A revised classification of Chinese Davalliaceae based on new evidence from molecular phylogenetics and morphological characteristics

Although the phylogenetic framework of Davalliaceae is known, the classification of Chinese Davalliaceae is still controversial. In this study, a molecular phylogenetic tree of 60 accessions, including 29 species produced in China, was constructed using five plastid DNA markers—atpB, atpB-rbcL, rbcL, rbcL-accD, and accD. New data on studied specimens, field investigations, and scanning electron microscopy analysis of leaf epidermis and spores were used to reclassify Chinese Davalliaceae. The taxonomic position of Davallia canariensis was confirmed based on new evidence and a new key to sections of Chinese Davalliaceae was proposed. The taxonomically controversial genus Paradavallodes was confirmed as a polyphyletic group, and it was assigned to Davallia sect. Trogostolon and Davallia sect. Davallodes. Further, species endemic to China were delimited, 21 species were admitted to six sections of Davallia, two new combinations were proposed, two new synonyms were defined and a new key to Chinese species of Davalliaceae was presented.


Introduction
Davalliaceae is a small epiphytic leptosporangiate fern family distributed in the tropics and subtropics of the Old World [1]. The circumscription and classifications of its genera have changed frequently since the establishment of this family. Previous morphology-based classifications divided Davalliaceae into 1-10 genera and approximately 49-130 species . Classification of Davalliaceae differs significantly across different studies (Fig 1). Tsutsumi & Kato [22][23], Tsutsumi, Zhang & Kato [24], Chen [25][26], and Liu & Schneider [27] used the molecular characteristics of Davalliaceae for their classification. Tsutsumi  Chinese davalloid ferns are mainly distributed in the south and southwest of China; they are especially abundant in the rainforest and limestone regions of Yunnan, Guangxi, and Guangdong Provinces. However, the classification of Chinese Davalliaceae is still controversial. Ching [6,[8][9][10] and Ching et al. [7] divided Chinese Davalliaceae into nine genera, including 40 species, and published a new genus, Paradavallodes, which was later incorporated into Araiostegia by Holttum [13]. Nooteboom [17] revealed three genera with 14 species after identifying and examining Chinese Davalliaceae specimens from the major Chinese herbariums. Wu & Wang [28] revised Chinese Davalliaceae and listed five genera with 31 species (including nine species produced only in China). Xing et al. [20] also reviewed the classification of Chinese Davalliaceae and divided it into four genera with 17 species based on new morphological evidence. A consensus on the delimitation of Chinese Davalliaceae species is therefore difficult, especially for endemic species. For instance, Davallia brevisora Ching, which was believed to be an endemic species only distributed in Yunnan and Guangxi, was reconsidered as a special form of D. denticulata (Burm. f.) Mett. ex Kuhn by Nooteboom [17] or incorporated into D. sinensis (H. Christ) Ching [20]; however, other scholars [7,28] insisted that it should remain separated as an independent endemic species. D. subsolida Ching, an endemic species of Orchid Island described by Ching et al. [7], which was afterwards incorporated into D. solida (G. Forst.) [20,28], fitted the "F" morphological form of the D. repens (L.f.) Kuhn complex [26]. Other endemic species, such as D. austro-sinica Ching, D. amabilis Ching, and Paradavallodes chingiae (Ching) Ching were rarely collected and examined since they were described, leading to their uncertain positions. Although molecular phylogenetic analyses would be helpful for understanding the taxonomic classification positions of these species, only 14 species produced in China have been investigated by Tsutsumi & Kato [21][22] Tsutsumi et al. [23][24] and Chen [25][26]. An overall molecular phylogenetic study of Chinese Davalliaceae is therefore necessary to address these questions.
Recent studies [29][30] showed that the characteristics of the cuticular layer of leaf epidermis and spore ornamentation were key features for the classification of Davalliaceae at the genera/species level. The classification according to morphological traits of the leaf epidermal cuticular layer was similar to that based on the molecular phylogenetic tree presented by Tsutsumi & Kato [22][23] and Tsutsumi et al. [24]. Several species complexes with taxonomic controversy showed informative variations in spore ornamentation. Thus, the two morphological features are important taxonomic characters [30]. In the present study, a complementary phylogenetic tree of Davalliaceae was reconstructed by mainly focusing on species for which molecular materials were unavailable in previous studies. Further, results from field investigations, specimen examinations, observations on leaf epidermal cuticular layer and spore ornamentation, and molecular phylogenetics were integrated to obtain extensive taxonomic evidence to address the controversial taxonomic classification of Chinese Davalliaceae.

