Reassessment of the Phylogeny and Systematics of Chinese Parnassia (Celastraceae): A Thorough Investigation Using Whole Plastomes and Nuclear Ribosomal DNA

Parnassia L., a perennial herbaceous genus in the family Celastraceae, consists of about 60 species and is mainly distributed in the Pan-Himalayan and surrounding mountainous regions. The taxonomic position and phylogenetic relationships of the genus are still controversial. Herein, we reassessed the taxonomic status of Parnassia and its intra- and inter-generic phylogeny within Celastraceae. To that end, we sequenced and assembled the whole plastid genomes and nuclear ribosomal DNA (nrDNA) of 48 species (74 individuals), including 25 species of Parnassia and 23 species from other genera of Celastraceae. We integrated high throughput sequence data with advanced statistical toolkits and performed the analyses. Our results supported the Angiosperm Phylogeny Group IV (APG IV) taxonomy which kept the genus to the family Celastraceae. Although there were topological conflicts between plastid and nrDNA phylogenetic trees, Parnassia was fully supported as a monophyletic group in all cases. We presented a first attempt to estimate the divergence of Parnassia, and molecular clock analysis indicated that the diversification occurred during the Eocene. The molecular phylogenetic results confirmed numerous taxonomic revisions, revealing that the morphological characters used in Parnassia taxonomy and systematics might have evolved multiple times. In addition, we speculated that hybridization/introgression might exist during genus evolution, which needs to be further studied. Similarly, more in-depth studies will clarify the diversification of characters and species evolution models of this genus.


