Evolution of Reproductive Traits and Implications for Adaptation and Diversiﬁcation in the Yam Genus Dioscorea L.

: Dioscorea is a pantropical monocotyledonous genus encompassing several well-known tuber crops and medicinal plants. It possesses remarkable morphological diversity, especially in reproductive characteristics, which are suggested to play important roles in species adaptation and diversiﬁcation. Yet there have been few studies that consider the evolutionary pattern followed by these characters in this genus. In this study, the phylogenetic relationships among Chinese yams were reconstructed from ﬁve chloroplast and two mitochondrial DNA sequences. The evolutionary histories of bulbil possession, inﬂorescence architecture, the color of the male ﬂowers and the degree of male ﬂower opening were reconstructed. The results suggested that yam bulbils evolved after the divergence between D . sect. Testudinaria and other species of Dioscorea except for in D . sect. Stenophora and D . sect. Apodostemon . The evolutionary trend in the degree of male ﬂower opening ranged from fully open to nearly closed. Male ﬂowers with dark colors and panicles were shown to be derived in Dioscorea . These characteristics were found to be closely associated with the reproductive patterns and pollinating mechanisms of the Dioscorea species. The ﬁndings also shed light on the systematic relationships within this genus.


Introduction
Dioscorea L. is the largest genus of the Dioscoreaceae, comprising about 630 species [1], mainly distributed in tropical and subtropical areas, sometimes expanding to temperate regions [2]. Species of this genus are mostly dioecious vines with underground storage organs [3]. The edible starchy tuber makes it the third most important tropical tuberous crop globally [4], and the metabolite-rich rhizomes of some species are used as a source of pharmaceutical compounds [5]. Apart from its economic importance, Dioscorea is one of the most critical taxa in monocot systematics since it is laid near the basal position in the phylogenetic tree of monocotyledonous plants and has a number of similarities with dicots [6,7]. However, there are challenges in the taxonomic and systematic investigation of Dioscorea due to its great morphological diversity, dioecy and small flowers [2,8]. There are about 1600 taxonomic names attributed to Dioscorea, but most of them are considered synonyms [1], indicating the controversy on the circumscription of species boundaries. Moreover, the classification system of Dioscorea differs greatly among authors, and many of the proposed infrageneric taxa do not quite represent natural lineages [2]. Based on morphological and anatomic characteristics, Dioscorea species have been assigned to between 24 and 58 sections in traditional taxonomic studies [3,[9][10][11]. Recent phylogenetic analyses based on molecular data sets, including cpDNA regions and nuclear genes, have led to the recognition of 10-11 main clades in this genus [2,[12][13][14][15]. The results greatly simplified previous classification

Sampling
A total of 48 Dioscorea taxa were sampled in this study, covering all sections distributed in China, except the monotypic section D. sect. Stenocorea. The samples consisted of 43 species, 1 subspecies and 4 varieties. Among them, 47 taxa were collected in the field and the vouchers were deposited in the Herbarium of the Institute of Botany, Jiangsu Province and the Chinese Academy of Sciences, China (NAS). The sequences of the remaining species, D. wallichii J. D. Hooker, were downloaded from GenBank. We selected Tacca chantrieri Andre (Dioscoreaceae), a representative of the genus closely related to Dioscorea according to the recently built phylogeny of the Dioscoreaceae [14], as the outgroup in our analyses. The geographical origin, voucher specimen information and Gen-Bank accession numbers of all samples are listed in Table 1.

Sampling
A total of 48 Dioscorea taxa were sampled in this study, covering all sections distributed in China, except the monotypic section D. sect. Stenocorea. The samples consisted of 43 species, 1 subspecies and 4 varieties. Among them, 47 taxa were collected in the field and the vouchers were deposited in the Herbarium of the Institute of Botany, Jiangsu Province and the Chinese Academy of Sciences, China (NAS). The sequences of the remaining species, D. wallichii J. D. Hooker, were downloaded from GenBank. We selected Tacca chantrieri Andre (Dioscoreaceae), a representative of the genus closely related to Dioscorea according to the recently built phylogeny of the Dioscoreaceae [14], as the outgroup in our analyses. The geographical origin, voucher specimen information and GenBank accession numbers of all samples are listed in Table 1.

