Nuclear phylogenomics of Asteraceae with increased sampling provides new insights into convergent morphological and molecular evolution

Abstract

Convergent morphological evolution is widespread in flowering plants, and understanding this phenomenon relies on well-resolved phylogenies. Nuclear phylogenetic reconstruction using transcriptome datasets has been successful in various angiosperm groups, but it is limited to taxa with available fresh materials. Asteraceae, which are one of the two largest angiosperm families and are important for both ecosystems and human livelihood, show multiple examples of convergent evolution. Nuclear Asteraceae phylogenies have resolved relationships among most subfamilies and many tribes, but many phylogenetic and evolutionary questions regarding subtribes and genera remain, owing to limited sampling. Here, we increased the sampling for Asteraceae phylogenetic reconstruction using transcriptomes and genome-skimming datasets and produced nuclear phylogenetic trees with 706 species representing two-thirds of recognized subtribes. Ancestral character reconstruction supports multiple convergent evolutionary events in Asteraceae, with gains and losses of bilateral floral symmetry correlated with diversification of some subfamilies and smaller groups, respectively. Presence of the calyx-related pappus may have been especially important for the success of some subtribes and genera. Molecular evolutionary analyses support the likely contribution of duplications of MADS-box and TCP floral regulatory genes to innovations in floral morphology, including capitulum inflorescences and bilaterally symmetric flowers, potentially promoting the diversification of Asteraceae. Subsequent divergences and reductions in CYC2 gene expression are related to the gain and loss of zygomorphic flowers. This phylogenomic work with greater taxon sampling through inclusion of genome-skimming datasets reveals the feasibility of expanded evolutionary analyses using DNA samples for understanding convergent evolution.

Asteraceae are one of the two largest angiosperm families, with ∼28 000 species having bilaterally and/or radially symmetric flowers that form head inflorescences. This study reports a new phylogenomic strategy integrating low-coverage genome skimming and transcriptomic datasets from 706 species to provide a robust and expanded Asteraceae phylogeny. Evolutionary analyses reveal multiple origins of bilateral symmetry associated with distinct expression patterns of CYC2 paralogs.

Introduction

Convergent evolution (“parallel evolution”) is widespread, involving the same (or similar) phenotype(s) that originate in separate (distant) lineages from other phenotypes (Arendt and Reznick, 2008; Losos, 2011; Bergmann and Morinaga, 2019). Angiosperms are the largest and most ecologically and economically important land plant group (Christenhusz and Byng, 2016; Lughadha et al., 2016). They have experienced convergence of diverse phenotypes, including habit, morphological characters, and physiological processes, and such changes have probably promoted diversity (Donoghue, 2005; Sun et al., 2012; Sousa-Baena et al., 2018; Thorogood et al., 2018; Heyduk et al., 2019; Zhang et al., 2023). The flower is one of the most important characteristics of angiosperms, and convergence of floral characters is widespread and likely important for angiosperm evolution (Specht and Bartlett, 2009; Soltis and Soltis, 2014; Reyes et al., 2016; Brock et al., 2018). For example, bilateral floral symmetry (floral zygomorphy) originated at least 199 times and has been hypothesized to be a key morphological innovation contributing to the species richness of many angiosperm groups (Sargent, 2004; Reyes et al., 2016; Zhong et al., 2017). In addition, functional studies have uncovered key regulators of floral characters, including MADS-box genes for floral organ identity and TCP genes that control floral zygomorphy in distant families (Luo et al., 1999; Martín-Trillo and Cubas, 2010; Bowman et al., 2012; Chen et al., 2017; Irish, 2017).

Previous convergence studies of floral characters have generally involved some representative species from different orders or families of angiosperms (Knudsen and Tollston, 1995; Preston and Hileman, 2009; Endress, 2016; Reyes et al., 2016) or focused on a few closely related species (Smith and Rausher, 2011; Larter et al., 2018). In addition, analyses of floral characters have rarely been combined with analyses of the molecular evolution of relevant regulatory genes. Comprehensive morphological and molecular evolutionary studies of floral characters in large families that exhibit a high degree of floral diversity can improve our understanding of the pattern and molecular basis of these characters.

The sunflower family (Asteraceae), with more than 26 000 species corresponding to 7% of flowering plant diversity, is one of the two largest families of flowering plants (Bremer, 1994; Funk et al., 2005; Anderberg et al., 2007). It is important both economically and ecologically, containing well-known oilseed crops (sunflower), vegetables (lettuce, chicory, and artichoke), and ornamentals (Simpson, 2009). Asteraceae also exhibit highly diverse floral features (Jeffrey, 2009). The typical Asteraceae inflorescence is a capitulum (head) with radial symmetry, whereas individual flowers in the capitulum (florets) can differ in symmetry; specifically, disc and filiform florets have radial symmetry, and ray, ligulate, and bilateral florets display zygomorphy. In some Asteraceae tribes, multiple capitula form various types of compound inflorescences called capitulescences, analogous to compound inflorescences in other families. In addition, flowers of most Asteraceae species retain a persistent calyx called the pappus, which is also highly diverse, including capillary, plumose, scaly, and spiny types. These floral characters are thought to play important roles in Asteraceae diversification (Leppik, 1970; Sheldon and Burrows, 1973; Van der Pijl, 1982; Mani and Saravanan, 1999; Anderberg et al., 2007) and have originated multiple times during Asteraceae evolution (Panero et al., 2014; Zhang et al., 2021). However, the convergence pattern of these characters has not yet been comprehensively studied throughout the family, mainly because of the low resolution or sampling limitations of previous Asteraceae phylogenies.

Asteraceae have been classified into 16 subfamilies and 45 tribes on the basis of plastid and nuclear phylogenies (Panero and Funk, 2002, 2008; Anderberg et al., 2007; Panero et al., 2014; Panero and Crozier, 2016; Susanna et al., 2020). Recent phylogenomic studies with 253 or 244 Asteraceae species (Mandel et al., 2019; Zhang et al., 2021) resolved most relationships among subfamilies and tribes that have long been controversial, e.g., the recognition of Vernonioideae and Doroniceae. However, these two studies sampled fewer than 100 subtribes from a total of ∼190 subtribes. Many subtribes and genera whose diverse floral characters may have experienced convergence and are important for phylogeny and morphological evolution (Supplemental Figures 1 and 2) were not included. Hence, insufficient sampling in these ranks may have affected the accuracy of evolutionary analyses of floral characters. Greater Asteraceae taxon sampling for phylogenomic studies is needed to include a larger proportion of subtribes and more genera and thus improve the phylogeny and the assessment of convergent evolution in floral characters.

Recent phylogenomic analyses have shown the effectiveness of using nuclear genes from transcriptomes in phylogenetic research on large angiosperm groups (Wickett et al., 2014; Xiang et al., 2017; Shen et al., 2019; Shi et al., 2020; Zhang et al., 2020, 2023; Zhao et al., 2021). However, transcriptome sequencing requires fresh materials and is relatively costly. On the other hand, millions of herbarium specimens (Zeng et al., 2018) and other dried materials used in plastid phylogenetic analyses (e.g., Li et al., 2019, 2021) have not been sufficiently utilized for nuclear phylogenies. Thus, the ability to increase sample numbers by including these materials should dramatically benefit nuclear phylogenetic studies. Low-coverage whole-genome sequencing called genome skimming (GS, usually with ∼1× or slightly higher coverage) has been widely used to obtain sequences of plastid and mitochondrial genomes and other high-copy DNA (Straub et al., 2012; Malé et al., 2014; Dodsworth, 2015). Higher-coverage (generally >10×) GS was used to retrieve nuclear genes from Vitaceae (Liu et al., 2021b) or from a few species to supplement transcriptome datasets in nuclear phylogenetic reconstruction (Zhang et al., 2020, 2021; Zhao et al., 2021), demonstrating that relatively high-coverage GS datasets can yield nuclear genes for phylogenetic studies (Guo et al., 2023). Two recent studies with 41 and 24 species, respectively (Pouchon et al., 2018; Vargas et al., 2019), suggested that shallow GS datasets usually contained ∼80% of reads from the nuclear genome. However, it is not known whether shallow GS can provide sufficient nuclear markers for large-scale phylogenetic studies of species-rich groups, nor is it known which ratio of GS to transcriptome datasets is effective for obtaining well-resolved phylogenies.

The flower homeotic MADS-box genes and TCP genes in the CYCLOIDEA (CYC) and CINCINNATA (CIN) clades encode transcription factors that regulate plant development and floral symmetry (Luo et al., 1996, 1999; Nath et al., 2003; Crawford et al., 2004; Ma, 2005; Dreni and Zhang, 2016; Chen et al., 2017; Callens et al., 2018). Functional studies of Asteraceae homologs of the ABCE MADS-box genes (Kotilainen et al., 2000; Shulga et al., 2008; Broholm et al., 2010; Ruokolainen et al., 2010a, 2010b; Goloveshkina et al., 2012; Zhang et al., 2017; Shchennikova et al., 2018; Zhao et al., 2020) and the CYC and CIN genes (Fambrini et al., 2011; Tähtiharju et al., 2012; Juntheikki-Palovaara et al., 2014; Huang et al., 2016c; Elomaa et al., 2018; Yu, 2020; Zhao et al., 2020) have revealed the functions of some genes in control of floral organ identities and symmetries. In addition, Asteraceae homologs of ABCE and TCP genes have expanded via duplications, as revealed by studies focusing on one or a few species or on a specific gene class (Yu et al., 1999; Broholm et al., 2010; Ruokolainen et al., 2010b; Zhang et al., 2017; Shchennikova et al., 2018; Zhao et al., 2020). However, possible connections between the evolution of regulatory genes and the convergent evolution of floral characters in Asteraceae have not been comprehensively investigated in previous analyses.