Specimen study
We reviewed about 1700 davalloid specimens from 11 herbariums (including K, BM, E, US, PE, KUN, IBSC, HITBC, CDBI, IBK, and GXMI), and the sampling location of each specimen examined was plotted on global and China maps by using ArcGIS v. 10.1 (ESRI Inc., Redlands, CA, USA) (Fig 2). Type specimens were thoroughly examined.

Field investigation
Field observation and sampling were widely conducted in the Canary Islands, and in Yunnan, Guangxi, Tibet, Guangdong, and Gansu Provinces of China since 2004. The geographical coordinates of sampling sites are listed in S1 Table and marked in Fig 2, and new field findings were recorded. No specific permissions were required for these locations/activities because the field studies did not involve endangered or protected species. Living plants were transplanted to the greenhouse of South China Botanical Garden, and some of them were pressed into voucher specimens, which were stored in IBSC for verification and reference.
Obtained were assembled in Sequencher v. 4.14, aligned using Clustal X v. 2.0 [37], and then edited manually using Bioedit v. 7.1.3 [38]. Phylogenetic trees using the sequences obtained from the combined markers (atpB, atpB-rbcL, rbcL, rbcL-accD, and accD) were constructed using maximum parsimony (MP) and Markov chain Monte Carlo Bayesian inference (BI). MrModeltest2 v. 2.3 [39] was used to select the general time reversible with a proportion of invariable sites and gamma distributed rate variation among sites (GTR+I+G) as the best fit molecular evolution model for the MP and BI analyses. The MP analyses were performed using PAUP � 4.0b10 [40], treating gaps as missing data and using the heuristic search options with 1000 random replicates and tree-bisection-reconnection branch swapping. All characteristics were unordered and equally weighted. For BI, trees were generated for 1,000,000 generations with sampling at every 100 generations. Four chains were used with a random initial tree. For each individual data partition and for the combined dataset, the first 2500 sampled trees were discarded as burn-in to ensure that the chains reached stationarity. Nodes receiving bootstrap support <70% in the MP analyses or posterior probability (PP) <0.95 in the BI analyses were not considered well supported.
Afterwards, molecular phylogenetic analyses integrated with leaf epidermal characters (observed under scanning electron microscopy) was studied. Eight types of cuticular layer characteristics of leaf epidermis in Davalliaceae (A, Sinuate, fine, unordered stripes. B, wavy, thick, and tightly joined stripes. C, cavities visible in stripes. D, compound stripes. E, sinuolate and thick stripes, hunch shallow. F, sinuolate stripes, stripes shortened to the apophysis. G, zigzag stripes, fine stripes. H, sinuolate, fine stripes) [29] were plotted the on the present phylogenetic tree. Thirty-three accessions, for which both molecular of present study and cuticular layer of leaf epidermis data were available [29], were examined.  (Fig 3), we confirmed that the morphological characters examined were discrepant between A. pulchra (A. Henry13069, K) and A. pulchra (Wallich259, K): the former has hook-shaped or oval ultimate lobes with imbricated scales densely borne on rhizome, while the latter has linear ultimate lobes with imbricated scales sparsely cling to the rhizome. According to the Chinese Virtual Herbarium (http://q.plantphoto.cn/), the type specimen photo of A. pulchra collected by Ching was not actually "Wallich259" but "A. Henry13069". Furthermore, we found that the right lamina of "Wallich259" had no distinguishing features from the lamina of "Colonel Dyas". In fact, A. pulchra is rather variable, and its size may differ greatly among different habitats the size may differ greatly [20]. As we observed in the wild, A. (= D.) pseudocystopteris is often a young stage of A. pulchra, which is the only Davalliaceae species able to grow terrestrially, and commonly presents linear ultimate lobes, imbricated scales sparsely cling to the rhizome, and rachis colour varying from pale straw to greenish. Hence, we treated A. (= D.) pseudocystopteris as a synonym of A. pulchra. However, A. beddomei, A. yunnanensis (Chist) Cop., and A. imbricata were not treated as synonyms of A. pulchra because they have distinct morphological characters (see list in the Classification section). This was corroborated by the subsequent phylogenetic analysis.

