Iterative allogamy–autogamy transitions drive actual and incipient speciation during the ongoing evolutionary radiation within the orchid genus Epipactis (Orchidaceae)

Abstract

Background and Aims

The terrestrial orchid genus Epipactis has become a model system for the study of speciation via transitions from allogamy to autogamy, but close phylogenetic relationships have proven difficult to resolve through Sanger sequencing.

Methods

We analysed with restriction site-associated sequencing (RAD-seq) 108 plants representing 29 named taxa that together span the genus, focusing on section Epipactis. Our filtered matrix of 12 543 single nucleotide polymorphisms was used to generate an unrooted network and a rooted, well-supported likelihood tree. We further inferred genetic structure through a co-ancestry heat map and admixture analysis, and estimated inbreeding coefficients per sample.

Key Results

The 27 named taxa of the ingroup were resolved as 11 genuine, geographically widespread species: four dominantly allogamous and seven dominantly autogamous. A single comparatively allogamous species, E. helleborine, is the direct ancestor of most of the remaining species, though one of the derived autogams has generated one further autogamous species. An assessment of shared ancestry suggested only sporadic hybridization between the re-circumscribed species. Taxa with the greatest inclination towards autogamy show less, if any, admixture, whereas the gene pools of more allogamous species contain a mixture alleles found in the autogams.

Conclusions

This clade is presently undergoing an evolutionary radiation driven by a wide spectrum of genotypic, phenotypic and environmental factors. Epipactis helleborine has also frequently generated many local variants showing inclinations toward autogamy (and occasionally cleistogamy), best viewed as incipient speciation from within the genetic background provided by E. helleborine, which thus becomes an example of a convincingly paraphyletic species. Autogams are often as widespread and ecologically successful as allogams.

Introduction

Self-fertilization (‘selfing’) is an important evolutionary mechanism that can provide selective leverage under challenging ecological conditions via transmission advantage and reproductive assurance (Stebbins, 1957; Takebayashi and Morrell, 2001; Igic and Busch, 2013; Wright et al., 2013). The Eurasian genus Epipactis Zinn. (tribe Neottieae, family Orchidaceae) provides an important example of a clade widely regarded as comparatively rich in selfing species (Claessens and Kleynen, 2011), its hypothesized elevated speciation rate having been attributed to autogamy-triggered isolation (Richards, 1982; Robatsch, 1995; Pridgeon and Light, 2005).

Genera Orchidacearum considered the genus to contain approx. 15 species (Wood, 2005), where – according to a molecular phylogenetic study based on nuclear ribosomal internal transcribed spacer (nrITS) and four plastid regions (Bateman et al., 2005) – half of these species constitute a paraphyletic group (the so-called ‘section Arthrochilum’) relative to the monophyletic section Epipactis (more commonly known as the Epipactis helleborine alliance/aggregate). Section Epipactis has a natural distribution that is confined to Eurasia, though the most widespread species, E. helleborine sensu stricto (s.s.), has become an increasingly common occupier of anthropogenic habitats (Rewicz et al., 2018), to the extent that it is viewed as an invasive adventive in North America (Squirrell et al., 2001; Light and MacConaill, 2006; Kolanowska, 2013). Within section Epipactis, modest levels of morphological variation exist among sometimes geographically restricted ‘swarms’ (Fig. 1) – variation that some taxonomists have converted into as many as 65 species (Delforge, 2016) in Europe and Asia Minor, at least partly by invoking the selfing inclination of the genus.

Fig. 1.

Flowers of 28 Epipactis taxa representing the morphological variation displayed by the studied genus. Taxa with a verified specific rank in our analyses are preceded by ‘E.’; the remaining taxa are best viewed as infraspecific, but further research using modern methods is needed before the most appropriate infraspecific rank for each of these taxa can be determined. Taxa: A, E. veratrifolia, Cyprus; B, E. palustris, Hungary; C, E. atrorubens, Hungary; D, E. phyllanthes, Belgium; E, persica, Turkey; F, exilis, Bulgaria; G, E. purpurata, Hungary; H, pseudopurpurata, Hungary; I, E. leptochila, Hungary; J, peitzii, Hungary; K, neglecta, Hungary; L, futakii, Hungary; M, E. greuteri, Greece; N, nauosaensis, Greece; O, densifolia, Turkey; P, E. pontica, Hungary; Q, E. muelleri, Hungary; R, voethii, Hungary; S, neerlandica, Denmark; T, renzii, Denmark; U, distans, Serbia; V, E. helleborine, Hungary; W, minor, Hungary; X, mecsekensis, Hungary; Y, tallosii, Hungary; Z, nordeniorum, Hungary; 1, albensis, Hungary; 2, bugacensis, Hungary. Flowers are shown at the same scale. Images: A–C, E, G, I–L, O, Q, R and U–2 by A. Molnár V.; D, F, H, M, N, P, S and T by M. Óvári.

Members of section Epipactis collectively occupy a wide range of soils and habitat types, and have become a model system for the study of tritrophic mycorrhizal interactions, wherein the orchid parasitizes adjacent trees via plumbing provided by fungal intermediaries (Bidartondo et al., 2004). The alliance has also enabled studies of the active transition from autotrophic to facultative mycoheterotrophic nutrition (Selosse et al., 2004; Julou et al., 2005; Schiebold et al., 2017). More importantly in the context of the present study, compared with members of ‘section Arthrochilum’, those species attributed to section Epipactis differ among each other less markedly in morphology (Fig. 1) and show negligible divergence in both nrITS and plastid regions (Bateman et al., 2005; Hollingsworth et al., 2006; Drouzas et al., 2017; Zhou and Jin, 2018). Nonetheless, collectively they exhibit the full range of reproductive modes from entomophilous allogamy approximating Hardy–Weinberg equilibrium through enhanced geitonogamy and facultative autogamy to near-obligate autogamy by means of cleistogamy (Richards, 1986; Robatsch, 1995; Ehlers and Pedersen, 2000; Squirrell et al., 2002; Pridgeon and Light, 2005; Hollingsworth et al., 2006; Tranchida-Lombardo et al., 2011). Thus, the section may provide an example of the evolution of a mating system in which species could in theory evolve towards either complete selfing or complete outcrossing, depending on the balance between automatic selection promoting self-fertilization and potential costs of inbreeding such as inbreeding depression and reduced adaptability (Charlesworth and Charlesworth, 1987; Goodwillie et al., 2005; Brys and Jacquemyn, 2016).

Further studies involving population-level allozyme and sequencing analyses (Harris and Abbott, 1997; Ehlers and Pedersen, 2000; Pedersen and Ehlers, 2000; Squirrell et al., 2002; Bateman et al., 2005; Hollingsworth et al., 2006; Tranchida-Lombardo et al., 2011) strongly suggested that the near-obligate autogams had originated iteratively across Europe from within the more widespread, dominantly allogamous E. helleborine s.s., leading section Epipactis to become a textbook case of speciation via the transition from allogamy to autogamy (Proctor and Yeo, 1973; Richards, 1986). Some authors further suggested that this transition in breeding system had a reliable polarity, allogams never arising from autogams (Richards, 1982), and that autogamous Epipactis could therefore be viewed as evolutionary dead-ends (Hollingsworth et al., 2006; Bateman, 2009, 2012a, b; Claessens and Kleynen, 2016). Some authors also argued that a clade containing substantial numbers of putative species, yet possessing so little molecular divergence in otherwise polymorphic regions that an origin >1 Myr ago is unlikely, offers a particularly high probability that it is currently evolving at a rate sufficiently high to constitute an active evolutionary radiation (sensuBateman, 1999).

