Phylogenomics reveals widespread hybridization and polyploidization in Henckelia (Gesneriaceae)

Abstract

Background and Aims

Hybridization has long been recognized as an important process for plant evolution and is often accompanied by polyploidization, another prominent force in generating biodiversity. Despite its pivotal importance in evolution, the actual prevalence and distribution of hybridization across the tree of life remain unclear.

Methods

We used whole-genome shotgun (WGS) sequencing and cytological data to investigate the evolutionary history of Henckelia, a large genus in the family Gesneriaceae with a high frequency of suspected hybridization and polyploidization events. We generated WGS sequencing data at about 10× coverage for 26 Chinese Henckelia species plus one Sri Lankan species. To untangle the hybridization history, we separately extracted whole plastomes and thousands of single-copy nuclear genes from the sequencing data, and reconstructed phylogenies based on both nuclear and plastid data. We also explored sources of both genealogical and cytonuclear conflicts and identified signals of hybridization and introgression within our phylogenomic dataset using several statistical methods. Additionally, to test the polyploidization history, we evaluated chromosome counts for 45 populations of the 27 Henckelia species studied.

Key Results

We obtained well-supported phylogenetic relationships using both concatenation- and coalescent-based methods. However, the nuclear phylogenies were highly inconsistent with the plastid phylogeny, and we observed intensive discordance among nuclear gene trees. Further analyses suggested that both incomplete lineage sorting and gene flow contributed to the observed cytonuclear and genealogical discordance. Our analyses of introgression and phylogenetic networks revealed a complex history of hybridization within the genus Henckelia. In addition, based on chromosome counts for 27 Henckelia species, we found independent polyploidization events occurred within Henckelia after different hybridization events.

Conclusions

Our findings demonstrated that hybridization and polyploidization are common in Henckelia. Furthermore, our results revealed that H. oblongifolia is not a member of the redefined Henckelia and they suggested several other taxonomic treatments in this genus.

INTRODUCTION

The importance of hybridization in the evolutionary process is now well established (Seehausen, 2004; Mallet, 2005; Abbott et al., 2013; Folk et al., 2018; Taylor and Larson, 2019). Usually, hybridization can be conceptually divided into two groups. In one case, hybridization transfers only part of the genome from a donor species to a recipient species without speciation, often termed introgression or introgressive hybridization (Seehausen, 2004; Mallet, 2005). In the other case, hybridization between two species can promote the formation of new species with or without a change in ploidy level (Mallet, 2007; Soltis and Soltis, 2009). However, homoploid hybrid speciation (i.e. without a change in ploidy) is generally considered to be relatively rare and difficult (Buerkle et al., 2000; Schumer et al., 2014). In most cases, hybrid speciation is accompanied by polyploidization as polyploidy results in rapid reproductive isolation from parental lineages (Soltis and Soltis, 2009). Nevertheless, either form of hybridization can immediately elevate standing genetic variation (Barrett and Schluter, 2008; Seehausen, 2013); thus, both theoretical and empirical studies support the idea that hybridization can promote an increased rate of evolution (Stelkens et al., 2014; Marques et al., 2019). For example, hybridization has often been linked to adaptive radiations, such as those in Darwin’s finches (Lamichhaney et al., 2015), East African cichlids (Irisarri et al., 2018), Heliconius butterflies (Edelman et al., 2019) and Hawaiian silverswords (Barrier et al., 1999). The significant evolutionary consequences of hybridization have prompted interest in the process by evolutionary biologists for hundreds of years (Goulet et al., 2017). Nevertheless, the frequency and distribution of hybridization across the tree of life remain unclear owing to a lack of studies of many lineages.

Hybridization can leave discordant signals across the genome, and these signals can persist over time. This genealogical signature in the genome provides a basis for the detection of hybridization using molecular sequences. However, several other biological events, such as incomplete lineage sorting (ILS) and gene duplication and loss, may result in similar genomic footprints (Degnan and Rosenberg, 2009). Furthermore, genomes often reflect combinations of these events (Cai et al., 2021; Morales-Briones et al., 2021), especially in clades that have undergone rapid radiations, presenting challenges for hybridization detection and phylogenetic inference assuming a strictly bifurcating tree. Therefore, dissecting the root causes of observed phylogenetic discordances allows for a more thorough understanding of the underlying processes and facilitates accurate parameter estimation for phylogenetic inference. Recently, methods to test and distinguish among alternative evolutionary hypotheses have been developed. These include methods to estimate phylogenetic networks while accounting for ILS and hybridization simultaneously (e.g. Yu and Nakhleh, 2015; Wen et al., 2018) and methods to detect introgression based on biallelic site frequency spectra (e.g. Durand et al., 2011; Malinsky et al., 2018). Another approach to test hybridization as a source of phylogenetic conflict is to simulate data under the multispecies coalescent (MSC), estimate a distribution of expected variation and then compare the fit of real data to alternative hypotheses (e.g. Folk et al., 2017; Olave et al., 2017; Allman et al., 2022). However, no method accounts for all factors simultaneously, and each has limitations (Hibbins and Hahn, 2022). Therefore, the combined application of multiple methods is of fundamental importance for understanding evolutionary processes and for accurate phylogenetic reconstruction (Blair and Ané, 2020).