Specimen study and field investigation
After contrasting the types of Davallia cyclindrica Ching (Wang Q.W.74303, PE, published time: 1959), D. bullata Wall. ex Hook. (Wallich258, K, published time: 1829), and about 10 specimens of D. bullata from K and E, we confirmed there were presence of no distinguishable features between D. cyclindrica Ching and D. bullata (Fig 3). Further, both species were distributed in the Himalayan region without geographical isolation.
Davallia austro-sinica and D. amabilis are both endemic Chinese species and were synchronously described by Ching et al. [7]. According to the description published by these authors, D. austro-sinica plant height is 30 cm and they present quadripinnate laminae, whereas D. amabilis plant height is above 100 cm and presents quadripinnate to 5-pinnate-pinnatifid laminae. To our knowledge, no studies have thoroughly compared these species since their description because of their extremely rare distributions.
During our field survey in Napo County of Guangxi Province, Davallia amabilis were found to be morphologically variable: the length of their mature lamina varied from 20 cm to more than 100 cm, indicating that lamina length was unstable in D. amabilis. Further, the 20-cm-thick laminae, which were similar to those of D. austro-sinica, were quadripinnate, and the 100-cm-thick laminae were quadripinnate to 5-pinnate-pinnatifid (specimens MA057 and MA058 in IBSC, which corresponded to different laminae of the same plant). Thus, in this case, species delimitation was difficult.
Davallia brevisora is an endemic Chinese species, and the delimitation between this species and D. sinensis is controversial. Nooteboom [17] regarded D. brevisora as a special form of D. denticulata and considered D. sinensis as a synonym of D. solida. Wu & Wang [28] also treated D. sinensis as a synonym of D. solida, but considered D. brevisora as a separate species. Xing et al. [20] treated D. brevisora as a synonym of D. sinensis but considered it distinct from D. solida. The key distinction between D. sinensis and D. brevisora is that the tubular indusium in the former is twice as long as wide, whereas the latter has cup-shaped indusium, which is slightly longer than wide or almost as long as wide [7].
Davallia brevisora has an extremely narrow distribution area and it is found only in Yunan and Guangxi Provinces. During the field survey in Malipo, Yunan, members of this species were found growing as epiphytes on an ancient tree. Two different forms were synchronously growing in the same plant: one with cup-shaped indusium and the other with tubular indusium (specimens MA024 and MA055 in IBSC, which corresponed to different laminae ofthe same plant). Therefore, the shape of the indusium is not a stable character for taxonomic delimitation. No critical circumscription is available to separate D. sinensis and D. brevisora.
To confirm whether the two species are the same, we individually extracted the DNA of both lamina forms and performed phylogenetic analyses.

Molecular phylogenetic analyses
The phylogenetic tree of Davalliaceae (with 60 accessions) generated using the five markers (atpB, atpB-rbcL, rbcL, rbcL-accD, and accD) comprised 5368 nucleotides, of which 1069 were variable (19.91%) and 687 were phylogenetically informative (12.80%). The MP analysis based on this dataset yielded one MP tree of 1779 steps with a consistency index of 0.6695 and a retention index of 0.8193. The tree obtained from BI analysis had a similar topology to the MP strict consensus tree (Fig 4).
The phylogenetic tree based on the new datasets (Fig 4) structurally resembled that obtained based on atpB-rbcL-accD, nuclear LFY intron 1, and gapCp intron [21], as it com-