Here, we integrated GS of varying volumes with transcriptome datasets to perform nuclear phylogenomic analyses of Asteraceae and assess the phylogenetic utility of such GS datasets. Nuclear phylogenetic reconstruction of 706 Asteraceae species (16 subfamilies, 41 tribes, and 144 subtribe-level groups) provided improved phylogenetic relationships for many subtribes and genera. Molecular clock estimates of divergence times suggested possible environmental factors that have contributed to Asteraceae evolution. The newly established Asteraceae phylogenetic trees enabled morphological character analyses for convergence of several floral traits at different levels. Molecular evolution analyses of MADS-box and TCP gene families were performed to detect duplication in Asteraceae and to test whether duplications or other changes contributed to the convergence of related floral characters.

Results and discussion

Asteraceae phylogeny and divergence times based on transcriptomes and GS datasets

To assess the phylogenetic effectiveness of a new approach that integrated transcriptomes and GS datasets, we performed phylogenetic analyses with assessments using both simulated and empirical Asteraceae datasets. The simulation analysis showed that transcriptome-like datasets for 30%–50% of taxa combined with GS-like datasets for the remaining taxa resulted in highly supported phylogenetic relationships (Supplemental Figure 3; see supplemental information for details). Using empirical Asteraceae transcriptomes and GS datasets (Supplemental Tables 1–3, Figure 1, and Supplemental Figure 4; see supplemental information for details), we obtained a 1094-gene set in which 50% of genes were found in 61% of the GS datasets ≥2 Gbp and 90% of the genes were recovered from transcriptomes. Phylogenetic analysis using a 686-gene subset after removal of paralogs and genes with relatively high missing rates generated a coalescent tree of Asteraceae that was supported by similarly high bootstrap support (BS) values (average ≥92 and median 100) in clades with 0%–90% of GS datasets (see supplemental information), suggesting that up to 90% GS datasets in combination with transcriptomes can generally provide sufficient phylogenetic resolution (Figure 1E). Clades with 450–550 sampled genes (66%–80% of the 686-gene set) generally had high BS values (mean 91 and median 100), and GS datasets that had at least one-third of the 686 genes yielded well-supported phylogenetic placements (Figure 1F). Other factors, such as mismapping, base-calling error of GS reads, and number of removed paralogs, had small effects on the phylogenetic resolution (Supplemental Figures 5–9; see supplemental information). These multiple analyses support the idea that GS datasets with sufficient sequencing volume (>2.1 Gbp for Asteraceae) and numbers of sampled genes (at least >36%; stronger with >65%) can be included for generally effective phylogenetic reconstruction, enabling the sampling of more taxa representing a greater number of lineages.

Figure 1.

Assessments of shallow genome-skimming datasets from 526 Asteraceae samples.

(A) A histogram shows the distribution of sequencing volumes of GS datasets. The vertical lines indicate the average C value (converted to a genome size of 3.6 Gbp) of Asteraceae species and the average size (1.8 Gbp) of 31 sequenced Asteraceae genomes.

(B) A cumulative curve of dataset number shows the distribution of sequencing volumes of GS datasets. The blue dashed lines indicate that 50% and 80% of the datasets have sequencing volumes greater than 3.2 Gbp and 2.1 Gbp, respectively.

(C) A histogram and a cumulative curve show the distribution of the number of detected genes with length ≥300 bp in the GS datasets.

(D) A box plot shows the distribution of detected genes with length ≥300 bp in groups of GS datasets with different sequencing volumes.

(E) A box plot shows the distribution of bootstrap support (BS) values in clades with different proportions of GS datasets.

(F) A box plot shows the distribution of BS values in clades with different average numbers of genes.

The intervals used in the x axes of (D), (E), and (F) are left-open and right-closed. ns, non-significant (P > 0.05); ∗P ≤ 0.05; ∗∗P ≤ 0.01 (Mann–Whitney U test).

In the phylogenetic analyses of Asteraceae with 706 species, coalescent trees reconstructed using the six datasets were largely similar, with a few differences in clades without high support and among closely related species, suggesting that the different proportions of missing data at the subtribe level did not significantly affect the resolution and robustness of the phylogenetic trees (see supplemental information). The Asteraceae phylogeny here recovered 15 of 16 recently defined subfamilies (Susanna et al., 2020) and generally agreed with recent phylogenetic studies (Huang et al., 2016b; Mandel et al., 2019; Zhang et al., 2021) (Supplemental Figures 1 and 2), with the exception of Carduoideae, Tarchonanthoideae, Wunderlichioideae, Gochnatioideae, and Cyclolepis. These phylogenetic trees maximally supported the sister relationship of Tarchonanthoideae and Carduoideae (Figure 2 and Supplemental Figures 2, 9, and 10) and did not support the subfamily position of Tarchonanthoideae proposed by Mandel et al. (2019) and Susanna et al. (2020). In addition, these phylogenetic trees placed Wunderlichioideae and Gochnatioideae as separate clades in a grade with moderate to strong support. Previously, Cyclolepis was placed as sister to either Wunderlichioideae or Gochnatioideae with support values <70 (Mandel et al., 2019) or included in Gochnatioideae (Susanna et al., 2020). Here, the consistent phylogenetic placement of Cyclolepis as a separate lineage from Wunderlichioideae and Gochnatioideae, with weak to strong support, suggested that Cyclolepis might represent a new subfamily of Asteraceae (Figure 2 and Supplemental Figure 9).

Figure 2.

Summary phylogenetic tree of non-Asteroideae groups of Asteraceae.

A summary of coalescent trees from six gene sets with different numbers of genes (Supplemental Figure 9), showing relationships of non-Asteroideae subtribes and higher-rank groups. The symbols on the branches indicate the summary of the bootstrap support values (BS) and posterior probabilities (LPP) of the corresponding clades, as explained at the upper left corner. The dashed lines indicate lineages with conflicting positions among the coalescent trees. Tips with only one species are represented by solid circles. Recently reported classification information (Susanna et al., 2020) is shown on the right side of the tree: OUT, outgroups; ST, Stifftioideae; INC. SED., incertae sedis group; WU, Wunderlichioideae; TA, Tarchonanthoideae; DI, Dicomoideae.

The largest subfamily, Asteroideae, was shown to have two well-supported supertribes, i.e., Helianthodae and Senecionodae. The monophyly and phylogenetic relationships of most Asteroideae tribes were consistent with previous nuclear phylogenies (Mandel et al., 2019; Zhang et al., 2021), supporting the non-monophyly of Millerieae and Neurolaeneae revealed in Zhang et al. (2021) (Figures 2 and 3; Supplemental Figures 2 and 9). However, Tageteae were divided into two clades in our results, with the monogeneric subtribe Jaumeinae (designated as Tageteae clade I) separate from other Tageteae as a deeply divergent clade in the Heliantheae alliance (Figure 3 and Supplemental Figure 9). This result suggests that Jaumeinae might represent a new tribe of the Heliantheae alliance. In addition, our phylogenetic trees placed the Jaumeinae and subtribe Neurolaeninae of Neurolaeneae as the earliest divergent lineages, respectively, of two major subclades of the Heliantheae alliance, HA1 and HA2 (Figure 3). Five tribes within the Heliantheae alliance, Madieae, Chaenactideae, Bahieae, Eupatorieae, and Perityleae, were well resolved and formed a clade (the MCBEP clade). Also, our results placed the tribe Calenduleae as a sister group to all other tribes of Senecionodae except Doroniceae, unlike the previous placement of Calenduleae as sister to Astereae and Gnaphalieae (Mandel et al., 2019; Zhang et al., 2021).

Figure 3.

Summary phylogenetic tree of the subfamily Asteroideae of Asteraceae.

A summary of coalescent trees from six gene sets with different numbers of genes (Supplemental Figure 9), showing the relationships of subtribes and higher-rank groups of Asteroideae. The symbols on the branches indicate the summary of the bootstrap support values (BS) and posterior probabilities (LPP) of the corresponding clades. The dashed line indicates conflicting topologies among coalescent trees. Tips with only one species are indicated by solid circles. Recently reported classification information (Susanna et al., 2020) is shown at the right side of the tree: Athr, Athroismeae; Hele, Helenieae; Neur I, Neurolaeneae clade 1; Neur II, Neurolaeneae clade 2; Tage I, Tageteae clade 1; Tage II, Tageteae clade 2; Mill I, Millerieae clade 1; Mill II, Millerieae clade 2; Bahi, Bahieae; Ast., Asterodae; Senec., Senecionodae. Two supertribes recognized by us and several important clades mentioned in this study are indicated with labels on the tree: MCBEP clade, the clade composed of Madieae, Chaenactideae, Bahieae, Eupatorieae, and Perityleae; HA1, Heliantheae alliance clade 1; HA2, Heliantheae alliance clade 2.