INTRODUCTION
Parnassia L. (Celastraceae) comprises approximately 60 perennial herbaceous species, and is widely distributed in the Arctic and temperate regions of the Northern Hemisphere (Phillips, 1982;Hultgard, 1987;Gu and Hultgård, 2001;Simmons, 2004). Most Parnassia species are endemic to China and have limited distribution, with >30 species only restricted to the Pan-Himalayan region (Hultgard, 1987), which is considered as an important diversity center of Parnassia (Handel-Mazzetti, 1941;Phillips, 1982;Ku, 1987;Wu, 2003;Simmons, 2004). The phylogenetic position of the genus is debatable, and prior studies showed that Parnassia should be treated as a member of Saxifragaceae (Hooker and Thomson, 1858;Engler, 1930;Cronquist and Takhtadzh " ian, 1981;Gu and Hultgård, 2001). However, Parnassia had also been suggested being closely related to Droseraceae morphologically (especially with the characteristics gynoecium; Pace, 1912;Dahlgren, 1980), to Hypericaceae (based on androecium and Chemotaxonomy through flavonoids; Arber, 1913;Jay, 1970), to Nymphaeaceae (based on vegetative characters; Hallier, 1901), and to Ochnaceae (based on peculiar staminodes; Bennett, 1869). Many systematists (e.g., Drude, 1875;Murbeck, 1918;Sharma, 1968;Klopfer, 1972;Hultgard, 1987) even suggested that Parnassia should be recognized as a separate family or order (e.g., Takhtadzhian and Takhtajan, 1997;Wu, 2003). The taxonomic debate over the genus seems to have ended over the last decades, especially with the advancement in molecular systematics. In contrast to the long-debated morphology-based systematics, more consistent results of the phylogenetic position of Parnassia were revealed by molecular studies, which suggested that Parnassia might be more closely related to Celastraceae than other families (Savolainen et al., 1997;Soltis et al., , 2000Simmons et al., 2001). Other studies subsequently confirmed this result, and Parnassia has been recovered as an early derived lineage of Celastraceae (Zhang and Simmons, 2006;Simmons et al., 2012;Bacon et al., 2016;Li et al., 2019). However, all these studies were based on a limited number of species, especially for those from its center of distribution and differentiation. Similarly, the molecular DNA markers used in these studies also lacked sufficient phylogenetic signals (Fischer and Steel, 2009;Philippe et al., 2011). Therefore, it is essential to provide molecular markers with sufficient informative characters to reconstruct the phylogenetic relationship of Parnassia.
Besides the debatable phylogenetic position of Parnassia, infrageneric systematics of the genus were also problematic. The distribution of Parnassia species in the Pan-Himalayan region often overlaps, and considerable morphological variation may occur within the same species. Scholars often have a different understanding of taxonomic traits, which led to different opinions about the infrageneric classification of the genus, including different ranks as section and series (Drude, 1875;Franchet, 1897;Engler, 1930;Handel-Mazzetti, 1941;Ku, 1987;Wu, 2003Wu, , 2005Shu, 2017), and a plethora of taxonomic revisions (Turner and Veldkamp, 2001;Wu et al., 2004Wu et al., , 2008Wu et al., , 2009Shu andZhang, 2016, 2017;Wang et al., 2018;Yu et al., 2018;Ma et al., 2020;Dai et al., 2021). However, most of these investigations were based on morphological characters and need to be validated with molecular phylogenetic studies. Molecular phylogenetics for the genus based on nuclear ITS as well plastidial trnL-F and trnT-L markers suggested that Parnassia is monophyletic and that parallel evolution of morphological traits may occur (Wu, 2005). Yu (2019) reconstructed the phylogeny of Parnassia with 19 high polymorphic fragments of the plastid genome and suggested that the complete plastid genome dataset would be more comprehensive. Different studies have revealed that phylogenetic resolution and reliability seem to be related to the taxon as well as gene sampling (Wortley et al., 2005;Philippe et al., 2011). Since prior molecular phylogeny analyses on Parnassia are inadequate, there is an urgent need to construct a stable and precise phylogenetic tree to substantiate or negate the hypotheses made about Parnassia infrageneric systematics.
With the rapid development of next-generation sequencing technology, the amount of data used in phylogenetic research increased exponentially. Large-scale genomic data, such as plastid genome sequences and nuclear ribosomal DNA (nrDNA) data, can provide more phylogenetic informative sites than those of only a few DNA fragments. Recently, the plastid genome data has been successfully applied to resolve long-standing debates in phylogenetic positions of different taxa: e.g., the phylogenetic positions of Paeoniaceae within Saxifragales (Dong et al., 2018), Selaginellaceae , and sanguinolenta group (Zhang H. R. et al., 2020). In addition, the plastid genome has also been employed to resolve the phylogenetic relationships of various plant lineages (Zhai et al., 2019;Liu et al., 2020;Ma et al., 2021). However, few plastid genomes have been sequenced for Parnassia and Celastraceae so far, which hinders the process of phylogenetic study of Parnassia (e.g., Wu, 2005;Yang et al., 2012).
With the development of molecular systematics, the phylogenetic research of Parnassia has made rapid progress. However, there is still a lack of large-scale data containing substantial phylogenetic signals (e.g., the plastid genome) for inferring the phylogenetic relationship of the genus. In addition, a large number of taxonomic revisions of Parnassia in recent years need to be confirmed by more reliable molecular systematics. Therefore, the present study attempts to resolve infrageneric phylogenetic questions about Parnassia using the whole plastid genomes and nrDNA (18S-ITS1-5.8S-ITS2-26S) data. The sampled species included seven of the nine sections of Chinese Parnassia (Ku, 1995;Wu, 2003) and covered the major species of Parnassia in the Pan-Himalayan region. The aims of this study are (1) to provide a robust and highly resolved phylogenetic tree for the phylogenetic position of Parnassia; and (2) to reassess the reliability of morphological features that have been widely used for taxonomic and phylogenetic considerations.

Sampling, Genome Sequencing, and Annotation
We sampled a total of 74 individuals, including 25 species (48 individuals) of Parnassia and 23 species (24 individuals) from all other 14 genera of Celastraceae, and two outgroups (Supplementary Table 1). To show the relationships within the species, we included two or more individuals from different localities for some species. Voucher specimens for the species sampled were deposited in the Qinghai-Tibetan Plateau Museum of Biology (HNWP), Northwest Institute of Plateau Biology, Chinese Academy of Sciences. Besides these 74 individuals, we retrieved 89 plastid genomes from the GenBank (Supplementary Table 2). In the final dataset, most orders of Superrosids were included, and two individuals of Asteraceae were selected as outgroups. We re-annotated the downloaded plastid genome sequence files using PGA (Qu et al., 2019) to ensure the accuracy of the gene coding sequences (CDSs) data obtained from GenBank in the analysis.
We extracted total genomic DNA using about 10 mg of silicadried leaf tissue through modified CTAB protocols (Doyle and Doyle, 1987). The extracted DNAs of all the individuals were then sent to Novogene (Beijing, China). Genomic DNA library was generated using NEB Next R Ultra TM DNA Library Prep Kit for Illumina (NEB, United States) following the manufacturer's recommendations, and index codes were added to each sample and sequenced on an Illumina HiSeq 2500 sequencer (San Diego, CA, United States) using the paired-end option (2 × 150 bp). Sequencing quality was assessed through FastQC v. 0.11.8 (Andrews, 2010). Where necessary, we used Trimmomatic v. 0.33 (Bolger et al., 2014) to clean the sequencing data. The plastid genomes and nrDNA (18S-ITS1-5.8S-ITS2-26S) were assembled de novo using GetOrganelle (Jin et al., 2020) and then visually assembled, resulting in bandage v. 0.8.1 (Wick et al., 2015). Annotation of plastid genome was performed through the online program GeSeq 1 (Tillich et al., 2017)