There exists interest well beyond the realm of orchid studies in resolving the general questions of (1) whether transitions from allogamy to autogamy can be iterative (e.g. Squirrell et al., 2002); (2) whether autogams can speciate to form further autogams (Stebbins, 1970; Takebayashi and Morrell, 2001; Bateman, 2009; Igic and Busch, 2013; Wright et al., 2013); and (3) whether autogams might even undergo ‘reverse speciation’ to form novel allogams. Capturing such transitions in the midst of a genuine evolutionary diversification would be an additional bonus (e.g. Bateman et al., 1998; Bateman, 1999).

Recently developed molecular techniques collectively termed next-generation sequencing (NGS; Schloss, 2008; Olson et al., 2016) offer the opportunity to bring to bear on such questions far larger numbers of genome-wide, phylogenetically informative characters and thereby improve resolution of phylogenetic relationships among such controversial taxa as section Epipactis (Kircher and Kelso, 2010; Harrison and Kidner, 2011). Among these techniques, restriction site-associated sequencing (RAD-seq) utilizes the Illumina high-throughput sequencing technique (Baird et al., 2008) to screen hundreds of thousands of base pairs of genomic DNA for informative sites [single nucleotide polymorphisms (SNPs)] sampled throughout the whole genome, yielding incomparably more data than conventional candidate gene sequencing (Rubin et al., 2012; Pante et al., 2015). RAD-seq is increasingly being employed in studies of the plant kingdom (e.g. Eaton and Ree, 2013; Paun et al., 2016; Ren et al., 2017; Trucchi et al., 2017; Heckenhauer et al., 2018). Having applied the RAD-seq technique to the other orchid genera (Bateman et al., 2018; Brandrud et al., 2019), we have now turned our attention to Epipactis section Epipactis. We sampled the genus extensively across Europe and (to a lesser degree) Asia Minor, initially within a framework provided by the divisive taxonomy of Delforge (2016). Here, we selected a sub-set of samples for detailed analysis that gave greater emphasis than previous studies to material from the less intensively researched eastern European taxa. Our main objectives in pursuing the present study were as follows: (1) to use well-founded molecular estimates of monophyly and disparity as a guide to determine which of the many named taxa have the strongest support to be recognized as bona fide species; (2) to determine with greater confidence the relationships among those species; (3) to explore any phylogenetic patterns that can be discerned from the putative universal ancestor of this taxonomic diversity, E. helleborine; and (4) to combine those phylogenetic insights with knowledge of the reproductive biology of taxa in order to test previous hypotheses relevant to the broader discipline of evolutionary biology. More specifically, we explore whether (a) few or many autogamous taxa have arisen iteratively across Europe from within the widespread allogam E. helleborine s.s.; (b) the transition from an allogamous to an autogamous lineage is irreversible, the converse transition never taking place; and (c) autogams are evolutionary dead-ends, no autogam ever giving rise to further autogamous species.

We conclude by speculating on the geographic regions of origin of the derived species and whether section Epipactis has indeed acquired the rare scientific credentials necessary to be viewed as undergoing an active evolutionary radiation.

MATERIALS AND METHODS

Sampling

Previous molecular phylogenetic analyses (notably Bateman et al., 2005) were used to select as outgroups from within the apparently paraphyletic section Arthrochilum two successively diverging species, E. veratrifolia and E. palustris. Within the ingroup (i.e. section Epipactis), the 29 samples selected of E. helleborine s.s. – the putative progenitor of the autogamous species – extended west–east from the coast of Wales to the eastern shore of the Black Sea (Fig. 2). Within this area, we selected a further 77 ingroup samples, together encompassing 26 named taxa. Each taxon was represented by between one and six samples (mean 2.9 ± 1.4), with a concentration of sampling in Eastern Europe. Taxa were chosen to encompass the full range of breeding systems from putative obligate allogams to putative obligate autogams (Table 1). Field-collected samples of leaf tissue were immediately immersed in sachets of fine-grained silica gel, while an open flower from each plant was placed in 96 % ethanol as a voucher to be deposited in the herbarium of the University of Debrecen (DE).

Fig. 2.

Collecting locations of ingroup samples. Taxon codes follow Table 1. The map layer was extracted from GADM version 1.0 (available from https://gadm.org).

Table 1.

Species of Epipactis sampled for this study

Species	Distribution (countries*)	Sampled countries*	No. of studied populations	Pollination mode	Abbreviation in analyses
Epipactis albensis	AT, CZ, DE, HU, PL, RO, SK	HU, RO	4	Obligate autogam	alben
Epipactis atrorubens	Throughout Europe and TR	SK, LT	2	Facultative allogam	atror
Epipactis bugacensis	HU	HU	2	Obligate autogam	bugac
Epipactis distans	AT, CZ, DE, FR, HR, HU, IT, LT, PL, SI, SK	AT, BG, GR, RS	4	Facultative allogam	dista
Epipactis dunensis	GB	GB	2	Facultative autogam	dunen
Epipactis exilis	BG, ES, FR, GR, HR, HU, IT, RO, SK	BG, GR, HU	4	Facultative autogam	exili
Epipactis futakii	HU,SK	HU, SK	3	Obligate autogam	futak
Epipactis greuteri	AT, BG, CZ, DE, GR, HR, HU, RO, SI, SK	AT, GR, SK	3	Obligate autogam	greut
Epipactis helleborine s.s.	Throughout Europe and TR	AT, BE, BG, CH, DE, DK, FR, GB, GR, HU, RO, RS, RU, SK, TR	29	Allogam	helle
Epipactis leptochila	Throughout Europeand TR	DE, GB, HU, RU, SI	5	Facultative autogam	lepto
Epipactis lusitanica	ES, PT	PT	1	Allogam	lusit
Epipactis mecsekensis	HU	HU	1	Facultative autogam	mecse
Epipactis muelleri	AT, CZ, DE, ES, FR, HR, HU, IT, PL, SI, SK	DE, HU, SI, SK	4	Obligate autogam	muell
Epipactis naousoensis	IT, GR	GR	1	Facultative autogam	naous
Epipactis neerlandica	BE, DE, DK, FR, GB, NL	DK, GB, NL	3	Allogam	neerl
Epipactis neglecta	AT, BE, CH, CZ, DE, FR, GB, GR, HR, HU, IT, ME, RS, SI, SK	DE, GB, HU, RO	5	Facultative autogam	negle
Epipactis nordeniorum	AT, HR, HU, SI	HU	2	Facultative autogam	norde
Epipactis palustris	Throughout Europe and TR	NL	1	Allogam	palus
Epipactis peitzii	DE, HU	DE, HU	4	Obligate autogam	peitz
Epipactis persica	RO, TR	TR	2	Facultative autogam	pers
Epipactis phyllanthes	BE, DK, ES, FR, GB, NL	BE, GB	3	Obligate autogam	phyll
Epipactis pontica	AT, BG, CZ, GE, GR, HR, HU, IT, RO, SI, SK, TR	BG, HU, RO, SK	6	Facultative autogam	ponti
Epipactis pseudopurpurata	HU, SK	HU, SK	3	Obligate autogam	psepu
Epipactis purpurata	Whole of Europe	BG, GB, HU, SK	4	Allogam	purpu
Epipactis renzii	DK	DK	1	Obligate autogam	renzi
Epipactis rhodanensis	AT, CH, DE, FR, IT	AT, DE, FR	3	Obligate autogam	rhoda
Epipactis tallosii	CZ, HU, IT, RO, SK	HU	2	Facultative autogam	tallo
Epipactis tynensis	GB	GB	1	Facultative autogam	tynen
Epipactis veratrifolia	Middle East	CY	1	Allogam	verat
Epipactis voethii	AT, CZ, HU, SK	HU, SK	2	Facultative autogam	voeth

The general distribution, regions and populations sampled, pollination mode and abbreviated name in analyses are given for each taxon.