Molecular phylogenetic analyses integrated with the leaf epidermal cuticular layer characters in Davalliaceae
Previous studies suggested that Davalliaceae can be divided into nine groups based on the cuticular layer characteristics of the leaf epidermis [29]. The examinations performed here revealed that the group represented by only Davallia canariensis, which was also characterised by zigzag stripes, fine stripes (Fig 5, Type G), should be included as Type G. The relationships among taxa classified according to the characteristics of leaf epidermis were unusually similar to that based on molecular phylogenies (Fig 6): The cuticular of Davallia sect. Humata has visible cavities in stripes (Figs 5 and 6, Type C), whereas that of Davallia sect. Cordisquama has wavy, thick, and tightly joined stripes (Figs 5 and 6, Type B). Davallia sect. Davallodes has sinuate (Figs 5 and 6, Type A) or zigzag (Figs 5 and 6, Type G) stripes or stripes shortened to the apophysis (Figs 5 and 6, Type F). These results indicated that the evolution of leaf epidermal cuticular layer is likely an immediate external reflection of the molecular evolution of these species. Therefore, the cuticular characteristics of leaf epidermis are important traits for the section classification of Davallia.

Significance of our study to the taxonomy on section level of Davallia
Recently, Tsutsumi et al. [21] conducted molecular phylogenetic analysis and showed that the unstable phylogenetic position of the clade represented only by Davallia canariensis led to the inconsistent relationships among clades with respect to the markers analysed, and therefore the species tree derived from all datasets did not resolve the relationships. However, several close relatives of D. canariensis, such as Araiostegia pseudocystopteris (Kunze) Cop., A. beddomei, Paradavallodes kansuense Ching, and P. membranulosum (Wall. ex Hook.) Ching, were absent in the previous phylogenetic study. After these close relatives were included, phylogenetic clades with high PP were obtained andthe clade represented only by D. canariensis was merged into the clade represented by sect. Davallodes.In the present study, molecular data and leaf epidermis analysis results both indicated that D. canariensis is a member of Davallia sect. Davallodes. D. canariensis is generally accepted as the type of the genus Davallia [27]. However, if this species is classified into Davallia sect. Davallodes, the name of the section should be replaced by "Davallia sect. Davallia". The new section classification is listed under "Classification".

Paradavallodes is a polyphyletic group
Genus Paradavallodes, comprising four species, was published by Ching [9]. Copeland [1] once divided Paradavallodes membranulosum (Wall. ex Hook.) Ching and P. multidentatum (Hook. et Bak.) Ching into Davallodes and Araiostegia respectively. Ching assigned these species to Paradavallodes based on their similar distribution (Himalayan region) and morphological characters (lamina pubescent). The examination of leaf epidermis under the scanning electronic microscope (SEM) showed that the cuticular layer of P. membranulosum and P. kansuense had sinuolate stripes, stripes shortened to the apophysis (Figs 5, 6 and 7, Type F), whereas that of P. multidentatum, a member of Davallia sect. Davallodes [21], had zigzag stripes, fine stripes (Figs 5 and 6, Type G). However, P. chingiae (Ching) Ching differed from both species by the sinuolate cuticular layer, thick stripes, and shallow hunch (Fig 7, Type E, belonging to Davallia sect. Trogostolon). The morphologically closest species to P. chingiae with similar leaf epidermis characteristics is Humata assamica (Bedd.) C. Chr. (Fig 7). Further, its macromorphological traits resembled those of H. assamica: both have broad lanceolate lamina, homomorphic basal and upper pinnae, and a narrow wing borne on their petiole. The spore ornamentation characters were conspicuously different between P. chingiae (verrucate/ lophate) and other species of Paradavallodes (Figs 5 and 7). In fact, P. chingiae is the only Davalliaceae species which has lophate spore ornamentation. This indicated the particularity of the species and the necessity of it being treated as a separated species. Based on these findings, we suggest that P. chingiae belongs to Davallia sect. Trogostolon rather than Davallia sect.

