RAD‐seq data provide new insights into biogeography, diversity anomaly, and species delimitation in eastern Asian–North American disjunct clade Benthamidia of Cornus (Cornaceae)

The big‐bracted dogwood clade Benthamidia of Cornus is a typical example of the well‐known eastern Asia (EA) and North America (NA) floristic disjunction, with greater species diversity in EA than in NA. The lineage provides an opportunity to explore factors contributing to the plant diversity unevenness between EA and NA and test hypotheses on the origin of disjunct distribution from a phylogenetic perspective. We generated RAD‐seq data, conducted phylogenomic and biogeographic analyses for the clade with sampling of all species (9) and subspecies (10) currently recognized in floras. We also assessed species delineation and calculated phylogenetic diversity to evaluate the diversity unevenness between EA and NA. Finally, we examined variation of diversification rates and ecological niches on the phylogeny to explore potential causes underlying the observed diversity pattern. Our results revealed phylogenetic relationships congruent with previous studies and suggested a trans‐Beringian ancestral distribution of the clade Benthamidia in the mid‐Oligocene, dispersal from Mexico to eastern United States in the mid‐Miocene, and early diversification of the EA clade in SW China. Our results also confirmed greater phylogenetic diversity and diversification rate of the EA clade. Species delimitation analysis suggested 17 species in the clade Benthamidia, including all recognized subspecies. By integrating the results of molecular data with morphology, we proposed to retain the subspecies without changing their ranks. Our data suggested increased diversification rate in EA as an intrinsic factor explaining the greater species diversity in the region driven mainly by biogeographic isolation and partially by niche divergence.


Introduction
The disjunct distribution of deciduous forests between eastern Asia (EA) and North America (NA), with species richness biased toward EA, represents a phytogeographic pattern of discontinuous distribution and an outstanding example of diversity disparity between geographic regions with similar climate in the Northern Hemisphere (Boufford & Sponberg, 1983;Tiffney 1985;Guo, 1999;Qian & Ricklefs, 1999;Qian et al., 2017).There is a large number of temperate vascular plant genera or clades confined to the two regions which are similar in areal size but differ dramatically in species richness and phylogenetic diversity (PD) (Li, 1952;Wu, 1983;Qian & Ricklefs, 2000;Hong & Blackmore, 2015;Hu et al., 2022).Such large biogeographic and biodiversity patterns are results of historical processes and an outcome of differences between the two regions in ecological and evolutionary processes driven by long-term interplay of extrinsic (environmental) and intrinsic (biological) factors.Major factors proposed to explain the diversity pattern are several, including greater allopatric speciation in EA promoted by its more complex topology (Axelrod et al., 1996;Qian & Ricklefs, 2000), greater net diversification due to less extinction in EA facilitated by more refugia from its complex topological structure (Adams, 2009) and more extinction in NA due to glaciation in its northern regions and aridification in the western regions (Latham & Ricklefs, 1993;Qian & Ricklefs, 2000;Eronen et al., 2012;Ricklefs & He, 2016), older age in EA allowing diversification in the area before spreading into NA (Latham & Ricklefs, 1993), and broad connection between tropical and temperate floras in southern EA, allowing immigrants from the south (Axelrod et al., 1996) which is largely lacking in NA.It is likely true that the outstanding diversity anomaly observed today is a result of a combination of all these factors.However, in each disjunct genus, the diversity pattern may vary and was caused by one or more of these factors in different combinations.Therefore, exploring the roles of these factors in different taxa is a prerequisite for a clear understanding of the mechanisms of the global pattern (e.g., the relative importance of the possible factors).
Phylogenetic and biogeographic studies of many disjunct lineages revealed that "Out of Asia" or "Out of Old World" was predominant over "Out of America" or "Out of New World" migrations (Donoghue & Smith, 2004;Wen et al., 2010Wen et al., , 2016;;Harris et al., 2013), supporting a major role of the age factor (older in EA) on the diversity anomaly.On the other hand, evidence supporting greater diversification rate in EA has been scanty.Comparisons between disjunct sisters in limited number of taxa were not congruent among studies and genera regarding their relative diversification rate (e.g., Xiang et al., 2004, on 10 genera;Yan et al., 2018, on Lysimachieae;Lindelof et al., 2020, on clade Swida of Cornus; Zhou et al., 2022, on Torreya;Melton et al., 2022).These studies indicate that the diversity pattern and underlying causes in disjunct taxa are not uniform.The roles of ecological niche divergence in driving the evolution and diversification of the disjunct taxa also appear to be inconsistent.Comparisons of climatic niches for 31 EA-eastern North America (ENA) disjunct genera supported a general trend of overall niche conservatism of congeners but greater niche space occupied by EA species compared to their close relatives in ENA (Melton et al., 2022), supporting greater opportunity for speciation of the EA lineages and importance of historical geographic isolation or other factors, rather than climatic niche divergence, in driving evolutionary divergence of the sister counterparts.However, a few studies tracking the evolution of ecological niches showed that disjunct counterparts apparently diverged in one or more specific climatic variables (Du et al., 2020;Lindelof et al., 2020;Zhou et al., 2020).The evidence further supported variation of underlying causes of diversification among disjunct lineages.More studies comparing PD, diversification rate, and ecological niches between the disjunct counterparts would be helpful to better understanding the intrinsic and extrinsic factors underlying the diversity anomaly.
The big-bracted dogwoods (BB) (clade Benthamidia of Cornus L.; Du et al., 2023) represent a typical clade of EA-NA disjunction with more species recognized in EA, providing an opportunity to evaluate the diversity anomaly and explore the intrinsic factors (e.g., lineage age and diversification rate) and extrinsic factors (e.g., ecological niche and geography) underlying the diversity pattern.The clade consists of three species in NA (Cornus nuttallii in western North America (WNA), Cornus disciflora in Mexico and Central America (MCA), and Cornus florida with one subspecies in ENA and one subspecies in Mexico) (Murrell & Poindexter, 2016) and six species (Cornus kousa with two subspecies, Cornus multinervosa, Cornus elliptica, Cornus hongkongensis with six subspecies, Cornus capitata, and Cornus sunhangii) in EA based on Flora of China (Xiang & Boufford, 2005) and recent literature (Lv et al., 2019), occurring mostly in high elevations of mountains (200-3200 m) except C. florida from ENA and C. kousa from EA which also occur in lowland (Figs.S1, S2).Species of the group differ from other clades in Cornus in having head inflorescence subtended by usually four large showy bracts (four to six in C. nuttallii from WNA; bracts in C. disciflora from MCA are deciduous before expansion).Most of the species are evergreen while four are deciduous including C. florida and C. nuttallii from NA and C. kousa and C. multinervosa from EA.The three NA species are clearly defined and easily to distinguish by their morphology and distribution.For instance, the large showy bracts are lost (bracts deciduous before expansion) in C. disciflora that occurs in MCA, whereas C. nuttallii is restricted to the Pacific Northwest of NA and has four to six showy bracts and C. florida is widely distributed in eastern NA and has four showy bracts.The bracts of C. florida subsp.urbiniana is distinguishable by its apical cohesion of the showy bracts during the expansion.However, the EA subclade has long been debated on species delimitation, especially surrounding the complex groups of C. capitata and C. hongkongensis, due to continuous or complicated variation of taxonomic features (e.g., leaf texture, shape, trichome morphology and distribution on leaf surfaces, leaf cuticles, inflorescence morphology, and fruit color).
The eastern Asian clade Syncarpea extends from Nepal, Bhutan, northern India, Myanmar, Laos and Vietnam, northward to China, Korea, and Japan.The number of species recognized in the group has been varied from 4 to 12 (Fang, 1953;Xiang, 1987;Hu et al., 1990;Xiang & Boufford, 2005).Hu et al. (1990) recognized 10 species (as a distinct genus Dendrobenthamia), whereas Xiang (1987) treated several of the species as subspecies and recognized four species (C.capitata, C. multinervosa, C. kousa, and C. hongkongensis), with four subspecies in C. capitata and six subspecies in C. hongkongensis.Xiang & Boufford (2005) later raised a subspecies of C. capitata to a species status (C.elliptica) and recognized five species in Flora of China for the clade.The subspecies taxa in C. hongkongensis were retained but subspecies in C. capitata were not recognized.The C. hongkongensis species complex exhibits a pattern of morphological variation that overlaps among subspecies but is associated with geographic distribution (Fig. S2).This prompted Xiang (1987) to downrank six species to subspecies.Recently, a new species C. sunhangii from the Himalayan region within the range of C. capitata was published (Lv et al., 2019) and Cornus "brevipedunculata," previously merged in C. capitata (Xiang, 1987;Xiang & Boufford, 2005) was proposed to be resurrected (Du et al., 2023).
In previous phylogenetic studies of Cornus, Cornaceae, or Cornales, the big-bracted species clade had incomplete species sampling and/or a single individual per species was included (e.g., Xiang et al., 2006, based on the combined matK-ITS data; Xiang et al., 2008, based   .These studies often resolved a well-supported EA clade but the NA species were either weakly supported as a clade or resolved as being paraphyletic.Exception is the most recent study of Du et al. (2023) where all species were included and the NA species were resolved into a strongly supported clade.Furthermore, the relationships among species within each clade were either incompletely resolved or some nodes were not strongly supported.In this study, we conducted a detailed phylogenomic study, using double digest Restriction Site Associated DNA Sequencing (ddRAD-seq), focusing on the big-bracted clade Benthamidia with broader sampling of populations collected from the natural range of all recognized species and subspecies.We hope to provide a solid phylogenetic basis for evaluating species delimitation and comparing PD, diversification rate, and climatic niches between the EA and NA clades to gain insights into the intrinsic and extrinsic factors underlying observed diversity pattern.Using the phylogeny as a framework, we also reconstructed biogeographic history to test the hypothesis for the origin of the disjunct distribution of the clade derived from the biogeographic analysis of the entire Cornus clade and to gain insights into the past exchanges of plants between Eurasia and North America (Du et al., 2023).

