Whole‐genome analyses disentangle reticulate evolution of primroses in a biodiversity hotspot

Summary

• Biodiversity hotspots, such as the Caucasus mountains, provide unprecedented opportunities for understanding the evolutionary processes that shape species diversity and richness. Therefore, we investigated the evolution of Primula sect. Primula, a clade with a high degree of endemism in the Caucasus.

• We performed phylogenetic and network analyses of whole‐genome resequencing data from the entire nuclear genome, the entire chloroplast genome, and the entire heterostyly supergene. The different characteristics of the genomic partitions and the resulting phylogenetic incongruences enabled us to disentangle evolutionary histories resulting from tokogenetic vs cladogenetic processes. We provide the first phylogeny inferred from the heterostyly supergene that includes all species of Primula sect. Primula.

• Our results identified recurrent admixture at deep nodes between lineages in the Caucasus as the cause of non‐monophyly in Primula. Biogeographic analyses support the ‘out‐of‐the‐Caucasus’ hypothesis, emphasizing the importance of this hotspot as a cradle for biodiversity.

• Our findings provide novel insights into causal processes of phylogenetic discordance, demonstrating that genome‐wide analyses from partitions with contrasting genetic characteristics and broad geographic sampling are crucial for disentangling the diversification of species‐rich clades in biodiversity hotspots.

Introduction

Biodiversity hotspots, which are regions with high levels of species diversity and endemism (Myers et al., ²⁰⁰⁰; Habel et al., ²⁰¹⁹), are central to our understanding of how species evolve and diversify through time. These ‘evolutionary hotspots’ often harbor both old and new lineages that played a key role in generating extant species richness, thus representing a reservoir of evolutionary history and genetic diversity (Molina‐Venegas et al., ²⁰¹⁷; Zhang et al., ²⁰²¹). Among the known diversification centers, mountains contribute disproportionately to the abundance of biodiversity on Earth (Rahbek et al., ^2019b). Mountain regions represent the centers of diversity for many organisms, and studies of clades with high degrees of endemism in mountains can enhance our ability to reconstruct the evolutionary and biogeographic history of these regions (Dagallier et al., ²⁰²⁰; Perrigo et al., ²⁰²⁰; Chiocchio et al., ²⁰²¹). This is due, in part, to mountains providing novel and ecologically diverse habitats, functioning as refugia during cycles of glaciations, and acting both as corridors and as barriers of species dispersal (Rahbek et al., ^2019a). Furthermore, the repeated upward and downward movements of mountain habitats and climatic zones, particularly with respect to climatic dynamics in the Quaternary Period, lead to rapid range shifts bringing previously isolated taxa into contact and potentially triggering admixture and hybridization (Comes & Kadereit, 1998; Antonelli et al., ²⁰¹⁸; Rahbek et al., ^2019a). Consequently, admixture and hybridization may be among the key processes responsible for higher plant diversity in mountains (Stebbins, 1959; Soltis & Soltis, ²⁰⁰⁹; Baack et al., ²⁰¹⁵; Yang et al., ²⁰¹⁹; Stubbs et al., ^2020b).

Recurring admixture and speciation between previously isolated lineages is well documented in several montane plant clades such as Primula sect. Aleuritia (Primulaceae, Guggisberg et al., ²⁰⁰⁹), Ranunculus (Ranunculaceae, Emadzade et al., ²⁰¹⁵), Cupressus (Cupressaceae, Ma et al., ²⁰¹⁹), Micranthes (Saxifragaceae, Stubbs et al., ^2020a), and Polemonium (Polemoniaceae, Rose et al., ²⁰²¹). This phenomenon presents a challenge for reconstructing evolutionary relationships as strictly bifurcating hierarchies (Funk, 1985). Thus, admixture and its genomic consequences must be taken into consideration when reconstructing the evolutionary histories of plant clades. For example, recently admixed individuals may exhibit greater genome‐wide similarity with a distantly related introgressive donor lineage than with the population from which they are derived (Runemark et al., ²⁰¹⁹). Disentangling network‐like evolutionary histories may require not only sampling many different regions across genomes, as is common for multispecies coalescent (MSC) methods (Xu & Yang, 2016; Sukumaran & Knowles, ²⁰¹⁷), but also incorporating information from distinct genomic compartments. For instance, plastomes and regions of suppressed recombination, such as inversions and supergenes, can retain historical evolutionary patterns that have been elsewhere eroded by admixture and recombination (Brandvain et al., ²⁰¹⁴; Schumer et al., ²⁰¹⁸; Li et al., ²⁰¹⁹). However, comparative phylogenetic approaches examining evolutionary history across entire genomes (e.g. whole‐genome resequencing (WGR)) remain scarce.

Primula (primroses, Primulaceae) is ideally suited for studying complex evolutionary processes due to its unique biological and geographic attributes and the extensive scientific knowledge accumulated for this mostly northern and montane clade of c. 550 species (de Vos et al., ^2014a). Primroses have been extensively studied in terms of their reproductive, floral, and pollination biology, as well as their ecology, genetics, biogeography, and taxonomy for over a century (Darwin, 1862; Ernst, ¹⁹²⁵; Smith & Fletcher, ¹⁹⁴⁷; Valentine, ¹⁹⁵²; Richards, ²⁰⁰³; Theodoridis et al., ²⁰¹³; de Vos et al., ^2014b; Cocker et al., ²⁰¹⁸; Keller et al., ²⁰²¹; Mora‐Carrera et al., ²⁰²¹). This persistent interest in Primula is due, in part, to primroses serving as the primary model for the study of heterostyly, a complex genetic, reproductive, and morphological syndrome that occurs in at least 28 angiosperm families (Ganders, 1979). In primroses, heterostyly is controlled by the S‐locus, or heterostyly supergene, which is composed of five adjacent genes held together via hemizygosity, preventing recombination between them (Nowak et al., ²⁰¹⁵; Huu et al., ²⁰¹⁶; Li et al., ²⁰¹⁶; Potente et al., ²⁰²²). The S‐locus is present only in the genome of S‐morph individuals, characterized by flowers with short styles and long anthers, while it is absent in L‐morph individuals, characterized by flowers with the reciprocal spatial arrangement of female and male sexual organs.