Hybridization is particularly pervasive in plants, with ~25 % of all plant species involved in hybridization (Mallet, 2005). Notably, the family Gesneriaceae shows a high level of hybridization propensity (Whitney et al., 2010). Interspecific hybrids between gesneriads are frequently reported in both nature and glasshouse conditions (e.g. Arisum, 1964; Morley, 1976; Smith et al., 1996; Puglisi et al., 2011; Qiu et al., 2011; Afkhami-Sarvestani et al., 2012; de Araujo et al., 2021). Recent molecular studies have detected intensive introgression signals in a wide range of lineages of Gesneriaceae (e.g. de Pillon et al., 2013; de Villiers et al., 2013; Smith et al., 2017; Roberts and Roalson, 2018; Kleinkopf et al., 2019; Ke et al., 2022). However, little is known about the contribution of hybridization to the evolution of the redefined Henckelia, a genus in the family Gesneriaceae. Under the most recent circumscription (Weber et al., 2011), this genus includes all species formerly assigned to Henckelia sect. Henckelia, Chirita sect. Chirita (excluding species belonging to Damrongia) and the monotypic genus Hemiboeopsis. After its redefinition, Henckelia comprises more than 70 species, most of which are distributed in the Himalayas and Hengduan mountains, with others extending to Southeast Asia, Sri Lanka and southern India (Weber et al., 2011; Middleton et al., 2013; Janeesha and Nampy, 2020). Although the redefinition of this genus was based mainly on molecular evidence, the monophyly of the group is still uncertain because molecular data were lacking for many species in the study of Weber et al. (2011). In addition, previous cytological data have shown that the chromosome number varies among taxa in the redefined Henckelia, even though cytological studies have been performed for fewer than one-fifth of the species (Christie et al., 2012). The high karyotypic variation might reflect a complex hybridization history within this genus because hybridization often leads to karyotype rearrangement (Rieseberg, 2001; Schubert, 2007). Nevertheless, the hybridization history of this redefined genus has never been studied.

Here, we generated whole-genome shotgun (WGS) sequencing data at about 10× coverage for almost all Chinese Henckelia species and used this dataset to untangle the hybridization history within the genus. WGS sequencing can provide sequences across the whole nuclear genome as well as the organellar genome and thus is particularly well suited to resolve complicated evolutionary histories. Notably, the recently developed bioinformatic pipelines that use BLAST searches to assemble small targeted regions of the genome make the WGS sequencing approach more promising for phylogenomics, even when the sequencing depth is low (Allen et al., 2017, 2018). We separately extracted whole plastomes and thousands of single-copy nuclear genes from those sequencing data, and reconstructed phylogenies based on both nuclear and plastid data. We explored sources of both genealogical and cytonuclear conflicts using a wide range of state-of-the-art methods, and identified signals of hybridization and introgression within our phylogenomic dataset. Additionally, we evaluated the chromosome number in this genus. Combining analyses of phylogenetic discordance, introgression, phylogenetic networks and cytological data, we obtained reliable evidence that hybridization and polyploidy are widespread across the evolutionary history of Henckelia and may have played an important role in the diversification of this group.

MATERIALS AND METHODS

Sample collection, DNA extraction and sequencing

In this study, 27 Henckelia species were collected, including almost all morphological species of this genus in China (except two rare species, H. pycnantha and H. fasciculiflora, only represented by type specimens at present) and one species in Sri Lanka. In many cases, multiple geographically distant populations (one to two individuals from each population) were sampled for each species, and a total of 90 individuals from 86 populations were collected (Fig. 1 and Supplementary Data Table S1). Additionally, seven species from different genera were included as outgroup species. DNA extraction and WGS sequencing were performed following our previously described methods (Feng et al., 2020a; Ke et al., 2022). Previous empirical studies have shown that a minimum of about 10× coverage is optimal for high-quality nuclear orthologous gene assembly via WGS sequencing (Allen et al., 2017; Zhang et al., 2019; Liu et al., 2021). Therefore, in this study, all samples were sequenced with at least 10× coverage. New sequence reads were deposited in the NCBI Sequence Read Archive (SRA) and the accession numbers can be found in Supplementary Data Table S1.

Fig. 1.

Sampling locations and phylogeny of Henckelia. (A) Geographical origin of all Henckelia samples used in this study. (B) The population-level phylogeny of Henckelia inferred by ASTRAL-III based on matrix_5 (see Supplementary Data Table S2). Numbers above internal branches show local posterior probabilities, except that the full support is not shown.

Plastome assembly, annotation, alignment and phylogenetic analyses

Raw reads were cleaned using Trimmomatic 0.39 (Bolger et al., 2014) by removing low-quality bases from the beginning (LEADING:3) and end (TRAILING:3), by removing reads shorter than 36 bp (MINLEN:36), and by removing reads with a quality score of <15 using a sliding window approach (SLIDINGWINDOW:4:15). The clean data were then submitted to GetOrganelle v.1.7.5 (Jin et al., 2020) for de novo plastome assembly under default parameters. While this approach allowed most plastomes to be completely assembled, a few individuals (Supplementary Data Table S1) still had one to several gaps. The scaffolds for these samples were therefore produced as draft plastid genomes. All complete or draft plastomes were then annotated using the online program GeSeq (Tillich et al., 2017) based on the annotated genomes of two species, Primulina huaijiensis (NC_036413) and Streptocarpus teitensis (NC_037184).

DNA sequences for 85 coding genes (Supplementary Data Table S2) were extracted for chloroplast phylogenetic reconstruction. Each gene was aligned using MAFFT v.7.427 (Katoh and Standley, 2013) with the ‘-auto’ option. The alignments were trimmed using TrimAl v.1.4.15 (Capella-Gutiérrez et al., 2009) with a gap-threshold value of 0.2 (-gt 0.2). Trimmed alignments were combined in a single super-matrix to infer a maximum-likelihood (ML) phylogenetic tree using IQ-TREE v.2.1.3 (Nguyen et al., 2015). In IQ-TREE, the best-fitting partitioning scheme as well as the best-fitting models for each partition were determined using ModelFinder (Kalyaanamoorthy et al., 2017), and 1000 ultrafast bootstrap (UFBoot) replicates were used to obtain nodal support values.