Taxon sampling and RAD sequencing
We sampled 79 individuals (Table S1) representing all the known species and subspecies of the BB clade (17 taxa) and nine samples representing seven species from the remaining three major clades of the dogwood genus as outgroups.Total DNAs were extracted from silica-dried leaves using a microprep CTAB protocol of Doyle (1991), with slight modification described in Xiang et al. (1998).The checking of DNA quality and the preparation of ddRAD-seq library followed Peterson et al. (2012) and Zhou et al. (2018).We used two restriction enzymes, PstI-HF and MspI, to digest 350 ng of total DNA.Fragments of 400-600 bp were selected and used in polymerase chain reactions to construct the library.Single-end sequencing of 150 bp of the library were performed on Illumina HiSeq 2500 (San Diego, USA) at Genomic Sciences Laboratory of North Carolina State University.

RAD-seq data analysis
We used iPyRAD v.0.6.15 (Eaton & Overcast, 2018) to filter and assemble the sequences into individual clusters/loci.After removal of the adapters and recognition sequences, the obtained sequences were 105 bp in length per read.In filtering step, low-quality base calls of Q < 20 were replaced by letter "N"s and reads containing more than five ambiguous bases ("N"s) were discarded.Clusters with a minimum depth of read coverage greater than six in an individual sample were used to build matrix.We used 0.85 as the cluster threshold of sequence similarity for assembly of loci, a maximum of 0.5 as the frequency of the shared heterozygosities per locus, and two as the maximum alleles per locus in an individual sample in the consensus sequence construction for diploid organisms because all taxa of the study with reported chromosome counts are diploid (Eyde, 1988;Murrell, 1993; polyploid in Cornus has been reported only for Cornus unalaschkensis and occasionally for Cornus canadensis of the dwarf dogwood clade; Bain & Denford, 1979;Shearer & Ranney, 2013).We used the plastid genome sequence of Cornus capitata (GenBank accession no.MG524990), a member of the EA clade, as the reference and the assembly method of "denovo-reference" in ipyRAD to exclude plastid DNA loci from our data.We then built four data matrices (labeled as M40-M70) by filtering the initial matrix to allow different levels of missing data, according to the minimum representation of loci among the 88 samples.For instance, the matrix M50 contains loci or reads present in at least 50% of samples.The percentage of missing data and the number of loci in each matrix are listed in Table 1.The matrices are available at https://doi:10.5061/dryad.hhmgqnkp0.

Phylogenetic analyses of concatenated sequences
We first performed phylogenetic analyses of the concatenated sequences for each matrix using maximum likelihood (ML) method implemented in RAxML v.8.2.10 (Stamatakis, 2014) and Bayesian method in MrBayes v.3.2.2 (Ronquist et al., 2012).We used general-time-reversible (GTR) + Gamma model in both analyses as suggested by model test with jModelTest v.2.1.10( Darriba et al., 2012).The RAxML analyses were conducted with 1000 bootstrap replicates, whereas the Bayesian analyses were performed for 1.0 × 10 8 generations with tree sampling in every thousand generations.The log files derived from the Bayesian analyses were examined in Tracer v.1.7.0 (Rambaut et al., 2018) to verify convergence and sampling sufficiency, as indicated by stationarity of the likelihood scores of the trees.The effective sample sizes (ESS) were shown to exceed 200.We discarded the first 10% of trees as burnin, and used the remaining trees to estimate posterior probability in TreeAnnotator (Drummond et al., 2012).We visualized the ML and Bayesian trees in FigTree v. 1.4.3 (Rambaut, 2014).We also used a phylogenetic method that does not force treelike relationships to analyze our data to see if phylogenetic networks within species complexes are present.As the M50 matrix showed the highest support values in the phylogenetic analysis, we performed Neighbor-net analysis in SplitTree4 (Huson & Bryant, 2014) to reconstruct possible network-like relationships within clade Benthamidia using the M50 unlinked SNP data (available at https://doi:10.5061/dryad.hhmgqnkp0).et al., 2010) to estimate the number of genetic subgroups.The analysis focused on the C. hongkongensis-C.elliptica complex because not all relationships within the complex were strongly supported in the phylogeny (Fig. 1).We investigated K values between 1 and 10 with admixture model and correlated allele frequencies by repeating five independent Markov Chain Monte Carlo (MCMC) chains of 200 000 replicates, each with a 10% burnin.The outputs from STRUCTURE analyses with different K values were parsed and visualized using Structure Harvester (Earl & vonHoldt, 2012) and CLUMPAK server (Kopelman et al., 2015) to determine the optimal K value (number of genetic groups).