Sequence Alignment and Dataset Feature Evaluation
To reconstruct the phylogeny of Parnassia and its phylogenetic position among the Superrosids, we generated seven different datasets (datasets 1-7). We used 71 shared CDSs based on 88 annotated plastomes (83 plastomes from NCBI and five newly generated) to reassess the reliability of the morphological-based phylogenetic position of Parnassia (see details in Supplementary  Table 3). Similarly, we used 73 shared CDSs data based on 80 annotated plastomes (six downloaded ones and the rest are newly generated) to reconstruct the infrageneric systematic of Parnassia (see details in Supplementary Table 3). To extract the shared CDSs, we used the program PhyloSuite v1.2.2 [see details in Zhang D. et al. (2020)]. We generated three different pairs of chloroplast datasets: the concatenated CDSs (datasets 1 and 4), utilizing only the first and second codon sites of the concatenated CDSs (datasets 2 and 5), and the third codon site of the concatenated CDSs (datasets 3 and 6). Additionally, we generated the nrDNA (18S-ITS1-5.8S-ITS2-26S; 5,879 bp) to explore the phylogenetic relationships of Parnassia (dataset 7). The alignment of datasets was conducted using MAFFT V.7.409 with the codon matching strategy (Katoh and Standley, 2013). To check the suitability of all the datasets, we performed the sequence substitution saturation test through DAMBE7 (Xia, 2018).

Phylogenetic Analyses
We used the maximum likelihood (ML) and Bayesian inference (BI) statistics to reconstruct the Parnassia phylogenetic tree. Before tree reconstructions, we specified the GTR + G + I model inferred by jModelTest 2.1.6 (Darriba et al., 2012) Table 4). For the ML analysis, we used the program RAxML 8.2.12 (Stamatakis, 2014), utilizing the option of 1,000 rapid bootstrap replicates. Similarly, BI analyses were carried out by MrBayes v 3.2.7a (Ronquist et al., 2012) implemented in the CIPRES Science Gateway V 3.3 (Miller et al., 2010). Each BI analysis was conducted with two independent runs and four Monte Carlo Markov chains (MCMCs) of ten million generations, and trees were sampled every 1,000 generations. The first 25% of the sampled trees were discarded as burn-in, and the remaining were used to generate the consensus tree and calculate the Bayesian posterior probabilities (PP).