DNA Extraction, Amplification and Sequencing
The total genomic DNA was extracted from fresh or silica-gel dried leaves following a modified CTAB method [23] and stored at −20 • C before amplification. Five chloroplast markers, matK, rbcL, trnL-F, psbA-trnH and rpl36-rps8, plus two mitochondrial markers, nad1 and rps3, were chosen, taking evolutionary rate and ease of sequencing into account. The primers used for matK and rbcL amplification followed Gao et al.'s recommendations [24]. The amplification of trnL-F used primers described by Taberlet et al. [25]. The primers for psbA-trnH and rpl36-rps8 were newly designed based on homologous sequences of D. elephantipes (L'Hér.) Engl. on Gen-Bank. The primers for the nad1 gene were designed based on the sequence of Oryza sativa Linn. The rps3 gene was amplified according to Laroche and Bousquet [26]. The sequences of all primers used in this study are detailed in Table 2. Table 2. Markers and primers used in this study.

Marker
Primer Name Direction Primer Sequence (5 to 3 ) The PCR reaction mixture contained 40 ng of genomic DNA template, 2.5 mmol/L MgCl 2 , 1 × Mg-free DNA polymerase buffer, 0.12 mmol/L dNTPs, 0.3 mmol/L of each primer, 1 U Taq DNA polymerase and water added accordingly to a final volume of 50 µL. The PCR program for matK, rbcL, trnL-F, nad1 and rps3 was as follows: a 3 min premelt at 94 • C, followed by 35 cycles of 45 s denaturation at 94 • C, 30 s annealing at 58 • C and a 1.5 min extension at 72 • C, plus a final extension of 5 min at 72 • C. For psbA-trnH and rpl36-rps8, the PCR reaction included an initial denaturation at 94 • C for 3 min, followed by 35 cycles of denaturation at 94 • C for 30 s, annealing at 54 • C for 30 s, extension at 72 • C for 80 s and a final extension at 72 • C for 5 min. The PCR products were examined electrophoretically using 0.8-1.2% agarose gels and purified using a TIAN gel Midi Purification Kit (TIANGEN Biotech, Beijing, China). The purified products were sequenced using the same primer pairs with PCR. All of the sequences obtained were subjected to a BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi (accessed on 15 April 2020)) search to detect contamination and nonspecific amplification. After confirmation, the newly generated sequences were submitted to GenBank.

Phylogenetic Analyses
The assembled sequences were aligned using MAFFT v.7 [27] with default settings and then adjusted manually for accuracy in Geneious R9 9.1.8 (https://www.geneious.com (accessed on 4 July 2020)). Character gaps were treated as missing data. Phylogenetic relationships were reconstructed using maximum likelihood (ML), maximum parsimony (MP) and Bayesian inference (BI) approaches. ML analysis was firstly performed to build a single-gene tree in online CIPRES Science Gateway v.3.3 (http://www.phylo.org/ (accessed on 18 Octobor 2020)) [28], using RAxML-HPC v8.2.10 [29,30] under the GTR + G DNA substitution model [31]. One hundred rapid bootstrap replicates were generated for each single gene matrix [32]. Since there was no strong conflict among individual markers, we conducted further ML analysis using a concatenated matrix of all seven loci under the GTR + G model as recommended by jModelTest v2.1 [33]. Bootstrap analyses were used to evaluate the support for each clade with 1000 bootstrap replicates. MP analysis was conducted in PAUP* version 4.0b10 [34]. All characters were equally weighted. Trees were inferred using the heuristic search option with tree bisection probabilities (TBR) swapping and 1000 replicates of random addition. Ten trees were held in each step during stepwise addition. The maximum number of trees was set to 10,000, and all parsimonious trees were saved. The tree length (TL), consistency index (CI), retention index (RI) and rescaled consistency index (RC) were calculated for each maximum parsimony tree. The bootstrap results were summarized in a 50% majority-rule consensus cladogram. BI analysis was implemented using MrBayes version 3.2.6. [35]. A Markov Chain Monte Carlo (MCMC) analysis was run with two independent chains with a random starting tree for 100 million generations, sampling one tree every 1000 generations. The first 25% generations were discarded as burn-in, and the remaining trees were used to construct a consensus tree with a 50% majority rule and obtain the posterior probabilities.