Species delimitation and coalescent species tree estimation
We assessed the species delimitation and inferred species tree for the clade Benthamidia using the Bayesian coalescentbased method implemented in BPP v.4.3 (Rannala & Yang, 2003, 2017;Yang & Rannala, 2014;Yang, 2015).BPP analyzes each locus separately under independent genealogical histories of loci (Rannala & Yang, 2003, 2017).When the taxa of each sample are annotated, BPP delimitates the number of species in the analyzed taxa.The taxa belonging to the same species will be collapsed into a single branch in the resulting phylogeny.
We first assessed the number of taxa/genetic groups in the Benthamidia clade using a "blind" A11 analysis without the annotation of individuals to taxa.The A11 analysis means joint analysis of species delimitation and species-tree inference (speciesdelimitation = 1, speciestree = 1; Yang & Rannala, 2014).For this analysis, we used the data set containing all 79 samples of the Benthamidia clade from the Fig. 1.Phylogenetic tree inferred from RAD-seq M50 matrix for 88 samples of the big-bracted dogwoods (BB) and outgroup species using RAxML.Maximum likelihood bootstrap support values are indicated at the nodes.Bayesian posterior probabilities from analysis using MrBayes are shown above the branch.The branch lengths of the dwarf dogwoods and the big-bracted dogwoods are shortened.The two clades of BB and four groups of Cornus are annotated.The eight main subclades of the Cornus hongkongensis-C.elliptica species complex are indicated with S1-S8.Subclades within subspecies corresponding to the genetic groups identified by A11 analysis of BPP are indicated by "I" and "II."Distribution areas of the taxa are indicated in parentheses.
above-mentioned matrix M50.We then performed a normal A11 analysis to determine the number of species in the Benthamidia clade with the annotation of individuals to taxa.We performed the above analyses in BPP with the MCMC chains of 2.0 × 10 6 iterations, burn-in of 4 × 10 4 iterations, and sampling of trees every 20 iterations.Two independent runs were performed for each analysis to check for consistency.

Divergence time estimation
We used a reduced set of 25 samples, representing 10 genetic groups in the C. hongkongensis-C.elliptica complex, 10 remaining taxa in the Benthamidia clade, and five outgroup taxa, to estimate divergence times in BEAST v.1.10.4 (Suchard et al., 2018).We built a matrix containing loci/reads that were present in at least 50% of the 25 samples using ipyRAD.In the reduced matrix, each taxon was represented by a single individual and the accession that has the lowest amount of missing data within the taxon was chosen.Three fossil constraints were applied to calibrate divergence times of the molecular phylogeny.The root node including the Benthamidia clade and the outgroups was constrained using a uniform prior of 75-84.2 million years ago (Ma) based on the crown age of Cornaceae s. s. (consisting of only Cornus) estimated from the BEAST analysis of Cornales using six fossil calibrations (Fu et al., 2019).Two internal nodes of the phylogeny were constrained using the lognormal prior.Specifically, the node separating the outgroup cornelian cherries (the Macrocarpium clade) and the Benthamidia-Arctocrania clade was constrained using the oldest fossil of the cornelian cherries (Cornus cf.piggae; Atkinson et al., 2016) from the Campanian (~73 Ma) as the lower bound of the 95% confidence interval (CI) and the minimum age of the root node (75 Ma) as the upper bound.The crown node of the clade Benthamidia (splitting of the North American and eastern Asian subclades) was constrained using 27.82 Ma as the lower bound, based on the oldest fruit fossil of the Benthamidia clade from the mid-Oligocene, Cynoxylon caronii (Mai & Walther, 1978) that represents the North American subclade based on its morphology (Eyde, 1988).The fossil fruit is similar to the fruits in species of the North American subclade Cynoxylon, representing a minimum time of its divergence from the Asian sister.We used the early Oligocene (33.9 Ma) as the upper bound of 95% CI for the node to constrain age largely within the Oligocene.The dating analysis implemented the lognormal relaxed molecular clock using the GTR substitution model with gamma site heterogeneity, four rate categories, and an MCMC length of 2 × 10 8 generations with tree sampling every thousand generations.The lognormal prior was chosen because the ages of the calibrated nodes are more likely to be close to the fossil ages constraining the lower bounds than further away from them, which fits the distribution pattern of the lognormal priors.The Birth-Death model could predict patterns of species diversity over time intervals.We, therefore, used the Birth-Death process tree prior in this analysis to study how species change through time.The log file of the analysis was examined with Tracer v.1.7.0 (Rambaut et al., 2018) to check for convergence of stationary distribution based on the distribution of likelihood scores of the sampled trees and the values of ESS (ESS > 200).After discarding the first 10% of trees as burn-in, the remaining trees were used to generate a maximum clade credibility tree using TreeAnnotator v.1.8.0 (Drummond et al., 2012).The reduced matrix is available at https://doi:10.5061/dryad.hhmgqnkp0.

Inference of biogeographic histories
The extant species of BB are separated in EA and North and Central America, but two fruit fossils of the clade were reported from Europe (EU) (Eyde, 1988;Mai & Walther, 1978).We added the two European fossil taxa onto the dated molecular phylogeny according to their ages and affinities.One of the fossils, Dendrobenthamia tegeliensis dated to the Pliocene (Eyde, 1988), was considered closely related to C. kousa (Eyde, 1988).We placed it next to Cornus kousa subsp.chinensis based on that the divergence time of the two subspecies of C. kousa is likely older than the fossil according to the result from dating analysis.The other fossil, C. caronii dated to the mid-Oligocene (Mai & Walther, 1978), was considered closely related to the extant North American subclade (Eyde, 1988), which has a fruit stone (endocarp) with an ellipsoid shape, like that of the simple/disjunct fruits of clade Cynoxylon.The endocarp of the compound/fused fruit (characteristic of the Asian subclade Syncarpea) has an irregular shape.We, therefore, placed the fossil on the stem of Cynoxylon.The taxa were scored for presence in one or more of the following five areas that covered the entire range of the clade Benthamidia: EA, EU, ENA, WNA, and MCA.The range was divided into these regions based on the endemism of individual species.We conducted the biogeographic analysis in RASP v.4.0 (Yu et al., 2020) based on this modified dated tree including the two fossils and the dispersal-extinction-cladogenesis (DEC) model (Ree & Smith, 2008) to reconstruct ancestral distributions at each node of the tree and infer vicariance and dispersal events.The DEC model was chosen based on the result of the biogeographic model test performed using the BioGeoBEARS (Matzke, 2013) available in RASP, which compared among six methods.The result from the model test suggested that the DEC model was the best fit for our data.In the analysis with DEC, we implemented time slice-dependent dispersal constraints between areas according to the geological data, following Du et al. (2023) (Table S2).In order to obtain a more detailed biogeographic history of the EA subclade Syncarpea, we conducted another DEC analysis focusing on this subclade by dividing EA into six areas according to species distributions that are usually restricted to one of these areas: the Himalayas, Southwest China, South China, East China, Central China, and Korea and Japan (Fig. S2).Southwest China is narrowly defined to include Yunnan, Sichuan, and Guizhou; a rugged and mountainous region, transitioning between the Tibet Plateau and the eastern hills and plains; including the Hengduan Mts., the Sichuan Basin, and the Yungui Plateau.South China is defined to include Hainan, Guangxi, and Guangdong (containing the Nanling Mts).Central China is here defined to include Henan, Shanxi, and Hubei (containing the Dabashan Mts, Dabieshan Mts, and Shennongjia Mts), while East China is defined to include Jiangxi, Zhejiang, and Fujian (containing the Wuyishan Mts).