Within primroses, the relationships of Primula sect. Primula, a section consisting of seven species and 13 subspecies (all diploid except for Primula grandis) and a high degree of endemism in the Caucasus region, remain unresolved. Importantly, investigating relationships in Primula sect. Primula is of significance because it has been hypothesized that this clade originated in an eastern refugium in the Caucasus and spread westward into Europe (Richards, 2003; Volkova et al., ²⁰²⁰). Although the Caucasus region is considered an important reservoir of biodiversity, studies investigating speciation patterns within this region, its biogeographic role in Eurasia, and overall knowledge of the biodiversity of this area are limited (Myers et al., ²⁰⁰⁰; Zazanashvili, ²⁰⁰⁹; Koch et al., ²⁰¹⁶; Mumladze et al., ²⁰²⁰).

Previous phylogenetic analysis suggested a lack of reciprocal monophyly and conflict in the evolutionary relationships within this clade among the three widespread species, that is, Primula elatior (P. el.), Primula veris (P. ve.), and Primula vulgaris (Schmidt‐Lebuhn et al., ²⁰¹²). This study was hampered by three factors: it employed limited geographic sampling and only three genetic markers (rps16‐trnK, trnS‐trnG, and internal transcribed spacer); the relatively young age of Primula sect. Primula (2.46 million years ago (Ma), 95% highest posterior density: 3.268–1.527 Ma; de Vos et al., ^2014a) and short branch lengths within this clade, which might imply a recent, rapid radiation and/or incomplete lineage sorting (ILS); and hybridization that has been extensively documented in both natural populations and artificial crosses (Darwin, 1868; Smith & Fletcher, ¹⁹⁴⁷; Valentine, ^{1947, 1952, 1961}; Woodell, ¹⁹⁶⁹; Kálmán et al., ²⁰⁰⁴; Keller et al., ^{2016, 2021}; Tendal et al., ²⁰¹⁸). Overall, it is unclear whether the recovered non‐monophyly was the result of limited sampling, limited genetic resolution, ILS, and/or indicative of reticulate evolution.

In this study, we conducted widespread geographic sampling of Primula sect. Primula across Eurasia (Fig. 1a) and generated novel WGR data for 106 specimens to assess the evolutionary relationships and causal processes that underlie the previously identified non‐monophyly in this clade. We specifically investigated whether the evolution in Primula sect. Primula was shaped by speciation in the Caucasus biodiversity hotspot before westward dispersal into Europe, reflecting an ‘out‐of‐Caucasus’ biogeographic pattern (Richards, 2003; Volkova et al., ²⁰²⁰). Compared to previous research that found the three widespread species to be non‐monophyletic, our analyses with genome‐scale data recovered two of these three species as monophyletic. Our analyses suggest that the remaining non‐monophyly in this section is primarily driven by past admixture between lineages in the Caucasus and confirmed the Caucasus as the center of origin for Primula sect. Primula. The research presented here highlights the importance of this understudied biodiversity hotspot in shaping species phylogenetic history and diversity and demonstrates that dense genomic and extensive geographic sampling can provide resolution in recalcitrant groups.

Fig. 1.

Overview of Primula sect. Primula. (a) Approximate distribution of the three widespread Primula sect. Primula species across Eurasia. Colored shapes indicate sampling localities and correspond to species and subspecies. Inset map shows Caucasus collections. Countries where samples were collected are identified by the two‐letter country code. (b) Species and subspecies used in this study. (c) Neighbor‐Net split graphs based on quartet distances from 3623 genome fragment trees with three widespread species highlighted. Non‐monophyletic subspecies are distinguished by the two‐letter country code. P. ve., P. veris; P. vu., P. vulgaris; P. el., P. elatior; EN, England; CH, Switzerland; SK, Slovakia; GE, Georgia; TR, Turkey; ES, Spain; IR, Iran; RU, Russia.

Materials and Methods

Geographic sampling

Targeted fieldwork was conducted in 2019 to collect accessions covering the taxonomic and geographic diversity of Primula sect. Primula L. (Fig. 1a,b). Every currently recognized species and subspecies of Primula sect. Primula (following Richards, ²⁰⁰³ and Schmidt‐Lebuhn et al., ²⁰¹²) was obtained from at least one population, and from every population we collected two individuals (Supporting Information Table S1). Additionally, eight outgroup accessions used in previous phylogenetic analyses (Schmidt‐Lebuhn et al., ²⁰¹²; de Vos et al., ^2014a) were obtained from the University of Zurich frozen tissue collection (Table S1). This resulted in 98 ingroup and eight outgroup samples, totaling 106 specimens (see Methods S1 for more details).

DNA extraction and WGR

For all samples, DNA was extracted using the Maxwell Rapid Sample Concentrator 48 and the Maxwell RSC PureFood GMO and Authentication Kit. Manufacturer's instructions were followed with the exclusion of Proteinase K. Library preparation and WGR were conducted at Rapid Genomics (Gainesville, FL, USA; rapid‐genomics.com). Samples were sequenced on the Illumina NovaSeq 6000 S4 (paired‐end 150 bp reads) to an average depth of 20×.

Read alignment, variant calling, and filtering

Raw paired‐end reads were mapped to the chromosome‐scale P. veris L. reference genome (haploid assembly, containing the S‐locus; Potente et al., ²⁰²²) using Bwa v.0.7.17 (Li & Durbin, 2009) and the ‐mem algorithm. BAM files were then sorted, marked for duplicates, and indexed using picard 2.7.1 (Broad Institute, 2019) and SAMtools v.1.3 (Li, 2011). Mapping to a high‐quality reference genome allows for an increase in the number of reads used, a more even distribution of mapped reads, and high precision in the location and order of sites and genomic regions of interest, such as repetitive sites (Lowry et al., ²⁰¹⁷; Pfeifer, ²⁰¹⁷).