Character Coding and Ancestral State Reconstruction
Four characters closely related to vegetative or sexual reproduction were selected for ancestral character state reconstruction, namely bulbil formation, inflorescence structure, floral color and opening degree of the perianth. Morphological data were obtained from direct observations and the literature. The evolutionary history of each of the four characters was traced over the Bayesian 50% majority-rule tree using MP approaches available in Mesquite 3.5.1 [36]. The character states were treated as unordered and equally weighted. Morphological characters and their states were coded as follows: a. bulbil absence (0), presence (1); b. inflorescence umbel (0), spike (1), raceme (2) or panicle (3); c. floral color dark (e.g., violet, purplish red, orange) (0) or light (e.g., white, pale yellow, green) (1); d.
perianth opening degree open (0), half-closed (1) or closed (2). For species in which the floral color changed during blooming, we used the color at the mature stage as the character state. The resulting codes for the sampled species are summarized in Table 3. Table 3. Data matrix of morphological characters used in this study.

Species
Character

Phylogenetic Analyses
According to our analyses, the chloroplast segment trnL-F exhibited the highest variability among the seven molecular markers, followed by matK and nad1 (see Table 4). The final concatenated matrix of all seven markers consisted of 7218 base pairs containing 1157 variable sites and 614 parsimony-informative sites. The molecular reconstruction based on the combined matrix gave rise to a highly resolved phylogenetic tree with moderate to strong support. There was no significant incongruence in topology or support values among trees generated from different methods. Thus, only the ML tree is presented in Figure 2.

Ancestral Character State Analyses
The ancestral state of four reproductive features of the Dioscorea species was reconstructed based on MP methods. As for character "a" the absence or presence of bulbils at the axil), state "0" (absent) was suggested to be the plesiomorphic state for Dioscorea The ancestral state of four reproductive features of the Dioscorea species was reconstructed based on MP methods. As for character "a" the absence or presence of bulbils at the axil), state "0" (absent) was suggested to be the plesiomorphic state for Dioscorea (Figure 3). No species of the outgroup Tacca and D. sect. Stenophora, the earliest diverging group in Dioscorea, produces bulbils. The trait distribution in the phylogenetic tree suggests that bulbils appeared after the divergence of D. sect. Stenophora, but were subsequently lost several times independently, in D. sect. Combilium, D. sect. Lasiophyton and several species in D. sect. Enantiophyllum (D. wallichii Hook. f., D. aspersa Prain et Burkill). Bulbils persisted in most species in D. sect. Botryosicyos, D. sect. Opsophyton and D. sect. Enantiophyllum. The ancestral state of the inflorescence architecture was inferred to be a spike ( Figure  4). Racemes and panicles were inferred to be derived states, with the former occurring sporadically in different clades and the latter occurring mainly in D. sect. Botryosicyos and D. sect. Enantiophyllum. Spike occurred as a homoplasy in D. sect. Shannicorea, while other clades did not show a consistent type of male inflorescence. The ancestral state of the inflorescence architecture was inferred to be a spike (Figure 4). Racemes and panicles were inferred to be derived states, with the former occurring sporadically in different clades and the latter occurring mainly in D. sect. Botryosicyos and D. sect. Enantiophyllum. Spike occurred as a homoplasy in D. sect. Shannicorea, while other clades did not show a consistent type of male inflorescence.   The light color of the male flowers was uncovered as the ancestral state of Dioscorea ( Figure 5). The dark color evolved independently several times in D. sect. Stenophora and D. sect. Opsophyton. Most species with dark colored male flowers occurred in D. sect. Stenophora. There were also some species in D. sect. Shannicorea and D. sect. Enantiophyllum that possessed male flowers that were purple in color or had a brown stripe on them. This distribution probably reflects multiple convergence events in response to similar pollination strategies. The light color of the male flowers was uncovered as the ancestral state of Dioscorea ( Figure 5). The dark color evolved independently several times in D. sect. Stenophora and D. sect. Opsophyton. Most species with dark colored male flowers occurred in D. sect. Stenophora. There were also some species in D. sect. Shannicorea and D. sect. Enantiophyllum that possessed male flowers that were purple in color or had a brown stripe on them. This distribution probably reflects multiple convergence events in response to similar pollination strategies.