Our results provided placements for 26 and 28 subtribes that were not sampled in Mandel et al. (2019) and Zhang et al. (2021), respectively, and suggested that 26 out of 77 subtribes with two or more sampled species were not monophyletic. We also sampled nine genera that were previously unplaced and resolved their subtribal positions. This phylogenetic study significantly improves our phylogenetic understanding of Asteraceae and facilitates evolutionary analyses of morphological characters, such as floral characters (more information on Asteraceae phylogeny is provided in the supplemental information). A coalescent tree of 306 Asteraceae species with high-coverage genomic and transcriptomic datasets was reconstructed using the 686-gene set (Supplemental Figure 11) and was compared with the tree containing additional species with GS datasets (706 Asteraceae species; Supplemental Figure 9). The two trees were highly consistent with each other in topology and support values for the shared taxa. A concordance factor analysis between gene trees and the coalescent tree from the 686-gene set (Supplemental Figure 12) suggested a high concordance between the gene trees and coalescent trees. A comparison of coalescent trees from seven new gene sets with different missing rates of genes and species (Supplemental Figures 13–19) indicated that missing data were not a major problem in our Asteraceae phylogeny from integrated datasets (see also supplemental information).

Our analyses using Asteraceae datasets including 43% of species with transcriptomes demonstrated the effectiveness of combining transcriptomes with GS datasets, which benefit from available dried samples at herbaria and other sources. In addition, GS allows access to a larger number of nuclear genes than Hyb-Seq methods that use probes for given sets of genes (e.g., the Angiosperms353 kit), resulting in generally well-resolved coalescent trees (see Guo et al., 2023). Furthermore, the GS data contain numerous reads from other nuclear genes and can be used for studies of multi-gene families and non-coding sequences, among others. This approach is therefore an efficient way to reconstruct large phylogenies for use in evolutionary analyses that include a greater number of lineages in species-rich groups. Our analyses also indicated that clades with relatively low proportions of GS datasets generally had higher resolution, whereas subtribes with only or high proportions of GS datasets, e.g., Artemisiinae, Asterinae, FLAG, and Tussilagininae, were less well resolved (Supplemental Table 1 and Supplemental Figure 9). Therefore, inclusion of sufficient transcriptome datasets in major clades is important for phylogenetic resolution when integrating transcriptome and GS datasets, providing guidance for the design of future studies using this approach.

Molecular clock estimation of divergence times (Figure 4; Supplemental Figures 20 and 21) showed that Asteraceae originated in the Late Cretaceous (∼74.84 million years ago [Ma]) and then rapidly diverged within a narrow window of ∼13 million years (72.99–59.49 Ma), producing all currently recognized subfamilies. To test the effects of different calibration scenarios on the time estimates for Asteraceae, we included six alternative calibration scenarios (Supplemental Table 4), as detailed in the supplemental information and Supplemental Figures 20 and 22–27. Our results showed that most tribes of non-Asteroideae subfamilies with more than one tribe diverged from 61 to 50 Ma, with the tribes Mutisieae, Onoserideae, Nassauvieae, and Cardueae having relatively old ages ranging from 61 to 57 Ma. However, Oldenburgieae and Tarchonantheae diverged relatively recently at ∼31.5 Ma in the Oligocene. In Asteroideae, most tribes diverged from ∼55 to ∼49 Ma, whereas the tribes of the Heliantheae alliance had younger stem ages of ∼46–39 Ma. Crown ages of tribes in the Heliantheae alliance and those of most other Asteroideae tribes were comparable (∼40–33 Ma), except for those of Senecioneae and Anthemideae (∼51.41 Ma and 44.24 Ma, respectively). However, the divergence times of subtribes within Senecioneae and Anthemideae (Supplemental Figures 20 and 21) were comparable to those of subtribes in other tribes of Senecionodae. The divergence times estimated here for the family, subfamilies, and tribes were intermediate compared with previously reported ages (Barreda et al., 2015; Fu et al., 2016; Huang et al., 2016b; Panero, 2016; Panero and Crozier, 2016; Mandel et al., 2019; Zhang et al., 2021) (see supplemental information). In the non-Asteroideae tribes, most subtribes diverged in the Middle to Late Eocene (51–34 Ma) (Supplemental Figures 20 and 21). In addition, divergence time patterns suggested that the gradual global cooling by ∼8°C from the Early Eocene to the Eocene–Oligocene boundary (50–34 Ma) (Zachos et al., 2001; Westerhold et al., 2020) and the subsequent expansion of open habitats may have contributed to diversification within most clades (such as tribes) of Asteraceae. Furthermore, such diversification seemed to be associated with shifts in the dominant insect groups that were related to pollination of Asteraceae.

Figure 4.

Estimated divergence times of Asteraceae.

(A) A simplified chronogram of Asteraceae showing the divergence times of subfamilies and tribes. The chronogram was reconstructed using the penalized likelihood method and calibration scenario 1 with 10 fossils and one secondary calibration node (Supplemental Table 4 and Supplemental Figure 20). The vertical blue and green bars on the chronogram indicate the period of rapid divergence among subfamilies and the times of crown ages of most Asteraceae tribes, respectively. The numbers on the branches indicate the divergence times of the corresponding nodes. The subfamilies and the Heliantheae alliance (Susanna et al., 2020) are indicated on the right side of the chronogram: Ba, Barnadesioideae; Fa, Famatinanthoideae; Wu, Wunderlichioideae; St, Stifftioideae; Mu, Mutisioideae; Unp., incertae sedis group; Go, Gochnatioideae; He, Hecastoclidoideae; Pe, Pertyoideae; Ta, Tarchonanthoideae; Ca, Carduoideae; Di, Dicomoideae; Gy, Gymnarrhenoideae; Ve, Vernonioideae; Ci, Cichorioideae; Co, Corymbioideae. The photographs on the left side show representatives of large Asteraceae subfamilies: from top to bottom, these are Aster mongolicus (Asteroideae, supertribe Senecionodae), Helianthus annuus (Asteroideae supertribe Helianthodae), Rafinesquia neomexicana (Cichorioideae), Baccharoides lasiopus (Vernonioideae), and Echinopsis sp. (Carduoideae).

(B) The time bar of the chronogram and global temperature changes from 66 Ma to the present, showing the global cooling and habitat shift related to the diversification of subfamilies and tribes of Asteraceae.

(C) An area of typical open habitat in the Mojave Desert in southern California, USA, which is the home of hundreds of Asteraceae species.

Multiple convergent evolutionary transitions of reproductive characters in Asteraceae

We mapped four reproductive characters onto the phylogeny (Supplemental Table 5), i.e., floret type (also about floral symmetry), capitulescence type, corolla color, and pappus type, to explore the evolutionary histories of these characters in Asteraceae. Our analyses showed that the ancestor of Asteraceae had a loose corymbiform cymose capitulescence composed of capitula (heads) that contained only actinomorphic disc florets with yellow corollas and capillary pappi (Supplemental Figures 28–31). Previous ancestral reconstructions of floral characters generally focused on character transitions at the subfamily or tribe levels and showed that some floral characters have evolved multiple times in parallel, but these studies included smaller numbers of species (61 in Panero et al., 2014 and 244 in Zhang et al., 2021) and lacked many subtribes. Here we sampled almost three times as many species as in Zhang et al. (2021) using nuclear phylogeny and >10 times as that in Panero et al. (2014) using plastid phylogeny. This greater sampling enabled more extensive analyses of the convergent evolution of floral characters in Asteraceae.

We found seven major convergent gains of floral zygomorphy (bilateral floral symmetry) early in the histories of seven subfamilies, producing outer ray florets in Asteroideae and Vernonioideae, ligulate florets throughout the capitulum in Cichorioideae and Pertyoideae, and bilateral florets in Barnadesioideae, Stifftioideae, and Mutisioideae (Figure 5 and Supplemental Figure 28). The only known fossil capitulescence of Asteraceae, Raiguenrayun cura, had ray-like florets and was proposed to be closely related to core Asteraceae (Barreda et al., 2010, 2012; Panero et al., 2014; Rivera et al., 2016; Zhang et al., 2021). However, the character analysis here suggests that the most recent common ancestors (MRCAs) of core Asteraceae and of core Asteraceae plus the sister group Famatinanthoideae had only actinomorphic florets. Therefore, this fossil might represent a stem group of core Asteraceae with an independent transition to zygomorphic florets. In all Cichorioideae subtribes except the earliest divergent Warioniinae, all florets had shifted from actinomorphic to zygomorphic ligulate florets. In addition to the seven major gains of floral zygomorphy, two recent gains of floral zygomorphy were also found in Oldenburgia (Oldenburgieae) and Cnicothamnus (Gochnatioideae) (Figure 5 and Supplemental Figure 28).

Figure 5.

Convergent evolution of florets and floral symmetry in Asteraceae.