Nuclear orthologous sequence assembly, alignment and phylogenetic analyses

aTRAM v.2.4.3 (Allen et al., 2018) was used to assemble protein-coding gene sequences from our clean WGS sequencing data. To obtain a suitable set of query sequences, 7177 single-copy orthologous genes were identified from one genome (H. pumila) and three transcriptomes (H. anachoreta, H. dielsii and H. tibetica) using OrthoFinder v.2.2.7 (Emms and Kelly, 2015). All four reference assemblies were generated in our other study (L.-H. Yang et al., unpubl. data) and have been submitted to NCBI (accession numbers are PRJNA882983, SRR14802384, SRR14802379 and SRR14802376, respectively). Among these 7177 single-copy genes, 5393 genes with at least 200 and no more than 600 amino acids were selected as the targeted sequences for aTRAM to assemble potentially orthologous genes. The aTRAM analysis was run for three iterations, using ABySS v.1.5 (Simpson et al., 2009) for de novo assembly and using the protein sequences of the 5393 genes from the H. pumila genome for tblastn searches. Following assembly, the exons of each resulting contig (found in the ‘filtered’ file output) were extracted using the atram_stitcher.py script provided in aTRAM. For assembled exons that were at least 50 % of the length of targeted sequences, a reciprocal best hit (RBH) check was performed by blastx searches of our assembled exons against all translated proteins of the H. pumila genome. If the best hit for an assembled locus was not the corresponding targeted gene, it was removed from the assembly. Additionally, an assembled locus was discarded if it had two or more BLAST hits on the reference genome because it might represent a paralogous gene (Zhou et al., 2022).

Three nuclear nucleotide datasets were built using loci that were retained after the RBH check, matrix_2, matrix_3 and matrix_4, with 50, 30 and 10 % missing taxa, respectively, to test the impact of missing data. Additionally, to avoid possible biases from genes with low or high substitution rates, matrix_5 was built by removing from matrix_2 genes with the 20 % slowest and 20 % fastest rates of evolution, as calculated using IQ-TREE. Each gene sequence was aligned using MAFFT v.7.427 with the ‘-auto’ option. Thereafter, the alignments were trimmed. TAPER v.1.0.0 (Zhang et al., 2021) was used to remove potential sequencing and assembly errors with default parameters. Additionally, a cleaning step was performed using TrimAl v.1.4.15 to remove spurious sequences. Sequences for each species were selected if they had at least 30 % of sites overlapping with 75 % of the rest of the sequence matrix (-seqoverlap 30 and -resoverlap 0.75 options). Finally, columns containing gaps (‘-’) or missing data (‘N’) or both missing data and gaps for more than 80 % of the present sequences were deleted using a custom script (remove_missingdata_gap.py; available in Supplementary Data).

Both concatenation- and coalescent-based methods were used to reconstruct the nuclear phylogeny of the genus Henckelia. The concatenated matrix was analysed using the same method used for the plastid ML analysis. For coalescent-based estimations, ASTRAL-III v.5.6.3 (Zhang et al., 2018) was used to estimate species trees by summarizing a set of gene trees. Thus, the best ML gene tree was estimated from the unpartitioned alignment for each gene in IQ-TREE v.2.1.3 using the best-fitting models selected by ModelFinder. Before they were submitted to ASTRAL-III, gene trees were treated using Newick Utilities (Junier and Zdobnow, 2010) to collapse the nodes with UFBoot values of <10. In ASTRAL-III, local posterior probabilities (LPPs) were used to assess the support for each node.

In addition to the population-level phylogenies (with each sample represented as a tip) estimated above, species-level phylogenies were estimated by assigning multiple individuals to the corresponding species in ASTRAL-III (Rabiee et al., 2019). However, IQ-TREE does not have this function; thus, the species-level topology was summarized manually from the population-level phylogeny. Based on our morphological observations and population-level phylogenetic results, we found that the two species complexes, H. dielsii complex (including H. briggsioides, H. dielsii and H. monantha; Supplementary Data Fig. S1) and H. ceratoscyphus complex (including H. auriculata, H. ceratoscyphus, H. multinervia and H. nanxiheensis; Fig. S2), might have taxonomic errors. Therefore, to reduce the influence of taxonomic error, all samples identified as H. multinervia, H. nanxiheensis and H. auriculata were assigned to H. ceratoscyphus, and all samples identified as H. monantha and H. briggsioides were assigned to H. dielsii.

Detection and dissection of genealogical and cytonuclear conflict

To assess the degree of genealogical concordance in our phylogenomic dataset, the gene concordance factor (gCF) and site concordance factor (sCF) were estimated (Minh et al., 2020). We focus only on the conflict signal at each node on the nuclear species-level tree. To reduce the conflict signal resulting from sequence composition heterogeneity and missing data, DNA sequences were extracted from matrix_5, in which sequences with extremely high or low rates of evolution were removed, and further discarded loci with missing data for more than 10 % of taxa. These treatments resulted in a matrix (i.e. matrix_6 in Supplementary Data Table S2) with 686 loci and at least 26 taxa and provide a basis for the detection of true conflicts. An ML gene tree was estimated for each gene of matrix_6 following the same method used before. Then, gCF and sCF were estimated using IQ-TREE with these gene trees and the concatenated sequence of matrix_6 against the ASTRAL-III topology.

Gene tree discordance can be caused by gene flow or ILS. To assess the level of ILS on each edge of the species tree, the parameter theta was estimated (Cai et al., 2021). Theta was calculated by dividing the branch lengths estimated from IQ-TREE (mutation units) by those estimated from ASTRAL-III (coalescent units) (Cai et al., 2021). To further measure the goodness-of-fit of the coalescent model with ILS explaining the gene tree discordance, the approach developed by Allman et al. (2022) was applied. Briefly, this approach involves obtaining a quartet count concordance factor (qcCF) and computing P-values to determine the fit of the gene trees to a species tree under the MSC model. qcCFs were calculated for all 686 gene trees in matrix_6 using the function quartetTreeTestInd in the R package MSCquartets (Rhodes et al., 2021). Then, these qcCFs were used to draw simplex plots in MSCquartets. In these plots, qcCF values that are distant from the model lines indicate that ILS alone cannot account for gene tree variation (Allman et al., 2022).