Molecular Dating Analyses and Ancestral Area Reconstruction
We used the concatenated CDSs of chloroplast (dataset 4) and estimated the divergence time of Parnassia utilizing the Bayesian statistics as implemented in BEAST v1.10.4 (Drummond et al., 2012). The parameters were set as GTR + I + Γ model and uncorrelated relaxed lognormal clock. The birth-death speciation model of the tree prior was selected according to the value of marginal likelihood, which was estimated by path sampling/stepping-stone sampling. The chain length of MCMC generations and the sampling frequency were set to 200,000,000 and 20,000, respectively. Generally, there may be many factors for the error in molecular clock analysis (e.g., taxon and gene sampling numbers, fossil calibrations strategy), which lead to bias in divergence time estimates (Sauquet et al., 2012;Bacon et al., 2016;Foster et al., 2017;Barba-Montoya et al., 2018). Due to the lack of reliable fossil evidence for the small perennial herbaceous plants of Parnassia (Coope et al., 1961), we selected four fossil records of its close relative taxa within Celastraceae as calibrations point to get a more reliable estimated divergence time (Figure 3). The strategies of divergence time estimation were as follows: Firstly, we selected the four most reliable evaluated fossils (Bacon et al., 2016) of relative taxa as fossil calibrations to avoid the generation of misleading results. Secondly, we employed a more informative dataset to reduce bias caused by a single gene (Zhai et al., 2019). We used four different times constraints based on the fossil records: (1) exponential priors (offset = 50.5, mean = 0.5) were set for the stem node of the monotypic Catha (following Poole and Wilkinson, 1999); (2) normal prior (mean = 26, St. dev. = 1) was set for the crown age of Celastrus + Tripterygium clade (following Xi et al., 2012;Bacon et al., 2016;Zhu et al., 2020); (3) exponential priors (offset = 56.5, mean = 0.833) were set for the stem node of Glyptopetalum + Euonymus (following Pitman and Rowan, 2012); and (4) secondary age was estimated from Xi et al. (2012) and Bacon et al. (2016), and a normal prior (mean = 90.6, St. dev. = 8.1) was set for the crown clade of Celastrales. We used BEAUti to generate the XML input files for BEAST, and the runs were performed on the CIPRES website (Miller et al., 2010). A total of four independent runs were followed in parallel, and the log files were combined using LogCombiner v1.10.4 (Drummond et al., 2012). We used Tracer v1.7 (Rambaut et al., 2018) to check the convergence of effective sample size (ESS). Lastly, the maximum clade credibility (MCC) tree with median heights was generated using TreeAnnotator v1.10.4 with the initial 10% trees discarded as burn-in. The final tree was visualized and edited using FigTree v.1.4.3 3 .
To assess the possible ancestral distribution regions of Parnassia, we performed an ancestral region reconstruction analysis in RASP v4.2 . The distribution regions of Parnassia were defined as belonging to the Pan-Himalayan and/or non-Pan-Himalayan. According to the distribution of species, we set the parameter "Max area" to two. The biogeographic models DEC, DIVALIKE, and BAYAREALIKE with/without J-parameter were tested by the R package "BioGeoBEARS" (Matzke, 2014), and the best-fit model was selected based on the Akaike Information Criterion (AIC) values.

Plastomes and Nuclear Ribosomal DNA Assemblies
We obtained a total of 778.1 Gb sequencing data for all the 74 individuals with an average of 10.51 Gb ranging from 4.7 Gb (Ribes glaciale Wallich) to 15.86 Gb (Parnassia nubicola Wallich ex Royle) (Supplementary Table 1 (4). Numbers associated with branches represent the mean estimated divergence time (Mya) and the blue bars correspond to the 95% highest posterior density (HPD) of divergence time. Red stars indicate four calibrating points. Pl, Pleistocene; Plio., Pliocene. (A-E) represent radical leaf, staminode, sepal, petal, and anther connective type, respectively. It should be noted that the characters shown in the figure only represent the characteristic types rather than the actual performance of the species. For a description of the specific feature types, please see Supplementary  Table 5. of a typical quadripartite structure, similar to the previously published genomes of Celastraceae (Supplementary Figure 1). The total number of annotated genes varied between 113 and 115, including 79-81 protein-coding genes, four rRNA genes, and 30 tRNA genes, respectively. Overall, the GC content in these species ranged from 38.10% in S. sinomontana to 36.90% in P. arborea (Supplementary Table 1).

Phylogenetic Position of Parnassia
The results about the suitability of the datasets used to investigate the phylogenetic position of Parnassia showed that the I ss values were significantly smaller than the I ss−c values (Supplementary Table 4). This suggested that these datasets had not reached substitution saturation and could be used for phylogenetic analyses (Xia et al., 2003). Though the phylogenetic trees were reconstructed with different datasets and different methods, all the results showed highly congruent topologies. Similarly, the relationships among different taxa were highly supported, suggesting that large-scale genomic data indeed significantly improved the resolution of phylogenetic trees (details see in Figure 1 and Supplementary Figures 2-5).