Variant calling and filtering followed Ravinet et al. (2021). Variant calling was performed by BCFtools v.1.8 multiallelic caller (Li, 2011; Danecek et al., ²⁰²¹) and set to output only variant sites. Repetitive sequence regions were masked based on the annotation from Potente et al. (2022). Filtering was performed using Vcftools v.0.1.14 (Danecek et al., ²⁰¹¹) and set to remove indels and include only single nucleotide polymorphisms (SNPs) occurring in at least 80% of individuals, with a minimum site quality of 30, and a mean site depth between 5× and 60×.

Genomic fragment trees

We divided the genome into 50‐kb non‐overlapping genome fragments (GFs) from the 11 chromosomes in the P. veris assembly with custom scripts. We selected every other GF, resulting in 3623 alignments. Iq‐tree v.2.1.2 (Nguyen et al., ²⁰¹⁵) was used to estimate individual GF trees. For each alignment, Modelfinder (Kalyaanamoorthy et al., ²⁰¹⁷) was used to select the best nucleotide substitution model (adjusted for ascertainment bias) and otherwise default parameters were used. We generated statistical support for branches with 1000 ultrafast bootstraps (BS) (Minh et al., ²⁰¹³; Chernomor et al., ²⁰¹⁶).

Network analyses

We created a species network under the coalescent model with Network inference Algorithm via Neighbour Net Using Quartet distance (Nanuq; Allman et al., ²⁰¹⁹) from 3623 GF trees. Nanuq assumes a coalescent process, and therefore accommodates ILS and provides a model‐based interpretation of reticulation that justifies a hybridization network (Allman et al., ²⁰¹⁹). Species networks were estimated using the Nanuq algorithm as implemented in the R package mscquartets (Allman et al., ²⁰¹⁹). We selected an alpha of 5e−11 to impose a high standard for evidence of hybridization and a beta of 0.05, which requires stronger evidence for any resolution of the four‐taxon network (Methods S2). Using the software Splitstree4 (Huson & Bryant, 2006), a splits graph was inferred from the resulting Nanuq quartet distance matrix.

Phylogenetic inference

Phylogenetic relationships between samples were determined using maximum likelihood (ML). For the ML analysis, due to computational limitations, we randomly selected and concatenated 400 GFs from across the genome, spanning 20 million sites in total. This alignment was used for ML analysis in Iq‐tree with the same parameters as mentioned in the Genomic Fragment Trees section. We also constructed a MSC phylogeny. The MSC phylogeny, which assumes ILS as the primary driver of gene tree discordance, was estimated in the program Astral‐III v.5.7.3 (Zhang et al., ²⁰¹⁸; Rabiee et al., ²⁰¹⁹) with all 3623 GF trees used as input. As recommended, GF tree branches with support < 10 BS were collapsed before inputting into Astral‐III. Support for each branch was assessed with local posterior probability (LPP) (Sayyari & Mirarab, 2016).

Plastid phylogeny

Phylogenetic incongruence between plant nuclear and organellar genomes can provide evidence of introgression (Folk et al., ²⁰¹⁷; Lee‐Yaw et al., ²⁰¹⁹; Morales‐Briones et al., ²⁰²¹; Zhou et al., ²⁰²²). To investigate phylogenetic signals from the plastid genome, reads from each individual were mapped to the plastid genome of P. veris NC 031428 (Zhou et al., ²⁰¹⁶) using the BWA‐mem algorithm using default parameters. The plastid variant call format (VCF) required different filtering from the nuclear genome due to the plastome being both haploid and present in large copy numbers in each sample (Meleshko et al., ²⁰²¹). Filtering was performed by Vcftools to keep SNPs occurring in at least 50% of individuals with a minimum quality score of 15 and mean site and genotype depths between 10× and 400×. The filtered VCF retained 147 446 out of 152 682 sites. The plastome phylogeny was estimated using a ML and conducted in IQ‐Tree using the same parameters as for the nuclear analysis.

S‐locus phylogeny

When considering just recombination, the S‐locus is equivalent to both individual nonrecombining loci in the nuclear genome and the entire nonrecombining plastome. However, the S‐locus is larger (c. 300 kb) than most other nonrecombining individual nuclear loci, and therefore the S‐locus can provide far greater genetic resolution than much smaller individual nuclear loci. Additionally, unlike the maternally inherited plastome, the S‐locus can be either maternally or paternally inherited, and thus captures both maternal and paternal evolutionary patterns. Finally, as a result of the increased rate of fixation of mutations by drift in the S‐locus caused by its small effective population size and the decrease in genetic polymorphisms due to linkage between nearly neutral and non‐neutral mutations, the S‐locus is expected to have lower levels of ILS than other parts of the genome (Bergero & Charlesworth, 2009; Willyard et al., ²⁰⁰⁹; Hobolth et al., ²⁰¹¹; Pease & Hahn, ²⁰¹³; Martin & Jiggins, ²⁰¹⁷; Sackton & Clark, ²⁰¹⁹).

We inferred the phylogeny of the S‐locus to provide resolution and insight into the evolution of nonrecombining regions. The location of the S‐locus, containing 255 577 sites, was obtained from the P. veris genome assembly (Potente et al., ²⁰²²). We reduced our total alignment to only accessions containing the heterostyly supergene (i.e. plants identified in the field as being S‐morphs) and one outgroup. Then, after trimming the alignment to the location of S‐locus and masking repetitive regions, we removed specimens with a mean depth lower than 5×. We filtered this region to keep SNPs occurring in at least 75% of individuals, with a minimum quality score of 30, and a mean site and genotype depths of between 5× and 30×.

Discordance

Discordance analyses between the nuclear genome, chloroplast, and S‐locus were performed and visualized as tanglegrams using the cophylo function of the R package phytools (Revell, 2012). Additionally, we explored gene and sequence conflict using the gene‐concordance factor (gCF) and site‐concordance factor (sCF) in IQ‐Tree (Nguyen et al., ²⁰¹⁵; Minh et al., ²⁰²⁰) based on the GF trees, GF sequence alignments, and all phylogenies. We used default settings, except for increasing the number of sampled quartets for computing sCF to 10 000.