Two additional approaches were applied to dissect the causes of cytonuclear conflicts. First, as described by Folk et al. (2017), a coalescent simulation was used to test whether ILS alone could explain the incongruence between the plastome tree and the nuclear species tree. Briefly, 10 000 plastome trees were simulated under the coalescent model within DendroPy v.4.4.0 (Sukumaran and Holder, 2010) using the ASTRAL-III tree estimated based on matrix_5 as the model tree. All branch lengths of the ASTRAL-III tree were scaled by four to account for the haploidy and maternal inheritance of the plastome, following Folk et al. (2017). The frequency of bipartitions in simulated plastid trees was then annotated on the observed chloroplast phylogeny using Raxmal v.8.2.9 (Stamatakis, 2014). In the scenario of ILS alone, relationships in the empirical plastid tree should be present in the simulated plastid trees with a high frequency. By contrast, if gene flow is present, unique clades in the empirical plastid tree should be absent or detected at a low frequency in the simulated plastid trees. Additionally, we compared the number of extra lineages in observed and simulated plastid trees using the function DeepCoalCount_tree in PhyloNet v.3.8.0 (Wen et al., 2018). When gene flow is present, more extra lineages are expected in the observed tree relative to simulated trees (Olave et al., 2017).

Phylogenetic network inference and introgression tests

To verify the contribution of gene flow to discordance, the hybridization history in Henckelia was evaluated by inferring the phylogenetic network and testing introgression. Phylogenetic networks were compared using the maximum pseudolikelihood method (MPL; Yu and Nakhleh, 2015) implemented in PhyloNet v.3.8.0. The MPL method infers networks under the network MSC model, accounting for ILS; however, it is computationally intensive when large numbers of taxa are included. Therefore, matrix_6 was reduced to matrix_7 by removing sequences of all outgroup species other than Microchirita hamosa and discarding the genes that were not found in the outgroup species M. hamosa. Gene trees were estimated again for each locus of matrix_7. For the MPL analysis, network searches were performed, allowing for zero to six reticulations and using only nodes in the rooted ML gene trees that had UFBoot support values of at least 50.

To further explore introgression between taxa, Patterson’s D-statistic, or the ABBA–BABA statistic (Durand et al., 2011), was evaluated using Dtrios implemented in Dsuite (Malinsky et al., 2021). D-statistics were calculated for all fitted trios across the ASTRAL-III tree by fixing M. hamosa as the outgroup. Dsuite needs an input file with VCF format; thus, SNP-sites (Page et al., 2016) was used to generate a VCF file containing all single nucleotide polymorphisms of the concatenated matrix_7. The Benjamini–Hochberg correction (P < 0.05) for multiple tests was used to obtain false discovery rate-corrected P-values. However, a single gene flow event between ancestral lineages can lead to multiple significant D-statistics, and thus it is hard to disentangle these correlated signatures of introgression. Therefore, the f-branch or f_b(C) metric was used to assign gene flow to specific internal branches on a phylogeny (Malinsky et al., 2018). The f-branch statistic is a summary of the f₄ admixture ratio (or f-statistic; closely related to D-statistic; Durand et al. 2011), as described by Martin et al. (2013). The f₄-ratio was calculated using Fbranch implemented in Dsuite for all groups of species that fit the topology ((A, B), C) in the ASTRAL-III tree, with the outgroup species fixed as M. hamosa. After applying a Benjamini–Hochberg correction (P < 0.05) for multiple tests, all significant f_b(C) values were plotted using the script dtools.py supplied in Dsuite.

Chromosome counting

Hybridization often leads to the formation of allopolyploidy. Accordingly, the ploidy of Henckelia species was evaluated by counting chromosomes. P for chromosome studies were collected from the field and subsequently cultivated in the South China National Botanical Garden or Gesneriad Conservation Center of China. Cytological experiments were carried out following our previously described procedures (Yang et al., 2020). Multiple chromosome counts were obtained for each specimen and species, when possible.

RESULTS

DNA sequencing and characteristics of the datasets

WGS sequencing produced a mean of 69 636 679 reads (range, 64 420 214–89 529 124) per sample after quality control (Supplementary Data Table S1). Average sequencing coverage was 12× (range, 11–15×) based on the 897.04-Mb genome of H. pumila.

Using these sequence reads, we obtained 73 complete chloroplast genomes and 24 draft plastomes of Henckelia and several outgroup species (Supplementary Data Table S1). We successfully extracted sequences of almost all 85 protein-coding genes from both complete and draft plastomes, although the draft plastomes had one to several gaps. The concatenated chloroplast (cp)DNA matrix (matrix_1 in Table S2) had an aligned length of 74 057 bp with only 0.7 % missing data. The length and proportion of missing data for each plastid gene are presented in Table S3.

Of the 5393 targeted nuclear genes, aTRAM assembled a mean of 5249 candidate orthologues (range, 4800–5357; Supplementary Data Table S1). Among these, lengths of 4537 loci (range, 2484–5244), on average, were at least 50 % of the length of the reference (Table S1). After data filtering to remove potentially non-orthologous contigs and gene sets with more than 50 % missing taxa, 1976 putative orthologues (matrix_2) were retained for phylogenetic analysis. To test the impact of missing data, we built another two nuclear datasets with no more than 30 % (matrix_3) and 10 % (matrix_4) missing taxa. In addition, we built matrix_5 by removing genes with the 20 % slowest and 20 % fastest rates of evolution in matrix_2 to evaluate biases from extreme evolutionary rates. The characteristics of the concatenated alignment for each nuclear dataset are summarized in Table S2.