(A) Evolution of floret composition of capitula and the floral symmetries of Asteraceae subtribes recognized in the ancestral character reconstruction of florets (Supplemental Figure 28); branch colors indicate floret types. Branches with dashed lines have uncertain floret types. Classification information is shown on the right side, with subtribes in black or gray, tribes in color, and subfamilies in color and all capital letters (CI, Cichorioideae; INC. SED., incertae sedis group). Only the tribe names are shown when all subtribes of a tribe have the same floret type. Tribes in bold and subtribes in bold and black indicate groups that are related to the transition of floret types in our analysis. The solid black blocks on branches and the solid red blocks near tips indicate a change in floral symmetry from radial to bilateral on the corresponding branches or in the corresponding groups, respectively. The black asterisks on branches and the red asterisks near tips indicate a change in floral symmetry from bilateral to radial on the corresponding branches or in the corresponding groups, respectively. The circled numbers mark seven important gains of floral zygomorphy, and the corresponding transitions of floret types are shown in (B) with the same circled numbers. The CYC2 gene duplications detected in this study (Supplemental Figure 39) are shown on the branches. The letters b, c, d, e, and g in blue indicate no or low expression of the corresponding CYC2 gene in more than 90% of the species in the clades, and b, c, d, e, and g in red indicate no or low expression of the corresponding CYC2 gene in all species of the clades included in this study.

(B) Evolutionary patterns of floret types revealed through ancestral character reconstruction, showing the number of times that convergent evolution occurred among different floret types and the convergent gains or losses of floral zygomorphy. The arrows indicate floret transitions, and different colors of arrows represent various changes in floral symmetry. The numbers of convergent transitions among floret types are shown in parentheses near the corresponding arrows, and the names of groups related to the corresponding transitions are shown near the arrows, with the number of involved species in parentheses after the taxon names. The colors of floret types and the groups match the branch colors in (A). Three types of floret transition are marked with circled numbers corresponding to the seven important gains of floral zygomorphy in (A).

Floral zygomorphy was lost at least 46 times, mainly in recently diverged groups in Vernonioideae and Asteroideae (Figure 5 and Supplemental Figure 28). Floral zygomorphy was frequently lost through the loss of the whole ray florets (or the reversion of ray to disc florets), with at least 27 occurrences mainly at the tribe level. The blade of the corolla of a ray floret is called the limb, whereas a floret without the limb is called filiform floret; the limb was lost at least 17 times, generally at the subtribe and genus levels (Figure 5 and Supplemental Figure 28). The transition from ligulate or bilateral florets to disc florets was observed only twice, suggesting that these zygomorphic flowers might involve some molecular mechanisms different from those of ray florets.

Our analyses showed that convergent evolution has also occurred in the evolution of capitulescence structure in Asteraceae (Supplemental Figure 29). All backbone nodes of the Asteraceae phylogeny were shown to have loose corymbiform cymose capitulescences. In particular, the loose corymbiform cymose capitulescence of the MRCA of core Asteraceae and the sister group fit well with the structure of the fossil capitulescence of R. cura, a likely stem group of core Asteraceae (Barreda et al., 2010, 2012; Panero and Crozier, 2016; Zhang et al., 2021). Reduction from a loose corymbiform cymose capitulescence to a solitary capitulum was detected at least 70 times, mainly at the subtribe and genus levels (Supplemental Figure 29). Transitions from solitary capitulescences to open cymose structures were rarer, with 12 such transitions found in Arctotideae, Anthemideae, Astereae, and others (Supplemental Figure 29). There were also transitions to more complex capitulescences, the most common type being the transition from a loose cymose to a dense cymose capitulescence, which occurred at least 32 times, mostly at the subtribe level (Supplemental Figure 29). We also found 14 convergent transitions to densely clustered capitula at the genus level in different tribes, including those from loose cymose capitulescences (nine times), dense cymose capitulescences (twice), and solitary capitula (three times). Extremely compressed globose compound capitula, also known as secondary heads, were gained three times in the sampled taxa and separately derived from dense cymose capitulescences in Echinopsinae (Cardueae) and Craspedia (Gnaphalieae) or from loose cymose capitulescences in Sphaeranthus (Inuleae).

Our analyses also found convergent evolution of corolla color in Asteraceae and revealed a complex evolutionary history of this character (Supplemental Figure 30). Corolla color shifted frequently at the genus or species level, including ∼20 transitions from the Asteraceae ancestral yellow corolla to purple, mainly in Asteroideae and Cichorioideae. Fourteen transitions from yellow to white were detected (mostly in Asteroideae), and further transitions from white to purple were detected 17 times (e.g., in Eupatorieae). There have been more than 10 reversions from white or purple to yellow (e.g., some subtribes of Astereae and Lactuca). An even more complex pattern of shifts in corolla color was found in Astereae, in which some reversions from white to yellow rays were followed by another transition to purple rays in the subtribes Symphyotrichinae (Symphotrichum) and Machaerantherinae (separately in Corethrogyne and Xanthisma) and the unplaced genus Eurybia. In addition, transitions of disc corolla color from yellow to purple and corresponding reversions were found in several groups that have only disc florets, including Inuleae and Cardueae.

Our analyses showed that the pappus, a modified calyx that plays important roles in achene (single-seeded fruit) dispersal, has experienced many convergent evolutionary events from the ancestral capillary pappus to different pappus types, and the derived pappi also have shifted convergently (Supplemental Figure 31). A capillary pappus was also found in the fossil capitula of R. cura, consistent with the pappus type of MRCAs of core Asteraceae and core Asteraceae plus Famatinanthoideae. Sixteen transitions from a capillary to a plumose pappus were generally found in non-Asteroideae lineages, e.g., Mutisieae, Cardueae, and Cichorieae. Thirty-two transitions to a scaly pappus were detected in Astereae, Arctotideae, Cardueae, Cichorieae, and other tribes. Other gains of the scaly pappus have occurred in lineages of the Anthemideae and the Heliantheae alliance that previously produced epappose achenes. An awn-like pappus evolved once in the clade composed of Heliantheae, Coreopsideae, and Neurolaeneae I, with several subsequent transitions to scaly pappi. Unlike the frequent transitions from capillary to other types of pappus in Asteraceae, the reverse transition to a capillary pappus was detected fewer than 10 times, including only one case from a plumose to a capillary pappus in Jurinea. Losses of different kinds of pappi were observed 22 times, except for the plumose pappus, suggesting that the plumose pappus may have played important roles in the corresponding lineages, such as dispersal of achenes, and has thus been under selection. The complex convergent evolution of the pappus suggests that various pappus types were strongly selected and have played an important role in the success of Asteraceae.

We found that convergent transitions of some floral characters were related to species-rich groups and may have been responsible for increases in the species numbers of some clades compared with their sister groups, suggesting that convergent evolution of characters may have played an important role in diversification of Asteraceae. Asteroideae, the largest subfamily (∼17 000 species), has gained floral zygomorphy, whereas its sister subfamily Corymbioideae contains only nine species that produce capitula with only actinomorphic disc florets, the plesiomorphic condition (Figure 5). The fact that zygomorphic florets are found in ∼70% of Asteraceae species suggests that they likely promoted diversification via increased interaction with pollinators. The zygomorphic flowers of Asteraceae developed enlarged limbs that increased floral display, possibly contributing to the diversification of these groups. However, there are also cases in which the convergent loss of zygomorphic flowers was related to increases in species diversity. For example, ray florets were lost in Eupatorieae, with ∼2400 species, in comparison with its sister group Perityleae, which have retained ray florets and only contain ∼84 species, suggesting that the loss of ray florets (and thus the flower-like morphology of capitula) has not negatively impacted the diversity of Eupatorieae (Figure 5). The various parallel losses of floral zygomorphy suggest that alternative pollination strategies may have allowed such losses to reduce the cost of producing large corollas. Moreover, our study showed that frequent convergent transitions of the pappus in the Heliantheae alliance significantly increased pappus diversity, which may in turn have increased the adaptability in achene dispersal and contributed to species divergence in the Heliantheae alliance (Supplemental Figure 31).

Duplication history of Asteraceae floral MADS-box genes and implications for inflorescence and flower evolution

We reconstructed the phylogeny of MADS-box homologs of ABCE genes from 103 Asteraceae species and outgroups, revealing duplications of these genes in Asteraceae (Figure 6 and Supplemental Figures 32–36). Duplications (or triplication) involving Asteraceae and its sister Calyceraceae were found in the FUL clade of the APETALA1 (AP1) group, the AP3/DEF clade of the APETALA3 (AP3) group, and the STK clade of the AGAMOUS (AG) group. Additional duplications involving all Asteraceae were found in AP1/CAL and AGL79 clades of the AP1 group and the SEP3 clade of the SEPELLATA (SEP) group. The ancestor of the core Asteraceae experienced duplication of the paralog of Gerbera GSUA2, producing the AGL79-like and M41 clades with subsequent loss of AGL79-like genes in Asteroideae. In Asteroideae, we detected duplications in all five groups of ABCE genes in species of the Heliantheae alliance and sometimes also in its sister tribe Athroismeae. The former type was found in SQUA3, CAL, and FUL clades of the AP1 group, the AG clade of the AG group, and the RCD7 clade of the SEP group, and the latter type was detected in the M41 clade of the AP1 group, the AP3-1 clade of the AP3 group, the PI clade of the PISTILLATA (PI) group, and the AG2 clade of the AG group. Our results generally agreed with previous phylogenetic analyses of MADS-box ABCE genes that used a much smaller number of species (Zahn et al., 2005; Broholm et al., 2010; Ruokolainen et al., 2010b; Zhang et al., 2017). Our broader analysis also provided more specific information about previously identified duplications and detected some new duplications in Asteraceae. For example, our results provided a more specific position of duplication of SEP3 homologs and showed that AST.SEP3 genes are specific to Asterales rather than to Asteraceae. Our results also showed that most duplications detected here occurred in five clades that exhibit high levels of floral morphological diversity (solid circles with different colors in Figure 6), suggesting that duplication of floral regulator genes may have contributed to floral diversification in Asteraceae (see supplemental information for details).