Gene flow and hybrid network analyses

We quantified the contribution of gene flow and hybridization in Primula sect. Primula using D‐statistics. D‐statistic tests (also known as ABBA‐BABA tests) detect gene flow in the presence of ILS through testing for an imbalance in the number of discordant SNPs among quartets of species (Patterson et al., ²⁰¹²). We used the package Dsuite, which applies D‐statistics in a phylogenetic context (Malinsky et al., ²⁰²¹). We specifically employed the Fbranch statistic (f _b) introduced in Malinsky et al. (2018), which combines the f ₄‐ratio with a species phylogeny to assign gene flow evidence to specific branches, with 0 being no admixture and 1 being completely admixed. The f _b statistic differs from Patterson's D in that it provides phylogenetically‐based estimates of excess allele sharing. Thus, f _b statistics are particularly powerful for complex groups because, unlike Patterson's D, it assigns gene flow evidence to specific branches in a phylogeny (Malinsky et al., ²⁰²¹; De‐Kayne, ²⁰²²). Dsuite was run with default parameters with the entire genomic dataset and the Dtrios and Fbranch commands. One drawback of phylogeny‐based methods for investigating admixture is their reliance on the ‘true’ species tree. Thus, we ran the Dsuite analysis for both the ML and MSC phylogenies. We used D _FOIL (Pease & Hahn, 2015) to confirm admixture events identified by the f _b statistic and checked for the possibility of ‘ghost’ lineages influencing our results following the methods of Hibbins & Hahn (2021b) (Methods S3).

Ancestral range estimations

Ancestral range estimations (ARE) were employed to test the hypothesis that Primula sect. Primula originated in the Caucasus (Richards, 2003; Volkova et al., ²⁰²⁰). Given that no fossil calibration points are available within our sampled taxa, we conducted an ARE analysis on the ML phylogeny. We chose this approach because multiple studies have found that ancestral state reconstructions are more accurate when performed on phylograms compared with time‐calibrated phylogenies (Litsios & Salamin, 2012; Cusimano & Renner, ²⁰¹⁴; Wilson et al., ²⁰²²). The ARE analyses were performed with a phylogeny reduced to 19 individuals representing monophyletic groups (Fig. 2). We delineated the following six geographic areas for ARE based on bioclimatic zones, current distribution patterns, and geographic barriers: Alps (A), British Isles (B), Caucasus (C), Iberian Peninsula (I), Carpathians I, and/or Siberia/northern Asia (S).

Fig. 2.

Phylogenomic analysis and ancestral range estimations resolve relationships and recover the Caucasus as the ancestral range of Primula sect. Primula. Topology constructed from the nuclear genome using maximum likelihood (ML). Branches with < 100% bootstrap (BS) are labeled. Pie charts at nodes indicate the probability of that node being recovered in only the Caucasus (black), Caucasus plus the Alps, British Isles, Iberian Peninsula, Carpathians, and/or Siberia/northern Asia (gray), or any areas outside of the Caucasus (white) in the ancestral range estimation analysis. Full pie charts in Supporting Information Fig. S6. Outgroups not shown. P. ve., P. veris; P. vu., P. vulgaris; P. el., P. elatior; EN, England; CH, Switzerland; SK, Slovakia; GE, Georgia; TR, Turkey; ES, Spain; IR, Iran; RU, Russia.

We used the phytools functions make.simmap and fitpolyMk to conduct an ARE analysis under a Bayesian framework. This method was selected because it allows for polymorphic states and can accommodate phylograms. We tested all four models: all transitions occurring at the same rate, backward and forward transitions occurring at the same rate, all transitions occurring at different rates, and the transient model where the acquisition and loss of a state occur at different rates. The best model was selected based on the Akaike information criterion. Stochastic mapping was simulated 1000 times, summarized, and plotted on the phylogram with the posterior probability (PP) of each node occurring in each range. We also investigated whether ARE with a time‐tree recovered similar ancestral ranges as those obtained with a phylogram. We constructed an ultrametric tree through relative dating by putting a tight normal prior on the root of Primula sect. Primula and conducted parallel analyses (Methods S4).

Isolation‐by‐distance

It is expected that populations become less genetically related with increasing geographic distance, that is, a pattern of isolation‐by‐distance (IBD) (Wright, 1943; Hutchison & Templeton, ¹⁹⁹⁹; Twyford et al., ²⁰²⁰). Intraspecific structure can result from IBD because geographically close populations tend to be more genetically similar than populations located farther away (Wright, 1943; Hutchison & Templeton, ¹⁹⁹⁹). We therefore tested for IBD using a Mantel test between the matrix of nuclear genetic distances and a matrix of geographic distances made from collection localities using the R package adegenet v.2.1.5 (Jombart, 2008; Jombart & Ahmed, ²⁰¹¹). If the Mantel test results in a significant P‐value (P < 0.05) and a strong correlation coefficient (r ≥ 0.7), this is suggestive of IBD and the potentially major role of geographic isolation in lineage diversification. We tested for IBD among all populations within P. veris, P. vulgaris Huds., and P. elatior (L.) Hill. Additionally, because the Caucasus endemics were recovered as sister to the Caucasus P. elatior subspecies in the phylogenetic and network analyses (see the Results section), we tested for IBD in a group consisting of P. elatior subspecies plus all four Caucasus endemic species. To visualize the correlation between genetic and geographic distances, densities of distances were plotted using a two‐dimensional kernel density estimation.

Results

Read alignment, variant calling, and filtering

We generated WGR data for 98 Primula sect Primula accessions and eight outgroups (total 106 samples). Paired‐end reads were mapped to the haploid reference genome assembly of P. veris containing the S‐locus (Potente et al., ²⁰²²) using Bwa v.0.7.17 (Li & Durbin, 2009). On average, 96.1% ± 1.7 of ingroup reads aligned to the P. veris reference genome (91.8% ± 15.4 with outgroups). The average coverage for each ingroup sample was 20.2× ± 4.5 (19.5× ± 5.0 with outgroups). After base calling, the total dataset had 103 365 340 sites, and after filtering the number of high‐quality variant sites was 15 114 651.