Distribution data and pollination mode information are from the comprehensive online monograph of the association Arbeitskreis Heimische Orchideen Bayern e.V.W, which treats all named taxa as species and bases pollination mode on personal observations without making judgements regarding their biological validity. Available at: http://www.aho-bayern.de/epipactis/fs_epipactis_1.html

*Country codes are given according to their ISO standard.

RAD-seq library preparation and SNP filtering

The silica gel-stored tissue was used as a template for DNA extraction following the protocol detailed by Sramkó et al. (2014). A modified cetyltrimethyl ammonium bromide (CTAB) protocol was used with RNase treatment to isolate high molecular weight genomic DNA. Checks on 1.8 % agarose gel enabled removal of any samples showing signs of fragmentation and/or RNA contamination, and identified a positive correlation between the period of storage and degree of genomic DNA degradation. Double-stranded DNA (dsDNA) contents of all acceptable samples were assessed using a Qubit v.3.0 fluorimeter (Thermo-Fisher Scientific Inc., USA).

Three single-digest RAD libraries were prepared from between 40 and 60 individuals (including repeats where judged necessary). For each accession, 210 ng of dsDNA was digested with SbfI-HF enzyme (New England Biolabs Inc., USA), reflecting the comparatively large genome size of E. helleborine (1C = approx. 14 pg: Leitch et al., 2009). The protocol of library preparation largely followed Paun et al. (2016) but with the minor modifications described by Trucchi et al. (2017). The only deviation from these past protocols was applying a different regime of sonication using Bioruptor Pico (Diagenode, Belgium), which in this case involved three cycles of 45 s ‘on’ and 45 s ‘off’ at 6 °C. After library control, the libraries were submitted to the VBCF NGS Unit (www.vbcf.ac.at/ngs) for sequencing on an Illumina HiSeq as 100 bp single-end reads.

Raw Illumina reads were first demultiplexed based on index reads with BamIndexDecoder v.1.03 (included in the Picard Illumina2Bam package, available from http://gq1.github.io/illumina2bam/). Stacks v.1.44 (Catchen et al., 2011) was used to further process the RAD-seq data. The reads were de-multiplexed and quality checked with the process_radtags program under the following settings: remove any uncalled base, discard both reads with low quality scores and rescue barcodes, and cut sites with a maximum of one mismatch. Loci were produced de novo, only allowing ‘sequence stacks’ to be formed with a minimum depth of five reads. One mismatch was allowed when merging loci within individuals (i.e. setting ‘M’ in denovo_map.pl of Stacks), but two mismatches were allowed when merging loci among individuals (i.e. setting ‘n’). With export_sql.pl of Stacks, we further extracted from this catalogue a set of loci that occurred in at least 50 % of the individuals and contained between one and 15 SNPs, yielding 3927 RAD-seq loci. A consensus haplotype of each of these loci was further used to build a synthetic reference, including each locus as a ‘chromosome’. To better accommodate the biological and technical variation in the data (i.e. in coverage per individual and locus), the raw reads of all individuals were then mapped back to this reference using BOWTIE2 v.2.2.6 (Langmead and Salzberg, 2012). We then followed the bioinformatics pipeline used by Heckenhauer et al. (2018), sorting the aligned sam files by reference co-ordinate and adding read groups using Picard tools v.2.9.2 (available from http://broadinstitute.github.io/picard), followed by local realigning around indels using the Genome Analysis Toolkit v.3.7.0 (McKenna et al., 2010). Finally, SNPs were called from the realigned bam files using default settings for ref_map.pl of Stacks (Catchen et al., 2011). A VCF file was produced using Stacks’ populations and the file was further filtered using VCFtools v.0.1.14 (Danecek et al., 2011) to only contain SNPs present in at least 90 % of the individuals. The vcf file was finally converted to phylip format with PGDSpider v.2.1.1.3 (Lischer and Excoffier, 2012), using IUPAC symbols to represent heterozygosity.

A measure of inbreeding was estimated for each individual using a method of moments with VCFtools (option -het). The results were drawn as beanplots (Kampstra, 2008) for each of four putative breeding groups determined a priori on anecdotal evidence of contrasts in gynostemium morphology (i.e. obligate allogams, predominantly allogams, predominantly autogams and obligate autogams) in R v.3.1.2, run under Rstudio v.0.98.1102. The statistical significance of differences in distributions of the inbreeding coefficient F between the prior groups was estimated with Mann–Whitney tests in R [command wilcox.test()].

To run admixture analyses, we used the genotype-free method of calling variants implemented in ANGSD v.0.910 (Korneliussen et al., 2014) on the indel-realigned mapping files. Here, we excluded the two outgroup accessions, retained only reads with base and mapping qualities of at least 20 (-minMapQ 20 -minQ 20) and implemented a SAMtools-derived algorithm (-GL 1) when inferring alleles from genotype likelihoods (-doMajorMinor 1). We kept only variable sites at P < 0.000001 (-SNP_pval 1e-06) that had data for at least 50 % of individuals and showed a minor allele frequency of 0.02. Finally, we estimated allele frequencies (-doMaf 2) and output three possible genotypes in a beagle genotype likelihood format (-doGlf 2).

Phylogenetic analyses of SNP data

For phylogenetic tree reconstruction, we used maximum likelihood (ML), as implemented in the web-based version of PhyML v.3.0 (Guindon et al., 2010) with automatic model selection (Lefort et al., 2017). This found the GTR + G model of sequence evolution to be most appropriate for the present data set. Branch robustness was tested using the approximate likelihood-ratio test (aLRT) approach (Anisimova and Gascuel, 2006), where branch support was categorized as ‘strong’ (aLRT ≥0.95), ‘moderate’ (0.95 >aLRT ≥0.90), ‘weak’ (0.90 >aLRT ≥0.81) or ‘none’ (0.80 >aLRT). The resulting ML phylogenetic tree was visualized using PRESTO (available at: www.atgc-montpellier.fr/presto/).

To explore possible reticulations within the phylogeny, we also built a phylogenetic network using the uncorrected-p distance in a NeighbourNet analysis, as implemented in SplitsTree v.4.14.4. This approach provides a better representation of reticulate evolution, which is likely to be present when outcrossing species are considered (Huson and Bryant, 2006).

Lastly, shared ancestry of Epipactis taxa was explored on a sub-sampled data set of unlinked SNPs (i.e. selecting one SNP per locus) with fineRADstructure v.0.2 (Malinsky et al., 2018) using default settings.

To assess the genetic structure of the ingroup, we performed an admixture analysis with NGSadmix v.32 based on the genotype likelihoods data set obtained from ANGSD. Ten independent runs from different starting seeds were carried out for values of K ranging from 1 to 22. The optimal number of clusters (K) was determined by implementing the Evanno method (Evanno et al., 2005), whereas the graphical representations of the results were generated as barplots using the statistical software package R.

RESULTS AND DISCUSSION

Phylogenetic overview

The average number of Illumina reads per accession retained after demultiplexing and quality filtering was 1.9 (s.d. = 0.8) million. When allowing for the presence of 10 % missing values, the 108 Epipactis accessions representing 29 species collectively produced 12 543 filtered SNPs for use in downstream phylogenetic analyses. The average coverage across variants and individuals was 125× (s.d. 60×). The variant calling method of ANGSD inferred genotype likelihoods at 9415 sites that passed the criteria listed in the Materials and Methods.