Phylogenetic relationships and conflicts in Henckelia

The phylogeny inferred by IQ-TREE based on matrix_2 was the best supported, with only one node within the H. dielsii complex supported by a UFBoot value of 93 (Supplementary Data Fig. S3), while other phylogenies possessed several moderately supported nodes (LPP < 0.95 or UFBoot < 95; Fig. 1 and Figs S4–S7). The tree topologies showed different degrees of conflict depending on the input matrixes and estimation methods (Fig. S8). Despite these conflicts, in all phylogenetic analyses based on either plastid or different nuclear matrixes, all Henckelia species formed a monophyletic group with 100 % support, except for H. oblongifolia, assigned to the outgroup clade (Fig. 1 and Figs S3–S7). In addition, we sampled multiple individuals per species, providing a basis for evaluating the monophyly of current morphological species. As expected, most species were recovered as monophyletic with full support when data for multiple individuals were available, except for the two species complexes (i.e. the H. ceratoscyphus complex and H. dielsii complex; Fig. 1). In the H. ceratoscyphus complex, the current H. ceratoscyphus was paraphyletic; however, all samples of this species together with H. multinervia, H. nanxiheensis and H. auriculata formed a fully supported monophyletic group in all phylogenies (Fig. 1 and Figs S3–S7). Similarly, in the H. dielsii complex, H. briggsioides and H. monantha clustered together, and H. dielsii was paraphyletic; however, this species complex was a monophyletic group with full support in all phylogenies (Fig. 1 and Figs S3–S7).

After assigning multiple individuals to the corresponding species, we found that the ML topologies inferred by IQ-TREE based on different nuclear matrixes were identical and largely consistent with the topologies inferred by ASTRAL-III (Supplementary Data Fig. S9). The main conflicts between the ML and ASTRAL-III topologies involved the placements of H. medogensis and H. walkerae. In addition, the ASTRAL-III topologies inferred from different nuclear matrixes differed in the placement of H. lallanii. However, the plastid backbone phylogeny was in substantial conflict with the nuclear species tree, particularly in the placements of H. connata, H. shuii, H. fruticola, H. inaequalifolia, H. lachenensis and H. dielsii (Fig. 2A). Despite massive cytonuclear incongruences, both plastid and nuclear phylogenies could be divided into five clades, referred to as Clades I–V (Fig. 2A). Clades I, III and IV were each represented by one species (i.e. H. longisepala, H. medogensis and H. walkerae, respectively). Clade II included three species mainly distributed in the eastern Himalaya Mountains, while Clade V included most Henckelia species found in southern China.

Fig. 2.

Detecting and dissecting cytonuclear discordance. (A) Comparison of the nuclear (left) and chloroplast (right) phylogenetic trees and summary of the plastome tree simulation results. The nuclear phylogeny was estimated by ASTRAL-III based on matrix_5 (see Supplementary Data Table S2), and the chloroplast phylogenetic tree was estimated by IQ-TREE based on matrix_1 (see Table S2). Different colour dots at each node indicate different support values (i.e. local posterior probability for the ASTRAL-III tree and ultrafast bootstrap value for the IQ-TREE tree), defined at the bottom right. Dashed lines connect the same species in the two trees. Red internal branches indicate unique clades of the chloroplast topology compared to the nuclear species tree. Numbers above internal branches show the clade frequency (%) of the simulated plastid genes. (B) Histogram of the number of extra lineages required for gene trees simulated under a strict coalescent model. Green bars show the distribution of simulated plastome trees. The vertical red line represents the number of extra lineages observed in plastid gene trees. LPP, local posterior probability.

ILS alone cannot explain cytonuclear or genealogical discordance

To dissect the cause of observed conflicts between nuclear and chloroplast trees, we applied two approaches. A coalescent simulation revealed all conflicting plastid bipartition frequencies at or near zero in the 10 000 simulated plastid trees, except that 31 % of simulated trees agreed with a weakly supported sister relationship between H. lallanii and H. infundibuliformis (Fig. 2A). This result thus favoured a hybridization scenario as the cause of most of the observed cytonuclear incongruences. The hybridization scenario was also supported by an analysis based on the approach of Olave et al. (2017), which showed that the number of extra lineages (19) in observed plastid trees falls within the upper tail of the distribution (0–21 with a median value of 9) of extra lineages for simulated plastid trees generated under a strict coalescent process (Fig. 2B).

We also assessed the strength and causes of genealogical discordance. The gCF and sCF metrics showed substantial conflicts at most internodes across the species tree. We found that 18 of 27 internodes show gCF values of <50 % (Fig. 3A), most of which were distributed across the Henckelia clade. The sCF values for most nodes were relatively high, with an average of 48.28 %. If there is no consistent information in an alignment, we expect a roughly equal proportion of sites supporting each of the possible relationships of a quartet, leading to an sCF value of ~33 %. The relatively high sCF values detected here indicate that the high levels of conflict mostly resulted from biological processes (e.g. hybridization and/or ILS), rather than a lack of informative characters. Therefore, we further applied two methods to assess the cause of the observed gene tree discordance. Theta revealed the different extent of ILS at different edges across the species tree (Fig. 3B), with values ranging from 0.006 to 0.526. ILS was highest on the branch leading to H. urticifolia and H. infundibuliformis and lowest on the branch leading to the whole Henckelia clade (excluding H. longisepala). In addition, we used the pipeline of Allman et al. (2022) to check the goodness-of-fit for how well an ILS model explains the gene tree discordance. Both T3 and T1 simplex plots showed that most empirical quartets lie close to the model lines (Fig. 3C), while a number of points were far from the model lines (Fig. 3C). Applying the Holm–Bonferroni method (P < 0.05), 4.59 and 4.69 % test results for the T3 and T1 models, respectively, rejected the MSC hypothesis. This suggests that ILS alone cannot explain the gene tree discordance, and more complicated models, such as those including gene flow, are needed.

Fig. 3.

Detecting and dissecting genealogical discordance. (A) Degree of conflict between nuclear gene trees and the species tree at each node, as evaluated by gene concordance factor (gCF) and site concordance factor (sCF) values. gCF and sCF (expressed as percentages) are given above branches in the order gCF/sCF. Different components of gCF and sCF are shown as pie charts above and below each node, respectively. (B) Theta parameter estimation. (C) Simplex plots of quartet count concordance factors for the 686 gene trees.