Estimation of PD and diversification rate
We estimated Faith's PD and net diversification rates in the EA and NA clades to confirm the diversity inequality between the two regions and explore the role of diversification rate as an underlying potential mechanism.We calculated Faith's PD using the "picante" package (Kembel et al., 2010) in R based on the dated phylogeny from BEAST.The 95% CIs of PD were calculated from profile likelihoods.The net diversification rates at each node were estimated in R with the packages of "MEDUSA" (Alfaro et al., 2009) and "geiger" (Harmon et al., 2008), based on the same chronogram used in the estimation of PD.The analysis was conducted using the corrected Akaike information criterion (AICc) and optimal MEDUSA model.We performed these analyses twice, with one including subspecies and the other excluding subspecies to see if results are different.

Comparison of ecological niche and assessment of phylogenetic signal
To explore the role of ecological niche on diversification, we assessed relationship between niche evolution and shifts in diversification rate.We first performed species pairwise comparison of niche similarity using Schoener's D and a standardized measure of Hellinger distance (I) calculated in the R package ENMTools 1.0 (Warren et al., 2021).Niche variables for each species were estimated using ENMTools based on georeferenced herbarium records in the databases of Global Biodiversity Information Facility (https://www.gbif.org) and National Plant Specimen Resource Center (www.cvh.ac.cn) (see Data S1 for citations).We also used ENMTools for grid screening and reserved only one distribution point for each grid.Two species with less than ten records were not included in this analysis (i.e., Cornus "brevipedunculata" and Cornus sunhangii).Furthermore, we used ENMTools to trim the obtained set of climatic variables by removing variables with Pearson correlation coefficient ≥0.85 to reduce collinearity.Eight relatively uncorrelated climatic variables (BIO1, BIO5, BIO6, BIO7, BIO12, BIO13, BIO14, and BIO15) were retained and used to calculate niche similarity indices between species based on the data points of each species.Measures of "0" indicate that the two species under comparison do not overlap at all for the predicted environmental tolerances, whereas measures of "1.0" indicate that the two species completely overlap in their predicated niches (i.e., all grid cells are estimated to be equally suitable for both species).
Next, we extracted the values of elevation and 19 precipitation and temperature variables from the WorldClim v.2.1 data set (Fick & Hijmans, 2017) for each occurrence point using Phytools (Revell, 2012).We used the mean values of these variables within species to reconstruct history of niche evolution on the dated phylogeny from BEAST using fastAnc in Phytools for each niche variable.The results allow direct visualization of niche similarity or divergence among species based on branch colors.We further performed a test of phylogenetic signal for each of the 20 niche variables on the dated phylogeny in Phytools using Pagel's λ (Pagel, 1999) and Blomberg's K (Blomberg et al., 2003) statistics.Values of the statistics less than 1.0 suggest variables being less similar among species than expected from their phylogenetic relationships, whereas values greater than or equal to 1.0 indicate strong phylogenetic signal.

Phylogenetic analysis of concatenated RAD-seq data
The 88 samples had an average of 713 680 reads passed the quality threshold (range from 63 066 to 2 562 467 reads).The filtered reads were clustered into an average of 105 125 clusters per sample (range from 10 628 to 535 317) and had a mean depth of 29.59 (range from 3.61 to 90.88).Among these, an average of 11 710 clusters per sample had a depth greater than six.After the filtering steps, we obtained an initial matrix containing 121 830 loci.Although matrices M40-M70 varied in the number of loci (Table 1), the results of phylogenetic analyses from them showed topologies that were highly similar (trees not shown).The tree from the M50 matrix (1477 loci) had the highest bootstrap support values (Fig. 1).All accessions from the same taxa clustered together, and each of the four major clades of Cornus was recovered.Relationships among them were resolved as the same as that found in Du et al. (2023).
The result of Neighbor-net analysis within clade Benthamidia showed that accessions from the same taxon clustered together (Fig. 2) and the presences of three subclades within the American clade Cynoxylon, six subclades within the eastern Asian clade Syncarpea, and eight subclades within the C. hongkongensis-C.elliptica complex.These results are consistent with the phylogenetic tree based on the concatenated RAD-seq data (Fig. 1).

STRUCTURE analyses of the C. hongkongensis-C. elliptica complex
The results of STRUCTURE analysis of 34 samples of C. hongkongensis and C. elliptica showed that the highest ΔK value was 7.6 when k was 4, while the ΔK values of all the other Ks were below 2.0.An optimal K of four suggests that the sampled individuals were divided among four genetic groups (Fig. 3).Group (1) included samples representing Cornus hongkongensis subsp.tonkinensis, Cornus hongkongensis subsp.melanotricha, and Cornus hongkongensis subsp.gigantea; Group (2) included samples representing Cornus hongkongensis subsp.ferruginea; Group (3) included samples representing Cornus hongkongensis subsp.hongkongensis; and Group (4) included samples representing C. elliptica.Samples representing C. hongkongensis subsp.elegans combines the genetic components of Cornus hongkongensis subsp.hongkongensis and C. elliptica, with the proportion of C. hongkongensis subsp.hongkongensis slightly greater (Fig. 3).Because the result from species delineation analysis using BPP (see below) suggested 10 genetic groups, we also report the result from STRUCTURE analysis using "k = 10."Its likelihood Ln′(K) was −588 117, much lower than −124 304 of "k = 4." The result with "k = 10" revealed the six subspecies of C. hongkongensis and C. ellipica (subclades in Fig. 1) in distinct genetic groups, however, with samples of C. hongkongensis subsp.tokinensis splitting into two genetic types.One of the types has approximately 2/5 of the genetic component from an unknown ancestor.The result from "k = 10" also shows that C. hongkongensis subsp.gigantea has approximately 1/3 of the genetic component from C. hongkongensis subsp.melanotricha, and C. hongkongensis subsp.elegans has approximately 1/7 of the genetic component from C. elliptica.