Evolution of Organs for Vegetative Propagation and Diversity in the Reproductive Strategy
Vegetative propagation is a crucial mechanism that ensures plants maintain and spread their population, allowing plants to survive unfavorable conditions for sexual reproduction [37]. Dioscorea species possess multiple organs involved in vegetative propagation, including rhizomes, horizontal or vertical tubers and aerial bulbils, as a result of adaptation to various environmental conditions. In the early diverged D. sect. Stenophora, the underground parts consisted of perennial branched horizontal rhizomes. Perennial to annual starchy tubers took the place of rhizomes in subsequently evolved lineages to undertake the function of asexual reproduction. Rhizomatous species are supposed to be more adaptive to long growing seasons in shaded habitats [38], whereas tubers can evade destruction by animals and allow adaptation to warm and humid climates [39]. The possession of tubers was suggested to play an important role in species diversification of Dioscorea [2]. However, tubers are mainly storage organs for plants to live through winter and germinate in for the next growing season; the reproductive efficiency and dispersal ability of tubers are limited.
The emergence of aerial bulbils in some Dioscorea species greatly changed the propagation patterns of these species. The bulbil of Dioscorea is a minor storage organ compared to the underground tuber, arising from the axils of leaves or inflorescences and varying in form and size in different species [40]. On the basis of morphological and anatomical studies, bulbils are interpreted as modified axillary branches and miniature reproductions of the rhizome [11,41]. Functioning as a means of vegetative propagation, bulbils are released from the plants after maturation and then dispersed by gravity or by water in streams, sometimes over long distances, like seeds. Bulbils can easily germinate and grow independently, giving rise to large numbers of progeny. Because of these advantageous characteristics, species that produce bulbils are usually more vigorous and more competitive than species that reproduce only by seeds, as observed in Allium (Alliaceae) [42] and Butomus (Butomaceae) [43].
As mentioned above, the vegetative propagation strategies vary among the different lineages of Dioscorea, and the production of bulbils is a derived character in most clades of this genus. This may have greatly influenced the expansion of the Dioscorea species. In Poa alpina L. (Poaceae), bulbil-producing plants are better adapted to higher elevations compared with seed-producing plants. Therefore, this species is able to occupy a range of ecological niches by means of different reproductive modes [44]. Bulbils in Fritillaria (Liliaceae) are suspected to be an adaptation to underground dispersal [38]. In Dioscorea, the rhizomatous taxa are mainly distributed in South and East Asia, and many of them are endemic species only found in restricted areas, while the species that produce bulbils occupy a wide range of habitats worldwide [2]. D. bulbifera, the only wild species exhibiting worldwide distribution in Dioscorea, produces huge numbers of bulbils compared to other species. Bulbils of D. sansibarensis Pax are buoyant and spread through water flow, achieving wide distribution in African valleys [11]. Here, it is postulated that the production of bulbils improved the adaptive ability of Dioscorea species and promoted their expansion.