Network analyses

We created a species network under the coalescent model with Nanuq from 3623 GF trees. Unlike other methods employing splits graphs, which should simply be viewed as exploratory devices and cannot be used to identify the underlying biological mechanisms causing network‐like relationships, Nanuq provides a firm model‐based interpretation of reticulation (Allman et al., ²⁰¹⁹). The resulting neighbor‐net splits graph showed high levels of reticulation along the backbone of Primula sect. Primula, and less reticulation at shallower nodes (Fig. 1c).

Phylogenetic analysis

The ML and MSC phylogenies were well resolved and received high support (100% BS, 1.0 LPP) at most nodes (Figs 2, S1). Although there was high branch support throughout both nuclear genome phylogenies, there was low gCF and sCF at deeper nodes in both phylogenies (Fig. S2). When both gCF and sCF are low, this is indicative of discordant signal (i.e. ILS and/or introgression), rather than stochastic error from limited information (Lanfear, 2018; Minh et al., ²⁰²⁰). Of note, there is overall a slightly higher gCF and sCF for the nuclear ML topology compared to the MSC topology. The ML and MSC phylogenies varied solely in the relationships of Primula megaseifolia, P. el. lofthousei, and P. el. pallasii (Figs 2, S1). Notably, gCF at these nodes was low for both phylogenies, while sCF was high for the ML topology and low for the MSC topology. Additionally, the branch supporting the sister relationship of P. megaseifolia, P. el. lofthousei, and P. el. pallasii in the MSC phylogeny is the only node along the backbone with < 1.0 LPP (=0.96).

Overall, both analyses were largely congruent and both supported P. veris and P. vulgaris as monophyletic, whereas P. elatior was non‐monophyletic. Furthermore, P. grandis, the one Caucasus endemic species for which we sampled more than one population, was also monophyletic. In addition to the non‐monophyly recovered in P. elatior, non‐monophyly was also recovered at the subspecies level. For instance, the autonymous subspecies (P. el. elatior, P. ve. veris, and P. vu. vulgaris) are all non‐monophyletic in the ML analysis, with the descendants of the most recent common ancestor (MRCA) of each of these widespread subspecies including the narrow endemic subspecies P. el. intricata (ES), P. ve. canescens (ES) plus P. ve. columnae (ES and CH), and P. vu. balearica (ES), respectively (Fig. 2). Consequently, in downstream analyses requiring species monophyly, these narrow endemic subspecies are included in the autonymous subspecies (see Fig. 2).

Chloroplast phylogeny

The phylogenetic analysis of the entire chloroplast genome resulted in a well‐resolved topology, where most branches had BS = 100% (Fig. S3). However, in the chloroplast topology, all three widespread species are recovered as non‐monophyletic (Fig. 3), similar to previous studies with only two plastid spacer regions (Schmidt‐Lebuhn et al., ²⁰¹²). Furthermore, two specimens of P. el. elatior are recovered among the P. vu. vulgaris samples, and one specimen of P. vu. vulgaris is recovered among P. el. elatior samples (Fig. S3). Additionally, the chloroplast topology is in strong conflict with the nuclear phylogeny, with incongruent relationships being recovered at both deep and shallow relationships (Fig. 3). The gCF and sCF for the plastid topology are extremely low with gCF being zero or close to zero at most nodes (Fig. S2), indicative of the high degree of incongruence between the nuclear and chloroplast topologies.

Fig. 3.

Tanglegram comparing topologies of S‐locus, nuclear genome, and chloroplast cladograms in Primula sect. Primula. Monophyletic groups are reduced to one tip per taxon. Full chloroplast and S‐locus phylogenies in Supporting information Figs S3, S4. P. ve., P. veris; P. vu., P. vulgaris; P. el., P. elatior; GE, Georgia; TR, Turkey.

S‐locus phylogeny

If ancient admixture occurred in the evolutionary history of Primula sect. Primula, we would expect the nuclear genome and S‐locus phylogenies to be incongruent at deep nodes in the phylogeny. Overall, the S‐locus topology is well resolved and largely aligns with the nuclear phylogeny at shallow nodes (Figs 3, S4). However, we did recover hard incongruence at deeper nodes. One example of deep incongruence pertains to P. megaseifolia. In the S‐locus topology, P. el. pallasii and P. el. lofthousei are sister to a clade containing P. megaseifolia. By contrast, in the nuclear genome topology, P. megaseifolia is sister to a clade containing all other species in the section, excluding P. grandis. Another example of deep incongruence is the relationships of P. vulgaris and P. veris. In the nuclear genome topology, P. vulgaris is sister to a clade consisting of Primula juliae, P. el. cordifolia, and P. el. lecuophylla, which is then sister to P. veris. However, in the phylogenetic analysis of the S‐locus, P. vulgaris is sister to a clade consisting of P. elatior plus the Caucasus species (excluding P. el. pseudoelatior), while P. veris is sister to all other species in the section.

Gene flow analyses

Low f _b values were recovered for many species (Fig. 4), corroborating the widespread reticulate evolution at deeper evolutionary time scales recovered in the network analysis (Fig. 1c). However, an excess of shared alleles was seen in multiple species pairs in Primula sect. Primula; therefore, we will focus on the two highest f _b values (f _b ≥ 0.140), signifying the highest excess of shared alleles (Fig. 4; Table S2). Of note, these same species pairs were also identified by high f _b values when using the MSC topology (Fig. S5). The highest f _b value was recovered between P. el. pseudoelatior and the ancestor of P. ve. macrocalyx and P. ve. columnae TR (f _b = 0.159). The next highest number of shared alleles was recovered between P. megaseifolia and the MRCA of the P. el. meyeri, P. el. elatior, and Primula renifolia clade (f _b = 0.140), indicative of admixture between P. megaseifolia and the MRCA and/or the descendants of the renifolia clade (P. el. meyeri, P. el. elatior, and P. renifolia). These admixture events were confirmed by D _FOIL (Table S3), and we did not recover any evidence suggesting the impact of a ‘ghost’ lineage on these results (Table S4). Finally, the f ₄‐ratio is prone to false positives (Eaton et al., ²⁰¹⁵), and admixture fractions are dependent on demography and thus can be skewed by demographic events. It is therefore possible that the recovered admixture results are artifacts of the analysis and/or a complex demographic history. However, it has been shown that the f ₄‐ratio, on which the f _b statistic is based, is robust under most demographic scenarios (Durand et al., ²⁰¹¹; Patterson et al., ²⁰¹²). Furthermore, our results remained robust under different combinations of taxa (Fig. S5) and alternative approaches such as network inference (Fig. 1c).