Species delimitation and coalescent species tree estimation
The two independent runs for each of the analyses produced consistent results.In the "blind" A11 analysis of the 79 samples of the clade Benthamidia, 27 genetic groups were delimited in BPP (for details of sample assignments, see Table S3).The "species tree" contains 10 genetic groups from the C. hongkongensis-C.elliptica complex and the remaining 17 genetic groups from other taxa of the clade (Fig. S3).The species relationships within the BB clade shown in this "species tree" were the same as the "global" phylogeny from RAxML and Bayesian analyses containing all samples based on the concatenated RAD-seq data (Fig. 1).However, relationships of subspecies within the C. hongkongensis-C.ellipcia clade were not identical to those revealed in Fig. 1 due to a different placement of C. hongkongensis subsp.tokinensis.This species tree shows Cornus kongkongensis subsp.tokinensis being sister of the clade Cornus kongkongensis subsp.melanotricha-Cornus kongkongensis subsp.gigantea (Fig. S3), rather than being the sister of the clade consisting of the remaining subspecies and C. elliptica (Fig. 1).
The normal A11 analysis with samples annotated to taxa identified 17 "species" within the clade Benthamidia, each represented by a single branch (Fig. 4).The 10 genetic groups in the C. hongkongensis-C.elliptica complex were sorted into seven species matching the six respective subspecies of C. hongkongensis and the species C. elliptica, suggesting raising these subspecies to the species rank.Genetic groups of the other eight big-bracted species were sorted into their respective branches.For the three American species, the four genetic groups of C. florida collapsed into two branches corresponding to the two current subspecies, suggesting raising the subspecies rank to species.The two genetic groups of either C. nuttallii or C. disciflora were combined into one branch.In the EA clade, C. multinervosa and the three existing species of the C. capitata complex were sorted into four respective branches.The two subspecies of C. kousa remained as two branches, with the three genetic groups of Cornus kousa subsp.chinensis collapsed into one branch while Cornus kousa subsp.kousa remained as a single branch, suggesting recognition of two species.

Divergence time estimation
The result of divergence time analysis (Fig. 5) showed that the divergence of the BB clade from its sister group the dwarf dogwoods was 66.36 Ma (95% highest posterior density (HPD): 60.51-71.66Ma), near the end of the

Inference of biogeographic history
The result of biogeographic analysis using the DEC model showed that the common ancestor of the BB clade Benthamidia was most likely widespread in four areas (EA-A, EU-B, WNA-D, and Mexico-Central America-E; ABDE) in the mid-Oligocene (Fig. 6).The ancestral ranges of the Cynoxylon clade and the crown clade of its extant species were inferred to be ABDE and ADE, respectively, in the late Oligocene.Clade Syncarpea was shown to have an ancestral distribution in eastern Asia before the mid-Miocene and experienced two episodes of rapid diversification, one in the mid-Miocene, resulting in the presently recognized species and the other in the Miocene-Pliocene boundary, resulting in the various subspecies in C. hongkongensis, C. elliptica, and subspecies in C. kousa.Divergence of the two subspecies of C. florida was also inferred to have occurred in the Pliocene-Pleistocene boundary from a common ancestor widespread from eastern NA to Mexico (Fig. 6).The biogeographic analyses inferred two dispersal events and four vicariance events in the biogeographic history of clade Benthamidia.The vicariance events involved the splits of the wide distributed ancestors during speciation in late Oligocene-early Miocene and the divergence of subspecies in C. kousa and C. florida in the Pliocene or Pleistocene.The dispersal events involved spread from EA to EU of C. kousa, and the spread from Mexico to eastern NA of C. florida, both in the Neogene.
The biogeographic analysis focusing on the eastern Asian clade Syncarpea suggested that the clade have experienced six dispersal and four vicariance events within Eurasia (Fig. 7).The ancestor of clade Syncarpea had the most likely ancestral distribution in Southwest China.

Diversification rate and PD
The estimated net diversification rates and their 95% CIs within the BB clade Benthamidia are shown in Fig. 8.The results showed a general pattern of rate increases in the mid-Miocene in EA in the lineage diversifying to C. multinervosa, C. kousa, and C. hongkongensis and its two subclades of subspecies.The results also showed rate decreases in the clade Cynoxylon in NA and a subclade of the Asian clade Syncarpea that speciated into C. capitata, C. sunhangii, and C. "brevipedunculata" (Fig. 8).The diversification rates evolved from 0.069 (95% CI: 0.039-0.11) in the common ancestor of the clade Benthamidia to 0.083 (95% CI: 0.043-0.142) in the common ancestor of clade Syncarpea in EA; from which, the rates further increased to 0.093, 0.105, and 0.119 in the three successive upper nodes in the subclade C. multinervosa-C.kousa-C.hongkongensis.In contrast, the rates dropped from 0.083 to 0.023 in the other subclade of the Asian species, and to 0.028 and 0.027 in the NA clade.When subspecies were excluded in the analysis, the diversification rate of the EA clade (0.043) was still three times higher than that of the NA clade (0.015).The Faith's PD for the clade Benthamidia was estimated as 216.62 and the PD of the EA and NA clades were estimated as 143.19 and 73.42, respectively, showing greater diversity in EA, approximately twice the level as in NA (Fig. 8).When subspecies were excluded in this analysis, the PDs for the clade Benthamidia, the EA subclade, and the NA subclade were lowered to 175, 104, and 71, respectively, but the pattern of variation is consistent, still showing a greater level of diversity in EA, 1.5 times the level as in NA.

Niche similarity and phylogenetic signal
The result of niche similarity comparisons showed a range of 0.21-1.0for Schoener's D and 0.19-1.0for Hellinger Distance I values (Table 2).In general, the D and I values were similar but varied in whether D or I were slightly greater for the same species pair comparison.Among all the comparisons, C. nuttallii and Cornus kousa subsp.kousa showed the greatest differences with other species, with D and I values < 0.65, and most <0.2), suggesting the mountain regions in southern China and Mexican plateau are similar in overall climate, and species in each region occupy highly similar environments and do not evidently diverge in climatic niches.
A closer examination of the niche variables via comparison of altitude and 19 bioclimatic variables among species were generally congruent with the findings from the similarity comparisons of total niches, showing divergence between C. kousa subsp.chinensis and C. kousa subsp.kousa in six of the 20 variables and three of the eight uncorrelated variables, whereas divergence between C. florida subsp.florida and C. nuttallii in 15 of the 20 variables and seven of the eight uncorrelated variables (Figs.S4-S23).Species from mountains of southern China and Mexican plateau showed similarities in five of the eight uncorrelated variables between species and regions and divergence in three of the eight variables between regions.Five of the 20 variables exhibited heterogeneity among these species and within regions, especially BIO12 (Annual Precipitation), BIO14 Fig. 6.Reconstructed ancestral distributions for the big-bracted dogwoods from dispersal-extinction-cladogenesis analysis in RASP using the BEAST tree including fossils.Distribution ranges are indicated as follows: A, eastern Asia (EA); B, Europe (EU); C, eastern North America (ENA); D, western North America (WNA); E, Mexico and Central America (MCA).Asterisks indicate fossil species.Pie charts represent the marginal probabilities for each alternative ancestral area.The most likely areas are annotated.Red arrows indicate the direction of two dispersal events.Blue arrows indicate four vicariance events.The time scale is in millions of years.The map above shows distribution areas and dispersal routes.
(Precipitation of Driest Month), and Altitude, varying between terminal sister lineages.For the eight uncorrelated variables, species and subspecies within each region were similar for most of the variables although they showed differences in one or two variables except for C. disciflora and Cornus florida subsp.urbiniana which differed in four variables (Figs.S15, S17, S19, and S21).All the sister branches on the phylogeny showed differences in mean value of one or more of the 20 variables (Figs.S4-S23).Among the eight uncorrelated variables (BIO1,5,6,7,12,13,14,and 15), BIO12 showed differences in nine terminal sister branches, BIO14 and BIO15 in seven terminal sister branches, BIO5 and BIO6 in five terminal sisters, while BIO7 and BIO13 diverged in threeterminal sisters and BIO1 diverged in four terminal sisters.
Tests of phylogenetic signal using Pagel′s λ and Blomberg′s K statistics for the 20 variables revealed strong phylogenetic signals for only two bioclimatic variables (BIO2 Mean Diurnal Range and BIO18 Precipitation of Warmest Quarter) (Table S4), indicating the evolution of the clade Benthamidia has been shaped by phylogenetic conservatism of these two factors and closely related species are more similar in these two bioclimatic factors than expected from more distantly related species.A majority of the bioclimatic factors were, however, not detected for significant phylogenetic signals, indicating they are not more similar in closely related species than in more distantly related species.This is consistent with the result of niche similarity test.