As would be anticipated for a phylogeny based on such a large volume of informative characters, most of the branches on the resulting ML tree (Fig. 3) receive either strong or moderate statistical support (aLRT ≥90). In summary, as expected, branches separating the outgroups from the ingroup are much longer than those within the ingroup. The tree (described in greater detail below) reveals several monophyletic groups subtended by comparatively long branches that are nested with a paraphyletic (arguably polyphyletic) E. helleborine. The deeper branches are considerably shorter; three receive only weak statistical support and a further three collapse.

Fig. 3.

Maximum likelihood phylogram depicting the evolutionary relationship of the studied Epipactis species (left) and ‘heat map’ of shared ancestry (right) based on RADseq data. On the tree, unsupported branches (aLRT <80) are collapsed, weakly supported branches (80 ≤aLRT <90) are dashed, whereas moderately to strongly supported branches (aLRT ≥90) are presented as a continuous line. Sample codes follow Supplementary data Table S1 suffixed by a two-letter country code. Major groups of samples identified as monophyletic entities are grouped by grey shading and identified by upper case letters (B–K), whereas main branches referenced in the text are indicated by Roman numerals. Samples of the progenitor species, forming the paraphyletic group A, are highlighted in red. Branches leading to the two outgroup samples are severely truncated to facilitate viewing. The ‘heat map’ indicates the level of shared ancestry between samples according to the attached key, where yellow denotes the lowest and dark blue the highest co-ancestry values.

The weak confidence in the relationships recovered between some of these early-divergent lineages is even more evident in the star shape of the unrooted network (Fig. 4). Nonetheless, major groups of samples identified as monophyletic in the rooted tree remain evident, once again subtended by comparatively long branches. Relationships among the individual samples are strikingly similar, and only two samples occupy placements significantly different from those evident in the rooted tree. The Bulgarian sample attributed to E. ‘exilis’ (223) is placed well below the other ‘exilis’ samples, and the English plant attributed by some workers to the supposedly entirely Continental E. ‘neglecta’ (348) is no longer placed within the E. leptochila clade that includes the remaining four samples attributed to ‘neglecta’; instead, it is placed close to the origin of the network.

Fig. 4.

Phylogenetic network depicting the genetic relationship of the studied Epipactis samples presented as a NeighbourNet network based on uncorrected-p distance. The dots represent an edge with a sample (codes are given according to Supplementary data Table S1); red dots represent samples of the progenitor species. The branches leading to the two outgroup samples – E. palustris (palus445) and E. veratrifolia (verat362) – are severely truncated to facilitate viewing.

One further sample merits explanation because of its unexpected phylogenetic placement. A Russian sample (165) that when collected was attributed to E. helleborine is actually nested phylogenetically well within E. leptochila sensu lato (s.l.), adjacent to a Russian sample of E. leptochila s.s. (166). This proximity leads us to believe that sample 165 was misidentified when collected, actually being a plant of E. leptochila that possesses an unusually well-developed viscidium. Admittedly, E. leptochila has only recently been formally reported from Russia (Fateryga et al., 2015).

Having attempted to account for these ‘rogue’ samples, we can now proceed to use the remaining samples to attempt an optimal circumscription of species within section Epipactis. Named taxa regarded by us as bona fide species are preceded by ‘E.’; taxa regarded by us as infraspecific are simply given as single epithets (e.g. ‘exilis’).

Re-circumscription of species

Our exploration of the more evolutionary aspects of this study will be made simpler and more comprehensible if we begin the more detailed discussion with our conclusions regarding species delimitation. In doing so, it is important to bear in mind that the genetic disparities between any of the ingroup plants shown in Fig. 3 are much smaller than those detected between most other closely related species of European orchids – even the major lineages within Ophrys offer greater collective disparity (Bateman et al., 2003, 2018). It is also important to note that the application of monophyly at species level as one component of a lineage-based species concept remains controversial (for a comparison of approx. 24 species concepts, see Mayden, 1997). In particular, if a genuinely ancestral species remains extant and the daughter species arose only recently, the species containing the ancestor(s) of the derived species is likely to be resolved as paraphyletic rather than monophyletic.

The more relevant of Bateman’s (2012a) rules for converting a tree into species are, briefly: (1) recognize only monophyletic groups (here necessarily amended as ‘recognize monophyletic groups except where a paraphyletic ancestral species remains clearly identifiable’); (2) preferentially divide the tree at branches that are relatively robust (and usually comparatively long within the context of the tree in question); and (3) preferentially divide the tree in a way that minimizes the need to (a) create new names (a low risk in this case, given the huge number of epithets already available within section Epipactis) and (b) create new combinations of existing names.

Based on applying these rules to the topology and comparative branch lengths of Figs 3 and 4, we conclude that optimally 11 ingroup species can be recognized, ten of them ostensibly monophyletic and seven of them dominantly autogamous. These groups of samples that comprise these species are labelled A–K on Fig. 3; additional internal branches of potential interest are labelled in roman numerals (I–XVI). These 11 entities delimited within the ingroup are also evident in the admixture results (Fig. 5), which find optimal peaks in clustering the accessions in four, seven and 15 groups (Supplementary data Fig. S1) according to the deltaK method of Evanno et al. (2005).

Fig. 5.

Barplots of results of admixture analyses based on 9415 polymorphic sites inferred with ANGSD. The method of Evanno et al. (2005) identified four, seven and 15 groups as best clustering patterns across the ingroup samples. Taxa codes follow Table 1. Major groups of samples identified in Fig. 3 are indicated by upper case letters (B–K) and are compared with the progenitor paraphyletic group A.

It is immediately clear from Fig. 3 that most of the autogamous lineages are embedded within the ancestral plexus that is E. helleborine (species A). The samples attributed by us to E. helleborine are distributed throughout much of the tree, though concentrated in the derived portion of the tree above branch VII. In the network (Fig. 4) they are placed around the centre of the graph on shorter branches. The admixture results (Fig. 5) show this lineage to be a mixture of alleles present in all other genetic pools. Imposing rigorous monophyly on E. helleborine s.s. would require us to allocate the 29 samples of E. helleborine among no less than 16 ‘species’. Given that all of these ‘species’ would probably be indistinguishable using morphological criteria (or even using standard candidate gene sequencing), we have little option but to continue to accept E. helleborine as a most likely paraphyletic species. In addition, ongoing uncertainty rests over two samples located at important nodes positioned low in the tree: might the single analysed sample (108) of the Iberian allogam E. lusitanica (species B) be more justifiably attributed to E. helleborine? Or conversely, could the single Turkish plant attached above branch V, diverging immediately below species F and initially attributed by us to E. helleborine (173), actually represent a new and as yet unrecognized allogamous species?

The nine remaining derived species are distributed throughout the tree. All are nucleated around at least one well-known and comparatively widespread taxon (when selecting the preferred epithet for each species, we adopted the apparently earliest name but did not perform in-depth nomenclatural research). Within our spectrum of analysed taxa that were treated as ‘species’ by Delforge (2016), four of the nine bona fide species contain samples attributed to only one Delforgean species: E. atrorubens (C), E. greuteri (G), E. pontica (H) and E. muelleri (I). Between two and four Delforgean species are encompassed by the five remaining species: E. phyllanthes (D), E. purpurata (E), E. leptochila (F), E. albensis (J) and E. dunensis (K). All of these groups are both monophyletic (Fig. 3) and genetically relatively homogeneous (Fig. 5) (Rule 1); it is mainly the criterion of comparatively long molecular branches subtending species (Rule 2) that has dictated these amalgamations of putative species.