Fig. 4.

Evidence of gene flow in Primula sect. Primula. Results of the Fbranch (f _b) analysis in Dsuite with maximum likelihood (ML) nuclear genome phylogeny. The values in the matrix signify excess allele sharing between the branch identified on the tree on the y‐axis, which includes internal branches, and the species identified on the x‐axis, relative to its sister branch. The light violet to dark purple gradient in the matrix refers to excess allele sharing, and white squares correspond to 0, nonsignificant values, or correspond to tests that are not consistent with the phylogeny. The two squares marking the highest number of shared alleles are designated with A and B. The corresponding admixture event signified by that square is drawn onto each tree. (A) Excess of shared alleles between Primula megaseifolia and the ancestor of P. el. elatior, P. el. meyeri, and P. renifolia. (B) Excess of shared alleles between P. el. pseudoelatior and the ancestor of P. ve. macrocalyx and P. columnae. P. ve., P. veris, P. vu., P. vulgaris; P. el., P. elatior; GE, Georgia; TR, Turkey.

Ancestral range estimation

The ARE analysis suggests that there is strong support for the origin of Primula sect. Primula in the Caucasus (Figs 2, S6, S7; Table S5). The transient model was selected as the best fit to the data in the ARE analysis. This analysis supported an origin in the Caucasus for the MRCA of Primula sect. Primula, with the Caucasus being recovered with the highest PP of 0.482. The next most likely range, albeit with a much lower probability, also included the Caucasus and consisted of the Caucasus plus Iberian Peninsula (PP = 0.077) (Table S5). Additionally, all nodes along the backbone of the phylogeny identified the Caucasus as the ancestral range with the highest probability (Fig. 2). The chronogram analysis recovered congruent results and inferred the Caucasus as the ancestral range (Fig. S7).

Isolation‐by‐distance

Our analysis revealed strong IBD patterns within both P. vulgaris and P. veris, but not within P. elatior (Fig. 5). The IBD analyses for both P. vulgaris and P. veris recovered significant and strong correlations between genetic and geographic distance. Primula vulgaris showed a single high‐density cloud suggesting a continuous cline of differentiation. However, the IBD analysis with P. veris showed two distinct high‐density clouds, suggesting that these are differentiated populations. These discontinuities may be an artifact of our sampling or could be indicative of intraspecific structure. Overall, both P. vulgaris and P. veris conform to the expectation of IBD. By contrast, both analyses with P. elatior recovered three distinct high‐density regions and low correlation (r < 0.4) (Fig. 5). Additionally, both IBD analyses with the P. elatior subspecies indicated a patched pattern of differentiation among populations. Due to the non‐monophyly of P. elatior (Fig. 2), this result is not surprising. However, we evaluated IBD for P. elatior with the Caucasus endemic species expecting to recover a stronger signal of IBD due to the close phylogenetic relationships and geographic overlap of these species. In fact, we found the opposite pattern, whereby there was a lack of significance and a much weaker correlation in the P. elatior analyses that included the Caucasus endemics. This suggests that geographically proximal populations of these species in the Caucasus are genetically distant (Fig. 4, high‐density cloud in upper left corner of plot) and supports the role of biogeographic processes in species diversification.

Fig. 5.

Scatterplots with two‐dimensional kernel density estimation illustrating the relationship between genetic and geographic distance. We tested for isolation‐by‐distance (IBD) using a Mantel test. When correlations are significant (Mantel test, P < 0.05), regression lines are drawn. If the Mantel test results in a significant P‐value and a strong correlation coefficient (r ≥ 0.7), this is suggestive of IBD. A single high‐density cloud suggests a continuous cline of differentiation, while multiple discontinuous high‐density regions suggest differentiated populations. A significant and strong positive correlation between geographic and genetic distance is recovered for both Primula vulgaris and Primula veris, supporting a pattern of intraspecific IBD. By contrast, there was a (very) weak correlation (r < 0.4) between genetic and geographic distances for Primula elatior. Furthermore, despite the close phylogenetic relationships of P. elatior and the Caucasus endemics, there was a lack of significance and a very weak correlation in the analysis combining P. elatior, Primula grandis, Primula juliae, Primula megaseifolia, and Primula renifolia.

Discussion

We used WGR in conjunction with widespread geographic sampling to reconstruct a highly resolved Primula sect. Primula phylogeny and investigate the biogeographic patterns and role of the Caucasus biodiversity hotspot in the evolution of this section. Our results provide an updated evolutionary framework for Primula sect. Primula and identify the Caucasus as the center of origin for this clade. Using an array of analytical approaches, we found substantial evidence of reticulate evolution primarily driven by deep lineage admixture in the Caucasus (Figs 1, 2, 3). Overall, our results are consistent with the studies of other montane plants that find widespread reticulate evolution in clades occurring in mountain regions (Guggisberg et al., ²⁰⁰⁹; Emadzade et al., ²⁰¹⁵; Rose et al., ²⁰²¹) and highlight the importance of studying centers of diversification in resolving species evolutionary history.