Figure 6.

Duplications of MIKC^C type MADS-box genes with functions in floral organ identity in the ABCE model.

The tree on the left shows a simplified phylogeny of the Asteraceae groups included in this analysis; the solid circles on the tree and the arrows with dashed lines indicate the whole-genome duplication (WGD) events reported in Huang et al. (2016a) and Zhang et al. (2021) (C + A: the WGD event at the common ancestor of Calyceraceae and Asteraceae; A + H: the WGD event at the common ancestor of Athroismeae and the Heliantheae alliance). The tips of the tree show the abbreviated names of taxa (Meny, Menyanthaceae; Good, Goodeniaceae; Caly, Calyceraceae; Barn, Barnadesioideae; Muti, Mutisioideae plus Stifftioideae; Card, Carduoideae plus Tarchonanthoideae; Vern, Vernonioideae; Cich, Cichorioideae; Sene, Senecionodae; Inul, Inuleae; Arth, Athroismeae; Hele, Helenieae; HA1, Heliantheae alliance clade 1; HA2, Heliantheae alliance clade 2). The symbols to the right of the taxon names represent three floral characters of corresponding groups from ancestral character reconstruction, with explanations of the characters shown above the phylogeny. Four gene trees to the right of the floret illustrations show the phylogenetic results of several types of MADS-box genes analyzed in this work (colored names at the top); the gene names in black on the left of the branches indicate genes reported in previous work, and the gene names in magenta indicate new gene copies identified in this study. The names of clades used in this study are indicated on the left of the clades in green bold-italic font. The question mark near the node of the FUL clade in the APETALA1 tree indicates that the corresponding node was not well resolved in our results.

The capitulum (head) is a determinate inflorescence with sessile florets and is a most remarkable characteristic shared by all Asteraceae species. It differs from the compressed thyrsoid inflorescence of Calyceraceae (sister of Asteraceae) in lacking cymose units (branches with several pedicellate flowers) and a terminal flower (Pozner et al., 2012). The Gerbera GRCD2 gene (a SEP1/2 homolog) was found to be required for normal inflorescence meristem (IM) identity (Uimari et al., 2004), such that anti-GRCD2 Gerbera plants with reduced GRCD2 expression produced inner IMs from positions of floral meristems (FMs) (Uimari et al., 2004; Teeri et al., 2006; Zhang et al., 2017). Such abnormal inflorescences resemble the cymose units of the Calyceraceae inflorescence and may represent the ancestral morphology of Asteraceae and Calyceraceae (Pozner et al., 2012). Also, downregulation of a close GRCD2 paralog, GRCD7, resulted in a similar phenotype, indicating that both GRCD2 and GRCD7 are important for determination of the Gerbera capitulum IM (Zhang et al., 2017). Our analysis showed that GRCD2 and GRCD7 were duplicated in the common ancestor of core Asteraceae (Figure 6), suggesting that their functions are probably conserved in core Asteraceae. In addition, reduced GRCD7 expression was observed in a GRCD 4/5 double RNAi mutant with the transition of FM to IM (Zhang et al., 2017). Our analysis showed that GRCD5 (the other copy is GRCD8) arose from a duplication in the common ancestor of Asteraceae (Figure 6), suggesting that GRCD5 may have contributed to the origin of the capitulum in Asteraceae by regulating GRCD7 and its homologs in Asteraceae subfamilies, including Barnadesioideae.

We also detected duplications of ABC-class MADS-box genes at the ancestor of core Asteraceae, Asteraceae, or the common ancestor of Asteraceae and Calyceraceae, with implications for evolution of the Asteraceae capitulum. For example, early expression of Gerbera DEFICIENS homologs in the capitulum is associated with the formation of floral primordia, suggesting that these genes might participate in specification of capitulum identity (Yu et al., 1999). Our analysis showed that the DEFICIENS homologs originated from a duplication in the common ancestor of Calyceraceae and Asteraceae, suggesting that their function in the Asteraceae capitulum could be related to that in the compressed thyrsoid inflorescences of Calyceraceae. Furthermore, homologs of the Arabidopsis AP/CAL, FUL, and AGL79 genes with a role in meristem determination were duplicated at the ancestor of Asteraceae or the common ancestor of Calyceraceae and Asteraceae (Figure 6). Therefore, these duplications of ABCE-class MADS-box genes may have played an important role in the origin and evolution of the capitulum in Asteraceae.

The present phylogenetic study of Asteraceae MADS-box genes with greater sampling showed that the A- and E-class MADS-box genes have experienced greater expansion in Asteraceae than the B- and C-class MADS-box genes. Besides their role in development of the capitulum, A- and E-class genes have also been implicated in regulation of Asteraceae pappus development. For example, GRCD3 from Gerbera hybrida, an ortholog of AGL6, was highly expressed in the pappus and was proposed to contribute to its development (Zhang et al., 2017). Furthermore, an A-class gene from the SQUA3 clade (Figure 6) was preferentially expressed in the pappus bristles of Lactuca sativa (Ning et al., 2019) and Taraxacum officinale (Vijverberg et al., 2021), suggesting an important role in pappus development. These results and the duplications in Asteraceae suggest that expansion of both A- and E-class genes (Figure 6) in Asteraceae may be related to the evolution and diversity of the pappus.

Molecular phylogeny of Asteraceae TCP genes: Duplications of CYC2 genes predate multiple gains of zygomorphic florets

The TCP genes are classified into the CYC and CIN clades on the basis of phylogenic analysis and include members that regulate plant shoot development and floral symmetry (Luo et al., 1996, 1999; Nath et al., 2003; Crawford et al., 2004). Functional studies of Asteraceae CYC and CIN homologs in G. hybrida, Helianthus annuus, Senecio vulgaris, and Chrysanthemum morifolium have revealed the functions of some genes in controlling floral organ identities and symmetries (Fambrini et al., 2011; Tähtiharju et al., 2012; Juntheikki-Palovaara et al., 2014; Huang et al., 2016c; Elomaa et al., 2018; Yu, 2020; Zhao et al., 2020) (see below for more details).

To examine the evolutionary history of Asteraceae CYC/CIN genes and investigate possible links to the evolution of floral symmetry, we performed phylogenetic analyses of CYC/CIN homologs from 103 Asteraceae species. The results showed that Asteraceae members of the CYC clade formed three classes, CYC1, CYC2, and CYC3 (Figure 7, as indicated at the top of the gene tree; Supplemental Figure 37), consistent with previous results showing that they resulted from the gamma triplication shared by core eudicots (Tähtiharju et al., 2012; Zhao et al., 2020). Our analysis found four duplications of CYC1 homologs, six duplications of CYC2 homologs, and four duplications of CYC3 homologs. The Asteraceae CIN-like genes formed three clades in our analyses, the GhCIN1/2, GhCIN10, and GhCIN3/4/5 groups (Figure 7 and Supplemental Figure 38), in agreement with previous findings (Zhao et al., 2020). We detected three duplications of GhCIN1/2 homologs, five duplications of GhCIN10 homologs, and seven duplications of GhCIN3/4/5 homologs. The CYC and CIN genes experienced multiple duplications shared by Asteraceae and Calyceraceae or involving Asteraceae, core Asteraceae, and the Heliantheae alliance, leading to the expansion of CYC/CIN genes in Asteraceae and potentially contributing to the convergent evolution of floret types and floral symmetries in Asteraceae, as shown in our character analyses.

Figure 7.

Phylogenetic analysis of CYCLOIDEA-like and CINCINNATA-like genes of the TCP family and relative expression levels of CYC2 genes in several Asteraceae species.

(A) Phylogenetic analysis of Asteraceae CYCLOIDEA-like and CINCINNATA-like genes of the TCP family. The tree on the left shows a simplified phylogeny of the groups included in the analysis; the solid circles and arrows with dashed lines indicate the WGD events reported in Huang et al. (2016a) and Zhang et al. (2021) (C + A: the WGD event at the common ancestor of Calyceraceae and Asteraceae). To the right of the tips of the tree are abbreviated taxon names (Meny, Menyanthaceae; Good, Goodeniaceae; Caly, Calyceraceae; Barn, Barnadesioideae; Muti, Mutisioideae plus Stifftioideae; Card, Carduoideae plus Tarchonanthoideae; Vern, Vernonioideae; Cich, Cichorioideae; Sene, Senecioneae; Anth, Anthemideae; Gnap, Gnaphalieae; Aste, Astereae; Inul, Inuleae; Hele, Helenieae; HA1, Heliantheae alliance clade 1; HA2, Heliantheae alliance clade 2). The symbols near the tips of the tree represent floret types of corresponding groups from the results of ancestral character reconstruction, and their explanations are shown under the phylogenetic tree. The gene trees to the right of the floret illustrations show the phylogenetic results of CYC-like and CIN-like genes analyzed in this work; the names of previously reported genes are shown in black at the left of the branches, and the names of new gene copies identified in this study are shown in magenta. The names of clades used in this study are indicated on the left of the clades in green bold-italic font. The question marks near nodes indicate that the corresponding nodes were not well resolved in our results.