For example, in species K (E. dunensis), samples attributed to rhodanensis and bugacensis are intermingled, thus failing the criterion of monophyly, but could in theory be combined into a single monophyletic taxon, which would be sister to E. dunensis (apparently monophyletic) and its localized segregate tynensis (monophyly not tested, as only one sample was included in our analysis). However, the very short branch lengths within the group (Fig. 3) and the unique genetic cluster identified in the admixture analyses (Fig. 5) clearly show that these samples constitute a single cohesive biological entity, particularly when considering that together these eight samples of species K extend from Britain eastward as far as Hungary and share a clear habitat preference for disturbed sandy soils. In the case of species E (E. purpurata), four widely geographically separate samples attributed to E. purpurata form a paraphyletic group subtending a monophyletic (Fig. 3) group of three samples of the eastern European ‘local endemic’ obligate autogam pseudopurpurata, but comparative branch lengths indicate that these two taxa should be treated as a single species. They also form a single gene pool (Fig. 5). Species F (E. leptochila) encompasses three additional named taxa, neglecta, peitzii and futakii, that have often been recognized as species but were grouped within E. leptochila as subspecies by Kreutz (2004) and as varieties by Delforge (2016). Some of these cases may indicate situations where incipient speciation has just begun and could in theory lead rapidly to distinct lineages (analogous to groups G, H and I in Fig. 3). Certainly, some of these groups appear on present evidence to be monophyletic and, although they tend to be geographically localized, samples are in some cases at least 100 km apart. Such clusters suggest a modest degree of divergence and the potential to form independent lineages in the future, though we suspect that most such groups are in practice rapidly drawn back into the general genetic background and fail to evolve the credentials needed for subspecies or species status. The lack of genetic distance separating these often apparently monophyletic groups and the remaining populations forming the species they are embedded in is simply reflecting the lack of time elapsed since their separation, but the monophyly of the accessions – often sampled hundred kilometres apart from each other – is a good sign that some sort of group within the ‘gene pool’ has formed. We do not yet know the fate of this ‘group’; it could easily return to the genetic background from which it emerged. Therefore, we cannot accept these entities at the rank of species.

Other Delforgean species clearly cannot be separated at species level from E. helleborine s.s., either because they are phylogenetically intermingled with samples of E. helleborine, thus failing Rule 1 (distans), or because they are placed close to unequivocal samples of E. helleborine s.s., thus failing Rule 2 (e.g. neerlandica, naousaensis and voethii). Although absent from our analysis, E. microphylla has already proven its qualifications as an autogamous species in previous molecular studies (Bateman et al., 2005; Hollingsworth et al., 2006; Tranchida-Lombardo et al., 2011; Zhou and Jin, 2018). With that exception, we suspect that most, but probably not all, of the 39 putative species of section Epipactis that were listed by Delforge (2016) but not included in our analysis would similarly fail to qualify as bona fide species if analysed via RAD-seq, each instead being placed phylogenetically either within one of the derived species B–K or, in some cases,within the ancestral plexus that is E. helleborine. As is the case in several other genera of Eurasian orchids, severe taxonomic inflation has occurred in recent years, driven largely by intensity of interest within the orchidological community. Of course, the challenge remaining to systematic biology is to generate morphometric matrices of equal rigour and taxonomic breadth, in order to assess the degree of genus-wide congruence between genotype and phenotype. Thus far, with very few geographically restricted exceptions (e.g. Tyteca and Dufrene, 1994), the morphology of Epipactis has been seriously under-researched using any method beyond traditional herbarium taxonomy. Morphometric studies could be particularly effective in apportioning named taxa that do not merit species rank among the ranks of subspecies (e.g. E. helleborine subsp. neerlandica) or varieties (e.g. E. helleborine subsp. neerlandica var. renzii) – a topic to which we return in the next section (Pedersen and Ehlers, 2000; Brys and Jacquemyn, 2016; Jacquemyn et al., 2018).

Ancestry and breeding system during speciation

Branches subtending these ten derived monophyletic species are reliably considerably longer than those that form the backbone of the phylogeny, suggesting that delimitation of these species can be discussed with greater confidence than relationships among them. This result is unsurprising, as previous sequencing studies have given meaningful support to only one major branch: branch VII, which separates the early-divergent species E. phyllanthes, E. purpurata and E. leptochila (also E. microphylla) from the more derived clade containing the majority of the remaining species, was evident previously in both nrITS sequences (Bateman et al., 2005; Hollingsworth et al., 2006; Zhou and Jin, 2018) and a marker in the plastid rbcL–accD spacer (Tranchida-Lombardo et al., 2011). Interestingly, this is the branch above which incontrovertible specimens of E. helleborine are confined (Fig. 3). Moreover, of the three species that diverge below this point on Fig. 3, two (E. atrorubens and E. purpurata) are dominantly allogamous; only E. phyllanthes is autogamous (though previous candidate gene studies show that the autogam E. microphylla also diverges below this point). In contrast, all five species that diverged subsequently from within the E. helleborine alliance are at least facultative autogams (E. leptochila, E. greuteri, E. muelleri, E. albensis and E. dunensis). Although in most cases the autogamous species have as their closest relatives allogamous species, there is one exception – the eastern European E. albensis is sister to E. dunensis, a species traditionally viewed as western European. This relationship implies that one of these autogamous species most probably gave rise to the other, though it does not indicate which is the more probable ancestor. Admittedly, cladistic principles state that both of these autogamous sister species could in theory have originated from an allogamy that has since become extinct, though such a scenario appears to us highly improbable. In summary, our initial hypotheses that (a) autogams evolved from allogams and (b) autogams are evolutionary dead-ends do constitute good guidelines, as was previously believed; however, we have also shown that they apparently fall short of being unbreakable rules.

In addition to bona fide monophyletic species derived from E. helleborine, there are signs of incipient speciation within the clade subtended by branch XI (Fig. 3), yielding taxa most appropriately treated as subspecies that have not (yet) gained clear molecular monophyly or become clearly distinct morphologically from E. helleborine. One of these examples, located above branch XII, is the localized eastern European facultative autogam voethii, intermingled with a co-occurring sample of typical E. helleborine from western Slovakia. The other example, located above branch XIV, is the localized eastern European, facultative autogam distans, again intermingled with a co-occurring sample of typical E. helleborine from a lower elevation on the same Bulgarian mountain. Also of interest is neerlandica, located above branch XIII and here, as expected, shown to be a coastal ecotype of E. helleborine (Brys and Jacquemyn, 2016). Within this clade, geography dominates over both taxonomy and breeding system. The Danish plant of the allogam neerlandica is sister to, and molecularly very similar to, the Danish plant attributed to the rare and reputedly obligate autogam renzii. The Welsh duneland plant attributed to neerlandica has a similarly close relationship with a plant attributed to E. helleborine sampled from nearby back-dune woodland. These results endorse previous, more detailed investigations of this duneland ecotype (Jacquemyn et al., 2016, 2018), but are the first to demonstrate a close relationship between the Welsh and Dutch/Danish populations in a molecular phylogenetic context. Further down the tree, above branch VIII, a Greek sample of E. helleborine s.s. is shown to be sister to the molecularly similar facultative autogam nauosaensis, reputedly endemic to Greece and Bulgaria and said to ‘often form hybrid swarms with E. helleborine’ (Delforge, 2016: p. 72) – an observation readily explained as these few localized populations are, in fact, a minor variant within E. helleborine.