Evolution of Floral Characters and Their Relation to Sexual Propagation
Flower evolution is strictly linked to pollination strategy [45,46]. Dioscorea is predominantly dioecious and allogamous; thus sexual propagation in this genus relies on the outcrossing process. Moreover, the sticky nature of Dioscorea pollen grains prevents transport by wind, and, consequently, the flowers of Dioscorea species are pollinated mainly by insects [47]. The diverse flower exhibition characteristics of this genus were most likely shaped by pollinator-mediated selection.
According to our analyses, spikes are reconstructed as the plesiomorphy of Dioscorea, while panicles appear mainly in D. sect. Enantiophyllum as a derived inflorescence type. This contrasts with the opinion presented by Burkill [11], who suggested that spikes and racemes evolved from panicles. The distribution patterns of inflorescence types among species do not mirror phylogenetic relationships, with closely related lineages displaying different types, except D. sect. Shannicorea. Such a pronounced polymorphism might have been driven by pollinators. Inflorescence architecture is related to the mode and efficiency of pollination. For instance, changes in inflorescence architecture were associated with the transition from biotic (insect) to abiotic (wind) pollination in Schiedea salicaria Hbd. (Caryophyllaceae) [48], and affected pollinator behavior and mating success in Spiranthes sinensis (Pers.) Ames (Orchidaceae) [49]. Panicles are made of multiple spikes or racemes to enlarge the volume of the flower and enhance the attraction of pollinators. They possess two to three times the number of flowers and amount of pollen as spikes or racemes of the same length. The function of panicles in insect attraction compared with other types of inflorescences is worthy of investigation in Dioscorea.
In our results, most species of Dioscorea possess male flowers of light color, and dark color originated several times independently in different lineages. It is well known that floral color is among the most important visual signals in pollinator attraction. For example, hawkmoth and hummingbird pollinators prefer yellow and red morphs of Mimulus aurantiacus Curtis, respectively [50]. The reflectance spectrum of the dark perianth generated by UV-light is recognizable to insects [51]. Dark color and light color perianth could be visually discriminated by insects and thus will affect the visiting choice of different pollinators [52]. An alternative explanation for floral color polymorphisms among closely related species is that color divergence evolves in response to interspecific competition for pollinators as a means to decrease interspecific pollinator movements [53]. Indeed, the color of perianth in Dioscorea includes white, yellow, green, orange, red and purple, sometimes changing from one to another during flower development. How these variations of floral colors influence the category and behavior of the pollinators will be an interesting subject to investigate.
The evolutionary trend of the perianth opening degree in Dioscorea is from wholly opened to nearly closed. This character may also influence the category of pollinators. As discovered in the study of Merianieae (Melastomataceae), corollas of most buzz-bee syndrome species are widely open, forming bowl-shaped flowers, whereas they are more closed and form urceolate to pseudo-campanulate flowers in vertebrate-pollinated species [54]. The same was also observed in Erica (Ericaceae) [55] and Ruellia (Acanthaceae) [56]. Zhao et al. [57] reported that D. nipponica subsp. rosthornii is pollinated by halictids. In D. sect. Enantiophyllum species, male panicles were found usually to be made up of closed flowers that restrict visitations only to small insects. The pollinators of species such as D. rotundata Poir., D. japonica Thunb. and D. polystachya Turcz. were purported to be thrips [47,[58][59][60]. Therefore, the evolution of male flower opening degree in Dioscorea may be a result of pollinator specialization.
Actually, as reported in previous studies, the pollinators of Dioscorea species include Coleoptera, Diptera, Hymenoptera, Hemiptera, Thripidae, Thysanoptera, Halictus and Andrena. [47,60]. The diversification of floral morphological characters in this genus is thought to be promoted by pollination-associated adaptations. Intensive investigation of the pollination strategies in other lineages is needed to better understand the adaptive evolution of reproduction in Dioscorea.