Phylogenetic Relationship of Parnassia
All the plastome datasets (datasets 4, 5, and 6) generated consistent trees with identical topologies of Parnassia, except for some uncertain clades (Figure 2 and Supplementary Figures 6-9). Phylogenetic trees showed that Parnassia was sister to the rest of Celastraceae with high support (BS = 100, PP = 1). In addition, all individuals of Parnassia were clustered together in all cases, indicating a well-supported monophyly of the genus. Parnassia sect. Nectarotrilobos Drude, the largest section in this genus, shows paraphyly, where some species of sect. The topology of BI and ML analysis, reconstructed with the nrDNA sequence data, showed the same results. However, it was incongruent with the phylogeny retrieved by plastomes datasets (Supplementary Figures 6-9). In the nrDNA phylogenetic tree, the position of Clade I and Clade II was reversed (Figure 2). Three species of sect. Nectarotrilobos (P. chinensis (Parnassia venusta Jien, Parnassia submysorensis J. T. Pan) formed a highly supported clade (BS = 100, PP = 1). Similarly, P. longshengensis, P. gansuensis, and Parnassia wightiana Wallich ex Wight & Arnott were also recovered as a clade with enough support (BS = 94, PP = 1). Besides, concordant results of both the plastome-based and the nrDNA phylogenetic trees were shown in some branches (Figure 2), i.e., individuals of Clade I, II, and III were strongly recovered with the same topology in the nrDNA tree (except for P. lutea of Clade I).
According to the AICc_wt value, the best model was BAYAREALIKE + J. The common ancestor of Parnassia probably originated in the Pan-Himalayan region (Supplementary Figure 10). The initial divergence of Parnassia took place at the Eocene and emerged as Clade A and Clade B. The diversification of Clade A was estimated to be in the non-Pan-Himalayan region, which was different from Clade B (Supplementary  Figure 10).  2020)]. Here, we used the plastome and nrDNA datasets to deeply reassess the phylogenetic position of Parnassia within Superrosids and provide a resolved phylogenetic tree of its species. Our results highly supported the monophyly of Parnassia; however, the infrageneric systematics of the genus showed incongruencies.

Phylogenetic Position of Parnassia in Superrosids
We reconstructed a well-resolved phylogenetic tree for 17 orders of Superrosids. The COM clade topology of all the phylogenetic trees revealed that Celastrales was sister to Malpighiales with high-support, and then this clade is sister to Oxalidales. The COM clade topology (Figure 1) obtained here was consistent with Zhang and Simmons (2006); Soltis et al. (2007), and Li et al. (2019), nevertheless, incongruent with many other previous studies (Qiu et al., 2010;Soltis et al., 2011;Ruhfel et al., 2014;Chen et al., 2016;Group et al., 2016;Sun et al., 2016). Although the sequence data used for phylogenetic analysis of COM clade has significantly increased in recent years, their systematics are still problematic or at least did not obtain well-supported topologies [details see in Li et al. (2019) and here in this study].
The phylogenetic trees based on different datasets to investigate the phylogenetic position of Saxifragales and Vitales were incongruent (Figure 1 and Supplementary Figures 2-5). The phylogenetic trees of dataset 3 recovered a similar topology as proposed by Li et al. (2019; Supplementary  Figures 4, 5). However, the trees of datasets 1 and 2 showed reverse topologies of the phylogenetic position between Saxifragales and Vitales (Supplementary Figures 1-3). All the phylogenetic trees revealed that Parnassia was closely related to Celastrales (Celastraceae: Figure 1 and Supplementary  Figures 2-5). Therefore, our results rejected the hypothesis based on morphology, which considered Parnassia was close to Saxifragaceae or other families (e.g., Hooker and Thomson, 1858;Hallier, 1901;Pace, 1912;Arber, 1913;Engler, 1930;Thorne, 1968;Spongberg, 1972;Bensel and Palser, 1975;Dahlgren, 1980;Cronquist and Takhtadzh " ian, 1981;Ku, 1987;Gu and Hultgård, 2001). Previous studies have questioned the relationship between Parnassia and Saxifragaceae. Grund and Jensen's (1981) comparative investigations of serological characteristics of seed proteins revealed that there was no relationship between Parnassia and Saxifragaceae. In addition, Matthews and Endress (2005) compared the floral structures of Celastrales, expounded 15 shared floral features between Parnassia and Celastraceae, and described five different floral structural characteristics that distinguished Parnassia from Saxifragaceae. Soltis et al. (1990) found that the relationship between Parnassia and Brexia was closer than Saxifragaceae by using rbcL. It was also confirmed in other later molecular phylogenetic researches (Chase et al., 1993;Morgan and Soltis, 1993;Soltis et al., , 2000Koontz and Soltis, 1999). Similarly, our results corroborated that Parnassia, belonging to Celastrales, was closely related to Celastraceae.
To investigate the phylogenetic position of Parnassia, we sampled species of Parnassia and 15 genera of Celastrales distributed in China and reconstructed the phylogenetic relationship. Two families were included in Celastrales: Celastraceae and Lepidobotryaceae. In previous investigations, Parnassia and Celastraceae were resolved as one clade according to the rbcL (Savolainen et al., 1997), and Lepidobotryaceae was sister to this clade (also see Zhang and Simmons, 2006). The floral morphological study revealed that Lepidobotryaceae was more distant from Parnassia and Celastraceae, and these three taxa were also supported as a group with some shared floral features (Matthews and Endress, 2005). Nevertheless, the relationship between Parnassia and Celastraceae was still controversial (details see introduction). In this study, plastome phylogenetic suggested that Parnassia was sister to Celastraceae with high support, which was consistent with the results of Simmons et al. (2012), Bacon et al. (2016), Chen et al. (2016), and Li et al. (2019). However, the nrDNA data clustered Parnassia and Microtropis together and then resolved them as sister groups to the other genera of Celastraceae (also see Simmons et al., 2012 andSun et al., 2016). Discordance between the plastid genome and nrDNA analyses is a common phenomenon in molecular phylogenetic studies (see Lee-Yaw et al., 2019;Chen et al., 2020;Liu et al., 2020;Ma et al., 2021;Uckele et al., 2021), which was usually suggested to be a result of horizontal gene transfer (Stegemann et al., 2012), hybridization (Rieseberg et al., 1991(Rieseberg et al., , 1996 and incomplete lineage sorting (Degnan and Rosenberg, 2009).
In addition to the doubtful phylogenetic relationship between Parnassia and Microtropis shown in this study, the relationship among Parnassia, Perrottetia, and Mortonia is also unclear. Previous studies revealed that (Parnassia, Lepuropetalon) + (Perrottetia, Mortonia) were sister to other genera of Celastraceae, suggesting that Parnassia should be considered as an early derived lineage of Celastraceae (Simmons et al., 2001;Sun et al., 2016). Meanwhile, several studies demonstrated that Parnassia was sister to all the remaining genera of Celastraceae (Simmons et al., 2012;Bacon et al., 2016;Chen et al., 2016;Li et al., 2019). Here, it was suggested that the topology of trees based on plastome have higher resolution in comparison with the nrDNA. With the combination of the previous studies and our results of the plastome tree, we speculated that Parnassia was a basalmost taxon of Celastraceae. Nevertheless, it is still not mature enough to make clearer hypotheses before resolving the phylogenetic relationships among Parnassia, Mortonia, Perrottetia, and Microtropis.