Davallodes.
The molecular phylogenetic study showed that P. multidentatum, Araiostegia pulchra, A. imbricata, A. yunnanensis, and all members of Davallodes (in a narrow sense) clustered into one clade. Paradavallodes membranulosum and P. kansuense clustered into another clade. These two clades are sister groups with high PP (0.99; Fig 4). Although P. kansuense has similar morphological characters to those of P. multidentatum [20], they are phylogenetically distant. Thus, P. kansuense should be classified as separated species rather than reserved as a synonym of P. multidentatum. All these species are nested in the Davallia sect. Davallodes (Fig 4). Integrated evidence therefore indicates that Paradavallodes is a polyphyletic group. Paradavallodes multidentatum, P. membranulosum, and P. kansuense should be assigned to Davallia sect. Davallodes, while P. chingiae should be assigned to Davallia sect. Trogostolon.

Phylogenetic position of controversial species produced in China
Clarification of A. pulchra taxonomic status. The ambiguous species delimitation of A. pulchra stems from the absence of a type specimen. Although it was first published without Classification of Chinese Davalliaceae specifying the type, Nooteboom [15] specified "Wallich259", and its collection time and morphological features fit well with the published time and morphological description of A. pulchra. At the same time, the author defined the broad classification concept of A. pulchra [15]: Araiostegia beddomei, A. pulchra, A. pseudocystopteris, A. imbricata, and A. yunnanensis, which were previously separated [7,28,41], were incorporated into species A. pulchra. Field surveys showed that A. pulchra (= pseudocystopteris) had a particular terrestrial life form, thereby differing from other members of Araiostegia, which required its separation. The original life form during the evolution of epiphytes in Davalliaceae and related ferns is terrestrial [23]. This is also confirmed by the position of A. pulchra (= pseudocystopteris), which is located at the base of the phylogenetic tree (Fig 4). According to Flora of Tibet [41], the purple rachis is a key character to distinguish A. beddomei from other species. However, Wu & Wang [28] selected "the density, arrangement, wrinkle of the scales", and "the shape of ultimate pinna" rather than "purple rachis" as the key characters. Interestingly, the special individuals of A. pulchra, which fit the description of Wu & Wang [28], but grow with purple rachis, are difficult to recognise. The particular A. pulchra individual examined here was collected from Binchuan, Yunnan. The A. pulchra from Tsutsumi 's study [21] and A. beddomei from Tibet were also analysed in the present phylogenetic study to decipher their phylogenetic relationships among them. The endemic Chinese species A. imbricata was also examined. The phylogenetic tree (Fig 4) showed that A. pulchra from Binchuan, Yunnan and A. beddomei from Tibet clustered into a clade; A. imbricata and A. pulchra from Sikkim clustered into another clade. This suggested that the purple rachis is a key taxonomic trait in the classification of Araiostegia, as plants with purple rachis are closely related. The distant phylogenetic relationship between A. pseudocystopteris and A. imbricata further supported that they should be separated. Taken together with the results of previous specimen examinations, we conclude that the classification of A. (= Davallodes) imbricata should be reserved. The different phylogenetic positions obtained for A. pulchra (from Sikkim) and A. pseudocystopteris might be because Tsutsumi et al. [21] used the broad classification concept of A. pulchra that was actually unsupported by molecular data.
Phylogenetic position of D. sinensis (= D. brevisora). Sequence of Davallia sinensis and D. brevisora were not different further supporting that D. brevisora is a synonym of D. sinensis. Molecular data (Fig 4) griffithiana, H. tyermanii, H. henryana, and  H. platylepis. These four species are extremely similar in morphology. Ching [7] delimited them based on the indusium shape and attached position of indusium base. According to the broad classification concept of Nooteboom [17], these species were merged into one species.
