as pointed out by Gao et al. [63], some specimens of A. angustius from Guizhou and Hubei, China, had been previously misidentified as A. loczyanum Rapaics or A. pterocaule (both belonging to ser. Volubilia) by Handel-Mazzetti [64] and Tamura and Lauener [6]. From its karyotypic constitution A. angustius may be of an allopolyploid origin [63]. Considering their close morphological similarity and somewhat overlapping geographical distribution we infer that one of the parents of A. angustius is very likely A. finetianum. The origin of A. angustius is an interesting problem worthy of further investigations.
Aconitum ser. Longicassidata (Clade D: PP/MP/ML = 1.00/90%/88% in Fig 3) is the most complex group which comprises species from five different series classified by Tamura [1]: ser. Longibracteolata, ser. Longicassidata, ser. Lycoctonia, ser. Ranunculoidea, and ser. Reclinata. The series Umbrosa proposed by Kadota [7] should also be transferred to here. Among them, ser. Longibracteolata and ser. Reclinata are both unispecific. The close similarity between A. monticola and A. krylovii Steinb. (ser. Lycoctonia) in morphology has been noted by Steinberg [65]. Aconitum barbatum and A. kirinense Nakai (ser. Longicassidata) have also been considered to be closely related to each other [6]; Handel-Mazzetti [64] even placed the latter in synonymy with the former. Most notably, while Tamura and Lauener [6] established ser. Ranunculoidea to accommodate A. ranunculoides Turcz. ex Ledeb. and A. ajanense Steinb., they pointed out that the latter species was also near to A. umbrosum (ser. Lycoctonia), a species which they considered to be closely related to A. gigas var. hondoense (= A. iinumae Kadota). These opinions are largely supported by our molecular results (Fig 3). Aconitum reclinatum A. Gray is the only representative of subgen. Lycoctonum in the New World [1,6]. Our molecular phylogeny reveals it to be closely related to several Asian species, suggesting that it may have migrated from Asia to eastern North America.
It is noteworthy that two accessions of Aconitum barbatum var. barbatum (GQ150 and ZY69) and A. gigas var. hondoense show significant incongruences between the nrDNA and cpDNA datasets. The former taxon is widespread in northeastern and central China and the Far East of Russia (Primorye). Four accessions of it are included in our analyses, including one (GQ95) from Shaanxi in central China, two (GQ150 and ZY69) from Jilin in northeastern China, and one (SG786) from Primorye. On the nrDNA tree (Fig 2), the four accessions cluster together with A. gigas var. hondoense and then are nested within ser. Longicassidata (Clade D in Fig 3). On the cpDNA tree (Fig 1), GQ95 and SG786 are still grouped with species of ser. Longicassidata (although this series do not form a clade in the cpDNA tree), whereas GQ150 and ZY69 cluster with species of ser. Volubilia (Clade C in Fig 3). All the topology tests indicate significant incongruence between the two phylogenetic hypotheses of GQ150 and ZY69 (Table 5). However, both GQ150 and ZY69 are typical A. barbatum var. barbatum in morphology, showing no obvious difference from GQ95 and SG786, and no intermediates between A. barbatum var. barbatum and any species of ser. Volubilia have thus far been found. A visual examination of the nrDNA sequences of A. barbatum var. barbatum shows that there are no more than three nucleotide differences among its four accessions, and those of ZY69 (with incongruent placements) and SG786 (with congruent placement) are even totally identical. This seems to be a typical pattern of gene tree incongruence caused by chloroplast capture: the cytoplasm of GQ150 and ZY69 has been replaced by that of a certain member of ser. Volubilia probably via introgression [66][67][68]. This hypothesis is further indicated by the geographical distribution of A. barbatum var. barbatum and species of ser. Volubilia. Members in ser. Volubilia occur mainly in southeastern and northeastern China, the Korean Peninsula, Japan, and the Far East of Russia, with their distribution largely overlapping with that of A. barbatum var. barbatum. In the field A. barbatum var. barbatum is often found to grow in the neighborhood of some species of ser. Volubilia, e.g., A. alboviolaceum Kom. (pers. observ.).
Similar to the case with the two accessions of Aconitum barbatum var. barbatum, GQ150 and ZY69, A. gigas var. hondoense is nested in ser. Longicassidata on the nrDNA tree (Fig 2) but in ser. Volubilia on the cpDNA tree (Fig 1). The topology tests also indicate significant incongruence between the two phylogenetic hypotheses (Table 5). Regrettably only one accession of this variety is included in this study. A more extensive sampling is needed to explore the overall phylogenetic pattern of this taxon and the exact cause(s) of the incongruence of the accession sampled here. From a morphological perspective, A. gigas var. hondoense is closely related to A. umbrosum, and thus should be placed in ser. Longicassidata construed here.

Taxonomic treatment
The first formal phylogeny-based classification of Aconitum subgen. Lycoctonum is presented below (also see Fig 3), which involves segregating both sect. Galeata and sect. Fletcherum from this subgenus as two independent subgenera of their own within the genus Aconitum, reinstating one series (ser. Crassiflora) and abolishing six (ser. Laevia, ser. Longibracteolata, ser. Micrantha, ser. Ranunculoidea, ser. Reclinata, and ser. Umbrosa) within sect. Lycoctonum. The series affiliation of some species within the section is adjusted accordingly. We include only the more significant synonyms, for a more complete synonymy see Tamura [1].
Note. This series has been fairly well defined previously by Tamura and Lauener [6] and Tamura [1]. Although A. pteropus Nakai, a species from Korea, is not included in our molecular analyses because of unavailability of DNA material, we agree with Tamura and Lauener [6] that it should belong to this series from a morphological perspective. Morphologically this species is closely similar to A. pterocaule. Series  Description. Stem erect, subscapose or leaning, sometimes trailing. Leaves obicular-reniform or reniform-pentagonal, 3-parted or subpedatifid to partite, central leaf segment broadly rhombic or cuneate-rhombic, sometimes 3-parted nearly to midvein, lateral segments obliquely flabellate. Inflorescence lax or densely racemiform; pedicels spreading or appressed pubescent; bracteoles below the middle or near the base of pedicels. Flowers yellow, white or pale blue-violet; upper sepal elongate conical, cylindrical or tubulose, often with a beak; petal lip linear, conspicuous, spur short or elongate, circinate or semi-coiled, shorter than, nearly as long as or longer than the lip.
Note. This series is greatly expanded to include several species which have been previously placed in four other series by Tamura and Lauener [6] and Tamura [1]: Aconitum reclinatum (the single species in ser. Reclinata), A. sukaczevii (the single species in ser. Longibracteolata), A. ajanense, A. ranunculoides (both in ser. Ranunculoidea), A. umbrosum, A. monticola and A. krylovii (all in ser. Lycoctonia). Another two species, including A. crassifolium Steinb. (ser. Ranunculoidea) from the Far East of Russia and A. puchonroenicum Uyeki & Satake (ser. Lycoctonia) from Korea, should also belong to this series, but this needs to be verified by using molecular data.
Among the nine species (Aconitum asahikawaense Kadota, A. gigas, A. hiroshi-igarashii Kadota, A. ikedae Kadota, A. mashikense, A. soyaense Kadota, A. tatewaki Miyabe, A. umezawae Kadota, A. umbrosum) placed by Kadota [7,8] in his ser. Umbrosa, only A. gigas var. hondoense and A. umbrosum are included in our molecular analyses. It is somewhat strange to us that when Kadota [8]  Note. Although the Iranian species Aconitum iranshahri H. Riedl is not included in our molecular analyses, it should belong to this series from its close morphological similarity with A. orientale.