Biogeography and Divergence Estimation of Chinese Parnassia
The divergence time in this study is congruent with the previous geographical study of Parnassia (Wu, 2005). However, due to the limited fossil calibrations points (lack of reliable fossil records of Parnassia), the results of divergence estimation in this study should be treated cautiously. Wu (2005) discussed the geographical distribution of Parnassia and supposed that its origin was in the early tertiary periods or even earlier. Bell et al. (2010) estimated the divergence times of key angiosperm lineages and showed that the crown-group age of Parnassia was estimated as 29 (17-43) Ma (exponential priors) and 34 (19-48) Ma (Lognormal priors). Here, the diversification of Parnassia was estimated to be 39.97 Mya (95% HPD = 28.13-52.93 Mya, Figure 3), which was earlier than the result of Bell et al. (2010;34 Mya, 95% HPD = 19-48 Mya). The limited taxon coverage may reduce the accuracy of divergence time estimation [only one individual representing Parnassia in Bell et al. (2010) and Gao et al. (2017)], which may partially account for the slight inconsistency.
Most species in this study were distributed in the southeastern Qinghai-Tibetan Plateau and its surrounding mountainous regions. The ancestral area inference results indicated that the ancestral lineage of Parnassia probably originated in the Pan-Himalayan region. The north of Tibet localized uplift to the near present heights during the Eocene, while northwestern Yunnan formed its present topography during the late Eocene/early Oligocene (Spicer et al., 2021). Our results of divergence time estimation indicated that the genus Parnassia began to diverge and spread to non-Pan-Himalayan areas during the Eocene. Combining with the Qinghai-Tibetan Plateau uplift, the global climate oscillations led to a significant change in plant diversity during the Eocene-Oligocene transition period (Abels et al., 2011;Deng et al., 2020). As a result, the differentiation Parnassia species began to appear at Oligocene in the Pan-Himalayan region, and then mountain building and Asian monsoon may have combined to promote species diversification and colonization in the early to middle Miocene (Xing and Ree, 2017;Ding et al., 2020). Finally, the final uplift and climate oscillation of the Qinghai-Tibet Plateau in the late Miocene or Pliocene has jointly driven the rapid diversification of Parnassia species (Chen et al., 2012;Zhou et al., 2013). This may partially explain the high species richness of Parnassia species in the Pan-Himalayan region. Similar to previous studies in this space-time (Sun et al., 2017;Lu et al., 2018), it is probable that the large-scale expansion occurred after species diversification of Parnassia and accelerated the colonization of adjacent areas.