(B) Relative expression levels of six CYC2 genes in eight Asteraceae species determined by qRT–PCR. The phylogenetic relationships among these species are shown on the left. The genus and tribe names of corresponding species are shown to the right of the branch tips. The background colors of the names indicate the floral symmetries of the corresponding genera, with explanations shown below the tree. The symbols near the names represent the floret types of corresponding species used in the qRT–PCR analysis. The phylogenetic relationships of six CYC2 genes are provided below the bar charts, with colored circles indicating duplications of CYC2 genes as shown in (A). The bar charts show the relative expression level of CYC2a and five CYC2 genes that regulate floral symmetry, i.e., CYC2b–CYC2e and CYC2g. Bar height indicates the mean relative expression level of three replicates, with the highest bar of each species set to 1.00. Expression levels of the two paralogs CYC2e1 and CYC2e2 in Chrysanthemum are shown as the left and right bars, respectively, in the same cell. The expression levels of CYC2b2 and CYC2e2 of Zinnia and CYC2b2 of Gerbera are provided, but the expression of the other close paralog of each gene was not detected or was less than 0.01. N.D. indicates that expression was not detected.

Our phylogenetic analysis of Asteraceae CYC2 genes, including representatives from Gerbera (GhCYC2, GhCYC3, GhCYC5, and GhCYC4/9), Senecio (RAY1, RAY2, and RAY3), and Helianthus (HaCYC2c, HaCYC2d, and HaCYC2g), revealed several duplications, resulting in six clades shared by core Asteraceae (CYC2a–CYC2e and CYC2g; green in Figure 7). These results were largely consistent with previous analyses using genes from 50 Asteraceae representing 20 tribes in seven subfamilies (Chen et al., 2018). Note that previous studies of genes from different Asteraceae species have sometimes used different letters for various CYC2 genes, such that the same letter does not necessarily indicate orthology (see Figure 7 for gene orthology). An early duplication of the single copy shared with other core eudicots occurred at the ancestor of Asteraceae and Calyceraceae; one of the duplicates is the ancestor of the CYC2a clade, including the Gerbera GhCYC7 gene that is mainly expressed in the inflorescence stem during the early developmental stages of ray-like outer florets (Tähtiharju et al., 2012). The sister copy of CYC2a was duplicated at the level of the Asteraceae ancestor, and both resulting gene copies experienced further duplication (triplication) at the ancestor of core Asteraceae, generating the ancestors of the CYC2b–CYC2e and CYC2g clades (Figure 7).

Previous molecular and genetic studies in several Asteraceae species support the function of CYC2b–CYC2e and CYC2g genes in development of ray florets (Fambrini et al., 2011, 2014; Huang et al., 2016c; Bello et al., 2017; Chen et al., 2018; Shen et al., 2021). The CYC2 duplication shared by Asteraceae and Calyceraceae generated CYC2a and another copy. The gain of zygomorphic florets in the two earliest divergent subfamilies may have been dependent on the single-copy paralog of CYC2a; however, the subsequent duplication and the increased copies of CYC2b–CYC2e/CYC2g genes in core Asteraceae and possible functional innovations of the paralogs probably contributed to additional convergent gains of zygomorphic florets. This idea is further supported by the newly detected Heliantheae-alliance-specific duplication with a newly detected CYC2e2 copy in the CYC2e clade, which was not recognized in previous phylogenetic studies (Chen et al., 2018), and with the observed preferential expression of one of these genes (HaCYC2e) in Helianthus ray florets (Tähtiharju et al., 2012).

In summary, in each of the CYC1, CYC2, and CYC3 groups, there were duplications shared by most Asteraceae subfamilies (core Asteraceae or earlier) and by the Heliantheae alliance, suggesting that these duplications may have enhanced the function for development of zygomorphic florets. In particular, as mentioned above, CYC2 genes play important roles in establishment of bilateral floral symmetry in Asteraceae, suggesting that the duplications of CYC2 genes in Asteraceae may have contributed to the convergent gains of floral zygomorphy in several large subfamilies. The CIN genes also participate in Asteraceae floral corolla development, and duplications of CIN genes may also have contributed to the establishment of floral zygomorphy in Asteraceae.

Relative expression levels of CYC2 genes support roles in floral zygomorphy

Previous studies (Huang et al., 2016c; Chen et al., 2018) and our analyses here (Figure 7A) recognized six Asteraceae CYC2 clades (CYC2a–CYC2e and CYC2g). These results suggest that the early duplication of CYC2 may have facilitated functional diversification of the ancestor of CYC2b–CYC2e/CYC2g in support of zygomorphic floret development, and further duplications of these genes in core Asteraceae may have enhanced this function. Our analyses revealed that transcripts of different CYC2 genes or gene combinations were detected in different clades with zygomorphic florets in Asteraceae, suggesting that different CYC2 gene functions may provide part of the molecular basis for convergent gains of Asteraceae zygomorphic florets. In addition, low expression of some of these genes, particularly in lineages that have experienced loss of zygomorphic florets, suggests that reduced CYC2 expression may have been the underlying cause of the corresponding morphological convergence. To test the hypothesis that divergence in CYC2 gene expression might be important for the morphological convergence of floral symmetry in Asteraceae, we first expanded the phylogenetic analyses of CYC2 genes using seven available Asteraceae genome sequences and all transcriptomes with detected floral-specific MADS-box genes as a positive control (249 Asteraceae species in total) (Supplemental Table 6; Figure 7A; Supplemental Figures 39 and 40). We analyzed the relative expression levels of CYC2 genes in seven tribes using quantitative RT–PCR (qRT–PCR), including zygomorphic florets of six species and actinomorphic florets of two species that lack zygomorphic florets (Figure 7B and Supplemental Figure 41).

Asteroideae are the largest subfamily of Asteraceae, containing more than 17 000 species, and have experienced both gain of highly differentiated ray florets and multiple subsequent losses of ray florets in some tribes or smaller groups (Figure 5). Several studies of CYC2 function in ray floret development have been performed using members of Asteroideae and support a role for five CYC2 genes in the regulation of floral symmetry (CYC2b–CYC2e and CYC2g, but probably not CYC2a) (Huang et al., 2016c; Bello et al., 2017; Chen et al., 2018). Our analysis of Asteroideae transcriptomes detected each of five CYC2 genes that regulate floral symmetry (CYC2b–CYC2e and CYC2g) in members of some tribes with rays, e.g., Heliantheae and Calenduleae (Supplemental Figure 40). In the supertribe Helianthodae, CYC2b, CYC2d, and CYC2e genes were frequently detected in clades with zygomorphic florets (Supplemental Figure 40). Furthermore, in a qRT–PCR analysis of ray florets of Zinnia (Heliantheae), expression levels of CYC2b, CYC2d, and CYC2e were 5–25 times as high as that of CYC2a, which may not regulate floral symmetry (Figure 7B and Supplemental Figure 41), consistent with the transcriptome data. In previous analyses of three tribes in the HA II clade of Helianthodae, CYC2c was shown to be important for the development of ray florets in Helianthus (Heliantheae) and was expressed in ray florets of Bidens (Coreopsideae) and Tagetes (Tageteae) (Supplemental Figure 40) (Fambrini et al., 2011; Chen et al., 2018). Our analysis showed that the expression level of CYC2c in the ray florets of Zinnia (Heliantheae) was four times as high as that of CYC2a (Figure 7B and Supplemental Figure 41), supporting a role for CYC2c. These analyses suggest that the combination of CYC2b, CYC2d, and CYC2e genes probably plays important roles in the development of zygomorphic florets in the Helianthodae supertribe, and CYC2c may also contribute to floral zygomorphy of some tribes in the HA II clade. Compared with a previous analysis, the expression pattern of CYC2 genes in ray florets of Zinnia (Heliantheae) was similar to that of Tagetes (Tageteae) rather than Helianthus (Heliantheae) from the same tribe, suggesting possible divergence of the CYC2 gene expression pattern within the same tribe (Chen et al., 2018).

In the phylotranscriptomic analysis of Senecionodae, the other supertribe of Asteroideae, CYC2b, CYC2d, CYC2e, and CYC2g transcripts were frequently detected. The expression levels of CYC2b and CYC2e measured by qRT–PCR were significantly higher than that of CYC2a in Symphyotrichum ray florets (Astereae), whereas the expression levels of CYC2d and CYC2g were slightly higher than that of CYC2a (Figure 7B and Supplemental Figure 41). In ray florets of Chrysanthemum lavandulifolium (Anthemideae), the expression levels of CYC2e1 and CYC2e2 (two CYC2e paralogs) were two to three times as high as that of CYC2a, whereas CYC2d was expressed at a slightly higher level than CYC2a, similar to the expression patterns of the Symphyotrichum CYC2 genes. However, no expression of CYC2b or CYC2g was detected in ray florets of Chrysanthemum, suggesting divergence of CYC2 gene expression between Astereae and Anthemideae. A previous study of another Chrysanthemum species, i.e., C. morifolium, showed that CYC2g was highly expressed and that the level of CYC2b expression was relatively low but detectable in ray florets (Chen et al., 2018), suggesting divergence of CYC2 gene expression in ray florets of even closely related species in the same genus.