Species delimitation in EA clade Syncarpea and relationships within the BBs
The eastern Asian clade Syncarpea was classified in Flora of China by Xiang and Boufford (2005) into five species (Cornus capitata, Cornus multinervosa, Cornus kousa, Cornus hongkongensis, and Cornus elliptica), with Cornus hongkongensis further classified into six subspecies and C. kousa into two subspecies.All of the subspecies in C. hongkongensis were initially published as species (Fang, 1953;Hu et al., 1990).Xiang & Boufford (2005) also merged Cornus "brevipedunculata" into C. capitata and separated Cornus capitata subsp.angustata out as C. elliptica.Recently, a new species Cornus sunhangii was published from the Himalayan region within the range of C. capitata (Lv et al., 2019).This species is morphologically similar to C. capitata, but was considered to differ from C. capitata in infructescence, peduncle, and trichomes (Lv et al., 2019).Our results from the A11 analysis of BPP for species delimitation based on the multispecies coalescent (MSC) method supported 13 species in the EA clade (Fig. 5), corresponding to 13 clades in the phylogeny (Figs. 1, 2, and 4).These included all of the species previously recognized in Xiang & Boufford (2005), as well as the subspecies of C. kousa, C. capitata, and C. hongkongensis, such as Cornus kousa subsp.chinensis, Cornus capitata subsp.brevipedunculata, Cornus hongkongensis subsp.ferrugnea, Cornus hongkongensis subsp.melanotricha, Cornus hongkongensis subsp.tokinensis, Cornus hongkongensis subsp.gigantea, Cornus hongkongensis subsp.hongkongensis, and Cornus hongkongensis subsp.elegans.These results support resurrecting the species status of the subspecies in C. hongkongensis and C. capitata subsp.brevipedunculata and upranking the two subspecies of C. florida and two subspecies of C. kousa to species.Results from STRUCTURE, Neighbor-Net, and divergence time analyses suggest that introgression or gene flow are present among the more recently diverged species within the C. hongkongensis-C.elliptica clade.Morphologically, there are overlaps among these taxa as noted in Xiang (1987) and Xiang & Boufford (2005), adding evidence for interspecies introgression.The recent divergence, introgression, and morphological overlap of these taxa within the C. hongkongensis-C.elliptica complex may as well support them having the subspecies status (see arguments below).Although the BPP analyses of molecular data under the MSC model have been commonly used as a genetic and quantitative approach for species delimitation to provide evidence supporting changes of species' taxonomy (e.g., Barley et al., 2018;Luo et al., 2018;Chambers & Hillis, 2020), it has been noted that the genetic breaks identified by the MSC-based method does not necessarily represent species boundaries, but may represent population structures within species (Sukumaran & Knowles, 2017;Barley et al., 2018;Luo et al., 2018;Leaché et al., 2019;Chambers & Hillis, 2020).Thus, delineating species solely based on the results from the BPP analysis could potentially lead to oversplitting of species within a species complex.Similarly, delineating species following the phylogenetic species concept (PSC) that is based on historical monophyly (Wheeler & Meier, 2000) of individuals/populations can also 8. Phylogenetic diversity (PD) and diversification rates of major lineages within the big-bracted dogwoods (BB) inferred from the BEAST chronogram.PD of BB and its eastern Asian (EA) and North American (NA) clades are shown in square brackets.Diversification rates are in bold next to the nodes.Their 95% confidence intervals are in parenthesis.The red branches indicate increases of rates in the EA clade.The habits (deciduous or evergreen) of each taxa are annotated.potentially lead to taxonomic inflation, where monophyletic structures within species, such as known subspecies or subclades, are raised to species, causing an explosion in the number of species (Isaac et al., 2004;Zachos & Lovari, 2013).
In the case with the C. hongkongensis-C.elliptica complex, although the species delimitation analysis by BPP suggested seven species in the C. hongkongensis-C.elliptica complex (Fig. 4) and each of these "species" were resolved as a monophyletic group on the phylogeny (Fig. 1), classification of the complex into seven species may lead to oversplitting of the species complex because the subclades in the complex may represent diverging populations within a species as stated above.Based on that the divergence times of these subclades within the species complex were much more recent (in the Pliocene) compared to those of other species of the EA clade (in the mid-Miocene; Fig. 5), that introgression is present among them (Figs. 2, 3), and that they are morphologically not easy to distinguish (Table S5) and there are overlaps or continuity among them for the diagnostic features, these taxa in the C. hongkongensis complex may be better treated as subspecies, if an integrative approach is used.They were initially published as species based on some specimens showing differences of various parts, but the variations were found to overlap among them and were associated with geographic distribution, based on the examination of large number of herbarium specimens (Xiang, 1987).These taxa are largely allopatric with some overlapping at the edges of their range for some of them.Cornus hongkongensis subsp.ferruginea appears to be within the range of Cornus hongkongensis subsp.hongkongensis, but it occurs at lower altitudes (Fig. S23).Such case well fits the subspecies concept adopted in Ellison et al. (2014) defining subspecies as "a group of individuals only in cases where there is strong supporting evidence of incomplete differentiation, distinct geographic distribution, at least one clearly fixed phenotypic difference, or genetic differentiation that confers the possible evolutionary potential for speciation to occur." Subspecies in other species of the Benthamidia clade, such as Cornus florida subsp.urbinana and C. kousa subsp.chinensis, were also identified as species in the species delimitation analysis (Fig. 4).The subspecies within these species also diverged in the Pliocene (Fig. 5).They are geographically allopatric and morphologically overlapping (Xiang, 1987, and personal observation;Murrell & Poindexter, 2016), similar to those within C. hongkongensis, also favoring the subspecies status.The two subspecies of C. kousa are largely disjunct in Japan-Korea and Central China (Xiang & Boufford, 2005), while the two subspecies of C. florida are disjunct in ENA and Mexico (Murrell & Poindexter, 2016).However, occasional gene flow could occur via long-distance drupe dispersal by birds.These subspecies are probably on their ways of diverging into species.As for C. capitata subsp.brevipedunculata, due to the early divergence (mid-Miocene) of the lineage, we tentatively accept its species status.This taxon is rare in the herbaria and fields and is poorly known.Further analyses of molecular and quantitative morphological data with extensive sampling including the type specimens or type locations and contact zones between the delimited subspecies of all species will be helpful to test the species delimitation in this study.These analyses may reveal additional subclades within or outside of the species and subspecies resolved in this study.A rank-free PhyloCode (Cantino & de Queiroz, 2020) based classification of the BBs using clade names will be preferred which has the benefits of stabilizing names and being predictive of phylogenetic relationships.We tentatively keep the six subspecies of C. hongkongensis and treat C. elliptica as a subspecies within C. hongkongensis.But the alternative sevenspecies scenario of the C. hongkongensis-C.elliptica complex warrants further consideration.A list of species and subspecies of the BBs with diagnostic morphological features is provided in Table S5.
The species relationships within the Benthamidia clade of Cornus revealed in the present study were consistent with the finding in a recent study of Cornales using 353 nuclear genes (Thomas et al., 2021), but incongruent with the finding in the study of Cornales using the plastid genome sequences (Fu et al., 2019), which supported paraphyly of the North American species, with Cornus nuttallii from WNA being the sister of the remaining species of the big-bracted clade.This incongruence was also revealed in the recent study on the entire Cornus clade (Du et al., 2023) and was explained by incomplete lineage sorting associated with early rapid diversification of clade leading to three subclades, one in EA (subclade Syncarpea), one in WNA (C.nuttallii), and one in MCA-ENA (C.disciflora-C.florida).Rapid divergence of these three lineages was also supported by our data from divergence time analysis (Fig. 5).The three NA species are morphologically distinct, but the two temperate deciduous species C. nuttallii and C. florida have long been considered more closely related to each other than to the evergreen Mexican species C. disciflora which was often classified as a separate subgenus due to the lack of expanded petaloid bracts (see Eyde, 1988;Murrell, 1993;Xiang et al., 1993Xiang et al., , 1996Xiang et al., , 2006)).Cornus disciflora is morphologically similar to C. florida in having distinct fruits, four small bracts protective to the inflorescence buds in winter.The bracts are enlarged and petaloid in the spring to attach pollinators in C. florida but fall off before expansion in C. disciflora.Cornus nuttallii is distinct from them by having fruits packed tightly into angular shape, four to six bracts, and bracts nonprotective to the inflorescence buds in winter (inflorescent buds exposed in winter; like evergreen species C. capitata and C. hongkongensis in EA).Our study clearly shows that C. florida is more closely related to C. disciflora than to C. nuttallii (Figs. 1, 2, and 4), indicating that previous taxonomists have overemphasized the early deciduous feature of bracts in C. disciflora.