Infrageneric Relationships of Parnassia
We mainly focused on resolving the phylogenetic relationship of Parnassia species and advancing some supportive hypotheses about its systematics. Our results recovered a well-supported monophyletic clade of Parnassia by using both the plastome and the nrDNA datasets (Figure 2). Notably, most of the plastomebased topology branches reached a 100% support rate, which significantly improved the phylogenetic resolution (Figure 2).
The molecular systematics of Parnassia have focused on the phylogenetic position problems, and there were few studies on the phylogenetic relationship within this genus (Wu, 2005;Yang et al., 2012;Yu, 2019). In this study, more comprehensive plastome and nrDNA data were used to elaborate the phylogeny of Parnassia for the first time. Consistent with previous studies, Parnassia has been fully supported as a monophyletic group. Clade I (P. lutea, P. tibetana, P. nubicola, P. oreophila, P. trinervis, P. filchneri, and P. viridiflora) consisted of seven species of sect. Nectarotrilobos, forming a monophyletic branch (details see in Wu, 2005). Parnassia lutea, without the cauline leaves, was considered to represent the early colonizer of sect. Nectarotrilobos to high-altitude areas (Wu, 2003(Wu, , 2005. Although P. lutea was located at the base of Clade I, reflecting it was an earlier element of Clade I, our results did not support its more ancient hypothesis within the genus. Ma et al. (2020) proposed that P. tibetana was synonymous to P. nubicola. Likewise, P. tibetana has been resolved as nested within the individuals of P. nubicola, and we did not find any significant molecular differences between P. tibetana and P. nubicola (Figure 2). Compared to our results of P. oreophila, Yu (2019) also demonstrated that P. oreophila, P. nubicola, P. laxmannii, P. trinervis, and P. cabulica belonged to the same clade. As inferred from the distribution range and divergence time, P. trinervis, P. filchneri, and P. viridiflora may be the species differentiated from P. oreophila in the process of colonizing to high altitude during Miocene. Hence, we herein supported the taxonomic revisions of P. tibetana and P. nubicola in Ma et al. (2020). Individuals of P. trinervis, P. filchneri, and P. viridiflora were nested within each other (Figure 2). During field investigation, we found that the distribution range of P. trinervis overlapped with P. viridiflora. Furthermore, the petal color of these two species changed from dark green to light green and white, and even different colors might occur in the same population (petal color was the key character to distinguish these two species). Therefore, we supported the recommendation that P. viridiflora should be treated as the synonym of P. trinervis by Yang et al. (2012) [also in ]. For P. filchneri, a follow-up study is recommended to assess the relationship with P. trinervis. Previous studies suggested that P. palustris appeared earlier than other species of Parnassia and should be a relict species (Wu, 2003(Wu, , 2005. Divergence time estimation supported this speculation (Figure 3), and we suggested that P. palustris was the ancestor species of Pan-Himalayan region.
Clade II consisted of three species of sect. Nectarotrilobos (P. pusilla, P. mysorensis, and P. deqenensis). For P. deqenensis, Wu (2005) suggested reducing "it" as the synonym of P. trinervis, while  proposed reducing "it" as the synonym of P. pusilla. Our results here showed that the three species in Clade II were grouped into one branch in all trees, while individuals were nested within each other in phylogenetic trees (Figure 2). Based on the above results, we agreed with the taxonomic revision of . Meanwhile, further studies with P. mysorensis and P. pusilla are suggested.
Clade III consisted of three species of sect. Nectarotrilobos subsect. Xiphosandra (P. brevistyla, P. delavayi, and P. leptophylla). Yu et al. (2018) found that the continuous characteristics were mistakenly used in distinguishing these three species. Hence, P. brevistyla and P. leptophylla were designated as synonyms of P. delavayi. In this study, the results of phylogenetic relationships showed that individuals of these three species were nested within each other (Figure 2), supporting the taxonomic revision of Yu et al. (2018).
There were widespread inconsistencies in the topological structure of the plastid trees and the nrDNA trees (Figure 2), such as the systemic position of P. noemiae, P. lutea, P. foliosa, P. chinensis, P. venusta, P. submysorensis, P. longshengensis, P. gansuensis, and P. wightiana were changed. The individuals of P. wightiana were resolved as polyphyletic in plastid trees (Figure 2), which could be attributed to the previously reported hybridization of this species (Yu, 2019). Furthermore, P. longipetala was not monophyletic as well in plastid trees, but clustered together in nrDNA trees, suggesting that incomplete lineage sorting, hybridization, or gene introgression events might have occurred with its related sympatric species. After comparing with many previous studies about the genus (e.g., Chen et al., 2020;Liu et al., 2020;Ma et al., 2021;Uckele et al., 2021), we speculated that widespread hybridization or introgression events might occur in Parnassia species distributed in and around the Hengduan Mountains (e.g., P. venusta, P. longshengensis, P. wightiana, P. submysorensis, and P. chinensis). Wu (2005) and Yu (2019) explored the phylogeny of Parnassia based on morphology. However, the results showed that almost all characters (such as the shape of petal margin and the number of staminodes lobs) that have been used for taxonomy and systematic might have undergone parallel or even reverse evolution [details in Shu (2017)]. Our phylogenetic results showed that except for the characters of basal leaves and anther connective, other characteristics may not represent the evolutionary relationship of Parnassia (Figure 3). The sect. Nectarotrilobos, sect. Saxifragastrum, sect. Fimbripetalum, and sect. Nectaroquinquelobos were resolved as polyphyletic. Shu (2017) proposed reducing P. venusta and Parnassia degeensis as synonyms of Parnassia cacuminum. However, in this study, individuals of P. venusta and P. degeensis were not clustered together, which was inconsistent with traditional taxonomy revisions (Shu, 2017). The molecular phylogenetic analyses showed that a single species with continuous quantitative traits might be divided into several different species in the morphological study of Parnassia, e.g., P. brevistyla, P. delavayi, and P. leptophylla .
In addition, different species with similar characters may be merged into one species, e.g., P. venusta, P. degeensis, and P. cacuminum (see details in Figure 3; . Characters (e.g., staminodes 3-lobed) of Parnassia might evolve at multiple times, and introgression or hybridization events might have happened in the process of evolution. This may explain why the trend of morphological character evolution of Parnassia has always been controversial in the traditional taxonomy.