Complex patterns of hybridization and introgression within Henckelia

The network with five reticulation instances was the most reliable based on the ranking of pseudologlikelihood values. This optimal network recovered two lineages (i.e. Clade II + III and Clade V), H. connata and H. dielsii as descendants of different hybridization events, and also included a hybrid edge connecting to non-terminal branches (Fig. 4). However, pseudologlikelihood values for the networks with four (Loglik = −576 503.20) and six (Loglik = −576 437.18) reticulations were very similar to those of the best network (Loglik = −576 402.15). Therefore, we considered these two suboptimal networks with four and six reticulations. We found that the positions of the reticulations varied across these three networks; however, the hybrid origin of H. connata was consistently observed, with inheritance probabilities near 0.5 in all three networks.

Fig. 4.

Species networks inferred from the pseudolikelihood method with a maximum of four to six reticulations. Clade designations and numbers correspond to those shown in Fig. 2.

We further identified introgression signals against the ASTRAL-III species tree (Fig. 3A) using D-statistics. Approximately 42 % (553 out of 1330) of the D-statistics in our dataset were significantly different from zero (Bonferroni-corrected P < 0.05; Supplementary Data Table S4; Fig. 5), providing strong support for gene flow between different extant species pairs. These 553 combinations involved almost all Henckelia species included in the analysis. We detected introgression between closely related species as well as distantly related species, indicating that gene flow might have occurred throughout the evolutionary history of Henckelia. In total, 61 f_b(C) scores were significantly elevated at a Bonferroni-corrected P < 0.05, and 10 of the 40 branches in the phylogeny showed significant excess allele sharing with at least one other species C (Fig. S10). For example, the highest percentages of gene flow were detected between H. inaequalifolia and H. fruticola [f_b(C) = 20.97 %] and between H. inaequalifolia and H. dielsii [(f_b(C) = 20.65 %]. We also detected several ancestral introgression events connecting different hierarchical branches within Henckelia, such as the clade including H. inaequalifolia, H. xinpingensis and H. tibetica and the clade including H. dielsii, H. lachenensis, H. grandifolia and H. speciosa.

Fig. 5.

Heatmap depicting introgression. D-statistics were calculated using the ASTRAL-III tree estimated from matrix_5 (see Supplementary Data Table S2). The most significant D-statistic found for two species is indicated by heatmap cells. Red and blue indicate high and low values, respectively. Increasing colour saturation indicates greater significance based on P-values, and white cells indicate a lack of information.

Chromosome number

We evaluated chromosome counts for 45 populations of the studied 27 Henckelia species (Supplementary Data Fig. S11 and Table S5). For 15 species, these are the first reported chromosome counts, i.e. H. longisepala (2n = 16), H. lallanii (2n = 34), H. infundibuliformis (2n = 34), H. medogensis (2n = 34), H. ceratoscyphus (2n = 18), H. connata (2n = 34), H. forrestii (2n = 18), H. lachenensis (2n = 18), H. shuii (2n = 34), H. fruticola (2n = 34), H. puerensis (2n = 18), H. inaequalifolia (2n = 18), H. xinpingensis (2n = 18), H. tibetica (2n = 18) and H. briggsioides (2n = 18).

DISCUSSION

Phylogenomic perspectives on the evolutionary relationships within Henckelia

Species-level relationships within the redefined Henckelia have not been resolved, with only a few studies including very limited Henckelia species and a small number of gene regions (e.g. ITS and trnL–trnF; Weber et al., 2011; Roalson and Roberts, 2016; Middleton et al., 2018; Li et al., 2022). Here, we used WGS sequencing to generate massive amounts of genomic data to reconstruct the phylogenetic relationships among all Chinese Henckelia species. We successfully recovered thousands of nuclear single-copy genes and whole plastome sequences with about 10× coverage. Using these datasets, we verified the monophyly of most current morphological species and generally obtained highly supported relationships (LPP > 0.95 or UFBoot > 95). Specifically, our phylogenies revealed that H. longisepala is sister to the rest of the genus, consistent with some previous results (Roalson and Roberts, 2016; Middleton et al., 2018) but in conflict with the results of Weber et al. (2011). The early divergence of H. longisepala in Henckelia is consistent with its distinct morphological characteristics (i.e. spathulate calyx lobes with a rounded apex, not found in other Henckelia species). In addition, we observed several other subclades that are well supported by both molecular data and morphological characters. For example, the sister relationships between H. pumila and the H. ceratoscyphus complex and between H. xinpingensis and H. tibetica were consistently recovered in all phylogenies with full support (Fig. 2 and Supplementary Data Fig. S9). Both H. pumila and the H. ceratoscyphus complex were characterized by horn-like calyx lobes, while both H. xinpingensis and H. tibetica were characterized by slender stolons. Moreover, we identified several clades corresponding to geographical distributions. For example, Clades II and III included three and one species, respectively, that were mainly distributed in the eastern Himalaya Mountains; Clade IV included one species distributed in Sri Lanka; and Clade V included most species in southern China. Overall, our phylogenies show broad-scale clustering by geography or morphology.

Our phylogenomic analyses placed H. oblongifolia outside the redefined Henckelia. This species was recently transferred from Chirita to Henckelia during the remodelling of Chirita and its associated genera based on molecular data; however, molecular data for H. oblongifolia were missing in that previous study (Weber et al., 2011). In our study, we collected three samples of H. oblongifolia including one (YLH1046) from its type locality, Chittagong, Bangladesh, and two (YLH1088 and YLH1092) from the southeastern Himalaya. Unexpectedly, while all three accessions formed a fully supported monophyletic group, they were assigned to the outgroup clade in both our plastid and nuclear phylogenies (Fig. 1 and Supplementary Data Figs S3–S7). These results suggest that H. oblongifolia is not a member of the redefined Henckelia. In fact, H. oblongifolia was once published as a new species in Lysionotus (i.e. Lysionotus bijantiae; Borah and Joe, 2018). Recently, Cai et al. (2020) found that L. bijantiae is H. oblongifolia, and the name L. bijantiae was thus treated as a synonym under H. oblongifolia. However, although Cai et al. (2020) reported this taxonomic mistake, the real identity of H. oblongifolia is still uncertain according to our current results. Further studies are needed to accurately determine the generic placement of this species.