Diversity anomaly between EA and NA and underlying causes
An overall greater species richness and PD in the flora of EA compared to the flora of NA is rooted in the diversity of the component lineages of each flora.Examining the diversity pattern and underlying causes of lineages occurring in both floras can shed light on the relative contributions of lineage age and diversification rate to this large-scale biodiversity pattern.These factors are not mutually exclusive but likely vary in their importance in generating unequal diversity between EA and NA in different lineages, as shown in previous studies (e.g., Lindelof et al., 2020;Zhou & Xiang, 2022).To obtain a global view of these factors on the diversity disparity between EA and NA, data from more, ideally all, individual lineages occurring in these areas are needed.In this study of the BB clade, the greater species richness in EA than in NA is congruent with a greater PD measured based on the RAD-seq data.Our results from diversification rate estimation based on a dated phylogeny also revealed a higher rate of net diversification in EA than in NA (Fig. 8).Without evidence from fossils to suggest differential level of extinction between the two regions, our data suggest that a greater speciation rate in EA was most likely the major factor contributing to the greater diversity of the BBs in EA.
Evidence supporting greater speciation/diversification in EA was also shown in Lysimachieae tribe (Primulaceae) (Yan et al., 2018), Torreya (Taxaceae) (Zhou et al., 2022), and many disjunct lineages (e.g., Asarum, Illicium, Panax, and sect.Cyrta of Styrax reviewed in Xiang et al., 2004; Picea, Tsuga, Woodwardia, and Zelkova reviewed in Harris et al., 2013).In the BBs, the greater speciation rate in EA was likely promoted by geographic isolation.The evergreen species of the BBs (clade Benthamidia) mainly occur at latitudes between 20°N and 30°N, in mountains of southern China extending to adjacent countries in Asia and in Mexican plateau extending to Central America.All evergreen species have high level of overall niche similarities (D and I values > 0.9; Table 2), but are largely restricted to allopatric areas of the mountains (see distribution maps in Figs.S1, S2).Southern China (south of Qinling Mountain and Huaihe River) is more than twice the size of the Mexican plateau and can be subdivided into Southwest, South, Central, and East parts.The Asian species are largely restricted to mountains in one of these areas.The evidence suggests more opportunities for geographic isolation in promoting greater speciation in EA, particularly in southern China and its adjacent areas (e.g., India and Vietnam), leading to greater diversity of evergreen species (four species in EA vs. 1 species in NA).It must be noted that although the overall niche similarity among the evergreen species in EA is high, these species (i.e., C. capitata, C. sunhangii, C. hongkongensis-C.elliptica complex) do differ to some extents in the elevation of their distributions (Fig. S23 S20).This observation suggests ecological niche shift may have, as well, been a minor driver, in concert with geographic isolation, for speciation of some EA species.
A recent study by Hu et al. (2022) found that China had preserved a higher proportion of genera that originated before the Miocene than the United States, which was explained by both the complex topography in China that provided numerous refugia for ancient lineages and less extinction during climate cooling and glaciation.They suggested that southern China bears the signature of both old and new diversification and is a diversity center for both anciently and recently originated genera.This hypothesis is supported by the findings of a recent spatial phylogenetic study that showed most grid cells in China with significantly high phylogenetic endemism were located in the mountainous regions of southern China which contain both ancient relictual lineages and recently evolved lineages (Zhang et al., 2022).Mountainous regions with high topographical diversity harbor numerous microhabitats, and these are not only excellent refugia to buffer disturbances but also cradles for allopatric speciation (Wen et al., 2014).Our study adds another example supporting these findings.