Cichorioideae contain one tribe (Cichorieae) with ∼1500 species, almost all of which have only zygomorphic ligulate florets. In the phylotranscriptomic analysis, CYC2e and CYC2b were detected in many species of the subfamily, and CYC2c and CYC2g were detected in a smaller number of species (Supplemental Figure 40). Our qRT–PCR analysis of the (ligulate) florets of Sonchus showed that CYC2b was highly expressed, with detectable expression of CYC2e (Figure 7B and Supplemental Figure 41). In a previous analysis of Taraxacum (Cichorieae) ligulate florets, CYC2b and CYC2e were more highly expressed than CYC2c and CYC2g (Chen et al., 2018). Both the phylotranscriptomic and qRT–PCR expression analyses suggested that the combination of CYC2b and CYC2e might be important for the ligulate florets of Cichorioideae, with possible contributions of CYC2c and CYC2g to the development of ligulate florets in some Cichorioideae species. In Arctotideae of the subfamily Vernonioideae, a tribe with ray florets, phylotranscriptomic analysis showed that CYC2b and CYC2e transcripts were frequently detected and that CYC2g was detected in one species (Supplemental Figure 40). Consistent with these findings, qRT–PCR analysis showed that CYC2b and CYC2e were highly expressed in ray florets of Gazania in this tribe, with CYC2c expression detected at a lower level (Figure 7B and Supplemental Figure 41). In the subfamily Mutisioideae, which contains many species with bilateral florets, we frequently detected CYC2b, CYC2c, and CYC2e transcripts (Supplemental Figure 40). In addition, in qRT–PCR analysis of Gerbera from this subfamily, CYC2b, CYC2c, and CYC2e were more highly expressed in the bilateral florets than CYC2a (Figure 7B and Supplemental Figure 41). Thus, the expression patterns of CYC2 genes are similar in these three subfamilies.

qRT–PCR analyses of florets (all actinomorphic) from Liatris (Eupatorieae) indicated that CYC2b and CYC2e were expressed at approximately one-third to one-half the level of CYC2a, and CYC2d was expressed at an even lower level (Figure 7B and Supplemental Figure 41). CYC2d was rarely detected in the phylotranscriptomic analysis of actinomorphic clades of Helianthodae, especially Eupatorieae and closely related tribes (Supplemental Figure 40). Therefore, low expression of these three genes, especially CYC2d, may be associated with losses of floral zygomorphy in Helianthodae, particularly in Eupatorieae and closely related tribes. In Senecionodae of Asteroideae, CYC2c was not detected in two genomes (C. morifolium and Artemisia annua) and 71 transcriptomes of Astereae, Anthemideae, Gnaphalieae, and Senecioneae (Figure 5 and Supplemental Figure 40). In addition, CYC2c expression was not detected in the ray florets of Chrysanthemum (Anthemideae) and Symphyotrichum (Astereae) or in actinomorphic florets of Artemisia and Crossostephium (both Anthemideae) (Figure 7B and Supplemental Figure 41) (Chen et al., 2018). In the qRT–PCR analysis of actinomorphic florets of Artemisia, CYC2e was expressed at a relatively low level compared with that in ray florets of the closely related genus Chrysanthemum (Figure 7B and Supplemental Figure 41). In the subfamily Carduoideae, most of whose species have only actinomorphic florets, the phylotranscriptomic analysis did not detect CYC2c, CYC2d, or CYC2g transcripts (Supplemental Figure 40). These expression results suggest that convergent losses of floral zygomorphy in many lineages of Asteraceae may be associated with low or no expression of some CYC2 genes that are highly expressed in related lineages that have zygomorphic florets.

The phylotranscriptomic and qRT–PCR analyses here showed that three early-divergent Asteraceae clades with zygomorphic florets, i.e., Mutisioideae, Cichorioideae, and Arctotideae, have similar expression patterns of CYC2 genes, suggesting that there may have been a common underlying molecular precondition for morphological convergence. However, in the largest subfamily, Asteroideae, several divergences in CYC2 expression patterns were observed among different clades with zygomorphic florets, consistent with studies of CYC2-like genes in other angiosperm lineages with zygomorphic flowers, e.g., Gesneriaceae (Song et al., 2009; Yang et al., 2012, 2015; Liu et al., 2021a). In addition, our analysis showed that losses of floral zygomorphy in different Asteraceae lineages may be associated with low to absent expression of different CYC2 genes. Our results on divergence in CYC2 gene expression patterns in multiple Asteraceae lineages with or without zygomorphic florets provide new insights into the molecular basis of the convergent evolution of floral symmetry.

Concluding remarks and implications

Here, we increased taxon sampling for phylogenetic reconstruction of Asteraceae by integrating GS datasets with transcriptome datasets for ∼43% of taxa. The assessment of GS datasets and phylogenetic analyses showed that such combined datasets with sufficient transcriptome fractions can yield highly supported phylogenies, and the inclusion of GS datasets from readily available herbarium specimens and other dried materials enables greater taxon sampling at a relatively low cost. The expanded Asteraceae nuclear phylogenies provide a framework for analyses of divergence times, convergent evolution of floral characters, and duplication and loss patterns of genes that regulate floral development.

Asteraceae are estimated to have originated in the Late Cretaceous, with the divergence of all extant subfamilies occurring from the Late Cretaceous to the Early Paleocene. Most lineages within subfamilies diverged starting in the Early Eocene, matching the global temperature decrease after the Early Eocene Climatic Optimum and the subsequent appearance and expansion of various open habitats (Jacobs et al., 1999; Retallack, 2001; Strömberg, 2004, 2005; Strömberg et al., 2013; Bellosi and Krause, 2014; Azevedo et al., 2020), suggesting that open habitats and cool climates may have played important roles in diversification of Asteraceae.

Character analyses revealed more detailed evolutionary patterns of four reproductive characters compared with previous work (Panero et al., 2014; Zhang et al., 2021) and demonstrated that the convergence of these characters was very common in Asteraceae. We found that many convergent events have occurred in shallow lineages and potentially have facilitated the recent divergences in these groups, thus contributing to the species richness of Asteraceae. We suggest that the transition to capitula with various floret types and an enlarged corolla resembling that of typical simple flowers may have helped to attract pollinators and could have promoted Asteraceae diversification. The combination of convergent evolution in reproductive characters has likely enhanced the adaptation of various Asteraceae lineages to a wide spectrum of environments, thus promoting the diversification and species richness of Asteraceae.

Morphological transitions are likely supported by changes in underlying molecular regulatory mechanisms. The MADS-box and TCP genes experienced multiple duplications (Figures 6 and 7) that were coincident with several whole-genome duplication (WGD) events in Asteraceae, and the expansion of these genes may have promoted the differentiation and increased complexity of reproductive characters, which in turn have likely enhanced pollinator attraction and achene dispersal, accelerating the diversification of Asteraceae. Duplications of CYC2 genes and divergent expression patterns of the five CYC2 genes that regulate ray floret development have probably contributed to the convergent evolution of floral symmetry and thus to the diversification of many Asteraceae groups.

Methods

Asteraceae datasets from transcriptomes and genomes

We sampled 310 transcriptome datasets and two genomic datasets from 306 Asteraceae species and five outgroups (collectively called transcriptomes in the following discussion), including 64 newly sequenced for this study, 212 from our previous studies (Liu et al., 2015; Huang et al., 2016b; Zhang et al., 2021), and 34 from GenBank. In addition, GS datasets from 526 Asteraceae samples and four outgroups were newly sequenced for this study, producing 0.5–5 Gbp data for each dataset (Figure 1A). Our samples included 706 Asteraceae species that represented all 16 currently recognized subfamilies (Susanna et al., 2020) and 41 of 45 recognized tribes (Anderberg et al., 2007; Funk et al., 2009; Panero et al., 2014; Susanna et al., 2020). The sampling covered 125 subtribes and 12 genera that have not previously been placed in a subtribe.

Sequencing and selection of nuclear gene sets

DNA and RNA extraction, library preparation, and sequencing were performed as described in previous works (Huang et al., 2016b; Li et al., 2019; Zhang et al., 2021). Transcriptomes were assembled using Trinity 2.9.1 (Grabherr et al., 2011) as described previously (Huang et al., 2016a; Zhang et al., 2021). After examination of the completeness and quality of the transcriptomes and genomes using BUSCO 3.0.2 (Simão et al., 2015), we chose high-quality transcriptomes (Supplemental Table 1 and Supplemental Figure 8) from ten Asteraceae species with which to identify orthogroups using OrthoFinder v2.3.3 (Emms and Kelly, 2015). For phylogenetic analyses, 1094 core orthogroups that have one copy in each of the 10 Asteraceae representatives were obtained and used as queries to identify (1) homologs from transcriptomes and public genomes using HaMStR v13.2.16 (Ebersberger et al., 2009) with a cutoff of 1e−20 for BLAST and hmmster searches and (2) homologous reads from GS datasets using BBmap tools (Bushnell, 2015) and the script reads2sam2consensus_baits.py (Vargas et al., 2019). The depth, length, and completeness of mapped genes were calculated (Supplemental Tables 7–9), and genes with length <300 bp were removed because of their low power for resolution and low completeness.