Field survey data showed that Humata tyermanii has become one of the key species in Guangxi Province. Notably, its indusium, similar to that of H. platylepis (Bak.) Ching, is semicircular. Humata platylepis is distinguished by lamina height of 35 cm, and by thick and flat rhizome. However, the shape of the indusium is not clearly different between H. tyermanii and H. platylepis. Phylogenetic data (Fig 4) showed that H. griffithiana (Hook.) C. Chr., H. tyermanii, and H. platylepis clustered into a clade. Further, the leaf epidermis and spore ornamentation characters were almost similar among the four species [29][30]. Like members of the H. repens (L. f.) Diels complex [26], H. griffithiana might undergo hybridisation and polyploidisation, leading to its high phenotypic plasticity. These results supported that H. griffithiana, H. tyermanii, H. platylepis, and H. henryana should be treated as members of the H. griffithiana complex. Considering the complexity of these species, further studies regarding their evolutionary history are needed.
Phylogenetic position of D. cyclindrica. According to the depiction of this Chinese endemic species [7], its architype-Davallia bullata-is found in K. Some scholars believed that D. bullata, D. cyclindrica Ching and D. mariesii Moore ex Baker were all synonyms of D. trichomanoides Blume [16][17]20] and had several common characters: 3-8 mm rhizome, tripinnate lamina, flat and acicular scales. Field surveys and specimen examination confirmed D. cyclindrica a synonym of D. bullata and could be distinguished from D. mariesii. The former is characterised by the reddish brown scales of the tender stem, which is linear with subulate head, and hook-shaped or oval ultimate lobes; in contrast, the latter shows light brown or grey scales, above a considerably broader base, evenly narrowed toward the apex, and linear or sickle-shaped ultimate lobes.
The above positions are further supported by molecular data (Fig 4), showing that D. cyclindrica (= D. bullata), D. petelotti Tard. -Blot & C. Chr., and D. trichomanoides clustered into a clade. D. cyclindrica is separated from the other two species, indicating that it should be retained as a species. Davallia mariesii had a close relationship to H. griffithiana (variable indusia), but not to D. cyclindrica. Thus, the shape of indusia does not always reveal the evolutionary relationships of close relatives (see also H. griffithiana and D. divaricata), and therefore this character was excluded from these species identification. Thus, D. cyclindrica is a new synonym of D. bullata.
Phylogenetic relationships among D. austro-sinica, D. formosana, and D. divaricata. These three species are extremely similar in morphology and distinguishing them based on external morphology is difficult. The key character to distinguish them is indusium shape. In the phylogenetic tree, they clustered into a monophyletic group (Fig 4). Nooteboom [17] treated Davallia formosana (Hayata) M. Kato & Tsutsumi as a synonym of D. divaricata Blume. Wu & Wang [28] excluded D. divaricata from the species produced in China. After examining the specimens of D. divaricata and D. formosana in K and IBSC, we found that D. divaricata is distinguished by cup-shaped indusia, which are slightly longer than wide or almost as long and wide, whereas D. formosana has tubular indusia twice as long as wide. D. austro-sinica and D. divaricata have similar indusia. However, the phylogenetic tree (Fig 4) showed that these two species are closer to each other than to D. divaricata. This also suggests that the shape of indusia is an unstable or ambiguous classification character. Thus, D. austrosinica and D. formosana should be treated as synonyms of D. divaricata.
Davallia subsolida. Davallia subsolida is an endemic species of Orchid Island, where D. repens is polymorphic due to polyploidy [7,20,26]. After examining the type specimens of D. subsolida in PE, we found that the species fitted well with the "F" morphological form of the members of the D. repens complex presented by Chen et al. [26] (see Fig 2 in their study). The phylogenetic position of the "F" morphological form nested in Clade X of D. repens. Furthermore, leaf epidermis SEM observations (Figs 5 and 7) showed that D. subsolida (Type C) was distinct from D. solida (Type D) and the latter is not distributed in China. Type C has a specific leaf epidermis characteristic of Davallia sect. Humata [29]. Thus, we merged D. subsolida into the D. repens complex based on geographical distribution, molecular data and leaf epidermis cuticular layer (SEM observations).

Classification
The present study integrated the new findings from field surveys, specimens examination, the structural characteristics of leaf epidermis cuticular layer, and molecular phylogenetics, and provides a new key to identify the six section and 21 species of Davallia produced in China, which is presented below (Tables 1 and 2).