CONCLUSION
Here we utilized the whole plastome and nrDNA (5,879 bp) markers to reconstruct the position of Parnassia within the Superrosids and phylogenetic relationships of Chinese Parnassia species. Our results corroborated that Parnassia is most closely related to Celastraceae, while follow-up studies are still needed to determine the relationship between Parnassia and several related genera. However, there was incongruence between the plastome and nuclear phylogenetic trees. In addition, we presented a first attempt to use the fossil records of Celastraceae to estimate the divergence of Chinese Parnassia, which was strongly consistent with the expected results of previous biogeographic studies. Moreover, the molecular phylogenetic results confirmed numerous taxonomic revisions and suggested that many traits have been widely used for classification, and that systematic considerations were the results of multiple evolutions. Our results will provide valuable insights into the taxonomic study of Parnassia and expand the foundation for further exploration of the evolutionary diversity of the genus.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/Supplementary Material.

AUTHOR CONTRIBUTIONS
F-QZ: conceptualization, funding acquisition, and resources. S-LC: funding acquisition, resources, and supervision. M-ZX: formal analysis, data curation, and writing -original draft. YL: resources, formal analysis, and writing -review and editing. J-YY, X-FC, and HX: resources and formal analysis. GK: formal analysis and writing -review and editing. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS
We thank Pengcheng Fu (Luoyang Normal University), Qingbo Gao (Chinese Academy of Sciences), and Liukun Jia (Chinese Academy of Sciences) for their assistance in material collection. We also thank Xiaolei Ma (Chinese Academy of Sciences) for his help with the specimen image.

SUPPLEMENTARY MATERIAL
The