Biogeographic history of the BBs
The extant species of the clade Benthamidia exhibit the typical EA-NA disjunction pattern whose origin has attracted much research (e.g., Xiang et al., 1998;Xiang & Solits, 2001;Donoghue & Smith, 2004;Wen, 1999;Wen et al., 2010Wen et al., , 2016;;Harris et al., 2013).Our results from historical biogeographic analysis using the DEC method with a time slice model are consistent to those of Du et al. (2023), suggested a trans-Beringian distribution of its ancestor in Eurasia and WNA (including Mexico) in the mid-Oligocene (ABDE in Fig. 6), supporting trans-Beringian exchanges of plants during the end of Paleogene.The ancestral distributions revealed from our biogeographic analysis (Fig. 6) suggest a most likely peripheral origin of the EA clade Syncarpea within the widespread ancestor of Benthamidia.The intercontinental disjunction of clades Syncarpea and Cynoxylon was then likely a result of extinction of Cynoxylon in Eurasia, as evidenced by the presence of Oligocene fossil species Cynoxylon caronii.This mode of origin of the EA-NA disjunction is an exception of the "out of Asia" pattern found in previous synthesis (Donoghue & Smith, 2004) and does not fit either the dispersal or vicariance model.
Following the disjunction, our results suggested that the ancestor of clade Cynoxylon diversified into three extant species in WNA and Mexico in the late Oligocene (divergence of C. nuttallii) and in the mid-Miocene (divergence of C. florida and C. disciflora) via vicariances, followed by extinction in EA.One of the species, C. florida, subsequently dispersed into ENA.The dispersal likely occurred before the climatic cooling in the Pliocene that had promoted the grassland formation separating forests in ENA and WNA (Tiffney, 1985).The grassland formation resulted in the isolation of two subspecies of C. florida in ENA and eastern Mountains of Mexico.This history suggests a close affinity of the mountain flora of eastern Mexico to the deciduous forests of eastern United States and the presence of a potential corridor for plant migration between Mexico and eastern United States during the early Neogene.It also suggests a more ancient mountain flora in WNA and Mexico compared to deciduous forests in ENA.
The eastern Asian clade Syncarpea split from the NA clade in the late Oligocene, with an extant distribution extending from the Himalayas northeastward to southern China, Korea, and Japan (Xiang & Boufford, 2005).Our biogeographic analysis suggests that the extant species of the clade had a common ancestor likely in Southwest China during the mid-Miocene (Fig. 7).The latest fossil and tectonic evidence suggest that uplift of the Qinghai-Tibet Plateau led to the formation of the Hengduan Mts. in Southwest China since the Eocene-Oligocene transition (Su et al., 2019), providing diverse microhabitats for plant lineages to diversify (Ding et al., 2020)

Fig. 2 .
Fig. 2. Phylogenetic network of clade Benthamidia generated by Neighbor-net analysis of unlinked SNP data using SplitTree4.The two main clades Cynoxylon and Syncarpea are annotated.

Fig. 3 .
Fig. 3. Result of STRUCTURE analysis using unlinked SNP data of Cornus hongkongensis and Cornus elliptica samples.Eight subclades of the C. hongkongensis-C.elliptica species complex in the phylogenetic tree of Figure 1 are indicated with arrows.The distribution areas of each taxon are annotated.The best K value is 4. When K is 10, each of the eight subclades has a unique genetic component.

Fig. 4 .
Fig. 4. The results of species delimitation analysis of the big-bracted dogwoods (BB) and outgroup species of Cornus using BPP.The Cornus hongkongensis-Cornus elliptica species complex, two clades of BB, and four groups of Cornus are annotated.Taxa belonging to the same species are collapsed into a single branch.Each of the six subspecies of C. hongkongensis forms a single branch.The two subspecies of C. kousa or C. florida are not collapsed into a single branch.
5. The niches of four deciduous taxa in the warm-temperate and subtropical regions, C. kousa subsp.chinensis and C. kousa subsp.kousa from EA and C. florida subsp.florida and C. nuttallii from NA, were substantially divergent.The niches of C. florida subsp.florida and C. nuttallii separated in ENA and WNA had very low level of overlapping

Fig. 5 .
Fig. 5. Time-calibrated phylogeny inferred from RAD-seq data for the big-bracted dogwoods (BB) and outgroup species using BEAST.Numbers 1-3 highlighted in brown solid circles indicate fossil-calibrated nodes.Median values of divergence time in millions of years are shown at each node.Blue bars and values in parentheses indicate the 95% highest posterior density values.The four groups of Cornus and two clades of BB are indicated.

Fig. 7 .
Fig. 7. Reconstructed ancestral distributions for Clade Syncarpea of the big-bracted dogwoods from dispersal-extinctioncladogenesis analysis in RASP using the BEAST tree including fossil.Distribution ranges are indicated as follows: A, Southwest China; B, the Himalayas; C, South China; D, East China; E, Central China; F, Korea and Japan; G, Europe.Regions A, C, D, and E are subdivisions of southern China.Asterisk indicates fossil species.Pie charts show the marginal probabilities of each alternative ancestral area.The most likely areas are annotated.Red arrows indicate the direction of six dispersal events.Blue arrows indicate four vicariance events.The time scale is in million of years.The map above shows distribution areas and dispersal routes.
) and some climatic parameters (such as BIO5 of Max Temperature of Warmest Month, Fig. S8; BIO6 of Min Temperature of Coldest Month, Fig. S9; BIO12 of Annual Precipitation, Fig. S15; BIO14 of Precipitation of Driest Month, Fig. S17; and BIO17 of Precipitation of Driest Quarter, Fig.

Table 1
Its two immediate descendant subclades also had the most likely ancestral distributions in Southwest China.One of the subclades (consisting of C. "brevipedunculata"-C. sunhangii-C.capitata) expanded range from Southwest China to the Himalayas since the middle Miocene.Whereas the other subclade speciated in Southwest China forming C. multinervosa and its sister clade (consisting of C. kousa-C.hongkongensis).The latter spread from Southwest China to South China and Central China, and then diverged into two lineages in Central China and South-Southwest China.The C. kousa lineage later dispersed from Central China to Korea and Japan, while the C. hongkongensis-C.elliptica lineage vicarianced between Southwest China and South China during the late Miocene.Two taxa C. hongkongensis subsp.elegans and C. elliptica further expanded from South China to East China.During the Pliocene, C. kousa split into two lineages, one in Korea and Japan, and one extended from China to EU.

Table 2
Similarity tests of niches using Schoener's D and a measure derived from Hellinger distance (I) in ENMTools The lower left is D, and the upper right is I. Measures of 0 indicate that species-predicted environmental tolerances do not overlap at all, whereas measures of 1 indicate that all grid cells are estimated to be equally suitable for both species.If values of both D and I are equal to or greater than 0.90, they are marked in boldfaces.Species from mountain regions in southern China (C.elliptica, C. hongkongensis, C. multinervosa, and C. capitata) and Mexican plateau (C.disciflora and C. florida subsp.urbiniana) have high values of niche similarity between each pair of species.
, leading to the origins of C. "brevipedunculata," C. sunhangii, and C. capitata in Southwest China.Another lineage of clade Syncarpea dispersed eastward to South China and further to Central and East China during the middle Miocene, resulting in diversification of this lineage into C. multinervosa, C. kousa, and the C. hongkongensis-C.elliptica complex.This history suggests that Southwest China might act as a museum during the global climatic cooling in the Oligocene but as a cradle for floral expansion during the Miocene.