To examine the phylogenetic relationships of GS datasets and transcriptomes from the same species, we used taxon coverage by the GS datasets of at least 80% of the subtribes as a criterion to select 529 of the 1094 orthogroups for reconstruction of a preliminary tree with all samples (supplemental information and Supplemental Figure 7). In the preliminary tree, GS datasets and transcriptomes from the same species generally had sister relationships or very close positions. Therefore, to reduce the complexity of the analyses, the 1094 orthogroups from different datasets of the same species (including 102 transcriptomes and 146 GS datasets from 120 species, Supplemental Table 1) were merged in subsequent analyses. In the merged datasets, the proportion of Asteraceae species with transcriptomes was 43% (306 of 706) (Supplemental Tables 1 and 2). To test the robustness of the phylogenetic trees and the effects of different rates of missing data at subtribe or higher levels, we first generated four subsets of the 1094 orthogroups composed of genes with different extents of taxon coverage at the subtribe level: 1010 (≥82% of subtribes), 903 (≥85% of subtribes), 811 (≥88% of subtribes), and 686 (≥90% of subtribes) orthogroups.

To reduce the effects of putative non-orthologs (PNOs) on the phylogenetic resolution, we used tree-based comparison to identify putative orthologs. To identify PNOs, we used published Asteraceae phylogenetic trees (Mandel et al., 2019; Zhang et al., 2021) and our preliminary tree as a guide and used the 686 orthogroups, in part because they have relatively low amounts of missing data and the corresponding single-gene trees are generally sufficient for distinguishing orthologs from paralogs. The single-gene trees of the 686 orthogroups were examined, and sequences (from either transcriptome or GS datasets) that were placed outside Asteraceae or had significant conflicts with monophyly of Asteroideae, Barnadesioideae, and Mutisioideae were considered PNOs and were removed, resulting in the 686-gene set. Sequences with extremely long branch lengths (>10 times longer than the sister clade) in gene trees are generally contaminants and were also removed. The same putative PNOs and sequences with long branches were also removed from the other four orthogroups mentioned above, resulting in four gene sets with 1094, 1010, 903, and 811 genes. In addition, for molecular clock estimates, which use a supermatrix dataset and are prone to systematic errors with many hundreds of genes (Xi et al., 2016), we selected a 334-gene set with 95% coverage of subtribes (each gene identified in ≥95% of subtribes) (Supplemental Table 3) starting from the 686-gene set.

Phylogenetic reconstruction, divergence time estimation, and morphological character evolution

Gene sequences were aligned using MAFFT v7.407 (Katoh and Standley, 2013) with automatic strategy selection and trimmed using trimAl v1.4.rev22 (Capella-Gutiérrez et al., 2009) with removal of sites with 80% or more gaps. Phylogenetic trees were reconstructed for each gene (orthogroup) in simulated and empirical datasets using RAxML v8.2.12 (Stamatakis, 2014) with the GTRGAMMA model, and support values for each gene tree were inferred from 100 rapid bootstrap replicates. Coalescent trees were generated using Astral v5.14.2 (Zhang et al., 2018) with inference of both local posterior probabilities and multi-locus BS values. A supermatrix tree was reconstructed from a concatenated matrix composed of sequences from the 334-gene set in RAxML v8.2.12 (Stamatakis, 2014) with the GTRGAMMA model and 100 rapid bootstrap replicates. For divergence time estimation, another topology-fixed supermatrix tree and 100 topology-fixed bootstrap trees were generated with the same parameters but using the 686-gene coalescent tree as the guide tree.

Ten fossils and one secondary calibration node (scenario 1 in Supplemental Table 4) were used in the calibrations of Asteraceae divergence time estimation using penalized likelihood methods in TreePL (Smith and O’Meara, 2012) on the basis of the topology-fixed supermatrix tree (see supplemental information for details). The 95% HPD of divergence time was estimated on the basis of 100 topology-fixed bootstrap trees with calibration scenario 1. Six alternative calibration scenarios were used for time estimation to investigate the influence of calibration differences (scenarios 2–7 in Supplemental Table 4). Ancestral state reconstruction in Asteraceae was performed for four reproductive traits—floret type, capitulescence type, corolla color, and pappus type—using a maximum-likelihood method in the R package phytools v0.7-47 (Revell, 2012) and the 686-gene coalescent tree (see Supplemental Table 5 for character-coding information).

Phylogenetic analyses of the MADS-box and TCP transcription factor families

MADS-box and TCP transcription factor sequences of Arabidopsis thaliana downloaded from TAIR (Berardini et al., 2015) were used as queries to obtain their homologs from seven publicly available Asteraceae genomes and 96 Asteraceae transcriptomes (Supplemental Table 1 and Supplemental Figure 5; see also supplemental information). We obtained candidate MADS-box and TCP sequences using BLASTP (Camacho et al., 2009) with an e-value cutoff of 1e−5. Candidates with SRF-TF and K-box domains for the MADS-box family and a TCP domain for the TCP family were identified using hmmsearch in HMMER3 (Eddy, 2011) and the Pfam database (Mistry et al., 2021) with an e-value cutoff of 1e−5 and retained for further analyses. After removal of alternatively spliced sequences, the protein sequences were aligned using MAFFT v7.407 (Katoh and Standley, 2013) with the --dash parameter, and corresponding coding sequence alignments were generated using the tranalign program of EMBOSS 6.6.0 (Rice et al., 2000); the alignments were manually adjusted, and regions with low levels of similarity were removed. Gene family trees were reconstructed using RAxML v8.2.12 (Stamatakis, 2014) with the GTRGAMMA model and 100 bootstrap replicates. When analyzing the relative expression of CYC2 genes, because transcriptomes do not sample genomes completely, particularly for genes with low expression levels, lack of detection could be due to low expression. As a control for transcriptome quality and the inclusion of floral transcripts, we also detected the expression of floral-specific MADS-box genes and only considered the 249 transcriptomes that contained transcripts for one or more floral-specific MADS-box genes for transcript detection of CYC2 genes.

qRT–PCR of CYC2 genes

Eight species in seven tribes of four subfamilies were selected for qRT–PCR analysis of relative expression levels of CYC2 genes in florets, i.e., A. annua, C. lavandulifolium, Gazania rigens, Gerbera hybrid, Liatris scariosa, Sonchus asper, Symphyotrichum hybrid, and Zinnia elegans (Supplemental Table 10). Total RNA was isolated from zygomorphic florets of opening capitula of six species with zygomorphic florets and from actinomorphic florets of two species without zygomorphic florets (Figure 7B). Total RNA from each species was normalized to 50 ng/μL, and cDNA was synthesized using a HiScript III 1st Strand cDNA Synthesis Kit (Vazyme, Nanjing, China) after digestion of genomic DNA. The products were diluted five times with double-distilled water. Primers were designed for each gene according to the gene sequences obtained from the transcriptomes (Supplemental Table 10). Because the CYC2c sequence was not found in transcriptomes of Astereae and Anthemideae, CYC2c expression in species of these tribes was tested using general primers designed according to the CYC2c gene sequences of Asteroideae species (Supplemental Table 10). The reactions were performed in a 20-μl system with 2 μl of cDNA template using AceQ qPCR SYBR Green Master Mix (Vazyme). The ACTIN7 gene of each species was used as the reference gene for normalization, and there were three technical replicates for each gene.

Data and code availability

Raw reads generated in this study have been deposited in the NCBI SRA database under BioProject PRJNA1077795 and in the National Genomics Data Center under BioProject PRJCA023669.

Funding

This work was supported by funds from the Eberly College of Sciences and the Huck Institutes of the Life Sciences at the Pennsylvania State University, the Hunan Normal University and by the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB31000000), the Large-Scale Scientific Facilities of the Chinese Academy of Sciences (2017-LSFGBOWS-02), and the National Natural Science Foundation of China (nos. 32270229, 31870179, 31570204, 31270237, 31070167, 30670148). Additional support was provided by the Key Project at Central Government Level: the Ability Establishment of Sustainable Use of Valuable Chinese Medicine Resources (no. 2060302), National Plant Specimen Resource Bank (no. E0117G1001), Survey of Wildlife Resources in Key Areas of Tibet (no. ZL202203601), and the International Partnership Program of CAS (no.151853KYSB20190027). Some of the GS experiments were performed at the Laboratory of Molecular Biology of Germplasm Bank of Wild Species in Southwest China, Kunming Institute of Botany, CAS. No conflict of interest is declared.

Author contributions

H.M. and T.G. designed the project. H.M., T.G., and L.-M.G. provided funds. H.M., T.G., L.-M.G., J.Y., C.Z., G.Z., B.J., J.L.P., J.C., and Z.-R.Z. provided the plant materials and generated sequence data. G.Z. and H.M. performed the data analyses. G.Z. wrote the first draft of the manuscript. H.M., G.Z., J.L.P., L.-M.G., B.J., and T.G. revised the manuscript.

Published: February 25, 2024