Genome‐scale data resolve the evolutionary history of Primula sect. Primula and provide insight into the biogeography of Caucasus biodiversity hotspot

In previous analyses of Primula sect. Primula, both deep and shallow relationships were uncertain (Schmidt‐Lebuhn et al., ²⁰¹²). By contrast, both the ML and the MSC phylogenies from this study were well resolved and supported highly similar topologies (Figs 2, S1). However, the network analysis implies widespread reticulate evolution at deeper evolutionary time scales in this clade (Fig. 1c). Although all phylogenies are well resolved, both the ML and the MSC phylogenies have extremely short branch lengths along the backbone (Figs 2, S1). Due to the young age of this section, this could be indicative of a recent, rapid radiation, as seen in other groups of European Primulaceae (Soldanella, Comes & Kadereit, 2003; Primula sect. Auricula; Zhang et al., ²⁰⁰⁴; Androsace, Roquet et al., ²⁰¹³). Consequently, ILS, a well‐documented product of rapid radiations, likely played an important role in the evolutionary history of Primula sect. Primula, in addition to hybridization, a conclusion further supported by the low gCF and sCF values estimated throughout the phylogeny (Fig. S2). It is notoriously difficult to distinguish the signals of ILS and introgression (Cai et al., ²⁰²¹; Hibbins & Hahn, ^2021a; Esquerré et al., ²⁰²²), but by employing three programs that account for ILS (i.e. Nanuq, Astral, Dsuite), our results suggest that both introgression and ILS have played a significant role in shaping the inferred evolutionary relationships in this clade. Overall, the complexity of cladogenetic and tokogenetic processes within this clade is not well represented in a bifurcating topology.

Despite the fact that the reticulate processes shaping lineage evolution may not be adequately captured in the phylogenetic analysis, at shallower nodes the network and phylogenetic analyses support congruent results. One key finding concerning shallower relationships is whether the three widespread Primula sect. Primula species are monophyletic or not. In our analyses, based on the entire nuclear genome, we found that P. veris and P. vulgaris are monophyletic, while P. elatior is non‐monophyletic. This differs from previous research that found non‐monophyly in all three species (Schmidt‐Lebuhn et al., ²⁰¹²); it is therefore likely that the previously‐recovered non‐monophyly of P. veris and P. vulgaris was an artifact of limited phylogenetic information, caused by insufficient genomic data. By contrast, the non‐monophyly of P. elatior recovered in both studies seems indicative of reticulate evolution. Consequently, the evolutionary history of P. elatior is more complex, differing considerably from P. vulgaris and P. veris (see ‘Cytonuclear discordance is indicative of admixture in P. elatior ’ in the Discussion section).

It has been hypothesized, but never tested, that the Primula sect. Primula clade originated in an eastern refugia in the Caucasus and Asia and spread westward into Europe (Richards, 2003; Volkova et al., ²⁰²⁰). Our study provides multiple lines of evidence confirming this hypothesis. First, in the ARE analysis, the Caucasus region is identified as the ancestral range of Primula sect. Primula. Second, both P. veris and P. vulgaris show strong declining genetic similarity with increasing geographic distance, that is, IBD, correlated to the increased genetic distances between the Caucasian and the western European populations (Fig. 5). Taken together, these results support an ‘out‐of‐Caucasus’ model of evolution for the ancestral lineages of Primula sect. Primula and for P. veris and P. vulgaris. These findings are consistent with other studies, which have highlighted the role of the Caucasus biodiversity hotspot as a source for European populations (Hewitt, 2000; Grimm & Denk, ²⁰¹⁴; Volkova et al., ²⁰²⁰). However, the model of evolution for P. elatior is less straightforward, and processes of reticulation can be invoked to explain the evolution of this species. Specifically, our results imply that the non‐monophyly of P. elatior can be attributed, in large part, to admixture between the Caucasian P. elatior lineages and Caucasus endemic species (Figs 1c–4).

Cytonuclear discordance is indicative of admixture in P. elatior

Many of our results point toward gene flow as the cause of the non‐monophyly of P. elatior. Specifically, as cytonuclear discordance is indicative of hybridization and/or introgression (Chan et al., ²⁰²¹; Wang et al., ²⁰²¹), the incongruences between the nuclear and chloroplast topologies highlight multiple examples of reticulate evolution (Fig. 3). We will focus on three of the most striking cases of strongly supported (> 90% BS) incongruences and draw upon our other analyses to highlight putative cases of reticulate evolution.

Perhaps the most conspicuous example of cytonuclear discordance is seen in P. vu. sibthorpii and P. el. meyeri. In the nuclear phylogeny, both the Georgian and Turkish populations of each species are recovered in clades with their respective conspecifics (Figs 2, 3). However, in the chloroplast phylogeny, the Georgian populations of P. vu. sibthorpii and P. el. meyeri are recovered as sister taxa, while Turkish populations of each subspecies maintain congruent relationships to those seen in the nuclear topology (Fig. 3). We were unable to detect any other strong evidence of gene flow between P. vu. sibthorpii and P. el. meyeri, which aligns with studies showing organelle capture aligning with geography and occurring in the absence of detectable nuclear introgression (Rieseberg & Soltis, 1991; Stegemann et al., ²⁰¹²). Therefore, the surprising sister relationship of the Georgian populations of P. vu. sibthorpii and P. el. meyeri in the chloroplast topology is strongly indicative of plastome introgression between these subspecies.

Another hard incongruence between the nuclear and chloroplast topologies concerns the relations of the Caucasus‐endemic P. megaseifolia (Fig. 3). In the nuclear ML analysis, P. megaseifolia is sister to all of Primula sect. Primula, excluding P. grandis, while in the plastid phylogeny, P. megaseifolia is sister to P. el. pallasii (Fig. 3). In both nuclear phylogenies, P. megaseifolia is on a very short branch, which is indicative of ILS or introgression plus ILS. Due to the ML analysis being sensitive to ILS, it is quite possible that the recovery of P. megaseifolia as sister to the rest of the section is an artifact of ILS. However, the history of reticulate evolution suggested by topological incongruence recovered in the MSC, chloroplast, and S‐locus phylogenies was supported by other analyses. First, D‐statistics recovered a high level of shared alleles between P. megaseifolia and the MRCA of the renifolia clade (Figs 4, S5). Second, in the network analysis, which accommodates ILS, P. megaseifolia is recovered as an intermediate between P. el. pallasii and P. el. lofthousei and the renifolia clade (Fig. 1c).