Infrageneric Relationships and Taxonomic States of Several Species
The phylogenetic framework reconstructed in this study was consistent with those obtained in previous studies [2,12,14,15]. The relationships indicated by molecular information were generally consistent with infrageneric division by traditional systematics.
The monophyly of Dioscorea sect. Stenophora has been certified in a number of molecular studies using various datasets. This is also supported by morphological evidence since species of D. sect. Stenophora show underground organ type, chromosome number and pollen type that set them apart from other lineages of Dioscorea. As indicated in this study, no species of D. sect. Stenophora generated bulbils, and they all had male flowers with fully open perianths, thus giving further support to the isolated position of this section within the Dioscorea genus.
Most species of Dioscorea sect. Lasiophyton have been transferred into D. sect. Botryosicyos [22]. However, there are clear morphological differences between these two taxa. Species of D. sect. Lasiophyton have compound leaves with three palmate leaflets that are also palmately veined, and all six stamens of the male flower are fertile; in contrast, in D. sect. Botryosicyos, leaves usually have more than three leaflets, are pinnately veined, and have three stamens alternating with three staminodes. In this study, the only Chinese species of D. sect. Lasiophyton, D. hispida, was resolved as the sister lineage to the Botryosicyos clade, supporting the exclusion of this species from D. sect. Botryosicyos by Ding and Gilbert [22].
Dioscorea tentaculigera was assigned to D. sect. Shannicorea by Burkill [11]. Its male flowers are light in color and fully open. However, D. tentaculigera produces bulbils, and its male flowers are sessile, which is an obvious departure from other species in D. sect. Shannicorea. The phylogenetic position of D. tentaculigera was not resolved in our analyses, as has been the case in previous studies (e.g., [2,[12][13][14]). Maurin et al. [20] placed this species outside of the Enantiophyllum clade based on six cpDNAs; Soto Gomez et al. [61] positioned it as a sister to the Mediterranean clade based on 260 nuclear genes; and Noda et al. [15] suggested that it should be treated as a distinct section according to four cpDNAs. More evidence from molecular and morphological data is needed to draw a conclusion about the correct classification of this species.
Accessions of Dioscorea sect. Enantiophyllum were found to cluster into a highly supported monophyletic clade. There are several synapomorphies of this group: (i) stem twining to the right, (ii) leaves generally opposite, (iii) seeds winged all round, (iv) bulbils normally present, (v) male inflorescence panicles and (vi) perianth white to yellow and nearly closed when blooming. The limits of this section were found to be reasonable for a natural lineage [2,12,14,15,17].
The phylogenetic analyses also shed light on the taxonomy of certain species. Dioscorea nipponica subsp. rosthornii was separated from D. nipponica subsp. nipponica by a cork layer of rhizomes and male flowers that were pedicellate [62], as well as a difference in chromosome number [24,39]. However, they have not been resolved as sister groups in any molecular phylogenetic trees generated to date [12,14,24]. A similar situation was observed with D. subcalva var. submolis, which showed differences from D. subcalva var. subcalva that included sparse leaf blades and longer infructescences. We noticed that these characters varied in a continuous range between individuals among populations. The relationships between these two varieties and other species of D. sect. Shannicorea remain unresolved by molecular analyses [15,17]. Further investigation with sufficient samples may help us understand the internal relationships within this section and the circumscription of the species.

Conclusions
This study provides the first phylogenetic analyses focusing on the evolution of four main reproductive traits in Dioscorea. The development of bulbils in late-diverged lineages of Dioscorea has diversified the mode of vegetative propagation, which is supposed to have greatly improved reproductive efficiency and heightened plants' ability to occupy new habitats. Spikes, male flowers in light color and wholly open perianth are reconstructed as plesiomorphic in Dioscorea, whereas panicles, dark color flowers and nearly closed perianth are suggested derived states. The extraordinary variation of floral characters in this genus appears to be a consequence of adaptive evolution driven by pollinators. A broader sampling of Dioscorea, especially new-world species, together with their character information and extensive pollination studies among species with different floral, types are necessary to achieve a better understanding of the diversification mechanism of this genus.
On the other hand, this work provides a suitable phylogenetic framework for the revision of infrageneric classification and species delimitation in Dioscorea. Some of the morphological characters show consistency in a certain clade. Species in D. sect. Stenophora do not produce bulbils, and their male flowers are always wholly open; D. sect. Shanni-corea is characterized by no bulbils, spikes, light color and wholly open flowers; D. sect. Enantiophyllum is distinguished by producing bulbils, with flowers in light color and nearly closed. According to our results, bulbils and perianth opening degree are informative taxonomic features at the section level in Dioscorea. The monophyletic status of several sections proposed in classical taxonomy, such as D. sect. Stenophora and D. sect. Enantiophyllum, were supported by morphological synapomorphies besides molecular data. Morphological characters analyzed in the present work can also explain the unexpected position of D. tentaculigera in phylogenetic trees. In addition, our results suggest the necessity of reconsideration of some infraspecific taxa as well.