Pervasive hybridization and polyploidization in the evolutionary history of Henckelia

We detected a strong signature of hybridization in the genomes of Henckelia species. Following hybridization, the chromosome numbers in several species doubled, indicating that allopolyploidization occurred in this plant genus. Below, we discuss several lines of evidence for the hybridization and polyploidization history within Henckelia.

First, detecting and dissecting the root causes of genealogical and cytonuclear conflicts in our phylogenomic dataset provide preliminary evidence of reticulate evolution in Henckelia. Genome-scale data offer an excellent opportunity for resolving relationships across the tree of life; however, these data also open the door to phylogenetic conflicts, which usually arise from biological processes, such as hybridization, ILS, or gene duplication and loss (Jeffroy et al., 2006; Degnan and Rosenberg, 2009; Cai et al., 2021; Morales-Briones et al., 2021). In this study, we observed strong discordances between the plastid and nuclear species trees as well as extensive conflicts among nuclear gene trees. Our stringent data filtering aimed at improving the signal-to-noise ratio should have minimized the effect of systematic errors on conflicts. In addition, bias caused by paralogues should not be a serious problem in our study because we selected strict single-copy genes using two steps: an orthogroup analysis and RBH check. ILS was once considered the most prominent source of phylogenetic conflicts because it is an unavoidable consequence of neutral population processes (Degnan and Rosenberg, 2009). However, we detected only two nodes with theta values >0.1, indicating a high level of ILS (Cai et al., 2021). Most nodes showed strong genealogical discordance (i.e. gCF < 50; Fig. 3A) but a low level of ILS (theta < 0.01; Fig. 3B). These results indicate that ILS cannot fully account for the observed genealogical discordances. This conclusion was further supported by an analysis of qcCF values (Fig. 3C). We applied two state-of-the-art pipelines [i.e. those of Folk et al. (2017) and Olave et al. (2017)] to dissect the root causes of cytonuclear discordance. Both of these analyses (Fig. 2) suggested that gene flow contributed to the observed discordances. Therefore, our results are in agreement with the increasing number of phylogenomic studies showing that gene flow might represent a common and important source of observed phylogenetic conflicts (Smith et al., 2020; Cai et al., 2021; Morales-Briones et al., 2021).

Second, phylogenetic network and introgression analyses revealed a complicated pattern of hybridization within Henckelia, involving almost all species studied here. We found recurrent reticulation events at both deep and shallow levels across the phylogeny (Figs 4 and 5 and Supplementary Data Fig. S10), indicating that admixture events were common throughout the evolutionary history of Henckelia. In particular, the ancient hybridization events detected in all D-statistic, f-branch statistic and PhyloNet analyses might have a profound effect on the diversification of this group (Folk et al., 2018; Marques et al., 2019). However, despite evidence for widespread hybridization within Henckelia, the exact parents and clades involved in the hybridization events should be interpreted with caution owing to the limited taxon sampling from the genus, which has more than 70 species. We obtained similar pseudologlikelihood values from three different PhyloNet analyses (Fig. 4), indicating no overwhelming support for an optimal network. The uncertainty of the PhyloNet result might result from the intrinsic nature of the MPL method used in our study. While the MPL method can be applied to large datasets, it is a heuristic method based only on the summaries of the gene tree topologies (Yu and Nakhleh, 2015). This heuristic method often suffers from unidentifiability issues when many different combinations of network parameters are equally consistent with the real data (Pardi and Scornavacca, 2015; Hibbins and Hahn, 2022). This is a particular issue in systems with a complicated hybridization history (Pardi and Scornavacca, 2015). Using the full-likelihood method with more complex models might resolve the unidentifiability problems (Yang and Flouri, 2022). However, the computational costs make this method unrealistic for large datasets, like that used in our study. This suggests that the complex reticulate evolution within Henckelia cannot be accurately inferred by established methods. Furthermore, the difficulty in determining the number and position of reticulation in our study could be explained by the effects of ‘ghost’ lineages (i.e. extinct, unknown or unsampled lineages) (Marcussen et al., 2015; Tricou et al., 2022), particularly as only about one-third of extant species were sampled. Notably, the networks inferred by PhyloNet included edges connecting non-terminal branches (Fig. 4), indicating reticulations with ‘ghost’ taxa. Similarly, the f-branch statistic identified multiple gene flow events involving ancestral or unsampled taxa (Supplementary Data Fig. S10). To more precisely elucidate which species are of hybrid origin and which parental species gave rise to these hybrids, further sampling is needed. Nevertheless, the number of hybridization events should increase with further sampling.