All together, we have strong evidence for genetic exchange between the lineages consisting of P. megaseifolia, P. el. Pallasii, and P. el. lofthousei, and the renifolia clade. Although we cannot infer directionality from our methods, due to P. megaseifolia varying throughout our analyses between sister relationships with P. el. pallasii and P. el. lofthousei and the renifolia clade (Figs 1c, 3, 4), it is possible that the ancestor of P. megaseifolia was of hybrid origin between these two lineages. Of note, one requirement of speciation is reproductive isolation (Wang et al., ²⁰²¹), and P. megaseifolia is the earliest flowering species within Primula sect. Primula (Richards, 2003), a well‐established form of reproductive isolation in angiosperms (Lowry et al., ²⁰⁰⁸; Baack et al., ²⁰¹⁵).

A final example of reticulate evolution in Primula sect. Primula is P. el. pseudoelatior, a subspecies of P. elatior endemic to the Caucasus region. In every phylogenetic analysis, P. el. pseudoelatior was recovered as sister to P. veris, and in the Dsuite analysis, P. el. pseudoelatior had the highest amount of introgressed alleles of any species (Fig. 4), a result corroborated by D _FOIL (Table S3). Although both f _b statistics and D‐statistics explicitly attempt to distinguish ILS from introgression, the f _b considers admixture in a phylogenetic context. In our case, this allowed us to deduce that the introgressed alleles in P. el. pseudoelatior are shared with the Caucasus P. veris subspecies. The high gCF and sCF at this node (Fig. S2) and long branch lengths in both the ML and MSC analyses (Figs 2, S2) suggest that ILS alone cannot account for this discrepancy. Taken together, our analyses imply that the ancestor of P. el. pseudoelatior has undergone gene flow with the ancestor of the Caucasus P. veris subspecies. Overall, we have strong evidence from multiple analyses suggesting a prominent role for reticulate evolution in Primula sect. Primula and specifically P. elatior.

Nonrecombining genomic region recovers nuclear incongruence

We have established that admixture played a significant role in the evolution of Primula sect. Primula, which has resulted in a mosaic of phylogenomic signals in this clade. However, it has been demonstrated that the most accurate reconstruction of ancient branching events will be maintained within regions of the nuclear genome that undergo reduced recombination (Schumer et al., ²⁰¹⁸; Li et al., ²⁰¹⁹). Specifically, with complete lineage sorting and without gene flow between taxa, the nuclear and S‐locus phylogenies should recover highly concordant relationships, while incongruence between the phylogenies of these genomes would be evidence of ILS and/or hybridization (Pease & Hahn, 2013; Meleshko et al., ²⁰²¹; Rose et al., ²⁰²¹). We therefore provided the first phylogeny ever inferred from the heterostyly supergene including all species of Primula sect. Primula. Overall, we found the S‐locus phylogeny to be well resolved and largely congruent with the nuclear genome topology at shallow nodes. However, there was strong incongruence between the S‐locus and the nuclear genome topologies in the backbone relationships of Primula sect. Primula (Fig. 3). One potential cause of this incongruence is ILS. The S‐locus, with suppressed recombination and lower effective population size, generally exhibits less ILS, as seen in sex chromosomes and other regions of low recombination (Willyard et al., ²⁰⁰⁹; Pease & Hahn, ²⁰¹³; Rifkin et al., ²⁰¹⁹; Sackton & Clark, ²⁰¹⁹). The incongruence between the S‐locus and nuclear genome topologies could also be due to S‐locus introgression or to the nuclear genome including high‐recombination regions that are enriched for signals of gene flow, as documented in studies with both empirical (Brandvain et al., ²⁰¹⁴; Li et al., ²⁰¹⁹) and simulated (Schierup & Hein, 2000, p. 200; Leaché et al., ²⁰¹⁴) data. The latter explanation would suggest admixture at deeper evolutionary time scales in Primula sect. Primula, aligning with the results from the network analysis (Fig. 1c).

Conclusions

Genome‐scale data allow us to rigorously test hypotheses previously generated from analyses utilizing limited genetic datasets. In this study, we provided the first phylogenetic analysis for primroses using sequence data from the entire nuclear genome, the entire chloroplast genome, and the entire S‐locus, the latter representing the first ever clade‐wide phylogenetic analysis based on this supergene. Future work focusing on species demographic responses to Quaternary climate fluctuations would provide a deeper understanding of the timing and mode of lineage admixture and diversification, and thus identify specific biogeographic processes shaping the evolutionary history of this section (Theodoridis et al., ²⁰¹⁷; Folk et al., ²⁰¹⁹).

Author contributions

RLS, ST, BK and EC designed the research. RLS, BK, EM‐C, NY, EL‐B, GT, FC and JK collected the data. RLS analyzed the data. RLS, ST, BK, EM‐C, GP, EL‐B, DARE, NY and EC interpreted the data. All authors contributed to the writing of the manuscript. RLS, ST and EC contributed equally to this work.

Supporting information

Fig. S1 Multispecies coalescent phylogeny of Primula sect. Primula made in Astral‐III.

Fig. S2 gCF and sCF for all phylogenies.

Fig. S3 Chloroplast phylogeny of Primula sect. Primula.

Fig. S4 S‐locus phylogeny of Primula sect. Primula.

Fig. S5 Fbranch (f _b) analysis in Dsuite with multispecies coalescent topology.

Fig. S6 Ancestral range estimation under a Bayesian framework with phylogram.

Fig. S7 Ancestral range estimation under a Bayesian framework with chronogram.

Methods S1 Geographic sampling.

Methods S2 Species networks estimation with Nanuq.

Methods S3 Supplementary D‐statistic analyses.

Methods S4 Ancestral range estimation.

Table S1 Primula specimens used in the study.

Table S2 Fbranch values from Dsuite.

Table S3 Results from D _FOIL.

Table S4 D‐statistics, f4‐ratios, Z‐score, and f4‐ratio from Dsuite for all species trios.

Table S5 Posterior probabilities for every state in ancestral state estimation with phylogram.

Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

Data availability

New sequences are deposited in NCBI's sequence read archive (https://www.ncbi.nlm.nih.gov/sra); accession numbers: SAMN31083866–SAMN31083971 (BioProject ID: PRJNA885283). Sequence alignments, gene trees, and species trees are deposited in the Dryad Digital Repository: 10.5061/dryad.b2rbnzsgz.