All of the bona fide species present in our analysis of section Epipactis are geographically widespread, extending across at least half of Europe. The most extensive are E. helleborine and E. atrorubens – the only species to reach as far northward as Norway. In contrast, only E. helleborine and E. phyllanthes (together with E. microphylla) occur frequently at sub-alpine altitudes in the Mediterranean Basin. Central Europe has marginally the highest species diversity, where species with a western bias (E. leptochila, E. muelleri and E. dunensis) overlap with species with an eastern bias (notably E. pontica) and others have the centres of their distributions (E. greuteri and E. albensis). As would be expected (e.g. Burns-Balogh et al., 1987), taxa apparently undergoing incipient speciation (distans and especially voethii) consist of fewer, more localized populations.

With regard to seeking likely centres of origin of these derived species, there is obvious interest in exploring the ‘background’ genetic structure within the ancestor of most E. helleborine. This task was made rather more difficult by the realization that the two easternmost samples attributed by their collectors to E. helleborine yielded substantially deviant RAD-seq profiles. As already noted, we suspect that sample 165 from the Caucasus Mountains was a morphologically misidentified plant actually attributable to E. leptochila – a species only recently been reported from Russia (Fateryga et al., 2015). The western Turkish sample 173, isolated in the phylogeny on a comparatively poorly supported branch V immediately below E. leptochila (Fig. 3), presents a more ambiguous challenge. We could continue to view this plant as the earliest divergent of our E. helleborine samples, but alternatively it could be treated as a geographically disparate, as yet unnamed allogam – analogous with, and geographically mirroring, the Iberian endemic E. lusitanica located near the base of the tree (species B). Allogamous populations of Asia Minor evidently merit further investigation.

As expected from mainstream evolutionary theory (Wright et al., 2013), lineages of all autogamous species (as re-circumscribed here) form monophyletic entities on our phylogenetic tree (Fig. 3), suggesting that gene flow with their progenitor, E. helleborine s.s., is limited or absent (Brys and Jacquemyn, 2016). Although inbreeding coefficients were calculated from our RAD-seq data primarily as a proxy for comparative levels of selfing, we recognize that our estimates of F, based on proportions of heterozygous loci, are also influenced by effective population sizes, both current and in the recent past (i.e. bottlenecks). We also note the comparatively small number of individuals analysed per taxon. We have less concern regarding the effect of multiple ramets developing from a single genet through division of rhizomes, as this phenomenon is far more prevalent in section Arthrochilum, in which occasional albino-flowered plants demonstrate just how extensive a single genet can become.

If interpreted primarily in the context of inbreeding, our F-values indicate a tendency for frequent selfing in all species initially hypothesized to be obligately or predominantly autogamous (Fig. 6). However, all inbreeding coefficients recovered were positive within a theoretical scale of –1 to +1 for F-values, indicating an excess of homozygotes even in the case of allogamous taxa; values for some individuals approximate the mean value of facultative allogams (Fig. 6). In these cases, a positive F-value can reflect the often small effective population sizes in Epipactis species, though it can also be the result of increased levels of geitonogamy, perhaps prompted by less favourable environmental factors. Previous studies reported much narrower genetic variability in self-pollinating Epipactis taxa compared with the outcrossing progenitor species E. helleborine s.s. (Ehlers and Pedersen, 2000; Pedersen and Ehlers, 2000; Squirrell et al., 2002; Hollingsworth et al., 2006). A similar pattern is evident in the present study, as E. helleborine s.s. occupies a much greater genetic space on the phylogenetic network than any other species (Fig. 4). Moreover, all but one of the autogamous species arise from within this genetic variation, thus arguing for the universality of E. helleborine s.s. as the progenitor for Epipactis species originating above node IV.

Fig. 6.

Beanplots of estimated inbreeding coefficient F for four hypothetical a priori categories of breeding system in Epipactis section Epipactis. The index F reported here is estimated per individual using a method of moments implemented in VCFtools. The significance of differences in distribution between categories calculated with a Mann–Whitney test is indicated as ***P < 0.001 and as ‘n.s.’ if P > 0.05.

In some cases, the geographic origin of the allogamous sister sample of a monophyletic autogamous species allows inference of their approximate geographical origin. For example, the predominantly central European clade XVI has as sister samples E. helleborine s.s. accessions collected in Hungary, hinting at a central European origin of this clade that has since greatly expanded its geographic distribution as far as the British Isles. Similarly, the predominantly Balkanic species E. greuteri (species G) has sister lineages from exclusively Greek accessions of E. helleborine s.s. and the Greek infraspecific endemic naousoensis making up the monophyletic clade VIII (Fig. 3). We can conclude a southern Balkanic origin of E. greuteri, which then successfully colonized suitable habitats northward to central Germany.

Because in several cases the samples of a given autogamous species were collected from geographically distinct locations but nonetheless formed a monophyletic group, we can also argue for a single origin of these species and for their subsequent spread. This ecological success is partly the result of being able to colonize environments where the outcrossing progenitor has a disadvantage because of the limited availability of pollinators (cf. ‘reproductive assurance hypothesis’: Wright et al., 2013; Barrett et al., 2014). For example, E. pontica is a typical plant of shaded beech forests from northern Turkey to eastern Austria, where the most characteristic pollinating insects of E. helleborine s.s., wasps, are usually less frequent (Claessens and Kleynen, 2011). Improved colonization ability of self-pollinating plants has been reported in the literature (Baker, 1955; Wright et al., 2013) and might have contributed significantly to the large geographic area that autogamous Epipactis species can occupy.

Allogamy vs. autogamy: transition or gradation?

We initially rather boldly divided the taxa in our analysis among four categories: obligate and facultative allogamy, and facultative and obligate autogamy. The inbreeding coefficients calculated by us yielded statistically significant differences between some prior categories but not between facultative and obligate autogams; also, all four categories registered notable outliers (Fig. 6). Outliers were less evident when the F-values were reallocated to groups A–J as re-circumscribed here (Figs 3 and 4). When viewed as a posteriori groups, mean F-values per taxon decline in a gradual progression from 0.85 to 0.65, before falling to lower levels of approx. 0.5 in E. atrorubens plus E. helleborine neerlandica and approx. 0.3 in the clearly more allogamous E. helleborine helleborine (Supplementary data Fig. S2) (Bateman, 2019). We are increasingly inclined to view categories of breeding system as gradational, despite the fact that previous population genetic studies that encompassed numerous putative species (Squirrell et al., 2002; Hollingsworth et al., 2006) have suggested a profound contrast between species that are dominantly allogamous and those that are dominantly autogamous – the latter reliably show fixed homozygosity in allozyme profiles. Squirrell et al. (2002: p. 1963) further noted that ‘With each generation of complete selfing, homozygosity increases by 50%. In this fashion, a large genetic distance arises rapidly between progenitor and derivative species.’ Additionally, Brys and Jacquemyn (2016) found that rapid evolution of selfing can be achieved by outbreeding depression acting within Epipactis populations, thus effectively isolating autogamous forms from outbreeders. We believe that this phenomenon is evident in the comparative branch lengths of our phylogeny; a much shorter molecular distance separates the root of the tree from individuals of the allogams E. helleborine, E. lusitanica and E. atrorubens than from the various autogams. Admittedly, it is also a longer distance to the morphologically allogamous E. purpurata, but allozyme data have suggested that this species is more vulnerable to geitonogamy (Hollingsworth et al., 2006).