Finally, karyotype variation in Henckelia might reflect the complex hybridization history within this genus. Hybridization is often accompanied by polyploidization and leads to allopolyploidy (Mallet, 2007; Soltis and Soltis, 2009). To test the polyploidization history of the redefined Henckelia, we counted the chromosomes of 45 populations belonging to 27 species (Supplementary Data Fig. S11 and Table S5). Previously published cytological data showed that the redefined Henckelia is highly heterogeneous in chromosome number, including 2n = 8, 18, 20, 22, 28, 32, 34 and 54 (reviewed in Christie et al., 2012). Our results further expanded the diversity of karyotypes in this genus to include 2n = 16 of H. longisepala. However, the highly heterogeneous chromosome number in Henckelia might, at least in part, be a product of error resulting from the difficulty of chromosome counting in Gesneriaceae (Möller and Kiehn, 2003) or taxonomic error. For example, we demonstrated that H. oblongifolia with 2n = 32 is not a true member of Henckelia. Christie et al. (2012) regarded 2n = 18 as the basic chromosome number in Henckelia. However, our current phylogenies identified H. longisepala with 2n = 16 as the basal species in Henckelia. In spite of this, we agree with Christie et al. (2012) that 2n = 18 probably represents the basic chromosome number in this genus for three reasons: (1) taxon sampling in our study was incomplete, (2) the ancestral karyotype of the Old World Gesneriaceae (i.e. subfamily Didymocarpoideae) is also 2n = 18 (Wang et al., 1998; Möller and Kiehn, 2003) and (3) 2n = 18 is the most prevalent (18 out of 33 species) karyotype in the redefined Henckelia. Therefore, 2n = 16 in Henckelia might be a derived karyotype from 2n = 18 by chromosome fusion, and the second most prevalent karyotype, 2n = 34, might be a polyploid in Henckelia. Seven species possessed this karyotype. Interestingly, all seven of these species were involved in different reticulations in the phylogenetic network analysis (Fig. 4), indicating that they are allopolyploids with several independent origins. Several other species were reported to have chromosome numbers of 2n = 28, 32 and 54 (reviewed in Christie et al., 2012). If these cytological results are accurate, these species could also be polyploids in Henckelia. Therefore, both ours and previous results revealed that polyploidization is a common phenomenon during the diversification of the genus Henckelia.

The high prevalence of hybridization in Henckelia is not surprising given the high frequency of this process in the family Gesneriaceae (Arisum, 1964; Morley, 1976; Smith et al., 1996; Puglisi et al., 2011; Qiu et al., 2011; Afkhami-Sarvestani et al., 2012; de Pillon et al., 2013; de Villiers et al., 2013; Smith et al., 2017; Roberts and Roalson, 2018; Kleinkopf et al., 2019; de Araujo et al., 2021). Extensive hybridization in Henckelia can probably be explained by a lack of complete reproductive isolation in Gesneriaceae (Zhang et al., 2017; Ramírez-Aguirre et al., 2019; Feng et al., 2020b). Therefore, artificial hybrids can be easily produced using different parental species, even from different genera (Lü et al., 2017). In addition, unlike most other genera of Gesneriaceae, which often have a restricted distribution, most species in Henckelia possess a relatively wide geographical distribution, providing more opportunities for contact and interbreeding.

It is worth noting that we detected introgression signals between species whose current geographical distributions have no overlaps. For example, the three endemic species (i.e. H. infundibuliformis, H. medogensis and H. lallanii) from the eastern Himalaya show different extents of introgression with species restricted to southwestern China (e.g. H. xinpingensis, H. inaequalifolia and H. fruticola; Fig. 5). This result indicates that the distributions of these species might have overlaps during a historical period with ecological optimum, or, alternatively, these introgression signals might be inherited from their ancestors whose distributions are overlapped. In particular, the situation that the descendants of an ancestral hybridization are allopatric and highly disjunctive in modern times has been confirmed in several other plant lineages (Klein and Kadereit, 2016; Marques et al., 2016; Folk et al., 2017, 2018).

Taxonomic implications

Our sampling of multiple populations provides perspective on the current species circumscription in Henckelia. Taking the phylogenetic and cytological results and morphological observations together, we suggest treating the H. ceratoscyphus species complex (including H. ceratoscyphus, H. nanxiheensis, H. multinervia and H. auriculata) and the H. dielsii complex (including H. briggsioides, H. dielsii and H. monantha) as one species each, although we acknowledge that the recircumscribed H. ceratoscyphus will be polymorphic. A detailed discussion of these taxonomic suggestions is provided in the Supplementary Data Note S1.

CONCLUSIONS

We have provided the first phylogenomic perspective on the evolutionary relationships among all Chinese Henckelia species and demonstrated the monophyly of most morphological species. We observed intensive discordances between the plastid and nuclear trees as well as conflicts among nuclear gene trees. Based on analyses of the root causes of observed discordances, introgression, phylogenetic networks and cytological data, we provide strong evidence that hybridization and polyploidization were widespread in the evolutionary history of Henckelia. However, more comprehensive taxon sampling is needed to present a complete phylogenetic framework for this genus and for a more precise elucidation of the reticulate evolutionary history. In addition, the integrative and stepwise approach taken here provides an empirical example of using low-coverage WGS sequencing data to extract single-copy genes for plant phylogenomics.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1. List of species, voucher information, and summary of whole genome sequencing and aTRAM assemblies. Table S2. Characteristics of all datasets used in this study after several steps of trimming. Table S3. Lengths and proportion of missing data of each of the 85 chloroplast protein coding genes. Table S4. Detailed results of the D-statistic and f-statistic tests. Table S5. Chromosome number of Henckelia species studied in this study. Fig. S1. Morphological variation of the Henckelia dielsii complex. Fig. S2. Morphological variation of the Henckelia ceratoscyphus complex. Fig. S3. Phylogenies of Henckelia inferred from IQTREE and ASTRAL-III based on matrix_2. Fig. S4. Phylogenies of Henckelia inferred from IQTREE and ASTRAL-III based on matrix_3. Fig. S5. Phylogenies of Henckelia inferred from IQTREE and ASTRAL-III based on matrix_4. Fig. S6. Phylogenies of Henckelia inferred from IQTREE and ASTRAL-III based on matrix_5. Fig. S7. Phylogeny of Henckelia inferred from IQTREE based on matrix_1. Fig. S8. Robinson–Foulds distances among different phylogenies estimated from different methods and datasets. Fig. S9. Topology comparisons among all nuclear species-level phylogenies. Fig. S10. Heatmap depicting the gene flow among phylogeny branches estimated with the f-branch statistic. Fig. S11. Somatic metaphase chromosome numbers of Henckelia species studied here. Note S1. Taxonomic treatments.

CONFLICT OF INTEREST

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

FUNDING

This work was supported by the National Natural Science Foundation of China (31900178), Biological Resources Programme, Chinese Academy of Sciences (KFJ-BRP-007-004) and the foundation from South China Botanical Garden, Chinese Academy of Sciences (QNXM-07).