Although some authors have argued that some species (including E. helleborine) are obligately allogamous, both crossing experiments (Tałałaj and Brzosko, 2008; Jacquemyn et al., 2018) and some pollinator exclusion experiments (Brantjes, 1981; Ehlers et al., 2002; Tałałaj and Brzosko, 2008) demonstrated that intrinsic sterility barriers facilitating self-incompatibility are absent across the genus; these species are self-fertile. Unsurprisingly, given the elongate inflorescences bearing numerous large flowers, effective self-pollination via geitonogamy occurs frequently (Ehlers and Olesen, 1997; Kropf and Renner, 2008; Tałałaj and Brzosko, 2008; Claessens and Kleynen, 2016), leading Bateman (2019) to argue that geitonogamy is most likely to be the main cause of the positive F-values reported here (Fig. 6; Supplementary data Fig. S2) and that even the taxa categorized a priori as ‘allogams’ are likely to undergo a majority of self-pollination events. Fruit-set percentages are predictably high in both the autogamy and allogamy categories, as all species secrete considerable quantities of nectar and insect-attracting volatiles (Brodmann et al., 2008).

Whether wholly obligate autogamy occurs is also questionable (Fig. 6; Supplementary data Fig. S2). On the spectrum of allogamy to autogamy transition, some of the supposed ‘species’ of section Epipactis, particularly those based on a small number of geographically localized populations, are prone to cleistogamy; some have a proportion of cleistogamous populations (e.g. E. greuteri and E. phyllanthes), whereas futakii (placed within E. leptochila in Fig. 3) is predominantly cleistogamous and shows some morphological adaptations to cleistogamy (Fig. 1L). These taxa, showing the extreme end of the spectrum as near-obligately autogamous (Richards, 1982, 1986), appear on our tree as monophyletic entities terminating comparatively long branches. These long branches, presumably reflecting accelerated retention of mutations and/or fixation of single alleles at many loci, may reflect rapid isolation of autogamous species, as their separation from their genetic background (e.g. E. leptochila in the case of futakii) is fairly recent in evolutionary terms. We doubt, however, that cleistogamy alone is adequate grounds for separation at the species level, as the appearance of cleistogamous individuals within dominantly autogamous species makes it difficult to accept cleistogamy (which is in any case generally imperfectly expressed) as a pre-eminent criterion for species delimitation. Multiple samples of most of the dominantly cleistogamous taxa such as futakii (within E. leptochila: Fig. 3) and pseudopurpurata (within E. purpurata) not only formed monophyletic groups but also differed little if at all in genotype, despite the vast number of SNPs included in our analysis; these are expected properties of a lineage.

On the above reproductive spectrum, the primary morphological indicators of autogamy are said to be a reduced viscidium (a sphere of milky adhesive liquid surrounded by a viscidial membrane connected to the pollinarium with the function of adhering it to the pollinator) separating the pollinaria from the broad, adhesive stigmatic surface below and more friable pollinia capable of disaggregation (Richards, 1986; Claessens and Kleynen, 2011). Nonetheless, the relevance of environmental factors should not be underestimated. The actual pollination syndrome can be influenced by floral ontogeny, environment or both (Bateman, 2012b; Claessens and Kleynen, 2016). In particular, the quantity and effectiveness of the viscidial glue decrease during anthesis, as does the physical integrity of the pollinia. Thus, Epipactis flowers are well adapted for a mixed breeding system that emphasizes allogamy (or geitonogamy, providing de facto autogamy) early in anthesis but incurs an increased risk of autogamy with time. Warm temperatures, drying winds and/or low humidity can all conspire to accelerate this process (e.g. Claessens et al., 1998; Pedersen and Ehlers, 2000), thereby causing contrasting frequencies of autogamy in successive flowering seasons. Consequently, some taxa that possess the suite of floral features regarded as being indicative of autogamy have in practice been shown to possess population genetic profiles typical of allogams: an example is the British youngiana (Hollingsworth et al., 2006).

Close morphological similarities make the confident identification of hybrids difficult in the absence of diagnostic molecular markers. Also, the difficulty of cultivating species of section Epipactis (in contrast to section Arthrochilum) discourages attempts at artificial crossing. Nonetheless, hybrids between bona fide species within section Epipactis are often reported in nature (Vlčko et al., 2003; Batoušek, 2010). The limited genetic evidence of gene flow between species reliably indicates flow from allogams (typically E. helleborine) into autogams, including E. dunensis tynensis and E. phyllanthes (Hollingsworth et al., 2006), but such hybridization has also been inferred through morphological means in the case of E. purpurata × albensis (Jakubska-Busse et al., 2017).

Our shared ancestry analysis carried out using fineRADstructure (Fig. 3) offers a direct test of possible hybridization events; SNPs shared between samples at the tips of the phylogenetic tree are indicated on the ‘heat map’. Higher levels of shared SNPs located close to the left-hand margin of this heat map are assumed to indicate higher levels of shared ancestry between phylogenetically closely related entities – an expectation if lineages are phylogenetically well isolated from the rest of the samples. In contrast, shared ancestry deeper in the heat map should represent recent gene flow between phylogenetically distant entities.

Although we deliberately excluded from our sample set any suspected hybrids, it is nonetheless reassuring to see little evidence of gene flow in the heat map (Fig. 3). The highest levels of shared SNPs are indicated for the autogamous species E. phyllanthes (species D), E. greuteri (G), E. pontica (H) and E. muelleri (I). Less pronounced elevated levels of shared ancestry are evident in other autogamous lineages, including E. dunensis (species J), E. albensis (K) and E. leptochila (F). These findings, plus the relatively homogeneous gene pools revealed by the admixture results (Fig. 5), indicate a comparatively cohesive genetic background of the autogamous lineages, a conclusion that accords with previous population genetic studies (Ehlers and Pedersen, 2000; Pedersen and Ehlers, 2000; Squirrell et al., 2002; Hollingsworth et al., 2006). Much denser genetic sampling is needed to determine whether there is significant gene flow between E. helleborine and the predominantly autogamous species. Such data are needed to test the ‘evolutionary detour’ hypothesis in the genus formulated by Hollingsworth et al. (2006), who argued that genes permitting a more autogamy-friendly gynostemium morphology might initiate increasingly autogamous lineages after entering the novel heterozygous background characteristic of an allogam.

There are nonetheless some examples of possible gene flow evident in the fineRADstructure analysis (Fig. 3). The most significant is an indication of higher co-ancestry than expected between nauosaensis (sample 251) and a typical E. helleborine (sample 258) from the same mountain in Greece. The second most localized sign of shared co-ancestry involves a German sample of E. leptochila s.s. (sample 334) and a Turkish sample of E. phyllanthes (sample 393), followed by sharing of SNPs between a British E. helleborine (sample 344) and French E. dunensis (sample 465); we assume these somewhat higher co-ancestry values to be coincidental.

Indications of gene flow between phylogenetically distantly related samples are otherwise absent from our data set. This comes as a surprise, as sample 348 of the present study represents the controversial Risborough population in England that has variously been suggested as being E. leptochila leptochila, E. leptochila neglecta or E. leptochila introgressed by E. helleborine. The strongly contrasting phylogenetic placement of this plant in Figs 3 and 4, and its location well below the remaining 17 samples of E. leptochila in Fig. 3, may indicate that the introgression hypothesis is the most likely, though there is little evidence of gene flow in the shared ancestry analysis. A possible explanation of this finding is that a genotype closely similar to the Risborough population of E. helleborine is not represented in our data set, so the method cannot possibly find traces of gene flow. Whatever the explanation for the actual gene flow patterns reported here for given samples, the shared ancestry analysis detected only limited and localized gene flow between phylogenetically distant samples in section Epipactis.

However, local-scale studies indicated that even a modest amount of gene flow between dominantly allogamous and autogamous species should be sufficient to maintain a degree of heterozygosity within the autogams (Durka, 2002; Hollingsworth et al., 2006). An additional point made by Claessens and Kleynen (2011) is that airborne pollinators (most commonly wasps in most of the allogams) are by no means the only insects that routinely inhabit Epipactis plants, and that the friable nature of the pollinia can in practice permit transfer of pollen tetrads by much smaller insects. Indeed, friability could also lead to mixed matings, pollen tubes from different source plants competing to reach the several thousand ovules held within a particular Epipactis ovary.

Clearly, the transition from allogamy to autogamy is far more complex and labile than a simple binary switch. However, our data support the view that species with a tendency to be autogamous are – despite occasional allogamous pollination or hybridization – isolated from the genetic background provided by E. helleborine. If the switch to autogamy happened relatively long ago in the evolutionary history of the genus, the autogamous lineages might have gained severe genetic isolation from their genetically rather polymorphic progenitor species.

Does speciation in the E. helleborine alliance constitute an active evolutionary radiation?

After reviewing in detail prior definitions and underlying concepts, Bateman (1999: p. 441) defined an evolutionary radiation as ‘a large surplus in the rate of natality over the rate of mortality for species and/or character states within a specified clade through a specified time interval.’ Given the extreme paucity of fossil orchids – none is definitively known from Epipactis or any other member of tribe Neottieae, though the amber-entombed Eocene pollinarium Succinanthera bears some resemblance (Poinar and Rasmussen, 2017) – we are poorly placed to comment on the rate of species mortality within Epipactis section Epipactis. This means that we cannot conclusively differentiate between a genuine radiation and a lineage that simply indulges in an unusually high rate of species turnover.

However, we do now possess conclusive evidence that section Epipactis constitutes a clade, that all of the speciation events within the section have occurred comparatively recently, that each such event is underpinned by a molecular branch considerably longer than those within the group and that several unequivocal species have emerged from within a single ancestral species, E. helleborine s.s. This impression is further enhanced by the NeighbourNet analysis (Fig. 4), which clearly depicts comparatively long-branch clusters of samples, each cluster constituting a single re-circumscribed species, radiating outward from a central core of comparatively short-branch allogams. Thus, we have in place most of the features that we would hope to see in a genuine radiation. Also, the fact that we can identify examples of incipient speciation that are employing the same evolutionary mechanisms as the lineages that have already achieved speciation strongly suggests that we could legitimately brand this radiation as ongoing and thus active.

When used in its evolutionary context, the term radiation is rarely widely separated from the term adaptive (Bateman, 1999). Referring to the drivers that allow the transition from allogamy to autogamy in the genus – reduced size of rostellum and increased friability of pollinium – these represent a combination of developmental genetic, epigenetic and ecophenotypic influences. The highly iterative trend toward autogamy could be taken as evidence of prolonged selection pressure, but equally it could be viewed simply as weak control of rostellum development leading to genetic drift (Hollingsworth et al., 2006) combined with environmentally driven desiccation of the pollinarium (R. Bateman, unpubl. obs.).

In the case of the sister pairing of re-circumscribed autogamous species, all of the named taxa now encompassed by E. albensis flower later than those encompassed by E. dunensis, and most also prefer less alkaline soils. Given that many Neottieae species have been shown to maintain contrasting yet broadly predictable mycorrhizal communities (e.g. Bidartondo et al., 2004; Selosse et al., 2004; Jacquemyn et al., 2016, 2018; Schiebold et al., 2017), it seems likely to one of us (R.M.B.) that mycorrhizal switching may have enabled colonization of contrasting soils and habitats, thereby at least assisting – if not driving – speciation (Bateman, 2012b). Certainly, mycorrhizally mediated nutrient sequestration is greater in species that preferentially inhabit shadier woodlands (Schiebold et al., 2017).

Shady woodlands with comparatively impoverished ground floras are the preferred habitats of most of the autogams. This observation has led some authors to conclude that the autogams are adapted to these habitats because insect faunas are likely to be similarly impoverished and hence pollinator visits are likely to be less frequent (e.g. Claessens and Kleynen, 2011, 2016). In light of our data presented above, the frequency of successful pollination events must be of little if any evolutionary or ecological significance, given the huge number of seeds produced by each fertilized Epipactis capsule; only if pollination fails completely through a considerable period of time will the population be under serious threat of extirpation through non-renewal.

Conclusions

It is notoriously difficult to identify with any confidence ancestor–descendant relationships (rather than the less informative sister species relationships) at the species level and to identify with adequate rigour an evolutionary radiation. Nonetheless, it is highly probable that E. helleborine is the ancestor of at least ten recently derived species, the majority of them near-obligate autogams. Genuine speciation events – as opposed to those where the embryonic ‘pro-species’ is drawn back into the parental gene pool or becomes extinct – can derive one autogamous species from another, though the majority of autogamous species are likely to prove to be evolutionary dead-ends. As noted by Claessens and Kleynen (2011), comparison of the dominantly allogamous species vs. the dominantly autogamous species within section Epipactis does not reveal any great disadvantage incurred by the latter, despite the theoretical ravages of inbreeding depression. It is possible that inbreeding depression is offset by outbreeding depression, as in the allogamous terrestrial orchid Gymnadenia conopsea (Sletvold et al., 2012), or that inbreeding depression is purged from the autogams by selection operating on unmasked deleterious recessives. With the exception of the ancestral plexus that is E. helleborine, the average distributional areas or population sizes do not differ greatly between the allogams and the autogams. However, E. helleborine does occupy a greater range of phylogenetic space (Figs 3–5), suggesting that the combination of its greater genetic diversity and its predominance of comparatively allogamous populations allows it to function far more successfully as a source of further novel species.

The vast majority of the (usually more geographically localized) taxa awarded species status by some systematists on the basis of perceived morphological differences actually merit the rank of subspecies, variety or forma. The great frequency and rapidity with which populations that are increasingly oriented toward autogamy can emerge from within dominantly allogamous species, and the difficulty of using floral morphology alone to determine within species the comparative frequency of allogamy vs. autogamy, together dissuade us from placing too much taxonomic emphasis on perceived breeding system. One of the many advantages of acquiring genetic data is their ability to summarize the recent breeding history of the populations under scrutiny, such that, as here, species concepts can be applied that employ both monophyly and genetic cohesion. It is unsurprising that the status of E. helleborine as the near-universal parent of the remaining species must excuse it from the general requirement for monophyly.

Although Epipactis section Epipactis has become an increasingly well-understood model system for the studies conducted at the interface of evolution and ecology, several disparate factors contribute to speciation within the section. The comparative significance of genetic vs. epigenetic or ecophenotypic influences on phenotype, pre-adaptation to adaptation, selection vs. drift or saltation and of partnerships with pollinators vs. those with mycorrhizae, all remain open for further informed discussion.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: accessions used in this study. Figure S1: plot of the ad hoc statistic deltaK used to identify gene pools within genetic data. Figure S2: box plot showing means and sample standard deviations of the inbreeding coefficient per re-circumscribed taxon within the E. helleborine group.

Funding

This work was supported by the Hungarian Scientific Research Fund (OTKA PD109686 to G.S.) and the Austrian Science Fund (FWF Y661-B16 to O.P.).